1904566 - Perma layout/reftests/svg/filters/feDropShadow-01.svg == layout/reftests/svg/filters/feDropShadow-01-ref.svg | image comparison, max difference: 97, number of differing pixels: 22869 when Gecko 129 merges to beta on 2024-07-08

Status: VERIFIED FIXED

(Core :: Graphics: WebRender, defect)

Description

[Serban Stanca [:SerbanS]]

Central-as-beta-simulation

How to run these simulations

Failure log

Comment 1

[Daniel Holbert [:dholbert]]

I'm adjusting component to match bug 1896503, the regressor here.

Adding a few more notes while I'm here... from the log:

REFTEST INFO | REFTEST fuzzy test (55, 14033) <= (97, 22869) <= (98, 16467)
REFTEST TEST-UNEXPECTED-FAIL | layout/reftests/svg/filters/feDropShadow-01.svg == layout/reftests/svg/filters/feDropShadow-01-ref.svg |
  image comparison, max difference: 97, number of differing pixels: 22869

The allowed range there seems to come from this annotation:
fuzzy-if(!useDrawSnapshot,55-98,14033-16467)
...and the mismatching pixel count (22869) is very outside the allowable range there (for this annotation and all of the annotations on this line of the manifest).

That seems to be a substantial increase in the number of mismatching pixels on any platform/config. Not sure if that's concerning or not.

Comment 2

[Ashley Hale [:ahale]]

I'll look into this regression today.

Comment 3

[Ashley Hale [:ahale]]

https://searchfox.org/mozilla-central/source/layout/reftests/svg/filters/reftest.list#118
Excerpt: fuzzy-if(winWidget,0-2,0-25) fuzzy-if(!useDrawSnapshot,55-98,14033-16467) fuzzy-if(gfxSVGFEDropShadow&&!gfxSVGFEOffset,97-98,21575-22951) fuzzy-if(gfxSVGFEOffset,0-2,0-1000000) == feDropShadow-01.svg feDropShadow-01-ref.svg

The value 22869 for failed pixels is within the range I was seeing with gfxSVGFEDropShadow without gfxSVGFEOffset (where it ends up comparing the result from the blob fallback in Gecko in the feGaussianBlur+feOffset case, with the result from native WebRender rendering of SVGFEDropShadow), but gfx.webrender.svg-filter-effects is nightly build only ( https://searchfox.org/mozilla-central/rev/67956a15835fc397334668525d528d471b19b273/modules/libpref/init/StaticPrefList.yaml#6530 ), so I am not yet certain why gfxSVGFEDropShadow conditional would differ from the actual state of the prefs.

I'll continue debugging, but suffice to say this is within tolerances I expect for this particular scenario, but why it is behaving this way is unclear.

Note: There is no firm guideline on blur accuracy in SVG or CSS specs so we're given a lot of leeway on deviations like this, so at worst we expand the fuzzy range to fuzzy(0-98,0-22951), which would still flag any major bugs in feDropShadow rendering, but I'd prefer to have a tighter tolerance and understand this case.

Comment 4

[Ashley Hale [:ahale]]

The cause turned out to be a missing pref check in FilterInstance::BuildWebRenderFilters, so it was executing the SVGFE code path even if the top level pref was off (the individual prefs for particular filters still worked fine, but it wasn't supposed to get that far outside of Nightly builds).

Comment 5

[Ashley Hale [:ahale]]

Waiting for phabricator to catch up so I can land fix D214912

Comment 6

[Ashley Hale [:ahale]]

Attachment: Bug 1904566 - check gfx.webrender.svg-filter-effects pref before building SVGFE filters r?#gfx-reviewers

Comment 7

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/91d635836bae
check gfx.webrender.svg-filter-effects pref before building SVGFE filters r=gfx-reviewers,gw

Comment 8

[Sandor Molnar[:smolnar]]

Backed out for causing reftest failures

Backout link: https://hg.mozilla.org/integration/autoland/rev/52dafeb18b382afe8a927a7295774fe2efee2220

Push with failures

Failure log -> REFTEST TEST-UNEXPECTED-FAIL | layout/reftests/svg/filters/feGaussianBlur-4.svg == layout/reftests/svg/filters/feGaussianBlur-4-ref.svg

Comment 9

[Donal Meehan [:dmeehan]]

:ahale, the Fx129 soft code freeze starts on Thursday 2024-07-04, ahead of Fx129 merging to beta on Monday 2024-07-08.
There is little time to fix this before next week.

Comment 10

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/7b206bd8dfb4
check gfx.webrender.svg-filter-effects pref before building SVGFE filters r=gfx-reviewers,gw

Comment 11

[Norisz Fay [:noriszfay]]

https://hg.mozilla.org/mozilla-central/rev/7b206bd8dfb4

Comment 12

[Norisz Fay [:noriszfay]]

The patch fixed the initial issues but other failures seem to have appeared as mentioned in Bug 1905354. Not closing it as verified for now, in case a patch will be submitted here

Comment 13

[Cosmin Sabou [:CosminS]]

Patch posted in https://bugzilla.mozilla.org/show_bug.cgi?id=1905354#c8

Comment 14

[Intermittent Failures Robot]

4 failures in 3199 pushes (0.001 failures/push) were associated with this bug in the last 7 days.

Repository breakdown:

• try: 4

Platform and build breakdown:

• macosx1015-64-devedition-qr: 1
  • opt: 1
• windows11-32-2009-qr: 1
  • debug: 1
• linux1804-64-qr: 1
  • debug: 1
• linux1804-64-tsan-qr: 1
  • opt: 1

For more details, see:
https://treeherder.mozilla.org/intermittent-failures/bugdetails?bug=1904566&startday=2024-07-01&endday=2024-07-07&tree=all

Comment 15

[BugBot [:suhaib / :marco/ :calixte]]

Set release status flags based on info from the regressing bug 1896503