5.9 KiB
pmacs roadmap — July 2026
Date: 2026-07-07. Produced from a five-way codebase/docs/memory sweep (core editing + persistence, LSP surface, GPU parity, extensibility + terminal, deferred-work inventory across all framing docs).
Decision (2026-07-07): push Arc 1 (LSP utility surface), with Arc 2 (editing table stakes) items interleaved between sub-arcs.
State assessment (as of main @ ccb0ff6, protocol v14)
Strong. CRDT multi-frontend core; both frontends at near input/render parity (GPU exceeds TUI: minimap, wavy squiggles); LSP data layer — 13 language ids / ~8 servers, rename + cross-file definition + formatting fully working, semantic tokens + inlay hints rendering; package manager (git installs, lockfile, resolver, SHA-256 hardening); MCP client; async worker + PTY subprocess substrate; bundled line-oriented REPL package; headless GPU render CI; help system; buffer-list UI; atomic save; persisted minibuffer history.
Dark matter — built but unwired (highest leverage).
- Complete completion framework (
src/completion_framework.rs: providers for lsp/snippets/project-symbols/dabbrev; store;CompletionViewpopup insrc/completion.rs) — no keybinding, no typed-char trigger, popup never instantiated, no GPU message. HoverView(src/hover.rs),SignatureView(src/signature.rs), document-highlight store — plumbing only. Hover / signature / references / document-symbols surface as one-line modeline summaries (builtin/runtime/lsp.luamarks each "future UX work").- Code actions apply the first action blindly — no picker.
- Semantic-token pull is refresh/manual-only (no pull on attach/edit) — semantic styling can silently never appear. Arguably a bug.
- Wire-declared but unproduced protocol families:
FoldState,BlockAdornments,ResourceOffer(no fold engine / blame / diff source).
Absent. Query-replace, kill ring (single slot only), keyboard
macros, rectangles, registers, comment/uncomment, auto-indent,
snippets, auto-pairing; desktop-save/session restore, recentf,
saveplace, autosave, backups, crash recovery; theming beyond syntax
captures (all chrome hardcoded per-frontend; GPU font hardcoded);
terminal emulation (src/ansi.rs parses but discards alt-screen /
cursor addressing — no grid/scrollback); compile/grep/shell-command
modes; DAP debugging; GPU splits / multi-buffer / auto-reconnect.
Arcs, ranked by value-per-effort
Arc 1 — LSP utility surface: "light up the dark matter" ← ACTIVE
Data layer is done; only UI is missing.
- 1a. In-buffer completion popup (first). Trigger on typing, TAB/RET accept, both frontends. GPU needs a wire message — mirror the minibuffer-dropdown pattern (protocol v12). Framework + popup view already exist.
- 1b. Panels: hover popup, code-action picker, references list,
document-symbol outline. Generalize the buffer-list UI pattern
(
*buffer-list*buffer-local bindings) into a reusable list-buffer idiom. - 1c. Semantic-token auto-pull fix (small): pull on attach + on edit-flush, like inlay hints already do.
- 1d. Signature-help auto-trigger on
(.
Arc 2 — Editing table stakes (interleave with Arc 1)
Each small, core-only, frontend-agnostic: query-replace (isearch
exists; search.rs has no replace API), real kill ring + M-y,
comment/uncomment, auto-indent on newline, auto-pairing.
Arc 3 — Persistence/serialization
Desktop-save (buffer set + layout + cursors → restore), recentf,
saveplace, autosave + crash recovery, optional backups. Generalize the
$XDG_STATE_HOME/pmacs/ pattern from minibuffer history. Framing
question: what is a "session" in a daemon world; do CRDT snapshots
ride along.
Arc 4 — Themes + extensibility surface
Extend pmacs.theme from syntax captures to named UI faces (modeline,
minibuffer, gutter, selection, status band); wire GPU chrome to it —
Q#UX1 lesson applies: rendering is frontend-local but control is
daemon-owned, so a wire channel (ThemeFacts-style) is needed. Add
pmacs.gpu.set_font (designed, never built) and a Lua
statusline-segment API.
Arc 5 — Terminal, staged
- Stage 1: compile-mode / grep-mode / shell-command on the existing
PTY + ANSI + REPL-package substrate (line-oriented output buffer,
error-regex jump-to-file,
M-x compile). Cheap, transformative. - Stage 2 (vterm): extend
ansi.rsinto a 2D grid model (alt-screen, cursor addressing, scrollback — parser already recognizes and discards these), grid-backed buffer view, GPU rendering question (grid cells vs text buffer).
Arc 6 — Folding (keystone gutter rider)
Fold engine (tree-sitter fold ranges) unblocks gutter fold markers +
the unproduced FoldState wire family + a visible feature. Git gutter
markers similarly just need a diff source.
Arc 7 — Debugging (DAP)
Greenfield but newly unblocked: gutter (breakpoint signs), process substrate (DAP = JSON-RPC over stdio, same shape as the LSP client), list-buffer panels from Arc 1b (stack/variables). Frame after Arc 1 ships — panels are reused here.
Arc 8 — GPU structural parity
Splits / multi-buffer (largest unscoped design problem), auto-reconnect after daemon restart, cursor blink, font/theme config (overlaps Arc 4).
Housekeeping (do opportunistically)
- Merge PR #91 (gutter click classification + fit guard).
- Delete or mark-superseded stale docs:
pmacs-gpu-editing-perf-handoff.md(freeze fixed by PR #60 arc),session-5-stale-styling-handover.md(fixed, task #25 closed),python_experiment.md,#run.sh#,semantic-frontend-protocol.md.local-bak. V0.2-PREREQUISITES.mdcited twice by CHANGELOG but missing on disk — reconstruct or unlink.- README still leads with "What v0.1 ships with" — refresh.
- F-016 lua_bindings split: ~5–8 tranches left;
editor.rs(7k lines) andpmacs-gpu/main.rs(7.6k lines) splits are named follow-up arcs. - Deferred backlogs live in each framing doc's Deferred section; the consolidated sweep is reflected above.