On Tuesday 20 May 2008 13:05:43 Will Coleda wrote:
> The original problem description is a little off: this works for core
> Integer type, but not for the parrotObject subclass.
>
> I was able to use gdb to break in the function called by the cmp op,
> and then step through to find which actual function this was being
> sent to via MMD. It was using Integer's 'cmp', in the DEFAULT branch,
> which assumes things about the storage of SELF which are apparently
> not true.
>
> I suspect that if this is the right fix (see below), that we need to
> replace this usage throughout the core PMCs, or subclassing is going
> to break in all sorts of odd places.
>
> $ svn diff src
> Index: src/pmc/integer.pmc
> ===================================================================
> --- src/pmc/integer.pmc (revision 27433)
> +++ src/pmc/integer.pmc (working copy)
> @@ -1174,7 +1174,7 @@
> }
> MMD_DEFAULT: {
> /* int or undef */
> - const INTVAL diff = PMC_int_val(SELF)
> + const INTVAL diff = VTABLE_get_integer(INTERP, SELF)
> - VTABLE_get_integer(INTERP, value);
> return diff > 0 ? 1 : diff < 0 ? -1 : 0;
> }
This is correct. PMC_*_val must die.
-- c
