Replace filter outsets with bounds mapping in Blink.

Replace filter outsets with bounds mapping in Blink.

This mirrors SkImageFilter::filterBounds, and is more flexible.

Also included are some additional unit tests, and changes to layout test
baselines:
* css3/filters/effect-reference-repaint-offset.html
  Progression: now only repaints the offset area.
* fast/repaint/*
  Minor: rounding for inexact values is more conservative (blur radius).

BUG=600821
Committed: https://crrev.com/42b8b79d8538d508b2d660e3dab12e8d0839bfaf
Cr-Commit-Position: refs/heads/master@{#386527}

[jbroman, 2016-04-05 21:13:45]

Description was changed from

==========
WIP: Replace filter outsets with bounds mapping in Blink.

BUG=
==========

to

==========
WIP: Replace filter outsets with bounds mapping in Blink.

BUG=600821
==========

[jbroman, 2016-04-05 21:13:57]

Description was changed from

==========
WIP: Replace filter outsets with bounds mapping in Blink.

BUG=600821
==========

to

==========
Replace filter outsets with bounds mapping in Blink.

BUG=600821
==========

[jbroman, 2016-04-05 21:14:10]

Patchset #5 (id:80001) has been deleted

[jbroman, 2016-04-06 17:56:31]

Description was changed from

==========
Replace filter outsets with bounds mapping in Blink.

BUG=600821
==========

to

==========
Replace filter outsets with bounds mapping in Blink.

This mirrors SkImageFilter::filterBounds, and is more flexible.

Also included are some additional unit tests, and changes to layout test
baselines:
* css3/filters/effect-reference-repaint-offset.html
  Progression: now only repaints the offset area.
* fast/repaint/*
  Minor: rounding for inexact values is more conservative (blur radius).

BUG=600821
==========

[jbroman, 2016-04-06 17:56:55]

jbroman@chromium.org changed reviewers:
+ senorblanco@chromium.org

[Stephen White, 2016-04-06 20:44:11]

Thanks for taking this on!

https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
File third_party/WebKit/Source/core/paint/PaintLayer.h (right):

https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
third_party/WebKit/Source/core/paint/PaintLayer.h:477: FloatRect
mapRectForFilter(const FloatRect&, bool forward = true) const;
I couldn't find any callsites that use forward "false". If so, could you remove
the param? If you really think you'll need it, you can always add it later.

https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
third_party/WebKit/Source/core/paint/PaintLayer.h:480: LayoutRect
mapRectForFilter(const LayoutRect&, bool forward = true) const;
Naming nit: could we call this one mapLayoutRectForFilter(), just to be clear?

[jbroman, 2016-04-06 21:51:41]

https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
File third_party/WebKit/Source/core/paint/PaintLayer.h (right):

https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
third_party/WebKit/Source/core/paint/PaintLayer.h:477: FloatRect
mapRectForFilter(const FloatRect&, bool forward = true) const;
On 2016/04/06 at 20:44:11, Stephen White wrote:
> I couldn't find any callsites that use forward "false". If so, could you
remove the param? If you really think you'll need it, you can always add it
later.

Alright, done.

https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
third_party/WebKit/Source/core/paint/PaintLayer.h:480: LayoutRect
mapRectForFilter(const LayoutRect&, bool forward = true) const;
On 2016/04/06 at 20:44:11, Stephen White wrote:
> Naming nit: could we call this one mapLayoutRectForFilter(), just to be clear?

Done.

[jbroman, 2016-04-08 18:22:06]

On 2016/04/06 at 21:51:41, jbroman wrote:
>
https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
> File third_party/WebKit/Source/core/paint/PaintLayer.h (right):
> 
>
https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
> third_party/WebKit/Source/core/paint/PaintLayer.h:477: FloatRect
mapRectForFilter(const FloatRect&, bool forward = true) const;
> On 2016/04/06 at 20:44:11, Stephen White wrote:
> > I couldn't find any callsites that use forward "false". If so, could you
remove the param? If you really think you'll need it, you can always add it
later.
> 
> Alright, done.
> 
>
https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
> third_party/WebKit/Source/core/paint/PaintLayer.h:480: LayoutRect
mapRectForFilter(const LayoutRect&, bool forward = true) const;
> On 2016/04/06 at 20:44:11, Stephen White wrote:
> > Naming nit: could we call this one mapLayoutRectForFilter(), just to be
clear?
> 
> Done.

ping

[Stephen White, 2016-04-08 22:15:57]

On 2016/04/08 18:22:06, jbroman wrote:
> On 2016/04/06 at 21:51:41, jbroman wrote:
> >
>
https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
> > File third_party/WebKit/Source/core/paint/PaintLayer.h (right):
> > 
> >
>
https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
> > third_party/WebKit/Source/core/paint/PaintLayer.h:477: FloatRect
> mapRectForFilter(const FloatRect&, bool forward = true) const;
> > On 2016/04/06 at 20:44:11, Stephen White wrote:
> > > I couldn't find any callsites that use forward "false". If so, could you
> remove the param? If you really think you'll need it, you can always add it
> later.
> > 
> > Alright, done.
> > 
> >
>
https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
> > third_party/WebKit/Source/core/paint/PaintLayer.h:480: LayoutRect
> mapRectForFilter(const LayoutRect&, bool forward = true) const;
> > On 2016/04/06 at 20:44:11, Stephen White wrote:
> > > Naming nit: could we call this one mapLayoutRectForFilter(), just to be
> clear?
> > 
> > Done.
> 
> ping

You've got me worried that our conservative outsets implementation was masking
some bugs wrt culling.

Could you try to conjure a test case that exercises this? Something like an
feOffset that would push objects that would be culled into visible range might
work.

[jbroman, 2016-04-11 17:34:36]

On 2016/04/08 at 22:15:57, senorblanco wrote:
> On 2016/04/08 18:22:06, jbroman wrote:
> > On 2016/04/06 at 21:51:41, jbroman wrote:
> > >
> >
https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
> > > File third_party/WebKit/Source/core/paint/PaintLayer.h (right):
> > > 
> > >
> >
https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
> > > third_party/WebKit/Source/core/paint/PaintLayer.h:477: FloatRect
> > mapRectForFilter(const FloatRect&, bool forward = true) const;
> > > On 2016/04/06 at 20:44:11, Stephen White wrote:
> > > > I couldn't find any callsites that use forward "false". If so, could you
> > remove the param? If you really think you'll need it, you can always add it
> > later.
> > > 
> > > Alright, done.
> > > 
> > >
> >
https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
> > > third_party/WebKit/Source/core/paint/PaintLayer.h:480: LayoutRect
> > mapRectForFilter(const LayoutRect&, bool forward = true) const;
> > > On 2016/04/06 at 20:44:11, Stephen White wrote:
> > > > Naming nit: could we call this one mapLayoutRectForFilter(), just to be
> > clear?
> > > 
> > > Done.
> > 
> > ping
> 
> You've got me worried that our conservative outsets implementation was masking
some bugs wrt culling.
> 
> Could you try to conjure a test case that exercises this? Something like an
feOffset that would push objects that would be culled into visible range might
work.

Filed crbug.com/602341. This behaviour is the same both with and without this
CL; I think we don't do this mapping, but just get saved by the prepaint margin
in real world scenarios.

[Stephen White, 2016-04-11 18:13:16]

On 2016/04/11 17:34:36, jbroman wrote:
> On 2016/04/08 at 22:15:57, senorblanco wrote:
> > On 2016/04/08 18:22:06, jbroman wrote:
> > > On 2016/04/06 at 21:51:41, jbroman wrote:
> > > >
> > >
>
https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
> > > > File third_party/WebKit/Source/core/paint/PaintLayer.h (right):
> > > > 
> > > >
> > >
>
https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
> > > > third_party/WebKit/Source/core/paint/PaintLayer.h:477: FloatRect
> > > mapRectForFilter(const FloatRect&, bool forward = true) const;
> > > > On 2016/04/06 at 20:44:11, Stephen White wrote:
> > > > > I couldn't find any callsites that use forward "false". If so, could
you
> > > remove the param? If you really think you'll need it, you can always add
it
> > > later.
> > > > 
> > > > Alright, done.
> > > > 
> > > >
> > >
>
https://codereview.chromium.org/1855303002/diff/160001/third_party/WebKit/Sou...
> > > > third_party/WebKit/Source/core/paint/PaintLayer.h:480: LayoutRect
> > > mapRectForFilter(const LayoutRect&, bool forward = true) const;
> > > > On 2016/04/06 at 20:44:11, Stephen White wrote:
> > > > > Naming nit: could we call this one mapLayoutRectForFilter(), just to
be
> > > clear?
> > > > 
> > > > Done.
> > > 
> > > ping
> > 
> > You've got me worried that our conservative outsets implementation was
masking
> some bugs wrt culling.
> > 
> > Could you try to conjure a test case that exercises this? Something like an
> feOffset that would push objects that would be culled into visible range might
> work.
> 
> Filed crbug.com/602341. This behaviour is the same both with and without this
> CL; I think we don't do this mapping, but just get saved by the prepaint
margin
> in real world scenarios.

OK, in that case, let's tackle it separately. Thanks for the test case.

LGTM

[jbroman, 2016-04-11 20:22:30]

The CQ bit was checked by jbroman@chromium.org

[commit-bot: I haz the power, 2016-04-11 20:23:03]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1855303002/180001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1855303002/180001

[commit-bot: I haz the power, 2016-04-12 00:11:34]

Committed patchset #9 (id:180001)

[commit-bot: I haz the power, 2016-04-12 00:12:37]

Description was changed from

==========
Replace filter outsets with bounds mapping in Blink.

This mirrors SkImageFilter::filterBounds, and is more flexible.

Also included are some additional unit tests, and changes to layout test
baselines:
* css3/filters/effect-reference-repaint-offset.html
  Progression: now only repaints the offset area.
* fast/repaint/*
  Minor: rounding for inexact values is more conservative (blur radius).

BUG=600821
==========

to

==========
Replace filter outsets with bounds mapping in Blink.

This mirrors SkImageFilter::filterBounds, and is more flexible.

Also included are some additional unit tests, and changes to layout test
baselines:
* css3/filters/effect-reference-repaint-offset.html
  Progression: now only repaints the offset area.
* fast/repaint/*
  Minor: rounding for inexact values is more conservative (blur radius).

BUG=600821

Committed: https://crrev.com/42b8b79d8538d508b2d660e3dab12e8d0839bfaf
Cr-Commit-Position: refs/heads/master@{#386527}
==========

[commit-bot: I haz the power, 2016-04-12 00:12:38]

Patchset 9 (id:??) landed as
https://crrev.com/42b8b79d8538d508b2d660e3dab12e8d0839bfaf
Cr-Commit-Position: refs/heads/master@{#386527}