From @cpansprout

The presence of the code block in the second program should make no
difference. The first result (0) is correct.

$ perl5.21.3 -le '$_ =3D "rdvark"; /(?=3D[A-Za-z0-9_])a*?/g; warn pos'
0 at -e line 1.
$ perl5.21.3 -le '$_ =3D "rdvark"; /(?{})(?=3D[A-Za-z0-9_])a*?/g; warn pos'
3 at -e line 1.

That 3 should be 0, too.

A bisect points to this​:

commit 77ebeeb
Author​: Karl Williamson <public@​khwilliamson.com>
Date​: Mon Sep 30 14​:06​:23 2013 -0600

  PATCH​: [perl #120041] regcomp.c missing parens and broken STCLASS

(Sorry, I should have noticed this before the release of 5.20.)

Most importantly, I need a workaround. :-)

--

Father Chrysostomos

From @khwilliamson

On 08/02/2014 12​:01 PM, Father Chrysostomos (via RT) wrote​:

# New Ticket Created by Father Chrysostomos
# Please include the string​: [perl #122460]
# in the subject line of all future correspondence about this issue.
# <URL​: https://rt-archive.perl.org/perl5/Ticket/Display.html?id=122460 >

The presence of the code block in the second program should make no
difference. The first result (0) is correct.

$ perl5.21.3 -le '$_ =3D "rdvark"; /(?=3D[A-Za-z0-9_])a*?/g; warn pos'
0 at -e line 1.
$ perl5.21.3 -le '$_ =3D "rdvark"; /(?{})(?=3D[A-Za-z0-9_])a*?/g; warn pos'
3 at -e line 1.

That 3 should be 0, too.

A bisect points to this​:

commit 77ebeeb
Author​: Karl Williamson <public@​khwilliamson.com>
Date​: Mon Sep 30 14​:06​:23 2013 -0600

 PATCH​: \[perl \#120041\] regcomp\.c missing parens and broken STCLASS

(Sorry, I should have noticed this before the release of 5.20.)

Most importantly, I need a workaround. :-)

Not having looked into this at all, but knowing that 5.20.1 is due this
month, if this were to get into that release, would that be a good
enough workaround?

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

From @khwilliamson

Now fixed by 842a9d2
which I vote to backport into 5.20.1
--
Karl Williamson

@khwilliamson - Status changed from 'open' to 'resolved'

From @khwilliamson

Reopening so I can instead mark it as pending release
--
Karl Williamson

@khwilliamson - Status changed from 'resolved' to 'open'

@khwilliamson - Status changed from 'open' to 'pending release'

From @cpansprout

On Fri Aug 08 11​:52​:10 2014, khw wrote​:

Now fixed by 842a9d2
which I vote to backport into 5.20.1

Thank you. It gets my vote, too.

On Sat Aug 02 20​:07​:50 2014, public@​khwilliamson.com wrote​:

Not having looked into this at all, but knowing that 5.20.1 is due
this
month, if this were to get into that release, would that be a good
enough workaround?

Yes, that would be okay. However, I would still like to get my module working with 5.20.0 if possible, so I have some questions​:

$ perl5.21.3 -Mre=debug -e 'my $a = "rdvark"; $a =~ /(?=[A-Za-z0-9_])a*?/g;'
...
stclass ANYOF[0-9A-Z_a-z] minlen 0
...
$ perl5.21.3 -Mre=debug -e 'my $a = "rdvark"; $a =~ /(?{})(?=[A-Za-z0-9_])a*?/g;'
...
stclass ANYOF[a] minlen 0 with eval
...

(These one-liners are from before the fix.)

It seems that the stclass optimisation is behaving differently depending on the presence or absence of an eval. (I find that surprising. I thought the optimisation would just be turned on or off.)

What can I add to the pattern to turn off the optimisation altogether? Under what circumstances does this bug occur? (I.e., where does (?=) have to be in the pattern, etc.) Is it only with (?=...) or does it apply to other assertions, too?

--

Father Chrysostomos

From [Unknown Contact. See original ticket]

On Fri Aug 08 11​:52​:10 2014, khw wrote​:

Now fixed by 842a9d2
which I vote to backport into 5.20.1

Thank you. It gets my vote, too.

On Sat Aug 02 20​:07​:50 2014, public@​khwilliamson.com wrote​:

Not having looked into this at all, but knowing that 5.20.1 is due
this
month, if this were to get into that release, would that be a good
enough workaround?

Yes, that would be okay. However, I would still like to get my module working with 5.20.0 if possible, so I have some questions​:

$ perl5.21.3 -Mre=debug -e 'my $a = "rdvark"; $a =~ /(?=[A-Za-z0-9_])a*?/g;'
...
stclass ANYOF[0-9A-Z_a-z] minlen 0
...
$ perl5.21.3 -Mre=debug -e 'my $a = "rdvark"; $a =~ /(?{})(?=[A-Za-z0-9_])a*?/g;'
...
stclass ANYOF[a] minlen 0 with eval
...

(These one-liners are from before the fix.)

It seems that the stclass optimisation is behaving differently depending on the presence or absence of an eval. (I find that surprising. I thought the optimisation would just be turned on or off.)

What can I add to the pattern to turn off the optimisation altogether? Under what circumstances does this bug occur? (I.e., where does (?=) have to be in the pattern, etc.) Is it only with (?=...) or does it apply to other assertions, too?

--

Father Chrysostomos

From @cpansprout

On Sat Aug 09 09​:49​:38 2014, sprout wrote​:

What can I add to the pattern to turn off the optimisation altogether?
Under what circumstances does this bug occur? (I.e., where does (?=)
have to be in the pattern, etc.) Is it only with (?=...) or does it
apply to other assertions, too?

My meagre experimentation suggests that it only occurs with (?=...) and that adding (??{}) to the pattern will disable the bug. So I will use that for now....

--

Father Chrysostomos

From [Unknown Contact. See original ticket]

On Sat Aug 09 09​:49​:38 2014, sprout wrote​:

What can I add to the pattern to turn off the optimisation altogether?
Under what circumstances does this bug occur? (I.e., where does (?=)
have to be in the pattern, etc.) Is it only with (?=...) or does it
apply to other assertions, too?

My meagre experimentation suggests that it only occurs with (?=...) and that adding (??{}) to the pattern will disable the bug. So I will use that for now....

--

Father Chrysostomos

From @steve-m-hay

On 9 August 2014 17​:49, Father Chrysostomos via RT
<perlbug-comment@​perl.org> wrote​:

On Fri Aug 08 11​:52​:10 2014, khw wrote​:

Now fixed by 842a9d2
which I vote to backport into 5.20.1

Thank you. It gets my vote, too.

+1 from me too, and now cherry-picked into maint-5.20.

From @khwilliamson

Thanks for submitting this ticket

The issue should be resolved with the release today of Perl v5.22. If you find that the problem persists, feel free to reopen this ticket

--
Karl Williamson for the Perl 5 porters team

@khwilliamson - Status changed from 'pending release' to 'resolved'