Hi kir,

Out of curiosity I played around with this a bit more (using Carp, Carp::Always, and Carp::Clan; I didn't get around to Devel::SimpleTrace or Acme::JavaTrace), but I discovered that none of these modules gave me enough control, there was always something off about the error messages. So I came to the same conclusion as you, that I'd have to build my own stack trace, and I found Devel::StackTrace. Here's what I came up with:

package L;
use warnings;
use strict;

use Devel::StackTrace;
our $DYING = 0;

sub include {
    my $file = shift;
    unless (my $return = do $file) {
        my $e;
        if ($@) { $e = "Couldn't parse $file: $@" }
        elsif (!defined $return) { $e = "Couldn't do $file: $!\n" }
        elsif (!$return) { $e = "Couldn't run $file\n" }
        if ($e) {
            # don't add a new stack trace if we're already dying
            die $@ if $DYING++;
            die join '', $e, map {"\t".$_->as_string."\n"}
                Devel::StackTrace->new(ignore_package=>__PACKAGE__)->f
+rames;
        }
    }
}

1;
[download]

Output:

Couldn't parse X.pm: Blammo at X.pm line 9.
    L::include('X.pm') called at B.pm line 8
    L::include('B.pm') called at A.pm line 9
    A::load_b('foo') called at A.pm line 11
    L::include('A.pm') called at test.pl line 8
[download]

(If you wanted to get a stack trace from inside the failing file, you'd have to install a handler there, too.)

Because with "do" there is no way to distinguish an error (like file not found) from the script just returning undef (is this right?).

The normal Perl convention is that files that are required/used end on a true value (often 1;). If your config files return a true value the same way, then you'll be able to distinguish those cases with the above code.

Hope this helps,
-- Hauke D