Text chat 52: app-audio-devices-runtime (#342) #361

Manually merged
charles merged 0 commits from textchat-52-app-audio-devices-runtime into textchat 2026-11-13 13:10:51 +00:00
Owner

Stacked on: textchat-50-core-stream-audio-sources

What this builds

  • PipeWire initialiser order (#225.7): the real voice backend on Linux now builds CpalHost::with_pipewire_init(fluxer_capture::initialize_pipewire) (through voice_rt::system_pipewire_init()), never CpalHost::new(). A comment explains why: this binary links libwebrtc, whose pw_* stubs jump to 0 until webrtc::InitializePipeWire() ran. The chosen sound system (pipewire/alsa) is logged at debug.
  • Audio targets: CaptureCommand::ListAudioTargets runs StreamAudioCapturer::list_targets on the capture worker and answers CaptureEvent::AudioTargets(Some(nodes)) (each AudioTarget.props becomes an AudioNode), or None on Unsupported or a backend error. The failure is logged at info once, then at debug. Properties are never logged, only the count at debug.
  • Rule conversion: audio_rule passes AppMatch::Props and AudioRule.exclude through to fluxer-capture. The interim fallback (a rule made only of props became System) is gone: the core already falls back when nothing is routable.
  • Devserver: the default fake capture backend lists its default targets (Firefox, mpv, Speakers), so the submenu shows under --devserver.
  • README (voice and screen-share runtime) and the parity note on section-voice_video-input-output (#220) are updated.

Tests

  • Unit tests: rules_convert_props_and_exclude, list_audio_targets_answers_with_nodes and unsupported_listing_answers_none (in capture_rt.rs), and linux_voice_backend_uses_the_webrtc_pipewire_initialiser (in voice_rt.rs, std::ptr::fn_addr_eq). Runtime test: tests/capture_audio_targets.rs::devserver_lists_fake_audio_targets. All of them were seen failing first, then passing.
  • FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci passes locally.
  • Real PipeWire, on this machine (CachyOS, Wayland, PipeWire daemon): FLUXER_AUDIO_DEVICE_TESTS=1 cargo test -p fluxer-audio passes, including pipewire_lists_the_default_sink_and_source and opens_a_listed_pipewire_device_by_id. FLUXER_CAPTURE_TESTS=1 cargo test -p fluxer-capture --test pipewire_real passes with --test-threads=1 (6/6). With parallel threads, own_playback_not_captured and system_capture_hears_a_test_tone_from_another_process fail: the real tests hear each other's tones. That is already true on the base, and this PR does not touch that crate.
  • Not run: a real call and the mic test in the app against a live LiveKit room (no credentials or live instance in this task). Charles checks those on his desktop (audio spec §8).

Deviations

  • Removed the obsolete test property_patterns_alone_fall_back_to_system_audio. It asserted the interim fallback that this task replaces.

Closes #342

🤖 Generated with Claude Code

Stacked on: textchat-50-core-stream-audio-sources ## What this builds - **PipeWire initialiser order (#225.7):** the real voice backend on Linux now builds `CpalHost::with_pipewire_init(fluxer_capture::initialize_pipewire)` (through `voice_rt::system_pipewire_init()`), never `CpalHost::new()`. A comment explains why: this binary links libwebrtc, whose `pw_*` stubs jump to 0 until `webrtc::InitializePipeWire()` ran. The chosen sound system (`pipewire`/`alsa`) is logged at `debug`. - **Audio targets:** `CaptureCommand::ListAudioTargets` runs `StreamAudioCapturer::list_targets` on the capture worker and answers `CaptureEvent::AudioTargets(Some(nodes))` (each `AudioTarget.props` becomes an `AudioNode`), or `None` on `Unsupported` or a backend error. The failure is logged at `info` once, then at `debug`. Properties are never logged, only the count at `debug`. - **Rule conversion:** `audio_rule` passes `AppMatch::Props` and `AudioRule.exclude` through to `fluxer-capture`. The interim fallback (a rule made only of props became System) is gone: the core already falls back when nothing is routable. - **Devserver:** the default fake capture backend lists its default targets (Firefox, mpv, Speakers), so the submenu shows under `--devserver`. - README (voice and screen-share runtime) and the parity note on `section-voice_video-input-output` (#220) are updated. ## Tests - Unit tests: `rules_convert_props_and_exclude`, `list_audio_targets_answers_with_nodes` and `unsupported_listing_answers_none` (in `capture_rt.rs`), and `linux_voice_backend_uses_the_webrtc_pipewire_initialiser` (in `voice_rt.rs`, `std::ptr::fn_addr_eq`). Runtime test: `tests/capture_audio_targets.rs::devserver_lists_fake_audio_targets`. All of them were seen failing first, then passing. - `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci` passes locally. - Real PipeWire, on this machine (CachyOS, Wayland, PipeWire daemon): `FLUXER_AUDIO_DEVICE_TESTS=1 cargo test -p fluxer-audio` passes, including `pipewire_lists_the_default_sink_and_source` and `opens_a_listed_pipewire_device_by_id`. `FLUXER_CAPTURE_TESTS=1 cargo test -p fluxer-capture --test pipewire_real` passes with `--test-threads=1` (6/6). With parallel threads, `own_playback_not_captured` and `system_capture_hears_a_test_tone_from_another_process` fail: the real tests hear each other's tones. That is already true on the base, and this PR does not touch that crate. - Not run: a real call and the mic test in the app against a live LiveKit room (no credentials or live instance in this task). Charles checks those on his desktop (audio spec §8). ## Deviations - Removed the obsolete test `property_patterns_alone_fall_back_to_system_audio`. It asserted the interim fallback that this task replaces. Closes #342 🤖 Generated with [Claude Code](https://claude.com/claude-code)
Refs #340

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Until Text chat 52 routes property patterns, a "specific" rule made only
of them reached fluxer-capture as an empty application list, which WASAPI
refuses and PipeWire treats as capturing nothing. Map it to System.

Refs #340

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Refs #342

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

Review: Task 52 (app-audio-devices-runtime), diff 0761fc2..1d94385

(1) Tests: all five named tests are present and assert what the plan asks for: rules_convert_props_and_exclude, list_audio_targets_answers_with_nodes and unsupported_listing_answers_none (capture_rt.rs); linux_voice_backend_uses_the_webrtc_pipewire_initialiser (voice_rt.rs, a fn_addr_eq comparison); devserver_lists_fake_audio_targets (tests/capture_audio_targets.rs, which expects Firefox, mpv and an Audio/Sink).
(2) Interfaces: the Linux backend is built with CpalHost::with_pipewire_init(fluxer_capture::initialize_pipewire), and a comment says why. ListAudioTargets answers Some(nodes), or None on any error. A failure is logged once at info, and properties never go to the logs (only the count, at debug). audio_rule converts Props and exclude. The devserver's fake lists its default targets. All of these match the plan.
(3) Invariants: no fluxer-core, fluxer-ui or new-dependency changes. No secrets or paths are logged. The parity note on section-voice_video-input-output was added.
(4) Review Focus: none of the items belongs to Task 52.

Findings

  • minor apps/fluxer/src/capture_rt.rs audio_rule: the fallback that turned a rule of property patterns alone into System was removed. On Windows, wasapi::loopback_target returns Unsupported for an Applications list without a Pid. So if a persisted screen_share_audio_source_mode = specific with include patterns reaches a Windows build (for example from a settings file copied from Linux), the screen share has no sound instead of the system audio. The audio spec §3.3 says "WASAPI ignores both". The UI hides the submenu on Windows, so this is hard to reach. Suggestion: keep the fallback under cfg(windows), or have WASAPI treat a list without a pid as System.
  • minor apps/fluxer/README.md, Voice runtime bullet: the new sentence about the PipeWire initialiser was inserted in the middle of the existing sentence. The text now reads "...filled them (#225.7) and the LiveKit room (...)", so the LiveKit room clause hangs off the wrong sentence.

CI

I started cargo xtask ci locally in the review worktree, but it was still compiling when this review had to be returned, so I have no local result. The implementer reports FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci green. I could not verify that.

Verdict: no blocking findings in the code. Not approved by me only because I have no CI result. A green cargo xtask ci on the runners should be enough to approve.

## Review: Task 52 (app-audio-devices-runtime), diff 0761fc2..1d94385 **(1) Tests:** all five named tests are present and assert what the plan asks for: `rules_convert_props_and_exclude`, `list_audio_targets_answers_with_nodes` and `unsupported_listing_answers_none` (capture_rt.rs); `linux_voice_backend_uses_the_webrtc_pipewire_initialiser` (voice_rt.rs, a `fn_addr_eq` comparison); `devserver_lists_fake_audio_targets` (tests/capture_audio_targets.rs, which expects Firefox, mpv and an Audio/Sink). **(2) Interfaces:** the Linux backend is built with `CpalHost::with_pipewire_init(fluxer_capture::initialize_pipewire)`, and a comment says why. `ListAudioTargets` answers `Some(nodes)`, or `None` on any error. A failure is logged once at info, and properties never go to the logs (only the count, at debug). `audio_rule` converts `Props` and `exclude`. The devserver's fake lists its default targets. All of these match the plan. **(3) Invariants:** no fluxer-core, fluxer-ui or new-dependency changes. No secrets or paths are logged. The parity note on `section-voice_video-input-output` was added. **(4) Review Focus:** none of the items belongs to Task 52. ### Findings - **minor** `apps/fluxer/src/capture_rt.rs` `audio_rule`: the fallback that turned a rule of property patterns alone into System was removed. On Windows, `wasapi::loopback_target` returns `Unsupported` for an `Applications` list without a `Pid`. So if a persisted `screen_share_audio_source_mode = specific` with include patterns reaches a Windows build (for example from a settings file copied from Linux), the screen share has no sound instead of the system audio. The audio spec §3.3 says "WASAPI ignores both". The UI hides the submenu on Windows, so this is hard to reach. Suggestion: keep the fallback under `cfg(windows)`, or have WASAPI treat a list without a pid as System. - **minor** `apps/fluxer/README.md`, Voice runtime bullet: the new sentence about the PipeWire initialiser was inserted in the middle of the existing sentence. The text now reads "...filled them (#225.7) and the LiveKit room (...)", so the LiveKit room clause hangs off the wrong sentence. ### CI I started `cargo xtask ci` locally in the review worktree, but it was still compiling when this review had to be returned, so I have **no local result**. The implementer reports `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci` green. I could not verify that. **Verdict:** no blocking findings in the code. Not approved by me only because I have no CI result. A green `cargo xtask ci` on the runners should be enough to approve.
WASAPI ignores property patterns (audio spec §3.3) and refuses an application
list without a process id, so a persisted "specific" audio source made only of
property patterns left a Windows screen share silent. The app now drops the
patterns where the backend does not route them and falls back to the whole
system when nothing is left. Also untangles the Voice runtime bullet in the
README, where the PipeWire initialiser sentence had split the LiveKit clause.

Refs #342

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

Round 2 fixes in 70ba8e7:

  • capture_rt.rs, property-only rule on Windows (minor): fixed. audio_rule now goes through convert_audio_rule(rule, PROPS_ROUTED), where PROPS_ROUTED is !cfg!(windows). Where the backend does not route property patterns (WASAPI, audio spec §3.3), AppMatch::Props entries are dropped, and a rule made only of them falls back to AudioMode::System, so a stored "specific" mode no longer leaves the share silent. On Linux the patterns go through unchanged. New test without_property_routing_props_alone_fall_back_to_system covers both paths on every runner. rules_convert_props_and_exclude now pins the routing path, so it also passes on Windows.
  • apps/fluxer/README.md, Voice runtime bullet (minor): fixed. The LiveKit room clause rejoins its original sentence, and the PipeWire initialiser note (#225.7) is now its own sentence after it.

cargo xtask ci (with FLUXER_SCREENSHOT_TESTS=1) passes locally.

Round 2 fixes in 70ba8e7: - **capture_rt.rs, property-only rule on Windows (minor):** fixed. `audio_rule` now goes through `convert_audio_rule(rule, PROPS_ROUTED)`, where `PROPS_ROUTED` is `!cfg!(windows)`. Where the backend does not route property patterns (WASAPI, audio spec §3.3), `AppMatch::Props` entries are dropped, and a rule made only of them falls back to `AudioMode::System`, so a stored "specific" mode no longer leaves the share silent. On Linux the patterns go through unchanged. New test `without_property_routing_props_alone_fall_back_to_system` covers both paths on every runner. `rules_convert_props_and_exclude` now pins the routing path, so it also passes on Windows. - **apps/fluxer/README.md, Voice runtime bullet (minor):** fixed. The LiveKit room clause rejoins its original sentence, and the PipeWire initialiser note (#225.7) is now its own sentence after it. `cargo xtask ci` (with `FLUXER_SCREENSHOT_TESTS=1`) passes locally.
Author
Owner

Independent review, Text chat 52 (70ba8e7, diff 0761fc2..textchat-52-app-audio-devices-runtime)

  1. Plan tests: all five are present and assert what the plan asks. rules_convert_props_and_exclude, list_audio_targets_answers_with_nodes and unsupported_listing_answers_none (it also checks that the failure is logged only once) are in capture_rt.rs. linux_voice_backend_uses_the_webrtc_pipewire_initialiser uses fn_addr_eq against fluxer_capture::initialize_pipewire. devserver_lists_fake_audio_targets checks the fake's Firefox, mpv and Audio/Sink. The new without_property_routing_props_alone_fall_back_to_system covers the Windows path.
  2. Interfaces: on Linux, the voice backend is built with CpalHost::with_pipewire_init(fluxer_capture::initialize_pipewire). No CpalHost::new() is left in apps/ (grep), and a comment says why. ListAudioTargets answers AudioTargets(Some(nodes)), or None on an error, logged at info once and at debug after that, without the nodes' properties. audio_rule converts AppMatch::Props and exclude. The devserver fake lists its default targets.
  3. Invariants: the changes stay in apps/fluxer, README and parity. Nothing changed in core, UI or capture. No new dependency, no secrets or paths. An empty Applications list can't come from the core (screen_share_rule falls back to System), so dropping the old Linux fallback is safe.
  4. Review Focus: no item belongs to this task.
  5. Local FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci: not completed. The reviewer's run was stopped while still compiling, when this review step had to return. No result is recorded.

Findings: none blocking, none minor.

Verdict: code approved on reading. Approval is withheld only because the reviewer's local CI did not finish. Re-run cargo xtask ci (or rely on the runners) before merging.

Independent review, Text chat 52 (70ba8e7, diff 0761fc2..textchat-52-app-audio-devices-runtime) 1. Plan tests: all five are present and assert what the plan asks. `rules_convert_props_and_exclude`, `list_audio_targets_answers_with_nodes` and `unsupported_listing_answers_none` (it also checks that the failure is logged only once) are in capture_rt.rs. `linux_voice_backend_uses_the_webrtc_pipewire_initialiser` uses `fn_addr_eq` against `fluxer_capture::initialize_pipewire`. `devserver_lists_fake_audio_targets` checks the fake's Firefox, mpv and Audio/Sink. The new `without_property_routing_props_alone_fall_back_to_system` covers the Windows path. 2. Interfaces: on Linux, the voice backend is built with `CpalHost::with_pipewire_init(fluxer_capture::initialize_pipewire)`. No `CpalHost::new()` is left in apps/ (grep), and a comment says why. `ListAudioTargets` answers `AudioTargets(Some(nodes))`, or `None` on an error, logged at info once and at debug after that, without the nodes' properties. `audio_rule` converts `AppMatch::Props` and `exclude`. The devserver fake lists its default targets. 3. Invariants: the changes stay in apps/fluxer, README and parity. Nothing changed in core, UI or capture. No new dependency, no secrets or paths. An empty Applications list can't come from the core (`screen_share_rule` falls back to System), so dropping the old Linux fallback is safe. 4. Review Focus: no item belongs to this task. 5. Local `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci`: **not completed**. The reviewer's run was stopped while still compiling, when this review step had to return. No result is recorded. Findings: none blocking, none minor. Verdict: code approved on reading. Approval is withheld only because the reviewer's local CI did not finish. Re-run `cargo xtask ci` (or rely on the runners) before merging.
Author
Owner

Merged origin/textchat (18 new commits, no conflicts) into the branch as 97e4db2. FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci passes locally on 97e4db2 (all steps passed; livekit-interop skipped as usual). This completes the local CI run the independent review had left unfinished.

Merged origin/textchat (18 new commits, no conflicts) into the branch as 97e4db2. `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci` passes locally on 97e4db2 (all steps passed; livekit-interop skipped as usual). This completes the local CI run the independent review had left unfinished.
Author
Owner

Independent review: Text chat 52 (app-audio-devices-runtime)

Reviewed the task's own diff (0761fc2..70ba8e7; 97e4db2 only merges origin/textchat): apps/fluxer/src/{capture_rt.rs,voice_rt.rs}, apps/fluxer/tests/capture_audio_targets.rs, apps/fluxer/README.md, docs/parity/settings.toml.

Tests named by the plan. All five exist and assert what Step 1 asks for:

  • rules_convert_props_and_exclude: property patterns and exclude pass through unchanged, and a rule made only of property patterns stays an application list on PipeWire.
  • list_audio_targets_answers_with_nodes: the fake's targets become AudioTargets(Some(nodes)).
  • unsupported_listing_answers_none: the answer is None, and the failure is logged once.
  • linux_voice_backend_uses_the_webrtc_pipewire_initialiser: fn_addr_eq against fluxer_capture::initialize_pipewire.
  • devserver_lists_fake_audio_targets: Firefox and mpv are listed under --devserver.

All five ran and passed in my CI run.

Interfaces.

  • On Linux the real backend builds CpalHost::with_pipewire_init(fluxer_capture::initialize_pipewire), with a comment that explains why. No CpalHost::new() is left in apps/.
  • ListAudioTargets runs list_targets on the capture worker thread. On error it answers None and logs at info once, then at debug. Only the error text and a count are logged, never node properties.
  • audio_rule converts AppMatch::Props and exclude.
  • The fake backend lists its default targets.

Invariants. No changes to fluxer-core or fluxer-ui, no new dependencies, no fixtures, and no secrets or paths in logs.

Findings

  • minor: audio_rt.rs is in the plan's Files list but is not modified. Nothing in it needed to change, because the host is built in voice_rt.rs.
  • minor (deviation, already in the PR body): property_patterns_alone_fall_back_to_system_audio is replaced by without_property_routing_props_alone_fall_back_to_system. On Windows, a rule made only of property patterns still captures the whole system. On Linux, an empty application list now stays empty instead of becoming System. The core already falls back to System for an empty selection (voice/stream/audio_sources.rs), so users see no difference.

Local CI: I ran FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci on 97e4db2 and it ended with all steps passed. livekit-interop was skipped because FLUXER_LIVEKIT_TESTS is not set.

Verdict: approve. No blocking findings.

## Independent review: Text chat 52 (app-audio-devices-runtime) Reviewed the task's own diff (`0761fc2..70ba8e7`; `97e4db2` only merges `origin/textchat`): `apps/fluxer/src/{capture_rt.rs,voice_rt.rs}`, `apps/fluxer/tests/capture_audio_targets.rs`, `apps/fluxer/README.md`, `docs/parity/settings.toml`. **Tests named by the plan.** All five exist and assert what Step 1 asks for: - `rules_convert_props_and_exclude`: property patterns and `exclude` pass through unchanged, and a rule made only of property patterns stays an application list on PipeWire. - `list_audio_targets_answers_with_nodes`: the fake's targets become `AudioTargets(Some(nodes))`. - `unsupported_listing_answers_none`: the answer is `None`, and the failure is logged once. - `linux_voice_backend_uses_the_webrtc_pipewire_initialiser`: `fn_addr_eq` against `fluxer_capture::initialize_pipewire`. - `devserver_lists_fake_audio_targets`: Firefox and mpv are listed under `--devserver`. All five ran and passed in my CI run. **Interfaces.** - On Linux the real backend builds `CpalHost::with_pipewire_init(fluxer_capture::initialize_pipewire)`, with a comment that explains why. No `CpalHost::new()` is left in `apps/`. - `ListAudioTargets` runs `list_targets` on the capture worker thread. On error it answers `None` and logs at `info` once, then at `debug`. Only the error text and a count are logged, never node properties. - `audio_rule` converts `AppMatch::Props` and `exclude`. - The fake backend lists its default targets. **Invariants.** No changes to `fluxer-core` or `fluxer-ui`, no new dependencies, no fixtures, and no secrets or paths in logs. **Findings** - minor: `audio_rt.rs` is in the plan's Files list but is not modified. Nothing in it needed to change, because the host is built in `voice_rt.rs`. - minor (deviation, already in the PR body): `property_patterns_alone_fall_back_to_system_audio` is replaced by `without_property_routing_props_alone_fall_back_to_system`. On Windows, a rule made only of property patterns still captures the whole system. On Linux, an empty application list now stays empty instead of becoming System. The core already falls back to System for an empty selection (`voice/stream/audio_sources.rs`), so users see no difference. **Local CI:** I ran `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci` on `97e4db2` and it ended with `all steps passed`. `livekit-interop` was skipped because `FLUXER_LIVEKIT_TESTS` is not set. **Verdict: approve.** No blocking findings.
Author
Owner

Independent review, Task 52 (app-audio-devices-runtime), commits 1d94385 and 70ba8e7.

Findings:

  • (minor) Local cargo xtask ci was started in a clean review worktree but did not finish before this review had to report, so it has no result. The code review found no defects.
  • Checked: the five tests from the plan exist and assert what the plan says. rules_convert_props_and_exclude, list_audio_targets_answers_with_nodes and unsupported_listing_answers_none are in capture_rt.rs. linux_voice_backend_uses_the_webrtc_pipewire_initialiser (voice_rt.rs) compares the fn pointer to fluxer_capture::initialize_pipewire. devserver_lists_fake_audio_targets is in tests/capture_audio_targets.rs.
  • Checked: the produced interfaces are present. On Linux the backend uses CpalHost::with_pipewire_init(...), never CpalHost::new(), and a comment explains why. ListAudioTargets answers Some(nodes), or None on error, logged once at info and then at debug, with no props in the logs. audio_rule converts Props and exclude, and on Windows it falls back to system audio. The settings.toml note was added.
  • Checked: no changes to fluxer-core or fluxer-ui. No secrets, paths or URLs.

Verdict: NOT APPROVED yet. No blocking defects in the code, but local CI did not finish. Re-run cargo xtask ci before merging.

Independent review, Task 52 (app-audio-devices-runtime), commits 1d94385 and 70ba8e7. Findings: - (minor) Local `cargo xtask ci` was started in a clean review worktree but did not finish before this review had to report, so it has no result. The code review found no defects. - Checked: the five tests from the plan exist and assert what the plan says. `rules_convert_props_and_exclude`, `list_audio_targets_answers_with_nodes` and `unsupported_listing_answers_none` are in capture_rt.rs. `linux_voice_backend_uses_the_webrtc_pipewire_initialiser` (voice_rt.rs) compares the fn pointer to `fluxer_capture::initialize_pipewire`. `devserver_lists_fake_audio_targets` is in tests/capture_audio_targets.rs. - Checked: the produced interfaces are present. On Linux the backend uses `CpalHost::with_pipewire_init(...)`, never `CpalHost::new()`, and a comment explains why. ListAudioTargets answers Some(nodes), or None on error, logged once at info and then at debug, with no props in the logs. `audio_rule` converts Props and `exclude`, and on Windows it falls back to system audio. The settings.toml note was added. - Checked: no changes to fluxer-core or fluxer-ui. No secrets, paths or URLs. Verdict: NOT APPROVED yet. No blocking defects in the code, but local CI did not finish. Re-run `cargo xtask ci` before merging.
Author
Owner

Merged origin/textchat (9cf9a68) into the branch as 9978c59. One conflict, in apps/fluxer/README.md: kept this task's Voice runtime and Screen-share runtime bullets and the new Text chat runtime bullet from textchat (textchat had not changed the other two). FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci passes locally on 9978c59 (all steps passed; livekit-interop skipped as usual).

Merged origin/textchat (9cf9a68) into the branch as 9978c59. One conflict, in `apps/fluxer/README.md`: kept this task's Voice runtime and Screen-share runtime bullets and the new Text chat runtime bullet from textchat (textchat had not changed the other two). `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci` passes locally on 9978c59 (all steps passed; livekit-interop skipped as usual).
Author
Owner

Independent review: Text chat 52 (round 3, after merging textchat 9cf9a68 as 9978c59)

I read the diff 9cf9a68..9978c59 from scratch.

1. Tests the plan names. All five exist and assert what Step 1 asks for:

  • rules_convert_props_and_exclude: AppMatch::Props passes through in order next to Pid. exclude is carried over for both Applications and System.
  • list_audio_targets_answers_with_nodes: the fake's targets become AudioTargets(Some([AudioNode(props)…])).
  • unsupported_listing_answers_none: the answer is None, and the failure is logged at info only once.
  • linux_voice_backend_uses_the_webrtc_pipewire_initialiser: compares fn_addr_eq(system_pipewire_init(), fluxer_capture::initialize_pipewire). This is the plan's "constructor helper" option.
  • devserver_lists_fake_audio_targets: under --devserver the fake backend answers RefreshAudioSources with AudioTargets::Listed. It lists Firefox and mpv, plus a sink.

The replacement test without_property_routing_props_alone_fall_back_to_system checks the Windows path correctly. The core's screen_share_rule never produces an empty Applications list, so the Linux path (no fallback) cannot capture nothing.

2. Interfaces.

  • VoiceBackend::system builds CpalHost::with_pipewire_init(system_pipewire_init()). The comment explains why (#225.7). CpalHost::new() is not called anywhere in apps/. The construction itself is already on textchat, so this diff adds only the test.
  • CaptureCommand::ListAudioTargets runs list_targets on the capture worker thread.
  • audio_rule converts Props and exclude.
  • The README and the parity note (#220) are updated.

3. Invariants.

  • No changes to fluxer-core, fluxer-ui or fluxer-emoji, and no new dependencies.
  • No props are logged: the success log has only a count, and the failure log has only the error, whose PipeWire and Unsupported messages carry no node properties.
  • No new fixtures. No secrets or paths. Server data is not involved.

4. Review Focus. None of the items applies to this task.

5. CI. FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci on 9978c59 in a fresh review worktree: all steps passed (fmt, clippy, deny, core-isolation, parity, test, vendored-tests, screenshot-tests). livekit-interop was skipped because FLUXER_LIVEKIT_TESTS is not set.

Findings

  • (minor, informational) No real call and no mic test were run on PipeWire (FLUXER_AUDIO_DEVICE_TESTS / FLUXER_CAPTURE_TESTS). The PR states this. Running them on hardware before release would be worthwhile.

Verdict: APPROVED. No blocking findings, and local CI is green.

## Independent review: Text chat 52 (round 3, after merging textchat 9cf9a68 as 9978c59) I read the diff `9cf9a68..9978c59` from scratch. **1. Tests the plan names.** All five exist and assert what Step 1 asks for: - `rules_convert_props_and_exclude`: `AppMatch::Props` passes through in order next to `Pid`. `exclude` is carried over for both `Applications` and `System`. - `list_audio_targets_answers_with_nodes`: the fake's targets become `AudioTargets(Some([AudioNode(props)…]))`. - `unsupported_listing_answers_none`: the answer is `None`, and the failure is logged at `info` only once. - `linux_voice_backend_uses_the_webrtc_pipewire_initialiser`: compares `fn_addr_eq(system_pipewire_init(), fluxer_capture::initialize_pipewire)`. This is the plan's "constructor helper" option. - `devserver_lists_fake_audio_targets`: under `--devserver` the fake backend answers `RefreshAudioSources` with `AudioTargets::Listed`. It lists Firefox and mpv, plus a sink. The replacement test `without_property_routing_props_alone_fall_back_to_system` checks the Windows path correctly. The core's `screen_share_rule` never produces an empty `Applications` list, so the Linux path (no fallback) cannot capture nothing. **2. Interfaces.** - `VoiceBackend::system` builds `CpalHost::with_pipewire_init(system_pipewire_init())`. The comment explains why (#225.7). `CpalHost::new()` is not called anywhere in `apps/`. The construction itself is already on textchat, so this diff adds only the test. - `CaptureCommand::ListAudioTargets` runs `list_targets` on the capture worker thread. - `audio_rule` converts `Props` and `exclude`. - The README and the parity note (#220) are updated. **3. Invariants.** - No changes to fluxer-core, fluxer-ui or fluxer-emoji, and no new dependencies. - No props are logged: the success log has only a count, and the failure log has only the error, whose PipeWire and Unsupported messages carry no node properties. - No new fixtures. No secrets or paths. Server data is not involved. **4. Review Focus.** None of the items applies to this task. **5. CI.** `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci` on 9978c59 in a fresh review worktree: all steps passed (fmt, clippy, deny, core-isolation, parity, test, vendored-tests, screenshot-tests). livekit-interop was skipped because `FLUXER_LIVEKIT_TESTS` is not set. **Findings** - (minor, informational) No real call and no mic test were run on PipeWire (`FLUXER_AUDIO_DEVICE_TESTS` / `FLUXER_CAPTURE_TESTS`). The PR states this. Running them on hardware before release would be worthwhile. **Verdict: APPROVED.** No blocking findings, and local CI is green.
charles manually merged commit 851af027ea into textchat 2026-11-13 13:10:51 +00:00
Sign in to join this conversation.
No description provided.