Skip to content

Fix browser frame handoff and isolate first-frame CI evidence - #62

Merged
Fectivnfy112357 merged 2 commits into
mainfrom
fix/browser-frame-subscriber-handoff
Nov 20, 2026
Merged

Fectivnfy112357 merged 2 commits into
mainfrom
fix/browser-frame-subscriber-handoff

Conversation

@Fectivnfy112357

Copy link
Copy Markdown
Contributor

Confirmed: a replacement browser watcher can lose its frames

The last watcher used to await the provider stop before removing its subscriber and handler entries. A replacement could join or start in that gap, after which the old cleanup deleted the replacement's state. The test parks the fake provider's stop, starts a replacement, releases the stop, and delivers a distinct frame. On the unchanged runtime the replacement did not receive it.

Detach the old entries synchronously, then order the new start after the same provider/target stop. Canceled replacements can leave the wait immediately. Stale callbacks and finalizers must still own the exact subscription entry and provider. No timeout changes, skipped tests, new cache, or changes to explicit page close semantics.

Separately: first-frame CI evidence

Main run 207 macOS timed out waiting for a first frame after tab metadata arrived. We have not established that it hit the subscriber race. A separate test-only commit captures bounded operation timing and frame counts on failure, and namespaces the transcript's existing CDP artifacts so other suites cannot replace them. A local synthetic dropped-inbound-frame control failed the unchanged assertion and the evidence distinguished the received wire frame from the absent panel frame.

Verification

  • Deterministic watcher tests: parked successful/failed stop, abort before stop settles, stale callback, overlapping watchers, close while parked, and pending start before teardown.
  • Linux: 68 browser-service unit tests and 23 browser live-navigation/screencast/two-page/session-release tests passed; focused real agent transcript scenario passed with the isolated fixture for the diagnostic commit.
  • WebUI test/server typechecks and source validation passed locally. Draft pending hosted CI; keep the separate pre-existing Windows native teardown issue open.

@Fectivnfy112357
Fectivnfy112357 marked this pull request as ready for review November 20, 2026 20:52
@Fectivnfy112357
Fectivnfy112357 merged commit 1b9a00d into main Nov 20, 2026
18 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant