Re: Avoiding scm_protect_object

[Andy Wingo]

Hi,

I just checked in an implementation of scm_glib_gc_protect_object, and
its unprotect counterpart, into the guile-gnome repository. It seems to
fix the GC problems that were first found by Gregory.

I think my thoughts on toggle refs were a bit half-baked though. I think
we are more or less fine as we are now. To collect the closure-object
cycle I think that we should make the objects mark any closures
connected to signals. Other closures, such as idle callbacks, should
continue to be marked by the scm_glib_gc_protect_object, with the
invalidate() function calling unprotect_object. (Currently all closures
are marked by scm_glib_gc_protect_object.)

Here would be a nice time for people to run scripts/pull-packages, to
update their checkouts :)

Cheers,

Andy
-- 
http://wingolog.org/