Commit r279987 caused 7 new unexpected failures on GTK and WPE ports:
Regressions: Unexpected image-only failures (7)
fast/selectors/input-with-selection-pseudo-element.html [ ImageOnlyFailure ]
imported/blink/fast/table/whitespace-between-elements-with-table-cell-display-3.html [ ImageOnlyFailure ]
imported/w3c/web-platform-tests/css/css-color-adjust/rendering/dark-color-scheme/color-scheme-change-checkbox.html [ ImageOnlyFailure ]
imported/w3c/web-platform-tests/css/css-color-adjust/rendering/dark-color-scheme/color-scheme-iframe-background-mismatch-alpha.html [ ImageOnlyFailure ]
imported/w3c/web-platform-tests/css/css-color-adjust/rendering/dark-color-scheme/color-scheme-iframe-background-mismatch-opaque.html [ ImageOnlyFailure ]
imported/w3c/web-platform-tests/css/css-color-adjust/rendering/dark-color-scheme/color-scheme-visited-link-initial.html [ ImageOnlyFailure ]
imported/w3c/web-platform-tests/html/rendering/widgets/button-layout/propagate-text-decoration.html [ ImageOnlyFailure ]
Looking at the diffs seems that there is some difference on the rounding of the corners of some boxes in some cases and also differences on the color of the text or background in other cases
Check here the results:
`imported/w3c/web-platform-tests/css/css-color-adjust/rendering/dark-color-scheme/*` ones - I see we have a Pass override for glib and otherwise it's expected to fail. Which is... interesting, but I suppose we should match Safari here.
The rounding ones - I need to check again if there's a way to only do the rounding with non-`none` appearance. Makes sense, that seems like an actual regression.
The 5 WPT tests for css-color-adjust (imported WPT tests) tests fail also on the Mac port, and they started to fail on GTK/WPE after enabling this CSS property, so it seems there is a pre-exist bug somewhere on the common code for this (not related with r279987 )
The other 2 failures seem new failures that are not reported on the Mac port:
- fast/selectors/input-with-selection-pseudo-element.html [ ImageOnlyFailure ]
- imported/blink/fast/table/whitespace-between-elements-with-table-cell-display-3.html [ ImageOnlyFailure ]
Which are the ones related with some difference on the rounding of the corners of the boxes,
(In reply to Carlos Alberto Lopez Perez from comment #2)
> An update:
> The 5 WPT tests for css-color-adjust (imported WPT tests) tests fail also on
> the Mac port, and they started to fail on GTK/WPE after enabling this CSS
> property, so it seems there is a pre-exist bug somewhere on the common code
> for this (not related with r279987 )
See bug 214455
imported/w3c/web-platform-tests/html/rendering/widgets/button-layout/propagate-text-decoration.html is also reported on the Mac port (no bug assigned), so I assume this was also a pre-existent failure not related with r279987
Updated test expectations on r280159
Mm, I think fixing the rounding would still make sense. Thanks for updating the tests!
Found the reason for the rounding. https://bugs.webkit.org/show_bug.cgi?id=173572 calls adjustTextFieldStyle() even if the element is styled:
+ case TextFieldPart:
+ adjustTextFieldStyle(styleResolver, style, element);
I wonder if that's still relevant or if it was only relevant before for RenderThemeGtk? Because then we could remove that for an easy fix.
Created attachment 435095 [details]
Comment on attachment 435095 [details]
Let's see if setting cq+ myself works.
Committed r280742 (240328@main): <https://commits.webkit.org/240328@main>
All reviewed patches have been landed. Closing bug and clearing flags on attachment 435095 [details].