Re: Core dump running PL/Perl installcheck with bleadperl [PATCH]

On Sun, Mar 07, 2010 at 12:11:26PM -0500, Tom Lane wrote:
> Tim Bunce <Tim(dot)Bunce(at)pobox(dot)com> writes:
> > I encountered a core dump running PL/Perl installcheck with a very
> > recent git HEAD of PostgreSQL and a not quite so recent git HEAD of perl.
>
> > The cause is a subtle difference between SvTYPE(sv) == SVt_RV and
> > SvROK(sv). The former is checking a low-level implementation detail
> > while the later is directly checking "does this sv contains a reference".
>
> Hmm. Seems like this patch begs the question: if checking SvTYPE(*svp)
> isn't safe, why is it safe to look at SvTYPE(SvRV(*svp))? Shouldn't the
> tests against SVt_PVHV be made more abstract as well?

Some SvTYPE values, like SVt_RV, allow the SV to hold one of a number of
different kinds of things. Others, like SVt_PVHV, don't.

No, I don't like it either but that's the way the "Jenga tower made of
yaks" (to use a phrase recently coined by one of the perl maintainers)
has grown. Something like an SvRVHVOK(sv) would be welcome sugar.

Tim.