Support mask images for filter-based -webkit-box-reflect.

Support mask images for filter-based -webkit-box-reflect.

This is implemented by having a SkPictureImageFilter which draws the provided
mask image (e.g. a gradient), which is then applied to the reflection.

A bunch of math is needed to make sure the mask is correctly sized and positioned,
because the property is actually fairly flexible and allows a nine-piece image
with insets for slicing and outsets for expanding.

BUG=436475
Committed: https://crrev.com/98e094a1f352965d9ea46601f672b3555b56f330
Cr-Commit-Position: refs/heads/master@{#389151}

[jbroman, 2016-04-19 17:22:30]

Description was changed from

==========
WIP: Mask images for reflections!
==========

to

==========
Support mask images for filter-based -webkit-box-reflect.

This is implemented by having a SkPictureImageFilter which draws the provided
mask image (e.g. a gradient), which is then applied to the reflection.

A bunch of math is needed to make sure the mask is correctly sized and
positioned,
because the property is actually fairly flexible and allows a nine-piece image
with insets for slicing and outsets for expanding.
==========

[jbroman, 2016-04-19 17:23:34]

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

[jbroman, 2016-04-19 17:23:35]

The current patch set shows the layout test failures that would exist if this
feature were turned on today. Mostly it's just minor subpixel differences due
to it being rasterized slightly differently. There are some issues that look
like compositor-side bounds issues.

I intend to remove the RuntimeEnabledFeatures.in change before landing; it's
there only so that you can look at the current status of layout tests.

[jbroman, 2016-04-19 17:23:48]

Description was changed from

==========
Support mask images for filter-based -webkit-box-reflect.

This is implemented by having a SkPictureImageFilter which draws the provided
mask image (e.g. a gradient), which is then applied to the reflection.

A bunch of math is needed to make sure the mask is correctly sized and
positioned,
because the property is actually fairly flexible and allows a nine-piece image
with insets for slicing and outsets for expanding.
==========

to

==========
Support mask images for filter-based -webkit-box-reflect.

This is implemented by having a SkPictureImageFilter which draws the provided
mask image (e.g. a gradient), which is then applied to the reflection.

A bunch of math is needed to make sure the mask is correctly sized and
positioned,
because the property is actually fairly flexible and allows a nine-piece image
with insets for slicing and outsets for expanding.

BUG=436475
==========

[Stephen White, 2016-04-20 21:33:57]

https://codereview.chromium.org/1879223002/diff/140001/third_party/WebKit/Sou...
File third_party/WebKit/Source/core/paint/BoxReflectionUtils.cpp (right):

https://codereview.chromium.org/1879223002/diff/140001/third_party/WebKit/Sou...
third_party/WebKit/Source/core/paint/BoxReflectionUtils.cpp:22: const auto*
reflectStyle = style.boxReflect();
This might be personal style, but I find "auto" here hurts readability more than
it helps conciseness. What do you think about putting the actual type here?

https://codereview.chromium.org/1879223002/diff/140001/third_party/WebKit/Sou...
third_party/WebKit/Source/core/paint/BoxReflectionUtils.cpp:56: const auto&
maskNinePiece = reflectStyle->mask();
Same here.

https://codereview.chromium.org/1879223002/diff/140001/third_party/WebKit/Sou...
File
third_party/WebKit/Source/platform/graphics/filters/SkiaImageFilterBuilder.cpp
(right):

https://codereview.chromium.org/1879223002/diff/140001/third_party/WebKit/Sou...
third_party/WebKit/Source/platform/graphics/filters/SkiaImageFilterBuilder.cpp:192:
SkImageFilter::CropRect cropRect(maskPicture->cullRect());
Just wondering why a crop rect is necessary here. The SPIF's output should be
the picture's cullRect() in size, and SXIF should produce an output of the same
size. Is that not the case?

[jbroman, 2016-04-21 16:52:05]

https://codereview.chromium.org/1879223002/diff/140001/third_party/WebKit/Sou...
File third_party/WebKit/Source/core/paint/BoxReflectionUtils.cpp (right):

https://codereview.chromium.org/1879223002/diff/140001/third_party/WebKit/Sou...
third_party/WebKit/Source/core/paint/BoxReflectionUtils.cpp:22: const auto*
reflectStyle = style.boxReflect();
On 2016/04/20 at 21:33:56, Stephen White wrote:
> This might be personal style, but I find "auto" here hurts readability more
than it helps conciseness. What do you think about putting the actual type here?

Alright. The type is "StyleReflection", so the line becomes the amusingly
repetitive "const StyleReflection* reflectStyle = style.boxReflect()";

https://codereview.chromium.org/1879223002/diff/140001/third_party/WebKit/Sou...
third_party/WebKit/Source/core/paint/BoxReflectionUtils.cpp:56: const auto&
maskNinePiece = reflectStyle->mask();
On 2016/04/20 at 21:33:56, Stephen White wrote:
> Same here.

Done.

https://codereview.chromium.org/1879223002/diff/140001/third_party/WebKit/Sou...
File
third_party/WebKit/Source/platform/graphics/filters/SkiaImageFilterBuilder.cpp
(right):

https://codereview.chromium.org/1879223002/diff/140001/third_party/WebKit/Sou...
third_party/WebKit/Source/platform/graphics/filters/SkiaImageFilterBuilder.cpp:192:
SkImageFilter::CropRect cropRect(maskPicture->cullRect());
On 2016/04/20 at 21:33:56, Stephen White wrote:
> Just wondering why a crop rect is necessary here. The SPIF's output should be
the picture's cullRect() in size, and SXIF should produce an output of the same
size. Is that not the case?

My reading of SXIF is that it unites the source and destination rect when
choosing the default output size. The input image can, due to visual overflow,
be larger than the mask image, but since we definitely won't draw anything
outside the mask, it's unnecessary. SXIF could conceivably come to this
conclusion on its own by checking the particular xfer mode being used, but it
doesn't seem to do so.

[Stephen White, 2016-04-21 16:59:29]

LGTM w/skia bug reference added

https://codereview.chromium.org/1879223002/diff/140001/third_party/WebKit/Sou...
File
third_party/WebKit/Source/platform/graphics/filters/SkiaImageFilterBuilder.cpp
(right):

https://codereview.chromium.org/1879223002/diff/140001/third_party/WebKit/Sou...
third_party/WebKit/Source/platform/graphics/filters/SkiaImageFilterBuilder.cpp:192:
SkImageFilter::CropRect cropRect(maskPicture->cullRect());
On 2016/04/21 16:52:05, jbroman wrote:
> On 2016/04/20 at 21:33:56, Stephen White wrote:
> > Just wondering why a crop rect is necessary here. The SPIF's output should
be
> the picture's cullRect() in size, and SXIF should produce an output of the
same
> size. Is that not the case?
> 
> My reading of SXIF is that it unites the source and destination rect when
> choosing the default output size. The input image can, due to visual overflow,
> be larger than the mask image, but since we definitely won't draw anything
> outside the mask, it's unnecessary. SXIF could conceivably come to this
> conclusion on its own by checking the particular xfer mode being used, but it
> doesn't seem to do so.

Oh right, that makes sense. Would you log a skia bug to add that optimization,
and reference it here?

[jbroman, 2016-04-22 15:23:51]

On 2016/04/21 at 16:59:29, senorblanco wrote:
> LGTM w/skia bug reference added
> 
>
https://codereview.chromium.org/1879223002/diff/140001/third_party/WebKit/Sou...
> File
third_party/WebKit/Source/platform/graphics/filters/SkiaImageFilterBuilder.cpp
(right):
> 
>
https://codereview.chromium.org/1879223002/diff/140001/third_party/WebKit/Sou...
>
third_party/WebKit/Source/platform/graphics/filters/SkiaImageFilterBuilder.cpp:192:
SkImageFilter::CropRect cropRect(maskPicture->cullRect());
> On 2016/04/21 16:52:05, jbroman wrote:
> > On 2016/04/20 at 21:33:56, Stephen White wrote:
> > > Just wondering why a crop rect is necessary here. The SPIF's output should
be
> > the picture's cullRect() in size, and SXIF should produce an output of the
same
> > size. Is that not the case?
> > 
> > My reading of SXIF is that it unites the source and destination rect when
> > choosing the default output size. The input image can, due to visual
overflow,
> > be larger than the mask image, but since we definitely won't draw anything
> > outside the mask, it's unnecessary. SXIF could conceivably come to this
> > conclusion on its own by checking the particular xfer mode being used, but
it
> > doesn't seem to do so.
> 
> Oh right, that makes sense. Would you log a skia bug to add that optimization,
and reference it here?

Filed skbug.com/5210.

FYI, I'll land this CL, but this effort will probably pause for awhile while we
figure out how to respond to the SkMatrixImageFilter removal and associated Skia
concerns.

[jbroman, 2016-04-22 15:25:37]

The CQ bit was checked by jbroman@chromium.org

[jbroman, 2016-04-22 15:25:38]

The patchset sent to the CQ was uploaded after l-g-t-m from
senorblanco@chromium.org
Link to the patchset: https://codereview.chromium.org/1879223002/#ps180001
(title: " ")

[jbroman, 2016-04-22 15:25:47]

The CQ bit was unchecked by jbroman@chromium.org

[jbroman, 2016-04-22 15:27:15]

The CQ bit was checked by jbroman@chromium.org

[jbroman, 2016-04-22 15:27:16]

The patchset sent to the CQ was uploaded after l-g-t-m from
senorblanco@chromium.org
Link to the patchset: https://codereview.chromium.org/1879223002/#ps200001
(title: "add comment referring to skbug.com/5210")

[commit-bot: I haz the power, 2016-04-22 15:27:30]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1879223002/200001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1879223002/200001

[commit-bot: I haz the power, 2016-04-22 17:23:55]

Committed patchset #11 (id:200001)

[commit-bot: I haz the power, 2016-04-22 19:49:14]

Description was changed from

==========
Support mask images for filter-based -webkit-box-reflect.

This is implemented by having a SkPictureImageFilter which draws the provided
mask image (e.g. a gradient), which is then applied to the reflection.

A bunch of math is needed to make sure the mask is correctly sized and
positioned,
because the property is actually fairly flexible and allows a nine-piece image
with insets for slicing and outsets for expanding.

BUG=436475
==========

to

==========
Support mask images for filter-based -webkit-box-reflect.

This is implemented by having a SkPictureImageFilter which draws the provided
mask image (e.g. a gradient), which is then applied to the reflection.

A bunch of math is needed to make sure the mask is correctly sized and
positioned,
because the property is actually fairly flexible and allows a nine-piece image
with insets for slicing and outsets for expanding.

BUG=436475

Committed: https://crrev.com/98e094a1f352965d9ea46601f672b3555b56f330
Cr-Commit-Position: refs/heads/master@{#389151}
==========

[commit-bot: I haz the power, 2016-04-22 19:49:15]

Patchset 11 (id:??) landed as
https://crrev.com/98e094a1f352965d9ea46601f672b3555b56f330
Cr-Commit-Position: refs/heads/master@{#389151}