I suspect (but don't know for sure) that your "bombs my machine" problem is due to the cyclic nature of: run script ... load module ... run B::Xref ... run script ... load module ... run B::Xref ... ad infinitum.

There's some other (potential) issues, which include: you're currently finding all modules using @ARGV, not just the ones loaded before your module; splitting the output from B::Xref on whitespace when pathnames may contain whitespace; portability of /dev/null; assuming modules have a .pm extension.

I created module PM::1082204::Log::Reproducible in ~/tmp/PM/1082204/Log/. This is intended to mirror your Log::Reproducible. Here's some notes:

• I've used a BEGIN block, instead of a subroutine called from import(), so that there's no reliance on import() being called and the @ARGV checks are done first.
• I've taken the code from the beginning of the calling script (up to, but not including, the use Log::Reproducible) and saved to a File::Temp temporary file. B::Xref operates on this temporary file which fixes the cyclic/infinite loop issue mentioned above.
• I've used IPC::Open3 to deal with the /dev/null issue: STDERR output is accessible from <CERR> but that filehandle isn't read.
• B::Xref discards a leading './' in the filename path but retains instances of embedded '/./'. I've handled this (e.g. /^(?:\.[\\\/]?)?(.*)$/) but there may be better ways to do this.
• There's quite a few tweaks I could envisage you making to this; the code here should at least provide a framework as a starting point.

Here's the PM::1082204::Log::Reproducible code:

package PM::1082204::Log::Reproducible;

use 5.010;
use strict;
use warnings;
use autodie;

use File::Temp ();
use IPC::Open3;

BEGIN {
    my $code = do { open my $fh, '<', $0; local $/; <$fh> };
    my ($code_to_test) = $code =~ /(\A .*?) use \s+ @{[__PACKAGE__]}/s
+x;
    my ($temp_fh, $temp_filename) = File::Temp::tempfile();
    print $temp_fh $code_to_test;

    local(*CIN, *COUT, *CERR);
    my $cmd = "$^X -MO=Xref,-r $temp_filename";
    my $pid = open3(\*CIN, \*COUT, \*CERR, $cmd);

    my $re = '(?:' . join('|' => map { /^(?:\.[\\\/]?)?(.*)$/; "\Q$1" 
+} @INC)
           . ')[\\\/]?(\S+?)(?:\.\S+)?\s';
    my %argv_modules;

    for (<COUT>) {
        next unless /\@\s+ARGV/;
        (my $module) = /$re/;
        $module =~ s{[\\\/]}{::}g;
        ++$argv_modules{$module};
    }

    waitpid $pid, 0;

    my @warn_modules = sort keys %argv_modules;

    if (@warn_modules) {
        warn "WARNING:\n",
            "Modules using'\@ARGV' before " . __PACKAGE__ . " loaded:\
+n";
        warn "\t$_\n" for @warn_modules;
    }
}

1;
[download]

In addition, the Qwerty module, in ~/tmp/PM/1082204/Log/X X/, tests for pathnames with whitespace:

package Qwerty;

use 5.010;
use strict;
use warnings;

sub xxx {
    say "ARGV not empty." if @ARGV;
}

1;
[download]

And, the ModuleWithoutPmExtension, in ~/tmp/, tests pretty much what it says:

package ModuleWithoutPmExtension;

sub test {
    print "I'm ", __PACKAGE__, "\n";
    print "\@ARGV has elements.\n" if @ARGV;
}

1;
[download]

I ran the tests from ~/tmp. Here's pm_1082204_b_xref.pl:

#!/usr/bin/env perl

use strict;
use warnings;

BEGIN {
    require 'ModuleWithoutPmExtension';
    ModuleWithoutPmExtension::test();
}

use lib './../tmp/./PM/1082204/Log/X X';
use Qwerty;

{
    use Getopt::Long;

    use PM::1082204::Log::Reproducible;
}

use Getopt::Std;
[download]

Output:

I'm ModuleWithoutPmExtension
WARNING:
Modules using'@ARGV' before PM::1082204::Log::Reproducible loaded:
    Getopt::Long
    ModuleWithoutPmExtension
    Qwerty
[download]

Getopt::Std isn't in the list. If you load it earlier in the code, it does appear in the list.

Hey Ken,
Thanks so much for this great and thorough suggestion. I incorporated your code into v0.8.2 of Log::Reproducible to check for potential conflicts along side another subroutine that checks for known conflicts (the list of which is currently empty, but I'm sure there are some out there). I really appreciate your contribution!

Thanks again,
Mike

"Thanks so much for this great and thorough suggestion. I incorporated your code into ... I really appreciate your contribution!"

You're welcome; and thanks for the attribution.

-- Ken