I'm guessing that this is from the cron job on Jeff's workstation rather than the builder since https://chromium-review.googlesource.com/c/444149/ isn't committed yet -- so the easiest way to stop this without logging into Jeff's workstation would be to export another CL after this?

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/056c8fa959aa437e99d8a7f4ad1d717bebc6c81a

commit 056c8fa959aa437e99d8a7f4ad1d717bebc6c81a
Author: qyearsley <qyearsley@chromium.org>
Date: Fri Feb 17 06:31:59 2017

Trivial exportable change in wpt.

TBR=jeffcarp
BUG= 693350 

Review-Url: https://codereview.chromium.org/2699063002
Cr-Commit-Position: refs/heads/master@{#451247}

[modify] https://crrev.com/056c8fa959aa437e99d8a7f4ad1d717bebc6c81a/third_party/WebKit/LayoutTests/external/wpt/referrer-policy/README.md

Just got back, I logged into my workstation and stopped the cron job. I'll investigate this first thing tomorrow morning and send out a postmortem.

I found the root of the problem - our internal WPT mirror fell behind GitHub. Compare:

https://chromium.googlesource.com/external/w3c/web-platform-tests/+log/master
https://github.com/w3c/web-platform-tests/commits/master

This made it so that the 'most recent exportable commit' stayed constant, resulting in PRs of that commit being made repeatedly.

I'm working around the problem locally by pointing local_wpt.py at github/master instead of origin/master. There are two exportable commits that need to be exported.

We should get a sense for whether this is something that might happen in the future. Since we're pulling both the internal and GitHub repositories as part of the WPT Export script, and the GitHub remote must be pulled so that we can create PRs, it might make sense to not use the internal repo anymore in the WPT Export script.

Interesting and subtle :-) Yeah, it seems like it wouldn't be any worse to directly use the GitHub repo instead of our mirror for export. (For import, what would happen when our mirror and GitHub get out of sync? We'll just have more delay in tests getting imported, right?)

One major mystery: it seems like our mirror stopped getting the latest commits at the same time that the first 'Do not guess UTF8 encoding' PR was made; were those two things related? Could there have been something about that patch that affected the googlesource.om mirror? And what mechanism is responsible for keeping mirrors synchronized, anyway?

I wrote up a quick postmortem on this:
https://docs.google.com/document/d/1cttcS0NNfUHKvCbbMrrqdOyZZ7_3LbXYhc2VvrhO3v4/edit?usp=sharing

Lowering priority since it is no longer an emergency. I'll keep this bug open until  issue 693651  (which eliminates the possibility of this happening) is landed.

Marking as fixed since  issue 693651  is now fixed, and we believe that the same thing wouldn't happen in exactly the same way :-)

Thanks for dealing with this so swiftly!