boxed re-registration key freeing

[Kevin Ryde <user42 zip com au>]

With the current cvs the foo.pl below gets an error,

    type Gtk2::Border is not registered with Glib-Perl at foo.pl line 8.

Chasing it down in gdb it looks like Gtk2.xs register.xsh registers
Gtk2::Border with gperl_register_boxed, and then the Gtk2::Entry boot
code re-registers it.  On that second registration the info_by_gtype
update frees the boxed_info of the first registration, but the package
name string "Gtk2::Border" in that boxed_info is still in use as the key
in info_by_package.  Changing g_hash_table_insert to
g_hash_table_replace to update the key in info_by_package seems to do
the trick.

The GType of GtkBorder is unchanging, so the info_by_gtype doesn't care
between _insert or _replace, if I'm not mistaken.

Is GtkEntry.xs meant to register Gtk2::Border, or can it be left to
register.xsh?


use strict;
use warnings;
use Gtk2 '-init';

Gtk2::Entry->new;
print "Gtk2::Border->copy is ",Gtk2::Border->can('copy'),"\n";

my $pspec = Glib::ParamSpec->boxed ('foo','foo','blurb',
                                    'Gtk2::Border',
                                    ['readable']);
print "pspec ",$pspec,"\n";
print "value_type ",$pspec->get_value_type,"\n";

Index: GBoxed.xs
===================================================================
--- GBoxed.xs	(revision 1062)
+++ GBoxed.xs	(working copy)
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2003-2005 by the gtk2-perl team (see the file AUTHORS for
+ * Copyright (C) 2003-2005, 2009 by the gtk2-perl team (see the file AUTHORS for
  * the full list)
  *
  * This library is free software; you can redistribute it and/or modify it
@@ -186,12 +186,17 @@
 						         NULL);
 	}
 	boxed_info = boxed_info_new (gtype, package, wrapper_class);
+
 	/* We need to insert into info_by_package first because there might
 	 * otherwise be trouble if we overwrite an entry: inserting into
 	 * info_by_gtype frees the boxed_info of the overwritten entry, so that
 	 * boxed_info->package is no longer valid at this point.
+	 *
+	 * Note also it's g_hash_table_replace() for info_by_package,
+	 * because the old key string in the old boxed_info will be freed
+	 * when info_by_gtype updates the value there.
 	 */
-	g_hash_table_insert (info_by_package, boxed_info->package, boxed_info);
+	g_hash_table_replace (info_by_package, boxed_info->package, boxed_info);
 	g_hash_table_insert (info_by_gtype, (gpointer) gtype, boxed_info);
 
 	/* GBoxed types are plain structures, so it would be really