Dispose of MIDIAccessInitializer on contextDestroyed() notification.

Dispose of MIDIAccessInitializer on contextDestroyed() notification.

If a contextDestroyed() notification is issued, dispose of the
MIDIAccessInitializer then. With Oilpan enabled, this can be the last
opportunity to dispose via the ExecutionContext, as the weak reference
to the context may well have been cleared by the time the prefinalizer
runs.

R=haraken
BUG=420515
Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=189204

[sof, 2015-01-29 16:47:16]

sigbjornf@opera.com changed reviewers:
+ oilpan-reviews@chromium.org

[sof, 2015-01-29 16:47:17]

Please take a look.

Not particularly satisfying, but I don't see a way to avoid doing this.

[haraken, 2015-01-29 16:55:32]

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

[haraken, 2015-01-29 16:55:32]

LGTM.

Ideally we should unregister the prefinalizer when contextDestroyed is called,
but we cannot do that because LifecycleObserver still needs the prefinalizer...

[sof, 2015-01-29 17:00:39]

On 2015/01/29 16:55:32, haraken wrote:
> LGTM.
> 
> Ideally we should unregister the prefinalizer when contextDestroyed is called,
> but we cannot do that because LifecycleObserver still needs the
prefinalizer...

Maybe we can put a prefinalizer on MIDIController instead, which clears out?

[haraken, 2015-01-29 17:02:02]

> > Ideally we should unregister the prefinalizer when contextDestroyed is
called,
> > but we cannot do that because LifecycleObserver still needs the
> prefinalizer...
> 
> Maybe we can put a prefinalizer on MIDIController instead, which clears out?

Sounds nicer.

[sof, 2015-01-29 17:04:57]

On 2015/01/29 17:02:02, haraken wrote:
> > > Ideally we should unregister the prefinalizer when contextDestroyed is
> called,
> > > but we cannot do that because LifecycleObserver still needs the
> > prefinalizer...
> > 
> > Maybe we can put a prefinalizer on MIDIController instead, which clears out?
> 
> Sounds nicer.

Hmm, it doesn't directly keep track of the registered-with-client object set, so
that switch in approach couldn't be done readily.

[haraken, 2015-01-29 17:06:45]

On 2015/01/29 17:04:57, sof wrote:
> On 2015/01/29 17:02:02, haraken wrote:
> > > > Ideally we should unregister the prefinalizer when contextDestroyed is
> > called,
> > > > but we cannot do that because LifecycleObserver still needs the
> > > prefinalizer...
> > > 
> > > Maybe we can put a prefinalizer on MIDIController instead, which clears
out?
> > 
> > Sounds nicer.
> 
> Hmm, it doesn't directly keep track of the registered-with-client object set,
so
> that switch in approach couldn't be done readily.

OK. Then probably we can land this CL as is. Once we remove a pre-finalizer from
LifecycleObserver, we can unregister MIDIAccessInitializer's prefinalizer when
contextDestroyed() is called.

[sof, 2015-01-29 17:07:11]

The CQ bit was checked by sigbjornf@opera.com

[commit-bot: I haz the power, 2015-01-29 17:08:00]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/864943003/1

[commit-bot: I haz the power, 2015-01-29 18:30:55]

Committed patchset #1 (id:1) as
https://src.chromium.org/viewvc/blink?view=rev&revision=189204