Fix resize issues when using threaded compositing

Fix resize issues when using threaded compositing

Threaded compositing doesn't ever call DoDeferredUpdate, so never has a chance
to send the resize ack via an update message back to the browser. Fix by piping
through didBeginFrame (called at the beginning of a main thread frame) and
didFinishFrame (called during the client side OnSwapBuffers when a main thread
frame has been committed). This patch queues up update messages during
didBeginFrame and then sends them out during the corresponding didFinishFrame.

BUG=103776
TEST=run with threaded compositing, resize page multiple times

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=110599

[enne (OOO), 2011-11-10 20:29:28]

This patch depends on piman's http://codereview.chromium.org/8498036 change by
reusing the pending update queue.

The WebKit side of the patch is here:
https://bugs.webkit.org/show_bug.cgi?id=72041

[jamesr, 2011-11-10 22:55:05]

lgtm

[nduca, 2011-11-11 06:01:08]

http://codereview.chromium.org/8528006/diff/1/content/renderer/render_widget.cc
File content/renderer/render_widget.cc (left):

http://codereview.chromium.org/8528006/diff/1/content/renderer/render_widget....
content/renderer/render_widget.cc:406: void RenderWidget::OnSwapBuffersPosted()
{
Can we get rid of this completely? E.g. never use the swapbuffers posted
interface.

I'd really like to remove the web_view_ pointer on
WebGraphcsContext3DCommandBufferImpl, which is the route via which these are
traditionally called.

http://codereview.chromium.org/8528006/diff/1/content/renderer/render_widget.cc
File content/renderer/render_widget.cc (right):

http://codereview.chromium.org/8528006/diff/1/content/renderer/render_widget....
content/renderer/render_widget.cc:1036: void RenderWidget::didFinishFrame()
One thing that nags me here is that Antoine's patch sends udpates on the swap
callback, which is the GPU process finishing the frame.

This does so when the compositor thread has submitted the frame to the gpu
process.

Which are we trying to do? Both? One? The other? I'm a bit confused.

[nduca, 2011-11-12 03:22:06]

http://codereview.chromium.org/8528006/diff/6001/content/renderer/render_widg...
File content/renderer/render_widget.cc (right):

http://codereview.chromium.org/8528006/diff/6001/content/renderer/render_widg...
content/renderer/render_widget.cc:975: void
RenderWidget::didCompleteSwapbuffers()
Should this be conditional on compositor thread mode? Or do we not send this
when the CCSingleThreadProxy is active? We need to eventually call this all the
time, is there any #ifdef we should put in here to anticipate that behavior
change and avoid having to do a triple-commit?

[nduca, 2011-11-12 03:22:13]

Oh, also, LGTM

[enne (OOO), 2011-11-12 19:29:19]

http://codereview.chromium.org/8528006/diff/6001/content/renderer/render_widg...
File content/renderer/render_widget.cc (right):

http://codereview.chromium.org/8528006/diff/6001/content/renderer/render_widg...
content/renderer/render_widget.cc:975: void
RenderWidget::didCompleteSwapbuffers()
On 2011/11/12 03:22:06, nduca wrote:
> Should this be conditional on compositor thread mode? Or do we not send this
> when the CCSingleThreadProxy is active? We need to eventually call this all
the
> time, is there any #ifdef we should put in here to anticipate that behavior
> change and avoid having to do a triple-commit?

This function is robust to being called multiple times.  So, you should be able
to remove the render view pointer and things should Just Work (TM).

[jam, 2011-11-17 22:56:12]

lgtm

http://codereview.chromium.org/8528006/diff/9001/content/renderer/render_widg...
File content/renderer/render_widget.cc (right):

http://codereview.chromium.org/8528006/diff/9001/content/renderer/render_widg...
content/renderer/render_widget.cc:971: void
RenderWidget::didCommitAndDrawCompositorFrame()
nit: here and below, brace bracket on the previous line per the chrome style
guide

[jbates, 2012-04-16 23:41:35]

http://codereview.chromium.org/8528006/diff/9001/content/renderer/render_widg...
File content/renderer/render_widget.cc (right):

http://codereview.chromium.org/8528006/diff/9001/content/renderer/render_widg...
content/renderer/render_widget.cc:980: if (!next_paint_flags_ &&
!plugin_window_moves_.size())
This will blow up if someone adds new flags to ViewHostMsg_UpdateRect_Flags.