Bug 218655 – Reversed transform animation not applied alongside other transform animations

• Component: Animations
• Status: RESOLVED FIXED

[Antoine Quint]

Created attachment 413425 [details]
Test

Consider this example:

div {
    width: 100px;
    height: 100px;
    margin-left: 100px;
    margin-top: 100px;
    background-color: black;

    animation: scale 2s forwards, rotate 500ms 1s reverse;
}

@keyframes scale {
    50% { scale: 2 }
}

@keyframes rotate {
    100% { rotate: 90deg }
}

In Safari, the rotate animation won't be performed because it's reversed and we disable accelerated animations when the playback rate is not 1 (see bug 211839). We should ensure that we only run accelerated transform animations if all of the accelerated transform animations can be accelerated.

[Radar WebKit Bug Importer]

<rdar://problem/71116284>

[Antoine Quint]

We should also test we switch to software mode if already accelerated and a non-accelerated animation is added.

[Antoine Quint]

I have the case where we try to add an animation that we think should be accelerated but fails to run accelerated due to the internals of GraphicsLayerCA working fine.

I also need to handle the case where we determine the animation cannot be run accelerated before even involving GraphicsLayerCA, such as a steps() timing function.

[Antoine Quint]

Created attachment 417627 [details]
Patch

[Simon Fraser (smfr)]

Comment on attachment 417627 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=417627&action=review

> Source/WebCore/animation/DocumentTimeline.cpp:443
> +    m_isApplyingPendingAcceleratedAnimations = true;

SetForScope<>

> Source/WebCore/animation/DocumentTimeline.cpp:466
> +        if (keyframeEffect.failedToRunAcceleratedTransformRelatedAnimation()) {

It's a bit weird to have the very specific failedToRunAcceleratedTransformRelatedAnimation() function. Could applyPendingAcceleratedActions() instead return an OptionSet<> of what it did?

> Source/WebCore/animation/KeyframeEffect.cpp:1773
> +    if (downcast<DocumentTimeline>(animation()->timeline())->isApplyingPendingAcceleratedAnimations()) {

Not a fan of code like this that needs to know what's on its call stack.

[Antoine Quint]

(In reply to Simon Fraser (smfr) from comment #5)
> Comment on attachment 417627 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=417627&action=review
> 
> > Source/WebCore/animation/DocumentTimeline.cpp:466
> > +        if (keyframeEffect.failedToRunAcceleratedTransformRelatedAnimation()) {
> 
> It's a bit weird to have the very specific
> failedToRunAcceleratedTransformRelatedAnimation() function. Could
> applyPendingAcceleratedActions() instead return an OptionSet<> of what it
> did?

Yes, I'll do this.

> > Source/WebCore/animation/KeyframeEffect.cpp:1773
> > +    if (downcast<DocumentTimeline>(animation()->timeline())->isApplyingPendingAcceleratedAnimations()) {
> 
> Not a fan of code like this that needs to know what's on its call stack.

Me neither actually, but was a bit lazy there I confess. I'll make it so that a parameter is passed down to effectFailedToRunAcceleratedTransformRelatedAnimation() to indicate whether the change should be made synchronously or as part of an accelerated action.

[Antoine Quint]

Created attachment 417682 [details]
Patch

[Antoine Quint]

Created attachment 417690 [details]
Patch

[EWS]

Committed r271524: <https://trac.webkit.org/changeset/271524>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 417690 [details].