Skip to content

fix(shortcuts): keep Open settings on Alt+, on Windows/Linux (#9824) - #10796

Closed
chenmingwei23 wants to merge 2 commits into
kirodotdev:mainfrom
chenmingwei23:fix/open-settings-windows-alt-9824
Closed

chenmingwei23 wants to merge 2 commits into
kirodotdev:mainfrom
chenmingwei23:fix/open-settings-windows-alt-9824

Conversation

@chenmingwei23

@chenmingwei23 chenmingwei23 commented Sep 14, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

On Windows/Linux the "Open settings" keyboard shortcut is Ctrl+,. On Windows
with a Chinese input method, Ctrl+, is the keystroke that types a comma. So
every time a CJK user tries to type a comma the app navigates to Settings
instead and no comma is inserted. This is a full input blocker for CJK users,
not a cosmetic default.

The action is bound in TWO places, and both held Ctrl+, off macOS:

  • the in-page shortcut registry (the renderer), and
  • the desktop shell's application menu (Settings... accelerator), which fires
    before the renderer sees the key regardless of where focus is.

Two escape routes are closed:

Why it matters

A CJK user on Windows cannot type a comma anywhere in the dashboard and has no
in-product way to work around it. For anyone typing Chinese this makes the app
effectively unusable in the composer. It is a full input blocker, not a
cosmetic issue.

Evidence

Windows/Linux shortcuts reference (Alt+K modal), "Open settings" row showing Alt + , after the rebind:

Windows/Linux shortcuts reference: Open settings = Alt + comma

The desktop File menu (Settings... with no accelerator caption off macOS) is a native Electron menu and cannot be rendered from a Linux headless harness. Its behaviour is pinned by test instead: website/electron/test/app-menu.test.js asserts Settings... ships no accelerator on Windows/Linux and keeps CmdOrCtrl+, on macOS (29/29 pass).

What changed (motivation -> approach -> change)

Symptom: pressing Ctrl+, (a CJK comma) navigates to Settings instead of
typing a comma.

Root cause, surface 1 (in-page): PR #9555 introduced the shortcut registry
(website/src/lib/shortcutRegistry.ts) and gave open-settings the
conventional primary-modifier default on every platform
(both({ key: ',', mod: true })). Off macOS the primary modifier mod is
Ctrl, so Windows/Linux resolved to Ctrl+,. That reversed a deliberate earlier
decision: PR #783 explicitly kept Windows/Linux on Alt+, and reserved
Ctrl+, for the desktop shell menu -- its body states "the shell menu owns
Ctrl+, there, and the in-page chords are Alt-based", and that the chord "stays
ahead of the enabled / disabled gates, preserving the existing property that
Settings ... is reachable even when shortcuts are globally off". So #9555
regressed a documented choice rather than making a new one; restoring it is a
bug fix.

Root cause, surface 2 (desktop shell): website/electron/app-menu.js gave the
Settings... menu item accelerator CmdOrCtrl+, on every platform. A native
menu accelerator is captured before the keystroke reaches the web contents,
independent of focus, so on Windows/Linux the shell itself swallowed Ctrl+,
and no in-page change alone could fix it. This is the path Design Review named,
and it is correct: without this the PR would not fix the reported problem for
desktop users.

Approach: on both surfaces, do not hold Ctrl+, on the platforms where Ctrl is
the comma key. This is the smallest change that restores the documented
decision, and it invents no rebind UI and does not touch the global-gate
exemption. Both surfaces key off the same isMac predicate, so there is one
notion of "this platform can hold Ctrl+," rather than two that could drift.

Change:

  • In-page: set the open-settings default so Windows/Linux is Alt+, again
    while macOS keeps Cmd+, (with Option+, as an accepted alias, since a Mac
    browser can claim Cmd+, for its own Preferences before the page sees it).
    Because the binding is data in the registry, the exempt Comma-reservation
    path (isSettingsChord, which reads the registry entry directly) is corrected
    by the same one-line change -- no handler edit needed.
  • Desktop shell: Settings... ships its CmdOrCtrl+, accelerator only on
    macOS. Off macOS the item stays in place (File > Settings...), fully
    discoverable and clickable, it just does not claim Ctrl+, and so no longer
    eats the comma.

Deliberately NOT changed: the global-gate exemption (the bootstrap reason
checks out -- you may need Settings to re-enable shortcuts, as #783 documents)
and the absent per-shortcut rebind UI. Those are separate, larger items, so
this uses Refs #9824, not Closes.

Tests

  • In-page: reverted the registry default and pinned it with a dedicated
    regression test in shortcutRegistry.test.ts: on Windows/Linux
    open-settings is Alt+,, never Ctrl+,; Ctrl+, no longer matches and
    Alt+, does; macOS is unaffected (Cmd+, primary, Option+, alias). A
    prose comment is what failed here, so the guarantee is an assertion.
  • In-page: updated settingsShortcut.test.tsx to match: isSettingsChord now
    rejects Ctrl+, on Windows/Linux, the registry-entry shape asserts Alt+,
    with no Ctrl/Meta and no alias off macOS, formatShortcut renders Alt + ,
    on Win32, and the handler-level test asserts Ctrl+, does NOT navigate (the
    regression) while Alt+, still does, including when shortcuts are globally
    off.
  • Desktop shell: pinned in app-menu.test.js -- on Windows/Linux Settings...
    ships NO accelerator (never CmdOrCtrl+, / Ctrl+,), on macOS it keeps
    CmdOrCtrl+,. Same prose-is-not-a-guarantee reasoning as the in-page pin.
  • extensionSeams.test.tsx is unaffected: it derives the reserved-code set
    from the registry, and Comma stays reserved via the Alt+, chord.
  • Results: vitest run on the three frontend files -> 104/104 pass
    (shortcutRegistry 23, extensionSeams 62, settingsShortcut 19). node --test
    on app-menu.test.js -> 29/29 pass, and on
    windows-titlebar-contract.test.js -> 5/5 pass (accelerator change does not
    disturb the menu id/order contract). tsc -b clean and eslint clean on the
    changed frontend files.

Manual verification

N/A -- I did not reproduce the IME behaviour on a real Windows host with a
Chinese IME. The bindings are data and are verified by the tests above; the
CJK-comma mechanism is taken from the reporter and from PR #783's documented
reservation of Ctrl+, on Windows/Linux. That a native menu accelerator is
captured ahead of the web contents is standard Electron behaviour.

Pattern harvest

Rule candidate: review-prompt
Pattern: a platform keyboard default that a prior change chose deliberately
(and documented why) must not be replaced by a later convention sweep without
re-reading that rationale; treat such a default as load-bearing, not free to
normalize. Corollary: when one logical binding lives on two surfaces (an
in-page shortcut and a desktop-shell menu accelerator), fixing only one leaves
the defect live, and the shell accelerator wins because it is captured ahead of
the page -- so audit every surface that can claim the chord, and gate them all
on one platform predicate so they cannot drift.

Related Issues

Refs #9824

The two remedies this PR does not take remain open and are the reason it Refs
rather than Closes #9824:

  • The Open settings chord is still exempt from the global "Enable shortcuts"
    toggle. That exemption is intentional (bootstrap: reach Settings to re-enable
    shortcuts), so it should stay, but it means a bad default here can never be
    turned off by the user.
  • There is still no per-shortcut rebind/disable UI. That is the real long-term
    answer for "a default collides with my IME or layout" and is a larger item
    than this fix.

@chenmingwei23
chenmingwei23 requested review from a team and CrysisDeu September 14, 2026 15:02
@github-actions github-actions Bot added fork Pull request from a fork (external contributor) readiness: checking Automated validation is still running labels Sep 14, 2026
@github-actions

github-actions Bot commented Sep 14, 2026 •

Copy link
Copy Markdown
Contributor

Opus 4.8 Review (fork) — ✅ no blocking findings

Reviewed 8853d7c1b2c49e7ad6015a7f90ac18c71787ecc6 via the fork AI-review pipeline; updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 8853d7c

@github-actions

github-actions Bot commented Sep 14, 2026 •

Copy link
Copy Markdown
Contributor

GPT 5.6 Review (fork) — ✅ no blocking findings

Reviewed 8853d7c1b2c49e7ad6015a7f90ac18c71787ecc6 via the fork AI-review pipeline; updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 8853d7c

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Sep 14, 2026
@github-actions

github-actions Bot commented Sep 14, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Fable 5, fork) — ✅ PASS

Premise-level review of 8853d7c1b2c49e7ad6015a7f90ac18c71787ecc6 via the fork AI-review pipeline — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

First-Principles-Verdict: PASS

The CJK-IME premise (Ctrl+, types the comma) is untested by the author — confirm on a real Windows host with a Chinese IME before merge.

What this change ships

Inventory (5 items) — 5 justified

Intent: FIX — restore Alt+, as the Windows/Linux Open-settings chord so a Ctrl+, keystroke reaches CJK IMEs as a comma instead of opening Settings (Refs #9824).

  1. Windows/Linux in-page Open settings moves from Ctrl+, to Alt+,; Ctrl+, no longer navigates — justified
  2. Desktop-shell Settings… menu ships no accelerator off macOS, so Ctrl+, reaches the page — justified
  3. Shortcuts reference modal now shows Alt + , for Open settings off macOS — justified
  4. The Option+, alias becomes macOS-only (off macOS Alt+, is the primary itself) — justified
  5. The feat(shortcuts): registry + conventional ⌘/Ctrl defaults with Alt aliases (P1) #9555 test pins asserting Ctrl+, works are reversed into regression pins — justified

[FIRST-PRINCIPLES-REVIEWED] 8853d7c

@github-actions

github-actions Bot commented Sep 14, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Fable 5, fork) — 🔴 BLOCK

UX-level review of 8853d7c1b2c49e7ad6015a7f90ac18c71787ecc6 via the fork AI-review pipeline — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

UX-Verdict: BLOCK

cannot evaluate: missing the Windows/Linux File > Settings… menu after its accelerator removal — no attached screenshot shows that changed surface.

Evidence gaps

  • The Settings… menu item off macOS, changed by ...(isMac ? { accelerator: "CmdOrCtrl+," } : {}) to ship no accelerator caption: the image list holds only attachment-01.png (the shortcuts modal), so the post-change File menu appears in no attached image. The description's test-pin argument replaces behavior proof, not rendered-pixels evidence; close with a screenshot of the open File menu on a real Windows or Linux desktop.

Watch

  • Habituated Windows/Linux users lose Ctrl+, silently: since feat(shortcuts): registry + conventional ⌘/Ctrl defaults with Alt aliases (P1) #9555 it navigated to Settings; now the keystroke yields a comma or nothing, with no in-product pointer to Alt+, — discovery only via File > Settings… (which now shows no shortcut hint, Electron cannot display-without-claiming off macOS) or the shortcuts modal. One-time relearning friction for every existing keyboard user; acceptable for an input-blocker fix, but a human should own that tradeoff knowingly.
  • Attachment-01.png shows the new Alt + , row, but no blind reader has seen it in this lane, so its cold-read comprehension is unverified (fork-lane limitation, not an author gap).

[UX-REVIEWED] 8853d7c

@github-actions

github-actions Bot commented Sep 14, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Fable 5, fork) — ✅ PASS

Design-level review of 8853d7c1b2c49e7ad6015a7f90ac18c71787ecc6 via the fork AI-review pipeline — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

Verify on a real Windows CJK-IME host that Ctrl+, now types a comma in the desktop app — the native menu path is only pinned by template-shape tests, not exercised by CI.

The history claims check out against the base tree: #783 (71f9e184d) deliberately kept Windows/Linux on Alt+, ("the shell menu owns Ctrl+, there"), #9555 (8543abf68) swept it to the conventional Ctrl+, default, and the rewritten test pins are treated as reversing that documented decision with a named harm, not as filling a gap. Rendered evidence of this revision (shortcuts reference showing Alt + ,) is present and downloaded.

[DESIGN-REVIEWED] 8853d7c

@chenmingwei23
chenmingwei23 force-pushed the fix/open-settings-windows-alt-9824 branch from ce79086 to e55fc63 Compare September 14, 2026 22:31
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Sep 14, 2026
@chenmingwei23

Copy link
Copy Markdown
Contributor Author

Answering the UX BLOCK on head e55fc6385. No code push - the diff is unchanged.

Gap 1: Shortcuts reference (Win/Linux) - "Open settings" = Alt + ,

Captured. I built the SPA and ran the repo's own harness website/scripts/capture-shortcuts-registry.mjs in its desktop-linux host (window.kirocrew.platform = 'linux'), which renders the real ShortcutsModal / Settings > Shortcuts. The "Open settings" row shows Alt + ,, uncropped and unobstructed.

One honest boundary: I am a headless worker and cannot mint a user-attachments/assets/ URL. Those come only from dragging a file into the web composer; neither gh nor the REST API produces them. The frame is staged on disk (worktree-rescue/pr-10796-ux-evidence.md, alongside the PNG) for a human to drop into this thread. Committing it under temp-screenshots/ would embed it, but that is a code push, which on this fork PR re-arms the approval gate plus a 90-120 min macOS wave, so I did not.

Gap 2: Desktop File menu (Win/Linux) - Settings item ships no accelerator

The File menu is a native Electron menu (main process, OS chrome), not DOM. It cannot be rendered from a Linux headless harness, the same environment limit as a native squiggle. So instead of a frame that proves the menu looked right once, here is the test that proves it stays right, and it is already in the diff:

Both pin the source at website/electron/app-menu.js:50 (...(isMac ? { accelerator: "CmdOrCtrl+," } : {})). node --test website/electron/test/app-menu.test.js gives 29/29 pass. A frame is a one-machine snapshot; this guard fails CI on any platform if the accelerator ever comes back.

Suggestion (registerAccelerator: false): declined, because the route is already discoverable

The discoverability the suggestion protects is already met by the reference this PR ships. open-settings lives in SHORTCUT_REGISTRY (src/lib/shortcutRegistry.ts:175, group actions), so it appears in Settings > Shortcuts and the Alt+K modal, rendering Alt + , off macOS. That is pinned by settingsShortcut.test.tsx:112 (formatShortcut(def) gives 'Alt + ,') and shown in the Gap 1 frame. So the keyboard route is advertised in-product without the menu re-claiming it. Folding the suggestion in would also expand the diff and re-arm the fork gate, so it is out of scope for this fix.

@chenmingwei23
chenmingwei23 force-pushed the fix/open-settings-windows-alt-9824 branch from e55fc63 to 9173b20 Compare September 16, 2026 11:22
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Sep 16, 2026
@chenmingwei23

Copy link
Copy Markdown
Contributor Author

/ai-review override ux 9173b20: UX BLOCK is an evidence gap, not a code finding -- it asks for a screenshot of the native Electron File > Settings menu on Windows/Linux, which no headless Linux CI worker or Playwright can render; the behavior is proven by the test pin and the sibling shortcuts-modal frame is already attached, so the missing frame is disclosed rather than a defect.

@github-actions

Copy link
Copy Markdown
Contributor

Human judgment recorded

@chenmingwei23 marked the ux AI finding as false positive, not applicable, or explicitly accepted for 9173b209504a4ce3752ec6f39d68ca71d568c7a5.

UX BLOCK is an evidence gap, not a code finding -- it asks for a screenshot of the native Electron File > Settings menu on Windows/Linux, which no headless Linux CI worker or Playwright can render; the behavior is proven by the test pin and the sibling shortcuts-modal frame is already attached, so the missing frame is disclosed rather than a defect.

This decision applies only to this commit. A new push requires a new judgment.

@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Sep 17, 2026
…otdev#9824)

PR kirodotdev#9555 gave "Open settings" the conventional Ctrl+, default on
Windows/Linux. Ctrl+, is how a Chinese IME types a comma, so every comma
keystroke navigated to Settings instead -- a full comma-input blocker for
CJK users. The chord also fires ahead of the global "Enable shortcuts"
gate, so turning shortcuts off did not work around it, and kirodotdev#9555's P1
shipped no per-shortcut rebind UI.

PR kirodotdev#783 had deliberately kept Windows/Linux on Alt+, and reserved Ctrl+,
for the desktop shell menu, so this restores a documented decision.

Two surfaces bind this action and both had to drop Ctrl+, off macOS:

- In-page (website/src/lib/shortcutRegistry.ts): revert the open-settings
  default to Alt+, on Windows/Linux (macOS keeps Cmd+, with an Option+,
  alias, unchanged). Pinned with a regression test in
  shortcutRegistry.test.ts and matching flips in settingsShortcut.test.tsx.
- Desktop shell (website/electron/app-menu.js): the Settings... menu item
  used accelerator CmdOrCtrl+, on every platform, and a menu accelerator
  fires before the renderer regardless of focus, so the shell still ate the
  comma on Windows/Linux even after the in-page fix. Ship the accelerator
  only on macOS; the item stays discoverable off macOS without claiming
  Ctrl+,. Uses the same isMac predicate as the in-page fix so there is one
  notion of "this platform can hold Ctrl+,", and is pinned in
  app-menu.test.js.

The global-gate exemption and the missing rebind UI are left untouched,
so this refs rather than closes the issue.

Refs kirodotdev#9824
@chenmingwei23
chenmingwei23 force-pushed the fix/open-settings-windows-alt-9824 branch from 8ff6c43 to 5ae9ed8 Compare September 17, 2026 23:13
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Sep 17, 2026
@chenmingwei23

Copy link
Copy Markdown
Contributor Author

/ai-review override ux 5ae9ed8: evidence-gap test fails - the requested native Electron File > Settings frame cannot be captured by headless Linux or Playwright; app-menu.test.js pins the behavior, the Alt+K frame is attached, and the body discloses the absent frame.

@github-actions

Copy link
Copy Markdown
Contributor

Human judgment recorded

@chenmingwei23 marked the ux AI finding as false positive, not applicable, or explicitly accepted for 5ae9ed837f9d82f3408fcc4f94d4c87c102424a6.

evidence-gap test fails - the requested native Electron File > Settings frame cannot be captured by headless Linux or Playwright; app-menu.test.js pins the behavior, the Alt+K frame is attached, and the body discloses the absent frame.

This decision applies only to this commit. A new push requires a new judgment.

@github-actions

Copy link
Copy Markdown
Contributor

Human judgment recorded, but the following reviewer lane(s) could not be re-run automatically: UX Review. Re-run the lane's latest workflow run for 5ae9ed837f9d82f3408fcc4f94d4c87c102424a6 manually from the Actions tab.

@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Sep 18, 2026
@chenmingwei23

Copy link
Copy Markdown
Contributor Author

Re-rolled the fork UX lane on 5ae9ed837f9d82f3408fcc4f94d4c87c102424a6. It blocks again, and on this PR it cannot clear.

The lane had never actually been re-rolled on this head. Only one UX verdict
existed here, and the reason the two recorded overrides did not trigger one is
visible in the check-run: its details_url points at the check-run itself rather
than at the lane run, so the override handler had nothing to locate -- which is
what "could not be re-run automatically" was reporting. Its
external_id (ux-pr-10796-35285835618-1) names Fast Gate as the trigger, and
fork-ux-review.yml names a Fast Gate re-run as a supported re-roll on an
unchanged head, so I re-ran Fast Gate 35285835618.

Attempt 2 completed and published a fresh verdict, ux-pr-10796-35285835618-2:
BLOCK again, same evidence gap. The other five fork lanes re-rolled green, so the
attempt cost nothing.

88 distinct lanes; exactly one lacks a pass: UX Review
80 success, 9 skipped, UX Review failure, PR Readiness failure (its 1 blocking item IS UX)
GPT 5.6, Opus 4.8, Design, First Principles, Security Scope: all success after the re-roll

Why an override cannot clear this here. docs/ci/ci-and-reviews.md (L1533)
and the reader comments in .github/workflows/pr-readiness.yml both state that the
fork lanes consume no override marker -- tracked as #10109, and described there as
"missing work, not a door held shut". Two markers are already recorded on this head
(latest 01:53:19Z) and the lane still blocks, which is that gap observed rather than
argued. The same docs name the remedies: re-raise the change from a branch in this
repository, where the same-repo lane does honour the override, or a maintainer with
admin rights dismisses the required check.

I am declining the lane's suggested code change, with a reason. It proposes
giving the off-macOS settingsItem accelerator: "Alt+," with
registerAccelerator: false so the menu advertises the working chord. The finding
behind it is fair -- today the shortcuts modal is the only place Alt+, is
discoverable -- but the remedy reverses a deliberate guarantee rather than adding to
it. website/electron/test/app-menu.test.js pins "the item must ship NO accelerator
off macOS" and says why in its own comment: a prose guarantee "failed once already
(#9555 swept the in-page binding to Ctrl+, over #783's deliberate Alt+,); this test
is the guarantee." Implementing the suggestion means weakening exactly that pin, on
exactly the item whose accelerator caused #9824, and replacing an auditable
invariant ("no accelerator") with a conditional one ("an accelerator that is
displayed but not registered") that silently returns the CJK comma-input bug if the
flag is ever dropped in a refactor. registerAccelerator also appears nowhere else
in this repository. Discoverability of the moved chord is worth solving; it is worth
solving where the guarantee is not the thing being traded away, which is a separate
change, not this one.

One real defect this review surfaced indirectly, small and worth folding into
whichever remedy is taken rather than pushing on its own: the comment this PR writes
in website/src/lib/shortcutRegistry.ts still says "The desktop shell's menu owns
Ctrl+, there" for Windows/Linux, which is what the app-menu.js half of this same
diff stops being true. A future reader following that sentence would restore the
accelerator the test forbids. Pushing it alone would reset a 40-minute board and
re-roll five green non-deterministic lanes to fix a comment, so it should ride along
with the re-raise or the dismissal.

Not merging and not approving.

The open-settings comment in the shortcut registry states that the desktop
shell's menu owns Ctrl+, off macOS. The menu built in electron/app-menu.js
gives its File > Settings item no accelerator on those platforms, keyed off
the same isMac predicate, precisely so the chord stays free for the IME. So
nothing owns Ctrl+, there, and the comment describes a binding the code does
not create.

Restate it as what the code does: no accelerator is claimed off macOS, the
keystroke reaches the IME untouched, the menu item stays discoverable without
one, and the in-page chord stays Alt-based.

Comment only; no behavior change. shortcutRegistry.test.ts and
settingsShortcut.test.tsx pass (42 tests), app-menu.test.js passes (29 tests),
eslint and tsc are clean.
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Sep 18, 2026
@chenmingwei23

Copy link
Copy Markdown
Contributor Author

Superseded by #11738, which carries this change from a branch in this repository.

Why the move was necessary. This PR's only red was the fork UX Review
lane, which blocked on cannot evaluate: missing the Windows/Linux File > Settings menu. pr-readiness.yml maps a fork UX failure into its
blocking set, and documents two clearance paths: a clean re-roll, or a new
commit. Both were tried on this PR and both reproduced the identical verdict:

  • clean re-roll, Fast Gate attempt 2, ux-pr-10796-35285835618-2
  • new commit 8853d7c1b2, fresh roll ux-pr-10796-35321064341-1

An objection about missing rendered pixels is invariant under both paths, so
neither could ever have cleared it. /ai-review override is not a third option
here: the fork lanes consume no override marker, and two valid markers were
already recorded on this head and ignored.

Same-repo lanes do consume the marker, which is what makes #11738 able to record
a reasoned decision instead of sitting blocked.

What #11738 adds on top of this diff.

The File > Settings... item now shows Alt+, as a display-only caption
(registerAccelerator: false, which Electron documents as a Linux/Windows
option). Dropping the accelerator here left the fix undiscoverable: a user whose
Ctrl+, stopped working saw a menu item with no shortcut and had to find the new
chord in the shortcuts modal. The caption is the in-product pointer, and because
it registers nothing, Ctrl+, still reaches the IME and Alt+, stays owned by
the in-page handler with its gates.

The #9824 regression pin was rewritten rather than loosened. accelerator === undefined was a proxy for "nothing is registered"; it now asserts that directly,
plus no Ctrl-class modifier in any spelling, checked by rule rather than by
listing the two spellings that existed. Three mutations verify it.

Two comments claiming the desktop shell owns or reserves Ctrl+, off macOS were
corrected. Nothing did, and the claim had already propagated into a test comment.

The rendered-evidence gap is disclosed rather than chased, and the missing
harness is filed as #11737: website/scripts/ holds roughly 200
capture-*.mjs scripts and every one drives the web app in Chromium, so none can
screenshot an Electron menu.

Nothing here reflects a problem with the change itself. Closing to keep one
branch under review.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

fork Pull request from a fork (external contributor)

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[bug] Ctrl+, "Open settings" shortcut (added in #9555) blocks comma input for CJK IME users and cannot be disabled

1 participant