Bug 139975 – Update css3/flexbox/auto-margins.html after r177774

• Component: New Bugs
• Status: RESOLVED FIXED

[Myles C. Maxfield]

Update css3/flexbox/auto-margins.html after r177774

[Myles C. Maxfield]

Created attachment 243783 [details]
Patch

[Darin Adler]

Comment on attachment 243783 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=243783&action=review

> LayoutTests/ChangeLog:12
> +        This is a subpixel problem. Flexbox uses SimpleLineLayout for the entire
> +        box and therefore does all its math in floating point, while the
> +        -expected uses the regular renderer hierarchy and therefore fixed-point
> +        math. This patch updates the test to use Ahem so that the character
> +        widths divide the same way in both cases.

Does that mean that I could see a difference if a page dynamically changed layout between these two techniques? That sounds like a bug!

I understand the desire to not make this test sensitive to this bug, but what about fixing the bug?

[alan]

Comment on attachment 243783 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=243783&action=review

> LayoutTests/ChangeLog:10
> +        This is a subpixel problem. Flexbox uses SimpleLineLayout for the entire
> +        box and therefore does all its math in floating point, while the
> +        -expected uses the regular renderer hierarchy and therefore fixed-point

Simple line layout snaps content to pixels at painting time the same way as any other InlineBox/RenderBox does. Whether it uses float or LayoutUnit while layouting shouldn't make any difference (other than in some edge cases caused by precession loss -where we might push content to the wrong position at halfway values). I think this is more about table layout vs. rest of the rendering layout, where table still uses integral positioning.

[Myles C. Maxfield]

Committed r178063: <http://trac.webkit.org/changeset/178063>

[Myles C. Maxfield]

Zalan says we already have a few bugs about subpixelifying tables, so I'm going to land this.

[alan]

(In reply to comment #5)
> Zalan says we already have a few bugs about subpixelifying tables, so I'm
> going to land this.
bug 132010