1697629 - Fission + sw-wr: UNEXPECTED-FAIL | testing/marionette/harness/marionette_harness/tests/unit/test_reftest.py TestReftest.test_cache_multiple_sizes | AssertionError: 'FAIL' != 'PASS'

Status: RESOLVED FIXED

(Core :: Graphics: WebRender, defect)

Description

[Chris Peterson [:cpeterson]]

In bug 1697195, we tried to auto-enable SW-WR with Fission is enabled. Unfortunately, some tests failed when run with Fission + SW-WR mode:

Marionette failures:

https://treeherder.mozilla.org/jobs?repo=autoland&searchStr=linux%2C18.04%2Cx64%2Copt%2Ctest-linux1804-64%2Fopt-marionette-headless-fis-e10s%2Cmnh-fis&tochange=a24d9f6247aa4211e227c56db389df0255fc7cc7&fromchange=03e3558727b689d50eefeb19f0d927341d8f2255&selectedTaskRun=M9DmAYu0RHeQ11LnVoLMsw.0).

https://treeherder.mozilla.org/logviewer?job_id=332576581&repo=autoland&lineNumber=57773

TEST-UNEXPECTED-FAIL | testing/marionette/harness/marionette_harness/tests/unit/test_reftest.py TestReftest.test_cache_multiple_sizes | AssertionError: 'FAIL' != 'PASS'

I don't know if this test failure requires Fission + SW-WR or just SW-WR.

This bug may be caused by or fixed by Puppeteer bug 1693011.

Comment 1

[Chris Peterson [:cpeterson]]

Sending this bug to gfx-triage.

Does this Marionette test fail when is SW-WR enabled without Fission? This seems like a SW-WR bug.

Comment 2

[Brad Werth [:bradwerth]]

I will attempt to reproduce.

Comment 3

[Chris Peterson [:cpeterson]]

Tracking for Fission M7a Beta. TBD whether SW-WR will block Fission MVP.

Comment 4

[Brad Werth [:bradwerth]]

Can't reproduce on macOS. I'll keep needinfo open while I rebuild for Linux and try there. We'll keep this in triage.

Comment 5

[Brad Werth [:bradwerth]]

I can't reproduce this on Linux with the patches from Bug 1697195 applied. I believe this bug is specific to automation.

Comment 6

[Brad Werth [:bradwerth]]

Reproduction command should be: ./mach test testing/marionette/harness/marionette_harness/tests/unit/test_reftest.py

Comment 7

[Jim Mathies [:jimm]]

Managed to reproduce this on try - https://treeherder.mozilla.org/jobs?repo=try&revision=20f6177b9dec25b04df8ec6d513e3660b04b4f07&selectedTaskRun=Dqc3l9PFTFyYLkhbgzpS8g.0

Comment 8

[Jim Mathies [:jimm]]

Note this is the headless ('MnH') set of test runs. Might be interesting to understand what might be going wrong but I don't think this is a priority.

Comment 9

[Matt Woodrow (:mattwoodrow)]

Attachment: Bug 1697629 - Allocate a temporary framebuffer the size of the widget when running in headless mode, so that reftest readback works correctly. r?lsalzman

Comment 10

[Matt Woodrow (:mattwoodrow)]

HeadlessCompositorWidget doesn't let us map a widget DT (since there is no widget), and we instead did a fallback of allocating a 2x2 temporary framebuffer.

Readback on this asserts in a debug build (since we try to readpixels outside the 2x2 bounds), and also fails this test since it was expecting a not-equal result and the difference was also outside the 2x2 bounds (at 600,600).

Comment 11

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/33bce767b192
Allocate a temporary framebuffer the size of the widget when running in headless mode, so that reftest readback works correctly. r=lsalzman

Comment 12

[Sandor Molnar[:smolnar]]

https://hg.mozilla.org/mozilla-central/rev/33bce767b192