On 21.01.2013, at 12:44, Nicolas Cellier <nicolas.cellier.aka.nice at gmail.com> wrote:

> Ah ah, I wanted to ask if 3 timesRepeat garbageCollect was a kind of
> magical thinking or a known limitation of our collector :)
> 
> I noticed several usages of such snippet, in ReleaseBuilder and
> ScriptLoader and WeakSetInspectorTest, but oh, the last has my
> initials stamped :(
> 
> I also wonder what are the effects of flushing the caches...
> 
> As I understand it, MCFileBasedRepository flushAllCaches will ignore
> the copy already in package-cache and force a re-load from network. It
> is already present above in code, so let's consider the effect is
> benign.
> (Personnaly I would only trigger a reload action in exception
> handling, because if network is unreliable, there is a high risk to
> replace a finally valid load by an invalid one).
> 
> I have no idea of what (MCDefinition clearInstances) does, all I see
> is that 'instances' variable is weak, so I presume it is supposed to
> auto-clean. While chasing the pointers of MCDefinition
> allSubInstances, I noticed a lot of MC related scories in Object's
> DependentFields.
> So we can consider this cache flushing as a workaround, but should
> better chase the buggish dependents...
> 
> Nicolas

Oh, I didn't even notice the clearInstances. That's definitely not needed as they are weak, as you say.

- Bert -

> 
> 2013/1/21 Bert Freudenberg <bert at freudenbergs.de>:
>> On 21.01.2013, at 14:05, commits at source.squeak.org wrote:
>> 
>>> 3 timesRepeat: [ Smalltalk garbageCollect].
>> 
>> Doing this 3 times is superstitious. It won't clean up more than doing it once.
>> 
>> IMHO an explicit GC is not necessary at all - what purpose does it serve?
>> 
>> - Bert -
>> 
>> 
>