Win32: Use of // after a successful /(?{})/ results in strange effect

[p5pRT]

Migrated from rt.perl.org#134246 (status was 'open')

Searchable as RT134246$

[p5pRT]

From @haukex

Hi,

As first reported by PerlMonks user Eily at
http​://www.perlmonks.org/?node_id=11102239 :

The output of the attached script under Linux, Perls 5.20.0 through
5.30.0, is as expected​:
--8<--
Regex!
Item​: Bar
Item​: Foo
Regex!
-->8--

However, under Strawberry Perl 5.30.0.1-64bit on Win 7, the output is​:
--8<--
Regex!

Item​: Bar

Item​: Foo

Use of uninitialized value in say at strangebug.pl line 4.

Foo

-->8--

This seems to be some really strange memory (stack? dunno) issue that I
can't further diagnose myself. I also haven't been able to make the
example code much shorter yet.

Regards,
-- Hauke D

[p5pRT]

From @haukex

strangebug.pl

[p5pRT]

From @haukex

On Mon, 01 Jul 2019 10​:25​:48 -0700, haukex@​zero-g.net wrote​:

However, under Strawberry Perl 5.30.0.1-64bit on Win 7, the output is​:
--8<--
Regex!

Item​: Bar

...

Sorry, the output under Windows doesn't really have those extra blank lines, seems my mail client didn't like me pasting CRLF into the mail (looked fine before sending).

Regards,
-- Hauke D

[p5pRT]

From [Unknown Contact. See original ticket]

On Mon, 01 Jul 2019 10​:25​:48 -0700, haukex@​zero-g.net wrote​:

However, under Strawberry Perl 5.30.0.1-64bit on Win 7, the output is​:
--8<--
Regex!

Item​: Bar

...

Sorry, the output under Windows doesn't really have those extra blank lines, seems my mail client didn't like me pasting CRLF into the mail (looked fine before sending).

Regards,
-- Hauke D

[p5pRT]

From @sisyphus

On my Windows 7 box I have stable perl versions 5.x.0 for x in the range
10..30.
The strange behaviour begins with perl-5.18.0 and persists all the way
through to 5.30.0.
The strange behaviour is not dependent upon architecture (at least not the
32bit/64bit, ivtype, ntype variants that I have), nor can it be attributed
to a change compiler behaviour as my perl-5.18.0 was built with the same
gcc-4.7.0 that built the earlier versions.

Hope that helps narrow it down a bit.

Cheers,
Rob

On Tue, Jul 2, 2019 at 3​:28 AM Hauke D via RT <perlbug-comment@​perl.org>
wrote​:

On Mon, 01 Jul 2019 10​:25​:48 -0700, haukex@​zero-g.net wrote​:

However, under Strawberry Perl 5.30.0.1-64bit on Win 7, the output is​:
--8<--
Regex!

Item​: Bar

...

Sorry, the output under Windows doesn't really have those extra blank
lines, seems my mail client didn't like me pasting CRLF into the mail
(looked fine before sending).

Regards,
-- Hauke D

[p5pRT]

From @sisyphus

Sorry - I failed to mention that I (obviously) removed the "use 5.020;"
from the demo script and replaced "say" with "print" & "\n".

Cheers,
Rob

On Tue, Jul 2, 2019 at 12​:23 PM sisyphus <sisyphus359@​gmail.com> wrote​:

On my Windows 7 box I have stable perl versions 5.x.0 for x in the range
10..30.
The strange behaviour begins with perl-5.18.0 and persists all the way
through to 5.30.0.
The strange behaviour is not dependent upon architecture (at least not the
32bit/64bit, ivtype, ntype variants that I have), nor can it be attributed
to a change compiler behaviour as my perl-5.18.0 was built with the same
gcc-4.7.0 that built the earlier versions.

Hope that helps narrow it down a bit.

Cheers,
Rob

On Tue, Jul 2, 2019 at 3​:28 AM Hauke D via RT <perlbug-comment@​perl.org>
wrote​:

On Mon, 01 Jul 2019 10​:25​:48 -0700, haukex@​zero-g.net wrote​:

However, under Strawberry Perl 5.30.0.1-64bit on Win 7, the output is​:
--8<--
Regex!

Item​: Bar

...

Sorry, the output under Windows doesn't really have those extra blank
lines, seems my mail client didn't like me pasting CRLF into the mail
(looked fine before sending).

Regards,
-- Hauke D

[p5pRT]

From @hvds

I haven't investigated what's going on here, but something similar happens with this simpler testcase​:

% perl -wle '(my $x = "x") =~ /.(?{ print "Regex!" })/; func(); sub func { "Foo" =~ //i }'
Regex!

%

Running with -Mre=debug shows that we really are attempting to re-invoke the regexp, but the EVAL node is somehow printing an empty string - at least with my (Ubuntu) system perl, a threaded non-debugging perl-5.26.1 (+ 67 Ubuntu patches). Reinvoking at the top level (without wrapping in the function call) prints "Regex!" twice as expected.

I can't reproduce the problem with an unthreaded debugging 5.30.0.

I haven't got time to dig further right now, but it feels like clarifying relevant build options would be a useful first step.

Hugo

[p5pRT]

The RT System itself - Status changed from 'new' to 'open'