Steve Hay <[EMAIL PROTECTED]> writes:
>I have an XSUB that returns a list of things like this:
>
>void
>foo()
> PPCODE:
> {
> EXTEND(SP, 2);
> PUSHs(sv_2mortal(newSViv(42)));
> PUSHs(sv_2mortal(newSVpvn("Forty-two", 9)));
> XSRETURN(2);
> }
>
>It works fine as it is, but I though that the following might be a
>little easier on the eye:
>
>void
>foo()
> PPCODE:
> {
> EXTEND(SP, 2);
> PUSHi(42);
> PUSHp("Forty-two", 9);
> XSRETURN(2);
> }
>
>However, when I try to build the latter I get this compiler error:
>
>'targ' : undeclared identifier
>
>Having hunted around a bit for examples of other things using PUSHi()
>and PUSHp(), I found that the constant() XSUB in a typical
>"const-xs.inc" file produed by ExtUtils::Constants contains a PREINIT:
>section like this:
>
> PREINIT:
>#ifdef dXSTARG
> dXSTARG; /* Faster if we have it. */
>#else
> dTARGET;
>#endif
>
>and sure enough, if I add that to my XSUB then it builds and works OK
>once more.
>
>Why is this necessary,
PUSHi et. al. are core perl utils and (most) perl ops have a TARG SV
>and how was I supposed to know it?
Well 'targ' : undeclared identifier seems to have pushed you in the
direction.
>Neither
>dTARGET nor dXSTARG seem to be documented anywhere,
dTARGET is an internals thing.
dXSTARG didn't exist until "recently".
>and there is no
>mention of the need for such things in the documentation of PUSHi() etc.
>
>Aside from all that, is the PUSHi()/PUSHp() example above OK? It
>doesn't appear to leak, but I'm always wary of removing calls to
>sv_2mortal().
It it doesn't leak or give "attempt to free ..." warnings
it is probably okay.
(Devel::Leak may help here.)
>I seem to recall that the "targ" stuff has a different
>"mortality" scheme relating to the OP tree, but I'm not sure.
I haven't used them much myself. As I understand it the OP that does the
XS call now has an SV associated with it (from the PAD?). This will be
undef'ed on scope exit along with other my-ish SVs (I think).
As such I think there may be issues if you recurse (that is
obscure use of XS - but with Tk callbacks it happens a lot).
