Smalltalk › Pharo › Pharo Smalltalk Users

Finalization question

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

10 messages Options

Udo Schneider

Aug 02, 2014; 11:16am

Finalization question

667 posts

All,

I just stumbled over something which thought was pretty easy - turned
out it isn't as easy as I thought or I'm simply _______ (fill in
expression of choice like "too dumb", "blind", etc.)

I have an object which starts a background process to do it's job. This
object holds a reference to the running proccess in an instance variable.

My goal is to #terminate that proccess once the object isn't needed
anymore (before GC).

Using WeakRegistry for the object itself doesn't work. E.g.

MyObject>>#initialize
super initialize.
WeakRegistry default register: self.
myProcess := self startProcess.

MyObject>>#finalize
myProcess terminate. "Doesn't work. myProcess is already nil"

I think this is due to all the ivars of the object already being nil
when #finalize is begin called. As far as I read the code this is due to
Object>>#executor returning a shallow copy of the object and thus being
by intention ...

So how to proceed? Did I miss something obvious? Is overwriting
#executor the way to go (in terms of "does it work" and "doesn't it
break something")?

Best Regards,

Udo

Ben Coman

Aug 02, 2014; 1:33pm

Re: Finalization question

5447 posts

Udo Schneider wrote:

> All,
>
> I just stumbled over something which thought was pretty easy - turned
> out it isn't as easy as I thought or I'm simply _______ (fill in
> expression of choice like "too dumb", "blind", etc.)
>
> I have an object which starts a background process to do it's job.
> This object holds a reference to the running proccess in an instance
> variable.
>
> My goal is to #terminate that proccess once the object isn't needed
> anymore (before GC).
>
> Using WeakRegistry for the object itself doesn't work. E.g.
>
> MyObject>>#initialize
> super initialize.
> WeakRegistry default register: self.
> myProcess := self startProcess.
>
> MyObject>>#finalize
> myProcess terminate. "Doesn't work. myProcess is already nil"
>
> I think this is due to all the ivars of the object already being nil
> when #finalize is begin called. As far as I read the code this is due
> to Object>>#executor returning a shallow copy of the object and thus
> being by intention ...
>
> So how to proceed? Did I miss something obvious? Is overwriting
> #executor the way to go (in terms of "does it work" and "doesn't it
> break something")?
>
> Best Regards,
>
> Udo
>
>
>
>
>

... [show rest of quote]

I don't know much about finalization of Weak constructs, but maybe an
alternative. I assume you have something like..
startProcess
[
"do stuff"
] fork.

so how about...
startProcess
[ [
"do stuff"
myProcess = nil.
] whileFalse ] fork.

cheers -ben

Udo Schneider

Aug 04, 2014; 4:25pm

Re: Finalization question

667 posts

On 02.08.2014 15:33, Ben Coman wrote:

> I don't know much about finalization of Weak constructs, but maybe an
> alternative. I assume you have something like..
> startProcess
> [ "do stuff" ] fork.
>
> so how about...
> startProcess
> [ [
> "do stuff"
> myProcess = nil.
> ] whileFalse ] fork.
>
> cheers -ben
>
>

... [show rest of quote]

That's a clever way to do it ... only it doesn't work :-(

My current understanding is that "self" in the block refers to the
original instance where "myProcess" is still assigned. "self" in
#finalize is the "executor" object which is (by default) a #shallowCopy
of the object where ivar are nil.

I currently have the feeling I'm missing something. Finalization
shouldn't be that hard ... or may I'm simply spoiled comming from a
Dolphin Smalltalk background :-)

Thanks for your help,

Udo

stepharo

Aug 04, 2014; 7:35pm

Re: Finalization question

5159 posts

In reply to this post by Udo Schneider

We should write a chapter on finalization but since I do not know it
well (and I forget)...
igor will answer you.

Stef
On 2/8/14 13:16, Udo Schneider wrote:

... [show rest of quote]

Udo Schneider

Aug 05, 2014; 9:22am

Re: Finalization question

667 posts

On 04.08.2014 21:35, stepharo wrote:> We should write a chapter on
finalization but since I do not know it
> well (and I forget)...
> igor will answer you.
>
> Stef

I found "the" bug. It seems it's a side effect of finalization,
WeakRegistry and cyclic references. I'll write a TestCase for that.

Thanks,

Udo

Santiago Bragagnolo

Aug 05, 2014; 10:10am

Re: Finalization question

219 posts

TaskIT has a feature named StickyReference, for that propouse, is not trivial at all, because when you fork you are using a block that has a reference to a context. In the end even if your variable is nil, your process is being pointed by the context.

What i did in TaskIT is to have a witness object, that mixed with a ghost proxy (the one you will be pointing) can do the magic.

If what you want in your process variable is not really a reference to the process or if you have enough with a partial proxy based in DNU, things can go easier, you just need to register that third object into the weak resgistry.

If you check out taskit commit 20 you will find a mechanism based in a DNU based proxy. If you checkout the last stable version you will find the witness/ghost proxy based one. In any case, look for the method asStickyReference. It will lead you straight to the mechanism.

Gofer new

smalltalkhubUser: 'CAR' project: 'TaskIT';

configuration;

load.

(#ConfigurationOfTaskIT asClass project version: #stable) load: 'full'

2014-08-05 11:22 GMT+02:00 Udo Schneider <[hidden email]>:

On <a href="tel:04.08.2014%2021" value="+33408201421" target="_blank">04.08.2014 21:35, stepharo wrote:> We should write a chapter on finalization but since I do not know it

> well (and I forget)...
> igor will answer you.
>
> Stef

I found "the" bug. It seems it's a side effect of finalization, WeakRegistry and cyclic references. I'll write a TestCase for that.

Thanks,

Udo

... [show rest of quote]

Udo Schneider

Aug 05, 2014; 12:08pm

Re: Finalization question

667 posts

On 05.08.2014 12:10, Santiago Bragagnolo wrote:> TaskIT has a feature
named StickyReference, for that propouse, is not
> trivial at all, because when you fork you are using a block that has a
> reference to a context. In the end even if your variable is nil, your
> process is being pointed by the context.
You're feeding my deepest fears :-)

I also reduced the testcases to even show a (IMHO) faulty behavior w/o
any process at all. Simple references are sufficient.

Still I'm wondering what's the official way to solve that problem. IMHO
objects with forked background processes which want to terminate those
processes implicitly using finalization isn't something uncommon ...

CU,

Udo

Udo Schneider

Aug 05, 2014; 2:35pm

Re: Finalization question

667 posts

On 05.08.2014 14:08, Udo Schneider wrote:
> I also reduced the testcases to even show a (IMHO) faulty behavior w/o
> any process at all. Simple references are sufficient.
I just created a small TestCase showing the (IHMO) faulty behavior. I
hope the intention is clear from the package/var names.

Gofer it
smalltalkhubUser: 'UdoSchneider' project: 'Playground';
package: 'CyclicFinalizationBug';
load.

Check out the tests in CyclicFinalizationBugTestCase.

Hope this helps.

CU,

Udo

Ben Coman

Aug 06, 2014; 12:36am

Re: Finalization question

5447 posts

Udo Schneider wrote:

> On 05.08.2014 14:08, Udo Schneider wrote:
>> I also reduced the testcases to even show a (IMHO) faulty behavior w/o
>> any process at all. Simple references are sufficient.
> I just created a small TestCase showing the (IHMO) faulty behavior. I
> hope the intention is clear from the package/var names.
>
> Gofer it
> smalltalkhubUser: 'UdoSchneider' project: 'Playground';
> package: 'CyclicFinalizationBug';
> load.
>
> Check out the tests in CyclicFinalizationBugTestCase.
>
> Hope this helps.
>
> CU,
>
> Udo
>
>
>
>

... [show rest of quote]

Is there a Fogbugz case to track it?
cheers -ben

Udo Schneider

Aug 06, 2014; 1:38pm

Re: Finalization question

667 posts

On 06.08.2014 02:36, Ben Coman wrote:
> Is there a Fogbugz case to track it?
> cheers -ben

There is now:
https://pharo.fogbugz.com/f/cases/13785/Cyclic-Referencing-self-prevents-an-object-from-being-finalized

CU,

Udo