Re: scheme closures: crash during garbage collection

[gregory benison]

> Initial reactions:
>  - using the glib loop is really distasteful -- you have no guarantee
>    that the program will use the loop

Agreed.  I have since figured out that everything I was trying to do
in the glib main loop can be done just as easily in the protected-list
mark function, so that the actions I was trying to defer until some
future invocation of the glib loop are instead deferred until the next
gc mark phase, which seems a lot cleaner to me.

> Also, consider glib/REFCOUNTING:
>
>      1. A closure connected to a GObject signal should not be collected
> until
>         either the closure is disconnected, or the object itself (not the
>         wrapper) goes out of existence.
>
>      2. Attaching closures to an object should not prevent the object from
> being
>         freed. Otherwise, any object with a connected signal will live
> forever.
>         Closures should only have a weak reference on an object.

I think there can be a solution to this issue without any big change
in the way guile-gnome deals with refcounting and memory management.
I am aiming for a substitute for scm_gc_protect_object() /
scm_gc_unprotect_object() which has the same semantics but is safe to
call during GC, and I think I have it.  The patch I am implementing,
now that I've eliminated the glib-loop tie-in weirdness, is not even
gobject-specific.  Aside from the fact that it uses a GList to manage
a data structure, it wouldn't even need to be linked against glib.

Below is a short test program and its output with debugging messages
turned on.  (This test program would ordinarily trigger the fatal call
to scm_gc_unprotect_object.)  Note that the scheme callback gets
marked properly as long as the gobject is around and when the gobject
is garbage-collected, the scheme procedure ceases being marked.

-------------------------------------- test program
--------------------------------

(use-modules (gnome gtk))
(display "\n\n *** Setting up a gobject and attaching a scheme
callback *** \n\n")
(define obj (make <gobject>))
(connect obj 'notify (lambda args #f))
(display "\n\n *** Calling (gc) a few times *** \n\n")
(gc)(gc)(gc)
(display "\n\n *** Clobbering the scheme reference to the gobject *** \n\n")
(set! obj #f)
(display "\n\n *** Calling (gc) a few times *** \n\n")
(gc)(gc)(gc)(gc)
(display "\n\n *** All done!! *** \n\n")

--------------------------------- output below -------------------------------

*** Setting up a gobject and attaching a scheme callback ***

gclosure.c:   protecting new closure 0xb7dce4e8 of GuileGClosure 0x8e51c58
gc_bridge.c: scheduling object 0x8e51c58 for protection


*** Calling (gc) a few times ***

gc_bridge.c:  ---- entering GC mark function for bridge 0x8e54048 ----
gc_bridge.c: inserting object 0x8e51c58 into protected list
gclosure.c: gclosure_mark: marking SCM 0xb7dce4e8 of GuileGClosure 0x8e51c58
gc_bridge.c:  ---- entering GC mark function for bridge 0x8e54048 ----
gclosure.c: gclosure_mark: marking SCM 0xb7dce4e8 of GuileGClosure 0x8e51c58
gc_bridge.c:  ---- entering GC mark function for bridge 0x8e54048 ----
gclosure.c: gclosure_mark: marking SCM 0xb7dce4e8 of GuileGClosure 0x8e51c58


*** Clobbering the scheme reference to the gobject ***



*** Calling (gc) a few times ***

gc_bridge.c:  ---- entering GC mark function for bridge 0x8e54048 ----
gclosure.c: gclosure_mark: marking SCM 0xb7dce4e8 of GuileGClosure 0x8e51c58
gclosure.c:   unprotecting closure 0xb7dce4e8 of GuileGClosure 0x8e51c58
gc_bridge.c: scheduling object 0x8e51c58 for de-protection
gc_bridge.c:  ---- entering GC mark function for bridge 0x8e54048 ----
gc_bridge.c: excising object 0x8e51c58 from protected list
gc_bridge.c:  ---- entering GC mark function for bridge 0x8e54048 ----
gc_bridge.c:  ---- entering GC mark function for bridge 0x8e54048 ----


*** All done!! ***