Terminating a Process when an object dies

> Hi all,
>
>
> @Sven, yes of course I forgot the WeakArray to save the hard ref in my
> previous explanation.
>
>
> 2013/4/23 Igor Stasenko <[hidden email]>
>>
>> On 23 April 2013 18:02, Sean P. DeNigris <[hidden email]> wrote:
>> > Igor Stasenko wrote
>> >> but what you expecting?
>> >> you refer to 'self' multiple times. sure thing it cannot be GC-ed,
>> >> neither process will die because
>> >> of plenty of references on stack.
>> >> Even block closure which you creating for fork will have home context
>> >> which keeps reference to 'self',
>> >> e.g. receiver.
>> >
>> > I'm expecting magic apparently because I am stupid when it comes to
>> > processes ;) I changed all the "self"s to array first, which also didn't
>> > work, but then I read your statement about the home context...
>> >
>> > Luc's class-side trick seems to work:
>> > Timer>>start
>> >
>> >         | array |
>> >         array := WeakArray with: self.
>> >         process := self class createProcessFor: array. "tried passing
>> > self here,
>> > but it seemed to keep the process around"
>> >
>> of course, because the home context (will be the method below) of
>> closure captures method's arguments,
>> and holding them strongly..
>>
>>
>> > Timer class>>createProcessFor: aWeakArray
>> >
>> >         [ [ aWeakArray first notNil ] whileTrue: [
>> >                 | timer |
>> >                 aWeakArray first interval asDelay wait.
>> >                 "We have to check again if the timer was garbage
>> > collected during the
>> > wait"
>> >                 aWeakArray first ifNotNil: [
>> >                         aWeakArray first value: aWeakArray first value +
>> > 1.
>> >                         aWeakArray first announcer announce:
>> > (TimerValueChanged to: aWeakArray
>> > first value) ] ] ] fork.
>> >
>> > Wow, it would seem that an object using a process internally that is
>> > only
>> > valid when the object is around is a fairly common use case. I can't
>> > believe
>> > how complicated this is :/
>> >
>> it is not apparent to me why this is common use case. :)
>>
>> you can also simply pass a copy of 'self' to process.
>> as long as copy accessing same announcer, your ticker process will
>> work perfectly.
>> then all you would need to do is:
>>
>> | copy |
>> copy := self copy.
>
>
> @Igor.
> It will not work here I think.
> because you make the shallow copy before initializing the process and the
> finalize message is then sent to the copy.

> Anyway, I always fight with the finalization mechanism.
>
> With Noury we wrote an ActiveObject class some time ago that has an internal
> process that is stopped and the activeobject is finalized.
> The only thing is to be **really** carreful when giving the block to the
> ActiveObject to not introduce references.
>
> Here one of the test case that is green:
>
> testDoesStopBlockUponFinalizationForLowestPriority
> |counter startSemaphore finalizationSemaphore process |
> counter := 0.
> startSemaphore := Semaphore new.
> finalizationSemaphore := Semaphore new.
> activeObject := ActiveObject
> do: [
> startSemaphore signal.
> [(Delay forMilliseconds: 50) wait]repeat
> ]
> ensure: [
> counter := 1.
> finalizationSemaphore signal.
> ].
> activeObject
> priority: Processor lowestPriority;
> start.
> self deny: (startSemaphore waitTimeoutSeconds: 2).
> process := activeObject process.
> activeObject := nil.
> Smalltalk garbageCollect.
> self deny: (finalizationSemaphore waitTimeoutSeconds: 2).
> self assert: counter = 1.
> self assert: process isTerminated
>
>
> Here the whole code:
>
> MCHttpRepository
> location: 'http://car.mines-douai.fr/squeaksource/ReusableBricks'
> user: ''
> password: ''
>
> Luc
>
>
>>
>> process := self class createProcessFor: copy.
>> WeakRegistry default add: self executor: copy.
>>
>> and implement:
>>
>> finalize
>>   process terminate.
>>   process := nil.
>>
>>
>> Then createProcessFor: will be simple:
>>
>> Timer class>>createProcessFor: timer
>>  [
>> [ timer wait. " make Demeter a bit happier"
>>   timer tick. "this should increment counter and announce
>> TimerValueChanged,
>>                      like that we're not make Demeter unhappy"
>> ] whileTrue   ] fork.
>>
>> (actually , to make mr.Demeter completely happy, i would do
>>   [ timer waitAndTick ] whileTrue.
>> )
>>
>> and, tick, will be something like that:
>>
>> tick
>>   process ifNil: [ ^ false ].
>>   self incrementCounter.
>>   self announce: (MyGreatAnnouncement with: whatever).
>>   ^ true.
>>
>> cheerz :)
>>
>> > -----
>> > Cheers,
>> > Sean
>> > --
>> > View this message in context:
>> > http://forum.world.st/Terminating-a-Process-when-an-object-dies-tp4683036p4683144.html
>> > Sent from the Pharo Smalltalk Users mailing list archive at Nabble.com.
>> >
>>
>>
>>
>> --
>> Best regards,
>> Igor Stasenko.
>>
>

> On 23 Apr 2013, at 19:59, "Sean P. DeNigris" <[hidden email]> wrote:
>
>> Sean P. DeNigris wrote
>>> If you run TimerTest>>#testStart from Nautilus with an open Process
>>> Browser (with auto-update enabled), you'll see the process appear and then
>>> disappear.
>>
>> Although if I evaluate the snippet in the class comment of SpdLedMorph, the
>> process hangs around. Arggghhhh. This is crazy!!!
>
> I loaded your code and I have the impression that it works, both in the test and in a workspace.
> But it drives me crazy that my version does not seem to work.
> I will have to come back to this tomorrow with a fresh mind.
>
> Sven