Data race with CancellableContinuationImpl.parentHandle

[shmuelr]

Hey there,

Our internal Java TSAN tests inside Google picked up a race condition on the develop branch with CancellableContinuationImpl.parentHandle, likely introduced in this commit that changes parentHandle to be non-volatile.

According to our stack trace, the race happens at
line 494 parentHandle = NonDisposableHandle

WARNING: ThreadSanitizer: data race (pid=4318)
Write of size 4 at 0x0000cf6bf808 by thread T30 (mutexes: write M379709747054803224, write M380272697008225000, write M379709747035779072, write M380272696989200768, write M357473223960208968, write M361976823587580648, write M498773662268158696, write M565201756770664040):
#0 kotlinx.coroutines.CancellableContinuationImpl.detachChild$kotlinx_coroutines_kotlinx_coroutines()V CancellableContinuationImpl.kt:494
#1 kotlinx.coroutines.CancellableContinuationImpl.detachChildIfNonResuable()V CancellableContinuationImpl.kt:485
#2 kotlinx.coroutines.CancellableContinuationImpl.resumeImpl(Ljava/lang/Object;ILkotlin/jvm/functions/Function1;)V CancellableContinuationImpl.kt:427
#3 kotlinx.coroutines.CancellableContinuationImpl.resumeImpl$default(Lkotlinx/coroutines/CancellableContinuationImpl;Ljava/lang/Object;ILkotlin/jvm/functions/Function1;ILjava/lang/Object;)V CancellableContinuationImpl.kt:420
#4 kotlinx.coroutines.CancellableContinuationImpl.resumeWith(Ljava/lang/Object;)V CancellableContinuationImpl.kt:325
#5 kotlinx.coroutines.guava.ToContinuation.run()V ListenableFuture.kt:266

Previous read of size 4 at 0x0000cf6bf808 by thread T9:
#0 kotlinx.coroutines.CancellableContinuationImpl.getResult()Ljava/lang/Object; CancellableContinuationImpl.kt:267
#1 kotlinx.coroutines.guava.ListenableFutureKt.await(Lcom/google/common/util/concurrent/ListenableFuture;Lkotlin/coroutines/Continuation;)Ljava/lang/Object; ListenableFuture.kt:517
#2

It appears that parentHandle can be raced at when using ListenableFuture extensions for Coroutines, (possibly because LF calls getResult() which triggers the race at line 267).

Thanks!

[qwwdfsad]

Thanks for the report, good to know our code is externally tested for data-races :)

It is benign data-race and it's safe to suppress, I've added the corresponding comment