Bugzilla

One observation... It seems like this is reliably associated with `canvas` operations, at least in our four crashtest failures.

Looking at the orange-factor history for the crashtest testsuites, it looks like `canvas` is the common thread.
Crashtest runs:
- The two `crashtest-swr-6` failures are during https://searchfox.org/mozilla-central/rev/e66cff951667dacd0faa95dfde830564a58a8a3f/dom/media/test/crashtests/1378826.html which draws to a canvas.
- The `crashtest-swr-4` failure is during https://searchfox.org/mozilla-central/rev/e66cff951667dacd0faa95dfde830564a58a8a3f/dom/canvas/crashtests/360293-1.html
- The `crashtest-11` failure is during https://searchfox.org/mozilla-central/rev/e66cff951667dacd0faa95dfde830564a58a8a3f/gfx/tests/crashtests/766422-2.html
...and all three of those tests use an HTML `<canvas>` with some scripted drawing onto it.

The mochitest failures don't have as clear of a connection to canvas.  We do have two failures on `mochitest-a11y-1proc` where the issue is during https://searchfox.org/mozilla-central/rev/e66cff951667dacd0faa95dfde830564a58a8a3f/accessible/tests/mochitest/elm/test_HTMLSpec.html which does include a section for testing `HTML:canvas` -- but the other mochitest failures don't seem related to canvas.  (Two of our mochitest failures are mis-stars, too -- a `mochitest-plain-spi-nw-1` failure and a `mochitest-browser-chrome-swr-16` failure).

And then, as noted before, the reftest failures are nearly all during `drawSnapshot` calls ([this API](https://searchfox.org/mozilla-central/rev/e66cff951667dacd0faa95dfde830564a58a8a3f/dom/chrome-webidl/WindowGlobalActors.webidl#127-148)), aside from the one that I linked in comment 59 where we failed a bit earlier.

So: focusing just on the (few) crashtest results and the (many) reftest results, it seems like this is typically a failure during canvas operations or `drawSnapshot`.

One observation... It seems like this is reliably associated with `canvas` operations, at least in our four crashtest failures.

Looking at the orange-factor history for the crashtest testsuites, it looks like `canvas` is the common thread.
Crashtest runs:
- The two `crashtest-swr-6` failures are during https://searchfox.org/mozilla-central/rev/e66cff951667dacd0faa95dfde830564a58a8a3f/dom/media/test/crashtests/1378826.html which draws to a canvas.
- The `crashtest-swr-4` failure is during https://searchfox.org/mozilla-central/rev/e66cff951667dacd0faa95dfde830564a58a8a3f/dom/canvas/crashtests/360293-1.html
- The `crashtest-11` failure is during https://searchfox.org/mozilla-central/rev/e66cff951667dacd0faa95dfde830564a58a8a3f/gfx/tests/crashtests/766422-2.html
...and all three of those tests use an HTML `<canvas>` with some scripted drawing onto it.

The mochitest failures don't have as clear of a connection to canvas.  We do have two failures on `mochitest-a11y-1proc` where the issue is during https://searchfox.org/mozilla-central/rev/e66cff951667dacd0faa95dfde830564a58a8a3f/accessible/tests/mochitest/elm/test_HTMLSpec.html which does include a section for testing `HTML:canvas` -- but the other mochitest failures don't seem related to canvas.  (Two of our mochitest failures are mis-stars, too -- a `mochitest-plain-spi-nw-1` failure and a `mochitest-browser-chrome-swr-16` failure).

And then, as noted before, the reftest failures are nearly all during ~`drawSnapshot`~ `drawWindow` calls aside from the one that I linked in comment 59 where we failed a bit earlier. [EDIT: I initially said `drawSnapshot` but I meant `drawWindow`.]

So: focusing just on the (few) crashtest results and the (many) reftest results, it seems like this is typically a failure during canvas operations or `drawWindow`.