Fix recording shortcuts dropping out during Secure Input - #965
Conversation
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
Greptile SummaryThis PR adds a Carbon-based path for eligible recording shortcuts so they continue working while Secure Input suppresses event-tap delivery.
|
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 222b8a131c
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if down { | ||
| guard self.pressed.insert(id).inserted else { return } // no autorepeat |
There was a problem hiding this comment.
If the machine sleeps, switches sessions, or otherwise loses the release callback while a registered chord is held, its ID remains in pressed, so every subsequent press of that shortcut is rejected here as autorepeat. Event-tap recovery also deliberately preserves state while hasPressedShortcut is true, leaving the shortcut disabled until a settings change or manual reinitialization calls releaseAll; clear or reconcile registered press state during session/wake recovery.
Useful? React with 👍 / 👎.
1 participant
Description
Option+Space kept stopping working for me on Tahoe. I tracked it down to ChatGPT holding Secure Event Input: FluidVoice's event tap could still report itself as enabled, but it wasn't receiving the shortcut.
This registers recording shortcuts like Option+Space with
RegisterEventHotKeyand sends press/release events through the existing dictation logic. I tested the patched app on the same Mac with Secure Input still active, and Option+Space now starts recording again.Registration is exclusive, so a conflicting owner produces an error and keeps the event-tap fallback available. Entering shortcut capture or reinitializing the manager also stops a held Automatic-mode recording, even if the press was short.
The change also handles the bits around registration: avoiding duplicate actions from the event tap, ignoring key repeats, finishing held shortcuts before settings change, and temporarily unregistering shortcuts while editing them. Event-tap recovery leaves active registered shortcuts alone. Secure Input state is logged separately from event-tap health. Sleep/wake and session changes finish interrupted presses and clear stale held-key state, so losing a release callback does not swallow the next press.
Type of Change
Related Issue or Discussion
Related to #849, especially the report that Option+Space fails while modifier-only shortcuts still work. I reproduced that symptom, though I can't confirm that report had the same cause.
Also related to the event-tap reliability work in #657. This gives supported key combinations a path that doesn't depend on the tap; it doesn't fix the Function-key failure described there. I'm leaving both as related issues rather than claiming to close them.
Testing
legacy_swiftui_aspect_ratioviolations in unrelated UI files.HotkeyShortcutTestsandRegisteredHotkeysTests). The standalone 13-test suite also passed.The new tests cover exclusive registration requests, native conflicts and retry after the owner releases the shortcut, interruption release context, missing key-up recovery across all four sleep/session notifications, registration cleanup against the native API, routing without event-tap events, repeat suppression, modifier-first releases, failed registrations and retry, shortcut editing, stale events, and balancing held shortcuts.
The missing-release test failed before adding the lifecycle observers and passes afterward. It also checks repeated recovery notifications, late releases, and the next complete press.
I checked the regression tests against temporary copies with each fix reverted; both caught the old behavior.
I also launched the debug app and tested physical Option+Space presses with Secure Input active. The shortcut reached the dictation callback; after granting the debug app microphone permission, recording worked. Hold/automatic mode, sleep/wake, and VNC recovery still need hands-on testing.
Screenshots / Video
The ContentView change only hooks shortcut editing into registration cleanup; it doesn't change the layout or controls.
Notes
Plain keys, modifier-only shortcuts, Fn combinations, and mouse buttons keep the existing event-tap path. If native registration fails, the app logs the error and falls back to the tap. Those paths can still be affected by Secure Input.
This doesn't disable Secure Input or override another app's ownership of a shortcut. Cancel and paste-last also keep their existing context-dependent behavior.
docs/REGISTERED_HOTKEYS.mdincludes the design and remaining acceptance checks.Tests/run_registered_hotkey_tests.sh --liveruns a short Option+Space-only probe without recording audio.