Re: GObject reference counts

["muppet" <scott asofyet org>]

Robert G. Werner said:
> Andrew Pimlott wrote:
> [snip]
> > Although implementing GC semantics will be a witch for you, I'm
> > certain the current behavior will bite other programmers hundreds of
> > times, possibly in subtle ways.  This will ultimately waste more time than
> > the developers are saving, and sour people on gtk2-perl.
> I kind of agree with you.  However,  I really can feel sympathy for  the
> point of view that trying to implement the correctly GCd versions  of
> GObject is a _hard_ problem.

GObject is not so bad.  GtkObject by itself is not so bad.  handling the
idiosyncrasies of both at the same time is what's tricky.


> Maybe,  though,  it is something that can be added later as an option?
>   Getting gtk+2_perl up an running and then implementing an
> alternative view of Gobject in perl (what I mean is some way that perl
> objects and the gtk+ objects would have a one to one corespondance)  later
> seems like a reasonable approach.

in my estimation, that's already been done.  if we wait any longer to do it
right, we'll be breaking an ever-larger body of code when the release with the
new "correct" semantics comes out.

for example, i'm getting ready to write *three* extension library wrappers
that all layer atop Gtk2.  i don't want to have to come back in two months and
make major renovations.


this is a really big bee in my bonnet, so i will sandbox on something to do
with the reference counting semantics for a week or two and post my findings.


-- 
muppet <scott asofyet org>