On Sat, Jan 03, 2026 at 05:29:03PM +0000, James Cook wrote:
> Andrew Hewus Fresh <andrew@afresh1.com> wrote:
> > Yet another perl update, this time to 5.42.0.  There is talk about a .1
> > release, but it didn't sound like anything that should block this update.
> 
> There is a regression in 5.42.0: a segfault which is fairly easy to
> encounter. Here is a simple way to trigger it:
> 
> 	perl -c -wE 'for my ($x, $y) (Bar->foo) {}'
> 
> This causes perl 5.42.0 to segfault. Perl 5.40.0 says "-e syntax OK".
> 
> I've included the upstream fix as a patch. I would be grateful if it
> could be included in the next OpenBSD release.

Seems like a good idea, I'll read more about it and test it out.

Thank you!


> The problem is documented here: https://github.com/perl/perl5/issues/23405
> The below patch is upstream commit 96673a4b, which (when I asked in October)
> people on #perl in libera.chat thought was a reasonable thing to apply to
> 5.40.0.
> 
> It is perhaps worth noting that the October perl5-porters mailing list
> post about a 5.42.1 maintenance release suggested it would be worth it
> even if it only included that commit.
> 
> Although I have only minimally tested this on OpenBSD, I can say
> that on NetBSD, the full perl test suite passed when I applied that
> same change to the 5.40.0 source (and it's the patch is in pkgsrc).
> 
> (BTW, is there a reliable way to build/test only gnu/usr.bin/perl in
> OpenBSD's source tree? In order to test this patch, I just followed the
> release(8) instructions excluding xenocara, which feels like overkill,
> but I didn't want to make assumptions (e.g: I can just follow perl's
> build instructions) or spend too much time digging into the build
> system internals.)
> 
> -- 
> James
> 
> diff /usr/src
> path + /usr/src
> commit - 287000ffa12fe34ef71791da513207ea48a4b740
> blob - f616532c491cebc2d5dc89f124135ff4a9841281
> file + gnu/usr.bin/perl/op.c
> --- gnu/usr.bin/perl/op.c
> +++ gnu/usr.bin/perl/op.c
> @@ -9665,7 +9665,7 @@ S_op_is_cv_xsub(pTHX_ OP *o, XSUBADDR_t xsub)
>          }
>  
>          case OP_PADCV:
> -            cv = (CV *)PAD_SVl(o->op_targ);
> +            cv = find_lexical_cv(o->op_targ);
>              assert(cv && SvTYPE(cv) == SVt_PVCV);
>              break;
>  
> @@ -9683,10 +9683,18 @@ S_op_is_cv_xsub(pTHX_ OP *o, XSUBADDR_t xsub)
>  static bool
>  S_op_is_call_to_cv_xsub(pTHX_ OP *o, XSUBADDR_t xsub)
>  {
> -    if(o->op_type != OP_ENTERSUB)
> +    if (o->op_type != OP_ENTERSUB)
>          return false;
>  
> -    OP *cvop = cLISTOPx(cUNOPo->op_first)->op_last;
> +    /* entersub may be a UNOP, not a LISTOP, so we can't just use op_last */
> +    OP *aop = cUNOPo->op_first;
> +    if (!OpHAS_SIBLING(aop)) {
> +        aop = cUNOPx(aop)->op_first;
> +    }
> +    aop = OpSIBLING(aop);
> +    OP *cvop;
> +    for (cvop = aop; OpHAS_SIBLING(cvop); cvop = OpSIBLING(cvop)) ;
> +
>      return op_is_cv_xsub(cvop, xsub);
>  }
>  
> commit - 287000ffa12fe34ef71791da513207ea48a4b740
> blob - 2f6790aee7758e16c2c86374648e22a9304ca4fe
> file + gnu/usr.bin/perl/t/op/for-many.t
> --- gnu/usr.bin/perl/t/op/for-many.t
> +++ gnu/usr.bin/perl/t/op/for-many.t
> @@ -498,4 +498,17 @@ is($continue, 'xx', 'continue reached twice');
>      is("@have", "Pointy end Up Flamey end Down", 'for my ($one, $two)');
>  }
>  
> +# GH #23405 - segfaults when compiling 2-var for loops
> +{
> +    my $dummy = sub {};
> +    for my ($x, $y) (main->$dummy) {}
> +    pass '2-var for does not crash on method calls';
> +
> +    my sub dummy {}
> +    sub {
> +        for my ($x, $y) (dummy) {}
> +    }->();
> +    pass '2-var for does not crash on lexical sub calls';
> +}
> +
>  done_testing();

-- 
andrew

Full-time system administration is a delicate balance 
    between proactiveness and laziness.
                      --  jhorwitz from use.perl.org