Don't cancel outstanding connect jobs when "flushing" socket pool.

Don't cancel outstanding connect jobs when "flushing" socket pool.
BUG=50565
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=54155

[eroman, 2010-07-29 06:55:09]

See the bug comment for more details.

[Mike Belshe, 2010-07-29 09:18:42]

http://codereview.chromium.org/3028034/diff/1/2
File net/socket/client_socket_pool_base.cc (left):

http://codereview.chromium.org/3028034/diff/1/2#oldcode656
net/socket/client_socket_pool_base.cc:656: CancelAllConnectJobs();
This might not be the best fix.

Another way to go would be to add code to do the notifications.

I believe it would go into CancelAllConnectJobs and look like this:

  while (group.pending_requests.size()) {
    scoped_ptr<const Request> request(
        RemoveRequestFromQueue(
          group.pending_requests.begin(),
          &group.pending_requests));
    InvokeUserCallbackLater(r->handle(), r->callback(), ERR_ABORTED);
  }


If you don't do it this way, then we're not being 100% clean on our flushing,
which effects benchmarks and such.

[eroman, 2010-07-29 18:27:00]

Agreed.

My goal was to get a quick fix in (since the current problem makes chrome
unusable), and then do a more involved follow-up change which aborts and has the
necessary unit-tests.

Are you OK with that ordering, or do you want me to tackle the abort in this
change?

[Mike Belshe, 2010-07-29 18:28:10]

On 2010/07/29 18:27:00, eroman wrote:
> Agreed.
> 
> My goal was to get a quick fix in (since the current problem makes chrome
> unusable), and then do a more involved follow-up change which aborts and has
the
> necessary unit-tests.
> 
> Are you OK with that ordering, or do you want me to tackle the abort in this
> change?

LGTM for Chrome 6.