Re: [README] [IMPORTANT] rep-gtk and gtk+3

[Christopher Roy Bratusek <zanghar freenet de>]

Am Thu, 12 Aug 2010 15:08:05 -0400
schrieb Clinton Ebadi <clinton unknownlamer org>:

> Christopher Roy Bratusek <zanghar freenet de> writes:
> 
> > Hi fine folks,
> >
> > so GTK+ 3.0 will be in town around middle of September. Sawfish 3.0 which is
> > *currently* scheduled for June 2011 will be using it (or rather GTK+ 3.2 which is
> > released in March 2011). Either way: the glue-code generator for rep-gtk is rather
> > ugly.
> >
> > So there are two opportunities (actually there are three, but the third is rejected):
> >
> > a) port g-wrap to librep and get bindings to c-apps easily. Difficulty ****, examples:
> > only g-wrap itself.
> 
> IIRC g-wrap is more or less unmaintained now so this is probably a bad
> idea (Guile 2.0 has a dynamic FFI built in based upon libffi).

librep has basic ffi bindings, which may be extended.

> > b) glue librep and gobject-introspection (gis) and get bindings to gis-using apps
> > easily (basically everything gnome-related). Difficulty **, examples: python-gtk,
> > gtkmm (and maybe other bindings) already gis, so we may learn from them.
> 
> I'd take a look to what guile-gnome is doing as well.

the last time i've check it has been using g-wrap.

> 
> > Well I would prefer b). The problem is: the missing manpower. So this is also a call
> > to contributors. (ehh.. to YOU - YOU there and YOU (who want's to leave the show now))
> >
> > Just to complete the list: c) would be switching to GUILE. Rejected. There are various
> > reason for rejecting the switch. (...)
> 
> Actually, is there any reason to not retarget librep onto the guile-vm?
> By this I mean, if patches were to magically appear would they be
> rejected?

90%: yes… basically I've never rejected "magically appearing patches", but I doubt on
this thing.

> By targetting the Guile VM REP would gain the ability to call Guile
> functions... Scheme and REP are similar enough that it seems
> straightforward enough (the only hairy issue is probably nil as false --
> but Guile has a sort-of solution to this now for the emacs-lisp compiler
> by having a special #nil value that satisfies both false? and
> null?... ugly from a type theoretic standpoint but multi-language
> support is in the early stages).
> 
> I'm familiar enough with the Guile internals to do this, but I've yet to
> really look around librep.
> 

Well… I'm not sure how useful it is to use librep and then depend on guile doing the
stuff. FFI and GIS are better options if you ask me.

Regards,
Chris