1210797 - Print the stack trace of all threads before killing firefox when a test times out on tbpl.

Status: NEW

(Core :: General, defect)

Description

[Nicolas Silva [:nical]]

I can't overstate how useful this would be. When the harness is about to kill gecko because of a test timeout, if it could first connect to the running process(es) with gdb, and print their stack traces to the log, debugging test timouts would be a whole lot easier!

Comment 1

[(not currently active) Ted Mielczarek]

I replied to this on IRC, but we should already have this behavior if our test harnesses are working properly. For example, in Mochitest:
https://dxr.mozilla.org/mozilla-central/rev/2c1fb007137dcb68b1862a79553b53f1a34c99c3/testing/mochitest/runtests.py#1562

When we hit a timeout we attempt to kill the process in a way that triggers Breakpad, which writes a minidump. We then run minidump_stackwalk on the dump which prints every thread's stack.

Are you seeing this not work somewhere?