Bugzilla

Looks like mak landed a patch that has the same effect as toggling that pref off, here:
https://hg.mozilla.org/integration/autoland/rev/13ee6ecc1f427ec1b553b9542c298d34643d26aa

I generated a bunch of tsvgx talos runs before/after that commit and it seems to have helped:
https://treeherder.mozilla.org/perfherder/compare?originalProject=autoland&newProject=autoland&newRevision=99a4c5ae7bcd8ac307ad9a448b8b29d1f6c7809e&framework=1&originalRevision=350b696788a174ceeba0e05ac3ef2be7ddd7c5af&page=1&filter=tsvgx

With the current data (39 runs before, 41 runs after), it shows a 1.39% improvement in tsvgx; and the subtest view shows a ~3% improvement in `5.xml` and `6.xml`, the ones that were most-affected (possibly just the ones that are at risk of overlapping a lazily-scheduled task?), per comment 7.

We can watch the charts for a few more days, but I think it's likely we can consider this fixed.  (To the extent that there's any remaining regression here associated with this places change, it's small enough not to worry about, and pretty-clearly not *actually* something that's impacting the paint speed that this test is generally trying to measure.)

Looks like mak landed a patch that has the same effect as toggling that pref off, here:
https://hg.mozilla.org/integration/autoland/rev/13ee6ecc1f427ec1b553b9542c298d34643d26aa

I generated a bunch of tsvgx talos runs before/after that commit, and that data confirms that it seems to have helped:
https://treeherder.mozilla.org/perfherder/compare?originalProject=autoland&newProject=autoland&newRevision=99a4c5ae7bcd8ac307ad9a448b8b29d1f6c7809e&framework=1&originalRevision=350b696788a174ceeba0e05ac3ef2be7ddd7c5af&page=1&filter=tsvgx

With the current data (39 runs before, 41 runs after), it shows a 1.39% improvement in tsvgx; and the subtest view shows a ~3% improvement in `5.xml` and `6.xml`, the ones that were most-affected (possibly just the ones that are at risk of overlapping a lazily-scheduled task?), per comment 7.

(The values might change slightly as more data accumulates, since I'm going to trigger 10-20 more runs to smooth out the impact of a few outlier data-points. But I don't suspect they'll change much.)

We can watch the charts for a few more days, but I think it's likely we can consider this fixed.  (To the extent that there's any remaining regression here associated with this places change, it's small enough not to worry about, and pretty-clearly not *actually* something that's impacting the paint speed that this test is generally trying to measure.)