On 8/2/19 12:14 PM, Francois Gouget wrote:
On Wed, 31 Jul 2019, Jacek Caban wrote: [...]
debugger.c:1143: Test failed: Got 10958580 failures from child process. debugger.c:1091: Test failed: unexpected event pid
This is not related to the patch. It looks like we need to increase the timeout in existing tests.
Do you mean the 2 minutes test timeout? We really don't want the tests to take that long.
No, it's a timeout inside test code. I will take care of it.
On slightly related note, I feel like Vista is overrepresented on TestBot. It's the least interesting Windows version. While XP is still interesting sometimes, I don't see any value in testing on Vista. Debugger tests are an example of a case where Vista seems to behave worse than any other Windows version (XP including) and it just adds extra useless work of working around its weirdness. I'd suggest to plan to get rid of Vista from TestBot. Currently, due to multiple locale VMs, it has the highest number of VMs.
I don't think that matters. For tests I would say the only results that matter are 0 test failures, and 1+ test failures. Running a test more times on a given platform would not change that, except for random failures which are probably random on any platform anyway.
Debugger tests are an example, where I had to add a work around for random failures seem to be Vista only (like, it sometimes creates a weird, superfluous thread in a random moment). Whenever possible, we'd like to fix random failures on interesting platforms. TestBot is currently more likely to reveal rare failures on Vista than on win10, which is opposite to bug priorities IMHO.
So if test failures on Vista are not important the only option is to not run the tests on that platform at all.
That would be my preference. I don't think it's urgent nor important, but I think it would be nice to steer in that direction (and, at very least, not add any more Vista VMs in the future).
Thanks,
Jacek