g_type_is_a() shortcut (Re: Optimizations)

[Tim Janik <timj gtk org>]

On Tue, 3 Feb 2004, Stefan Westerfeld wrote:

> * No patches: performance of the unmodified build.
>
> stefan@luna1:~/src/beast-performance-eval/beast/tests$ perftest
> 0.980416 seconds for 10000 invocations => 10199.751787 invocations/second, 0.000098 sec per invocation

> stefan@luna1:~/src/beast-performance-eval/beast/tests$ ./perftest
> 0.613326 seconds for 10000 invocations => 16304.543860 invocations/second, 0.000061 sec per invocation
>
> * Patch #9: That was just a try - I think we can trivially say that g_type_is_a
>   is TRUE if the arguments match. I am not sure if its worth the extra code,
>   though.
>
> --- gtype.orig.c        2004-02-03 01:42:50.000000000 +0100
> +++ gtype.c     2004-02-03 01:40:39.000000000 +0100
> @@ -2341,6 +2341,9 @@ g_type_is_a (GType type,
>    TypeNode *node, *iface_node;
>    gboolean is_a;
>
> +  if (type == iface_type)
> +    return TRUE;
> +
>    node = lookup_type_node_I (type);
>    iface_node = lookup_type_node_I (iface_type);
>    is_a = node && iface_node && type_node_conforms_to_U (node, iface_node, TRUE, TRUE);

this changes behaviour in that even invalid type ids can succeed an is_a
check. apart from that, this trivial check is what we try to optimize
with the sophisticated macro version in gtype.h (Patch #7 being part
of which), so i don't expect a relevant benefit from this.

>
>    Cu... Stefan

---
ciaoTJ