mirror of
https://github.com/stablyai/orca.git
synced 2026-09-22 00:02:31 +00:00
Neil
e7563c63f1
* fix: fence browser recovery to attach inventory placements * refactor: name the attach-inventory fence and make its test deterministic Extract the placement check into isPlacedAsObservedAtAttach so the recovery filter stays a flat list of named predicates, and document that omitting pagePlacementsAtAttach recovers against unfenced live state. Replace the 30-microtask drain in the post-attach regression with the handler's own completion: attach only settles after recovery returns, so awaiting the dispatch orders the assertions instead of guessing at a microtask count. Verified by forcing the fence open: both regressions fail (the post-attach one in 60ms on a retired placement) and the other 27 still pass. * test: settle the attach handler even when the regression fails early The barrier ran inline, so a waitFor timeout or the placement guard left the attach handler parked on a promise nothing awaited. Hoist it into settleAttach and call it from a finally as well; cleanup is guarded and the dispatch promise is already settled, so the second call is a no-op.