Maybe you and Jenda should avoid benchmarking tools that you don't really master.

Your first attempt was perfectly valid as a solution when performance is not an issue. But if it becomes one, then loading the entire document in memory when XML::Twig is specifically designed to avoid this, is kinda lame don't you think?

The code below is probably not faster than what you have, but a least it should not use too much memory.

#!/usr/bin/perl

use strict;
use warnings;

use XML::Twig;

my $t= XML::Twig->new( keep_spaces => 1, comments => 'drop',
                       twig_handlers => { _all_ => sub { $_[0]->flush 
+}}
                     )
                ->parsefile( "test_comments.xml")
    ;
[download]

Thanks. I knew that there must be a way to print as you go, but, not ever needing that feature, couldn't pull it out of my hat (and I meant to mention that earlier, but oh well...and I think I have actually used flush() by accident when I was looking for purge() ). Ok, "wrong solution" was maybe too strong..."not the quickest" or "not if performance is a big concern" maybe would've been more appropriate. But I don't think not mastering something will ever keep me from attempting to benchmark it :-)

I would rather say wrong tool than wrong solution. Which (as mirod's response shows) doesn't mean it's not possible to use XML::Twig efectively, but rather that it was designed for a different kind of tasks. Which definitely doesn't mean the module itself is bad. Far from that and sorry if my comment sounded that way.