Use CondVar+Lock+bool instead of WaitableEvent for MessagePumpDefault on posix

Use CondVar+Lock+bool instead of WaitableEvent for MessagePumpDefault on posix

BUG=412137

[darin (slow to review), 2014-09-09 06:21:06]

darin@chromium.org changed reviewers:
+ darin@chromium.org

[darin (slow to review), 2014-09-09 06:21:07]

https://codereview.chromium.org/558553002/diff/1/base/message_loop/message_pu...
File base/message_loop/message_pump_default.cc (right):

https://codereview.chromium.org/558553002/diff/1/base/message_loop/message_pu...
base/message_loop/message_pump_default.cc:109: cv_.TimedWait(max_time);
TODO: you'll need to protect this against spurious wake-ups.

[jamesr, 2014-09-09 06:26:28]

On 2014/09/09 06:21:07, darin wrote:
>
https://codereview.chromium.org/558553002/diff/1/base/message_loop/message_pu...
> File base/message_loop/message_pump_default.cc (right):
> 
>
https://codereview.chromium.org/558553002/diff/1/base/message_loop/message_pu...
> base/message_loop/message_pump_default.cc:109: cv_.TimedWait(max_time);
> TODO: you'll need to protect this against spurious wake-ups.

Do I? If we're doing a timed wait and we wake up too early, the only thing we'll
do is try to reload the work queue (which will take a lock, swap an empty queue
for an empty queue, then return) and then look at the clock and call TimedWait()
again with an updated value.  Seems like what we'd want to do anyway.  To avoid
this, I'd have to look at the clock again in this function if needs_wake_up_
isn't set which seems bad.

[darin (slow to review), 2014-09-09 06:32:03]

On 2014/09/09 06:26:28, jamesr wrote:
> On 2014/09/09 06:21:07, darin wrote:
> >
>
https://codereview.chromium.org/558553002/diff/1/base/message_loop/message_pu...
> > File base/message_loop/message_pump_default.cc (right):
> > 
> >
>
https://codereview.chromium.org/558553002/diff/1/base/message_loop/message_pu...
> > base/message_loop/message_pump_default.cc:109: cv_.TimedWait(max_time);
> > TODO: you'll need to protect this against spurious wake-ups.
> 
> Do I? If we're doing a timed wait and we wake up too early, the only thing
we'll
> do is try to reload the work queue (which will take a lock, swap an empty
queue
> for an empty queue, then return) and then look at the clock and call
TimedWait()
> again with an updated value.  Seems like what we'd want to do anyway.  To
avoid
> this, I'd have to look at the clock again in this function if needs_wake_up_
> isn't set which seems bad.

Good point. I guess it just needs that as a comment :)

[jamesr, 2014-09-22 17:42:28]

https://codereview.chromium.org/558553002/diff/1/base/message_loop/message_pu...
File base/message_loop/message_pump_default.cc (right):

https://codereview.chromium.org/558553002/diff/1/base/message_loop/message_pu...
base/message_loop/message_pump_default.cc:120: needs_wake_up_ = true;
try adding:

if (needs_wake_up_)
  return;

to avoid the CV signal