pos SCALAR
       pos     Returns the offset of where the last "m//g" search
               left off for the variable in question ($_ is used
               when the variable is not specified).  May be modi-
               fied to change that offset.  Such modification
               will also influence the "\G" zero-width assertion
               in regular expressions.  See perlre and perlop.

[download]

The error

panic: sv_pos_b2u: bad byte offset at [...].

[download]

-Mark

while (m/\[\[\w+\]\]/) {
    my $oldpos = pos;
    m/\[\[(\w+)\]\]/;

    my $expansion = $helper->($1);
    pos = $oldpos;
    s/[[$1]]/$expansion/;
}
[download]

Grr, that'll teach me to make changes without testing. The substitution needs to have those square brackets escaped:

while (m/\[\[\w+\]\]/) {
    my $oldpos = pos;
    m/\[\[(\w+)\]\]/;
    
    my $expansion = $helper->($1);
    pos = $oldpos;
    s/\[\[$1\]\]/$expansion/;
}
[download]

Off the top of my head, you might try introducing some interpolation to see if that would make a difference:

s/\[\[(\w+)\]\]/$helper->("$1")/eg;

s/\[\[(\w+)\]\]/"@{[$helper->($1)]}"/eg;

s/\[\[(\w+)\]\]/my $match = $1; $helper->($match)/eg;

You're right - a simple test case doesn't have any problems. It turns out that our (I'm working with v_thunder on this project) helper function was trampling over $_, which caused the failed assertion only in combination with some other buggy code.

Rewriting the helper to use a different local variable makes the original code work as expected.

I'm having trouble producing a simple case that shows the problem. I think we just happened to hit a case where buggy code happened to work on earlier perls, but doesn't in 5.8. :)

This code works fine for me.

#!/usr/local/bin/perl
 
require 5.008;
use strict;
my $helper=sub { print @_ };
$_="[[This]] [[is]] [[a]] [[test]]";
s/(\w+)/$helper->($1)/eg;
[download]

Hello again,

After some testing, I am even more baffled than before. After attempting to create some small test cases, I'm sure there is a perl bug lurking here. Here are our (teichman's and my) findings:

This code will break:

#!/usr/bin/perl -w

$data = { key1 => 'value1',
      key2 => 'value2', };

sub macro_replace {
    $_ = shift;

    my $helper = sub {
    $_ = shift;
    return $data->{$_} if $data->{$_};
    return undef;
    };

    chomp (my $pwd = `pwd`);
    s/\[\[(pwd)\]\]/$pwd/g;
    s/\[\[(\w+)\]\]/@{[$helper->($1)]}/g;
    return $_;
}

print macro_replace ("blah [[key1]] [[key2]] [[pwd]] blah blah\n");
[download]

Note the use of $_ in the helper function. Changing this to use another variable makes it all work as expected, as teichman said.

Strangely, however, taking out the pwd substitution (i.e., commenting out lines 15 and 16) seems to stop it from crashing as well, but only in this test case. I haven't been able to reproduce the exact behavior we see in our program yet.

This is being tested on Red Hat Linux 8.0 and SuSE Linux 8.1, both of which come with perl 5.8.0 (+/- a few patches of their own--which I believe are all unrelated to this).

Thanks for the useful tips!

-Dan

I've a similar problem on RH8. I'm trying to port a a running application, and the xml parser is returning the same panic error when my handler returns. I tried unsuccessfully to use the variable substitution you mentioned, so I was wondering if you're any futher along.

s{ <\?nm(             # opening angle bracket
    (?:   # Non-backreffing grouping paren
    [^>'"] *  # 0 or more things that are neither > nor ' nor "
    |     # or else
    ".*?" |     # a section between double quotes (stingy match)
    |     # or else
    '.*?' |     # a section between single quotes (stingy match)
    ) +              # repetire ad libitum
    #  hm.... are null tags <> legal? XXX 
    )\?>          # closing angle bracket
}{my $var = $1; my_handler($r, $var)}geisx;  # mutate

[download]