(In reply to Emilio Cobos Álvarez (:emilio) from comment #6)
If you can reproduce on debian, running
mozregression --good 81 --bad 82 and posting the result here would be immensely helpful, see the docs for more info, and let me know if you get stuck or need any help.
Oh, I didn't know such tool existed. For a moment I thought it would do a complete build at each step, but i see it just downloads pre-compiled nightly builds. Nice!
I had to expand the bisect range since I was able to easily reproduce the issue on the very first step and it told me
Build was expected to be good! The initial good/bad range seems incorrect.
In the end it found this: https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=8fad18d87c946b2652d9284b91e5fb2424266d2f&tochange=8396d94496e0a24c3edfb4a4862d14ddb8528396
Could it make sense?
I'm attaching the full log.
I'll look to run it again, just to be sure I end up with the same result.
To me it seems to be some timing issue like a race condition. I even tried to debug it using the devtools but the scripts used in the page are really big and Firefox ends up using a lot of memory and gets killed by OOM.
I think we might try to report the issue to Google using the "Send feedback" feature (the button next to your Google account photo in any Analytics page).