fix tsan errors #745
fix tsan errors #745
Conversation
| @@ -31,3 +31,7 @@ race:crossbeam_deque*steal | |||
| race:Backup::next_sleeper | |||
| race:Backup::set_next_sleeper | |||
| race:WorkerEntry::set_next_sleeper | |||
|
|
|||
| # This ignores a false positive caused by `thread::park()`/`thread::unpark()`. | |||
There was a problem hiding this comment.
How would you know it's a false positive?
There was a problem hiding this comment.
There is some discussion on it here: rust-lang/rust#54806 (comment)
The PR that changed the implementation of thread::unpark() is a small change we all agreed on that it must be correct, and the only explanation is that the error is caused by tsan not understanding fences (it only understands atomic operations).
The reported data race is in Condvar::drop. We only use SeqCst operations in thread::park() and thread::unpark() so the problem must be somewhere else.
There was a problem hiding this comment.
But race:pthread_cond_destroy syntax silents all tsan errors for pthread_cond_destroy, isn't it?
There was a problem hiding this comment.
True. I'm not sure if we can suppress a more specific set of cases.
ghost
mentioned this pull request
ghost
requested a review
from 
|
@carllerche How does that look? Can we merge? |
The last few builds have been failing due to thread sanitizer errors.
See rust-lang/rust#54806 (comment) for why this is happening.
This PR adds a rule to the tsan suppresions list.