Re: [gtk-list] Re: Garbage Collection and other things (belated)

[Kenneth Albanowski <kjahds kjahds com>]

On Fri, 17 Apr 1998 attila@stalphonsos.com wrote:

> I don't mean to keep beating on this, but have you read Boehm's papers
> or looked at his work?  It's pretty impressive, and has been ported to a
> lot of environments.  

Sorry, yes, I should have said that. I've not explored it deeply, but I'm
familiar with the package.

It certainly does work -- in much the same way threading & dynamic loading
worked before people put the work into making it work everywhere. The
final step -- making it available to everyone on all platforms, may be
simple, very difficult, or perhaps infeasable. That I cannot answer. 

> In any event, I, for one, do not think that you can't implement this
> portably or reliably, especially in the context of something like gtk,
> or whatever gtk turns into.  You do have to really think about it,
> though, and it's much, much harder to take something built around
> reference counts and change it to use another scheme than just doing it
> a different way from the start.

It depends on the level: any generalized GC package in C has to walk over
the entire address space, stack, data, etc. It's possible to create a much
more compact and reliable (in the sense of having less probabalistic
behaviour) GC in certain contexts: for example, the naieve one in PerlGtk
simply walks a hash table every now and then, thus the behavior is very
closely constrained. More importantly, the portability is absolutely
guaranteed. 

-- 
Kenneth Albanowski (kjahds@kjahds.com, CIS: 70705,126)