747718 - ColorLayer optimization doesn't account for rounded rect clipping

Status: RESOLVED FIXED

(Core :: Layout, defect)

Description

[Matt Woodrow (:mattwoodrow)]

Attachment: Don't convert items with rounded rect clips into ColorLayers

This causes an existing test to fail too, but only with win7 (USE_WIDGET_LAYERS disabled, so visible area == invalidation area), and DLBI (reduced invalidation area).

This fixes the problem, and adds an explicit test that should fail everywhere without the fix.

Comment 1

[Matt Woodrow (:mattwoodrow)]

I'm a little bit worried about the method I used to disable the ColorLayer in the reference image (adding obscured content behind the solid color), since it's a case that probably *should* still create a ColorLayer if someone bothers to optimize this.

This would silently change the test to be useless.

Any ideas for more resilient methods to avoid a ColorLayer? I think we might have a similar issue with tests assuming a canvas object will force an active layer - no longer true I believe?

Comment 2

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Comment on attachment 617296 [details] [diff] [review]
Don't convert items with rounded rect clips into ColorLayers

Review of attachment 617296 [details] [diff] [review]:
-----------------------------------------------------------------

In the reference testcase, how about including two small boxes on top of the background, one on top of the other, the lower one with border:white and the upper one with border:blue? I can't see us ever optimizing the white one away.

Comment 3

[Matt Woodrow (:mattwoodrow)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/3ca6df58d05d

Comment 4

[:Ms2ger (he/him; ⌚ UTC+1/+2)]

https://hg.mozilla.org/mozilla-central/rev/3ca6df58d05d