Open Bug 1329781 Opened 4 years ago Updated 4 years ago
On Linux64 debug crashtest runs, shutdown crashes are incorrrectly reported as having happened during the final crashtest ("REFTEST PROCESS-CRASH | [test-filename] | application terminated ...")
This treeherder crashtest log includes an abort that happens during shutdown -- after "REFTEST SUITE-END | Shutdown": https://treeherder.mozilla.org/logviewer.html#?job_id=65504183&repo=mozilla-inbound&lineNumber=123897 But, there's a problem -- the reftest harness mistakenly blames the final crashtest (by name) for the abort! This happens on Linux64, but not on mac, for some reason. It's not yet clear to me what the key difference is. On linux, the error is reported as: TEST-UNEXPECTED-FAIL | file:///home/worker/workspace/build/tests/reftest/tests/xpcom/string/crashtests/1113005.html | application terminated with exit code -139 REFTEST PROCESS-CRASH | file:///home/worker/workspace/build/tests/reftest/tests/xpcom/string/crashtests/1113005.html | application crashed [None] ...whereas on mac, it's reported as: REFTEST PROCESS-CRASH | Main app process exited normally | application crashed [@ NS_ShutdownAtomTable()] This caused some confusion in bug 1276669 (the bug filed on this intermittent issue) -- philor updated the title to mention the test's name (because he ran into the first log which mentions the test's name). But then that confused me today when I hit the intermittent on Mac, because the test's name is (correctly) not mentioned alongside the error in the log there. Ideally we should be consistent and we should *not* include the test's name in this case. Maybe we should clear whatever variable stores the test's name, when we get to printing REFTEST SUITE-END | Shutdown?
Summary: On Linux64 debug crashtest runs, shutdown crashes are incorrrectly reported as being during the final test → On Linux64 debug crashtest runs, shutdown crashes are incorrrectly reported as having happened during the final crashtest ("REFTEST PROCESS-CRASH | [test-filename] | application terminated ...")
For archival purposes, here's a gzipped full log from the affected Linux64 run that I linked in comment 0.
And for comparison, here's a link to the Mac testrun that I briefly quoted in comment 0, which reports the issue more correctly (IMO) -- *without* the test name in the crash log line: https://treeherder.mozilla.org/#/jobs?repo=autoland&revision=c7ce652655932785c40843f979a3f1e964e56f21&selectedJob=65758557 I'm attaching this full log for reference/archival purposes as well.
Attachment #8825171 - Attachment description: sample linux log which has the issue → sample linux log with this shutdown crash, which mistakenly attributes it to the final crashtest 1113005.html (bad)
You need to log in before you can comment on or make changes to this bug.