[richiemcilroy]

Fixes the intermittent sync-tests (macos-latest) failures in audio::tests::a_replay_resumes_a_parked_render and audio::tests::a_stale_render_is_released_once_the_next_render_finishes (added in #2521).

Cause

PlaybackAudioCache::with_idle_release calls memory_pressure::watch() (crates/editor/src/audio.rs:1786), which installs a real process-wide DISPATCH_SOURCE_TYPE_MEMORYPRESSURE source. Its handler calls release_playback_audio_caches() (audio.rs:1761), which empties every cache in the process.

The tests guard the simulated release in export_and_memory_pressure_release_every_cached_render with the CACHE_RELEASE write lock (audio.rs:4724), but nothing guards the real OS source. On a loaded macOS runner a warn-level pressure event can arrive at any time, and every test that asserts a cached render is still there fails together:

• a_replay_resumes_a_parked_render (audio.rs:5378): the entry is gone, so from_cache misses and the replay builds a fresh render.
• a_stale_render_is_released_once_the_next_render_finishes (audio.rs:5269): the paused render is freed before the 50 ms check.

In run 2755 both tests failed within 62 ms of each other, which fits one process-wide release and not two independent races. The product behaviour is intended: under memory pressure the parked render should be dropped and the next play renders fresh.

Fix

memory_pressure::watch() returns early under cfg!(test), so the editor's unit tests no longer receive host memory pressure events. The release path is still covered by calling on_memory_pressure directly in export_and_memory_pressure_release_every_cached_render. Non-test builds are unchanged. No assertions, sleeps or timeouts were changed.

Verification

CI on this PR: A/V Sync Tests passed on macOS, Ubuntu and Windows (45 checks green). export_and_memory_pressure_release_every_cached_render still covers the release path by calling on_memory_pressure directly. Before the fix, 3 of 14 macOS sync-tests runs on unrelated PRs failed these two tests, and two of the failures landed within 62 ms of each other in one run, which points to a single process-wide release.

Confidence Score: 5/5

The PR appears safe to merge; it isolates unit tests without changing production behavior.

Summary

Prevents macOS memory pressure events from clearing audio caches during editor unit tests.

• Editor unit tests skip the host memory-pressure watcher.

_{Reviews (1) · Last reviewed commit: "fix(editor): keep the OS memory pressure..." · Reviewed by Greptile}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
cap-status	Ready	Preview	Oct 23, 2026 11:41pm UTC

[capy-ai]

Capy found no issues.

View 1 other finding in Capy.