[WIP] Adding support for <iframe>s to slimming paint.

[WIP] Adding support for <iframe>s to slimming paint.

This CL does the following to add support for <iframe>s to slimming paint:

1. The per-RenderView ViewDisplayList has been moved from RenderView to
   the local root FrameView, so a sub-frame's paint chunks will be appended
   to appropriate places in the display list.

2. Added TranslationDisplayItem. This display item allows nesting child
   elements inside a translated context.

3. RenderPart painting code moved to paint/PartPainter. Proper chunk recorder
   added.

4. Partially implement paint chunks for FrameView barely to pass fast/block
   layout tests.

[trchen, 2014-11-01 00:27:12]

trchen@chromium.org changed reviewers:
+ chrishtr@chromium.org, leviw@chromium.org, schenney@google.com

[trchen, 2014-11-01 00:27:12]

Hello, this is a WIP for adding support for <iframe>s to slimming paint. A few
implementation details haven't been decided yet. (There are a few display items
don't belong to specific render object / phase. Needs to assign proper
identifier for them.) Please take a look and let me know your opinion. Thanks!

[leviw_travelin_and_unemployed, 2014-11-01 00:29:01]

I feel like it would make this easier to review if the PartPainter refactor was
in another patch.

[chrishtr, 2014-11-01 23:44:11]

https://codereview.chromium.org/698743002/diff/1/Source/core/frame/FrameView.h
File Source/core/frame/FrameView.h (right):

https://codereview.chromium.org/698743002/diff/1/Source/core/frame/FrameView....
Source/core/frame/FrameView.h:521: ViewDisplayList& viewDisplayList()
Don't inline this.

https://codereview.chromium.org/698743002/diff/1/Source/core/paint/DrawingRec...
File Source/core/paint/DrawingRecorder.cpp (right):

https://codereview.chromium.org/698743002/diff/1/Source/core/paint/DrawingRec...
Source/core/paint/DrawingRecorder.cpp:57:
ViewDisplayList::fromRenderObject(m_renderer).add(drawingItem.release());
I hadn't mentioned it yet, but since your CL forces the issue, we should put the
DisplayItemList (and rename it to that instead of ViewDisplayList) on
paintInvalidationContainer->layer()->graphicsLayerBacking(). If we do that then
the rest we get for free, e.g. traversing across non-composited vs composited
iframes, or eventually turning off compositing below the root document entirely.


GraphicsLayer will also be the right entry point for sending the DisplayList to
cc as we implement the rest of Slimming Paint, and allow us to easily get to the
phase 1 launch point of painting existing cc::Layers w/Slimming Paint.

https://codereview.chromium.org/698743002/diff/1/Source/core/paint/PartPainter.h
File Source/core/paint/PartPainter.h (right):

https://codereview.chromium.org/698743002/diff/1/Source/core/paint/PartPainte...
Source/core/paint/PartPainter.h:14: class PartPainter {
Like Levi said, please split off the PartPainter refactor into a separate CL and
commit it first.

https://codereview.chromium.org/698743002/diff/1/Source/core/paint/Translatio...
File Source/core/paint/TranslationRecorder.cpp (right):

https://codereview.chromium.org/698743002/diff/1/Source/core/paint/Translatio...
Source/core/paint/TranslationRecorder.cpp:13: void
TranslationDisplayItem::replay(GraphicsContext* context)
Scale, rotate and translate are just convenience wrappers around setting an
affine transform (which in turn maps to SkMatrix).
Let's just have an affine transform display item, and put the convenience
wrapper API methods on the AffineTransformRecorder.

[trchen, 2014-11-02 10:36:46]

https://codereview.chromium.org/698743002/diff/1/Source/core/frame/FrameView.h
File Source/core/frame/FrameView.h (right):

https://codereview.chromium.org/698743002/diff/1/Source/core/frame/FrameView....
Source/core/frame/FrameView.h:521: ViewDisplayList& viewDisplayList()
On 2014/11/01 23:44:10, chrishtr wrote:
> Don't inline this.

Acknowledged.

https://codereview.chromium.org/698743002/diff/1/Source/core/paint/DrawingRec...
File Source/core/paint/DrawingRecorder.cpp (right):

https://codereview.chromium.org/698743002/diff/1/Source/core/paint/DrawingRec...
Source/core/paint/DrawingRecorder.cpp:57:
ViewDisplayList::fromRenderObject(m_renderer).add(drawingItem.release());
On 2014/11/01 23:44:10, chrishtr wrote:
> I hadn't mentioned it yet, but since your CL forces the issue, we should put
the
> DisplayItemList (and rename it to that instead of ViewDisplayList) on
> paintInvalidationContainer->layer()->graphicsLayerBacking(). If we do that
then
> the rest we get for free, e.g. traversing across non-composited vs composited
> iframes, or eventually turning off compositing below the root document
entirely.
> 
> 
> GraphicsLayer will also be the right entry point for sending the DisplayList
to
> cc as we implement the rest of Slimming Paint, and allow us to easily get to
the
> phase 1 launch point of painting existing cc::Layers w/Slimming Paint.

Ah ha, that makes sense. I will revise as you planned.

https://codereview.chromium.org/698743002/diff/1/Source/core/paint/PartPainter.h
File Source/core/paint/PartPainter.h (right):

https://codereview.chromium.org/698743002/diff/1/Source/core/paint/PartPainte...
Source/core/paint/PartPainter.h:14: class PartPainter {
On 2014/11/01 23:44:10, chrishtr wrote:
> Like Levi said, please split off the PartPainter refactor into a separate CL
and
> commit it first.

Acknowledged.

https://codereview.chromium.org/698743002/diff/1/Source/core/paint/Translatio...
File Source/core/paint/TranslationRecorder.cpp (right):

https://codereview.chromium.org/698743002/diff/1/Source/core/paint/Translatio...
Source/core/paint/TranslationRecorder.cpp:13: void
TranslationDisplayItem::replay(GraphicsContext* context)
On 2014/11/01 23:44:11, chrishtr wrote:
> Scale, rotate and translate are just convenience wrappers around setting an
> affine transform (which in turn maps to SkMatrix).
> Let's just have an affine transform display item, and put the convenience
> wrapper API methods on the AffineTransformRecorder.

That makes sense. I was thinking that 2D translation is one important degenerate
case that worth optimizing. (We can implement it without having to save
context.) But I guess we can still use a TransformRecorder to do the recording,
and optimize if we detected the transform was a 2D translation.