Shell 35: ui-in-window-voice (#827) #898

Closed
charles wants to merge 0 commits from shell-35-ui-in-window-voice into shell
Owner

Stacked on: shell-11-core-keybinds-actions-combos, shell-32-ui-daemon-window-lifecycle (both already contained in origin/shell when this branch was cut; the merges were no-ops)

What this builds (plan Task 35, spec §8.4)

  • app::window_input is now the window's event filter: every key and mouse-button event (not left/right click) becomes WindowEvent::Input { input, captured, then }, with input a fluxer_core::shell::ShortcutInput and then what the old window_event makes of it.
  • One table from iced's physical key codes to DOM codes (keybinds::dom_code: iced's code names, Super* → Meta*); mouse buttons by DOM number (keybinds::mouse_button_number: 1 middle, 3 back, 4 forward); the layout's character as layout_key, so "M" follows the layout.
  • keybinds::VoiceKeys (in Local) runs the core Matcher over keybinds::window_bindings(&App): logged in, the combos of app.shell.keybinds for the armed actions (shell::keybinds::armed), filtered by available (Answer/Decline only while a call rings, Toggle soundboard / compact call view only in voice) and by delivered_by_backend (a back end is Started, the action is global-capable and the combo's global switch is on; on the portal a hold with a modified trigger stays the window's, spec §8.4). Paused while Settings → Keybinds records.
  • Typing: the ten actions have the original's DEFAULT Typing (voice_typing); a press a widget captured counts only with Ctrl or Alt, and its release is dropped too, so a bare-letter push-to-talk key does not fire while typing (as today).
  • keybinds::voice_actions: push-to-talk → UserAction::PttKey (window hold source), push-to-mute → ShellEvent::Shortcut, mute / deafen / disconnect / transmit-mode toggle → the store's actions, answer / decline on the ringing call, the soundboard popout, the compact call's chat (local.call.chat_open, the header's chat button, as COMPACT_VOICE_CALL_EXPANSION_TOGGLE). A press the voice keys took goes no further (Escape declines a ringing call instead of running the ladder; otherwise the ladder is unchanged). Everything held is let go on focus loss (KeybindManager.ts:889-893).
  • The fixed is_ptt_key path in app.rs is removed.

Tests

  • crates/fluxer-ui/tests/keybinds_voice.rs (8 tests, through TestDaemon with real iced events): ctrl_shift_m_and_d_toggle, ctrl_enter_answers_and_escape_declines_only_while_ringing, soundboard_and_compact_only_in_voice, custom_binding_replaces_default, ptt_hold_and_focus_loss_release, bare_key_not_while_typing, mouse_side_button_binding, not_double_fired_when_backend_active. Seen failing (8/8) with the matcher's output cut off, passing with it.
  • Round 3 fixes: debug_prints_no_held_key (Debug of VoiceKeys and Local holding a key and a mouse button prints neither the raw code, decimal or hex, nor Some(3) nor KeyV; a window input message prints neither code nor text), hold_released_when_no_longer_armed (PTT released on the snapshot that switches to voice activity; the answer chord no longer held once the call stops ringing), and fluxer-core voice::settings::tests::stored_logo_key_reads_as_the_dom_code. All three seen failing on the merged tree before the fix.
  • FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci green locally. No desktop-gated tests apply; none run. No screenshots changed.

Deviations / notes for the reviewer

  • Touches crates/fluxer-ui/src/message.rs (one new WindowEvent::Input variant), outside the task's Files list; needed to carry the input.
  • "A text input has the focus" is read as "a widget captured the key" (the same proxy as today's push-to-talk path), not the async focus operation the chord table uses, so presses and releases stay in order.
  • Window push-to-mute goes as ShellEvent::Shortcut (core has no window source for it); the window's matcher releases it on focus loss itself.
  • ShortcutInput/KeyCombo are never logged; VoiceKeys has a manual Debug (the matcher's redacted one, the suppressed actions, only taken.len(), finish_non_exhaustive), and WindowEvent::Input's then is a Then newtype printing Then(…), so neither VoiceKeys, Local nor a window message prints a key (spec §12).
  • A push-to-talk key stored as SuperLeft/SuperRight (the native recorder's earlier names) reads as MetaLeft/MetaRight in VoiceSettings::from_stored (fluxer-core, one place), so it keeps matching after the recorder switched to the DOM names.
  • VoiceKeys::sync also runs with every snapshot: a hold whose action is no longer armed or available (transmit mode left push-to-talk, the ringing call ended) is released at once, not on the next input.
  • A custom voice binding on Ctrl+E/G/S is not seen by the window (those chords stay the expression picker's).
  • Windows: nothing platform-specific; the new test file is cfg(not(windows)) like the other shell suites.

Strings copied: none (no new UI text).

Closes #827

🤖 Generated with Claude Code

https://claude.ai/code/session_0161qMQMsCyJhjYw4gqa6e4K

Stacked on: shell-11-core-keybinds-actions-combos, shell-32-ui-daemon-window-lifecycle (both already contained in origin/shell when this branch was cut; the merges were no-ops) ## What this builds (plan Task 35, spec §8.4) - `app::window_input` is now the window's event filter: every key and mouse-button event (not left/right click) becomes `WindowEvent::Input { input, captured, then }`, with `input` a `fluxer_core::shell::ShortcutInput` and `then` what the old `window_event` makes of it. - One table from iced's physical key codes to DOM codes (`keybinds::dom_code`: iced's code names, `Super*` → `Meta*`); mouse buttons by DOM number (`keybinds::mouse_button_number`: 1 middle, 3 back, 4 forward); the layout's character as `layout_key`, so "M" follows the layout. - `keybinds::VoiceKeys` (in `Local`) runs the core `Matcher` over `keybinds::window_bindings(&App)`: logged in, the combos of `app.shell.keybinds` for the armed actions (`shell::keybinds::armed`), filtered by `available` (Answer/Decline only while a call rings, Toggle soundboard / compact call view only in voice) and by `delivered_by_backend` (a back end is `Started`, the action is global-capable and the combo's global switch is on; on the portal a hold with a modified trigger stays the window's, spec §8.4). Paused while Settings → Keybinds records. - Typing: the ten actions have the original's `DEFAULT` `Typing` (`voice_typing`); a press a widget captured counts only with Ctrl or Alt, and its release is dropped too, so a bare-letter push-to-talk key does not fire while typing (as today). - `keybinds::voice_actions`: push-to-talk → `UserAction::PttKey` (window hold source), push-to-mute → `ShellEvent::Shortcut`, mute / deafen / disconnect / transmit-mode toggle → the store's actions, answer / decline on the ringing call, the soundboard popout, the compact call's chat (`local.call.chat_open`, the header's chat button, as `COMPACT_VOICE_CALL_EXPANSION_TOGGLE`). A press the voice keys took goes no further (Escape declines a ringing call instead of running the ladder; otherwise the ladder is unchanged). Everything held is let go on focus loss (`KeybindManager.ts:889-893`). - The fixed `is_ptt_key` path in `app.rs` is removed. ## Tests - `crates/fluxer-ui/tests/keybinds_voice.rs` (8 tests, through `TestDaemon` with real iced events): `ctrl_shift_m_and_d_toggle`, `ctrl_enter_answers_and_escape_declines_only_while_ringing`, `soundboard_and_compact_only_in_voice`, `custom_binding_replaces_default`, `ptt_hold_and_focus_loss_release`, `bare_key_not_while_typing`, `mouse_side_button_binding`, `not_double_fired_when_backend_active`. Seen failing (8/8) with the matcher's output cut off, passing with it. - Round 3 fixes: `debug_prints_no_held_key` (Debug of `VoiceKeys` and `Local` holding a key and a mouse button prints neither the raw code, decimal or hex, nor `Some(3)` nor `KeyV`; a window input message prints neither code nor text), `hold_released_when_no_longer_armed` (PTT released on the snapshot that switches to voice activity; the answer chord no longer held once the call stops ringing), and fluxer-core `voice::settings::tests::stored_logo_key_reads_as_the_dom_code`. All three seen failing on the merged tree before the fix. - `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci` green locally. No desktop-gated tests apply; none run. No screenshots changed. ## Deviations / notes for the reviewer - Touches `crates/fluxer-ui/src/message.rs` (one new `WindowEvent::Input` variant), outside the task's Files list; needed to carry the input. - "A text input has the focus" is read as "a widget captured the key" (the same proxy as today's push-to-talk path), not the async focus operation the chord table uses, so presses and releases stay in order. - Window push-to-mute goes as `ShellEvent::Shortcut` (core has no window source for it); the window's matcher releases it on focus loss itself. - `ShortcutInput`/`KeyCombo` are never logged; `VoiceKeys` has a manual `Debug` (the matcher's redacted one, the suppressed actions, only `taken.len()`, `finish_non_exhaustive`), and `WindowEvent::Input`'s `then` is a `Then` newtype printing `Then(…)`, so neither `VoiceKeys`, `Local` nor a window message prints a key (spec §12). - A push-to-talk key stored as `SuperLeft`/`SuperRight` (the native recorder's earlier names) reads as `MetaLeft`/`MetaRight` in `VoiceSettings::from_stored` (fluxer-core, one place), so it keeps matching after the recorder switched to the DOM names. - `VoiceKeys::sync` also runs with every snapshot: a hold whose action is no longer armed or available (transmit mode left push-to-talk, the ringing call ended) is released at once, not on the next input. - A custom voice binding on Ctrl+E/G/S is not seen by the window (those chords stay the expression picker's). - Windows: nothing platform-specific; the new test file is `cfg(not(windows))` like the other shell suites. Strings copied: none (no new UI text). Closes #827 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_0161qMQMsCyJhjYw4gqa6e4K
The window feeds its key and mouse-button events, as ShortcutInputs (DOM
codes from iced's key codes through one table), to the core matcher over
the bindings of app.shell.keybinds for the ten actions of spec §8.1,
leaving to a running back end the global combos it delivers. Answer and
decline only while a call rings (Escape keeps its ladder otherwise), the
soundboard and the compact call view only in voice, holds released on
focus loss, a bare key from a text input does not fire. Replaces the
fixed is_ptt_key path.

Refs #827

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

Review: Shell 35 (ui-in-window-voice), commit 2dfe7e52

Scope: git diff d48106dc..origin/shell-35-ui-in-window-voice (README, app.rs, keybinds.rs, message.rs, tests/keybinds_voice.rs).

Checked and fine

  • All 8 tests the plan names exist in crates/fluxer-ui/tests/keybinds_voice.rs. They drive real iced events through TestDaemon and assert the spec §8.1 defaults: Ctrl+Shift+M/D, Ctrl+Enter, Escape, Ctrl+Shift+B, Ctrl+Shift+V, and mouse buttons 3 and 4.
  • Interfaces: is_ptt_key is removed. Window key and mouse events reach the core Matcher through one dom_code table. Answer and Decline work only while calls.incoming is set. Soundboard and compact view work only in voice. Holds are released on Unfocused. Escape declines while a call rings and skips the ladder; otherwise the ladder runs (the test asserts both).
  • Invariants: no I/O was added to fluxer-ui. No tracing:: call is near key data. ShortcutInput's and Matcher's Debug output is redacted.
  • Parity spot-checks against the fork:
    • InputKeybind.ts:680-727 (defaults and global flags) matches.
    • KeybindManager.ts:457-469 (isActionAvailableForLocalShortcut) matches.
    • KeybindManager.ts:880-893 (blur releases local holds) matches.
  • Gate: FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci -> EXIT=0. The first run failed only on fluxer-platform::update_download::errors_never_contain_the_url (TooLarge vs Network). That test reuses a released port, which another process on this machine can bind, so the failure is unrelated to this diff. The rerun was fully green, keybinds_voice included.

Findings

  1. [blocking] delivered_by_backend treats Started(Portal) as "the portal delivers every global combo", whatever the portal's state (keybinds.rs, delivered_by_backend). Core's wanted() runs the portal whenever it is the session's selection, not only once it is bound. So PortalState::NotSetUp, Declined, Binding, Reconnecting, Error and Unsupported all exist while the back end runs, and in each of them the portal delivers nothing.

    • What breaks: on a fresh Wayland session, the global-on defaults Ctrl+Shift+M and Ctrl+Shift+D do nothing, in the window or anywhere else. An unmodified push-to-talk key does nothing either. That includes one set on the Voice page or migrated from an older file, because set_ptt_key makes it global, as setPrimaryCustomKeybindCombo does.
    • This regresses the old is_ptt_key path, which always worked in the window. It also contradicts the original's strings that core already ships: PTT_NOT_SET_UP ("Push-to-talk only works while {productName} is focused until you set up system-wide shortcuts"), PTT_DECLINED, and the row hint WORKS_WHILE_FOCUSED.
    • The same happens once the portal is Bound: an action with no assigned trigger (SYSTEM_WIDE_NOT_ASSIGNED), or with a trigger the desktop changed, loses its in-app key.
    • For reference, the original never routes holds through hooks on the portal (KeybindManager.ts:592-606, hooksActive && backend !== 'portal').
    • Suggested fix: on BackendKind::Portal, count a combo as delivered only when app.shell.system_shortcuts.portal.state is Bound { triggers } and triggers.get(action) is Some. Keep the modified-hold exception. Add a test case for Started(Portal) with NotSetUp.
  2. [minor] Two key-name tables. The window now matches dom_code (SuperLeft -> MetaLeft), but the Voice page still records keys with processing::key_name (format!("{code:?}"), which gives SuperLeft). A push-to-talk key recorded on Super on the Voice page therefore never fires, though it did before this change. The plan says "through one table": key_name should use keybinds::dom_code (or migrate SuperLeft/SuperRight).

  3. [minor] Typing rule vs the original. voice_typing returns DEFAULT for all ten actions, so a key a widget captured fires only with Ctrl or Alt. The original lets every voice_and_video keybind through from an editable element (KeybindManager.ts:1985-1986, shouldAllowLocalShortcutFromEditable). Two consequences:

    • Escape in the focused composer while a call rings does not decline the call.
    • A bare-key custom Toggle mute does not fire while typing.

    The plan explicitly keeps only the bare-key push-to-talk rule, so this is a parity gap, not a plan violation. Record it in the PR or open a follow-up issue.

  4. [minor] Misplaced doc comment (app.rs, around fn voice_events). The new function was inserted between send's doc comment and send. voice_events now carries send's "Puts event into the runtime's queue…" text, and send has no doc comment.

Implementer deviations 1 to 5 (message.rs variant, the captured-as-typing proxy, push-to-mute as ShellEvent::Shortcut, Super->Meta, Ctrl+E/G/S left to the picker) are acceptable apart from the points above. Deviation 4 is finding 2.

Verdict: changes requested (1 blocking: in-window voice keys are dead on an unbound portal).

## Review: Shell 35 (ui-in-window-voice), commit 2dfe7e52 Scope: `git diff d48106dc..origin/shell-35-ui-in-window-voice` (README, app.rs, keybinds.rs, message.rs, tests/keybinds_voice.rs). **Checked and fine** - All 8 tests the plan names exist in `crates/fluxer-ui/tests/keybinds_voice.rs`. They drive real iced events through `TestDaemon` and assert the spec §8.1 defaults: Ctrl+Shift+M/D, Ctrl+Enter, Escape, Ctrl+Shift+B, Ctrl+Shift+V, and mouse buttons 3 and 4. - Interfaces: `is_ptt_key` is removed. Window key and mouse events reach the core `Matcher` through one `dom_code` table. Answer and Decline work only while `calls.incoming` is set. Soundboard and compact view work only in voice. Holds are released on `Unfocused`. Escape declines while a call rings and skips the ladder; otherwise the ladder runs (the test asserts both). - Invariants: no I/O was added to fluxer-ui. No `tracing::` call is near key data. `ShortcutInput`'s and `Matcher`'s `Debug` output is redacted. - Parity spot-checks against the fork: - `InputKeybind.ts:680-727` (defaults and global flags) matches. - `KeybindManager.ts:457-469` (`isActionAvailableForLocalShortcut`) matches. - `KeybindManager.ts:880-893` (blur releases local holds) matches. - Gate: `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci` -> EXIT=0. The first run failed only on `fluxer-platform::update_download::errors_never_contain_the_url` (TooLarge vs Network). That test reuses a released port, which another process on this machine can bind, so the failure is unrelated to this diff. The rerun was fully green, keybinds_voice included. **Findings** 1. **[blocking] `delivered_by_backend` treats `Started(Portal)` as "the portal delivers every global combo", whatever the portal's state** (keybinds.rs, `delivered_by_backend`). Core's `wanted()` runs the portal whenever it is the session's selection, not only once it is bound. So `PortalState::NotSetUp`, `Declined`, `Binding`, `Reconnecting`, `Error` and `Unsupported` all exist while the back end runs, and in each of them the portal delivers nothing. - What breaks: on a fresh Wayland session, the global-on defaults Ctrl+Shift+M and Ctrl+Shift+D do nothing, in the window or anywhere else. An unmodified push-to-talk key does nothing either. That includes one set on the Voice page or migrated from an older file, because `set_ptt_key` makes it global, as `setPrimaryCustomKeybindCombo` does. - This regresses the old `is_ptt_key` path, which always worked in the window. It also contradicts the original's strings that core already ships: `PTT_NOT_SET_UP` ("Push-to-talk only works while {productName} is focused until you set up system-wide shortcuts"), `PTT_DECLINED`, and the row hint `WORKS_WHILE_FOCUSED`. - The same happens once the portal is `Bound`: an action with no assigned trigger (`SYSTEM_WIDE_NOT_ASSIGNED`), or with a trigger the desktop changed, loses its in-app key. - For reference, the original never routes holds through hooks on the portal (`KeybindManager.ts:592-606`, `hooksActive && backend !== 'portal'`). - Suggested fix: on `BackendKind::Portal`, count a combo as delivered only when `app.shell.system_shortcuts.portal.state` is `Bound { triggers }` and `triggers.get(action)` is `Some`. Keep the modified-hold exception. Add a test case for `Started(Portal)` with `NotSetUp`. 2. **[minor] Two key-name tables.** The window now matches `dom_code` (`SuperLeft` -> `MetaLeft`), but the Voice page still records keys with `processing::key_name` (`format!("{code:?}")`, which gives `SuperLeft`). A push-to-talk key recorded on Super on the Voice page therefore never fires, though it did before this change. The plan says "through one table": `key_name` should use `keybinds::dom_code` (or migrate `SuperLeft`/`SuperRight`). 3. **[minor] Typing rule vs the original.** `voice_typing` returns `DEFAULT` for all ten actions, so a key a widget captured fires only with Ctrl or Alt. The original lets every `voice_and_video` keybind through from an editable element (`KeybindManager.ts:1985-1986`, `shouldAllowLocalShortcutFromEditable`). Two consequences: - Escape in the focused composer while a call rings does not decline the call. - A bare-key custom Toggle mute does not fire while typing. The plan explicitly keeps only the bare-key push-to-talk rule, so this is a parity gap, not a plan violation. Record it in the PR or open a follow-up issue. 4. **[minor] Misplaced doc comment** (app.rs, around `fn voice_events`). The new function was inserted between `send`'s doc comment and `send`. `voice_events` now carries `send`'s "Puts `event` into the runtime's queue…" text, and `send` has no doc comment. Implementer deviations 1 to 5 (message.rs variant, the captured-as-typing proxy, push-to-mute as `ShellEvent::Shortcut`, Super->Meta, Ctrl+E/G/S left to the picker) are acceptable apart from the points above. Deviation 4 is finding 2. **Verdict: changes requested** (1 blocking: in-window voice keys are dead on an unbound portal).
The window left every global combo to the portal back end whenever it ran,
though the portal delivers nothing until bound and nothing for an action
the desktop gave no trigger: on a fresh Wayland session Ctrl+Shift+M/D and
a global push-to-talk key did nothing. Count a combo as delivered on the
portal only when it is Bound with a trigger for the action.

The Voice page now records keys through keybinds::dom_code, so a Super key
is stored as MetaLeft, as the window matches it. send's doc comment is back
on send.

Refs #827

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0161qMQMsCyJhjYw4gqa6e4K
Author
Owner

Review round 1 addressed in 11befa0f (local FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci green).

Blocking, delivered_by_backend on the portal: fixed. On Started(Portal) a combo now counts as delivered only when system_shortcuts.portal.state is Bound and triggers.get(action) is Some; NotSetUp, Binding, Declined, Reconnecting, Error and Unsupported leave every combo to the window, as does a bound portal for an action without a trigger. The hold-with-modifiers rule is unchanged on top. New test portal_not_bound_leaves_keys_to_the_window (Ctrl+Shift+M and a bare global F8 push-to-talk key across five unbound states, plus bound-without-trigger for mute); not_double_fired_when_backend_active now binds the portal with triggers for its portal case.

Minor, two key-name tables: fixed. processing::key_name now goes through keybinds::dom_code, so a Super key records as MetaLeft. Test voice_page_records_keys_as_the_window_matches_them.

Minor, voice_typing parity gap: deferred to follow-up #907 (the original's voice_and_video section passes through from editable elements; doing it also replaces the bare_key_not_while_typing test that pins today's behaviour).

Minor, doc comment in app.rs: fixed; the queue doc is back on send, and voice_events keeps only its own.

README's voice-keys paragraph updated to match.

Review round 1 addressed in 11befa0f (local `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci` green). **Blocking, `delivered_by_backend` on the portal:** fixed. On `Started(Portal)` a combo now counts as delivered only when `system_shortcuts.portal.state` is `Bound` and `triggers.get(action)` is `Some`; NotSetUp, Binding, Declined, Reconnecting, Error and Unsupported leave every combo to the window, as does a bound portal for an action without a trigger. The hold-with-modifiers rule is unchanged on top. New test `portal_not_bound_leaves_keys_to_the_window` (Ctrl+Shift+M and a bare global F8 push-to-talk key across five unbound states, plus bound-without-trigger for mute); `not_double_fired_when_backend_active` now binds the portal with triggers for its portal case. **Minor, two key-name tables:** fixed. `processing::key_name` now goes through `keybinds::dom_code`, so a Super key records as `MetaLeft`. Test `voice_page_records_keys_as_the_window_matches_them`. **Minor, `voice_typing` parity gap:** deferred to follow-up #907 (the original's `voice_and_video` section passes through from editable elements; doing it also replaces the `bare_key_not_while_typing` test that pins today's behaviour). **Minor, doc comment in `app.rs`:** fixed; the queue doc is back on `send`, and `voice_events` keeps only its own. README's voice-keys paragraph updated to match.
Author
Owner

Independent review, Task 35 (round 2), commit 11befa0f

Reviewed git diff d48106dc..origin/shell-35-ui-in-window-voice cold. Local FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci on 11befa0f: all steps passed.

What checks out:

  • All eight tests the plan names exist in crates/fluxer-ui/tests/keybinds_voice.rs and assert what the plan says. The round-1 fix (portal_not_bound_leaves_keys_to_the_window, a portal that is bound but has no trigger) is right: delivered_by_backend now needs Started(kind), the action to be global-capable, combo.global and, on the portal, a bound trigger.
  • Interfaces: is_ptt_key is gone. Keys go through one table (dom_code, which processing::key_name uses too). Answer/Decline only work while calls.incoming is set, and Escape keeps its ladder otherwise. Soundboard and compact view only work in voice. Holds are released on Unfocused. Recording pauses the matcher.
  • Invariants: nothing in fluxer-ui does I/O. ShortcutInput/KeyCombo Debug output is redacted. No tracing:: near key data. No secrets.
  • Parity spot-checks against the fork: the defaults Ctrl+Shift+M/D (InputKeybind.ts:681-694), Ctrl+Enter / Escape / Ctrl+Shift+B / Ctrl+Shift+V (InputKeybind.ts:697-723), answer/decline gated on an incoming call (KeybindManager.ts:457-468), local holds released on blur (KeybindManager.ts:868-887), and soundboard/compact as press-only handlers (defaultHandlers.ts:273-282).

Findings

  1. [blocking] Holding a bare printable push-to-talk key types it into the composer (a regression). In app.rs WindowEvent::Input, the then event is dropped only when the voice keys produced a press (took). Key-repeat presses of a held PTT key make the matcher produce nothing (correctly), so then (WindowEvent::Key { text: Some("v") }) runs, and the Daily use 37b type-to-start path writes the letter into the composer and focuses it. The base's is_ptt_key took every press of the PTT key, repeats included. I reproduced it with a throwaway test on 11befa0f (PTT on KeyV, push-to-talk mode, then a press with text: Some("v") followed by two repeat: true presses): first=[PttKey { down: true }], repeats=[ComposerEdited { text: "v" }, ComposerEdited { text: "vv" }]. So talking with a letter or backquote PTT key fills the user's draft once key repeat starts. The suite misses it because key_event always builds text: None. Fix: also swallow presses of a key the matcher holds as the press of a matched combo (for example, keep the raw codes of active presses in VoiceKeys and report took for their repeats), and add a test with text: Some(..) and repeat: true.
  2. [minor] What "a hold on the portal with a modified trigger" means is open to two readings. delivered_by_backend decides from the client combo's modifiers, not from the trigger the desktop assigned. The test binds PTT with the portal trigger CTRL+SHIFT+F12, yet expects bare F8 to belong to the portal. Spec §8.4 does not say which one; worth one line in the doc comment, or a question to settle it.
  3. [minor] The compact call view toggle acts even when the call is not the channel on screen. The original dispatches COMPACT_VOICE_CALL_EXPANSION_TOGGLE with currentChannelId, and only the open voice channel or DM view listens (GuildChannelView.tsx:263-271). Here local.call.chat_open flips whenever the user is in voice, even while another channel is shown.
  4. [minor] mouse_button_number maps Button::Other(n) straight to n. winit's Other values are platform button indices, not DOM MouseEvent.button numbers. This is harmless today (the recorder offers 1, 3 and 4 only), but a combo stored by the original with another number would not match.
  5. [minor, deferred] The editable-focus parity gap (the original lets voice_and_video shortcuts through from editables, KeybindManager.ts:1985-1987) is tracked in #907. That is fine.

Verdict: changes requested. One blocking finding (1). CI is green.

## Independent review, Task 35 (round 2), commit 11befa0f Reviewed `git diff d48106dc..origin/shell-35-ui-in-window-voice` cold. Local `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci` on 11befa0f: **all steps passed**. What checks out: - All eight tests the plan names exist in `crates/fluxer-ui/tests/keybinds_voice.rs` and assert what the plan says. The round-1 fix (`portal_not_bound_leaves_keys_to_the_window`, a portal that is bound but has no trigger) is right: `delivered_by_backend` now needs `Started(kind)`, the action to be global-capable, `combo.global` and, on the portal, a bound trigger. - Interfaces: `is_ptt_key` is gone. Keys go through one table (`dom_code`, which `processing::key_name` uses too). Answer/Decline only work while `calls.incoming` is set, and Escape keeps its ladder otherwise. Soundboard and compact view only work in voice. Holds are released on `Unfocused`. Recording pauses the matcher. - Invariants: nothing in `fluxer-ui` does I/O. `ShortcutInput`/`KeyCombo` Debug output is redacted. No `tracing::` near key data. No secrets. - Parity spot-checks against the fork: the defaults Ctrl+Shift+M/D (`InputKeybind.ts:681-694`), Ctrl+Enter / Escape / Ctrl+Shift+B / Ctrl+Shift+V (`InputKeybind.ts:697-723`), answer/decline gated on an incoming call (`KeybindManager.ts:457-468`), local holds released on blur (`KeybindManager.ts:868-887`), and soundboard/compact as press-only handlers (`defaultHandlers.ts:273-282`). ### Findings 1. **[blocking] Holding a bare printable push-to-talk key types it into the composer (a regression).** In `app.rs` `WindowEvent::Input`, the `then` event is dropped only when the voice keys produced a *press* (`took`). Key-repeat presses of a held PTT key make the matcher produce nothing (correctly), so `then` (`WindowEvent::Key { text: Some("v") }`) runs, and the Daily use 37b type-to-start path writes the letter into the composer and focuses it. The base's `is_ptt_key` took every press of the PTT key, repeats included. I reproduced it with a throwaway test on 11befa0f (PTT on KeyV, push-to-talk mode, then a press with `text: Some("v")` followed by two `repeat: true` presses): `first=[PttKey { down: true }]`, `repeats=[ComposerEdited { text: "v" }, ComposerEdited { text: "vv" }]`. So talking with a letter or backquote PTT key fills the user's draft once key repeat starts. The suite misses it because `key_event` always builds `text: None`. Fix: also swallow presses of a key the matcher holds as the press of a matched combo (for example, keep the raw codes of active presses in `VoiceKeys` and report `took` for their repeats), and add a test with `text: Some(..)` and `repeat: true`. 2. **[minor] What "a hold on the portal with a modified trigger" means is open to two readings.** `delivered_by_backend` decides from the *client* combo's modifiers, not from the trigger the desktop assigned. The test binds PTT with the portal trigger `CTRL+SHIFT+F12`, yet expects bare F8 to belong to the portal. Spec §8.4 does not say which one; worth one line in the doc comment, or a question to settle it. 3. **[minor] The compact call view toggle acts even when the call is not the channel on screen.** The original dispatches `COMPACT_VOICE_CALL_EXPANSION_TOGGLE` with `currentChannelId`, and only the open voice channel or DM view listens (`GuildChannelView.tsx:263-271`). Here `local.call.chat_open` flips whenever the user is in voice, even while another channel is shown. 4. **[minor] `mouse_button_number` maps `Button::Other(n)` straight to `n`.** winit's `Other` values are platform button indices, not DOM `MouseEvent.button` numbers. This is harmless today (the recorder offers 1, 3 and 4 only), but a combo stored by the original with another number would not match. 5. **[minor, deferred]** The editable-focus parity gap (the original lets `voice_and_video` shortcuts through from editables, `KeybindManager.ts:1985-1987`) is tracked in #907. That is fine. **Verdict: changes requested.** One blocking finding (1). CI is green.
- VoiceKeys::feed answers Fed { events, took }: it remembers the keys and
  buttons whose press it took until they go up, so their key repeats are
  taken too and a held bare push-to-talk key no longer types into the
  composer.
- delivered_by_backend reads "a hold with a modified trigger" (spec 8.4)
  from the trigger the desktop assigned, as the original's
  triggerHasModifiers does, not from the combo's modifiers.
- ToggleCompactCallView toggles the chat only with the voice channel's call
  open (GuildChannelView.tsx:263-271).
- mouse_button_number maps only middle, back and forward; winit's other
  indices are not DOM numbers.

Refs #827

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0161qMQMsCyJhjYw4gqa6e4K
Author
Owner

Round 2 review fixes, commit d47c90a3 (cargo xtask ci with screenshots green locally):

  • Blocking (app.rs: held bare PTT key types into the composer): fixed. VoiceKeys::feed now returns keybinds::Fed { events, took }. It remembers each key or button whose press it took until that key goes up, so the key repeats of a held push-to-talk key are taken too and never reach the type-to-start path. A focus loss clears the list. New test held_ptt_key_repeats_do_not_type sends real presses with text: Some("v") and repeat: true: no ComposerEdited while PTT is held, a different key still types, and the same key types normally when PTT is not armed. It failed on 11befa0f with [ComposerEdited { text: "v" }].
  • Minor (delivered_by_backend, portal hold with a modified trigger): settled. The function now reads the trigger the desktop assigned, using the original rule triggerHasModifiers (SystemShortcutsSection.tsx:269-271: longer than one character and containing +), instead of the combo's modifiers. If the trigger has modifiers, every combo of that hold stays with the window. If the trigger is a bare key, the portal handles the hold. Tests updated: a CTRL+SHIFT+F12 trigger leaves bare F8 to the window, and an F9 trigger gives the portal both F8 and Ctrl+V.
  • Minor (ToggleCompactCallView acting anywhere in voice): fixed. The shortcut only flips local.call.chat_open when the open view is the call of the voice channel the user is in (call::model is Call). This matches GuildChannelView.tsx:263-271. The native client has no compact-call chat toggle for DMs, so the shortcut does nothing there. The test now checks both cases: another channel open (no change) and the voice channel open (toggles).
  • Minor (mouse_button_number Other(n)): fixed. Button::Other(_) now maps to None, so only middle, back and forward (1, 3, 4) can be bound. New test only_the_dom_mouse_buttons_bind.
  • Minor (voice_typing parity gap): left as is, deferred to #907 as agreed.

README updated to match.

Round 2 review fixes, commit d47c90a3 (`cargo xtask ci` with screenshots green locally): - **Blocking (app.rs: held bare PTT key types into the composer):** fixed. `VoiceKeys::feed` now returns `keybinds::Fed { events, took }`. It remembers each key or button whose press it took until that key goes up, so the key repeats of a held push-to-talk key are taken too and never reach the type-to-start path. A focus loss clears the list. New test `held_ptt_key_repeats_do_not_type` sends real presses with `text: Some("v")` and `repeat: true`: no `ComposerEdited` while PTT is held, a different key still types, and the same key types normally when PTT is not armed. It failed on 11befa0f with `[ComposerEdited { text: "v" }]`. - **Minor (delivered_by_backend, portal hold with a modified trigger):** settled. The function now reads the trigger the desktop assigned, using the original rule `triggerHasModifiers` (`SystemShortcutsSection.tsx:269-271`: longer than one character and containing `+`), instead of the combo's modifiers. If the trigger has modifiers, every combo of that hold stays with the window. If the trigger is a bare key, the portal handles the hold. Tests updated: a `CTRL+SHIFT+F12` trigger leaves bare F8 to the window, and an `F9` trigger gives the portal both F8 and Ctrl+V. - **Minor (ToggleCompactCallView acting anywhere in voice):** fixed. The shortcut only flips `local.call.chat_open` when the open view is the call of the voice channel the user is in (`call::model` is `Call`). This matches `GuildChannelView.tsx:263-271`. The native client has no compact-call chat toggle for DMs, so the shortcut does nothing there. The test now checks both cases: another channel open (no change) and the voice channel open (toggles). - **Minor (mouse_button_number Other(n)):** fixed. `Button::Other(_)` now maps to `None`, so only middle, back and forward (1, 3, 4) can be bound. New test `only_the_dom_mouse_buttons_bind`. - **Minor (voice_typing parity gap):** left as is, deferred to #907 as agreed. README updated to match.
Author
Owner

Independent review, round 3: Task 35 (ui-in-window-voice) at d47c90a3

Scope: git diff d48106dc..origin/shell-35-ui-in-window-voice (keybinds.rs, app.rs, message.rs, voice_video/processing.rs, README, tests/keybinds_voice.rs).

Gate: FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci run locally in a fresh review worktree on d47c90a3: all steps passed (EXIT=0). keybinds_voice ran, no failures.

Checked and OK

  • All eight tests the plan names exist and assert what it says (ctrl_shift_m_and_d_toggle, ctrl_enter_answers_and_escape_declines_only_while_ringing, soundboard_and_compact_only_in_voice, custom_binding_replaces_default, ptt_hold_and_focus_loss_release, bare_key_not_while_typing, mouse_side_button_binding, not_double_fired_when_backend_active). There are also new tests for the round-2 fixes: held_ptt_key_repeats_do_not_type, portal_not_bound_leaves_keys_to_the_window, only_the_dom_mouse_buttons_bind.
  • Interfaces: the fixed is_ptt_key path has been removed. Keys and mouse buttons now go through one table (dom_code, mouse_button_number) into the core Matcher over armed actions that are available and not delivered_by_backend. Answer and decline only work while a call rings, so Escape keeps its ladder otherwise. The soundboard and compact view work only in voice. Holds are released on Unfocused.
  • Round-2 blocker is fixed: key repeats of a held, taken PTT key no longer reach the composer.
  • Parity spot-checks against the fork:
    • The blur releases local holds as KeybindManager.ts onWindowBlur does (~l.870-893).
    • triggerHasModifiers matches SystemShortcutsSection.tsx:269-271 (length > 1 && includes('+')).
    • Answer and decline are gated on the incoming call, as isActionAvailableForLocalShortcut (KeybindManager.ts:457-460) does.
    • The DOM mouse numbers are 1 middle, 3 back and 4 forward.
  • Crate rules: no new I/O in fluxer-ui and no tracing near key data. ShortcutInput, KeyCombo, Modifiers and Matcher keep their redacted Debug.

Findings

  1. [blocking] VoiceKeys derives Debug, and that output contains key identifiers (crates/fluxer-ui/src/keybinds.rs, #[derive(Clone, Debug, Default)] pub struct VoiceKeys).
    • The new taken: Vec<(u32, Option<u8>)> field holds raw_code, an FNV-1a hash of the DOM code, plus the mouse button number. The hash is trivially reversible over the small set of key codes.
    • Global Constraints (spec §12) say: "no key code, name or text in ... Debug output". The struct's own doc comment says "Its Debug prints no key".
    • Local (which derives Debug) now carries this field, so any {:?} of Local would print which key is held.
    • Nothing logs it today, but this is exactly the invariant Review Focus 1 guards.
    • Fix: a manual Debug impl like Matcher's, e.g. print matcher, suppressed and taken.len() and use finish_non_exhaustive. A one-line test asserting the Debug output of a VoiceKeys holding a key contains no digits of its raw code would also help.
  2. [minor] Stored push-to-talk keys named SuperLeft/SuperRight are no longer matched (voice_video/processing.rs::key_name).
    • The recorder now stores the logo key as MetaLeft/MetaRight, which matches the original's KeyboardEvent.code.
    • A value the native client recorded before this change (SuperLeft) silently stops matching.
    • Consider mapping Super* to Meta* when the setting is read, or note it in the PR.
  3. [minor] The bindings are re-synced only on the next input (VoiceKeys::sync runs from feed).
    • If the armed set changes while a hold is down, the matcher's release for it comes only with the next key or mouse event. Examples: the transmit mode switches away from push-to-talk, or a ringing call ends while the answer chord is held.
    • Harmless for PttKey, because the core ignores it outside PTT mode, but worth a comment or a follow-up.

Verdict: changes requested. One blocking finding (key data in Debug, #1). The fix is small. Everything else, CI included, is green.

## Independent review, round 3: Task 35 (ui-in-window-voice) at d47c90a3 Scope: `git diff d48106dc..origin/shell-35-ui-in-window-voice` (keybinds.rs, app.rs, message.rs, voice_video/processing.rs, README, tests/keybinds_voice.rs). **Gate:** `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci` run locally in a fresh review worktree on d47c90a3: **all steps passed (EXIT=0)**. keybinds_voice ran, no failures. **Checked and OK** - All eight tests the plan names exist and assert what it says (`ctrl_shift_m_and_d_toggle`, `ctrl_enter_answers_and_escape_declines_only_while_ringing`, `soundboard_and_compact_only_in_voice`, `custom_binding_replaces_default`, `ptt_hold_and_focus_loss_release`, `bare_key_not_while_typing`, `mouse_side_button_binding`, `not_double_fired_when_backend_active`). There are also new tests for the round-2 fixes: `held_ptt_key_repeats_do_not_type`, `portal_not_bound_leaves_keys_to_the_window`, `only_the_dom_mouse_buttons_bind`. - Interfaces: the fixed `is_ptt_key` path has been removed. Keys and mouse buttons now go through one table (`dom_code`, `mouse_button_number`) into the core `Matcher` over `armed` actions that are `available` and not `delivered_by_backend`. Answer and decline only work while a call rings, so Escape keeps its ladder otherwise. The soundboard and compact view work only in voice. Holds are released on `Unfocused`. - Round-2 blocker is fixed: key repeats of a held, taken PTT key no longer reach the composer. - Parity spot-checks against the fork: - The blur releases local holds as `KeybindManager.ts` onWindowBlur does (~l.870-893). - `triggerHasModifiers` matches `SystemShortcutsSection.tsx:269-271` (`length > 1 && includes('+')`). - Answer and decline are gated on the incoming call, as `isActionAvailableForLocalShortcut` (`KeybindManager.ts:457-460`) does. - The DOM mouse numbers are 1 middle, 3 back and 4 forward. - Crate rules: no new I/O in fluxer-ui and no `tracing` near key data. `ShortcutInput`, `KeyCombo`, `Modifiers` and `Matcher` keep their redacted Debug. **Findings** 1. **[blocking] `VoiceKeys` derives `Debug`, and that output contains key identifiers** (`crates/fluxer-ui/src/keybinds.rs`, `#[derive(Clone, Debug, Default)] pub struct VoiceKeys`). - The new `taken: Vec<(u32, Option<u8>)>` field holds `raw_code`, an FNV-1a hash of the DOM code, plus the mouse button number. The hash is trivially reversible over the small set of key codes. - Global Constraints (spec §12) say: "no key code, name or text in ... `Debug` output". The struct's own doc comment says "Its `Debug` prints no key". - `Local` (which derives Debug) now carries this field, so any `{:?}` of `Local` would print which key is held. - Nothing logs it today, but this is exactly the invariant Review Focus 1 guards. - Fix: a manual `Debug` impl like `Matcher`'s, e.g. print `matcher`, `suppressed` and `taken.len()` and use `finish_non_exhaustive`. A one-line test asserting the Debug output of a `VoiceKeys` holding a key contains no digits of its raw code would also help. 2. **[minor] Stored push-to-talk keys named `SuperLeft`/`SuperRight` are no longer matched** (`voice_video/processing.rs::key_name`). - The recorder now stores the logo key as `MetaLeft`/`MetaRight`, which matches the original's `KeyboardEvent.code`. - A value the native client recorded before this change (`SuperLeft`) silently stops matching. - Consider mapping `Super*` to `Meta*` when the setting is read, or note it in the PR. 3. **[minor] The bindings are re-synced only on the next input** (`VoiceKeys::sync` runs from `feed`). - If the armed set changes while a hold is down, the matcher's release for it comes only with the next key or mouse event. Examples: the transmit mode switches away from push-to-talk, or a ringing call ends while the answer chord is held. - Harmless for `PttKey`, because the core ignores it outside PTT mode, but worth a comment or a follow-up. **Verdict: changes requested.** One blocking finding (key data in `Debug`, #1). The fix is small. Everything else, CI included, is green.
Round 3 findings of PR #898:
- VoiceKeys gets a manual Debug (matcher, suppressed, taken.len(),
  finish_non_exhaustive): the derived one printed the taken keys' raw
  codes and mouse button numbers (spec §12). WindowEvent::Input's `then`
  is now a `Then` newtype whose Debug prints `Then(…)`, not the key event
  with its code and text.
- A push-to-talk key stored as SuperLeft/SuperRight (the native
  recorder's earlier names) reads as MetaLeft/MetaRight in
  VoiceSettings::from_stored, so it keeps matching.
- VoiceKeys::sync runs with every snapshot, so a hold whose action is no
  longer armed or available (transmit mode left push-to-talk, the ringing
  call ended) is released at once, not on the next input.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Author
Owner

Round 3 findings fixed: all in a9cab9e5 (after merging origin/shell at 6a0eac65).

  1. [blocking] Key data in VoiceKeys's Debug: manual Debug in Matcher's style (matcher, suppressed, taken as its length, finish_non_exhaustive). Test debug_prints_no_held_key (keybinds_voice.rs): with PTT on KeyV and Deafen on mouse Back both held, {:?} of VoiceKeys and of Local contain neither the raw code (decimal, hex lower and upper), nor Some(3), nor KeyV. It failed on the derived impl (taken: [(992671918, None), (0, Some(3))]).
    Audit of the other types this branch added or changed: WindowEvent::Input printed its then (a WindowEvent::Key/Chord with code and text). Its then is now message::Then(Box<WindowEvent>), whose Debug prints Then(…); the same test checks that a window input message prints neither KeyV nor "v". Checked and fine: Fed (actions and bools only); ShortcutInput, KeyCombo, Matcher (already redacted); ShortcutEvent (action and pressed only); Local (fine now through VoiceKeys); processing::key_name and dom_code return values and are never logged. There is no tracing and no Display in the branch's diff.
    Older, not this branch's: WindowEvent::Key/Chord/ChordIn and keybinds::Chord derive Debug with code and text (Text chat / Daily use 35), and so do VoiceSettings/VoiceSettingChange::PttKey (core). Untouched here; candidates for a follow-up.
  2. [minor] Stored SuperLeft/SuperRight: mapped to MetaLeft/MetaRight in one place, VoiceSettings::take_stored (from_stored, fluxer-core), before the keybind sync and the platform's [keybinds] migration read it. Test voice::settings::tests::stored_logo_key_reads_as_the_dom_code, which failed before the fix.
  3. [minor] Re-sync only on the next input: VoiceKeys::sync is now public and runs in the shell's UiMessage::Snapshot handler, which is where the armed set changes. It is pure and cheap, because Matcher::set_bindings returns early on an unchanged set. Its releases go through voice_actions. Test hold_released_when_no_longer_armed covers both cases: PTT held, then a snapshot in voice-activity mode gives PttKey { down: false } at once and the key-up later does nothing more; the answer chord held, then the call stops ringing and AnswerCall is no longer held (new VoiceKeys::is_held). It failed before. No change outside the UI crate was needed.

Gate: FLUXER_SCREENSHOT_TESTS=1 dbus-run-session -- cargo xtask ci gave ==> xtask ci: all steps passed. READMEs (fluxer-ui, fluxer-core) and the PR description are updated.

Round 3 findings fixed: all in a9cab9e5 (after merging origin/shell at 6a0eac65). 1. **[blocking] Key data in `VoiceKeys`'s `Debug`**: manual `Debug` in `Matcher`'s style (`matcher`, `suppressed`, `taken` as its length, `finish_non_exhaustive`). Test `debug_prints_no_held_key` (keybinds_voice.rs): with PTT on `KeyV` and Deafen on mouse Back both held, `{:?}` of `VoiceKeys` and of `Local` contain neither the raw code (decimal, hex lower and upper), nor `Some(3)`, nor `KeyV`. It failed on the derived impl (`taken: [(992671918, None), (0, Some(3))]`). Audit of the other types this branch added or changed: `WindowEvent::Input` printed its `then` (a `WindowEvent::Key`/`Chord` with code and text). Its `then` is now `message::Then(Box<WindowEvent>)`, whose `Debug` prints `Then(…)`; the same test checks that a window input message prints neither `KeyV` nor `"v"`. Checked and fine: `Fed` (actions and bools only); `ShortcutInput`, `KeyCombo`, `Matcher` (already redacted); `ShortcutEvent` (action and pressed only); `Local` (fine now through `VoiceKeys`); `processing::key_name` and `dom_code` return values and are never logged. There is no `tracing` and no `Display` in the branch's diff. Older, not this branch's: `WindowEvent::Key`/`Chord`/`ChordIn` and `keybinds::Chord` derive `Debug` with code and text (Text chat / Daily use 35), and so do `VoiceSettings`/`VoiceSettingChange::PttKey` (core). Untouched here; candidates for a follow-up. 2. **[minor] Stored `SuperLeft`/`SuperRight`**: mapped to `MetaLeft`/`MetaRight` in one place, `VoiceSettings::take_stored` (`from_stored`, fluxer-core), before the keybind sync and the platform's `[keybinds]` migration read it. Test `voice::settings::tests::stored_logo_key_reads_as_the_dom_code`, which failed before the fix. 3. **[minor] Re-sync only on the next input**: `VoiceKeys::sync` is now public and runs in the shell's `UiMessage::Snapshot` handler, which is where the armed set changes. It is pure and cheap, because `Matcher::set_bindings` returns early on an unchanged set. Its releases go through `voice_actions`. Test `hold_released_when_no_longer_armed` covers both cases: PTT held, then a snapshot in voice-activity mode gives `PttKey { down: false }` at once and the key-up later does nothing more; the answer chord held, then the call stops ringing and `AnswerCall` is no longer held (new `VoiceKeys::is_held`). It failed before. No change outside the UI crate was needed. Gate: `FLUXER_SCREENSHOT_TESTS=1 dbus-run-session -- cargo xtask ci` gave `==> xtask ci: all steps passed`. READMEs (fluxer-ui, fluxer-core) and the PR description are updated.
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 22:58:23 +00:00

Pull request closed

Sign in to join this conversation.
No description provided.