Mac: Correctly show/hide compositor on navigations.

Mac: Correctly show/hide compositor on navigations.

The problem was that unlike plugins, the compositor surface is not destroyed on navigation, only a variable is_gpu_rendering_active is updated. The fix is to tell the RenderWidgetHostView if this variable changed and then to adapt the visibility of the view rendering the compositing layer accordingly.

Also, destroy its AcceleratedSurfaceContainer when the compositor is destroyed.

BUG=52134, 51748
TEST=go to http://webkit.org/blog/386/3d-transforms/, then hit back. page should redraw on back button. go forward, should work too.

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=57135

[Nico, 2010-08-24 01:57:28]

This CL is much smaller than the file list suggests.

kbr: Please review.
Pink, Stuart: FYI (review the hole punching CL instead :-) )

[Ken Russell, 2010-08-24 02:44:51]

LGTM. Thanks for cleaning up the plugin window handle management. One comment.

http://codereview.chromium.org/3132038/diff/4001/5003
File chrome/browser/renderer_host/render_widget_host.cc (right):

http://codereview.chromium.org/3132038/diff/4001/5003#newcode929
chrome/browser/renderer_host/render_widget_host.cc:929:
view_->GpuRenderingStateDidChange();
Should probably do a NULL check on view_ as is done in some of the related
methods.

[Nico, 2010-08-24 02:46:22]

thanks!

http://codereview.chromium.org/3132038/diff/4001/5003
File chrome/browser/renderer_host/render_widget_host.cc (right):

http://codereview.chromium.org/3132038/diff/4001/5003#newcode929
chrome/browser/renderer_host/render_widget_host.cc:929:
view_->GpuRenderingStateDidChange();
On 2010/08/24 02:44:51, kbr wrote:
> Should probably do a NULL check on view_ as is done in some of the related
> methods.
> 

Done.