1404433 - Get coverage build with Clang on Linux green

Status: RESOLVED FIXED

(Testing :: Code Coverage, enhancement)

Description

[Marco Castelluccio [:marco]]

There are quite a few failures, e.g. see https://treeherder.mozilla.org/#/jobs?repo=try&revision=e30d8c3c59a9737f1825e81e6901774f562ede00 or https://treeherder.mozilla.org/#/jobs?repo=try&revision=014e361716000d0dee4427bf7195316a2aba7b21.

Most of them are timeouts, some of them are due to leak checking (that we could disable like we do for asan).

Greg, do you think you could help here? You have a lot of experience as you've done the same with the GCC build too!

Comment 1

[Greg Mierzwinski [:sparky]]

Sure thing! I'll get to it within the next couple weeks. I'm leaving a ni? open here for myself.

Comment 2

[Greg Mierzwinski [:sparky]]

Here's a test run for the latest changes (including the clang patches): https://treeherder.mozilla.org/#/jobs?repo=try&revision=73dc9ddd7ee0e3b72c0d80bbe67b7a462760de16

Comment 3

[Greg Mierzwinski [:sparky]]

Good run without RUSTFLAG: https://treeherder.mozilla.org/#/jobs?repo=try&revision=ab4006e68da88eb2705d40af90f1b98022f977d6

Good run with RUSTFLAG: https://treeherder.mozilla.org/#/jobs?repo=try&revision=26556adb8b2f4def3a372f398bbf63d2f1bdbf16

Comment 4

[Greg Mierzwinski [:sparky]]

So, the build that uses the RUSTFLAG option is failing. But the regular clang build without it is working. There are many timeouts though and only a handful of bugs from the looks of it, which is promising.

Comment 5

[Marco Castelluccio [:marco]]

(In reply to Greg Mierzwinski [:sparky] from comment #4)
> So, the build that uses the RUSTFLAG option is failing. But the regular
> clang build without it is working. There are many timeouts though and only a
> handful of bugs from the looks of it, which is promising.

Strange, the build with the Rust flags is failing because it doesn't recognize the "profile" option ([task 2017-12-16T05:14:18.647Z] 05:14:18     INFO -  error: unknown codegen option: `profile`). So it seems it is not actually using the custom Rust build.

Comment 6

[Greg Mierzwinski [:sparky]]

Here's a linux64 clang coverage run without any task timeouts, only bugs to fix: https://treeherder.mozilla.org/#/jobs?repo=try&revision=f60c1e4bbf038e3333ca22e79fc667fbebd869f8

Should we file bugs for each of them?

Comment 7

[Marco Castelluccio [:marco]]

(In reply to Greg Mierzwinski [:sparky] from comment #6)
> Here's a linux64 clang coverage run without any task timeouts, only bugs to
> fix:
> https://treeherder.mozilla.org/#/
> jobs?repo=try&revision=f60c1e4bbf038e3333ca22e79fc667fbebd869f8
> 
> Should we file bugs for each of them?

One failure is due to a wrong year (2017 instead of 2018), that's an interesting one :)
Most of the failures in web-platform tests are timeouts, we could try increasing the timeout multiplier like I did in https://bugzilla.mozilla.org/page.cgi?id=splinter.html&bug=1418283&attachment=8933264 (I increased it to 4, maybe we can increase it a little more).
Also, we need to add the "--llvm" parameter to the gcov invocation.

With these changes, could you try respinning another try build and re-running the failures 5 times (so we make sure they are not intermittents)?

Comment 8

[Marco Castelluccio [:marco]]

With all the changes that we have landed recently, we can call this fixed.