my $pat = qr/a/;
'aaaa'  =~ /$pat/gp or die;
print "MATCH=[${^MATCH}]\n";
[download]

Dave.

Now given ticket RT #118213

Dave.

It appears to depend on qr.

my $pat = qr/a/p;
'aaaa'  =~ /$pat/g or die;
print "MATCH=[${^MATCH}]\n";
[download]

Appears to work OK on 5.18.0.

Yes, I overlooked the fact that $pat in my original example was pre-compiled with qr//.

It's possible for either or both of pos or  ${^MATCH} to be undefined after a match, which is easily demonstrated (Perl version 5.14):

>perl -wMstrict -le
"my @strs = (qw(123x foo ***), '');
 ;;
 for my $s (@strs) {
   $s =~ m{ \d+ }xmsgp;
   printf qq{for '%s': pos == %s, \${^MATCH} eq '%s' \n}, $s,
     pos($s) // 'undefined', ${^MATCH} // 'undefined',
     ;
   }
"
for '123x': pos == 3, ${^MATCH} eq '123'
for 'foo': pos == undefined, ${^MATCH} eq 'undefined'
for '***': pos == undefined, ${^MATCH} eq 'undefined'
for '': pos == undefined, ${^MATCH} eq 'undefined'
[download]

Rather than Perl, I think the most likely explanation is that the data you are processing has changed.

Update: ... and the length of an undef is undefined.

G'day nbtrap,

Welcome to the monastery.

I'm using 5.18.0. Without any context or knowledge of the values of $buf, $pat, $self or $sub, I am unable to reproduce your problem.

pos() and ${^MATCH} are returning the values I would expect:

$ perl -Mstrict -Mwarnings -E '
    my $buf = q{aXa};
    while ($buf =~ /a/gp) {
        say q{pos($buf):};
        say pos($buf);
        say q{${^MATCH}:};
        say ${^MATCH};
    }
'
pos($buf):
1
${^MATCH}:
a
pos($buf):
3
${^MATCH}:
a
[download]

perldelta (for 5.18.0) mentions a change to ${^MATCH} under Internal Changes which may be relevant.

perldoc.perl.org hasn't updated to 5.18.0 yet. If you're using online documentation, I suggest using the CPAN perl 5.18.0 doco. (The commandline perldoc utility is up-to-date.)

You may get some hints for creating a short, self-contained script that reproduces your problem from "How do I post a question effectively?".

Thank you. I'm aware of the changes documented in the 5.18 manual pages, but none of them seemed to address this problem. And I know a working script would have been very helpful, but I couldn't get this to work outside the context of my rather large program, so I decided to cross my fingers and hope someone else might be experiencing the same thing.