As I mentioned - this code has been running fine in a production server for 18 months or more - the previous code definitely worked.

I'm quite willing to believe there is something else in the environment that is getting in the way - I was just very surprised (and, happily, judging by ppls reactions so far, so are they) that the my() variable wasn't filled in.

And, yes, I've tested that fill_cache gets values and puts them into %cache. The problem is that the values don't hang around once fill_cache() returns. I don't know why :-)

Output from full code below is here. First set of statements is with my(), second with our(). Full apache restart just before each request.

A normal request involves hitting the cache for password then again for a list of allowed groups (in two separate Auth* handlers) and checking against the Apache require group directive. You can see that the group lookup isn't even mentioned on the successful run as it is already cached.

[18981] get_pass: not in cache - retrieving
[18981] fill_cache: setting cache [andrewo] pw:[andrewo] groups:[enabl
+ed]
[18981] get_pass: password now: andrewo
[18981] get_groups: not in cache - retrieving
[18981] fill_cache: setting cache [andrewo] pw:[andrewo] groups:[enabl
+ed]
[18981] get_groups: groups now: user undef!
[Tue Sep 20 17:15:51 2005] [error] Can't use an undefined value as an 
+ARRAY reference at
/usr/local/share/perl/5.8.4/Oriel/Apache/UserCache.pm line 56.\n

[19054] get_pass: not in cache - retrieving
[19054] fill_cache: setting cache [andrewo] pw:[andrewo] groups:[enabl
+ed]
[19054] get_pass: password now: andrewo
[download]

And the full UserCache module (with added warns):

package My::Apache::UserCache;

use strict;
use Apache::Constants qw(:common);
use My::User;
require Exporter;
our @ISA = qw(Exporter);
our @EXPORT_OK = qw(get_pass get_groups);

use constant CACHE_TIMEOUT => 60; # seconds

#
# username => { timestamp => unixtime,
#               password => blah,
#               groups => [groups], },
#
our %cache = ();

sub fill_cache {
  my $user = shift;
  my $now = time;
  my ($password, @groups) = Oriel::User->get_pw_groups($user );
  # enforce at least 1 character passwords
  return 0 unless defined $password && length($password);
  warn "[$$] fill_cache: setting cache [$user] pw:[$password] groups:[
+".join(',',@groups)."]\n";
  $cache{$user} = { timestamp => $now,
                    password => $password,
                    groups => [@groups] };
}
sub get_pass {
  my $user = shift;
  my $now = time;
  return $cache{$user}->{password}
    if ( exists $cache{$user} &&
         $cache{$user}->{timestamp} &&
         ($now - $cache{$user}->{timestamp}) < CACHE_TIMEOUT );
  warn "[$$] get_pass: not in cache - retrieving\n";
  return undef unless fill_cache( $user );
  warn "[$$] get_pass: password now: ".(defined cache{$user} ? $cache{
+$user}->{password} : 'user undef!')."\n";
  $cache{$user}->{password};
}
sub get_groups {
  my $user = shift;
  my $now = time;
  return @{$cache{$user}->{groups}}
    if ( exists $cache{$user} &&
         $cache{$user}->{timestamp} &&
         ($now - $cache{$user}->{timestamp}) < CACHE_TIMEOUT );
  warn "[$$] get_groups: not in cache - retrieving\n";
  return () unless fill_cache( $user );
  warn "[$$] get_groups: groups now: ".(defined $cache{$user} ? @{$cac
+he{$user}->{groups}} : 'user undef!')."\n";
  @{$cache{$user}->{groups}};
}

1;
[download]

Its weird. Any ideas?

OK, I've narrowed this down to something funky in Exporter. Removing all traces of Exporter from the UserCache module and calling subs by fully qualified name in the Apache Auth* handlers has completely fixed it and life has returned to normal - ie my() works as expected.

I'm going to have much more of a poke through the mod_perl mailing lists now and see if this is a manifestation of the accidental closure issue.

Thanks for helping me bounce this around. If anyone has any ideas why this behaviour happens when using Exporter as above then please please let me know :)

Cheers!

It would not be an accidental closure, since what you have here is an intentional closure. If using Exporter breaks the closure, the right thing to do is to check the Perl changelog for bits about that and if you don't see anything, make a small example and post it first on the mod_perl list and then on perl5-porters.