Shell 11: core-keybinds-actions-combos (#803) #854

Closed
charles wants to merge 0 commits from shell-11-core-keybinds-actions-combos into shell
Owner

What this builds

fluxer-core::shell::keybinds (spec §8.1, §8.2; plan Task 11):

  • ShortcutAction::{id, from_id, label, kind, global_capable, default_combos} with the original's ids, labels and defaults (feat/input/state/InputKeybind.ts:679-728, 851-922; labels from feat/input/state/input_keybind/shared.ts and feat/voice/utils/VoiceMessageDescriptors.ts). ShortcutKind::{Press, Hold}.
  • Keybinds { custom } in app.shell.keybinds: effective, stored (by id, ALL order), apply_stored (same entry shape as Task 7's StoredKeybinds; unknown ids and combos without a trigger skipped), ptt_key / set_ptt_key.
  • Matcher (app.shell.matcher), a port of D/GlobalShortcutMatcher.ts: the modifier rule, the layout key, numpad by code only, mouse buttons, modifier-only and both sides, reconcileModifierFlags, release keyed on the key that pressed, set_bindings releasing changed bindings, reset, and set_paused (releases only while the recorder is open). Two combos of one action press it once and release it on the last.
  • armed(&App) (RuntimeKeybinds.ts:29-36): every press action, plus push-to-talk in push-to-talk mode or push-to-mute in voice activity. The matcher holds the global combos of the armed, global-capable actions, kept in sync after every event.
  • conflicts(&Keybinds) (combosLooseEqual).
  • KeybindAction: Record, Captured, Clear, SetGlobal, ResetDefaults, CancelRecord.
  • ShellEvent::ShortcutInput → matcher → ShellEvent::Shortcut follow-ups. A pressed Shortcut (matched or from the portal) → ToggleSelfMute, ToggleSelfDeaf, LeaveVoice or the transmit-mode toggle. Session(Locked | Suspending) and Backend(Stopped | Failed) reset the matcher and release every global hold.
  • voice::transmit hold sources (HoldSources.ts): HoldSource::{Window, Global}. Push-to-talk's release delay starts only when the last source lets go, and a focus loss releases only the window's source. Push-to-mute is voice-activity-only, mutes while held (targetMute = held || self_mute, through effective_mute and the announced self_mute) and releases at once. UserAction::PttKey stays as the window source.
  • VoiceSettings::ptt_key is the first push-to-talk combo. A key set on Voice & video, or loaded from an older file, becomes that combo (global, as setPrimaryCustomKeybindCombo). Editing push-to-talk's combos writes the key back through SetVoiceSetting(PttKey). A mouse button is named Mouse{n}, so push-to-talk still applies with a button-only binding.

Strings copied from the original: the ten action labels above ("Toggle mute", "Toggle deafen", "Push to talk", "Push to mute", "Toggle voice activity", "Disconnect from voice", "Answer the incoming call", "Decline the incoming call", "Toggle the soundboard", "Expand or collapse compact call view").

How it was tested

  • crates/fluxer-core/tests/shell_keybinds.rs has the plan's tests: actions_have_original_ids_and_defaults, the matcher_cases table (bare key with modifiers held, exact modifiers, layout key, numpad, mouse buttons, modifier-only, both sides, a flag-dropped modifier, release after a modifier change, two combos of one action), only_mode_holds_armed, ptt_release_delay_only_on_last_source, push_to_mute_ors_with_self_mute, conflicts_found, ptt_key_setting_is_first_combo, lock_and_backend_stop_release_all, the proptest any_input_sequence_then_reset_leaves_nothing_held and combo_debug_prints_no_key. Extra tests: matcher_bindings_changed_while_held_release, press_actions_map_to_voice_actions and keybind_actions_edit_the_bindings.
  • The existing push-to-talk tests in voice_settings.rs pass unchanged.
  • FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci passes locally. No tests that need a desktop: this is core only.

Deviations from the plan

  • Persistence of Keybinds. LocalSettings::keybinds (StoredKeybinds) comes from Task 7, which is not in this branch's base. So the keybinds are not read from or written to settings.toml here. Only the push-to-talk key persists, through [voice.settings].ptt_key. Keybinds::apply_stored takes the same entry shape as StoredKeybinds, so the join is one line on load and one in persist once both are on shell.
  • voice/settings.rs is unchanged. The two-way ptt_key sync lives in the keybinds reducer instead. voice/local.rs changed by one line, so that effective_mute includes push-to-mute.
  • No key tables in core. hookKeyNameToDomCode accepts any DOM code of the original's key tables, and those tables belong to fluxer-platform (Task 25). Here, a name shaped like a DOM code stands for one. With back ends that report layout_key only for single characters (the ShortcutInput doc), the behaviour is the same.
  • tests/shell_ports.rs. The stub-reducer test no longer sends the keybind, shortcut and raw-input events, which now have behaviour. The default-state test ignores the matcher, which holds the default global combos once loaded.
  • Recorder input. Raw back-end inputs are not turned into recorded combos here. The recorder takes Captured from the UI, and back-end capture is Task 12/34's Capture command.

Not run here

Nothing is platform-specific, so there is nothing for Charles to check by hand.

Base: 158c90dd (origin/shell, which already contains Task 6). The base gate had one failure, the known flaky window_input_updates_last_input (#709), which passed on five reruns.

Stacked on: shell-6-core-shell-ports-settings (already merged into shell)

Closes #803

🤖 Generated with Claude Code

https://claude.ai/code/session_0161qMQMsCyJhjYw4gqa6e4K

## What this builds `fluxer-core::shell::keybinds` (spec §8.1, §8.2; plan Task 11): - `ShortcutAction::{id, from_id, label, kind, global_capable, default_combos}` with the original's ids, labels and defaults (`feat/input/state/InputKeybind.ts:679-728, 851-922`; labels from `feat/input/state/input_keybind/shared.ts` and `feat/voice/utils/VoiceMessageDescriptors.ts`). `ShortcutKind::{Press, Hold}`. - `Keybinds { custom }` in `app.shell.keybinds`: `effective`, `stored` (by id, `ALL` order), `apply_stored` (same entry shape as Task 7's `StoredKeybinds`; unknown ids and combos without a trigger skipped), `ptt_key` / `set_ptt_key`. - `Matcher` (`app.shell.matcher`), a port of `D/GlobalShortcutMatcher.ts`: the modifier rule, the layout key, numpad by code only, mouse buttons, modifier-only and both sides, `reconcileModifierFlags`, release keyed on the key that pressed, `set_bindings` releasing changed bindings, `reset`, and `set_paused` (releases only while the recorder is open). Two combos of one action press it once and release it on the last. - `armed(&App)` (`RuntimeKeybinds.ts:29-36`): every press action, plus push-to-talk in push-to-talk mode or push-to-mute in voice activity. The matcher holds the global combos of the armed, global-capable actions, kept in sync after every event. - `conflicts(&Keybinds)` (`combosLooseEqual`). - `KeybindAction`: `Record`, `Captured`, `Clear`, `SetGlobal`, `ResetDefaults`, `CancelRecord`. - `ShellEvent::ShortcutInput` → matcher → `ShellEvent::Shortcut` follow-ups. A pressed `Shortcut` (matched or from the portal) → `ToggleSelfMute`, `ToggleSelfDeaf`, `LeaveVoice` or the transmit-mode toggle. `Session(Locked | Suspending)` and `Backend(Stopped | Failed)` reset the matcher and release every global hold. - `voice::transmit` hold sources (`HoldSources.ts`): `HoldSource::{Window, Global}`. Push-to-talk's release delay starts only when the last source lets go, and a focus loss releases only the window's source. Push-to-mute is voice-activity-only, mutes while held (`targetMute = held || self_mute`, through `effective_mute` and the announced `self_mute`) and releases at once. `UserAction::PttKey` stays as the window source. - `VoiceSettings::ptt_key` is the first push-to-talk combo. A key set on Voice & video, or loaded from an older file, becomes that combo (global, as `setPrimaryCustomKeybindCombo`). Editing push-to-talk's combos writes the key back through `SetVoiceSetting(PttKey)`. A mouse button is named `Mouse{n}`, so push-to-talk still applies with a button-only binding. Strings copied from the original: the ten action labels above ("Toggle mute", "Toggle deafen", "Push to talk", "Push to mute", "Toggle voice activity", "Disconnect from voice", "Answer the incoming call", "Decline the incoming call", "Toggle the soundboard", "Expand or collapse compact call view"). ## How it was tested - `crates/fluxer-core/tests/shell_keybinds.rs` has the plan's tests: `actions_have_original_ids_and_defaults`, the `matcher_cases` table (bare key with modifiers held, exact modifiers, layout key, numpad, mouse buttons, modifier-only, both sides, a flag-dropped modifier, release after a modifier change, two combos of one action), `only_mode_holds_armed`, `ptt_release_delay_only_on_last_source`, `push_to_mute_ors_with_self_mute`, `conflicts_found`, `ptt_key_setting_is_first_combo`, `lock_and_backend_stop_release_all`, the proptest `any_input_sequence_then_reset_leaves_nothing_held` and `combo_debug_prints_no_key`. Extra tests: `matcher_bindings_changed_while_held_release`, `press_actions_map_to_voice_actions` and `keybind_actions_edit_the_bindings`. - The existing push-to-talk tests in `voice_settings.rs` pass unchanged. - `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci` passes locally. No tests that need a desktop: this is core only. ## Deviations from the plan - **Persistence of `Keybinds`.** `LocalSettings::keybinds` (`StoredKeybinds`) comes from Task 7, which is not in this branch's base. So the keybinds are not read from or written to `settings.toml` here. Only the push-to-talk key persists, through `[voice.settings].ptt_key`. `Keybinds::apply_stored` takes the same entry shape as `StoredKeybinds`, so the join is one line on load and one in `persist` once both are on `shell`. - **`voice/settings.rs` is unchanged.** The two-way `ptt_key` sync lives in the keybinds reducer instead. `voice/local.rs` changed by one line, so that `effective_mute` includes push-to-mute. - **No key tables in core.** `hookKeyNameToDomCode` accepts any DOM code of the original's key tables, and those tables belong to `fluxer-platform` (Task 25). Here, a name shaped like a DOM code stands for one. With back ends that report `layout_key` only for single characters (the `ShortcutInput` doc), the behaviour is the same. - **`tests/shell_ports.rs`.** The stub-reducer test no longer sends the keybind, shortcut and raw-input events, which now have behaviour. The default-state test ignores the matcher, which holds the default global combos once loaded. - **Recorder input.** Raw back-end inputs are not turned into recorded combos here. The recorder takes `Captured` from the UI, and back-end capture is Task 12/34's `Capture` command. ## Not run here Nothing is platform-specific, so there is nothing for Charles to check by hand. Base: `158c90dd` (origin/shell, which already contains Task 6). The base gate had one failure, the known flaky `window_input_updates_last_input` (#709), which passed on five reruns. Stacked on: shell-6-core-shell-ports-settings (already merged into shell) Closes #803 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_0161qMQMsCyJhjYw4gqa6e4K
Refs #803

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0161qMQMsCyJhjYw4gqa6e4K
ShortcutAction ids, labels, kinds and defaults as the original
(feat/input/state/InputKeybind.ts), Keybinds with stored/apply_stored and
effective, the matcher of D/GlobalShortcutMatcher.ts, the holds armed per
transmit mode (RuntimeKeybinds.ts), conflicts (combosLooseEqual), and hold
sources in voice::transmit (HoldSources.ts): window and global push-to-talk,
push-to-mute, release on lock, suspend and back end stop. VoiceSettings::ptt_key
is kept as the first push-to-talk combo.

Refs #803

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0161qMQMsCyJhjYw4gqa6e4K
Author
Owner

Independent review: Shell 11 (keybinds: actions, combos, matcher, holds)

Reviewed git diff 158c90dd..origin/shell-11-core-keybinds-actions-combos (head 65b860c3) cold.

Plan tests: every named test exists in tests/shell_keybinds.rs and asserts what the plan says: actions_have_original_ids_and_defaults, the matcher_cases table (bare key, exact modifiers, layout key/AZERTY, numpad by code, mouse buttons, modifier-only, both sides, release after a modifier change), only_mode_holds_armed, ptt_release_delay_only_on_last_source, push_to_mute_ors_with_self_mute, conflicts_found, ptt_key_setting_is_first_combo, lock_and_backend_stop_release_all, the proptest any_input_sequence_then_reset_leaves_nothing_held and combo_debug_prints_no_key.
Interfaces: ShortcutAction::{ALL, id, label, kind, global_capable, default_combos}, Keybinds { custom } with stored/apply_stored/effective, Matcher::{feed, reset}, armed(&App), conflicts(&Keybinds), HoldSource::{Window, Global}, the KeybindAction variants: all present as written.
Invariants: core imports nothing forbidden and does no I/O. No tracing call near key data. Debug of Keybinds, Matcher and ShellState prints no combo (tested). The only slice index (combos[index + 1..]) is in bounds. apply_stored skips unknown ids and combos with nothing to trigger on, so file data cannot panic.
Parity checks against the fork: defaults and global: true (InputKeybind.ts:679-728, 851-922); bindingFromCombo, keyTriggerMatches, modifiersMatch and handleKey/handleMouse incl. RELEASE_ONLY (fluxer_desktop/src/main/GlobalShortcutMatcher.ts); the per-mode hold lists (RuntimeKeybinds.ts:29-36); PTT delay on the last source and PTM released at once (HoldSources.ts:36-56); labels (Disconnect from voice, Toggle the soundboard, ...); setPrimaryCustomKeybindCombo global default (UserVoiceTab.tsx:358-367). All match.
The implementer's three questions: (a) the US-layout code for a combo without a code is the original's own combo.code ?? usLayoutKeyToDomCode(combo.key), so it is correct. (b) Follow-up ordering is sound: matched ShortcutEvents are applied after the input event, so reduce_transmit takes the holds and keybinds::reduce the press actions, in that order. On lock and back-end stop, transmit releases first and the matcher-reset releases that follow do nothing. (c) Positional source ids: see finding 1.
Review Focus 2/3 (this task's part): every global hold is released on lock, suspend and back-end stop or failure, including portal-only holds that never went through the matcher. The proptest covers reset.

Findings

  1. minor (parity): Matcher::set_bindings keys a press on its position in the binding list. The original keys it on sourceId plus bindingSignature. So when an earlier action's global combos change while a hold is down, a still-held binding whose combo did not change is released. Example: toggling Toggle mute's "Global shortcut" while push-to-talk is held starts the PTT release delay. The original keeps it held. This errs towards releasing, never towards sticking, and needs a settings edit mid-hold. It can be fixed by comparing the new list's equal Binding (action, trigger and modifiers) instead of the one at the same index.
  2. minor (process): Keybinds is not yet joined to LocalSettings ([keybinds] comes from Task 7, same wave), and no follow-up issue for the join is linked from this PR. CLAUDE.md requires one before merge. Until the join, a PTT combo with modifiers, modifier-only or both-sides is rebuilt from the bare ptt_key code at the next start (the Loaded sync). No UI can create such combos before Task 36, so no user data is lost today.
  3. minor (scope note): Session(Locked | Suspending) and Backend(Stopped) release the Global source only. A window PTT key still held relies on WindowFocus(false) to be released. The plan's "release everything" is met for global holds. The window source is Task 35's (focus loss), so this is recorded only.
  4. minor (docs): the README voice section still says push-to-mute is "Not carried over from the original". It now is (the line after the changed PTT paragraph).
  5. minor: VoiceSettings::ptt_key can now be Mouse{n}. Until Task 35 replaces is_ptt_key, the in-window path and the Voice & video field treat it as an unmatched key name. This is harmless, but the field displays it raw.

Gate: FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci in a fresh review worktree: all steps passed (exit 0; shell_keybinds 13 tests + proptest ok).

Verdict: APPROVED (0 blocking, 5 minor; please open the follow-up issue for finding 2 before merge).

## Independent review: Shell 11 (keybinds: actions, combos, matcher, holds) Reviewed `git diff 158c90dd..origin/shell-11-core-keybinds-actions-combos` (head 65b860c3) cold. **Plan tests:** every named test exists in `tests/shell_keybinds.rs` and asserts what the plan says: `actions_have_original_ids_and_defaults`, the `matcher_cases` table (bare key, exact modifiers, layout key/AZERTY, numpad by code, mouse buttons, modifier-only, both sides, release after a modifier change), `only_mode_holds_armed`, `ptt_release_delay_only_on_last_source`, `push_to_mute_ors_with_self_mute`, `conflicts_found`, `ptt_key_setting_is_first_combo`, `lock_and_backend_stop_release_all`, the proptest `any_input_sequence_then_reset_leaves_nothing_held` and `combo_debug_prints_no_key`. **Interfaces:** `ShortcutAction::{ALL, id, label, kind, global_capable, default_combos}`, `Keybinds { custom }` with `stored`/`apply_stored`/`effective`, `Matcher::{feed, reset}`, `armed(&App)`, `conflicts(&Keybinds)`, `HoldSource::{Window, Global}`, the `KeybindAction` variants: all present as written. **Invariants:** core imports nothing forbidden and does no I/O. No `tracing` call near key data. `Debug` of `Keybinds`, `Matcher` and `ShellState` prints no combo (tested). The only slice index (`combos[index + 1..]`) is in bounds. `apply_stored` skips unknown ids and combos with nothing to trigger on, so file data cannot panic. **Parity checks against the fork:** defaults and `global: true` (`InputKeybind.ts:679-728, 851-922`); `bindingFromCombo`, `keyTriggerMatches`, `modifiersMatch` and `handleKey`/`handleMouse` incl. RELEASE_ONLY (`fluxer_desktop/src/main/GlobalShortcutMatcher.ts`); the per-mode hold lists (`RuntimeKeybinds.ts:29-36`); PTT delay on the last source and PTM released at once (`HoldSources.ts:36-56`); labels (`Disconnect from voice`, `Toggle the soundboard`, ...); `setPrimaryCustomKeybindCombo` global default (`UserVoiceTab.tsx:358-367`). All match. **The implementer's three questions:** (a) the US-layout code for a combo without a code is the original's own `combo.code ?? usLayoutKeyToDomCode(combo.key)`, so it is correct. (b) Follow-up ordering is sound: matched `ShortcutEvent`s are applied after the input event, so `reduce_transmit` takes the holds and `keybinds::reduce` the press actions, in that order. On lock and back-end stop, transmit releases first and the matcher-reset releases that follow do nothing. (c) Positional source ids: see finding 1. **Review Focus 2/3 (this task's part):** every global hold is released on lock, suspend and back-end stop or failure, including portal-only holds that never went through the matcher. The proptest covers reset. ### Findings 1. **minor (parity):** `Matcher::set_bindings` keys a press on its position in the binding list. The original keys it on `sourceId` plus `bindingSignature`. So when an earlier action's global combos change while a hold is down, a still-held binding whose combo did not change is released. Example: toggling Toggle mute's "Global shortcut" while push-to-talk is held starts the PTT release delay. The original keeps it held. This errs towards releasing, never towards sticking, and needs a settings edit mid-hold. It can be fixed by comparing the new list's equal `Binding` (action, trigger and modifiers) instead of the one at the same index. 2. **minor (process):** `Keybinds` is not yet joined to `LocalSettings` (`[keybinds]` comes from Task 7, same wave), and no `follow-up` issue for the join is linked from this PR. CLAUDE.md requires one before merge. Until the join, a PTT combo with modifiers, modifier-only or both-sides is rebuilt from the bare `ptt_key` code at the next start (the `Loaded` sync). No UI can create such combos before Task 36, so no user data is lost today. 3. **minor (scope note):** `Session(Locked | Suspending)` and `Backend(Stopped)` release the `Global` source only. A window PTT key still held relies on `WindowFocus(false)` to be released. The plan's "release everything" is met for global holds. The window source is Task 35's (focus loss), so this is recorded only. 4. **minor (docs):** the README voice section still says push-to-mute is "Not carried over from the original". It now is (the line after the changed PTT paragraph). 5. **minor:** `VoiceSettings::ptt_key` can now be `Mouse{n}`. Until Task 35 replaces `is_ptt_key`, the in-window path and the Voice & video field treat it as an unmatched key name. This is harmless, but the field displays it raw. **Gate:** `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci` in a fresh review worktree: **all steps passed** (exit 0; `shell_keybinds` 13 tests + proptest ok). **Verdict: APPROVED** (0 blocking, 5 minor; please open the follow-up issue for finding 2 before merge).
Author
Owner

Merged into shell by hand (the merge API returns 405); the commits are in the branch.

Merged into `shell` by hand (the merge API returns 405); the commits are in the branch.
charles closed this pull request 2026-11-18 06:40:05 +00:00
Author
Owner

Independent re-review: Shell 11 (keybinds: actions, combos, matcher, holds)

Scope: git diff 158c90dd6..origin/shell-11-core-keybinds-actions-combos (head 26a635c05), Task 11's own commits 80a9222ec, 65b860c30, d215c7968, 26a635c05. The Task 1 and Task 9 code that was merged in was not reviewed here.

Gate: FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci in a fresh review worktree: all steps passed (exit 0). shell_keybinds: 13 of 13 pass.

What I checked

  1. Plan tests: all ten exist and assert what the plan asks: actions_have_original_ids_and_defaults, the matcher_cases table (bare key with modifiers, exact modifiers, layout key (AZERTY), numpad, mouse buttons, modifier-only, both sides, release after a modifier change, and the extras: stale flags, two combos of one action), only_mode_holds_armed, ptt_release_delay_only_on_last_source, push_to_mute_ors_with_self_mute, conflicts_found, ptt_key_setting_is_first_combo, lock_and_backend_stop_release_all (Locked, Suspending, Stopped, Failed), the proptest any_input_sequence_then_reset_leaves_nothing_held, and combo_debug_prints_no_key.
  2. Interfaces: ShortcutAction::{ALL, id, from_id, label, kind, global_capable, default_combos}, ShortcutKind::{Press, Hold}, Keybinds { custom } with effective/stored/apply_stored, Matcher::{feed, reset, set_bindings, set_paused, is_held}, armed(&App), conflicts(&Keybinds) -> Vec<(action, action)>, HoldSource::{Window, Global}, and the KeybindAction variants all exist as the plan writes them. KeyCombo has the plan's ten fields.
  3. Invariants: no iced, wgpu, winit, reqwest, tokio-tungstenite or I/O in fluxer-core. No tracing, log or println in the diff. Debug of Keybinds and Matcher prints no combo, and Matcher's prints only counts and actions. No new fixtures, no secrets.
  4. Parity spot checks against the fork (all match):
    • ids, defaults and global: true for mute and deafen: InputKeybind.ts:679-728, 851-922. Ctrl+Shift+M/D global; Ctrl+Enter, Escape, Ctrl+Shift+B, Ctrl+Shift+V with KeyV; empty key for push-to-talk, push-to-mute, toggle voice activity and disconnect.
    • labels: VoiceMessageDescriptors.ts and the *_DESCRIPTORs ("Toggle deafen", "Disconnect from voice", "Expand or collapse compact call view", ...).
    • HOLD_ACTIONS_FOR_PTT_MODE / _VOICE_ACTIVITY_MODE: RuntimeKeybinds.ts:29-36.
    • D/GlobalShortcutMatcher.ts: modifiersMatch, keyTriggerMatches, heldModifierCode, reconcileModifierFlags, releaseUnsatisfiedModifierOnly, satisfiedModifierOnlyBindings, isModifierOnlySatisfied, releaseAll and the RELEASE_ONLY policy are ported line for line.
  5. Review Focus 2 and 3: the proptest resets after random input in both transmit modes. Push-to-talk is let go on lock, suspend, back-end stop and back-end failure (global source), and on focus loss (window source only). The release delay starts only when the last source lets go, and push-to-mute releases at once. The KDE debounce belongs to Task 28.

Findings

  • [minor] crates/fluxer-core/src/shell/keybinds.rs Matcher::set_bindings: a held binding is identified by its position in the list, while the original uses sourceId plus bindingSignature. Suppose push-to-talk is held through a global combo and the user edits an earlier action's combos (for example, removes Toggle mute's). Push-to-talk's binding moves to a new index, so it is released and is not pressed again until the key is pressed again. This was raised in the first review. It is rare and fails safe (the mic mutes), so it is not blocking.
  • [minor] Cross-task gap, now on shell: LocalSettings::keybinds (StoredKeybinds, Shell 7) is read and written by fluxer-platform but is never joined to app.shell.keybinds. Nothing calls Keybinds::apply_stored with it on load, and keybind edits are never written back. Every binding recorded in Settings -> Keybinds except the push-to-talk key is therefore lost at restart. The README of this PR states the deferral, but no follow-up issue existed, so I opened #925 (label follow-up, milestone Desktop shell). It must land before Task 36 (Settings -> Keybinds) is usable.
  • [minor] A push-to-talk combo recorded with only a non-ASCII layout key and no code (for example key: "é", code: None) has no ptt_key name. The edit then sets VoiceSettings::ptt_key to None, which turns push-to-talk off (ptt_active needs a key), even though the combo is still bound. Back ends always report a code, so this is an edge case.
  • [minor] crates/fluxer-core/README.md still says push-to-mute is "Not carried over", but the Keybinds section added below it now implements push-to-mute. Raised before.
  • [minor] The plan lists voice/settings.rs as modified. ptt_key is instead kept in sync by the keybind reducer (on Loaded and SetVoiceSetting(PttKey)). The behaviour asked for ("reads and writes the first custom PushToTalk combo") holds and is tested. Recorded only so the deviation is visible.

Verdict: APPROVED (0 blocking, 5 minor; gate green). Deferred work: #925.

## Independent re-review: Shell 11 (keybinds: actions, combos, matcher, holds) Scope: `git diff 158c90dd6..origin/shell-11-core-keybinds-actions-combos` (head 26a635c05), Task 11's own commits 80a9222ec, 65b860c30, d215c7968, 26a635c05. The Task 1 and Task 9 code that was merged in was not reviewed here. **Gate:** `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci` in a fresh review worktree: **all steps passed** (exit 0). `shell_keybinds`: 13 of 13 pass. **What I checked** 1. Plan tests: all ten exist and assert what the plan asks: `actions_have_original_ids_and_defaults`, the `matcher_cases` table (bare key with modifiers, exact modifiers, layout key (AZERTY), numpad, mouse buttons, modifier-only, both sides, release after a modifier change, and the extras: stale flags, two combos of one action), `only_mode_holds_armed`, `ptt_release_delay_only_on_last_source`, `push_to_mute_ors_with_self_mute`, `conflicts_found`, `ptt_key_setting_is_first_combo`, `lock_and_backend_stop_release_all` (Locked, Suspending, Stopped, Failed), the proptest `any_input_sequence_then_reset_leaves_nothing_held`, and `combo_debug_prints_no_key`. 2. Interfaces: `ShortcutAction::{ALL, id, from_id, label, kind, global_capable, default_combos}`, `ShortcutKind::{Press, Hold}`, `Keybinds { custom }` with `effective`/`stored`/`apply_stored`, `Matcher::{feed, reset, set_bindings, set_paused, is_held}`, `armed(&App)`, `conflicts(&Keybinds) -> Vec<(action, action)>`, `HoldSource::{Window, Global}`, and the `KeybindAction` variants all exist as the plan writes them. `KeyCombo` has the plan's ten fields. 3. Invariants: no iced, wgpu, winit, reqwest, tokio-tungstenite or I/O in `fluxer-core`. No `tracing`, `log` or `println` in the diff. `Debug` of `Keybinds` and `Matcher` prints no combo, and `Matcher`'s prints only counts and actions. No new fixtures, no secrets. 4. Parity spot checks against the fork (all match): - ids, defaults and `global: true` for mute and deafen: `InputKeybind.ts:679-728, 851-922`. Ctrl+Shift+M/D global; Ctrl+Enter, Escape, Ctrl+Shift+B, Ctrl+Shift+V with `KeyV`; empty `key` for push-to-talk, push-to-mute, toggle voice activity and disconnect. - labels: `VoiceMessageDescriptors.ts` and the `*_DESCRIPTOR`s ("Toggle deafen", "Disconnect from voice", "Expand or collapse compact call view", ...). - `HOLD_ACTIONS_FOR_PTT_MODE` / `_VOICE_ACTIVITY_MODE`: `RuntimeKeybinds.ts:29-36`. - `D/GlobalShortcutMatcher.ts`: `modifiersMatch`, `keyTriggerMatches`, `heldModifierCode`, `reconcileModifierFlags`, `releaseUnsatisfiedModifierOnly`, `satisfiedModifierOnlyBindings`, `isModifierOnlySatisfied`, `releaseAll` and the RELEASE_ONLY policy are ported line for line. 5. Review Focus 2 and 3: the proptest resets after random input in both transmit modes. Push-to-talk is let go on lock, suspend, back-end stop and back-end failure (global source), and on focus loss (window source only). The release delay starts only when the last source lets go, and push-to-mute releases at once. The KDE debounce belongs to Task 28. **Findings** - [minor] `crates/fluxer-core/src/shell/keybinds.rs` `Matcher::set_bindings`: a held binding is identified by its position in the list, while the original uses `sourceId` plus `bindingSignature`. Suppose push-to-talk is held through a global combo and the user edits an earlier action's combos (for example, removes Toggle mute's). Push-to-talk's binding moves to a new index, so it is released and is not pressed again until the key is pressed again. This was raised in the first review. It is rare and fails safe (the mic mutes), so it is not blocking. - [minor] Cross-task gap, now on `shell`: `LocalSettings::keybinds` (`StoredKeybinds`, Shell 7) is read and written by `fluxer-platform` but is never joined to `app.shell.keybinds`. Nothing calls `Keybinds::apply_stored` with it on load, and keybind edits are never written back. Every binding recorded in Settings -> Keybinds except the push-to-talk key is therefore lost at restart. The README of this PR states the deferral, but no follow-up issue existed, so I opened **#925** (label `follow-up`, milestone Desktop shell). It must land before Task 36 (Settings -> Keybinds) is usable. - [minor] A push-to-talk combo recorded with only a non-ASCII layout key and no `code` (for example `key: "é"`, `code: None`) has no `ptt_key` name. The edit then sets `VoiceSettings::ptt_key` to `None`, which turns push-to-talk off (`ptt_active` needs a key), even though the combo is still bound. Back ends always report a `code`, so this is an edge case. - [minor] `crates/fluxer-core/README.md` still says push-to-mute is "Not carried over", but the Keybinds section added below it now implements push-to-mute. Raised before. - [minor] The plan lists `voice/settings.rs` as modified. `ptt_key` is instead kept in sync by the keybind reducer (on `Loaded` and `SetVoiceSetting(PttKey)`). The behaviour asked for ("reads and writes the first custom PushToTalk combo") holds and is tested. Recorded only so the deviation is visible. **Verdict: APPROVED** (0 blocking, 5 minor; gate green). Deferred work: #925.

Pull request closed

Sign in to join this conversation.
No description provided.