Yes i have confirmed that

This is perl 5, version 20, subversion 1 (v5.20.1) built for MSWin32-x
+86-multi-thread-64int
(with 1 registered patch, see perl -V for more detail)

Copyright 1987-2014, Larry Wall

Binary build 2000 [298557] provided by ActiveState http://www.ActiveSt
+ate.com
Built Oct 15 2014 22:10:49
[download]

you do realize require is a run time operation? and it modifies the global symbol table(GST)? did you realize that Text::CSV_XS REQUIRES Exporter? I understand that modifying the GST in threads is a very BAD thing.

that said i was able to modify your code so it doesnt fail. although this stops at 1000, i ran it to 10,000 without fail. "use strict; use warnings;" caused me to notice that your $csv in $outth $inth was not declared too, again a modifcation of the GST.

#!/usr/bin/perl
use strict; use warnings; 

# if these are commented it will fail 
use Carp qw( croak );
use IO::Handle; 

#g1 set ... uncommenting these makes it run in 70% of the time (oneway
+=0)
# but it doesnt fail if left comment 
#use IO::File;
#use DynaLoader ();
#use Exporter; 


use threads;
use Thread::Queue;

my $joinset=0;  # =1 still fails 
my $oneway=0; # =1 runs in 35% of time (g1 commented)
require Text::CSV_XS if ($oneway);

#require Text::CSV_XS;

my $input = 'input.csv';
my $output = 'output.csv';
my @cols = qw(a b c d);
my $q;
my $count = 0;
my $bdt=time; 
while(1) {
        $q = Thread::Queue->new;
        my $inth = threads->create(\&inputthread);
        my $outth = threads->create(\&outputthread);
if ($joinset){
        $inth->join();
        $q->enqueue(undef);
        }
        $outth->join() or die("thread crashed?\n");
$inth->join() unless ($joinset);
        $count++;
        print "$count\n";
last if ($count>=1000);
}

print 'end:'.(time-$bdt)."\n";

sub inputthread {
require Text::CSV_XS unless ($oneway);
#        require Text::CSV_XS;
        my $csv = Text::CSV_XS->new ({ binary => 1, eol => "\n", sep_c
+har => ",", quote_char => '"', escape_char => '"', empty_is_undef => 
+0, decode_utf8 => 0 });
        $csv->column_names(@cols);
        my $fh;
        open($fh,'<',$input);
        binmode($fh);
        my @rows = ();
        while(1) {
                my $data;
                if($csv) {
                        $data = $csv->getline_hr($fh);
                        unless($data) {
                                last if($csv->eof);
                                die("Failed to parse CSV line: ".$csv-
+>error_diag."\n");
                        }
                }
                else {
                        die("Data parsed another way\n");
                }
                my @arr = values(%$data);
                push(@rows,\@arr);
        }
        close($fh);
        $q->enqueue(\@rows);
$q->enqueue(undef) unless ($joinset);
}

sub outputthread {
        my $fh;
        open($fh,'>',$output);
        binmode($fh);
        my $csv;
require Text::CSV_XS unless ($oneway);
#        require Text::CSV_XS;
        $csv = Text::CSV_XS->new ({ binary => 1, eol => "\n", sep_char
+ => ',', quote_char => '"', escape_char => '"' });
        while(1) {
                my $rows = $q->dequeue;
                last unless(defined $rows);
                foreach my $data(@$rows) {
                        if($csv) {
                                $csv->print($fh,$data);
                        }
                        else {
                                die("Data output another way\n");
                        }
                }
        }
        close($fh);
        return 1;
}
[download]

I added the g1 uses when joinset was "essentialy 1" (well missing that kinda code) and it went farther but still failed. so i thought about what happens at join, my theory is that is when "garbage collection" takes place on the now unreferenced variables in $inth. I suspect it was destroying things that $outth will still reference, maybe because they existed in the GST when some use under "require Text::CSV_XS" took place in $outth. by deferring the $inth join to after the $outth join those items were still "valid?" in the GST

i found this very interesting and decided to look further, you may want to play with these mods to get a better understanding. commenting "use croak" and "use IO::Handle" produce interesting errors.

funny my perlmonks quote here was "Clear questions and runnable codeget the best and fastest answer". I started just to see if i could even run your code..

edit:global $cst was in $inth