That's very strange. This file should be built as part of components_unittests, and they are on the CQ AFAICT.

I can compile components_unittests locally with https://crrev.com/2373663002 patched in.  Maybe this file is not built as part of components_unittests after all?

Never mind, I got confused about rebasing and patches and reverts.  Indeed compiling components_unittest fails with the patch.

See https://codereview.chromium.org/2494153002/# for an toy example of a break in this file failing on the CQ. I'll patch locally to see what is going on.

Okay, good :). I tried to look at the actual output of the trybots you were running, and it looks like BidirectionalStream* tests were never run (https://chromium-swarm.appspot.com/task?id=3268d95331aea810&refresh=10&show_raw=1). The last diff uploaded in the CL is actually before my change landed. Does the CQ cache results for some period of time? If so, and you ran the test before my change but then submitted after my change, that would explain the issue. That seems wrong though.

Yes, #5 explains it, thank you for investigating.  https://crrev.com/2492703002
by gcasto@chromium.org landed on Nov 10 at 13:44:49-0800.  Later,
https://crrev.com/2373663002 by bnc@chromium.org landed on Nov 11 at
08:21:12-0800, but its trybots ran before the first CL landed, so they passed.
On the other hand, if I apply https://crrev.com/2373663002 to current tip of
tree, compilation fails on most trybots, see https://crrev.com/2500963004.

Trybot results are cached for some amount of time, this is intentional, so I'm
closing this issue as WAI.  In fact, even if they were not cached at all, this
race could happen if the first CL lands after the trybots start for the second
CL.