macOS 14.4 Regression: HTML elements in a WebView (Legacy) do not repaint when their width/height change

[Jonathan Deutsch]

Created attachment 470332 [details]
Screenshot of reproduction

SUMMARY:
Using the old WebView class, elements with 3D transforms do not repaint when their size changes programmatically.

STEPS TO REPRODUCE:
1. Create an app that uses an WebView (the pre-WKWebView one)
2. Have code like:

    <div id="elm" style="transform: rotateY(0deg); height: 100px; width: 100px; background-color:#ccc"></div>
    <script>
        window.setTimeout(function() {
            elm.style.height = "10px";
        }, 1000);
    </script>

EXPECTED RESULTS:
The height of the div should change from 100px to 10px after one second.

ACTUAL RESULTS:
It does not change, the div continues to look like it is 100px.
(If you do certain things that trigger a repaint though (like click in the webview), you will eventually see it.)

FILES:
I have attached a screenshot comparing WebView(Legacy) and WKWebView, along with an Xcode project that reproduces the problem.

NOTES:
- This is a *regression* in macOS 14.4. 
- It does not occur in macOS 14.3.
- It does not occur with WKWebViews.
- It does not occur if you remove the 'transform: rotateY(0deg)' style.

HARM:
This affects all users of Tumult Hype: http://tumult.com/hype/. Resizing elements is a critical function of the application!

[Jonathan Deutsch]

Created attachment 470333 [details]
Reproduction Xcode project

[Radar WebKit Bug Importer]

<rdar://problem/124564409>

[Simon Fraser (smfr)]

This only happens if the layer has no painted content; putting a   in the div works around the bug. Nothing seems to trigger a compositing update.

[Simon Fraser (smfr)]

Probably a regression from bug 84393.

[Simon Fraser (smfr)]

Pull request: https://github.com/WebKit/WebKit/pull/27589

[EWS]

Committed 277850@main (1d9888c85e20): <https://commits.webkit.org/277850@main>

Reviewed commits have been landed. Closing PR #27589 and removing active labels.

[Jonathan Deutsch]

I'm still able to reproduce this bug.  Please see the attached video '270898-still-repro.mov'.

I downloaded the 278547@main webkit nightly, removed the quarantine bits, and set the DYLD_FRAMEWORK_PATH to the Release folder path.  I added an NSLog to look at the com.apple.WebKit and verify I was in fact linked against a different framework.

If my setup in doing this is wrong please let me know.  How did you verify the fix?


(Side note: I changed the delay in the RenderBug.html file to 3 seconds instead of 1 second -- there was something in the WKWebView load that must have tickled it to re-render within this time when linked against the nightly. When at 3s, the bug still reproduces when it should not.)

[Jonathan Deutsch]

Created attachment 471332 [details]
270898-still-repro.mov