539290 - Test harness should look for minidumps from child (plugin) processes

Status: RESOLVED FIXED

(Core :: IPC, defect)

Description

[Benjamin Smedberg]

The automation.py test harness should look for minidumps from child (plugin) processes even if the main process exits with a success code. In this case, it should not print TEST-UNEXPECTED-FAIL (because the crash may be expected from a test), but should simply print the minidump. The leak harness will catch the unexpected crashes with "failed to print leak summary".

Comment 1

[(not currently active) Ted Mielczarek]

I suggested that we make the parent process log something to the PID log so the harness would know to look for minidumps. Currently it only checks if the main process exits with a non-zero return code.

Comment 2

[Benjamin Smedberg]

Unfortunately that would require some way to match up minidumps with PIDs. Going for the simplest solution, I'd make "application crashed (minidump found)" be "INFO" instead of "TEST-UNEXPECTED-FAIL"

Comment 3

[Benjamin Smedberg]

Attachment: Dump minidumps for all child crashes, rev. 1

This dumps the crash data for all crashes, whether or not the crash was "expected" or not.

Comment 4

[(not currently active) Ted Mielczarek]

Well, no, I just meant we could leave a marker in the PID log that some child process crashed, to know to look for minidumps. But I guess it doesn't matter. I'm not wild about making the "crashed (minidump found)" bit not TEST-UNEXPECTED-FAIL, because then it won't show up in the brief tinderbox log.

Comment 5

[Benjamin Smedberg]

Attachment: Dump minidumps for all child crashes, rev. 2

This patch has the alternate problem where crashes will show up in the brief log even if they are "expected"... it needs to be combined with the tinderbox change in bug 483061 to look for the PROCESS-CRASH string.

Comment 6

[(not currently active) Ted Mielczarek]

Comment on attachment 421325 [details] [diff] [review]
Dump minidumps for all child crashes, rev. 2

This is fine for now, I just know (from experience) that people will see the crash in the brief log and blame the orange on that no matter what.

Comment 7

[Benjamin Smedberg]

http://hg.mozilla.org/mozilla-central/rev/e828ac500665