If a GraphicsLayerChromium goes away during the middle of paint, its

If a GraphicsLayerChromium goes away during the middle of paint, its
CCLayerImpl will be kept alive in the render surface layer list, but its render
surface will be cleared.  updateCompositorResources and drawLayers is not
robust to this.

Rather than trying to modify the destruction order, just add safety checks.

[nduca, 2011-09-09 23:24:09]

http://codereview.chromium.org/7866017/diff/1/Source/WebCore/platform/graphic...
File Source/WebCore/platform/graphics/chromium/LayerRendererChromium.cpp
(right):

http://codereview.chromium.org/7866017/diff/1/Source/WebCore/platform/graphic...
Source/WebCore/platform/graphics/chromium/LayerRendererChromium.cpp:979: if
(!ccLayerImpl)
(drive by) How do we prevent this crap from happening in the future? It seems to
me that rendersurfaces are crazy fragile as implemented.

[jamesr, 2011-09-09 23:28:51]

On 2011/09/09 23:24:09, nduca wrote:
>
http://codereview.chromium.org/7866017/diff/1/Source/WebCore/platform/graphic...
> File Source/WebCore/platform/graphics/chromium/LayerRendererChromium.cpp
> (right):
> 
>
http://codereview.chromium.org/7866017/diff/1/Source/WebCore/platform/graphic...
> Source/WebCore/platform/graphics/chromium/LayerRendererChromium.cpp:979: if
> (!ccLayerImpl)
> (drive by) How do we prevent this crap from happening in the future? It seems
to
> me that rendersurfaces are crazy fragile as implemented.

This is 14, where we didn't have 2 types of RenderSurface so the LayerChromium
and CCLayerImpl trees both pointed to the same RS's

on 15+ the rendersurfaces are decoupled so this isn't so bad

[enne_, 2011-09-09 23:34:32]

On 2011/09/09 23:24:09, nduca wrote:
> (drive by) How do we prevent this crap from happening in the future? It seems
to
> me that rendersurfaces are crazy fragile as implemented.

Unit tests.

[nduca, 2011-09-09 23:37:23]

Hehe sounds good (and LGTM). Apologies in advance for having an opinion on
something I know a bit too little about...

This smells like more of the "RenderSurfaces" conflate the "curent frame's
drawing list" with "we want to have a pool of rendertargets so we don't allocate
them each frame."

Naievely, If we built the drawing list every frame, then just looked aside to a
cache for available rendertargets, wouldn't we be simplify things a bit?