1717627 - Move devtools tests to Linux WebRender

Status: RESOLVED FIXED

(Core :: Graphics: WebRender, task, P3)

Comment 1

[Andrew Osmond [:aosmond] (he/him)]

Attachment: Bug 1717627 - Move mochitest-devtools to run on WebRender only on Linux.

Comment 2

[Andrew Osmond [:aosmond] (he/him)]

Attachment: Bug 1717627 - Move Linux devtools tests to WebRender.

--- target_task_set@d009f0738a
+++ target_task_set@ao_ci_linux_devtools
+test-linux1804-64-asan-qr/opt-mochitest-devtools-chrome-e10s
-test-linux1804-64-asan/opt-mochitest-devtools-chrome-e10s
+test-linux1804-64-qr/debug-mochitest-devtools-chrome-e10s
+test-linux1804-64-qr/debug-mochitest-devtools-chrome-fis-e10s
+test-linux1804-64-qr/debug-mochitest-devtools-chrome-swr-e10s
+test-linux1804-64-qr/opt-mochitest-devtools-chrome-e10s
+test-linux1804-64-qr/opt-mochitest-devtools-chrome-fis-e10s
-test-linux1804-64/debug-mochitest-devtools-chrome-e10s
-test-linux1804-64/debug-mochitest-devtools-chrome-fis-e10s
-test-linux1804-64/opt-mochitest-devtools-chrome-a11y-checks-e10s
-test-linux1804-64/opt-mochitest-devtools-chrome-e10s
-test-linux1804-64/opt-mochitest-devtools-chrome-fis-e10s
+test-macosx1015-64-qr/debug-mochitest-devtools-chrome-e10s

mozilla-central

--- target_task_set@d009f0738a
+++ target_task_set@ao_ci_linux_devtools
+test-linux1804-64-asan-qr/opt-mochitest-devtools-chrome-e10s
-test-linux1804-64-asan/opt-mochitest-devtools-chrome-e10s
+test-linux1804-64-ccov-qr/opt-mochitest-devtools-chrome-e10s
-test-linux1804-64-ccov/opt-mochitest-devtools-chrome-e10s
+test-linux1804-64-qr/debug-mochitest-devtools-chrome-e10s
+test-linux1804-64-qr/debug-mochitest-devtools-chrome-fis-e10s
+test-linux1804-64-qr/debug-mochitest-devtools-chrome-swr-e10s
+test-linux1804-64-shippable-qr/opt-mochitest-devtools-chrome-e10s
+test-linux1804-64-shippable-qr/opt-mochitest-devtools-chrome-fis-e10s
-test-linux1804-64-shippable/opt-mochitest-devtools-chrome-a11y-checks-e10s
-test-linux1804-64-shippable/opt-mochitest-devtools-chrome-e10s
-test-linux1804-64-shippable/opt-mochitest-devtools-chrome-fis-e10s
-test-linux1804-64/debug-mochitest-devtools-chrome-e10s
-test-linux1804-64/debug-mochitest-devtools-chrome-fis-e10s
+test-macosx1015-64-qr/debug-mochitest-devtools-chrome-e10s

fixup

Comment 3

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/044115544bf6
Move Linux devtools tests to WebRender. r=ahal

Comment 4

[Andrew Osmond [:aosmond] (he/him)]

Attachment: Bug 1717627 - Part 2. Follow up test fuzzing.

Comment 5

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/74c20eb1d866
Part 2. Follow up test fuzzing. r=gfx-reviewers,jrmuizel

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/mozilla-central/rev/ee40230413eb
Part 2. Follow up test fuzzing. r=gfx-reviewers,jrmuizel

Comment 7

[Andreea Pavel [:apavel]]

https://hg.mozilla.org/mozilla-central/rev/044115544bf6
https://hg.mozilla.org/mozilla-central/rev/ee40230413eb

Comment 8

[Atila Butkovits]

https://hg.mozilla.org/mozilla-central/rev/74c20eb1d866

Comment 9

[Joel Maher ( :jmaher ) (UTC -8)]

Comment on attachment 9231910 [details]
Bug 1717627 - Move Linux devtools tests to WebRender.

ESR Uplift Approval Request

• If this is not a sec:{high,crit} bug, please state case for ESR consideration: adjusting testing to being on webrender only; this is 99% the case for 91, and the tests are only 60% migrated.
• User impact if declined:
• Fix Landed on Version: 92
• Risk to taking this patch: Low
• Why is the change risky/not risky? (and alternatives if risky):
• String or UUID changes made by this patch:

Comment 10

[Ryan VanderMeulen [:RyanVM]]

Comment on attachment 9231910 [details]
Bug 1717627 - Move Linux devtools tests to WebRender.

Test-only change, approved for esr91.

Comment 11

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/releases/mozilla-esr91/rev/4b75c908a01d
https://hg.mozilla.org/releases/mozilla-esr91/rev/1e2939816531

Comment 12

[Julian Descottes [:jdescottes]]

Hi Andrew,

Sorry for the ni? on a closed bug, but since we moved DevTools tests on webrender, we have a lot of intermittent failures, and I'm not sure what is the best way to address them.

We had a first class of failures where we seemed to interact with the test page too quickly. After seeing one error related to a missing "presShell" available (see this comment), I started waiting for a presShell on top of the regular BrowserTestUtils.browserLoaded. It seemed to fix a whole class of DevTools intermittents. But it feels like this should either be addressed on platform side. Whatever BrowserTestUtils.browserLoaded currently waits for should probably still work with webrender without adding extra code?

And there are still intermittents which fail despite this "fix", for instance https://bugzilla.mozilla.org/show_bug.cgi?id=1721938 . We have a DOM element, in the DevTools toolbox. The element is visible, we can get its bounding client rect etc... But when we try to simulate a mousedown on it using nsDOMWindowUtils::SendMouseEvent, nothing happens. I checked the event listener was properly added before we simulate the event, I also added event listeners on the containing window, to check if we were missing the element for some reason...

My next step will be to start adding logs on the C++ side of the event simulation, but in case you have any idea of what might cause this, it might speed up the investigation.

Thanks!

Comment 13

[Andrew Osmond [:aosmond] (he/him)]

Hm. Well, Linux + llvmpipe + HW-WR is much slower than Linux + Basic, Linux + gfx hw + HW-WR and Linux + Software WR. So for tests that are more timing sensitive, I would perhaps expect more failures.

For the browser-chrome tests, I had to switch it over to use SW-WR only on Linux because the timing was just way too painful. Perhaps devtools would benefit from a similar treatment? I think largely they will act the same from a devtools perspective (aside from timing) so I'm not that worried about loss of coverage.

You can see how I managed that in bug 1721945:

https://searchfox.org/mozilla-central/rev/7bafa9b9c2156018ec1d410194f9bf4b5b23e77f/taskcluster/ci/test/mochitest.yml#125

Comment 14

[Julian Descottes [:jdescottes]]

That would explain why devtools tests seemed the only ones impacted by that. Our tests are really similar to browser chrome mochitests, so yes I assume we should do the same. I will file bug for that, thanks!

Comment 15

[Julian Descottes [:jdescottes]]

I thought a bit more about this, and I don't really understand why this would be related to a performance problem. Helpers such as BrowserLoaded have been working fine on very slow platforms (debug, ccov etc...), so it seems odd that they start having timing issues on opt builds.

It rather feels that there is now a delay between the events we wait for in BrowserLoaded and the actual moment where you can interact with the page. I will file another bug for BrowserLoaded, hoping we can find a way to fix this helper and make it wait for the correct event.