ServiceWorker: Unregistering active version should mark it redundant

ServiceWorker: Unregistering active version should mark it redundant

Spec issue:

https://github.com/slightlyoff/ServiceWorker/issues/353

[michaeln, 2014-07-03 00:48:54]

this makes sense to me

[jsbell, 2014-07-07 20:03:45]

On 2014/07/03 00:48:54, michaeln wrote:
> this makes sense to me

The spec was updated to do this slightly differently:

https://slightlyoff.github.io/ServiceWorker/spec/service_worker/index.html#un...

* if registration.installingWorker -> set it redundant
* if registration.waitingWorker -> set it redundant
* resolve the promise
* wait until no document is using registration
* if registration.activeWorker -> set it redundant
* delete the registration

... so I suspect the fix will need to be more complex, but I need to stare at
the code a bit longer to see how our impl matches the algorithm.

[michaeln, 2014-07-07 22:06:38]

On 2014/07/07 20:03:45, jsbell wrote:
> On 2014/07/03 00:48:54, michaeln wrote:
> > this makes sense to me
> 
> The spec was updated to do this slightly differently:
> 
>
https://slightlyoff.github.io/ServiceWorker/spec/service_worker/index.html#un...
> 
> * if registration.installingWorker -> set it redundant
> * if registration.waitingWorker -> set it redundant
> * resolve the promise
> * wait until no document is using registration
> * if registration.activeWorker -> set it redundant
> * delete the registration
> 
> ... so I suspect the fix will need to be more complex, but I need to stare at
> the code a bit longer to see how our impl matches the algorithm.

there are a lot of hands poking into this aspect of things, see the recently
added OnNoControllees listerner method
https://codereview.chromium.org/355163003/ 

a call to SetStatus(redundant) in the Storage::OnNoControllees might work?

[jsbell, 2014-07-07 22:21:14]

On 2014/07/07 22:06:38, michaeln wrote:
> there are a lot of hands poking into this aspect of things, see the recently
> added OnNoControllees listerner method
> https://codereview.chromium.org/355163003/ 
> 
> a call to SetStatus(redundant) in the Storage::OnNoControllees might work?

Thanks - sounds promising!

I don't know (without poking deeper) if that's guaranteed to be called for all
registrations, or merely for registrations that had > 0 controllees then lost
them. It would make sense if it's the latter, at which point then a
SetStatus(redundant) would match the spec, I think.

If one of those many hands poking at this chunk of code can speak confidently
about the lifecycle and wants to take this, please do. :)

[michaeln, 2014-07-07 23:49:10]

> Thanks - sounds promising!
> 
> I don't know (without poking deeper) if that's guaranteed to be called for all
> registrations, or merely for registrations that had > 0 controllees then lost
> them. It would make sense if it's the latter, at which point then a
> SetStatus(redundant) would match the spec, I think.
> 
> If one of those many hands poking at this chunk of code can speak confidently
> about the lifecycle and wants to take this, please do. :)

I can pick this one up.

[falken, 2014-07-08 02:08:40]

On 2014/07/07 23:49:10, michaeln wrote:
> > Thanks - sounds promising!
> > 
> > I don't know (without poking deeper) if that's guaranteed to be called for
all
> > registrations, or merely for registrations that had > 0 controllees then
lost
> > them. It would make sense if it's the latter, at which point then a
> > SetStatus(redundant) would match the spec, I think.
> > 
> > If one of those many hands poking at this chunk of code can speak
confidently
> > about the lifecycle and wants to take this, please do. :)
> 
> I can pick this one up.

Chatted with Michael, I can take this (as part of crbug.com/388095)