Mandoline: ConnectToApplication and ConnectToService should fail gracefully

Mandoline: ConnectToApplication and ConnectToService should fail gracefully

It seems ApplicationImpl::ConnectToApplication and ApplicationImpl::ConnectToService are written mostly with the assumption that connections can always be established. However, late arriving connection requests can fail during tear down. This CL ensures that these methods fail gracefully.

This fixes some test flakiness.

My understanding of what's going on is the view_manager is running on one thread and the shell is running on another (presumably the test thread?). When the test completes, the shell goes away, but the view_manager is still running.

BUG=496935
TBR=beng@chromium.org, sky@chromium.org

Committed: https://crrev.com/d663b3298e865fbbe2d21ee2d6ca4a148096f103
Cr-Commit-Position: refs/heads/master@{#337663}

[Fady Samuel, 2015-07-07 18:16:27]

fsamuel@chromium.org changed reviewers:
+ msw@chromium.org

[msw, 2015-07-07 18:33:37]

msw@chromium.org changed reviewers:
+ ben@chromium.org, sky@chromium.org

[msw, 2015-07-07 18:33:37]

lgtm for a temporary workaround. I'm sure Ben, Scott or John will have a better
idea of the long-term fix here. Perhaps the View Manager itself could more
gracefully handle the shell being destroyed amid its initialization? Perhaps
apps ought to ack before the shell can actually shutdown?

[Fady Samuel, 2015-07-07 18:35:59]

On 2015/07/07 18:33:37, msw wrote:
> lgtm for a temporary workaround. I'm sure Ben, Scott or John will have a
better
> idea of the long-term fix here. Perhaps the View Manager itself could more
> gracefully handle the shell being destroyed amid its initialization? Perhaps
> apps ought to ack before the shell can actually shutdown?

For the record, I think the ViewManager could definitely be more graceful. I
don't think apps ACKs before shell shutdown is a good idea because apps may have
hung during their lifetime. A hung mojo app shouldn't block shutdown, I think.
Maybe we could set a deadline for ACKs, thus making shutdown ACKs best effort
only work?

[Fady Samuel, 2015-07-07 18:37:07]

The CQ bit was checked by fsamuel@chromium.org

[commit-bot: I haz the power, 2015-07-07 18:38:26]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1229553002/1

[commit-bot: I haz the power, 2015-07-07 19:30:45]

Committed patchset #1 (id:1)

[commit-bot: I haz the power, 2015-07-07 19:31:35]

Patchset 1 (id:??) landed as
https://crrev.com/d663b3298e865fbbe2d21ee2d6ca4a148096f103
Cr-Commit-Position: refs/heads/master@{#337663}