Re: Reördering SVt_* constants; "internal" SV type

"Paul \"LeoNerd\" Evans" <[email protected]> Mon, 29 Jun 2026 23:00:58 +0100

Newsgroups	gmane.comp.lang.perl.perl5.porters
Message-ID	<[email protected]>

On Thu, 18 Jun 2026 14:45:20 +0200
"H.Merijn Brand" via perl5-porters <[email protected]> wrote:

> I am scared it would break a lot of CPAN/XS

Actually, having read them over I think a lot of your examples prove my
exact point ;)

For example:

> DBI.xs:    if (!SvROK(ref) || SvTYPE(SvRV(ref)) > SVt_PVMG) /* XXX LV */

Yup. This line here is attempting to check that `ref` is definitely a
reference to a perl scalar; so it's going to reject anything that isn't
ROK or anything whose RV's SvTYPE is greater than SVt_PVMG. But it
therefore misses the SVt_PVLV case. In fact its author is even aware of
this, having written  /* XXX LV */  in the comment there (unless they
did mean to write the Roman numbers 30, 53. But I doubt it ;) )

Having skim-read most of the rest of the other examples, I think every
single one of them is going to fall foul of the exact issue I am trying
to address here. Namely, that they will have a false positive and
believe that an SVt_INVLIST is a perl scalar (when in fact it is not),
and a false negative and believe that an SVt_PVLV is not a perl scalar
(when in fact it is and should be treated as such).

I do not believe that the simple adjustment of moving SVt_INVLIST out
of that range and moving SVt_PVLV up next to SVt_PVMG is going to
change the behaviour of any of these previous comparisons which were
already correct, because if they were correct before they'll still be
correct afterwards. However, I think it will stand a good chance of
suddenly making coincidentally correct a lot of code that previously
missed those two edge cases.

-- 
Paul "LeoNerd" Evans

[email protected]
http://www.leonerd.org.uk/  |  https://metacpan.org/author/PEVANS

signature.asc (application/pgp-signature, 228 B)