Bug 237732 – [css-grid] Grid contents sometimes layout with the wrong width

• Status: RESOLVED
• Resolution: FIXED
• Component: Layout and Rendering

[Matt Woodrow]

<rdar://88512506>

Resizing the browser on the provided URL sometimes lays out the main article text to the wrong width, where it doesn't line wrap at all and just extends off the right side of the page.

It looks like we call RenderGrid::computeIntrinsicLogicalWidths and that results in us calling layoutIfNeeded on the child RenderBlock with an estimated width (which turns out to be wrong).

Now that we've mutated the child and laid it out incorrectly, nothing marks the RenderGrid itself as needing layout, so we never do a proper layout with the final grid column sizes to fix this again.

The bug happens intermittently, since anytime we do a full layout of the grid, we also layout the children correctly.

[Matt Woodrow]

*** Bug 237710 has been marked as a duplicate of this bug. ***

[Matt Woodrow]

Bug 237710 has a reduced test case for this: https://bugs.webkit.org/attachment.cgi?id=454343

This is a regression from https://bugs.webkit.org/show_bug.cgi?id=232140

[Matt Woodrow]

It seems like we shouldn't be laying out children of a RenderGrid from (const!) computeIntrinsicLogicalWidths using estimated grid breadths.

Could we cache the intrinsic sizes when we do a full layout, and reuse those if we haven't marked the RenderGrid as needing layout?

A quick prototype of that seems to work.

[Matt Woodrow]

Looks like this showed up previously, but went away due to a change in the flex code. This is new variant of the same existing issue.

https://bugs.webkit.org/show_bug.cgi?id=209282

[Matt Woodrow]

Created attachment 454926 [details]
Patch

[Javier Fernandez]

Comment on attachment 454926 [details]
Patch

The change looks fine, but there are a few tests that regressed.

[Matt Woodrow]

Created attachment 455971 [details]
Patch

[EWS]

Committed r292079 (249006@main): <https://commits.webkit.org/249006@main>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 455971 [details].