| Summary: | Removing the transform CSS property from the SVG element does not cause invalidation | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Product: | WebKit | Reporter: | Pavel Panchekha <me> | ||||||||
| Component: | SVG | Assignee: | Said Abou-Hallawa <sabouhallawa> | ||||||||
| Status: | RESOLVED FIXED | ||||||||||
| Severity: | Minor | CC: | bfulgham, changseok, dino, esprehn+autocc, ews-watchlist, fmalita, glenn, gyuyoung.kim, kondapallykalyan, pdr, sabouhallawa, schenney, sergio, simon.fraser, webkit-bug-importer, zalan, zimmermann | ||||||||
| Priority: | P2 | Keywords: | InRadar | ||||||||
| Version: | Safari 14 | ||||||||||
| Hardware: | Mac (Intel) | ||||||||||
| OS: | macOS 11 | ||||||||||
| Attachments: |
|
||||||||||
Created attachment 428355 [details]
Patch
Created attachment 428384 [details]
Patch
(In reply to Said Abou-Hallawa from comment #3) > Created attachment 428384 [details] > Patch I forgot to add waitUntilDone() and notifyDone() in the layout test. Because the test requires two page updates, this would have made the test flaky. Committed r277385 (237641@main): <https://commits.webkit.org/237641@main> All reviewed patches have been landed. Closing bug and clearing flags on attachment 428384 [details]. |
Created attachment 427700 [details] Minimized test case that reproduces the bug When an SVG "transform" property is changed on two tested SVG elements, the Webkit layout engine (in Safari 14.1) does not fully update the layout. Consider the attached test file. It uses a CSS `max-width` media query to trigger a change to the `transform` property of an SVG `text` element; in a window narrow than 800px, part of the `text` is offscreen, while in a wider window none of it is. If you start with a wide window (the whole text is onscreen), resize the screen to narrow it (so that part of it is offscreen), and then widen again, the text stays partially offscreen. The same works in reverse. You can confirm that this is an underinvalidation bug by starting wide, narrowing, and then refreshing. The refresh changes the layout of the page—it triggers a from-scratch layout pass which circumvents underinvalidation bugs. I've confirmed that the issue is absent in Firefox (88) and Chrome (90) on macOS.