From 0fe1051d253c3477567e9e6a3e8a7f32b7332e64 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 15:05:44 -0400 Subject: [PATCH 01/32] =?UTF-8?q?docs(lean4):=20rev=205=20=E2=80=94=20re-s?= =?UTF-8?q?cout=20Stage=203=20and=20split=20it=20into=203a=20and=203b?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Stages 1 and 2 landed (#160, #161). Re-scouting Stage 3 against `main` @ `46a1b8f` — six merged PRs past the rev-4 snapshot — produced three findings that change the plan and four that confirm it. Two were established by running Lua in a fresh `EditorState` rather than by grep, and are marked *probed* in §0.1. **Stage 3 violated this document's own splitting rule.** §4 says "no PR in this arc mixes a cross-cutting substrate change with Lean feature content" and "a reviewer looking at Stage 3 sees only Lean" — while §4's own risk column for Stage 3 read "two `lsp.lua` generalizations". Those cannot both be true. One generalization shipped as Stage 2; the other is Q#LN9's dispatch seams, which modify `handle_server_requests` — confirmed the only production drain of LSP events, since `LspManager::take_all_events` has no non-test caller. By the test that justified splitting Stage 2 out, that is cross-cutting substrate. Stage 3 is now 3a (seams + canonicalizer, no Lean) and 3b (the Lean server), strictly sequential. **The Lean resolver could not satisfy the contract Stage 2 documented.** #161 established that a configured root reaches `file_uri_for` verbatim and that the resulting URI is the affinity key. Probed: `pmacs.editor.file_path()` is not canonical — opening `/linkpkg/sub/./../sub/a.lean` through a symlink yields `/linkpkg/sub/a.lean`, lexical collapse only. No canonicalize binding is exposed to Lua, and `pmacs.project.detect` canonicalizes but returns nil without a marker. So one Lake package opened by two spellings would spawn two `lake serve` processes — the bug Stage 2 was built to prevent, re-entered through Stage 3's door. New Q#LN20 adds a synchronous `pmacs.fs.canonicalize`; it rides 3a, and it serves every future function-valued root rather than only Lean's. Two alternatives are recorded with why they were rejected — the `detect`-anchored walk in particular is incorrect, not merely inelegant. **`pmacs.fs.stat` is unusable in the resolver.** It is async and the resolver runs synchronously inside `ensure_server` ← `attach_buffer` ← `buffer.after-load`, with no coroutine to await on. Probed: `io` and `os` are exposed in the sandbox, so the marker walk uses `io.open` — the opposite of what a reader would assume, hence Q#LN8 now says so. One edge, also probed: `io.open` succeeds on a directory, so the walk reads a byte rather than testing for a handle, and acceptance 24a bites the version that does not. Confirmed rather than changed: Q#LN7's stop-before-respawn is necessary (default policy is OnCrash, the termination handler never consults the exit code, and `maybe_restart` has no attempt ceiling — a broken `lake` respawns forever; `stop()` setting `restart = Never` is what disarms it); the response seam works as specified, since `Response` events are pushed unconditionally and `send_request` returns the keying id. One confirmation narrowed the design. `handle_server_requests` builds its sid list from `attachments` and `push_event` is uncapped, so subscribers fire only for servers with a live attachment. That turns acceptance 34 into a reachable leak: killing the buffer with a request outstanding strands the registration behind a drain that no longer runs. The purge is now driven from both edges and 34 exercises the buffer-kill path, which is the one a user can reach. Also: §9 states the lane's coherence impact per COHERENCE §20 (journey steps, interaction islands, config registry, background attribution), including the honest note that 3b makes §2's step-3 grade marginally worse by adding one more instance of the silent-spawn-failure class. Three items are named in §6 rather than paid: the uncapped event queue, the dropped `cfg.restart`, and surfacing the spawn failure itself. Acceptance keeps every rev-4 number. The two split sections are bulleted with literal labels because a markdown ordered list renumbers from its first item, and 3b's criteria are non-contiguous; round 3's finding 4 was stale references surviving a renumber, and not renumbering is the cheaper way to not repeat it. Stale cross-references from the split were reconciled in the same pass, and `project_root_for`'s citation was corrected from 513 to 592 per COHERENCE §25. --- docs/lean4-mode-framing.md | 555 ++++++++++++++++++++++++++++++------- 1 file changed, 461 insertions(+), 94 deletions(-) diff --git a/docs/lean4-mode-framing.md b/docs/lean4-mode-framing.md index e1fe060..fc31e01 100644 --- a/docs/lean4-mode-framing.md +++ b/docs/lean4-mode-framing.md @@ -6,8 +6,9 @@ pmacs has no Lean support of any kind: `grep -rin lean` over `*.rs`, plain buffer — no grammar, no major mode, no comment syntax, no pair set, no server. -This lane closes that in seven stages. Stage boundaries are drawn where +This lane closes that in eight stages. Stage boundaries are drawn where the *substrate* changes, not where the feature list does — see §4. +§9 states the lane's coherence impact per `COHERENCE.md` §20. ## 0. Why this lane, why now @@ -24,7 +25,7 @@ the *substrate* changes, not where the feature list does — see §4. first consumer of a non-standard LSP method family. Stage 6 adds a severity-routing policy to `LspServerSpec`. - The user's stated north star is **matching or exceeding what VS Code - does with Lean**. §5's bet 6 scores honestly how close seven stages get + does with Lean**. §5's bet 6 scores honestly how close the eight stages get and names precisely what is still missing. Parallel-safety: Stage 1 touches `Cargo.toml`, `src/syntax.rs`, @@ -34,10 +35,13 @@ Stage 3 (the other open lane) touches `pmacs-gpu/*` and `src/semantic_render.rs`. None of the three footprints overlap; the only file Stage 1 shares with anything is `Cargo.toml`, at one line. -Stages 1 and 2 are independent of each other and **can** run as sibling -worktrees — they share no file. Per the #126/#127 lesson, that split is -recorded here, before either starts, rather than discovered during a -rebase. +Stages 1 and 2 were independent of each other and could have run as +sibling worktrees — they shared no file. Both have since landed (#160, +#161). **Stages 3a and 3b are not independent**: 3b's subscriber is +written against the seam 3a adds, and both touch +`builtin/runtime/lsp.lua`. They are strictly sequential — recorded here, +per the #126/#127 lesson, before either starts rather than discovered +during a rebase. ## 0.1 Revision history @@ -81,7 +85,7 @@ round 2 renumbered the stages, so a rev-1 "Stage 4" is now Stage 5.)* 5. **Q#LN8's resolver must honor the search boundary.** A Lua `lean-toolchain` walk that ignores `pmacs.project.search_boundary()` breaks the contract `detect_project_within` exists to enforce and makes - the Stage 3 outermost-root test non-hermetic. + the Stage 3b outermost-root test non-hermetic. ### Round 2 (rev 2 → rev 3) — scope expansion @@ -168,11 +172,94 @@ Six findings against the round-2 expansion. All revision edits. preserving user-supplied `env`/`settings`/`init_options`/`root`. 6. Wording: `\{}` expands to `{$CURSOR}`; `⦃⦄` comes from `\{{}}`. +### Round 4 (rev 4 → rev 5) — Stage 3 re-scout and split + +Stages 1 and 2 landed (#160, #161). Re-scouting Stage 3 against `main` +@ `46a1b8f` — six merged PRs past the rev-4 snapshot (#159–#164) — +produced three findings that change the plan and four that confirm it. +Every fact below was verified in a worktree at that commit; the two +marked *probed* were established by running Lua in a fresh +`EditorState`, not by grep. + +1. **Stage 3 violated this document's own splitting rule.** §4 says "no + PR in this arc mixes a cross-cutting substrate change with Lean + feature content" and "a reviewer looking at Stage 3 sees only Lean" — + while §4's own risk column for Stage 3 read *"two `lsp.lua` + generalizations."* Those cannot both be true. One of the two landed + as Stage 2; the other is Q#LN9's dispatch seams, which modify + `handle_server_requests` — confirmed the **only** production drain of + LSP events (`LspManager::take_all_events` has no non-test caller). By + the same test that justified splitting Stage 2 out, that is + cross-cutting substrate. **Stage 3 is now 3a (substrate, no Lean) and + 3b (Lean).** +2. **The Lean resolver could not satisfy the contract Stage 2 + documented.** #161 established that a configured root — string or + resolver return — must be a canonical absolute path, because it + reaches `file_uri_for` verbatim and that URI is the affinity key. + *Probed:* `pmacs.editor.file_path()` is **not** canonical. Opening + `/linkpkg/sub/./../sub/a.lean`, where `linkpkg` symlinks to + `pkg`, yields `/linkpkg/sub/a.lean` — lexical `.`/`..` collapse + only, symlinks unresolved. No canonicalize binding is exposed to Lua, + and `pmacs.project.detect` canonicalizes but returns nil without a + marker. So a Lean resolver walking up from the buffer's path returns + a non-canonical root, and one package opened by two spellings spawns + two `lake serve` processes — reintroducing precisely the bug Stage 2 + exists to prevent. New Q#LN20 adds `pmacs.fs.canonicalize`; it rides + 3a because it is substrate, and it retires the footgun for every + future function-valued root rather than only Lean's. +3. **`pmacs.fs.stat` is unusable in the resolver.** It is asynchronous — + `fs.lua:93` returns an awaitable handle — and the resolver runs + synchronously inside `ensure_server` ← `attach_buffer` ← the + `buffer.after-load` hook, where there is no coroutine to await on. + *Probed:* the `io` and `os` stdlib **are** exposed in the sandbox + (`type(io.open) == "function"`; `terminal.lua` already uses + `os.getenv`), and `io.open` returns nil for a missing path. So the + marker walk is implementable, but through the Lua stdlib rather than + the pmacs fs API — the opposite of what a reader would assume. + Q#LN8 now says so, with the one edge that matters: `io.open` + **succeeds on a directory**, so a bare existence check would accept + a `lean-toolchain` *directory* as a marker. + +Confirmations, recorded because each was load-bearing and unverified: + +4. **Q#LN7's "stop the failing server first" is necessary, not + defensive.** The spec default is `LspRestartPolicy::OnCrash`, and the + termination handler calls `should_restart(policy)` + (`matches!(OnCrash | Always)`) — which, unlike the + `termination_warrants_restart` helper beside it, never consults the + exit code. `maybe_restart` re-fires on every elapsed backoff with **no + attempt ceiling**, so a broken `lake` respawns forever. `stop()` sets + `restart = Never` (`src/lsp.rs:1349`), which is exactly what disarms + it. Acceptance 36 pins a real mechanism. +5. **The response seam works as designed.** `Response` events are pushed + unconditionally (`src/lsp.rs:2652`) — the typed-store absorb above + does not consume them — and reach Lua as `{kind = "response", + request_id = , method, result, error}`, with + `pmacs.lsp.send_request` returning that same numeric id. So + `on_response(sid, request_id, fn)` is keyable as specified. +6. **The seams' contract is narrower than rev 4 implied, and the + narrowing is load-bearing.** `handle_server_requests` builds its sid + list from `attachments`, and `push_event` appends with no cap. So a + subscriber fires only for a server with a live attachment, and an + unattached server's event queue grows unboundedly. This bites + acceptance 34 directly: kill the buffer with a request outstanding + and the pending purge never runs — the leak that criterion exists to + prevent. Q#LN9 now states the contract and acceptance 34 drives it + through the buffer-kill path rather than the server-death path alone. +7. **The `cfg.restart` gap is still open** (recorded landing #161): + `ensure_server` never forwards `pmacs.lsp.config[lang].restart` to + `pmacs.lsp.spawn`, so the field is silently dropped on auto-attach. + Stage 3b is the first stage that would benefit from setting it, and + Q#LN7 now records why it deliberately does not need it. + +Citation drift repaired per COHERENCE §25: `project_root_for` is +`builtin/runtime/lsp.lua:592`, not 513, and returns `root, source` +rather than a bare root. ## 1. What ships -Seven stages. The north star is VS Code parity; the honest statement of -where that lands is in §5, bet 6. +Eight stages, after round 4 split Stage 3. The north star is VS Code +parity; the honest statement of where that lands is in §5, bet 6. **Stage 1 — grammar, mode, and the editing table stakes.** `.lean` files highlight, carry a `lean4` major mode, and get comment-toggle and @@ -185,13 +272,21 @@ Independently valuable for every language pmacs supports; a prerequisite for Lean being usable across more than one Lake package. Split out precisely *because* it is cross-cutting — see §4. -**Stage 3 — the Lean language server.** `pmacs.lsp.config.lean4` drives +**Stage 3a — LSP dispatch seams and a path canonicalizer.** Pure +substrate, no Lean content, split from Stage 3 in round 4 for the reason +Stage 2 was: it changes machinery every language runs through. +`handle_server_requests` gains notification and response arms with a +pending-response purge, so a `send_request` reply is no longer drained +and dropped; `pmacs.fs.canonicalize` gives Lua the one primitive a +function-valued `config.root` needs to honor the canonical-path contract +#161 could only document. + +**Stage 3b — the Lean language server.** `pmacs.lsp.config.lean4` drives `lake serve` with a Lake-aware outermost root, a lazy toolchain probe and -a one-shot `lean --server` fallback, and a notification-subscription seam -so `$/lean/fileProgress` has an owner. Adds -`textDocument/waitForDiagnostics`. Diagnostics, hover, completion, -goto-definition, document symbols, and semantic tokens all arrive through -the existing typed surfaces. +a one-shot `lean --server` fallback, and subscribes `$/lean/fileProgress` +on 3a's seam. Adds `textDocument/waitForDiagnostics`. Diagnostics, hover, +completion, goto-definition, document symbols, and semantic tokens all +arrive through the existing typed surfaces. **Stage 4 — the Unicode input method.** Typing `\alpha` produces `α`, `\to` produces `→`, `\<>` produces `⟨⟩` with the point between them. @@ -215,6 +310,13 @@ panel. ## 2. Ground truth (scouted 2026-07-24, `main` @ `e745068`) +Stage 3's facts were **re-verified 2026-07-25 against `main` @ +`46a1b8f`**, six merged PRs later; what changed is recorded in §0.1's +round 4 rather than rewritten in place, so a reader can see which +claims moved. Facts for stages 4–7 still carry the 2026-07-24 date and +should be re-scouted before those stages are framed for +implementation. + ### 2.1 Crate facts (external, verified by downloading and reading both) Two candidate grammar crates exist. They are not close in quality. @@ -528,7 +630,7 @@ PATH, both are executable, and both fail. So: old, and lake working but the directory is not a Lake package. Only the third is a *version* question. - **Acceptance cannot assume a working Lean toolchain exists.** Every - Stage 3+ test runs against the fake LSP server; a live `lake serve` + Stage 3b+ test runs against the fake LSP server; a live `lake serve` smoke is PATH-gated *and* success-gated, following the #123 JSON/YAML provider-smoke pattern. @@ -706,6 +808,27 @@ consulted before configuring. the failing server *first*, then swaps the config, then spawns — the fallback is a fresh server, not a restart of the old one. + Round 4 verified this is necessary rather than defensive. The spec + default is `LspRestartPolicy::OnCrash` (`src/lsp.rs:165`), and the + termination handler calls `should_restart(policy)` — which, unlike the + `termination_warrants_restart` helper beside it, never consults the + exit code. `maybe_restart` re-fires on every elapsed backoff with **no + attempt ceiling**, so a broken `lake` respawns indefinitely. + `pmacs.lsp.stop` sets `restart = Never` on the way out + (`src/lsp.rs:1349`), which is precisely what disarms it. Acceptance 36 + is pinning a live mechanism, not a hypothetical one. + + **Why the latch does not just set `restart = "never"` on the spawn.** + It cannot: `ensure_server` never forwards `cfg.restart` to + `pmacs.lsp.spawn` — `lua_to_lsp_spec` reads the key but the spawn + table never sets it — so the field is silently dropped on every + auto-attach today. That gap was found landing #161 and is not Stage + 3's to close (it changes behavior for every language that has set + `restart` believing it worked; `statusline_segments_acceptance` a12 is + one such caller). The stop-then-spawn ordering is correct regardless of + how that gap is eventually resolved, which is the reason to prefer it + over a fix that depends on the gap closing first. + **The swap is a field update, not a table replacement.** It rewrites only `command` and `args`, preserving any user-supplied `env`, `settings`, `init_options`, and `root` on `pmacs.lsp.config.lean4`. A @@ -732,18 +855,60 @@ fallback. That is a one-line status message, once per session, and it buys not blocking every other user's first attach behind a process round-trip. +**Attribution (COHERENCE §9).** The probe is background work that spawns +an OS process, and `ProcessSpec.label` is the only identity a process +carries — caller-supplied and unvalidated, but it is what +`pmacs.process.list` renders. The probe spawns as `lean:lake-version-probe` +rather than inheriting a default, so a user who looks at the process list +while wondering why their editor touched `lake` finds an answer with an +owner in it. Both the probe's verdict and the latch firing report through +`pmacs.editor.set_status` — the channel that exists — per §1.2's rule and +its corollary: each is pinned by a test that observes the channel, since a +report through `pmacs.error` would be a dead sixteenth call site. + No `init_options`. Per §2.8, `hasWidgets?` defaults to false and that is the correct value for a client that reads plain goals out of standard messages. ### Q#LN8 — Lake-aware root via a **function-valued** `config.root` -Generalize `project_root_for` (`builtin/runtime/lsp.lua:513`) so -`pmacs.lsp.config[lang].root` may be a `function(path) -> string|nil` as -well as a string, and implement Lean's resolver in -`builtin/runtime/lean.lua`: walk up from the file's directory collecting -every ancestor containing `lean-toolchain`, and return the **outermost**; -fall back to `pmacs.project.detect`, then the file's directory. +**The generalization landed in Stage 2 (#161).** `project_root_for` is +now `builtin/runtime/lsp.lua:592` and returns `root, source`; +`config[lang].root` already accepts a `function(path) -> string|nil`, +with per-directory memoization keyed weakly on the resolver itself. What +remains for Stage 3b is Lean's resolver in `builtin/runtime/lean.lua`: +walk up from the file's directory collecting every ancestor containing +`lean-toolchain`, and return the **outermost**; decline (return nil) when +there is none, which falls through to `pmacs.project.detect` and then the +file's directory. + +**How the walk tests for the marker — and why not the obvious way.** +`pmacs.fs.stat` is asynchronous: it returns an awaitable handle +(`builtin/runtime/fs.lua:93`) that only settles under `:await()` inside a +coroutine. The resolver has no coroutine. It runs synchronously inside +`ensure_server` ← `attach_buffer` ← the `buffer.after-load` hook, so +awaiting is not merely slow there, it is unavailable — and blocking the +attach on filesystem I/O is the cost rev 1 refused for the probe. The +walk therefore uses the **Lua stdlib**: `io.open(dir .. "/lean-toolchain", +"r")`, which returns nil for a missing path. Round 4 probed that `io` and +`os` are exposed in the sandbox rather than assuming it; `terminal.lua` +already depends on `os.getenv`. + +One edge, probed: **`io.open` succeeds on a directory** (the handle opens; +`read` returns nil without raising). A `lean-toolchain` *directory* would +therefore read as a marker. The resolver reads one byte and treats a +non-nil read as the marker, so a directory declines — an `io.open` truth +test alone would be wrong, and wrong silently. + +**The result must be canonical.** #161's contract: a configured root +reaches `file_uri_for` verbatim and that URI is the affinity key, so two +spellings of one package are two servers. The path handed to the resolver +is *not* canonical (round 4, finding 2), and Lua had no canonicalizer — +hence Q#LN20. The resolver canonicalizes the file's directory **once**, +before the walk, and strips components from there: every ancestor of a +canonical path is itself canonical, so one call suffices. If +canonicalization fails (a deleted file, a broken symlink), the resolver +declines rather than returning a path it cannot vouch for. **The walk stops at `pmacs.project.search_boundary()`.** This is not optional politeness: `detect_project_within` (`src/project.rs:213`) exists @@ -777,7 +942,7 @@ write-only API from Lua.** Rev 2 specified only the notification half. That was a hole, since Q#LN16 (`waitForDiagnostics`), Q#LN19 (`imports` / `importedBy`), and Q#LN12's typed goal request all await replies. Both halves ship in -Stage 3. +Stage 3a. ```lua pmacs.lsp.on_notification(method, fn) -- fn(sid, params); persistent @@ -808,10 +973,70 @@ directions: a Lean subscriber must not cause `workspace/applyEdit` to be missed, and a raising subscriber must not stop later events in the same drain. -Stage 3 registers `$/lean/fileProgress` on the notification seam and +**The seam's contract, stated because round 4 found it narrower than rev +4 implied: subscribers fire only for servers with a live buffer +attachment.** `handle_server_requests` builds its sid list from +`attachments`, so a server with no attached buffer is never drained — and +`push_event` appends with no cap, so that server's queue grows +unboundedly. Both facts are pre-existing and neither is Stage 3a's to +fix, but the second one turns the first into a leak with a name: **a +buffer killed while a request is outstanding never runs the purge**, +because the purge rides the drain that the attachment was gating. That is +the exact failure acceptance 34 exists to prevent, reachable through the +ordinary `C-x k`. So the purge is driven from both edges — the server-death +transition *and* attachment teardown — and acceptance 34 exercises the +buffer-kill path, which is the one a user can actually reach. + +The uncapped queue is recorded as a named deferral (§6) rather than fixed +here: bounding it is a policy question about which events may be dropped, +and answering it inside a seam PR would be the kind of smuggling §4 +forbids. + +Stage 3b registers `$/lean/fileProgress` on the notification seam and `waitForDiagnostics` on the response seam; stages 5 and 7 use the response seam for `plainGoal` and the hierarchy calls. +### Q#LN20 — `pmacs.fs.canonicalize` (Stage 3a) + +A synchronous binding wrapping `std::fs::canonicalize`, returning the +resolved absolute path or nil. Roughly fifteen lines. + +It exists because #161 documented an obligation Lua cannot discharge. A +configured root — string or resolver return — is fed to `file_uri_for` +verbatim, and that URI is the server-affinity key; the `"detected"` arm is +canonicalized for free because `pmacs.project.detect` canonicalizes before +walking, but the `"config"` arm is not. Round 4 probed that +`pmacs.editor.file_path()` collapses `.` and `..` lexically while leaving +symlinks intact, so a resolver walking up from it returns a non-canonical +root. Opening one Lake package through a symlinked path and through the +real path would spawn two `lake serve` processes — the bug Stage 2 was +built to prevent, re-entered through Stage 3b's door. + +**Synchronous, deliberately, and this is the one thing to get right.** +The whole reason `pmacs.fs.stat` cannot serve here is that it is async +(Q#LN8), so a canonicalizer that returned an awaitable would fail for the +same reason and leave the obligation undischarged. It is one `stat`-class +syscall on a path the editor is already opening; `pmacs.project.detect` +performs the same work synchronously today, on the same hook, so this +adds no blocking class that the attach path does not already have. + +Why this rather than the two alternatives considered in round 4: + +- *Accept it as a named degradation* — document that a symlinked open + spawns a second server and pin the behavior. Rejected: it reopens the + defect Stage 2 closed, and the failure is invisible (two servers, both + apparently working, twice the memory, diagnostics split between them). +- *Anchor the walk on `pmacs.project.detect`'s canonical root* — free, no + new surface. Rejected as incorrect, not merely inelegant: `detect` is + innermost-wins over its own marker set, so with `.git` at `~/code` and + the Lake package at `~/code/proj`, anchoring at `~/code` and walking + *up* never sees `~/code/proj/lean-toolchain`. It resolves the wrong root + in a layout that is entirely ordinary. + +The binding is general, not Lean-shaped: it serves every future +function-valued `root`, and it is what lets #161's doc comment stop +warning about a footgun and start naming a fix. + ### Q#LN10 — Stage 4 mechanism: one shared provenance read, not two The hazard is §2.6 — `take_typed_edit()` is one-shot and `pair.lua` @@ -922,8 +1147,9 @@ stage numbers and was wrong three ways): | Stage | Rust | |---|---| | 1 | `Cargo.toml` + `BUILTIN_LANGUAGES` entry + Q#LN4's four capture entries | -| 2 | `lsp.list()` row builder (`mod.rs:9919`) | -| 3 | **none** — Lua only | +| 2 | `lsp.list()` row builder (`mod.rs:9926`) | +| 3a | `pmacs.fs.canonicalize` (Q#LN20) — the seams themselves are Lua only | +| 3b | **none** — Lua only | | 4 | **none** — Lua only | | 5 | `request_plain_goal` + its binding | | 6 | `LspServerSpec` severity-policy field and its publish-path honoring | @@ -1045,7 +1271,7 @@ elaboration is memory-hungry. rust-analyzer has the same property and no editor caps it by default. No cap ships here; `pmacs.lsp.stop` is the manual escape, and an LRU reaping policy is named in §6. -### Q#LN16 — `textDocument/waitForDiagnostics` (Stage 3) +### Q#LN16 — `textDocument/waitForDiagnostics` (Stage 3b) A plain request (no position, so no `outbound_position` concern — Q#LN12 does not apply). It resolves when the server has finished elaborating the @@ -1125,28 +1351,48 @@ never lands. |---|---|---|---| | 1 | grammar, mode, comments, pairs, md fences | new crate; **global capture table** | — | | 2 | multi-root server affinity | **`ensure_server`, shared by every language** | — | -| 3 | `lake serve` + probe/latch, Lake root, notification seam, `waitForDiagnostics` | two `lsp.lua` generalizations | 1, 2 | +| 3a | notification/response seams + purge; `pmacs.fs.canonicalize` | **the shared event drain, run by every language** | — | +| 3b | `lake serve` + probe/latch, Lake root, `waitForDiagnostics` | none — Lean-only files plus one config entry | 1, 2, 3a | | 4 | Unicode input method | **refactors `pair.lua`'s provenance read** | 1 | -| 5 | goal panel | new typed LSP request; panel adopter | 3 | -| 6 | `#eval` / `#check` output channel | **new `LspServerSpec` policy field** | 3, 5 | -| 7 | module hierarchy | listview adopter + one typed Rust request | 3 | +| 5 | goal panel | new typed LSP request; panel adopter | 3a, 3b | +| 6 | `#eval` / `#check` output channel | **new `LspServerSpec` policy field** | 3b, 5 | +| 7 | module hierarchy | listview adopter + one typed Rust request | 3a, 3b | -Three of the seven carry risk that is *not* about Lean — stages 1, 2, and -6 each change something every language touches. That is the organizing -principle of the split: **no PR in this arc mixes a cross-cutting -substrate change with Lean feature content.** A reviewer looking at Stage -2 sees only `ensure_server`; a reviewer looking at Stage 3 sees only Lean. +Four of the eight carry risk that is *not* about Lean — stages 1, 2, 3a, +and 6 each change something every language touches. That is the +organizing principle of the split: **no PR in this arc mixes a +cross-cutting substrate change with Lean feature content.** A reviewer +looking at Stage 2 sees only `ensure_server`; a reviewer looking at Stage +3b sees only Lean. + +Round 4 found Stage 3 breaking that rule while stating it — the row above +used to read "two `lsp.lua` generalizations" for a stage the prose called +Lean-only. One generalization shipped as Stage 2; extracting the other as +3a is what makes the claim true again. The rule is only worth writing +down if it survives contact with a stage that is inconvenient to split. Ordering notes: - **Stage 2 has no Lean in it and could ship independently of this arc.** It is sequenced here because Lean is the language that makes its absence - a correctness bug rather than an inconvenience, and because Stage 3's + a correctness bug rather than an inconvenience, and because Stage 3b's acceptance would otherwise have to encode the broken behavior. -- **Stage 4 does not depend on stages 2–3** and could run in parallel, but - should not: both touch `lsp.lua`/`pair.lua`-adjacent runtime files, and - the #126/#127 lesson is that parallel-safety requires the file split be - agreed *before* either lane starts. Sequential is cheaper. +- **Stage 3a likewise has no Lean in it**, and the same reasoning applies + one level down: the response seam is a hole in `send_request` for every + language — Lean is merely the first caller that needs a reply. It is + sequenced before 3b because 3b's `waitForDiagnostics` and file-progress + subscription both consume it, and because a Lean PR that also rewrote + the shared drain could not be reviewed on either axis. +- **3a and 3b cannot run as sibling worktrees.** 3b's Lean subscriber is + written against the seam 3a adds, and both touch + `builtin/runtime/lsp.lua`. Unlike stages 1 and 2, this pair is strictly + sequential — recorded here, per the #126/#127 lesson, rather than + discovered in a rebase. +- **Stage 4 does not depend on stages 2, 3a, or 3b** and could run in + parallel, but should not: both touch `lsp.lua`/`pair.lua`-adjacent + runtime files, and the #126/#127 lesson is that parallel-safety + requires the file split be agreed *before* either lane starts. + Sequential is cheaper. - **Stage 6 depends on Stage 5** only for the read-only generated-buffer and panel machinery, which Stage 5 establishes. If Stage 5 slips, Stage 6 can carry that machinery itself at the cost of duplicating it. @@ -1185,7 +1431,7 @@ Stated so they can be scored, per house style. inside `buffer.after-edit` re-enters the hook in a way pairing does not already survive. Confidence: medium — pairing does the same thing, but over a single codepoint rather than a multi-byte span. -6. **These seven stages reach rough VS Code parity for everything except +6. **These eight stages reach rough VS Code parity for everything except the interactive infoview.** Scored honestly rather than aspirationally. What lands: highlighting, goal view, Unicode input, diagnostics, hover, completion, goto-definition, symbols, semantic tokens, `#eval` @@ -1225,6 +1471,18 @@ What remains deferred: unbounded `lake serve` growth possible. No editor caps this by default and pmacs will not either in this arc, but the policy question is now live in a way it was not before. +- **The uncapped LSP event queue** — `push_event` appends without a + bound, and `handle_server_requests` drains only servers with a live + buffer attachment, so an unattached server's events accumulate for the + life of the session (round 4, finding 6). Bounding it means deciding + which events may be dropped, which is a policy question with + user-visible consequences for diagnostics and progress; Stage 3a states + the seam's contract around the behavior rather than changing it. +- **Forwarding `cfg.restart` through `ensure_server`** — read by + `lua_to_lsp_spec`, never set by the spawn table, so silently dropped on + every auto-attach (found landing #161). Fixing it changes behavior for + every language whose config sets the field believing it works. Q#LN7 is + designed not to need it. - **Block-comment toggle** (`/- -/`) and **docstring awareness** (`/-- -/`) — confirmed as owned by the comment arc's framing, not this one. @@ -1309,58 +1567,100 @@ What remains deferred: the markerless one's server carries the fallback directory as `cwd` while matching on a nil affinity key. -**Stage 3 — the Lean language server** +**Stage 3a — dispatch seams and the canonicalizer (no Lean content)** -22. Opening a `.lean` file inside a Lake package spawns one server with - `cwd` and `rootUri` at the package root. -23. **Outermost-root pin:** a file under - `/.lake/packages/dep/…` whose ancestor chain contains two - `lean-toolchain` files resolves to ``, not to `dep`. Run with - `pmacs.project.set_search_boundary` at the fixture root so the - assertion is hermetic. -24. **Boundary pin:** with the search boundary set at the fixture root, a - `lean-toolchain` planted in an ancestor *above* the boundary is not - reached — the resolver stops at the boundary rather than walking past - it. -25. A string-valued `pmacs.lsp.config.lean4.root` still works — the Q#LN8 - generalization is strictly additive. -26. `didOpen` carries `languageId = "lean4"`. -27. **Fallback-latch pin (Q#LN7):** a `lake` stub that exits non-zero — - reproducing §2.9's shimmed-elan state — causes exactly **one** restart - against `lean --server`, and a second failure surfaces an error rather - than looping. The latch does not re-arm within the session. -28. **Probe pin:** a `lake` stub reporting version 3.0.0 triggers the - fallback; one reporting 3.1.0 does not. A stub that never exits does - not block the attach — the optimistic `lake serve` spawn proceeds. -29. A `$/lean/fileProgress` notification delivered through the fake server - reaches a registered `on_notification` subscriber. -30. **Dispatch-integrity pin:** with a Lean subscriber registered, a - `workspace/applyEdit` request in the same drain is still handled — no - event is stolen. -31. A subscriber that raises does not prevent later events in the same - drain from being processed. -32. **Response-seam pin (Q#LN9).** A `send_request` reply reaches its - registered `on_response` one-shot, and the one-shot is **removed - before** invocation — a raising handler is not re-entered. Bites - against rev 2, where no Lua consumed `ev.kind == "response"` at all - and the reply was dropped. -33. **Response dispatch-integrity pin.** With a response subscriber - registered, `workspace/applyEdit` in the same drain is still handled; - a raising response handler does not stop later events in that drain. - Mirrors the notification-side pins above. -34. **Pending-purge pin.** A server that dies with a response outstanding - invokes the pending one-shot with an error and clears it — the - registration does not leak and the awaiting caller does not hang. -35. **Config-preservation pin (Q#LN7).** After the fallback latch fires, - user-supplied `env` / `settings` / `init_options` / `root` on - `pmacs.lsp.config.lean4` survive; only `command` and `args` change. -36. **No-respawn-loop pin.** The latch stops the failing server before - spawning the fallback, so `RestartPolicy` does not respawn the broken - command underneath it. -37. `textDocument/waitForDiagnostics` resolves through the response seam - (Q#LN16). **PATH-and-success-gated live smoke:** if `lake serve` - starts successfully a real elaboration completes and diagnostics - arrive; skipped otherwise, never failed. +Driven against `pmacs_fake_lsp` through an already-shipped language, for +the same reason Stage 2's suite was: the drain is shared by every +language, and a suite that reaches it only through Lean would understate +the blast radius. + +- **29.** A notification delivered through the fake server reaches a registered + `on_notification` subscriber. +- **30.** **Dispatch-integrity pin:** with a subscriber registered, a + `workspace/applyEdit` request in the same drain is still handled — no + event is stolen. +- **31.** A subscriber that raises does not prevent later events in the same + drain from being processed. +- **32.** **Response-seam pin (Q#LN9).** A `send_request` reply reaches its + registered `on_response` one-shot, and the one-shot is **removed + before** invocation — a raising handler is not re-entered. Bites + against rev 2, where no Lua consumed `ev.kind == "response"` at all + and the reply was dropped. +- **33.** **Response dispatch-integrity pin.** With a response subscriber + registered, `workspace/applyEdit` in the same drain is still handled; + a raising response handler does not stop later events in that drain. + Mirrors the notification-side pins above. +- **34.** **Pending-purge pin, both edges.** A server that dies with a response + outstanding invokes the pending one-shot with an error and clears it. + **And** — the case round 4 found reachable and rev 4 missed — killing + the *buffer* with a request outstanding does the same, rather than + stranding the registration behind a drain that no longer runs for + that server. The second half must be shown to fail against a + purge wired only to the server-death transition; otherwise this + criterion is satisfied by the implementation that leaks. +- **34a.** **Canonicalizer pin (Q#LN20).** `pmacs.fs.canonicalize` resolves a + symlinked and dot-segmented path to the same string as the real path, + and returns nil for a nonexistent one. Fixture builds the symlink + rather than assuming one exists. +- **34b.** **Affinity-through-canonicalization pin.** With a function-valued + `root` that canonicalizes, the same project opened by its real path + and through a symlink reuses **one** server. Falsified by a resolver + that returns the path verbatim, which yields two — this is the + regression Q#LN20 exists to prevent, so it is asserted at the + affinity layer, not just at the binding. + +**Stage 3b — the Lean language server** + +- **22.** Opening a `.lean` file inside a Lake package spawns one server with + `cwd` and `rootUri` at the package root. +- **23.** **Outermost-root pin:** a file under + `/.lake/packages/dep/…` whose ancestor chain contains two + `lean-toolchain` files resolves to ``, not to `dep`. Run with + `pmacs.project.set_search_boundary` at the fixture root so the + assertion is hermetic. +- **24.** **Boundary pin:** with the search boundary set at the fixture root, a + `lean-toolchain` planted in an ancestor *above* the boundary is not + reached — the resolver stops at the boundary rather than walking past + it. +- **24a.** **Marker-is-a-file pin (Q#LN8).** A `lean-toolchain` + *directory* does not mark a root. Bites against the bare `io.open` + truth test, which round 4 probed succeeds on directories — the shape + that would pass every other criterion here while being wrong. +- **25.** A string-valued `pmacs.lsp.config.lean4.root` still works — the Q#LN8 + generalization is strictly additive. +- **26.** `didOpen` carries `languageId = "lean4"`. +- **27.** **Fallback-latch pin (Q#LN7):** a `lake` stub that exits non-zero — + reproducing §2.9's shimmed-elan state — causes exactly **one** restart + against `lean --server`, and a second failure surfaces an error rather + than looping. The latch does not re-arm within the session. +- **28.** **Probe pin:** a `lake` stub reporting version 3.0.0 triggers the + fallback; one reporting 3.1.0 does not. A stub that never exits does + not block the attach — the optimistic `lake serve` spawn proceeds. +- **35.** **Config-preservation pin (Q#LN7).** After the fallback latch fires, + user-supplied `env` / `settings` / `init_options` / `root` on + `pmacs.lsp.config.lean4` survive; only `command` and `args` change. +- **36.** **No-respawn-loop pin.** The latch stops the failing server before + spawning the fallback, so `RestartPolicy` does not respawn the broken + command underneath it. +- **36a.** **Attribution pin (COHERENCE §9/§1.2).** The probe process + appears in `pmacs.process.list` under a Lean-owned label, and the + latch firing leaves a status-line trace. Both assert through the + channel a user can actually observe; a report added through + `pmacs.error` alone must fail this. +- **37.** `textDocument/waitForDiagnostics` resolves through the response seam + (Q#LN16). **PATH-and-success-gated live smoke:** if `lake serve` + starts successfully a real elaboration completes and diagnostics + arrive; skipped otherwise, never failed. + +These two sections are bulleted with explicit labels rather than +numbered, because the split leaves each stage's criteria non-contiguous +(3b runs 22–28 then 35–37) and a markdown ordered list renumbers from +its first item regardless of what is written. Keeping the labels literal +means **every rev-4 number still denotes what it denoted in rev 4** — +"acceptance 34", "acceptance 27" — and the four criteria added in this +revision take letter suffixes rather than displacing anything. Round 3's +finding 4 was stale cross-references surviving a renumber; not +renumbering is the cheaper way to not repeat it. **Stage 4 — the Unicode input method** @@ -1447,7 +1747,7 @@ What remains deferred: - **#146 (HTML+CSS)** — the global capture table, and the requirement to pin retro-paint in both directions. Q#LN4 is that lesson applied. - **#123 (JSON/YAML)** — declarative `pmacs.lsp.config` entries with a - fake-server delivery proof plus PATH-gated live smokes. Stage 3 follows + fake-server delivery proof plus PATH-gated live smokes. Stage 3b follows it, with the extra success-gate §2.9 forces. - **#110 (auto-pairing)** — `take_typed_edit()` provenance, the fail-closed discipline on transformed source edits, and Q#AP1's optimistic-classifier @@ -1465,3 +1765,70 @@ What remains deferred: which Q#LN17 registers into. - **#94/#95 (LSP panels)** — `pmacs.listview.open` and the references/outline panel shape that Stage 7 reuses wholesale. + +## 9. Coherence impact (COHERENCE §20) + +Required of every framing since #163. Stated for stages 3a and 3b, the +work this revision authorizes; the earlier stages predate the rule and +are not retrofitted here. + +**Sections served.** §1.2 (the silence asymmetry) primarily, and §7 +(first-class workspaces) indirectly — per-root affinity is the workspace +concern arriving one language at a time. §9 (worker identity) is touched +but not advanced. + +**Golden journey (§2).** No step is touched. Neither stage changes what +happens between launching pmacs and editing a file; Lean is not on the +journey's critical path, and 3a is invisible to a user who has no Lean +installed. Stage 3b does make §2's step-3 grade slightly *worse* in one +narrow way, and it is honest to say so: a preconfigured-but-missing +`lake` is one more instance of the silent-spawn-failure class, on a +toolchain many users will not have. Q#LN7's status-line reports on the +probe verdict and the latch cover the Lean-specific paths, but they do +not fix the general failure — that remains Priority 1 work with its own +framing, as §1.2's frequency note already records. + +**Interaction islands (§6).** None added. Stage 3b introduces no keymap, +no modal surface, and no dispatch shadow. Its one user-facing command +(`M-x lean-wait-for-diagnostics`, Q#LN16) registers through the ordinary +command table and is reachable from `M-x` like everything else. + +**Config registry (§11).** Neither stage adds a `pmacs.config` option. +`pmacs.lsp.config.lean4` joins the existing declarative server table +alongside sixteen other languages — deliberately *not* the typed registry, +because moving one language's entry there while the other sixteen stay +put would fragment the surface rather than unify it. Migrating +`pmacs.lsp.config` wholesale is a config-arc concern; this lane must not +create a precedent that makes it harder. Stage 4's `lean.abbrev` gate is +where this arc does enter the registry, and Q#LN10 already commits to the +`editing.auto-pair` shape. + +**Background-work attribution (§9).** Three pieces of background work, +each with a named owner and an observable trace: + +| Work | Identity | Trace | +|---|---|---| +| `lake --version` probe | `ProcessSpec.label = "lean:lake-version-probe"`, visible in `pmacs.process.list` | status line on a verdict that triggers fallback | +| the fallback latch | the server it stops/spawns is already in `pmacs.lsp.list()` | status line on firing | +| root resolution | none — synchronous, inside the attach | status line on resolver failure (shipped #161) | + +This is attribution within the identity layer §9 says is absent, not a +fix for its absence: the probe carries a label because +`ProcessSpec.label` is the only field available, and §9's own ground +truth calls that "caller-supplied, unvalidated convention." Owner/purpose +/parent fields remain unbuilt, and nothing here joins the four activity +planes. What this lane commits to is not *worsening* the ratio — every +background action it adds is nameable in some user-visible view on the +day it ships. + +**Debt this revision retires.** Q#LN20 closes the gap #161 could only +document: a configured root reaching `file_uri_for` uncanonicalized. That +was coherence debt of exactly §1.3's compounding kind — a correct +substrate with a footgun the next caller was expected to disarm by +reading a comment. + +**Debt this revision names rather than pays.** Three, all in §6: the +uncapped event queue, the dropped `cfg.restart`, and — unchanged from +#161 — surfacing the spawn failure itself. Each is a behavior change for +languages other than Lean, and §4's rule is what keeps them out of a Lean +PR. From a516a46359286c4c6503d9ba43b37e051eb1b1e3 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 15:17:51 -0400 Subject: [PATCH 02/32] =?UTF-8?q?docs(lean4):=20rev=205=20round=201=20?= =?UTF-8?q?=E2=80=94=20fix=20the=20empty-marker=20case,=20sweep=20citation?= =?UTF-8?q?s?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Two review findings, both revision edits. **Q#LN8's marker test was wrong in the other direction.** Rev 5 fixed the directory case by reading a byte and requiring a non-nil read — but an **empty** `lean-toolchain` reads nil at EOF too, so that rule declines a marker that exists, silently, falling through to `pmacs.project.detect`. Marker semantics here are `lean4-mode`'s `locate-dominating-file` semantics: existence, not content, and a `lean-toolchain` can legitimately be empty. The discriminator is `read`'s second return, probed on LuaJIT 2.1: | Path | `io.open` | `f:read(1)` | Verdict | |---|---|---|---| | file with content | handle | `"l"`, no error | marker | | empty file | handle | `nil`, no error | marker | | directory | handle | `nil`, `"Is a directory"` | decline | | missing | `nil` | — | decline | So `local data, err = f:read(1)`, declining only on a non-nil `err`. The rule needs no per-platform re-probe: both directory behaviors are declines, since a platform whose `fopen` refuses a directory fails at `io.open` and one that opens it fails at `read`. There is no platform on which a directory both opens and yields a byte. Acceptance gains **24b** (an empty `lean-toolchain` marks a root) beside 24a, with the obligation that each be shown to fail against the implementation satisfying only the other. A suite carrying just one is satisfied by a resolver silently wrong for the other case — which is precisely how rev 5's first answer got written. **Citation sweep.** Round 4 stated the `project_root_for` correction in §0.1 without editing the citation in §2.5; the correction and the fix are different acts, and noting one is not doing the other. Review caught a second stale citation (`handle_server_requests` at :1448), which prompted a sweep of every `file:line` from §2.4 onward. Four more were stale. All six: `project_root_for` 513 → 592, `ensure_server` 527 → 610, `handle_server_requests` 1448 → 1549, `take_typed_edit` 12798 → 12827, `pair.lua` 213 → 229, `compile.lua` 264 → 266. Six others were verified good and left alone, listed in §0.1 so the next sweep knows what has already been checked. Q#LN15's present-tense "the change is small and spans two files" now reads as past tense with its PR number, since that stage landed. Its pre-#161 line numbers stay as written — historical record, not navigation. --- docs/lean4-mode-framing.md | 72 +++++++++++++++++++++++++++++++------- 1 file changed, 59 insertions(+), 13 deletions(-) diff --git a/docs/lean4-mode-framing.md b/docs/lean4-mode-framing.md index fc31e01..317d74f 100644 --- a/docs/lean4-mode-framing.md +++ b/docs/lean4-mode-framing.md @@ -252,9 +252,21 @@ Confirmations, recorded because each was load-bearing and unverified: Stage 3b is the first stage that would benefit from setting it, and Q#LN7 now records why it deliberately does not need it. -Citation drift repaired per COHERENCE §25: `project_root_for` is -`builtin/runtime/lsp.lua:592`, not 513, and returns `root, source` -rather than a bare root. +Citation drift repaired per COHERENCE §25. Round 4's first pass stated +the `project_root_for` correction in this section without editing the +citation in §2.5 — the correction and the fix are different acts, and +noting one is not doing the other. Review caught a second stale citation +(`handle_server_requests`), which prompted a full sweep of every +`file:line` from §2.4 onward; it found four more. All six: +`project_root_for` 513 → **592** (and it now returns `root, source` +rather than a bare root), `ensure_server` 527 → **610**, +`handle_server_requests` 1448 → **1549**, `take_typed_edit` +12798 → **12827**, `pair.lua` 213 → **229**, and `compile.lua` +264 → **266**. Verified good and left alone: `listview.lua:138`, +`src/lsp.rs:264`, `src/diag.rs:50`, `src/process.rs:193`, +`src/project.rs:145`, and the `mod.rs` binding-block citations. The pre-#161 line numbers inside Q#LN15 are +left as written: that stage has landed and its citations are historical +record, not navigation. ## 1. What ships @@ -465,7 +477,7 @@ and pin it.* params }` and `Response { id, result, error, method }` variants. Unknown server methods are delivered, not dropped. - **But `events_take` has exactly one consumer**: `handle_server_requests` - at `builtin/runtime/lsp.lua:1448`, driven off `pmacs._async.tick`. It + at `builtin/runtime/lsp.lua:1549`, driven off `pmacs._async.tick`. It `take`s — a drain. Its `if/elseif` chain handles five `request` methods and `initialized`, and **ignores every `notification` and every `response`**. A second module calling `events_take` would steal events @@ -481,7 +493,7 @@ and pin it.* ### 2.5 Project-root detection -`project_root_for` (`builtin/runtime/lsp.lua:513`) resolves: +`project_root_for` (`builtin/runtime/lsp.lua:592`) resolves: `pmacs.lsp.config[language].root` → `pmacs.project.detect` → the file's own directory. Two gaps for Lean: @@ -509,7 +521,7 @@ directory. Two gaps for Lean: then reports import errors for the whole file. Third, and the reason Stage 2 exists: `ensure_server` -(`builtin/runtime/lsp.lua:527`) reuses any live server with a matching +(`builtin/runtime/lsp.lua:610`) reuses any live server with a matching `language_id` regardless of the new file's project, so **the first `.lean` file opened fixes the root for every later `.lean` file.** For most languages that is an inconvenience; for Lean, where `lake serve` is bound @@ -526,10 +538,10 @@ changes loose-file behavior for every language. `builtin/runtime/pair.lua` is the whole precedent for "react to a typed character": subscribe to `buffer.after-edit`, gate on -`ed.this_command() == "buffer.self-insert"` (`pair.lua:213`), then take the +`ed.this_command() == "buffer.self-insert"` (`pair.lua:229`), then take the exact provenance record. -`pmacs.editor.take_typed_edit()` (`src/lua_bindings/mod.rs:12798`) returns +`pmacs.editor.take_typed_edit()` (`src/lua_bindings/mod.rs:12827`) returns `{ buffer, window, codepoint, char, requested_start, requested_end, effective_start, effective_end, inserted_len, post_cursor, clean }` — or nil. Its doc comment is explicit: @@ -558,7 +570,7 @@ cross-peer-degraded**. Lean's `⟨⟩` is outside that set. shows the adopter shape, gated on `spec.display == "panel"`. `pmacs.window.params()` and `pmacs.window.quit()` complete the surface. - Read-only generated buffers use the listview idiom, documented at - `builtin/runtime/compile.lua:264`: an erroring `pmacs.buffer.add_intercept` + `builtin/runtime/compile.lua:266`: an erroring `pmacs.buffer.add_intercept` for user edits, with module writes passing `{ bypass_intercept = true }`. - **Note for whoever picks this up on another machine:** the ledgers are stale about this. `docs/active-work.md:57` still heads the lane "Stage 1 @@ -896,9 +908,36 @@ already depends on `os.getenv`. One edge, probed: **`io.open` succeeds on a directory** (the handle opens; `read` returns nil without raising). A `lean-toolchain` *directory* would -therefore read as a marker. The resolver reads one byte and treats a -non-nil read as the marker, so a directory declines — an `io.open` truth -test alone would be wrong, and wrong silently. +therefore read as a marker under an `io.open` truth test — wrong, and +wrong silently. + +The fix is **not** "read a byte and require it to be non-nil", which was +this section's first answer and is wrong in the other direction: an +**empty** `lean-toolchain` file also reads nil at EOF, so that rule +declines a marker that exists. Marker semantics here are `lean4-mode`'s +`locate-dominating-file` semantics — *existence*, not content — and a +`lean-toolchain` can legitimately be empty. The discriminator is +`read`'s **second** return, probed on LuaJIT 2.1: + +| Path | `io.open` | `f:read(1)` | Verdict | +|---|---|---|---| +| file with content | handle | `"l"`, no error | marker | +| **empty file** | handle | `nil`, **no error** | **marker** | +| directory | handle | `nil`, `"Is a directory"` | decline | +| missing | `nil` | — | decline | + +So: `local data, err = f:read(1)` and decline only on a non-nil `err`. +The rule is robust across platforms without needing to be re-probed on +each, because both directory behaviors are declines — a platform whose +`fopen` refuses a directory outright fails at `io.open`, and one that +opens it fails at `read`. There is no platform on which a directory both +opens and yields a byte. + +Acceptance 24a and 24b pin the two halves, and each must be shown to +fail against the implementation that satisfies only the other — +otherwise "handles directories" is satisfiable by the version that +breaks empty files, which is exactly how this section's first answer got +written. **The result must be canonical.** #161's contract: a configured root reaches `file_uri_for` verbatim and that URI is the affinity key, so two @@ -1204,7 +1243,7 @@ rough edge but a correctness failure: `lake serve` is bound to one Lake package, so the second package a user opens gets a server that cannot resolve its imports. -The change is small and spans two files: +The change was small and spanned two files (Stage 2, landed as #161): - **`src/lua_bindings/mod.rs:9919`** — the `lsp.list()` row builder sets `id`/`label`/`language_id`/`command`/`state`/`attempt`. Add `root_uri` @@ -1626,6 +1665,13 @@ the blast radius. *directory* does not mark a root. Bites against the bare `io.open` truth test, which round 4 probed succeeds on directories — the shape that would pass every other criterion here while being wrong. +- **24b.** **Empty-marker pin (Q#LN8).** An **empty** `lean-toolchain` + file *does* mark a root — marker semantics are existence, not content. + Bites against the read-a-byte-and-require-non-nil rule, which declines + it at EOF. 24a and 24b must each be shown to fail against the + implementation that satisfies only the other; a suite carrying just + one of them is satisfied by a resolver that is silently wrong for the + other case. - **25.** A string-valued `pmacs.lsp.config.lean4.root` still works — the Q#LN8 generalization is strictly additive. - **26.** `didOpen` carries `languageId = "lean4"`. From 12236b265da4eebb1bf973965920b56cc5b2c3b5 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 15:37:48 -0400 Subject: [PATCH 03/32] feat(lsp): notification/response dispatch seams and fs.canonicalize MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Arc 8 Stage 3a (framing Q#LN9, Q#LN20). No Lean content: this changes the event drain every LSP language runs through, and is split from the Lean server work for the reason Stage 2 was. **The seams.** `handle_server_requests` handled five `request` methods and `initialized`, dropping every `notification` and `response` on the floor. Dropping responses made `pmacs.lsp.send_request` a write-only API from Lua — the reply was drained and discarded, so nothing outside Rust's typed stores could consume one. Two new arms route to `pmacs.lsp.on_notification(method, fn)` (persistent, method-keyed) and `pmacs.lsp.on_response(sid, request_id, fn)` (one-shot). Both extend the existing loop rather than opening a second `events_take` caller, which would steal events from it. A one-shot is removed **before** invocation, so a raising handler cannot be re-entered. Every subscriber is `pcall`ed and a raise reports through `pmacs.editor.set_status` per COHERENCE §1.2 — not `pmacs.error`, which is defined nowhere in production. The notification list's length is captured before the walk so a subscriber registering another cannot extend the list being iterated. **The purge is driven off `pmacs.lsp.list()`, not off a death event.** The framing said acceptance 34's second edge was a killed buffer. That was wrong, and scouting the implementation is what caught it: pmacs fires exactly five hooks (`buffer.after-edit`, `buffer.after-load`, `buffer.after-switch`, `frontend.detached`, `process.after-tick`) and there is no buffer-kill hook at all, so `lsp.lua` never tears an attachment down and the drain keeps reaching that server. No leak there. The real leak is a different path with the same root cause. The drain builds its sid list from `attachments`, and `attach_buffer` drops a sid from that table the moment `server_is_live` reports false — rebuilding against a fresh server. So the `crashed` / `stopped` event that should trigger the purge is precisely the one most likely to go undrained. A purge wired to that event leaks exactly when it matters. `pmacs.lsp.list()` enumerates the manager directly and is unaffected by attachment bookkeeping, so the purge polls it after each drain: a sid that is absent, terminal, or running a **new generation** settles its pending one-shots with an error. The generation check uses the `attempt` field, because a crash-then-restart reuses the sid — without it a one-shot would sit waiting on a reply the dead generation owed. **`pmacs.fs.canonicalize`** (Q#LN20) is the one synchronous function on `pmacs.fs`, and synchronous is the point: its consumer is a function-valued `config.root` called from `ensure_server` <- `attach_buffer` <- `buffer.after-load`, where there is no coroutine and `pmacs.fs.stat`'s awaitable handle is unusable. It is installed from `install_async` rather than `install_project` purely for load order — `make_workspace` runs after `fs.lua` is evaluated, so a canonicalizer placed there reads nil. Acceptance: `tests/lsp_dispatch_seams_acceptance.rs`, 14 tests, driven against `pmacs_fake_lsp` through rust so nothing needs a toolchain. Dispatch integrity is exercised at real co-occurrence — the fake server writes `workspace/applyEdit` and the `executeCommand` reply back to back, so both land in one `events_take` batch. 34b asserts affinity survives a symlinked open and is paired with its own falsification: the same resolver minus the canonicalize call spawns two servers, so the positive case cannot be vacuous. --- builtin/runtime/fs.lua | 24 + builtin/runtime/lsp.lua | 162 ++++++ src/lua_bindings/mod.rs | 39 ++ tests/lsp_dispatch_seams_acceptance.rs | 667 +++++++++++++++++++++++++ 4 files changed, 892 insertions(+) create mode 100644 tests/lsp_dispatch_seams_acceptance.rs diff --git a/builtin/runtime/fs.lua b/builtin/runtime/fs.lua index 02ca064..77fee1f 100644 --- a/builtin/runtime/fs.lua +++ b/builtin/runtime/fs.lua @@ -284,4 +284,28 @@ function fs.watch(path, callback, opts) return watch end +-- pmacs.fs.canonicalize(path) -> string | nil +-- +-- Arc 8 Stage 3a (framing Q#LN20). The **only synchronous** function on +-- this module, and deliberately so: its consumer is a function-valued +-- `pmacs.lsp.config[lang].root`, invoked from `ensure_server` <- +-- `attach_buffer` <- the `buffer.after-load` hook, where there is no +-- coroutine and therefore nothing to `:await()` on. Every other +-- primitive here returns a Handle; this one cannot, or it would be +-- unusable at the one call site that needs it — the same trap +-- `pmacs.fs.stat` falls into for that caller. +-- +-- Resolves symlinks and `.` / `..`, returning an absolute path, or nil +-- if the path does not exist or cannot be resolved. Nil is a normal +-- answer, not an error: callers routinely ask about paths that may have +-- been deleted. +-- +-- Why it exists: a configured LSP root reaches `file_uri_for` verbatim +-- and that URI is the server-affinity key (PR #161), so one project +-- opened through a symlink and through its real path would otherwise +-- spawn two servers. `pmacs.editor.file_path()` collapses `.` and `..` +-- lexically but leaves symlinks intact, so the resolver cannot get a +-- canonical path any other way. +fs.canonicalize = pmacs._fs.canonicalize + pmacs.fs = fs diff --git a/builtin/runtime/lsp.lua b/builtin/runtime/lsp.lua index 6021134..eac5753 100644 --- a/builtin/runtime/lsp.lua +++ b/builtin/runtime/lsp.lua @@ -1546,6 +1546,160 @@ end -- itself is unaffected. Server ids are snapshotted before the loop -- because `apply_workspace_edit` → `find_or_open` can attach a new -- buffer mid-iteration (mutating `attachments`). +-- Server-originated notification / response seams (framing Q#LN9) ------- +-- +-- Before this, `handle_server_requests` handled five `request` methods +-- and `initialized`, and dropped every `notification` and `response` on +-- the floor. Dropping responses made `pmacs.lsp.send_request` a +-- write-only API from Lua: the reply was drained and discarded, so +-- nothing outside Rust's typed stores could ever consume one. +-- +-- Both seams route through the *existing* drain. A second +-- `events_take` caller would steal events from this one — `take_events` +-- removes the queue — so any new consumer must extend this loop rather +-- than open its own. +-- +-- method -> array of subscriber fns. Persistent; `pmacs.hook` has no +-- `remove` and neither does this, deliberately matching it. +local notification_subs = {} +-- tostring(sid) -> { [request_id] = { fn = fn, attempt = n } }. One-shot. +local pending_responses = {} + +local function report_subscriber_error(what, err) + local msg = string.format("LSP: %s subscriber failed: %s", what, + tostring(err)) + -- COHERENCE §1.2: a pcall around background wiring must report, not + -- discard. `pmacs.editor.set_status` is the channel that exists; + -- `pmacs.error` is referenced by fifteen call sites and defined + -- nowhere in production, so it rides along rather than standing alone. + pcall(pmacs.editor.set_status, msg) + if pmacs.error then pcall(pmacs.error, msg) end +end + +-- Current spawn attempt for `sid`, or nil if the manager has forgotten +-- it. A restart reuses the sid but bumps the attempt, which is how a +-- pending one-shot tells "my server is still here" from "my server died +-- and a new generation took its id". +local function server_attempt(sid) + local skey = tostring(sid) + for _, info in ipairs(pmacs.lsp.list()) do + if tostring(info.id) == skey then + return info.attempt or 0, info.state and info.state.kind + end + end + return nil, nil +end + +-- fn(sid, params); persistent, fires for every server. +function pmacs.lsp.on_notification(method, fn) + if type(method) ~= "string" or type(fn) ~= "function" then + error("pmacs.lsp.on_notification(method, fn): want string, function") + end + local subs = notification_subs[method] + if not subs then + subs = {} + notification_subs[method] = subs + end + subs[#subs + 1] = fn +end + +-- fn(result, err); ONE-SHOT, keyed to the exact request. +-- `request_id` is what `pmacs.lsp.send_request` returned. +function pmacs.lsp.on_response(sid, request_id, fn) + if not sid or type(request_id) ~= "number" or type(fn) ~= "function" then + error("pmacs.lsp.on_response(sid, request_id, fn): want sid, number, function") + end + local skey = tostring(sid) + local pend = pending_responses[skey] + if not pend then + pend = {} + pending_responses[skey] = pend + end + -- The attempt is captured at registration so a restart under the same + -- sid purges this entry rather than leaving it waiting on a reply the + -- dead generation was going to send. + pend[request_id] = { fn = fn, attempt = server_attempt(sid) or 0 } +end + +local function dispatch_notification(sid, ev) + local subs = notification_subs[ev.method] + if not subs then return end + -- Length captured up front: a subscriber that registers another one + -- must not be able to extend the list being walked. + local n = #subs + for i = 1, n do + local ok, err = pcall(subs[i], sid, ev.params) + if not ok then + report_subscriber_error("notification " .. tostring(ev.method), err) + end + end +end + +local function deliver_response(sid, ev) + local skey = tostring(sid) + local pend = pending_responses[skey] + if not pend then return end + local entry = pend[ev.request_id] + if not entry then return end + -- Removed BEFORE invocation: a handler that raises must not be + -- re-entered by a later event carrying the same id. + pend[ev.request_id] = nil + if next(pend) == nil then pending_responses[skey] = nil end + local ok, err = pcall(entry.fn, ev.result, ev.error) + if not ok then + report_subscriber_error("response " .. tostring(ev.method), err) + end +end + +-- Settle every one-shot whose server can no longer answer it. +-- +-- Deliberately driven off `pmacs.lsp.list()` and NOT off a death event +-- observed in the drain, because the drain cannot be relied on to reach +-- the server in question: `handle_server_requests` builds its sid list +-- from `attachments`, and a sid leaves that table whenever +-- `attach_buffer` finds it dead and rebuilds the attachment against a +-- fresh server. So the very event that should trigger the purge — +-- `crashed` / `stopped` — is the one most likely to go undrained. A +-- one-shot settled only by the drain would leak exactly when it matters. +-- +-- `pmacs.lsp.list()` enumerates the manager directly and is unaffected +-- by attachment bookkeeping, which is what makes it the right authority. +local function purge_dead_pending() + if next(pending_responses) == nil then return end + local ok, rows = pcall(pmacs.lsp.list) + -- A failed enumeration is not evidence that every server died; leaving + -- the registrations alone is the safe read of "we don't know". + if not ok or not rows then return end + local alive = {} + for _, info in ipairs(rows) do + local kind = info.state and info.state.kind + if kind ~= "crashed" and kind ~= "stopped" then + alive[tostring(info.id)] = info.attempt or 0 + end + end + for skey, pend in pairs(pending_responses) do + local attempt = alive[skey] + local dead = {} + for rid, entry in pairs(pend) do + -- Absent or terminal, or the same sid running a NEW generation: + -- in every case the request this entry awaits is unanswerable. + if attempt == nil or attempt ~= entry.attempt then + dead[#dead + 1] = rid + end + end + for _, rid in ipairs(dead) do + local entry = pend[rid] + pend[rid] = nil + local ok_h, err = pcall(entry.fn, nil, + { message = "server gone before response" }) + if not ok_h then + report_subscriber_error("response purge", err) + end + end + if next(pend) == nil then pending_responses[skey] = nil end + end +end + local function handle_server_requests() local sids, seen = {}, {} for _, rec in pairs(attachments) do @@ -1598,6 +1752,10 @@ local function handle_server_requests() -- LSP spells the field "unregisterations". pcall(unregister_file_watchers, sid, ev.params and ev.params.unregisterations) + elseif ev.kind == "notification" then + dispatch_notification(sid, ev) + elseif ev.kind == "response" then + deliver_response(sid, ev) elseif ev.kind == "initialized" then -- Buffers attach before the server finishes initializing, so -- the pulls in `attach_buffer` are no-ops for the FIRST file @@ -1620,6 +1778,10 @@ if pmacs._async and pmacs._async.tick then pmacs._async.tick = function(...) local ret = _prior_async_tick(...) pcall(handle_server_requests) + -- After the drain, so a response delivered this tick settles its + -- one-shot normally rather than being purged as "server gone" in the + -- same pass when the server died right after answering. + pcall(purge_dead_pending) pcall(flush_due_did_changes) return ret end diff --git a/src/lua_bindings/mod.rs b/src/lua_bindings/mod.rs index 3a92520..369b810 100644 --- a/src/lua_bindings/mod.rs +++ b/src/lua_bindings/mod.rs @@ -6519,6 +6519,45 @@ pub fn install_async( ) -> mlua::Result<()> { lua.set_app_data(runtime.clone()); let pmacs: Table = lua.globals().get("pmacs")?; + + // Arc 8 Stage 3a (framing Q#LN20): the one *synchronous* filesystem + // primitive Lua has. `pmacs.fs` is otherwise an async, handle- + // returning surface built in `builtin/runtime/fs.lua`, so this + // arrives through a private table that file re-exports rather than + // joining the `_dispatch_fs_*` family it would not belong to. + // + // Installed here, alongside those dispatchers, purely for load + // order: `make_async_runtime` runs before `fs.lua` is evaluated, + // whereas `install_project` — the other plausible home — runs after + // it, so a canonicalizer placed there is nil when `fs.lua` reads it. + // + // Synchronous on purpose, and that is the whole point. The consumer + // is a function-valued `pmacs.lsp.config[lang].root`, which + // `project_root_for` calls from `ensure_server` <- `attach_buffer` + // <- the `buffer.after-load` hook — no coroutine, nothing to await + // on. An awaitable canonicalizer would be unusable there for exactly + // the reason `pmacs.fs.stat` already is, leaving #161's + // canonical-root obligation undischarged. The cost is one syscall on + // a path the editor is already opening; `pmacs.project.detect` + // canonicalizes synchronously on the same hook today. + { + let fs_priv = lua.create_table()?; + fs_priv.set( + "canonicalize", + lua.create_function(|_, path: String| { + // nil rather than an error for a path that cannot be + // resolved: asking about a deleted file or a broken + // symlink is ordinary, and raising would surface through + // `resolve_root_fn`'s pcall as a config bug, which it is + // not. + Ok(std::fs::canonicalize(&path) + .ok() + .map(|p| p.display().to_string())) + })?, + )?; + pmacs.set("_fs", fs_priv)?; + } + let async_mod = lua.create_table()?; { diff --git a/tests/lsp_dispatch_seams_acceptance.rs b/tests/lsp_dispatch_seams_acceptance.rs new file mode 100644 index 0000000..d240ba4 --- /dev/null +++ b/tests/lsp_dispatch_seams_acceptance.rs @@ -0,0 +1,667 @@ +//! Arc 8 Stage 3a acceptance — LSP notification/response dispatch seams +//! and `pmacs.fs.canonicalize`. +//! +//! `docs/lean4-mode-framing.md` Q#LN9 and Q#LN20, acceptance 29–34 plus +//! 34a/34b. +//! +//! This suite deliberately contains **no Lean content**. +//! `handle_server_requests` (`builtin/runtime/lsp.lua`) is the single +//! LSP event drain for every language in pmacs, so the change is +//! exercised through an already-shipped language driven against +//! `pmacs_fake_lsp`. A suite that reached the drain only through Lean +//! would understate the blast radius — the same reasoning that shaped +//! Stage 2's suite. +//! +//! Every fixture calls `pmacs.project.set_search_boundary` at its own +//! tempdir root, so a stray marker above the temp directory cannot make +//! a "markerless" case silently detected. + +use std::path::{Path, PathBuf}; +use std::time::Duration; + +use pmacs::editor::EditorState; + +fn exec(state: &EditorState, source: &str) { + state.lua_host.lua().load(source.to_owned()).exec().unwrap(); +} + +fn eval(state: &EditorState, source: &str) -> T { + state.lua_host.lua().load(source.to_owned()).eval().unwrap() +} + +fn fake_lsp_path() -> String { + env!("CARGO_BIN_EXE_pmacs_fake_lsp").to_owned() +} + +/// A fresh editor with the shipped language configs cleared, so the only +/// server any test can spawn is the fake one it configures itself. +fn editor() -> EditorState { + let state = EditorState::new(); + exec(&state, "pmacs.lsp.config = {}"); + state +} + +fn lua_str(path: &Path) -> String { + path.display() + .to_string() + .replace('\\', "\\\\") + .replace('"', "\\\"") +} + +struct Fixture { + _dir: tempfile::TempDir, + root: PathBuf, +} + +impl Fixture { + fn new() -> Self { + let dir = tempfile::tempdir().unwrap(); + let root = std::fs::canonicalize(dir.path()).unwrap(); + Self { _dir: dir, root } + } + + fn write(&self, rel: &str, contents: &str) -> PathBuf { + let path = self.root.join(rel); + std::fs::create_dir_all(path.parent().unwrap()).unwrap(); + std::fs::write(&path, contents).unwrap(); + path + } + + fn dir(&self, rel: &str) -> PathBuf { + self.root.join(rel) + } + + fn bind(&self, state: &EditorState) { + exec( + state, + &format!( + "pmacs.project.set_search_boundary(\"{}\")", + lua_str(&self.root) + ), + ); + } +} + +fn configure(state: &EditorState, language: &str) { + exec( + state, + &format!( + "pmacs.lsp.config.{language} = {{ command = \"{}\" }}", + fake_lsp_path() + ), + ); +} + +fn open(state: &EditorState, path: &Path) { + exec( + state, + &format!("pmacs.buffer.find_or_open(\"{}\")", lua_str(path)), + ); +} + +/// `tick_async` is what drives the drain: `handle_server_requests` is +/// wrapped onto `pmacs._async.tick`, so a settle loop without it moves +/// the LSP state machine while never delivering a single event to Lua. +fn settle(state: &mut EditorState) { + for _ in 0..8 { + state.tick_processes(); + state.tick_lsp(); + state.tick_async(); + std::thread::sleep(Duration::from_millis(2)); + } +} + +/// A rust project with one file, an attached fake server, and the +/// probes below installed. Returns the opened file's path. +fn attached_rust(state: &mut EditorState, fx: &Fixture) -> PathBuf { + fx.write("proj/Cargo.toml", "[package]\nname = \"p\"\n"); + let file = fx.write("proj/src/main.rs", "fn main() {}\nlet x = 1;\n"); + fx.bind(state); + configure(state, "rust"); + open(state, &file); + settle(state); + file +} + +/// The sid of the single live server, as a Lua expression fragment. +const THE_SID: &str = "pmacs.lsp.list()[1].id"; + +// --------------------------------------------------------------------------- +// Acceptance 29 — a notification reaches a registered subscriber. +// --------------------------------------------------------------------------- + +#[test] +fn acc29_notification_reaches_a_registered_subscriber() { + let fx = Fixture::new(); + let mut state = editor(); + // Registered BEFORE the open, so the didOpen-triggered `pmacs/echo` + // is in the first drain. + exec( + &state, + r#" + _G.seen = {} + pmacs.lsp.on_notification("pmacs/echo", function(sid, params) + _G.seen[#_G.seen + 1] = tostring(params and params.uri) + end) + "#, + ); + attached_rust(&mut state, &fx); + + let n: i64 = eval(&state, "return #_G.seen"); + assert!( + n >= 1, + "expected at least one pmacs/echo notification, got {n}" + ); + let first: String = eval(&state, "return _G.seen[1]"); + assert!( + first.starts_with("file://") && first.ends_with("main.rs"), + "subscriber got the document uri; saw {first:?}" + ); +} + +#[test] +fn acc29_subscriber_for_an_unsent_method_does_not_fire() { + let fx = Fixture::new(); + let mut state = editor(); + exec( + &state, + r#" + _G.hits = 0 + pmacs.lsp.on_notification("pmacs/never", function() _G.hits = _G.hits + 1 end) + "#, + ); + attached_rust(&mut state, &fx); + + // Non-vacuity for acc29: the seam is method-keyed, not a firehose. + // Without this, a subscriber invoked for every notification would + // pass the test above while being wrong. + let hits: i64 = eval(&state, "return _G.hits"); + assert_eq!(hits, 0, "a subscriber must only fire for its own method"); +} + +// --------------------------------------------------------------------------- +// Acceptance 30 + 33 — dispatch integrity: with subscribers registered, +// a `workspace/applyEdit` request in the same drain is still handled. +// +// The fake server writes the applyEdit request and the executeCommand +// response back to back, so both land in one `events_take` batch. That +// co-occurrence is the point: a seam that consumed the batch, or that +// returned early, would starve the `request` arms that share it. +// --------------------------------------------------------------------------- + +fn drive_apply_edit(state: &mut EditorState, file: &Path) { + exec( + state, + &format!( + r#" + local sid = {THE_SID} + local uri = "file://{}" + _G.rid = pmacs.lsp.send_request(sid, "workspace/executeCommand", {{ + command = "pmacs.fake.applyEdit", + arguments = {{ uri }}, + }}) + _G.response_hits = 0 + pmacs.lsp.on_response(sid, _G.rid, function(result, err) + _G.response_hits = _G.response_hits + 1 + end) + "#, + lua_str(file) + ), + ); + settle(state); +} + +fn buffer_text(state: &EditorState) -> String { + eval( + state, + "local b = pmacs.window.buffer() return b:slice(0, b:len())", + ) +} + +#[test] +fn acc30_apply_edit_still_handled_with_a_notification_subscriber() { + let fx = Fixture::new(); + let mut state = editor(); + exec( + &state, + r#" + _G.notes = 0 + pmacs.lsp.on_notification("pmacs/echo", function() _G.notes = _G.notes + 1 end) + "#, + ); + let file = attached_rust(&mut state, &fx); + assert!( + eval::(&state, "return _G.notes") >= 1, + "precondition: the notification subscriber is actually firing" + ); + + drive_apply_edit(&mut state, &file); + + assert!( + buffer_text(&state).contains("ED2"), + "workspace/applyEdit must still be applied with a subscriber \ + registered; buffer was {:?}", + buffer_text(&state) + ); +} + +#[test] +fn acc33_apply_edit_still_handled_with_a_response_subscriber() { + let fx = Fixture::new(); + let mut state = editor(); + let file = attached_rust(&mut state, &fx); + drive_apply_edit(&mut state, &file); + + // Both halves in one drain: the response was delivered to its + // one-shot AND the server-originated request was serviced. + assert_eq!( + eval::(&state, "return _G.response_hits"), + 1, + "the executeCommand response reaches its one-shot" + ); + assert!( + buffer_text(&state).contains("ED2"), + "workspace/applyEdit must still be applied with a response \ + subscriber registered; buffer was {:?}", + buffer_text(&state) + ); +} + +// --------------------------------------------------------------------------- +// Acceptance 31 — a raising subscriber does not stop later events in the +// same drain (and does not stop the `request` arms either). +// --------------------------------------------------------------------------- + +#[test] +fn acc31_raising_notification_subscriber_does_not_stop_the_drain() { + let fx = Fixture::new(); + let mut state = editor(); + exec( + &state, + r#" + _G.second_hits = 0 + pmacs.lsp.on_notification("pmacs/echo", function() + error("subscriber blew up") + end) + pmacs.lsp.on_notification("pmacs/echo", function() + _G.second_hits = _G.second_hits + 1 + end) + "#, + ); + let file = attached_rust(&mut state, &fx); + + assert!( + eval::(&state, "return _G.second_hits") >= 1, + "a raising subscriber must not starve the ones after it" + ); + + // And the shared `request` arms still run in a later drain. + drive_apply_edit(&mut state, &file); + assert!( + buffer_text(&state).contains("ED2"), + "a raising subscriber must not stop workspace/applyEdit" + ); +} + +#[test] +fn acc33_raising_response_handler_does_not_stop_the_drain() { + let fx = Fixture::new(); + let mut state = editor(); + let file = attached_rust(&mut state, &fx); + exec( + &state, + &format!( + r#" + local sid = {THE_SID} + _G.notes_after = 0 + pmacs.lsp.on_notification("pmacs/echo", function() + _G.notes_after = _G.notes_after + 1 + end) + local rid = pmacs.lsp.send_request(sid, "workspace/executeCommand", {{ + command = "pmacs.fake.applyEdit", + arguments = {{ "file://{}" }}, + }}) + pmacs.lsp.on_response(sid, rid, function() error("handler blew up") end) + "#, + lua_str(&file) + ), + ); + settle(&mut state); + + assert!( + buffer_text(&state).contains("ED2"), + "a raising response handler must not stop workspace/applyEdit in \ + the same drain" + ); +} + +// --------------------------------------------------------------------------- +// Acceptance 32 — the one-shot is removed BEFORE invocation. +// +// Observed rather than asserted structurally: the handler raises, and +// the server is then stopped. If removal happened only on a clean +// return — or not at all — the purge below would invoke the same handler +// a second time with an error. The count is what pins it. +// --------------------------------------------------------------------------- + +#[test] +fn acc32_response_one_shot_is_removed_before_invocation() { + let fx = Fixture::new(); + let mut state = editor(); + attached_rust(&mut state, &fx); + exec( + &state, + &format!( + r#" + local sid = {THE_SID} + _G.calls = 0 + local rid = pmacs.lsp.send_request(sid, "test/ping", {{ v = 1 }}) + pmacs.lsp.on_response(sid, rid, function(result, err) + _G.calls = _G.calls + 1 + error("handler raises after being removed") + end) + "# + ), + ); + settle(&mut state); + assert_eq!( + eval::(&state, "return _G.calls"), + 1, + "the one-shot fires exactly once for its reply" + ); + + exec(&state, &format!("pmacs.lsp.stop({THE_SID})")); + settle(&mut state); + assert_eq!( + eval::(&state, "return _G.calls"), + 1, + "a delivered one-shot must not be re-invoked by the purge — it \ + was removed before the raising handler ran, not after" + ); +} + +#[test] +fn acc32_response_carries_the_servers_result() { + let fx = Fixture::new(); + let mut state = editor(); + attached_rust(&mut state, &fx); + exec( + &state, + &format!( + r#" + local sid = {THE_SID} + _G.echoed = nil + _G.saw_err = "unset" + local rid = pmacs.lsp.send_request(sid, "test/ping", {{ v = 42 }}) + pmacs.lsp.on_response(sid, rid, function(result, err) + _G.echoed = result and result.echo and result.echo.v + _G.saw_err = tostring(err) + end) + "# + ), + ); + settle(&mut state); + + // Non-vacuity: without this the seam could "fire" with nil payloads + // and every count-based assertion above would still pass. + assert_eq!( + eval::(&state, "return _G.echoed or -1"), + 42, + "the handler receives the server's result payload" + ); + assert_eq!( + eval::(&state, "return _G.saw_err"), + "nil", + "a successful reply passes nil for err" + ); +} + +// --------------------------------------------------------------------------- +// Acceptance 34 — the pending purge, driven off `pmacs.lsp.list()` and +// NOT off a death event seen in the drain. +// +// The second test is the load-bearing one. `handle_server_requests` +// builds its sid list from `attachments`, so a server that is in no +// attachment is never drained — and its `stopped` event is therefore +// never seen. A purge wired to that event leaks exactly there. +// --------------------------------------------------------------------------- + +#[test] +fn acc34_purge_settles_a_pending_one_shot_when_the_server_dies() { + let fx = Fixture::new(); + let mut state = editor(); + attached_rust(&mut state, &fx); + exec( + &state, + &format!( + r#" + local sid = {THE_SID} + _G.err_msg = "never called" + -- A method the fake server answers only after a delay would + -- be ideal; instead the server is stopped in the same breath, + -- so the reply can never arrive. + local rid = pmacs.lsp.send_request(sid, "test/slow", {{}}) + pmacs.lsp.on_response(sid, rid, function(result, err) + _G.err_msg = tostring(err and err.message) + end) + pmacs.lsp.stop(sid) + "# + ), + ); + settle(&mut state); + + let msg: String = eval(&state, "return _G.err_msg"); + assert!( + msg.contains("server gone") || msg == "nil", + "a pending one-shot must be settled, not left waiting; saw {msg:?}" + ); + assert_ne!( + msg, "never called", + "the one-shot was never settled — it leaked" + ); +} + +#[test] +fn acc34_purge_reaches_a_server_that_is_in_no_attachment() { + let fx = Fixture::new(); + let mut state = editor(); + fx.bind(&state); + // Spawned directly, never attached to a buffer. `attachments` is + // empty, so `handle_server_requests` never visits this sid and its + // `stopped` event is never drained. + exec( + &state, + &format!( + r#" + _G.settled = "never called" + local sid = pmacs.lsp.spawn({{ + label = "orphan", + language_id = "rust", + command = "{}", + args = {{}}, + }}) + _G.orphan = sid + "#, + fake_lsp_path() + ), + ); + settle(&mut state); + + let attached: i64 = eval( + &state, + "local n = 0 for _ in pairs(_G) do n = n + 1 end return n", + ); + assert!(attached > 0, "lua globals are readable"); + + exec( + &state, + r#" + local rid = pmacs.lsp.send_request(_G.orphan, "test/slow", {}) + pmacs.lsp.on_response(_G.orphan, rid, function(result, err) + _G.settled = tostring(err and err.message) + end) + pmacs.lsp.stop(_G.orphan) + "#, + ); + settle(&mut state); + + let settled: String = eval(&state, "return _G.settled"); + assert_ne!( + settled, "never called", + "the purge must not depend on the drain reaching this server — \ + it is in no attachment, so the drain never does" + ); + assert!( + settled.contains("server gone"), + "settled with the purge's error; saw {settled:?}" + ); +} + +// --------------------------------------------------------------------------- +// Acceptance 34a — `pmacs.fs.canonicalize` (Q#LN20). +// --------------------------------------------------------------------------- + +#[test] +fn acc34a_canonicalize_resolves_symlinks_and_dot_segments() { + let fx = Fixture::new(); + fx.write("pkg/sub/a.txt", "x\n"); + // Built here rather than assumed: the whole point is the symlink. + std::os::unix::fs::symlink(fx.dir("pkg"), fx.dir("linkpkg")).unwrap(); + let state = editor(); + + let noncanon = format!("{}/sub/./../sub/a.txt", fx.dir("linkpkg").display()); + let got: String = eval( + &state, + &format!("return tostring(pmacs.fs.canonicalize(\"{noncanon}\"))"), + ); + let want = fx.root.join("pkg/sub/a.txt").display().to_string(); + assert_eq!(got, want, "symlink and dot segments both resolved"); + + // Falsification for 34b: the uncanonicalized spelling really is + // different, so the affinity test below is not vacuous. + assert_ne!(noncanon, want); +} + +#[test] +fn acc34a_canonicalize_returns_nil_for_a_missing_path() { + let fx = Fixture::new(); + let state = editor(); + let missing = fx.dir("nope/not-here").display().to_string(); + let got: String = eval( + &state, + &format!("return tostring(pmacs.fs.canonicalize(\"{missing}\"))"), + ); + assert_eq!( + got, "nil", + "a nonexistent path declines rather than raising" + ); +} + +// --------------------------------------------------------------------------- +// Acceptance 34b — affinity survives a symlinked open. +// +// Asserted at the affinity layer, not just at the binding: the +// regression Q#LN20 exists to prevent is *two servers for one project*, +// and only this shape observes it. +// --------------------------------------------------------------------------- + +fn server_count(state: &EditorState) -> i64 { + eval(state, "return #pmacs.lsp.list()") +} + +#[test] +fn acc34b_canonicalizing_resolver_reuses_one_server_across_a_symlink() { + let fx = Fixture::new(); + fx.write("proj/Cargo.toml", "[package]\nname = \"p\"\n"); + let real = fx.write("proj/src/main.rs", "fn main() {}\n"); + std::os::unix::fs::symlink(fx.dir("proj"), fx.dir("linkproj")).unwrap(); + let linked = fx.dir("linkproj").join("src/main.rs"); + + let mut state = editor(); + fx.bind(&state); + exec( + &state, + &format!( + r#" + pmacs.lsp.config.rust = {{ + command = "{}", + root = function(path) + local dir = path:match("^(.*)/[^/]*$") + if not dir then return nil end + -- Walk up to the directory holding Cargo.toml, then + -- canonicalize — the Q#LN8 shape Stage 3b will use. + while dir and #dir > 0 do + local f = io.open(dir .. "/Cargo.toml", "r") + if f then + f:close() + return pmacs.fs.canonicalize(dir) + end + dir = dir:match("^(.*)/[^/]*$") + end + return nil + end, + }} + "#, + fake_lsp_path() + ), + ); + + open(&state, &real); + settle(&mut state); + assert_eq!(server_count(&state), 1, "the real path spawns one server"); + + open(&state, &linked); + settle(&mut state); + assert_eq!( + server_count(&state), + 1, + "the symlinked path must reuse the same server — two here is the \ + exact regression Q#LN20 exists to prevent" + ); +} + +#[test] +fn acc34b_falsified_by_a_resolver_that_skips_canonicalization() { + let fx = Fixture::new(); + fx.write("proj/Cargo.toml", "[package]\nname = \"p\"\n"); + let real = fx.write("proj/src/main.rs", "fn main() {}\n"); + std::os::unix::fs::symlink(fx.dir("proj"), fx.dir("linkproj")).unwrap(); + let linked = fx.dir("linkproj").join("src/main.rs"); + + let mut state = editor(); + fx.bind(&state); + // Same resolver, minus the canonicalize call. This is the bite: if + // it also produced one server, the test above would be vacuous and + // `pmacs.fs.canonicalize` would be doing nothing. + exec( + &state, + &format!( + r#" + pmacs.lsp.config.rust = {{ + command = "{}", + root = function(path) + local dir = path:match("^(.*)/[^/]*$") + while dir and #dir > 0 do + local f = io.open(dir .. "/Cargo.toml", "r") + if f then f:close() return dir end + dir = dir:match("^(.*)/[^/]*$") + end + return nil + end, + }} + "#, + fake_lsp_path() + ), + ); + + open(&state, &real); + settle(&mut state); + open(&state, &linked); + settle(&mut state); + assert_eq!( + server_count(&state), + 2, + "without canonicalization the two spellings key differently and \ + spawn two servers — this is what 34b's positive case rules out" + ); +} From 027c7d18e079c0abedcd90b9b0c995aa7f64c66e Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 15:58:34 -0400 Subject: [PATCH 04/32] test(lsp): name acc32 for what it pins; label the unpinned guard MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Two corrections found by bite-testing the suite rather than by reading it. **Acceptance 32 was mis-named.** It claimed to pin "the one-shot is removed BEFORE invocation". Biting that — moving the removal after the `pcall` — still passes, because `pcall` catches the raise either way and the removal runs regardless. The before/after ordering is unobservable unless a handler re-enters the drain, and nothing does. What the test actually pins is that removal is **unconditional**: the bite that gates it on `if ok then` fails 2 != 1, because the surviving registration is invoked a second time by the purge. Renamed and re-commented to say so. The implementation still removes before invoking, which is the right defensive order; it is simply not what the assertion proves. **The purge's generation check is defensive and untested**, now labelled in place instead of reading as covered. Reaching it needs a crash and its restart to both land in a gap with no `_async.tick`; the backoff is 500ms, so any tick in that window sees `crashed` and the absent-or-terminal test fires first. Every attempt to stage it deterministically exercised the `crashed` path instead. It stays as insurance for a stalled editor, and says that about itself. Bites recorded, all against the committed tree: - removal gated on a clean return -> acc32 fails (2 != 1). - purge driven by a `crashed`/`stopped` event seen in the drain, the design the framing originally implied -> the no-attachment case fails ("never called"), while the attached case still passes. That is the discrimination acceptance 34's second half exists for. - a resolver without `pmacs.fs.canonicalize` -> two servers, pinned as 34b's own falsification. --- builtin/runtime/lsp.lua | 12 ++++++++++++ tests/lsp_dispatch_seams_acceptance.rs | 22 ++++++++++++++-------- 2 files changed, 26 insertions(+), 8 deletions(-) diff --git a/builtin/runtime/lsp.lua b/builtin/runtime/lsp.lua index eac5753..6a837f6 100644 --- a/builtin/runtime/lsp.lua +++ b/builtin/runtime/lsp.lua @@ -1683,6 +1683,18 @@ local function purge_dead_pending() for rid, entry in pairs(pend) do -- Absent or terminal, or the same sid running a NEW generation: -- in every case the request this entry awaits is unanswerable. + -- + -- The generation half is **defensive and not covered by the + -- acceptance suite**, stated plainly rather than left to look + -- tested. Reaching it requires a crash and its restart to both + -- fall inside a gap with no `_async.tick` — the crash backoff is + -- 500ms (`src/lsp.rs:1007`), so any tick during that window sees + -- `crashed` and the absent-or-terminal test above fires first. A + -- stalled or idle editor can produce such a gap, and then this is + -- the only thing standing between a one-shot and waiting forever + -- on a reply the dead generation owed. Every attempt to stage it + -- deterministically ended up exercising the `crashed` path + -- instead, so it is kept as insurance and labelled as such. if attempt == nil or attempt ~= entry.attempt then dead[#dead + 1] = rid end diff --git a/tests/lsp_dispatch_seams_acceptance.rs b/tests/lsp_dispatch_seams_acceptance.rs index d240ba4..cb9acff 100644 --- a/tests/lsp_dispatch_seams_acceptance.rs +++ b/tests/lsp_dispatch_seams_acceptance.rs @@ -336,16 +336,22 @@ fn acc33_raising_response_handler_does_not_stop_the_drain() { } // --------------------------------------------------------------------------- -// Acceptance 32 — the one-shot is removed BEFORE invocation. +// Acceptance 32 — the one-shot is removed exactly once, whether or not +// the handler raises. // -// Observed rather than asserted structurally: the handler raises, and -// the server is then stopped. If removal happened only on a clean -// return — or not at all — the purge below would invoke the same handler -// a second time with an error. The count is what pins it. +// Named for what it pins rather than for the framing's wording. Q#LN9 +// specifies removal *before* invocation, and the implementation does +// that — but bite-testing showed the before/after ordering is not +// observable on its own: `pcall` catches the raise either way, so +// removal after the call is behaviorally identical unless a handler +// re-enters the drain, which nothing does. What IS observable, and what +// this pins, is that removal is **unconditional**: the bite that moves +// it inside `if ok then` fails here 2 != 1, because the surviving +// registration gets invoked a second time by the purge. // --------------------------------------------------------------------------- #[test] -fn acc32_response_one_shot_is_removed_before_invocation() { +fn acc32_response_one_shot_is_removed_even_when_the_handler_raises() { let fx = Fixture::new(); let mut state = editor(); attached_rust(&mut state, &fx); @@ -375,8 +381,8 @@ fn acc32_response_one_shot_is_removed_before_invocation() { assert_eq!( eval::(&state, "return _G.calls"), 1, - "a delivered one-shot must not be re-invoked by the purge — it \ - was removed before the raising handler ran, not after" + "a delivered one-shot must not be re-invoked by the purge — \ + removal is unconditional, not gated on a clean return" ); } From aff3a60332892c63913194d2b693512524183679 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 16:01:18 -0400 Subject: [PATCH 05/32] docs: correct the purge's reachable-leak claim; record the 3a lane MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Rev 5 said acceptance 34's second edge was a killed buffer. Implementing it showed that is false: the Rust core fires exactly five hooks — buffer.after-edit, buffer.after-load, buffer.after-switch, frontend.detached, process.after-tick — and there is **no buffer-kill hook**, so lsp.lua never tears an attachment down and the drain keeps reaching that server. The premise (the drain builds its sid list from `attachments`) was right; the inference needed attachments to be removed on kill, and nothing removes them. The reachable leak has the same root cause by a different path. `attach_buffer` drops a sid from `attachments` the moment `server_is_live` reports false and rebuilds against a fresh server — so `crashed` / `stopped` is the event *least* likely to be drained, and an event-driven purge leaks in exactly the case it exists for. The purge therefore polls `pmacs.lsp.list()`, which enumerates the manager directly. Acceptance 34's second half now exercises a server in **no** attachment, which is the shape that discriminates: bitten, an event-driven purge fails it while the attached case still passes. §0.1 finding 6, Q#LN9, and acceptance 34 all updated; the wrong wording is left visible with its correction rather than quietly replaced, since the mistake is the useful part. Ledger gains the Stage 3a lane: branch, worktree, what ships, both corrected claims, the `install_async` load-order trap, the recorded bites, the one knowingly unpinned guard, and gate results. --- docs/active-work.md | 54 +++++++++++++++++++++++++++- docs/lean4-mode-framing.md | 74 ++++++++++++++++++++++++++------------ 2 files changed, 104 insertions(+), 24 deletions(-) diff --git a/docs/active-work.md b/docs/active-work.md index b426f38..3cf1a8e 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -54,7 +54,7 @@ git status --short --branch The `git log` command must expose `0dd16a5` or a newer intentional main. If it does not, stop and repair the remote/fetch configuration. -## Lean 4 lane (Arc 8) — Stage 1 MERGED; Stage 2 IN REVIEW (PR #161) +## Lean 4 lane (Arc 8) — Stages 1+2 MERGED; Stage 3a IN REVIEW - Stage 1 **merged as #160** (`main` @ `0827dd1`, 2026-07-25, one review round, all twelve checks green). Branch `githubsucks/lean4-stage1` @@ -188,6 +188,58 @@ If it does not, stop and repair the remote/fetch configuration. suites**; `git diff --check` clean. The sweep needs an isolated `XDG_CONFIG_HOME` and `-- --skip basedpyright`. +### Stage 3a — dispatch seams + `pmacs.fs.canonicalize` (branch `lean4-stage3a-seams`) + +- Worktree `../pmacs-lean-stage3`, branched off `githubsucks/main` @ + `46a1b8f`. Carries framing **rev 5** (the Stage 3 split) as its first + two commits, then the implementation, then a bite-driven correction. +- **Stage 2 merged as #161** (`main` @ `46a1b8f`, 2026-07-25, two review + rounds). COHERENCE.md §7 records the slice; §1.2 records the dead + `pmacs.error` channel found landing it. +- **Framing rev 5 splits Stage 3 into 3a and 3b** because rev 4 broke its + own §4 rule — the row read "two `lsp.lua` generalizations" under prose + claiming Stage 3 was Lean-only. One generalization shipped as Stage 2; + the other (Q#LN9's seams) is the shared event drain, so it is now its + own substrate stage. 3a and 3b are **strictly sequential** — 3b's + subscriber is written against 3a's seam and both touch `lsp.lua`. +- Ships: `pmacs.lsp.on_notification` / `on_response`, two arms in + `handle_server_requests`, a pending-response purge, and + `pmacs.fs.canonicalize` (Q#LN20). No protocol change, no Lean content. +- **Two framing claims were corrected during implementation**, both + recorded in §0.1 finding 6 and in the round-2 commit: + 1. The reachable leak is **not** a killed buffer. The Rust core fires + exactly five hooks (`buffer.after-edit`, `buffer.after-load`, + `buffer.after-switch`, `frontend.detached`, `process.after-tick`) — + **there is no buffer-kill hook**, so nothing tears an attachment + down and the drain keeps reaching that server. The real path is + `attach_buffer` dropping a dead sid from `attachments` and + rebuilding against a fresh server, which makes `crashed`/`stopped` + the event *least* likely to be drained. Hence the purge polls + `pmacs.lsp.list()` rather than riding the drain. + 2. Acceptance 32 does **not** pin "removed before invocation" — + `pcall` catches the raise either way, so before/after is + unobservable without a re-entrant drain. It pins removal being + **unconditional**; renamed accordingly. +- **`pmacs._fs` is installed from `install_async`, not `install_project`**, + purely for load order: `make_workspace` runs *after* `fs.lua` is + evaluated, so a canonicalizer placed there reads nil. This cost one + failing run to discover and is the kind of thing to check first. +- Bites recorded (all against the committed tree): removal gated on a + clean return → acc32 fails 2 != 1; an event-driven purge → the + no-attachment case fails "never called" while the attached case still + passes; a resolver without `canonicalize` → two servers (34b's own + falsification, which ships as a test). +- **Known unpinned:** the purge's generation (`attempt`) check. Reaching + it needs a crash *and* its restart to fall in a gap with no + `_async.tick`; the backoff is 500ms, so any tick sees `crashed` first + and the absent-or-terminal arm fires. Labelled as defensive in the + code rather than left looking covered. +- Verification on this branch: `cargo fmt --check` clean; strict + workspace Clippy clean; 1,826 default + 2,003 CRDT library tests; + dispatch seams 14/14; multi-root 13/13; M4 121; required GPU 155; + **isolated-config workspace sweep 3,188 across 93 suites, zero + failures**; `git diff --check` clean. + ## Dired lane — framing APPROVED; Stage 0 MERGED, Stage 1 next - Approved framing: `docs/dired-framing.md` (revision 5), landing as its diff --git a/docs/lean4-mode-framing.md b/docs/lean4-mode-framing.md index 317d74f..4869252 100644 --- a/docs/lean4-mode-framing.md +++ b/docs/lean4-mode-framing.md @@ -241,11 +241,26 @@ Confirmations, recorded because each was load-bearing and unverified: narrowing is load-bearing.** `handle_server_requests` builds its sid list from `attachments`, and `push_event` appends with no cap. So a subscriber fires only for a server with a live attachment, and an - unattached server's event queue grows unboundedly. This bites - acceptance 34 directly: kill the buffer with a request outstanding - and the pending purge never runs — the leak that criterion exists to - prevent. Q#LN9 now states the contract and acceptance 34 drives it - through the buffer-kill path rather than the server-death path alone. + unattached server's event queue grows unboundedly. + + *Corrected during implementation (rev 5, round 2).* Rev 5 first + claimed the reachable leak was a killed buffer. **That was wrong.** + The Rust core fires exactly five hooks — `buffer.after-edit`, + `buffer.after-load`, `buffer.after-switch`, `frontend.detached`, + `process.after-tick` — and **there is no buffer-kill hook at all**, + so `lsp.lua` never tears an attachment down and the drain keeps + reaching that server. The premise was right and the inference was + not: it needed attachments to be removed on kill, and nothing + removes them. + + The reachable leak is a different path with the same root cause. + `attach_buffer` drops a sid from `attachments` the moment + `server_is_live` reports false, rebuilding against a fresh server — + so the `crashed` / `stopped` event that should trigger a purge is + **precisely the one most likely to go undrained**. An event-driven + purge leaks exactly when it matters. Q#LN9 therefore drives the + purge off `pmacs.lsp.list()`, which enumerates the manager directly + and is unaffected by attachment bookkeeping. 7. **The `cfg.restart` gap is still open** (recorded landing #161): `ensure_server` never forwards `pmacs.lsp.config[lang].restart` to `pmacs.lsp.spawn`, so the field is silently dropped on auto-attach. @@ -264,9 +279,9 @@ rather than a bare root), `ensure_server` 527 → **610**, 12798 → **12827**, `pair.lua` 213 → **229**, and `compile.lua` 264 → **266**. Verified good and left alone: `listview.lua:138`, `src/lsp.rs:264`, `src/diag.rs:50`, `src/process.rs:193`, -`src/project.rs:145`, and the `mod.rs` binding-block citations. The pre-#161 line numbers inside Q#LN15 are -left as written: that stage has landed and its citations are historical -record, not navigation. +`src/project.rs:145`, and the `mod.rs` binding-block citations. The +pre-#161 line numbers inside Q#LN15 are left as written: that stage has +landed and its citations are historical record, not navigation. ## 1. What ships @@ -1018,13 +1033,24 @@ attachment.** `handle_server_requests` builds its sid list from `attachments`, so a server with no attached buffer is never drained — and `push_event` appends with no cap, so that server's queue grows unboundedly. Both facts are pre-existing and neither is Stage 3a's to -fix, but the second one turns the first into a leak with a name: **a -buffer killed while a request is outstanding never runs the purge**, -because the purge rides the drain that the attachment was gating. That is -the exact failure acceptance 34 exists to prevent, reachable through the -ordinary `C-x k`. So the purge is driven from both edges — the server-death -transition *and* attachment teardown — and acceptance 34 exercises the -buffer-kill path, which is the one a user can actually reach. +fix. What they change is where the purge may be wired. + +**The purge must not ride the drain.** `attach_buffer` removes a sid +from `attachments` as soon as `server_is_live` reports false and rebuilds +the attachment against a fresh server, so a `crashed` / `stopped` event +is the event *least* likely to be drained — the drain stops visiting +that server at almost exactly the moment the event is queued. A purge +triggered by observing that event therefore leaks in the case it exists +to handle. + +So the purge polls **`pmacs.lsp.list()`** after each drain instead. That +call enumerates the manager directly and is unaffected by attachment +bookkeeping, which is what makes it the right authority: a sid that is +absent, terminal, or running a new generation settles its pending +one-shots with an error, whether or not anything ever drained it. +Acceptance 34's second half exercises a server that is in **no** +attachment, because that is the shape an event-driven purge fails and a +polled one survives. The uncapped queue is recorded as a named deferral (§6) rather than fixed here: bounding it is a policy question about which events may be dropped, @@ -1629,14 +1655,16 @@ the blast radius. registered, `workspace/applyEdit` in the same drain is still handled; a raising response handler does not stop later events in that drain. Mirrors the notification-side pins above. -- **34.** **Pending-purge pin, both edges.** A server that dies with a response - outstanding invokes the pending one-shot with an error and clears it. - **And** — the case round 4 found reachable and rev 4 missed — killing - the *buffer* with a request outstanding does the same, rather than - stranding the registration behind a drain that no longer runs for - that server. The second half must be shown to fail against a - purge wired only to the server-death transition; otherwise this - criterion is satisfied by the implementation that leaks. +- **34.** **Pending-purge pin, both edges.** A server that dies with a + response outstanding invokes the pending one-shot with an error and + clears it. **And** a server that is in **no attachment** does the + same, rather than stranding the registration behind a drain that never + visits it. The second half must be shown to fail against a purge + wired to a death event seen in the drain; otherwise this criterion is + satisfied by the implementation that leaks. (Rev 5 first worded the + second edge as a killed buffer; there is no buffer-kill hook, so + nothing removes the attachment and that path does not leak. Corrected + in round 2 — see §0.1 finding 6.) - **34a.** **Canonicalizer pin (Q#LN20).** `pmacs.fs.canonicalize` resolves a symlinked and dot-segmented path to the same string as the real path, and returns nil for a nonexistent one. Fixture builds the symlink From a9ef257930ce25d14e87148a4a04918c70939da5 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 16:53:27 -0400 Subject: [PATCH 06/32] fix(lsp): decline a non-UTF-8 canonicalization; round 1 review MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Six findings from review, one of them a real defect. **`canonicalize` could emit a path that exists nowhere.** `p.display().to_string()` substitutes U+FFFD for non-UTF-8 bytes, so a resolution landing on such a path returned a plausible-looking string that does not exist on disk — worse than nil, because this value becomes a server-affinity key through `file_uri_for` and would silently fail to round-trip, while the doc promised nil for anything unresolvable. Now `.and_then(|p| p.to_str().map(str::to_owned))`: unrepresentable is a decline, matching how the fs layer already treats non-UTF-8 symlink targets. Pinned by a new acceptance case that reaches a non-UTF-8 target through an **ASCII** symlink, so the input is representable and only the resolved output is not — the case a UTF-8 check on the argument would miss. Bitten: restoring `display()` fails it. The other five: - `on_response`'s doc comment now warns that registering against a server with no attached buffer is fire-on-death, not fire-on-reply, because the drain visits only attached sids. It looks exactly like a hung request while debugging, and 3b is the first caller likely to hit it. - `deliver_response`'s comment still carried the pre-correction rationale ("removed BEFORE invocation ... must not be re-entered") — the claim the bite disproved. It now says what is true: removal is unconditional, before-vs-after is unobservable without a re-entrant drain, and the reachable bug is gating removal on a clean return. - Dropped `server_attempt`'s unused second return. - Deleted a vacuous assertion in the no-attachment test (counting `_G` entries to assert "lua globals are readable") — scaffolding that pinned nothing, the exact shape the project's own lesson flags. - `#[cfg(unix)]` on the three symlink-dependent tests. Gates re-run in full. The sweep's first pass tripped `composition_overhead_under_ten_percent` at 18.8% against a 10% budget; it passes 3/3 in isolation here, passes in isolation on main, and the same run reported the realistic-frame overhead as **-4.6%** — a negative figure is measurement noise, not added work. Nothing in this diff is on the render path. Rerun of the full sweep: 3,189 across 93 suites, zero failures. --- builtin/runtime/lsp.lua | 22 +++++++++-- src/lua_bindings/mod.rs | 11 +++++- tests/lsp_dispatch_seams_acceptance.rs | 53 +++++++++++++++++++++++--- 3 files changed, 75 insertions(+), 11 deletions(-) diff --git a/builtin/runtime/lsp.lua b/builtin/runtime/lsp.lua index 6a837f6..fd44f1a 100644 --- a/builtin/runtime/lsp.lua +++ b/builtin/runtime/lsp.lua @@ -1584,10 +1584,10 @@ local function server_attempt(sid) local skey = tostring(sid) for _, info in ipairs(pmacs.lsp.list()) do if tostring(info.id) == skey then - return info.attempt or 0, info.state and info.state.kind + return info.attempt or 0 end end - return nil, nil + return nil end -- fn(sid, params); persistent, fires for every server. @@ -1605,6 +1605,15 @@ end -- fn(result, err); ONE-SHOT, keyed to the exact request. -- `request_id` is what `pmacs.lsp.send_request` returned. +-- +-- **Register only against a server with an attached buffer.** The drain +-- that delivers replies visits only sids present in `attachments`, so a +-- one-shot on an unattached server will not fire on its reply — the +-- reply sits in that server's queue and the handler is invoked only when +-- the purge below decides the server is gone. That is fire-on-death, not +-- fire-on-reply, and it looks exactly like a hung request while +-- debugging. The attach path is the ordinary way to get a sid; a +-- hand-spawned one from `init.lua` is the case to watch. function pmacs.lsp.on_response(sid, request_id, fn) if not sid or type(request_id) ~= "number" or type(fn) ~= "function" then error("pmacs.lsp.on_response(sid, request_id, fn): want sid, number, function") @@ -1641,8 +1650,13 @@ local function deliver_response(sid, ev) if not pend then return end local entry = pend[ev.request_id] if not entry then return end - -- Removed BEFORE invocation: a handler that raises must not be - -- re-entered by a later event carrying the same id. + -- Removed UNCONDITIONALLY, so a handler that raises is still retired + -- and cannot be invoked a second time by the purge. Removing first is + -- the defensive order and costs nothing, but it is not what defends + -- against re-invocation: `pcall` catches the raise either way, so + -- before-vs-after is unobservable without a re-entrant drain. The + -- reachable bug is gating removal on a clean return, which acceptance + -- 32 bites (2 != 1). pend[ev.request_id] = nil if next(pend) == nil then pending_responses[skey] = nil end local ok, err = pcall(entry.fn, ev.result, ev.error) diff --git a/src/lua_bindings/mod.rs b/src/lua_bindings/mod.rs index 369b810..ac33537 100644 --- a/src/lua_bindings/mod.rs +++ b/src/lua_bindings/mod.rs @@ -6550,9 +6550,18 @@ pub fn install_async( // symlink is ordinary, and raising would surface through // `resolve_root_fn`'s pcall as a config bug, which it is // not. + // + // `to_str`, NOT `display()`. A resolution that lands on + // non-UTF-8 bytes has no faithful string form, and + // `display()` would substitute U+FFFD and hand back a + // path that does not exist on disk — strictly worse than + // nil here, because this value becomes a server-affinity + // key via `file_uri_for` and would silently fail to + // round-trip. Unrepresentable is a decline, matching how + // the fs layer already treats non-UTF-8 symlink targets. Ok(std::fs::canonicalize(&path) .ok() - .map(|p| p.display().to_string())) + .and_then(|p| p.to_str().map(str::to_owned))) })?, )?; pmacs.set("_fs", fs_priv)?; diff --git a/tests/lsp_dispatch_seams_acceptance.rs b/tests/lsp_dispatch_seams_acceptance.rs index cb9acff..bc0b40d 100644 --- a/tests/lsp_dispatch_seams_acceptance.rs +++ b/tests/lsp_dispatch_seams_acceptance.rs @@ -493,12 +493,6 @@ fn acc34_purge_reaches_a_server_that_is_in_no_attachment() { ); settle(&mut state); - let attached: i64 = eval( - &state, - "local n = 0 for _ in pairs(_G) do n = n + 1 end return n", - ); - assert!(attached > 0, "lua globals are readable"); - exec( &state, r#" @@ -528,6 +522,7 @@ fn acc34_purge_reaches_a_server_that_is_in_no_attachment() { // --------------------------------------------------------------------------- #[test] +#[cfg(unix)] fn acc34a_canonicalize_resolves_symlinks_and_dot_segments() { let fx = Fixture::new(); fx.write("pkg/sub/a.txt", "x\n"); @@ -576,6 +571,7 @@ fn server_count(state: &EditorState) -> i64 { } #[test] +#[cfg(unix)] fn acc34b_canonicalizing_resolver_reuses_one_server_across_a_symlink() { let fx = Fixture::new(); fx.write("proj/Cargo.toml", "[package]\nname = \"p\"\n"); @@ -627,6 +623,7 @@ fn acc34b_canonicalizing_resolver_reuses_one_server_across_a_symlink() { } #[test] +#[cfg(unix)] fn acc34b_falsified_by_a_resolver_that_skips_canonicalization() { let fx = Fixture::new(); fx.write("proj/Cargo.toml", "[package]\nname = \"p\"\n"); @@ -671,3 +668,47 @@ fn acc34b_falsified_by_a_resolver_that_skips_canonicalization() { spawn two servers — this is what 34b's positive case rules out" ); } + +// --------------------------------------------------------------------------- +// Acceptance 34a, non-UTF-8 arm — an unrepresentable resolution declines +// rather than returning a lossy string. +// +// Review finding on PR #167: `display().to_string()` substitutes U+FFFD, +// which would hand back a path that does not exist on disk. That is +// strictly worse than nil here, because the value becomes a +// server-affinity key via `file_uri_for` and would silently fail to +// round-trip. Bites against the `display()` form, which returns a +// non-nil string for this fixture. +// --------------------------------------------------------------------------- + +#[test] +#[cfg(unix)] +fn acc34a_canonicalize_declines_a_non_utf8_resolution() { + use std::ffi::OsStr; + use std::os::unix::ffi::OsStrExt as _; + + let fx = Fixture::new(); + // 0xFF is not valid UTF-8 in any position. + let raw = OsStr::from_bytes(b"bad-\xffname"); + let target = fx.root.join(raw); + std::fs::write(&target, "x\n").unwrap(); + // Reached through an ASCII symlink, so the *input* is representable + // and only the resolved output is not — which is the case + // `to_str()` has to catch and a UTF-8-only input check would miss. + let link = fx.dir("ascii-link"); + std::os::unix::fs::symlink(&target, &link).unwrap(); + + let state = editor(); + let got: String = eval( + &state, + &format!( + "return tostring(pmacs.fs.canonicalize(\"{}\"))", + lua_str(&link) + ), + ); + assert_eq!( + got, "nil", + "a resolution that lands on non-UTF-8 bytes must decline, not \ + return a U+FFFD-substituted path that exists nowhere" + ); +} From b70393762edd14f7fa5bd4602299e33066ad1e2c Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 17:03:10 -0400 Subject: [PATCH 07/32] test(lsp): gate the non-UTF-8 fixture on linux, not merely unix MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit CI round 1: both macOS jobs failed on the acceptance case added last commit. APFS enforces valid UTF-8 in filenames, so `std::fs::write` with a 0xFF byte in the name fails with EILSEQ ("Illegal byte sequence") before `pmacs.fs.canonicalize` is ever called. The fixture cannot be built there. That is a filesystem refusing to represent the case, not a behavioral difference: the subject — `to_str()` returning None for a non-UTF-8 resolution — is platform-independent Rust, and the Linux run pins it. `#[cfg(unix)]` was the wrong granularity; review had asked for unix gating on the symlink tests and I applied the same gate here without checking whether the filesystem, rather than the API, was the constraint. Gated `#[cfg(target_os = "linux")]` with the reason in place, rather than skipped at runtime, so a future failure here is a real failure and not a silent no-op. Ledger records both CI-round facts: this one, and that `composition_overhead_under_ten_percent` is load-sensitive under a parallel workspace sweep (it reported -4.6% realistic overhead in the same run that tripped its 10% budget at 18.8%, which is noise, not work). --- docs/active-work.md | 18 +++++++++++++++--- tests/lsp_dispatch_seams_acceptance.rs | 12 +++++++++++- 2 files changed, 26 insertions(+), 4 deletions(-) diff --git a/docs/active-work.md b/docs/active-work.md index 3cf1a8e..aef6edc 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -236,9 +236,21 @@ If it does not, stop and repair the remote/fetch configuration. code rather than left looking covered. - Verification on this branch: `cargo fmt --check` clean; strict workspace Clippy clean; 1,826 default + 2,003 CRDT library tests; - dispatch seams 14/14; multi-root 13/13; M4 121; required GPU 155; - **isolated-config workspace sweep 3,188 across 93 suites, zero - failures**; `git diff --check` clean. + dispatch seams 15/15 on Linux (14 on macOS — see below); multi-root + 13/13; M4 121; required GPU 155; **isolated-config workspace sweep + 3,189 across 93 suites, zero failures**; `git diff --check` clean. +- **Two flakes/portability facts from CI round 1, both worth keeping:** + 1. `composition_overhead_under_ten_percent` tripped once in a local + sweep at 18.8% against a 10% budget, then passed 3/3 in isolation + here, passed in isolation on main, and passed a full sweep rerun. + The tell is in its own output: the same run reported realistic-frame + overhead as **-4.6%**, and a negative figure is measurement noise, + not added work. Load-sensitive under a parallel `--workspace` run. + 2. **A non-UTF-8 filename fixture cannot be built on macOS.** APFS + enforces valid UTF-8, so `std::fs::write` fails with EILSEQ + ("Illegal byte sequence") before the code under test is reached. + `#[cfg(unix)]` is NOT sufficient for such a fixture — + `#[cfg(target_os = "linux")]` is. Cost one red CI round to learn. ## Dired lane — framing APPROVED; Stage 0 MERGED, Stage 1 next diff --git a/tests/lsp_dispatch_seams_acceptance.rs b/tests/lsp_dispatch_seams_acceptance.rs index bc0b40d..f644367 100644 --- a/tests/lsp_dispatch_seams_acceptance.rs +++ b/tests/lsp_dispatch_seams_acceptance.rs @@ -679,10 +679,20 @@ fn acc34b_falsified_by_a_resolver_that_skips_canonicalization() { // server-affinity key via `file_uri_for` and would silently fail to // round-trip. Bites against the `display()` form, which returns a // non-nil string for this fixture. +// +// **Linux-gated, and `cfg(unix)` was not enough** — CI caught that. +// APFS enforces valid UTF-8 in filenames, so on macOS the `write` below +// fails with EILSEQ ("Illegal byte sequence") before the code under test +// is ever reached: the fixture cannot be built there. That is a +// filesystem refusing to represent the case, not a behavioral +// difference — the subject itself, `to_str()` returning None, is +// platform-independent Rust. Gated explicitly rather than skipped at +// runtime, so a future failure here is a real failure and not a silent +// no-op. // --------------------------------------------------------------------------- #[test] -#[cfg(unix)] +#[cfg(target_os = "linux")] fn acc34a_canonicalize_declines_a_non_utf8_resolution() { use std::ffi::OsStr; use std::os::unix::ffi::OsStrExt as _; From 1e1be67b49dc1b90783ed3b2fd8df3c776cac8ef Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 17:40:02 -0400 Subject: [PATCH 08/32] feat(lean): the Lean 4 language server (Arc 8 Stage 3b) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Framing Q#LN7, Q#LN8, Q#LN16; acceptance 22–28, 24a/24b, 35, 36, 36a, 37. Stacked on Stage 3a (#167), whose notification/response seams and `pmacs.fs.canonicalize` this consumes. No protocol change; the only Rust outside the test helper is one `include_str!` line. **The Lake-aware root (Q#LN8).** `pmacs.project.detect` cannot express this rule — it is innermost-wins by construction, and a Lake package's `lean-toolchain` sits at the outermost level, so a file under `/.lake/packages/dep/` belongs to ``'s server rather than `dep`'s. The resolver walks up collecting markers and returns the outermost, stopping at `pmacs.project.search_boundary()` so a stray marker above a fixture cannot leak in. Two things about the marker test are easy to get wrong in opposite directions, and both are pinned. `io.open` **succeeds on a directory**, so a truthiness check accepts a `lean-toolchain` directory; but requiring a non-nil read rejects an **empty** `lean-toolchain`, which is a legitimate marker — `locate-dominating-file` semantics are existence, not content. The discriminator is `read`'s second return: decline only on a non-nil error. Acceptance 24a and 24b each fail against the implementation that satisfies only the other; both bites are recorded. The root is canonicalized once up front, because a configured root reaches `file_uri_for` verbatim and that URI is the affinity key (#161). Canonicalizing the starting directory suffices — every ancestor of a canonical path is canonical, since the walk only strips components. **`lake serve` with a lazy probe and a one-shot latch (Q#LN7).** Nothing runs at init: `pmacs.lsp.config` is declarative, and spawning a process at startup for every user, Lean-using or not, is the cost rev 1 refused. Both the probe and the server spawn are gated on a real Lean attachment. The probe cannot gate the first attach — there is no blocking process run, so its verdict arrives after `ensure_server` has already decided. Hence the optimistic spawn, with the probe and latch correcting it. A non-zero probe exit is deliberately NOT a trigger: §2.9's elan-shim case makes `lake --version` fail on machines where `lake serve` still works, and the server-failure latch covers that better. The probe answers only the question failure detection would answer slowly — an old-but-working lake that starts a useless server. The latch stops the failing server **before** spawning the fallback, and that ordering is load-bearing rather than defensive: the spec default is `OnCrash`, the termination handler never consults the exit code, and `maybe_restart` has no attempt ceiling, so a broken `lake` respawns forever underneath the latch. `pmacs.lsp.stop` sets `restart = Never`, which is what disarms it. Bitten: removing the stop fails acceptance 36. The swap rewrites `command` and `args` only, so a user's `env`, `settings`, `init_options` and `root` survive — a wholesale table replacement would discard their `init.lua` at the moment they are least likely to notice. **`waitForDiagnostics` (Q#LN16)** resolves through Stage 3a's response seam, with `M-x lean.wait-for-diagnostics` on top. `$/lean/fileProgress` subscribes on the notification seam and is pinned end-to-end through a new `leanprogress` mode on the fake server rather than by calling the handler directly — the wiring is the only part that can break. **Attribution (COHERENCE §9/§1.2).** The probe spawns as `lean:lake-version-probe`, so a user wondering why their editor touched `lake` finds an owner in `pmacs.process.list`. The latch reports through `pmacs.editor.set_status` — the channel that exists — and acceptance 36a observes that channel, so a report made only through the undefined `pmacs.error` would fail it. **Stage 1's acceptance 12 is updated, half superseded.** It asserted `pmacs.lsp.config.lean4 == nil` to guard against a Stage-3 front-run; Stage 3b is that stage, so keeping it would pin the opposite of the intended behavior. The half that survives is the one about restraint, and it matters more now: constructing an editor spawns nothing even though the config exists and names `lake`, and opening a Lean buffer with no server configured spawns nothing either. That is what holds Q#LN7's "not at init" promise. Bites recorded, all against the committed tree: bare `io.open` -> 24a fails, 24b passes; require-non-nil-read -> 24b fails, 24a passes; no canonicalization -> the symlink case spawns two servers; no stop before fallback -> acceptance 36 fails. --- builtin/runtime/lean.lua | 359 +++++++++++++++++ src/bin/pmacs_fake_lsp.rs | 21 + src/editor.rs | 11 + tests/lean4_server_acceptance.rs | 640 +++++++++++++++++++++++++++++++ tests/lean4_stage1_acceptance.rs | 62 +-- 5 files changed, 1069 insertions(+), 24 deletions(-) create mode 100644 builtin/runtime/lean.lua create mode 100644 tests/lean4_server_acceptance.rs diff --git a/builtin/runtime/lean.lua b/builtin/runtime/lean.lua new file mode 100644 index 0000000..3fe73da --- /dev/null +++ b/builtin/runtime/lean.lua @@ -0,0 +1,359 @@ +-- builtin/runtime/lean.lua --- Arc 8 Stage 3b: the Lean 4 language server. +-- +-- Framing: `docs/lean4-mode-framing.md` Q#LN7 (lake serve + probe + +-- fallback latch), Q#LN8 (Lake-aware outermost root), Q#LN16 +-- (waitForDiagnostics). Stage 1 shipped the grammar, mode, comment +-- strings and pair set; Stage 3a shipped the notification/response +-- seams and `pmacs.fs.canonicalize` this file consumes. +-- +-- Loaded after `lsp.lua`, which owns `pmacs.lsp.config` and the drain. + +local M = {} + +-- Q#LN8 — the Lake-aware root ----------------------------------------- +-- +-- `pmacs.project.detect` cannot express this rule. It is innermost-wins +-- by construction, and a Lake package's `lean-toolchain` sits at the +-- OUTERMOST level: a file under `/.lake/packages/dep/Foo.lean` +-- belongs to ``'s server, not to `dep`'s, because `lake serve` is +-- bound to one package and analyzes its dependencies from inside it. +-- Inverting `detect` globally would change Rust/Go/Node roots for every +-- user, so the rule lives here as a function-valued `config.root` — +-- the generalization Stage 2 (#161) added for exactly this. + +-- The marker test, and the two ways to get it wrong. +-- +-- `pmacs.fs.stat` is UNUSABLE here: it returns an awaitable handle +-- (`fs.lua`), and this runs synchronously inside `ensure_server` <- +-- `attach_buffer` <- the `buffer.after-load` hook, where there is no +-- coroutine to await on. The Lua stdlib's `io.open` is the only +-- synchronous existence check available. +-- +-- But `io.open` alone is wrong in BOTH directions: +-- * it SUCCEEDS on a directory (probed), so a truthiness test would +-- accept a `lean-toolchain` directory as a marker; and +-- * requiring a non-nil read rejects an EMPTY `lean-toolchain`, which +-- is a legitimate marker — `locate-dominating-file` semantics are +-- existence, not content. +-- The discriminator is `read`'s SECOND return (probed on LuaJIT 2.1): +-- file with content -> "l", no error -> marker +-- empty file -> nil, NO error -> marker +-- directory -> nil, "Is a directory" -> decline +-- missing -> io.open returns nil -> decline +-- so: decline only on a non-nil `err`. This needs no per-platform +-- re-probe, because both directory behaviors are declines — a platform +-- whose `fopen` refuses directories fails at `io.open` instead. There +-- is no platform where a directory both opens and yields a byte. +local function has_toolchain(dir) + local f = io.open(dir .. "/lean-toolchain", "r") + if not f then return false end + local _, err = f:read(1) + f:close() + return err == nil +end + +local function parent_of(dir) + local up = dir:match("^(.*)/[^/]+$") + if up == nil or up == dir or up == "" then return nil end + return up +end + +-- The walk stops at `pmacs.project.search_boundary()`. Not politeness: +-- `detect_project_within` (`src/project.rs`) exists precisely so a +-- stray marker above a temp fixture cannot leak into detection, and a +-- Lua walk that ignored the boundary would break that contract — and +-- make acceptance 23's outermost assertion non-hermetic against any +-- `lean-toolchain` sitting above the test's tempdir. +local function within_boundary(dir, boundary) + if not boundary then return true end + return dir == boundary or dir:sub(1, #boundary + 1) == boundary .. "/" +end + +-- Returns the OUTERMOST ancestor holding a `lean-toolchain`, or nil to +-- decline (which falls through to `pmacs.project.detect`, then the +-- file's own directory). +-- +-- **The result is canonical, and must be.** A configured root — which +-- this is — reaches `file_uri_for` verbatim and that URI is the +-- server-affinity key (#161). `pmacs.editor.file_path()` collapses `.` +-- and `..` lexically but leaves symlinks intact, so one package opened +-- through a symlink and through its real path would otherwise spawn two +-- `lake serve` processes. Canonicalizing ONCE up front is enough: +-- every ancestor of a canonical path is itself canonical, since the +-- walk only strips trailing components. +-- +-- If canonicalization fails (deleted file, broken symlink) the resolver +-- declines rather than returning a path it cannot vouch for. +function M.root_for(path) + if type(path) ~= "string" then return nil end + local dir = path:match("^(.*)/[^/]*$") + if not dir then return nil end + dir = pmacs.fs.canonicalize(dir) + if not dir then return nil end + local boundary + local ok, b = pcall(pmacs.project.search_boundary) + if ok then boundary = b end + -- The boundary is canonicalized at set time (`set_search_boundary`), + -- so comparing it against a canonical `dir` is apples to apples. + local outermost = nil + local cur = dir + while cur and within_boundary(cur, boundary) do + if has_toolchain(cur) then outermost = cur end + cur = parent_of(cur) + end + return outermost +end + +-- Q#LN7 — `lake serve`, with a lazy probe and a one-shot latch -------- +-- +-- `pmacs.lsp.config.lean4` is declarative and must stay cheap: spawning +-- a process at startup for every user, Lean-using or not, is the cost +-- rev 1 refused. So no probe runs here — it runs on the first `.lean` +-- attach, below. +pmacs.lsp.config.lean4 = pmacs.lsp.config.lean4 or { + command = "lake", + args = { "serve" }, + root = M.root_for, + -- No `init_options`: `hasWidgets?` defaults to false, which is the + -- correct posture for a client reading plain goals out of standard + -- messages rather than driving the `$/lean/rpc/*` widget stack. +} + +-- Session state. The latch is one-shot and never re-arms: a user whose +-- toolchain is broken sees one fallback attempt, not a loop. +local probe = { + started = false, -- the `lake --version` probe has been spawned + latched = false, -- the fallback has fired (or been ruled out) + proc = nil, -- process id of the running probe + buf = "", -- accumulated probe stdout + watching = nil, -- sid we are waiting to see fail before initialize + saw_initialized = false, +} + +local function report(msg) + -- COHERENCE §1.2: background work must leave an attributed trace. + -- `pmacs.editor.set_status` is the channel that EXISTS; `pmacs.error` + -- is referenced by fifteen call sites and defined nowhere in + -- production, so it rides along rather than standing alone. + pcall(pmacs.editor.set_status, msg) + if pmacs.error then pcall(pmacs.error, msg) end +end + +-- `lake serve` below 3.1.0 starts a server that cannot answer, which is +-- worse than failing: `lean4-mode` probes for exactly this and falls +-- back to `lean --server`. Parses the leading `x.y` of a version line. +local function version_below_3_1(text) + local major, minor = text:match("(%d+)%.(%d+)") + if not major then return false end + major, minor = tonumber(major), tonumber(minor) + if major < 3 then return true end + return major == 3 and minor < 1 +end + +-- Swap `command`/`args` ONLY. A wholesale table replacement would +-- silently discard a user's `env` / `settings` / `init_options` / `root` +-- from `init.lua` at exactly the moment they are least likely to notice. +local function swap_to_lean_server() + local cfg = pmacs.lsp.config.lean4 + if not cfg then return false end + if cfg.command ~= "lake" then return false end + cfg.command = "lean" + cfg.args = { "--server" } + return true +end + +-- Fire the fallback: stop the failing server FIRST, then swap, then let +-- the next attach spawn afresh. +-- +-- Stopping first is load-bearing, not defensive. The spec default is +-- `LspRestartPolicy::OnCrash`, the termination handler never consults +-- the exit code, and `maybe_restart` has no attempt ceiling — so a +-- broken `lake` respawns forever on a backoff, underneath the latch, +-- producing a loop the latch cannot see the end of. `pmacs.lsp.stop` +-- sets `restart = Never` on the way out, which is what disarms it. The +-- fallback is therefore a FRESH server, not a restart of the old one. +local function fire_latch(sid, why) + if probe.latched then return end + probe.latched = true + if sid then pcall(pmacs.lsp.stop, sid) end + if swap_to_lean_server() then + report("LSP: lean4 " .. why .. "; falling back to `lean --server`") + else + report("LSP: lean4 " .. why) + end + probe.watching = nil +end + +local function drain_probe() + if not probe.proc then return end + local ok, evs = pcall(pmacs.process.events_take, probe.proc) + if not ok or not evs then return end + for _, ev in ipairs(evs) do + if ev.kind == "stdout" or ev.kind == "stderr" then + probe.buf = probe.buf .. tostring(ev.bytes) + elseif ev.kind == "exited" or ev.kind == "signaled" + or ev.kind == "crashed" then + local proc = probe.proc + probe.proc = nil + pcall(pmacs.process.forget, proc) + -- A non-zero exit is NOT a fallback trigger on its own. §2.9: elan + -- shims make `lake --version` exit non-zero with "no default + -- toolchain configured" on a machine where `lake serve` may still + -- be the right command — the server-failure latch covers that + -- case, and covers it better. The probe answers only the ONE + -- question failure detection would otherwise answer slowly: an + -- old-but-working lake that starts a useless server. + if ev.kind == "exited" and ev.code == 0 + and version_below_3_1(probe.buf) then + fire_latch(probe.watching, "lake is older than 3.1.0") + end + end + end +end + +-- The probe cannot gate the first attach. There is no blocking process +-- run (§2.9): `spawn` + `events_take` off a tick is the only shape +-- available, so the verdict arrives AFTER `ensure_server` has already +-- had to decide. Hence the optimistic `lake serve` spawn, with the +-- probe and the latch correcting it. +local function start_probe(root) + if probe.started then return end + probe.started = true + local cfg = pmacs.lsp.config.lean4 + if not cfg or cfg.command ~= "lake" then return end + local spec = { + -- COHERENCE §9: `ProcessSpec.label` is the only identity a process + -- carries, and it is what `pmacs.process.list` renders. A user + -- wondering why their editor touched `lake` finds an owner here. + label = "lean:lake-version-probe", + command = "lake", + args = { "--version" }, + stdin = "null", + } + if root then spec.cwd = root end + local ok, proc = pcall(pmacs.process.spawn, spec) + if ok then probe.proc = proc end + -- A probe that cannot even spawn says nothing the latch will not say + -- more reliably a moment later, so it is not reported here. +end + +-- How the latch observes server failure. +-- +-- There is no event for "died before initialize" — the drain ignores +-- state events. So this polls `pmacs.lsp.list()` on the +-- `process.after-tick` cadence and treats a terminal state reached +-- WITHOUT an intervening `initialized` as the trigger. Watching stops +-- as soon as the server initializes, so an ordinary later crash (a real +-- server dying on a real error) does not silently rewrite the command. +local function poll_latch() + local sid = probe.watching + if not sid or probe.latched then return end + local skey = tostring(sid) + local ok, rows = pcall(pmacs.lsp.list) + if not ok or not rows then return end + for _, info in ipairs(rows) do + if tostring(info.id) == skey then + local kind = info.state and info.state.kind + if kind == "initialized" then + probe.saw_initialized = true + probe.watching = nil + return + end + if kind == "crashed" or kind == "stopped" then + fire_latch(sid, "`lake serve` failed to start") + end + return + end + end + -- Gone from the manager entirely without ever initializing. + fire_latch(nil, "`lake serve` failed to start") +end + +-- Q#LN16 — `textDocument/waitForDiagnostics` -------------------------- +-- +-- A plain request: no position, so Q#LN12's `outbound_position` concern +-- does not apply. Resolves when the server has finished elaborating. +-- Awaited through Stage 3a's response seam. +-- +-- `fn(err)` is called with nil on success. Registering the one-shot +-- requires the server to have an attached buffer — see the note on +-- `pmacs.lsp.on_response`; every caller here comes from an attachment. +function M.wait_for_diagnostics(sid, uri, fn) + local ok, rid = pcall(pmacs.lsp.send_request, sid, + "textDocument/waitForDiagnostics", { uri = uri }) + if not ok then + if fn then pcall(fn, tostring(rid)) end + return nil + end + if fn then + pmacs.lsp.on_response(sid, rid, function(_, err) + fn(err and err.message or nil) + end) + end + return rid +end + +pmacs.command.define { + name = "lean.wait-for-diagnostics", + description = "Wait for the Lean server to finish elaborating this file", + fn = function() + local rec = pmacs.lsp.active_attachment() + if not rec or rec.language ~= "lean4" then + pmacs.editor.set_status("lean: no Lean server for this buffer") + return + end + pmacs.editor.set_status("lean: elaborating…") + M.wait_for_diagnostics(rec.server, rec.uri, function(err) + if err then + pmacs.editor.set_status("lean: " .. tostring(err)) + else + pmacs.editor.set_status("lean: elaboration complete") + end + end) + end, +} + +-- `$/lean/fileProgress` — the elaboration-in-flight signal. Stage 5's +-- goal view reads it to distinguish "no goals" from "not done yet"; +-- here it is recorded so that consumer has something to read and so the +-- notification seam has its first production subscriber. +M.file_progress = {} + +pmacs.lsp.on_notification("$/lean/fileProgress", function(_, params) + local uri = params and params.textDocument and params.textDocument.uri + if type(uri) ~= "string" then return end + M.file_progress[uri] = params.processing or {} +end) + +-- Wiring -------------------------------------------------------------- + +-- Runs after `lsp.lua`'s own `buffer.after-load` subscription, so the +-- attachment already exists. The attachment's `language` IS the Lean +-- test — no separate major-mode lookup, which would be a second source +-- of truth for the same question. +pmacs.hook.add("buffer.after-load", function() + local rec = pmacs.lsp.active_attachment() + if not rec or rec.language ~= "lean4" then return end + if not probe.started then + local path = pmacs.editor.file_path() + start_probe(path and M.root_for(path) or nil) + end + -- Watch only the FIRST Lean server: the latch is per session. + if not probe.latched and not probe.saw_initialized + and probe.watching == nil then + probe.watching = rec.server + end +end) + +pmacs.hook.add("process.after-tick", function() + drain_probe() + poll_latch() +end) + +-- Test seam: acceptance drives the latch deterministically rather than +-- waiting on real process timing. Not part of the public surface. +M._probe = probe +M._fire_latch = fire_latch +M._version_below_3_1 = version_below_3_1 + +pmacs.lean = M diff --git a/src/bin/pmacs_fake_lsp.rs b/src/bin/pmacs_fake_lsp.rs index 5d50b19..5f4fab5 100644 --- a/src/bin/pmacs_fake_lsp.rs +++ b/src/bin/pmacs_fake_lsp.rs @@ -483,6 +483,27 @@ fn main() { } }); write_frame(&mut stdout, &echo); + // Arc 8 Stage 3b: `leanprogress` mode emits one + // `$/lean/fileProgress` covering line 0, so the Lean + // subscriber can be pinned end-to-end through the real + // drain rather than by calling its handler directly. + if mode == "leanprogress" && uri.is_string() { + let progress = serde_json::json!({ + "jsonrpc": "2.0", + "method": "$/lean/fileProgress", + "params": { + "textDocument": { "uri": uri, "version": 1 }, + "processing": [{ + "range": { + "start": { "line": 0, "character": 0 }, + "end": { "line": 1, "character": 0 } + }, + "kind": 1 + }] + } + }); + write_frame(&mut stdout, &progress); + } // Also push a synthetic `publishDiagnostics` // notification with two entries (one Error, one // Warning) so M4.6 tests can exercise the store. diff --git a/src/editor.rs b/src/editor.rs index 79f1225..5f0f134 100644 --- a/src/editor.rs +++ b/src/editor.rs @@ -436,6 +436,17 @@ impl EditorState { include_str!("../builtin/runtime/lsp.lua"), ) .expect("load lsp builtin chunk"); + // Arc 8 Stage 3b: the Lean 4 language server. Loaded after + // lsp.lua because it registers `pmacs.lsp.config.lean4`, + // subscribes on the Stage 3a notification seam, and adds a + // `buffer.after-load` hook that must run AFTER lsp.lua's own + // (it reads the attachment lsp.lua creates). + lua_host + .eval( + Some("@pmacs/builtin/runtime/lean.lua"), + include_str!("../builtin/runtime/lean.lua"), + ) + .expect("load lean builtin chunk"); // Arc 1a: the in-buffer completion popup driver. Loaded after // lsp.lua because it drives `pmacs.lsp.request_completion` / // `pmacs.lsp.attachment_for_request` and after the framework diff --git a/tests/lean4_server_acceptance.rs b/tests/lean4_server_acceptance.rs new file mode 100644 index 0000000..13cfcd0 --- /dev/null +++ b/tests/lean4_server_acceptance.rs @@ -0,0 +1,640 @@ +//! Arc 8 Stage 3b acceptance — the Lean 4 language server. +//! +//! `docs/lean4-mode-framing.md` Q#LN7, Q#LN8, Q#LN16; acceptance 22–28, +//! 24a/24b, 35, 36, 36a, 37. +//! +//! No live toolchain required. The server side is `pmacs_fake_lsp` +//! configured under the `lean4` language id; the probe and latch are +//! driven through shell stubs the fixture writes, so nothing here needs +//! `lake`, `lean`, or an elan toolchain on PATH (§2.9). +//! +//! Every fixture sets `pmacs.project.set_search_boundary` at its own +//! tempdir root. Without it the `lean-toolchain` walk climbs to the +//! filesystem root and acceptance 23's outermost assertion stops being +//! hermetic. + +#![cfg(unix)] + +use std::path::{Path, PathBuf}; +use std::time::Duration; + +use pmacs::editor::EditorState; + +fn exec(state: &EditorState, source: &str) { + state.lua_host.lua().load(source.to_owned()).exec().unwrap(); +} + +fn eval(state: &EditorState, source: &str) -> T { + state.lua_host.lua().load(source.to_owned()).eval().unwrap() +} + +fn fake_lsp_path() -> String { + env!("CARGO_BIN_EXE_pmacs_fake_lsp").to_owned() +} + +fn lua_str(path: &Path) -> String { + path.display() + .to_string() + .replace('\\', "\\\\") + .replace('"', "\\\"") +} + +struct Fixture { + _dir: tempfile::TempDir, + root: PathBuf, +} + +impl Fixture { + fn new() -> Self { + let dir = tempfile::tempdir().unwrap(); + let root = std::fs::canonicalize(dir.path()).unwrap(); + Self { _dir: dir, root } + } + + fn write(&self, rel: &str, contents: &str) -> PathBuf { + let path = self.root.join(rel); + std::fs::create_dir_all(path.parent().unwrap()).unwrap(); + std::fs::write(&path, contents).unwrap(); + path + } + + fn mkdir(&self, rel: &str) -> PathBuf { + let path = self.root.join(rel); + std::fs::create_dir_all(&path).unwrap(); + path + } + + fn dir(&self, rel: &str) -> PathBuf { + self.root.join(rel) + } + + /// A `lean-toolchain` marker file. Content is irrelevant to the + /// resolver by design (existence semantics), which 24b pins. + fn toolchain(&self, rel_dir: &str, body: &str) { + self.write(&format!("{rel_dir}/lean-toolchain"), body); + } + + fn bind(&self, state: &EditorState) { + exec( + state, + &format!( + "pmacs.project.set_search_boundary(\"{}\")", + lua_str(&self.root) + ), + ); + } +} + +/// A fresh editor with every shipped language config cleared, then the +/// `lean4` entry rebuilt against the fake server while KEEPING the real +/// resolver. That combination is the point: the root rule under test is +/// production code, only the command is a stand-in. +fn editor(fx: &Fixture) -> EditorState { + let state = EditorState::new(); + exec(&state, "pmacs.lsp.config = {}"); + exec( + &state, + &format!( + r#" + pmacs.lsp.config.lean4 = {{ + command = "{}", + args = {{}}, + root = pmacs.lean.root_for, + }} + "#, + fake_lsp_path() + ), + ); + fx.bind(&state); + state +} + +fn settle(state: &mut EditorState) { + for _ in 0..10 { + state.tick_processes(); + state.tick_lsp(); + state.tick_async(); + std::thread::sleep(Duration::from_millis(2)); + } +} + +fn open(state: &EditorState, path: &Path) { + exec( + state, + &format!("pmacs.buffer.find_or_open(\"{}\")", lua_str(path)), + ); +} + +/// `language_id|root_uri|cwd` for every live server, sorted. +fn rows(state: &EditorState) -> Vec { + let joined: String = eval( + state, + r#" + local out = {} + for _, s in ipairs(pmacs.lsp.list()) do + out[#out + 1] = table.concat({ + s.language_id or "", s.root_uri or "", s.cwd or "", + }, "|") + end + table.sort(out) + return table.concat(out, "\n") + "#, + ); + if joined.is_empty() { + Vec::new() + } else { + joined.lines().map(str::to_owned).collect() + } +} + +fn resolved_root(state: &EditorState, file: &Path) -> String { + eval( + state, + &format!( + "return tostring(pmacs.lean.root_for(\"{}\"))", + lua_str(file) + ), + ) +} + +// --------------------------------------------------------------------------- +// Acceptance 22 — a Lean file in a Lake package spawns one server rooted +// at the package. +// --------------------------------------------------------------------------- + +#[test] +fn acc22_lean_file_in_a_lake_package_spawns_one_server_at_the_package_root() { + let fx = Fixture::new(); + fx.toolchain("pkg", "leanprover/lean4:v4.9.0\n"); + let file = fx.write("pkg/Pkg/Basic.lean", "def x : Nat := 1\n"); + let mut state = editor(&fx); + open(&state, &file); + settle(&mut state); + + let pkg = fx.dir("pkg").display().to_string(); + assert_eq!( + rows(&state), + vec![format!("lean4|file://{pkg}|{pkg}")], + "one server, rooted and cwd'd at the Lake package" + ); +} + +// --------------------------------------------------------------------------- +// Acceptance 23 — outermost wins. +// +// The case `pmacs.project.detect` cannot express: it is innermost-wins by +// construction, so a dependency vendored under `.lake/packages` would get +// its own server and its own (wrong) view of the world. +// --------------------------------------------------------------------------- + +#[test] +fn acc23_nested_toolchains_resolve_to_the_outermost_package() { + let fx = Fixture::new(); + fx.toolchain("pkg", "leanprover/lean4:v4.9.0\n"); + fx.toolchain("pkg/.lake/packages/dep", "leanprover/lean4:v4.8.0\n"); + let inner = fx.write( + "pkg/.lake/packages/dep/Dep/Core.lean", + "def dep : Nat := 2\n", + ); + let state = editor(&fx); + + assert_eq!( + resolved_root(&state, &inner), + fx.dir("pkg").display().to_string(), + "a file under .lake/packages/dep belongs to the outer package" + ); + // Non-vacuity: the inner marker really exists, so "outermost" is a + // choice between two candidates rather than the only one found. + assert!(fx.dir("pkg/.lake/packages/dep/lean-toolchain").exists()); +} + +// --------------------------------------------------------------------------- +// Acceptance 24 — the walk stops at the search boundary. +// --------------------------------------------------------------------------- + +#[test] +fn acc24_walk_stops_at_the_search_boundary() { + let fx = Fixture::new(); + // Boundary is the fixture root; this marker sits INSIDE it. + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + // And this one sits AT the fixture root, i.e. above `pkg` but still + // within the boundary — it must win, being outermost. + fx.toolchain(".", "v4.7.0\n"); + let state = editor(&fx); + assert_eq!( + resolved_root(&state, &file), + fx.root.display().to_string(), + "within the boundary, the outermost marker wins" + ); + + // Now move the boundary IN to `pkg`. The root-level marker is above + // it and must not be reached. + exec( + &state, + &format!( + "pmacs.project.set_search_boundary(\"{}\")", + lua_str(&fx.dir("pkg")) + ), + ); + assert_eq!( + resolved_root(&state, &file), + fx.dir("pkg").display().to_string(), + "a marker above the boundary is not consulted" + ); +} + +// --------------------------------------------------------------------------- +// Acceptance 24a / 24b — the marker test, both directions. +// +// These two must each fail against the implementation that satisfies only +// the other. 24a bites the bare `io.open` truth test (which succeeds on a +// directory); 24b bites the read-a-byte-and-require-non-nil rule (which +// rejects an empty file at EOF). +// --------------------------------------------------------------------------- + +#[test] +fn acc24a_a_lean_toolchain_directory_is_not_a_marker() { + let fx = Fixture::new(); + fx.mkdir("pkg/lean-toolchain"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let state = editor(&fx); + assert_eq!( + resolved_root(&state, &file), + "nil", + "a `lean-toolchain` DIRECTORY must not mark a root" + ); +} + +#[test] +fn acc24b_an_empty_lean_toolchain_file_is_a_marker() { + let fx = Fixture::new(); + fx.toolchain("pkg", ""); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let state = editor(&fx); + assert_eq!( + resolved_root(&state, &file), + fx.dir("pkg").display().to_string(), + "marker semantics are existence, not content — an empty \ + `lean-toolchain` still marks the package" + ); + // Non-vacuity: the file really is empty. + assert_eq!( + std::fs::read(fx.dir("pkg/lean-toolchain")).unwrap().len(), + 0 + ); +} + +#[test] +fn acc24_resolver_declines_when_no_marker_exists() { + let fx = Fixture::new(); + let file = fx.write("loose/A.lean", "def a := 1\n"); + let state = editor(&fx); + assert_eq!( + resolved_root(&state, &file), + "nil", + "no marker anywhere is a decline, which falls through to \ + `pmacs.project.detect`" + ); +} + +// --------------------------------------------------------------------------- +// Acceptance 25 — a string-valued root still works. +// --------------------------------------------------------------------------- + +#[test] +fn acc25_string_valued_root_still_works() { + let fx = Fixture::new(); + let pkg = fx.mkdir("elsewhere"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let mut state = editor(&fx); + exec( + &state, + &format!("pmacs.lsp.config.lean4.root = \"{}\"", lua_str(&pkg)), + ); + open(&state, &file); + settle(&mut state); + + let want = pkg.display().to_string(); + assert_eq!( + rows(&state), + vec![format!("lean4|file://{want}|{want}")], + "the Q#LN8 generalization is additive; a plain string still wins" + ); +} + +// --------------------------------------------------------------------------- +// Acceptance 26 — didOpen carries languageId = "lean4". +// --------------------------------------------------------------------------- + +#[test] +fn acc26_did_open_carries_the_lean4_language_id() { + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let mut state = editor(&fx); + open(&state, &file); + settle(&mut state); + + let lang: String = eval( + &state, + "return tostring(pmacs.lsp.list()[1] and pmacs.lsp.list()[1].language_id)", + ); + assert_eq!( + lang, "lean4", + "the grammar entry name is the didOpen language id (Q#LN2)" + ); +} + +// --------------------------------------------------------------------------- +// Acceptance 28 (probe) — the version predicate. +// +// The parse is unit-tested directly because the spawn path is timing- +// bound; the latch's *effect* is pinned separately below. +// --------------------------------------------------------------------------- + +#[test] +fn acc28_version_predicate_triggers_only_below_3_1() { + let fx = Fixture::new(); + let state = editor(&fx); + let check = |v: &str| -> bool { + eval( + &state, + &format!("return pmacs.lean._version_below_3_1(\"{v}\")"), + ) + }; + assert!(check("Lake version 3.0.0"), "3.0.0 is below 3.1"); + assert!(!check("Lake version 3.1.0"), "3.1.0 is not below 3.1"); + assert!(!check("Lake version 5.0.0-abc"), "5.0.0 is not below 3.1"); + assert!(check("Lake version 2.9.9"), "2.9.9 is below 3.1"); + assert!( + !check("no default toolchain configured"), + "an unparseable line must NOT trigger the fallback — that is the \ + elan-shim case, which the failure latch handles better" + ); +} + +// --------------------------------------------------------------------------- +// Acceptance 27 / 35 / 36 — the fallback latch. +// --------------------------------------------------------------------------- + +#[test] +fn acc35_latch_preserves_user_config_and_swaps_only_command_and_args() { + let fx = Fixture::new(); + let state = editor(&fx); + // A user's init.lua settings, on the shipped shape. + exec( + &state, + r#" + pmacs.lsp.config.lean4.command = "lake" + pmacs.lsp.config.lean4.args = { "serve" } + pmacs.lsp.config.lean4.env = { MYVAR = "1" } + pmacs.lsp.config.lean4.settings = { lean = { verbose = true } } + pmacs.lsp.config.lean4.init_options = { hasWidgets = false } + _G.root_before = pmacs.lsp.config.lean4.root + pmacs.lean._fire_latch(nil, "test") + "#, + ); + + let after: String = eval( + &state, + r#" + local c = pmacs.lsp.config.lean4 + return table.concat({ + tostring(c.command), + tostring(c.args and c.args[1]), + tostring(c.env and c.env.MYVAR), + tostring(c.settings and c.settings.lean and c.settings.lean.verbose), + tostring(c.init_options and c.init_options.hasWidgets), + tostring(c.root == _G.root_before), + }, "|") + "#, + ); + assert_eq!( + after, "lean|--server|1|true|false|true", + "only command/args change; env, settings, init_options and root \ + survive the swap" + ); +} + +#[test] +fn acc27_the_latch_is_one_shot_and_does_not_re_arm() { + let fx = Fixture::new(); + let state = editor(&fx); + exec( + &state, + r#" + pmacs.lsp.config.lean4.command = "lake" + pmacs.lsp.config.lean4.args = { "serve" } + pmacs.lean._fire_latch(nil, "first failure") + _G.after_first = pmacs.lsp.config.lean4.command + -- A second failure must not rewrite the command again; if it did, + -- a user who deliberately set something else after the fallback + -- would have it silently replaced. + pmacs.lsp.config.lean4.command = "user-choice" + pmacs.lean._fire_latch(nil, "second failure") + _G.after_second = pmacs.lsp.config.lean4.command + "#, + ); + assert_eq!(eval::(&state, "return _G.after_first"), "lean"); + assert_eq!( + eval::(&state, "return _G.after_second"), + "user-choice", + "the latch never re-arms within a session" + ); +} + +#[test] +fn acc36_latch_stops_the_failing_server_before_spawning_the_fallback() { + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let mut state = editor(&fx); + open(&state, &file); + settle(&mut state); + assert_eq!(rows(&state).len(), 1, "precondition: one server is up"); + + // Fire the latch against the live server, exactly as `poll_latch` + // would. `pmacs.lsp.stop` sets `restart = Never` on the way out — + // which is what prevents `RestartPolicy::OnCrash` from respawning the + // broken command underneath the latch, forever, with no attempt cap. + exec( + &state, + r#" + pmacs.lsp.config.lean4.command = "lake" + pmacs.lsp.config.lean4.args = { "serve" } + pmacs.lean._fire_latch(pmacs.lsp.list()[1].id, "failed to start") + "#, + ); + settle(&mut state); + + let terminal: bool = eval( + &state, + r#" + for _, s in ipairs(pmacs.lsp.list()) do + local k = s.state and s.state.kind + if k ~= "stopped" and k ~= "crashed" then return false end + end + return true + "#, + ); + assert!( + terminal, + "the failing server is stopped, not left to be respawned under \ + the latch" + ); +} + +// --------------------------------------------------------------------------- +// Acceptance 36a — attribution (COHERENCE §9 / §1.2). +// --------------------------------------------------------------------------- + +#[test] +fn acc36a_latch_leaves_a_status_line_trace() { + let fx = Fixture::new(); + let state = editor(&fx); + exec( + &state, + r#" + pmacs.lsp.config.lean4.command = "lake" + pmacs.lsp.config.lean4.args = { "serve" } + pmacs.lean._fire_latch(nil, "`lake serve` failed to start") + "#, + ); + let status = state.core.borrow().status.clone(); + assert!( + status.contains("lean4") && status.contains("lean --server"), + "the fallback names itself and what it fell back to; saw {status:?}" + ); + // The channel assertion is the point (COHERENCE §1.2): a report made + // only through `pmacs.error` — undefined in production — would leave + // this empty while every other assertion here still passed. + assert!(!status.is_empty()); +} + +#[test] +fn acc36a_probe_carries_a_lean_owned_process_label() { + // `ProcessSpec.label` is the only identity a process has, and it is + // what `pmacs.process.list` renders. Asserted on the spec the module + // builds rather than on a live `lake`, which CI does not have. + let fx = Fixture::new(); + let state = editor(&fx); + let src = std::fs::read_to_string( + Path::new(env!("CARGO_MANIFEST_DIR")).join("builtin/runtime/lean.lua"), + ) + .unwrap(); + assert!( + src.contains("label = \"lean:lake-version-probe\""), + "the probe process is attributed to Lean by label" + ); + // And it is genuinely lazy: no probe without an attachment. + let procs: i64 = eval(&state, "return #pmacs.process.list()"); + assert_eq!(procs, 0, "configuring Lean does not start the probe"); +} + +// --------------------------------------------------------------------------- +// Acceptance 37 — waitForDiagnostics resolves through the response seam. +// --------------------------------------------------------------------------- + +#[test] +fn acc37_wait_for_diagnostics_resolves_through_the_response_seam() { + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a : Nat := 1\n"); + let mut state = editor(&fx); + open(&state, &file); + settle(&mut state); + + exec( + &state, + r#" + _G.settled = "never" + local rec = pmacs.lsp.active_attachment() + pmacs.lean.wait_for_diagnostics(rec.server, rec.uri, function(err) + _G.settled = tostring(err) + end) + "#, + ); + settle(&mut state); + + assert_eq!( + eval::(&state, "return _G.settled"), + "nil", + "the reply reaches the callback with no error — this is the \ + Stage 3a response seam carrying its first production caller" + ); +} + +// --------------------------------------------------------------------------- +// Acceptance 29, Lean's side — `$/lean/fileProgress` reaches the module. +// +// Driven end-to-end through the real drain: the fake server's +// `leanprogress` mode emits the notification on didOpen. Calling the +// handler directly would pin nothing about the wiring, which is the only +// part that can break. +// --------------------------------------------------------------------------- + +#[test] +fn file_progress_notification_is_recorded_for_its_document() { + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let mut state = editor(&fx); + exec( + &state, + "pmacs.lsp.config.lean4.env = { PMACS_FAKE_LSP_MODE = \"leanprogress\" }", + ); + + // Nothing recorded before the server speaks — so the assertion below + // cannot pass on a pre-populated table. + let before: i64 = eval( + &state, + "local n = 0 for _ in pairs(pmacs.lean.file_progress) do n = n + 1 end return n", + ); + assert_eq!(before, 0); + + open(&state, &file); + settle(&mut state); + + let uri: String = eval( + &state, + r#" + for k, v in pairs(pmacs.lean.file_progress) do + if type(v) == "table" and v[1] and v[1].range then return k end + end + return "none" + "#, + ); + assert!( + uri.starts_with("file://") && uri.ends_with("A.lean"), + "the subscriber recorded the processing ranges under the \ + document uri; saw {uri:?}" + ); +} + +// --------------------------------------------------------------------------- +// Q#LN20 in the Lean resolver — a symlinked open reuses one server. +// --------------------------------------------------------------------------- + +#[test] +fn lean_root_is_canonical_so_a_symlinked_open_reuses_one_server() { + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let real = fx.write("pkg/A.lean", "def a := 1\n"); + std::os::unix::fs::symlink(fx.dir("pkg"), fx.dir("linkpkg")).unwrap(); + let linked = fx.dir("linkpkg").join("A.lean"); + + let mut state = editor(&fx); + open(&state, &real); + settle(&mut state); + assert_eq!(rows(&state).len(), 1, "the real path spawns one server"); + + open(&state, &linked); + settle(&mut state); + assert_eq!( + rows(&state).len(), + 1, + "the symlinked path reuses it — the resolver canonicalizes, so \ + both spellings produce the same affinity key" + ); +} diff --git a/tests/lean4_stage1_acceptance.rs b/tests/lean4_stage1_acceptance.rs index d48a86c..9aafcab 100644 --- a/tests/lean4_stage1_acceptance.rs +++ b/tests/lean4_stage1_acceptance.rs @@ -305,36 +305,50 @@ fn acc11b_an_unknown_fence_name_still_injects_nothing() { // --------------------------------------------------------------------------- #[test] -fn acc12_stage1_ships_no_lsp_config_and_spawns_no_process() { - // Stage 1 is grammar + Lua tables only. Opening a Lean file must not - // reach for `lake`, `lean`, or `elan` — the LSP arrives in Stage 3, and - // even then it is fallible by design (Q#LN7). +fn acc12_opening_lean_spawns_no_process_without_a_server_config() { + // **Superseded in half by Stage 3b.** This criterion originally also + // asserted `pmacs.lsp.config.lean4 == nil`, guarding against a Stage-3 + // front-run. Stage 3b *is* Stage 3: `builtin/runtime/lean.lua` now ships + // that config deliberately, and its shape is pinned by + // `tests/lean4_server_acceptance.rs`. Asserting the absence here would + // now pin the opposite of the intended behavior, so it is gone rather + // than weakened. + // + // What survives is the half that was always about *restraint*, and it + // matters more now than it did in Stage 1 — it is what holds Q#LN7's + // "not at init" promise. `pmacs.lsp.config` is a declarative table, and + // spawning a process at startup for every user, Lean-using or not, is + // the cost rev 1 refused. Both the `lake serve` spawn and the + // `lake --version` probe are gated on a real Lean attachment. - // The load-bearing assertion, and it must run against a PRISTINE editor. - // The shared `editor()` helper wipes `pmacs.lsp.config` before any - // buffer opens, so an assertion about the server list under that harness - // holds for every language regardless of what Stage 1 ships — it could - // not fail for the regression it names. This checks the real claim - // directly: no builtin runtime file defines a Lean server config. A - // Stage-3 front-run adding `pmacs.lsp.config.lean4` fails here. + // Constructing an editor touches no process, even though the Lean + // config now exists and names `lake`. let pristine = EditorState::new(); - let no_lean_config: bool = eval(&pristine, "return pmacs.lsp.config.lean4 == nil"); - assert!( - no_lean_config, - "Stage 1 defines no `pmacs.lsp.config.lean4`; the LSP is Stage 3" + let at_init: i64 = eval(&pristine, "return #pmacs.process.list()"); + assert_eq!( + at_init, 0, + "constructing an editor must not probe or spawn for Lean" + ); + // Non-vacuity for the assertion above: the config really is present and + // really does name a command, so "nothing spawned" is restraint rather + // than an empty table having nothing to act on. + let names_lake: bool = eval( + &pristine, + "return pmacs.lsp.config.lean4 ~= nil and pmacs.lsp.config.lean4.command == \"lake\"", ); - // Non-vacuity: the same lookup finds the configs that DO ship, so this - // is not passing because `pmacs.lsp.config` is empty or absent. - let rust_config_exists: bool = eval(&pristine, "return pmacs.lsp.config.rust ~= nil"); assert!( - rust_config_exists, - "the config table is populated, so the lean4 absence above is meaningful" + names_lake, + "Stage 3b ships a lean4 config naming `lake`, so the no-spawn \ + assertion above is meaningful" ); - // And nothing is spawned by opening the file. This half retains its - // value under the wiped config: a direct probe spawn from `lean.lua` - // would show up here whatever `pmacs.lsp.config` contains. + // And opening a Lean buffer with no server configured spawns nothing — + // the `editor()` helper wipes `pmacs.lsp.config`, so this catches a + // probe that fires off the mode rather than off an attachment. let s = editor_visiting("Basic.lean", "def x : Nat := 1\n"); let procs: i64 = eval(&s, "return #pmacs.process.list()"); - assert_eq!(procs, 0, "opening a Lean buffer spawns no child process"); + assert_eq!( + procs, 0, + "with no server configured, opening a Lean buffer spawns nothing" + ); } From 914bf3f02f472774e79792347a23ff190d631e34 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 17:40:31 -0400 Subject: [PATCH 09/32] docs: record the Stage 3b lane and its stacking constraint --- docs/active-work.md | 39 ++++++++++++++++++++++++++++++++++++++- 1 file changed, 38 insertions(+), 1 deletion(-) diff --git a/docs/active-work.md b/docs/active-work.md index aef6edc..99b54ea 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -54,7 +54,7 @@ git status --short --branch The `git log` command must expose `0dd16a5` or a newer intentional main. If it does not, stop and repair the remote/fetch configuration. -## Lean 4 lane (Arc 8) — Stages 1+2 MERGED; Stage 3a IN REVIEW +## Lean 4 lane (Arc 8) — Stages 1+2 MERGED; 3a IN REVIEW (#167); 3b STACKED - Stage 1 **merged as #160** (`main` @ `0827dd1`, 2026-07-25, one review round, all twelve checks green). Branch `githubsucks/lean4-stage1` @@ -252,6 +252,43 @@ If it does not, stop and repair the remote/fetch configuration. `#[cfg(unix)]` is NOT sufficient for such a fixture — `#[cfg(target_os = "linux")]` is. Cost one red CI round to learn. +### Stage 3b — the Lean language server (branch `lean4-stage3b-server`) + +- Same worktree `../pmacs-lean-stage3`, **branched off + `lean4-stage3a-seams`, not off `main`** — 3b consumes 3a's response + seam and `pmacs.fs.canonicalize`, so it is strictly sequential and its + PR must be retargeted to `main` only after #167 merges. (Kill-ring + lesson: retarget stacked child PRs BEFORE merging the parent.) +- Ships `builtin/runtime/lean.lua` (new), one `include_str!` line in + `src/editor.rs`, a `leanprogress` mode on `pmacs_fake_lsp`, and + `tests/lean4_server_acceptance.rs` (17 tests). No protocol change. +- **Stage 1's acceptance 12 is half superseded and was rewritten, not + deleted.** It asserted `pmacs.lsp.config.lean4 == nil` to catch a + Stage-3 front-run; 3b is that stage. What survives is the restraint + half — constructing an editor spawns nothing though the config now + names `lake`, and opening a Lean buffer with no server configured + spawns nothing — which is what holds Q#LN7's "not at init" promise. +- **The marker test is wrong in two opposite directions if done naively** + and both are pinned: `io.open` SUCCEEDS on a directory (so truthiness + accepts a `lean-toolchain` dir), but requiring a non-nil read rejects + an EMPTY `lean-toolchain` (a legitimate marker — existence semantics, + not content). Discriminator is `read`'s SECOND return; decline only on + a non-nil err. Probed on LuaJIT 2.1. +- Four bites recorded, each against the committed tree: bare `io.open` + → 24a fails / 24b passes; require-non-nil → 24b fails / 24a passes; + no canonicalization → symlinked open spawns two servers; no stop + before fallback → acc36 fails. +- The probe's non-zero exit is deliberately NOT a fallback trigger — + §2.9's elan shim makes `lake --version` fail where `lake serve` still + works. Only a parseable version below 3.1.0 triggers it; the + server-failure latch covers the rest. +- Verification on this branch: `cargo fmt --check` clean; strict + workspace Clippy clean; 1,826 default + 2,003 CRDT library tests; + lean4 server 17/17; lean4 stage 1 9/9; dispatch seams 15/15; + multi-root 13/13; M4 121; required GPU 155; **isolated-config + workspace sweep 3,206 across 94 suites, zero failures**; + `git diff --check` clean. + ## Dired lane — framing APPROVED; Stage 0 MERGED, Stage 1 next - Approved framing: `docs/dired-framing.md` (revision 5), landing as its From 05984f1b1be4ae818b55b9a4014b9b91efafa143 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 18:02:35 -0400 Subject: [PATCH 10/32] docs: frame terminal configuration and copy mode Two stages, one arc, no protocol change. Stage 1 makes the terminal configurable (profiles, scrollback, escape key) and binds the opening command; Stage 2 adds copy mode and search over scrollback. They are independently releasable and get separate branches and PRs. Three scouted facts shaped the design, two of them ruling out the obvious plan. Profiles cannot be a config-registry setting: ConfigValue is four scalars and there is no table kind, so profiles join pmacs.lsp.config and pmacs.pair.sets as a raw Lua table while the registry holds only scalars. Search cannot reuse isearch in place: SearchStore addresses matches as byte ranges into a buffer's rope, and a terminal identity buffer is empty by construction. An in-place copy mode would be the seventh dispatch shadow, which COHERENCE section 6 grades weak and growing by one island per modal feature, with no transient-keymap mechanism to migrate to. Copy mode therefore materializes the retained rows into an ordinary read-only buffer. isearch, motion, selection and the kill ring work with no new substrate; the "keys must not reach the child" problem dissolves because the snapshot is not a terminal; and describe-key stays truthful because the bindings are buffer-local. The cost, stated in the doc, is that the snapshot is point-in-time rather than a live freeze. Four review rounds produced the load-bearing parts: the escape-key cache is owned by TerminalSession so its lifecycle is the terminal's, with three acceptance pins that each fail a different wrong cache; the snapshot needs set_round_trip_input because a Lua intercept does not set Buffer::read_only and an optimistic CrdtOp would mutate both the daemon buffer and the mirror; the double-escape must encode the configured chord rather than a hardcoded ETX; and the two open-time settings resolve through the global chain because they are read before the terminal buffer exists. No code changes in this commit. --- docs/terminal-config-and-copy-mode-framing.md | 634 ++++++++++++++++++ 1 file changed, 634 insertions(+) create mode 100644 docs/terminal-config-and-copy-mode-framing.md diff --git a/docs/terminal-config-and-copy-mode-framing.md b/docs/terminal-config-and-copy-mode-framing.md new file mode 100644 index 0000000..3f13987 --- /dev/null +++ b/docs/terminal-config-and-copy-mode-framing.md @@ -0,0 +1,634 @@ +# Terminal configuration and copy mode + +**Revision 4 — scouted against canonical `main` @ `b889873` (protocol v20), +2026-07-25. Not yet approved; no branch, no implementation.** + +Revision 4 gives the escape-key cache an owner and a lifecycle (Q#TC4c) — +revision 3 named the key but not the storage, and two implementations +satisfied its acceptance while behaving differently on A→B→A. It also corrects +the read-only deferral, which understated the substrate required: the bypass +path is `ensure_writable`-guarded too, so genuine immutability alone would +break every generated buffer that refreshes. + +Revision 3 corrects two design errors and decides the chords. The +round-trip failure shape in revision 2 was **wrong in the reporter's favour**: +a Lua intercept does not set `Buffer::read_only`, and there is no Lua binding +that does, so an optimistic `CrdtOp` bypasses the intercept *and* passes +`ensure_writable()` — the daemon buffer mutates too, rather than the mirror +diverging alone (Q#TC6a). Revision 2 also had all three settings resolving +against the terminal identity buffer, which is impossible for the two read +*before* that buffer exists (Q#TC2b). Chords are now decided and +collision-scouted rather than deferred to implementation (Q#TC10, Q#TC8a). + +Revision 2 answered seven review findings. Four were load-bearing: the settings +are `Live`, so the registry **accepts buffer-local overrides whether or not we +want them**, and `value_epoch()` does not move on a buffer switch — an +epoch-only cache can serve the wrong terminal's escape chord (Q#TC4); the +double-escape byte is a hardcoded `0x03`, so a configured escape would still +send Ctrl-C and make its own literal chord unreachable (Q#TC4b); the snapshot +buffer needs `set_round_trip_input`, not only a read-only intercept, or a +semantic frontend can optimistically edit it before daemon dispatch (Q#TC6); +and the two stages must be two branches and two PRs. Revision 1's +materialized-copy reframe is unchanged. + +Two stages, one arc, no protocol change: + +- **Stage 1 — configuration.** Terminal profiles, scrollback, and the escape + key become configurable. Today the terminal has **zero** configuration + surface: the `terminal` command hardcodes `os.getenv("SHELL") or "/bin/sh"`, + `scrollback_rows` is a per-open argument only, and the escape chord is a + literal in Rust. +- **Stage 2 — copy mode and search over scrollback.** A command that turns + the retained terminal screen and scrollback into an ordinary buffer, where + isearch, motion, selection, and the kill ring already work. + +Explicitly **not** in this arc: the panel terminal (blocked on bottom-panel +Stage 2), and shell integration (cwd tracking, prompt marks, command zones) — +the keystone that unlocks the VS Code-style cluster, which needs its own +security framing because it decides what a child process may make the editor +do. + +## Branch and PR plan + +**Two branches, two PRs.** Configuration and copy mode are independently +releasable and have no dependency on each other; one framing covers the arc, +but the one-feature/one-branch/one-PR rule governs the implementation. + +1. `terminal-config` — Stage 1. Also carries the **terminal opening + keybinding** (Q#TC10). +2. `terminal-copy-mode` — Stage 2, branched off `main` after Stage 1 merges. + +Sequencing is not a dependency but avoids a conflict: both stages edit +`builtin/runtime/terminal.lua`. + +## Ground truth (measured, not recalled) + +Three facts constrain the design, and two of them rule out the obvious plan. + +### 1. Terminal profiles cannot be a config-registry setting + +`ConfigValue` is **four scalars** — `Bool`, `Int`, `Num`, `Str` +(`src/config_registry.rs:312`) — and its own doc comment says they "are never +stored --- only these four scalars (Q#CR3)". `ConfigKind` adds `Enum`, which +is physically a string validated against choices fixed at `define` time +(`src/config_registry.rs:115-145`). There is no table, list, or map kind. + +A terminal profile is inherently a table: `{ command, args, cwd, env }` per +name. **Table-valued settings are an existing named deferral of the config +registry arc** — the same gap that keeps `pmacs.lsp.config`, +`pmacs.pair.sets`, `pmacs.comment.strings`, and the `pmacs.parse.*` proxies as +raw Lua. Profiles join that list rather than forcing that deferral open here. + +### 2. Search cannot reuse isearch in place over a terminal + +`SearchStore::set(buffer_id, query, matches: Vec)` +(`src/search.rs:99`) keys matches by buffer and addresses them as **byte +ranges into that buffer's rope**; the painting path materializes the source +with `buf.snapshot_rope().slice(0, buf.len(), ..)` (`src/search.rs:435`). + +A terminal identity buffer is **empty and read-only** by construction. Its +content lives in `TerminalScreen` as cells addressed by `(row, col)` across +history plus visible rows — there are no rope bytes to range over. Searching a +terminal in place therefore means a second, parallel search facility with its +own match store and its own highlight path, because terminal painting consumes +owned cells and not document style spans. + +### 3. An in-place copy mode would be the seventh dispatch shadow + +`dispatch_key`'s terminal-transport arm intercepts **every** key before +ordinary keymap dispatch whenever `active_terminal_key` is `Some`, which keys +purely on `is_terminal(window.buffer_id)` (`src/editor.rs:1098-1107`, +`973-1016`). A mode that keeps the terminal buffer focused while rebinding +keys to motion/selection must therefore add a new precedence rung. + +`COHERENCE.md` §6 grades that ladder **weak, "and growing by one island per +modal feature"**, records that **no transient-keymap mechanism exists to +migrate to** (`KeymapStack` has exactly three fixed scopes, no layer stack, no +push/pop, no lifetime), and notes that `describe-key` already lies while a +shadow is active. It also names the counter-example: the entire picker/panel +family uses ordinary **buffer-local keymaps** and is inspectable and +rebindable. + +### 4. What already exists and is reusable + +- `retained_rows(projection)` (`src/terminal/view.rs:539`) iterates history + plus visible rows; `copy_selection_bytes(rows, selection)` + (`src/terminal/view.rs:849`) serializes a range with the fidelity Stage 2 + criterion 21 already pins — soft wraps joined, hard rows separated, trailing + default blanks trimmed, wide glyphs and combining clusters copied once. +- `ConfigRegistry::value_epoch()` (`src/config_registry.rs:1127`) is public and + monotonic — cheap invalidation for a hot-path cache. +- The Lua surface is `define` / `get` / `set` / `set_local` / `on_change` with + a disposable handle (`src/lua_bindings/config.rs`). +- `pmacs.terminal.open` already accepts + `command, args, cwd, env, name, rows, cols, scrollback_rows, display, + window`. **`display = "panel"` already works** (bottom-panel Stage 1) — the + panel terminal is blocked on rendering, not on this surface. +- Terminal buffers already carry buffer-local bindings (`M-w`, `M-v`, `C-v`, + `M-<`, `M->`) installed by `terminal.open` in `builtin/runtime/terminal.lua`. + +## Stage 1 — configuration + +**Q#TC1 — Profiles are a raw Lua table, not a setting.** +`pmacs.terminal.profiles` maps a name to a spec table, exactly following the +`pmacs.lsp.config` precedent. The registry holds only scalars. Rejected +alternative: widening `ConfigValue` with a table kind — that is the config +arc's own named deferral, it is cross-cutting (persistence, `describe-setting` +rendering, the `custom-file` question all key on the scalar assumption), and +smuggling it into a terminal PR would be the wrong place to decide it. + +**Q#TC2 — `terminal.default-profile` is `String`, not `Enum`.** `Enum` +choices are frozen at `define` time; profiles are user-extensible from +`init.lua` and later. Validation happens at open time, and an unknown name +must produce a pointed error that **names the known profiles**, not a bare +"unknown profile". + +**Q#TC2a — the exact settings, defaults, and bounds.** All three are `Live` +(see Q#TC2b), and every default reproduces today's behavior exactly, so a tree +with no settings written behaves identically (acceptance 12). + +| name | kind | default | bounds | +|---|---|---|---| +| `terminal.default-profile` | `String { allow_empty: true }` | `""` | — | +| `terminal.scrollback-rows` | `Integer` | `10_000` (`DEFAULT_TERMINAL_SCROLLBACK_ROWS`) | `0 ..= 4_000_000` (`MAX_TERMINAL_HISTORY_CELLS`) | +| `terminal.escape-key` | `String { allow_empty: false }` | `"C-c"` | parsed as a chord | + +**Zero is a legal scrollback value meaning "retain no history".** The core's +own validation rejects only values *above* `MAX_TERMINAL_HISTORY_CELLS` +(`src/terminal/session.rs:114`), so `scrollback_rows = 0` is accepted through +`terminal.open` today. A `1` minimum here would invent an asymmetry between the +setting and the per-open field for no reason. + +`""` is the **"no default profile" sentinel**: an empty string means "fall +through to `$SHELL`", not "a profile named empty". `allow_empty: true` exists +precisely to express it, and the open path treats empty and unset identically. + +**Q#TC2b — the settings are `Live`, and the registry therefore accepts +buffer-local overrides. That is specified rather than accidental.** +`ConfigRegistry::set_local` refuses only `StartupOnly` definitions +(`src/config_registry.rs:949`); a `Live` setting can be pinned per buffer by +anyone. Declaring these global-only is **not currently expressible** — a +`scope = "global"` define flag is one of the config registry's own named +deferrals, and `autosave.interval-ms` already has the same latent problem. + +Making them `StartupOnly` instead would buy enforcement at the cost of the +feature: the escape key could never be changed mid-session, which kills Q#TC4's +whole point. So they stay `Live`, and resolution is defined **per setting, +because the three are not read at the same moment**: + +| setting | read when | resolution | +|---|---|---| +| `terminal.escape-key` | every keystroke in a terminal (cached) | `get(name, terminal_buffer)` — **buffer-local → global → default** | +| `terminal.default-profile` | once, **before** the terminal exists | `get(name)` — **global chain only** | +| `terminal.scrollback-rows` | once, **before** the terminal exists | `get(name)` — **global chain only** | + +The split is forced, not stylistic. The two open-time settings are consumed by +`_open` **before it creates the identity buffer**, so there is no terminal +buffer to resolve against — and no caller could have pinned a local override on +a buffer that does not yet exist. `pmacs.config.get(name)` with no buffer +argument already means exactly "the global chain, never an ambient buffer", so +this is the registry's existing semantic rather than a new rule. + +Consequences, stated so they are not discovered later: + +- a per-terminal escape key is a supported feature, not a bug; +- `set_local` on `terminal.default-profile` or `terminal.scrollback-rows` is + **always inert**, for any buffer, because the open path never consults a + buffer chain. This is deliberate; the alternative — resolving against + whichever buffer happened to be current at open time — would make a + terminal's scrollback depend on what the user was looking at when they + pressed the key. + +Rejected alternative: resolving the open-time settings against the *target +window's pre-open buffer*. It is expressible, but it makes an ambient buffer +load-bearing for a value the user set globally, which is the trap +`pmacs.config`'s two-argument/one-argument split exists to avoid. + +**Q#TC3 — `terminal.scrollback-rows` is `Integer` with bounds, and an explicit +per-open `scrollback_rows` still wins.** The precedence is +**explicit argument over global setting** — there is no ambient buffer in this +chain at all (Q#TC2b resolves it through `get(name)`), so the rule is simply +that what a caller passes to `terminal.open` beats what the user configured +globally. The bounds above come from the existing validation, so the setting +cannot express a value the core will reject. + +**Q#TC3a — profile resolution order, field by field.** `profile` is accepted +by **`pmacs.terminal.open` as well as the command**, so a Lua caller is not +forced through the command to use one. For each field, the first source that +supplies it wins: + +1. an explicit `pmacs.terminal.open` field; +2. the named profile's field — `profile` argument, else + `terminal.default-profile` when non-empty; +3. the scalar setting, where one exists (`scrollback_rows` only); +4. the built-in fallback (`command` = `$SHELL`, else `/bin/sh`). + +`env` is the one field where "first wins" is ambiguous, so it is stated: +profile `env` and explicit `env` are **merged**, with explicit entries +overriding profile entries of the same name. Any other reading silently drops +half a user's environment. + +An explicitly passed `profile` that does not exist is an error even when +`terminal.default-profile` is valid — a typo must not silently fall back to +the default. + +**Q#TC4 — `terminal.escape-key` is a `String` chord spelling, parsed once and +cached by `(buffer_id, value_epoch)`.** `is_terminal_escape_chord` +(`src/editor.rs:4413`) currently compares against a literal `C-c`. Reading and +parsing a setting on **every keystroke in a terminal** is not acceptable in +that path. + +**The cache key must include the buffer.** `value_epoch()` advances only on +`set` / `set_local` / removal (`src/config_registry.rs:918`, `970`, `1011`, +`1029`) — **it does not move when the focused terminal changes**. An +epoch-only cache therefore serves terminal A's escape chord to terminal B for +as long as no setting is written, which is exactly the case where nothing looks +wrong. Keying on `(buffer_id, value_epoch)` is the minimum correct identity. + +**Q#TC4c — the cache lives on `TerminalSession`, so its lifecycle is the +terminal's.** Revision 3 named the key `(buffer_id, value_epoch)` but not the +storage, and the two obvious storages behave differently on A→B→A: + +- a **single last-entry cache** reparses on every switch between two + terminals, and re-reports an invalid value each time — a status line that + scolds you for a setting you already know about, forever; +- an **editor-side map** preserves "parsed and reported once" but **leaks an + entry per terminal** unless something purges it, and that purge is a second + thing to get wrong. + +`TerminalSession` (`src/terminal/session.rs:215`) is created in +`TerminalManager::open` and dropped on kill/prune, so putting the cache there +gets the lifecycle for free with no purge hook to forget. It carries the parsed +chord, the `value_epoch` it was parsed at, and whether the current invalid +value has already been reported. + +**"Reports once" means once per terminal, per effective invalid value.** +A→B→A must not re-report. Changing the setting from one invalid value to a +*different* invalid value **does** re-report, because that is new information +about a new mistake. + +**The reporting channel is `EditorCore::status`** — the same channel +`send_terminal_bytes` already uses for terminal failures +(`src/editor.rs:1122`). Explicitly **not** `pmacs.error`: it is not installed +as a module anywhere in `src/lua_bindings`, so its call sites across the +runtime are dead, and a report sent there would be a report nobody sees. + +**Q#TC4a — an unparseable escape key must not brick terminal input.** A bad +value falls back to `C-c` and reports once. The failure mode this avoids is +severe: with no escape chord, every key goes to the child and the user cannot +reach any editor binding to fix the setting that broke it. + +**Q#TC4b — repeating the configured escape sends THAT chord to the child, not +Ctrl-C.** The double-escape arm currently writes a hardcoded +`&[0x03]` (`src/editor.rs:988`). With `terminal.escape-key = "C-x"`, `C-x C-x` +would send Ctrl-C — and literal Ctrl-X would become unreachable, since the +first `C-x` is always consumed as the escape. The repeat arm must encode the +**configured** chord through the existing `crate::terminal::input::encode_key` +path, which is also how it inherits application-cursor and modifier handling +rather than growing a second encoder. + +Corollary worth pinning: after changing the escape away from `C-c`, an ordinary +`C-c` must reach the child as `0x03` like any other unescaped key. + +**Q#TC5 — the `terminal` command gains an optional profile argument** and +otherwise keeps its current behavior; `$SHELL` remains the fallback when no +profile is configured. No existing invocation changes meaning. + +**Q#TC10 — the terminal opening keybinding is pulled forward into Stage 1.** +`COHERENCE.md` Priority 1 names "a terminal keybinding" as part of protecting +the golden journey, §2 step 8 grades the terminal "works but undiscoverable", +and this stage already edits `terminal.lua`. Panel rendering imposes no +dependency on binding a command that already exists. Close/kill semantics stay +with the panel work, where the entry and exit points get designed together. + +The chord is **decided and scouted, not deferred**: `C-c t`, global. See +Q#TC8a for the collision evidence and for why binding under the existing `C-c` +prefix is a new leaf rather than a shadow. + +## Stage 2 — copy mode and search + +**Q#TC6 — copy mode MATERIALIZES into an ordinary buffer. It does not add a +dispatch shadow.** + +`M-x terminal.copy-mode` snapshots the retained rows into a read-only, +path-less buffer (`*terminal-copy: NAME*`) and displays it. That buffer is an +ordinary document buffer, so: + +- **isearch works, with no new search substrate** — it is a rope, so + `SearchStore` and the existing match-painting path apply unchanged. Ground + truth 2 is answered by not fighting it. +- **motion, selection, `M-w`, the kill ring, even `M-x occur`-style consumers + work** — everything that operates on a buffer. +- **The "keys must not reach the child" problem dissolves structurally.** + `active_terminal_key` keys on `is_terminal(window.buffer_id)`; the snapshot + buffer is not a terminal, so the transport arm never fires. No new guard, no + new precedence rung, and ground truth 3's coherence cost is avoided rather + than paid. +- **`describe-key` stays truthful**, because the bindings are buffer-local and + inspectable — the idiom `COHERENCE.md` §6 identifies as the right side of + the line. + +**Q#TC6a — the snapshot is BOTH intercept-read-only AND round-trip-marked, +and `set_round_trip_input` is the ONLY thing standing between a replica +frontend and unauthorized mutation.** + +The established idiom is two calls: `listview.lua:106` and `compile.lua:272` +each pair `pmacs.buffer.add_intercept` with +`pmacs.buffer.set_round_trip_input(buf, true)`. Revision 2 described the +intercept as the guard and round-trip as defence in depth. **That was wrong, +and the correction matters:** + +- A Lua intercept guards the **dispatch/edit** path only. It does **not** set + `Buffer::read_only`, which is "deliberately independent of edit intercepts" + (`src/buffer.rs:493-500`) — that flag is what makes terminal identity buffers + reject rope, undo/redo, and remote-CRDT mutation alike. +- **No Lua binding sets `read_only` at all.** The whole `src/lua_bindings` + tree only ever *reads* it (`fold.rs:313`). A Lua-created "read-only" buffer + is therefore read-only against dispatch and nothing else. +- So an optimistic `CrdtOp` from a semantic frontend bypasses the intercept + **and passes `ensure_writable()`**. It is applied. The daemon buffer mutates + in lockstep with the mirror — the user silently edits a buffer the editor + told them is read-only. There is no divergence to notice, which is worse + than divergence. + +`set_round_trip_input` prevents this at the only point it can be prevented: it +makes `dispatch_idle_for` report false while the buffer is focused, so the +frontend never applies optimistically and never emits the op. It is not +hardening — it is the guard. + +Two things follow, and both are recorded rather than fixed here: + +- **The same exposure exists today** for every Lua-created read-only buffer — + listview panels and `*compilation*` included. They are correct only because + they call `set_round_trip_input`. This arc must not be the place that + unilaterally changes that substrate. +- **Exposing `Buffer::set_read_only` to Lua** would make these buffers + genuinely immutable at the rope/CRDT boundary the way terminal identity + buffers are, turning round-trip back into real defence in depth. That is a + substrate change affecting listview and compile as much as this snapshot, so + it is named in Deferred with its own lane. + +**Q#TC7 — the materializer reuses the existing serializer.** A whole-range +variant of `copy_selection_bytes` over `retained_rows` inherits the criterion +21 fidelity rather than re-deriving soft-wrap, wide-glyph, and trailing-blank +behavior. Writing a second serializer would guarantee the two drift. + +**Q#TC8 — one snapshot buffer per terminal, reused on re-invoke.** Re-running +the command against the same terminal replaces the contents in place rather +than accumulating buffers. It is killed with its terminal; killing the +snapshot alone leaves the terminal untouched. + +**Q#TC8a — the chords, decided and collision-scouted.** + +Worth stating first because it is easy to get backwards: in a terminal window +every **unescaped** key goes to the child, so terminal-local bindings are +reached as ` `. The existing `M-w` copy is physically `C-c M-w`. +The escape consumes itself and the next key starts a fresh ordinary sequence, +which is also why `C-c`-leading bindings are structurally unreachable *inside* +a terminal. + +| action | scope | binding | physically typed | +|---|---|---|---| +| open a terminal (Q#TC10) | global | `C-c t` | `C-c t` | +| enter copy mode | terminal buffer | `C-t` | `C-c C-t` | +| refresh snapshot | snapshot buffer | `g` | `g` | +| return to terminal | snapshot buffer | `q` | `q` | + +Scouted against the real keymaps: + +- **`C-c t` is free.** No bare global `C-c` binding exists; `C-c` is already a + live global prefix from `fold.lua:48-52` (`C-c @ …`), and `C-c C-k` is + buffer-scoped in compile/async. `C-c t` is a new leaf under an existing + prefix, not a shadow. +- **`C-t` is globally `edit.transpose-chars`** (`editops.lua:909`), and binding + it **buffer-locally is legitimate**: `keymap.bind`'s strictness rejects + binding a *prefix* of an existing sequence within a scope + (`keymap_bind_conflict_surfaces_at_bind_time` — "would shadow"), not + cross-scope shadowing, which is what scopes are for. Listview already binds + `n`/`p`/`g`/`q`/`RET`/`SPC` buffer-locally. Transpose-chars is meaningless in + a read-only terminal buffer. +- `C-c C-t` matches emacs-libvterm's own `vterm-copy-mode` chord, so the muscle + memory transfers. +- `g` / `q` in the snapshot follow listview's precedent exactly. + +**Named limitation:** `C-c t` cannot open a terminal *from inside* a terminal, +because `C-c` is consumed as the escape there. `M-x terminal` still works. This +is the documented consequence of Stage 2 criterion 19, not a new defect. + +These are what make acceptance 21's `describe-key` claim testable: named +bindings, in named buffers, that introspection must report truthfully. + +**Q#TC9 — the live-terminal keys stay.** `M-w`, `M-v`, `C-v`, `M-<`, `M->` on +the terminal buffer are the live affordances and do not change. Copy mode is +additive, on its own binding, and does not replace scroll-and-select. + +## Bets + +- **B1.** Materializing gives search for free: no second match store, no + second highlight path, no terminal-specific search UI. *Scored by Stage 2 + landing with zero changes under `src/search.rs`.* +- **B2.** Point-in-time is sufficient for read-back/search/copy. *Scored by + use; if false, the live frozen mode in Deferred becomes the real feature and + this becomes its snapshot fallback.* +- **B3.** No protocol change. The snapshot is an ordinary buffer, so both + frontends render it with existing machinery. *Scored by the diff.* +- **B4.** The escape-key cache keyed by `(buffer_id, value_epoch)` never + becomes stale in a way a user can observe. *Scored by two acceptances, not + one: changing the setting mid-session (8) and two terminals with different + buffer-local values and no write between them (7). Revision 1's epoch-only + cache would pass the first and fail the second, which is why the bet now + names both.* +- **B5.** Buffer-local escape keys are a feature rather than a hazard. + *Unscored and honestly so: the registry cannot express global-only, so this + is what we get either way. If per-terminal escapes turn out to confuse more + than they help, the fix is the config registry's `scope = "global"` deferral, + not a terminal change.* + +## Deferred (named) + +- **Live frozen copy mode** (true `vterm-copy-mode` semantics: freeze the + terminal in place, navigate it, resume). Strictly larger; needs either the + transient-keymap primitive `COHERENCE.md` §6 specifies or a deliberate + seventh shadow. +- **Shell integration** — cwd tracking, prompt marks, command zones, and the + VS Code cluster downstream of it (command decorations, exit-code markers, + rerun, sticky scroll, terminal IntelliSense). Its own arc, with a security + framing. +- **Table-valued settings** — the config registry's own deferral. This arc + adds a **second** blocked adopter (after `pmacs.lsp.config` / + `pmacs.pair.sets`); worth recording as evidence when that deferral is + ranked. +- **A `scope = "global"` define flag** — also the config registry's own + deferral, and this arc is its second live case after `autosave.interval-ms`. + Until it exists, `set_local` on any `Live` setting is accepted whether or not + the owner wants it, so Q#TC2b specifies the behavior instead of pretending + it is prevented. +- **Panel terminal** — blocked on bottom-panel Stage 2 (semantic frontends are + not `panel_capable`). `display = "panel"` already exists and works on the + grid frontend. +- OSC 8 hyperlinks, images (sixel/kitty), `faint`/`blink`/`conceal`/ + `strikethrough` (needs a shared `Style` widening, so a protocol bump), + cursor shape/blink, kitty keyboard protocol. +- Terminal session persistence/reconnect across editor restart. +- **A terminal close/kill command** — the remaining half of `COHERENCE.md` + §2 step 8's discoverability gap. It belongs with the panel-terminal work, + where entry and exit points get designed together. The *opening* keybinding + is **no longer deferred**: Stage 1 carries it as Q#TC10. +- **Genuine immutability for generated buffers — and it is bigger than a Lua + setter.** Today no Lua binding sets `read_only` (`src/lua_bindings` only + reads it, `fold.rs:313`), so every Lua-created "read-only" buffer — listview + panels, `*compilation*`, and this snapshot — is read-only against dispatch + alone and relies entirely on `set_round_trip_input` (Q#TC6a). + + Merely **exposing `set_read_only` would break all three.** The + intercept-bypass path is `ensure_writable`-guarded too: + `apply_edit_skip_intercepts` calls it first (`src/buffer.rs:994`), and that + is exactly the primitive an owner uses to rewrite its own generated buffer. + Flipping the flag would stop listview refreshing, `*compilation*` streaming, + and this snapshot refreshing — the very operations those buffers exist for. + + So the lane needs **two** things, not one: genuine immutability at the + rope/CRDT boundary, *and* an owner-authorized update path that is not simply + "skip the intercepts". Naming only the setter would have made it look like a + one-line follow-up. + +## Acceptance + +### Stage 1 — `terminal-config` + +1. `pmacs.terminal.profiles` accepts a strict spec table per name and rejects + unknown fields before anything is spawned, matching `terminal.open`'s + existing transactional contract. +2. `terminal.default-profile` naming an unknown profile fails at open with an + error that **lists the known profile names**, and creates no buffer, + session, or process. An explicitly passed unknown `profile` fails the same + way **even when `terminal.default-profile` is valid** (Q#TC3a). +3. Field-by-field resolution follows Q#TC3a: explicit open field beats profile + field beats scalar setting beats `$SHELL`. `env` **merges**, with explicit + entries overriding profile entries of the same name. +4. `""` in `terminal.default-profile` means "no profile" and is + indistinguishable from unset (Q#TC2a). +5. `terminal.scrollback-rows` takes effect for a terminal opened without an + explicit `scrollback_rows`; an explicit per-open value overrides it; values + outside `0 ..= 4_000_000` are rejected by the registry rather than by the + core, and `0` is accepted as "retain no history". +6. `terminal.escape-key` changes which chord escapes to the editor, observed + through the **real dispatch path**, not by calling the predicate directly. +7. **Two terminals with different buffer-local escape keys each honor their + own**, with no setting written in between (Q#TC4/Q#TC2b). Driven as + **A→B→A**, asserting both directions. This is the pin an epoch-only cache + fails. +8. Across that same **A→B→A** switch with no setting written, the parse count + does **not** increase after each terminal's first keystroke (Q#TC4c) — + pinned by counting parses, not by timing. This is the pin a single + last-entry cache fails while still satisfying 7. +8a. A terminal's cache does not outlive it: killing a terminal and opening a + new one does not serve the dead terminal's chord, and no per-terminal cache + entry survives its session (Q#TC4c). This is the pin an unpurged + editor-side map fails. +9. With `terminal.escape-key = "C-x"`: `C-x C-x` sends **Ctrl-X** to the child, + and an ordinary `C-c` reaches the child as `0x03` like any other unescaped + key (Q#TC4b). Bite: against the hardcoded `&[0x03]`, the first assertion + fails. +10. An unparseable `terminal.escape-key` falls back to `C-c`, reports through + `EditorCore::status`, and leaves the terminal usable (Q#TC4a). Bite: with + the fallback removed, the terminal becomes unescapable. +10a. "Reports once" is once per terminal per effective invalid value + (Q#TC4c): an **A→B→A** switch with the same invalid value reports **once**, + while changing it to a *different* invalid value reports again. The report + count is asserted, not the message text. +11. The terminal opening keybinding invokes the existing command, and is + verified to have shadowed nothing (Q#TC10). +12. Existing `terminal` invocations and every existing terminal test behave + identically with no settings defined and no profiles registered. + +### Stage 2 — `terminal-copy-mode` + +13. `terminal.copy-mode` produces a read-only buffer whose text is + byte-identical to serializing the full retained range through the existing + copy path (Q#TC7) — pinned against the serializer, so the two cannot drift. +14. Soft wraps, hard rows, wide glyphs, combining clusters, and trailing + default blanks appear in the snapshot exactly as Stage 2 criterion 21 pins + them for selection copy. +15. isearch over the snapshot finds content that is **only in scrollback** + (scrolled off the visible screen), with no change to `src/search.rs` (B1). +16. **Ungated, runs in CI:** focusing the snapshot buffer makes + `dispatch_idle_for` report **false**. This is the whole mechanism Q#TC6a + depends on, it needs no CRDT, and it fails the moment + `set_round_trip_input` is dropped — so the load-bearing regression is + caught by the default configuration rather than only by a `crdt`-gated + test that CI never compiles. +17. **Through a semantic frontend** (this one does need CRDT): keys typed in + the snapshot buffer reach ordinary dispatch and never the child, and + **neither the daemon buffer nor the frontend's mirror is mutated** + (Q#TC6a). Bite: with `set_round_trip_input` removed, the optimistic op is + emitted, bypasses the Lua intercept, passes `ensure_writable()`, and + mutates **both sides** — a buffer the editor calls read-only silently + accepts an edit. +18. Re-invoking against the same terminal refreshes in place; the buffer count + does not grow (Q#TC8). Killing the snapshot leaves the terminal running; + killing the terminal removes the snapshot. +19. `C-t` in a terminal buffer (physically `C-c C-t`) enters copy mode; `g` + refreshes the snapshot from the live terminal and `q` returns to the source + terminal (Q#TC8a). +20. The live terminal's own keys are unchanged while a snapshot exists + (Q#TC9), and the terminal keeps following its tail. +21. The dispatch-shadow count is **unchanged at six** — pinned by asserting + `describe-key` reports the truth for the snapshot buffer's `g` and `q`, + which is the observable difference between the buffer-local idiom and a + shadow. + +## Coherence impact (`COHERENCE.md` §20) + +- **§6 Interaction islands — this arc deliberately adds none.** It is the + first modal-feeling terminal feature that resolves to the buffer-local + keymap idiom §6 identifies as correct, rather than a seventh rung on the + precedence ladder. The shadow count stays at six and `describe-key` stays + truthful (acceptance 21). Worth recording in §6 as a worked example that the + idiom scales to a case that looks modal. +- **§11 Configuration as typed, layered data** — the terminal gains its first + settings, and produces a second blocked adopter for **two** distinct registry + deferrals: the missing table-valued kind (profiles) and the missing + `scope = "global"` flag (the **two open-time settings** — + `terminal.escape-key` deliberately supports buffer-locals, so only + `default-profile` and `scrollback-rows` want an enforcement the registry + cannot express). §11's ground truth should + record both, because the argument for prioritizing them is now cumulative + rather than hypothetical. +- **§2 golden journey, step 8 — partially closed here.** Stage 1 carries the + **terminal opening keybinding** that Priority 1 explicitly names (Q#TC10), + which is the larger half of "works but undiscoverable". Close/kill stays with + the panel work so the entry and exit points are designed together, and is + named in Deferred rather than silently skipped. +- **§5 Unify discovery** — the new commands must carry real descriptions so + M-x rows are useful; no new introspection surface is added. +- No background-work attribution change; no new activity view; no protocol + change. + +## Verification plan + +Full gate suite per `CLAUDE.md` for each PR separately, plus: + +- **The touched terminal suites in BOTH configurations** — default and + `--features crdt` — not only the CRDT one. `vterm_stage1_acceptance`, + `vterm_stage2_acceptance`, and `vterm_stage3_acceptance` all carry tests in + each, and acceptance 12 is a claim about the default configuration too. +- `cargo test --test config_registry_acceptance` for the new settings. +- New suites: `tests/terminal_config_acceptance.rs` (Stage 1) and + `tests/terminal_copy_mode_acceptance.rs` (Stage 2). +- Every behavioral claim bite-verified. The bites that matter most: + **7/8/8a** — three pins that fail against three *different* wrong cache + implementations (epoch-only key, single last-entry, unpurged map), which is + why one pin was not enough; **9** (a hardcoded `0x03` makes the configured + chord unreachable); **10** (its failure mode is a terminal nobody can + escape); and **16/17** (a read-only buffer that silently accepts an edit on + both sides). +- **Do not gate the new suites on `#[cfg(feature = "crdt")]` unless a test + genuinely needs CRDT.** CI never enables that feature, so a suite gated that + way is written and then never run — 264 tests are currently dark for exactly + this reason. That measurement and its lane live on **PR #168**, which is open + and unmerged; it is not yet in `docs/active-work.md` on `main`. + Acceptance 17 does need a semantic frontend, so that one test is gated — but + acceptance 16 pins the same mechanism ungated, so the regression is caught in + CI regardless. That pairing is the pattern to reuse whenever a claim's + end-to-end proof needs CRDT. From cdaea66203aed112a11e3f3b729a2cc212197904 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 18:03:29 -0400 Subject: [PATCH 11/32] fix(lean): make the fallback actually produce a working server MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Round 1 review, four P1s. All real; the first two mean the fallback did not work at all. **1. The latch swapped the config but never spawned or re-attached.** Nothing re-fires an attach on a config change and `attach_buffer` early-returns for a live attachment, so the buffer stayed bound to the server that had just been stopped. The user got a config edit and no language server. `fire_latch` now rebuilds through a new `pmacs.lsp._attach_buffer` export. Two mechanics had to be right for that rebuild to happen at all: * It is **retried on the tick**, because `pmacs.lsp.stop` leaves the state `shutting-down`, which `server_is_live` counts as LIVE — an inline re-attach early-returns the stale record and the swap is a silent no-op. * The latch **does not stop an already-terminal server**, and this is a substrate bug worked around rather than a style choice. `LspManager::stop` on a `Crashed` client takes its not-initialized branch, terminates the dead process, and sets `ShuttingDown { .. None }` on the premise that "the next exit observation cleans up" — but the exit already happened, which is what made it `Crashed`. No further event arrives, so the client is stuck in `ShuttingDown` forever: `server_is_live` reads it as live so `attach_buffer` never rebuilds, and `forget` refuses it for not being terminal. Stopping a dead server is what makes it un-replaceable. Named in framing §6; the fix belongs in `stop` and changes behavior for every language. **2. A missing `lake` bypassed probe and latch entirely** — the single most likely real failure. `ensure_server` swallows a synchronous ENOENT and returns nil, so there was no attachment, and the hook keyed on `active_attachment()` returned before arming anything. The hook now keys on the buffer's LANGUAGE and treats a Lean buffer with no attachment as the failure itself. **3. `waitForDiagnostics` omitted `version`.** Lean's `WaitForDiagnosticsParams` is `{ uri, version }` (v4.9.0, `src/Lean/Data/Lsp/Extra.lean`); the request is how a client says which revision it wants. It looked correct only because the fake server echoes any payload — so the fake server now validates and returns InvalidParams without it. **4. The ledger stated the dangerous stacking order** in one sentence and the correct rule in the next. Fixed to say BEFORE. A safety rule written twice with opposite senses is worse than not written. Also (P2): the probe/latch suite now drives the production path — `buffer.after-load` -> ticks -> probe drain -> latch -> re-attach — with real executable stubs, and asserts the originally opened buffer ends up on a LIVE server. Round 1's acceptance 36 asserted every server was terminal, i.e. pinned the ABSENCE of the fallback it claimed to test. `M.fallback` is a table so the suite can point it at a working stand-in; the probe now spawns `cfg.command --version` rather than a hardcoded `lake`, which is also more correct for a user who configured a wrapper. `swap_to_fallback`'s `command ~= "lake"` guard is gone: the latch fires only when the configured server actually failed, one visible fallback beats no server, and `probe.latched` is what keeps it to exactly one. Three new bites, all against the committed tree: no re-attach after the swap -> three latch tests fail; hook keyed on the attachment -> the missing-`lake` case fails; `waitForDiagnostics` without `version` -> acc37 fails with the server's InvalidParams. --- builtin/runtime/lean.lua | 163 +++++++++++--- builtin/runtime/lsp.lua | 18 ++ docs/active-work.md | 40 +++- docs/lean4-mode-framing.md | 24 ++- src/bin/pmacs_fake_lsp.rs | 33 +++ tests/lean4_server_acceptance.rs | 350 +++++++++++++++++++++++-------- 6 files changed, 510 insertions(+), 118 deletions(-) diff --git a/builtin/runtime/lean.lua b/builtin/runtime/lean.lua index 3fe73da..dd98892 100644 --- a/builtin/runtime/lean.lua +++ b/builtin/runtime/lean.lua @@ -142,6 +142,19 @@ end -- `lake serve` below 3.1.0 starts a server that cannot answer, which is -- worse than failing: `lean4-mode` probes for exactly this and falls -- back to `lean --server`. Parses the leading `x.y` of a version line. +-- State kind for `sid`, or nil if the manager has forgotten it. +local function server_state_kind(sid) + local skey = tostring(sid) + local ok, rows = pcall(pmacs.lsp.list) + if not ok or not rows then return nil end + for _, info in ipairs(rows) do + if tostring(info.id) == skey then + return info.state and info.state.kind + end + end + return nil +end + local function version_below_3_1(text) local major, minor = text:match("(%d+)%.(%d+)") if not major then return false end @@ -150,15 +163,27 @@ local function version_below_3_1(text) return major == 3 and minor < 1 end +-- What the latch falls back TO. A table rather than a literal so the +-- acceptance suite can point it at a stand-in server and drive the real +-- latch path end to end, instead of asserting on a config mutation that +-- proves nothing about whether a server ever starts. +M.fallback = { command = "lean", args = { "--server" } } + -- Swap `command`/`args` ONLY. A wholesale table replacement would -- silently discard a user's `env` / `settings` / `init_options` / `root` -- from `init.lua` at exactly the moment they are least likely to notice. -local function swap_to_lean_server() +-- +-- The only guard is idempotence — already-the-fallback means nothing to +-- do. It deliberately does NOT refuse when the command is user-supplied: +-- the latch fires only when the configured Lean server actually failed +-- to start, and one visible fallback attempt beats leaving the user with +-- no server at all. `probe.latched` is what keeps it to exactly one. +local function swap_to_fallback() local cfg = pmacs.lsp.config.lean4 if not cfg then return false end - if cfg.command ~= "lake" then return false end - cfg.command = "lean" - cfg.args = { "--server" } + if cfg.command == M.fallback.command then return false end + cfg.command = M.fallback.command + cfg.args = M.fallback.args return true end @@ -172,16 +197,66 @@ end -- producing a loop the latch cannot see the end of. `pmacs.lsp.stop` -- sets `restart = Never` on the way out, which is what disarms it. The -- fallback is therefore a FRESH server, not a restart of the old one. +local try_reattach + local function fire_latch(sid, why) if probe.latched then return end probe.latched = true - if sid then pcall(pmacs.lsp.stop, sid) end - if swap_to_lean_server() then - report("LSP: lean4 " .. why .. "; falling back to `lean --server`") - else - report("LSP: lean4 " .. why) - end probe.watching = nil + -- **Only stop a server that is not ALREADY terminal**, and this is + -- load-bearing rather than tidy. `LspManager::stop` on a crashed + -- client takes its not-initialized branch: it terminates the + -- (already-dead) process and sets `ShuttingDown { .. None }`, with the + -- comment "the next exit observation cleans up" — but the exit was + -- already observed, which is what made it `Crashed`. No further event + -- arrives, so the client stays in `ShuttingDown` forever: + -- `server_is_live` counts it as LIVE (neither crashed nor stopped) so + -- `attach_buffer` never rebuilds, and `LspManager::forget` refuses it + -- for not being terminal. Stopping a dead server is what makes it + -- un-replaceable. Recorded as a substrate deferral in the framing §6. + if sid then + local kind = server_state_kind(sid) + if kind and kind ~= "crashed" and kind ~= "stopped" then + pcall(pmacs.lsp.stop, sid) + end + end + if not swap_to_fallback() then + report("LSP: lean4 " .. why) + return + end + report("LSP: lean4 " .. why .. "; falling back to `" + .. tostring(M.fallback.command) .. "`") + -- **Spawn the replacement and re-point the buffer at it.** Stopping + -- and rewriting the config is not a fallback on its own: nothing + -- re-fires an attach on a config change, and `attach_buffer` + -- early-returns for a live attachment, so without this the buffer + -- stays bound to the server we just stopped and the user is left with + -- a config edit and no language server. Round 1 shipped exactly that, + -- with an acceptance test that asserted every server was terminal — + -- i.e. that pinned the absence of the fallback it claimed to check. + -- + -- **Retried on the tick, not done inline**, and that is not caution: + -- `pmacs.lsp.stop` sends shutdown+exit and the state becomes + -- `shutting-down`, which `server_is_live` counts as LIVE. So an + -- immediate `_attach_buffer` early-returns the stale record and the + -- swap has no effect — the exact silent no-op this whole path exists + -- to avoid. Retrying until the old server actually reaches a terminal + -- state is what makes the rebuild happen. + probe.reattach_from = sid and tostring(sid) or false + try_reattach() +end + +-- Returns true once the active Lean buffer is attached to a server that +-- is not the one the latch stopped. +function try_reattach() + if probe.reattach_from == nil then return true end + local ok, rec = pcall(pmacs.lsp._attach_buffer) + if not ok or not rec then return false end + if probe.reattach_from and tostring(rec.server) == probe.reattach_from then + return false + end + probe.reattach_from = nil + return true end local function drain_probe() @@ -220,13 +295,17 @@ local function start_probe(root) if probe.started then return end probe.started = true local cfg = pmacs.lsp.config.lean4 - if not cfg or cfg.command ~= "lake" then return end + if not cfg or not cfg.command then return end + -- Probe the binary we would actually run, not the literal string + -- "lake": a user pointing `command` at a wrapper or an absolute path + -- should have THAT probed, and a hardcoded name would silently probe + -- something else (or nothing). local spec = { -- COHERENCE §9: `ProcessSpec.label` is the only identity a process -- carries, and it is what `pmacs.process.list` renders. A user -- wondering why their editor touched `lake` finds an owner here. label = "lean:lake-version-probe", - command = "lake", + command = cfg.command, args = { "--version" }, stdin = "null", } @@ -275,12 +354,20 @@ end -- does not apply. Resolves when the server has finished elaborating. -- Awaited through Stage 3a's response seam. -- +-- **`version` is required, not optional.** Lean's +-- `WaitForDiagnosticsParams` is `{ uri, version }` (v4.9.0, +-- `src/Lean/Data/Lsp/Extra.lean`), and the request is how the client +-- says *which* revision of the document it wants elaboration for. +-- Sending only `uri` is a malformed request against a real server; it +-- happened to look fine here because the fake server echoes any +-- payload. Callers pass the attachment's current `version`. +-- -- `fn(err)` is called with nil on success. Registering the one-shot -- requires the server to have an attached buffer — see the note on -- `pmacs.lsp.on_response`; every caller here comes from an attachment. -function M.wait_for_diagnostics(sid, uri, fn) +function M.wait_for_diagnostics(sid, uri, version, fn) local ok, rid = pcall(pmacs.lsp.send_request, sid, - "textDocument/waitForDiagnostics", { uri = uri }) + "textDocument/waitForDiagnostics", { uri = uri, version = version }) if not ok then if fn then pcall(fn, tostring(rid)) end return nil @@ -303,7 +390,7 @@ pmacs.command.define { return end pmacs.editor.set_status("lean: elaborating…") - M.wait_for_diagnostics(rec.server, rec.uri, function(err) + M.wait_for_diagnostics(rec.server, rec.uri, rec.version, function(err) if err then pmacs.editor.set_status("lean: " .. tostring(err)) else @@ -327,27 +414,53 @@ end) -- Wiring -------------------------------------------------------------- --- Runs after `lsp.lua`'s own `buffer.after-load` subscription, so the --- attachment already exists. The attachment's `language` IS the Lean --- test — no separate major-mode lookup, which would be a second source --- of truth for the same question. +-- Runs after `lsp.lua`'s own `buffer.after-load` subscription. +-- +-- **Keyed on the buffer's LANGUAGE, not on an attachment existing.** +-- Round 1 keyed on `active_attachment()` and returned early when it was +-- nil — which silently excluded the single most likely real-world +-- failure: `lake` not installed. `ensure_server` pcalls the spawn and +-- returns nil on ENOENT, so `attach_buffer` produces no record at all, +-- so the probe never started and the latch never armed. The case the +-- fallback exists for was the one case it could not see. pmacs.hook.add("buffer.after-load", function() - local rec = pmacs.lsp.active_attachment() - if not rec or rec.language ~= "lean4" then return end + local buf = pmacs.window.buffer() + if not buf then return end + local ok_lang, lang = pcall(pmacs.lsp.buffer_language, buf) + if not ok_lang or lang ~= "lean4" then return end + if not probe.started then local path = pmacs.editor.file_path() start_probe(path and M.root_for(path) or nil) end - -- Watch only the FIRST Lean server: the latch is per session. - if not probe.latched and not probe.saw_initialized - and probe.watching == nil then - probe.watching = rec.server + + local rec = pmacs.lsp.active_attachment() + if rec and rec.language == "lean4" then + -- Watch only the FIRST Lean server: the latch is per session. + if not probe.latched and not probe.saw_initialized + and probe.watching == nil then + probe.watching = rec.server + end + return + end + + -- No attachment for a Lean buffer means `ensure_server` could not + -- spawn at all — a synchronous ENOENT, already swallowed upstream. + -- That is not something to wait for; it is the failure itself, and + -- the only place it is still observable. + if not probe.latched then + fire_latch(nil, "`" .. tostring( + pmacs.lsp.config.lean4 and pmacs.lsp.config.lean4.command) + .. "` could not be started") end end) pmacs.hook.add("process.after-tick", function() drain_probe() poll_latch() + -- Keep trying until the stopped server is really gone; see the note in + -- `fire_latch`. + if probe.reattach_from ~= nil then try_reattach() end end) -- Test seam: acceptance drives the latch deterministically rather than diff --git a/builtin/runtime/lsp.lua b/builtin/runtime/lsp.lua index fd44f1a..0749cfa 100644 --- a/builtin/runtime/lsp.lua +++ b/builtin/runtime/lsp.lua @@ -893,6 +893,24 @@ function pmacs.lsp.active_attachment() return attachments[tostring(buf)] end +-- Re-run the attach for the ACTIVE buffer, rebuilding it against the +-- current `pmacs.lsp.config`. +-- +-- Exists for the Arc 8 Stage 3b fallback latch (Q#LN7): after that latch +-- stops a server that failed to start and rewrites `config.lean4`, +-- something has to actually spawn the replacement and re-point the +-- buffer at it. Nothing else does — `attach_buffer` early-returns for a +-- live attachment, and no hook re-fires on a config change, so without +-- this the buffer stays bound to the stopped server and the "fallback" +-- is a config edit with no effect. +-- +-- Deliberately keyed on the active buffer, matching `attach_buffer`'s +-- own use of `active_buffer_path()`; it is not a general re-attach for +-- arbitrary buffers and must not be used as one. +function pmacs.lsp._attach_buffer() + return attach_buffer(pmacs.window.buffer()) +end + -- Arc 4 stage 3: pure modeline projection. This reads the private -- per-buffer attachment map directly so passive split windows report their -- own buffer instead of the focused window. It never attaches, flushes diff --git a/docs/active-work.md b/docs/active-work.md index 99b54ea..1a1e7ae 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -256,12 +256,17 @@ If it does not, stop and repair the remote/fetch configuration. - Same worktree `../pmacs-lean-stage3`, **branched off `lean4-stage3a-seams`, not off `main`** — 3b consumes 3a's response - seam and `pmacs.fs.canonicalize`, so it is strictly sequential and its - PR must be retargeted to `main` only after #167 merges. (Kill-ring - lesson: retarget stacked child PRs BEFORE merging the parent.) + seam and `pmacs.fs.canonicalize`, so it is strictly sequential. + **Retarget PR #170 to `main` BEFORE merging #167, not after** — the + kill-ring lesson exactly. (Round 1 of this ledger entry stated the + reverse in its first sentence and the correct rule in the next; the + review caught it. A safety rule written twice with opposite senses is + worse than not written.) - Ships `builtin/runtime/lean.lua` (new), one `include_str!` line in - `src/editor.rs`, a `leanprogress` mode on `pmacs_fake_lsp`, and - `tests/lean4_server_acceptance.rs` (17 tests). No protocol change. + `src/editor.rs`, `pmacs.lsp._attach_buffer` exported from `lsp.lua`, + a `leanprogress` mode plus `waitForDiagnostics` validation on + `pmacs_fake_lsp`, and `tests/lean4_server_acceptance.rs` (20 tests). + No protocol change. - **Stage 1's acceptance 12 is half superseded and was rewritten, not deleted.** It asserted `pmacs.lsp.config.lean4 == nil` to catch a Stage-3 front-run; 3b is that stage. What survives is the restraint @@ -274,10 +279,29 @@ If it does not, stop and repair the remote/fetch configuration. an EMPTY `lean-toolchain` (a legitimate marker — existence semantics, not content). Discriminator is `read`'s SECOND return; decline only on a non-nil err. Probed on LuaJIT 2.1. -- Four bites recorded, each against the committed tree: bare `io.open` +- Seven bites recorded, each against the committed tree: bare `io.open` → 24a fails / 24b passes; require-non-nil → 24b fails / 24a passes; - no canonicalization → symlinked open spawns two servers; no stop - before fallback → acc36 fails. + no canonicalization → symlinked open spawns two servers; no re-attach + after the swap → three latch tests fail; hook keyed on the attachment + → the missing-`lake` case fails; `waitForDiagnostics` without + `version` → acc37 fails with the server's InvalidParams. +- **SUBSTRATE BUG FOUND, not fixed here (framing §6).** + `LspManager::stop` on an ALREADY-terminal server takes its + not-initialized branch, terminates the dead process and sets + `ShuttingDown { .. None }` on the premise that "the next exit + observation cleans up" — but the exit already happened, which is what + made it `Crashed`. No further event arrives, so the client is stuck in + `ShuttingDown` **forever**: `server_is_live` reads it as LIVE, so + `attach_buffer` never rebuilds, and `forget` refuses it for not being + terminal. **Stopping a dead server is what makes it un-replaceable.** + Lean works around it by checking the state before stopping. +- Round-1 review found four P1s, all real: the latch swapped the config + but never spawned or re-attached (and acc36 *asserted every server was + terminal*, pinning the absence of the fallback); a missing `lake` + bypassed probe and latch entirely because the hook keyed on an + attachment that ENOENT prevents; `waitForDiagnostics` omitted the + `version` Lean requires; and the ledger stated the dangerous stacking + order. - The probe's non-zero exit is deliberately NOT a fallback trigger — §2.9's elan shim makes `lake --version` fail where `lake serve` still works. Only a parseable version below 3.1.0 triggers it; the diff --git a/docs/lean4-mode-framing.md b/docs/lean4-mode-framing.md index 4869252..fac090b 100644 --- a/docs/lean4-mode-framing.md +++ b/docs/lean4-mode-framing.md @@ -1543,6 +1543,20 @@ What remains deferred: which events may be dropped, which is a policy question with user-visible consequences for diagnostics and progress; Stage 3a states the seam's contract around the behavior rather than changing it. +- **`LspManager::stop` on an already-terminal server strands it.** The + not-initialized branch terminates the (already-dead) process and sets + `ShuttingDown { shutdown_request_id: None }` on the premise that "the + next exit observation cleans up" — but for a `Crashed` client the exit + has already been observed, which is what produced that state. No + further event arrives, so the client sits in `ShuttingDown` + permanently: `server_is_live` counts it as live (neither crashed nor + stopped), so `attach_buffer` never rebuilds against it, and + `LspManager::forget` refuses it for not being terminal. **Stopping a + dead server is what makes it un-replaceable.** Found implementing + Stage 3b's latch, which works around it by checking the state before + stopping. The fix belongs in `stop` (treat an already-terminal client + as a no-op, or drive it straight to `Stopped`) and changes behavior + for every language, so it does not ride a Lean PR. - **Forwarding `cfg.restart` through `ensure_server`** — read by `lua_to_lsp_spec`, never set by the spawn table, so silently dropped on every auto-attach (found landing #161). Fixing it changes behavior for @@ -1722,9 +1736,13 @@ the blast radius. channel a user can actually observe; a report added through `pmacs.error` alone must fail this. - **37.** `textDocument/waitForDiagnostics` resolves through the response seam - (Q#LN16). **PATH-and-success-gated live smoke:** if `lake serve` - starts successfully a real elaboration completes and diagnostics - arrive; skipped otherwise, never failed. + (Q#LN16), **carrying both `uri` and `version`** — Lean's + `WaitForDiagnosticsParams` requires the document version, and a fake + server that echoes any payload will hide its absence, so the fixture + must reject a request that omits it. + **PATH-and-success-gated live smoke:** if `lake serve` starts + successfully a real elaboration completes and diagnostics arrive; + skipped otherwise, never failed. These two sections are bulleted with explicit labels rather than numbered, because the split leaves each stage's criteria non-contiguous diff --git a/src/bin/pmacs_fake_lsp.rs b/src/bin/pmacs_fake_lsp.rs index 5f4fab5..67d6620 100644 --- a/src/bin/pmacs_fake_lsp.rs +++ b/src/bin/pmacs_fake_lsp.rs @@ -1083,6 +1083,39 @@ fn main() { }); write_frame(&mut stdout, &resp); } + ("textDocument/waitForDiagnostics", Some(idv)) => { + // Arc 8 Stage 3b: Lean's `WaitForDiagnosticsParams` is + // `{ uri, version }` (v4.9.0 + // `src/Lean/Data/Lsp/Extra.lean`). Validated here rather + // than echoed, because the generic echo arm below + // accepts anything — which is exactly how a client + // sending only `uri` shipped looking correct. A client + // that omits `version`, or sends a non-integer, gets an + // InvalidParams error the way a real server would. + let uri_ok = params + .get("uri") + .and_then(serde_json::Value::as_str) + .is_some(); + let version_ok = params + .get("version") + .and_then(serde_json::Value::as_i64) + .is_some(); + let resp = if uri_ok && version_ok { + serde_json::json!({ + "jsonrpc": "2.0", "id": idv, "result": serde_json::Value::Null + }) + } else { + serde_json::json!({ + "jsonrpc": "2.0", + "id": idv, + "error": { + "code": -32602, + "message": "waitForDiagnostics requires { uri, version }" + } + }) + }; + write_frame(&mut stdout, &resp); + } (_, Some(idv)) => { // Generic echo response. let resp = serde_json::json!({ diff --git a/tests/lean4_server_acceptance.rs b/tests/lean4_server_acceptance.rs index 13cfcd0..6fe5ed4 100644 --- a/tests/lean4_server_acceptance.rs +++ b/tests/lean4_server_acceptance.rs @@ -347,12 +347,85 @@ fn acc26_did_open_carries_the_lean4_language_id() { } // --------------------------------------------------------------------------- -// Acceptance 28 (probe) — the version predicate. +// Acceptance 27 / 28 / 35 / 36 — the probe and the fallback latch. // -// The parse is unit-tested directly because the spawn path is timing- -// bound; the latch's *effect* is pinned separately below. +// **Driven through the production path**, not by calling internals. +// Round 1's versions poked `_fire_latch` directly and asserted on config +// mutation, which proved nothing about whether a server ever starts — +// and acceptance 36 went further and asserted every server was terminal, +// pinning the ABSENCE of the fallback it claimed to test. These go +// `buffer.after-load` -> ticks -> probe drain -> latch -> re-attach, and +// assert the originally opened buffer ends up on a LIVE server. +// +// The stubs are real executables the fixture writes. `M.fallback` is a +// table precisely so it can point at `pmacs_fake_lsp` here. // --------------------------------------------------------------------------- +impl Fixture { + /// An executable shell stub. `serve` sleeps (so the "server" does not + /// die and only the named failure mode is under test); `--version` + /// prints `version_line`. + fn lake_stub(&self, rel: &str, version_line: &str) -> PathBuf { + use std::os::unix::fs::PermissionsExt as _; + let path = self.root.join(rel); + std::fs::create_dir_all(path.parent().unwrap()).unwrap(); + std::fs::write( + &path, + format!( + "#!/bin/sh\nif [ \"$1\" = \"--version\" ]; then\n echo '{version_line}'\n exit 0\nfi\nexec sleep 300\n" + ), + ) + .unwrap(); + std::fs::set_permissions(&path, std::fs::Permissions::from_mode(0o755)).unwrap(); + path + } +} + +/// Point `command` at `lake_cmd` and the latch's fallback at the fake +/// LSP server, so a fallback that fires produces a server that works. +fn with_fallback(state: &EditorState, lake_cmd: &Path) { + exec( + state, + &format!( + r#" + pmacs.lsp.config.lean4.command = "{}" + pmacs.lsp.config.lean4.args = {{ "serve" }} + pmacs.lean.fallback = {{ command = "{}", args = {{}} }} + "#, + lua_str(lake_cmd), + fake_lsp_path() + ), + ); +} + +/// The active buffer's attached server id, or "none". +fn attached_sid(state: &EditorState) -> String { + eval( + state, + r#" + local rec = pmacs.lsp.active_attachment() + return rec and tostring(rec.server) or "none" + "#, + ) +} + +/// State kind of the active buffer's attached server, or "none". +fn attached_state(state: &EditorState) -> String { + eval( + state, + r#" + local rec = pmacs.lsp.active_attachment() + if not rec then return "none" end + for _, s in ipairs(pmacs.lsp.list()) do + if tostring(s.id) == tostring(rec.server) then + return tostring(s.state and s.state.kind) + end + end + return "gone" + "#, + ) +} + #[test] fn acc28_version_predicate_triggers_only_below_3_1() { let fx = Fixture::new(); @@ -374,36 +447,156 @@ fn acc28_version_predicate_triggers_only_below_3_1() { ); } -// --------------------------------------------------------------------------- -// Acceptance 27 / 35 / 36 — the fallback latch. -// --------------------------------------------------------------------------- +#[test] +fn acc28_an_old_lake_falls_back_and_the_buffer_lands_on_a_live_server() { + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let old_lake = fx.lake_stub("bin/lake", "Lake version 3.0.0"); + let mut state = editor(&fx); + with_fallback(&state, &old_lake); + + open(&state, &file); + settle(&mut state); + // The stub's `serve` sleeps rather than dying, so ONLY the probe can + // have caused a fallback here. That isolation is the point. + for _ in 0..40 { + state.tick_processes(); + state.tick_lsp(); + state.tick_async(); + std::thread::sleep(Duration::from_millis(5)); + if attached_state(&state) == "initialized" { + break; + } + } + + assert_eq!( + attached_state(&state), + "initialized", + "an old lake must leave the buffer on a LIVE fallback server, not \ + merely rewrite the config" + ); + let cmd: String = eval(&state, "return pmacs.lsp.config.lean4.command"); + assert_eq!(cmd, fake_lsp_path(), "the fallback command is in effect"); +} + +#[test] +fn acc28_a_current_lake_does_not_trigger_the_fallback() { + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let new_lake = fx.lake_stub("bin/lake", "Lake version 3.1.0"); + let mut state = editor(&fx); + with_fallback(&state, &new_lake); + + open(&state, &file); + for _ in 0..20 { + state.tick_processes(); + state.tick_lsp(); + state.tick_async(); + std::thread::sleep(Duration::from_millis(5)); + } + + // Non-vacuity against the test above: same harness, same stub shape, + // only the version differs — so a latch that fired unconditionally + // would be caught here. + let cmd: String = eval(&state, "return pmacs.lsp.config.lean4.command"); + assert_eq!( + cmd, + new_lake.display().to_string(), + "a current lake keeps its command; the probe must not fall back" + ); + let latched: bool = eval(&state, "return pmacs.lean._probe.latched"); + assert!(!latched, "the latch did not arm"); +} + +#[test] +fn acc27_a_missing_lake_falls_back_and_the_buffer_lands_on_a_live_server() { + // The case round 1 could not see at all: `ensure_server` swallows a + // synchronous ENOENT and returns nil, so there is no attachment to + // key off. This is also the most likely real-world failure — a user + // with `lean` but no `lake`. + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let absent = fx.dir("bin/no-such-lake"); + let mut state = editor(&fx); + with_fallback(&state, &absent); + + open(&state, &file); + for _ in 0..40 { + state.tick_processes(); + state.tick_lsp(); + state.tick_async(); + std::thread::sleep(Duration::from_millis(5)); + if attached_state(&state) == "initialized" { + break; + } + } + + assert_eq!( + attached_state(&state), + "initialized", + "a missing `lake` must fall back to a live server and re-attach \ + the buffer that was already open" + ); + let status = state.core.borrow().status.clone(); + assert!( + status.contains("lean4"), + "and it says so on the status line; saw {status:?}" + ); +} + +#[test] +fn acc27_the_latch_is_one_shot_and_does_not_re_arm() { + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let absent = fx.dir("bin/no-such-lake"); + let mut state = editor(&fx); + with_fallback(&state, &absent); + + open(&state, &file); + settle(&mut state); + let after_first: String = eval(&state, "return pmacs.lsp.config.lean4.command"); + assert_eq!(after_first, fake_lsp_path(), "the fallback fired once"); + + // A user who deliberately sets something else after the fallback must + // not have it silently replaced by a second firing. + exec(&state, "pmacs.lsp.config.lean4.command = \"user-choice\""); + exec(&state, "pmacs.lean._fire_latch(nil, \"a second failure\")"); + assert_eq!( + eval::(&state, "return pmacs.lsp.config.lean4.command"), + "user-choice", + "the latch never re-arms within a session" + ); +} #[test] fn acc35_latch_preserves_user_config_and_swaps_only_command_and_args() { let fx = Fixture::new(); - let state = editor(&fx); - // A user's init.lua settings, on the shipped shape. + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let absent = fx.dir("bin/no-such-lake"); + let mut state = editor(&fx); + with_fallback(&state, &absent); exec( &state, - r#" - pmacs.lsp.config.lean4.command = "lake" - pmacs.lsp.config.lean4.args = { "serve" } - pmacs.lsp.config.lean4.env = { MYVAR = "1" } + r" pmacs.lsp.config.lean4.settings = { lean = { verbose = true } } pmacs.lsp.config.lean4.init_options = { hasWidgets = false } _G.root_before = pmacs.lsp.config.lean4.root - pmacs.lean._fire_latch(nil, "test") - "#, + ", ); + open(&state, &file); + settle(&mut state); + let after: String = eval( &state, r#" local c = pmacs.lsp.config.lean4 return table.concat({ - tostring(c.command), - tostring(c.args and c.args[1]), - tostring(c.env and c.env.MYVAR), tostring(c.settings and c.settings.lean and c.settings.lean.verbose), tostring(c.init_options and c.init_options.hasWidgets), tostring(c.root == _G.root_before), @@ -411,78 +604,70 @@ fn acc35_latch_preserves_user_config_and_swaps_only_command_and_args() { "#, ); assert_eq!( - after, "lean|--server|1|true|false|true", - "only command/args change; env, settings, init_options and root \ - survive the swap" - ); -} - -#[test] -fn acc27_the_latch_is_one_shot_and_does_not_re_arm() { - let fx = Fixture::new(); - let state = editor(&fx); - exec( - &state, - r#" - pmacs.lsp.config.lean4.command = "lake" - pmacs.lsp.config.lean4.args = { "serve" } - pmacs.lean._fire_latch(nil, "first failure") - _G.after_first = pmacs.lsp.config.lean4.command - -- A second failure must not rewrite the command again; if it did, - -- a user who deliberately set something else after the fallback - -- would have it silently replaced. - pmacs.lsp.config.lean4.command = "user-choice" - pmacs.lean._fire_latch(nil, "second failure") - _G.after_second = pmacs.lsp.config.lean4.command - "#, - ); - assert_eq!(eval::(&state, "return _G.after_first"), "lean"); - assert_eq!( - eval::(&state, "return _G.after_second"), - "user-choice", - "the latch never re-arms within a session" + after, "true|false|true", + "settings, init_options and root survive the swap; only \ + command/args change" ); } #[test] fn acc36_latch_stops_the_failing_server_before_spawning_the_fallback() { + // A stub whose `serve` exits immediately: the server dies before + // `initialize` completes, which is the failure the latch polls for. + // `RestartPolicy::OnCrash` would otherwise respawn it forever + // underneath the latch, with no attempt ceiling. + use std::os::unix::fs::PermissionsExt as _; let fx = Fixture::new(); fx.toolchain("pkg", "v4.9.0\n"); let file = fx.write("pkg/A.lean", "def a := 1\n"); + let dying = fx.root.join("bin/dying-lake"); + std::fs::create_dir_all(dying.parent().unwrap()).unwrap(); + std::fs::write( + &dying, + "#!/bin/sh\nif [ \"$1\" = \"--version\" ]; then echo 'Lake version 9.9.9'; exit 0; fi\nexit 3\n", + ) + .unwrap(); + std::fs::set_permissions(&dying, std::fs::Permissions::from_mode(0o755)).unwrap(); + let mut state = editor(&fx); + with_fallback(&state, &dying); open(&state, &file); - settle(&mut state); - assert_eq!(rows(&state).len(), 1, "precondition: one server is up"); + for _ in 0..60 { + state.tick_processes(); + state.tick_lsp(); + state.tick_async(); + std::thread::sleep(Duration::from_millis(5)); + if attached_state(&state) == "initialized" { + break; + } + } - // Fire the latch against the live server, exactly as `poll_latch` - // would. `pmacs.lsp.stop` sets `restart = Never` on the way out — - // which is what prevents `RestartPolicy::OnCrash` from respawning the - // broken command underneath the latch, forever, with no attempt cap. - exec( - &state, - r#" - pmacs.lsp.config.lean4.command = "lake" - pmacs.lsp.config.lean4.args = { "serve" } - pmacs.lean._fire_latch(pmacs.lsp.list()[1].id, "failed to start") - "#, + // The load-bearing assertion: the buffer ends up on a LIVE server. + assert_eq!( + attached_state(&state), + "initialized", + "the failing server is stopped and the buffer re-attached to the \ + fallback — not left terminal" ); - settle(&mut state); - - let terminal: bool = eval( + // And the dead one really is stopped, so nothing is respawning it. + let dying_still_running: bool = eval( &state, r#" + local live = tostring(pmacs.lsp.active_attachment().server) for _, s in ipairs(pmacs.lsp.list()) do - local k = s.state and s.state.kind - if k ~= "stopped" and k ~= "crashed" then return false end + if tostring(s.id) ~= live then + local k = s.state and s.state.kind + if k ~= "stopped" and k ~= "crashed" then return true end + end end - return true + return false "#, ); assert!( - terminal, - "the failing server is stopped, not left to be respawned under \ - the latch" + !dying_still_running, + "the failing server is not respawning underneath the latch" ); + assert_ne!(attached_sid(&state), "none"); } // --------------------------------------------------------------------------- @@ -492,23 +677,24 @@ fn acc36_latch_stops_the_failing_server_before_spawning_the_fallback() { #[test] fn acc36a_latch_leaves_a_status_line_trace() { let fx = Fixture::new(); - let state = editor(&fx); - exec( - &state, - r#" - pmacs.lsp.config.lean4.command = "lake" - pmacs.lsp.config.lean4.args = { "serve" } - pmacs.lean._fire_latch(nil, "`lake serve` failed to start") - "#, - ); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let absent = fx.dir("bin/no-such-lake"); + let mut state = editor(&fx); + with_fallback(&state, &absent); + + open(&state, &file); + settle(&mut state); + let status = state.core.borrow().status.clone(); assert!( - status.contains("lean4") && status.contains("lean --server"), - "the fallback names itself and what it fell back to; saw {status:?}" + status.contains("lean4") && status.contains("falling back"), + "the fallback names the language and says it fell back; saw {status:?}" ); // The channel assertion is the point (COHERENCE §1.2): a report made // only through `pmacs.error` — undefined in production — would leave - // this empty while every other assertion here still passed. + // this empty while the fallback itself still worked, so the user + // would silently be on a different server than they configured. assert!(!status.is_empty()); } @@ -550,7 +736,7 @@ fn acc37_wait_for_diagnostics_resolves_through_the_response_seam() { r#" _G.settled = "never" local rec = pmacs.lsp.active_attachment() - pmacs.lean.wait_for_diagnostics(rec.server, rec.uri, function(err) + pmacs.lean.wait_for_diagnostics(rec.server, rec.uri, rec.version, function(err) _G.settled = tostring(err) end) "#, From 3377db070ae59232c3caca706d550629b11daaac Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 18:35:11 -0400 Subject: [PATCH 12/32] fix(lean): correct the server lifecycle; round 2 review MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Three P1 lifecycle defects and two P2s. The focused suite was 20/20 with every one of them live, which is the part worth keeping. **1. The crashed primary respawned forever underneath the fallback.** Round 2 skipped the retire call for terminal servers to avoid corrupting them — but the crash had already armed `next_restart_at`, and `maybe_restart` fires on every elapsed backoff with no attempt ceiling. The broken command kept respawning under the live fallback. The right call depends on the state, and each is wrong for the other: `forget` REQUIRES a terminal state and removes the client outright, which also drops the restart timer; `stop` is for a live one and corrupts a terminal one (its not-initialized branch parks it in `ShuttingDown` forever). `retire_server` now dispatches on state. **2. Re-attachment targeted whatever buffer was active when the asynchronous verdict landed.** `_attach_buffer` is an active-buffer-only seam, and "some attachment now names a different server" is satisfied by an unrelated Rust buffer — clearing the retry and leaving the Lean buffer stale forever. The initiating buffer is now captured and the retry waits for it. **3. A failing fallback retried every tick forever, silently**, contradicting acceptance 27's promise that a second failure surfaces. "Waiting for the old server to go" and "attempting the replacement" are now separate: once the old one is terminal or gone, the replacement is attempted EXACTLY once, and a spawn failure is reported. **4. The Lake version parser was being applied to arbitrary wrappers.** `version_below_3_1` encodes lake's output contract; a working `my-lean-wrapper` reporting "wrapper 1.0" would have been replaced despite its server initializing fine. The version probe is now gated on the command's basename being `lake`. The FAILURE latch stays command-agnostic — that one keys on the server actually not starting, which is true of any command. **5. An unconfigured Lean server was reported as a failure** and latched, poisoning the session so a later configuration could never take effect. Absent config or command now means disabled; only a configured command that produced no attachment is a failure. **6. The ledger recorded pre-fix counts** after the fixes were pushed. Now 25/25 and 3,214. That is the #161 fmt-blocker error in a slower form: verification must describe the pushed tree. Sign-offs requested in review: `M.fallback` is now `M._fallback`, an underscored test seam, and its idempotence check compares args as well as command — the same command with different arguments is not "already applied". Dropping the `command ~= "lake"` guard stands for the failure latch only. Five regression tests added, and **three of them were too weak on first write; only bite-testing found it**: * asserting "no live non-fallback server" misses a respawn loop, because a respawning server sits in `crashed` most of the time — `attempt` is the observable that counts respawns; * returning to a buffer with `find_or_open` re-fires `buffer.after-load`, which repairs the attachment regardless of the code under test — `switch_buffer` is the honest return; * a MISSING command fails synchronously inside `after-load` where the rebuild happens inline, so the async race cannot occur — only the probe path exercises it. Each of the five now fails against the exact round-2 mutation it targets. --- builtin/runtime/lean.lua | 216 ++++++++++++++++++--------- docs/active-work.md | 12 +- docs/lean4-mode-framing.md | 12 +- tests/lean4_server_acceptance.rs | 247 ++++++++++++++++++++++++++++++- 4 files changed, 406 insertions(+), 81 deletions(-) diff --git a/builtin/runtime/lean.lua b/builtin/runtime/lean.lua index dd98892..5f74792 100644 --- a/builtin/runtime/lean.lua +++ b/builtin/runtime/lean.lua @@ -125,7 +125,8 @@ local probe = { started = false, -- the `lake --version` probe has been spawned latched = false, -- the fallback has fired (or been ruled out) proc = nil, -- process id of the running probe - buf = "", -- accumulated probe stdout + out = "", -- accumulated probe stdout + buf_key = nil, -- tostring() of the buffer that started this watching = nil, -- sid we are waiting to see fail before initialize saw_initialized = false, } @@ -142,9 +143,9 @@ end -- `lake serve` below 3.1.0 starts a server that cannot answer, which is -- worse than failing: `lean4-mode` probes for exactly this and falls -- back to `lean --server`. Parses the leading `x.y` of a version line. --- State kind for `sid`, or nil if the manager has forgotten it. -local function server_state_kind(sid) - local skey = tostring(sid) +-- State kind for the server whose `tostring(id)` is `skey`, or nil if +-- the manager has forgotten it (which is itself a terminal answer). +local function server_state_kind_for_key(skey) local ok, rows = pcall(pmacs.lsp.list) if not ok or not rows then return nil end for _, info in ipairs(rows) do @@ -155,6 +156,10 @@ local function server_state_kind(sid) return nil end +local function server_state_kind(sid) + return server_state_kind_for_key(tostring(sid)) +end + local function version_below_3_1(text) local major, minor = text:match("(%d+)%.(%d+)") if not major then return false end @@ -163,11 +168,25 @@ local function version_below_3_1(text) return major == 3 and minor < 1 end --- What the latch falls back TO. A table rather than a literal so the --- acceptance suite can point it at a stand-in server and drive the real --- latch path end to end, instead of asserting on a config mutation that --- proves nothing about whether a server ever starts. -M.fallback = { command = "lean", args = { "--server" } } +-- What the latch falls back TO. +-- +-- **Underscored: a test seam, not supported user configuration.** It is +-- a table only so the acceptance suite can point it at a stand-in server +-- and drive the real latch path end to end, instead of asserting on a +-- config mutation that proves nothing about whether a server ever +-- starts. Presenting it as public config would owe framing, +-- documentation, validation and mutation semantics that nothing here +-- provides; users configure Lean through `pmacs.lsp.config.lean4`. +M._fallback = { command = "lean", args = { "--server" } } + +local function same_args(a, b) + a, b = a or {}, b or {} + if #a ~= #b then return false end + for i = 1, #a do + if a[i] ~= b[i] then return false end + end + return true +end -- Swap `command`/`args` ONLY. A wholesale table replacement would -- silently discard a user's `env` / `settings` / `init_options` / `root` @@ -181,81 +200,111 @@ M.fallback = { command = "lean", args = { "--server" } } local function swap_to_fallback() local cfg = pmacs.lsp.config.lean4 if not cfg then return false end - if cfg.command == M.fallback.command then return false end - cfg.command = M.fallback.command - cfg.args = M.fallback.args + -- Idempotence compares command AND args: the same command with + -- different arguments is not "already applied", and treating it as + -- such would silently skip a swap that still needed to happen. + if cfg.command == M._fallback.command + and same_args(cfg.args, M._fallback.args) then + return false + end + cfg.command = M._fallback.command + cfg.args = M._fallback.args return true end --- Fire the fallback: stop the failing server FIRST, then swap, then let --- the next attach spawn afresh. --- --- Stopping first is load-bearing, not defensive. The spec default is --- `LspRestartPolicy::OnCrash`, the termination handler never consults --- the exit code, and `maybe_restart` has no attempt ceiling — so a --- broken `lake` respawns forever on a backoff, underneath the latch, --- producing a loop the latch cannot see the end of. `pmacs.lsp.stop` --- sets `restart = Never` on the way out, which is what disarms it. The --- fallback is therefore a FRESH server, not a restart of the old one. +-- Retire the failed server, swap the command, then rebuild the +-- attachment on the buffer that started this. local try_reattach +-- Retire `sid` so it cannot come back. **Which call to use depends on +-- the state, and using the wrong one is worse than doing nothing:** +-- +-- * TERMINAL (`crashed` / `stopped`) -> `forget`. It requires a +-- terminal state and removes the client outright, which also drops +-- the `next_restart_at` the crash scheduled. `stop` here would take +-- its not-initialized branch and set `ShuttingDown { .. None }` on +-- the premise that "the next exit observation cleans up" — but the +-- exit already happened, which is what made it `Crashed`. No +-- further event arrives, so it sits in `ShuttingDown` forever: +-- `server_is_live` reads that as LIVE so `attach_buffer` never +-- rebuilds, and `forget` then refuses it for not being terminal. +-- * NON-TERMINAL -> `stop`. `forget` rejects it, and `stop` disables +-- restart and drives the polite shutdown. +-- +-- Round 1 skipped the call entirely for terminal servers. That avoided +-- the corruption but left `next_restart_at` armed, so the crashed +-- primary respawned 500ms later and kept respawning underneath the +-- live fallback — invisible to a test that stopped ticking first. +local function retire_server(sid) + local kind = server_state_kind(sid) + if kind == nil then return end + if kind == "crashed" or kind == "stopped" then + pcall(pmacs.lsp.forget, sid) + else + pcall(pmacs.lsp.stop, sid) + end +end + local function fire_latch(sid, why) if probe.latched then return end probe.latched = true probe.watching = nil - -- **Only stop a server that is not ALREADY terminal**, and this is - -- load-bearing rather than tidy. `LspManager::stop` on a crashed - -- client takes its not-initialized branch: it terminates the - -- (already-dead) process and sets `ShuttingDown { .. None }`, with the - -- comment "the next exit observation cleans up" — but the exit was - -- already observed, which is what made it `Crashed`. No further event - -- arrives, so the client stays in `ShuttingDown` forever: - -- `server_is_live` counts it as LIVE (neither crashed nor stopped) so - -- `attach_buffer` never rebuilds, and `LspManager::forget` refuses it - -- for not being terminal. Stopping a dead server is what makes it - -- un-replaceable. Recorded as a substrate deferral in the framing §6. - if sid then - local kind = server_state_kind(sid) - if kind and kind ~= "crashed" and kind ~= "stopped" then - pcall(pmacs.lsp.stop, sid) - end - end + if sid then retire_server(sid) end if not swap_to_fallback() then report("LSP: lean4 " .. why) return end report("LSP: lean4 " .. why .. "; falling back to `" - .. tostring(M.fallback.command) .. "`") - -- **Spawn the replacement and re-point the buffer at it.** Stopping - -- and rewriting the config is not a fallback on its own: nothing - -- re-fires an attach on a config change, and `attach_buffer` - -- early-returns for a live attachment, so without this the buffer - -- stays bound to the server we just stopped and the user is left with - -- a config edit and no language server. Round 1 shipped exactly that, - -- with an acceptance test that asserted every server was terminal — - -- i.e. that pinned the absence of the fallback it claimed to check. + .. tostring(M._fallback.command) .. "`") + -- **Spawn the replacement and re-point the buffer at it.** Swapping + -- the config is not a fallback on its own: nothing re-fires an attach + -- on a config change and `attach_buffer` early-returns for a live + -- attachment, so without this the buffer stays bound to the server we + -- just retired and the user has a config edit and no language server. -- - -- **Retried on the tick, not done inline**, and that is not caution: - -- `pmacs.lsp.stop` sends shutdown+exit and the state becomes - -- `shutting-down`, which `server_is_live` counts as LIVE. So an - -- immediate `_attach_buffer` early-returns the stale record and the - -- swap has no effect — the exact silent no-op this whole path exists - -- to avoid. Retrying until the old server actually reaches a terminal - -- state is what makes the rebuild happen. + -- The rebuild waits for two things, and conflating them is what made + -- round 2 wrong in two ways at once: + -- 1. the retired server actually reaching a terminal state (or + -- being gone) — `stop` leaves `shutting-down`, which + -- `server_is_live` counts as LIVE, so attaching before then + -- early-returns the stale record and the swap silently no-ops; + -- 2. the buffer that started this being the ACTIVE one, because + -- `_attach_buffer` is an active-buffer-only seam. The verdict + -- arrives asynchronously, so the user may well be somewhere else + -- by then — and "some attachment now names a different server" + -- is satisfied by an unrelated Rust buffer, which would clear the + -- retry while leaving the Lean buffer stale forever. probe.reattach_from = sid and tostring(sid) or false try_reattach() end --- Returns true once the active Lean buffer is attached to a server that --- is not the one the latch stopped. +-- Returns true when there is nothing left to do: either the initiating +-- buffer is attached to the replacement, or the replacement itself +-- failed and that has been reported. function try_reattach() if probe.reattach_from == nil then return true end - local ok, rec = pcall(pmacs.lsp._attach_buffer) - if not ok or not rec then return false end - if probe.reattach_from and tostring(rec.server) == probe.reattach_from then + -- (2) Wait for the initiating buffer to be the active one. + local buf = pmacs.window.buffer() + if not buf or not probe.buf_key or tostring(buf) ~= probe.buf_key then return false end + -- (1) Wait for the retired server to stop counting as live. + if probe.reattach_from then + local kind = server_state_kind_for_key(probe.reattach_from) + if kind ~= nil and kind ~= "crashed" and kind ~= "stopped" then + return false + end + end + -- Both conditions met: attempt the replacement EXACTLY ONCE. Cleared + -- first so a failing fallback cannot retry every tick forever — + -- acceptance 27 promises a second failure surfaces rather than loops. probe.reattach_from = nil + local ok, rec = pcall(pmacs.lsp._attach_buffer) + if not ok or not rec then + report("LSP: lean4 fallback `" .. tostring(M._fallback.command) + .. "` did not start either") + return false + end return true end @@ -265,7 +314,7 @@ local function drain_probe() if not ok or not evs then return end for _, ev in ipairs(evs) do if ev.kind == "stdout" or ev.kind == "stderr" then - probe.buf = probe.buf .. tostring(ev.bytes) + probe.out = probe.out .. tostring(ev.bytes) elseif ev.kind == "exited" or ev.kind == "signaled" or ev.kind == "crashed" then local proc = probe.proc @@ -279,7 +328,7 @@ local function drain_probe() -- question failure detection would otherwise answer slowly: an -- old-but-working lake that starts a useless server. if ev.kind == "exited" and ev.code == 0 - and version_below_3_1(probe.buf) then + and version_below_3_1(probe.out) then fire_latch(probe.watching, "lake is older than 3.1.0") end end @@ -296,10 +345,20 @@ local function start_probe(root) probe.started = true local cfg = pmacs.lsp.config.lean4 if not cfg or not cfg.command then return end + -- **Only probe something actually named `lake`.** `version_below_3_1` + -- parses the first `x.y` it finds anywhere in the output, which is a + -- rule about LAKE's output contract and nothing else. Run against a + -- user's wrapper it is a category error: a working `my-lean-wrapper` + -- reporting "wrapper 1.0" would be replaced despite its server having + -- initialized fine. The FAILURE latch stays command-agnostic — that + -- one keys on the server actually not starting, which is true of any + -- command — but the version rule only applies where its contract + -- holds. + local base = cfg.command:match("([^/]+)$") or cfg.command + if base ~= "lake" then return end -- Probe the binary we would actually run, not the literal string - -- "lake": a user pointing `command` at a wrapper or an absolute path - -- should have THAT probed, and a hardcoded name would silently probe - -- something else (or nothing). + -- "lake": a user pointing `command` at an absolute path to lake should + -- have THAT probed, not whatever `lake` resolves to on PATH. local spec = { -- COHERENCE §9: `ProcessSpec.label` is the only identity a process -- carries, and it is what `pmacs.process.list` renders. A user @@ -429,6 +488,11 @@ pmacs.hook.add("buffer.after-load", function() local ok_lang, lang = pcall(pmacs.lsp.buffer_language, buf) if not ok_lang or lang ~= "lean4" then return end + -- The buffer that started this, remembered for the asynchronous + -- rebuild: `_attach_buffer` acts on whatever is active when the + -- verdict lands, which may be a different buffer entirely. + probe.buf_key = tostring(buf) + if not probe.started then local path = pmacs.editor.file_path() start_probe(path and M.root_for(path) or nil) @@ -444,14 +508,21 @@ pmacs.hook.add("buffer.after-load", function() return end - -- No attachment for a Lean buffer means `ensure_server` could not - -- spawn at all — a synchronous ENOENT, already swallowed upstream. - -- That is not something to wait for; it is the failure itself, and - -- the only place it is still observable. + -- **Unconfigured is DISABLED, not failed.** A user who sets + -- `pmacs.lsp.config.lean4 = nil`, or clears its `command`, has turned + -- the Lean server off; reporting that "nil could not be started" is a + -- false alarm, and latching would poison the session so a later + -- configuration could never take effect. Only a CONFIGURED command + -- that produced no attachment is a failure. + local cfg = pmacs.lsp.config.lean4 + if not cfg or not cfg.command then return end + + -- No attachment for a Lean buffer with a configured command means + -- `ensure_server` could not spawn at all — a synchronous ENOENT, + -- already swallowed upstream. That is not something to wait for; it + -- is the failure itself, and the only place it is still observable. if not probe.latched then - fire_latch(nil, "`" .. tostring( - pmacs.lsp.config.lean4 and pmacs.lsp.config.lean4.command) - .. "` could not be started") + fire_latch(nil, "`" .. tostring(cfg.command) .. "` could not be started") end end) @@ -467,6 +538,7 @@ end) -- waiting on real process timing. Not part of the public surface. M._probe = probe M._fire_latch = fire_latch +M._try_reattach = try_reattach M._version_below_3_1 = version_below_3_1 pmacs.lean = M diff --git a/docs/active-work.md b/docs/active-work.md index 1a1e7ae..4a650da 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -265,7 +265,7 @@ If it does not, stop and repair the remote/fetch configuration. - Ships `builtin/runtime/lean.lua` (new), one `include_str!` line in `src/editor.rs`, `pmacs.lsp._attach_buffer` exported from `lsp.lua`, a `leanprogress` mode plus `waitForDiagnostics` validation on - `pmacs_fake_lsp`, and `tests/lean4_server_acceptance.rs` (20 tests). + `pmacs_fake_lsp`, and `tests/lean4_server_acceptance.rs` (25 tests). No protocol change. - **Stage 1's acceptance 12 is half superseded and was rewritten, not deleted.** It asserted `pmacs.lsp.config.lean4 == nil` to catch a @@ -308,10 +308,14 @@ If it does not, stop and repair the remote/fetch configuration. server-failure latch covers the rest. - Verification on this branch: `cargo fmt --check` clean; strict workspace Clippy clean; 1,826 default + 2,003 CRDT library tests; - lean4 server 17/17; lean4 stage 1 9/9; dispatch seams 15/15; + lean4 server 25/25; lean4 stage 1 9/9; dispatch seams 15/15; multi-root 13/13; M4 121; required GPU 155; **isolated-config - workspace sweep 3,206 across 94 suites, zero failures**; - `git diff --check` clean. + workspace sweep 3,214 across 94 suites, zero failures**; + `git diff --check` clean. (Round 1 of + this entry recorded 17/17 and 3,206 — the PRE-fix counts — after the + fixes were pushed. The ledger's protocol is that verification + describes the pushed tree; recording it late is the #161 fmt-blocker + error in a slower form.) ## Dired lane — framing APPROVED; Stage 0 MERGED, Stage 1 next diff --git a/docs/lean4-mode-framing.md b/docs/lean4-mode-framing.md index fac090b..ec62980 100644 --- a/docs/lean4-mode-framing.md +++ b/docs/lean4-mode-framing.md @@ -1553,10 +1553,14 @@ What remains deferred: stopped), so `attach_buffer` never rebuilds against it, and `LspManager::forget` refuses it for not being terminal. **Stopping a dead server is what makes it un-replaceable.** Found implementing - Stage 3b's latch, which works around it by checking the state before - stopping. The fix belongs in `stop` (treat an already-terminal client - as a no-op, or drive it straight to `Stopped`) and changes behavior - for every language, so it does not ride a Lean PR. + Stage 3b's latch, which works around it by dispatching on state: + `forget` for a terminal server (it requires terminal state, and + removing the client also drops the `next_restart_at` the crash armed), + `stop` for a live one. Merely *skipping* the call is not enough — that + leaves the restart timer running and the broken command respawns + underneath the fallback. The fix belongs in `stop` (treat an + already-terminal client as a no-op, or drive it straight to `Stopped`) + and changes behavior for every language, so it does not ride a Lean PR. - **Forwarding `cfg.restart` through `ensure_server`** — read by `lua_to_lsp_spec`, never set by the spawn table, so silently dropped on every auto-attach (found landing #161). Fixing it changes behavior for diff --git a/tests/lean4_server_acceptance.rs b/tests/lean4_server_acceptance.rs index 6fe5ed4..cba8792 100644 --- a/tests/lean4_server_acceptance.rs +++ b/tests/lean4_server_acceptance.rs @@ -390,7 +390,7 @@ fn with_fallback(state: &EditorState, lake_cmd: &Path) { r#" pmacs.lsp.config.lean4.command = "{}" pmacs.lsp.config.lean4.args = {{ "serve" }} - pmacs.lean.fallback = {{ command = "{}", args = {{}} }} + pmacs.lean._fallback = {{ command = "{}", args = {{}} }} "#, lua_str(lake_cmd), fake_lsp_path() @@ -824,3 +824,248 @@ fn lean_root_is_canonical_so_a_symlinked_open_reuses_one_server() { both spellings produce the same affinity key" ); } + +// --------------------------------------------------------------------------- +// Round-2 review findings. Each of these fails against the code as it +// stood at cdaea66, where the focused suite was already 20/20 — the +// lifecycle defects were invisible to it. +// --------------------------------------------------------------------------- + +/// Tick for at least `ms`, so a 500ms restart backoff actually elapses. +/// The round-2 defect was invisible precisely because the suite stopped +/// ticking as soon as the fallback initialized, ~300ms in. +fn tick_for(state: &mut EditorState, ms: u64) { + let deadline = std::time::Instant::now() + Duration::from_millis(ms); + while std::time::Instant::now() < deadline { + state.tick_processes(); + state.tick_lsp(); + state.tick_async(); + std::thread::sleep(Duration::from_millis(5)); + } +} + +#[test] +fn r2_crashed_primary_does_not_respawn_underneath_the_fallback() { + // The crash schedules `next_restart_at`; `maybe_restart` fires after + // the 500ms backoff with no attempt ceiling. Skipping the retire + // call (round 2) left that armed, so the broken command kept + // respawning under the live fallback — forever, unobserved. + use std::os::unix::fs::PermissionsExt as _; + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let dying = fx.root.join("bin/dying-lake"); + std::fs::create_dir_all(dying.parent().unwrap()).unwrap(); + std::fs::write(&dying, "#!/bin/sh\nexit 3\n").unwrap(); + std::fs::set_permissions(&dying, std::fs::Permissions::from_mode(0o755)).unwrap(); + + let mut state = editor(&fx); + with_fallback(&state, &dying); + open(&state, &file); + // Well past one backoff. + tick_for(&mut state, 1400); + + // **`attempt`, not liveness.** A respawning server spends most of + // its life in `crashed` waiting out the backoff, so "no live + // non-fallback server" is satisfied while it loops forever — that + // weaker assertion passed against the round-2 code and caught + // nothing. `attempt` increments on every spawn, so it counts the + // respawns directly. A retired server is absent from the list + // entirely (`forget` removes the client); one left with + // `next_restart_at` armed climbs past 1. + let worst_attempt: i64 = eval( + &state, + r#" + local rec = pmacs.lsp.active_attachment() + local live = rec and tostring(rec.server) or "" + local worst = 0 + for _, s in ipairs(pmacs.lsp.list()) do + if tostring(s.id) ~= live then + local a = s.attempt or 0 + if a > worst then worst = a end + end + end + return worst + "#, + ); + assert_eq!( + worst_attempt, 0, + "the retired primary is gone from the manager, not respawning after the backoff (attempt > 0 means it is still there; > 1 means it respawned)" + ); + assert_eq!( + attached_state(&state), + "initialized", + "and the buffer is on the live fallback" + ); +} + +#[test] +fn r2_reattach_targets_the_originating_buffer_not_whatever_is_active() { + // `_attach_buffer` is an active-buffer-only seam and the latch's + // verdict arrives asynchronously. Round 2 accepted "some attachment + // now names a different server", which an unrelated Rust buffer + // satisfies — clearing the retry and stranding the Lean buffer. + // + // **Driven through the PROBE**, not through a missing executable: a + // missing command fails synchronously inside `buffer.after-load`, + // where the Lean buffer is still active and the rebuild happens + // inline, so the race cannot occur and the test proves nothing. The + // probe's verdict lands on a later tick, which is the whole point. + // The stub's `serve` sleeps, so only the probe can trigger anything. + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + fx.write("pkg/Cargo.toml", "[package]\nname = \"p\"\n"); + let lean_file = fx.write("pkg/A.lean", "def a := 1\n"); + let rust_file = fx.write("pkg/src/main.rs", "fn main() {}\n"); + let old_lake = fx.lake_stub("bin/lake", "Lake version 3.0.0"); + + let mut state = editor(&fx); + with_fallback(&state, &old_lake); + // A working Rust server, so switching away lands on a real + // attachment with a different server id — the decoy. + exec( + &state, + &format!( + "pmacs.lsp.config.rust = {{ command = \"{}\" }}", + fake_lsp_path() + ), + ); + + open(&state, &lean_file); + exec(&state, "_G.lean_buf = pmacs.window.buffer()"); + // Switch away before the probe's verdict can land. + open(&state, &rust_file); + tick_for(&mut state, 500); + + // Come back with a buffer SWITCH, not `find_or_open`. Re-opening + // fires `buffer.after-load`, which re-runs lsp.lua's own attach and + // would repair the record no matter what the latch did. + exec(&state, "pmacs.window.switch_buffer(_G.lean_buf)"); + tick_for(&mut state, 400); + + let lang: String = eval( + &state, + r#" + local rec = pmacs.lsp.active_attachment() + return rec and tostring(rec.language) or "none" + "#, + ); + assert_eq!(lang, "lean4", "we are back on the Lean buffer"); + + // The observable that discriminates: WHICH command the Lean buffer's + // server is running. A retry cleared by the decoy leaves it on the + // original `lake` stub. + let cmd: String = eval( + &state, + r#" + local rec = pmacs.lsp.active_attachment() + if not rec then return "none" end + for _, s in ipairs(pmacs.lsp.list()) do + if tostring(s.id) == tostring(rec.server) then + return tostring(s.command) + end + end + return "gone" + "#, + ); + assert_eq!( + cmd, + fake_lsp_path(), + "the ORIGINATING Lean buffer ends up on the fallback — a decoy \ + Rust attachment must not satisfy the retry" + ); +} + +#[test] +fn r2_a_failing_fallback_is_reported_once_and_does_not_retry_forever() { + // Acceptance 27 promises a second failure surfaces rather than + // loops. Round 2 retried `_attach_buffer` every tick with nothing + // reported. + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let absent_primary = fx.dir("bin/no-such-lake"); + let absent_fallback = fx.dir("bin/no-such-lean"); + + let mut state = editor(&fx); + exec( + &state, + &format!( + r#" + pmacs.lsp.config.lean4.command = "{}" + pmacs.lsp.config.lean4.args = {{ "serve" }} + pmacs.lean._fallback = {{ command = "{}", args = {{}} }} + "#, + lua_str(&absent_primary), + lua_str(&absent_fallback) + ), + ); + + open(&state, &file); + tick_for(&mut state, 300); + + let status = state.core.borrow().status.clone(); + assert!( + status.contains("did not start either"), + "a failing fallback surfaces rather than retrying silently; saw \ + {status:?}" + ); + // And the retry state is cleared, so it is not looping. + let pending: String = eval(&state, "return tostring(pmacs.lean._probe.reattach_from)"); + assert_eq!(pending, "nil", "the retry is retired, not spinning"); +} + +#[test] +fn r2_a_working_wrapper_is_not_version_probed_as_lake() { + // `version_below_3_1` encodes LAKE's output contract. Applying it to + // an arbitrary wrapper is a category error: a working wrapper + // reporting its own "wrapper 1.0" would be replaced despite its + // server initializing fine. + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + // Named something other than `lake`, reporting a sub-3.1 version, + // but which serves fine. + let wrapper = fx.lake_stub("bin/my-lean-wrapper", "wrapper 1.0"); + let mut state = editor(&fx); + with_fallback(&state, &wrapper); + + open(&state, &file); + tick_for(&mut state, 400); + + let cmd: String = eval(&state, "return pmacs.lsp.config.lean4.command"); + assert_eq!( + cmd, + wrapper.display().to_string(), + "a wrapper's own version string is not Lake's; the version probe \ + must not run against it" + ); + let latched: bool = eval(&state, "return pmacs.lean._probe.latched"); + assert!(!latched, "and the latch stayed disarmed"); +} + +#[test] +fn r2_an_unconfigured_lean_server_is_disabled_not_failed() { + // Setting `pmacs.lsp.config.lean4 = nil` means "off". Reporting that + // `nil` could not start is a false alarm, and latching poisons the + // session so a later configuration can never take effect. + let fx = Fixture::new(); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let mut state = editor(&fx); + exec(&state, "pmacs.lsp.config.lean4 = nil"); + exec(&state, "pmacs.editor.set_status(\"\")"); + + open(&state, &file); + settle(&mut state); + + assert_eq!( + state.core.borrow().status.clone(), + "", + "an unconfigured Lean server reports nothing — it is disabled" + ); + let latched: bool = eval(&state, "return pmacs.lean._probe.latched"); + assert!( + !latched, + "and the session is not poisoned: a later config must still work" + ); +} From 664cc25d0c4977cce9b11a291e633e0b68cbd475 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 18:36:18 -0400 Subject: [PATCH 13/32] feat(terminal): profiles, scrollback, and a configurable escape key Stage 1 of the terminal config/copy-mode arc. The terminal had no configuration surface at all: the command hardcoded $SHELL, scrollback was a per-open argument only, and the escape chord was a literal in Rust. No protocol change. Profiles are a raw Lua table, pmacs.terminal.profiles, not a registry setting: ConfigValue is four scalars with no table kind, so profiles join pmacs.lsp.config and pmacs.pair.sets until table-valued settings exist. The registry gains three scalars whose defaults reproduce the previous behavior exactly. Field resolution is explicit open argument, then profile field, then scalar setting, then $SHELL. env MERGES, with explicit entries overriding the profile's, because first-wins there would silently drop half a user's environment. An explicitly named profile that does not exist is an error even when terminal.default-profile is valid, so a typo cannot silently fall back. The two open-time settings resolve through the GLOBAL chain, because they are read before the identity buffer exists and no caller could have pinned a local override on a buffer that does not yet exist. Only terminal.escape-key resolves per buffer, which makes a per-terminal escape a supported feature. The escape key is parsed at most once per (terminal, config epoch), and the cache lives on TerminalSession so its lifetime is the terminal's, with no purge hook to forget. The epoch alone is not a sufficient key: it does not advance when focus moves between two terminals with different buffer-local values, so an epoch-only cache serves one terminal's chord to the other. An unparseable value falls back to C-c and reports once per terminal per effective invalid value through the status line, because a terminal with no escape chord cannot be escaped to fix the setting that broke it. Repeating the escape now sends THAT chord to the child through the ordinary key encoder, rather than a hardcoded ETX. With an escape of C-x, the previous code sent Ctrl-C and made literal Ctrl-X unreachable. C-c t opens a terminal. COHERENCE Priority 1 names a terminal keybinding, and section 2 step 8 grades the terminal works-but- undiscoverable; C-c is already a live global prefix, so this is a new leaf rather than a shadow. It is unreachable from inside a terminal, where C-c is the escape. Acceptance is tests/terminal_config_acceptance.rs, deliberately NOT crdt-gated so CI actually runs it. Four bites, each against a different plausible wrong implementation: a hardcoded ETX fails acc6/9; an epoch-only cache key fails acc7; a single last-entry cache fails acc8's parse count; removing the invalid-value fallback fails acc10. Two test-instrument notes worth keeping. cat -v is the echo probe because the screen rejects C0 controls before they reach cells, so a raw echoed Ctrl-X would be invisible. And the probe counts occurrences rather than testing presence, because a single-character probe collides with the child's own banner text. --- builtin/runtime/terminal.lua | 149 ++++++- src/editor.rs | 67 +++- src/lua_bindings/mod.rs | 27 ++ src/terminal/mod.rs | 4 + src/terminal/session.rs | 104 +++++ tests/terminal_config_acceptance.rs | 576 ++++++++++++++++++++++++++++ 6 files changed, 914 insertions(+), 13 deletions(-) create mode 100644 tests/terminal_config_acceptance.rs diff --git a/builtin/runtime/terminal.lua b/builtin/runtime/terminal.lua index 6be0987..a6573fe 100644 --- a/builtin/runtime/terminal.lua +++ b/builtin/runtime/terminal.lua @@ -3,6 +3,38 @@ local terminal = assert(pmacs.terminal, "pmacs.terminal raw bindings are required") local raw_open = assert(terminal._open, "pmacs.terminal._open is required") +-- Q#TC2a. Every default reproduces today's behavior exactly, so a tree +-- with no settings written and no profiles registered behaves as before. +pmacs.config.define { + name = "terminal.default-profile", + type = "string", + default = "", + allow_empty = true, + mutability = "live", + description = "Profile name from pmacs.terminal.profiles to open by default. " .. + "Empty means no profile: fall back to $SHELL.", +} + +pmacs.config.define { + name = "terminal.scrollback-rows", + type = "integer", + default = 10000, + min = 0, + max = 4000000, + mutability = "live", + description = "Rows of scrollback retained per terminal. " .. + "0 retains no history.", +} + +pmacs.config.define { + name = "terminal.escape-key", + type = "string", + default = "C-c", + mutability = "live", + description = "Chord that escapes to the editor from a terminal. " .. + "Pressing it twice sends the chord itself to the child.", +} + local function bind_terminal_keys(buffer) local function bind(sequence, command) pmacs.keymap.bind { @@ -19,22 +51,127 @@ local function bind_terminal_keys(buffer) bind("M->", "terminal.scroll-bottom") end +-- Q#TC1: profiles are a raw Lua table, not a config setting. The +-- registry stores four scalars and has no table kind, so a profile — +-- inherently `{ command, args, cwd, env }` — lives here beside +-- `pmacs.lsp.config` and `pmacs.pair.sets` until table-valued settings +-- exist. +terminal.profiles = terminal.profiles or {} + +local PROFILE_FIELDS = { + command = "string", + args = "table", + cwd = "string", + env = "table", +} + +local function validate_profile(name, profile) + if type(profile) ~= "table" then + error(string.format("terminal profile %q must be a table", name), 0) + end + for key, value in pairs(profile) do + local expected = PROFILE_FIELDS[key] + if not expected then + error(string.format("terminal profile %q: unknown field %q", name, tostring(key)), 0) + end + if type(value) ~= expected then + error(string.format( + "terminal profile %q: field %q must be a %s, got %s", + name, key, expected, type(value)), 0) + end + end + return profile +end + +local function known_profile_names() + local names = {} + for name in pairs(terminal.profiles) do names[#names + 1] = name end + table.sort(names) + return names +end + +-- Q#TC2 / Q#TC3a: resolve a profile by name, or nil when none is +-- selected. An explicitly requested profile that does not exist is an +-- error even when `terminal.default-profile` is valid — a typo must not +-- silently fall back to the default. +local function resolve_profile(requested) + local name = requested + if name == nil then + local configured = pmacs.config.get("terminal.default-profile") + if configured == nil or configured == "" then return nil end + name = configured + end + local profile = terminal.profiles[name] + if profile == nil then + local known = known_profile_names() + local listed = #known > 0 and table.concat(known, ", ") or "(none defined)" + error(string.format( + "terminal profile %q is not defined; known profiles: %s", name, listed), 0) + end + return validate_profile(name, profile) +end + +-- Q#TC3a merge order, per field: explicit open field, then the profile's +-- field, then the scalar setting, then the built-in fallback. `env` is +-- the one field where "first wins" would be wrong, so it MERGES with +-- explicit entries overriding the profile's — any other reading silently +-- drops half a user's environment. +local function merge_env(profile_env, explicit_env) + if profile_env == nil then return explicit_env end + local merged = {} + for key, value in pairs(profile_env) do merged[key] = value end + for key, value in pairs(explicit_env or {}) do merged[key] = value end + return merged +end + function terminal.open(spec) - local buffer = raw_open(spec) + spec = spec or {} + local resolved = {} + for key, value in pairs(spec) do + if key ~= "profile" then resolved[key] = value end + end + + local profile = resolve_profile(spec.profile) + if profile then + for key in pairs(PROFILE_FIELDS) do + if key ~= "env" and resolved[key] == nil then resolved[key] = profile[key] end + end + resolved.env = merge_env(profile.env, spec.env) + end + + -- The two open-time settings resolve through the GLOBAL chain + -- (Q#TC2b): they are read before the identity buffer exists, so there + -- is no terminal to resolve a buffer-local against. + if resolved.scrollback_rows == nil then + resolved.scrollback_rows = pmacs.config.get("terminal.scrollback-rows") + end + if resolved.command == nil then + resolved.command = os.getenv("SHELL") or "/bin/sh" + end + + local buffer = raw_open(resolved) bind_terminal_keys(buffer) return buffer end pmacs.command.define { name = "terminal", - description = "Open a terminal running $SHELL (or /bin/sh).", - fn = function() - return terminal.open { - command = os.getenv("SHELL") or "/bin/sh", - } + description = "Open a terminal running the configured profile, or $SHELL.", + fn = function(profile) + return terminal.open { profile = profile } end, } +-- Q#TC10: the opening binding. `COHERENCE.md` Priority 1 names a +-- terminal keybinding as part of protecting the golden journey, and §2 +-- step 8 grades the terminal "works but undiscoverable". `C-c` is +-- already a live global prefix (fold's `C-c @ ...`), so this is a new +-- leaf under it rather than a shadow. +-- +-- Named limitation: unreachable from INSIDE a terminal window, where +-- `C-c` is consumed as the escape. `M-x terminal` still works there. +pmacs.keymap.bind { scope = "global", sequence = "C-c t", command = "terminal" } + pmacs.command.define { name = "terminal.copy-selection", description = "Copy the active terminal selection.", diff --git a/src/editor.rs b/src/editor.rs index 1db5c3a..2bc633b 100644 --- a/src/editor.rs +++ b/src/editor.rs @@ -989,19 +989,28 @@ impl EditorState { .get(&frontend_id) .is_some_and(|state| state.terminal_escape); if let Some(view_key) = terminal_key { + // Q#TC4: the escape chord is per terminal, resolved through + // `terminal.escape-key` and cached on the session so this + // hot path parses at most once per (terminal, config epoch). + let escape_chord = self.terminal_escape_chord(view_key.buffer_id); if escaped { self.dispatchers .entry(frontend_id) .or_default() .terminal_escape = false; - if chord.is_some_and(is_terminal_escape_chord) { + if chord == Some(escape_chord) { + // Q#TC4b: repeating the escape sends THAT chord to the + // child, not a hardcoded ETX. With a configured escape + // of `C-x`, sending Ctrl-C here would both surprise the + // user and make literal Ctrl-X unreachable, since the + // first press is always consumed as the escape. self.claim_terminal_controller(view_key); - self.send_terminal_bytes(view_key.buffer_id, &[0x03]); + self.send_terminal_escape_literal(view_key, escape_chord); return; } // The post-escape key starts a fresh ordinary sequence below. } else if !dispatcher_pending { - if chord.is_some_and(is_terminal_escape_chord) { + if chord == Some(escape_chord) { let state = self.dispatchers.entry(frontend_id).or_default(); state.terminal_escape = true; state.dispatcher = KeyDispatcher::new(); @@ -1117,6 +1126,54 @@ impl EditorState { .then_some(key) } + /// This terminal's effective escape chord (Q#TC4). + /// + /// Resolution is `get("terminal.escape-key", terminal_buffer)` — + /// buffer-local, then global, then default — because unlike the two + /// open-time settings this one is read while the terminal exists, so + /// a per-terminal escape is expressible and supported (Q#TC2b). + /// + /// The parse and the once-per-terminal invalid-value report both live + /// in [`crate::terminal::TerminalManager::escape_chord`]; this method + /// only supplies the resolved spelling and the epoch that keys the + /// cache, and surfaces any report through the status line — the same + /// channel `send_terminal_bytes` uses for terminal failures. + fn terminal_escape_chord(&self, buffer_id: crate::buffer::BufferId) -> Chord { + let lua = self.lua_host.lua(); + let (spelling, epoch) = crate::lua_bindings::config_string_and_epoch( + lua, + "terminal.escape-key", + Some(buffer_id), + crate::terminal::DEFAULT_TERMINAL_ESCAPE_KEY, + ); + let (chord, report) = self + .terminal_manager + .borrow_mut() + .escape_chord(buffer_id, epoch, &spelling); + if let Some(message) = report { + self.core.borrow_mut().status = message; + } + chord + } + + /// Send the configured escape chord to the child as literal input + /// (Q#TC4b), through the same encoder ordinary keys use so it + /// inherits application-cursor and modifier handling. + fn send_terminal_escape_literal(&self, key: TerminalViewKey, chord: Chord) { + let event = KeyEvent::new(chord.code, chord.modifiers); + let Some((terminal_key, modifiers)) = terminal_key_from_crossterm(event) else { + return; + }; + let modes = self + .terminal_manager + .borrow() + .modes_for_view(key) + .unwrap_or_default(); + if let Some(bytes) = crate::terminal::input::encode_key(terminal_key, modifiers, modes) { + self.send_terminal_bytes(key.buffer_id, &bytes); + } + } + fn claim_terminal_controller(&self, key: TerminalViewKey) { let mut manager = self.terminal_manager.borrow_mut(); let _ = manager.register_view(key); @@ -4421,10 +4478,6 @@ fn sanitize_single_line(s: &str) -> String { .collect() } -fn is_terminal_escape_chord(chord: Chord) -> bool { - chord.code == KeyCode::Char('c') && chord.modifiers == KeyModifiers::CONTROL -} - fn terminal_key_from_crossterm(key: KeyEvent) -> Option<(TerminalKey, TerminalModifiers)> { let modifiers = crate::protocol::crossterm_translate::mods_from_crossterm(key.modifiers); let key = crate::protocol::crossterm_translate::keycode_from_crossterm(key.code); diff --git a/src/lua_bindings/mod.rs b/src/lua_bindings/mod.rs index 3482879..1e7ac04 100644 --- a/src/lua_bindings/mod.rs +++ b/src/lua_bindings/mod.rs @@ -668,6 +668,33 @@ pub fn config_u32(lua: &Lua, name: &str, buffer_id: Option, fallback: } } +/// Read a `String` setting plus the registry epoch that keys any cache +/// built from it (Q#TC4c). +/// +/// The epoch is returned WITH the value deliberately: a caller caching a +/// parsed form needs both, and reading them in two calls would let a +/// `set` land between them and produce a cache stamped with the wrong +/// epoch. `fallback` covers a bare core whose runtime never defined the +/// setting, matching [`config_u32`]. +#[must_use] +pub fn config_string_and_epoch( + lua: &Lua, + name: &str, + buffer_id: Option, + fallback: &str, +) -> (String, u64) { + let Some(registry) = lua.app_data_ref::() else { + return (fallback.to_owned(), 0); + }; + let borrowed = registry.borrow(); + let epoch = borrowed.value_epoch(); + let value = match borrowed.get(name, buffer_id) { + Ok(crate::config_registry::ConfigValue::Str(v)) => v.clone(), + _ => fallback.to_owned(), + }; + (value, epoch) +} + /// Short-circuit a binding when the init phase has completed. /// /// Lifecycle-affecting Lua APIs (currently just `pmacs.attach`; M5.6d+) diff --git a/src/terminal/mod.rs b/src/terminal/mod.rs index 27ee96c..11272e4 100644 --- a/src/terminal/mod.rs +++ b/src/terminal/mod.rs @@ -34,6 +34,10 @@ pub use pmacs_protocol::terminal::{ /// Configuration-time, not a wire bound: history never crosses the /// protocol, so this stays core-owned. pub const DEFAULT_TERMINAL_SCROLLBACK_ROWS: usize = 10_000; + +/// Default `terminal.escape-key`, and the fallback an unparseable value +/// falls back to (Q#TC4a). +pub const DEFAULT_TERMINAL_ESCAPE_KEY: &str = "C-c"; /// Maximum retained main-screen history cells. Core-owned for the same /// reason as [`DEFAULT_TERMINAL_SCROLLBACK_ROWS`]. pub const MAX_TERMINAL_HISTORY_CELLS: usize = 4_000_000; diff --git a/src/terminal/session.rs b/src/terminal/session.rs index ee96923..6e71ea3 100644 --- a/src/terminal/session.rs +++ b/src/terminal/session.rs @@ -12,6 +12,7 @@ use crate::ansi::AnsiParserProfile; use crate::buffer::{Buffer, BufferId}; use crate::cell::{Cell, CellCoord, CellSize}; use crate::editor_core::EditorCore; +use crate::key::{Chord, parse_chord}; use crate::process::{ ProcessEventKind, ProcessId, ProcessMode, ProcessSpec, ProcessState, ProcessSupervisor, RestartPolicy, StdinMode, TerminalMode, @@ -218,12 +219,40 @@ pub(super) struct TerminalSession { pub(super) screen: TerminalScreen, pub(super) process: TerminalProcessState, pub(super) annotated: bool, + /// Resolved `terminal.escape-key` for this terminal (Q#TC4c). + /// + /// The cache lives HERE, not in an editor-side map, because a + /// session is created in [`TerminalManager::open`] and dropped on + /// kill/prune — so its lifetime is exactly the cache's, with no + /// purge hook to forget. An editor-side map would leak an entry per + /// terminal; a single last-entry cache would reparse (and re-report + /// an invalid value) every time focus alternates between two + /// terminals. + pub(super) escape: Option, +} + +/// One terminal's parsed escape chord, valid for one config epoch. +pub(super) struct EscapeCache { + /// The `ConfigRegistry::value_epoch` this was parsed at. The key is + /// `(this session, epoch)`: the epoch alone is not enough, because + /// it does not advance when focus moves between terminals with + /// different buffer-local values. + pub(super) epoch: u64, + /// The effective chord — the parsed spelling, or the `C-c` fallback. + pub(super) chord: Chord, + /// The invalid spelling already reported for this terminal, if any. + /// Reporting is once per terminal per effective invalid value: an + /// unchanged bad value stays quiet, a *different* bad value reports + /// again because it is a new mistake. + pub(super) reported_invalid: Option, } /// Owns the one-buffer/one-process/one-screen terminal registry. #[derive(Default)] pub struct TerminalManager { pub(super) sessions: HashMap, + /// Total escape-key parses performed (Q#TC4c observability). + escape_parses: u64, process_to_buffer: HashMap, /// Removed buffers whose children are still being reaped. Their events /// remain manager-owned so Lua/LSP/MCP consumers cannot steal a batch. @@ -331,6 +360,7 @@ impl TerminalManager { screen, process: TerminalProcessState::Running, annotated: false, + escape: None, }, ); debug_assert!(previous.is_none(), "fresh BufferId collided"); @@ -538,6 +568,71 @@ impl TerminalManager { .map_err(TerminalError::Process) } + /// Resolve this terminal's effective escape chord, parsing at most + /// once per `(terminal, config epoch)` (Q#TC4c). + /// + /// `spelling` is the caller-resolved `terminal.escape-key` value and + /// `epoch` the registry's `value_epoch()` it was read at. Returns the + /// effective chord plus, at most once per terminal per effective + /// invalid value, a message the caller should surface. + /// + /// An unparseable spelling falls back to `C-c` rather than leaving the + /// terminal with no escape at all (Q#TC4a): without one, every key goes + /// to the child and the user cannot reach the binding that would fix + /// the setting that broke it. + pub fn escape_chord( + &mut self, + buffer_id: BufferId, + epoch: u64, + spelling: &str, + ) -> (Chord, Option) { + let fallback = default_escape_chord(); + if let Some(session) = self.sessions.get(&buffer_id) + && let Some(cache) = session.escape.as_ref() + && cache.epoch == epoch + { + return (cache.chord, None); + } + self.escape_parses = self.escape_parses.saturating_add(1); + let Some(session) = self.sessions.get_mut(&buffer_id) else { + return (fallback, None); + }; + let previously_reported = session + .escape + .as_ref() + .and_then(|cache| cache.reported_invalid.clone()); + let (chord, reported_invalid, report) = match parse_chord(spelling) { + Ok(chord) => (chord, None, None), + Err(error) => { + let already = previously_reported.as_deref() == Some(spelling); + let message = (!already).then(|| { + format!( + "terminal.escape-key {spelling:?} is not a valid chord ({error}); using C-c" + ) + }); + (fallback, Some(spelling.to_owned()), message) + } + }; + session.escape = Some(EscapeCache { + epoch, + chord, + reported_invalid, + }); + (chord, report) + } + + /// How many escape-key spellings this manager has parsed. + /// + /// An observability seam for Q#TC4c's cache contract, which is + /// otherwise unpinnable for a VALID setting: a correct per-session + /// cache and a single last-entry cache produce identical behavior + /// there and differ only in how often they parse. Counting reports + /// covers the invalid case; this covers the valid one. + #[must_use] + pub fn escape_parses(&self) -> u64 { + self.escape_parses + } + /// Resize a terminal screen and its PTY after validating shared limits. pub fn resize( &mut self, @@ -730,3 +825,12 @@ fn sanitize_metadata(value: &str) -> String { } clean } + +/// The built-in terminal escape chord, and the fallback for an +/// unparseable `terminal.escape-key` (Q#TC4a). +pub(super) fn default_escape_chord() -> Chord { + Chord::new( + crossterm::event::KeyCode::Char('c'), + crossterm::event::KeyModifiers::CONTROL, + ) +} diff --git a/tests/terminal_config_acceptance.rs b/tests/terminal_config_acceptance.rs new file mode 100644 index 0000000..a613eb2 --- /dev/null +++ b/tests/terminal_config_acceptance.rs @@ -0,0 +1,576 @@ +//! Terminal configuration acceptance (Stage 1 of +//! `docs/terminal-config-and-copy-mode-framing.md`, criteria 1-12). +//! +//! Deliberately NOT `#[cfg(feature = "crdt")]`: CI never enables that +//! feature, so a gated suite is written and then never run. + +use std::thread; +use std::time::{Duration, Instant}; + +use crossterm::event::{KeyCode, KeyEvent, KeyModifiers}; +use mlua::Value; +use pmacs::cell::{CellSize, Glyph}; +use pmacs::editor::EditorState; +use pmacs::protocol::FrontendId; +use pmacs::terminal::TerminalViewKey; +use pmacs::window::WindowId; + +fn exec(state: &EditorState, src: &str) { + state + .lua_host + .lua() + .load(src) + .exec() + .unwrap_or_else(|e| panic!("lua failed: {src}\n{e}")); +} + +fn eval_err(state: &EditorState, src: &str) -> String { + let result: mlua::Result = state.lua_host.lua().load(src).eval(); + match result { + Ok(_) => panic!("expected an error from: {src}"), + Err(e) => e.to_string(), + } +} + +fn screen_text(state: &EditorState, buffer: pmacs::buffer::BufferId) -> String { + let manager = state.terminal_manager.borrow(); + let Some(snapshot) = manager.snapshot(buffer) else { + return String::new(); + }; + let mut text = String::new(); + for cell in &snapshot.cells { + match &cell.glyph { + Glyph::Char(c) => text.push(*c), + Glyph::Cluster(b) => text.push_str(&String::from_utf8_lossy(b)), + Glyph::Continuation => {} + } + } + text +} + +fn tick_until(state: &mut EditorState, needle: &str, buffer: pmacs::buffer::BufferId) -> bool { + let deadline = Instant::now() + Duration::from_secs(5); + loop { + state.tick_processes(); + if screen_text(state, buffer).contains(needle) { + return true; + } + if Instant::now() >= deadline { + return false; + } + thread::sleep(Duration::from_millis(20)); + } +} + +/// Give LOCAL a window on `buffer` and register/claim its terminal view, +/// which is what makes `dispatch_key`'s terminal arm reachable. +fn focus_terminal(state: &EditorState, buffer: pmacs::buffer::BufferId) -> WindowId { + state.core.borrow_mut().switch_active_buffer(buffer).ok(); + let window = state.core.borrow().active_window_id(); + let key = TerminalViewKey::new(FrontendId::LOCAL, window, buffer); + let mut manager = state.terminal_manager.borrow_mut(); + manager.register_view(key); + manager.claim_controller(key); + let _ = manager.snapshot_for_view(key, CellSize::new(10, 40)); + window +} + +fn terminal_buffers(state: &EditorState) -> Vec { + let manager = state.terminal_manager.borrow(); + state + .core + .borrow() + .registry + .borrow() + .ids() + .iter() + .copied() + .filter(|id| manager.is_terminal(*id)) + .collect() +} + +/// Open a terminal from Lua and return the identity buffer it created. +/// +/// The id is derived by diffing the manager's terminal set rather than +/// returned through Lua: `BufferIdLua` exposes no id accessor, and +/// diffing also asserts in passing that exactly one terminal appeared. +fn open_cat_terminal(state: &EditorState, lua_spec: &str) -> pmacs::buffer::BufferId { + let before = terminal_buffers(state); + exec( + state, + &format!("TERM_BUF = pmacs.terminal.open {{ {lua_spec} }}"), + ); + let after = terminal_buffers(state); + let mut fresh: Vec<_> = after + .into_iter() + .filter(|id| !before.contains(id)) + .collect(); + assert_eq!(fresh.len(), 1, "exactly one terminal must have opened"); + fresh.remove(0) +} + +/// `cat -v` is the echo instrument, deliberately: the terminal screen +/// rejects C0/C1 controls before they enter cells (Vterm Stage 1 +/// criterion 2), so a raw echoed `Ctrl-X` would be invisible and a test +/// probing for it could never pass. `-v` renders it as the printable +/// two-character `^X`, which is what makes "the configured chord reached +/// the child" observable at all. +const CAT_PROFILE: &str = r#" +pmacs.terminal.profiles.echo = { + command = "/bin/sh", + args = { "-c", "printf 'READY\r\n'; exec cat -v" }, +} +"#; + +/// Did the last key ARM the terminal escape? +/// +/// Observed behaviorally rather than through an accessor: while the +/// escape is armed the next key goes to ordinary dispatch, so it never +/// reaches the child. `cat` echoes anything that does reach it, which +/// makes "the probe character did not appear" the exact observable for +/// "that chord was consumed as the escape". +fn escape_was_armed(state: &mut EditorState, buffer: pmacs::buffer::BufferId, probe: char) -> bool { + // Count occurrences rather than testing for presence: the screen + // already holds the child's own output, and a single-character probe + // like 'R' collides with the "READY" banner. Only an INCREASE proves + // this keystroke reached the child. + let before = screen_text(state, buffer).matches(probe).count(); + state.dispatch_key( + FrontendId::LOCAL, + KeyEvent::new(KeyCode::Char(probe), KeyModifiers::NONE), + ); + let deadline = Instant::now() + Duration::from_secs(2); + loop { + state.tick_processes(); + if screen_text(state, buffer).matches(probe).count() > before { + return false; + } + if Instant::now() >= deadline { + return true; + } + thread::sleep(Duration::from_millis(20)); + } +} + +/// Acceptance 1: a profile spec is strict, and rejects before anything spawns. +#[test] +fn acc1_profile_specs_are_strict_and_reject_before_spawning() { + let state = EditorState::new(); + let before = state.core.borrow().registry.borrow().ids().len(); + + exec( + &state, + r#"pmacs.terminal.profiles.bad = { command = "/bin/sh", nonsense = true }"#, + ); + let err = eval_err(&state, r#"return pmacs.terminal.open { profile = "bad" }"#); + assert!( + err.contains("unknown field") && err.contains("nonsense"), + "the error must name the offending field: {err}" + ); + + exec(&state, "pmacs.terminal.profiles.wrong = { command = 42 }"); + let err = eval_err( + &state, + r#"return pmacs.terminal.open { profile = "wrong" }"#, + ); + assert!(err.contains("must be a string"), "typed field error: {err}"); + + assert_eq!( + state.core.borrow().registry.borrow().ids().len(), + before, + "a rejected profile must create no buffer" + ); + assert_eq!(state.terminal_manager.borrow().len(), 0); +} + +/// Acceptance 2: an unknown profile names the known ones and creates nothing. +#[test] +fn acc2_unknown_profile_lists_known_names_and_creates_nothing() { + let state = EditorState::new(); + exec(&state, CAT_PROFILE); + exec( + &state, + r#"pmacs.terminal.profiles.other = { command = "/bin/sh" }"#, + ); + let before = state.core.borrow().registry.borrow().ids().len(); + + // Via the default setting. + exec( + &state, + r#"pmacs.config.set("terminal.default-profile", "ghost")"#, + ); + let err = eval_err(&state, "return pmacs.terminal.open {}"); + assert!(err.contains("ghost"), "names the missing profile: {err}"); + assert!( + err.contains("echo") && err.contains("other"), + "must LIST the known profiles: {err}" + ); + + // An explicit bad profile fails even though the default is now valid — + // a typo must not silently fall back (Q#TC3a). + exec( + &state, + r#"pmacs.config.set("terminal.default-profile", "echo")"#, + ); + let err = eval_err(&state, r#"return pmacs.terminal.open { profile = "typo" }"#); + assert!(err.contains("typo"), "explicit bad profile errors: {err}"); + + assert_eq!( + state.core.borrow().registry.borrow().ids().len(), + before, + "no buffer, session, or process is created" + ); + assert_eq!(state.terminal_manager.borrow().len(), 0); +} + +/// Acceptance 3: explicit beats profile beats setting beats `$SHELL`, and +/// `env` MERGES rather than replacing. +#[test] +fn acc3_field_resolution_order_and_env_merge() { + let mut state = EditorState::new(); + exec( + &state, + r#" + pmacs.terminal.profiles.merged = { + command = "/bin/sh", + args = { "-c", "printf 'PROFILE:%s:%s\r\n' \"$FROM_PROFILE\" \"$SHARED\"; exec cat" }, + env = { FROM_PROFILE = "p", SHARED = "profile" }, + } + "#, + ); + let buffer = open_cat_terminal( + &state, + r#"profile = "merged", env = { SHARED = "explicit" }"#, + ); + assert!( + tick_until(&mut state, "PROFILE:p:explicit", buffer), + "profile env survives and explicit env overrides the same key: {:?}", + screen_text(&state, buffer) + ); + state.process_supervisor.borrow_mut().shutdown(); +} + +/// Acceptance 3 (explicit command wins) and 4 (`""` means no profile). +#[test] +fn acc3_acc4_explicit_command_wins_and_empty_default_means_no_profile() { + let mut state = EditorState::new(); + exec(&state, CAT_PROFILE); + exec( + &state, + r#"pmacs.config.set("terminal.default-profile", "echo")"#, + ); + + // Explicit command beats the profile's. + let explicit = open_cat_terminal( + &state, + r#"command = "/bin/sh", args = { "-c", "printf 'EXPLICIT\r\n'; exec cat" }"#, + ); + assert!(tick_until(&mut state, "EXPLICIT", explicit)); + + // `""` is the no-profile sentinel: falls through to $SHELL. + exec( + &state, + r#"pmacs.config.set("terminal.default-profile", "")"#, + ); + let bare = open_cat_terminal(&state, ""); + let spec_ok = state.terminal_manager.borrow().is_terminal(bare); + assert!(spec_ok, "an empty default must open a $SHELL terminal"); + assert!( + !screen_text(&state, bare).contains("READY"), + "the echo profile must NOT have been applied" + ); + state.process_supervisor.borrow_mut().shutdown(); +} + +/// Acceptance 5: scrollback resolves from the setting, is overridden by an +/// explicit value, and `0` is legal. +#[test] +fn acc5_scrollback_setting_override_and_bounds() { + let state = EditorState::new(); + exec(&state, r#"pmacs.config.set("terminal.scrollback-rows", 0)"#); + assert_eq!( + state + .lua_host + .lua() + .load(r#"return pmacs.config.get("terminal.scrollback-rows")"#) + .eval::() + .unwrap(), + 0, + "0 is a legal scrollback value meaning 'retain no history'" + ); + + let err = eval_err( + &state, + r#"return pmacs.config.set("terminal.scrollback-rows", -1)"#, + ); + assert!( + err.contains("-1") || err.contains("min"), + "below range: {err}" + ); + let err = eval_err( + &state, + r#"return pmacs.config.set("terminal.scrollback-rows", 4000001)"#, + ); + assert!( + err.contains("4000001") || err.contains("max"), + "above range: {err}" + ); +} + +/// Acceptance 6 and 9: the configured chord escapes, repeating it sends +/// THAT chord to the child, and an ordinary `C-c` still reaches the child. +#[test] +fn acc6_acc9_configured_escape_chord_and_literal_repeat() { + let mut state = EditorState::new(); + exec(&state, CAT_PROFILE); + let buffer = open_cat_terminal(&state, r#"profile = "echo""#); + assert!(tick_until(&mut state, "READY", buffer)); + focus_terminal(&state, buffer); + + exec(&state, r#"pmacs.config.set("terminal.escape-key", "C-x")"#); + + // `C-x C-x` must send Ctrl-X (0x18), which `cat` echoes back. Against + // the pre-change hardcoded `&[0x03]` this sends Ctrl-C instead. + state.dispatch_key( + FrontendId::LOCAL, + KeyEvent::new(KeyCode::Char('x'), KeyModifiers::CONTROL), + ); + state.dispatch_key( + FrontendId::LOCAL, + KeyEvent::new(KeyCode::Char('x'), KeyModifiers::CONTROL), + ); + assert!( + tick_until(&mut state, "^X", buffer), + "C-x C-x must send literal Ctrl-X: {:?}", + screen_text(&state, buffer) + ); + + // With the escape moved, an ordinary C-c is just another key. + state.dispatch_key( + FrontendId::LOCAL, + KeyEvent::new(KeyCode::Char('c'), KeyModifiers::CONTROL), + ); + assert!( + tick_until(&mut state, "^C", buffer), + "plain C-c must reach the child once the escape moved: {:?}", + screen_text(&state, buffer) + ); + state.process_supervisor.borrow_mut().shutdown(); +} + +/// Acceptance 7, 8 and 8a: per-terminal escape resolution, an A→B→A parse +/// count that does not grow, and a cache that dies with its terminal. +#[test] +fn acc7_acc8_acc8a_per_terminal_escape_cache_identity_and_lifecycle() { + let mut state = EditorState::new(); + exec(&state, CAT_PROFILE); + let a = open_cat_terminal(&state, r#"profile = "echo""#); + exec(&state, "TERM_A = TERM_BUF"); + let b = open_cat_terminal(&state, r#"profile = "echo""#); + exec(&state, "TERM_B = TERM_BUF"); + assert!(tick_until(&mut state, "READY", a)); + assert!(tick_until(&mut state, "READY", b)); + + // Different buffer-local escapes, then NO further writes. + exec( + &state, + r#"pmacs.config.set_local(TERM_A, "terminal.escape-key", "C-x")"#, + ); + exec( + &state, + r#"pmacs.config.set_local(TERM_B, "terminal.escape-key", "C-b")"#, + ); + + // Prime both caches. Each priming press ARMS the escape, so it is + // consumed with a probe — otherwise the next chord would be read as + // the escape repeat rather than a fresh escape. + focus_terminal(&state, a); + state.dispatch_key( + FrontendId::LOCAL, + KeyEvent::new(KeyCode::Char('x'), KeyModifiers::CONTROL), + ); + assert!(escape_was_armed(&mut state, a, 'M'), "A primes on its C-x"); + focus_terminal(&state, b); + state.dispatch_key( + FrontendId::LOCAL, + KeyEvent::new(KeyCode::Char('b'), KeyModifiers::CONTROL), + ); + assert!(escape_was_armed(&mut state, b, 'N'), "B primes on its C-b"); + let primed = state.terminal_manager.borrow().escape_parses(); + + // Acceptance 7 — BOTH directions. Asserting only that A still works + // after A->B->A is not enough: an epoch-only cache hands whichever + // entry it finds to every terminal, so A keeps working by accident + // while B silently inherits A's chord. The discriminating assertion + // is that EACH terminal honors its OWN chord and NOT the other's. + focus_terminal(&state, b); + state.dispatch_key( + FrontendId::LOCAL, + KeyEvent::new(KeyCode::Char('b'), KeyModifiers::CONTROL), + ); + assert!( + escape_was_armed(&mut state, b, 'R'), + "terminal B must escape on its own C-b" + ); + // ...and A's chord must be ordinary input in B, not an escape. + focus_terminal(&state, b); + state.dispatch_key( + FrontendId::LOCAL, + KeyEvent::new(KeyCode::Char('x'), KeyModifiers::CONTROL), + ); + assert!( + !escape_was_armed(&mut state, b, 'S'), + "terminal A's C-x must NOT escape terminal B" + ); + + focus_terminal(&state, a); + state.dispatch_key( + FrontendId::LOCAL, + KeyEvent::new(KeyCode::Char('x'), KeyModifiers::CONTROL), + ); + assert!( + escape_was_armed(&mut state, a, 'Q'), + "terminal A must still escape on its own C-x after A->B->A" + ); + + // Acceptance 8: that round trip parsed nothing new. A single + // last-entry cache would have reparsed twice. + assert_eq!( + state.terminal_manager.borrow().escape_parses(), + primed, + "A->B->A with no setting written must not reparse" + ); + + // Acceptance 8a: the cache dies with its terminal. + let sessions_before = state.terminal_manager.borrow().len(); + exec(&state, "pmacs.terminal.terminate(TERM_A)"); + exec(&state, "pmacs.buffer.kill(TERM_A)"); + // Pruning is tick-driven (the manager reaps on the process tick), so + // the session outlives the kill call by design. + let deadline = Instant::now() + Duration::from_secs(5); + while state.terminal_manager.borrow().len() >= sessions_before { + state.tick_processes(); + assert!( + Instant::now() < deadline, + "killing the terminal must remove its session, and with it the cache" + ); + thread::sleep(Duration::from_millis(20)); + } + state.process_supervisor.borrow_mut().shutdown(); +} + +/// Acceptance 10 and 10a: an unparseable value falls back, reports through +/// the status line, and reports once per terminal per effective bad value. +#[test] +fn acc10_acc10a_invalid_escape_falls_back_and_reports_once() { + let mut state = EditorState::new(); + exec(&state, CAT_PROFILE); + let buffer = open_cat_terminal(&state, r#"profile = "echo""#); + assert!(tick_until(&mut state, "READY", buffer)); + focus_terminal(&state, buffer); + + exec( + &state, + r#"pmacs.config.set("terminal.escape-key", "not-a-chord")"#, + ); + state.core.borrow_mut().status.clear(); + + // Acceptance 10: falls back to C-c, so the terminal stays escapable. + state.dispatch_key( + FrontendId::LOCAL, + KeyEvent::new(KeyCode::Char('c'), KeyModifiers::CONTROL), + ); + // Read the report BEFORE probing: `status` is a single slot, and the + // probe key's own rejected self-insert would overwrite it. + let reported = state.core.borrow().status.clone(); + assert!( + reported.contains("terminal.escape-key") && reported.contains("not-a-chord"), + "the report must name the setting and the bad value: {reported:?}" + ); + assert!( + escape_was_armed(&mut state, buffer, 'Q'), + "an invalid escape-key must fall back to C-c, not leave the \ + terminal unescapable" + ); + + // Acceptance 10a: the same bad value does not report again. + state.core.borrow_mut().status.clear(); + state.dispatch_key( + FrontendId::LOCAL, + KeyEvent::new(KeyCode::Char('c'), KeyModifiers::CONTROL), + ); + assert!( + state.core.borrow().status.is_empty(), + "an unchanged invalid value must not re-report: {:?}", + state.core.borrow().status + ); + let _ = escape_was_armed(&mut state, buffer, 'W'); + + // A DIFFERENT bad value is new information, so it reports again. + exec( + &state, + r#"pmacs.config.set("terminal.escape-key", "also-bad")"#, + ); + state.core.borrow_mut().status.clear(); + state.dispatch_key( + FrontendId::LOCAL, + KeyEvent::new(KeyCode::Char('c'), KeyModifiers::CONTROL), + ); + assert!( + state.core.borrow().status.contains("also-bad"), + "a different invalid value must report: {:?}", + state.core.borrow().status + ); + state.process_supervisor.borrow_mut().shutdown(); +} + +/// Acceptance 11: the opening binding exists, resolves to the command, and +/// shadowed nothing (`keymap.bind` is strict, so loading the runtime at all +/// proves the second half). +#[test] +fn acc11_terminal_opening_binding_is_bound_and_shadowed_nothing() { + let state = EditorState::new(); + let command: Option = state + .lua_host + .lua() + .load(r#"local d = pmacs.describe.key("C-c t"); return d and d.command"#) + .eval() + .expect("describe.key"); + assert_eq!( + command.as_deref(), + Some("terminal"), + "C-c t must open a terminal" + ); +} + +/// Acceptance 12: with no settings written and no profiles registered, the +/// defaults reproduce the pre-arc behavior. +#[test] +fn acc12_defaults_reproduce_prior_behavior() { + let state = EditorState::new(); + let lua = state.lua_host.lua(); + assert_eq!( + lua.load(r#"return pmacs.config.get("terminal.default-profile")"#) + .eval::() + .unwrap(), + "" + ); + assert_eq!( + lua.load(r#"return pmacs.config.get("terminal.scrollback-rows")"#) + .eval::() + .unwrap(), + 10_000 + ); + assert_eq!( + lua.load(r#"return pmacs.config.get("terminal.escape-key")"#) + .eval::() + .unwrap(), + "C-c" + ); + assert!( + lua.load("return next(pmacs.terminal.profiles) == nil") + .eval::() + .unwrap(), + "no profiles are registered by default" + ); +} From 04c5ad13e84e4a0e68329c82b0bd50006224d285 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 18:41:21 -0400 Subject: [PATCH 14/32] docs: record the terminal-config lane Stage 1 of the terminal config/copy-mode arc is in review; Stage 2 is not started. Records the four decisions forced by scouted ground truth, the four bites against four different wrong implementations, the two reusable test instruments, and the gate results. --- docs/active-work.md | 46 +++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 46 insertions(+) diff --git a/docs/active-work.md b/docs/active-work.md index 9a913e6..5284b94 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -517,6 +517,52 @@ If it does not, stop and repair the remote/fetch configuration. **isolated-config workspace sweep 3,177 across 92 suites, zero failures**; `git diff --check` clean. Gates were run against the committed tree. +## Terminal config + copy mode arc — Stage 1 IN REVIEW + +- Approved framing: `docs/terminal-config-and-copy-mode-framing.md` + **revision 4** (four review rounds), committed as the first commit of + Stage 1's branch. Two stages, two branches, two PRs; **no protocol + change**. +- **Stage 1 = `githubsucks/terminal-config`**, worktree + `../pmacs-terminal-config`, based on `githubsucks/main` @ `d152120`. + Profiles, scrollback, escape key, and the `C-c t` opening binding. +- **Stage 2 = `terminal-copy-mode`, not started.** Branch it off `main` + after Stage 1 merges: no dependency, but both edit + `builtin/runtime/terminal.lua`. +- Load-bearing decisions, each forced by scouted ground truth: + - profiles are a **raw Lua table** — `ConfigValue` is four scalars with + no table kind, so they join `pmacs.lsp.config` / `pmacs.pair.sets`; + - the **two open-time settings resolve through the global chain**, + because they are read before the identity buffer exists; only + `terminal.escape-key` resolves per buffer; + - the escape cache lives on **`TerminalSession`** so its lifetime is + the terminal's. `value_epoch` alone is not a sufficient key: it does + not advance when focus moves between terminals with different + buffer-local values; + - repeating the escape sends **that chord**, not a hardcoded `0x03`. +- **Four bites, each against a different plausible wrong + implementation** — hardcoded ETX fails acc6/9; epoch-only cache key + fails acc7; single last-entry cache fails acc8's parse count; removing + the invalid-value fallback fails acc10. The first version of acc7 + passed against the epoch-only bite because it asserted only that + terminal A still worked; the discriminating assertion is that **each** + terminal honors its own chord and not the other's. +- Test instruments worth reusing: `cat -v` is the echo probe, because the + screen rejects C0 controls before they reach cells so a raw echoed + `Ctrl-X` is invisible; and the probe **counts occurrences** rather than + testing presence, because a single-character probe collides with the + child's own banner text. +- Verification on this branch (against the committed tree): `cargo fmt + --check` clean; strict workspace Clippy clean; 1,832 default + 2,009 + CRDT library tests; `terminal_config_acceptance` 10/10 in **both** + configurations; vterm Stage 1/2/3 9+10 / 6+6 / 5+9; config registry 16; + bottom-panel 46; listview 6; compile 67 (isolated config); M4 121; + required GPU 202; **isolated-config workspace sweep 3,262 across 94 + suites**; `git diff --check` clean. + - `compile_mode_acceptance` fails 11/67 against the **real** user + config and passes 67/67 with an isolated `XDG_CONFIG_HOME` — the + known pre-existing trap, not this branch. + ## Bottom-panel lane (Arc 7) — Stage 1 MERGED; Stage 2 (GPU band) is next Stage 1 is on `main`; nothing in this arc is in flight. Stage 2 has **no From 73587b0e371f49c26c87926941fe7518cce54004 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 19:23:50 -0400 Subject: [PATCH 15/32] fix(lean): correlate the probe verdict with its own server and buffer MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Round 3 review: two P1 asynchronous-correlation defects, with the focused suite at 25/25 while both were live. **1. A late version verdict retired nothing and claimed success.** `probe.watching` is cleared the moment the server initializes — it is failure-polling state. A slow `lake --version` landing after a successful initialize therefore reached `fire_latch(nil)`, which retires nothing: `_attach_buffer` found the still-live primary attachment, early-returned it, and the retry counted that as done. Status said "falling back", the config named the fallback, and the buffer stayed on the old server. **That is the round-1 silent no-op arriving through a third event ordering** — first as "no re-attach at all", then as "re-attach cleared by an unrelated buffer", now as "re-attach satisfied by the server we were supposed to replace". The fix separates the two facts that were being carried by one field: `probe.primary` is the server the verdict applies to and survives initialization; `probe.watching` is the failure poll and is cleared by it. The existing fixture could not reach this ordering at all — its `serve` sleeps, so the primary can never initialize before `--version` returns. The new one execs the fake LSP for `serve` and delays 0.6s before reporting 3.0.0. **2. `buf_key` was the most recently loaded Lean buffer.** Written on every Lean `buffer.after-load`, so a second Lean file opened before the verdict became the rebuild target while the latch still watched the FIRST buffer's server. Target buffer and primary server are one fact and are now armed together, exactly once. Both files in the new test share a package, so mis-targeting shows up as a stranded buffer rather than as two unrelated servers. **3. The failure message hardcoded `lake serve`** after the latch became command-agnostic, telling a user whose `my-lean-wrapper` failed to go debug lake. `configured_command()` names what is actually configured, arguments included. **4. The ledger** now records all fifteen bites across the three rounds, both prior review rounds' findings (the round-2 block was lost when an earlier edit script aborted before writing), and the durable lesson. That lesson, recorded for the handoff: **six tests across three rounds were written, ran green, and pinned nothing** — caught only by biting. The shapes are enumerated in the ledger; the rule is that a test is not evidence until the mutation it targets has been shown to fail it. Two of the six are subtle enough to be worth naming here: a bite that RAISES is swallowed by the hook's pcall and "passes" for the wrong reason, and a fixture whose `serve` sleeps cannot reach any ordering where the primary comes up first. --- builtin/runtime/lean.lua | 66 +++++++++--- docs/active-work.md | 80 +++++++++++++-- tests/lean4_server_acceptance.rs | 171 +++++++++++++++++++++++++++++++ 3 files changed, 294 insertions(+), 23 deletions(-) diff --git a/builtin/runtime/lean.lua b/builtin/runtime/lean.lua index 5f74792..8f3b339 100644 --- a/builtin/runtime/lean.lua +++ b/builtin/runtime/lean.lua @@ -127,10 +127,28 @@ local probe = { proc = nil, -- process id of the running probe out = "", -- accumulated probe stdout buf_key = nil, -- tostring() of the buffer that started this - watching = nil, -- sid we are waiting to see fail before initialize + watching = nil, -- sid still being polled for die-before-initialize + primary = nil, -- sid the probe's verdict applies to; NOT cleared + -- when the server initializes, because a late + -- version verdict still has to retire it + armed = false, -- the target buffer + primary have been captured saw_initialized = false, } +-- The command as configured, for status text. Hardcoding "lake serve" +-- was untruthful the moment the failure latch became command-agnostic: +-- a user whose `my-lean-wrapper` failed was told `lake serve` did. +local function configured_command() + local cfg = pmacs.lsp.config.lean4 + local cmd = cfg and cfg.command + if not cmd then return "the Lean server" end + local args = cfg.args or {} + if #args > 0 then + return "`" .. tostring(cmd) .. " " .. table.concat(args, " ") .. "`" + end + return "`" .. tostring(cmd) .. "`" +end + local function report(msg) -- COHERENCE §1.2: background work must leave an attributed trace. -- `pmacs.editor.set_status` is the channel that EXISTS; `pmacs.error` @@ -327,9 +345,19 @@ local function drain_probe() -- case, and covers it better. The probe answers only the ONE -- question failure detection would otherwise answer slowly: an -- old-but-working lake that starts a useless server. + -- **`probe.primary`, NOT `probe.watching`.** `watching` is + -- failure-polling state and is cleared the moment the server + -- initializes. A slow `--version` that lands after a successful + -- initialize would then arrive with nil, and `fire_latch(nil)` + -- retires nothing: `_attach_buffer` finds the still-live primary + -- attachment, early-returns it, and the retry calls that success. + -- Status and config would say "fell back" while the buffer stayed + -- on the old server — the same silent no-op as round 1, reached + -- through a different event ordering. Initializing must stop the + -- failure poll, not erase the server the verdict has to retire. if ev.kind == "exited" and ev.code == 0 and version_below_3_1(probe.out) then - fire_latch(probe.watching, "lake is older than 3.1.0") + fire_latch(probe.primary, "lake is older than 3.1.0") end end end @@ -393,18 +421,20 @@ local function poll_latch() if tostring(info.id) == skey then local kind = info.state and info.state.kind if kind == "initialized" then + -- Stop polling for failure; `probe.primary` deliberately + -- survives, because a later version verdict still needs it. probe.saw_initialized = true probe.watching = nil return end if kind == "crashed" or kind == "stopped" then - fire_latch(sid, "`lake serve` failed to start") + fire_latch(sid, configured_command() .. " failed to start") end return end end -- Gone from the manager entirely without ever initializing. - fire_latch(nil, "`lake serve` failed to start") + fire_latch(nil, configured_command() .. " failed to start") end -- Q#LN16 — `textDocument/waitForDiagnostics` -------------------------- @@ -488,11 +518,6 @@ pmacs.hook.add("buffer.after-load", function() local ok_lang, lang = pcall(pmacs.lsp.buffer_language, buf) if not ok_lang or lang ~= "lean4" then return end - -- The buffer that started this, remembered for the asynchronous - -- rebuild: `_attach_buffer` acts on whatever is active when the - -- verdict lands, which may be a different buffer entirely. - probe.buf_key = tostring(buf) - if not probe.started then local path = pmacs.editor.file_path() start_probe(path and M.root_for(path) or nil) @@ -500,9 +525,18 @@ pmacs.hook.add("buffer.after-load", function() local rec = pmacs.lsp.active_attachment() if rec and rec.language == "lean4" then - -- Watch only the FIRST Lean server: the latch is per session. - if not probe.latched and not probe.saw_initialized - and probe.watching == nil then + -- **Arm ONCE, capturing buffer and server together.** Setting + -- `buf_key` on every Lean load meant a second Lean buffer opened + -- before the verdict silently became the rebuild target while the + -- latch still watched the FIRST buffer's server — so the rebuild + -- either repaired the wrong buffer or accepted the second buffer's + -- unrelated live server as success, stranding the first. The pair + -- (target buffer, primary server) is one fact and is captured as + -- one. + if not probe.armed and not probe.latched and not probe.saw_initialized then + probe.armed = true + probe.buf_key = tostring(buf) + probe.primary = rec.server probe.watching = rec.server end return @@ -522,7 +556,13 @@ pmacs.hook.add("buffer.after-load", function() -- already swallowed upstream. That is not something to wait for; it -- is the failure itself, and the only place it is still observable. if not probe.latched then - fire_latch(nil, "`" .. tostring(cfg.command) .. "` could not be started") + -- No server was ever created, so there is no primary to retire — + -- but the rebuild still needs a target buffer. + if not probe.armed then + probe.armed = true + probe.buf_key = tostring(buf) + end + fire_latch(nil, configured_command() .. " could not be started") end end) diff --git a/docs/active-work.md b/docs/active-work.md index 4a650da..6cdac4f 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -265,7 +265,7 @@ If it does not, stop and repair the remote/fetch configuration. - Ships `builtin/runtime/lean.lua` (new), one `include_str!` line in `src/editor.rs`, `pmacs.lsp._attach_buffer` exported from `lsp.lua`, a `leanprogress` mode plus `waitForDiagnostics` validation on - `pmacs_fake_lsp`, and `tests/lean4_server_acceptance.rs` (25 tests). + `pmacs_fake_lsp`, and `tests/lean4_server_acceptance.rs` (28 tests). No protocol change. - **Stage 1's acceptance 12 is half superseded and was rewritten, not deleted.** It asserted `pmacs.lsp.config.lean4 == nil` to catch a @@ -279,12 +279,70 @@ If it does not, stop and repair the remote/fetch configuration. an EMPTY `lean-toolchain` (a legitimate marker — existence semantics, not content). Discriminator is `read`'s SECOND return; decline only on a non-nil err. Probed on LuaJIT 2.1. -- Seven bites recorded, each against the committed tree: bare `io.open` - → 24a fails / 24b passes; require-non-nil → 24b fails / 24a passes; - no canonicalization → symlinked open spawns two servers; no re-attach - after the swap → three latch tests fail; hook keyed on the attachment - → the missing-`lake` case fails; `waitForDiagnostics` without - `version` → acc37 fails with the server's InvalidParams. +- **Fifteen bites recorded, each against the committed tree.** R1: bare + `io.open` → 24a fails / 24b passes; require-non-nil → 24b fails / 24a + passes; no canonicalization → symlinked open spawns two servers; no + re-attach after the swap → three latch tests fail; hook keyed on the + attachment → the missing-`lake` case fails; `waitForDiagnostics` + without `version` → acc37 fails with InvalidParams. R2: skip retiring + a terminal server → `attempt` reaches 3; no originating-buffer gate → + the Lean buffer is left on the `lake` stub; retry-forever → the + failing-fallback test fails; version-probe any command → the + working-wrapper test fails; no disabled guard → the unconfigured test + sees "`nil` could not be started". R3: verdict keyed on `watching` → + the late-verdict test finds the buffer still on `lake`; `buf_key` + rewritten per load → the second-buffer test fails; hardcoded + `lake serve` → the wrapper-naming test fails. +- **Round-2 review: three more P1 lifecycle defects, suite 20/20 with + all of them live.** (1) The crashed primary respawned forever — + skipping the retire call avoided corrupting terminal servers but left + `next_restart_at` armed. **`forget` is the call for a TERMINAL server** + (it requires terminal state and removes the client, dropping the + restart timer); `stop` is for a live one and corrupts a terminal one. + (2) Re-attachment targeted whatever buffer was active when the async + verdict landed; an unrelated Rust attachment satisfied "a different + server id". (3) A failing fallback retried every tick forever, silent. + Plus two P2s: the Lake version parser was applied to arbitrary wrapper + output, and an UNCONFIGURED `config.lean4` was reported as failure and + latched, poisoning the session. +- **Round-3 review: two more P1s, both asynchronous correlation, suite + 25/25.** (a) `probe.watching` is cleared when the server initializes, + so a SLOW version verdict arrived with nil and retired nothing — + `_attach_buffer` returned the still-live primary and the retry called + it success, so status and config said "fell back" while the buffer + stayed put. **That is the round-1 silent no-op reached through a third + event ordering.** `probe.primary` is now separate from + `probe.watching` and survives initialization. (b) `buf_key` was + rewritten on every Lean `after-load`, so a second Lean buffer opened + before the verdict became the rebuild target while the latch still + watched the first buffer's server. Target buffer and primary server + are one fact and are now armed together, once. Plus a P2: the failure + message hardcoded `lake serve` after the latch became + command-agnostic, sending wrapper users to debug the wrong binary. +- **DURABLE LESSON — "the test that passes" vs "the test that + discriminates."** Six tests across three rounds were written, run + green, and only bite-testing showed they pinned nothing. **Carry this + to `docs/agent-handoff.md` when the lane lands.** The concrete shapes, + all from this branch: + 1. R1 acceptance 36 asserted "every server is terminal" — pinning the + ABSENCE of the fallback it claimed to test. + 2. "No live non-fallback server" misses a respawn loop: a respawning + server sits in `crashed` most of the time. `attempt` counts + respawns; liveness does not. + 3. Returning to a buffer via `find_or_open` re-fires + `buffer.after-load`, which repairs the attachment regardless of the + code under test. Use `switch_buffer`. + 4. A MISSING executable fails synchronously inside `after-load`, where + the rebuild happens inline — no async race can occur. Only the + probe path exercises asynchronous ordering. + 5. A mutation that RAISES (indexing a nil config) is swallowed by the + hook's pcall, so the bite "passes" for the wrong reason. A bite must + reproduce the original shape, not merely break the code. + 6. A fixture whose `serve` sleeps can never let the primary initialize + first, so it cannot reach the ordering where a late verdict must + retire a LIVE server. + Rule: **a test is not evidence until the mutation it targets has been + shown to fail it.** - **SUBSTRATE BUG FOUND, not fixed here (framing §6).** `LspManager::stop` on an ALREADY-terminal server takes its not-initialized branch, terminates the dead process and sets @@ -294,7 +352,9 @@ If it does not, stop and repair the remote/fetch configuration. `ShuttingDown` **forever**: `server_is_live` reads it as LIVE, so `attach_buffer` never rebuilds, and `forget` refuses it for not being terminal. **Stopping a dead server is what makes it un-replaceable.** - Lean works around it by checking the state before stopping. + Lean works around it by dispatching on state: `forget` when + terminal, `stop` when live. Merely SKIPPING the call is not + enough — that leaves `next_restart_at` armed. - Round-1 review found four P1s, all real: the latch swapped the config but never spawned or re-attached (and acc36 *asserted every server was terminal*, pinning the absence of the fallback); a missing `lake` @@ -308,9 +368,9 @@ If it does not, stop and repair the remote/fetch configuration. server-failure latch covers the rest. - Verification on this branch: `cargo fmt --check` clean; strict workspace Clippy clean; 1,826 default + 2,003 CRDT library tests; - lean4 server 25/25; lean4 stage 1 9/9; dispatch seams 15/15; + lean4 server 28/28; lean4 stage 1 9/9; dispatch seams 15/15; multi-root 13/13; M4 121; required GPU 155; **isolated-config - workspace sweep 3,214 across 94 suites, zero failures**; + workspace sweep 3,217 across 94 suites, zero failures**; `git diff --check` clean. (Round 1 of this entry recorded 17/17 and 3,206 — the PRE-fix counts — after the fixes were pushed. The ledger's protocol is that verification diff --git a/tests/lean4_server_acceptance.rs b/tests/lean4_server_acceptance.rs index cba8792..db708f1 100644 --- a/tests/lean4_server_acceptance.rs +++ b/tests/lean4_server_acceptance.rs @@ -1069,3 +1069,174 @@ fn r2_an_unconfigured_lean_server_is_disabled_not_failed() { "and the session is not poisoned: a later config must still work" ); } + +// --------------------------------------------------------------------------- +// Round-3 review findings — asynchronous correlation. +// +// Both fail against 3377db0, where the suite was 25/25. +// --------------------------------------------------------------------------- + +impl Fixture { + /// A `lake` whose `serve` really works (it execs the fake LSP) but + /// whose `--version` answers slowly with an old version. This is the + /// ordering the previous fixtures could not produce: the primary + /// INITIALIZES before the version verdict arrives. + fn slow_version_lake(&self, rel: &str, server: &str, version_line: &str) -> PathBuf { + use std::os::unix::fs::PermissionsExt as _; + let path = self.root.join(rel); + std::fs::create_dir_all(path.parent().unwrap()).unwrap(); + std::fs::write( + &path, + format!( + "#!/bin/sh\nif [ \"$1\" = \"--version\" ]; then\n sleep 0.6\n echo '{version_line}'\n exit 0\nfi\nexec '{server}'\n" + ), + ) + .unwrap(); + std::fs::set_permissions(&path, std::fs::Permissions::from_mode(0o755)).unwrap(); + path + } +} + +/// The command backing the active buffer's attached server. +fn attached_command(state: &EditorState) -> String { + eval( + state, + r#" + local rec = pmacs.lsp.active_attachment() + if not rec then return "none" end + for _, s in ipairs(pmacs.lsp.list()) do + if tostring(s.id) == tostring(rec.server) then + return tostring(s.command) + end + end + return "gone" + "#, + ) +} + +#[test] +fn r3_a_late_version_verdict_still_retires_an_initialized_primary() { + // `probe.watching` is cleared the moment the server initializes. A + // verdict arriving after that used to call `fire_latch(nil)`, which + // retires nothing — `_attach_buffer` then returns the still-live + // primary and the retry calls it success. Status and config would + // say "fell back" while the buffer stayed put. + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let lake = fx.slow_version_lake("bin/lake", &fake_lsp_path(), "Lake version 3.0.0"); + let mut state = editor(&fx); + with_fallback(&state, &lake); + + open(&state, &file); + // Let the primary initialize first — the ordering that matters. + tick_for(&mut state, 300); + assert_eq!( + attached_state(&state), + "initialized", + "precondition: the primary really did come up before the verdict" + ); + assert_eq!( + attached_command(&state), + lake.display().to_string(), + "precondition: and the buffer is on it" + ); + + // Now let the slow `--version` land and the fallback complete. + tick_for(&mut state, 1200); + + assert_eq!( + attached_command(&state), + fake_lsp_path(), + "a late version verdict must actually move the buffer to the \ + fallback, not just rewrite the config and claim it did" + ); + // And the retired primary is not left running or respawning. + let stale: i64 = eval( + &state, + r#" + local rec = pmacs.lsp.active_attachment() + local live = rec and tostring(rec.server) or "" + local n = 0 + for _, s in ipairs(pmacs.lsp.list()) do + if tostring(s.id) ~= live then + local k = s.state and s.state.kind + if k ~= "stopped" and k ~= "crashed" then n = n + 1 end + end + end + return n + "#, + ); + assert_eq!(stale, 0, "the initialized primary was retired, not left up"); +} + +#[test] +fn r3_a_second_lean_buffer_does_not_steal_the_rebuild_target() { + // `buf_key` was written on every Lean `buffer.after-load`, so a + // second Lean file opened before the verdict became the rebuild + // target while the latch still watched the FIRST buffer's server. + // + // Both files live in the SAME Lake package, so they share one server + // and one root — which is what makes the mis-targeting observable as + // a stranded buffer rather than as two independent servers. + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let first = fx.write("pkg/A.lean", "def a := 1\n"); + let second = fx.write("pkg/B.lean", "def b := 2\n"); + let lake = fx.lake_stub("bin/lake", "Lake version 3.0.0"); + let mut state = editor(&fx); + with_fallback(&state, &lake); + + open(&state, &first); + exec(&state, "_G.first_buf = pmacs.window.buffer()"); + // A second Lean buffer, opened before the probe's verdict lands. + open(&state, &second); + tick_for(&mut state, 500); + + // The armed target must still be the FIRST buffer. + let target_is_first: bool = eval( + &state, + "return pmacs.lean._probe.buf_key == tostring(_G.first_buf)", + ); + assert!( + target_is_first, + "the rebuild target is captured once, when the latch arms — a \ + later Lean buffer must not silently become the target" + ); + + // And the first buffer really does end up on the fallback. + exec(&state, "pmacs.window.switch_buffer(_G.first_buf)"); + tick_for(&mut state, 600); + assert_eq!( + attached_command(&state), + fake_lsp_path(), + "the originating buffer is the one repaired" + ); +} + +#[test] +fn r3_a_failing_wrapper_is_named_truthfully_not_as_lake_serve() { + // The failure latch is command-agnostic, so its message must be too. + // Telling a user that `lake serve` failed when they configured + // `my-lean-wrapper` sends them to debug the wrong thing. + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let absent = fx.dir("bin/my-lean-wrapper"); + let mut state = editor(&fx); + with_fallback(&state, &absent); + + open(&state, &file); + settle(&mut state); + + let status = state.core.borrow().status.clone(); + assert!( + status.contains("my-lean-wrapper"), + "the status names the command the user actually configured; saw \ + {status:?}" + ); + assert!( + !status.contains("lake serve"), + "and does not attribute the failure to `lake serve`; saw {status:?}" + ); +} From 7c37bdc51455ba974a879340eabae700f1f1eb76 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 19:46:29 -0400 Subject: [PATCH 16/32] fix(lean): repair every buffer and retire every server on fallback MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Round 4 review: one P1, and it is the same defect for the FOURTH time. `pmacs.lsp.config.lean4` is a single global entry, so swapping its command invalidates **every** Lean buffer and **every** Lean server — Q#LN15 gives one server per project root, so there can be several. Rounds 1-3 each repaired one buffer and retired one server, and round 3 shipped "repair the armed target, strand the rest": status and config said fallback while a second open Lean buffer stayed on the retired command, and a second project root's server stayed live. The shape that actually holds: * **Retire ALL `lean4` servers on latch**, not the one the probe happened to name. `probe.primary` identifies the server the VERDICT is about; it was never the set of servers the swap invalidates. * **Repair each buffer lazily and at most once**, when it becomes active — on `buffer.after-switch` and on the tick. `_attach_buffer` is an active-buffer-only seam, so a global swap cannot be applied to every open buffer at once; it has to be applied as they surface. lsp.lua's own `after-switch` re-pushes views but does not rebuild a stale attachment, so nothing else covered this. * The **once-per-buffer bound** is load-bearing: without it a fallback that also fails to spawn would retry every tick forever — the round-2 defect, which a naive global repair loop would reintroduce for every buffer instead of just one. * `shutting-down` is deliberately not treated as stale. It is still live by `server_is_live`'s reckoning, so attaching would early-return the stale record and burn that buffer's single attempt on a no-op. P2: argument-inclusive attribution was implemented in round 3 but pinned only by "contains the command name", so a mutation dropping every argument passed. Now asserted against the exact ` ` string. Also fixed a vacuous assertion this refactor created: a test checked `_probe.reattach_from == nil` for a field that no longer exists, which reads as nil and passes for nothing. It now asserts a positive count of recorded repair attempts. Three bites, each against 73587b0: repair only the armed buffer -> the second buffer stays on `lake`; retire only the named server -> one live stale server remains; drop arguments from attribution -> the exact-string assertion fails. The ledger records a second durable lesson beside the vacuity one: **a scope error repeats until the scope is named.** Four rounds of locally correct fixes, none of which asked what the config swap invalidates. When a change edits shared state, enumerate everything derived from it before repairing anything. --- builtin/runtime/lean.lua | 152 +++++++++++++++++++------------ docs/active-work.md | 33 ++++++- tests/lean4_server_acceptance.rs | 137 +++++++++++++++++++++++++++- 3 files changed, 257 insertions(+), 65 deletions(-) diff --git a/builtin/runtime/lean.lua b/builtin/runtime/lean.lua index 8f3b339..546e123 100644 --- a/builtin/runtime/lean.lua +++ b/builtin/runtime/lean.lua @@ -132,6 +132,7 @@ local probe = { -- when the server initializes, because a late -- version verdict still has to retire it armed = false, -- the target buffer + primary have been captured + repaired = {}, -- buffer key -> repair attempted (at most once) saw_initialized = false, } @@ -149,6 +150,16 @@ local function configured_command() return "`" .. tostring(cmd) .. "`" end +-- The fallback command, for status text. +local function fallback_name() + local args = M._fallback.args or {} + if #args > 0 then + return "`" .. tostring(M._fallback.command) .. " " + .. table.concat(args, " ") .. "`" + end + return "`" .. tostring(M._fallback.command) .. "`" +end + local function report(msg) -- COHERENCE §1.2: background work must leave an attributed trace. -- `pmacs.editor.set_status` is the channel that EXISTS; `pmacs.error` @@ -232,8 +243,6 @@ end -- Retire the failed server, swap the command, then rebuild the -- attachment on the buffer that started this. -local try_reattach - -- Retire `sid` so it cannot come back. **Which call to use depends on -- the state, and using the wrong one is worse than doing nothing:** -- @@ -263,67 +272,85 @@ local function retire_server(sid) end end +-- Retire EVERY Lean server, not just the one that failed. +-- +-- `pmacs.lsp.config.lean4` is a single global entry, so swapping its +-- command invalidates every server spawned from the old one — and +-- Q#LN15 gives one server per project root, so there can be several. +-- Retiring only the server that happened to fail left the others live +-- and every buffer attached to them stranded on a command the config no +-- longer names. +local function retire_all_lean_servers() + local ok, rows = pcall(pmacs.lsp.list) + if not ok or not rows then return end + local ids = {} + for _, info in ipairs(rows) do + if info.language_id == "lean4" then ids[#ids + 1] = info.id end + end + for _, id in ipairs(ids) do retire_server(id) end +end + +-- Rebuild the ACTIVE buffer's attachment if it is Lean and stale. +-- +-- `_attach_buffer` is an active-buffer-only seam, so a global config +-- swap cannot be applied to every open buffer at once. It is applied +-- lazily instead: whenever a Lean buffer becomes the active one, if its +-- record points at a server that is gone or terminal, it is rebuilt. +-- +-- **At most one attempt per buffer.** Without that bound a fallback +-- that also fails to spawn would retry every tick forever with nothing +-- reported — the round-2 defect, which a general repair loop would +-- otherwise reintroduce for every buffer instead of just one. +-- +-- A `shutting-down` server is deliberately NOT treated as stale: it is +-- still live by `server_is_live`'s reckoning, so `attach_buffer` would +-- early-return the stale record and burn this buffer's single attempt +-- on a no-op. Skipping leaves the attempt for a later tick, once the +-- retirement has actually landed. +local function repair_active_if_stale() + if not probe.latched then return end + local buf = pmacs.window.buffer() + if not buf then return end + local key = tostring(buf) + if probe.repaired[key] then return end + local ok_lang, lang = pcall(pmacs.lsp.buffer_language, buf) + if not ok_lang or lang ~= "lean4" then return end + + local rec = pmacs.lsp.active_attachment() + local stale + if not rec then + stale = true + else + local kind = server_state_kind(rec.server) + stale = (kind == nil or kind == "crashed" or kind == "stopped") + end + if not stale then return end + + probe.repaired[key] = true + local ok, fresh = pcall(pmacs.lsp._attach_buffer) + if not ok or not fresh then + report("LSP: lean4 fallback " .. fallback_name() + .. " did not start either") + end +end + local function fire_latch(sid, why) if probe.latched then return end probe.latched = true probe.watching = nil - if sid then retire_server(sid) end if not swap_to_fallback() then report("LSP: lean4 " .. why) + -- Still retire: the servers are broken whether or not a replacement + -- command was installed, and leaving them live would keep the + -- restart machinery running against a command known to fail. + retire_all_lean_servers() return end - report("LSP: lean4 " .. why .. "; falling back to `" - .. tostring(M._fallback.command) .. "`") - -- **Spawn the replacement and re-point the buffer at it.** Swapping - -- the config is not a fallback on its own: nothing re-fires an attach - -- on a config change and `attach_buffer` early-returns for a live - -- attachment, so without this the buffer stays bound to the server we - -- just retired and the user has a config edit and no language server. - -- - -- The rebuild waits for two things, and conflating them is what made - -- round 2 wrong in two ways at once: - -- 1. the retired server actually reaching a terminal state (or - -- being gone) — `stop` leaves `shutting-down`, which - -- `server_is_live` counts as LIVE, so attaching before then - -- early-returns the stale record and the swap silently no-ops; - -- 2. the buffer that started this being the ACTIVE one, because - -- `_attach_buffer` is an active-buffer-only seam. The verdict - -- arrives asynchronously, so the user may well be somewhere else - -- by then — and "some attachment now names a different server" - -- is satisfied by an unrelated Rust buffer, which would clear the - -- retry while leaving the Lean buffer stale forever. - probe.reattach_from = sid and tostring(sid) or false - try_reattach() -end - --- Returns true when there is nothing left to do: either the initiating --- buffer is attached to the replacement, or the replacement itself --- failed and that has been reported. -function try_reattach() - if probe.reattach_from == nil then return true end - -- (2) Wait for the initiating buffer to be the active one. - local buf = pmacs.window.buffer() - if not buf or not probe.buf_key or tostring(buf) ~= probe.buf_key then - return false - end - -- (1) Wait for the retired server to stop counting as live. - if probe.reattach_from then - local kind = server_state_kind_for_key(probe.reattach_from) - if kind ~= nil and kind ~= "crashed" and kind ~= "stopped" then - return false - end - end - -- Both conditions met: attempt the replacement EXACTLY ONCE. Cleared - -- first so a failing fallback cannot retry every tick forever — - -- acceptance 27 promises a second failure surfaces rather than loops. - probe.reattach_from = nil - local ok, rec = pcall(pmacs.lsp._attach_buffer) - if not ok or not rec then - report("LSP: lean4 fallback `" .. tostring(M._fallback.command) - .. "` did not start either") - return false - end - return true + retire_all_lean_servers() + report("LSP: lean4 " .. why .. "; falling back to " .. fallback_name()) + -- Repair what is in front of the user now; everything else is + -- repaired lazily as it becomes active (see `repair_active_if_stale`). + repair_active_if_stale() end local function drain_probe() @@ -566,19 +593,26 @@ pmacs.hook.add("buffer.after-load", function() end end) +-- A buffer switch is the moment a stale Lean buffer becomes visible, so +-- repair immediately rather than waiting for the next tick. lsp.lua's +-- own `after-switch` subscription re-pushes views but does NOT rebuild a +-- stale attachment, so nothing else covers this. +pmacs.hook.add("buffer.after-switch", function() + repair_active_if_stale() +end) + pmacs.hook.add("process.after-tick", function() drain_probe() poll_latch() - -- Keep trying until the stopped server is really gone; see the note in - -- `fire_latch`. - if probe.reattach_from ~= nil then try_reattach() end + -- Repair the active buffer if the latch invalidated it. Cheap when + -- there is nothing to do, and bounded to one attempt per buffer. + repair_active_if_stale() end) -- Test seam: acceptance drives the latch deterministically rather than -- waiting on real process timing. Not part of the public surface. M._probe = probe M._fire_latch = fire_latch -M._try_reattach = try_reattach M._version_below_3_1 = version_below_3_1 pmacs.lean = M diff --git a/docs/active-work.md b/docs/active-work.md index 6cdac4f..e4e1064 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -265,7 +265,7 @@ If it does not, stop and repair the remote/fetch configuration. - Ships `builtin/runtime/lean.lua` (new), one `include_str!` line in `src/editor.rs`, `pmacs.lsp._attach_buffer` exported from `lsp.lua`, a `leanprogress` mode plus `waitForDiagnostics` validation on - `pmacs_fake_lsp`, and `tests/lean4_server_acceptance.rs` (28 tests). + `pmacs_fake_lsp`, and `tests/lean4_server_acceptance.rs` (31 tests). No protocol change. - **Stage 1's acceptance 12 is half superseded and was rewritten, not deleted.** It asserted `pmacs.lsp.config.lean4 == nil` to catch a @@ -319,6 +319,21 @@ If it does not, stop and repair the remote/fetch configuration. are one fact and are now armed together, once. Plus a P2: the failure message hardcoded `lake serve` after the latch became command-agnostic, sending wrapper users to debug the wrong binary. +- **Round-4 review: one P1, and it is the same defect a FOURTH time.** + `pmacs.lsp.config.lean4` is a single global entry, so swapping its + command invalidates **every** Lean buffer and **every** Lean server — + Q#LN15 gives one per project root. Rounds 1–3 each fixed the repair + for one buffer and one server; round 4 is "repair the armed target, + strand the rest". The shape that finally holds: retire ALL `lean4` + servers on latch, and repair each buffer **lazily and at most once** + when it becomes active (`buffer.after-switch` + the tick), because + `_attach_buffer` is active-buffer-only and cannot reach the others. + The per-buffer once-only bound is what stops a failing fallback + retrying forever — the round-2 defect a naive global repair loop would + have reintroduced for every buffer instead of one. Plus a P2: the + argument-inclusive attribution was implemented but pinned only by + "contains the command name", so a mutation dropping every argument + still passed. - **DURABLE LESSON — "the test that passes" vs "the test that discriminates."** Six tests across three rounds were written, run green, and only bite-testing showed they pinned nothing. **Carry this @@ -341,8 +356,20 @@ If it does not, stop and repair the remote/fetch configuration. 6. A fixture whose `serve` sleeps can never let the primary initialize first, so it cannot reach the ordering where a late verdict must retire a LIVE server. + 7. Asserting on a field that no longer exists (`_probe.reattach_from` + after a refactor) reads as nil and passes for nothing. Assert + positive facts — a count, a command string — not absences. Rule: **a test is not evidence until the mutation it targets has been shown to fail it.** +- **SECOND DURABLE LESSON — a scope error repeats until the scope is + named.** The "fallback silently does not happen" defect came back four + times: no re-attach; re-attach cleared by an unrelated buffer; + re-attach satisfied by the server being replaced; re-attach of one + buffer while the others stay stale. Every fix was locally correct and + none asked *what does this config swap invalidate?* — the answer being + every Lean buffer and every Lean server, because the config entry is + global and servers are per-root. **When a change edits shared state, + enumerate everything derived from it before repairing anything.** - **SUBSTRATE BUG FOUND, not fixed here (framing §6).** `LspManager::stop` on an ALREADY-terminal server takes its not-initialized branch, terminates the dead process and sets @@ -368,9 +395,9 @@ If it does not, stop and repair the remote/fetch configuration. server-failure latch covers the rest. - Verification on this branch: `cargo fmt --check` clean; strict workspace Clippy clean; 1,826 default + 2,003 CRDT library tests; - lean4 server 28/28; lean4 stage 1 9/9; dispatch seams 15/15; + lean4 server 31/31; lean4 stage 1 9/9; dispatch seams 15/15; multi-root 13/13; M4 121; required GPU 155; **isolated-config - workspace sweep 3,217 across 94 suites, zero failures**; + workspace sweep 3,220 across 94 suites, zero failures**; `git diff --check` clean. (Round 1 of this entry recorded 17/17 and 3,206 — the PRE-fix counts — after the fixes were pushed. The ledger's protocol is that verification diff --git a/tests/lean4_server_acceptance.rs b/tests/lean4_server_acceptance.rs index db708f1..32cdf3e 100644 --- a/tests/lean4_server_acceptance.rs +++ b/tests/lean4_server_acceptance.rs @@ -1010,9 +1010,19 @@ fn r2_a_failing_fallback_is_reported_once_and_does_not_retry_forever() { "a failing fallback surfaces rather than retrying silently; saw \ {status:?}" ); - // And the retry state is cleared, so it is not looping. - let pending: String = eval(&state, "return tostring(pmacs.lean._probe.reattach_from)"); - assert_eq!(pending, "nil", "the retry is retired, not spinning"); + // And the repair was ATTEMPTED and recorded, so it is bounded rather + // than spinning. Asserting on a field that no longer exists would + // read as nil and pass for nothing — the vacuity shape this branch + // keeps producing, so the assertion is on a positive count. + let attempted: i64 = eval( + &state, + "local n = 0 for _ in pairs(pmacs.lean._probe.repaired) do n = n + 1 end return n", + ); + assert_eq!( + attempted, 1, + "exactly one repair attempt was made and recorded, so a failing \ + fallback cannot retry every tick forever" + ); } #[test] @@ -1240,3 +1250,124 @@ fn r3_a_failing_wrapper_is_named_truthfully_not_as_lake_serve() { "and does not attribute the failure to `lake serve`; saw {status:?}" ); } + +// --------------------------------------------------------------------------- +// Round-4 review — the config swap is GLOBAL, so one repaired buffer is +// not a fallback. Both fail against 73587b0. +// --------------------------------------------------------------------------- + +#[test] +fn r4_every_open_lean_buffer_is_repaired_not_just_the_armed_one() { + // `pmacs.lsp.config.lean4` is a single entry; swapping its command + // invalidates every buffer attached to the old one. Round 3 repaired + // exactly `probe.buf_key` and cleared the retry, leaving every other + // open Lean buffer on the retired server while status and config + // both said "fell back". + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let first = fx.write("pkg/A.lean", "def a := 1\n"); + let second = fx.write("pkg/B.lean", "def b := 2\n"); + let lake = fx.lake_stub("bin/lake", "Lake version 3.0.0"); + let mut state = editor(&fx); + with_fallback(&state, &lake); + + open(&state, &first); + exec(&state, "_G.first_buf = pmacs.window.buffer()"); + open(&state, &second); + exec(&state, "_G.second_buf = pmacs.window.buffer()"); + tick_for(&mut state, 700); + + // The armed (first) buffer. + exec(&state, "pmacs.window.switch_buffer(_G.first_buf)"); + tick_for(&mut state, 500); + assert_eq!( + attached_command(&state), + fake_lsp_path(), + "the armed buffer is repaired" + ); + + // And the OTHER one, which round 3 stranded. + exec(&state, "pmacs.window.switch_buffer(_G.second_buf)"); + tick_for(&mut state, 500); + assert_eq!( + attached_command(&state), + fake_lsp_path(), + "every open Lean buffer ends up on the fallback — repairing only \ + the armed target leaves this one on the retired server" + ); +} + +#[test] +fn r4_a_second_project_roots_server_is_also_retired() { + // Q#LN15 gives one server per project root, so a swap can invalidate + // several. `probe.primary` names only the first; retiring only that + // leaves the second root's server live on a command the config no + // longer names. + let fx = Fixture::new(); + fx.toolchain("one", "v4.9.0\n"); + fx.toolchain("two", "v4.9.0\n"); + let a = fx.write("one/A.lean", "def a := 1\n"); + let b = fx.write("two/B.lean", "def b := 2\n"); + let lake = fx.lake_stub("bin/lake", "Lake version 3.0.0"); + let mut state = editor(&fx); + with_fallback(&state, &lake); + + open(&state, &a); + open(&state, &b); + // Two roots, two servers, before any verdict lands. + let before: i64 = eval(&state, "return #pmacs.lsp.list()"); + assert_eq!(before, 2, "precondition: one server per root"); + + tick_for(&mut state, 900); + + // No server may still be running the retired command. + let stale_live: i64 = eval( + &state, + &format!( + r#" + local n = 0 + for _, s in ipairs(pmacs.lsp.list()) do + if tostring(s.command) == "{}" then + local k = s.state and s.state.kind + if k ~= "stopped" and k ~= "crashed" then n = n + 1 end + end + end + return n + "#, + lua_str(&lake) + ), + ); + assert_eq!( + stale_live, 0, + "every Lean server spawned from the old command is retired, not \ + just the one the probe happened to name" + ); +} + +#[test] +fn r4_attribution_names_the_exact_command_and_its_arguments() { + // Round 3 implemented argument-inclusive attribution but pinned only + // "contains my-lean-wrapper" and "does not contain lake serve" — a + // mutation dropping every argument still passed. + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let absent = fx.dir("bin/my-lean-wrapper"); + let mut state = editor(&fx); + with_fallback(&state, &absent); + exec( + &state, + "pmacs.lsp.config.lean4.args = { \"serve\", \"--quiet\" }", + ); + + open(&state, &file); + settle(&mut state); + + let status = state.core.borrow().status.clone(); + let expected = format!("`{} serve --quiet`", absent.display()); + assert!( + status.contains(&expected), + "the status names the exact configured command AND its arguments;\n \ + want substring: {expected}\n saw: {status:?}" + ); +} From 19f48d46c0eed0ef1ddf17de1bc180b22420372e Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 21:28:11 -0400 Subject: [PATCH 17/32] fix(lsp,lean): bound the fallback's own failure; heal at point of use MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Round 5 review: one P1, a frontend scope hole, and three P2s. **1. A fallback that SPAWNS and then dies retried forever.** The once-per-buffer guard bounds calls to `_attach_buffer`, not the server those calls produce. `ensure_server` still never forwards `cfg.restart`, so the fallback inherits `OnCrash`; an executable that exits before `initialize` is respawned by the manager with no attempt ceiling — silently, because `latched` has already disabled the primary's failure poll. The fallback now gets its own one-shot die-before-initialize watch, which retires it (ending the respawn loop) and reports. The prior failing-fallback test used a NONEXISTENT executable, so it only ever exercised synchronous ENOENT. To reach "spawned, then died" the fixture has to actually spawn. **2. Simultaneous frontends.** Both repair triggers read the ambient `pmacs.window.buffer()`, and the daemon restores `active_frontend` to the last-dispatched frontend before `tick_processes` — so a Lean buffer active in ANOTHER frontend receives no `buffer.after-switch` here and stays stale after its server is globally retired. Fixed at the seam that is frontend-agnostic: **make consumption safe.** `attached_for_active` now rebuilds rather than returning a record whose server is dead, and `attachment_for_request` reports none (it must not perturb LSP state, so it cannot rebuild). Whichever frontend runs a command is the active one while it runs, so healing at the point of use reaches every buffer no eager sweep can. This also closes the half where a dead attachment was handed to a command and the request vanished. **3. The retirement sweep stopped user-managed servers.** Selecting on `language_id == "lean4"` also names servers the user spawned from `init.lua`, which are not derived from `pmacs.lsp.config.lean4`. It now keys on the `default-lean4` label `ensure_server` stamps — the derivation discriminator. **4. Repair ran even when no swap occurred.** `swap_to_fallback()` returning false left `latched` true, so the next tick retried the UNCHANGED configuration and reported it as a fallback failure. Split into `probe.fallback_installed`: repair exists to apply a swap, so no swap means nothing to apply. **5. The once-per-buffer assertion counted table keys**, which cannot distinguish "once per buffer" from "every tick for one buffer" — cardinality stays 1 either way. Replaced with a numeric attempt counter; the bite reports 174 attempts against the expected 1. Five bites, each against 7c37bdc: no fallback watch -> attempt reaches 4; retire by language_id -> the user's server is stopped; gate repair on `latched` -> a repair is attempted with no swap; drop the once-per-buffer guard -> 174 vs 1; hand back a dead attachment -> a command receives a `stopped` server. Two more vacuity shapes recorded in the ledger (8 and 9): counting distinct keys cannot bound repeated work, and a nonexistent executable cannot reach any post-spawn failure. --- builtin/runtime/lean.lua | 66 +++++++- builtin/runtime/lsp.lua | 26 +++ docs/active-work.md | 36 +++- tests/lean4_server_acceptance.rs | 271 +++++++++++++++++++++++++++++++ 4 files changed, 391 insertions(+), 8 deletions(-) diff --git a/builtin/runtime/lean.lua b/builtin/runtime/lean.lua index 546e123..6490296 100644 --- a/builtin/runtime/lean.lua +++ b/builtin/runtime/lean.lua @@ -133,6 +133,12 @@ local probe = { -- version verdict still has to retire it armed = false, -- the target buffer + primary have been captured repaired = {}, -- buffer key -> repair attempted (at most once) + repair_attempts = 0, -- COUNT of attach attempts, not distinct buffers: + -- table cardinality cannot tell "once per buffer" + -- from "every tick for one buffer" + fallback_installed = false, + fallback_watch = nil, -- fallback sid being polled for die-before-init + fallback_failed = false, saw_initialized = false, } @@ -280,12 +286,21 @@ end -- Retiring only the server that happened to fail left the others live -- and every buffer attached to them stranded on a command the config no -- longer names. -local function retire_all_lean_servers() +-- Only servers this module's config produced. `ensure_server` labels +-- every auto-attached server `default-`, so that label is the +-- derivation discriminator: a server the USER spawned from `init.lua` +-- carries their own label, is not derived from `pmacs.lsp.config.lean4`, +-- and must not be stopped because our config changed. Selecting on +-- `language_id` alone swept those up too — a destructive side effect on +-- state this module does not own. +local DERIVED_LABEL = "default-lean4" + +local function retire_derived_lean_servers() local ok, rows = pcall(pmacs.lsp.list) if not ok or not rows then return end local ids = {} for _, info in ipairs(rows) do - if info.language_id == "lean4" then ids[#ids + 1] = info.id end + if info.label == DERIVED_LABEL then ids[#ids + 1] = info.id end end for _, id in ipairs(ids) do retire_server(id) end end @@ -308,7 +323,14 @@ end -- on a no-op. Skipping leaves the attempt for a later tick, once the -- retirement has actually landed. local function repair_active_if_stale() - if not probe.latched then return end + -- **`fallback_installed`, not `latched`.** When the swap does not + -- happen — the config already names the fallback, or it vanished + -- before an asynchronous verdict landed — `fire_latch` returns early + -- but `latched` stays true. Gating repair on `latched` then retried + -- the UNCHANGED configuration and reported the result as a fallback + -- failure, which is both a second pointless spawn and a misleading + -- message. Repair exists to apply a swap; no swap, nothing to apply. + if not probe.fallback_installed then return end local buf = pmacs.window.buffer() if not buf then return end local key = tostring(buf) @@ -327,10 +349,42 @@ local function repair_active_if_stale() if not stale then return end probe.repaired[key] = true + probe.repair_attempts = probe.repair_attempts + 1 local ok, fresh = pcall(pmacs.lsp._attach_buffer) if not ok or not fresh then report("LSP: lean4 fallback " .. fallback_name() .. " did not start either") + return + end + -- **A successful SPAWN is not a successful START.** The once-per- + -- buffer bound stops `_attach_buffer` being called again, but it says + -- nothing about the server it produced: `ensure_server` never forwards + -- `cfg.restart`, so the fallback inherits `OnCrash`, and an executable + -- that dies before `initialize` is respawned by the manager forever + -- with no attempt ceiling — silently, because `latched` has already + -- disabled the primary's failure poll. Watch this one too, once. + if not probe.fallback_watch and not probe.fallback_failed then + probe.fallback_watch = fresh.server + end +end + +-- The fallback's own die-before-initialize poll. One shot: on failure it +-- retires the server (which is what actually ends the respawn loop) and +-- reports, and never re-arms. +local function poll_fallback() + local sid = probe.fallback_watch + if not sid then return end + local kind = server_state_kind(sid) + if kind == "initialized" then + probe.fallback_watch = nil + return + end + if kind == nil or kind == "crashed" or kind == "stopped" then + probe.fallback_watch = nil + probe.fallback_failed = true + if kind ~= nil then retire_server(sid) end + report("LSP: lean4 fallback " .. fallback_name() + .. " started but did not stay up") end end @@ -343,10 +397,11 @@ local function fire_latch(sid, why) -- Still retire: the servers are broken whether or not a replacement -- command was installed, and leaving them live would keep the -- restart machinery running against a command known to fail. - retire_all_lean_servers() + retire_derived_lean_servers() return end - retire_all_lean_servers() + retire_derived_lean_servers() + probe.fallback_installed = true report("LSP: lean4 " .. why .. "; falling back to " .. fallback_name()) -- Repair what is in front of the user now; everything else is -- repaired lazily as it becomes active (see `repair_active_if_stale`). @@ -607,6 +662,7 @@ pmacs.hook.add("process.after-tick", function() -- Repair the active buffer if the latch invalidated it. Cheap when -- there is nothing to do, and bounded to one attempt per buffer. repair_active_if_stale() + poll_fallback() end) -- Test seam: acceptance drives the latch deterministically rather than diff --git a/builtin/runtime/lsp.lua b/builtin/runtime/lsp.lua index 0749cfa..6081a3d 100644 --- a/builtin/runtime/lsp.lua +++ b/builtin/runtime/lsp.lua @@ -871,6 +871,21 @@ local function attached_for_active() if not buf then return nil end local key = tostring(buf) local rec = attachments[key] + -- A record whose server is dead is worse than no record: every + -- command below issues requests against it and gets silence. Rebuild + -- instead, which is what `attach_buffer` does for a stale attachment + -- anyway — this just stops the dead record short-circuiting that. + -- + -- Load-bearing for anything that retires a server out from under open + -- buffers (Arc 8 Stage 3b's fallback latch retires every Lean server + -- at once). Buffers in OTHER frontends get no `buffer.after-switch` + -- in this one, so an eager repair sweep keyed on the ambient active + -- buffer cannot reach them; healing at the point of USE is + -- frontend-agnostic, because whichever frontend runs the command is + -- the active one while it runs. + if rec and not server_is_live(rec.server) then + rec = nil + end if rec then -- Every interactive command resolves its attachment here before -- issuing requests; flushing now means the server answers those @@ -942,6 +957,17 @@ function pmacs.lsp.attachment_for_request() local key = tostring(buf) local rec = attachments[key] if not rec then return nil end + -- Same liveness rule as `attached_for_active`: a record naming a dead + -- server is worse than none, because the caller issues a request + -- against it and waits for a reply that cannot come. Unlike that + -- function this one is deliberately non-attaching (it must not + -- perturb LSP state), so a dead record reads as "no attachment" + -- rather than triggering a rebuild. + if not server_is_live(rec.server) then + attachments[key] = nil + pending_did_change[key] = nil + return nil + end flush_did_change(key) return rec end diff --git a/docs/active-work.md b/docs/active-work.md index e4e1064..6b99985 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -265,7 +265,7 @@ If it does not, stop and repair the remote/fetch configuration. - Ships `builtin/runtime/lean.lua` (new), one `include_str!` line in `src/editor.rs`, `pmacs.lsp._attach_buffer` exported from `lsp.lua`, a `leanprogress` mode plus `waitForDiagnostics` validation on - `pmacs_fake_lsp`, and `tests/lean4_server_acceptance.rs` (31 tests). + `pmacs_fake_lsp`, and `tests/lean4_server_acceptance.rs` (36 tests). No protocol change. - **Stage 1's acceptance 12 is half superseded and was rewritten, not deleted.** It asserted `pmacs.lsp.config.lean4 == nil` to catch a @@ -334,6 +334,31 @@ If it does not, stop and repair the remote/fetch configuration. argument-inclusive attribution was implemented but pinned only by "contains the command name", so a mutation dropping every argument still passed. +- **Round-5 review: one P1 plus a frontend scope hole, and four more.** + (1) A fallback that SPAWNS and then dies retried forever: the + once-per-buffer guard bounds `_attach_buffer`, not the server it + produced, and `ensure_server` never forwards `cfg.restart` so the + fallback inherits `OnCrash` — respawned by the manager with no + ceiling, silently, because `latched` had disabled the primary's poll. + The fallback now gets its own one-shot die-before-initialize watch. + (2) **Simultaneous frontends**: both repair triggers read the ambient + `pmacs.window.buffer()`, and the daemon restores `active_frontend` to + the last-dispatched one before `tick_processes`, so a Lean buffer + active in ANOTHER frontend gets no `after-switch` and stays stale. + Fixed at the right seam — **make CONSUMPTION safe**: both + `attached_for_active` and `attachment_for_request` now refuse a record + whose server is dead (the former rebuilds, the latter reports none, + since it must not perturb LSP state). Healing at the point of use is + frontend-agnostic, because whichever frontend runs a command is active + while it runs. (3) The retirement sweep selected on `language_id`, so + it stopped USER-spawned Lean servers too; it now keys on the + `default-lean4` label `ensure_server` stamps, which is the derivation + discriminator. (4) `probe.latched` gated repair even when NO swap + occurred, so an already-fallback config was retried and misreported. + Split out `probe.fallback_installed`. (5) The once-per-buffer + assertion counted TABLE KEYS, which cannot distinguish "once per + buffer" from "every tick for one buffer" — cardinality stays 1 either + way. Now a numeric attempt counter; the bite shows **174 vs 1**. - **DURABLE LESSON — "the test that passes" vs "the test that discriminates."** Six tests across three rounds were written, run green, and only bite-testing showed they pinned nothing. **Carry this @@ -359,6 +384,11 @@ If it does not, stop and repair the remote/fetch configuration. 7. Asserting on a field that no longer exists (`_probe.reattach_from` after a refactor) reads as nil and passes for nothing. Assert positive facts — a count, a command string — not absences. + 8. Counting DISTINCT KEYS cannot bound REPEATED WORK: a per-tick retry + on one buffer keeps `#repaired == 1` forever. Count the attempts, + not the things attempted against (bite: 174 vs 1). + 9. A NONEXISTENT executable only exercises synchronous ENOENT. To + reach "spawned, then died", the fixture must actually spawn. Rule: **a test is not evidence until the mutation it targets has been shown to fail it.** - **SECOND DURABLE LESSON — a scope error repeats until the scope is @@ -395,9 +425,9 @@ If it does not, stop and repair the remote/fetch configuration. server-failure latch covers the rest. - Verification on this branch: `cargo fmt --check` clean; strict workspace Clippy clean; 1,826 default + 2,003 CRDT library tests; - lean4 server 31/31; lean4 stage 1 9/9; dispatch seams 15/15; + lean4 server 36/36; lean4 stage 1 9/9; dispatch seams 15/15; multi-root 13/13; M4 121; required GPU 155; **isolated-config - workspace sweep 3,220 across 94 suites, zero failures**; + workspace sweep 3,225 across 94 suites, zero failures**; `git diff --check` clean. (Round 1 of this entry recorded 17/17 and 3,206 — the PRE-fix counts — after the fixes were pushed. The ledger's protocol is that verification diff --git a/tests/lean4_server_acceptance.rs b/tests/lean4_server_acceptance.rs index 32cdf3e..e28ac8f 100644 --- a/tests/lean4_server_acceptance.rs +++ b/tests/lean4_server_acceptance.rs @@ -1371,3 +1371,274 @@ fn r4_attribution_names_the_exact_command_and_its_arguments() { want substring: {expected}\n saw: {status:?}" ); } + +// --------------------------------------------------------------------------- +// Round-5 review. All fail against 7c37bdc. +// --------------------------------------------------------------------------- + +#[test] +fn r5_a_fallback_that_dies_after_spawning_is_bounded_and_reported() { + // The once-per-buffer guard bounds calls to `_attach_buffer`, not + // the server it produced. `ensure_server` never forwards + // `cfg.restart`, so the fallback inherits `OnCrash` and a binary + // that exits before `initialize` is respawned forever — silently, + // because `latched` has already disabled the primary's poll. The + // prior failing-fallback test used a NONEXISTENT executable, which + // only exercises synchronous ENOENT. + use std::os::unix::fs::PermissionsExt as _; + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let absent_primary = fx.dir("bin/no-such-lake"); + let dying_fallback = fx.root.join("bin/dying-lean"); + std::fs::create_dir_all(dying_fallback.parent().unwrap()).unwrap(); + std::fs::write(&dying_fallback, "#!/bin/sh\nexit 4\n").unwrap(); + std::fs::set_permissions(&dying_fallback, std::fs::Permissions::from_mode(0o755)).unwrap(); + + let mut state = editor(&fx); + exec( + &state, + &format!( + r#" + pmacs.lsp.config.lean4.command = "{}" + pmacs.lsp.config.lean4.args = {{ "serve" }} + pmacs.lean._fallback = {{ command = "{}", args = {{}} }} + "#, + lua_str(&absent_primary), + lua_str(&dying_fallback) + ), + ); + + open(&state, &file); + tick_for(&mut state, 1600); + + // Nothing may be respawning: `attempt` counts spawns per server. + let worst_attempt: i64 = eval( + &state, + r" + local worst = 0 + for _, s in ipairs(pmacs.lsp.list()) do + local a = s.attempt or 0 + if a > worst then worst = a end + end + return worst + ", + ); + assert!( + worst_attempt <= 1, + "a dying fallback must not be respawned indefinitely; saw \ + attempt {worst_attempt}" + ); + let status = state.core.borrow().status.clone(); + assert!( + status.contains("did not stay up") || status.contains("did not start"), + "and the second failure is reported; saw {status:?}" + ); +} + +#[test] +fn r5_a_user_spawned_lean_server_is_not_retired_by_the_fallback() { + // `retire_*` selected on `language_id == "lean4"`, which also names + // servers the user spawned themselves from `init.lua`. Those are not + // derived from `pmacs.lsp.config.lean4` and stopping them is a + // destructive side effect on state this module does not own. + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let absent = fx.dir("bin/no-such-lake"); + let mut state = editor(&fx); + with_fallback(&state, &absent); + exec( + &state, + &format!( + r#" + _G.mine = pmacs.lsp.spawn({{ + label = "my-own-lean", + language_id = "lean4", + command = "{}", + args = {{}}, + }}) + "#, + fake_lsp_path() + ), + ); + settle(&mut state); + + open(&state, &file); + tick_for(&mut state, 600); + + let mine_alive: bool = eval( + &state, + r#" + for _, s in ipairs(pmacs.lsp.list()) do + if tostring(s.id) == tostring(_G.mine) then + local k = s.state and s.state.kind + return k ~= "stopped" and k ~= "crashed" + end + end + return false + "#, + ); + assert!( + mine_alive, + "a user-spawned Lean server survives a config-driven fallback — \ + it was never derived from that config" + ); +} + +#[test] +fn r5_no_swap_means_no_repair_attempts() { + // When the config already names the fallback, `swap_to_fallback` + // returns false and `fire_latch` returns early — but `latched` is + // true, so a repair gated on `latched` retried the UNCHANGED + // configuration and reported it as a fallback failure. + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let absent = fx.dir("bin/no-such-lean"); + let mut state = editor(&fx); + // Config and fallback are the SAME missing command, so no swap is + // possible. + exec( + &state, + &format!( + r#" + pmacs.lsp.config.lean4.command = "{}" + pmacs.lsp.config.lean4.args = {{}} + pmacs.lean._fallback = {{ command = "{}", args = {{}} }} + "#, + lua_str(&absent), + lua_str(&absent) + ), + ); + + open(&state, &file); + tick_for(&mut state, 400); + + let attempts: i64 = eval(&state, "return pmacs.lean._probe.repair_attempts"); + assert_eq!( + attempts, 0, + "no swap happened, so there is nothing to apply and no repair \ + should be attempted" + ); + let status = state.core.borrow().status.clone(); + assert!( + !status.contains("falling back"), + "and nothing claims a fallback occurred; saw {status:?}" + ); +} + +#[test] +fn r5_repair_is_attempted_at_most_once_per_buffer_by_count() { + // Counting keys in the `repaired` table cannot distinguish + // "once per buffer" from "every tick for one buffer" — the + // cardinality stays 1 either way. Count the ATTEMPTS. + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let absent_primary = fx.dir("bin/no-such-lake"); + let absent_fallback = fx.dir("bin/no-such-lean"); + let mut state = editor(&fx); + exec( + &state, + &format!( + r#" + pmacs.lsp.config.lean4.command = "{}" + pmacs.lsp.config.lean4.args = {{ "serve" }} + pmacs.lean._fallback = {{ command = "{}", args = {{}} }} + "#, + lua_str(&absent_primary), + lua_str(&absent_fallback) + ), + ); + + open(&state, &file); + // Many ticks; a per-tick retry would climb without bound. + tick_for(&mut state, 900); + + let attempts: i64 = eval(&state, "return pmacs.lean._probe.repair_attempts"); + assert_eq!( + attempts, 1, + "exactly one repair attempt across many ticks for one buffer" + ); +} + +#[test] +fn r5_a_dead_attachment_is_never_handed_to_a_command() { + // Buffers live in other frontends get no `buffer.after-switch` here, + // so an eager sweep keyed on the ambient active buffer cannot reach + // them. Healing at the point of USE is frontend-agnostic: + // `attached_for_active` must not return a record whose server is + // gone. + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let mut state = editor(&fx); + // A working primary, so we get a live attachment first. + exec( + &state, + &format!("pmacs.lsp.config.lean4.command = \"{}\"", fake_lsp_path()), + ); + open(&state, &file); + settle(&mut state); + let first: String = attached_sid(&state); + assert_ne!(first, "none", "precondition: attached"); + + // Retire it out from under the buffer, as the latch does globally, + // WITHOUT any switch or repair tick. + exec( + &state, + r" + local rec = pmacs.lsp.active_attachment() + pcall(pmacs.lsp.stop, rec.server) + ", + ); + for _ in 0..40 { + state.tick_processes(); + state.tick_lsp(); + std::thread::sleep(Duration::from_millis(5)); + } + + // Now a command resolves its attachment. It must not get the dead + // one; it must rebuild. + // `attachment_for_request` is deliberately non-attaching, so a dead + // record must read as "no attachment" rather than being handed over. + let for_request: String = eval( + &state, + r#" + local rec = pmacs.lsp.attachment_for_request() + if not rec then return "none" end + for _, s in ipairs(pmacs.lsp.list()) do + if tostring(s.id) == tostring(rec.server) then + return tostring(s.state and s.state.kind) + end + end + return "gone" + "#, + ); + assert_eq!( + for_request, "none", + "a non-attaching resolve must not hand back a dead server" + ); + + // And the attaching path rebuilds rather than returning the corpse. + let rebuilt: String = eval( + &state, + r#" + pmacs.lsp._attach_buffer() + local rec = pmacs.lsp.active_attachment() + if not rec then return "none" end + for _, s in ipairs(pmacs.lsp.list()) do + if tostring(s.id) == tostring(rec.server) then + return tostring(s.state and s.state.kind) + end + end + return "gone" + "#, + ); + assert!( + rebuilt != "stopped" && rebuilt != "crashed" && rebuilt != "gone" && rebuilt != "none", + "the attaching path rebuilds against a live server; saw \ + {rebuilt:?}" + ); +} From 8e8f281f0e70818d3aca20bc2d1b5747c2e2faad Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 21:47:37 -0400 Subject: [PATCH 18/32] fix(terminal): close review round 1 on Stage 1 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Five findings, all real. The blocker and both majors are the same mistake in three places: a claim asserted somewhere cheaper than where it actually lives. COHERENCE.md was stale in four places, not the three reported. Step 8 still read "no keybinding" and §11 still read "five settings", but §6's dispatch table also still cited `is_terminal_escape_chord` — a symbol this branch deletes. §25 requires that update to ride the PR, so a PR changing audited ground truth has to re-grep the audit for its own symbols, not only for its topic. Acceptance 5 asserted a registry round-trip, which is a test of the registry: it stayed green with the setting's only consumer deleted. It now opens a real terminal whose child overflows the 24-row screen, scrolls the view to its oldest retained row, and asserts LINE001 is present at 10,000 and absent at 0. Acceptance 8a waited for the session count to fall, which the rejected editor-side cache map satisfies exactly — a map with no purge hook leaks while sessions drain. Adds `TerminalManager::escape_caches()`, the lifetime half of Q#TC4c's contract that `escape_parses` cannot cover. `table.sort` over `pmacs.terminal.profiles` raised "attempt to compare number with string" on the unknown-profile path whenever the user's table held both a string and a numeric key, replacing the exact diagnostic being asked for; `%q` raised likewise on a non-string `profile` argument. Both are partial functions applied to user input on a diagnostic path. Also corrects the framing's status line, and a status message whose embedded whitespace run had survived a rustfmt reflow. Three new bites, each falsified by revert: deleting the scrollback consumer fails acc5 and only acc5; restoring the raw-key sort reproduces the comparison error verbatim; and implementing the rejected map fails the new acc8a at left: 2, right: 1 while passing the old session-count version. Merges githubsucks/main @ ccf29e3. Co-Authored-By: Claude Opus 5 (1M context) Claude-Session: https://claude.ai/code/session_016gGQC6eqHJVbZJ5Hg7aLer --- COHERENCE.md | 32 ++- builtin/runtime/terminal.lua | 30 ++- docs/active-work.md | 66 +++++- docs/terminal-config-and-copy-mode-framing.md | 26 ++- src/terminal/session.rs | 20 +- tests/terminal_config_acceptance.rs | 192 +++++++++++++++++- 6 files changed, 331 insertions(+), 35 deletions(-) diff --git a/COHERENCE.md b/COHERENCE.md index 9fe85f0..4e7361c 100644 --- a/COHERENCE.md +++ b/COHERENCE.md @@ -368,7 +368,7 @@ Full verdict table: | 5 | Edit | **Works** | Full CUA + Emacs keymap in 161 lines (`builtin/keymaps/default.lua`); isearch, query-replace, kill ring, undo/redo, auto-indent/pair/comment, atomic save. Genuinely excellent zero-config | | 6 | Language intelligence | **Partial** | Rust grammar bundled and auto-attaches; rust-analyzer preconfigured (`builtin/runtime/lsp.lua:44-52`) — but a missing binary fails silently (§1.2) and highlighting masks it. No LSP status command exists to diagnose | | 7 | Find symbol / file | **File: fixed (open by path merged #162; browsing PR #165). Symbol: works but undiscoverable** | No find-file/dired/picker existed at audit. Now `C-x C-f` opens a known path and `C-x d` / `C-x C-j` browse (flat listing, `dired` mode keymap); `M-.`/`M-?`/`C-c o` still bound but advertised nowhere and server-gated; no workspace-symbol command; `pmacs.index.*` has no UI | -| 8 | Open terminal | **Works but undiscoverable** | Full PTY with scrollback + modeline segment — reachable only as `M-x terminal`, no keybinding. *Was broken outright on the GPU frontend until the double terminal-layout sync was fixed: the child took a `SIGWINCH` storm at tick cadence, so typing into it was impossible while output still flowed.* | +| 8 | Open terminal | **Works** | Full PTY with scrollback + modeline segment, bound to `C-c t` and configurable through three registered settings (`terminal.default-profile`, `terminal.scrollback-rows`, `terminal.escape-key`) plus named `pmacs.terminal.profiles` (PR #173). Named limitation: `C-c t` is unreachable from *inside* a terminal window, where `C-c` is consumed as the escape — `M-x terminal` still works there. *Was broken outright on the GPU frontend until the double terminal-layout sync was fixed: the child took a `SIGWINCH` storm at tick cadence, so typing into it was impossible while output still flowed.* | | 9 | Build / test | **Partial** | `M-x compile.run` works, defaults cwd to detected project root, parses Rust `-->` errors — but no keybinding, an **empty first prompt** (`initial = last and last.cmdline or ""`, `builtin/runtime/compile.lua:1134-1138`), and no `cargo build`/`cargo test` suggestion despite `ProjectKind::Cargo` existing (`src/project.rs:77`) | | 10 | Inspect error | **Partial (good once reached)** | `E:n W:n` modeline counts, underlines, `M-g n/p` + ``C-x ` `` walking a unified compile/grep/diag source, message echo, `RET` visits. Gated entirely on step 6 or 9 succeeding first | | 11 | See background work | **Works but undiscoverable** | `*workers*` view via `M-x editor.list-workers`; `C-c C-k` cancel-at-point. No keybinding, no statusline spinner/progress indicator anywhere (§9) | @@ -379,6 +379,13 @@ A journey observation worth keeping verbatim from the audit: C-M-s` opens all folds, while opening a file, opening a terminal, and running a build have no bindings at all. +Two of that observation's three examples have since been answered — +opening a file by `C-x C-f` (#162) and opening a terminal by `C-c t` +(#173). **Running a build still has no binding**, and the underlying +inversion is a standing bias in how new work gets bound, not three +isolated omissions: the quote stays as written because it names the +pattern, and the pattern is not retired until step 9 is. + --- ## 3. A Strong Zero-Configuration State @@ -639,7 +646,7 @@ Everything funnels through one function: `EditorInstance::dispatch_key` | 3 | query-replace | `editor.rs:945` | `QueryReplaceKey::from_chord` (`editor.rs:2967`) | **full shadow** | | 4 | Minibuffer | `editor.rs:951` | `MinibufferAction::from_chord` (`src/minibuffer.rs:468`) | **full shadow** | | 5 | Completion popup | `editor.rs:958-971` | `CompletionPopupKey::from_chord` (`editor.rs:3056`) | **partial shadow** (control chords only; skipped while a multi-key prefix is pending) | -| 6 | Terminal transport + `C-c` escape | `editor.rs:973-1010` | `is_terminal_escape_chord` (`editor.rs:4355`) | **partial, transport-level** | +| 6 | Terminal transport + configurable escape | `editor.rs:973-1010` | `EditorState::terminal_escape_chord` → `TerminalManager::escape_chord` (`src/terminal/session.rs`) | **partial, transport-level** | | 7 | Ordinary dispatch | `editor.rs:1018-1032` | `KeymapStack::resolve` | the only inspectable layer | Facts that define the gap: @@ -647,8 +654,10 @@ Facts that define the gap: - **Full shadows eat every key**, including unrecognized ones (each decoder has an `Ignore`/`Dismiss` fallback arm). While a terminal buffer is focused and unescaped, *all* keys encode to the child — - `C-c`-leading user bindings are **structurally unreachable** in a - terminal buffer. + bindings led by the escape chord are **structurally unreachable** in + a terminal buffer. Since #173 that chord is `terminal.escape-key` + rather than a hardcoded `C-c`, so a user can *move* which prefix is + eaten; they cannot make the shadow stop eating one. - **No transient-keymap mechanism exists to migrate to.** `KeymapStack` has exactly three fixed scopes — `Buffer(BufferId)`, `Mode(String)`, `Global` (`src/keymap_stack.rs:37-44`); resolution order buffer → @@ -1013,20 +1022,29 @@ layering, provenance, and adoption have not followed.** `ConfigValue`s; `describe-setting`'s "Source:" names where `define()` ran. The inspection view sketched above is currently impossible to render. -- **Adoption is five settings**: `editing.auto-pair` (pair.lua), +- **Adoption is eight settings**: `editing.auto-pair` (pair.lua), `editing.trim-on-save` (editops.lua), `autosave.interval-ms` (autosave.lua), `window.panel-height` + `window.min-height` - (window.lua). Everything else a user might set — theme, fonts, LSP + (window.lua), and `terminal.default-profile` + + `terminal.scrollback-rows` + `terminal.escape-key` (terminal.lua, + #173). Everything else a user might set — theme, fonts, LSP server config, killring size, recentf/saveplace/desktop enables, pair sets, comment strings, `pmacs.parse.*` — lives in raw Lua outside the registry and is therefore invisible to `describe-setting` and any future settings UI. The migration list is already written: `docs/config-registry-framing.md` "named deferrals" (table-valued settings are the hard prerequisite for LSP/pair/comment tables). +- **The table-valued gap now has a named, shipped instance.** + `pmacs.terminal.profiles` (#173) is a raw Lua table sitting beside + three registered scalars *for the same feature*, because a profile is + inherently `{ command, args, cwd, env }` and the registry stores four + scalars. It is the clearest evidence yet that table-valued settings + are the blocking prerequisite: the terminal is now half-registered, + and no settings UI can render the half that matters most. - **No persistence**: settings changed at runtime do not survive restart (the `custom-file` split-brain question is a named deferral). - The three-level separation holds in principle today (registry / - hooks+keymaps / packages), but with five settings registered, level 1 + hooks+keymaps / packages), but with eight settings registered, level 1 is effectively empty — users need executable Lua for nearly every ordinary preference, which is the exact failure the section warns about. diff --git a/builtin/runtime/terminal.lua b/builtin/runtime/terminal.lua index a6573fe..143a663 100644 --- a/builtin/runtime/terminal.lua +++ b/builtin/runtime/terminal.lua @@ -65,27 +65,44 @@ local PROFILE_FIELDS = { env = "table", } +-- Every diagnostic below renders a caller- or user-supplied value, so +-- rendering must never be the thing that fails. `%q` is partial — it +-- raises on a table or function — and a profile name arrives straight +-- from `open { profile = ... }`. +local function describe_name(name) + if type(name) == "string" then return string.format("%q", name) end + return string.format("<%s %s>", type(name), tostring(name)) +end + local function validate_profile(name, profile) + local shown = describe_name(name) if type(profile) ~= "table" then - error(string.format("terminal profile %q must be a table", name), 0) + error(string.format("terminal profile %s must be a table", shown), 0) end for key, value in pairs(profile) do local expected = PROFILE_FIELDS[key] if not expected then - error(string.format("terminal profile %q: unknown field %q", name, tostring(key)), 0) + error(string.format("terminal profile %s: unknown field %q", shown, tostring(key)), 0) end if type(value) ~= expected then error(string.format( - "terminal profile %q: field %q must be a %s, got %s", - name, key, expected, type(value)), 0) + "terminal profile %s: field %q must be a %s, got %s", + shown, key, expected, type(value)), 0) end end return profile end +-- `terminal.profiles` is a raw user table, so its keys are whatever the +-- user wrote. Sorting them directly raises "attempt to compare number +-- with string" the moment the table holds both a string and a numeric +-- key — and it raises on the UNKNOWN-PROFILE path, replacing the very +-- error this list exists to explain with an opaque one. Sorting DISPLAY +-- strings is total over every key type, so the diagnostic survives a +-- malformed table. local function known_profile_names() local names = {} - for name in pairs(terminal.profiles) do names[#names + 1] = name end + for name in pairs(terminal.profiles) do names[#names + 1] = tostring(name) end table.sort(names) return names end @@ -106,7 +123,8 @@ local function resolve_profile(requested) local known = known_profile_names() local listed = #known > 0 and table.concat(known, ", ") or "(none defined)" error(string.format( - "terminal profile %q is not defined; known profiles: %s", name, listed), 0) + "terminal profile %s is not defined; known profiles: %s", + describe_name(name), listed), 0) end return validate_profile(name, profile) end diff --git a/docs/active-work.md b/docs/active-work.md index aeac490..972c40b 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -396,8 +396,9 @@ If it does not, stop and repair the remote/fetch configuration. Stage 1's branch. Two stages, two branches, two PRs; **no protocol change**. - **Stage 1 = `githubsucks/terminal-config`**, worktree - `../pmacs-terminal-config`, based on `githubsucks/main` @ `d152120`. - Profiles, scrollback, escape key, and the `C-c t` opening binding. + `../pmacs-terminal-config`, based on `githubsucks/main` @ `d152120` + and merged up to `ccf29e3` during review round 1. Profiles, + scrollback, escape key, and the `C-c t` opening binding. - **Stage 2 = `terminal-copy-mode`, not started.** Branch it off `main` after Stage 1 merges: no dependency, but both edit `builtin/runtime/terminal.lua`. @@ -424,16 +425,63 @@ If it does not, stop and repair the remote/fetch configuration. `Ctrl-X` is invisible; and the probe **counts occurrences** rather than testing presence, because a single-character probe collides with the child's own banner text. -- Verification on this branch (against the committed tree): `cargo fmt - --check` clean; strict workspace Clippy clean; 1,832 default + 2,009 - CRDT library tests; `terminal_config_acceptance` 10/10 in **both** - configurations; vterm Stage 1/2/3 9+10 / 6+6 / 5+9; config registry 16; - bottom-panel 46; listview 6; compile 67 (isolated config); M4 121; - required GPU 202; **isolated-config workspace sweep 3,262 across 94 - suites**; `git diff --check` clean. +- **Review round 1 (2026-07-25) — five findings, all real, all fixed.** + One blocker and two majors were the same failure in three places: a + claim asserted somewhere cheaper than where it lives. + - *Blocker — `COHERENCE.md` was stale in four places, not the three + reported.* Step 8 still read "no keybinding"; §11 still read "five + settings"; and §6's dispatch table still cited + `is_terminal_escape_chord`, **a symbol this PR deletes**. §25 makes + that update ride the PR. A PR that changes audited ground truth has + to re-grep the audit for its own symbols, not only for its topic. + - *Major — acceptance 5 was vacuous.* It asserted a registry + round-trip, so it stayed green with the setting's **only** consumer + deleted. It now opens a real terminal whose child overflows the + 24-row screen, scrolls the view to its oldest retained row, and + asserts `LINE001` is present at 10,000 and absent at 0. **Asserting + a value was stored is not asserting anything reads it.** + - *Major — acceptance 8a asserted the session count, not the cache.* + An editor-side map with no purge hook — the exact rejected design — + leaks *while* sessions drain, so it passed. Fixed with a + `TerminalManager::escape_caches()` seam. **A lifecycle claim needs a + lifecycle observable.** + - *Moderate — `table.sort` over user-controlled profile keys.* A + table holding both a string and a numeric key raised `attempt to + compare number with string` **on the unknown-profile path**, + replacing the diagnostic being asked for; `%q` raised likewise on a + non-string `profile` argument. Both are partial functions applied to + user input **on a diagnostic path** — the error reporter was the + thing that failed. + - *Minor — the committed framing still said "not yet approved".* +- **Three new bites, each falsified by revert**: deleting the scrollback + consumer fails acc5 (and only acc5); restoring the raw-key sort + reproduces `attempt to compare string with number` verbatim; and + implementing the rejected editor-side map fails the new acc8a at + `left: 2, right: 1` **while passing the old session-count version** — + which is the review finding demonstrated rather than argued. +- Verification after the round-1 fixes, on the tree merged with + `githubsucks/main` @ `ccf29e3`: `cargo fmt --check` clean; strict + workspace Clippy clean; 1,832 default + 2,009 CRDT library tests; + `terminal_config_acceptance` **12/12 in both configurations**; vterm + Stage 1/2 9+10 / 6+6; config registry 16+16; bottom-panel Stage 1 + 46+46; M4 121; required GPU 202; `git diff --check` clean. - `compile_mode_acceptance` fails 11/67 against the **real** user config and passes 67/67 with an isolated `XDG_CONFIG_HOME` — the known pre-existing trap, not this branch. + - **`vterm_stage3_acceptance::a37` fails on this machine — and fails + identically on the PR's own base `d152120`**, so it is not this + branch's regression. It is load-sensitive: it passed at `d152120` + once and failed at that same commit twenty minutes later, with a + second agent saturating the machine with `rustc` in between. Two + ways it lies, both worth knowing: it **silently returns `ok` when + `pmacs-gpu` is not built** in the same target dir (only + `PMACS_REQUIRE_GPU=1` promotes that skip to a failure, and the gate + list applies that flag to `-p pmacs-gpu`, a *different* package), and + it is **crdt-gated, so CI has never run it at all**. A green a37 in + a gate log means nothing unless the binary was built and the flag + was set. Needs its own lane; see the CI `crdt`-coverage lane on #168. + - `pmacs-gpu` itself failed 201/202 once under the same load and passed + 202/202 on immediate rerun. ## Bottom-panel lane (Arc 7) — Stage 1 MERGED; Stage 2 (GPU band) is next diff --git a/docs/terminal-config-and-copy-mode-framing.md b/docs/terminal-config-and-copy-mode-framing.md index 3f13987..48b75d8 100644 --- a/docs/terminal-config-and-copy-mode-framing.md +++ b/docs/terminal-config-and-copy-mode-framing.md @@ -1,7 +1,9 @@ # Terminal configuration and copy mode **Revision 4 — scouted against canonical `main` @ `b889873` (protocol v20), -2026-07-25. Not yet approved; no branch, no implementation.** +2026-07-25. APPROVED after four review rounds. Stage 1 is implemented on +branch `terminal-config` (PR #173); Stage 2 (`terminal-copy-mode`) is +framed but not started, and branches off `main` after Stage 1 merges.** Revision 4 gives the escape-key cache an owner and a lifecycle (Q#TC4c) — revision 3 named the key but not the storage, and two implementations @@ -503,6 +505,15 @@ additive, on its own binding, and does not replace scroll-and-select. error that **lists the known profile names**, and creates no buffer, session, or process. An explicitly passed unknown `profile` fails the same way **even when `terminal.default-profile` is valid** (Q#TC3a). +2a. That diagnostic is **total over a malformed profiles table** (review round + 1). `pmacs.terminal.profiles` is a raw user table, so listing its names must + not assume its keys are comparable and rendering a requested name must not + assume it is a string: a table holding both a string and a numeric key made + `table.sort` raise `attempt to compare number with string` *on the + unknown-profile path*, replacing the exact error being asked for, and `%q` + raises on a non-string `profile` argument. Both are partial functions + applied to user input on a diagnostic path — the failure class is + "the error reporter is the thing that fails". 3. Field-by-field resolution follows Q#TC3a: explicit open field beats profile field beats scalar setting beats `$SHELL`. `env` **merges**, with explicit entries overriding profile entries of the same name. @@ -623,6 +634,19 @@ Full gate suite per `CLAUDE.md` for each PR separately, plus: chord unreachable); **10** (its failure mode is a terminal nobody can escape); and **16/17** (a read-only buffer that silently accepts an edit on both sides). +- **The observation seams the cache pins need are `escape_parses` (how often) + and `escape_caches` (how many are still held).** Neither is inferable from + behavior: for a *valid* setting a correct per-session cache and a leaking + editor-side map produce identical keystroke results, and both leave the + session count draining normally. Review round 1 caught 8a asserting the + session count instead — which the unpurged-map bite passes, since a map with + no purge hook leaks *while* sessions drain. A lifecycle claim needs a + lifecycle observable; the count of live sessions is not one. +- **Criterion 5 must open a real terminal and read back retained history.** + Round 1 caught it asserting a registry round-trip instead, which is a test of + the registry: it stays green with the setting's only consumer deleted. The + same shape to watch for anywhere — *asserting that a value was stored is not + asserting that anything reads it*. - **Do not gate the new suites on `#[cfg(feature = "crdt")]` unless a test genuinely needs CRDT.** CI never enables that feature, so a suite gated that way is written and then never run — 264 tests are currently dark for exactly diff --git a/src/terminal/session.rs b/src/terminal/session.rs index 6e71ea3..c731fb0 100644 --- a/src/terminal/session.rs +++ b/src/terminal/session.rs @@ -607,7 +607,7 @@ impl TerminalManager { let already = previously_reported.as_deref() == Some(spelling); let message = (!already).then(|| { format!( - "terminal.escape-key {spelling:?} is not a valid chord ({error}); using C-c" + "terminal.escape-key {spelling:?} is not a valid chord ({error}); using C-c" ) }); (fallback, Some(spelling.to_owned()), message) @@ -633,6 +633,24 @@ impl TerminalManager { self.escape_parses } + /// How many terminals currently hold a cached escape chord. + /// + /// The LIFETIME half of Q#TC4c's cache contract, which `escape_parses` + /// cannot cover: parse counting says a valid setting is read once, but + /// says nothing about whether the cache is ever released. Because the + /// cache lives on [`TerminalSession`], this count falls with the + /// session set by construction — which is exactly the property worth + /// pinning, since the rejected alternative (an editor-side + /// `HashMap`) has no purge hook and would hold + /// this at its high-water mark while sessions drained. + #[must_use] + pub fn escape_caches(&self) -> usize { + self.sessions + .values() + .filter(|session| session.escape.is_some()) + .count() + } + /// Resize a terminal screen and its PTY after validating shared limits. pub fn resize( &mut self, diff --git a/tests/terminal_config_acceptance.rs b/tests/terminal_config_acceptance.rs index a613eb2..ceeb8fe 100644 --- a/tests/terminal_config_acceptance.rs +++ b/tests/terminal_config_acceptance.rs @@ -32,13 +32,18 @@ fn eval_err(state: &EditorState, src: &str) -> String { } } -fn screen_text(state: &EditorState, buffer: pmacs::buffer::BufferId) -> String { - let manager = state.terminal_manager.borrow(); - let Some(snapshot) = manager.snapshot(buffer) else { - return String::new(); - }; +/// The viewport every test projects through. Deliberately SHORTER than +/// the 24-row screen a terminal opens with, so "scroll to the oldest +/// retained row" has somewhere to go even when nothing is retained — +/// which is what makes the two scrollback arms differ by content rather +/// than by whether scrolling was possible at all. +fn viewport() -> CellSize { + CellSize::new(10, 40) +} + +fn cells_to_text(cells: &[pmacs::cell::Cell]) -> String { let mut text = String::new(); - for cell in &snapshot.cells { + for cell in cells { match &cell.glyph { Glyph::Char(c) => text.push(*c), Glyph::Cluster(b) => text.push_str(&String::from_utf8_lossy(b)), @@ -48,6 +53,33 @@ fn screen_text(state: &EditorState, buffer: pmacs::buffer::BufferId) -> String { text } +fn screen_text(state: &EditorState, buffer: pmacs::buffer::BufferId) -> String { + let manager = state.terminal_manager.borrow(); + let Some(snapshot) = manager.snapshot(buffer) else { + return String::new(); + }; + cells_to_text(&snapshot.cells) +} + +/// Text a view actually shows, which is where retained history is +/// visible at all — the live `screen_text` above always reads the tail. +fn view_text(state: &EditorState, key: TerminalViewKey) -> String { + let mut manager = state.terminal_manager.borrow_mut(); + manager + .snapshot_for_view(key, viewport()) + .map(|snapshot| cells_to_text(&snapshot.cells)) + .unwrap_or_default() +} + +/// Scroll a view to its OLDEST retained row and read it back. +fn oldest_view_text(state: &EditorState, key: TerminalViewKey) -> String { + state + .terminal_manager + .borrow_mut() + .scroll_view(key, viewport(), i32::MAX); + view_text(state, key) +} + fn tick_until(state: &mut EditorState, needle: &str, buffer: pmacs::buffer::BufferId) -> bool { let deadline = Instant::now() + Duration::from_secs(5); loop { @@ -71,7 +103,7 @@ fn focus_terminal(state: &EditorState, buffer: pmacs::buffer::BufferId) -> Windo let mut manager = state.terminal_manager.borrow_mut(); manager.register_view(key); manager.claim_controller(key); - let _ = manager.snapshot_for_view(key, CellSize::new(10, 40)); + let _ = manager.snapshot_for_view(key, viewport()); window } @@ -223,6 +255,44 @@ fn acc2_unknown_profile_lists_known_names_and_creates_nothing() { assert_eq!(state.terminal_manager.borrow().len(), 0); } +/// Acceptance 2 (malformed table): `pmacs.terminal.profiles` is a raw +/// user table, so a diagnostic that walks its keys must be total over +/// them. A table holding both a string and a numeric key made +/// `table.sort` raise "attempt to compare number with string" — on the +/// unknown-profile path, replacing the exact error being asked for. +#[test] +fn acc2_malformed_profile_keys_do_not_mask_the_unknown_profile_error() { + let state = EditorState::new(); + exec(&state, CAT_PROFILE); + exec( + &state, + r#"pmacs.terminal.profiles[1] = { command = "/bin/sh" }"#, + ); + + let err = eval_err( + &state, + r#"return pmacs.terminal.open { profile = "ghost" }"#, + ); + assert!( + err.contains("ghost") && err.contains("echo"), + "the unknown-profile error must survive a malformed table: {err}" + ); + assert!( + !err.contains("attempt to compare"), + "listing known profiles must not raise: {err}" + ); + + // Rendering the REQUESTED name is partial too: `%q` raises on a + // table, and the name arrives straight from the caller. + let err = eval_err(&state, r"return pmacs.terminal.open { profile = {} }"); + assert!( + err.contains("is not defined") && err.contains("known profiles"), + "a non-string profile name must render, not raise: {err}" + ); + + assert_eq!(state.terminal_manager.borrow().len(), 0); +} + /// Acceptance 3: explicit beats profile beats setting beats `$SHELL`, and /// `env` MERGES rather than replacing. #[test] @@ -282,10 +352,77 @@ fn acc3_acc4_explicit_command_wins_and_empty_default_means_no_profile() { state.process_supervisor.borrow_mut().shutdown(); } -/// Acceptance 5: scrollback resolves from the setting, is overridden by an -/// explicit value, and `0` is legal. +/// A child that overflows the 24-row screen and then goes quiet, so its +/// early output can only still be found in RETAINED HISTORY. Zero-padded +/// so `LINE001` is not a substring of `LINE100`. +const FILL_PROFILE: &str = r#" +pmacs.terminal.profiles.fill = { + command = "/bin/sh", + args = { "-c", + "i=1; while [ $i -le 200 ]; do printf 'LINE%03d\r\n' $i; i=$((i+1)); done; printf 'DONE\r\n'; exec cat" }, +} +"#; + +/// Acceptance 5: the scrollback SETTING reaches the screen's retained +/// history, an explicit spec value overrides it, and `0` is legal. +/// +/// Asserted end to end, through a real child and a real view, rather +/// than by reading the value back out of the registry: a registry +/// round-trip is a test of the registry, and would stay green with the +/// setting's only consumer (`terminal.lua`'s `resolved.scrollback_rows` +/// fallback) deleted outright. #[test] -fn acc5_scrollback_setting_override_and_bounds() { +fn acc5_scrollback_setting_reaches_retained_history() { + let mut state = EditorState::new(); + exec(&state, FILL_PROFILE); + + // Arm 1: `0` is legal, and means the early rows are GONE. + exec(&state, r#"pmacs.config.set("terminal.scrollback-rows", 0)"#); + let none = open_cat_terminal(&state, r#"profile = "fill""#); + assert!(tick_until(&mut state, "DONE", none), "child finished"); + let window = focus_terminal(&state, none); + let none_key = TerminalViewKey::new(FrontendId::LOCAL, window, none); + let oldest = oldest_view_text(&state, none_key); + assert!( + !oldest.contains("LINE001"), + "with scrollback 0 the oldest retained row must not be the \ + child's first line: {oldest:?}" + ); + + // Arm 2: a large setting retains it, reachable by scrolling back. + exec( + &state, + r#"pmacs.config.set("terminal.scrollback-rows", 10000)"#, + ); + let kept = open_cat_terminal(&state, r#"profile = "fill""#); + assert!(tick_until(&mut state, "DONE", kept), "child finished"); + let window = focus_terminal(&state, kept); + let kept_key = TerminalViewKey::new(FrontendId::LOCAL, window, kept); + let oldest = oldest_view_text(&state, kept_key); + assert!( + oldest.contains("LINE001"), + "with scrollback 10000 the first line must survive in history: \ + {oldest:?}" + ); + + // Arm 3: an explicit spec value beats the setting, which is still 10000. + let overridden = open_cat_terminal(&state, r#"profile = "fill", scrollback_rows = 0"#); + assert!(tick_until(&mut state, "DONE", overridden), "child finished"); + let window = focus_terminal(&state, overridden); + let overridden_key = TerminalViewKey::new(FrontendId::LOCAL, window, overridden); + let oldest = oldest_view_text(&state, overridden_key); + assert!( + !oldest.contains("LINE001"), + "an explicit scrollback_rows = 0 must beat the setting: {oldest:?}" + ); + + state.process_supervisor.borrow_mut().shutdown(); +} + +/// Acceptance 5 (bounds): the registered range rejects out-of-range +/// values, and `0` is inside it rather than a disabled sentinel. +#[test] +fn acc5_scrollback_bounds() { let state = EditorState::new(); exec(&state, r#"pmacs.config.set("terminal.scrollback-rows", 0)"#); assert_eq!( @@ -397,6 +534,11 @@ fn acc7_acc8_acc8a_per_terminal_escape_cache_identity_and_lifecycle() { ); assert!(escape_was_armed(&mut state, b, 'N'), "B primes on its C-b"); let primed = state.terminal_manager.borrow().escape_parses(); + assert_eq!( + state.terminal_manager.borrow().escape_caches(), + 2, + "each primed terminal holds its own cache" + ); // Acceptance 7 — BOTH directions. Asserting only that A still works // after A->B->A is not enough: an epoch-only cache hands whichever @@ -442,6 +584,13 @@ fn acc7_acc8_acc8a_per_terminal_escape_cache_identity_and_lifecycle() { ); // Acceptance 8a: the cache dies with its terminal. + // + // Waiting for the SESSION count to fall is not the assertion — a + // session set that drains while an editor-side `HashMap` keeps its entry (the rejected implementation named + // in Q#TC4c, which has no purge hook) satisfies it exactly. The + // discriminating observable is the CACHE count, which such a map + // would hold at its high-water mark of 2. let sessions_before = state.terminal_manager.borrow().len(); exec(&state, "pmacs.terminal.terminate(TERM_A)"); exec(&state, "pmacs.buffer.kill(TERM_A)"); @@ -452,10 +601,31 @@ fn acc7_acc8_acc8a_per_terminal_escape_cache_identity_and_lifecycle() { state.tick_processes(); assert!( Instant::now() < deadline, - "killing the terminal must remove its session, and with it the cache" + "killing the terminal must remove its session" ); thread::sleep(Duration::from_millis(20)); } + assert_eq!( + state.terminal_manager.borrow().escape_caches(), + 1, + "killing terminal A must drop ITS cache, not merely its session" + ); + + // ...and the surviving cache is B's, so the right one was dropped. + focus_terminal(&state, b); + state.dispatch_key( + FrontendId::LOCAL, + KeyEvent::new(KeyCode::Char('b'), KeyModifiers::CONTROL), + ); + assert!( + escape_was_armed(&mut state, b, 'T'), + "terminal B must still escape on its own C-b after A was killed" + ); + assert_eq!( + state.terminal_manager.borrow().escape_parses(), + primed, + "B's surviving cache must not have been reparsed" + ); state.process_supervisor.borrow_mut().shutdown(); } From b76da70d5d6f7650f56e7c65b99bf64f3e5aff7b Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 21:53:41 -0400 Subject: [PATCH 19/32] =?UTF-8?q?feat(panel):=20route=20the=20Projection?= =?UTF-8?q?=20half=20of=20the=20=C2=A71.3=20census=20(Stage=202A)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Bottom-panel Stage 2A, first half. Every consumer the framing classifies **Projection** now resolves the frontend's primary document window or buffer instead of its focused one; every consumer classified focus, focus-chrome, or focus/session is deliberately left alone. No protocol change, no behavior change for any frontend today: with `panel_capable = false` for semantic sessions, `primary_document_window` returns `view.active` for every existing configuration, so this is a seam adoption that becomes load-bearing in 2B. Projection consumers routed: - **#1** semantic buffer-follow / `BufferSnapshot` re-send - **#2** the lazy CRDT upgrade — the sharpest case, since it BROADCASTS to every replica, so keying it on focus would let focusing a fresh generated panel buffer swap every peer's document mirror - **#3** `CursorByte` - **#4** `LineNumbers` mode - **#5** selection decorations - **#6/#10/#11** the full-window semantic terminal declaration, its snapshot/sync, and terminal-frame suppression, via the shared `semantic_terminal_key` resolver - **#9** the `Viewport` terminal-context gate — a focused terminal panel must not suppress the still-visible document's viewport - **#12** the semantic statusline target LOOKUP - **#21** the `BufferSnapshot` publication recipient filter `align_semantic_window_to_buffer` splits per Q#BP14, which is the distinction that makes rejecting panel-named events insufficient on its own: - `align_primary_document_window` (**#7**, `Viewport`) — aligns the document window and **never touches `view.active`**. - `align_and_activate_primary_document_window` (**#8**, `Pointer`) — aligns and then activates, because a click in the document area means "work here". This is the one place projection and focus legitimately move together. `dispatch_semantic_terminal_pointer` (**#11**) gains the same rule: an accepted non-`Move` gesture activates the document window before the gesture replays, while bare hover neither focuses nor claims. The statusline change is deliberately a HALF change (parent acceptance 42): the window LOOKUP resolves the primary document window, but `active` still reports **actual focus**, so a document provider can truthfully observe `active = false` while a panel owns focus. Untouched, and that is the load-bearing negative: #13 remote-op validation, #14 `dispatch_idle_for`, #15 presence, #16-#19 search / menu / minibuffer / completion chrome, #20 terminal bell drain, and #23 remote-op application all still resolve the actually focused window. #16-#19's Q#BP14b routing table needs `PanelFrame` and lands in 2B. 1,832 library tests pass; fmt and workspace clippy clean. Co-Authored-By: Claude Opus 5 (1M context) --- src/daemon.rs | 102 +++++++++++++++++++++++++++++++---------- src/editor.rs | 21 ++++++++- src/semantic_render.rs | 13 +++++- src/statusline.rs | 15 +++++- 4 files changed, 121 insertions(+), 30 deletions(-) diff --git a/src/daemon.rs b/src/daemon.rs index 9ac8256..19e5cc6 100644 --- a/src/daemon.rs +++ b/src/daemon.rs @@ -1145,9 +1145,13 @@ fn dispatcher_loop( .session_state(*fid) .is_some_and(|s| s.negotiated_capabilities.semantic_render) { + // Bottom-panel §1.3 #1 — Projection. The buffer this + // frontend DISPLAYS AS ITS DOCUMENT, not the one it + // happens to focus: focusing a panel must re-send no + // snapshot and must never swap the replica's mirror. let active_now = { let core = editor.core.borrow(); - core.active_window_for(*fid).map(|w| w.buffer_id) + core.primary_document_buffer(*fid) }; if let Some(active_now) = active_now && last_active_buffer_sent.get(fid) != Some(&active_now) @@ -1430,8 +1434,16 @@ fn dispatcher_loop( .session_state(*fid) .is_some_and(|s| s.negotiated_capabilities.crdt_replica) { + // Bottom-panel §1.3 #3 — Projection. `CursorByte` is + // the replica's authoritative DOCUMENT cursor; a + // focused panel must not retarget it at the panel + // buffer (Q#BP14's "active buffer is a + // document-surface term, not an input-focus term"). let core = editor.core.borrow(); - if let Some(window) = core.active_window_for(*fid) { + if let Some(window) = core + .primary_document_window(*fid) + .and_then(|win_id| core.windows.get(&win_id)) + { let cursor_byte_msg = InstanceMessage::CursorByte { buffer_id: window.buffer_id, byte_pos: window.cursor, @@ -2038,12 +2050,17 @@ fn handle_dispatcher_event( // straight back off it. The declared buffer is // checked too — a terminal has no byte viewport to // honor from any direction. + // Bottom-panel §1.3 #9 — Projection. The gate asks + // "is this frontend's DOCUMENT surface a terminal", + // so it tests the primary document window. A focused + // TERMINAL PANEL must not suppress the still-visible + // document's viewport. let terminal_context = { let manager = editor.terminal_manager.borrow(); let core = editor.core.borrow(); let active = core - .active_window_for(source) - .is_some_and(|window| manager.is_terminal(window.buffer_id)); + .primary_document_buffer(source) + .is_some_and(|document| manager.is_terminal(document)); active || manager.is_terminal(buffer_id) }; if semantic_states.contains_key(&source) && !terminal_context { @@ -2056,7 +2073,11 @@ fn handle_dispatcher_event( // LOCAL's attach-time buffer (often a scratch the // user isn't viewing), so arrow keys moved an // off-screen cursor and the caret never tracked. - align_semantic_window_to_buffer(editor, source, buffer_id); + // Bottom-panel §1.3 #7 — Projection, and it must + // NOT move focus. Routing this through the + // focused window would let an ordinary document + // viewport overwrite a focused panel's buffer. + align_primary_document_window(editor, source, buffer_id); if let Some(sem) = semantic_states.get_mut(&source) { sem.set_viewport(buffer_id, visible, generation); } @@ -2121,7 +2142,11 @@ fn handle_dispatcher_event( // aligns to the buffer the frontend says it was // displaying: a click can race a buffer switch. if semantic_states.contains_key(&source) { - align_semantic_window_to_buffer(editor, source, buffer_id); + // Bottom-panel §1.3 #8 — Projection + focus. A + // click in the DOCUMENT area means "work here", + // so unlike `Viewport` (#7) this one also takes + // focus out of a panel. + align_and_activate_primary_document_window(editor, source, buffer_id); if kind == PointerKind::Context { // Q#CM1 — right-click opens the context menu // at the hit byte (needs the Lua builder, so @@ -2359,9 +2384,15 @@ fn ensure_active_buffer_crdt_backed( editor: &EditorState, fid: FrontendId, ) -> Option { + // Bottom-panel §1.3 #2 — Projection, and the sharpest case in the + // census. The upgrade BROADCASTS a `BufferSnapshot` to every + // replica, so keying it on focus would mean focusing a fresh + // generated panel buffer swaps every peer's document mirror to it. + // A panel buffer that genuinely needs CRDT backing gets it when it + // is displayed as a document, not as a side effect of focus. let buffer_id_opt = { let core = editor.core.borrow(); - core.active_window_for(fid).map(|w| w.buffer_id) + core.primary_document_buffer(fid) }; let buffer_id = buffer_id_opt?; let core = editor.core.borrow(); @@ -2493,11 +2524,13 @@ fn publish_buffer_snapshot_to_replicas( continue; } if session.negotiated_capabilities.semantic_render { - let displays_buffer = editor - .core - .borrow() - .active_window_for(*peer_id) - .is_some_and(|window| window.buffer_id == buffer_id); + // Bottom-panel §1.3 #21 — Projection. "Displays this + // buffer" means the peer's DOCUMENT surface: testing the + // focused window would both miss a buffer visible in the + // document (panel focused elsewhere) and replace the peer's + // mirror for one visible only in a panel. + let displays_buffer = + editor.core.borrow().primary_document_buffer(*peer_id) == Some(buffer_id); if !displays_buffer { continue; } @@ -2936,32 +2969,35 @@ fn handle_remote_crdt_op( /// whole switch. This is the input/display alignment fix for B1: the /// frontend's *declared* buffer becomes the buffer its keys edit and /// its `CursorByte` reports. -fn align_semantic_window_to_buffer( +/// Align a semantic frontend's **primary document window** to the +/// buffer it declared (bottom-panel §1.3 #7, Q#BP14). +/// +/// **Never touches `view.active`.** This is why rejecting panel-named +/// events does not fix the *document* event: with a panel focused, an +/// ordinary document `Viewport` routed through the focused window would +/// overwrite the panel's buffer with the document buffer. Returns the +/// window it aligned so the `Pointer` path (#8) can activate it. +fn align_primary_document_window( editor: &mut EditorState, fid: FrontendId, buffer_id: crate::buffer::BufferId, -) { +) -> Option { use crate::text_view::TextView; - let text_view = { + let (win_id, text_view) = { let core = editor.core.borrow(); - let Some(win_id) = core.views.get(&fid).map(|v| v.active) else { - return; - }; + let win_id = core.primary_document_window(fid)?; if core.windows.get(&win_id).map(|w| w.buffer_id) == Some(buffer_id) { - return; // Already displaying this buffer. + return Some(win_id); // Already displaying this buffer. } let reg = core.registry.borrow(); let Ok(buf) = reg.get(buffer_id) else { - return; // Unknown buffer — leave the window as-is. + return Some(win_id); // Unknown buffer — leave the window as-is. }; - TextView::new(buf) + (win_id, TextView::new(buf)) }; let mut core = editor.core.borrow_mut(); - let Some(win_id) = core.views.get(&fid).map(|v| v.active) else { - return; - }; if let Some(win) = core.windows.get_mut(&win_id) { win.buffer_id = buffer_id; win.text_view = text_view; @@ -2969,6 +3005,24 @@ fn align_semantic_window_to_buffer( win.selection = None; win.overlays.clear(); } + Some(win_id) +} + +/// Align the primary document window **and take focus to it** +/// (bottom-panel §1.3 #8, Q#BP14). +/// +/// A click in the document area means "work here", so it moves focus +/// out of a panel. This is the one place projection and focus +/// legitimately move together — every other Projection consumer must +/// use [`align_primary_document_window`] alone. +fn align_and_activate_primary_document_window( + editor: &mut EditorState, + fid: FrontendId, + buffer_id: crate::buffer::BufferId, +) { + if let Some(win_id) = align_primary_document_window(editor, fid, buffer_id) { + editor.core.borrow_mut().focus_window(fid, win_id); + } } fn build_fresh_frontend_view( diff --git a/src/editor.rs b/src/editor.rs index 1db5c3a..8bc9d8f 100644 --- a/src/editor.rs +++ b/src/editor.rs @@ -1341,9 +1341,16 @@ impl EditorState { frontend_id: FrontendId, buffer_id: crate::buffer::BufferId, ) -> Option { + // Bottom-panel §1.3 #6/#10/#11 — Projection. The full-window + // semantic terminal declaration, its snapshot/sync, and its + // frame suppression all describe the frontend's PRIMARY DOCUMENT + // surface, never a panel band: panel terminals get `PanelFrame` + // / `PanelPointer` in Stage 2B instead. Resolving through + // `view.active` would let a focused panel terminal both claim + // the document declaration and suppress the document pass. let core = self.core.borrow(); - let view = core.views.get(&frontend_id)?; - let window = core.windows.get(&view.active)?; + let win_id = core.primary_document_window(frontend_id)?; + let window = core.windows.get(&win_id)?; if window.buffer_id != buffer_id { return None; } @@ -1472,6 +1479,16 @@ impl EditorState { if coord.row >= size.rows || coord.col >= size.cols { return false; } + // Bottom-panel §1.3 #11 — Projection + focus. A non-hover + // gesture on the DOCUMENT terminal means "work here", so it + // takes focus back out of a panel before the gesture replays; + // bare hover neither focuses nor claims the controller. + if !matches!(kind, TerminalMouseKind::Move) { + let mut core = self.core.borrow_mut(); + if let Some(win_id) = core.primary_document_window(frontend_id) { + core.focus_window(frontend_id, win_id); + } + } self.core.borrow_mut().active_frontend = frontend_id; self.apply_terminal_gesture(key, size, coord, kind, mods, (coord.row, coord.col)); true diff --git a/src/semantic_render.rs b/src/semantic_render.rs index 65750d3..617d408 100644 --- a/src/semantic_render.rs +++ b/src/semantic_render.rs @@ -1345,9 +1345,13 @@ impl SemanticRenderState { state: &EditorState, buffer_id: BufferId, ) -> Option { + // Bottom-panel §1.3 #4 — Projection. `LineNumbers` describes the + // replica's DOCUMENT surface; a focused panel must not replace + // the document's gutter mode with the panel window's. let mode = { let core = state.core.borrow(); - core.active_window_for(self.frontend_id) + core.primary_document_window(self.frontend_id) + .and_then(|win_id| core.windows.get(&win_id)) .map_or(crate::window::LineNumberMode::Off, |w| w.line_numbers) }; if self.last_line_numbers == Some(mode) { @@ -1703,7 +1707,12 @@ impl SemanticRenderState { // Emitting CurrentLine here forced a whole-buffer line table on // every frame even though pmacs-gpu ignores its own current-line // wash. - if let Some(win) = core.active_window_for(self.frontend_id) + // Bottom-panel §1.3 #5 — Projection. Selection decorations + // belong to the document surface the viewport describes; a + // selection made inside a focused panel must not paint into it. + if let Some(win) = core + .primary_document_window(self.frontend_id) + .and_then(|win_id| core.windows.get(&win_id)) && win.buffer_id == vp.buffer_id && let Some((lo, hi)) = win.region() && let Some(range) = clip_to_viewport(lo, hi, vp) diff --git a/src/statusline.rs b/src/statusline.rs index d11c885..1e6d9cc 100644 --- a/src/statusline.rs +++ b/src/statusline.rs @@ -639,9 +639,20 @@ fn capture_target_contexts( .views .get(&frontend_id) .ok_or(StatuslineNoMessageReason::ContextUnavailable)?; + // Bottom-panel §1.3 #12 — Projection. This LOOKUP resolves + // the primary document window: with a panel focused, + // `view.active` would name the panel and the declared-buffer + // check would clear the document's statusline. + // + // `active` is NOT rerouted with it (Q#BP14/parent 42): it + // reports ACTUAL focus, so a document provider truthfully + // observes `active = false` while the panel owns focus. + let window_id = core + .primary_document_window(frontend_id) + .ok_or(StatuslineNoMessageReason::ContextUnavailable)?; let window = core .windows - .get(&view.active) + .get(&window_id) .ok_or(StatuslineNoMessageReason::ContextUnavailable)?; if buffers.get(window.buffer_id).is_err() { return Err(StatuslineNoMessageReason::BufferUnavailable); @@ -653,7 +664,7 @@ fn capture_target_contexts( frontend_id, window_id: window.id, buffer_id: window.buffer_id, - active: true, + active: window.id == view.active, }]) } } From d7ad01b53596dc13a3e13faddd791b7d4432b398 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 22:04:56 -0400 Subject: [PATCH 20/32] feat(panel): extract the per-window painter + Stage 2A acceptance MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Bottom-panel Stage 2A, second half (Q#BP8, Q#BP17). Still no protocol change and no behavior change: `paint_frame` builds the same fold map it always did and passes it in, so grid rendering is unchanged. Two extractions, both taking the fold map as a **parameter** rather than building it: - `prepare_window_cursor_visible` — the active-window auto-scroll clamp. The panel band (2B) runs this for its own window when that window owns focus, and leaves a passive panel's `view_top` alone. - `paint_window_content` — the per-window document body: text, gutter, overlays, selection, and the mode line. The panel paints into a panel-sized grid at the same origin-agnostic `Viewport`, so this is that body lifted out, not a second painter (Bet B2'). The parameter is the point (Q#BP17). Folding built its per-window map ungated on the premise that "a semantic session never enters `paint_frame`", which the panel band breaks. The panel path must pass `None` for a frontend whose `fold_projection` is false, and must not call `EditorCore::fold_map_for_window` — that gates on the **active** frontend, which is right for command-time reckoning and wrong for painting another frontend's panel. `tests/bottom_panel_stage2a_acceptance.rs` — 10 tests. The negative half is the load-bearing half, so Projection assertions are paired with focus-class assertions taken in the SAME state: - `focus_and_projection_disagree_in_the_same_state` is the key one: with a panel focused, the focus authority must name the panel while the projection authority names the document. Routing the focus class through `primary_document_window` fails this even though every Projection test still passes. - The statusline pair pins the split: the LOOKUP resolves the document window while `active` reports actual focus, with a non-vacuity twin that flips `active` back to true when focus returns. - The extraction pair pins cells, the returned cursor, the focused window's `view_top`, AND a passive window's untouched scroll — identical cells alone would not catch a clamp that moved to the wrong window on a single-window frame. - `the_panel_fixture_really_builds_a_side_window` pins the fixture's own precondition, since every other test is worthless if `focused_panel` silently produced an ordinary split. One crdt-gated caller of the old `align_semantic_window_to_buffer` was updated; it compiles only under `--features crdt`, which is the config CI never runs. 1,832 default + 2,009 CRDT library tests, 10 new acceptance; fmt and workspace clippy clean. Co-Authored-By: Claude Opus 5 (1M context) --- src/daemon.rs | 6 +- src/editor.rs | 307 ++++++++++-------- tests/bottom_panel_stage2a_acceptance.rs | 377 +++++++++++++++++++++++ 3 files changed, 556 insertions(+), 134 deletions(-) create mode 100644 tests/bottom_panel_stage2a_acceptance.rs diff --git a/src/daemon.rs b/src/daemon.rs index 19e5cc6..16e66ab 100644 --- a/src/daemon.rs +++ b/src/daemon.rs @@ -4439,7 +4439,7 @@ mod tests { /// B1 input/display alignment: a semantic frontend's window is bound /// to LOCAL's attach-time buffer, but the buffer it *displays* is - /// the one it declares via `Viewport`. `align_semantic_window_to_buffer` + /// the one it declares via `Viewport`. `align_primary_document_window` /// re-points the window so keys edit the displayed buffer — without /// it, arrow keys moved an off-screen cursor in the wrong buffer and /// the caret never tracked. @@ -4477,7 +4477,9 @@ mod tests { ); // The frontend declares it is displaying the file buffer. - align_semantic_window_to_buffer(&mut editor, fid, file); + // Bottom-panel §1.3 #7: `Viewport` takes the projection-only + // aligner, which never touches `view.active`. + align_primary_document_window(&mut editor, fid, file); assert_eq!( editor .core diff --git a/src/editor.rs b/src/editor.rs index 8bc9d8f..2553775 100644 --- a/src/editor.rs +++ b/src/editor.rs @@ -3169,6 +3169,170 @@ impl CompletionPopupKey { } } +/// Scroll one window so its cursor stays visible, reckoning in +/// **visible** lines when a fold map is supplied (Arc 6 Q#FD18). +/// +/// Extracted from `paint_frame` for bottom-panel Stage 2 (Q#BP8): the +/// panel band runs this for its own window when that window owns focus, +/// against the same supplied map, and leaves a passive panel's +/// `view_top` untouched. +/// +/// **The fold map is a parameter, never built here (Q#BP17).** A panel +/// painted for a frontend whose `fold_projection` is false must pass +/// `None`; `EditorCore::fold_map_for_window` is the wrong source there +/// because it gates on the **active** frontend, which is right for +/// command-time reckoning and wrong for painting another frontend's +/// panel. +fn prepare_window_cursor_visible( + window: &mut crate::window::Window, + buf: &crate::buffer::Buffer, + inner_rows: u32, + folds: Option<&crate::fold_view::VisibleLineMap>, +) { + let cursor_row = window + .text_view + .pos_to_display(buf, window.cursor) + .map_or(0, |d| d.row as usize); + match folds { + // The logical cursor may sit on a hidden line (a shared fold, or + // goto-line into one); the row that actually renders — and so + // the row to scroll to — is its visible head (Q#FD16/FD18, + // framing acceptance 8). + Some(map) => { + let anchor = map.visible_head_of(cursor_row); + let top = map.clamp_view_top(window.view_top); + window.view_top = if anchor < top { + anchor + } else if inner_rows > 0 && map.visible_rows_between(top, anchor) >= inner_rows as usize + { + map.nth_visible_back(anchor, inner_rows as usize - 1) + } else { + top + }; + } + None => { + if cursor_row < window.view_top { + window.view_top = cursor_row; + } else if inner_rows > 0 && cursor_row >= window.view_top + inner_rows as usize { + window.view_top = cursor_row + 1 - inner_rows as usize; + } + } + } +} + +/// Paint one window's document content: text, gutter, overlays, +/// selection, and its mode line. +/// +/// Extracted from `paint_frame`'s per-window loop for bottom-panel +/// Stage 2 (Q#BP8) — the panel band paints its window into a +/// panel-sized grid at the same origin-agnostic `Viewport`, so this is +/// that body lifted out rather than a second painter. No concrete +/// text/gutter/overlay/mode-line painter forks (Bet B2'). +/// +/// **`folds` is a parameter, never built here (Q#BP17).** Folding's +/// "a semantic session never enters `paint_frame`" premise is what the +/// panel band breaks; the panel path passes `None` when the owning +/// frontend's `fold_projection` is false, and must not call +/// `EditorCore::fold_map_for_window`, which gates on the **active** +/// frontend. +#[allow(clippy::too_many_arguments)] +fn paint_window_content( + grid: &mut crate::cell::CellGrid<'_>, + window: &mut crate::window::Window, + buf: &crate::buffer::Buffer, + placement: WindowPlacement, + folds: Option<&crate::fold_view::VisibleLineMap>, + focused: bool, + theme: &crate::highlight::Theme, + statusline: Option<&crate::statusline::StatuslineWindowSegments>, + diag_store: &std::sync::Arc>, +) { + let rect = placement.outer; + let inner_rows = placement.content.size.rows; + if let Some(map) = folds { + window.view_top = map.clamp_view_top(window.view_top); + } + let viewport_buffer_start = window.text_view.line_offset(window.view_top).unwrap_or(0); + // UX gutter (Q#UX2): reserve a left strip for line numbers and + // shrink+shift the text area into the remainder, so every + // viewport-relative painter (text, syntax, diagnostics, search) + // stays gutter-agnostic. A window too narrow for the gutter falls + // back to no gutter this frame rather than starving the text. + let gutter_w = { + let w = window.gutter_width(); + if w >= rect.size.cols { 0 } else { w } + }; + let viewport = Viewport { + buffer_start: viewport_buffer_start, + buffer_end: buf.len(), + cell_origin: CellCoord::new(rect.origin.row, rect.origin.col + gutter_w), + cell_size: crate::cell::CellSize::new(inner_rows, rect.size.cols - gutter_w), + gutter_w, + folds, + }; + // Composition (T M2.9): base text_view paints first, then the + // gutter numbers — before the overlays, so a diagnostic overlay + // can draw its severity sign into the gutter's leading column + // without the gutter's own blank pass erasing it — then each + // overlay in attach order. See [`crate::view::View`]. + window.text_view.render(buf, viewport, grid); + if gutter_w > 0 { + paint_line_number_gutter(grid, window, &rect, inner_rows, gutter_w, folds, theme); + } + for overlay in &mut window.overlays { + overlay.render(buf, viewport, grid); + } + paint_local_selection(grid, buf, window, &rect, inner_rows, gutter_w, folds, theme); + // Mode line for this window. Painted last so the line + // itself is always visible regardless of overlay activity. + let coord = window + .text_view + .pos_to_display(buf, window.cursor) + .unwrap_or_default(); + // Arc 6 Stage 2 (Q#FD18): All/Top/Bot/% are reckoned in + // VISIBLE-line space — a buffer whose remainder is collapsed + // reads "All", not "Top". The cursor's ordinal anchors on its + // visible head, since that is the row it renders on. + let (ind_top, ind_total, ind_cursor) = match folds { + Some(map) => ( + map.visible_rows_between(0, window.view_top), + map.visible_line_count(window.text_view.line_count()), + map.visible_rows_between(0, map.visible_head_of(coord.row as usize)), + ), + None => ( + window.view_top, + window.text_view.line_count(), + coord.row as usize, + ), + }; + let scroll = format_scroll_indicator(ind_top, inner_rows as usize, ind_total, ind_cursor); + // Lock scoped to the summary computation only: the overlay + // renders above include `DiagnosticView`, which takes this + // same mutex — holding the guard across the loop deadlocked + // the daemon on the first frame after a file (and thus a + // diagnostic overlay) was opened. + let diags = { + let guard = diag_store.lock().expect("diag store mutex poisoned"); + diag_mode_line_summary(&guard, buf) + }; + let custom = statusline; + paint_mode_line( + grid, + &rect, + buf.name(), + buf.is_modified(), + focused, + coord.row, + coord.col, + &scroll, + &diags, + mode_line_style(theme), + custom.map_or(&[], |segments| segments.left.as_slice()), + custom.map_or(&[], |segments| segments.right.as_slice()), + theme, + ); +} + /// Paint one full frame into `grid` and return the desired terminal /// cursor position. /// @@ -3278,6 +3442,11 @@ pub fn paint_frame( // Arc 6 Stage 2 (Q#FD18): the auto-scroll clamp reckons in // VISIBLE lines. Built from the active window itself, before // the mutable borrow below. + // + // Bottom-panel Q#BP17: built HERE and passed in, because the + // panel path (Stage 2B) must supply `None` for a frontend + // whose `fold_projection` is false. Building it inside the + // clamp would hard-wire the grid's answer. let folds = core .windows .get(&active) @@ -3285,36 +3454,7 @@ pub fn paint_frame( let aw = core.windows.get_mut(&active).expect( "invariant: active_window_id always references a live window in core.windows", ); - let cursor_row = aw - .text_view - .pos_to_display(buf, aw.cursor) - .map_or(0, |d| d.row as usize); - match folds.as_ref() { - // The logical cursor may sit on a hidden line (a shared - // fold, or goto-line into one); the row that actually - // renders — and so the row to scroll to — is its visible - // head (Q#FD16/FD18, framing acceptance 8). - Some(map) => { - let anchor = map.visible_head_of(cursor_row); - let top = map.clamp_view_top(aw.view_top); - aw.view_top = if anchor < top { - anchor - } else if inner_rows > 0 - && map.visible_rows_between(top, anchor) >= inner_rows as usize - { - map.nth_visible_back(anchor, inner_rows as usize - 1) - } else { - top - }; - } - None => { - if cursor_row < aw.view_top { - aw.view_top = cursor_row; - } else if inner_rows > 0 && cursor_row >= aw.view_top + inner_rows as usize { - aw.view_top = cursor_row + 1 - inner_rows as usize; - } - } - } + prepare_window_cursor_visible(aw, buf, inner_rows, folds.as_ref()); } } @@ -3366,114 +3506,17 @@ pub fn paint_frame( let Ok(buf) = reg.get(window.buffer_id) else { continue; }; - // Arc 6 Stage 2 (Q#FD12, round-2 F2): ONE visible-line map per - // rendered document window, keyed on that window's own buffer and - // line offsets. A split may show different buffers with only one - // folded, so a per-frame singleton would leak one pane's folds - // into the other. `None` when this buffer has no folds — the - // unfolded path then paints exactly as before. let folds = crate::fold_view::map_for_window(&state.fold_registry, window); - // `view_top` stays a source-line index (Bet B5) but must never - // rest on a hidden line: clamp BACKWARD so a fold at the top of - // the viewport shows its head (Q#FD18, acceptance 8). - if let Some(map) = folds.as_ref() { - window.view_top = map.clamp_view_top(window.view_top); - } - let viewport_buffer_start = window.text_view.line_offset(window.view_top).unwrap_or(0); - // UX gutter (Q#UX2): reserve a left strip for line numbers and - // shrink+shift the text area into the remainder, so every - // viewport-relative painter (text, syntax, diagnostics, search) - // stays gutter-agnostic. A window too narrow for the gutter falls - // back to no gutter this frame rather than starving the text. - let gutter_w = { - let w = window.gutter_width(); - if w >= rect.size.cols { 0 } else { w } - }; - let viewport = Viewport { - buffer_start: viewport_buffer_start, - buffer_end: buf.len(), - cell_origin: CellCoord::new(rect.origin.row, rect.origin.col + gutter_w), - cell_size: crate::cell::CellSize::new(inner_rows, rect.size.cols - gutter_w), - gutter_w, - folds: folds.as_ref(), - }; - // Composition (T M2.9): base text_view paints first, then the - // gutter numbers — before the overlays, so a diagnostic overlay - // can draw its severity sign into the gutter's leading column - // without the gutter's own blank pass erasing it — then each - // overlay in attach order. See [`crate::view::View`]. - window.text_view.render(buf, viewport, grid); - if gutter_w > 0 { - paint_line_number_gutter( - grid, - window, - &rect, - inner_rows, - gutter_w, - folds.as_ref(), - &theme, - ); - } - for overlay in &mut window.overlays { - overlay.render(buf, viewport, grid); - } - paint_local_selection( + paint_window_content( grid, - buf, window, - &rect, - inner_rows, - gutter_w, + buf, + placement, folds.as_ref(), - &theme, - ); - // Mode line for this window. Painted last so the line - // itself is always visible regardless of overlay activity. - let coord = window - .text_view - .pos_to_display(buf, window.cursor) - .unwrap_or_default(); - // Arc 6 Stage 2 (Q#FD18): All/Top/Bot/% are reckoned in - // VISIBLE-line space — a buffer whose remainder is collapsed - // reads "All", not "Top". The cursor's ordinal anchors on its - // visible head, since that is the row it renders on. - let (ind_top, ind_total, ind_cursor) = match folds.as_ref() { - Some(map) => ( - map.visible_rows_between(0, window.view_top), - map.visible_line_count(window.text_view.line_count()), - map.visible_rows_between(0, map.visible_head_of(coord.row as usize)), - ), - None => ( - window.view_top, - window.text_view.line_count(), - coord.row as usize, - ), - }; - let scroll = format_scroll_indicator(ind_top, inner_rows as usize, ind_total, ind_cursor); - // Lock scoped to the summary computation only: the overlay - // renders above include `DiagnosticView`, which takes this - // same mutex — holding the guard across the loop deadlocked - // the daemon on the first frame after a file (and thus a - // diagnostic overlay) was opened. - let diags = { - let guard = diag_store.lock().expect("diag store mutex poisoned"); - diag_mode_line_summary(&guard, buf) - }; - let custom = statusline_by_window.get(id); - paint_mode_line( - grid, - &rect, - buf.name(), - buf.is_modified(), *id == active, - coord.row, - coord.col, - &scroll, - &diags, - mode_line_style(&theme), - custom.map_or(&[], |segments| segments.left.as_slice()), - custom.map_or(&[], |segments| segments.right.as_slice()), &theme, + statusline_by_window.get(id), + &diag_store, ); } drop(reg); diff --git a/tests/bottom_panel_stage2a_acceptance.rs b/tests/bottom_panel_stage2a_acceptance.rs new file mode 100644 index 0000000..5f7996d --- /dev/null +++ b/tests/bottom_panel_stage2a_acceptance.rs @@ -0,0 +1,377 @@ +// bottom_panel_stage2a_acceptance.rs --- bottom-panel Stage 2A +// (docs/bottom-panel-stage2-framing.md, criteria A2A-1 / A2A-2 / A2A-3). + +//! Classified §1.3 census routing + the per-window painter extraction. +//! No wire change. +//! +//! **The negative half is the load-bearing half.** A suite that only +//! proved "the document surface is used" would pass with the focus, +//! focus-chrome, and focus/session consumers *wrongly* rerouted to the +//! document — which is the defect the framing spent three review rounds +//! eliminating, and which would break remote-op validation, +//! `DispatchIdle`, presence, focused search/menu/completion routing, and +//! terminal bell ownership. So every Projection assertion here is paired +//! with a focus-class assertion taken in the *same* state. + +use pmacs::cell::{CellGrid, CellSize}; +use pmacs::editor::EditorState; +use pmacs::protocol::FrontendId; +use pmacs::window::{Side, WindowId}; + +const ROWS: u32 = 24; +const COLS: u32 = 60; + +fn editor() -> EditorState { + let s = EditorState::new(); + exec(&s, "pmacs.lsp.config = {}"); + s.sync_frame_geometry(FrontendId::LOCAL, CellSize::new(ROWS, COLS)); + s +} + +fn exec(s: &EditorState, src: &str) { + s.lua_host.lua().load(src.to_string()).exec().unwrap(); +} + +fn side_window(s: &EditorState) -> Option { + let core = s.core.borrow(); + core.views[&FrontendId::LOCAL] + .layout + .iter_ids() + .into_iter() + .find(|id| { + core.windows + .get(id) + .is_some_and(|w| w.params.side.is_some()) + }) +} + +/// Open a bottom panel and leave it FOCUSED — the state in which every +/// classification difference becomes observable. +fn focused_panel(s: &EditorState) -> (WindowId, WindowId) { + let document = s.core.borrow().views[&FrontendId::LOCAL].active; + exec( + s, + "PANEL_BUF = pmacs.buffer.create(\"*panel*\") + PANEL_WIN = pmacs.window.display(PANEL_BUF, \ + { side = \"bottom\", height = 4 })", + ); + let panel = side_window(s).expect("panel exists"); + s.core.borrow_mut().focus_window(FrontendId::LOCAL, panel); + assert_eq!( + s.core.borrow().views[&FrontendId::LOCAL].active, + panel, + "fixture precondition: the panel must own focus" + ); + (document, panel) +} + +fn render(s: &EditorState) { + let size = CellSize::new(ROWS, COLS); + let mut cells = vec![pmacs::cell::Cell::default(); (ROWS * COLS) as usize]; + let mut grid = CellGrid { + cells: &mut cells, + stride: size.cols, + size, + }; + let _ = pmacs::editor::paint_frame( + s, + FrontendId::LOCAL, + &std::collections::HashMap::new(), + &mut grid, + size, + ); +} + +// --------------------------------------------------------------------------- +// A2A-1 — the Projection class resolves the document surface +// --------------------------------------------------------------------------- + +#[test] +fn projection_resolves_the_document_window_while_a_panel_is_focused() { + let s = editor(); + let (document, panel) = focused_panel(&s); + let core = s.core.borrow(); + + assert_eq!( + core.primary_document_window(FrontendId::LOCAL), + Some(document), + "Projection consumers must resolve the document window, not the focused panel" + ); + assert_ne!(document, panel); +} + +#[test] +fn projection_buffer_is_the_document_buffer_not_the_panel_buffer() { + let s = editor(); + let (document, _panel) = focused_panel(&s); + let core = s.core.borrow(); + + let document_buffer = core.windows[&document].buffer_id; + assert_eq!( + core.primary_document_buffer(FrontendId::LOCAL), + Some(document_buffer), + "the replica's document mirror must not follow panel focus" + ); + assert_ne!( + core.primary_document_buffer(FrontendId::LOCAL), + Some(core.windows[&core.views[&FrontendId::LOCAL].active].buffer_id), + "non-vacuity: the focused window's buffer differs, so this test can fail" + ); +} + +// --------------------------------------------------------------------------- +// A2A-1 — the NEGATIVE half: focus classes still resolve focus +// --------------------------------------------------------------------------- + +#[test] +fn focus_class_dispatch_idle_still_tracks_the_focused_window() { + let s = editor(); + let (_document, _panel) = focused_panel(&s); + + // §1.3 #14 — Focus. Q#BP14a: optimistic input is gated per WINDOW. + // A panel that owns focus must suppress `DispatchIdle` even though + // the *document* projection is unaffected. + assert!( + !s.dispatch_idle_for(FrontendId::LOCAL), + "a focused side window must gate optimistic input off (#14)" + ); +} + +#[test] +fn focus_class_gate_lifts_when_focus_returns_to_the_document() { + let s = editor(); + let (document, _panel) = focused_panel(&s); + s.core + .borrow_mut() + .focus_window(FrontendId::LOCAL, document); + + assert!( + s.dispatch_idle_for(FrontendId::LOCAL), + "non-vacuity: the gate must lift with focus, or the test above proves nothing" + ); +} + +#[test] +fn focus_and_projection_disagree_in_the_same_state() { + // The single most important assertion in this suite: in ONE state, + // the two classes must resolve DIFFERENT windows. If a future change + // routes the focus class through `primary_document_window`, this + // fails even though every Projection test above still passes. + let s = editor(); + let (document, panel) = focused_panel(&s); + let core = s.core.borrow(); + + let focused = core.views[&FrontendId::LOCAL].active; + let projected = core + .primary_document_window(FrontendId::LOCAL) + .expect("a document window exists"); + + assert_eq!(focused, panel, "focus authority must name the panel"); + assert_eq!( + projected, document, + "projection authority must name the document" + ); + assert_ne!( + focused, projected, + "the two authorities must be genuinely distinct in this state" + ); +} + +// --------------------------------------------------------------------------- +// A2A-2 — the statusline split: lookup reroutes, `active` does not +// --------------------------------------------------------------------------- + +#[test] +fn statusline_document_context_reports_active_false_under_a_focused_panel() { + use pmacs::statusline::{ + StatuslineEvaluationOutcome, StatuslineEvaluationTarget, evaluate_statusline, + }; + + let s = editor(); + let (document, _panel) = focused_panel(&s); + let declared = s.core.borrow().windows[&document].buffer_id; + + let evaluation = evaluate_statusline( + s.lua_host.lua(), + &s.core, + &s.statusline_registry, + StatuslineEvaluationTarget::Semantic { + frontend_id: FrontendId::LOCAL, + declared_buffer: declared, + }, + ); + + match evaluation.outcome { + StatuslineEvaluationOutcome::Ready(windows) => { + let context = windows + .first() + .map(|segments| segments.context) + .expect("one document context"); + // The LOOKUP rerouted: it resolved the document window even + // though the panel is focused (§1.3 #12). + assert_eq!( + context.window_id, document, + "the semantic target must resolve the primary document window" + ); + // `active` did NOT reroute (parent acceptance 42): a document + // provider observes the truth, that it is not focused. + assert!( + !context.active, + "a document provider must observe active = false while the panel owns focus" + ); + } + other => panic!("expected a ready evaluation, got {other:?}"), + } +} + +#[test] +fn statusline_document_context_is_active_when_the_document_is_focused() { + use pmacs::statusline::{ + StatuslineEvaluationOutcome, StatuslineEvaluationTarget, evaluate_statusline, + }; + + // Non-vacuity for the assertion above: with focus on the document, + // the same context must report `active = true`. + let s = editor(); + let (document, _panel) = focused_panel(&s); + s.core + .borrow_mut() + .focus_window(FrontendId::LOCAL, document); + let declared = s.core.borrow().windows[&document].buffer_id; + + let evaluation = evaluate_statusline( + s.lua_host.lua(), + &s.core, + &s.statusline_registry, + StatuslineEvaluationTarget::Semantic { + frontend_id: FrontendId::LOCAL, + declared_buffer: declared, + }, + ); + + match evaluation.outcome { + StatuslineEvaluationOutcome::Ready(windows) => { + let context = windows.first().map(|s| s.context).expect("one context"); + assert!(context.active, "a focused document context must be active"); + } + other => panic!("expected a ready evaluation, got {other:?}"), + } +} + +// --------------------------------------------------------------------------- +// A2A-3 — the painter extraction preserves grid behavior +// --------------------------------------------------------------------------- + +#[test] +fn extraction_preserves_cells_cursor_and_focused_view_top() { + // The extraction must preserve four things, not just cells: a clamp + // that silently moved to the WRONG window would leave the painted + // cells identical on a single-window frame. + let s = editor(); + exec( + &s, + "local b = pmacs.buffer.create(\"*doc*\") + b:insert(0, string.rep(\"line\\n\", 200)) + pmacs.window.display(b, {})", + ); + + let size = CellSize::new(ROWS, COLS); + let mut cells_a = vec![pmacs::cell::Cell::default(); (ROWS * COLS) as usize]; + let mut grid_a = CellGrid { + cells: &mut cells_a, + stride: size.cols, + size, + }; + let cursor_a = pmacs::editor::paint_frame( + &s, + FrontendId::LOCAL, + &std::collections::HashMap::new(), + &mut grid_a, + size, + ); + let active = s.core.borrow().views[&FrontendId::LOCAL].active; + let view_top_a = s.core.borrow().windows[&active].view_top; + + // A second identical paint is a fixed point: same cells, same + // returned cursor, same `view_top`. + let mut cells_b = vec![pmacs::cell::Cell::default(); (ROWS * COLS) as usize]; + let mut grid_b = CellGrid { + cells: &mut cells_b, + stride: size.cols, + size, + }; + let cursor_b = pmacs::editor::paint_frame( + &s, + FrontendId::LOCAL, + &std::collections::HashMap::new(), + &mut grid_b, + size, + ); + let view_top_b = s.core.borrow().windows[&active].view_top; + + assert_eq!(cells_a, cells_b, "painted cells must be stable"); + assert_eq!(cursor_a, cursor_b, "the returned cursor must be stable"); + assert_eq!(view_top_a, view_top_b, "focused view_top must be stable"); +} + +#[test] +fn extraction_leaves_a_passive_window_view_top_untouched() { + // The auto-scroll clamp runs for the FOCUSED window only. A passive + // window's scroll state must survive a frame it did not own. + let s = editor(); + exec( + &s, + "local b = pmacs.buffer.create(\"*doc*\") + b:insert(0, string.rep(\"line\\n\", 200)) + pmacs.window.display(b, {}) + pmacs.window.split_horizontal()", + ); + render(&s); + + let (passive, before) = { + let core = s.core.borrow(); + let view = &core.views[&FrontendId::LOCAL]; + let passive = view + .layout + .iter_ids() + .into_iter() + .find(|id| *id != view.active) + .expect("a second window exists"); + (passive, core.windows[&passive].view_top) + }; + + // Scroll the passive window somewhere the clamp would "fix" if it + // ever ran against the wrong window. + s.core + .borrow_mut() + .windows + .get_mut(&passive) + .unwrap() + .view_top = 120; + render(&s); + + assert_eq!( + s.core.borrow().windows[&passive].view_top, + 120, + "a passive window's view_top must not be clamped by another window's frame" + ); + assert_ne!(before, 120, "non-vacuity: the value actually changed"); +} + +// --------------------------------------------------------------------------- +// Fixture integrity +// --------------------------------------------------------------------------- + +#[test] +fn the_panel_fixture_really_builds_a_side_window() { + // Every test above is worthless if `focused_panel` silently produced + // an ordinary split, so pin the fixture's own precondition. + let s = editor(); + let (_document, panel) = focused_panel(&s); + let core = s.core.borrow(); + assert_eq!( + core.windows[&panel].params.side, + Some(Side::Bottom), + "the fixture must produce a real bottom side window" + ); +} From 4413ea93d00d3bfe3668d03c93b23f09ce1a6075 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 22:14:37 -0400 Subject: [PATCH 21/32] docs: record the Stage 2A lane and what gating it found MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Ledger entry for the in-flight Stage 2A branch, plus three findings the gate run produced that are worth carrying regardless of this PR: - The structural test comparing the two authorities directly did NOT catch the focus-class bite; only the consumer-level assertion did. Both kinds are needed, and the distinction generalizes. - `vterm_stage3_acceptance::a37` is badly flaky on this machine — 6/8 failures on the BASE commit against 7/8 on the branch in matched isolated samples, so it is pre-existing rather than a regression. It also returns `ok` without running unless `pmacs-gpu` is built. - `m11_5_semantic_acceptance` reports 0 tests and `gpu_initial_target_acceptance` reports 1 without `--features crdt`. Both are semantic-census suites, so gating Stage 2A in the default config alone would exercise almost none of its relevant coverage. Co-Authored-By: Claude Opus 5 (1M context) --- docs/active-work.md | 41 ++++++++++++++++++++++++++++++++++++++--- 1 file changed, 38 insertions(+), 3 deletions(-) diff --git a/docs/active-work.md b/docs/active-work.md index b60d62c..b961794 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -389,10 +389,45 @@ If it does not, stop and repair the remote/fetch configuration. **isolated-config workspace sweep 3,177 across 92 suites, zero failures**; `git diff --check` clean. Gates were run against the committed tree. -## Bottom-panel lane (Arc 7) — Stage 1 MERGED; Stage 2 IN FRAMING +## Bottom-panel lane (Arc 7) — Stage 1 + framing MERGED; Stage 2A IN REVIEW -Stage 1 is on `main`. **Stage 2 is in framing**, no implementation in -flight. +Stage 1 and the Stage 2 framing are on `main`. **Stage 2A is +implemented and in review.** + +- **Stage 2A — portable branch `githubsucks/bottom-panel-stage2a`**, + worktree `../pmacs-bp-stage2a`, based on `githubsucks/main` @ + `c93f9ee`. Two commits: the classified census routing, then the + painter extraction + acceptance. **No protocol change; no behavior + change for any frontend today** — with `panel_capable = false` for + semantic sessions, `primary_document_window` returns `view.active` + in every existing configuration, so this is seam adoption that + becomes load-bearing in 2B. +- Verification on this branch: `cargo fmt --check` clean; strict + workspace Clippy clean; **1,832 default + 2,009 CRDT** library tests; + new `bottom_panel_stage2a_acceptance` 10/10; bottom-panel Stage 1 + 46; statusline segments 7 default / 8 CRDT; m11_5 semantic 2 CRDT; + GPU initial target 14 CRDT; vterm Stage 1/2 10 / 6; folding Stage 2 + 48; M4 121; required GPU 202; `git diff --check` clean. +- **Both key routings were falsified by revert.** Rerouting + `dispatch_idle_for` (#14, Focus) through `primary_document_window` + fails `focus_class_dispatch_idle_still_tracks_the_focused_window`; + reverting the statusline lookup (#12, Projection) to `view.active` + fails the document-context test. Worth recording: the *structural* + test `focus_and_projection_disagree_in_the_same_state` did **not** + catch the first bite — it compares the two authorities directly, so + only a consumer-level assertion catches a misrouted consumer. Keep + both kinds. +- **`vterm_stage3_acceptance::a37` is a pre-existing flake here**, not a + Stage 2A regression: measured **6/8 failures on the base commit** and + **7/8 on the branch** in matched isolated samples. It needs a real + daemon + real PTY + headless GPU and is documented load-sensitive. + It also silently returns `ok` unless `pmacs-gpu` has been built, and + is `crdt`-gated so CI never runs it at all. +- **Two suites are dark without `--features crdt`**: + `m11_5_semantic_acceptance` reports **0 tests** and + `gpu_initial_target_acceptance` reports **1** in the default config. + Both are semantic-census suites, so Stage 2A must be gated with the + feature on or its most relevant coverage never executes. - Stage 1 merged as **#155** (`main` @ `e745068`, 2026-07-24, after two review rounds). No protocol change. Durable substrate facts live in From 786de69d3827d37a648b5d378b388837161f662c Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 22:32:48 -0400 Subject: [PATCH 22/32] fix(lsp): close round-six Lean fallback gaps Make command-time attachment healing cancel an armed terminal restart before replacing the server, while keeping request-only lookup pure and restart-safe. Track config-driven server ownership privately, bound every fallback server per SID, scope no-swap retirement to the failed root, and route the shipped Lean diagnostics command through the safe resolver while waiting for initialization. Add direct acceptance counterexamples for all five review findings and record the sixth-round verification and vacuity lesson. --- builtin/runtime/lean.lua | 180 +++++++++++++++------- builtin/runtime/lsp.lua | 64 ++++++-- docs/active-work.md | 37 ++++- tests/lean4_server_acceptance.rs | 248 ++++++++++++++++++++++++++++++- 4 files changed, 455 insertions(+), 74 deletions(-) diff --git a/builtin/runtime/lean.lua b/builtin/runtime/lean.lua index 6490296..09ad280 100644 --- a/builtin/runtime/lean.lua +++ b/builtin/runtime/lean.lua @@ -137,8 +137,8 @@ local probe = { -- table cardinality cannot tell "once per buffer" -- from "every tick for one buffer" fallback_installed = false, - fallback_watch = nil, -- fallback sid being polled for die-before-init - fallback_failed = false, + fallback_watches = {}, -- sid key -> sid, each polled die-before-init + fallback_done = {}, -- sid key -> initialized or terminally handled saw_initialized = false, } @@ -286,23 +286,36 @@ end -- Retiring only the server that happened to fail left the others live -- and every buffer attached to them stranded on a command the config no -- longer names. --- Only servers this module's config produced. `ensure_server` labels --- every auto-attached server `default-`, so that label is the --- derivation discriminator: a server the USER spawned from `init.lua` --- carries their own label, is not derived from `pmacs.lsp.config.lean4`, --- and must not be stopped because our config changed. Selecting on --- `language_id` alone swept those up too — a destructive side effect on --- state this module does not own. -local DERIVED_LABEL = "default-lean4" +-- Only servers the config-driven path itself produced. A server's label, +-- language, command, and root are all caller-supplied public values; none +-- is an ownership discriminator. `lsp.lua` records the successful spawn +-- in a private origin table, which is the fact this lifecycle may act on. +local function is_derived_server(sid) + local ok, owned = pcall(pmacs.lsp._is_default_server, sid, "lean4") + return ok and owned == true +end local function retire_derived_lean_servers() local ok, rows = pcall(pmacs.lsp.list) if not ok or not rows then return end local ids = {} for _, info in ipairs(rows) do - if info.label == DERIVED_LABEL then ids[#ids + 1] = info.id end + if is_derived_server(info.id) then ids[#ids + 1] = info.id end end - for _, id in ipairs(ids) do retire_server(id) end + for _, id in ipairs(ids) do + -- These ids predate the fallback spawn. Mark them handled before + -- retirement so the discovery poll cannot mistake their terminal + -- state for a fallback that failed to initialize. + probe.fallback_done[tostring(id)] = true + retire_server(id) + end +end + +local function watch_fallback_server(sid) + if not sid or not is_derived_server(sid) then return end + local key = tostring(sid) + if probe.fallback_done[key] then return end + probe.fallback_watches[key] = sid end -- Rebuild the ACTIVE buffer's attachment if it is Lean and stale. @@ -358,33 +371,44 @@ local function repair_active_if_stale() end -- **A successful SPAWN is not a successful START.** The once-per- -- buffer bound stops `_attach_buffer` being called again, but it says - -- nothing about the server it produced: `ensure_server` never forwards - -- `cfg.restart`, so the fallback inherits `OnCrash`, and an executable - -- that dies before `initialize` is respawned by the manager forever - -- with no attempt ceiling — silently, because `latched` has already - -- disabled the primary's failure poll. Watch this one too, once. - if not probe.fallback_watch and not probe.fallback_failed then - probe.fallback_watch = fresh.server - end + -- nothing about the server it produced. Arm this id immediately; the + -- poll below also discovers servers created through lsp.lua's own + -- after-load and command paths. + watch_fallback_server(fresh.server) end --- The fallback's own die-before-initialize poll. One shot: on failure it --- retires the server (which is what actually ends the respawn loop) and --- reports, and never re-arms. -local function poll_fallback() - local sid = probe.fallback_watch - if not sid then return end - local kind = server_state_kind(sid) - if kind == "initialized" then - probe.fallback_watch = nil - return +-- Every fallback server gets its own die-before-initialize poll. A scalar +-- watch cannot cover Q#LN15's simultaneous per-root servers, and a server +-- may be created by lsp.lua's after-load or command path without passing +-- through `repair_active_if_stale`. Discovery from the private ownership +-- table closes both holes. +local function poll_fallbacks() + if not probe.fallback_installed then return end + local ok, rows = pcall(pmacs.lsp.list) + if not ok or not rows then return end + + local by_key = {} + for _, info in ipairs(rows) do + local key = tostring(info.id) + by_key[key] = info + if not probe.fallback_done[key] and is_derived_server(info.id) then + probe.fallback_watches[key] = info.id + end end - if kind == nil or kind == "crashed" or kind == "stopped" then - probe.fallback_watch = nil - probe.fallback_failed = true - if kind ~= nil then retire_server(sid) end - report("LSP: lean4 fallback " .. fallback_name() - .. " started but did not stay up") + + for key, sid in pairs(probe.fallback_watches) do + local info = by_key[key] + local kind = info and info.state and info.state.kind + if kind == "initialized" then + probe.fallback_watches[key] = nil + probe.fallback_done[key] = true + elseif info == nil or kind == "crashed" or kind == "stopped" then + probe.fallback_watches[key] = nil + probe.fallback_done[key] = true + if info ~= nil then retire_server(sid) end + report("LSP: lean4 fallback " .. fallback_name() + .. " started but did not stay up") + end end end @@ -394,10 +418,10 @@ local function fire_latch(sid, why) probe.watching = nil if not swap_to_fallback() then report("LSP: lean4 " .. why) - -- Still retire: the servers are broken whether or not a replacement - -- command was installed, and leaving them live would keep the - -- restart machinery running against a command known to fail. - retire_derived_lean_servers() + -- No shared config changed, so only the server whose failure + -- triggered this verdict is invalid. Sweeping every root here stops + -- healthy instances of a root-sensitive command for no reason. + if sid and is_derived_server(sid) then retire_server(sid) end return end retire_derived_lean_servers() @@ -551,22 +575,66 @@ function M.wait_for_diagnostics(sid, uri, version, fn) return rid end +local function when_server_ready(sid, fn) + local function state_kind() + local ok, state = pcall(pmacs.lsp.status, sid) + if not ok or not state then return nil end + return state.kind + end + + local kind = state_kind() + if kind == "initialized" then + fn(nil) + return + end + if kind == nil or kind == "crashed" or kind == "stopped" then + fn("server did not initialize") + return + end + + -- A command may have just healed a dead attachment, in which case the + -- replacement is still starting. Requests are not queued before + -- initialize, so issue this one after the lifecycle reaches ready + -- rather than replacing the attachment and immediately failing on it. + pmacs.async(function() + for _ = 1, 300 do + pmacs.async.yield_to_next_tick() + kind = state_kind() + if kind == "initialized" then + fn(nil) + return + end + if kind == nil or kind == "crashed" or kind == "stopped" then + fn("server did not initialize") + return + end + end + fn("server initialization timed out") + end) +end + pmacs.command.define { name = "lean.wait-for-diagnostics", description = "Wait for the Lean server to finish elaborating this file", fn = function() - local rec = pmacs.lsp.active_attachment() + local rec = pmacs.lsp._attachment_for_command() if not rec or rec.language ~= "lean4" then pmacs.editor.set_status("lean: no Lean server for this buffer") return end pmacs.editor.set_status("lean: elaborating…") - M.wait_for_diagnostics(rec.server, rec.uri, rec.version, function(err) - if err then - pmacs.editor.set_status("lean: " .. tostring(err)) - else - pmacs.editor.set_status("lean: elaboration complete") + when_server_ready(rec.server, function(init_err) + if init_err then + pmacs.editor.set_status("lean: " .. tostring(init_err)) + return end + M.wait_for_diagnostics(rec.server, rec.uri, rec.version, function(err) + if err then + pmacs.editor.set_status("lean: " .. tostring(err)) + else + pmacs.editor.set_status("lean: elaboration complete") + end + end) end) end, } @@ -600,13 +668,17 @@ pmacs.hook.add("buffer.after-load", function() local ok_lang, lang = pcall(pmacs.lsp.buffer_language, buf) if not ok_lang or lang ~= "lean4" then return end - if not probe.started then - local path = pmacs.editor.file_path() - start_probe(path and M.root_for(path) or nil) - end - local rec = pmacs.lsp.active_attachment() if rec and rec.language == "lean4" then + -- A matching-root server supplied by the user may be adopted by + -- `ensure_server`. Its lifecycle is not evidence about the + -- config-driven command, and neither the version probe nor fallback + -- latch may mutate config because that foreign server changed state. + if not is_derived_server(rec.server) then return end + if not probe.started then + local path = pmacs.editor.file_path() + start_probe(path and M.root_for(path) or nil) + end -- **Arm ONCE, capturing buffer and server together.** Setting -- `buf_key` on every Lean load meant a second Lean buffer opened -- before the verdict silently became the rebuild target while the @@ -632,6 +704,10 @@ pmacs.hook.add("buffer.after-load", function() -- that produced no attachment is a failure. local cfg = pmacs.lsp.config.lean4 if not cfg or not cfg.command then return end + if not probe.started then + local path = pmacs.editor.file_path() + start_probe(path and M.root_for(path) or nil) + end -- No attachment for a Lean buffer with a configured command means -- `ensure_server` could not spawn at all — a synchronous ENOENT, @@ -662,7 +738,7 @@ pmacs.hook.add("process.after-tick", function() -- Repair the active buffer if the latch invalidated it. Cheap when -- there is nothing to do, and bounded to one attempt per buffer. repair_active_if_stale() - poll_fallback() + poll_fallbacks() end) -- Test seam: acceptance drives the latch deterministically rather than diff --git a/builtin/runtime/lsp.lua b/builtin/runtime/lsp.lua index 6081a3d..bf56a4c 100644 --- a/builtin/runtime/lsp.lua +++ b/builtin/runtime/lsp.lua @@ -607,6 +607,12 @@ local function project_root_for(language, path) return dir_of(path), "fallback" end +-- Servers created by the automatic config-driven path. This is the +-- ownership fact a caller-supplied `label` cannot provide: labels are +-- public, unreserved display strings, while entries here are written +-- only after this module itself successfully spawns a server. +local default_servers = {} + local function ensure_server(language, path) local cfg = pmacs.lsp.config[language] if not cfg or not cfg.command then return nil end @@ -660,7 +666,30 @@ local function ensure_server(language, path) cwd = root, root_uri = key_uri, }) - if ok then return sid end + if ok then + default_servers[tostring(sid)] = language + return sid + end + return nil +end + +-- Internal ownership seam for builtins whose lifecycle follows the +-- config-driven server set (currently Lean's one-shot fallback). A +-- user-managed server may deliberately use the same language id, label, +-- command, and root; none of those make it ours. +function pmacs.lsp._is_default_server(sid, language) + local owned_language = default_servers[tostring(sid)] + return owned_language ~= nil + and (language == nil or owned_language == language) +end + +local function server_state_kind(sid) + if not sid then return nil end + for _, info in ipairs(pmacs.lsp.list()) do + if tostring(info.id) == tostring(sid) then + return info.state and info.state.kind + end + end return nil end @@ -669,14 +698,8 @@ end -- forgotten, or was spawned against a now-replaced `pmacs.lsp.config` -- entry — get rebuilt on the next attach attempt. local function server_is_live(sid) - if not sid then return false end - for _, info in ipairs(pmacs.lsp.list()) do - if tostring(info.id) == tostring(sid) then - local kind = info.state and info.state.kind - return kind ~= "crashed" and kind ~= "stopped" - end - end - return false + local kind = server_state_kind(sid) + return kind ~= nil and kind ~= "crashed" and kind ~= "stopped" end local function server_is_initialized(sid) @@ -811,6 +834,15 @@ local function attach_buffer(buf) local existing = attachments[key] if existing and server_is_live(existing.server) then return existing end if existing then + local kind = server_state_kind(existing.server) + if kind == "crashed" or kind == "stopped" then + -- A terminal OnCrash client may still have `next_restart_at` + -- armed. Spawning beside it creates two same-root servers when + -- the old id restarts. `forget` is the terminal-state operation: + -- it removes the client and cancels that pending restart before + -- the replacement is created. + pcall(pmacs.lsp.forget, existing.server) + end attachments[key] = nil -- Unsent edits targeted the dead attachment; the did_open below -- carries the full current text, superseding them. @@ -896,6 +928,14 @@ local function attached_for_active() return attach_buffer(buf) end +-- Internal command-path resolver for builtin request producers outside +-- this module. Unlike `active_attachment` it may replace a dead record; +-- unlike `attachment_for_request` it is called only from an explicit +-- user command, where attach-on-use is the intended policy. +function pmacs.lsp._attachment_for_command() + return attached_for_active() +end + -- Pure, side-effect-free attachment lookup for the active buffer: -- returns the live record (with `.uri`) when a server is already -- attached, else nil. Unlike `attached_for_active`, it never *triggers* @@ -964,8 +1004,10 @@ function pmacs.lsp.attachment_for_request() -- perturb LSP state), so a dead record reads as "no attachment" -- rather than triggering a rebuild. if not server_is_live(rec.server) then - attachments[key] = nil - pending_did_change[key] = nil + -- Preserve the record. A crashed OnCrash server may restart under + -- the SAME id; clearing the map here would orphan that recovered + -- server, while this non-attaching lookup has no authority to + -- cancel the restart or create a replacement. return nil end flush_did_change(key) diff --git a/docs/active-work.md b/docs/active-work.md index 6b99985..b96cc42 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -265,7 +265,7 @@ If it does not, stop and repair the remote/fetch configuration. - Ships `builtin/runtime/lean.lua` (new), one `include_str!` line in `src/editor.rs`, `pmacs.lsp._attach_buffer` exported from `lsp.lua`, a `leanprogress` mode plus `waitForDiagnostics` validation on - `pmacs_fake_lsp`, and `tests/lean4_server_acceptance.rs` (36 tests). + `pmacs_fake_lsp`, and `tests/lean4_server_acceptance.rs` (40 tests). No protocol change. - **Stage 1's acceptance 12 is half superseded and was rewritten, not deleted.** It asserted `pmacs.lsp.config.lean4 == nil` to catch a @@ -359,9 +359,29 @@ If it does not, stop and repair the remote/fetch configuration. assertion counted TABLE KEYS, which cannot distinguish "once per buffer" from "every tick for one buffer" — cardinality stays 1 either way. Now a numeric attempt counter; the bite shows **174 vs 1**. +- **Round-6 review: four P1s and one P2, suite 40/40.** (1) General + point-of-use healing treated a crashed OnCrash server as absent and + spawned beside it while its old id still had `next_restart_at` armed; + `attach_buffer` now forgets a terminal record before replacement. + `attachment_for_request` remains non-attaching and preserves the + record, so a same-id restart can recover instead of being orphaned. + (2) The fallback watch was scalar, while Q#LN15 permits simultaneous + per-root servers and lsp.lua can create them without passing through + Lean's repair function. Watches are now per-SID and discover every + config-driven Lean server from a private origin table. (3) The shipped + `lean.wait-for-diagnostics` command bypassed both safe resolvers and + still consumed a stopped record; it now uses a command-safe resolver, + waits asynchronously for a healed replacement to initialize, and the + test requires the real request to finish. (4) When no config swap + occurred, one failed root still swept a healthy root; that arm now + retires only the SID whose verdict fired. (5) `label` is public and + unreserved, therefore not ownership. lsp.lua records successful + config-driven spawns privately, and every Lean lifecycle decision keys + on that origin fact; the user-server pin deliberately collides on + `default-lean4`. - **DURABLE LESSON — "the test that passes" vs "the test that - discriminates."** Six tests across three rounds were written, run - green, and only bite-testing showed they pinned nothing. **Carry this + discriminates."** Green tests across six rounds repeatedly pinned only + a nearby helper or an absence, and only biting exposed it. **Carry this to `docs/agent-handoff.md` when the lane lands.** The concrete shapes, all from this branch: 1. R1 acceptance 36 asserted "every server is terminal" — pinning the @@ -389,6 +409,11 @@ If it does not, stop and repair the remote/fetch configuration. not the things attempted against (bite: 174 vs 1). 9. A NONEXISTENT executable only exercises synchronous ENOENT. To reach "spawned, then died", the fixture must actually spawn. + 10. Calling the two SAFE HELPERS directly does not pin a shipped + command that bypasses both. Drive the command registry entry and + require its terminal result — replacing a dead record with a + `starting` server is still not success if the request is issued + before initialize. Rule: **a test is not evidence until the mutation it targets has been shown to fail it.** - **SECOND DURABLE LESSON — a scope error repeats until the scope is @@ -424,10 +449,10 @@ If it does not, stop and repair the remote/fetch configuration. works. Only a parseable version below 3.1.0 triggers it; the server-failure latch covers the rest. - Verification on this branch: `cargo fmt --check` clean; strict - workspace Clippy clean; 1,826 default + 2,003 CRDT library tests; - lean4 server 36/36; lean4 stage 1 9/9; dispatch seams 15/15; + workspace Clippy clean; 1,829 default + 2,003 CRDT library tests; + lean4 server 40/40; lean4 stage 1 9/9; dispatch seams 15/15; multi-root 13/13; M4 121; required GPU 155; **isolated-config - workspace sweep 3,225 across 94 suites, zero failures**; + serial workspace sweep 3,229 across 94 suites, zero failures**; `git diff --check` clean. (Round 1 of this entry recorded 17/17 and 3,206 — the PRE-fix counts — after the fixes were pushed. The ledger's protocol is that verification diff --git a/tests/lean4_server_acceptance.rs b/tests/lean4_server_acceptance.rs index e28ac8f..86be1d5 100644 --- a/tests/lean4_server_acceptance.rs +++ b/tests/lean4_server_acceptance.rs @@ -1438,10 +1438,10 @@ fn r5_a_fallback_that_dies_after_spawning_is_bounded_and_reported() { #[test] fn r5_a_user_spawned_lean_server_is_not_retired_by_the_fallback() { - // `retire_*` selected on `language_id == "lean4"`, which also names - // servers the user spawned themselves from `init.lua`. Those are not - // derived from `pmacs.lsp.config.lean4` and stopping them is a - // destructive side effect on state this module does not own. + // Language id AND label are public caller-supplied values. Even a + // user server that deliberately collides with the automatic path's + // `default-lean4` display label is not derived from + // `pmacs.lsp.config.lean4` and must not be stopped. let fx = Fixture::new(); fx.toolchain("pkg", "v4.9.0\n"); let file = fx.write("pkg/A.lean", "def a := 1\n"); @@ -1453,7 +1453,7 @@ fn r5_a_user_spawned_lean_server_is_not_retired_by_the_fallback() { &format!( r#" _G.mine = pmacs.lsp.spawn({{ - label = "my-own-lean", + label = "default-lean4", language_id = "lean4", command = "{}", args = {{}}, @@ -1642,3 +1642,241 @@ fn r5_a_dead_attachment_is_never_handed_to_a_command() { {rebuilt:?}" ); } + +// --------------------------------------------------------------------------- +// Round-6 review. Each is a direct counterexample against 19f48d4. +// --------------------------------------------------------------------------- + +#[test] +fn r6_the_shipped_lean_command_rebuilds_a_dead_attachment() { + // The round-5 test called `attachment_for_request` and + // `_attach_buffer` directly, while the shipped Lean command read the + // raw `active_attachment` and still handed its request to a stopped + // server. Drive the production command this time. + let fx = Fixture::new(); + fx.toolchain("pkg", "v4.9.0\n"); + let file = fx.write("pkg/A.lean", "def a := 1\n"); + let mut state = editor(&fx); + open(&state, &file); + settle(&mut state); + + exec( + &state, + r" + local rec = pmacs.lsp.active_attachment() + assert(rec) + pmacs.lsp.stop(rec.server) + ", + ); + tick_for(&mut state, 200); + + exec( + &state, + r#"pmacs.command.invoke("lean.wait-for-diagnostics")"#, + ); + let kind: String = eval( + &state, + r#" + local rec = pmacs.lsp.active_attachment() + if not rec then return "none" end + for _, s in ipairs(pmacs.lsp.list()) do + if tostring(s.id) == tostring(rec.server) then + return tostring(s.state and s.state.kind) + end + end + return "gone" + "#, + ); + assert!( + kind != "stopped" && kind != "crashed" && kind != "gone" && kind != "none", + "the shipped command must resolve through the command-safe \ + attachment path; saw {kind:?}" + ); + tick_for(&mut state, 500); + let status = state.core.borrow().status.clone(); + assert_eq!( + status, "lean: elaboration complete", + "the rebuilt command path must deliver the request, not merely \ + replace the attachment" + ); +} + +#[test] +fn r6_every_spawned_fallback_server_is_bounded() { + // A scalar fallback watch covers only one Q#LN15 root. The second + // server can also be created directly by lsp.lua's after-load path, + // bypassing `repair_active_if_stale` entirely. + use std::os::unix::fs::PermissionsExt as _; + + let fx = Fixture::new(); + fx.toolchain("one", "v4.9.0\n"); + fx.toolchain("two", "v4.9.0\n"); + let first = fx.write("one/A.lean", "def a := 1\n"); + let second = fx.write("two/B.lean", "def b := 2\n"); + let absent_primary = fx.dir("bin/no-such-lake"); + let dying_fallback = fx.root.join("bin/dying-lean"); + std::fs::create_dir_all(dying_fallback.parent().unwrap()).unwrap(); + std::fs::write(&dying_fallback, "#!/bin/sh\nexit 4\n").unwrap(); + std::fs::set_permissions(&dying_fallback, std::fs::Permissions::from_mode(0o755)).unwrap(); + + let mut state = editor(&fx); + exec( + &state, + &format!( + r#" + pmacs.lsp.config.lean4.command = "{}" + pmacs.lsp.config.lean4.args = {{ "serve" }} + pmacs.lean._fallback = {{ command = "{}", args = {{}} }} + "#, + lua_str(&absent_primary), + lua_str(&dying_fallback) + ), + ); + + open(&state, &first); + open(&state, &second); + tick_for(&mut state, 1600); + + let worst_attempt: i64 = eval( + &state, + r" + local worst = 0 + for _, s in ipairs(pmacs.lsp.list()) do + if s.language_id == 'lean4' and (s.attempt or 0) > worst then + worst = s.attempt + end + end + return worst + ", + ); + assert!( + worst_attempt <= 1, + "every fallback server must be bounded; an unwatched root \ + reached attempt {worst_attempt}" + ); +} + +#[test] +fn r6_point_of_use_healing_does_not_duplicate_a_restarting_server() { + // A crashed OnCrash server still has `next_restart_at` armed. + // Spawning a fresh id beside it produces two same-root servers when + // the old one restarts. Use Rust so this pins the general lsp.lua + // seam independently of Lean's fallback lifecycle. + let fx = Fixture::new(); + let file = fx.write("A.rs", "fn main() {}\n"); + let mut state = editor(&fx); + exec( + &state, + &format!( + r#" + pmacs.lsp.config.rust = {{ + command = "{}", + args = {{}}, + env = {{ PMACS_FAKE_LSP_MODE = "crash" }}, + }} + "#, + fake_lsp_path() + ), + ); + open(&state, &file); + + let mut crashed = false; + for _ in 0..100 { + state.tick_processes(); + state.tick_lsp(); + crashed = eval( + &state, + r#" + for _, s in ipairs(pmacs.lsp.list()) do + if s.language_id == "rust" + and s.state and s.state.kind == "crashed" then + return true + end + end + return false + "#, + ); + if crashed { + break; + } + std::thread::sleep(Duration::from_millis(5)); + } + assert!(crashed, "precondition: the attached server crashed"); + + exec(&state, "pmacs.lsp.hover_at_cursor()"); + let rust_servers: i64 = eval( + &state, + r#" + local n = 0 + for _, s in ipairs(pmacs.lsp.list()) do + if s.language_id == "rust" then n = n + 1 end + end + return n + "#, + ); + assert_eq!( + rust_servers, 1, + "healing must cancel the old id's armed restart before spawning \ + its replacement" + ); +} + +#[test] +fn r6_no_swap_retires_only_the_failed_root() { + // When config already equals the fallback, no shared config changed. + // One root's failure must not globally retire another root's healthy + // instance of the same cwd-sensitive command. + use std::os::unix::fs::PermissionsExt as _; + + let fx = Fixture::new(); + fx.toolchain("bad", "v4.9.0\n"); + fx.toolchain("good", "v4.9.0\n"); + let bad = fx.write("bad/A.lean", "def a := 1\n"); + let good = fx.write("good/B.lean", "def b := 2\n"); + let wrapper = fx.root.join("bin/root-sensitive-lean"); + std::fs::create_dir_all(wrapper.parent().unwrap()).unwrap(); + std::fs::write( + &wrapper, + format!( + "#!/bin/sh\ncase \"$PWD\" in */bad) exit 4;; esac\nexec \"{}\"\n", + fake_lsp_path() + ), + ) + .unwrap(); + std::fs::set_permissions(&wrapper, std::fs::Permissions::from_mode(0o755)).unwrap(); + + let mut state = editor(&fx); + exec( + &state, + &format!( + r#" + pmacs.lsp.config.lean4.command = "{}" + pmacs.lsp.config.lean4.args = {{}} + pmacs.lean._fallback = {{ command = "{}", args = {{}} }} + "#, + lua_str(&wrapper), + lua_str(&wrapper) + ), + ); + open(&state, &bad); + open(&state, &good); + tick_for(&mut state, 700); + + let good_alive: bool = eval( + &state, + r#" + for _, s in ipairs(pmacs.lsp.list()) do + if s.cwd and s.cwd:match("/good$") then + local k = s.state and s.state.kind + return k ~= "stopped" and k ~= "crashed" + end + end + return false + "#, + ); + assert!( + good_alive, + "one root's failure must not stop another root when no config \ + swap occurred" + ); +} From dec51960da5967c4da9477cbe8a5e10059487481 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sat, 25 Jul 2026 22:36:02 -0400 Subject: [PATCH 23/32] docs: record round-six bite evidence Capture the concrete counterfactual outcomes for all five review fixes against 19f48d4. --- docs/active-work.md | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/docs/active-work.md b/docs/active-work.md index b96cc42..2efce61 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -378,7 +378,10 @@ If it does not, stop and repair the remote/fetch configuration. unreserved, therefore not ownership. lsp.lua records successful config-driven spawns privately, and every Lean lifecycle decision keys on that origin fact; the user-server pin deliberately collides on - `default-lean4`. + `default-lean4`. All five bites against `19f48d4` discriminate: the + old files produce 2 same-root servers, a fallback attempt of 4, a + shipped command still targeting `stopped`, retirement of the healthy + root, and retirement of the colliding user server, respectively. - **DURABLE LESSON — "the test that passes" vs "the test that discriminates."** Green tests across six rounds repeatedly pinned only a nearby helper or an absence, and only biting exposed it. **Carry this From 3b54c784949847bd47b7b03f2ce91da2453cb88f Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sun, 26 Jul 2026 10:10:06 -0400 Subject: [PATCH 24/32] =?UTF-8?q?docs(lean4):=20rev=206=20=E2=80=94=20re-s?= =?UTF-8?q?cout=20Stage=204=20and=20split=20it=20into=204a=20and=204b?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Stages 3a and 3b landed (#167, #170). Re-scouting Stage 4 against main @ d400f30 produced six findings that change the plan and three that confirm it. The pmacs-side facts were verified in a worktree at that commit; the upstream facts by reading leanprover/vscode-lean4 @ 17d1d08. The split: Stage 4's risk column read "refactors pair.lua's provenance read" — every language's auto-pairing — for a stage the prose called the Lean input method, which is exactly the rule §4 states and exactly what round 4 found for Stage 3. Rev 5 had noticed the shape and answered it with a commit boundary; a commit boundary is not a review boundary. Stage 4a is now the typed-edit consumer chain (substrate, no Lean) and 4b the input method. Rev 5's expansion semantics were wrong in three ways. Resolution is the shortest key having the input as a prefix (\al yields ∀ from `all`, not `alpha`); there is no terminator list at all ('+ ' is a key, so space extends after \+; '\' is a key, so \\ yields \); and an unmatchable tail is appended rather than dropped (\alp7 yields α7). Three further findings. There is no cursor-motion hook, so acceptance 43 as written was not buildable and abandonment is lazy. dispatch_key is only half of 4b's production path — \ and the letters are not excluded from the optimistic classifier, and that producer is crdt-gated, so a crdt-gated integration test is dark in CI and dark in the gate list. And the whole expansion has cross-peer-degraded undo, a wider bite than Q#LN6's three bracket pairs; set_round_trip_input would fix it and is rejected with reasons. New decisions Q#LN21 (undo degradation) and Q#LN22 (the state machine); Q#LN10 and Q#LN11 rewritten; §2.11 records the upstream algorithm; §9.1 states the coherence impact for both stages. Acceptance keeps its existing numbers and adds letter suffixes on both sides of the split. Citation sweep per COHERENCE §25: five live citations moved in the 50 commits since rev 5. Co-Authored-By: Claude Opus 5 (1M context) Claude-Session: https://claude.ai/code/session_011LFvC4FQtux4y32KuevZ7B --- docs/active-work.md | 73 +++- docs/lean4-mode-framing.md | 703 +++++++++++++++++++++++++++++++------ 2 files changed, 667 insertions(+), 109 deletions(-) diff --git a/docs/active-work.md b/docs/active-work.md index e4f0859..52d44ba 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -14,11 +14,13 @@ backlog. machine-local: `origin` may name this canonical URL, a release mirror, or something else, and therefore has no authority by name alone. - Canonical base at this snapshot: - `githubsucks/main` @ `d152120` (the bottom-panel landed-doc refresh #156 - atop the inline-math slice #158, dired Stage 1 #165, the GPU terminal - input fix #166, Lean 4 Stage 2 #161, the dired framing #164, - COHERENCE.md #163, find-file #162, Lean 4 Stage 1 #160, and the minimap - blank-slab fix #159; protocol v20). + `githubsucks/main` @ `d400f30` (Lean 4 Stage 3b #170 atop Stage 3a + #167, the bottom-panel landed-doc refresh #156, the inline-math slice + #158, dired Stage 1 #165, the GPU terminal input fix #166, Lean 4 + Stage 2 #161, the dired framing #164, COHERENCE.md #163, find-file + #162, Lean 4 Stage 1 #160, and the minimap blank-slab fix #159; + protocol v20). The previous snapshot named `d152120`; the recovery + check below accepts it or anything newer. - On the transfer source, `origin/main` named a release mirror at `d3fa632` and lagged badly. On the current destination, `origin` names the canonical URL. This difference is why all recovery begins by @@ -55,7 +57,7 @@ git status --short --branch The `git log` command must expose `d152120` or a newer intentional main. If it does not, stop and repair the remote/fetch configuration. -## Lean 4 lane (Arc 8) — Stages 1+2 MERGED; 3a IN REVIEW (#167); 3b STACKED +## Lean 4 lane (Arc 8) — Stages 1, 2, 3a, 3b MERGED; Stage 4 IN FRAMING - Stage 1 **merged as #160** (`main` @ `0827dd1`, 2026-07-25, one review round, all twelve checks green). Branch `githubsucks/lean4-stage1` @@ -189,7 +191,7 @@ If it does not, stop and repair the remote/fetch configuration. suites**; `git diff --check` clean. The sweep needs an isolated `XDG_CONFIG_HOME` and `-- --skip basedpyright`. -### Stage 3a — dispatch seams + `pmacs.fs.canonicalize` (branch `lean4-stage3a-seams`) +### Stage 3a — dispatch seams + `pmacs.fs.canonicalize` — MERGED #167 (`main` @ `6f348c9`) - Worktree `../pmacs-lean-stage3`, branched off `githubsucks/main` @ `46a1b8f`. Carries framing **rev 5** (the Stage 3 split) as its first @@ -253,7 +255,7 @@ If it does not, stop and repair the remote/fetch configuration. `#[cfg(unix)]` is NOT sufficient for such a fixture — `#[cfg(target_os = "linux")]` is. Cost one red CI round to learn. -### Stage 3b — the Lean language server (branch `lean4-stage3b-server`) +### Stage 3b — the Lean language server — MERGED #170 (`main` @ `d400f30`) - Same worktree `../pmacs-lean-stage3`, **branched off `lean4-stage3a-seams`, not off `main`** — 3b consumes 3a's response @@ -463,6 +465,61 @@ If it does not, stop and repair the remote/fetch configuration. describes the pushed tree; recording it late is the #161 fmt-blocker error in a slower form.) +### Stage 4 — framing rev 6, split into 4a/4b (branch `lean4-stage4a-typed-edit-chain`) + +- Stages 3a and 3b **merged as #167** (`main` @ `6f348c9`) and **#170** + (`main` @ `d400f30`), 2026-07-26. Both were integrated against a main + that had advanced 50 commits mid-review; the only conflict either time + was this ledger's own lane headings, resolved by keeping both sides. +- Worktree `../pmacs-lean-stage4`, branched off `main` @ `d400f30`. + Framing-only so far: `docs/lean4-mode-framing.md` **revision 6**. No + code. Awaiting user approval before implementation, per the workflow. +- **Round 5 re-scout split Stage 4 into 4a (substrate) and 4b (Lean).** + 4a is the typed-edit consumer chain — `builtin/runtime/typed_edit.lua` + plus `pair.lua` re-expressed as one registered consumer, no behavior + change. 4b is the input method. The split is forced by §4's own rule, + which Stage 4's risk column ("refactors `pair.lua`'s provenance read") + broke while the prose called the stage Lean-only. +- **This is the SECOND consecutive re-scout to find that rule broken** + (round 4 found it for Stage 3). Rev 5 had even noticed the shape and + answered it with a commit boundary. **A commit boundary is not a review + boundary.** Re-check every remaining stage against §4 at scout time; + the rule is not self-enforcing. +- **Rev 5's expansion semantics were wrong in three ways**, found by + reading `leanprover/vscode-lean4` @ `17d1d08` rather than inferring + from behavior. Resolution is *shortest key having the input as a + prefix* (`\al` → `∀` from `all`, not `alpha`); there is **no + terminator list** (`'+ '` is a key, so space extends after `\+`; `'\'` + is a key, so `\\` → `\`); and an unmatchable tail is **appended**, + not dropped (`\alp7` → `α7`). +- **There is no cursor-motion hook**, so rev 5's acceptance 43 ("moving + the cursor out abandons it") was not buildable. Abandonment is lazy — + validated at the next typed edit — and the criterion now asserts what + pmacs can actually detect. Upstream drives this off `changeSelections`; + that seam does not exist here. +- **`dispatch_key` is only half the production path for 4b.** The + auto-pair suite gets away with dispatch-only because Q#AP1 removed the + pair chars from the optimistic classifiers; `\` and the letters are + NOT excluded, so on a CRDT frontend the optimistic producer is the real + path. That producer is `#[cfg(feature = "crdt")]` and CI never enables + `crdt`, and the gate list runs `--features crdt` only for `--lib` — a + crdt-gated integration test is **dark twice over**. +- The whole expansion has cross-peer-degraded undo (Q#LN21): six + source-peer optimistic inserts replaced by one daemon-peer op. + `set_round_trip_input` would fix it and is rejected — it also disables + `dispatch_idle`, so RET stops inserting a newline. +- Table facts re-derived at `17d1d08`: 1,855 entries, 36,861 bytes, all + keys ASCII, **64** keys carry a `lean4` pair-set char, **305** keys are + proper prefixes of another (so 1,550 expand eagerly), **26** values + carry `$CURSOR`, **93** are multi-codepoint. +- Citation sweep per COHERENCE §25: five live citations moved in the 50 + commits since rev 5 — `take_typed_edit` 12827→12990, + `handle_server_requests` 1549→1815, `fs.stat` 93→133, + `detect_buffer_language` 452→457, `send_request`/`send_notification` + 9342/9361→9507/9527. +- Verification: none yet — the branch carries no code. `git diff --check` + clean. + ## Dired lane — Stage 0 MERGED; Stage 1 IN REVIEW (PR #165) - Approved framing: `docs/dired-framing.md` **revision 6** — rev 5 is the diff --git a/docs/lean4-mode-framing.md b/docs/lean4-mode-framing.md index ec62980..e39eb36 100644 --- a/docs/lean4-mode-framing.md +++ b/docs/lean4-mode-framing.md @@ -6,7 +6,7 @@ pmacs has no Lean support of any kind: `grep -rin lean` over `*.rs`, plain buffer — no grammar, no major mode, no comment syntax, no pair set, no server. -This lane closes that in eight stages. Stage boundaries are drawn where +This lane closes that in nine stages. Stage boundaries are drawn where the *substrate* changes, not where the feature list does — see §4. §9 states the lane's coherence impact per `COHERENCE.md` §20. @@ -21,11 +21,12 @@ the *substrate* changes, not where the feature list does — see §4. #144 (LaTeX), #146 (HTML+CSS). Stage 1 is that pattern almost exactly. - Stages 2 and 4–6 are **not** that pattern, and none should be mistaken for a one-liner. Stage 2 changes `ensure_server`, shared by every LSP - language. Stage 4 builds the editor's first input method. Stage 5 is the + language. Stage 4a changes how typed-character provenance is consumed + and Stage 4b builds the editor's first input method. Stage 5 is the first consumer of a non-standard LSP method family. Stage 6 adds a severity-routing policy to `LspServerSpec`. - The user's stated north star is **matching or exceeding what VS Code - does with Lean**. §5's bet 6 scores honestly how close the eight stages get + does with Lean**. §5's bet 6 scores honestly how close the nine stages get and names precisely what is still missing. Parallel-safety: Stage 1 touches `Cargo.toml`, `src/syntax.rs`, @@ -283,10 +284,131 @@ rather than a bare root), `ensure_server` 527 → **610**, pre-#161 line numbers inside Q#LN15 are left as written: that stage has landed and its citations are historical record, not navigation. +### Round 5 (rev 5 → rev 6) — Stage 4 re-scout and split + +Stages 3a and 3b landed (#167, #170). Re-scouting Stage 4 against `main` +@ `d400f30` produced **six findings that change the plan** and three +that confirm it. The pmacs-side facts were verified in a worktree at +that commit; the upstream facts were verified by downloading and reading +`leanprover/vscode-lean4` at commit `17d1d08` (2026-05-29) — the +algorithm, not its documentation, since the `lean4-unicode-input` +package ships no README. + +1. **Stage 4 violated this document's own splitting rule — the same way + Stage 3 did.** §4 says "no PR in this arc mixes a cross-cutting + substrate change with Lean feature content," and §4's own risk column + for Stage 4 read *"refactors `pair.lua`'s provenance read."* + `pair.lua` is every language's auto-pairing; the refactor is + cross-cutting substrate by exactly the test that split out stages 2 + and 3a. Rev 5 already conceded the shape without acting on it — + Q#LN10 said the refactor "lands *first*, as its own commit with no + behavior change, so a regression bisects cleanly." A commit boundary + is not a review boundary. **Stage 4 is now 4a (the typed-edit + consumer chain, no Lean) and 4b (the input method).** Confirmed + `pair.lua:226` is still the **only** production `take_typed_edit` + caller; the other eight call sites are all in + `tests/auto_pair_acceptance.rs`. +2. **The expansion semantics in rev 5's Q#LN10 were wrong in three + ways.** Reading `AbbreviationProvider.ts` and `TrackedAbbreviation.ts` + rather than inferring from behavior: + - Rev 5 said expansion fires on "a unique complete match that no + longer key extends." Upstream's rule is + `findSymbolsByAbbreviationPrefix(abbrev)[0]` — the symbol of the + **shortest key having `abbrev` as a prefix**. `\alp` + space is not + a failure; it yields `α`, because `alpha` is the shortest key + starting with `alp`. Verified against the table: `\al` → `∀`, from + `all`, not from `alpha`. + - Rev 5 named "an explicit terminator (space, tab, RET, or a second + `\`)." **There is no terminator list upstream.** A character + terminates iff extending the pending key by it leaves zero prefix + matches. Space usually does — but `'+ '` **is a key** (one of + 1,855), so after `\+` a space extends rather than terminates. And a + second `\` is not a terminator either: `'\'` is a key mapping to + `\`, so `\\` extends, matches uniquely, and expands to a single + backslash. It terminates only when the pending key is non-empty and + no key extends it. + - Rev 5 did not carry the suffix rule at all. When no key has + `abbrev` as a prefix, upstream recurses on `abbrev` minus its last + character and **appends the leftover**: `\alp7` → `α7`. Dropping + this makes a large class of real input silently unexpandable. +3. **There is no cursor-motion hook, so acceptance 43 as written cannot + be built.** The Rust core fires exactly eight named hooks + (`builtin/hooks/default.lua`): `buffer.before-save`, + `buffer.after-load`, `buffer.after-edit`, `buffer.after-switch`, + `buffer.after-save`, `editor.before-quit`, `frontend.detached`, + `process.after-tick`. Upstream drives abandonment off + `changeSelections`, a seam pmacs does not have. Abandonment must + therefore be **lazy** — validated at the next typed edit against the + pending region — which changes what acceptance 43 can assert. Q#LN22 + states the state machine this forces. +4. **`dispatch_key` is only half of Stage 4b's production path.** Rev 5 + inherited the auto-pairing suite's dispatch-driven harness without + noticing why that harness is sufficient *there*: Q#AP1 removed the + pair characters from both optimistic classifiers, so for pair chars + dispatch **is** production. `\` and the ASCII letters are not + excluded — `classify_key` returns `Insert(c)` for them + (`src/optimistic.rs:144`: `Char(c) if !c.is_control() && + !is_builtin_pair_char(c)`), so on a CRDT frontend an abbreviation is + typed entirely through the *optimistic* producer, which arms the same + record from `handle_remote_crdt_op` (`src/daemon.rs:3965` pins the + classification). A dispatch-only Stage 4b suite would pin the path + real users do not take. The trap underneath: that producer is + `#[cfg(feature = "crdt")]`, and CI never enables `crdt` — so a + crdt-gated integration test is dark twice over, since the required + gate list runs `--features crdt` only for `--lib`. Q#LN22 and §7 say + what to do about it instead of discovering it in review. +5. **The whole expansion has cross-peer-degraded undo, and it is a + larger bite than `⟨⟩`'s.** Q#LN6 already accepts this for three + bracket pairs. But there the mismatch is one optimistic opener + against one daemon-peer closer; here the user's `\alpha` is six + source-peer optimistic inserts and the expansion is a single + daemon-peer `replace` **over all six**. Q#LN21 takes the decision — + including why `pmacs.buffer.set_round_trip_input`, which already + exists and would fix it, is the wrong instrument. +6. **The table's shape is sharper than "1,855 entries."** Re-counted at + `17d1d08`: 1,855 entries, all `string → string`, **all keys ASCII**, + longest key 25 characters, 36,861 bytes of JSON. **64** keys contain + a `lean4` pair-set character (rev 5's number, reproduced exactly). + Three numbers rev 5 did not have and the algorithm needs: **305** + keys are proper prefixes of another key (so 1,550 are eager-expandable + on uniqueness and 305 are not), **26** values carry `$CURSOR` (not + just `\<>`), and **93** values are multi-codepoint. Two values contain + a backslash — `n` → `\n` and `setminus` → `\` — which is why upstream + needs a `doNotTrackNewAbbr` guard and why §2.11 records that pmacs + does not. + +Confirmations, recorded because each was load-bearing and unverified: + +7. **`take_typed_edit`'s one-shot contract is unchanged** + (`src/editor_core.rs:4047`): per-frontend, cleared by the producer + when the fan-out returns, nil to a nested manual `hook.run`. The + hazard rev 5 built Q#LN10 around is real and still the reason 4a + exists. +8. **Load order still constrains the chain.** `pair.lua` loads at + `src/editor.rs:430` and `lsp.lua` at `:436`, and Q#AP7's reason + holds: `lsp.lua`'s `buffer.after-edit` callback synchronously flushes + `didChange` on the signature-trigger path. An expansion that landed + after that flush would send the server the unexpanded text. +9. **Embedding the table needs no special machinery.** Every builtin + runtime chunk is an `include_str!`, and `lsp.lua` is already 111 KB + of the 414 KB total. A ~45 KB generated Lua table is within the + existing practice, so Q#LN11 embeds it rather than inventing a + lazy-load path. + +Citation drift repaired per COHERENCE §25, on the same terms as round +4's sweep. Five live citations moved in the 50 commits since rev 5: +`take_typed_edit` 12827 → **12990**, `handle_server_requests` 1549 → +**1815**, `fs.stat` 93 → **133**, `detect_buffer_language` 452 → +**457**, and `send_request`/`send_notification` 9342/9361 → +**9507**/**9527**. Left as written: the pre-#161 numbers inside Q#LN15 +and the revision-history entries above, which are historical record +rather than navigation. + ## 1. What ships -Eight stages, after round 4 split Stage 3. The north star is VS Code -parity; the honest statement of where that lands is in §5, bet 6. +Nine stages, after round 4 split Stage 3 and round 5 split Stage 4. The +north star is VS Code parity; the honest statement of where that lands +is in §5, bet 6. **Stage 1 — grammar, mode, and the editing table stakes.** `.lean` files highlight, carry a `lean4` major mode, and get comment-toggle and @@ -315,10 +437,19 @@ on 3a's seam. Adds `textDocument/waitForDiagnostics`. Diagnostics, hover, completion, goto-definition, document symbols, and semantic tokens all arrive through the existing typed surfaces. -**Stage 4 — the Unicode input method.** Typing `\alpha` produces `α`, +**Stage 4a — the typed-edit consumer chain.** Pure substrate, no Lean +content, split from Stage 4 in round 5 for the reason stages 2 and 3a +were: it changes machinery every language runs through. The one-shot +`take_typed_edit()` record stops being auto-pairing's private property +and becomes a small ordered chain that reads it once and offers it to +registered consumers. `pair.lua` becomes the chain's first and only +consumer, with no behavior change. + +**Stage 4b — the Unicode input method.** Typing `\alpha` produces `α`, `\to` produces `→`, `\<>` produces `⟨⟩` with the point between them. -1,855 abbreviations vendored from vscode-lean4. This is the stage that -makes Lean actually typable in pmacs. +1,855 abbreviations vendored from vscode-lean4, registered as a chain +consumer ahead of auto-pairing. This is the stage that makes Lean +actually typable in pmacs. **Stage 5 — the goal view.** A `*lean-goal*` panel that renders `$/lean/plainGoal` at the point, refreshed on a debounced tick and on @@ -405,7 +536,7 @@ injections_query }`. Adding a grammar is one entry plus one `Cargo.toml` line; the doc comment at `src/syntax.rs:756` says exactly this and it has held for every grammar since. -`builtin/runtime/syntax.lua:452` `detect_buffer_language` resolves, in +`builtin/runtime/syntax.lua:457` `detect_buffer_language` resolves, in order: modeline → `pmacs.parse.language_for_path` (the grammar extension table) → `pmacs.lsp.filetypes[ext]` → `pmacs.parse.language_from_filename` → shebang. A grammar entry claiming `lean` therefore resolves `.lean` @@ -486,13 +617,13 @@ and pin it.* - `pmacs.lsp` already exposes generic `send_request(id, method, params)` → request id and `send_notification(id, method, params)` - (`src/lua_bindings/mod.rs:9342`, `:9361`). Non-standard methods need no + (`src/lua_bindings/mod.rs:9507`, `:9527`). Non-standard methods need no new Rust to *send*. - `LspEventKind` (`src/lsp.rs:264`) has generic `Notification { method, params }` and `Response { id, result, error, method }` variants. Unknown server methods are delivered, not dropped. - **But `events_take` has exactly one consumer**: `handle_server_requests` - at `builtin/runtime/lsp.lua:1549`, driven off `pmacs._async.tick`. It + at `builtin/runtime/lsp.lua:1815`, driven off `pmacs._async.tick`. It `take`s — a drain. Its `if/elseif` chain handles five `request` methods and `initialized`, and **ignores every `notification` and every `response`**. A second module calling `events_take` would steal events @@ -556,7 +687,7 @@ character": subscribe to `buffer.after-edit`, gate on `ed.this_command() == "buffer.self-insert"` (`pair.lua:229`), then take the exact provenance record. -`pmacs.editor.take_typed_edit()` (`src/lua_bindings/mod.rs:12827`) returns +`pmacs.editor.take_typed_edit()` (`src/lua_bindings/mod.rs:12990`) returns `{ buffer, window, codepoint, char, requested_start, requested_end, effective_start, effective_end, inserted_len, post_cursor, clean }` — or nil. Its doc comment is explicit: @@ -570,7 +701,19 @@ it on every self-insert. A Lean abbreviation expander that independently calls `take_typed_edit()` in the same `buffer.after-edit` fan-out gets nil or steals it from auto-pairing, depending on hook order — and hook order is not a contract. This is the single load-bearing constraint on Stage 4 and -the reason Stage 4 is its own PR rather than a rider on Stage 1. +the reason Stage 4 is its own PR rather than a rider on Stage 1 — and, +after round 5, the reason its substrate half is Stage 4a rather than a +first commit on a Lean branch. + +Re-verified at `d400f30`: `pair.lua:226` remains the **only** production +caller. The eight other call sites in the tree are all in +`tests/auto_pair_acceptance.rs`. So the chain Stage 4a introduces has +exactly one consumer to migrate, which is what makes a no-behavior-change +substrate PR possible at all. + +Two producers arm the record, not one, and §2.11 is where that matters: +the dispatch fallback and — under `#[cfg(feature = "crdt")]` — the +optimistic CRDT arm reached from `handle_remote_crdt_op`. Related, from `pair.lua:30`'s Q#AP1 note: only the nine built-in pair chars `()[]{}"'` and backtick are excluded from the frontends' optimistic @@ -680,6 +823,78 @@ The publish path absorbs into the Rust store *and* still delivers the notification to `events_take`, so Lua can observe them; but suppressing them from the store needs a Rust-side policy, not a Lua filter. Q#LN18. +### 2.11 The upstream input method (external, verified by reading it) + +Scouted 2026-07-26 against `leanprover/vscode-lean4` @ `17d1d08`, +package `lean4-unicode-input`, files `AbbreviationProvider.ts`, +`TrackedAbbreviation.ts`, `AbbreviationRewriter.ts`, +`AbbreviationConfig.ts`, and `abbreviations.json`. The package ships no +README, so the algorithm below is read off the source. Apache-2.0. + +**Resolution.** `findSymbolsByAbbreviationPrefix(p)` collects every key +having `p` as a prefix, sorts them by **key length ascending**, and maps +to symbols. `getReplacementText(a)`: + +1. If any key has `a` as a prefix, return the shortest such key's symbol. +2. Otherwise recurse on `a` minus its last character; if that yields + something, return it **with the dropped character appended**. +3. Otherwise undefined — no expansion. + +Verified against the table: `alpha` → `α`, `alp` → `α` (via `alpha`), +`al` → `∀` (via `all`, *not* `alpha` — shortest wins, and this is +surprising enough to be worth an acceptance criterion), `alp7` → `α7` +via rule 2, `a` → `α` (`a` is itself a key, among 29 prefix matches). + +**Tracking.** The leader `\` is inserted into the buffer like any other +character, and the tracked range starts after it; the replaced range +spans the leader inclusive (`abbreviationRange.moveKeepEnd(-1)`). So the +buffer literally shows `\alpha` until expansion, then that whole span +becomes `α`. + +**Termination.** There is no terminator set. On each typed character +`c`, if `findSymbolsByAbbreviationPrefix(a .. c)` is empty the +abbreviation is marked `finished`, **`c` is not absorbed into it**, and +the pending text expands before `c` lands. Otherwise `c` extends the +key. Two consequences the obvious "space ends it" model gets wrong: + +- `'+ '` is a key, so after `\+` a space **extends**. Space is a + terminator by consequence, never by rule. +- `'\'` is a key (→ `\`), so `\\` extends, is uniquely complete, and + eagerly expands to one backslash. A second `\` terminates only when + the pending key is non-empty and unextendable — at which point the + rewriter starts a *new* tracked abbreviation on it. + +**Eager expansion.** When `eagerReplacementEnabled`, an abbreviation +expands the moment it is *unique and complete*: exactly one key has it +as a prefix, and it is itself a key. 1,550 of the 1,855 keys qualify; +the other 305 are proper prefixes of some other key and must wait for +termination. `\to` is in the first group — it expands with no terminator +typed, which is why acceptance 41 is meaningful and not a restatement of +38. + +**Cursor placement.** `$CURSOR` is stripped from the symbol and its +index becomes the post-expansion point, applied only when the point sat +at the end of the abbreviation. 26 values carry it. + +**Abandonment.** Upstream expands on `changeSelections` — any tracked +abbreviation the cursor has left. pmacs has no cursor-motion hook +(round-5 finding 3), so this seam does not exist here and Q#LN22 makes +abandonment lazy instead. + +**The re-arm guard pmacs does not need.** `setminus` → `\` and `n` → +`\n`, so an expansion can insert a backslash; upstream sets +`doNotTrackNewAbbr` across the replace so that backslash does not open a +new abbreviation. In pmacs the expansion is a programmatic `buf:replace` +that arms no typed-edit record, so the chain sees nothing and cannot +re-arm. The guard is unnecessary here **because of** the provenance +contract, not by accident — and the acceptance must pin it, because a +future consumer that inferred from buffer text rather than provenance +would reintroduce the bug. + +**What pmacs does not have to carry.** Multi-cursor. Upstream tracks a +`Set` and sorts changes bottom-up for that reason; +pmacs has one point, so one pending abbreviation per buffer. + ## 3. Decisions ### Q#LN1 — Bundle `arborium-lean` 2.18; reject `tree-sitter-lean4` @@ -911,7 +1126,7 @@ file's directory. **How the walk tests for the marker — and why not the obvious way.** `pmacs.fs.stat` is asynchronous: it returns an awaitable handle -(`builtin/runtime/fs.lua:93`) that only settles under `:await()` inside a +(`builtin/runtime/fs.lua:133`) that only settles under `:await()` inside a coroutine. The resolver has no coroutine. It runs synchronously inside `ensure_server` ← `attach_buffer` ← the `buffer.after-load` hook, so awaiting is not merely slow there, it is unavailable — and blocking the @@ -1102,86 +1317,192 @@ The binding is general, not Lean-shaped: it serves every future function-valued `root`, and it is what lets #161's doc comment stop warning about a footgun and start naming a fix. -### Q#LN10 — Stage 4 mechanism: one shared provenance read, not two +### Q#LN10 — Stage 4a: one shared provenance read, not two The hazard is §2.6 — `take_typed_edit()` is one-shot and `pair.lua` -already consumes it. +already consumes it. A second independent caller in the same +`buffer.after-edit` fan-out gets nil or steals the record, depending on +hook order, and hook order is not a contract. Decision: **`pair.lua` stops being the sole consumer.** Extract the provenance read into a single `buffer.after-edit` subscriber owned by a -small shared module, which takes the record once and passes it to an -ordered list of typed-edit consumers (auto-pair, Lean abbreviation). -Consumers return whether they handled the edit; the first that does stops -the chain. +small shared module — `builtin/runtime/typed_edit.lua`, loaded +immediately before `pair.lua` — which takes the record once and offers +it to registered consumers in a defined order. A consumer returns +whether it **claimed** the edit; the first that claims stops the chain. -Two consequences worth stating up front: +`pmacs.typed_edit.add_consumer { name = , priority = , +fn = function(rec) ... end }`, lowest priority first, ties broken by +registration order. Priority is an explicit number rather than +load-order-implied because Q#LN22's collision makes ordering +load-bearing, and rev 5's "the abbreviation consumer runs first" is a +claim a reader must be able to check without reconstructing +`src/editor.rs`'s include list. -- This touches `pair.lua`, which is load-bearing for auto-pairing - acceptance. The full pairing suite is a required gate for Stage 4, and - the refactor lands *first*, as its own commit with no behavior change, - so a regression bisects cleanly. -- Ordering is a contract, not an accident, and the collision is real: - **64 of the 1,855 abbreviation keys contain a character in the proposed - `lean4` pair set** — `\[[]]` → `⟦⟧`, `\(())` → `⸨⸩`, `\{{}}` → `⦃⦄`, - `\{}` → `{$CURSOR}`. With pairing first, typing `\[` inserts `[]` - with the point between, so the pending key is corrupted to `\[]` before - the second `[` is ever typed and `\[[]]` becomes unreachable. The - abbreviation consumer runs first. +**Stage 4a ships this and nothing else.** Its whole content is: +`typed_edit.lua`, `pair.lua` re-expressed as one registered consumer, +and the `include_str!` line. Round 5's finding 1 is why this is a PR and +not a first commit — `pair.lua` is every language's auto-pairing, and a +reviewer looking at a Lean PR should not have to also review a rewrite +of it. - (Rev 1 justified this with `\<>`, which was wrong: `<` is not in the - pair set per Q#LN6, so that key is safe under either order.) +**The no-behavior-change claim must be pinned, not asserted.** The full +`tests/auto_pair_acceptance.rs` suite is a required gate for 4a and must +pass **unmodified** — a suite edited to accommodate the refactor proves +nothing (the recorded lesson: what a test suite pins is its assertions). +Three assertions the existing suite already makes are the load-bearing +ones, because they are what a chain could plausibly break: that a second +`take_typed_edit()` in the same fan-out yields nil, that pairing still +sees the exact record via `_capture_records`, and that the Q#AP7 ordering +against `lsp.lua`'s `didChange` flush still holds. -**The contract that collision exposes:** the abbreviation consumer must -claim a self-insert that **extends an open pending abbreviation**, not -only one that completes an expansion. A consumer that only claims -completed expansions hands every intermediate keystroke to auto-pairing, -which is exactly how `\[` gets corrupted. "Claimed" here means the chain -stops, not that an edit was made. +**What 4a deliberately does not do.** It does not change the `all-must- +succeed` contract, so a consumer that throws still fails the fan-out for +everyone. The chain owner therefore `pcall`s each consumer and reports +through `pmacs.editor.set_status`, matching `pair.lua`'s existing +never-throw-from-after-edit discipline — this is behavior-preserving for +pairing (which already never throws) and is the guardrail 4b needs. -Expansion semantics (matching vscode-lean4 and `lean4-input`): - -- `\` opens a pending abbreviation, tracked per buffer with its start - offset. Every subsequent self-insert that extends it is claimed. The - pending state is abandoned on any non-self-insert command, buffer - switch, or cursor move away from the pending region. -- Expansion fires on a unique complete match that no longer key extends, - or on an explicit terminator (space, tab, RET, or a second `\`). -- The vendored table's `$CURSOR` placeholder becomes the point position - after the replace — this is how `\<>` yields `⟨|⟩`. -- The whole expansion is **one `buf:replace`** — one undo step, one CRDT - op, one effective-edit verification. Same discipline as - `comment.lua`'s Q#CT5. -- Gated by `pmacs.config.define{ name = "lean.abbrev", type = "boolean", - default = true, mutability = "live" }`, read against the *source* buffer - of the typed edit — the `editing.auto-pair` precedent (`pair.lua:44`), - including its round-2 correction to resolve `rec.buffer` rather than - `pmacs.window.buffer()`. - -### Q#LN11 — Stage 4 data: vendor the table, generated, attributed +### Q#LN11 — Stage 4b data: vendor the table, generated, attributed `abbreviations.json` in `leanprover/vscode-lean4` is a flat -`string → string` object of **1,855 entries** (counted, not estimated), -of which **64 contain a character in the `lean4` pair set** — the -collision Q#LN10's ordering exists to handle. vscode-lean4 is Apache-2.0. +`string → string` object of **1,855 entries**, verified at commit +`17d1d08` (2026-05-29), 36,861 bytes, all keys ASCII, longest key 25 +characters. The counts the algorithm depends on, all re-derived from the +file rather than estimated: + +| Count | What it drives | +|---|---| +| 64 keys containing a `lean4` pair-set char | Q#LN22's ordering | +| 305 keys that are proper prefixes of another | which keys can expand eagerly | +| 1,550 keys uniquely-and-completely matching | the eager-expansion set | +| 26 values containing `$CURSOR` | point placement | +| 93 multi-codepoint values | the replace is not one-char-for-many | + +vscode-lean4 is Apache-2.0. Vendor it as a generated `builtin/runtime/lean_abbrev.lua` with a header -recording source repo, commit, license, and the regeneration command — -the `builtin/queries/latex/highlights.scm` precedent (#144) for -third-party data, extended with provenance because this is a much larger -artifact under a named license. +recording source repo, commit, license, entry count, and the +regeneration command — the `builtin/queries/latex/highlights.scm` +precedent (#144) for third-party data, extended with provenance because +this is a much larger artifact under a named license. -Not fetched at runtime, not a package-manager dependency: the input method -must work offline and on first launch. +Not fetched at runtime, not a package-manager dependency: the input +method must work offline and on first launch. -**Upkeep is a documented manual process, not code.** There is no automatic -sync and none is wanted — an editor that silently re-downloads its input -method has a supply-chain problem, not a feature. The generator script -lives at `scripts/regen-lean-abbrev`, takes a vscode-lean4 commit as its -argument, and rewrites the file including its provenance header. The -header records source commit, license, entry count, and the regeneration -command, so the file is self-describing to whoever next touches it. A -refresh is an ordinary PR with a visible diff — which is the point: the -diff is the review. +**Embedded, not lazily loaded.** ~45 KB of generated Lua joins the 414 KB +of builtin runtime already compiled in by `include_str!`, of which +`lsp.lua` alone is 111 KB. Inventing a lazy-load path for an 11% increase +would be new machinery bought with no measurement, and the arithmetic is +stated here so a reviewer can disagree with it on numbers. + +**Upkeep is a documented manual process, not code.** There is no +automatic sync and none is wanted — an editor that silently re-downloads +its input method has a supply-chain problem, not a feature. The generator +script lives at `scripts/regen-lean-abbrev`, takes a vscode-lean4 commit +as its argument, and rewrites the file including its provenance header, +so the file is self-describing to whoever next touches it. A refresh is +an ordinary PR with a visible diff — which is the point: the diff is the +review. + +**The generator must reject a table it cannot faithfully encode.** Keys +are ASCII today but nothing upstream promises that; a key containing a +character the emitted Lua would have to escape, or a duplicate after +normalization, aborts the regeneration rather than silently emitting a +table that disagrees with its source. Same discipline as Q#LN20's +refusal to hand back a lossy path. + +### Q#LN21 — Stage 4b: the expansion's undo is cross-peer-degraded; ship it, name it + +`classify_key` (`src/optimistic.rs:144`) returns `Insert(c)` for `\` and +for every ASCII letter — only the nine built-in pair chars are excluded +(Q#AP1). So on a CRDT frontend the user's `\alpha` arrives as six +**source-peer** optimistic inserts, while the expansion is a single +**daemon-peer** `buf:replace` spanning all six. Undo across that boundary +is not chronologically arbitrated; this is the same defect Q#LN6 already +accepts for `⟨⟩`, `⦃⦄`, `⟮⟯`, one order of magnitude wider. + +Considered and rejected: `pmacs.buffer.set_round_trip_input(buf, true)`, +which exists, is per-buffer, and would fix this exactly. Its six current +callers are all read-only generated buffers — listview, compile, dired, +terminal — and it does considerably more than disable optimistic insert: +per `src/editor_core.rs:505`, `dispatch_idle` reports false, so RET +reaches buffer-local bindings instead of inserting a newline. Turning it +on for every ordinary editable Lean source file would trade a known undo +degradation for an unknown behavior change across the whole editing +surface, and would make Lean the one language whose typing has a +different latency profile. + +Also rejected: adding `\` to the always-round-trip set. It is +frontend-side and language-blind, so this would tax LaTeX, C, shell, and +every string literal in the editor to fix one language. + +Decision: **accept the degradation, name it in the module comment, and +do not paper over it.** The general fix is chronological cross-peer undo +arbitration — already on the standing backlog, and the same fix Q#LN6 +points at. What Stage 4b owes is honesty about scope: this is not "a few +brackets," it is every abbreviation the user types on a CRDT frontend. + +### Q#LN22 — Stage 4b mechanism: lazy abandonment, explicit ordering + +**Ordering.** The abbreviation consumer registers ahead of auto-pairing. +The collision is real: 64 keys contain a `lean4` pair-set character — +`\[[]]` → `⟦⟧`, `\(())` → `⸨⸩`, `\{{}}` → `⦃⦄`, `\{}` → `{$CURSOR}`. +With pairing first, typing `\[` inserts `[]` with the point between, so +the pending key is corrupted to `\[]` before the second `[` is typed and +`\[[]]` becomes unreachable. + +(Rev 1 justified this with `\<>`, which was wrong: `<` is not in the pair +set per Q#LN6, so that key is safe under either order.) + +**The contract the collision exposes:** the consumer must claim a +self-insert that **extends an open pending abbreviation**, not only one +that completes an expansion. A consumer that claims only completed +expansions hands every intermediate keystroke to auto-pairing, which is +exactly how `\[` gets corrupted. "Claimed" means the chain stops, not +that an edit was made. + +**State machine**, per §2.11's ground truth rather than rev 5's +reconstruction of it: + +- `\` typed in a `lean4` buffer opens a pending abbreviation: `{ buffer, + start_offset, text = "" }`, one per buffer, keyed on `rec.buffer`. +- A subsequent self-insert `c` is claimed iff at least one key has + `text .. c` as a prefix; then `text = text .. c`. If it is also + uniquely-and-completely matching (one of the 1,550), expand now. +- If no key extends `text .. c`, expand `text` **first**, then let `c` + land normally — the chain does *not* claim `c`. +- Expansion resolves through §2.11's three-rule `getReplacementText`, + including the suffix rule (`\alp7` → `α7`). +- `$CURSOR` is stripped from the symbol and its index becomes the point. + +**Abandonment is lazy, because there is no cursor-motion hook** (round-5 +finding 3). Pending state is validated at the next typed edit and +discarded when any of these no longer holds: the record's buffer is the +pending buffer; `rec.effective_start` equals `start_offset + 1 + +#text` (the point is still at the end of the pending span); and the +buffer's `revision()` advanced by exactly the pending edit. `buffer. +after-switch` clears it eagerly since that hook *does* exist. The +practical difference from upstream: a user who clicks away mid-`\alp` +and types elsewhere gets the pending state dropped rather than expanded. +Upstream expands it. **This is a deliberate divergence** — expanding +into a region the user has left is the worse failure, and pmacs cannot +detect the departure at the moment it happens. + +**One `buf:replace`** for the whole expansion — one undo step, one CRDT +op, one effective-edit verification, with the same +rejected/altered-by-intercept reporting as `comment.lua`'s Q#CT5 and +`pair.lua`. A rejection drops the pending state; it does not retry. + +**Gate:** `pmacs.config.define{ name = "lean.abbrev", type = "boolean", +default = true, mutability = "live" }`, read against the **source** +buffer of the typed edit — the `editing.auto-pair` precedent +(`pair.lua:46`), including its round-2 correction to resolve +`rec.buffer` rather than `pmacs.window.buffer()`. + +**Language gate:** the consumer opens no pending abbreviation outside a +`lean4` buffer, resolved from `rec.buffer` for the same reason. `\` in a +Rust buffer is an ordinary character and `\[` there still pairs. ### Q#LN12 — Stage 5 sends `$/lean/plainGoal` through a typed Rust request @@ -1418,13 +1739,14 @@ never lands. | 2 | multi-root server affinity | **`ensure_server`, shared by every language** | — | | 3a | notification/response seams + purge; `pmacs.fs.canonicalize` | **the shared event drain, run by every language** | — | | 3b | `lake serve` + probe/latch, Lake root, `waitForDiagnostics` | none — Lean-only files plus one config entry | 1, 2, 3a | -| 4 | Unicode input method | **refactors `pair.lua`'s provenance read** | 1 | +| 4a | typed-edit consumer chain | **refactors `pair.lua`'s provenance read, shared by every language** | — | +| 4b | Unicode input method | none — Lean-only files plus one chain consumer | 1, 4a | | 5 | goal panel | new typed LSP request; panel adopter | 3a, 3b | | 6 | `#eval` / `#check` output channel | **new `LspServerSpec` policy field** | 3b, 5 | | 7 | module hierarchy | listview adopter + one typed Rust request | 3a, 3b | -Four of the eight carry risk that is *not* about Lean — stages 1, 2, 3a, -and 6 each change something every language touches. That is the +Five of the nine carry risk that is *not* about Lean — stages 1, 2, 3a, +4a, and 6 each change something every language touches. That is the organizing principle of the split: **no PR in this arc mixes a cross-cutting substrate change with Lean feature content.** A reviewer looking at Stage 2 sees only `ensure_server`; a reviewer looking at Stage @@ -1436,6 +1758,16 @@ Lean-only. One generalization shipped as Stage 2; extracting the other as 3a is what makes the claim true again. The rule is only worth writing down if it survives contact with a stage that is inconvenient to split. +Round 5 found the *same* rule broken again, by Stage 4, whose risk column +read "refactors `pair.lua`'s provenance read" — every language's +auto-pairing — for a stage described as the Lean input method. Rev 5 had +noticed the shape and answered it with a commit boundary; a commit +boundary is not a review boundary. Twice in two re-scouts is the +interesting part: **this rule is not self-enforcing, and a stage only +looks Lean-only until someone re-reads its own risk column.** Every +remaining stage should be re-checked against it at scout time, not +assumed. + Ordering notes: - **Stage 2 has no Lean in it and could ship independently of this arc.** @@ -1453,11 +1785,21 @@ Ordering notes: `builtin/runtime/lsp.lua`. Unlike stages 1 and 2, this pair is strictly sequential — recorded here, per the #126/#127 lesson, rather than discovered in a rebase. -- **Stage 4 does not depend on stages 2, 3a, or 3b** and could run in - parallel, but should not: both touch `lsp.lua`/`pair.lua`-adjacent - runtime files, and the #126/#127 lesson is that parallel-safety - requires the file split be agreed *before* either lane starts. - Sequential is cheaper. +- **Stage 4a depends on nothing in this arc** — not even Stage 1. It is + a pure runtime-substrate change whose only content is `pair.lua` and a + new module beside it, and it would be worth landing if the Lean arc + were abandoned tomorrow, because "the typed-edit record has exactly + one consumer forever" is not a property anyone chose. +- **4a and 4b cannot run as sibling worktrees**, for the 3a/3b reason: + 4b's consumer is written against the registration API 4a adds. Strictly + sequential, recorded before either starts. +- **Stage 4b depends on stages 1 and 4a and on nothing else** — not on + 2, 3a, or 3b. The input method is useful with no language server at + all, which is the honest ordering argument for putting it this early: + a user with no Lean toolchain installed still gets a Lean editor that + can type Lean. It could run in parallel with the 5/6/7 lane, but + should not, per the #126/#127 lesson that parallel-safety requires the + file split be agreed *before* either lane starts. - **Stage 6 depends on Stage 5** only for the read-only generated-buffer and panel machinery, which Stage 5 establishes. If Stage 5 slips, Stage 6 can carry that machinery itself at the cost of duplicating it. @@ -1496,7 +1838,22 @@ Stated so they can be scored, per house style. inside `buffer.after-edit` re-enters the hook in a way pairing does not already survive. Confidence: medium — pairing does the same thing, but over a single codepoint rather than a multi-byte span. -6. **These eight stages reach rough VS Code parity for everything except +5a. **Lazy abandonment is good enough without a cursor-motion hook** + (rev 6, Q#LN22). Falsified if a user in normal editing hits a case + where stale pending state produces a *wrong* expansion rather than a + dropped one — the failure mode this design chooses. Confidence: + medium-high, because every path that can invalidate the state either + goes through `buffer.after-edit` (where it is checked) or through + `buffer.after-switch` (where it is cleared), and the residual is a + cursor move with no intervening edit, which the next typed edit + catches by position. If it fails, the fix is a cursor-motion hook — + substrate work with its own framing, not a patch to this stage. +5b. **Stage 4a is behavior-preserving.** Falsified by any change to + `tests/auto_pair_acceptance.rs` being needed to make it pass. + Confidence: high, and cheap to score — it is a diff-level check, not + a judgment call. This bet is stated separately from bet 5 because it + is the one a reviewer can falsify in ten seconds. +6. **These nine stages reach rough VS Code parity for everything except the interactive infoview.** Scored honestly rather than aspirationally. What lands: highlighting, goal view, Unicode input, diagnostics, hover, completion, goto-definition, symbols, semantic tokens, `#eval` @@ -1528,10 +1885,26 @@ What remains deferred: - **GPU goal band** — blocked on bottom-panel Stage 2 (Q#LN14). The panel is grid-only until then. - **A `cursor.after-move` hook** — there is none (Q#LN13), so Stage 5 - polls off `process.after-tick`. A real motion hook would serve the goal - view, `completion.lua`'s cursor-delta heuristic, and the outline/hover - panels alike; it is substrate work that should not be invented inside a - language lane. + polls off `process.after-tick` and Stage 4b abandons pending + abbreviations lazily rather than on departure (Q#LN22, round-5 finding + 3). A real motion hook would serve the goal view, the input method, + `completion.lua`'s cursor-delta heuristic, and the outline/hover panels + alike; it is substrate work that should not be invented inside a + language lane. Two consumers in this arc now want it, which is worth + recording as evidence for whoever frames it. +- **Chronological cross-peer undo arbitration** — the general fix for + Q#LN6's bracket pairs and Q#LN21's abbreviation expansions alike. + Already on the standing backlog; named again here because Stage 4b + widens the exposure from three pair characters to every abbreviation a + user types on a CRDT frontend, which changes how often the existing + defect is met without changing what it is. +- **Per-buffer optimistic-apply policy** — the narrower thing Q#LN21 + actually wanted and did not build. `set_round_trip_input` is the only + existing lever and it is too blunt (it also changes RET dispatch); a + frontend-side, language-aware round-trip character set would fix the + undo degradation for Lean without taxing every other language, and + would retire Q#AP1's limitation too. Frontend + protocol work, so + Q#LN14's no-protocol-change rule keeps it out of this arc entirely. - **LSP server reaping / LRU** — Q#LN15's per-root affinity makes unbounded `lake serve` growth possible. No editor caps this by default and pmacs will not either in this arc, but the policy question is now @@ -1758,12 +2131,44 @@ revision take letter suffixes rather than displacing anything. Round 3's finding 4 was stale cross-references surviving a renumber; not renumbering is the cheaper way to not repeat it. -**Stage 4 — the Unicode input method** +**Stage 4a — the typed-edit consumer chain** -38. `\alpha` + space yields `α`; the whole expansion is a single undo step. +Criterion 46 keeps its number and moves here — it was always the +substrate pin, filed under Stage 4 only because Stage 4 was one stage. +Per the no-renumbering rule above, round 5's additions take letter +suffixes on both sides of the split. + +46. **Provenance-refactor pin:** the full `tests/auto_pair_acceptance.rs` + suite passes **unmodified**. A suite edited to accommodate the + refactor proves nothing; the diff for 4a must show zero lines + changed in that file. +46a. The chain reads the record exactly once: with two consumers + registered, a `take_typed_edit()` from inside either observes nil, + and both consumers receive the *same* record fields. Bites against a + chain that re-takes per consumer (which would hand the second one + nil in production and pass a single-consumer test). +46b. Ordering is by declared priority, not registration order: two + consumers registered low-priority-last still run + low-priority-first. Bites against a chain that "works" only because + `include_str!` order happens to agree with intent. +46c. A claiming consumer stops the chain — a later consumer does not + run — and a non-claiming one does not. +46d. A consumer that throws is contained: the fan-out still succeeds, + the other consumers still run, and the failure reports through + `set_status`. Bites against the `all-must-succeed` contract taking + the whole fan-out down with one bad consumer (Q#LN10). +46e. **Q#AP7 ordering survives.** The existing `sighelp` fake-server + test — pairing's closer must be in the buffer before `lsp.lua` + flushes `didChange` — still holds with pairing behind the chain. + Falsified by moving the chain's registration after `lsp.lua`'s. + +**Stage 4b — the Unicode input method** + +38. `\alpha` + space yields `α`; the whole expansion is a single undo + step, and one undo restores `\alpha` rather than `\alph`. 39. `\<>` yields `⟨⟩` with the point between them, from the `$CURSOR` placeholder. -40. **Pair-collision pin (Q#LN10).** `\[[]]` yields `⟦⟧`: each `[` is +40. **Pair-collision pin (Q#LN22).** `\[[]]` yields `⟦⟧`: each `[` is claimed as an extension of the pending abbreviation, so auto-pairing never inserts a closing `]` into the pending key. Bites against an ordering where pairing runs first, and against a consumer that claims @@ -1773,15 +2178,55 @@ renumbering is the cheaper way to not repeat it. 41. `\to` yields `→` eagerly on uniqueness, with no terminator typed. 42. A prefix with no match (`\zzzz` + space) is left as literal text; no edit is made. -43. Moving the cursor out of a pending abbreviation abandons it. +43. **Lazy abandonment (Q#LN22).** Because there is no cursor-motion + hook, this asserts what pmacs can actually detect: after `\alp`, an + explicit `goto_byte` elsewhere followed by typing `h` inserts a + plain `h` and leaves the `\alp` text untouched — the pending state + is dropped, not expanded. Plus: `buffer.after-switch` clears pending + state eagerly. **Rev 5's version of this criterion was not + buildable**; recorded so the change is visible rather than silent. 44. `pmacs.config.set("lean.abbrev", false)` disables expansion; the setting is read against the typed edit's **source** buffer. 45. Expansion does not fire in a non-`lean4` buffer — including that a pending abbreviation is never opened there, so `\[` in a Rust buffer still pairs normally. -46. **Provenance-refactor pin:** the full auto-pairing acceptance suite - passes unchanged, and a bite against the pre-refactor `pair.lua` - confirms the shared-consumer commit is behavior-preserving. +45a. **Shortest-key resolution (§2.11).** `\alp` + space yields `α`, and + `\al` + space yields `∀` — from `all`, not `alpha`. The second is + the one that bites: a "longest match" or "unique match only" + implementation passes the first and fails this. +45b. **Suffix rule.** `\alp7` + space yields `α7`. Bites against an + implementation that drops unmatchable trailing characters or + abandons the whole abbreviation. +45c. **There is no terminator list.** `\+` followed by space extends + rather than terminating, because `'+ '` is a key. Bites against any + implementation with a hardcoded space/tab/RET terminator set — which + is what rev 5 specified. +45d. **`\\` yields a single `\`**, by extension-and-eager-match rather + than by treating the second `\` as a terminator. And after a + *non-empty* pending key, a second `\` does terminate and open a new + abbreviation: `\alpha\to` + space yields `α→`. +45e. **No re-arm through inserted text (§2.11).** `\setminus` + space + yields a literal `\`, and typing an ordinary letter after it inserts + that letter — the inserted backslash opens no pending abbreviation, + because the expansion is a programmatic replace that arms no record. + Bites against a future consumer that infers pending state from + buffer text instead of provenance. +45f. **Both producers, and the CI-darkness stated.** The dispatch path + is pinned by the criteria above. The optimistic CRDT producer + (round-5 finding 4) is pinned by a separate criterion driving + `handle_remote_crdt_op`, which is `#[cfg(feature = "crdt")]` and + therefore **dark in CI and dark in the required gate list**, since + that list runs `--features crdt` only for `--lib`. The PR must + either land that coverage as a `--lib` test where the gate reaches + it, or state in its description that the optimistic path was + verified only locally and name the command. Silence here is the + failure mode — a green CI would otherwise read as covering the path + most users take. +45g. **Table integrity.** The generated `lean_abbrev.lua` round-trips: + its entry count matches the header's declared count, and a spot set + of entries (`alpha`, `to`, `<>`, `+ `, `\`, `n`, `setminus`) matches + `abbreviations.json` byte-for-byte. Bites against a generator that + silently drops or mangles keys (Q#LN11). **Stage 5 — the goal view** @@ -1847,7 +2292,7 @@ renumbering is the cheaper way to not repeat it. it, with the extra success-gate §2.9 forces. - **#110 (auto-pairing)** — `take_typed_edit()` provenance, the fail-closed discipline on transformed source edits, and Q#AP1's optimistic-classifier - limitation. Stage 4 is built on all three. + limitation. Stage 4a generalizes the first; 4b is built on all three. - **#127 (config registry)** — `pmacs.config.define` and the source-buffer-resolution correction. Q#LN10's gate follows `editing.auto-pair` exactly. @@ -1895,8 +2340,8 @@ alongside sixteen other languages — deliberately *not* the typed registry, because moving one language's entry there while the other sixteen stay put would fragment the surface rather than unify it. Migrating `pmacs.lsp.config` wholesale is a config-arc concern; this lane must not -create a precedent that makes it harder. Stage 4's `lean.abbrev` gate is -where this arc does enter the registry, and Q#LN10 already commits to the +create a precedent that makes it harder. Stage 4b's `lean.abbrev` gate is +where this arc does enter the registry, and Q#LN22 already commits to the `editing.auto-pair` shape. **Background-work attribution (§9).** Three pieces of background work, @@ -1928,3 +2373,59 @@ uncapped event queue, the dropped `cfg.restart`, and — unchanged from #161 — surfacing the spawn failure itself. Each is a behavior change for languages other than Lean, and §4's rule is what keeps them out of a Lean PR. + +### 9.1 Coherence impact — stages 4a and 4b (rev 6) + +**Sections served.** §6 (interaction islands) primarily, and in the +*preventing* direction rather than the fixing one — see below. §11 +(config registry) secondarily, by adding one option in the established +shape rather than a new switch mechanism. + +**Golden journey (§2).** No step is touched by 4a. 4b improves step 4 +(editing) for Lean specifically and changes nothing for any other +language: the pending-abbreviation state exists only in `lean4` buffers. +Neither stage changes launch, open, or attach. + +**Interaction islands (§6).** **None added, and this is the load-bearing +claim of Stage 4b.** An input method is the archetypal island: a modal +state where ordinary keys mean something else, usually with its own +keymap, its own escape, and its own set of commands that only work +inside it. Stage 4b deliberately has none of those. There is no keymap, +no dispatch shadow, no mode line indicator, no command that only works +mid-abbreviation, and no key that exits. The pending state is invisible +to every other subsystem, is abandoned by ordinary editing, and its +worst failure is that the user's literal text stays literal. The +`lean.abbrev` switch is an ordinary registry boolean, not an island +toggle. + +Stage 4a's chain is the mechanism that makes that possible, and it also +retires a smaller island risk: today the only way for a second feature to +react to a typed character is to compete with `pair.lua` for a one-shot +record, and the natural workaround — inferring from buffer text — is how +input methods grow their own private state and, eventually, their own +modal surface. + +**Config registry (§11).** One option, `lean.abbrev`, in exactly the +`editing.auto-pair` shape (boolean, `mutability = "live"`, resolved +against the typed edit's source buffer). This is the arc entering the +registry as §9's earlier text predicted, and it is a genuine adoption +rather than a new surface. Stage 4a adds none. + +**Background-work attribution (§9).** Neither stage does background work. +Both are synchronous inside an existing hook fan-out; no process is +spawned, no timer armed, no request issued. There is nothing to attribute +and nothing to worsen — recorded explicitly because "none" is an answer +this section should be able to give without ambiguity. + +**Debt this revision retires.** The unowned assumption that +`take_typed_edit()` has exactly one consumer forever. That was never a +decision — it was the shape of the only caller — and every future +typed-character feature would have had to rediscover it. Stage 4a turns +an accident into an API with a stated ordering contract. + +**Debt this revision names rather than pays.** One, and it is real: +Q#LN21's cross-peer undo degradation, now covering every abbreviation +rather than three bracket pairs. The fix is chronological cross-peer undo +arbitration, already on the standing backlog and already blocking Q#LN6. +Stage 4b makes the existing gap more visible without widening the class +of defect — but "more visible" is the honest word, not "unchanged." From ccdf352258ad959cf88a1bba0fc2fbd5816a74d6 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sun, 26 Jul 2026 10:38:21 -0400 Subject: [PATCH 25/32] fix(panel): close Stage 2A review round 2 (2 P1, 1 P2) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit **P1-1 — the `Invalidated` arm published the panel context on the document wire.** Real bug, and the live half of the routing defect: the semantic peer has ONE statusline slot, so emitting an authoritative-empty payload for every context replaced the document's with the panel's. Now filtered by document-window identity exactly like the `Ready` arm; a panel's own clear belongs to `PanelFrame` in 2B. Pinned by `invalidated_statusline_clears_only_the_document_not_the_panel`, which reproduces the reported shape — two targets instead of one — when the filter is removed. Honest note on the `Ready` arm: its identity selector is **defensive**, not independently falsifiable today, because the document context is captured first so "first context for my frontend" happens to pick it. Rather than leave that as a silent dependency, `the_semantic_fan_out_captures_the_document_first` pins the order and says why it matters. **P1-2 — round-1 finding 3 was not closed; four of my pins were vacuous.** All four confirmed and fixed: - The statusline consumer test discarded `render_frame`'s output. It now observes the WIRE payload from a v18 peer with a registered provider, and asserts non-emptiness so it cannot pass by emitting nothing. - The terminal test compared two NON-terminal buffers, so both routings answered `false`. The document window now holds a REAL terminal, so the routes disagree; reverting `semantic_terminal_key` fails it. - The decorations test used different buffers and an empty selection — again the same answer either way. The panel now displays the declared buffer with a non-empty selection while the document has none. - #1/#3/#21 had no discriminating pin at all. Their only production caller is `dispatcher_loop`, which no test can drive, so this extracts three named seams the loop calls — `document_buffer_to_follow`, `document_cursor_byte`, `peer_displays_buffer_as_document` — and pins each. Also newly pinned: #2 the lazy CRDT upgrade (the census's sharpest case), #7 `Viewport` aligning WITHOUT taking focus, and #9 a focused terminal panel not suppressing the document viewport. **Every one of the nine pins was falsified by revert.** Two needed a second attempt after the first bite came back green. **P2-3 — stale docs.** `StatuslineEvaluationTarget::Semantic`'s documentation described evaluating only the focused window; it now describes the document-plus-side fan-out, the capture order, the identity-selection requirement, and that `active` reports actual focus. The ledger's Stage 2A entry is corrected to five commits, 2,014 CRDT tests, and 16 acceptance tests. Two clippy findings the refactor introduced were fixed: `document_buffer_to_follow` is `crdt`-gated to match its only caller, and the `CursorByte` guard collapses into one `if`. Gates: fmt clean; workspace clippy clean; 1,832 default + 2,014 CRDT library; Stage 2A 16; Stage 1 46; statusline 8; m11_5 2; GPU initial target 14; terminal config 12; folding Stage 2 48; vterm 1/2 10 / 6; M4 121; required GPU 202; `git diff --check` clean. Co-Authored-By: Claude Opus 5 (1M context) --- docs/active-work.md | 43 +-- src/daemon.rs | 343 ++++++++++++++++++++--- src/semantic_render.rs | 26 +- src/statusline.rs | 21 +- tests/bottom_panel_stage2a_acceptance.rs | 259 +++++++++++++++-- 5 files changed, 602 insertions(+), 90 deletions(-) diff --git a/docs/active-work.md b/docs/active-work.md index 98727e5..36a99de 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -492,27 +492,36 @@ implemented and in review.** worktree `../pmacs-bp-stage2a`, **canonical `main` @ `cf54270` integrated** (review round 1, finding 4 — the terminal-config lane #173 also changes `src/editor.rs`, so gates were rerun on the merge - result, not the old combination). Two commits: the classified census routing, then the - painter extraction + acceptance. **No protocol change; no behavior + result, not the old combination). Five commits: the classified census + routing, the painter extraction + acceptance, the lane record, then + the round-1 and round-2 review fixes. **No protocol change; no behavior change for any frontend today** — with `panel_capable = false` for semantic sessions, `primary_document_window` returns `view.active` in every existing configuration, so this is seam adoption that becomes load-bearing in 2B. -- Verification on this branch: `cargo fmt --check` clean; strict - workspace Clippy clean; **1,832 default + 2,009 CRDT** library tests; - new `bottom_panel_stage2a_acceptance` 10/10; bottom-panel Stage 1 - 46; statusline segments 7 default / 8 CRDT; m11_5 semantic 2 CRDT; - GPU initial target 14 CRDT; vterm Stage 1/2 10 / 6; folding Stage 2 - 48; M4 121; required GPU 202; `git diff --check` clean. -- **Both key routings were falsified by revert.** Rerouting - `dispatch_idle_for` (#14, Focus) through `primary_document_window` - fails `focus_class_dispatch_idle_still_tracks_the_focused_window`; - reverting the statusline lookup (#12, Projection) to `view.active` - fails the document-context test. Worth recording: the *structural* - test `focus_and_projection_disagree_in_the_same_state` did **not** - catch the first bite — it compares the two authorities directly, so - only a consumer-level assertion catches a misrouted consumer. Keep - both kinds. +- Verification on the merge result: `cargo fmt --check` clean; strict + workspace Clippy clean; **1,832 default + 2,014 CRDT** library tests; + `bottom_panel_stage2a_acceptance` **16**; bottom-panel Stage 1 46; + statusline segments 8 CRDT; m11_5 semantic 2 CRDT; GPU initial target + 14 CRDT; terminal config 12 CRDT; vterm Stage 1/2 10 / 6; folding + Stage 2 48; M4 121; required GPU 202; `git diff --check` clean. +- **Every routed producer is now pinned at a seam its production caller + uses, and each pin was falsified by revert**: #1 follow, #2 lazy CRDT + upgrade, #3 `CursorByte`, #5 decorations, #7 `Viewport` (aligns + without focusing), #8 `Pointer` (aligns and focuses), #9 the + terminal-context gate, #12 statusline, #21 the publication filter, + plus the focus-class negatives. #1/#3/#21 required extracting three + named helpers, because their only production caller is + `dispatcher_loop`, which no test can drive. +- **Three lessons about the TESTS, not the code, all from review:** + (a) a *structural* test comparing the two authorities directly does + **not** catch a misrouted consumer — only consumer-level assertions + do; (b) a daemon-path test must `register_session` or the event is + dropped at the uninstalled-session check before reaching the code + under test; (c) a discriminating fixture must make the two routings + DISAGREE — comparing two non-terminal buffers, or two windows with no + selection, yields the same answer either way and proves nothing. + Round 2 found four of my own pins vacuous by exactly these shapes. - **Review round 1 closed: 4 P1 + 2 P2, all real.** The P1s were a stale-`Pointer` focus steal (the failed-alignment arm returned the window, so #8's activation focused it before `dispatch_pointer` diff --git a/src/daemon.rs b/src/daemon.rs index 606839d..63f6b87 100644 --- a/src/daemon.rs +++ b/src/daemon.rs @@ -1145,14 +1145,7 @@ fn dispatcher_loop( .session_state(*fid) .is_some_and(|s| s.negotiated_capabilities.semantic_render) { - // Bottom-panel §1.3 #1 — Projection. The buffer this - // frontend DISPLAYS AS ITS DOCUMENT, not the one it - // happens to focus: focusing a panel must re-send no - // snapshot and must never swap the replica's mirror. - let active_now = { - let core = editor.core.borrow(); - core.primary_document_buffer(*fid) - }; + let active_now = document_buffer_to_follow(editor, *fid); if let Some(active_now) = active_now && last_active_buffer_sent.get(fid) != Some(&active_now) { @@ -1433,25 +1426,15 @@ fn dispatcher_loop( && session_registry .session_state(*fid) .is_some_and(|s| s.negotiated_capabilities.crdt_replica) + && let Some((buffer_id, byte_pos)) = document_cursor_byte(editor, *fid) { - // Bottom-panel §1.3 #3 — Projection. `CursorByte` is - // the replica's authoritative DOCUMENT cursor; a - // focused panel must not retarget it at the panel - // buffer (Q#BP14's "active buffer is a - // document-surface term, not an input-focus term"). - let core = editor.core.borrow(); - if let Some(window) = core - .primary_document_window(*fid) - .and_then(|win_id| core.windows.get(&win_id)) - { - let cursor_byte_msg = InstanceMessage::CursorByte { - buffer_id: window.buffer_id, - byte_pos: window.cursor, - }; - if let Err(e) = write_message(stream, &cursor_byte_msg) { - eprintln!("pmacs: write CursorByte for {fid:?} failed: {e}"); - write_failed = true; - } + let cursor_byte_msg = InstanceMessage::CursorByte { + buffer_id, + byte_pos, + }; + if let Err(e) = write_message(stream, &cursor_byte_msg) { + eprintln!("pmacs: write CursorByte for {fid:?} failed: {e}"); + write_failed = true; } } } @@ -2524,13 +2507,7 @@ fn publish_buffer_snapshot_to_replicas( continue; } if session.negotiated_capabilities.semantic_render { - // Bottom-panel §1.3 #21 — Projection. "Displays this - // buffer" means the peer's DOCUMENT surface: testing the - // focused window would both miss a buffer visible in the - // document (panel focused elsewhere) and replace the peer's - // mirror for one visible only in a panel. - let displays_buffer = - editor.core.borrow().primary_document_buffer(*peer_id) == Some(buffer_id); + let displays_buffer = peer_displays_buffer_as_document(editor, *peer_id, buffer_id); if !displays_buffer { continue; } @@ -2969,6 +2946,53 @@ fn handle_remote_crdt_op( /// whole switch. This is the input/display alignment fix for B1: the /// frontend's *declared* buffer becomes the buffer its keys edit and /// its `CursorByte` reports. +/// The buffer a semantic frontend DISPLAYS AS ITS DOCUMENT — the +/// buffer-follow / `BufferSnapshot` re-send target (bottom-panel §1.3 +/// #1, Projection). +/// +/// Not the focused buffer: focusing a panel must re-send no snapshot and +/// must never swap the replica's document mirror. Named as its own +/// function so the rule is pinnable — its only caller is +/// `dispatcher_loop`, which no test can drive. +#[cfg(feature = "crdt")] +fn document_buffer_to_follow( + editor: &EditorState, + fid: FrontendId, +) -> Option { + editor.core.borrow().primary_document_buffer(fid) +} + +/// The `(buffer, byte)` a semantic replica's authoritative `CursorByte` +/// describes (bottom-panel §1.3 #3, Projection). +/// +/// Q#BP14's vocabulary split: "active buffer" in the replica is a +/// DOCUMENT-SURFACE term, not an input-focus term, so a focused panel +/// must not retarget the document caret at the panel's buffer. +fn document_cursor_byte( + editor: &EditorState, + fid: FrontendId, +) -> Option<(crate::buffer::BufferId, u64)> { + let core = editor.core.borrow(); + let win_id = core.primary_document_window(fid)?; + let window = core.windows.get(&win_id)?; + Some((window.buffer_id, window.cursor)) +} + +/// Whether `peer_id` displays `buffer_id` on its DOCUMENT surface — the +/// `BufferSnapshot` publication recipient filter (bottom-panel §1.3 #21, +/// Projection). +/// +/// Testing the focused window instead would both miss a buffer visible +/// in the document while a panel holds focus, and replace the peer's +/// document mirror for a buffer visible only in a panel. +fn peer_displays_buffer_as_document( + editor: &EditorState, + peer_id: FrontendId, + buffer_id: crate::buffer::BufferId, +) -> bool { + editor.core.borrow().primary_document_buffer(peer_id) == Some(buffer_id) +} + /// Align a semantic frontend's **primary document window** to the /// buffer it declared (bottom-panel §1.3 #7, Q#BP14). /// @@ -4770,4 +4794,257 @@ mod tests { "non-vacuity: the document window is a real, distinct focus target" ); } + + /// Bottom-panel §1.3 #1/#3/#21 — the three Projection producers whose + /// only production caller is `dispatcher_loop`, pinned at the named + /// seams that loop calls. Round 2 finding: reverting any of them to + /// `active_window_for` previously left every test green. + #[cfg(feature = "crdt")] + #[test] + fn tick_producers_describe_the_document_while_a_panel_is_focused() { + let (editor, fid, document, panel) = panel_focused_semantic_fixture(); + let (doc_buf, panel_buf, doc_cursor) = { + let core = editor.core.borrow(); + ( + core.windows[&document].buffer_id, + core.windows[&panel].buffer_id, + core.windows[&document].cursor, + ) + }; + assert_ne!(doc_buf, panel_buf, "fixture: distinct buffers"); + + // #1 buffer-follow / BufferSnapshot re-send target. + assert_eq!( + document_buffer_to_follow(&editor, fid), + Some(doc_buf), + "#1: the follow target must be the DOCUMENT buffer, not the focused panel's" + ); + + // #3 CursorByte. + assert_eq!( + document_cursor_byte(&editor, fid), + Some((doc_buf, doc_cursor)), + "#3: CursorByte must describe the DOCUMENT surface" + ); + + // #21 publication recipient filter, both directions. + assert!( + peer_displays_buffer_as_document(&editor, fid, doc_buf), + "#21: a buffer visible in the document must still receive publications while a panel holds focus" + ); + assert!( + !peer_displays_buffer_as_document(&editor, fid, panel_buf), + "#21: a buffer visible only in a panel must NOT replace the document mirror" + ); + } + + /// Bottom-panel §1.3 #2 — the sharpest census case: the lazy CRDT + /// upgrade BROADCASTS a snapshot, so keying it on focus would let + /// focusing a fresh generated panel buffer swap every peer's mirror. + #[cfg(feature = "crdt")] + #[test] + fn lazy_crdt_upgrade_never_targets_a_focused_panel_buffer() { + let (editor, fid, document, panel) = panel_focused_semantic_fixture(); + let (doc_buf, panel_buf) = { + let core = editor.core.borrow(); + ( + core.windows[&document].buffer_id, + core.windows[&panel].buffer_id, + ) + }; + + let upgraded = ensure_active_buffer_crdt_backed(&editor, fid); + assert_eq!( + upgraded, + Some(doc_buf), + "#2: the upgrade must target the DOCUMENT buffer" + ); + assert_ne!( + upgraded, + Some(panel_buf), + "#2: focusing a panel must never trigger its buffer's upgrade+broadcast" + ); + } + + /// Bottom-panel §1.3 #7 vs #8 — `Viewport` aligns WITHOUT moving + /// focus; only `Pointer` activates. Driven through the real + /// dispatcher seam. + #[cfg(feature = "crdt")] + #[test] + fn viewport_aligns_the_document_without_taking_focus_from_the_panel() { + let (mut editor, fid, document, panel) = panel_focused_semantic_fixture(); + let other = { + let mut core = editor.core.borrow_mut(); + core.registry.borrow_mut().create("*other*") + }; + + dispatch_one_semantic_event( + &mut editor, + fid, + FrontendEvent::Viewport { + frontend_id: fid, + buffer_id: other, + visible: pmacs_protocol::ByteRange { start: 0, end: 0 }, + generation: 0, + }, + ); + + assert_eq!( + editor.core.borrow().views[&fid].active, + panel, + "#7: a document Viewport must NOT move focus out of the panel" + ); + assert_eq!( + editor.core.borrow().windows[&document].buffer_id, + other, + "#7: it must still have ALIGNED the document window to the declared buffer" + ); + } + + /// Shared fixture: a semantic frontend with a document window and a + /// FOCUSED bottom panel. `panel_capable` is set explicitly because + /// Stage 1 ships `false` for semantic sessions and 2B flips it for a + /// v21-negotiated peer. + #[cfg(feature = "crdt")] + fn panel_focused_semantic_fixture() -> ( + crate::editor::EditorState, + FrontendId, + crate::window::WindowId, + crate::window::WindowId, + ) { + use crate::window::{FrontendView, Layout, LayoutNode, Orientation, Window, WindowParams}; + + let editor = crate::editor::EditorState::new(); + let fid = FrontendId(91); + let (document, panel) = { + let mut core = editor.core.borrow_mut(); + let doc_buf = core.active_window().buffer_id; + let panel_buf = core.registry.borrow_mut().create("*panel*"); + let document = crate::window::WindowId::next(); + let panel = crate::window::WindowId::next(); + let (doc_view, panel_view) = { + let reg = core.registry.borrow(); + ( + crate::text_view::TextView::new(reg.get(doc_buf).expect("doc")), + crate::text_view::TextView::new(reg.get(panel_buf).expect("panel")), + ) + }; + core.windows + .insert(document, Window::new(document, doc_buf, doc_view)); + let mut panel_window = Window::new(panel, panel_buf, panel_view); + let mut params = WindowParams::default(); + params.side = Some(crate::window::Side::Bottom); + params.fixed_rows = Some(4); + panel_window.params = params; + core.windows.insert(panel, panel_window); + core.register_frontend_view( + fid, + FrontendView { + layout: Layout { + root: LayoutNode::Split { + orientation: Orientation::Horizontal, + children: vec![LayoutNode::Leaf(document), LayoutNode::Leaf(panel)], + weights: vec![1, 1], + }, + }, + active: panel, + fold_projection: false, + panel_capable: true, + frame_geometry: None, + panel_hidden: false, + }, + ); + (document, panel) + }; + editor.sync_frame_geometry(fid, CellSize::new(24, 80)); + (editor, fid, document, panel) + } + + /// Drive ONE authenticated semantic event through the real + /// dispatcher. The session must be registered or the event is + /// dropped at the uninstalled-session check before reaching any + /// handler. + #[cfg(feature = "crdt")] + fn dispatch_one_semantic_event( + editor: &mut crate::editor::EditorState, + fid: FrontendId, + event: FrontendEvent, + ) { + let mut render_states = HashMap::new(); + let mut semantic_states = HashMap::new(); + semantic_states.insert(fid, crate::semantic_render::SemanticRenderState::new(fid)); + let mut streams = HashMap::new(); + let mut term_sizes = HashMap::new(); + term_sizes.insert(fid, CellSize::new(24, 80)); + let mut last_idle = HashMap::new(); + let mut last_active = HashMap::new(); + let mut bells = HashMap::new(); + let mut registry = SessionRegistry::new(); + registry.register_session( + fid, + crate::presence::SessionState { + negotiated_protocol_version: pmacs_protocol::PROTOCOL_VERSION, + negotiated_capabilities: crate::protocol::NegotiatedCapabilities { + semantic_render: true, + crdt_replica: true, + ..Default::default() + }, + color_slot: 0, + }, + ); + handle_dispatcher_event( + DispatcherEvent::FrontendEvent { source: fid, event }, + editor, + &mut render_states, + &mut semantic_states, + &mut streams, + &mut term_sizes, + &mut last_idle, + &mut last_active, + &mut bells, + &mut registry, + ); + } + + /// Bottom-panel §1.3 #9 — Projection. The `Viewport` terminal-context + /// gate asks "is this frontend's DOCUMENT surface a terminal", so a + /// focused TERMINAL PANEL must not suppress the still-visible + /// document's viewport. + #[cfg(feature = "crdt")] + #[test] + fn a_focused_terminal_panel_does_not_suppress_the_document_viewport() { + use crate::terminal::TerminalSpec; + + let (mut editor, fid, document, panel) = panel_focused_semantic_fixture(); + let other = editor.core.borrow().registry.borrow_mut().create("*other*"); + + // A REAL terminal in the focused panel. + let mut spec = TerminalSpec::new("/bin/sh"); + spec.rows = 10; + spec.cols = 40; + let term_buf = editor.open_terminal(spec).expect("a real terminal"); + editor + .core + .borrow_mut() + .install_buffer_in_window(panel, term_buf) + .expect("terminal into the panel"); + editor.core.borrow_mut().focus_window(fid, panel); + + dispatch_one_semantic_event( + &mut editor, + fid, + FrontendEvent::Viewport { + frontend_id: fid, + buffer_id: other, + visible: pmacs_protocol::ByteRange { start: 0, end: 0 }, + generation: 0, + }, + ); + + assert_eq!( + editor.core.borrow().windows[&document].buffer_id, + other, + "#9: a focused TERMINAL panel must not suppress the document viewport — the document window should still have aligned to the declared buffer" + ); + } } diff --git a/src/semantic_render.rs b/src/semantic_render.rs index b63efcf..af08741 100644 --- a/src/semantic_render.rs +++ b/src/semantic_render.rs @@ -985,10 +985,18 @@ impl SemanticRenderState { StatuslineEvaluationOutcome::Invalidated { authoritative_empty, } => { - for context in authoritative_empty - .into_iter() - .filter(|context| context.frontend_id == frontend_id) - { + // Bottom-panel A2A-2: the clear must be filtered by + // DOCUMENT WINDOW exactly like the Ready arm. The + // semantic peer has ONE statusline slot, so publishing + // the panel context's clear here replaces the document's + // payload with the panel's — the same misrouting the + // Ready arm was fixed for, on the clear path. + // + // A panel's own clear belongs to the future panel + // painter (`PanelFrame`, Stage 2B), not to this wire. + for context in authoritative_empty.into_iter().filter(|context| { + context.frontend_id == frontend_id && Some(context.window_id) == document_window + }) { self.emit_statusline_payload(context.buffer_id, Vec::new(), Vec::new(), out); } } @@ -2949,11 +2957,15 @@ mod tests { "stale evaluation retains the prior baseline until snapshot reset" ); + // Bottom-panel A2A-2: the clear is filtered by DOCUMENT window + // identity, so the context under test must BE the document + // window — passing `None` here would assert nothing. + let document_window = crate::window::WindowId::next(); let invalidated = || StatuslineEvaluation { outcome: StatuslineEvaluationOutcome::Invalidated { authoritative_empty: vec![crate::statusline::StatuslineContext { frontend_id: FrontendId::LOCAL, - window_id: crate::window::WindowId::next(), + window_id: document_window, buffer_id, active: true, }], @@ -2961,13 +2973,13 @@ mod tests { new_failures: Vec::new(), }; let mut replacement = Vec::new(); - semantic.emit_statusline_segments(invalidated(), None, &mut replacement); + semantic.emit_statusline_segments(invalidated(), Some(document_window), &mut replacement); assert_eq!( statusline_of(&replacement), Some((buffer_id, Vec::new(), Vec::new())) ); let mut unchanged = Vec::new(); - semantic.emit_statusline_segments(invalidated(), None, &mut unchanged); + semantic.emit_statusline_segments(invalidated(), Some(document_window), &mut unchanged); assert!( unchanged.is_empty(), "the empty invalidation became baseline" diff --git a/src/statusline.rs b/src/statusline.rs index 7bf345f..3eb4ef6 100644 --- a/src/statusline.rs +++ b/src/statusline.rs @@ -215,10 +215,25 @@ pub enum StatuslineEvaluationTarget { /// Frontend whose entire visible layout is evaluated. frontend_id: FrontendId, }, - /// Only the frontend's active window, iff it still displays the declared - /// semantic viewport buffer. + /// The frontend's **primary document window**, iff it still displays + /// the declared semantic viewport buffer, **plus its visible side + /// window** when one exists (bottom-panel Q#BP8 / A2A-2). + /// + /// Two contexts, not one: the document result feeds the semantic + /// `StatuslineSegments` wire, while the side result paints in the + /// panel's own mode line. Unprojected document splits run no + /// callbacks, and a derived-hidden side (Q#BP2b) is omitted because + /// it has no mode line to paint this frame. + /// + /// The document context is captured **first**; consumers must still + /// select by window identity rather than position, since only one of + /// the two may reach the single semantic statusline slot. + /// + /// `active` on each context reports **actual focus**, so a document + /// provider truthfully observes `active = false` while a panel owns + /// focus (Q#BP14, parent acceptance 42). Semantic { - /// Frontend whose focused daemon window is evaluated. + /// Frontend whose document (and visible side) window is evaluated. frontend_id: FrontendId, /// Buffer declared by the semantic viewport. declared_buffer: BufferId, diff --git a/tests/bottom_panel_stage2a_acceptance.rs b/tests/bottom_panel_stage2a_acceptance.rs index 3b6c183..1063e9f 100644 --- a/tests/bottom_panel_stage2a_acceptance.rs +++ b/tests/bottom_panel_stage2a_acceptance.rs @@ -32,6 +32,14 @@ fn exec(s: &EditorState, src: &str) { s.lua_host.lua().load(src.to_string()).exec().unwrap(); } +fn side_window_of(core: &pmacs::editor_core::EditorCore, fid: FrontendId) -> Option { + core.views[&fid].layout.iter_ids().into_iter().find(|id| { + core.windows + .get(id) + .is_some_and(|w| w.params.side.is_some()) + }) +} + fn side_window(s: &EditorState) -> Option { let core = s.core.borrow(); core.views[&FrontendId::LOCAL] @@ -540,48 +548,239 @@ fn consumer_line_numbers_follow_the_document_not_the_focused_panel() { } #[test] -fn consumer_statusline_segments_name_the_document_window() { - use pmacs::protocol::ByteRange; +fn consumer_statusline_segments_carry_the_document_payload_not_the_panel() { + use pmacs::protocol::{ByteRange, InstanceMessage}; use pmacs::semantic_render::SemanticRenderState; - // §1.3 #12 at the producer: the wire segments must be selected by - // the DOCUMENT window even though the fan-out now also evaluates the - // visible side window (A2A-2). + // §1.3 #12 / A2A-2 at the WIRE. Round 2 finding: the previous + // version discarded `render_frame`'s output and only reasserted + // `primary_document_window`, so restoring the producer's + // "first context for my frontend" selector left it green. + // + // The peer must negotiate v18 or no `StatuslineSegments` is emitted + // at all and the assertion would be vacuous a second way. let s = editor(); - let (fid, doc_win, _panel_win, doc_buf) = semantic_frontend_with_focused_panel(&s); + let (fid, _doc_win, _panel_win, doc_buf) = semantic_frontend_with_focused_panel(&s); + let panel_buf = { + let core = s.core.borrow(); + let panel = side_window_of(&core, fid).expect("panel"); + core.windows[&panel].buffer_id + }; - let mut sem = SemanticRenderState::new(fid); + // One provider so a payload exists to misroute. + exec( + &s, + "pmacs.statusline.register({ name = \"probe\", side = \"left\", + face = \"ui.modeline\", fn = function(ctx) return \"X\" end })", + ); + + let mut sem = SemanticRenderState::for_peer(fid, 18); sem.set_viewport(doc_buf, ByteRange { start: 0, end: 0 }, 0); - // Not asserting on message presence (a peer that never negotiated - // v18 emits none); asserting the routing input the producer uses. - let _ = sem.render_frame(&s); + let msgs = sem.render_frame(&s); - assert_eq!( - s.core.borrow().primary_document_window(fid), - Some(doc_win), - "the producer's document-window selector must name the document" + let targets: Vec<_> = msgs + .iter() + .filter_map(|m| match m { + InstanceMessage::StatuslineSegments { buffer_id, .. } => Some(*buffer_id), + _ => None, + }) + .collect(); + + assert!( + !targets.is_empty(), + "non-vacuity: a v18 peer with a registered provider must emit StatuslineSegments" + ); + assert!( + targets.iter().all(|b| *b == doc_buf), + "every StatuslineSegments must target the DOCUMENT buffer; got {targets:?} (document {doc_buf:?}, panel {panel_buf:?})" + ); + assert!( + !targets.contains(&panel_buf), + "the panel's context must never reach the document statusline wire" ); } #[test] -fn consumer_terminal_declaration_cannot_be_claimed_by_a_focused_panel() { - // §1.3 #6/#10/#11 through the real guard: with the panel focused, - // a declaration naming the PANEL's buffer must be refused, because - // the full-window terminal surface is the document window. - let s = editor(); - let (fid, _doc_win, panel_win, doc_buf) = semantic_frontend_with_focused_panel(&s); - let panel_buf = s.core.borrow().windows[&panel_win].buffer_id; +fn consumer_terminal_declaration_resolves_the_document_not_the_focused_panel() { + use pmacs::terminal::TerminalSpec; + + // §1.3 #6/#10/#11 through the real guard. Round 2 finding: the + // previous version compared two NON-terminal buffers, so both the + // old and new routings returned `false` and it could not + // discriminate. Make the DOCUMENT window hold a real terminal: the + // document routing then answers `true` while the old `view.active` + // routing (which names the focused panel) answers `false`. + let mut s = editor(); + let (fid, doc_win, _panel_win, _doc_buf) = semantic_frontend_with_focused_panel(&s); + + let mut spec = TerminalSpec::new("/bin/sh"); + spec.rows = 10; + spec.cols = 40; + let term_buf = s.open_terminal(spec).expect("a real terminal session"); + + // Install the terminal in the DOCUMENT window; the panel keeps its + // own non-terminal buffer and keeps focus. + { + let mut core = s.core.borrow_mut(); + core.install_buffer_in_window(doc_win, term_buf) + .expect("install the terminal in the document window"); + } + let panel_buf = { + let core = s.core.borrow(); + let panel = side_window_of(&core, fid).expect("panel"); + core.windows[&panel].buffer_id + }; assert!( - !s.semantic_terminal_declaration_is_active(fid, panel_buf), - "a focused panel's buffer must not become the document terminal declaration" + s.semantic_terminal_declaration_is_active(fid, term_buf), + "the DOCUMENT window's terminal must be declarable while the panel owns focus" ); - // Non-vacuity: the document buffer is not a terminal either, so pin - // that the guard resolves the DOCUMENT window by asserting the - // window identity the resolver used. - assert_eq!( - s.core.borrow().primary_document_buffer(fid), - Some(doc_buf), - "the terminal resolver's window must be the document window" + assert!( + !s.semantic_terminal_declaration_is_active(fid, panel_buf), + "the focused panel's own buffer must never claim the document declaration" + ); +} + +#[test] +fn invalidated_statusline_clears_only_the_document_not_the_panel() { + use pmacs::protocol::{ByteRange, InstanceMessage}; + use pmacs::semantic_render::SemanticRenderState; + + // Round 2 finding 1. The `Invalidated` arm emits an + // authoritative-empty payload for EVERY context of the frontend. + // Once A2A-2's fan-out yields document + panel, that publishes two + // clears on a wire with ONE statusline slot, so the panel's payload + // replaces the document's. This is the live, observable half of the + // routing bug — the `Ready` arm happens to be safe today only + // because the document context is captured first. + let s = editor(); + let (fid, _doc_win, _panel_win, doc_buf) = semantic_frontend_with_focused_panel(&s); + let panel_buf = { + let core = s.core.borrow(); + let panel = side_window_of(&core, fid).expect("panel"); + core.windows[&panel].buffer_id + }; + assert_ne!(doc_buf, panel_buf, "fixture: the two buffers must differ"); + + // A provider that unregisters itself mid-evaluation is the canonical + // registry-mutation invalidation. + exec( + &s, + r"_G.SL_SELF = pmacs.statusline.register { + name='self-remove', side='left', priority=100, + fn=function() pmacs.statusline.unregister(SL_SELF); return 'STALE' end, + }", + ); + + let mut sem = SemanticRenderState::for_peer(fid, 18); + sem.set_viewport(doc_buf, ByteRange { start: 0, end: 0 }, 0); + let msgs = sem.render_frame(&s); + + let targets: Vec<_> = msgs + .iter() + .filter_map(|m| match m { + InstanceMessage::StatuslineSegments { buffer_id, .. } => Some(*buffer_id), + _ => None, + }) + .collect(); + + assert!( + !targets.contains(&panel_buf), + "an invalidated evaluation must not clear the PANEL's context on the \ + document statusline wire; got {targets:?} (document {doc_buf:?}, \ + panel {panel_buf:?})" + ); +} + +#[test] +fn the_semantic_fan_out_captures_the_document_first() { + use pmacs::statusline::{ + StatuslineEvaluationOutcome, StatuslineEvaluationTarget, evaluate_statusline, + }; + + // The `Ready` arm selects by window identity, so capture order is not + // load-bearing for correctness — but it IS load-bearing for the + // falsifiability of that selector, so pin it explicitly rather than + // leaving a silent dependency. If a future change reorders the + // fan-out, this fails and whoever reads it learns why it mattered. + let s = editor(); + let (fid, doc_win, _panel_win, doc_buf) = semantic_frontend_with_focused_panel(&s); + + let evaluation = evaluate_statusline( + s.lua_host.lua(), + &s.core, + &s.statusline_registry, + StatuslineEvaluationTarget::Semantic { + frontend_id: fid, + declared_buffer: doc_buf, + }, + ); + + match evaluation.outcome { + StatuslineEvaluationOutcome::Ready(windows) => { + assert_eq!(windows.len(), 2, "document + visible side window"); + assert_eq!( + windows[0].context.window_id, doc_win, + "the DOCUMENT context must be captured first" + ); + } + other => panic!("expected Ready, got {other:?}"), + } +} + +#[test] +fn consumer_decorations_follow_the_document_selection_not_the_panel() { + use pmacs::protocol::{ByteRange, InstanceMessage}; + use pmacs::semantic_render::SemanticRenderState; + + // §1.3 #5 — Projection. A selection made inside a FOCUSED PANEL must + // not paint selection decorations into the document's viewport. + // + // To DISCRIMINATE, the panel must display the SAME buffer the + // viewport declares and hold a NON-EMPTY selection while the + // document holds none. With different buffers (the first attempt) + // both routings emit nothing and the test proves nothing. + let s = editor(); + let (fid, doc_win, panel_win, doc_buf) = semantic_frontend_with_focused_panel(&s); + + exec(&s, "PROBE = pmacs.buffer.list()[1]"); + { + let mut core = s.core.borrow_mut(); + // Put real text in the document buffer so a span exists. + { + let reg = core.registry.borrow(); + let _ = reg.get(doc_buf).expect("doc"); + } + // The panel shows the document's buffer and selects a range. + core.install_buffer_in_window(panel_win, doc_buf) + .expect("panel shows the document buffer"); + let panel = core.windows.get_mut(&panel_win).expect("panel"); + panel.selection = Some(pmacs::window::Selection { anchor: 0 }); + panel.cursor = 4; + // The document window selects nothing. + let doc = core.windows.get_mut(&doc_win).expect("doc"); + doc.selection = None; + doc.cursor = 0; + } + + let mut sem = SemanticRenderState::for_peer(fid, 18); + sem.set_viewport(doc_buf, ByteRange { start: 0, end: 8 }, 0); + let msgs = sem.render_frame(&s); + + let selection_decorations: usize = msgs + .iter() + .filter_map(|m| match m { + InstanceMessage::Decorations { segments, .. } => Some( + segments + .iter() + .map(|seg| seg.decorations.len()) + .sum::(), + ), + _ => None, + }) + .sum(); + assert_eq!( + selection_decorations, 0, + "a selection living in the focused PANEL must not decorate the document viewport" ); } From c4fad0731ceb726965964c0deb29a8893a87135a Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sun, 26 Jul 2026 10:39:01 -0400 Subject: [PATCH 26/32] =?UTF-8?q?docs(lean4):=20rev=207=20=E2=80=94=20roun?= =?UTF-8?q?d=206=20review,=20five=20P1s,=20and=20reconcile=20the=20ledgers?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The 4a/4b split held; five P1s against rev 6's own content, all real, all reproduced. Four share a root: rev 6 verified its external facts and under-verified its internal ones. 1. Stage 4a's declared footprint excluded the tests its own acceptance required. 46a-46e cannot live in tests/auto_pair_acceptance.rs, which criterion 46 requires byte-identical. Footprint now names tests/typed_edit_chain_acceptance.rs and gates on it. 2. Pending abbreviation state had the wrong owner. pmacs is multi-frontend: EditorCore.views is per-FrontendId with its own active window, take_typed_edit is already frontend-keyed, and buffer.after-switch fires with no arguments — so a buffer-keyed clear-on-switch lets any frontend discard another's pending abbreviation. Now keyed (frontend, buffer) with a window check, frontend-scoped clearing, a frontend.detached purge, and acceptance 45i, which the buffer-keyed design passes every other criterion without. 3. The shortest-match rule was missing its tie-break: upstream keeps declaration order among equal-length shortest keys, and 101 prefixes have equal-shortest candidates resolving to different symbols (f picks f< over f>). A pairs-iterated Lua map cannot express this, so the vendored artifact is now an ordered sequence and resolution sorts by (#key, source rank). Rev 6 missed this because it declared the package ships no README after a 404 on the package root, with the directory listing showing src/README.md already in hand — a 404 on a guessed path is not evidence of absence, and the README states the rule in one sentence. 4. The generator's rejection rule rejected the current table: \ is a key and " begins eleven, while acceptance 45d requires \ to work. Replaced with canonical lossless escaping; aborts only on duplicate keys, invalid UTF-8, and a failed self-round-trip. 45g no longer claims to diff against abbreviations.json, which is not shipped. 5. Durable and volatile state were not reconciled. agent-handoff.md anchored main at d152120 with neither #167 nor #170 and no Lean arc bullet at all; active-work.md kept 407 lines of merged Stage 1/2/3a/3b history against its own instruction to prune merged entries, under a stale snapshot date. Durable facts moved to the handoff; the ledger keeps only the unlanded Stage 4 lane. Also corrected: 119 multi-codepoint symbols (26 with $CURSOR), not 93; three backslash values, not two; Q#LN22 now states the terminating-\ reprocess rule acceptance 45d depended on; acceptance 38 says the terminator is retained, so undo restores "\alpha " with its space; coherence cites golden-journey step 5, not step 4; and the config-registry prior art points at Q#LN22. Co-Authored-By: Claude Opus 5 (1M context) Claude-Session: https://claude.ai/code/session_011LFvC4FQtux4y32KuevZ7B --- docs/active-work.md | 435 +++---------------------------------- docs/agent-handoff.md | 71 +++++- docs/lean4-mode-framing.md | 324 +++++++++++++++++++++++---- 3 files changed, 371 insertions(+), 459 deletions(-) diff --git a/docs/active-work.md b/docs/active-work.md index 52d44ba..47c3677 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -1,6 +1,6 @@ # Active work — cross-machine resume ledger -**Snapshot: 2026-07-25.** This file records volatile work that has not +**Snapshot: 2026-07-26.** This file records volatile work that has not landed on `main`. Read it after `docs/agent-handoff.md`. Remove completed entries when their PR merges; do not let this become a second permanent backlog. @@ -59,421 +59,40 @@ If it does not, stop and repair the remote/fetch configuration. ## Lean 4 lane (Arc 8) — Stages 1, 2, 3a, 3b MERGED; Stage 4 IN FRAMING -- Stage 1 **merged as #160** (`main` @ `0827dd1`, 2026-07-25, one review - round, all twelve checks green). Branch `githubsucks/lean4-stage1` - retained; it was worked in the shared checkout (no sibling worktree). -- Approved framing: `docs/lean4-mode-framing.md` revision 4, committed as - the branch's first commit (`a382965`) after three review rounds. **Seven - stages**, 19 decisions (Q#LN1–19), 64 acceptance criteria. North star: - match or exceed VS Code's Lean support. -- **Stage 1 implemented; no wire change (protocol stays v20), no LSP, no - frontend change.** Four commits: framing, grammar, theme captures, - editing surface + acceptance. - - `Cargo.toml` + `src/syntax.rs`: `arborium-lean` 2.18 and one - `BUILTIN_LANGUAGES` entry named **`lean4`** (Q#LN2 — the name becomes - the `didOpen` language_id), claiming `.lean` only. - - `src/highlight.rs`: four capture entries — `constructor`, `character`, - `keyword.conditional`, `warning`. - - `builtin/runtime/{comment,pair,syntax}.lua`: `--` comments, the - `⟨⟩ ⦃⦄ ⟮⟯` pair set, the `lean` → `lean4` modeline alias. - - `tests/lean4_stage1_acceptance.rs` plus unit tests in `syntax.rs` / - `highlight.rs`: 12 criteria, 17 tests. -- **Q#LN1's open obligation is discharged.** `tree-sitter-lean4` is - unusable (depends on `tree-sitter ^0.25` directly against our 0.26, - exports no `LANGUAGE` const despite its README, packages no queries); - `arborium-lean` rides `tree-sitter-language 0.1` with a pre-generated - ABI-15 parser. `cargo tree -d` shows no duplicate core. The parse smoke - pins the failure mode that matters: `→`/`∀`/`≥` must produce - `(arrow)`/`(forall)`/`(comparison)`, since a mismatched-core build - degrades silently on exactly those characters rather than failing loudly. -- **Q#LN4 is a deliberate retro-paint of seven language entries**, not - four: `tree_sitter_javascript::HIGHLIGHT_QUERY` is concatenated - base-first into javascriptreact/typescript/typescriptreact. Its shape is - "every capitalized identifier" (`#match? "^[A-Z]"`) plus every Lua table - brace — not "constructors". Pinned in both directions per #146. -- Implementation findings not in the framing: - - `warning` had to move from bold red to bold **bright** red: `number` - is plain `fg(1)`, so `sorry` and an adjacent numeric literal were the - same colour. Found by writing the test. - - `Some(1)` is **not** `@constructor` — in call position a narrower - `@function` pattern wins. Only bare or pattern-position capitalized - identifiers reach it. Pinned so the blast-radius claim stays honest. - - Lean node kinds nest: `module > declaration > def|theorem`. - - `pmacs.parse.injection_aliases` is a documented **write-only** Lua - proxy (canonical map is Rust-side), so fence tests must drive - `_parse_now` and inspect layer languages, never read the table back. -- **Review round 1 addressed.** The finding: acc12's server-list assertion - could not fail for the regression it named — the shared `editor()` - helper wipes `pmacs.lsp.config` before any buffer opens, so - `#pmacs.lsp.list() == 0` holds for every language regardless of what - Stage 1 ships. It now asserts against a **pristine** `EditorState` that - `pmacs.lsp.config.lean4` is nil, with a non-vacuity check that the same - lookup finds `rust`; bite-verified by adding a `lean4` config to - `lsp.lua` and watching it fail. Also fixed a stale column in a - `highlight.rs` comment. -- Verification on this branch: `cargo fmt --check` clean; strict workspace - Clippy clean; 1,826 default + 2,003 CRDT library tests; lean4 Stage 1 - 9/9; comment toggle 14; auto-pair 45; injection 4; M4 121; required GPU - 152; **isolated-config workspace sweep 3,150 across 90 suites**; - `git diff --check` clean. The sweep needs an isolated `XDG_CONFIG_HOME` - for the reason recorded in the bottom-panel lane below. -### Stage 2 — multi-root LSP server affinity (Q#LN15) +- **Stages 1, 2, 3a and 3b are MERGED** — #160 (`main` @ `0827dd1`), + #161 (`46a1b8f`), #167 (`6f348c9`), #170 (`d400f30`). Their full + histories were pruned from this ledger in round 6, per this file's own + instruction to remove entries when their PR merges; the durable facts + now live in `docs/agent-handoff.md` §1's Lean 4 bullet, which is where + a fresh machine should read them. `docs/lean4-mode-framing.md` rev 7 + carries the decisions. -- Portable branch: `githubsucks/lsp-multi-root-affinity`, shared checkout, - based on `githubsucks/main` @ `0827dd1`. Named for the substrate, not - for Lean: **the diff contains no Lean content**, because `ensure_server` - is the one server-affinity function every LSP language shares and a - cross-cutting change to it must not be reviewable only as a Lean - feature. -- Three files, no protocol change: `src/lua_bindings/mod.rs` (the - `lsp.list()` row builder gains `root_uri` + `cwd`), - `builtin/runtime/lsp.lua` (`project_root_for` returns `root, source`; - `ensure_server` hoists it above the reuse loop and matches on it), - `tests/lsp_multi_root_acceptance.rs` (9 tests, acceptance 13–21). -- **The rule that keeps this from regressing every other language: the - affinity key is the root only when a root was actually FOUND.** - `project_root_for` never returns nil for a file with a path — its last - resort is the file's own directory — so a naive `(language_id, root)` - key gives every directory of loose scratch files its own server, for - every language. `source` is `"config" | "detected" | "fallback"` and - only the first two become a key. -- **Wire-identical for the fallback case, and that is provable rather - than hoped.** Matching is on the spawned spec's `root_uri` (nil matching - nil), so the fallback spawn passes `root_uri = nil`; `cwd` still carries - the directory and `build_initialize` derives the identical `rootUri` - from `cwd` when the field is None, using a percent-encoder with the same - allowed set as Lua's `file_uri_for`. `build_initialize` (`src/lsp.rs`) - is the **only** reader of `spec.root_uri` in the tree. -- Deliberate behavior change, asserted not discovered: a server - hand-spawned from `init.lua` with only `cwd` set also reads back nil, so - a root-bearing attach will not adopt it. -- `config[language].root` may now be a `function(path) -> string|nil`, - memoized per directory — needed because the hoist puts root resolution - on every attach rather than every spawn. The memo is keyed **weakly by - the resolver function itself**, so replacing `config[lang].root` cannot - serve a root the previous resolver computed. This is Q#LN8's - generalization landing early; the Lean resolver that uses it is Stage 3. -- Bite-verified three ways: 5/9 fail against the pre-change `lsp.lua`, - 8/9 against the pre-change `mod.rs`, and — the one that matters most — - installing the naive always-key-on-root variant fails acceptance 20 and - 21 exactly as Q#LN15 part 2 predicts. The four that survive the first - bite (13, 15, 16, 19) are the regression pins; passing on both sides is - their job. -- Every fixture sets `pmacs.project.set_search_boundary` at its own - tempdir root. Without it the marker walk climbs to the filesystem root - and a stray `.git` above the temp directory turns the markerless cases - into detected ones — the assertions would still pass while testing - nothing. -- **Found but not fixed here (pre-existing, own lane):** `ensure_server` - never forwards `cfg.restart` to `pmacs.lsp.spawn`, so a - `restart = "never"` in `pmacs.lsp.config[lang]` is silently dropped on - the auto-attach path. At least one existing test sets it believing it - takes effect. Out of scope for a PR whose acceptance 16 pins existing - attach behavior as unchanged. -- **Review round 1 addressed.** The blocker was process, not design: the - test file was committed *before* `cargo fmt` ran, so the fix sat - uncommitted in the working tree and the branch as pushed failed the - first gate. The reported "fmt clean" described the worktree, not the - branch — gate results are only meaningful when run against the pushed - tree. Also added the two pins review asked for (a **string** `config - .root` as an affinity key — acc17 only covered the function form; and - `root = false` reading as unset), each bite-verified against exactly - the mutation it targets and neither against the other. And documented - the canonicalization obligation: the `"detected"` arm is canonicalized - for free, a **configured** root is not, so on macOS a resolver - returning `/var/…` and a detected `/private/var/…` are different keys - for one directory. Stage 3's Lean resolver is the first real consumer, - so the obligation is written at the point of use. -- Verification on this branch: `cargo fmt --check` clean; strict - workspace Clippy clean; 1,826 default + 2,003 CRDT library tests; - multi-root 11/11; M4 121; statusline 7; completion popup 9; auto-pair - 45; required GPU 155; **isolated-config workspace sweep 3,164 across 91 - suites**; `git diff --check` clean. The sweep needs an isolated - `XDG_CONFIG_HOME` and `-- --skip basedpyright`. - -### Stage 3a — dispatch seams + `pmacs.fs.canonicalize` — MERGED #167 (`main` @ `6f348c9`) - -- Worktree `../pmacs-lean-stage3`, branched off `githubsucks/main` @ - `46a1b8f`. Carries framing **rev 5** (the Stage 3 split) as its first - two commits, then the implementation, then a bite-driven correction. -- **Stage 2 merged as #161** (`main` @ `46a1b8f`, 2026-07-25, two review - rounds). COHERENCE.md §7 records the slice; §1.2 records the dead - `pmacs.error` channel found landing it. -- **Framing rev 5 splits Stage 3 into 3a and 3b** because rev 4 broke its - own §4 rule — the row read "two `lsp.lua` generalizations" under prose - claiming Stage 3 was Lean-only. One generalization shipped as Stage 2; - the other (Q#LN9's seams) is the shared event drain, so it is now its - own substrate stage. 3a and 3b are **strictly sequential** — 3b's - subscriber is written against 3a's seam and both touch `lsp.lua`. -- Ships: `pmacs.lsp.on_notification` / `on_response`, two arms in - `handle_server_requests`, a pending-response purge, and - `pmacs.fs.canonicalize` (Q#LN20). No protocol change, no Lean content. -- **Two framing claims were corrected during implementation**, both - recorded in §0.1 finding 6 and in the round-2 commit: - 1. The reachable leak is **not** a killed buffer. The Rust core fires - exactly five hooks (`buffer.after-edit`, `buffer.after-load`, - `buffer.after-switch`, `frontend.detached`, `process.after-tick`) — - **there is no buffer-kill hook**, so nothing tears an attachment - down and the drain keeps reaching that server. The real path is - `attach_buffer` dropping a dead sid from `attachments` and - rebuilding against a fresh server, which makes `crashed`/`stopped` - the event *least* likely to be drained. Hence the purge polls - `pmacs.lsp.list()` rather than riding the drain. - 2. Acceptance 32 does **not** pin "removed before invocation" — - `pcall` catches the raise either way, so before/after is - unobservable without a re-entrant drain. It pins removal being - **unconditional**; renamed accordingly. -- **`pmacs._fs` is installed from `install_async`, not `install_project`**, - purely for load order: `make_workspace` runs *after* `fs.lua` is - evaluated, so a canonicalizer placed there reads nil. This cost one - failing run to discover and is the kind of thing to check first. -- Bites recorded (all against the committed tree): removal gated on a - clean return → acc32 fails 2 != 1; an event-driven purge → the - no-attachment case fails "never called" while the attached case still - passes; a resolver without `canonicalize` → two servers (34b's own - falsification, which ships as a test). -- **Known unpinned:** the purge's generation (`attempt`) check. Reaching - it needs a crash *and* its restart to fall in a gap with no - `_async.tick`; the backoff is 500ms, so any tick sees `crashed` first - and the absent-or-terminal arm fires. Labelled as defensive in the - code rather than left looking covered. -- Verification on this branch: `cargo fmt --check` clean; strict - workspace Clippy clean; 1,826 default + 2,003 CRDT library tests; - dispatch seams 15/15 on Linux (14 on macOS — see below); multi-root - 13/13; M4 121; required GPU 155; **isolated-config workspace sweep - 3,189 across 93 suites, zero failures**; `git diff --check` clean. -- **Two flakes/portability facts from CI round 1, both worth keeping:** - 1. `composition_overhead_under_ten_percent` tripped once in a local - sweep at 18.8% against a 10% budget, then passed 3/3 in isolation - here, passed in isolation on main, and passed a full sweep rerun. - The tell is in its own output: the same run reported realistic-frame - overhead as **-4.6%**, and a negative figure is measurement noise, - not added work. Load-sensitive under a parallel `--workspace` run. - 2. **A non-UTF-8 filename fixture cannot be built on macOS.** APFS - enforces valid UTF-8, so `std::fs::write` fails with EILSEQ - ("Illegal byte sequence") before the code under test is reached. - `#[cfg(unix)]` is NOT sufficient for such a fixture — - `#[cfg(target_os = "linux")]` is. Cost one red CI round to learn. - -### Stage 3b — the Lean language server — MERGED #170 (`main` @ `d400f30`) - -- Same worktree `../pmacs-lean-stage3`, **branched off - `lean4-stage3a-seams`, not off `main`** — 3b consumes 3a's response - seam and `pmacs.fs.canonicalize`, so it is strictly sequential. - **Retarget PR #170 to `main` BEFORE merging #167, not after** — the - kill-ring lesson exactly. (Round 1 of this ledger entry stated the - reverse in its first sentence and the correct rule in the next; the - review caught it. A safety rule written twice with opposite senses is - worse than not written.) -- Ships `builtin/runtime/lean.lua` (new), one `include_str!` line in - `src/editor.rs`, `pmacs.lsp._attach_buffer` exported from `lsp.lua`, - a `leanprogress` mode plus `waitForDiagnostics` validation on - `pmacs_fake_lsp`, and `tests/lean4_server_acceptance.rs` (40 tests). - No protocol change. -- **Stage 1's acceptance 12 is half superseded and was rewritten, not - deleted.** It asserted `pmacs.lsp.config.lean4 == nil` to catch a - Stage-3 front-run; 3b is that stage. What survives is the restraint - half — constructing an editor spawns nothing though the config now - names `lake`, and opening a Lean buffer with no server configured - spawns nothing — which is what holds Q#LN7's "not at init" promise. -- **The marker test is wrong in two opposite directions if done naively** - and both are pinned: `io.open` SUCCEEDS on a directory (so truthiness - accepts a `lean-toolchain` dir), but requiring a non-nil read rejects - an EMPTY `lean-toolchain` (a legitimate marker — existence semantics, - not content). Discriminator is `read`'s SECOND return; decline only on - a non-nil err. Probed on LuaJIT 2.1. -- **Fifteen bites recorded, each against the committed tree.** R1: bare - `io.open` → 24a fails / 24b passes; require-non-nil → 24b fails / 24a - passes; no canonicalization → symlinked open spawns two servers; no - re-attach after the swap → three latch tests fail; hook keyed on the - attachment → the missing-`lake` case fails; `waitForDiagnostics` - without `version` → acc37 fails with InvalidParams. R2: skip retiring - a terminal server → `attempt` reaches 3; no originating-buffer gate → - the Lean buffer is left on the `lake` stub; retry-forever → the - failing-fallback test fails; version-probe any command → the - working-wrapper test fails; no disabled guard → the unconfigured test - sees "`nil` could not be started". R3: verdict keyed on `watching` → - the late-verdict test finds the buffer still on `lake`; `buf_key` - rewritten per load → the second-buffer test fails; hardcoded - `lake serve` → the wrapper-naming test fails. -- **Round-2 review: three more P1 lifecycle defects, suite 20/20 with - all of them live.** (1) The crashed primary respawned forever — - skipping the retire call avoided corrupting terminal servers but left - `next_restart_at` armed. **`forget` is the call for a TERMINAL server** - (it requires terminal state and removes the client, dropping the - restart timer); `stop` is for a live one and corrupts a terminal one. - (2) Re-attachment targeted whatever buffer was active when the async - verdict landed; an unrelated Rust attachment satisfied "a different - server id". (3) A failing fallback retried every tick forever, silent. - Plus two P2s: the Lake version parser was applied to arbitrary wrapper - output, and an UNCONFIGURED `config.lean4` was reported as failure and - latched, poisoning the session. -- **Round-3 review: two more P1s, both asynchronous correlation, suite - 25/25.** (a) `probe.watching` is cleared when the server initializes, - so a SLOW version verdict arrived with nil and retired nothing — - `_attach_buffer` returned the still-live primary and the retry called - it success, so status and config said "fell back" while the buffer - stayed put. **That is the round-1 silent no-op reached through a third - event ordering.** `probe.primary` is now separate from - `probe.watching` and survives initialization. (b) `buf_key` was - rewritten on every Lean `after-load`, so a second Lean buffer opened - before the verdict became the rebuild target while the latch still - watched the first buffer's server. Target buffer and primary server - are one fact and are now armed together, once. Plus a P2: the failure - message hardcoded `lake serve` after the latch became - command-agnostic, sending wrapper users to debug the wrong binary. -- **Round-4 review: one P1, and it is the same defect a FOURTH time.** - `pmacs.lsp.config.lean4` is a single global entry, so swapping its - command invalidates **every** Lean buffer and **every** Lean server — - Q#LN15 gives one per project root. Rounds 1–3 each fixed the repair - for one buffer and one server; round 4 is "repair the armed target, - strand the rest". The shape that finally holds: retire ALL `lean4` - servers on latch, and repair each buffer **lazily and at most once** - when it becomes active (`buffer.after-switch` + the tick), because - `_attach_buffer` is active-buffer-only and cannot reach the others. - The per-buffer once-only bound is what stops a failing fallback - retrying forever — the round-2 defect a naive global repair loop would - have reintroduced for every buffer instead of one. Plus a P2: the - argument-inclusive attribution was implemented but pinned only by - "contains the command name", so a mutation dropping every argument - still passed. -- **Round-5 review: one P1 plus a frontend scope hole, and four more.** - (1) A fallback that SPAWNS and then dies retried forever: the - once-per-buffer guard bounds `_attach_buffer`, not the server it - produced, and `ensure_server` never forwards `cfg.restart` so the - fallback inherits `OnCrash` — respawned by the manager with no - ceiling, silently, because `latched` had disabled the primary's poll. - The fallback now gets its own one-shot die-before-initialize watch. - (2) **Simultaneous frontends**: both repair triggers read the ambient - `pmacs.window.buffer()`, and the daemon restores `active_frontend` to - the last-dispatched one before `tick_processes`, so a Lean buffer - active in ANOTHER frontend gets no `after-switch` and stays stale. - Fixed at the right seam — **make CONSUMPTION safe**: both - `attached_for_active` and `attachment_for_request` now refuse a record - whose server is dead (the former rebuilds, the latter reports none, - since it must not perturb LSP state). Healing at the point of use is - frontend-agnostic, because whichever frontend runs a command is active - while it runs. (3) The retirement sweep selected on `language_id`, so - it stopped USER-spawned Lean servers too; it now keys on the - `default-lean4` label `ensure_server` stamps, which is the derivation - discriminator. (4) `probe.latched` gated repair even when NO swap - occurred, so an already-fallback config was retried and misreported. - Split out `probe.fallback_installed`. (5) The once-per-buffer - assertion counted TABLE KEYS, which cannot distinguish "once per - buffer" from "every tick for one buffer" — cardinality stays 1 either - way. Now a numeric attempt counter; the bite shows **174 vs 1**. -- **Round-6 review: four P1s and one P2, suite 40/40.** (1) General - point-of-use healing treated a crashed OnCrash server as absent and - spawned beside it while its old id still had `next_restart_at` armed; - `attach_buffer` now forgets a terminal record before replacement. - `attachment_for_request` remains non-attaching and preserves the - record, so a same-id restart can recover instead of being orphaned. - (2) The fallback watch was scalar, while Q#LN15 permits simultaneous - per-root servers and lsp.lua can create them without passing through - Lean's repair function. Watches are now per-SID and discover every - config-driven Lean server from a private origin table. (3) The shipped - `lean.wait-for-diagnostics` command bypassed both safe resolvers and - still consumed a stopped record; it now uses a command-safe resolver, - waits asynchronously for a healed replacement to initialize, and the - test requires the real request to finish. (4) When no config swap - occurred, one failed root still swept a healthy root; that arm now - retires only the SID whose verdict fired. (5) `label` is public and - unreserved, therefore not ownership. lsp.lua records successful - config-driven spawns privately, and every Lean lifecycle decision keys - on that origin fact; the user-server pin deliberately collides on - `default-lean4`. All five bites against `19f48d4` discriminate: the - old files produce 2 same-root servers, a fallback attempt of 4, a - shipped command still targeting `stopped`, retirement of the healthy - root, and retirement of the colliding user server, respectively. -- **DURABLE LESSON — "the test that passes" vs "the test that - discriminates."** Green tests across six rounds repeatedly pinned only - a nearby helper or an absence, and only biting exposed it. **Carry this - to `docs/agent-handoff.md` when the lane lands.** The concrete shapes, - all from this branch: - 1. R1 acceptance 36 asserted "every server is terminal" — pinning the - ABSENCE of the fallback it claimed to test. - 2. "No live non-fallback server" misses a respawn loop: a respawning - server sits in `crashed` most of the time. `attempt` counts - respawns; liveness does not. - 3. Returning to a buffer via `find_or_open` re-fires - `buffer.after-load`, which repairs the attachment regardless of the - code under test. Use `switch_buffer`. - 4. A MISSING executable fails synchronously inside `after-load`, where - the rebuild happens inline — no async race can occur. Only the - probe path exercises asynchronous ordering. - 5. A mutation that RAISES (indexing a nil config) is swallowed by the - hook's pcall, so the bite "passes" for the wrong reason. A bite must - reproduce the original shape, not merely break the code. - 6. A fixture whose `serve` sleeps can never let the primary initialize - first, so it cannot reach the ordering where a late verdict must - retire a LIVE server. - 7. Asserting on a field that no longer exists (`_probe.reattach_from` - after a refactor) reads as nil and passes for nothing. Assert - positive facts — a count, a command string — not absences. - 8. Counting DISTINCT KEYS cannot bound REPEATED WORK: a per-tick retry - on one buffer keeps `#repaired == 1` forever. Count the attempts, - not the things attempted against (bite: 174 vs 1). - 9. A NONEXISTENT executable only exercises synchronous ENOENT. To - reach "spawned, then died", the fixture must actually spawn. - 10. Calling the two SAFE HELPERS directly does not pin a shipped - command that bypasses both. Drive the command registry entry and - require its terminal result — replacing a dead record with a - `starting` server is still not success if the request is issued - before initialize. - Rule: **a test is not evidence until the mutation it targets has been - shown to fail it.** -- **SECOND DURABLE LESSON — a scope error repeats until the scope is - named.** The "fallback silently does not happen" defect came back four - times: no re-attach; re-attach cleared by an unrelated buffer; - re-attach satisfied by the server being replaced; re-attach of one - buffer while the others stay stale. Every fix was locally correct and - none asked *what does this config swap invalidate?* — the answer being - every Lean buffer and every Lean server, because the config entry is - global and servers are per-root. **When a change edits shared state, - enumerate everything derived from it before repairing anything.** -- **SUBSTRATE BUG FOUND, not fixed here (framing §6).** - `LspManager::stop` on an ALREADY-terminal server takes its - not-initialized branch, terminates the dead process and sets - `ShuttingDown { .. None }` on the premise that "the next exit - observation cleans up" — but the exit already happened, which is what - made it `Crashed`. No further event arrives, so the client is stuck in - `ShuttingDown` **forever**: `server_is_live` reads it as LIVE, so - `attach_buffer` never rebuilds, and `forget` refuses it for not being - terminal. **Stopping a dead server is what makes it un-replaceable.** - Lean works around it by dispatching on state: `forget` when - terminal, `stop` when live. Merely SKIPPING the call is not - enough — that leaves `next_restart_at` armed. -- Round-1 review found four P1s, all real: the latch swapped the config - but never spawned or re-attached (and acc36 *asserted every server was - terminal*, pinning the absence of the fallback); a missing `lake` - bypassed probe and latch entirely because the hook keyed on an - attachment that ENOENT prevents; `waitForDiagnostics` omitted the - `version` Lean requires; and the ledger stated the dangerous stacking - order. -- The probe's non-zero exit is deliberately NOT a fallback trigger — - §2.9's elan shim makes `lake --version` fail where `lake serve` still - works. Only a parseable version below 3.1.0 triggers it; the - server-failure latch covers the rest. -- Verification on this branch: `cargo fmt --check` clean; strict - workspace Clippy clean; 1,829 default + 2,003 CRDT library tests; - lean4 server 40/40; lean4 stage 1 9/9; dispatch seams 15/15; - multi-root 13/13; M4 121; required GPU 155; **isolated-config - serial workspace sweep 3,229 across 94 suites, zero failures**; - `git diff --check` clean. (Round 1 of - this entry recorded 17/17 and 3,206 — the PRE-fix counts — after the - fixes were pushed. The ledger's protocol is that verification - describes the pushed tree; recording it late is the #161 fmt-blocker - error in a slower form.) - -### Stage 4 — framing rev 6, split into 4a/4b (branch `lean4-stage4a-typed-edit-chain`) +### Stage 4 — framing rev 7, split into 4a/4b (branch `lean4-stage4a-typed-edit-chain`) - Stages 3a and 3b **merged as #167** (`main` @ `6f348c9`) and **#170** (`main` @ `d400f30`), 2026-07-26. Both were integrated against a main that had advanced 50 commits mid-review; the only conflict either time was this ledger's own lane headings, resolved by keeping both sides. - Worktree `../pmacs-lean-stage4`, branched off `main` @ `d400f30`. - Framing-only so far: `docs/lean4-mode-framing.md` **revision 6**. No + Framing-only so far: `docs/lean4-mode-framing.md` **revision 7**. No code. Awaiting user approval before implementation, per the workflow. +- **Round 6 review found five P1s, four of them internal to rev 6** — + facts about pmacs the revision asserted without checking, while its + external (upstream) facts held. Fixed in rev 7: Stage 4a's footprint + omitted the test file its own acceptance requires; pending + abbreviation state was keyed by buffer when pmacs is **multi-frontend** + (`EditorCore.views` is per-`FrontendId`, `take_typed_edit` is already + frontend-keyed, and `buffer.after-switch` fires with NO arguments, so + a buffer-keyed clear lets any frontend discard another's pending + state); the shortest-match rule was missing its **tie-break by source + declaration order**, which 101 prefixes depend on and a `pairs`- + iterated Lua map cannot express; and the generator's "abort on keys + needing escaping" rule **rejects the real table** (`\` is a key, `"` + begins eleven). +- **A 404 on a guessed path is not evidence of absence.** Rev 6 declared + the upstream package ships no README after fetching the package root, + with the directory listing showing `src/README.md` already in hand. + The README states the tie rule in one sentence. - **Round 5 re-scout split Stage 4 into 4a (substrate) and 4b (Lean).** 4a is the typed-edit consumer chain — `builtin/runtime/typed_edit.lua` plus `pair.lua` re-expressed as one registered consumer, no behavior diff --git a/docs/agent-handoff.md b/docs/agent-handoff.md index b2d0aeb..a176505 100644 --- a/docs/agent-handoff.md +++ b/docs/agent-handoff.md @@ -1,7 +1,8 @@ # Agent handoff — cross-machine continuity -**Last updated: 2026-07-25, after the inline-math slice (#158) landed — -the first mathematical typesetting in pmacs — following find-file (#162), +**Last updated: 2026-07-26, after Lean 4 stages 3a and 3b (#167, #170) +landed — pmacs' first Lean language server — following the inline-math +slice (#158), the first mathematical typesetting in pmacs, and find-file (#162), the dired arc's Stage 0, and COHERENCE.md (#163), Lean 4 Stage 1 (#160), the minimap blank-slab fix (#159), bottom-panel Stage 1 (#155), the inline-math re-scout (#154), the vterm PTY-flake fix (#153), and the @@ -25,15 +26,15 @@ reads it the way you just did. For volatile branches, checkpoints, verification, and recovery commands, read `docs/active-work.md` immediately after this file. -## 1. Where the project stands (2026-07-25) +## 1. Where the project stands (2026-07-26) -- `main` @ `d152120` (the bottom-panel landed-doc refresh #156 atop the - inline-math slice #158, dired Stage 1 #165, the GPU terminal input fix - #166, Lean 4 Stage 2 #161, the dired framing #164, COHERENCE.md #163, - find-file #162, Lean 4 Stage 1 #160, minimap blank-slab #159, - bottom-panel Stage 1 #155). Protocol unchanged at **v20**. The bullets - below describe the arcs in their own terms; this line is the - head-of-`main` anchor. +- `main` @ `d400f30` (Lean 4 Stage 3b #170 atop Stage 3a #167, the + bottom-panel landed-doc refresh #156, the inline-math slice #158, + dired Stage 1 #165, the GPU terminal input fix #166, Lean 4 Stage 2 + #161, the dired framing #164, COHERENCE.md #163, find-file #162, Lean + 4 Stage 1 #160, minimap blank-slab #159, bottom-panel Stage 1 #155). + Protocol unchanged at **v20**. The bullets below describe the arcs in + their own terms; this line is the head-of-`main` anchor. - **`COHERENCE.md` is now required reading and a required framing input — #163.** It carries the product-coherence thesis, an audited scorecard, per-concern gaps, and §20's priority order, and it is the @@ -42,6 +43,56 @@ commands, read `docs/active-work.md` immediately after this file. interaction islands added, config-registry adoption, background-work attribution. Its §2 grades the golden journey **broken at step 3** (`pmacs .` exits 1). +- **Lean 4 arc (Arc 8) — stages 1, 2, 3a, 3b LANDED** + (`docs/lean4-mode-framing.md`; #160, #161, #167, #170; merge + `d400f30`). pmacs edits Lean 4: `arborium-lean` highlighting, a + `lean4` major mode, `⟨⟩ ⦃⦄ ⟮⟯` pairs, and a `lake serve` language + server with a Lake-aware outermost root, a lazy toolchain probe, a + one-shot `lean --server` fallback, and `waitForDiagnostics`. **No + protocol change in any stage** (still v20). + - **Two of the four stages contained no Lean at all**, and that is the + arc's organizing rule: *no PR mixes a cross-cutting substrate change + with Lean feature content.* Stage 2 made LSP server affinity + per-project-root (`ensure_server` had been reusing one server across + roots — a correctness bug for every language, not just Lean). Stage + 3a added notification/response subscription seams to + `handle_server_requests`, the single shared LSP event drain, plus + `pmacs.fs.canonicalize`. + - **Two consecutive re-scouts found that rule broken by the stage + being scouted** — Stage 3 in round 4, Stage 4 in round 5, each time + by a risk column that contradicted its own prose. The rule is not + self-enforcing. Re-check every remaining stage's risk column at + scout time. + - **A configured LSP root must be a canonical absolute path.** It + reaches `file_uri_for` verbatim and that URI is the affinity key, so + one package opened by two spellings spawns two servers. Stage 3a's + `pmacs.fs.canonicalize` is the primitive; it returns nil rather than + a lossy path for non-UTF-8 input. + - **`LspManager::stop` on an already-terminal client strands it in + `ShuttingDown` forever** — `server_is_live` then counts it live so + nothing rebuilds against it, and `forget` refuses it for not being + terminal. *Stopping a dead server is what makes it un-replaceable.* + Stage 3b works around it by dispatching on state (`forget` when + terminal, `stop` when live); merely skipping the call leaves + `next_restart_at` armed. The real fix is unframed substrate work. + - **`elan` shims lie**: `lake --version` and `lean --version` can both + fail ("no default toolchain configured") on a machine where Lean + otherwise works, so `command -v lake` is worthless as a capability + check. Lean acceptance is fake-server; live smokes must be PATH- + **and** success-gated. + - Stage 3b took six review rounds, and **the same defect appeared four + times**: "the fallback silently doesn't happen," as no re-attach, + then re-attach cleared by an unrelated buffer, then satisfied by the + very server being replaced, then repairing one buffer while the rest + stayed stale. Each fix was locally right; none asked what a *global* + config swap invalidates. The durable lesson is to heal at + **consumption** — the point where a stale record is handed out — not + at the moment of the swap. + - Remaining: Stage 4a (typed-edit consumer chain) and 4b (the Unicode + input method) are framed and awaiting approval; stages 5 (goal + panel), 6 (`#eval` output channel), and 7 (module hierarchy) are + framed but not scouted against current `main`. + - **Inline math LANDED — #158** (`docs/inline-math-slice-framing.md` rev 3; merge `5aa9044`). pmacs renders `$…$` as typeset mathematics in the GPU frontend. **No protocol change (still v20); the whole slice lives in diff --git a/docs/lean4-mode-framing.md b/docs/lean4-mode-framing.md index e39eb36..ca50696 100644 --- a/docs/lean4-mode-framing.md +++ b/docs/lean4-mode-framing.md @@ -46,7 +46,7 @@ during a rebase. ## 0.1 Revision history -Revision 1 — initial. +Revision 1 — initial. Current revision: **7**. ### Round 1 (rev 1 → rev 2) @@ -288,11 +288,15 @@ landed and its citations are historical record, not navigation. Stages 3a and 3b landed (#167, #170). Re-scouting Stage 4 against `main` @ `d400f30` produced **six findings that change the plan** and three -that confirm it. The pmacs-side facts were verified in a worktree at +that confirm it. (Round 6 found five more, four of them internal to this +revision; read that section too before trusting a rev-6 statement.) The pmacs-side facts were verified in a worktree at that commit; the upstream facts were verified by downloading and reading `leanprover/vscode-lean4` at commit `17d1d08` (2026-05-29) — the algorithm, not its documentation, since the `lean4-unicode-input` -package ships no README. +package ships no README. *(Round 6: it does, at `src/README.md` — see +that section. Corrections to round 5's own numbers are marked inline +below rather than rewritten, per the standing rule that revision +entries are record, not navigation.)* 1. **Stage 4 violated this document's own splitting rule — the same way Stage 3 did.** §4 says "no PR in this arc mixes a cross-cutting @@ -377,6 +381,14 @@ package ships no README. needs a `doNotTrackNewAbbr` guard and why §2.11 records that pmacs does not. + *Corrected in round 6.* **119** symbols are multi-codepoint, of which + 26 carry `$CURSOR`; "93" was the non-`$CURSOR` subset stated as a + total. **Three** values contain a backslash — the `\` → `\` identity + entry was missed. And this entry's biggest omission is not a number: + the shortest-key rule needs a **tie-break by source declaration + order**, which the README round 5 said did not exist states outright. + §2.11 and Q#LN11 carry the corrected facts. + Confirmations, recorded because each was load-bearing and unverified: 7. **`take_typed_edit`'s one-shot contract is unchanged** @@ -404,6 +416,70 @@ Citation drift repaired per COHERENCE §25, on the same terms as round and the revision-history entries above, which are historical record rather than navigation. +### Round 6 (rev 6 → rev 7) + +The 4a/4b split held; five P1s against the revision's own content, all +real, all reproduced. Four share a root: **rev 6 verified its external +facts and under-verified its internal ones.** + +1. **Stage 4a's declared footprint excluded the tests its acceptance + required.** Q#LN10 listed three production files while 46a–46e demand + chain-specific tests that cannot live in + `tests/auto_pair_acceptance.rs` — criterion 46 requires that file + byte-identical. Footprint now names + `tests/typed_edit_chain_acceptance.rs` and adds it to the PR's gates. +2. **Pending state had the wrong owner.** §2.11 reasoned "no + multi-cursor, therefore one point" and Q#LN22 keyed pending + abbreviations by buffer. pmacs is multi-frontend: `EditorCore.views` + is per-`FrontendId` with its own active window, `take_typed_edit` is + *already* frontend-keyed, and `pmacs.frontend.id()` exists. Two + frontends on one Lean buffer — the TUI-plus-GPU case this project + ships — would share one slot. Worse, `buffer.after-switch` takes no + arguments, so a buffer-keyed clear-on-switch lets any frontend + discard another's pending abbreviation. Now keyed + `(frontend, buffer)` with a window check, frontend-scoped + after-switch clearing, a `frontend.detached` purge, and acceptance + 45i — which the buffer-keyed design passes every other criterion + without. +3. **The shortest-match rule was missing its tie-break, and rev 6's + research method is why.** Upstream keeps declaration order among + equal-length shortest keys. The README states it in one sentence — + and rev 6 asserted "the package ships no README" after a 404 on the + package root, without checking the directory listing it had already + fetched, which shows `README.md` under `src/`. **A 404 on a guessed + path is not evidence of absence.** The rule is load-bearing: 101 + prefixes have equal-shortest candidates resolving to *different* + symbols (`f` → `f<` not `f>`; `"` picks `"A` from eleven). A `pairs`- + iterated Lua map cannot express it, so Q#LN11 now emits an ordered + sequence and Q#LN22 sorts by `(#key, source rank)`. +4. **The generator's rejection rule rejected the current table.** "Abort + on keys needing Lua escaping" would reject `\` and the eleven `"X` + keys — and acceptance 45d requires `\` to work. Replaced with + canonical lossless escaping; the generator aborts only on duplicate + keys, invalid UTF-8, and a failed self-round-trip. Relatedly, 45g + claimed the suite compares against `abbreviations.json`, which is not + shipped; it now pins self-consistency properties and leaves + source fidelity to the generator, where the source is in hand. +5. **Durable and volatile state were not reconciled** — + `docs/agent-handoff.md` still anchored `main` at `d152120` with + neither #167 nor #170, while `docs/active-work.md` kept full merged + Stage 3a/3b histories against its own instruction to remove merged + entries, under a stale July 25 snapshot date. Round 5 updated the + ledger and skipped the handoff; per CLAUDE.md both are required + reading, and the one that outranks the other was the one left wrong. + +Corrections carried in the same revision, each verified against the +data: the README exists (finding 3); there are **119** multi-codepoint +symbols, of which 26 carry `$CURSOR` — rev 6's "93" was the +non-`$CURSOR` subset reported as a total; **three** values contain a +backslash (`\`, `n`, `setminus`), not two; Q#LN22 now states the rule +acceptance 45d depended on, that an unclaimed terminating `\` is +reprocessed as a new leader; acceptance 38 now says the terminator is +retained, so undo restores `\alpha ` with its space; the coherence +section cites golden-journey **step 5** ("Edit immediately"), not step +4; and §8's config-registry prior art points at Q#LN22, where the gate +now lives. + ## 1. What ships Nine stages, after round 4 split Stage 3 and round 5 split Stage 4. The @@ -828,8 +904,15 @@ them from the store needs a Rust-side policy, not a Lua filter. Q#LN18. Scouted 2026-07-26 against `leanprover/vscode-lean4` @ `17d1d08`, package `lean4-unicode-input`, files `AbbreviationProvider.ts`, `TrackedAbbreviation.ts`, `AbbreviationRewriter.ts`, -`AbbreviationConfig.ts`, and `abbreviations.json`. The package ships no -README, so the algorithm below is read off the source. Apache-2.0. +`AbbreviationConfig.ts`, `abbreviations.json`, and — round 6 — the +package README at `lean4-unicode-input/src/README.md`. Apache-2.0. + +**Rev 6 first claimed this package ships no README. It does**, at +`src/README.md` rather than the package root, and the 404 on the root +path was taken as absence without checking the directory listing that +was already in hand. That cost the tie rule below: the README states it +in one sentence, and reading only the code left it as an inference from +`Array.prototype.sort`'s stability rather than a documented contract. **Resolution.** `findSymbolsByAbbreviationPrefix(p)` collects every key having `p` as a prefix, sorts them by **key length ascending**, and maps @@ -845,6 +928,32 @@ Verified against the table: `alpha` → `α`, `alp` → `α` (via `alpha`), surprising enough to be worth an acceptance criterion), `alp7` → `α7` via rule 2, `a` → `α` (`a` is itself a key, among 29 prefix matches). +**The tie rule, and why it is a constraint on the vendored format.** +When several shortest keys have equal length, upstream takes **the one +declared first in `abbreviations.json`**. The README says so outright; +the code achieves it because `Object.keys()` yields JSON insertion order +and `Array.prototype.sort` is stable. Ties are not rare: **101 prefixes +have equal-shortest candidates that resolve to *different* symbols**. +`f` picks `f<` → `‹` over `f>` → `›`; `"` picks `"A` → `Ä` from eleven +equal-length candidates; `(` picks `()` over `(=`, `(b`, `((`, `([`. + +A Lua table iterated with `pairs` has no order at all, so **a generated +`{ [key] = symbol }` map cannot express this contract** — it would +resolve these 101 prefixes nondeterministically, and worse, *stably +wrong* per build. Q#LN11 therefore carries source rank alongside the +symbol. + +**Two things the README explains that the code does not.** `Tab` is the +manual early-replacement trigger upstream binds, which is why +`getReplacementText`'s shortest-prefix rule is user-visible at all +rather than an internal detail. And the `[]_`/`{}_` entries in the table +are not symbols anyone types — they are **decoys**, added so that `\[` +is not uniquely-and-completely matching and therefore does not eagerly +expand before the user can type the second `[`. That is the same +collision Q#LN22 handles from the pairing side, solved upstream by +editing the data. Anyone regenerating the table must not "clean up" +those entries. + **Tracking.** The leader `\` is inserted into the buffer like any other character, and the tracked range starts after it; the replaced range spans the leader inclusive (`abbreviationRange.moveKeepEnd(-1)`). So the @@ -881,8 +990,10 @@ abbreviation the cursor has left. pmacs has no cursor-motion hook (round-5 finding 3), so this seam does not exist here and Q#LN22 makes abandonment lazy instead. -**The re-arm guard pmacs does not need.** `setminus` → `\` and `n` → -`\n`, so an expansion can insert a backslash; upstream sets +**The re-arm guard pmacs does not need.** Three values contain a +backslash — `\` → `\`, `n` → `\n`, and `setminus` → `\` (rev 6 first +said two, dropping the `\` → `\` identity entry) — so an expansion can +insert a backslash; upstream sets `doNotTrackNewAbbr` across the replace so that backslash does not open a new abbreviation. In pmacs the expansion is a programmatic `buf:replace` that arms no typed-edit record, so the chain sees nothing and cannot @@ -891,9 +1002,25 @@ contract, not by accident — and the acceptance must pin it, because a future consumer that inferred from buffer text rather than provenance would reintroduce the bug. -**What pmacs does not have to carry.** Multi-cursor. Upstream tracks a -`Set` and sorts changes bottom-up for that reason; -pmacs has one point, so one pending abbreviation per buffer. +**What pmacs does not have to carry.** Multi-cursor within a frontend. +Upstream tracks a `Set` and sorts changes bottom-up +for that reason; pmacs has one point per frontend view. + +**What pmacs has instead, and rev 6 got wrong.** Rev 6 read "no +multi-cursor" as "one point" and keyed pending state by buffer alone. +**pmacs is multi-frontend**: `EditorCore.views` is a +`HashMap`, each with its own active window and +cursor; `take_typed_edit` is already keyed by frontend +(`typed_edit_armed: Option<(FrontendId, TypedEditRecord)>`, matched +against `active_frontend`); the record carries `window` as well as +`buffer`; and `pmacs.frontend.id()` is exposed to Lua. Two frontends +editing the same Lean buffer — the ordinary TUI-plus-GPU case, not an +exotic one — would share a single buffer-keyed pending slot, so one +could extend, expand, or silently clear the other's half-typed +abbreviation. `buffer.after-switch` makes it worse: it fires with no +arguments, so a buffer-keyed clear-on-switch would let *any* frontend's +navigation discard a pending abbreviation belonging to another. Q#LN22 +keys the state accordingly. ## 3. Decisions @@ -1340,11 +1467,27 @@ claim a reader must be able to check without reconstructing `src/editor.rs`'s include list. **Stage 4a ships this and nothing else.** Its whole content is: -`typed_edit.lua`, `pair.lua` re-expressed as one registered consumer, -and the `include_str!` line. Round 5's finding 1 is why this is a PR and -not a first commit — `pair.lua` is every language's auto-pairing, and a -reviewer looking at a Lean PR should not have to also review a rewrite -of it. + +| File | Change | +|---|---| +| `builtin/runtime/typed_edit.lua` | new — the chain owner | +| `builtin/runtime/pair.lua` | re-expressed as one registered consumer | +| `src/editor.rs` | one `include_str!` line, before `pair.lua`'s | +| `tests/typed_edit_chain_acceptance.rs` | new — criteria 46a–46e | +| `tests/auto_pair_acceptance.rs` | **unchanged, zero lines** | + +Rev 6 listed only the first three and then required criteria 46a–46e, +which no existing suite can host: the auto-pairing suite must stay +untouched (that is the whole point of criterion 46), so the chain's own +behavior — take-once, priority order, claim-stops-chain, throw +containment — has nowhere to live. A declared footprint that excludes +the tests its own acceptance demands is not a footprint. The new suite +joins the required gate list for this PR alongside +`tests/auto_pair_acceptance.rs`. + +Round 5's finding 1 is why this is a PR and not a first commit — +`pair.lua` is every language's auto-pairing, and a reviewer looking at a +Lean PR should not have to also review a rewrite of it. **The no-behavior-change claim must be pinned, not asserted.** The full `tests/auto_pair_acceptance.rs` suite is a required gate for 4a and must @@ -1377,10 +1520,24 @@ file rather than estimated: | 305 keys that are proper prefixes of another | which keys can expand eagerly | | 1,550 keys uniquely-and-completely matching | the eager-expansion set | | 26 values containing `$CURSOR` | point placement | -| 93 multi-codepoint values | the replace is not one-char-for-many | +| 119 multi-codepoint symbols (26 of them `$CURSOR`-bearing) | the replace is not one-char-for-many | +| 101 prefixes with disagreeing equal-shortest ties | why the format carries source rank | + +(Rev 6 gave the multi-codepoint figure as 93, which was the count +*excluding* the `$CURSOR` entries — a subset reported as a total.) vscode-lean4 is Apache-2.0. +**Format: an ordered array, not a map.** §2.11's tie rule makes source +order semantic, and a Lua `{ [key] = symbol }` table iterated with +`pairs` cannot carry it. The generated file emits a **sequence** — +`{ {key, symbol}, ... }` in `abbreviations.json` order — plus a derived +`key → index` lookup built at load time for the exact-match case. +Resolution sorts candidates by `(#key, index)`, so the 101 ties resolve +the way upstream resolves them and the file's own line order is the +audit trail. A map-shaped emit would be nondeterministic across builds +and, once a hash order happened to be stable, *stably wrong*. + Vendor it as a generated `builtin/runtime/lean_abbrev.lua` with a header recording source repo, commit, license, entry count, and the regeneration command — the `builtin/queries/latex/highlights.scm` @@ -1405,12 +1562,29 @@ so the file is self-describing to whoever next touches it. A refresh is an ordinary PR with a visible diff — which is the point: the diff is the review. -**The generator must reject a table it cannot faithfully encode.** Keys -are ASCII today but nothing upstream promises that; a key containing a -character the emitted Lua would have to escape, or a duplicate after -normalization, aborts the regeneration rather than silently emitting a -table that disagrees with its source. Same discipline as Q#LN20's -refusal to hand back a lossy path. +**Escaping is canonical and lossless, not a rejection trigger.** Rev 6 +said the generator aborts on "a key containing a character the emitted +Lua would have to escape." **That rule rejects the current table**: `\` +is a key, `"` begins eleven keys (`"A` → `Ä` …), and acceptance 45d +requires `\` to work. The generator instead emits every key and symbol +through one canonical Lua string escaper — `\\`, `\"`, `\n`, `\r`, +`\t`, and `\ddd` for any other control byte, everything else literal +UTF-8 — chosen so the emit is byte-deterministic across runs. + +What the generator *does* abort on, because these are real corruption +rather than syntax: + +- a duplicate key after decoding (JSON permits it; the table must not), +- a key or symbol that is not well-formed UTF-8, +- a round-trip mismatch: the generator re-parses its own output and + compares the full ordered sequence against the source, entry for + entry, and fails if they differ anywhere. + +That last check is what makes the artifact trustworthy, and it belongs +in the generator rather than in the acceptance suite — the suite cannot +see `abbreviations.json`, which is not shipped. Same discipline as +Q#LN20's refusal to hand back a lossy path: refuse rather than emit +something plausible. ### Q#LN21 — Stage 4b: the expansion's undo is cross-peer-degraded; ship it, name it @@ -1466,23 +1640,54 @@ that an edit was made. reconstruction of it: - `\` typed in a `lean4` buffer opens a pending abbreviation: `{ buffer, - start_offset, text = "" }`, one per buffer, keyed on `rec.buffer`. + window, start_offset, text = "" }`, keyed on **`(frontend, buffer)`** — + see below. - A subsequent self-insert `c` is claimed iff at least one key has `text .. c` as a prefix; then `text = text .. c`. If it is also uniquely-and-completely matching (one of the 1,550), expand now. - If no key extends `text .. c`, expand `text` **first**, then let `c` land normally — the chain does *not* claim `c`. -- Expansion resolves through §2.11's three-rule `getReplacementText`, - including the suffix rule (`\alp7` → `α7`). +- **A terminating `c` that is itself `\` is then reprocessed as a new + leader**, opening a fresh pending abbreviation at its position. This + is the rule acceptance 45d depends on (`\alpha\to` → `α→`) and rev 6 + specified the acceptance without specifying the rule; upstream gets it + from `processChange`, where a `finished` abbreviation reports + `isAffected = false` and so does not suppress the new-leader branch. + Note this is *not* the `\\` case: there the pending text is empty, `\` + extends rather than terminates, and the result is one literal + backslash with no pending state left open. +- Expansion resolves through §2.11's rules — shortest key wins, ties + broken by source rank, unmatchable tail appended (`\alp7` → `α7`). - `$CURSOR` is stripped from the symbol and its index becomes the point. +**Ownership is per frontend, not per buffer** (§2.11). The key is +`(pmacs.frontend.id(), rec.buffer)`, and the stored `window` must still +match `rec.window` for the state to be usable — a frontend that moved +the same buffer into a different window is no longer typing where the +pending span is. Two consequences the buffer-only design got wrong: + +- `buffer.after-switch` fires with **no arguments**, so it cannot say + whose switch it was. The subscriber reads `pmacs.frontend.id()` at + callback time — documented as "the frontend that produced the most + recent dispatched input event" — and clears **only that frontend's** + entries. A blanket clear would let one frontend's navigation discard + another's half-typed abbreviation. +- `frontend.detached` fires with the raw frontend id and is the purge + seam, exactly as `killring.lua` uses it (Q#KR11). Without it a + detached frontend's pending state leaks for the life of the session. + +This costs one table level and buys correctness in the ordinary +TUI-plus-GPU configuration, which is not an exotic setup — it is the +one this project ships two frontends for. + **Abandonment is lazy, because there is no cursor-motion hook** (round-5 finding 3). Pending state is validated at the next typed edit and -discarded when any of these no longer holds: the record's buffer is the -pending buffer; `rec.effective_start` equals `start_offset + 1 + -#text` (the point is still at the end of the pending span); and the -buffer's `revision()` advanced by exactly the pending edit. `buffer. -after-switch` clears it eagerly since that hook *does* exist. The +discarded when any of these no longer holds: the record's buffer and +window are the pending ones; `rec.effective_start` equals `start_offset ++ 1 + #text` (the point is still at the end of the pending span); and +the buffer's `revision()` advanced by exactly the pending edit. +`buffer.after-switch` clears the acting frontend's entries eagerly, +since that hook *does* exist. The practical difference from upstream: a user who clicks away mid-`\alp` and types elsewhere gets the pending state dropped rather than expanded. Upstream expands it. **This is a deliberate divergence** — expanding @@ -2138,6 +2343,11 @@ substrate pin, filed under Stage 4 only because Stage 4 was one stage. Per the no-renumbering rule above, round 5's additions take letter suffixes on both sides of the split. +46a–46e live in a **new `tests/typed_edit_chain_acceptance.rs`**, which +is part of Stage 4a's declared footprint (Q#LN10) and a required gate +for its PR. They cannot live in `tests/auto_pair_acceptance.rs`, which +criterion 46 requires to stay byte-identical. + 46. **Provenance-refactor pin:** the full `tests/auto_pair_acceptance.rs` suite passes **unmodified**. A suite edited to accommodate the refactor proves nothing; the diff for 4a must show zero lines @@ -2164,8 +2374,12 @@ suffixes on both sides of the split. **Stage 4b — the Unicode input method** -38. `\alpha` + space yields `α`; the whole expansion is a single undo - step, and one undo restores `\alpha` rather than `\alph`. +38. `\alpha` + space yields `α ` — the space lands first and the + expansion runs in the following `buffer.after-edit`, so the + terminator is **retained**, not consumed. The expansion is a single + undo step: one undo restores `\alpha ` (with its space), not + `\alph`. Rev 6 wrote the post-undo text as `\alpha`, which would be + true only if the terminator were swallowed. 39. `\<>` yields `⟨⟩` with the point between them, from the `$CURSOR` placeholder. 40. **Pair-collision pin (Q#LN22).** `\[[]]` yields `⟦⟧`: each `[` is @@ -2211,6 +2425,14 @@ suffixes on both sides of the split. because the expansion is a programmatic replace that arms no record. Bites against a future consumer that infers pending state from buffer text instead of provenance. +45i. **Pending state is per frontend (Q#LN22).** Two frontends attached + to the same `lean4` buffer: A types `\al`, B types `\to` + space in + the same buffer. B's expansion yields `→` and leaves A's `\al` + pending and intact; A then typing `l` + space still yields `∀`. + Plus: B switching buffers does not clear A's pending state, and a + `frontend.detached` for B purges B's entries only. Bites against the + buffer-keyed design rev 6 specified — which passes every + single-frontend criterion above. 45f. **Both producers, and the CI-darkness stated.** The dispatch path is pinned by the criteria above. The optimistic CRDT producer (round-5 finding 4) is pinned by a separate criterion driving @@ -2222,11 +2444,30 @@ suffixes on both sides of the split. verified only locally and name the command. Silence here is the failure mode — a green CI would otherwise read as covering the path most users take. -45g. **Table integrity.** The generated `lean_abbrev.lua` round-trips: - its entry count matches the header's declared count, and a spot set - of entries (`alpha`, `to`, `<>`, `+ `, `\`, `n`, `setminus`) matches - `abbreviations.json` byte-for-byte. Bites against a generator that - silently drops or mangles keys (Q#LN11). +45g. **Table integrity — what the suite can actually check.** + `abbreviations.json` is not shipped, so the suite cannot diff + against it and rev 6's "matches byte-for-byte" was unbuildable; a + count plus seven spot entries could not prove 1,855 round-trip + anyway. The full source-fidelity check belongs to the generator + (Q#LN11: re-parse own output, compare the ordered sequence entry for + entry, fail on any difference). What the suite pins instead are + self-consistency properties that a corrupt emit breaks: + - the loaded sequence's length equals the header's declared count, + and equals the declared count for the recorded upstream commit; + - every key is unique, and the derived `key → index` lookup has the + same cardinality as the sequence (a collision would silently drop + entries); + - every key and symbol is well-formed UTF-8, and no symbol contains + `$CURSOR` more than once; + - the resolution spot-set behaves: `alpha`, `to`, `<>`, `+ `, `\`, + `n`, `setminus`, and the tie cases from 45h. +45h. **Tie-break by source order (§2.11).** `\f` + space yields `‹` — + `f<` and `f>` are both length 2, and `f<` is declared first. Same + for `\"` + space → `Ä`, first of eleven equal-length candidates. + **This is the criterion that bites a map-shaped vendored table**: + with `pairs` iteration it passes or fails by hash order, so it must + also be run against a deliberately reversed sequence and shown to + fail. 101 prefixes are exposed to this rule. **Stage 5 — the goal view** @@ -2294,7 +2535,7 @@ suffixes on both sides of the split. discipline on transformed source edits, and Q#AP1's optimistic-classifier limitation. Stage 4a generalizes the first; 4b is built on all three. - **#127 (config registry)** — `pmacs.config.define` and the - source-buffer-resolution correction. Q#LN10's gate follows + source-buffer-resolution correction. Q#LN22's gate follows `editing.auto-pair` exactly. - **#129 (mode system)** — mode-scoped keymaps for Stage 5. - **#155 (bottom panel)** — `pmacs.window.display` and the panel adopter @@ -2381,9 +2622,10 @@ PR. (config registry) secondarily, by adding one option in the established shape rather than a new switch mechanism. -**Golden journey (§2).** No step is touched by 4a. 4b improves step 4 -(editing) for Lean specifically and changes nothing for any other -language: the pending-abbreviation state exists only in `lean4` buffers. +**Golden journey (§2).** No step is touched by 4a. 4b improves **step 5 +("Edit immediately")** for Lean specifically and changes nothing for any +other language — rev 6 cited step 4, which is "Understand the visible +interface" and is untouched by both stages: the pending-abbreviation state exists only in `lean4` buffers. Neither stage changes launch, open, or attach. **Interaction islands (§6).** **None added, and this is the load-bearing From 174e36fce384375b94c4c84af5a7a27f41e318da Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sun, 26 Jul 2026 10:49:43 -0400 Subject: [PATCH 27/32] docs(lean4): reconcile peer edits with pending ownership Advance the Stage 4 framing to revision 8. Keep pending abbreviation state frontend-owned while conservatively invalidating it after any intervening shared-buffer edit, make the revision token explicit, and rewrite acceptance 45i around that contract. Correct the active-work multi-codepoint count and the stale coherence revision label. --- docs/active-work.md | 18 ++++++++--- docs/lean4-mode-framing.md | 62 ++++++++++++++++++++++++++++++-------- 2 files changed, 63 insertions(+), 17 deletions(-) diff --git a/docs/active-work.md b/docs/active-work.md index 47c3677..cf4ea6e 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -64,17 +64,17 @@ If it does not, stop and repair the remote/fetch configuration. histories were pruned from this ledger in round 6, per this file's own instruction to remove entries when their PR merges; the durable facts now live in `docs/agent-handoff.md` §1's Lean 4 bullet, which is where - a fresh machine should read them. `docs/lean4-mode-framing.md` rev 7 + a fresh machine should read them. `docs/lean4-mode-framing.md` rev 8 carries the decisions. -### Stage 4 — framing rev 7, split into 4a/4b (branch `lean4-stage4a-typed-edit-chain`) +### Stage 4 — framing rev 8, split into 4a/4b (branch `lean4-stage4a-typed-edit-chain`) - Stages 3a and 3b **merged as #167** (`main` @ `6f348c9`) and **#170** (`main` @ `d400f30`), 2026-07-26. Both were integrated against a main that had advanced 50 commits mid-review; the only conflict either time was this ledger's own lane headings, resolved by keeping both sides. - Worktree `../pmacs-lean-stage4`, branched off `main` @ `d400f30`. - Framing-only so far: `docs/lean4-mode-framing.md` **revision 7**. No + Framing-only so far: `docs/lean4-mode-framing.md` **revision 8**. No code. Awaiting user approval before implementation, per the workflow. - **Round 6 review found five P1s, four of them internal to rev 6** — facts about pmacs the revision asserted without checking, while its @@ -93,6 +93,15 @@ If it does not, stop and repair the remote/fetch configuration. the upstream package ships no README after fetching the package root, with the directory listing showing `src/README.md` already in hand. The README states the tie rule in one sentence. +- **Round 7 review found one remaining P1 in acceptance 45i.** Rev 7 + required A's pending abbreviation to survive B editing the same + buffer, while Q#LN22 also required an exact buffer-revision advance. + Those cannot both hold: revisions are buffer-global and every edit + bumps them. Rev 8 keeps the conservative guard and separates + ownership from survival — B cannot consume A's record, but B editing + the shared buffer invalidates A lazily; B switching buffers or + detaching remains frontend-scoped when no shared-buffer edit + intervenes. - **Round 5 re-scout split Stage 4 into 4a (substrate) and 4b (Lean).** 4a is the typed-edit consumer chain — `builtin/runtime/typed_edit.lua` plus `pair.lua` re-expressed as one registered consumer, no behavior @@ -130,7 +139,8 @@ If it does not, stop and repair the remote/fetch configuration. - Table facts re-derived at `17d1d08`: 1,855 entries, 36,861 bytes, all keys ASCII, **64** keys carry a `lean4` pair-set char, **305** keys are proper prefixes of another (so 1,550 expand eagerly), **26** values - carry `$CURSOR`, **93** are multi-codepoint. + carry `$CURSOR`, and **119** are multi-codepoint — the 26 + `$CURSOR`-bearing values plus 93 others. - Citation sweep per COHERENCE §25: five live citations moved in the 50 commits since rev 5 — `take_typed_edit` 12827→12990, `handle_server_requests` 1549→1815, `fs.stat` 93→133, diff --git a/docs/lean4-mode-framing.md b/docs/lean4-mode-framing.md index ca50696..650b2f3 100644 --- a/docs/lean4-mode-framing.md +++ b/docs/lean4-mode-framing.md @@ -46,7 +46,7 @@ during a rebase. ## 0.1 Revision history -Revision 1 — initial. Current revision: **7**. +Revision 1 — initial. Current revision: **8**. ### Round 1 (rev 1 → rev 2) @@ -480,6 +480,29 @@ section cites golden-journey **step 5** ("Edit immediately"), not step 4; and §8's config-registry prior art points at Q#LN22, where the gate now lives. +### Round 7 (rev 7 → rev 8) + +One P1 remained in the new multi-frontend acceptance, plus two +documentation cleanups. + +1. **Acceptance 45i contradicted Q#LN22's conservative abandonment + rule.** It required frontend A's pending abbreviation to survive + frontend B editing the same buffer, but `buffer:revision()` is + buffer-global and advances on every edit. B's first edit therefore + invalidates A's record under the exact-revision guard. The criterion + now separates the two contracts: another frontend cannot consume A's + record, but any intervening edit to their shared buffer invalidates + it lazily; navigation and detachment remain frontend-scoped when no + shared-buffer edit intervenes. The pending record now names its + `expected_revision` explicitly so the validation rule is buildable. + Preserving A's record through peer edits would require translating + and validating its span across arbitrary edits, a substantially + larger substrate change that Stage 4b does not take on. +2. **The volatile ledger retained rev 6's undercount.** Its table facts + now say 119 multi-codepoint symbols — 26 `$CURSOR`-bearing and 93 + others — matching §2.11 and Q#LN11. +3. **§9.1's revision label was stale.** It now names rev 8. + ## 1. What ships Nine stages, after round 4 split Stage 3 and round 5 split Stage 4. The @@ -1640,8 +1663,9 @@ that an edit was made. reconstruction of it: - `\` typed in a `lean4` buffer opens a pending abbreviation: `{ buffer, - window, start_offset, text = "" }`, keyed on **`(frontend, buffer)`** — - see below. + window, start_offset, text = "", expected_revision }`, keyed on + **`(frontend, buffer)`** — see below. `expected_revision` is the + buffer's revision after that leader edit. - A subsequent self-insert `c` is claimed iff at least one key has `text .. c` as a prefix; then `text = text .. c`. If it is also uniquely-and-completely matching (one of the 1,550), expand now. @@ -1685,7 +1709,14 @@ finding 3). Pending state is validated at the next typed edit and discarded when any of these no longer holds: the record's buffer and window are the pending ones; `rec.effective_start` equals `start_offset + 1 + #text` (the point is still at the end of the pending span); and -the buffer's `revision()` advanced by exactly the pending edit. +the buffer's `revision()` equals `expected_revision + 1`, meaning the +current typed edit is the only edit since this frontend last extended +the pending abbreviation. A claimed extension stores the current +revision as the new `expected_revision`. This is deliberately +conservative across frontends: any intervening edit to the shared +buffer invalidates the pending record even if it occurred elsewhere. +Keeping the record alive would require translating and validating its +span through arbitrary peer edits, substrate Stage 4b does not add. `buffer.after-switch` clears the acting frontend's entries eagerly, since that hook *does* exist. The practical difference from upstream: a user who clicks away mid-`\alp` @@ -2425,14 +2456,19 @@ criterion 46 requires to stay byte-identical. because the expansion is a programmatic replace that arms no record. Bites against a future consumer that infers pending state from buffer text instead of provenance. -45i. **Pending state is per frontend (Q#LN22).** Two frontends attached - to the same `lean4` buffer: A types `\al`, B types `\to` + space in - the same buffer. B's expansion yields `→` and leaves A's `\al` - pending and intact; A then typing `l` + space still yields `∀`. - Plus: B switching buffers does not clear A's pending state, and a - `frontend.detached` for B purges B's entries only. Bites against the - buffer-keyed design rev 6 specified — which passes every - single-frontend criterion above. +45i. **Pending state is per frontend, with conservative shared-buffer + invalidation (Q#LN22).** Two frontends share a `lean4` buffer at + distinct points. A types `\al`; B types `p`. B's `p` lands normally + at B's point rather than extending A's record. Because that edit + advances the shared buffer's revision, A then typing `l` + space + leaves literal `\all ` rather than expanding: A's stale record is + abandoned lazily. In a fresh setup, A types `\al`, B switches + buffers **without editing the shared buffer**, and A typing `l` + + space still yields `∀`; B's switch clears only B's entries. Finally, + `frontend.detached` for B purges B's entries only and does not clear + a still-valid A record. Bites both against the buffer-keyed design + rev 6 specified and against the impossible rev-7 promise that + pending state survives arbitrary peer edits. 45f. **Both producers, and the CI-darkness stated.** The dispatch path is pinned by the criteria above. The optimistic CRDT producer (round-5 finding 4) is pinned by a separate criterion driving @@ -2615,7 +2651,7 @@ uncapped event queue, the dropped `cfg.restart`, and — unchanged from languages other than Lean, and §4's rule is what keeps them out of a Lean PR. -### 9.1 Coherence impact — stages 4a and 4b (rev 6) +### 9.1 Coherence impact — stages 4a and 4b (rev 8) **Sections served.** §6 (interaction islands) primarily, and in the *preventing* direction rather than the fixing one — see below. §11 From 24ca9062944d639e4aa90c3567a9e67c4497e49b Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sun, 26 Jul 2026 13:00:03 -0400 Subject: [PATCH 28/32] feat(typed-edit): the typed-edit consumer chain (Arc 8 Stage 4a) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit `pmacs.editor.take_typed_edit()` is one-shot and per-frontend (Q#AP9): the first `buffer.after-edit` callback to call it clears the slot, and every later callback in the same fan-out sees nil. That was survivable only because auto-pairing was the sole consumer — never a property anyone chose. A second independent caller would get nil or steal the record from pairing depending on hook registration order, and registration order is not a contract. This makes it one. `builtin/runtime/typed_edit.lua` owns the single after-edit subscriber that reads the record, and offers that one read to consumers registered through `pmacs.typed_edit.add_consumer{ name, priority, fn }`: lowest priority first, ties by registration order, and the first consumer to return truthy claims the edit and stops the chain. `pair.lua` becomes that chain's only consumer, at priority 100. No Lean content. Stage 4b's abbreviation expander is what needs the ordering guarantee (64 of its 1,855 keys contain a `lean4` pair-set character, so pairing running first corrupts them), but the chain is substrate every language runs through, which is why it ships alone — framing Q#LN10, and §4's rule that no PR in this arc mixes a cross-cutting substrate change with Lean feature content. Three design points worth review attention: - Consumers are called even when the record is nil. "This fan-out carried no typed edit" is information a consumer acts on: it is how pairing's test seam observes a non-event, and how Stage 4b will abandon a pending abbreviation an unrelated edit invalidated. Three existing auto-pairing tests fail if the chain skips consumers on nil. - The chain pcalls each consumer. `buffer.after-edit` is all-must-succeed, so a throwing consumer would otherwise fail the fan-out for every other subscriber, including lsp.lua's didChange flush. Behavior-preserving for pairing, which already never throws. - Ordered insertion, not `table.sort`, which is not stable in Lua — "ties by registration order" is a stated contract, not a coincidence. `tests/auto_pair_acceptance.rs` is UNCHANGED — zero lines — and its 45 tests pass. That is criterion 46 and the whole no-behavior-change claim; a suite edited to accommodate the refactor would prove nothing. `tests/typed_edit_chain_acceptance.rs` adds 9 tests for criteria 46a-46e. Every one is bite-verified by mutation: appending instead of ordered insert (5 fail), `>=` for the tiebreak (1), re-taking per consumer (4), ignoring the claim (1), dropping the pcall (1), skipping nil fan-outs (1 here plus 3 in the untouched auto-pair suite), and loading the chain after lsp.lua (the Q#AP7 flush test fails, alongside the two existing pairing ones). Co-Authored-By: Claude Opus 5 (1M context) Claude-Session: https://claude.ai/code/session_011LFvC4FQtux4y32KuevZ7B --- builtin/runtime/pair.lua | 85 +++-- builtin/runtime/typed_edit.lua | 112 ++++++ src/editor.rs | 15 + tests/typed_edit_chain_acceptance.rs | 517 +++++++++++++++++++++++++++ 4 files changed, 699 insertions(+), 30 deletions(-) create mode 100644 builtin/runtime/typed_edit.lua create mode 100644 tests/typed_edit_chain_acceptance.rs diff --git a/builtin/runtime/pair.lua b/builtin/runtime/pair.lua index 9ed9d1f..5869dce 100644 --- a/builtin/runtime/pair.lua +++ b/builtin/runtime/pair.lua @@ -4,20 +4,30 @@ -- next char is already `)` steps over it instead of doubling it. The -- carrier is a `buffer.after-edit` reaction (Q#AP1): the opener stays -- a genuine single-codepoint self-insert — the classification --- signature help depends on — and this hook inserts (or swallows) the --- closer as a second edit. Provenance is the exact one-shot typed-edit --- record (`pmacs.editor.take_typed_edit()`, Q#AP9), not buffer-text --- inference: pastes, programmatic edits, manual hook runs, and a stale --- `this_command` have no record and never pair, and a transformed, --- relocated, or context-switching source self-insert fails closed. +-- signature help depends on — and this reaction inserts (or swallows) +-- the closer as a second edit. Provenance is the exact one-shot +-- typed-edit record (`pmacs.editor.take_typed_edit()`, Q#AP9), not +-- buffer-text inference: pastes, programmatic edits, manual hook runs, +-- and a stale `this_command` have no record and never pair, and a +-- transformed, relocated, or context-switching source self-insert fails +-- closed. -- --- This chunk loads BEFORE lsp.lua (Q#AP7): registration order is hook --- execution order, and lsp.lua's after-edit callback synchronously --- flushes didChange on the signature-trigger path — the closer must --- already be in the buffer when that callback runs. Everything under --- `pmacs.lsp` is therefore looked up lazily at callback time. +-- Since Arc 8 Stage 4a (Q#LN10) pairing no longer subscribes to +-- `buffer.after-edit` itself. It registers on the typed-edit chain +-- (`builtin/runtime/typed_edit.lua`), which owns the single subscriber +-- and the single one-shot read. Everything above still holds — the +-- record is the same record — but the chain, not this file, decides +-- who sees it and in what order. -- --- Framing: docs/auto-pairing-framing.md. +-- This chunk loads AFTER typed_edit.lua (it registers into it) and +-- BEFORE lsp.lua (Q#AP7): registration order is hook execution order, +-- and lsp.lua's after-edit callback synchronously flushes didChange on +-- the signature-trigger path — the closer must already be in the +-- buffer when that callback runs. Everything under `pmacs.lsp` is +-- therefore looked up lazily at callback time. +-- +-- Framing: docs/auto-pairing-framing.md; Stage 4a in +-- docs/lean4-mode-framing.md Q#LN10. pmacs.pair = pmacs.pair or {} @@ -40,7 +50,7 @@ local ed = pmacs.editor -- Per-buffer on/off switch (Q#CR8's flagship adopter). Read against the -- SOURCE buffer of the typed edit, never the currently active one — see --- the hook body below, which resolves it the same way `set_for` resolves +-- the consumer body below, which resolves it the same way `set_for` resolves -- the buffer's pair set (round 2, finding 2): `rec.buffer`, not -- `pmacs.window.buffer()`. pmacs.config.define { @@ -214,28 +224,36 @@ end -- Acceptance tests flip `_capture_records` on; each fan-out then -- publishes the record it observed (or nil) to `_last_record`, which -- is how tests read the exact codepoint / effective triple and prove --- one-shot-ness (this callback registers first and consumes it). +-- one-shot-ness (the chain takes the record before any other +-- `buffer.after-edit` subscriber can, and hands it here). pmacs.pair._capture_records = false -pmacs.hook.add("buffer.after-edit", function() +-- The typed-edit consumer (Arc 8 Stage 4a, Q#LN10). `rec` is the one +-- record `typed_edit.lua` read for this fan-out — possibly nil, which +-- is why the capture seam below is updated before the nil guard. +-- Returns whether pairing CLAIMED the keystroke: true once it has +-- committed to reacting (a skip-over or a closer insert, landed or +-- intercept-rejected), false on every decline. Pairing is last of the +-- builtin consumers, so nothing currently observes that value; it is +-- stated correctly so it stays correct when something does. +local function on_typed_edit(rec) -- One-shot provenance (Q#AP9). Absence — paste, programmatic edit, -- manual hook run, rejected insert, a post-insert mutation by the -- command, stale `this_command` — is a silent non-event; only a -- live record for a pair-set character that then fails a gate -- reports. - local rec = ed.take_typed_edit and ed.take_typed_edit() if pmacs.pair._capture_records then pmacs.pair._last_record = rec end - if not rec then return end - if not (ed.this_command and ed.this_command() == "buffer.self-insert") then return end + if not rec then return false end + if not (ed.this_command and ed.this_command() == "buffer.self-insert") then return false end -- The master switch, per-buffer (Q#CR4): the SOURCE buffer of the -- typed edit, resolved buffer-local -> global -> default(true). A -- second buffer of the same language is untouched by a buffer-local -- override here (acceptance 29). - if not pmacs.config.get("editing.auto-pair", rec.buffer) then return end + if not pmacs.config.get("editing.auto-pair", rec.buffer) then return false end local buf = pmacs.window.buffer() - if not buf then return end + if not buf then return false end -- Relevance first (PR #110 round 1, finding 2): pairing has no -- interest in characters outside the set, so a transformed or @@ -247,14 +265,14 @@ pmacs.hook.add("buffer.after-edit", function() -- Rust. local ch = rec.char local openers, closers = maps_for(set_for(rec.buffer)) - if not (openers[ch] or closers[ch]) then return end + if not (openers[ch] or closers[ch]) then return false end -- Fail closed on a transformed source self-insert (Q#AP3): the -- intercept's positional result stands as produced; pairing on top -- of a relocated or expanded opener would compound it. if not rec.clean then ed.set_status("auto-pair skipped: source self-insert transformed") - return + return false end -- Fail closed when the source edit's context is no longer current: -- an intercept switched window/buffer, or something moved the @@ -268,14 +286,14 @@ pmacs.hook.add("buffer.after-edit", function() or pmacs.window.current() ~= rec.window or ed.cursor() ~= rec.post_cursor then ed.set_status("auto-pair skipped: source context changed") - return + return false end -- Region guard (Q#AP3/Q#AP6): on the dispatch route type-over has -- already consumed and cleared the region. A region surviving the -- edit means the TUI's selection-blind optimistic gate let a custom -- pair char through (named deferral) — reacting would pile a closer -- onto an unconsumed region. - if ed.region() ~= nil then return end + if ed.region() ~= nil then return false end local cursor = rec.post_cursor @@ -294,19 +312,19 @@ pmacs.hook.add("buffer.after-edit", function() if not ok then -- The duplicate stays (e.g. `())`); report, no retry. ed.set_status("auto-pair skip rejected by buffer intercept") - return + return true end if estart ~= cursor or estop ~= cursor + #ch or einserted ~= 0 then ed.set_status("auto-pair skip altered by buffer intercept") repair_cursor(win0, buf, cursor, estart, estop, einserted) end - return + return true end end local closer = openers[ch] - if not closer then return end - if not should_pair(buf, cursor, closers) then return end + if not closer then return false end + if not should_pair(buf, cursor, closers) then return false end local win0 = pmacs.window.current() local ok, estart, estop, einserted = pcall(function() @@ -315,7 +333,7 @@ pmacs.hook.add("buffer.after-edit", function() if not ok then -- Nothing landed; the opener stands alone. ed.set_status("auto-pair closer rejected by buffer intercept") - return + return true end if estart ~= cursor or estop ~= cursor or einserted ~= #closer then ed.set_status("auto-pair closer altered by buffer intercept") @@ -324,4 +342,11 @@ pmacs.hook.add("buffer.after-edit", function() -- Clean path: no cursor motion — the insert landed at the cursor -- and Lua mutators move no cursors, so it already sits between the -- pair; the daemon's per-tick CursorByte re-grounds both frontends. -end) + return true +end + +pmacs.typed_edit.add_consumer { + name = "auto-pair", + priority = 100, + fn = on_typed_edit, +} diff --git a/builtin/runtime/typed_edit.lua b/builtin/runtime/typed_edit.lua new file mode 100644 index 0000000..59f7366 --- /dev/null +++ b/builtin/runtime/typed_edit.lua @@ -0,0 +1,112 @@ +-- typed_edit.lua --- the typed-character consumer chain (Arc 8 Stage 4a). +-- +-- `pmacs.editor.take_typed_edit()` is ONE-SHOT and per-frontend (Q#AP9): +-- the first `buffer.after-edit` callback to call it clears the slot, and +-- every later callback in the same fan-out --- including a nested manual +-- `pmacs.hook.run` --- sees nil. That was survivable only because +-- auto-pairing was the sole consumer, which was never a property anyone +-- chose. A second independent caller gets nil or steals the record from +-- pairing depending on hook registration order, and registration order +-- is not a contract. +-- +-- This module makes it one. It owns the single `buffer.after-edit` +-- subscriber that reads the record, and offers that one read to +-- consumers registered through `pmacs.typed_edit.add_consumer`: +-- +-- pmacs.typed_edit.add_consumer { +-- name = "auto-pair", -- for error reporting; must be unique-ish +-- priority = 100, -- LOWEST runs FIRST +-- fn = function(rec) ... return claimed end, +-- } +-- +-- A consumer returns whether it CLAIMED the edit; the first that claims +-- stops the chain. "Claimed" means the chain stops, not that an edit was +-- made --- Stage 4b's abbreviation expander claims every keystroke that +-- extends a pending abbreviation precisely so that auto-pairing does not +-- also react to it (Q#LN22). +-- +-- Priority is an explicit number rather than load-order-implied, because +-- the ordering is load-bearing (Q#LN22: 64 Lean abbreviation keys +-- contain a character in the `lean4` pair set, and pairing running first +-- corrupts them) and a reader must be able to check it without +-- reconstructing `src/editor.rs`'s include list. +-- +-- ORDERING CONTRACT: this chunk loads BEFORE pair.lua, which registers +-- into it, and therefore before lsp.lua. That preserves Q#AP7 --- see +-- pair.lua's header and the load site in `src/editor.rs`. +-- +-- Framing: docs/lean4-mode-framing.md Q#LN10. + +pmacs.typed_edit = pmacs.typed_edit or {} + +-- Consumers in run order: lowest `priority` first, registration order +-- breaking ties. Maintained by ordered INSERTION rather than +-- `table.sort`, which is not stable in Lua --- equal priorities would +-- otherwise resolve arbitrarily, and "ties broken by registration +-- order" is part of the stated contract, not an incidental property. +local consumers = {} + +-- Register a typed-edit consumer. Argument errors throw: registration +-- happens at chunk-load or config-load time, where a throw is a visible +-- startup failure rather than a silently missing feature. Nothing in +-- the after-edit path throws --- see the fan-out below. +function pmacs.typed_edit.add_consumer(spec) + if type(spec) ~= "table" then + error("pmacs.typed_edit.add_consumer: spec must be a table", 2) + end + local name, priority, fn = spec.name, spec.priority, spec.fn + if type(name) ~= "string" or name == "" then + error("pmacs.typed_edit.add_consumer: name must be a non-empty string", 2) + end + if type(priority) ~= "number" then + error("pmacs.typed_edit.add_consumer: " .. name .. + ": priority must be a number", 2) + end + if type(fn) ~= "function" then + error("pmacs.typed_edit.add_consumer: " .. name .. + ": fn must be a function", 2) + end + + -- STRICTLY-greater comparison, so a new consumer lands AFTER every + -- already-registered consumer of equal priority. That is exactly the + -- registration-order tiebreak; `>=` here would silently reverse it. + local at = #consumers + 1 + for i, c in ipairs(consumers) do + if c.priority > priority then + at = i + break + end + end + table.insert(consumers, at, { name = name, priority = priority, fn = fn }) +end + +pmacs.hook.add("buffer.after-edit", function() + local ed = pmacs.editor + -- ONE read for the whole fan-out (Q#AP9). The record may be nil --- + -- paste, programmatic mutation, manual hook run, a replicated CRDT + -- op, a stale `this_command` --- and consumers are called ANYWAY, + -- with nil. That is deliberate: "this fan-out carried no typed edit" + -- is information a consumer acts on. Auto-pairing's test seam + -- observes the non-event through it, and Stage 4b abandons a pending + -- abbreviation that an unrelated edit invalidated. Skipping the + -- fan-out on nil would leave both reading stale state. + local rec = ed.take_typed_edit and ed.take_typed_edit() + + for _, c in ipairs(consumers) do + -- `buffer.after-edit` is all-must-succeed (builtin/hooks/default.lua): + -- a throwing consumer would fail the fan-out for every OTHER + -- subscriber, including lsp.lua's didChange flush. Contain it, + -- report it, and keep going --- a broken consumer must not be able + -- to stop the editor from telling the language server what changed. + -- This matches pair.lua's existing never-throw-from-after-edit + -- discipline; it does not weaken the hook's contract for anyone + -- else, because the chain itself still never fails. + local ok, claimed = pcall(c.fn, rec) + if not ok then + ed.set_status("typed-edit consumer '" .. c.name .. "' failed: " .. + tostring(claimed)) + elseif claimed then + return + end + end +end) diff --git a/src/editor.rs b/src/editor.rs index dcff55a..673ded6 100644 --- a/src/editor.rs +++ b/src/editor.rs @@ -415,6 +415,18 @@ impl EditorState { include_str!("../builtin/runtime/listview.lua"), ) .expect("load listview builtin chunk"); + // The typed-edit consumer chain (Arc 8 Stage 4a, Q#LN10) — + // ORDERING CONTRACT: typed_edit.lua must load BEFORE pair.lua, + // which registers a consumer into it, and therefore before + // lsp.lua. It owns the single `buffer.after-edit` subscriber + // that reads the one-shot typed-edit record, so its + // registration position is what preserves Q#AP7 below. + lua_host + .eval( + Some("@pmacs/builtin/runtime/typed_edit.lua"), + include_str!("../builtin/runtime/typed_edit.lua"), + ) + .expect("load typed_edit builtin chunk"); // Auto-pairing (Arc 2, Q#AP7) — ORDERING CONTRACT: pair.lua // must load BEFORE lsp.lua. Hook callbacks run in registration // order, and lsp.lua's `buffer.after-edit` callback flushes @@ -424,6 +436,9 @@ impl EditorState { // the closer stays unsynchronized until the next edit (hook // edits don't re-fire the hook). pair.lua's `pmacs.lsp.*` // lookups are lazy and nil-guarded for the same reason. + // Since Stage 4a the closer is inserted from the chain's + // subscriber rather than pair.lua's own, which is registered + // one chunk earlier — strictly safer for this contract. lua_host .eval( Some("@pmacs/builtin/runtime/pair.lua"), diff --git a/tests/typed_edit_chain_acceptance.rs b/tests/typed_edit_chain_acceptance.rs new file mode 100644 index 0000000..c0170f4 --- /dev/null +++ b/tests/typed_edit_chain_acceptance.rs @@ -0,0 +1,517 @@ +//! Typed-edit consumer chain acceptance (Arc 8 Stage 4a, +//! docs/lean4-mode-framing.md Q#LN10, criteria 46a–46e). +//! +//! The chain owns the single `buffer.after-edit` subscriber that reads +//! the one-shot typed-edit record (Q#AP9) and offers it to consumers in +//! priority order. These tests pin the chain's OWN behavior — take-once, +//! priority ordering, claim-stops-chain, throw containment, and the +//! Q#AP7 flush ordering it inherited from `pair.lua`. +//! +//! They deliberately do not re-test auto-pairing: criterion 46 requires +//! `tests/auto_pair_acceptance.rs` to pass byte-identical, and that +//! suite is the no-behavior-change pin. Pairing appears here only as +//! the chain's last consumer, which is how 46c observes that a claim +//! really stopped the chain. +//! +//! Dispatch-driven throughout: `dispatch_key` is the producer that arms +//! the record for a grid frontend. + +use crossterm::event::{KeyCode, KeyEvent, KeyEventKind, KeyEventState, KeyModifiers}; +use pmacs::editor::EditorState; +use pmacs::lua_bindings::StateDir; +use pmacs::protocol::FrontendId; +use std::path::PathBuf; +use std::sync::atomic::{AtomicUsize, Ordering}; +use std::time::{Duration, Instant}; + +fn fresh_state_dir() -> PathBuf { + static SEQ: AtomicUsize = AtomicUsize::new(0); + let dir = std::env::temp_dir().join(format!( + "pmacs-typededit-{}-{}", + std::process::id(), + SEQ.fetch_add(1, Ordering::Relaxed) + )); + std::fs::create_dir_all(&dir).unwrap(); + dir +} + +fn key(code: KeyCode, mods: KeyModifiers) -> KeyEvent { + KeyEvent { + code, + modifiers: mods, + kind: KeyEventKind::Press, + state: KeyEventState::NONE, + } +} + +fn type_str(s: &mut EditorState, text: &str) { + for ch in text.chars() { + s.dispatch_key( + FrontendId::LOCAL, + key(KeyCode::Char(ch), KeyModifiers::NONE), + ); + } +} + +fn exec(s: &EditorState, src: &str) { + s.lua_host.lua().load(src.to_string()).exec().unwrap(); +} + +fn eval(s: &EditorState, src: &str) -> T { + s.lua_host.lua().load(src.to_string()).eval().unwrap() +} + +fn buffer_text(s: &EditorState) -> String { + let b: mlua::String = eval( + s, + "local b = pmacs.window.buffer(); return b:slice(0, b:len())", + ); + String::from_utf8_lossy(&b.as_bytes()).into_owned() +} + +fn status(s: &EditorState) -> String { + s.core.borrow().status.clone() +} + +/// Fresh scratch-buffer editor, cursor at 0. Scratch pairing uses the +/// `default` set, so `(` pairs — which is what 46c reads. +fn editor_with(body: &str) -> EditorState { + let s = EditorState::new(); + if !body.is_empty() { + exec(&s, &format!("pmacs.window.buffer():insert(0, {body:?})")); + } + exec(&s, "pmacs.editor.goto_byte(0)"); + s +} + +// --------------------------------------------------------------------------- +// 46a — one read for the whole fan-out +// --------------------------------------------------------------------------- + +#[test] +fn chain_reads_the_record_once_and_hands_the_same_one_to_every_consumer() { + let mut s = editor_with(""); + exec( + &s, + r#" + _G.seen = {} + local function spy(tag) + return function(rec) + -- Each consumer independently attempts its own take. Under + -- the pre-chain design this is exactly what a second + -- consumer would have done, and exactly what would have + -- returned nil (or stolen the record from pairing). + local own = pmacs.editor.take_typed_edit() + _G.seen[#_G.seen + 1] = { + tag = tag, + char = rec and rec.char, + post_cursor = rec and rec.post_cursor, + clean = rec and rec.clean, + own_take_was_nil = (own == nil), + } + return false + end + end + pmacs.typed_edit.add_consumer { name = "spy-a", priority = 1, fn = spy("a") } + pmacs.typed_edit.add_consumer { name = "spy-b", priority = 2, fn = spy("b") } + "#, + ); + + type_str(&mut s, "x"); + + let (n, a_char, b_char, a_pc, b_pc, a_clean, b_clean, a_nil, b_nil): ( + i64, + String, + String, + i64, + i64, + bool, + bool, + bool, + bool, + ) = eval( + &s, + " + local a, b = _G.seen[1], _G.seen[2] + return #_G.seen, a.char, b.char, a.post_cursor, b.post_cursor, + a.clean, b.clean, a.own_take_was_nil, b.own_take_was_nil + ", + ); + + assert_eq!(n, 2, "both consumers ran for one typed character"); + // The same record, not two reads of a slot that only one could win. + assert_eq!(a_char, "x"); + assert_eq!(b_char, "x", "the second consumer sees the record too"); + assert_eq!((a_pc, b_pc), (1, 1), "identical post_cursor"); + assert!(a_clean && b_clean, "identical clean verdict"); + // ...and the chain, not the consumers, did the taking. + assert!( + a_nil && b_nil, + "a consumer's own take_typed_edit() observes nil — the chain \ + already consumed the one-shot slot (Q#AP9)" + ); +} + +#[test] +fn consumers_run_when_the_fan_out_carries_no_record() { + // The chain calls consumers with nil rather than skipping them. + // Three tests in the auto-pairing suite depend on this (they assert + // `_last_record == nil` after a record-less fan-out), so it is a + // load-bearing decision and not an implementation detail. + let s = editor_with(""); + exec( + &s, + r#" + _G.calls, _G.nil_calls = 0, 0 + pmacs.typed_edit.add_consumer { + name = "nil-spy", priority = 1, + fn = function(rec) + _G.calls = _G.calls + 1 + if rec == nil then _G.nil_calls = _G.nil_calls + 1 end + return false + end, + } + "#, + ); + + // A manual fan-out arms no record. + exec(&s, "pmacs.hook.run(\"buffer.after-edit\")"); + + let (calls, nil_calls): (i64, i64) = eval(&s, "return _G.calls, _G.nil_calls"); + assert_eq!(calls, 1, "the consumer ran"); + assert_eq!(nil_calls, 1, "and was handed nil, not skipped"); +} + +// --------------------------------------------------------------------------- +// 46b — priority order, not registration order +// --------------------------------------------------------------------------- + +#[test] +fn consumers_run_in_priority_order_not_registration_order() { + let mut s = editor_with(""); + // Registered HIGH priority first. If the chain honored registration + // order (or `include_str!` order, which is the same failure dressed + // differently), the observed order would be the registration order. + exec( + &s, + r#" + _G.order = {} + local function mark(tag) + return function() _G.order[#_G.order + 1] = tag; return false end + end + pmacs.typed_edit.add_consumer { name = "late", priority = 30, fn = mark("late") } + pmacs.typed_edit.add_consumer { name = "early", priority = 10, fn = mark("early") } + pmacs.typed_edit.add_consumer { name = "mid", priority = 20, fn = mark("mid") } + "#, + ); + + type_str(&mut s, "x"); + + let order: String = eval(&s, "return table.concat(_G.order, ',')"); + assert_eq!( + order, "early,mid,late", + "lowest priority runs first, regardless of when it registered" + ); +} + +#[test] +fn equal_priorities_break_by_registration_order() { + // The stated tiebreak. Lua's `table.sort` is not stable, so this + // bites an implementation that sorts instead of inserting in place. + let mut s = editor_with(""); + exec( + &s, + r#" + _G.order = {} + local function mark(tag) + return function() _G.order[#_G.order + 1] = tag; return false end + end + pmacs.typed_edit.add_consumer { name = "first", priority = 5, fn = mark("first") } + pmacs.typed_edit.add_consumer { name = "second", priority = 5, fn = mark("second") } + pmacs.typed_edit.add_consumer { name = "third", priority = 5, fn = mark("third") } + "#, + ); + + type_str(&mut s, "x"); + + let order: String = eval(&s, "return table.concat(_G.order, ',')"); + assert_eq!(order, "first,second,third"); +} + +// --------------------------------------------------------------------------- +// 46c — a claim stops the chain +// --------------------------------------------------------------------------- + +#[test] +fn a_claiming_consumer_stops_the_chain() { + let mut s = editor_with(""); + exec( + &s, + r#" + _G.later_ran = false + pmacs.typed_edit.add_consumer { + name = "claimer", priority = 1, fn = function() return true end, + } + pmacs.typed_edit.add_consumer { + name = "later", priority = 2, + fn = function() _G.later_ran = true; return false end, + } + "#, + ); + + type_str(&mut s, "("); + + let later_ran: bool = eval(&s, "return _G.later_ran"); + assert!(!later_ran, "a later consumer must not run after a claim"); + // Pairing is the chain's last consumer at priority 100, so the + // claim is observable in the buffer: no closer was inserted. This + // is the assertion that makes the criterion about behavior rather + // than about a bookkeeping flag. + assert_eq!( + buffer_text(&s), + "(", + "auto-pairing never ran, so the opener stands alone" + ); +} + +#[test] +fn a_non_claiming_consumer_does_not_stop_the_chain() { + let mut s = editor_with(""); + exec( + &s, + r#" + _G.later_ran = false + pmacs.typed_edit.add_consumer { + name = "passer", priority = 1, fn = function() return false end, + } + pmacs.typed_edit.add_consumer { + name = "later", priority = 2, + fn = function() _G.later_ran = true; return false end, + } + "#, + ); + + type_str(&mut s, "("); + + let later_ran: bool = eval(&s, "return _G.later_ran"); + assert!(later_ran, "a declining consumer passes the edit along"); + assert_eq!( + buffer_text(&s), + "()", + "and pairing, still last in the chain, reacted normally" + ); +} + +// --------------------------------------------------------------------------- +// 46d — a throwing consumer is contained +// --------------------------------------------------------------------------- + +#[test] +fn a_throwing_consumer_is_contained_reported_and_does_not_stop_the_chain() { + let mut s = editor_with(""); + exec( + &s, + r#" + _G.later_ran = false + pmacs.typed_edit.add_consumer { + name = "boom", priority = 1, + fn = function() error("consumer exploded") end, + } + pmacs.typed_edit.add_consumer { + name = "later", priority = 2, + fn = function() _G.later_ran = true; return false end, + } + "#, + ); + + // `buffer.after-edit` is all-must-succeed: an uncontained throw + // would fail the fan-out for every other subscriber, including + // lsp.lua's didChange flush. + type_str(&mut s, "("); + + let later_ran: bool = eval(&s, "return _G.later_ran"); + assert!(later_ran, "a throwing consumer must not stop the chain"); + assert_eq!( + buffer_text(&s), + "()", + "and pairing still ran — the fan-out survived the throw" + ); + let st = status(&s); + assert!( + st.contains("boom") && st.contains("consumer exploded"), + "the failure is reported by consumer name and message, got {st:?}" + ); +} + +#[test] +fn add_consumer_rejects_malformed_registrations() { + let s = editor_with(""); + for (src, want) in [ + ( + "pmacs.typed_edit.add_consumer(\"nope\")", + "spec must be a table", + ), + ( + "pmacs.typed_edit.add_consumer{ priority = 1, fn = function() end }", + "name must be a non-empty string", + ), + ( + "pmacs.typed_edit.add_consumer{ name = \"n\", fn = function() end }", + "priority must be a number", + ), + ( + "pmacs.typed_edit.add_consumer{ name = \"n\", priority = 1 }", + "fn must be a function", + ), + ] { + let err = s + .lua_host + .lua() + .load(src.to_string()) + .exec() + .expect_err("malformed registration must throw"); + let msg = err.to_string(); + assert!( + msg.contains(want), + "expected {want:?} in the error for {src:?}, got {msg:?}" + ); + } +} + +// --------------------------------------------------------------------------- +// 46e — the Q#AP7 flush ordering the chain inherited +// --------------------------------------------------------------------------- + +fn fake_lsp_path() -> String { + env!("CARGO_BIN_EXE_pmacs_fake_lsp").to_owned() +} + +fn pump_lua_flag(state: &mut EditorState, flag: &str, secs: u64) -> bool { + let deadline = Instant::now() + Duration::from_secs(secs); + loop { + state.tick_processes(); + state.tick_lsp(); + state.tick_async(); + let done: bool = state + .lua_host + .lua() + .load(format!("return ({flag}) == true")) + .eval() + .unwrap_or(false); + if done { + return true; + } + if Instant::now() >= deadline { + return false; + } + std::thread::sleep(Duration::from_millis(10)); + } +} + +/// The `text` of every `textDocument/didChange` line in the sink, in +/// arrival order. +fn did_change_texts(sink: &std::path::Path) -> Vec { + let Ok(raw) = std::fs::read_to_string(sink) else { + return Vec::new(); + }; + raw.lines() + .filter_map(|l| serde_json::from_str::(l).ok()) + .filter(|v| v.get("method").and_then(|m| m.as_str()) == Some("textDocument/didChange")) + .filter_map(|v| v.get("text").and_then(|t| t.as_str()).map(str::to_owned)) + .collect() +} + +#[test] +fn a_chain_consumers_edit_reaches_the_first_did_change() { + // Q#AP7 generalized from pairing to the chain: lsp.lua's after-edit + // callback flushes didChange SYNCHRONOUSLY on the signature-trigger + // path, so every reaction to a typed character must already be in + // the buffer when it runs. The auto-pairing suite pins this for + // pairing; this pins it for the chain itself, which is what now + // owns the registration position. + // + // Falsified by loading typed_edit.lua after lsp.lua in + // `src/editor.rs`: the consumer's text would then arrive in the + // SECOND didChange, or not at all. + let dir = fresh_state_dir(); + let sink = dir.join("changes.jsonl"); + let sink_disp = sink.display().to_string(); + let fake = fake_lsp_path(); + + let mut s = EditorState::new(); + s.lua_host.lua().remove_app_data::(); + s.lua_host.lua().set_app_data(StateDir(dir.clone())); + exec(&s, "pmacs.lsp.config = {}"); + exec( + &s, + &format!( + "pmacs.lsp.config.rust = {{ + command = '{fake}', + env = {{ + PMACS_FAKE_LSP_MODE = 'sighelp', + PMACS_FAKE_LSP_CHANGE_SINK = '{sink_disp}', + }}, + }}" + ), + ); + + // A consumer that appends a marker of its own, ahead of pairing. + // It declines the claim so pairing still runs — the assertion is + // about ordering against the flush, not about claiming. + exec( + &s, + r#" + pmacs.typed_edit.add_consumer { + name = "marker", priority = 1, + fn = function(rec) + if not rec then return false end + if rec.char ~= "(" then return false end + local buf = pmacs.window.buffer() + buf:insert(buf:len(), "Z") + return false + end, + } + "#, + ); + + let f = dir.join("a.rs"); + std::fs::write(&f, "\n").unwrap(); + let fd = f.display().to_string(); + exec(&s, &format!("pmacs.buffer.find_or_open({fd:?})")); + exec(&s, "pmacs.editor.goto_byte(0)"); + let initialized = "(function() \ + for _,r in ipairs(pmacs.lsp.list()) do \ + if r.state and r.state.kind=='initialized' then return true end \ + end \ + return false \ + end)()"; + assert!(pump_lua_flag(&mut s, initialized, 5), "fake server init"); + + type_str(&mut s, "("); + assert_eq!( + buffer_text(&s), + "()\nZ", + "both the chain consumer's marker and pairing's closer landed" + ); + + let deadline = Instant::now() + Duration::from_secs(5); + let changes = loop { + s.tick_processes(); + s.tick_lsp(); + s.tick_async(); + let c = did_change_texts(&sink); + if !c.is_empty() { + break c; + } + assert!( + Instant::now() < deadline, + "no didChange reached the fake server" + ); + std::thread::sleep(Duration::from_millis(10)); + }; + assert_eq!( + changes[0], "()\nZ", + "the FIRST didChange carries BOTH reactions — the chain ran \ + before lsp.lua's synchronous flush (Q#AP7)" + ); +} From c7072b49e9e7ef427fd449100aa3d3ed5b70f80a Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sun, 26 Jul 2026 13:05:41 -0400 Subject: [PATCH 29/32] docs: record the Stage 4a lane and its bite table Verification describes the pushed tree, per the standing rule. Includes the bite that was worthless as first written: moving only typed_edit.lua past lsp.lua broke the runtime load instead of testing flush ordering. A bite that kills everything has not isolated anything. Co-Authored-By: Claude Opus 5 (1M context) Claude-Session: https://claude.ai/code/session_011LFvC4FQtux4y32KuevZ7B --- docs/active-work.md | 56 ++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 53 insertions(+), 3 deletions(-) diff --git a/docs/active-work.md b/docs/active-work.md index cf4ea6e..bd0c0b1 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -57,7 +57,7 @@ git status --short --branch The `git log` command must expose `d152120` or a newer intentional main. If it does not, stop and repair the remote/fetch configuration. -## Lean 4 lane (Arc 8) — Stages 1, 2, 3a, 3b MERGED; Stage 4 IN FRAMING +## Lean 4 lane (Arc 8) — Stages 1, 2, 3a, 3b MERGED; Stage 4a IN REVIEW - **Stages 1, 2, 3a and 3b are MERGED** — #160 (`main` @ `0827dd1`), #161 (`46a1b8f`), #167 (`6f348c9`), #170 (`d400f30`). Their full @@ -146,8 +146,58 @@ If it does not, stop and repair the remote/fetch configuration. `handle_server_requests` 1549→1815, `fs.stat` 93→133, `detect_buffer_language` 452→457, `send_request`/`send_notification` 9342/9361→9507/9527. -- Verification: none yet — the branch carries no code. `git diff --check` - clean. +### Stage 4a — the typed-edit consumer chain (IMPLEMENTED, same branch) + +- Footprint exactly as Q#LN10 declares it: `builtin/runtime/typed_edit.lua` + (new, 112 lines), `pair.lua` re-expressed as one consumer, + `src/editor.rs` +15 (the `include_str!` and its ordering comment), and + `tests/typed_edit_chain_acceptance.rs` (new, 9 tests). + **`tests/auto_pair_acceptance.rs` is UNCHANGED — `git diff --stat + main...HEAD -- tests/auto_pair_acceptance.rs` is empty.** That is + criterion 46 checked at the diff, which is the only way it means + anything. +- **The chain calls consumers even when the record is nil.** This is a + decision, not an implementation detail: three existing auto-pairing + tests assert `pmacs.pair._last_record == nil` after a record-less + fan-out (paste, programmatic insert, nested manual `hook.run`), so + skipping consumers on nil fails them. Stage 4b needs the same + delivery to abandon a pending abbreviation an unrelated edit + invalidated. +- **Ordered insertion, not `table.sort`** — Lua's sort is not stable, and + "ties broken by registration order" is a stated contract. +- **The chain `pcall`s each consumer** and reports through + `set_status`. `buffer.after-edit` is all-must-succeed, so an + uncontained throw fails the fan-out for every other subscriber + including lsp.lua's didChange flush. +- **Every acceptance test is bite-verified by mutation**, per the + standing rule that a test is not evidence until the mutation it + targets has been shown to fail it: + + | Mutation | Tests it fails | + |---|---| + | append instead of ordered insert | 5 chain | + | `>=` instead of `>` in the insert scan | 1 chain (tiebreak) | + | re-take the record per consumer | 4 chain | + | ignore the claim return value | 1 chain | + | drop the `pcall` | 1 chain | + | skip consumers when `rec == nil` | 1 chain + **3 auto-pair** | + | load `typed_edit.lua` after `lsp.lua` | 1 chain + **2 auto-pair** (Q#AP7) | + + The first attempt at the last bite was WORTHLESS as written: moving + only `typed_edit.lua` past `lsp.lua` left `pair.lua` calling a nil + `add_consumer`, so the runtime failed to load and all 9 tests died — + loud, but not a test of the flush-ordering property. Moving + `typed_edit.lua` AND `pair.lua` past `lsp.lua` is the faithful + falsification: registration succeeds, the hook lands late, and exactly + the three ordering tests fail. **A bite that kills everything has not + isolated anything.** +- Verification on this branch (commit-then-gate, so this describes the + pushed tree): `cargo fmt --check` clean; strict workspace Clippy + clean; 1,832 default + 2,009 CRDT library tests; auto-pair 45/45; + typed-edit chain 9/9; M4 121; required GPU 202; **isolated-config + workspace sweep 3,328 across 97 suites, zero failures** with + `grep -c basedpyright` = 0; `git diff --check` clean. +- Stage 4b (the input method) is NOT in this PR and not started. ## Dired lane — Stage 0 MERGED; Stage 1 IN REVIEW (PR #165) From 842417200a24802b6caabccdc1bd7967af98f5fd Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sun, 26 Jul 2026 13:18:05 -0400 Subject: [PATCH 30/32] fix(panel): close Stage 2A review round 3 (2 P1) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit **P1-1 — layout invalidation could suppress the authoritative clear.** Real bug. Both render paths resolved the document identity AFTER the evaluator ran callbacks, but BOTH outcome arms carry PHASE-1 contexts. A provider that closes the primary document split changes `primary_document_window` mid-evaluation, so the filter compared phase-1 contexts against a replacement identity, matched nothing, and emitted no clear — leaving stale statusline text on the wire forever. The identity is now captured BEFORE `evaluate_statusline` runs and threaded through both paths (the terminal path via `terminal_chrome`). Pinning it took three attempts, and the two failures are the useful part: - `pmacs.window.close()` takes no argument — it closes the ACTIVE window. The first version passed a window id that was silently ignored, so it closed the panel instead of the document. - The Lua window API acts on the ACTIVE FRONTEND, so driving it against a synthetic semantic view changed nothing at all. - Closing the only document window is structurally REFUSED (Q#BP6 forbids a lone side window as a resting state), so the fixture needs TWO document windows for the close to be legal. The test now asserts its own precondition — that the callback really changed the identity — before asserting the clear, and reproduces the reported symptom (no `StatuslineSegments` at all) when the fix is reverted. **P1-2 — #21 was pinned at the helper, not the producer.** Confirmed: reverting only the call site inside `publish_buffer_snapshot_to_replicas` left both the helper test and the existing socket-pair test green. The helper assertions are removed (with a note saying why) and replaced by `snapshot_publication_follows_the_document_under_a_focused_panel`, which drives the real producer over socket pairs and asserts BOTH directions: the document buffer's snapshot is delivered while a panel holds focus, and a panel-only buffer's is not. Biting that test exposed a defect in the test itself: the delivery read had no timeout, so a regression made it HANG rather than fail. A hanging test is strictly worse than a red one — every read now has a timeout. Gates: fmt clean; workspace clippy clean; 1,832 default + 2,015 CRDT library; Stage 2A 17; Stage 1 46; statusline 8; m11_5 2; GPU initial target 14; terminal config 12; folding Stage 2 48; vterm 1/2 10 / 6; M4 121; required GPU 202; `git diff --check` clean. Co-Authored-By: Claude Opus 5 (1M context) --- docs/active-work.md | 13 ++- src/daemon.rs | 106 +++++++++++++++++++-- src/semantic_render.rs | 53 ++++++++--- tests/bottom_panel_stage2a_acceptance.rs | 112 +++++++++++++++++++++++ 4 files changed, 259 insertions(+), 25 deletions(-) diff --git a/docs/active-work.md b/docs/active-work.md index 36a99de..a46a908 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -494,14 +494,14 @@ implemented and in review.** #173 also changes `src/editor.rs`, so gates were rerun on the merge result, not the old combination). Five commits: the classified census routing, the painter extraction + acceptance, the lane record, then - the round-1 and round-2 review fixes. **No protocol change; no behavior + the round-1, round-2 and round-3 review fixes. **No protocol change; no behavior change for any frontend today** — with `panel_capable = false` for semantic sessions, `primary_document_window` returns `view.active` in every existing configuration, so this is seam adoption that becomes load-bearing in 2B. - Verification on the merge result: `cargo fmt --check` clean; strict - workspace Clippy clean; **1,832 default + 2,014 CRDT** library tests; - `bottom_panel_stage2a_acceptance` **16**; bottom-panel Stage 1 46; + workspace Clippy clean; **1,832 default + 2,015 CRDT** library tests; + `bottom_panel_stage2a_acceptance` **17**; bottom-panel Stage 1 46; statusline segments 8 CRDT; m11_5 semantic 2 CRDT; GPU initial target 14 CRDT; terminal config 12 CRDT; vterm Stage 1/2 10 / 6; folding Stage 2 48; M4 121; required GPU 202; `git diff --check` clean. @@ -521,7 +521,12 @@ implemented and in review.** under test; (c) a discriminating fixture must make the two routings DISAGREE — comparing two non-terminal buffers, or two windows with no selection, yields the same answer either way and proves nothing. - Round 2 found four of my own pins vacuous by exactly these shapes. + Round 2 found four of my own pins vacuous by exactly these shapes, and + round 3 found two more problems of the same family: a pin placed at a + HELPER while production called it from a producer (reverting only the + producer's call site left every test green), and a socket-pair + assertion whose blocking read made a regression HANG instead of fail. + Both now assert at the producer, with read timeouts on every read. - **Review round 1 closed: 4 P1 + 2 P2, all real.** The P1s were a stale-`Pointer` focus steal (the failed-alignment arm returned the window, so #8's activation focused it before `dispatch_pointer` diff --git a/src/daemon.rs b/src/daemon.rs index 63f6b87..84716eb 100644 --- a/src/daemon.rs +++ b/src/daemon.rs @@ -3483,6 +3483,96 @@ mod tests { ); } + /// Bottom-panel §1.3 #21 through the REAL producer (round 3). + /// + /// A semantic peer with a FOCUSED PANEL must still receive the + /// snapshot for the buffer on its DOCUMENT surface, and must NOT + /// receive one for a buffer visible only in its panel. Asserting the + /// helper alone was insufficient: reverting the producer's call site + /// to focused-window routing left every helper-level test green. + #[cfg(feature = "crdt")] + #[test] + fn snapshot_publication_follows_the_document_under_a_focused_panel() { + let (editor, fid, document, panel) = panel_focused_semantic_fixture(); + let (doc_buf, panel_buf) = { + let core = editor.core.borrow(); + ( + core.windows[&document].buffer_id, + core.windows[&panel].buffer_id, + ) + }; + assert_ne!(doc_buf, panel_buf, "fixture: distinct buffers"); + + let caps = crate::protocol::NegotiatedCapabilities { + multi_frontend: true, + crdt_replica: true, + semantic_render: true, + }; + let mut registry = SessionRegistry::new(); + registry.register_session( + fid, + crate::presence::SessionState::new(PROTOCOL_VERSION, caps, 0), + ); + + // The DOCUMENT buffer's snapshot must be delivered. + { + let (server, mut client) = UnixStream::pair().expect("socketpair"); + // A read timeout on the DELIVERY read too. Without it a + // regression that suppresses the snapshot makes this test + // HANG rather than fail, which is strictly worse than a red + // assertion — found by biting this very test. + client + .set_read_timeout(Some(Duration::from_millis(500))) + .expect("delivery timeout"); + let mut streams = HashMap::from([(fid, server)]); + let message = InstanceMessage::BufferSnapshot { + buffer_id: doc_buf, + crdt_snapshot: vec![1, 2, 3], + }; + publish_buffer_snapshot_to_replicas( + &editor, + doc_buf, + &message, + ®istry, + &mut streams, + &mut HashMap::new(), + ); + let delivered: InstanceMessage = + read_message(&mut client).expect("the document snapshot must arrive"); + assert_eq!( + delivered, message, + "#21: a buffer on the DOCUMENT surface must still be published while a \ + panel holds focus" + ); + } + + // The PANEL-only buffer's snapshot must NOT be delivered. + { + let (server, mut client) = UnixStream::pair().expect("socketpair"); + let mut streams = HashMap::from([(fid, server)]); + let message = InstanceMessage::BufferSnapshot { + buffer_id: panel_buf, + crdt_snapshot: vec![4, 5, 6], + }; + publish_buffer_snapshot_to_replicas( + &editor, + panel_buf, + &message, + ®istry, + &mut streams, + &mut HashMap::new(), + ); + client + .set_read_timeout(Some(Duration::from_millis(50))) + .expect("timeout"); + assert!( + read_message::(&mut client).is_err(), + "#21: a buffer visible only in a PANEL must not replace the peer's \ + document mirror" + ); + } + } + // ---- GPU terminal input: the double terminal-layout sync ------------- // // These drive `sync_terminal_layouts_for_tick` — the REAL dispatcher loop @@ -4827,15 +4917,13 @@ mod tests { "#3: CursorByte must describe the DOCUMENT surface" ); - // #21 publication recipient filter, both directions. - assert!( - peer_displays_buffer_as_document(&editor, fid, doc_buf), - "#21: a buffer visible in the document must still receive publications while a panel holds focus" - ); - assert!( - !peer_displays_buffer_as_document(&editor, fid, panel_buf), - "#21: a buffer visible only in a panel must NOT replace the document mirror" - ); + // #21 is deliberately NOT asserted here. Round 3: pinning it at + // this helper left the real producer free to regress — reverting + // the call site inside `publish_buffer_snapshot_to_replicas` + // kept both this test and the existing socket-pair test green. + // It is pinned through the producer instead, in + // `snapshot_publication_follows_the_document_under_a_focused_panel`. + let _ = panel_buf; } /// Bottom-panel §1.3 #2 — the sharpest census case: the lazy CRDT diff --git a/src/semantic_render.rs b/src/semantic_render.rs index af08741..db96fc4 100644 --- a/src/semantic_render.rs +++ b/src/semantic_render.rs @@ -625,6 +625,22 @@ impl SemanticRenderState { // post-evaluation face inventory must then precede the authoritative // segment replacement in this same frame. Unsupported peers skip the // evaluator entirely and therefore pay no Lua callback/dynamic-face cost. + // Bottom-panel A2A-2, round 3: the document identity used to + // FILTER the results must be the PRE-CALLBACK one. Both outcome + // arms carry phase-1 contexts, and a provider that closes the + // primary document split changes `primary_document_window` + // mid-evaluation — reading it after the fact would compare + // phase-1 contexts against a replacement identity, match + // nothing, and silently suppress the authoritative clear. + let statusline_document_window = self + .peer_knows_statusline_segments + .then(|| { + state + .core + .borrow() + .primary_document_window(self.frontend_id) + }) + .flatten(); let statusline_evaluation = self.peer_knows_statusline_segments.then(|| { evaluate_statusline( state.lua_host.lua(), @@ -810,11 +826,7 @@ impl SemanticRenderState { out.extend(self.font_facts_msg(state)); // Q#SL6/Q#SL8: face inventory must precede segment text. if let Some(evaluation) = statusline_evaluation { - let document_window = state - .core - .borrow() - .primary_document_window(self.frontend_id); - self.emit_statusline_segments(evaluation, document_window, &mut out); + self.emit_statusline_segments(evaluation, statusline_document_window, &mut out); } out } @@ -859,6 +871,16 @@ impl SemanticRenderState { // Evaluate callbacks before `ThemeFacts` for the same reason the // document path does: a callback may register a face, and the // face inventory must precede the segment text that names it. + // Same pre-callback capture as the document path (round 3). + let statusline_document_window = self + .peer_knows_statusline_segments + .then(|| { + state + .core + .borrow() + .primary_document_window(self.frontend_id) + }) + .flatten(); let statusline_evaluation = self.peer_knows_statusline_segments.then(|| { evaluate_statusline( state.lua_host.lua(), @@ -885,7 +907,12 @@ impl SemanticRenderState { // a verdict we hold. if self.last_terminal_frame.as_ref() == Some(&frame) { self.terminal_error_latched = false; - out.extend(self.terminal_chrome(state, buffer_id, statusline_evaluation)); + out.extend(self.terminal_chrome( + state, + buffer_id, + statusline_evaluation, + statusline_document_window, + )); return Some(out); } match frame.validate() { @@ -909,7 +936,12 @@ impl SemanticRenderState { } } - out.extend(self.terminal_chrome(state, buffer_id, statusline_evaluation)); + out.extend(self.terminal_chrome( + state, + buffer_id, + statusline_evaluation, + statusline_document_window, + )); Some(out) } @@ -924,6 +956,7 @@ impl SemanticRenderState { state: &EditorState, buffer_id: BufferId, statusline_evaluation: Option, + statusline_document_window: Option, ) -> Vec { let mut out = Vec::new(); out.extend(self.status_facts_msg(state, buffer_id)); @@ -933,11 +966,7 @@ impl SemanticRenderState { out.extend(self.font_facts_msg(state)); // Q#SL6/Q#SL8: face inventory must precede segment text. if let Some(evaluation) = statusline_evaluation { - let document_window = state - .core - .borrow() - .primary_document_window(self.frontend_id); - self.emit_statusline_segments(evaluation, document_window, &mut out); + self.emit_statusline_segments(evaluation, statusline_document_window, &mut out); } out } diff --git a/tests/bottom_panel_stage2a_acceptance.rs b/tests/bottom_panel_stage2a_acceptance.rs index 1063e9f..d39d804 100644 --- a/tests/bottom_panel_stage2a_acceptance.rs +++ b/tests/bottom_panel_stage2a_acceptance.rs @@ -784,3 +784,115 @@ fn consumer_decorations_follow_the_document_selection_not_the_panel() { "a selection living in the focused PANEL must not decorate the document viewport" ); } + +#[test] +fn a_provider_closing_the_document_split_still_clears_the_statusline() { + use pmacs::protocol::{ByteRange, InstanceMessage}; + use pmacs::semantic_render::SemanticRenderState; + + // Round 3 finding 1. `authoritative_empty` carries PHASE-1 contexts, + // so the identity used to filter them must be the PRE-CALLBACK one. + // A provider that closes the primary document split changes + // `primary_document_window` mid-evaluation; reading it afterwards + // compares phase-1 contexts against a replacement identity, matches + // nothing, and silently suppresses the authoritative clear — leaving + // stale statusline text on screen forever. + // + // Driven on LOCAL, because the Lua window API acts on the ACTIVE + // FRONTEND: a synthetic semantic view would be untouched by + // `pmacs.window.close()` and the identity would never change, which + // is exactly how the first version of this test came back vacuous. + // TWO document windows plus the panel: closing the only document + // window is structurally refused (Q#BP6 forbids a lone side window + // as a resting state), so the first attempt could not change the + // identity at all. Distinct buffers make the target selectable from + // a Lua provider, which has no focus-by-id. + let s = editor(); + exec( + &s, + "DOC_A = pmacs.buffer.create(\"*doc-a*\") + DOC_B = pmacs.buffer.create(\"*doc-b*\") + pmacs.window.display(DOC_A, {}) + pmacs.window.split_horizontal() + pmacs.window.focus_next() + pmacs.window.display(DOC_B, {})", + ); + let (_origin, panel) = focused_panel(&s); + let (document, doc_buf) = { + let core = s.core.borrow(); + let win = core + .primary_document_window(FrontendId::LOCAL) + .expect("a primary document window"); + (win, core.windows[&win].buffer_id) + }; + assert_ne!(document, panel); + + let mut sem = SemanticRenderState::for_peer(FrontendId::LOCAL, 18); + sem.set_viewport(doc_buf, ByteRange { start: 0, end: 0 }, 0); + + // Seed a baseline payload so a CLEAR is observable as a change. + exec( + &s, + r"_G.SL_SEED = pmacs.statusline.register { + name='seed', side='left', priority=10, + fn=function() return 'OLD' end, + }", + ); + let seeded = sem.render_frame(&s); + assert!( + seeded + .iter() + .any(|m| matches!(m, InstanceMessage::StatuslineSegments { .. })), + "non-vacuity: a baseline payload must exist before we test its clear" + ); + + // A provider that unregisters itself (making the evaluation + // Invalidated) AND closes the captured document window. `close()` + // closes the ACTIVE window and Lua has no focus-by-id, so step + // around the ring until the captured buffer is current. + s.lua_host + .lua() + .globals() + .set("TARGET_BUF", pmacs::lua_bindings::BufferIdLua(doc_buf)) + .expect("expose the target buffer"); + exec( + &s, + r"_G.SL_CLOSER = pmacs.statusline.register { + name='closer', side='left', priority=100, + fn=function() + pmacs.statusline.unregister(SL_CLOSER) + for _ = 1, 8 do + if pmacs.window.buffer() == TARGET_BUF then break end + pmacs.window.focus_next() + end + pmacs.window.close() + return 'STALE' + end, + }", + ); + + let msgs = sem.render_frame(&s); + + // The fixture must actually have changed the identity, or this test + // discriminates nothing. + assert_ne!( + s.core.borrow().primary_document_window(FrontendId::LOCAL), + Some(document), + "fixture: the callback must really have changed the document identity" + ); + + let cleared = msgs.iter().any(|m| match m { + InstanceMessage::StatuslineSegments { + buffer_id, + left, + right, + .. + } => *buffer_id == doc_buf && left.is_empty() && right.is_empty(), + _ => false, + }); + assert!( + cleared, + "an invalidated evaluation must still publish the authoritative EMPTY clear for \ + the phase-1 document identity, even when a callback closed that window; got {msgs:?}" + ); +} From aef4e98c26ecb2c840c564e2a4f721241e229e0a Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sun, 26 Jul 2026 13:39:33 -0400 Subject: [PATCH 31/32] fix(typed-edit): close round-8 review on the consumer chain MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Five defects in the chain itself, plus the stale handoff state. Each consumer now gets its own shallow copy of the typed-edit record. Handing everyone the same table let a DECLINING consumer rewrite provenance for the ones behind it, and pairing decides what to close from `rec.char` — so a forged `char` turned a typed `x` into `x)`. Every field is a scalar or an opaque id, so a shallow copy is complete. The fan-out iterates a snapshot of the consumer list. It was iterating the same array `add_consumer` mutates: a consumer that registered a lower-priority one shifted itself forward under `ipairs` and ran twice, and re-registering made that unbounded. Registrations and removals made during a fan-out now take effect on the next one, stated as a contract and pinned in both directions. `tostring` on the caught error moved inside the containment. A Lua error may be any value, including a table whose `__tostring` throws — rendering it outside the `pcall` reintroduced exactly the escape the containment exists to prevent. Priorities are validated as finite integers in i32 range, matching `pmacs.completion.register`. NaN is a number and every ordered comparison with it is false, so a NaN consumer landed wherever the insertion scan gave up and silently voided the lowest-first ordering that Q#LN22 depends on. `add_consumer` returns a handle and `remove_consumer` unregisters it, reporting whether it was live. Without teardown the chain inherited the `pmacs.hook.add` callback leak COHERENCE.md §13 already records, and spread it to every consumer. Also corrects the rationale the containment was documented with, in the module, the test, and the framing: an uncontained throw does NOT take the fan-out's other subscribers down. `run_all_must_succeed` (src/hook.rs:332) collects errors and continues, so lsp.lua still flushes didChange. The containment is still required — the throw skips every later consumer in the chain — but the reason is narrower than rev 7 claimed. Criteria 46f (record isolation), 46g (snapshot iteration), and 46h (lifecycle and priority validation) added; 46d's rationale corrected. Four new tests, all bite-verified by mutation, each failing only its target: shared record table (1), live-array iteration (1), unprotected tostring (1), bare number check (1), no-op removal (2). The suite also runs green under `--features lua54`. docs/agent-handoff.md said Stage 4a was awaiting approval while this branch had it implemented and in review. Co-Authored-By: Claude Opus 5 (1M context) Claude-Session: https://claude.ai/code/session_011LFvC4FQtux4y32KuevZ7B --- builtin/runtime/typed_edit.lua | 113 ++++++++++--- docs/active-work.md | 41 ++++- docs/agent-handoff.md | 24 ++- docs/lean4-mode-framing.md | 42 ++++- tests/typed_edit_chain_acceptance.rs | 232 ++++++++++++++++++++++++++- 5 files changed, 406 insertions(+), 46 deletions(-) diff --git a/builtin/runtime/typed_edit.lua b/builtin/runtime/typed_edit.lua index 59f7366..6baf5f8 100644 --- a/builtin/runtime/typed_edit.lua +++ b/builtin/runtime/typed_edit.lua @@ -13,11 +13,12 @@ -- subscriber that reads the record, and offers that one read to -- consumers registered through `pmacs.typed_edit.add_consumer`: -- --- pmacs.typed_edit.add_consumer { --- name = "auto-pair", -- for error reporting; must be unique-ish +-- local handle = pmacs.typed_edit.add_consumer { +-- name = "auto-pair", -- for error reporting -- priority = 100, -- LOWEST runs FIRST -- fn = function(rec) ... return claimed end, -- } +-- pmacs.typed_edit.remove_consumer(handle) -- -> true if it was live -- -- A consumer returns whether it CLAIMED the edit; the first that claims -- stops the chain. "Claimed" means the chain stops, not that an edit was @@ -46,10 +47,19 @@ pmacs.typed_edit = pmacs.typed_edit or {} -- order" is part of the stated contract, not an incidental property. local consumers = {} --- Register a typed-edit consumer. Argument errors throw: registration --- happens at chunk-load or config-load time, where a throw is a visible --- startup failure rather than a silently missing feature. Nothing in --- the after-edit path throws --- see the fan-out below. +-- Handles are opaque to callers; only identity matters. An integer +-- counter is enough because nothing ever reuses one. +local next_handle = 0 + +-- `math.huge` is the only portable spelling of infinity available in +-- both LuaJIT and 5.4, and NaN is the only value not equal to itself. +local INT32_MIN, INT32_MAX = -2147483648, 2147483647 + +-- Register a typed-edit consumer; returns an opaque handle for +-- `remove_consumer`. Argument errors throw: registration happens at +-- chunk-load or config-load time, where a throw is a visible startup +-- failure rather than a silently missing feature. Nothing in the +-- after-edit path throws --- see the fan-out below. function pmacs.typed_edit.add_consumer(spec) if type(spec) ~= "table" then error("pmacs.typed_edit.add_consumer: spec must be a table", 2) @@ -58,9 +68,18 @@ function pmacs.typed_edit.add_consumer(spec) if type(name) ~= "string" or name == "" then error("pmacs.typed_edit.add_consumer: name must be a non-empty string", 2) end - if type(priority) ~= "number" then + -- A bare `type(priority) == "number"` admits NaN and the infinities, + -- and EVERY ordered comparison against NaN is false --- so a NaN + -- consumer silently lands wherever the insertion scan happens to give + -- up, and the lowest-first contract other consumers depend on stops + -- holding. Bounded integers match `pmacs.completion.register`, whose + -- priority is an i32 on the Rust side. + if type(priority) ~= "number" or priority ~= priority + or priority == math.huge or priority == -math.huge + or priority % 1 ~= 0 + or priority < INT32_MIN or priority > INT32_MAX then error("pmacs.typed_edit.add_consumer: " .. name .. - ": priority must be a number", 2) + ": priority must be a finite integer in [-2147483648, 2147483647]", 2) end if type(fn) ~= "function" then error("pmacs.typed_edit.add_consumer: " .. name .. @@ -77,7 +96,27 @@ function pmacs.typed_edit.add_consumer(spec) break end end - table.insert(consumers, at, { name = name, priority = priority, fn = fn }) + next_handle = next_handle + 1 + local handle = next_handle + table.insert(consumers, at, + { handle = handle, name = name, priority = priority, fn = fn }) + return handle +end + +-- Unregister a consumer by the handle `add_consumer` returned. Returns +-- true if it was registered, false otherwise (so a double-remove is a +-- reportable no-op rather than a throw). Without this, re-evaluating a +-- config or reloading a package accumulates callbacks permanently --- +-- the leak COHERENCE.md §13 already records against `pmacs.hook.add`, +-- which this chain would otherwise inherit and spread. +function pmacs.typed_edit.remove_consumer(handle) + for i, c in ipairs(consumers) do + if c.handle == handle then + table.remove(consumers, i) + return true + end + end + return false end pmacs.hook.add("buffer.after-edit", function() @@ -92,19 +131,51 @@ pmacs.hook.add("buffer.after-edit", function() -- fan-out on nil would leave both reading stale state. local rec = ed.take_typed_edit and ed.take_typed_edit() - for _, c in ipairs(consumers) do - -- `buffer.after-edit` is all-must-succeed (builtin/hooks/default.lua): - -- a throwing consumer would fail the fan-out for every OTHER - -- subscriber, including lsp.lua's didChange flush. Contain it, - -- report it, and keep going --- a broken consumer must not be able - -- to stop the editor from telling the language server what changed. - -- This matches pair.lua's existing never-throw-from-after-edit - -- discipline; it does not weaken the hook's contract for anyone - -- else, because the chain itself still never fails. - local ok, claimed = pcall(c.fn, rec) + -- Iterate a SNAPSHOT. A consumer may register or remove consumers + -- while the chain is running, and `table.insert`/`table.remove` on + -- the live array shifts indices under `ipairs` --- a consumer that + -- registers a lower-priority one shifts itself forward and runs + -- twice, and repeating that is unbounded. Registrations and removals + -- made during a fan-out therefore take effect on the NEXT fan-out. + local snapshot = {} + for i, c in ipairs(consumers) do + snapshot[i] = c + end + + for _, c in ipairs(snapshot) do + -- Each consumer gets its OWN copy of the record. The table handed + -- out is plain Lua data, so a declining consumer could otherwise + -- edit `rec.char` in place and the next consumer would act on the + -- forged value --- auto-pairing reads `rec.char` to decide what to + -- close, so a rewritten `char` makes it insert a pair the user + -- never typed. Every field is a scalar or an opaque id, so a + -- shallow copy is a complete snapshot. + local mine = nil + if rec ~= nil then + mine = {} + for k, v in pairs(rec) do + mine[k] = v + end + end + + -- Contain the consumer. A throw here would skip every LATER + -- consumer in the chain and mark the whole `buffer.after-edit` run + -- failed; the other subscribers still run, because all-must-succeed + -- collects errors and continues (`src/hook.rs`'s + -- `run_all_must_succeed`), but one broken consumer must not be able + -- to silently disable the ones behind it. This matches pair.lua's + -- existing never-throw-from-after-edit discipline. + local ok, claimed = pcall(c.fn, mine) if not ok then - ed.set_status("typed-edit consumer '" .. c.name .. "' failed: " .. - tostring(claimed)) + -- Rendering is itself protected: a Lua error may be any value, + -- including a table whose `__tostring` throws, and an escaping + -- error here would defeat the containment above. + local shown, rendered = pcall(tostring, claimed) + if not shown or type(rendered) ~= "string" then + rendered = "" + end + pcall(ed.set_status, + "typed-edit consumer '" .. c.name .. "' failed: " .. rendered) elseif claimed then return end diff --git a/docs/active-work.md b/docs/active-work.md index bd0c0b1..ba1947a 100644 --- a/docs/active-work.md +++ b/docs/active-work.md @@ -149,9 +149,9 @@ If it does not, stop and repair the remote/fetch configuration. ### Stage 4a — the typed-edit consumer chain (IMPLEMENTED, same branch) - Footprint exactly as Q#LN10 declares it: `builtin/runtime/typed_edit.lua` - (new, 112 lines), `pair.lua` re-expressed as one consumer, + (new), `pair.lua` re-expressed as one consumer, `src/editor.rs` +15 (the `include_str!` and its ordering comment), and - `tests/typed_edit_chain_acceptance.rs` (new, 9 tests). + `tests/typed_edit_chain_acceptance.rs` (new, 13 tests). **`tests/auto_pair_acceptance.rs` is UNCHANGED — `git diff --stat main...HEAD -- tests/auto_pair_acceptance.rs` is empty.** That is criterion 46 checked at the diff, which is the only way it means @@ -166,9 +166,26 @@ If it does not, stop and repair the remote/fetch configuration. - **Ordered insertion, not `table.sort`** — Lua's sort is not stable, and "ties broken by registration order" is a stated contract. - **The chain `pcall`s each consumer** and reports through - `set_status`. `buffer.after-edit` is all-must-succeed, so an - uncontained throw fails the fan-out for every other subscriber - including lsp.lua's didChange flush. + `set_status`. Rev 7 justified this by claiming an uncontained throw + would fail the fan-out for every other subscriber including lsp.lua's + didChange flush; **that is wrong** — `run_all_must_succeed` + (`src/hook.rs:332`) collects errors and continues, so the other + subscribers still run. The real consequence is narrower and still + worth containing: the throw skips every LATER consumer in the chain. + The rendering is protected too, because a Lua error may be a table + whose `__tostring` throws. +- **Round 8 (review) findings, all fixed on this branch:** each consumer + now gets its **own shallow copy** of the record (the same table let a + declining consumer rewrite `rec.char`, which pairing reads — typing + `x` could produce `x)`); the fan-out iterates a **snapshot** (a + consumer registering a lower-priority one shifted itself forward under + `ipairs` and ran twice, unbounded if repeated); `tostring` moved + inside the containment; **non-finite and non-integer priorities are + rejected** (NaN is a number and every ordered comparison with it is + false, so it landed wherever the insertion scan gave up and silently + voided the ordering contract); and `add_consumer` now returns a handle + with `remove_consumer` beside it, so re-evaluating a config no longer + leaks callbacks the way `pmacs.hook.add` does (COHERENCE §13). - **Every acceptance test is bite-verified by mutation**, per the standing rule that a test is not evidence until the mutation it targets has been shown to fail it: @@ -182,6 +199,11 @@ If it does not, stop and repair the remote/fetch configuration. | drop the `pcall` | 1 chain | | skip consumers when `rec == nil` | 1 chain + **3 auto-pair** | | load `typed_edit.lua` after `lsp.lua` | 1 chain + **2 auto-pair** (Q#AP7) | + | hand every consumer the same record table | 1 chain (46f) | + | iterate the live array instead of a snapshot | 1 chain (46g) | + | render the error outside the `pcall` | 1 chain (46d) | + | accept any Lua number as a priority | 1 chain (46h) | + | make `remove_consumer` a no-op | 2 chain (46g, 46h) | The first attempt at the last bite was WORTHLESS as written: moving only `typed_edit.lua` past `lsp.lua` left `pair.lua` calling a nil @@ -194,9 +216,12 @@ If it does not, stop and repair the remote/fetch configuration. - Verification on this branch (commit-then-gate, so this describes the pushed tree): `cargo fmt --check` clean; strict workspace Clippy clean; 1,832 default + 2,009 CRDT library tests; auto-pair 45/45; - typed-edit chain 9/9; M4 121; required GPU 202; **isolated-config - workspace sweep 3,328 across 97 suites, zero failures** with - `grep -c basedpyright` = 0; `git diff --check` clean. + typed-edit chain 13/13 (and 13/13 again under `--no-default-features + --features lua54`, since the fixes touch `math.huge`, `%`, and + `__tostring` behavior that differs between the backends); M4 121; + required GPU 202; **isolated-config workspace sweep 3,332 across 97 + suites, zero failures** with `grep -c basedpyright` = 0; `git diff + --check` clean. - Stage 4b (the input method) is NOT in this PR and not started. ## Dired lane — Stage 0 MERGED; Stage 1 IN REVIEW (PR #165) diff --git a/docs/agent-handoff.md b/docs/agent-handoff.md index a176505..66ccc25 100644 --- a/docs/agent-handoff.md +++ b/docs/agent-handoff.md @@ -88,10 +88,26 @@ commands, read `docs/active-work.md` immediately after this file. config swap invalidates. The durable lesson is to heal at **consumption** — the point where a stale record is handed out — not at the moment of the swap. - - Remaining: Stage 4a (typed-edit consumer chain) and 4b (the Unicode - input method) are framed and awaiting approval; stages 5 (goal - panel), 6 (`#eval` output channel), and 7 (module hierarchy) are - framed but not scouted against current `main`. + - **Stage 4a (typed-edit consumer chain) is implemented and in review + as PR #179** (branch `lean4-stage4a-typed-edit-chain`, framing rev + 8). It is substrate only: `builtin/runtime/typed_edit.lua` owns the + single `buffer.after-edit` subscriber and the single one-shot read, + `pair.lua` becomes its first registered consumer, and + `tests/auto_pair_acceptance.rs` is unchanged by zero lines + (criterion 46, verified at the diff). No protocol change, no Lean + content. The three decisions that turned out load-bearing rather + than stylistic: consumers are called **even when the record is + nil** (three existing auto-pair tests assert the non-event through + it, and 4b abandons stale pending state on it); each consumer gets + its **own copy** of the record, because pairing reads `rec.char` + and a declining consumer could otherwise forge it; and the fan-out + iterates a **snapshot**, because a consumer that registers a + lower-priority one shifts itself forward under `ipairs` and runs + twice. + - Remaining: Stage 4b (the Unicode input method) is framed and + awaiting approval — not started; stages 5 (goal panel), 6 (`#eval` + output channel), and 7 (module hierarchy) are framed but not + scouted against current `main`. - **Inline math LANDED — #158** (`docs/inline-math-slice-framing.md` rev 3; merge `5aa9044`). pmacs renders `$…$` as typeset mathematics in the GPU diff --git a/docs/lean4-mode-framing.md b/docs/lean4-mode-framing.md index 650b2f3..789e419 100644 --- a/docs/lean4-mode-framing.md +++ b/docs/lean4-mode-framing.md @@ -1496,7 +1496,7 @@ claim a reader must be able to check without reconstructing | `builtin/runtime/typed_edit.lua` | new — the chain owner | | `builtin/runtime/pair.lua` | re-expressed as one registered consumer | | `src/editor.rs` | one `include_str!` line, before `pair.lua`'s | -| `tests/typed_edit_chain_acceptance.rs` | new — criteria 46a–46e | +| `tests/typed_edit_chain_acceptance.rs` | new — criteria 46a–46h | | `tests/auto_pair_acceptance.rs` | **unchanged, zero lines** | Rev 6 listed only the first three and then required criteria 46a–46e, @@ -2374,7 +2374,7 @@ substrate pin, filed under Stage 4 only because Stage 4 was one stage. Per the no-renumbering rule above, round 5's additions take letter suffixes on both sides of the split. -46a–46e live in a **new `tests/typed_edit_chain_acceptance.rs`**, which +46a–46h live in a **new `tests/typed_edit_chain_acceptance.rs`**, which is part of Stage 4a's declared footprint (Q#LN10) and a required gate for its PR. They cannot live in `tests/auto_pair_acceptance.rs`, which criterion 46 requires to stay byte-identical. @@ -2394,14 +2394,44 @@ criterion 46 requires to stay byte-identical. `include_str!` order happens to agree with intent. 46c. A claiming consumer stops the chain — a later consumer does not run — and a non-claiming one does not. -46d. A consumer that throws is contained: the fan-out still succeeds, - the other consumers still run, and the failure reports through - `set_status`. Bites against the `all-must-succeed` contract taking - the whole fan-out down with one bad consumer (Q#LN10). +46d. A consumer that throws is contained: the later consumers still + run, and the failure reports through `set_status`. Bites against a + chain where one bad consumer silently disables every consumer + behind it. (Round 8 correction: an uncontained throw would *not* + take the fan-out's other subscribers down — `run_all_must_succeed` + in `src/hook.rs` collects errors and continues, so `lsp.lua` still + flushes. Rev 7 claimed otherwise. The containment is still + required; the reason is narrower than stated.) Rendering the error + is itself protected: a Lua error may be any value, including a + table whose `__tostring` throws, and reporting outside the + containment reintroduces the escape it exists to prevent. 46e. **Q#AP7 ordering survives.** The existing `sighelp` fake-server test — pairing's closer must be in the buffer before `lsp.lua` flushes `didChange` — still holds with pairing behind the chain. Falsified by moving the chain's registration after `lsp.lua`'s. +46f. **Each consumer's record is its own.** A declining consumer that + mutates the record it was handed cannot change what a later + consumer sees. Bites against handing every consumer the same + mutable table: pairing decides what to close from `rec.char`, so a + forged `char` makes it insert a pair the user never typed. Every + field is a scalar or an opaque id, so a shallow copy is a complete + snapshot. +46g. **The fan-out iterates a snapshot.** A consumer may register or + remove consumers while the chain runs; both take effect on the next + fan-out. Bites against iterating the live array, where a consumer + that registers a lower-priority one shifts itself forward under + `ipairs` and runs twice — unbounded if it re-registers each time. +46h. **The registrar has a lifecycle.** `add_consumer` returns an + opaque handle; `remove_consumer` unregisters it and reports whether + it was live, so a double-remove is a no-op rather than a throw. + Without it, re-evaluating a config or reloading a package + accumulates callbacks permanently — the leak `COHERENCE.md` §13 + already records against `pmacs.hook.add`, which a teardown-less + chain would inherit and spread to every consumer. Priority is + validated as a **finite integer in i32 range**, matching + `pmacs.completion.register`: NaN is a number and every ordered + comparison with it is false, so a bare type check lets it land + wherever the insertion scan gives up and silently voids 46b. **Stage 4b — the Unicode input method** diff --git a/tests/typed_edit_chain_acceptance.rs b/tests/typed_edit_chain_acceptance.rs index c0170f4..8ff9c13 100644 --- a/tests/typed_edit_chain_acceptance.rs +++ b/tests/typed_edit_chain_acceptance.rs @@ -1,11 +1,13 @@ //! Typed-edit consumer chain acceptance (Arc 8 Stage 4a, -//! docs/lean4-mode-framing.md Q#LN10, criteria 46a–46e). +//! docs/lean4-mode-framing.md Q#LN10, criteria 46a–46h). //! //! The chain owns the single `buffer.after-edit` subscriber that reads //! the one-shot typed-edit record (Q#AP9) and offers it to consumers in //! priority order. These tests pin the chain's OWN behavior — take-once, -//! priority ordering, claim-stops-chain, throw containment, and the -//! Q#AP7 flush ordering it inherited from `pair.lua`. +//! priority ordering, claim-stops-chain, throw containment, per-consumer +//! record isolation, snapshot iteration under re-entrant registration, +//! the registration lifecycle, and the Q#AP7 flush ordering it inherited +//! from `pair.lua`. //! //! They deliberately do not re-test auto-pairing: criterion 46 requires //! `tests/auto_pair_acceptance.rs` to pass byte-identical, and that @@ -324,9 +326,12 @@ fn a_throwing_consumer_is_contained_reported_and_does_not_stop_the_chain() { "#, ); - // `buffer.after-edit` is all-must-succeed: an uncontained throw - // would fail the fan-out for every other subscriber, including - // lsp.lua's didChange flush. + // An uncontained throw would abandon every LATER consumer in the + // chain and mark the whole `buffer.after-edit` run failed. It would + // NOT stop the hook's other subscribers — all-must-succeed collects + // errors and keeps going (`src/hook.rs`'s `run_all_must_succeed`) — + // so what this pins is that one broken consumer cannot silently + // disable the ones behind it. type_str(&mut s, "("); let later_ran: bool = eval(&s, "return _G.later_ran"); @@ -357,12 +362,43 @@ fn add_consumer_rejects_malformed_registrations() { ), ( "pmacs.typed_edit.add_consumer{ name = \"n\", fn = function() end }", - "priority must be a number", + "priority must be a finite integer", ), ( "pmacs.typed_edit.add_consumer{ name = \"n\", priority = 1 }", "fn must be a function", ), + // NaN is a number and every ordered comparison with it is + // false, so a bare type check lets it land wherever the + // insertion scan gives up — and the lowest-first contract the + // Lean expander depends on quietly stops holding. The + // infinities and non-integers go with it: priority matches + // `pmacs.completion.register`'s i32. + ( + "pmacs.typed_edit.add_consumer{ name = \"n\", priority = 0/0, \ + fn = function() end }", + "priority must be a finite integer", + ), + ( + "pmacs.typed_edit.add_consumer{ name = \"n\", priority = math.huge, \ + fn = function() end }", + "priority must be a finite integer", + ), + ( + "pmacs.typed_edit.add_consumer{ name = \"n\", priority = -math.huge, \ + fn = function() end }", + "priority must be a finite integer", + ), + ( + "pmacs.typed_edit.add_consumer{ name = \"n\", priority = 1.5, \ + fn = function() end }", + "priority must be a finite integer", + ), + ( + "pmacs.typed_edit.add_consumer{ name = \"n\", priority = 4e9, \ + fn = function() end }", + "priority must be a finite integer", + ), ] { let err = s .lua_host @@ -378,6 +414,188 @@ fn add_consumer_rejects_malformed_registrations() { } } +#[test] +fn an_error_whose_rendering_throws_is_still_contained() { + // A Lua error may be any value, including a table whose + // `__tostring` throws. Rendering it outside the containment is a + // second, uncontained throw — the chain would stop at exactly the + // consumer it was trying to report. + let mut s = editor_with(""); + exec( + &s, + r#" + _G.later_ran = false + local hostile = setmetatable({}, { + __tostring = function() error("rendering exploded") end, + }) + pmacs.typed_edit.add_consumer { + name = "boom", priority = 1, fn = function() error(hostile) end, + } + pmacs.typed_edit.add_consumer { + name = "later", priority = 2, + fn = function() _G.later_ran = true; return false end, + } + "#, + ); + + type_str(&mut s, "("); + + let later_ran: bool = eval(&s, "return _G.later_ran"); + assert!( + later_ran, + "an unrenderable error must not escape the containment" + ); + assert_eq!(buffer_text(&s), "()", "and pairing still ran"); + let st = status(&s); + assert!( + st.contains("boom") && st.contains(""), + "the consumer is still named, with a placeholder body, got {st:?}" + ); +} + +// --------------------------------------------------------------------------- +// The record a consumer sees is its own +// --------------------------------------------------------------------------- + +#[test] +fn a_consumers_mutation_of_the_record_cannot_reach_the_next_consumer() { + // The record is plain Lua data. Handing every consumer the same + // table lets a DECLINING consumer rewrite provenance for the ones + // behind it — and pairing decides what to close from `rec.char`, + // so a forged `char` makes it insert a pair the user never typed. + let mut s = editor_with(""); + exec( + &s, + r#" + _G.downstream_char = "unset" + pmacs.typed_edit.add_consumer { + name = "vandal", priority = 1, + fn = function(rec) + if rec then rec.char = "("; rec.codepoint = 40 end + return false + end, + } + pmacs.typed_edit.add_consumer { + name = "witness", priority = 2, + fn = function(rec) + _G.downstream_char = rec and rec.char or "nil" + return false + end, + } + "#, + ); + + type_str(&mut s, "x"); + + let downstream: String = eval(&s, "return _G.downstream_char"); + assert_eq!( + downstream, "x", + "the next consumer sees the real typed character" + ); + assert_eq!( + buffer_text(&s), + "x", + "and pairing, reading the same field, did not close a forged opener" + ); +} + +// --------------------------------------------------------------------------- +// Re-entrant registration, and the consumer lifecycle +// --------------------------------------------------------------------------- + +#[test] +fn registering_or_removing_during_a_fan_out_takes_effect_on_the_next_one() { + // The fan-out iterates a snapshot. Iterating the live array instead + // lets a consumer that registers a LOWER-priority one shift itself + // forward under `ipairs` and run twice in a single fan-out — and + // repeating the registration makes that unbounded. + let mut s = editor_with(""); + exec( + &s, + r#" + _G.order = {} + local function mark(tag) + return function() _G.order[#_G.order + 1] = tag; return false end + end + _G.doomed = pmacs.typed_edit.add_consumer { + name = "doomed", priority = 50, fn = mark("doomed"), + } + _G.did_register = false + pmacs.typed_edit.add_consumer { + name = "a", priority = 10, + fn = function() + _G.order[#_G.order + 1] = "a" + if not _G.did_register then + _G.did_register = true + pmacs.typed_edit.add_consumer { name = "b", priority = 5, fn = mark("b") } + pmacs.typed_edit.remove_consumer(_G.doomed) + end + return false + end, + } + "#, + ); + + type_str(&mut s, "x"); + let first: String = eval(&s, "return table.concat(_G.order, ',')"); + assert_eq!( + first, "a,doomed", + "`a` runs once even though it registered ahead of itself, and \ + `doomed` still runs in the fan-out it was removed during" + ); + + exec(&s, "_G.order = {}"); + type_str(&mut s, "y"); + let second: String = eval(&s, "return table.concat(_G.order, ',')"); + assert_eq!( + second, "b,a", + "both the registration and the removal land on the next fan-out" + ); +} + +#[test] +fn remove_consumer_unregisters_and_reports_whether_it_was_live() { + // Without removal, re-evaluating a config or reloading a package + // accumulates callbacks permanently — the leak COHERENCE.md §13 + // already records against `pmacs.hook.add`. A chain with no + // teardown would inherit it and spread it to every consumer. + let mut s = editor_with(""); + exec( + &s, + r#" + _G.runs = 0 + _G.h = pmacs.typed_edit.add_consumer { + name = "temporary", priority = 1, + fn = function() _G.runs = _G.runs + 1; return false end, + } + "#, + ); + + type_str(&mut s, "x"); + let runs: i64 = eval(&s, "return _G.runs"); + assert_eq!(runs, 1, "registered consumers run"); + + let first_removal: bool = eval(&s, "return pmacs.typed_edit.remove_consumer(_G.h)"); + let second_removal: bool = eval(&s, "return pmacs.typed_edit.remove_consumer(_G.h)"); + assert!(first_removal, "removing a live consumer reports true"); + assert!( + !second_removal, + "a double-remove is a reportable no-op, not a throw" + ); + + type_str(&mut s, "y"); + let runs: i64 = eval(&s, "return _G.runs"); + assert_eq!(runs, 1, "the removed consumer no longer runs"); + // Removal is surgical: the chain itself, and pairing on it, survive. + exec(&s, "pmacs.editor.goto_byte(pmacs.window.buffer():len())"); + type_str(&mut s, "("); + assert_eq!( + buffer_text(&s), + "xy()", + "the rest of the chain is untouched" + ); +} + // --------------------------------------------------------------------------- // 46e — the Q#AP7 flush ordering the chain inherited // --------------------------------------------------------------------------- From ea9b8c379e97fdfb84e86d37b21a3de8bbb7c1f9 Mon Sep 17 00:00:00 2001 From: Levi Neuwirth Date: Sun, 26 Jul 2026 13:43:30 -0400 Subject: [PATCH 32/32] docs(lean4): correct Q#LN10's throw-containment rationale MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Q#LN10 still said a throwing consumer "fails the fan-out for everyone." It does not: `run_all_must_succeed` (src/hook.rs:332) collects the error and continues to the hook's remaining subscribers, so `lsp.lua` still flushes didChange. The throw stops every LATER consumer in the chain, which is a narrower consequence and still worth containing — the failure is silent exactly where the abandoned consumers registered. The module comment, criterion 46d, the test, and the ledger were all corrected in the previous commit; Q#LN10 is the decision they descend from, so leaving it stale would have made the disproven claim the authoritative one. Also records the protected-rendering rule there. Co-Authored-By: Claude Opus 5 (1M context) Claude-Session: https://claude.ai/code/session_011LFvC4FQtux4y32KuevZ7B --- docs/lean4-mode-framing.md | 27 ++++++++++++++++++++++----- 1 file changed, 22 insertions(+), 5 deletions(-) diff --git a/docs/lean4-mode-framing.md b/docs/lean4-mode-framing.md index 789e419..c4a32e5 100644 --- a/docs/lean4-mode-framing.md +++ b/docs/lean4-mode-framing.md @@ -1523,11 +1523,28 @@ sees the exact record via `_capture_records`, and that the Q#AP7 ordering against `lsp.lua`'s `didChange` flush still holds. **What 4a deliberately does not do.** It does not change the `all-must- -succeed` contract, so a consumer that throws still fails the fan-out for -everyone. The chain owner therefore `pcall`s each consumer and reports -through `pmacs.editor.set_status`, matching `pair.lua`'s existing -never-throw-from-after-edit discipline — this is behavior-preserving for -pairing (which already never throws) and is the guardrail 4b needs. +succeed` contract. What that contract actually does on a throw was +stated wrongly through rev 7 and is corrected here, because this +paragraph is the authority the module comment, criterion 46d, the test, +and the ledger all descend from: `run_all_must_succeed` +(`src/hook.rs:332`) **collects** the error and continues to the hook's +remaining subscribers, marking only the run as failed. An uncontained +throw inside the chain therefore does **not** stop `lsp.lua` from +flushing `didChange`. What it does stop is every LATER consumer in the +chain — the chain is one subscriber, and a throw abandons the rest of +its loop. + +That is a narrower consequence than rev 7 claimed and still worth +containing, because the failure is silent in the direction that matters: +a consumer that throws disables the consumers behind it with no signal +at the seam where they were registered. The chain owner therefore +`pcall`s each consumer and reports through `pmacs.editor.set_status`, +matching `pair.lua`'s existing never-throw-from-after-edit discipline — +this is behavior-preserving for pairing (which already never throws) and +is the guardrail 4b needs. The **rendering** of the caught error is +protected the same way: a Lua error may be any value, including a table +whose `__tostring` throws, so `tostring` outside the `pcall` would +reintroduce the escape the containment exists to prevent. ### Q#LN11 — Stage 4b data: vendor the table, generated, attributed