fix(scope): Scope.runAsync in digest and flush #1005
Conversation
| _zone.onTurnDone = () { | ||
| apply(); | ||
| // Do not leave any microtasks behind. | ||
| assert(_runAsyncHead == null); |
There was a problem hiding this comment.
Can we move this to the assert phase?
|
Per our conversation can we have a different API for runAsync which runs in digest vs the one which runs in flush phase? Perhaps runAsync and runAsyncFlush? |
BREAKING CHANGE: Previously a micro task registered in flush phase would cause a new digest cycle after the current digest cycle. The new behavior will cause an error. Closes dart-archive#984
BREAKING CHANGE: Microtasks scheduled in flush will trigger a new digest+flush cycle. Microtasks scheduled in digest will be executed in digest, counting towards the ScopeDigestTTL.
|
This PR fixes a bug where a microtask scheduled from inside of Scope.apply() will trigger another complete Scope.apply(). It should only trigger another digest loop or another flush. We patched this change into real applications and saw actions that originally had ~30 flushes drop to ~20. However, we were not able to see any significant performance gains as measured in wall-time. Applications which track the number of flushes are nevertheless nervous about this bug. This PR changes the life-cycle of the application is subtle ways which are difficult to debug. I feel that we need to land Scope.apply() debugging, e.g. #1114 before we are able to land this PR. |
|
I am looking into this again. I don't think the above code is right as it loops over apply. |
|
Replaced by #1140 |
No description provided.