cc: Get BeginFrameArgs for BeginMainFrame while inside an impl frame.

cc: Get BeginFrameArgs for BeginMainFrame while inside an impl frame.

This patch also adds asserts to make sure that the assumption that
ScheduledActionSendBeginMainFrame is only called inside an impl frame
holds.

Sending a BeginMainFrame requires BeginFrameArgs to be read from the
LTHI which are only valid inside an impl frame. Due to needing to call
BeginMainFrame asynchronously (as the comment inside the function
mentions), the call could theoretically occur outside an impl frame.
The likelihood of this effecting production is extremely low because
impl frames are large but this problem does increase flakiness of the
layer tree host unittests.

BUG=346230
R=brianderson,enne

[mithro-old, 2015-04-24 07:11:52]

mithro@mithis.com changed reviewers:
+ enne@chromium.org

[mithro-old, 2015-04-24 07:11:52]

This patch currently doesn't work because
SingleThreadProxy::ScheduledActionSendBeginMainFrame() does a PostTask which
means the impl frame can end before BeginMainFrame actually occurs. See below;


void SingleThreadProxy::ScheduledActionSendBeginMainFrame() {
  TRACE_EVENT0("cc", "SingleThreadProxy::ScheduledActionSendBeginMainFrame");
  // Although this proxy is single-threaded, it's problematic to synchronously
  // have BeginMainFrame happen after ScheduledActionSendBeginMainFrame.  This
  // could cause a commit to occur in between a series of SetNeedsCommit calls
  // (i.e. property modifications) causing some to fall on one frame and some to
  // fall on the next.  Doing it asynchronously instead matches the semantics of
  // ThreadProxy::SetNeedsCommit where SetNeedsCommit will not cause a
  // synchronous commit.
  DCHECK(inside_impl_frame_)
      << "BeginMainFrame should only be sent inside a BeginImplFrame";
  MainThreadTaskRunner()->PostTask(
      FROM_HERE,
      base::Bind(&SingleThreadProxy::BeginMainFrame,
                weak_factory_.GetWeakPtr()));
}

I'm uncertain if when this occurs, it is okay to drop the BeginMainFrame onto
the floor or if we need to actually abort in some way if that occurs.

[mithro-old, 2015-04-29 01:06:30]

mithro@mithis.com changed reviewers:
+ brianderson@chromium.org