1500575 - Can we do better than blob images for some clip-paths?

Status: RESOLVED FIXED

(Core :: Graphics: WebRender, enhancement)

Description

[Emilio Cobos Álvarez (:emilio)]

Bug 1500300 triggers a giant blob image as an image mask because the container has a fixed-position child underneath.

However it looks to me like we ough to be able to push a clip rect instead of a blob image, since the page is only using the inset() function.

I'm not familiar with all the clip-path setup in order to know whether the current behavior is just suboptimal or needed for correctness in some cases.

Markus, does the above sound reasonable?

Comment 1

[Emilio Cobos Álvarez (:emilio)]

(See comment 0)

Comment 2

[Emilio Cobos Álvarez (:emilio)]

Also, https://christopher-tierney.com/apotheke-la/ is a real-world page that would trigger this... I guess we should first measure once I get the tiled mask images landed.

Comment 3

[Markus Stange [:mstange]]

It would be very valuable to optimize inset() clip paths into rectangle clips. There should be no difference from a correctness perspective. (It's possible that we currently don't pixel-snap inset() clips, so they might have blurry edges, but that's not a desirable result anyway.)

Comment 4

[Emilio Cobos Álvarez (:emilio)]

(Just so it doesn't fall of my radar)

Comment 5

[Emilio Cobos Álvarez (:emilio)]

Attachment: Bug 1500575 - Map inset clip paths to WR complex clip regions.

Comment 6

[Emilio Cobos Álvarez (:emilio)]

Attachment: Testcase

Comment 7

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/9140cba9a8f8
Map inset clip paths to WR complex clip regions. r=kats

Comment 8

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/bebe2e4eab25
De-fuzz another webrender test. r=me

Comment 9

[Eliza Balazs [:ebalazs_]]

https://hg.mozilla.org/mozilla-central/rev/9140cba9a8f8
https://hg.mozilla.org/mozilla-central/rev/bebe2e4eab25