Closed Bug 1865654 Opened 1 year ago Closed 10 months ago

Intermittent pixel-rounding/border-image-width-4.html == pixel-rounding/border-image-width-0.html | single tracking bug

Categories

(Core :: Graphics: WebRender, defect)

defect

Tracking

()

RESOLVED FIXED
124 Branch
Tracking Status
firefox123 --- fixed
firefox124 --- fixed

People

(Reporter: intermittent-bug-filer, Assigned: dholbert)

References

Details

(Keywords: intermittent-failure, intermittent-testcase, Whiteboard: [stockwell unknown])

Attachments

(1 file)

Filed by: smolnar [at] mozilla.com
Parsed log: https://treeherder.mozilla.org/logviewer?job_id=436846768&repo=try
Full log: https://firefox-ci-tc.services.mozilla.com/api/queue/v1/task/X9Xz1yDIQ0uwg9PXoMC1yA/runs/0/artifacts/public/logs/live_backing.log
Reftest URL: https://hg.mozilla.org/mozilla-central/raw-file/tip/layout/tools/reftest/reftest-analyzer.xhtml#logurl=https://firefox-ci-tc.services.mozilla.com/api/queue/v1/task/X9Xz1yDIQ0uwg9PXoMC1yA/runs/0/artifacts/public/logs/live_backing.log&only_show_unexpected=1


REFTEST TEST-START | layout/reftests/pixel-rounding/border-image-width-4.html == layout/reftests/pixel-rounding/border-image-width-0.html
[task 2023-11-20T13:01:53.859Z] 13:01:53     INFO - REFTEST TEST-LOAD | file:///Z:/task_170048272076390/build/tests/reftest/tests/layout/reftests/pixel-rounding/border-image-width-4.html | 158 / 163 (96%)
[task 2023-11-20T13:01:53.938Z] 13:01:53     INFO - REFTEST INFO | REFTEST fuzzy test (0, 0) <= (112, 29) <= (112, 25)
[task 2023-11-20T13:01:54.215Z] 13:01:54     INFO - REFTEST TEST-UNEXPECTED-FAIL | layout/reftests/pixel-rounding/border-image-width-4.html == layout/reftests/pixel-rounding/border-image-width-0.html | image comparison, max difference: 112, number of differing pixels: 29
[task 2023-11-20T13:01:54.216Z] 13:01:54     INFO - REFTEST   IMAGE 1 (TEST)
INFO - REFTEST   IMAGE 2 (REFERENCE):
INFO - REFTEST TEST-END | layout/reftests/pixel-rounding/border-image-width-4.html == layout/reftests/pixel-rounding/border-image-width-0.html

This is just an incremental iteration of a known-fuzzy issue discussed in bug 1842623.

See bug 1842623 comment 9 in particular. We had an annotation with a differing-pixel-count of 21, and we bumped it to 25 in that bug (+4), and it looks like we need to bump it to 29 (+4 again) now.

Though also, it's interesting that we didn't hit this for months and now we've hit it 3 times in the past week somehow, based on the orange factor graph? All three of the failures seem to be on a single platform on a single try run:
Windows 11 x64 22H2 WebRender DevEdition opt
https://treeherder.mozilla.org/jobs?repo=try&revision=080000b5e337362a9b5887dad2b7f5ad31598111&searchStr=Windows%2C11%2Cx64%2C22H2%2CWebRender%2CDevEdition%2Copt%2CR3

...for the "staging release: 121.0b1" push. So maybe this is worse with some nightly-only feature preffed off, specifically on Windows, for some reason?

nical, do you know what might be going on (what preffable features might be turned off when we switch to beta) that makes extra pixels fuzzy here? And should we just update the annotation to reflect reality (s/25/29/), or do you want to poke at anything else?

Component: Layout → Graphics: WebRender
Depends on: 1842623
Flags: needinfo?(nical.bugzilla)

(I can also imagine this being some sort of spooky-action-at-a-distance, where a pref-twiddle for some earlier test (to enable an otherwise nightly-only-feature) turns out to induce the intermittent behavior here. That sort of thing could explain the beta-vs-nightly release-train change in behavior.)

I can't think of recent changes that could have affected borders, but forwarding the needinfo to Glenn since I haven't touched webrender in a long time.

This test seems to exacerbate float arithmetic issues. There is a lot going on in the border shader so even a driver update might have made things change slightly.
We should probably just increase the reftest fuzziness.

Flags: needinfo?(nical.bugzilla) → needinfo?(gwatson)

Nothing obvious I can think of that has changed recently in WR to cause that. I do recall that some of the Windows CI machines were recently updated (perhaps Windows version?) or something like that? Perhaps that might explain what changed?

Flags: needinfo?(gwatson)

This seems to be specific to this week's merge-to-beta, so it's something specific to that configuration, I think (presumably some feature or set of tests being disabled on beta, in a way that has a direct or indirect/downstream effect on this tests).

Here's the intermittent-failures view:
https://treeherder.mozilla.org/intermittent-failures/bugdetails?startday=2023-11-14&endday=2023-11-21&tree=all&failurehash=all&bug=1865654

Note that all of the failures are on try "nightly-as-beta/staging" releases, or on actual-beta.

Anyway: seems not-concerning, so we should just bump up the fuzzy pixel count to 29 for border-image-width-4.html and border-image-width-9.html. (This bug was filed for the former test, but about half of the failure logs have the same failure in the latter test too, e.g. this one.)

These failures still show 28 or 29 pixels mismatching, for border-image-width-4.html and/or border-image-width-9.html

Recent examples of each:
28px: https://treeherder.mozilla.org/logviewer?job_id=444046236&repo=mozilla-central&lineNumber=6278
29px: https://treeherder.mozilla.org/logviewer?job_id=443346131&repo=mozilla-beta&lineNumber=9596

Let's just bump the threshold to allow 29, since this increase-in-fuzziness is small and seems to be unconcerning as discussed above.

Assignee: nobody → dholbert
Status: NEW → ASSIGNED
Pushed by dholbert@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/4a31b5feaf19 Bump fuzzy threshold to reflect reality, for reftests border-image-width-4.html and -9.html. r=TYLin
Status: ASSIGNED → RESOLVED
Closed: 10 months ago
Resolution: --- → FIXED
Target Milestone: --- → 124 Branch
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: