I do think you're onto something. It looks to me like it might be something worth reporting to Lincholn Stein (CGI::Carp's author)... or to the p5p people, since CGI::Carp is a core module. Not sure which is more appropriate.

Here are a few tests that I found interesting, and which give results that are in some ways inconsistant with documentation.

use strict;
use warnings;
use CGI::Carp;


eval { die "Ouch. 1" };
print $@, "\n" if $@;

eval { die "Ouch. 2\n" };
print $@, "\n" if $@;

eval { CORE::die "Ouch. 3" };
print $@, "\n" if $@;

eval { CORE::die "Ouch 4.\n" };
print $@, "\n" if $@;

__Output__
Ouch. 1 at C:/Perl/lib/CGI/Carp.pm line 312.
Ouch. 2
Ouch. 3 at C:\Perl\scripts\mytest.pl line 14.
Ouch. 4
[download]

And now the same tests but with Carp.pm (not CGI::Carp)...

use strict;
use warnings;
use Carp;


eval { die "Ouch. 1" };
print $@, "\n" if $@;

eval { die "Ouch. 2\n" };
print $@, "\n" if $@;

eval { CORE::die "Ouch. 3" };
print $@, "\n" if $@;

eval { CORE::die "Ouch 4.\n" };
print $@, "\n" if $@;

__OUTPUT__
Ouch. 1 at C:\Perl\scripts\mytest.pl line 8.
Ouch. 2
Ouch. 3 at C:\Perl\scripts\mytest.pl line 14.
Ouch. 4
[download]

Clearly Carp.pm's output is what we want. And CGI::Carp's output is improper. This bug appears only in scripts that are evaled. mod_perl scripts are eval'ed too, so the problem shows up there too. The POD for CGI::Carp says that CGI::Carp is a CGI-friendly equivilant to Carp. But it turns out that CGI::Carp is not as good as Carp at reporting the correct line numbers. I would consider this a bug.

As for why a "\n" character in the die string suppresses the printing of line numbers, that's a mystery for which I can't find any documentation anywhere.