Re: boxed re-registration key freeing

[Kevin Ryde <user42 zip com au>]

muppet <scott asofyet org> writes:
> It looks like gperl_boxed_register() takes precautions to insert into
> the hash with the destructor second, to avoid this sort of problem.
> Why isn't that working?

g_hash_table_insert() on info_by_package leaves the key string set to
the one in the original boxed_info, but it's then freed.  I think
g_hash_table_replace() will change to the key string of the new
boxed_info.  Same string contents of course, new pointer.

Electric fence or the like might detect an access after free, but it
might depend on the hashing, and I don't have enough ram to run it up
that way :).

Some printfs at judicious points show garbage though, with the border
not the only one getting a re-registration at some point.

void hprint (gpointer key,
             gpointer value,
             gpointer user_data)
{
        printf (" %s[%p]", key, key);
}

        if (strcmp(package,"Gtk2::Border") == 0) {
                printf ("gperl_register_boxed %s %p\n",
                boxed_info->package, boxed_info->package);
        }

        {
                printf ("Lookup:");
                g_hash_table_foreach (info_by_package, hprint, NULL);
                printf ("\n");
        }