PRD parser expontentially slower with larger files

Hello.

I have a simple grammar (used in a POE Filter) like:

================
dictionary : '{' key_value(s /\|/) '}'
key_value : key '=' value
key : /[A-Za-z0-9_ ]+/
value : dictionary | string
string : /[^|\}]*/
================

with a valid test string like:

{reply_to={message_type=portfolio_get|private=QQQ}|portfolio_name=P|ordervalidate=OK}

which dumps like:

{
'ordervalidate' => 'OK',
'portfolio_name' => 'P',
'reply_to' => {
'message_type' => 'portfolio_get',
'private' => 'QQQ'
}
}



And a parser:

================
#!/usr/bin/perl

use strict;
use warnings;

use Parse::RecDescent;

$::RD_HINT = 1;
$::RD_ERRORS = 1;
$::RD_WARN = 1;

my $parser = new Parse::RecDescent(q{
dictionary : '{' key_value(s /\|/) '}'
{ $return = { map { $_->[0] => $_->[1] } @{$item[2]} } }
key_value : key '=' value
{ [ @item{qw(key value)} ] }
key : /[A-Za-z0-9_ ]+/
value : dictionary | string
string : /[^|\}]*/
});

while (<>) {
my $msg = $parser->dictionary($_);
# print Dumper($msg);
}
==============





However, it gets very slow with larger files. If I generate data files with adjustable size with:

==============
#!/usr/bin/perl

use strict;
use warnings;

print "{reply_to={message_type=portfolio_get|private=PP}|portfolio_name=PP|ordervalidate=OK|portfolio_positions={";

print join "|", map "portfolio_position$_={instrument_id={instrument_tag=$_|market=MMMM|feedcode=FFFF|isincode=IIII|underlying=UUUU|kind=Spot|currency=CUR|multiplier=2|assettype=Equities}|volume=$_|invested=$_|accrued=$_|fee=$_|commission=$_|change_in_volume=$_|change_in_invested=$_|change_in_accrued=$_|change_in_fee=$_|change_in_commission=$_|currency=CUR}", 1 .. shift;

print "}}|error=0}";
==============



I get execution times like

size filesiz time
10 3398 .25
20 6788 .42
50 16958 1.17
100 33920 3.29
200 69020 10.9
300 104120 24.1
400 139220 42.0
500 174320 63.8
1000 349832 264
1500 531332 605

and these are kernel user time, not wallclock - it's eating all CPU a for a long time!
And the files aren't *that* big.

How can I get it faster ? It seems already simple enough. Please help!


Sincerely.

/ Jan Sundberg