Bug 220848 – CSS aspect-ratio interpolation

• Status: RESOLVED
• Resolution: FIXED
• Component: Animations

[Manuel Rego Casasnovas]

This was discussed on the CSSWG (https://github.com/w3c/csswg-drafts/issues/4953)
and it's now in the spec: https://drafts.csswg.org/css-values/#combine-ratio

[Radar WebKit Bug Importer]

<rdar://problem/73744914>

[Antoine Quint]

Should be pretty simple to add support for this using DiscretePropertyWrapper in CSSPropertyAnimationWrapperMap::CSSPropertyAnimationWrapperMap().

[Antoine Quint]

Actually, this might be trickier since this property is set across different RenderStyle methods.

[Rob Buis]

Created attachment 423106 [details]
Patch

[EWS Watchlist]

This patch modifies the imported WPT tests. Please ensure that any changes on the tests (not coming from a WPT import) are exported to WPT. Please see https://trac.webkit.org/wiki/WPTExportProcess

[Rob Buis]

Created attachment 423121 [details]
Patch

[Rob Buis]

Created attachment 423123 [details]
Patch

[Rob Buis]

Created attachment 423129 [details]
Patch

[Antoine Quint]

Comment on attachment 423129 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=423129&action=review

Thanks for doing this, I've had it on my to-do list but I'm glad someone else did the work :) This looks good overall but I think this could be tightened up a bit before it gets r+.

> Source/WebCore/animation/CSSPropertyAnimation.cpp:1851
> +class PropertyWrapperAspectRatio : public AnimationPropertyWrapperBase {

I think this class can be marked "final"

> Source/WebCore/animation/CSSPropertyAnimation.cpp:1869
> +    bool canInterpolate(const RenderStyle* a, const RenderStyle* b) const override

I've made a pass through this entire file in r274409 to use from/to instead of a/b for canInterpolate(). This can also be marked "final".

> Source/WebCore/animation/CSSPropertyAnimation.cpp:1875
> +    void logBlend(const RenderStyle* a, const RenderStyle* b, const RenderStyle* result, double progress) const final

I've made a pass through this entire file in r274409 to use from/to instead of a/b for logBlend().

> Source/WebCore/animation/CSSPropertyAnimation.cpp:1881
> +    void blend(const CSSPropertyBlendingClient*, RenderStyle* dst, const RenderStyle* a, const RenderStyle* b, double progress) const override

I've made a pass through this entire file in r274409 to use from/to instead of a/b for blend(), as well as "destination" instead of "dst". This can also be marked "final".

> Source/WebCore/animation/CSSPropertyAnimation.cpp:1884
> +            || a->aspectRatioType() == AspectRatioType::Auto || b->aspectRatioType() == AspectRatioType::Auto) {

The code conditioned by this clause looks like it only applies to discrete animation, so should the clause use similar logic as canInterpolate()? 

I would switch the code in this method to have a canInterpolate() block where you do the regular case and return. And then have the rest of the function assert that progress is 0 or 1, as should be the case for a discrete animation (see CSSPropertyAnimation::blendProperties) and perform the discrete animation code.

> Source/WebCore/animation/CSSPropertyAnimation.cpp:1888
> +                dst->setAspectRatio(b->aspectRatioWidth(), b->aspectRatioHeight());

Maybe just:

auto* applicableStyle = progress < 0.5 ? from : to;
destination->setAspectRatio(applicableStyle->aspectRatioWidth(), applicableStyle->aspectRatioHeight());

> Source/WebCore/animation/CSSPropertyAnimation.cpp:1892
> +        double aspectRatioDst = WebCore::blend(log(a->logicalAspectRatio()), log(b->logicalAspectRatio()), progress);

"auto" instead of "double".

> Source/WebCore/animation/CSSPropertyAnimation.cpp:1894
> +        dst->setAspectRatioType(progress < 0.5 ? a->aspectRatioType() : b->aspectRatioType());

This line is duplicated in the if clause above and here. You should run it first before the code branches.

> Source/WebCore/animation/CSSPropertyAnimation.cpp:2196
> +        new PropertyWrapperAspectRatio(),

Other wrappers that don't take arguments are created without the ().

[Antoine Quint]

Actually, looking at the code again:

    double aspectRatioDst = WebCore::blend(log(a->logicalAspectRatio()), log(b->logicalAspectRatio()), progress);
    dst->setAspectRatio(exp(aspectRatioDst), 1);
    dst->setAspectRatioType(progress < 0.5 ? a->aspectRatioType() : b->aspectRatioType());

Since in the discrete case the progress is either 0 or 1, I think the blend function could be just those three lines. While it's a bit extra work to actually blend the values when we know it'll be one or the other, it simplifies the logic.

That assumes that canInterpolate() will correctly identify all cases where the animation should be discrete, which is its contract.

[Rob Buis]

Created attachment 423150 [details]
Patch

[Rob Buis]

Comment on attachment 423129 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=423129&action=review

>> Source/WebCore/animation/CSSPropertyAnimation.cpp:1851
>> +class PropertyWrapperAspectRatio : public AnimationPropertyWrapperBase {
> 
> I think this class can be marked "final"

Done.

>> Source/WebCore/animation/CSSPropertyAnimation.cpp:1869
>> +    bool canInterpolate(const RenderStyle* a, const RenderStyle* b) const override
> 
> I've made a pass through this entire file in r274409 to use from/to instead of a/b for canInterpolate(). This can also be marked "final".

Done.

>> Source/WebCore/animation/CSSPropertyAnimation.cpp:1875
>> +    void logBlend(const RenderStyle* a, const RenderStyle* b, const RenderStyle* result, double progress) const final
> 
> I've made a pass through this entire file in r274409 to use from/to instead of a/b for logBlend().

Done.

>> Source/WebCore/animation/CSSPropertyAnimation.cpp:1881
>> +    void blend(const CSSPropertyBlendingClient*, RenderStyle* dst, const RenderStyle* a, const RenderStyle* b, double progress) const override
> 
> I've made a pass through this entire file in r274409 to use from/to instead of a/b for blend(), as well as "destination" instead of "dst". This can also be marked "final".

Done.

>> Source/WebCore/animation/CSSPropertyAnimation.cpp:1884
>> +            || a->aspectRatioType() == AspectRatioType::Auto || b->aspectRatioType() == AspectRatioType::Auto) {
> 
> The code conditioned by this clause looks like it only applies to discrete animation, so should the clause use similar logic as canInterpolate()? 
> 
> I would switch the code in this method to have a canInterpolate() block where you do the regular case and return. And then have the rest of the function assert that progress is 0 or 1, as should be the case for a discrete animation (see CSSPropertyAnimation::blendProperties) and perform the discrete animation code.

Done.

>> Source/WebCore/animation/CSSPropertyAnimation.cpp:1888
>> +                dst->setAspectRatio(b->aspectRatioWidth(), b->aspectRatioHeight());
> 
> Maybe just:
> 
> auto* applicableStyle = progress < 0.5 ? from : to;
> destination->setAspectRatio(applicableStyle->aspectRatioWidth(), applicableStyle->aspectRatioHeight());

Done.

>> Source/WebCore/animation/CSSPropertyAnimation.cpp:1892
>> +        double aspectRatioDst = WebCore::blend(log(a->logicalAspectRatio()), log(b->logicalAspectRatio()), progress);
> 
> "auto" instead of "double".

Done.

>> Source/WebCore/animation/CSSPropertyAnimation.cpp:1894
>> +        dst->setAspectRatioType(progress < 0.5 ? a->aspectRatioType() : b->aspectRatioType());
> 
> This line is duplicated in the if clause above and here. You should run it first before the code branches.

Good point, done.

>> Source/WebCore/animation/CSSPropertyAnimation.cpp:2196
>> +        new PropertyWrapperAspectRatio(),
> 
> Other wrappers that don't take arguments are created without the ().

Sure, done.

[Rob Buis]

Created attachment 423151 [details]
Patch

[Rob Buis]

(In reply to Antoine Quint from comment #10)
> Actually, looking at the code again:
> 
>     double aspectRatioDst = WebCore::blend(log(a->logicalAspectRatio()),
> log(b->logicalAspectRatio()), progress);
>     dst->setAspectRatio(exp(aspectRatioDst), 1);
>     dst->setAspectRatioType(progress < 0.5 ? a->aspectRatioType() :
> b->aspectRatioType());
> 
> Since in the discrete case the progress is either 0 or 1, I think the blend
> function could be just those three lines. While it's a bit extra work to
> actually blend the values when we know it'll be one or the other, it
> simplifies the logic.
> 
> That assumes that canInterpolate() will correctly identify all cases where
> the animation should be discrete, which is its contract.

The problem is that logicalAspectRatio will assert for certain aspect ratio types (auto and autoZero). I had the test fail because of that in Debug earlier.

Do you know what causes the FAILs in the last test of aspect-ratio-interpolation.html?
I am assuming some functionality is missing, since other tests in /css/css-sizing/animation seem to have similar failures. If so I wonder if we should link to the relevant bug and keep this bug open?

[Antoine Quint]

(In reply to Rob Buis from comment #14)
> (In reply to Antoine Quint from comment #10)
> > Actually, looking at the code again:
> > 
> >     double aspectRatioDst = WebCore::blend(log(a->logicalAspectRatio()),
> > log(b->logicalAspectRatio()), progress);
> >     dst->setAspectRatio(exp(aspectRatioDst), 1);
> >     dst->setAspectRatioType(progress < 0.5 ? a->aspectRatioType() :
> > b->aspectRatioType());
> > 
> > Since in the discrete case the progress is either 0 or 1, I think the blend
> > function could be just those three lines. While it's a bit extra work to
> > actually blend the values when we know it'll be one or the other, it
> > simplifies the logic.
> > 
> > That assumes that canInterpolate() will correctly identify all cases where
> > the animation should be discrete, which is its contract.
> 
> The problem is that logicalAspectRatio will assert for certain aspect ratio
> types (auto and autoZero). I had the test fail because of that in Debug
> earlier.

Thanks for the explanation. That seems worthy of a comment in the commit.

> Do you know what causes the FAILs in the last test of
> aspect-ratio-interpolation.html?
> I am assuming some functionality is missing, since other tests in
> /css/css-sizing/animation seem to have similar failures. If so I wonder if
> we should link to the relevant bug and keep this bug open?

Yes, this is due to us not supporting composite operations yet, see bug 189299.

[Antoine Quint]

Comment on attachment 423151 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=423151&action=review

> Source/WebCore/animation/CSSPropertyAnimation.cpp:1897
> +        auto* applicableStyle = progress < 0.5 ? from : to;

This can be simply "progress ? to : from" since we've already asserted that progress is 0 or 1.

[Rob Buis]

Created attachment 423155 [details]
Patch

[EWS]

Committed r274415: <https://commits.webkit.org/r274415>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 423155 [details].

[Rob Buis]

(In reply to Antoine Quint from comment #15)
> (In reply to Rob Buis from comment #14)
> > (In reply to Antoine Quint from comment #10)
> > > Actually, looking at the code again:
> > > 
> > >     double aspectRatioDst = WebCore::blend(log(a->logicalAspectRatio()),
> > > log(b->logicalAspectRatio()), progress);
> > >     dst->setAspectRatio(exp(aspectRatioDst), 1);
> > >     dst->setAspectRatioType(progress < 0.5 ? a->aspectRatioType() :
> > > b->aspectRatioType());
> > > 
> > > Since in the discrete case the progress is either 0 or 1, I think the blend
> > > function could be just those three lines. While it's a bit extra work to
> > > actually blend the values when we know it'll be one or the other, it
> > > simplifies the logic.
> > > 
> > > That assumes that canInterpolate() will correctly identify all cases where
> > > the animation should be discrete, which is its contract.
> > 
> > The problem is that logicalAspectRatio will assert for certain aspect ratio
> > types (auto and autoZero). I had the test fail because of that in Debug
> > earlier.
> 
> Thanks for the explanation. That seems worthy of a comment in the commit.
> 
> > Do you know what causes the FAILs in the last test of
> > aspect-ratio-interpolation.html?
> > I am assuming some functionality is missing, since other tests in
> > /css/css-sizing/animation seem to have similar failures. If so I wonder if
> > we should link to the relevant bug and keep this bug open?
> 
> Yes, this is due to us not supporting composite operations yet, see bug
> 189299.

[Rob Buis]

(In reply to Antoine Quint from comment #15)
> > Do you know what causes the FAILs in the last test of
> > aspect-ratio-interpolation.html?
> > I am assuming some functionality is missing, since other tests in
> > /css/css-sizing/animation seem to have similar failures. If so I wonder if
> > we should link to the relevant bug and keep this bug open?
> 
> Yes, this is due to us not supporting composite operations yet, see bug
> 189299.

Thanks, I made this bug depend on 189299 and unassigned myself to get it out of my work queue for now.

[Rob Buis]

Remaining test failures were fixed by r243954.