Bug #110073 for Syntax-Feature-Try: Segmentation fault, for the record

Fri Dec 04 09:29:53 2015 EDAVIS [...] cpan.org - Ticket created

Subject:

Segmentation fault, for the record

On my system (Fedora 23) this script gives a segmentation fault. So far I have not been able to narrow it down further. I post it here for the record and to see if you can reproduce it. use syntax 'try'; foreach (2, 3, 4) { try { die "odd\n" if $_ % 2; } catch ($err) { next; } warn "got here $_\n"; }

Fri Dec 04 09:32:16 2015 EDAVIS [...] cpan.org - Correspondence added

Actually, here is a slightly reduced test case: $Carp::Internal{ 'Devel::Declare' } ||= 1; BEGIN { require 'Syntax/Feature/Try.pm'; return Syntax::Feature::Try->install; } foreach (2, 3, 4) { try { die "odd\n" if $_ % 2; } catch ($err) { next; } warn "got here $_\n"; }

Fri Dec 04 09:39:52 2015 tomas.zemres [...] gmail.com - Correspondence added

last,redo,next (loop constrolls) are not supported inside try/catch/finally blocks. See documentation: http://search.cpan.org/~tnt/Syntax-Feature-Try-1.002/lib/Syntax/Feature/Try.pm#CAVEATS Do not call "next" inside "catch" block. Use some workaround: - use local variable and call "next" outside. - or use perl "eval" instead of try/catch. Dne Pá 04.pro.2015 09:32:16, EDAVIS napsal(a): Show quoted text

> Actually, here is a slightly reduced test case: > > $Carp::Internal{ 'Devel::Declare' } ||= 1; > BEGIN { require 'Syntax/Feature/Try.pm'; return Syntax::Feature::Try-

> >install; }

> foreach (2, 3, 4) { > try { > die "odd\n" if $_ % 2; > } > catch ($err) { > next; > } > warn "got here $_\n"; > }

Fri Dec 04 09:39:52 2015 The RT System itself - Status changed from 'new' to 'open'

Mon Dec 07 09:23:15 2015 EDAVIS [...] cpan.org - Correspondence added

Thanks. Is there something that can be done to fail more safely if next is called inside a catch block? To die with a clear message rather than possibly segfault? Is it possible that 'next LABEL' might be made to work even if plain 'next' is unsupported?

Mon Dec 07 14:24:39 2015 SREZIC [...] cpan.org - Correspondence added

On 2015-12-07 09:23:15, EDAVIS wrote: Show quoted text

> Thanks. Is there something that can be done to fail more safely if > next is called inside a catch block? To die with a clear message > rather than possibly segfault? > > Is it possible that 'next LABEL' might be made to work even if plain > 'next' is unsupported?

There's more perl diagnostics when running the first example script with 5.23.5: $ perl5.23.5 /tmp/try.pl got here 2 got here 4 panic: av_extend_guts() negative count (-8).

Wed Dec 09 04:27:32 2015 tomas.zemres [...] gmail.com - Correspondence added

Dne Po 07.pro.2015 09:23:15, EDAVIS napsal(a): Show quoted text

> Thanks. Is there something that can be done to fail more safely if > next is called inside a catch block? To die with a clear message > rather than possibly segfault?

I'am not sure how to detect if "next" is called inside this block (or subroutine). because try/catch block are internally converted to anonymous subroutine and then perl throws warning "Exiting subroutine via next ..." Show quoted text

> Is it possible that 'next LABEL' might be made to work even if plain > 'next' is unsupported?

I don't know.

Tue Dec 15 11:27:01 2015 EDAVIS [...] cpan.org - Correspondence added

Do you think that the segmentation fault with try.pl indicates a bug either in the Syntax::Feature::Try XS code or in perl itself? Or do you think that since 'next' is not supported, the resulting perl interpreter crash is not a bug?