dpw / skinny-mutex Goto Github PK

This comment in cond_wait_cleanup didn't make sense to me:

/* Cancellation of pthread_cond_wait re-acquires fat->mutex,
 * and we have it pinned, so we can access the fat mutex
 * directly from the skinny mutex for once. */

So I decided to use a more brutal test for the wait cancellation, to see if it would pop (which it did). I pushed the test I made, and suggested fix (and comments as to why I believe the way you had it would fail) - and you can see the diff here:

master...december1981:master
(You'll note line 450: "assert(!fat->held);" in fat_mutex_release was also detected as a false positive).

Not making a pull request, as the changes were hacked together quickly to assist you in exposing the issue/working toward a proper fix.