Re: Interesting issue with managing GC of proxies across process boundaries.

[Daniel Silverstone <dsilvers@...>]

On Mon, Aug 13, 2012 at 11:09:45AM -0300, Roberto Ierusalimschy wrote:
> If you really need to know that object is "about" to be collected (that
> is, it was removed from the weak table but did not run its finalizer
> yet),

It's less about knowing that it's about to be collected and more about the fact
that code running between the 'disappearing as a value' and having the __gc
called will be confused.  Remember that this is multi-process and that the __gc
event is being propagated between processes running on independent timebases
:-)

> use an extra strong table that is a set of the tags not yet
> finalized. Tags are removed from this table by the corresponding
> finalizer. If a tag is in the weak table, it is well and alive.
> If a tag is not in the weak table but is in this extra table, its
> object is about to be finalized. If a tag is in neither table,
> its object is really dead.

That's essentially the approach I chose to implement (it's in soak testing
right now).

I maintain a strong table of tag -> nr-of-proxies-made and the __gc method for
the proxies now decrements that refcount and only propagates the __gc across
the link if it hits zero.

Assuming this soak-tests out as working nicely, I'll have to deploy it to my
Gitano instance and try it in real-world situations.

Thanks for your response, it has at least validated my final approach as 'not
completely insane' :-)

D.

-- 
Daniel Silverstone                         http://www.digital-scurf.org/
PGP mail accepted and encouraged.            Key Id: 3CCE BABE 206C 3B69