NativeWatcher: Emit events in the order fs.watch reported them - fix flaky integration tests - #1923
Open
robhogan wants to merge 1 commit into
Open
NativeWatcher: Emit events in the order fs.watch reported them - fix flaky integration tests#1923robhogan wants to merge 1 commit into
robhogan wants to merge 1 commit into
Conversation
meta-cla
Bot
added
the
CLA Signed
**Problem** Each raw `fs.watch` callback was dispatched fire-and-forget, and every handler awaits an `lstat` before deciding what to emit. Those calls complete in whatever order the libuv thread pool finishes them, so the order we emitted in could differ from the order the OS reported. Over a recursive delete of `app/moved-in/file.js`, the raw callback order was correct in 250/250 runs but emission diverged in 5/250 - most often `delete app` overtaking `delete app/moved-in`. **Fix** `#handleEvent` now resolves a raw event into the event to emit rather than emitting it directly, and the `fs.watch` callback appends the result to a promise chain. The `lstat` calls still start immediately and run concurrently - only emission is sequenced - so there is no throughput cost. Also fixes a missing call in the constructor's `isSupported` guard, which made the check always truthy. **Impact** No user-facing bug is known to result from the old behaviour. `NativeWatcher` is the default backend on macOS when Watchman is unavailable, but each handler stats the path at handling time rather than trusting the event, so a reordered pair still tends to settle on the correct final state. The demonstrated effect is intermittent failure of the watcher integration tests, where a straggling deletion arriving after the one the test waited on desynchronised every later assertion. Ordering is worth guaranteeing regardless - `WatcherBackend` consumers reasonably assume events on a path and its ancestors arrive as the OS reported them, and `recrawl` in particular asks the file map to re-crawl a subtree that later events may still refer to. **Test plan** 250/250 emissions correctly ordered after the change, against 245/250 before. `packages/metro-file-map/src/watchers/__tests__/integration-test.js` passed 50/50 runs with this change alone and the tests unmodified, against a 5-10% baseline failure rate on macOS.
robhogan
force-pushed
the
fix/native-watcher-preserves-order
branch
from
September 10, 2026 11:00
ad16021 to
414c96e
Compare
facebook-github-tools
Bot
added
the
Shared with Meta
Contributor
|
@GijsWeterings has imported this pull request. If you are a Meta employee, you can view this in D119493126. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
Each raw
fs.watchcallback was dispatched fire-and-forget, and every handler awaits anlstatbefore deciding what to emit. Those calls complete in whatever order the libuv thread pool finishes them, so the order we emitted in could differ from the order the OS reported. Over a recursive delete ofapp/moved-in/file.js, the raw callback order was correct in 250/250 runs but emission diverged in 5/250 - most oftendelete appovertakingdelete app/moved-in.In particular, this failed integration tests, necessitating retries, eg 8 recent examples that failed on the first attempt and relied on
retries: 3:Fix
#handleEventnow resolves a raw event into the event to emit rather than emitting it directly, and thefs.watchcallback appends the result to a promise chain. Thelstatcalls still start immediately and run concurrently - only emission is sequenced - so there is no throughput cost.Also fixes a missing call in the constructor's
isSupportedguard, which made the check always truthy.Impact
No user-facing bug is known to result from the old behaviour.
NativeWatcheris the default backend on macOS when Watchman is unavailable, but each handler stats the path at handling time rather than trusting the event, so a reordered pair still tends to settle on the correct final state. And though watchers report directory events from the OS, we don't actually use them inmetro-file-map- we prefer to synthesise directory events as files make directories (non-)empty.Changelog: Internal
Test plan
250/250 emissions correctly ordered after the change, against 245/250 before.
packages/metro-file-map/src/watchers/__tests__/integration-test.jspassed 50/50 runs with this change alone and the tests unmodified, against a 5-10% baseline failure rate on macOS.