Re: The taming of the before-gc-hook (was Re: Upcoming 1.3.2 release)

[Telford Tendys <telford@eng.uts.edu.au>]

> > To support encapsulation I think guile should (no *must*) support a
> > per object finit function.  Maybe as an option.
> > 
> > For example:
> > 
> > scm_set_finit(object, free) /* call free befor object is free'ed */
> > 
> > (Mabe this is already available?)
> 
> No, and, sorry, it is too expensive.

Would it be difficult to make an SMOB that contains exactly one SCM
object and one (lambda (x) ...) that acts as a finit function?
Since the SMOB naturally gets a chance to clean itself up it could call
the lambda to clean up resources in the SCM object. This would be
a bit hairy depending on the SMOB to get cleaned before the cells that
it might link to get reused, dunno how difficult that might be for
gc authors.

Anyhow, if such a thing was done then normal eval would be the same
speed and anyone who needs such a function could wrap one in this
type of SMOB (sacrifice speed and memory only in cases where the
facility is needed). Mind you, this is getting kind of similar to a guardian
anyhow, the main difference would be that with some tweaking, maybe one
of these SMOBs could allow transparent access to its contents (i.e.
get used as a normal SCM object itself). This would depend on the
lambda getting set at time of creation and never changing.

Just an idea...

	- Tel