Closed
Bug 1385311
Opened 7 years ago
Closed 7 years ago
Make sure ASAN and LSAN failures turn the job orange
Categories
(Release Engineering :: Applications: MozharnessCore, defect)
Release Engineering
Applications: MozharnessCore
Tracking
(firefox57 fixed)
RESOLVED
FIXED
Tracking | Status | |
---|---|---|
firefox57 | --- | fixed |
People
(Reporter: ahal, Assigned: ahal)
References
Details
Attachments
(1 file)
I think LSAN failures will still turn the job orange, but there are cases of ASAN failures remaining green. Here's an example: https://treeherder.mozilla.org/logviewer.html#?job_id=118977611&repo=try&lineNumber=2858 This was from a try push where I added a check for "ERROR: AddressSanitizer" to make the job orange: https://treeherder.mozilla.org/#/jobs?repo=try&revision=f06adacf7fa6275a44022c632d1eff4b22b627b5 I'm still not sure if that failure is intermittent or not, but it'll need to be fixed/disabled before a patch here can land. We should also run the mochitest selftests with an ASAN build and add a test to ensure the job turns orange. Bill, do you know how to reliably reproduce any kind of ASAN failure from a mochitest plain test (or direct me to someone who might know)?
Assignee | ||
Updated•7 years ago
|
Flags: needinfo?(wmccloskey)
Comment 1•7 years ago
|
||
You could probably do what BrowserTestUtils.crashBrowser() does and use ctypes to do a null deref. You'd have to use special powers to access it. There may even be some way to use ctypes to induce a use-after-free, but getting any testing would obviously be an improvement.
Flags: needinfo?(wmccloskey)
Assignee | ||
Comment 2•7 years ago
|
||
Thanks, I'll take a stab at a test next week. I was intending the test to be more for whether the task properly turns orange, rather than a test for the AddressSanitizer itself. So from the mozharness/mochitest perspective, it wouldn't be valuable to test more than 1 type of ASAN error. Of course we could expand the test's scope to include ASAN if we wanted to.
Assignee: nobody → ahalberstadt
Status: NEW → ASSIGNED
Comment 3•7 years ago
|
||
Yeah, but you could imagine if different errors caused different output then you'd need multiple tests. I guess they all just do the same thing right now.
Comment hidden (mozreview-request) |
Assignee | ||
Comment 5•7 years ago
|
||
(sigh, I need some more coffee.. sorry about all the bug spam) Luckily it looks like there weren't any real problems that slipped in because of this, just needed to skip those two tests that depend on crashreporter. Here's a good looking try run: https://treeherder.mozilla.org/#/jobs?repo=try&revision=02baffe6bc221ba883117d60e92aebee3bb5e85a
Comment 6•7 years ago
|
||
mozreview-review |
Comment on attachment 8893364 [details] Bug 1385311 - Make sure AddressSanitizer errors turn the task orange, https://reviewboard.mozilla.org/r/164462/#review169836 nice fix
Attachment #8893364 -
Flags: review?(jmaher) → review+
Pushed by ahalberstadt@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/b804014faa13 Make sure AddressSanitizer errors turn the task orange, r=jmaher
Comment 8•7 years ago
|
||
bugherder |
https://hg.mozilla.org/mozilla-central/rev/b804014faa13
You need to log in
before you can comment on or make changes to this bug.
Description
•