Same thing in the ToT bot, but going back to 22-23 June: https://build.chromium.org/p/chromium.fyi/builders/ClangToTWin%28dbg%29%20tester?numbuilds=200

Being flaky suggests some kind of race I suppose, but it reproduces reliably on my machine.

RenderFrameHostManagerTest.SwapProcessWithRelNoopenerAndTargetBlank is checking the value of document.referrer, expecting something like http://127.0.0.1:60967/click-noreferrer-links.html but getting an empty string instead.

If I move the WaitForLoadStop(new_shell->web_contents())) from below the test to right before it, it passes.

I suspect this was just always racy, and something in Clang made the timing of one of the racing code paths change.

Not sure what's up with the webrtc test though..

> Not sure what's up with the webrtc test though..

That one doesn't reproduce for me locally, but it's good for comparing what error messages are expected, and what aren't :-)

It seems most of the spew is expected, but I don't think audio_manager_win.cc(464)] GetPreferredAudioParameters failed: 88890004 is.

> It seems most of the spew is expected, but I don't think audio_manager_win.cc(464)] GetPreferredAudioParameters failed: 88890004 is.

Maybe that's a red herring actually: perhaps the bot doesn't have an audio device but my machine does, and the test should still work.


There's another error that doesn't show on my machine though:

4856:540:0628/210930:35449526:ERROR:audio_input_sync_writer.cc(237)] AISW: No room in fifo

Perhaps that's indicative of the error.

https://codereview.chromium.org/2113893002/ for RenderFrameHostManagerTest

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/d662b254af5421e0b53b67f561377657fcd67686

commit d662b254af5421e0b53b67f561377657fcd67686
Author: hans <hans@chromium.org>
Date: Fri Jul 01 14:37:50 2016

Attempt to un-flake RenderFrameHostManagerTest.SwapProcessWithRelNoopenerAndTargetBlank

This test has been failing flakily in Windows Clang debug builds for a while. I
can reproduce the failure locally, and this patch makes it go away.

It seems there's a race between document.referrer being set and the test
checking for it. Other tests in this file do "Wait for the cross-site
transition in the new tab to finish" before checking the referer, so let's do
that here too.

BUG= 624523 
CQ_INCLUDE_TRYBOTS=tryserver.chromium.linux:linux_site_isolation

Review-Url: https://codereview.chromium.org/2113893002
Cr-Commit-Position: refs/heads/master@{#403452}

[modify] https://crrev.com/d662b254af5421e0b53b67f561377657fcd67686/content/browser/frame_host/render_frame_host_manager_browsertest.cc

 bug 625024  covers WebRtcGetUserMediaBrowserTest.GetUserMediaWithMandatorySourceID -- that one might be unrelated to clang.

So let's close this out for now. Thanks for fixing the other bug!