This fixes http://code.google.com/p/chromium/issues/detail?id=772,...

This fixes http://code.google.com/p/chromium/issues/detail?id=772,
which was an issue with the browser UI thread entering a tight loop
at times. The thread inputs of the browser UI thread and the plugin thread
are attached due to the parent child relationship between the windows.

As a result at times the MsgWaitForMultipleObjectsEx API returns the
fact that there are messages in the queue (mouse messages). The subsequent
peek fails to return any messages causing us to enter a loop for a while.
This also happens when the plugin has capture and just before it releases capture, some mousemoves end up in the browser ui thread causing a tight loop.

The fix is to check if there are mouse messages in the queue by invoking
GetQueueStatus and attempting to peek them out with PM_NOREMOVE. If this
fails we call WaitMessage to block until the next message.

The other fix is to return true from ProcessNextWindowsMessage if PeekMessage returns false and there were sent messages in the queue, as
this is as expected. This will ensure that we go back up and peek again
instead of calling MsgWaitForMultipleObjectsEx again.

Bug=772
R=darin


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=2740

[darin (slow to review), 2008-09-30 23:54:08]

LGTM

Is it possible to add a test to message_loop_unittest.cc?

http://codereview.chromium.org/6021/diff/1/2
File base/message_pump_win.cc (right):

http://codereview.chromium.org/6021/diff/1/2#newcode216
Line 216: bool sent_messages_in_queue = false;
please add some comments to explain why this is needed.

http://codereview.chromium.org/6021/diff/1/2#newcode369
Line 369: // The thread inputs of the browser UI thread and the plugin thread
since this is generic code, please try to describe this issue in more generic
terms.  try to avoid mentioning browser UI thread and plugin thread of chrome. 
instead, talk about HWNDs in separate threads, connected with AttachThreadInput,
encountering this issue when one HWND captures all mouse events.

[ananta, 2008-10-01 00:06:39]

Hi Darin

I have updated the comments in the code. Based on our discussion I will look
into whether it is feasible to detach the thread inputs of the browser and the
plugin windows. If this works then it will get rid of the hassle with a hung
plugin taking down the browser hierarchy.

Thanks
Ananta