1379425 - stylo: Compositor animations may make us miss restyles/use outdated styles.

Status: VERIFIED DUPLICATE of bug 1371450

(Core :: CSS Parsing and Computation, defect, P1)

Description

[Emilio Cobos Álvarez (:emilio)]

I think this is the root cause of bug 1379203, at least, and may be related to bug 1371450 too.

I'll attach a test-case that shows the a few links rotating normal restyles (one element should be green at a time).

When there's a transform animation in effect (you can click the button to trigger one), if you select some text under the <div id="container">, for example, you eventually end up with multiple green links.

In a debug build, this also happens to trigger the "lazy computation forbidden" assertion.

This is, I think, due to multiple things:

 * The slippery handling of element snapshots described in bug 1378064 comment 15.
 * The fact that the post-traversal for animations follows _both_ the animation dirty descendants bit and the normal dirty descendants bit, regardless of the kind of restyles we're doing. This makes us update the styles before the normal traversal I think, and makes us throw away all the restyle state of elements _not_ affected by animations.

This is definitely wrong, but I need more info about how all this is supposed to work with the snapshot stuff, etc, in order to figure out what the correct fix is.

Comment 1

[Emilio Cobos Álvarez (:emilio)]

Attachment: Testcase

Comment 2

[Emilio Cobos Álvarez (:emilio)]

Attachment: Slightly more reduced (but harder to repro) test-case.

In this test-case, you just have to click on the other non-active links, though it reproduces less frequently than changing the DOM content of the container and selecting it.

Comment 3

[Darkspirit]

I'd never thought that my loading circle could be the cause. That's why all my attempts to make a reduced test case didn't work. Thanks so much!

Comment 4

[Emilio Cobos Álvarez (:emilio)]

(In reply to Darkspirit from comment #3)
> I'd never thought that my loading circle could be the cause. That's why all
> my attempts to make a reduced test case didn't work. Thanks so much!

Thank _you_ for pointing me to the problematic page! That was way easier to reduce than Inbox, I'd probably have given up on reducing Inbox actually :-)

Comment 5

[Darkspirit]

Blocking stylo-site-issues as this will affect our website.

Comment 6

[Darkspirit]

(Darkspirit from bug 1380543 comment 7)
> Now I tried it with a try build from bug 1371450 comment 13:
> https://treeherder.mozilla.org/#/jobs?repo=try&revision=2deb83ce288532ce5f2b1534ffe6623b6ad9a9b2
> > Linux x64 Stylo opt > tc[tier2] > B > target.tar.bz2 (stylo is enabled by default)

With the above build I can't reproduce the bug with the testcases and my dev website anymore. Can you confirm this?

But with today's Nightly 56 x64 20170713100259 @ Debian Testing (Linux 4.11.0-1-amd64, Radeon RX480) it's still possible to reproduce. I don't know which changes happened between those two builds, but one or more fixed it.

Comment 7

[Hiroyuki Ikezoe (:hiro)]

(In reply to Darkspirit from comment #6)
> (Darkspirit from bug 1380543 comment 7)
> > Now I tried it with a try build from bug 1371450 comment 13:
> > https://treeherder.mozilla.org/#/jobs?repo=try&revision=2deb83ce288532ce5f2b1534ffe6623b6ad9a9b2
> > > Linux x64 Stylo opt > tc[tier2] > B > target.tar.bz2 (stylo is enabled by default)
> 
> With the above build I can't reproduce the bug with the testcases and my dev
> website anymore. Can you confirm this?
> 
> But with today's Nightly 56 x64 20170713100259 @ Debian Testing (Linux
> 4.11.0-1-amd64, Radeon RX480) it's still possible to reproduce. I don't know
> which changes happened between those two builds, but one or more fixed it.

Thanks for the confirmation! You are right, I believe the try fixes this bug, actually I've been trying to fix this bug in that bug. :)

Comment 8

[Darkspirit]

See comment 6 + comment 7: Fixed in try build from bug 1371450 comment 13.

Comment 9

[Darkspirit]

Fixed in Nightly 56 x64 20170716100258 @ Debian Testing (Linux 4.11.0-1-amd64, Radeon RX480).