feat(zoom): GUI zoom over the font preference that already existed

Ctrl +/- had no effect whatsoever in the GPU frontend. Stage 1 (#219)
fixed what zoom did to the TUI; this is the other half.

NO RENDERING WORK. FontMetrics::scale already derived every GUI
dimension — code size, line height, status band, divider, menu rows,
minibuffer dropdown, gutter advance — and apply_font_facts already
re-metriced all seven buffers in one transaction. This drives the
preference that existed: two settings, three commands, and a restore.

Q#Z1 = (c). Relative zoom needs an origin and the daemon is built never
to know one — font_pref.rs is explicit that it "never learns metrics,
advances, or what resolves". Hardcoding 16.0 would put a pixel constant
on the daemon side; always sending a size would destroy the `None`
state for everyone who never zooms. A configured base is the only
option where the daemon still infers nothing, and the untouched path
stays byte-identical.

THREE THINGS REVIEW CAUGHT THAT REVISION 1 HAD WRONG.

Q#Z3 was not implementable as framed. `keymap_stack::Scope` is
Buffer | Mode | Global and carries no frontend identity, so "bind on
GPU frontends only" does not exist; and FrontendEvent has no
command-invocation variant, so the GPU cannot ask for a command by name
either. A global binding would capture the chord in the TUI and take
away the terminal's own zoom — the very thing the user is pressing it
for. Commands ship; the binding waits on capability-aware keymap
resolution, which is now a named follow-on rather than something
smuggled in here.

The restore seam did not exist. Builtins and init.lua both run BEFORE
install_state_dirs, so a pmacs.state.read at module load returns
nothing, always. saveplace and recentf never meet this because both
read lazily inside functions; zoom must apply with no user action,
which makes it this project's first eager state consumer. Restore lives
at the end of install_state_dirs — by definition the moment state
becomes readable, so it cannot be ordered wrongly and a future third
startup path gets it without knowing it had to ask.

Every size write clobbered the family. set_font replaces both fields
unconditionally, so { size = n } alone silently cleared a configured
family until restart.

BITTEN, THREE WAYS. Dropping family preservation fails 3 tests.
Reverting to the framing's own first parser `^(%d+)$` fails 4 including
the seam restore — it anchors to end-of-subject and rejects the
newline-terminated file the writer emits, which is the contradiction
review caught in the framing before it reached code. Hardcoding the
16.0 origin fails the base test.

Also recorded: a loaded crdt run failed two m6_1 PTY tests with
`stty -a output was: ""`. That is R4/R6's empty-content readiness
family, and it means the readiness-helper audit's scope is wider than
three wait_for_file copies under tests/ — src/process.rs's own tests
carry the shape. Undiagnosed, load-sensitive, green isolated and on a
quiet full run; a scope note for that lane, not a registry row, since
the registry judges red CI runs and these were local.

Verified: fmt, clippy, diff-check, --lib 1900/0, crdt 2085/0,
gui_zoom 13/13, journey 47/0, m4 150/0, gpu 221/0, full_grid 1/1.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
Levi Neuwirth 2026-08-06 16:20:46 +02:00
parent 218d2e7acd
commit aa99ab39d2
No known key found for this signature in database
5 changed files with 1136 additions and 0 deletions

190
builtin/runtime/zoom.lua Normal file
View File

@ -0,0 +1,190 @@
-- GUI zoom (QoL Stage 2, framing docs/gui-zoom-framing.md).
--
-- Drives the font preference that already exists: `pmacs.gpu.set_font`
-- writes it, `semantic_render` relays it as `InstanceMessage::FontFacts`
-- at protocol v17, and the GPU frontend owns every pixel consequence.
-- Nothing here knows a metric, an advance, or what resolves --- the
-- no-pixels invariant (src/font_pref.rs) holds through this module.
--
-- NO KEYBINDINGS, deliberately (Q#Z3). `keymap_stack::Scope` is
-- Buffer | Mode | Global and carries no frontend identity, so "bind
-- this on GPU frontends only" is not expressible; and `FrontendEvent`
-- has no command-invocation variant, so the GPU cannot ask for a
-- command by name either. A global binding would capture C-+/C-- in
-- the TUI and take away the terminal's own zoom --- the very thing the
-- user is pressing the key for. Commands are discoverable via M-x and
-- one line to bind in init.lua; capability-aware binding is its own
-- lane.
pmacs.zoom = pmacs.zoom or {}
-- Wire bounds, in logical px: `FONT_SIZE_CENTI_PX_RANGE` is 600..=7200
-- (pmacs-gpu/src/main.rs). Mirrored rather than imported because the
-- Lua range check is a UX courtesy; the frontend re-checks on arrival
-- because that side is deserialized protocol input.
local MIN_PX = 6.0
local MAX_PX = 72.0
local STATE_KEY = "gpu-zoom"
pmacs.config.define {
name = "ui.gpu-font-size-base",
description = "Logical-pixel size the first zoom step starts from when no font size is set.",
type = "number",
default = 16.0,
min = MIN_PX,
max = MAX_PX,
mutability = "live",
}
-- Lower bound is the QUANTIZER, not tidiness: `validate_font_size`
-- rounds to the nearest hundredth, so a step below 0.01 quantizes to
-- zero and "zoom in" silently does nothing forever. A negative step
-- would invert the commands --- zoom-in shrinking is not a malfunction
-- the user can diagnose, because the command still does something
-- coherent. Upper bound is the range span (72 - 6): a larger step can
-- only ever clamp.
pmacs.config.define {
name = "ui.gpu-zoom-step",
description = "Logical pixels added or removed per zoom step.",
type = "number",
default = 1.0,
min = 0.01,
max = 66.0,
mutability = "live",
}
-- Round to centi-pixel, the wire's unit. Doing this here keeps every
-- comparison and every round-trip in the quantized domain, so "n in,
-- n out" is exact addition rather than float drift.
local function quantize(px)
return math.floor(px * 100 + 0.5) / 100
end
-- The current size in logical px, or nil when the preference is unset.
-- nil is a REAL state (the frontend's own default), never inferred from
-- silence --- Q#TH7.
local function current_px()
return pmacs.gpu.font().size
end
-- Preserve the configured family on EVERY write. `set_font` replaces
-- both fields unconditionally, so `set_font { size = n }` alone would
-- clear a family the user set in init.lua, and they would get it back
-- only by restarting.
local function write_size(px)
local spec = { size = px }
local family = pmacs.gpu.font().family
if family then spec.family = family end
pmacs.gpu.set_font(spec)
end
local function save(px)
if not pmacs.state.available() then return end
pmacs.state.write(STATE_KEY, string.format("%d\n", math.floor(px * 100 + 0.5)))
end
local function forget()
if not pmacs.state.available() then return end
pmacs.state.write(STATE_KEY, "")
end
-- Step by `delta` logical px. Returns the new size, or nil plus a
-- reason.
local function step(delta)
local base = current_px() or pmacs.config.get("ui.gpu-font-size-base")
local want = quantize(base + delta)
if want < MIN_PX or want > MAX_PX then
-- Reject the WHOLE step rather than pinning to the boundary. This
-- is what keeps "n steps in, n steps out returns exactly" true at
-- the edges, which is precisely where a user steps back and forth.
-- It also mirrors `apply_font_facts`, which rejects an
-- out-of-range message outright rather than clamping it.
return nil, string.format(
"zoom: %.2f px is outside %.2f-%.2f; size unchanged", want, MIN_PX, MAX_PX)
end
write_size(want)
save(want)
return want
end
-- Named `increase`/`decrease` rather than `in`/`out`: `in` is a Lua
-- keyword, and `in_` reads like a workaround for one.
function pmacs.zoom.increase()
return step(pmacs.config.get("ui.gpu-zoom-step"))
end
function pmacs.zoom.decrease()
return step(-pmacs.config.get("ui.gpu-zoom-step"))
end
-- Reset returns the preference to NIL --- the frontend's own default ---
-- not to `ui.gpu-font-size-base`. The base is only the origin for a
-- first step; resetting to it would ship an explicit size that merely
-- happens to equal the default, making the untouched state unreachable
-- once a user has ever zoomed. Clearing the saved state too, because
-- "reset until restart" is not what the word says.
function pmacs.zoom.reset()
local family = pmacs.gpu.font().family
local spec = {}
if family then spec.family = family end
pmacs.gpu.set_font(spec)
forget()
return nil
end
-- Restore a saved zoom. Called from the Rust side AFTER
-- `install_state_dirs`, never at module load: builtins and init.lua both
-- run before state is wired up, so a read here would return nothing,
-- always. Zoom is this project's first EAGER state consumer --- saveplace
-- and recentf both read lazily inside functions and never meet this.
--
-- Whole-file parse, not a line iterator. `^(%d+)$` would reject the
-- newline-terminated file we write ourselves ($ anchors to end of
-- subject), and saveplace's `gmatch("([^\n]+)")` would accept the FIRST
-- line of a multi-line file --- fine for recentf, where a line is one
-- independent entry, wrong here, where the file IS the value.
function pmacs.zoom.restore()
if not pmacs.state.available() then return nil end
local text = pmacs.state.read(STATE_KEY)
if not text then return nil end
local centi = text:match("^(%d+)\n$")
if not centi then return nil end
centi = tonumber(centi)
-- Range-check before it can reach `set_font`. A syntactically fine
-- but out-of-range value --- hand-edited, or written by a future
-- version with a wider range --- would otherwise be rejected as a
-- whole message, leaving the user with neither the saved zoom nor an
-- explanation.
if centi < MIN_PX * 100 or centi > MAX_PX * 100 then return nil end
local px = centi / 100
write_size(px)
return px
end
pmacs.command.define {
name = "gpu.zoom-in",
description = "Increase the GPU frontend's font size by one step",
fn = function()
local px, why = pmacs.zoom.increase()
pmacs.editor.set_status(why or string.format("zoom: %.2f px", px))
end,
}
pmacs.command.define {
name = "gpu.zoom-out",
description = "Decrease the GPU frontend's font size by one step",
fn = function()
local px, why = pmacs.zoom.decrease()
pmacs.editor.set_status(why or string.format("zoom: %.2f px", px))
end,
}
pmacs.command.define {
name = "gpu.zoom-reset",
description = "Return the GPU frontend to its own default font size",
fn = function()
pmacs.zoom.reset()
pmacs.editor.set_status("zoom: reset to the frontend default")
end,
}

View File

@ -296,6 +296,81 @@ change to *when* `needs_full_grid` is set — the producer's triggers
were verified correct, along with per-frame geometry sync and were verified correct, along with per-frame geometry sync and
`view_top` reconciliation on shrink. `view_top` reconciliation on shrink.
## GUI zoom (QoL Stage 2) — IN FLIGHT
**Written with the lane's first commit, before the PR exists** — the
standing correction from #171 and #215.
- **Branch `gui-zoom`**, base `githubsucks/main` @ `218d2e7` (the #219
merge). `githubsucks/gui-zoom` is the authoritative tip.
Recover: `git fetch githubsucks && git checkout gui-zoom`.
- **Framing `docs/gui-zoom-framing.md` revision 4**, approved after
four review rounds. **Q#Z1 = (c)** configured base with `None`
preserved; **Q#Z2 = additive**; **Q#Z3 = (C)** commands only, no
default bindings; **Q#Z4** eager restore inside `install_state_dirs`.
### What it ships
`builtin/runtime/zoom.lua`: two settings (`ui.gpu-font-size-base`,
`ui.gpu-zoom-step`), three commands (`gpu.zoom-in` / `-out` /
`-reset`), and `pmacs.zoom.restore` called from `install_state_dirs`.
**No rendering work** — `FontMetrics::scale` already derived every GUI
dimension and `apply_font_facts` already re-metriced everything; this
drives the preference that existed.
### The three findings review caught, none of which was in revision 1
- **Q#Z3 was not implementable.** `keymap_stack::Scope` is
`Buffer | Mode | Global` with no frontend identity, and
`FrontendEvent` has no command-invocation variant — so neither "bind
on GPU only" nor "the GPU asks for a command" exists. Commands ship;
the binding waits on **capability-aware keymap resolution**, now a
named follow-on.
- **The restore seam did not exist.** Builtins and `init.lua` both run
*before* `install_state_dirs`, so a `pmacs.state.read` at module load
returns nothing, always. `saveplace` and `recentf` never meet this
because **both read lazily**; zoom must apply with no user action,
making it the **first eager state consumer**. Restore lives at the
end of `install_state_dirs` — by definition when state becomes
readable, so it cannot be mis-ordered or missed by a future third
startup path.
- **Every size write clobbered the family.** `set_font` replaces both
fields unconditionally, so `{ size = n }` alone silently cleared a
configured family until restart.
### Verification
13 acceptance tests, three bitten: dropping family preservation fails
3; reverting to the framing's first parser `^(%d+)$` fails 4 including
the seam restore (it anchors to end-of-subject and rejects the
newline-terminated file the writer emits); hardcoding the 16.0 origin
fails the base test.
### Not in scope
Stage 3 (long lines). Capability-aware keymap resolution — Q#Z3's
option (A), deliberately deferred rather than half-built. Per-buffer
zoom. Any change to `FontFacts` or the wire.
## Empty-content readiness, a fourth and fifth instance — FOR THE R6 AUDIT
Found 2026-08-06 while gating this lane, recorded here because it
widens an existing lane's scope rather than starting one.
A loaded `--features crdt` run failed `m6_1_pty_raw_mode_disables_kernel_echo`
and `m6_1_pty_canonical_mode_keeps_kernel_echo` with
**`stty -a output was: ""`** — read-before-write on the child's output.
That is the **same family as R4** (readiness predicate satisfied by an
empty file) and **R6** (readiness file never published), and it means
the readiness-helper audit's scope is not just three `wait_for_file`
copies under `tests/`: `src/process.rs`'s own tests carry the shape
too.
Both passed isolated and the full suite was green on a quiet machine,
so this is load-sensitive and **undiagnosed** — recorded as a scope
note for the audit, not as a registry row: these were local, and the
registry judges red **CI** runs.
## Tree primitive (P5) — MERGED as #217; adoption is the open work ## Tree primitive (P5) — MERGED as #217; adoption is the open work
**The lane is gone, not the work.** Rule 4 removes a lane after merge, **The lane is gone, not the work.** Rule 4 removes a lane after merge,

440
docs/gui-zoom-framing.md Normal file
View File

@ -0,0 +1,440 @@
# GUI zoom — QoL Stage 2
**Status: revision 4 — proposed, awaiting approval.** Q#Z1 approved as
**(c)** (configured base, `None` preserved) and Q#Z2 as **additive**.
Revision 4 fixes a self-contradiction in §5a1: the specified parser
`^(%d+)$` rejects the newline-terminated format specified beside it.
Whole-file `^(%d+)\n$`, with a verified case table and a valid-restore
witness. Revision 3 added the contract details a second pass asked for: §1 and §7
no longer claim a keybinding this stage does not ship, §3.1 bounds and
quantizes both settings, and §5a1 defines the persisted format and what
corrupted state does. Revision 2 answered four review findings that revision 1 got wrong or
left out: Z3 was **not implementable as written** (§4), Z4 had **no
working restore seam** (§5), every size write **silently clears a
configured family** (§5a), and reset/precedence semantics were
unspecified (§5b). Reported from daily-driver use
alongside Stage 1: `Ctrl +/-` has *no effect whatsoever* in the GPU
frontend. Stage 1 (#219) fixed what zoom did to the TUI; this stage is
the other half — making the GUI zoom at all.
**Most of this already exists**, which is why it is a short stage. It
is not, however, a trivial one: relative zoom needs a starting point,
and the daemon is deliberately built never to know one (§3).
---
## 1. What is already built
The whole mechanism ships today and is exercised by tests:
| piece | where | state |
|---|---|---|
| Scale that drives **everything** | `FontMetrics { scale, advance_ratio }`, `pmacs-gpu/src/main.rs:129`+ | code size, line height, status band, divider, menu rows, minibuffer dropdown, gutter advance all derive from it |
| Daemon-side preference | `src/font_pref.rs` — family + `size_centi_px` + monotonic `epoch` | written by `pmacs.gpu.set_font`, read per-frame by the producer |
| Wire relay | `InstanceMessage::FontFacts`, protocol v17 | bufferless; v16 peers get none |
| Frontend application | `apply_font_facts`, `pmacs-gpu/src/main.rs:8132` | one transaction: validate, resolve family, re-metric all seven buffers, reshape at retained scroll, re-follow caret |
| Getter | `pmacs.gpu.font()` | fresh table, `size` in logical px, key absent when unset |
| Validation | `600..=7200` centi-px (6.072.0 px), checked at both ends | Lua range-check is UX; the frontend re-checks because it is wire input |
There is even a byte-identity test that `(None, None)` reproduces the
never-set frame exactly.
**So this stage adds no rendering work.** It adds **commands**, a
starting point, and remembering — **and deliberately no default
keybindings**, per Q#Z3 (§4): the keymap cannot express "GPU frontends
only", and neither a protocol variant nor a rushed keymap change is
worth a default chord.
---
## 2. Q#Z1 — where does a *relative* zoom start? (the real question)
`src/font_pref.rs` is explicit:
> *"The daemon relays a PREFERENCE: it never learns metrics, advances,
> or what resolves (the no-pixels invariant); the frontend owns
> resolution and every pixel consequence."*
And `size_centi_px: None` is **a real state, not an absence** — "the
frontend's built-in default", which Q#TH7 established must never be
inferred from silence.
Zoom-in is `current + step`. On a fresh session `current` is `None`,
and the daemon cannot ask what the frontend resolved. Three ways out:
- **(a) Hard-code the frontend's default (16.0) as the origin.** One
line, and it puts a pixel constant on the daemon side — the exact
thing the no-pixels invariant forbids. It also silently breaks for
any frontend whose default differs, which is the situation the
invariant exists to allow.
- **(b) Always send an explicit size**, sourced from a config setting
defaulting to 16.0. Simple, but it **destroys the `None` state for
everyone**: a user who never zooms now gets an explicit size, the
frontend's own default query never runs, and Q#F5's always-shipped
all-default baseline stops being reachable.
- **(c) A configured *base*, with `None` preserved until the user
actually zooms.** `ui.gpu-font-size-base` (logical px, default 16.0)
is the documented origin for the *first* zoom step. Until a zoom
happens the preference stays `None` and the frontend's default is
used exactly as today.
**Recommendation: (c).** It is the only one where the daemon still
infers nothing — the base is a *user-facing preference about zooming*,
not the daemon guessing a frontend metric. A user whose frontend
default is not 16.0 changes one setting, which is a real answer rather
than a broken assumption. And the untouched path is byte-identical to
today, so the existing `(None, None)` identity test keeps its meaning.
**Q#Z1 — DECIDED: (c)**, on review.
---
## 3. Q#Z2 — step shape
Additive (`+1.0 px`) or multiplicative (`×1.1`)?
**DECIDED: additive**, on review — via `ui.gpu-zoom-step` (logical px,
default 1.0).
### 3.1 Both settings are validated, and the bounds are not cosmetic
The registry validates at `define` time (`min`/`max`, as
`autosave.interval-ms` does), so both settings are constrained rather
than checked at use:
| setting | bounds | why |
|---|---|---|
| `ui.gpu-font-size-base` | **6.00 72.00** logical px | exactly the wire range (`600..=7200` centi-px). A base outside it could never be sent, so the first zoom step would fail from a value the user was allowed to set |
| `ui.gpu-zoom-step` | **0.01 66.00** logical px, strictly positive | see below |
**The step's lower bound is the quantizer.** `validate_font_size`
range-checks the original value and then rounds to the nearest
hundredth, so a step below `0.01` quantizes to **zero** — "zoom in"
would do nothing, forever, with no error anywhere. `0.01` is one
centi-pixel, the smallest representable step.
**Zero and negative are excluded for a stronger reason than tidiness.**
A negative step **inverts** the commands: `gpu.zoom-in` would shrink.
That is not a malfunction the user can diagnose from the outside — the
command does something coherent, just the opposite of its name.
**The upper bound is the range span** (72.00 6.00). A step larger
than the whole domain can only ever clamp or be rejected, so permitting
it buys a setting that cannot be used.
**These bounds are what make the round-trip claim true.** "n steps in,
then n steps out, returns to exactly the starting value" holds because
the step is centi-pixel representable and addition is exact in that
domain — *provided no clamp occurred*, which is why §6 requires an
out-of-range zoom to leave the preference **unmutated** rather than
pinning it to the boundary. Pinning would silently break the round trip
at the edges, which is precisely where a user is most likely to be
stepping back and forth. The domain is integer hundredths of a pixel over
6.0..=72.0 — a small, bounded, quantized range where additive steps are
predictable and land on round numbers. Multiplicative stepping
accumulates rounding through the quantizer and makes "two in, two out"
fail to return to where you started, which is the property users
actually notice.
---
## 4. Q#Z3 — keys. Revision 1 was not implementable
Revision 1 said "do not install the binding on a non-GPU frontend."
**There is no such thing.** `keymap_stack::Scope` is exactly
`Buffer(BufferId) | Mode(String) | Global`, and neither `resolve()` nor
`keymap_tree` receives any frontend identity or capability. A global
binding is global — it would capture `C-+` / `C--` in the TUI, which is
the outcome §4 claimed to avoid.
Nor is there a way for the GPU to ask for a command by name:
`FrontendEvent` is `Key | Mouse | Resize | Paste | FocusGained |
FocusLost | Detach | CrdtOp`. There is **no command-invocation
variant**, so "the GPU intercepts the chord locally" cannot reach the
daemon-side preference that owns the value and its persistence.
So there are three real options, and none is free:
- **(A) Capability-aware keymap resolution.** Teach `Scope` — or
`resolve()`'s inputs — about the requesting frontend. Correct and
general, and it would serve every later frontend-specific binding.
But it touches the keymap core, every resolve call site, and the Lua
`bind` surface. That is its own lane, not a step inside a zoom lane.
- **(B) A new `FrontendEvent` for it.** Narrow and direct, but it is a
**protocol addition** — schema bump and negotiation — which §8
explicitly scopes out, and which buys a keybinding rather than a
capability.
- **(C) Ship commands now; bind later.** `gpu.zoom-in` / `-out` /
`-reset` exist and work from `M-x` and from `init.lua`, where a user
who runs the GPU can bind them themselves in one line. No core
change, no protocol change, no half-built capability.
**Recommendation: (C), with (A) recorded as the follow-on it implies.**
The value the report asked for is *zoom working at all*; the default
keybinding is the smaller half, and buying it with either a protocol
variant or a rushed keymap change trades a large permanent surface for
a small convenience. **(A) is the right eventual answer** — "this
binding applies to frontends with capability X" is a question this
project will keep asking — and it deserves its own framing rather than
being smuggled in here.
**What must NOT happen** is a global binding plus a command that
reports "not applicable" in the TUI. That is the Q#P3 fall-through
lesson and #217's flat-panel `TAB` finding: capturing a key to deliver
an apology removes a working behaviour — and here the working
behaviour is *the terminal's own zoom*, which is exactly what the user
is pressing the key for.
*Tempting and rejected:* "terminals swallow `C-+` anyway, so a global
binding is harmless in practice." Most do. Relying on most is how a
binding becomes a bug report from whoever runs the terminal that
doesn't.
## 5. Q#Z4 — remembering it, and the seam revision 1 did not have
The config registry has **no write-back** — settings are declared, not
saved. The project's remembered-state mechanism is `pmacs.state.read` /
`write` / `available` (`docs/persistence-framing.md`), used by
`saveplace` and `recentf`. The split is the right one:
- **`pmacs.config`** — what you *declared*: base size, step.
- **`pmacs.state`** — what you *arrived at*: the current zoom level.
### 5.1 Reading at module load is inert, and revision 1 assumed it was not
`EditorState::new()` / `open()` load the builtins and run `init.lua`.
**`install_state_dirs()` runs after that** — `src/editor.rs:3838` on
the local path, `src/daemon.rs:480` on the daemon path. A
`pmacs.state.read` at Lua module load therefore returns nothing, always.
`saveplace` and `recentf` never hit this because **both read lazily**,
inside functions called long after startup (`recentf.lua:27` in
`load_list`, `saveplace.lua:33` behind its `available()` guard). **Zoom
cannot be lazy**: its whole purpose is to apply with no user action,
before the first frame the GPU paints. It is this project's **first
eager state consumer**, which is why the seam has to be named rather
than assumed.
### 5.2 The seam
Restore must run **after `install_state_dirs()` on both startup paths**.
Neither existing post-install hook qualifies: `restore_desktop_if_armed`
is local-only by design (Q#DS9 keeps desktop restore out of the daemon,
which has a layout per attached frontend and none at construction).
**Recommendation: restore inside `install_state_dirs()` itself**, at its
end. Both paths already call it, exactly once, and it is by definition
the moment state becomes readable — so the restore cannot be ordered
wrongly and cannot be forgotten by a future third startup path. The
alternative, a new call added beside both existing call sites, is two
places a fourth path can miss.
**This must be tested against production ordering, not a direct call.**
A test that calls the restore helper itself proves nothing about when
it runs — the same shape as the `prepare_startup` note at
`src/editor.rs:3820`, where splitting the sequence was what kept a
deleted call from leaving every direct-call test green while shipping
nothing. The witness asserts that a state file written before startup
is reflected in `pmacs.gpu.font()` after the real startup sequence.
### 5a. Every size write must preserve the family
`pmacs.gpu.set_font` replaces **both** fields unconditionally
(`src/lua_bindings/mod.rs`: `pref.family = family; pref.size_centi_px =
size_centi_px;`). So `set_font { size = 18 }` **silently clears a
configured family** — a user with `family = "Iosevka"` in `init.lua`
loses it the first time they zoom, and gets it back only by restarting.
Zoom must therefore read the current family via `pmacs.gpu.font()` and
pass it back with every write. **Pinned by a custom-family → zoom →
`FontFacts` case** asserting the relayed message still carries the
family. This is a property of the setter's whole-message semantics, not
a zoom bug, and the test belongs with zoom because zoom is what makes
it reachable.
### 5a1. The persisted format, and what corrupted state does
**Format**, following `saveplace` and `recentf` rather than inventing
one: a single line of plain text holding the size in **centi-pixels**
as decimal digits, newline-terminated.
```
1800
```
Centi-pixels, not logical px, because that is the wire unit and the
already-quantized one — writing `18.0` would reintroduce a
float-parsing step and a second place for rounding to disagree with
`validate_font_size`.
No version prefix. The existing state files carry none, and a single
integer has no forward-compatibility question that a prefix would
answer; a future format change can be detected by the strict parse
below failing, which lands in exactly the same handling as corruption.
**Only the size is persisted.** Not the family — that is declared
config, and §5b's precedence rule applies to size alone.
**Validity** is two checks, both required:
1. **Strict whole-file parse:** `^(%d+)\n$` against the **entire file
contents**, not against a line pulled out of it.
Revision 2 specified `^(%d+)$`, which **contradicts the
newline-terminated format two paragraphs above it**: in Lua, `$`
anchors to end-of-subject, so `("1800\n"):match("^(%d+)$")` is
`nil` and the parser would reject every file it had itself written.
Borrowing `saveplace`'s `gmatch("([^\n]+)")` line iterator fixes
that but introduces a different hole — it happily returns the
**first** line of a multi-line file, so trailing garbage would be
accepted silently. That is tolerable for `recentf`, where each line
is an independent entry and skipping a bad one loses one path. It is
not tolerable here, where the file *is* one value and extra content
means the file is not what we wrote.
Whole-file matching closes both, verified against Lua 5.4:
| contents | `^(%d+)$` | `^(%d+)\n$` |
|---|---|---|
| `"1800\n"` — what we write | **nil** ✗ | `1800` ✓ |
| `"1800"` — no trailing newline | `1800` | nil ✓ |
| `"1800\n1900\n"` — multi-line | nil | nil ✓ |
| `"18.0\n"` — decimal | nil | nil ✓ |
| `" 1800\n"` — leading space | nil | nil ✓ |
| `"\n"` — empty | nil | nil ✓ |
The multi-line case needs **no separate line-count check**: the
trailing `$` after `\n` already requires the file to end there.
2. **Range:** the parsed value is within `600..=7200`. A syntactically
fine but out-of-range number — a hand-edited file, or a file written
by a future version with a wider range — must never reach
`set_font`, which would reject the whole message and leave the user
with neither the saved zoom nor an explanation.
**Malformed, out-of-range, empty, or multi-line state is treated as
absent.** Startup proceeds exactly as if nothing had been saved: the
`init.lua` size applies if there is one, otherwise `None`.
**Silently, and deliberately so.** It matches the precedent — both
existing consumers skip unparseable lines without comment — and the
failure is already self-evident and self-healing: the user sees their
zoom did not restore, and the next zoom rewrites the file correctly.
A startup diagnostic for a recoverable, self-announcing condition is
noise on every launch for a problem that fixes itself on the next
keystroke.
**The corrupt file is not rewritten on read.** It is left alone until
the next successful zoom overwrites it, so a user who wants to look at
what went wrong still can. Truncating it at startup would destroy the
only evidence of a bug we would then have no way to reproduce.
### 5b. Reset, clearing, and precedence
Revision 1 left all three unspecified. As recommended in review:
- **Reset returns size to `None`**, not to `ui.gpu-font-size-base`.
`None` is the real "frontend's own default" state (Q#F5/Q#TH7); the
base is only the *origin for the first step*. Resetting to the base
would send an explicit size that merely happens to equal the default,
making the untouched state unreachable once a user has ever zoomed —
and quietly breaking the `(None, None)` identity property.
- **Reset clears the saved zoom state**, so it does not resurrect on
the next launch. A reset that leaves the state file behind means
"reset until restart", which is not what the word says.
- **Valid saved state wins over an `init.lua` size.** The saved value
is a later, deliberate user action; the init value is the standing
default it was chosen against. This is the same precedence
`saveplace` already applies — a remembered position overrides where
opening the file would otherwise land.
- **The family is retained throughout**, per §5a: precedence applies to
*size only*, and an `init.lua` family is never overridden by restored
zoom state, because zoom state does not carry one.
## 6. Verification
- **No binding is installed at all** (Q#Z3 = C), so the witness is that
`C-+` / `C--` resolve to nothing in a default TUI session — the
terminal's own zoom is left alone.
- **Restore runs at the production seam.** Asserted through the real
startup sequence, not by calling the helper directly: a state file
written beforehand must be visible in `pmacs.gpu.font()` afterwards,
on **both** the local and daemon paths.
- **A zoom preserves a configured family** — custom family → zoom →
the relayed `FontFacts` still carries it (§5a).
- **Reset returns `None`, not the base**, and clears the saved state;
a subsequent launch is back to the frontend's own default.
- **Saved state beats an `init.lua` size, and never its family.**
- **Config bounds are enforced at `define`** — a zero, negative, or
sub-centi-pixel step is refused by the registry, not discovered as a
zoom that does nothing or runs backwards (§3.1).
- **A valid newline-terminated file restores.** `"1800\n"` — exactly
the bytes the writer produces — must round-trip through the real
startup sequence. This is the case revision 2's own parser would have
rejected, so it is asserted rather than assumed.
- **Malformed, out-of-range, empty, multi-line, and
missing-trailing-newline saved state each behave as absent**, and the
file is left intact for inspection (§5a1). The multi-line case is
pinned specifically: it is the one a line-iterator parser would have
accepted.
- **The round trip holds only where no clamp occurred**, so the
out-of-range case asserts the preference is left **unmutated** rather
than pinned to the boundary.
- **Zoom from the untouched state uses the configured base**, not a
hardcoded 16.0 — bitten by changing the base and asserting the first
step follows it.
- **Clamping at both ends of `6.0..=72.0`** reports rather than
silently pinning, and a zoom that would leave the range does not
mutate the preference at all (the `apply_font_facts` whole-message
rejection precedent).
- **`n` steps in then `n` steps out returns to the exact starting
value** — the property that fails under multiplicative stepping and
the reason Q#Z2 recommends additive.
- **The untouched path stays byte-identical**: with no zoom performed,
the existing `(None, None)` identity test must still hold.
- **Persistence round-trips**, and writes nothing when
`pmacs.state.available` is false.
---
## 7. Coherence impact (§20 requirement)
**Concern served: `COHERENCE.md` §16, Productize the Semantic Frontend
Architecture**, which names **frontend-specific typography** in its own
list of properties the semantic protocol should make visible as a
product advantage. A GPU frontend that cannot change its own type size
is that concern under-delivered.
**No scorecard change.** Row 16 reads **Strong** (`COHERENCE.md:112`);
this neither earns nor forfeits it — it uses the mechanism the row
already credits. Per §25 the row moves only with a PR that changes what
it asserts.
- **Journey steps touched:** none.
- **Interaction islands added:** none, and **no keybinding either**.
Revision 1 said "an ordinary global keymap binding, scoped by
frontend capability"; no such scoping exists (§4), so this stage
ships commands only, reachable from `M-x` and bindable by the user in
`init.lua`. Capability-aware binding is deferred to its own framing
and named in §8.
- **Config registry:** **yes, two settings**
`ui.gpu-font-size-base` and `ui.gpu-zoom-step`. This is the "place
for GUI settings" the report asked for; it already exists and this
stage is its next adopter.
- **Background-work attribution:** none.
---
## 8. Not in scope
Long lines (Stage 3). **Capability-aware keymap resolution** — Q#Z3's
option (A), which this stage defers rather than half-builds, and which
should be its own framing because it decides how every future
frontend-specific binding is expressed. A default keybinding for zoom,
which waits on it. Per-buffer or per-window zoom — this is one
global preference, as `font_pref` already is. Family selection, which
`set_font` already exposes and no one has asked to bind. Any change to
`FontFacts`, the protocol, or `apply_font_facts` — this stage drives
the existing mechanism and adds nothing to the wire. Config write-back,
which does not exist and which Q#Z4 deliberately routes around rather
than inventing.

View File

@ -752,6 +752,17 @@ impl EditorState {
include_str!("../builtin/runtime/help.lua"), include_str!("../builtin/runtime/help.lua"),
) )
.expect("load help builtin chunk"); .expect("load help builtin chunk");
// GUI zoom (QoL Stage 2). Defines `ui.gpu-font-size-base` /
// `ui.gpu-zoom-step` and the `gpu.zoom-*` commands, and exposes
// `pmacs.zoom.restore` for `install_state_dirs` to call once
// `pmacs.state` is readable. Deliberately binds NO keys: the
// keymap cannot express "GPU frontends only" (framing Q#Z3).
lua_host
.eval(
Some("@pmacs/builtin/runtime/zoom.lua"),
include_str!("../builtin/runtime/zoom.lua"),
)
.expect("load zoom builtin chunk");
// T M7.11 bundled-package bootstrap. Through M7.10 the REPL // T M7.11 bundled-package bootstrap. Through M7.10 the REPL
// was loaded directly via `eval(include_str!(...))`; the // was loaded directly via `eval(include_str!(...))`; the
// M7.11 deliverable migrates it to the package system so it // M7.11 deliverable migrates it to the package system so it
@ -988,6 +999,30 @@ impl EditorState {
.lua() .lua()
.set_app_data(crate::lua_bindings::StateDir(dir)); .set_app_data(crate::lua_bindings::StateDir(dir));
} }
// Restore a saved GUI zoom (QoL Stage 2, framing §5.2). HERE and
// not in a runtime module: builtins and `init.lua` both run
// during construction, BEFORE this function, so a
// `pmacs.state.read` at module load returns nothing every time.
// `saveplace` and `recentf` never meet that because both read
// lazily inside functions; zoom must apply with no user action,
// which makes it the first eager state consumer.
//
// Inside `install_state_dirs` rather than beside its two call
// sites (`prepare_startup` and the daemon) because this is by
// definition the moment state becomes readable — so it cannot be
// ordered wrongly, and a future third startup path gets it
// without knowing it had to ask.
//
// Best-effort: a failure here must not stop a session from
// starting over a font size.
if let Err(error) = self
.lua_host
.lua()
.load("if pmacs.zoom then pmacs.zoom.restore() end")
.exec()
{
eprintln!("pmacs: could not restore saved zoom: {error}");
}
} }
/// Final step of a **local, no-target** launch: greet an untouched /// Final step of a **local, no-target** launch: greet an untouched

View File

@ -0,0 +1,396 @@
//! GUI zoom acceptance (`QoL` Stage 2, `docs/gui-zoom-framing.md`).
//!
//! Zoom drives the font preference that already existed: `set_font`
//! writes it, `semantic_render` relays it as `FontFacts` at v17, and the
//! GPU frontend owns every pixel consequence. Nothing here knows a
//! metric — the no-pixels invariant holds through the whole feature.
//!
//! Each test gets its **own** bootstrap roots. `iso::roots()` is a pure
//! function of the build environment by design, so every suite sharing
//! it shares one state directory — fine when nobody writes, wrong here,
//! where the state file *is* the subject and parallel tests would
//! overwrite each other's fixture.
use std::path::{Path, PathBuf};
use pmacs::bootstrap::BootstrapRoots;
use pmacs::editor::EditorState;
/// Per-test roots plus the path zoom's state file will occupy
/// (`<state>/pmacs/gpu-zoom`, per `src/state.rs`).
fn roots_for(name: &str) -> (BootstrapRoots, PathBuf) {
let base = Path::new(env!("CARGO_TARGET_TMPDIR"))
.join("gui-zoom")
.join(name);
let _ = std::fs::remove_dir_all(&base);
let roots = BootstrapRoots::isolated_under(&base);
for (_, dir) in roots.child_env() {
std::fs::create_dir_all(&dir).expect("create controlled root");
}
let dir = roots.state_dir().expect("isolated roots have a state dir");
std::fs::create_dir_all(&dir).expect("create state dir");
(roots, dir.join("gpu-zoom"))
}
/// A session whose state dirs are installed — which is also what runs
/// `pmacs.zoom.restore`, so anything planted at `path` beforehand is
/// visible to it.
fn session(roots: &BootstrapRoots) -> EditorState {
let state = EditorState::new_with_roots(roots);
state.install_state_dirs();
state
}
fn eval<T: mlua::FromLuaMulti>(s: &EditorState, src: &str) -> T {
s.lua_host.lua().load(src.to_string()).eval().unwrap()
}
fn exec(s: &EditorState, src: &str) {
s.lua_host.lua().load(src.to_string()).exec().unwrap();
}
/// Current preference size in logical px, or `None` when unset — the
/// real "frontend's own default" state, never inferred from silence.
fn size(s: &EditorState) -> Option<f64> {
eval(s, "return pmacs.gpu.font().size")
}
fn family(s: &EditorState) -> Option<String> {
eval(s, "return pmacs.gpu.font().family")
}
// ---------------------------------------------------------------------------
// The starting point (Q#Z1 = c)
// ---------------------------------------------------------------------------
/// The first step starts from the CONFIGURED base, not a hardcoded
/// 16.0 — which is the whole reason (c) was chosen over (a). The daemon
/// never learns what the frontend resolved; it reads a user-facing
/// preference about zooming.
#[test]
fn the_first_step_starts_from_the_configured_base_not_a_constant() {
let (roots, _) = roots_for("first_step_base");
let s = session(&roots);
assert_eq!(size(&s), None, "premise: untouched means unset");
exec(&s, r#"pmacs.config.set("ui.gpu-font-size-base", 20.0)"#);
exec(&s, "pmacs.zoom.increase()");
assert_eq!(
size(&s),
Some(21.0),
"20.0 base + 1.0 step. A hardcoded origin would give 17.0"
);
}
/// Until a zoom happens the preference stays unset, so the frontend's
/// own default query still runs. This is what option (b) would have
/// destroyed for every user who never zooms.
#[test]
fn defining_the_settings_does_not_itself_set_a_size() {
let (roots, _) = roots_for("no_implicit_size");
let s = session(&roots);
assert_eq!(size(&s), None);
assert_eq!(family(&s), None);
}
// ---------------------------------------------------------------------------
// Step arithmetic (Q#Z2 = additive)
// ---------------------------------------------------------------------------
/// n steps in, then n steps out, returns to EXACTLY the starting value.
/// True because the step is centi-pixel representable and addition is
/// exact in that domain — the property multiplicative stepping loses.
#[test]
fn n_steps_in_then_n_out_returns_exactly() {
let (roots, _) = roots_for("round_trip");
let s = session(&roots);
exec(&s, r#"pmacs.config.set("ui.gpu-zoom-step", 0.37)"#);
exec(&s, "pmacs.zoom.increase()");
let start = size(&s).expect("a size exists after the first step");
for _ in 0..7 {
exec(&s, "pmacs.zoom.increase()");
}
for _ in 0..7 {
exec(&s, "pmacs.zoom.decrease()");
}
assert_eq!(
size(&s),
Some(start),
"exact return, not approximately — 0.37 is centi-pixel \
representable and the domain is quantized"
);
}
/// An out-of-range step leaves the preference **unmutated** rather than
/// pinning it to the boundary. Pinning would silently break the round
/// trip precisely at the edges, where a user steps back and forth most.
#[test]
fn a_step_past_the_boundary_changes_nothing_and_says_so() {
let (roots, _) = roots_for("boundary");
let s = session(&roots);
exec(&s, "pmacs.gpu.set_font { size = 71.5 }");
exec(&s, r#"pmacs.config.set("ui.gpu-zoom-step", 2.0)"#);
let why: Option<String> = eval(&s, "local _, why = pmacs.zoom.increase() return why");
assert_eq!(
size(&s),
Some(71.5),
"the preference is untouched, not pinned to 72.0"
);
assert!(
why.unwrap_or_default().contains("unchanged"),
"and the caller is told why"
);
}
// ---------------------------------------------------------------------------
// Config bounds (§3.1)
// ---------------------------------------------------------------------------
/// A zero, negative, or sub-centi-pixel step is refused by the REGISTRY,
/// not discovered later as a zoom that does nothing or runs backwards.
///
/// The negative case is the sharp one: it would invert the commands, and
/// `gpu.zoom-in` shrinking is not a malfunction a user can diagnose from
/// the outside — the command still does something coherent.
#[test]
fn an_unusable_step_is_refused_at_the_registry() {
let (roots, _) = roots_for("step_bounds");
let s = session(&roots);
for bad in ["0.0", "-1.0", "0.001"] {
let ok: bool = eval(
&s,
&format!(r#"return pcall(pmacs.config.set, "ui.gpu-zoom-step", {bad})"#),
);
assert!(!ok, "a step of {bad} must be refused");
}
// …and the smallest representable step is allowed.
let ok: bool = eval(
&s,
r#"return pcall(pmacs.config.set, "ui.gpu-zoom-step", 0.01)"#,
);
assert!(ok, "0.01 is one centi-pixel — the smallest real step");
}
/// The base is bounded to the wire range. A base outside it could never
/// be sent, so the first step would fail from a value the registry had
/// allowed the user to set.
#[test]
fn the_base_is_bounded_to_the_wire_range() {
let (roots, _) = roots_for("base_bounds");
let s = session(&roots);
for bad in ["5.99", "72.01"] {
let ok: bool = eval(
&s,
&format!(r#"return pcall(pmacs.config.set, "ui.gpu-font-size-base", {bad})"#),
);
assert!(!ok, "a base of {bad} is outside 6.00-72.00");
}
}
// ---------------------------------------------------------------------------
// The family clobber (§5a)
// ---------------------------------------------------------------------------
/// `set_font` replaces BOTH fields unconditionally, so a size-only write
/// would clear a family the user configured — and they would get it back
/// only by restarting. Every zoom write must carry the family through.
#[test]
fn a_zoom_preserves_a_configured_family() {
let (roots, _) = roots_for("family_kept");
let s = session(&roots);
exec(
&s,
r#"pmacs.gpu.set_font { family = "Iosevka", size = 18.0 }"#,
);
exec(&s, "pmacs.zoom.increase()");
assert_eq!(size(&s), Some(19.0));
assert_eq!(
family(&s).as_deref(),
Some("Iosevka"),
"the family survives a zoom — set_font replaces both fields, so \
a size-only write would silently drop it"
);
exec(&s, "pmacs.zoom.decrease()");
assert_eq!(
family(&s).as_deref(),
Some("Iosevka"),
"and every step after"
);
}
// ---------------------------------------------------------------------------
// Reset (§5b)
// ---------------------------------------------------------------------------
/// Reset returns the size to UNSET — the frontend's own default — not to
/// the configured base. Resetting to the base would ship an explicit
/// size that merely happens to equal the default, making the untouched
/// state unreachable once a user has ever zoomed.
#[test]
fn reset_returns_to_unset_not_to_the_base() {
let (roots, path) = roots_for("reset_unsets");
let s = session(&roots);
exec(&s, r#"pmacs.gpu.set_font { family = "Iosevka" }"#);
exec(&s, "pmacs.zoom.increase()");
assert!(size(&s).is_some(), "premise: a size is set");
assert!(path.exists(), "premise: the zoom was saved");
exec(&s, "pmacs.zoom.reset()");
assert_eq!(size(&s), None, "unset, not the base value");
assert_eq!(
family(&s).as_deref(),
Some("Iosevka"),
"reset is about size; a configured family is not collateral"
);
// Cleared, so it does not resurrect on the next launch — "reset
// until restart" is not what the word says.
let saved = std::fs::read_to_string(&path).unwrap_or_default();
assert!(
saved.trim().is_empty(),
"reset clears the saved zoom, found {saved:?}"
);
}
// ---------------------------------------------------------------------------
// Persistence and its parser (§5a1)
// ---------------------------------------------------------------------------
/// The exact bytes the writer produces round-trip. This is the case the
/// framing's first parser (`^(%d+)$`) would have REJECTED — it anchors
/// to end-of-subject, so it fails on the trailing newline — which is why
/// it is asserted rather than assumed.
#[test]
fn a_valid_newline_terminated_file_restores() {
let (roots, path) = roots_for("restore_valid");
std::fs::write(&path, "1800\n").expect("plant state");
let s = session(&roots);
assert_eq!(
size(&s),
Some(18.0),
"restored at the seam, from the bytes the writer emits"
);
}
/// What zoom writes is what zoom reads. Pins the two halves against each
/// other so a format change cannot land in one alone.
#[test]
fn the_written_format_is_the_format_that_parses() {
let (roots, path) = roots_for("format_round_trip");
{
let s = session(&roots);
exec(&s, "pmacs.gpu.set_font { size = 23.5 }");
exec(&s, "pmacs.zoom.increase()");
assert_eq!(size(&s), Some(24.5));
}
let raw = std::fs::read_to_string(&path).expect("state written");
assert_eq!(raw, "2450\n", "centi-pixels, newline-terminated");
let s2 = session(&roots);
assert_eq!(size(&s2), Some(24.5), "and a fresh session reads it back");
}
/// Malformed, out-of-range, and shape-wrong state all behave as absent,
/// and the file is left intact for inspection rather than truncated.
///
/// The multi-line case is the one that matters most: `saveplace`'s
/// `gmatch("([^\n]+)")` line iterator would happily accept its FIRST
/// line. That is right for `recentf`, where a line is one independent
/// entry, and wrong here, where the file IS the value.
#[test]
fn unparseable_or_out_of_range_state_behaves_as_absent() {
for (label, contents) in [
("multi-line", "1800\n1900\n"),
("no trailing newline", "1800"),
("decimal", "18.0\n"),
("leading space", " 1800\n"),
("empty", ""),
("bare newline", "\n"),
("non-numeric", "big\n"),
("above the wire range", "9999\n"),
("below the wire range", "100\n"),
] {
let (roots, path) = roots_for(&format!("bad_{}", label.replace(' ', "_")));
std::fs::write(&path, contents).expect("plant state");
let s = session(&roots);
assert_eq!(
size(&s),
None,
"{label}: {contents:?} must behave exactly as no saved state"
);
assert_eq!(
std::fs::read_to_string(&path).expect("still there"),
contents,
"{label}: the file is left intact — truncating it on read \
would destroy the only evidence of whatever wrote it"
);
}
}
/// Precedence: a valid saved zoom beats a size set during startup, and
/// never touches a configured family.
///
/// The saved value is a later, deliberate user action; the init value is
/// the standing default it was chosen against — the same precedence
/// `saveplace` already applies to a remembered position.
#[test]
fn saved_state_beats_a_startup_size_but_never_the_family() {
let (roots, path) = roots_for("precedence");
std::fs::write(&path, "2600\n").expect("plant state");
let s = EditorState::new_with_roots(&roots);
// Stand in for init.lua, which runs before state is installed.
exec(
&s,
r#"pmacs.gpu.set_font { family = "Iosevka", size = 12.0 }"#,
);
s.install_state_dirs();
assert_eq!(size(&s), Some(26.0), "the remembered zoom wins on SIZE");
assert_eq!(
family(&s).as_deref(),
Some("Iosevka"),
"and never on family — restored state does not carry one"
);
}
// ---------------------------------------------------------------------------
// No keybindings (Q#Z3 = C)
// ---------------------------------------------------------------------------
/// This stage ships commands and NO default bindings.
///
/// The keymap has no way to say "GPU frontends only" — `Scope` is
/// `Buffer | Mode | Global` and carries no frontend identity — so a
/// global binding would capture the chord in the TUI and take away the
/// terminal's own zoom, which is the very thing the user is pressing it
/// for. Better an unbound key than one that answers with an apology.
#[test]
fn the_commands_exist_and_nothing_is_bound() {
let (roots, _) = roots_for("no_bindings");
let s = session(&roots);
// `pmacs.command.list()` returns plain names, as `help.lua` reads it.
let names: Vec<String> = eval(&s, "return pmacs.command.list()");
for want in ["gpu.zoom-in", "gpu.zoom-out", "gpu.zoom-reset"] {
assert!(names.iter().any(|n| n == want), "{want} is discoverable");
}
let bound: Vec<String> = eval(
&s,
r#"local out = {}
for _, b in ipairs(pmacs.keymap.list()) do
if b.command and b.command:match("^gpu%.zoom") then
out[#out + 1] = string.format("%s -> %s (%s)", b.sequence, b.command, b.scope)
end
end
return out"#,
);
assert!(
bound.is_empty(),
"no zoom keybinding may be installed by default, found {bound:?}"
);
}