Re: Fw: Custom widget with GladeXML

[muppet <scott asofyet org>]

On Wednesday, October 1, 2003, at 10:42 AM, Chas Owens wrote:

> > : So it sounds to me like a stack pointer mismanagement issue, but I
> > : can't find anything that looks out of place.
> > :
>
> Should we be using GPerlCallback instead of mucking about with the 
> stack
> ourselves?  I am going to try adapting the code in GtkContainer.xs to 
> what i
> think I need in GladeXML.xs and see how far I can get before someone 
> stops me or
> my brain explodes (offically 26 hours awake now).

it should absolutely be using GPerlCallback and/or GPerlClosure.  from 
my initial once-over[1] it looks like it's using the old gtk-perl-style 
callbacks which accept a list for user data, but sometimes pushes all 
the elements onto the stack and sometimes pushes only one, and appears 
in most cases to leak the AV after all.  it should be fixed to use the 
Glib module's utilities or at least its style, for consistency.

the global custom_args hack should not be necessary -- that's what the 
user_data argument and GPerlCallback are for.

now, it may be quite a trick figuring out how to trigger the proper 
destruction of the callback objects, since no destroy notifiers are 
included; possibly the callback should be placed into object data of 
the object to which it is connected (with g_object_set_data_full).


as for why POPs returns what was on the beginning of the stack 
before...  i have no clue.  the code looks right --- PUSHMARK, load the 
stack, commit the new stack with PUTBACK, call_sv, read the mangled 
stack with SPAGAIN, and POPs... that should read from the end.  i don't 
get it.


[1] i admit, i've never looked at this code before.  i have next to no 
idea what's going on with libglade, so i've never had occasion to touch 
it.

--
muppet <scott at asofyet dot org>