Cleanup lifetime management in OSX FileSystemWatcher #52019
Conversation
|
Tagging subscribers to this area: @carlossanlop |
| CFTimeInterval latency, | ||
| FSEventStreamCreateFlags flags) | ||
| { | ||
| return FSEventStreamCreate(IntPtr.Zero, cb, IntPtr.Zero, pathsToWatch, sinceWhen, latency, flags); |
| @@ -138,6 +138,9 @@ private sealed class RunningInstance | |||
| // Callback delegate for the EventStream events | |||
| private readonly Interop.EventStream.FSEventStreamCallback _callback; | |||
There was a problem hiding this comment.
The new GCHandle is a counterpart for this delegate. The RunningInstance has to be kept alive for as long as the marshalled delegate is in use. There was nothing to guarantee that before this change.
I have also looked into replacing the delegate by a function pointer, but that was more involved change so I have abandoned it.
| try | ||
| { | ||
| // When we get here, we've requested to stop so cleanup the EventStream and unschedule from the RunLoop | ||
| Interop.EventStream.FSEventStreamStop(eventStream); |
There was a problem hiding this comment.
FSEventStreamStop does not throw any exceptions. This try/finally does not look necessary. It may be a left over from some thread abort hardening.
| SafeCreateHandle? path = null; | ||
| SafeCreateHandle? arrPaths = null; | ||
| bool cleanupGCHandle = false; | ||
| try |
There was a problem hiding this comment.
This refactoring is to ensure that everything gets disposed when anything fails.
May fix #30056