Re: [RFC] automatic property getters and setters

On Sat, 2004-06-12 at 15:23, muppet wrote:

> > While toying around with param_test.pl, I found a small glitch in
> > gperl_value_from_sv, though:  it never checks if the sv is NULL or undef
> > -- so returning undef from the custom GET_PROPERTY made it croak in
> > gperl_get_object_check.  The attached patch (hopefully) fixes that by
> > returning FALSE if the sv is not ok.
> 
> i had seen that, too, and meant to fix it but decided not to conflate
> things.
> 
> gperl_value_from_sv() and gperl_sv_from_value() should fail only by
> croaking and should always return TRUE.  this is something decided a
> long, long time ago, but the signatures were never unchanged to avoid
> breaking existing code.

Ah, I didn't know about that.

> and, rationalizing it to myself; g_value_init() sets a default value
> (whatever 0 means to that type) that should correspond roughly to
> whatever we would map undef to, so bailing out early should be fine, and
> should not be an error because we did map the sv (undef) to a value.
> 
>    if (!sv || !SvOK(sv))
>        return TRUE;
> 
> now, is it really a good idea?  there's an implicit behavior for most of
> those types; enums will croak on undef (currently) telling you that you
> sent an invalid value through.  this change will silently skip that
> warning.  of course, that warning was often hard to track down...  is
> that alright?

Yeah, I see your point.  What about doing that check in the
G_TYPE_OBJECT branch of the switch then?  If we don't check the SV,
gperl_get_object_check will always croak and returning undef from a
subroutine that is supposed to return an object will not work.

Bye,
-Torsten