Perfherder has detected a talos performance regression from push 60beb289fe927089b0c9931bd0d9b1e52eff2a03. As author of one of the patches included in that push, we need your help to address this regression.

Regressions:

Details of the alert can be found in the alert summary, including links to graphs and comparisons for each of the affected tests. Please follow our guide to handling regression bugs and let us know your plans within 3 business days, or the offending patch(es) will be backed out in accordance with our regression policy.

For more information on performance sheriffing please see our FAQ.

Is there a chance this could be caused by bug 1756468 instead? The identified push specifically did not touch shippable builds.

I've scheduled this test on https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=65d5050fe312fd1184e247f670ccbe22b017c648&tochange=491217cd470c4f441935ee770deb8fc303ceb2ab and that seems to match better.

Set release status flags based on info from the regressing bug 1756468

(In reply to Julien Cristau [:jcristau] from comment #1)

Is there a chance this could be caused by bug 1756468 instead? The identified push specifically did not touch shippable builds.

Yes! Thank you! Yesterday there were no job results after multiple tryings for Bug 1756468.

:jfkthame, since you are the author of the regressor, bug 1756468, could you take a look?
For more information, please visit auto_nag documentation.

This test does:

window.onload = function() {
  const TEST_WORD = "iqwdzx zzaعظأkvcg rvde";
  let div = document.getElementById("test");

  div.textContent = TEST_WORD;
  flush_layout();

  perf_start();
  div.textContent = build_text(TEST_WORD, 30, 800);
  flush_layout(div);
  perf_finish();
};

which means it's reflowing a block containing 24000 copies of the mixed Latin/Arabic TEST_WORD string.

It's not surprising that bug 1756468 would have some impact here, as it adds locking around the access to cached shaped-words in the fonts, and that's happening an awful lot of times here. I think unless we see significant regressions on more "normal" content, we should probably accept this for now.

Once more of the font-thread-safety work is landed, we can profile it with real-world workloads and try to squeeze out any particular hotspots that may show up.

:jfkthame, based on your last comment, should we accept this regression and close the bug?

I think so, but let's also ask @lsalzman if he has any further thoughts here -- Lee, does that sound OK to you?

I am okay with it.