Fix shutdown assert in new Mojo EDK.

Fix shutdown assert in new Mojo EDK.

The problem was that on IO thread shutdown there could be a pending task to call CloseOnIO. The deletion of this pending task can free the last reference to MessagePipeDispatcher (or other dispatchers) that are holding on to a RawChannel. The destructor would then assert because channel_ was alive. The fix is to detect this case and destruct the channel, since it's safe as we're on the IO thread.

I only saw this when running with --v8-pac-mojo-out-of-process.

BUG=561803
Committed: https://crrev.com/8c4d270033fcce330ccb80350446125692c8e0c4
Cr-Commit-Position: refs/heads/master@{#362161}

[jam, 2015-11-26 05:10:07]

Description was changed from

==========
Fix shutdown assert in new Mojo EDK.

The problem was that on IO thread shutdown there could be a pending task to call
CloseOnIO. The deletion of this pending task can free the last reference to
MessagePipeDispatcher (or other dispatchers) that are holding on to a
RawChannel. The destructor would then assert because channel_ was alive. The fix
is to detect this case and destruct the channel, since it's safe as we're on the
IO thread.

I only saw this when running with --v8-pac-mojo-out-of-process.

BUG=561803
==========

to

==========
Fix shutdown assert in new Mojo EDK.

The problem was that on IO thread shutdown there could be a pending task to call
CloseOnIO. The deletion of this pending task can free the last reference to
MessagePipeDispatcher (or other dispatchers) that are holding on to a
RawChannel. The destructor would then assert because channel_ was alive. The fix
is to detect this case and destruct the channel, since it's safe as we're on the
IO thread.

I only saw this when running with --v8-pac-mojo-out-of-process.

BUG=561803
==========

[jam, 2015-11-26 05:10:07]

jam@chromium.org changed reviewers:
+ yzshen@chromium.org

[yzshen1, 2015-11-30 17:02:06]

lgtm

[jam, 2015-11-30 17:03:07]

The CQ bit was checked by jam@chromium.org

[commit-bot: I haz the power, 2015-11-30 17:04:26]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1475113004/1
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1475113004/1

[commit-bot: I haz the power, 2015-11-30 18:16:09]

Description was changed from

==========
Fix shutdown assert in new Mojo EDK.

The problem was that on IO thread shutdown there could be a pending task to call
CloseOnIO. The deletion of this pending task can free the last reference to
MessagePipeDispatcher (or other dispatchers) that are holding on to a
RawChannel. The destructor would then assert because channel_ was alive. The fix
is to detect this case and destruct the channel, since it's safe as we're on the
IO thread.

I only saw this when running with --v8-pac-mojo-out-of-process.

BUG=561803
==========

to

==========
Fix shutdown assert in new Mojo EDK.

The problem was that on IO thread shutdown there could be a pending task to call
CloseOnIO. The deletion of this pending task can free the last reference to
MessagePipeDispatcher (or other dispatchers) that are holding on to a
RawChannel. The destructor would then assert because channel_ was alive. The fix
is to detect this case and destruct the channel, since it's safe as we're on the
IO thread.

I only saw this when running with --v8-pac-mojo-out-of-process.

BUG=561803
==========

[commit-bot: I haz the power, 2015-11-30 18:16:10]

Committed patchset #1 (id:1)

[commit-bot: I haz the power, 2015-11-30 18:17:27]

Description was changed from

==========
Fix shutdown assert in new Mojo EDK.

The problem was that on IO thread shutdown there could be a pending task to call
CloseOnIO. The deletion of this pending task can free the last reference to
MessagePipeDispatcher (or other dispatchers) that are holding on to a
RawChannel. The destructor would then assert because channel_ was alive. The fix
is to detect this case and destruct the channel, since it's safe as we're on the
IO thread.

I only saw this when running with --v8-pac-mojo-out-of-process.

BUG=561803
==========

to

==========
Fix shutdown assert in new Mojo EDK.

The problem was that on IO thread shutdown there could be a pending task to call
CloseOnIO. The deletion of this pending task can free the last reference to
MessagePipeDispatcher (or other dispatchers) that are holding on to a
RawChannel. The destructor would then assert because channel_ was alive. The fix
is to detect this case and destruct the channel, since it's safe as we're on the
IO thread.

I only saw this when running with --v8-pac-mojo-out-of-process.

BUG=561803

Committed: https://crrev.com/8c4d270033fcce330ccb80350446125692c8e0c4
Cr-Commit-Position: refs/heads/master@{#362161}
==========

[commit-bot: I haz the power, 2015-11-30 18:17:28]

Patchset 1 (id:??) landed as
https://crrev.com/8c4d270033fcce330ccb80350446125692c8e0c4
Cr-Commit-Position: refs/heads/master@{#362161}