longmess and shortmess do include caller information. The key is how deep you are in the stack, as Carp has a set of rules for deciding how much stack trace (if any) to include. I'm not 100% sure the rules are consistent across all the calls. Here's a test program that illustrates. Maybe a monk with more Carp knowledge can comment. I need to think about it some more ... I suspect there's some nonobvious consistency here.

use strict;
use Carp qw(cluck);

sub test_errors
{
    cluck "First test message";

    my $long  = Carp::longmess("Second test message");
    my $short = Carp::longmess("Third test message");

    print STDERR $long;
    print STDERR $short;

    test2();
}

sub test2
{
    cluck "Fourth test message";

    my $long  = Carp::longmess("Fifth test message");
    my $short = Carp::longmess("Sixth test message");

    print STDERR $long;
    print STDERR $short;
}

test_errors();
[download]

produces:

First test message at cluckt line 8
        main::test_errors() called at cluckt line 30
Second test message at cluckt line 30
Third test message at cluckt line 30
Fourth test message at cluckt line 21
        main::test2() called at cluckt line 16
        main::test_errors() called at cluckt line 30
Fifth test message at cluckt line 16
        main::test_errors() called at cluckt line 30
Sixth test message at cluckt line 16
        main::test_errors() called at cluckt line 30
[download]

Basically, longmess and shortmess appear to always include one less stack frame than cluck does. Maybe someone should cluck about this. ;-)

Note that I got the same results by calling longmess and shortmess directly in the print calls rather than getting the messages into local variables first.