SPUG: object refs in Perl 5.6.1 vs 5.8.4

Andrew Sweger andrew at sweger.net
Fri May 12 18:48:41 PDT 2006

I feel like I'm going nuts. Please show me how nuts I am. I have some code
that runs fine in development and promptly goes whacky in production.
First, the environments:

    Debian 3.1 (sarge)
      w/ Linux 2.4.27
    Perl 5.8.4
    Apache 1.3.33
    mod_perl (via libapache-mod-perl)

    Red Hat Enterprise Linux ES release 2.1 (Panama) (no choice)
      w/ linux 2.3.9-e.65smp (no choice)
    Perl 5.6.1 (no choice)
    Apache 1.3.27-10.ent (no choice)
    mod_perl 1.26-2 (no choice)

I cannot duplicate the production environment in development. Nor would I
do so except under threat of violent torture.

My code is a CGI::Application using HTML::Template. The problem only
appears when running under Apache/mod_perl(Apache::Registry) in
production. I do clean restarts of the Apache server after making each
code change (and sometimes complete stop/start cycles). The problem does
not appear when running the code from the cammand line. It all runs fine
in the dev environment.

It seems that object methods that are supposed to return objects from
another class (or just a simple hash ref) are suddenly just returning a
scalar string containing the name of the thing I was trying to get back.

Does that ring any bells yet?

For example, in my application's cgiapp_postrun callback (automatically
hooked by CGI::Application), I ask for $self->session() expecting a
hash-ref or undef, I get back q{session}. I ask for
$self->param('affiliate') expecting a My::Affiliate blessed object or
undef), I get back q{affiliate}. These later blow up when I attempt to
call further methods or de-refs on (for example) $sess->{'status'} which
dies complaining I tried to use a string as a hash ref while under the
influence of "strict refs" (strict is on *everywhere* in my code, except
where it specifically is not, and this ain't one of 'em).

    my $sess = $self->session();
    my $affiliate = $self->param('affiliate');

I'm dying for someone to say, "Ah, Andrew. You simpleton. Have you already
forgotten ___________________?"

Somebody at least say, "Wow, dude. That sure is weird." I'd almost swear
there's nothing in the code path that could be returning those dumb

Andrew B. Sweger -- The great thing about multitasking is that several
                                things can go wrong at once.

More information about the spug-list mailing list