966996 - Repeating linear gradients applied to giant boxes render slowly and incorrectly on Mac

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Markus Stange [:mstange]]

Attachment: testcase

The testcase has a scrollable area that contains a 600px * 80000px box that has a repeating linear gradient as its background. Scrolling the box is very slow and the gradient stops are in the wrong locations.

Comment 1

[Markus Stange [:mstange]]

Attachment: part 1: respect clip when calculating gradient stops

This fixes the performance when we're not scrolled down much.

Comment 2

[Markus Stange [:mstange]]

Attachment: part 2: Don't include t=0 when not necessary

We calculate t_min and t_max in such a way that they always include zero. That's unnecessary because our displayed part of the gradient may be far away from zero, so initialize t_min and t_max to +- Infinity so that they always get replaced by the UpdateLinearParametersToIncludePoint call for the extent rect's top left corner.

This fixes the case when the testcase is scrolled down a lot.

Comment 3

[Markus Stange [:mstange]]

Comment on attachment 8369438 [details] [diff] [review]
part 2: Don't include t=0 when not necessary

>+  MOZ_ASSERT(!isinf(t_min) && !isinf(t_max),
>+             "The first call to UpdateLinearParametersToIncludePoint should have made t_min and t_max non-infinite.");

I realize now that "finite" is the short way of saying "non-infinite". Will fix before checkin.

Comment 4

[Markus Stange [:mstange]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/28b77fb96e62
https://hg.mozilla.org/integration/mozilla-inbound/rev/7d6e5cd7fe20
Oops, I forgot to address comment 3. Oh well.

Comment 5

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/28b77fb96e62
https://hg.mozilla.org/mozilla-central/rev/7d6e5cd7fe20