Open Bug 1281321 Opened 5 years ago Updated 4 years ago
.ini:dom/indexed DB/test/unit/test _transaction _error .js | application crashed [@ mozalloc _abort(char const*)]
Filed by: firstname.lastname@example.org https://treeherder.mozilla.org/logviewer.html#?job_id=30421010&repo=mozilla-inbound http://archive.mozilla.org/pub/firefox/tinderbox-builds/mozilla-inbound-macosx64-debug/1466541840/mozilla-inbound_yosemite_r7-debug_test-xpcshell-bm107-tests1-macosx-build2463.txt.gz
Markus, can you explain what's going on here? I don't really understand why pthread thread local destruction code is calling into PR_Assert.
I don't know much about threads or thread life cycles, but here's what I found out: PR_NewThreadPrivateIndex lets you specify a TLS destructor. If one of those destructors uses PR_Assert, that might potentially the one we're hitting. Moreover, _PR_DestroyThreadPrivate has a few PR_Asserts, and it's what calls the registered destructors. _PR_DestroyThreadPrivate is called at the end of the thread's root function _pt_root if the thread is not "detached" (whatever that means), and from _pt_thread_death, which is registered using pthread_key_create (through _PT_PTHREAD_KEY_CREATE). My hunch is that the function "_pthread_tsd_cleanup" in our crash callstack indicates that our code is called through such a pthread_key_create-supplied destructor. I don't know why none of those functions show up as their own frame in the call stack.
Ok ... the next question is where is stderr for the test we're running? I don't know who owns xpcshell in automation, so I'll needinfo gps and he will pass it along.
File xpcshell test harness bugs at https://bugzilla.mozilla.org/enter_bug.cgi?product=Testing&component=XPCShell%20Harness. That's for local and automation issues.
I was kind of hoping for a name that I could needinfo here, rather than a place to file a bug that may or may not get looked at ...
File the bug. People are pretty good about triaging problems with test harnesses especially if the bug report points to an obvious problem. You can needinfo ted or me if you want.
Bulk assigning P3 to all open intermittent bugs without a priority set in Firefox components per bug 1298978.
Priority: -- → P3
You need to log in before you can comment on or make changes to this bug.