Re^13: Memory leak question

[BrowserUk]

in reply to Re^12: Memory leak question
in thread Memory leak question

I believe you are being bitten by regex engine leaks.

Here's what I discovered.

1. If I replace _iso8601rx() with the bare minimum to parse the date/time in the test, the memory leaks disappear completely.

   my %cache;
   sub _iso8601_rx {
      my($self,$rx) = @_;
      my $dmt       = $$self{'tz'};
      my $dmb       = $$dmt{'base'};
   
      return $cache{ $rx } if exists $cache{ $rx };
   }
   
   $cache{cdate} = '(?<y>\d\d\d\d)-(?<m>\d\d)-(?<d>\d\d)';
   $cache{ctime} = '(?<h>\d\d):(?<mn>\d\d):(?<s>\d\d)';
   $cache{fulldate} = "$cache{cdate}\\s+$cache{ctime}";
   
   1;
   

2. However, if I change that to using the fully expanded regexes, it goes back to leaking like a sieve:
   <Reveal this spoiler or all in this thread>

I thought that it was maybe the use of (so many) named captures, but I tried very hard to make them leak. A single regex with 175,000 named captures; matching /g against a string that contained 10,000 matches for them; in a (v.slow) loop. It grew very arge, but once it maxed out, it didn't leak at all.

So then I remembered that I'd seen the regex trie optimisation caused problems with large alternations, but disabling it didn't change things.

Then I thought to try your monster regexes in a standalone script and run them directly on the sample date in a loop:

#! perl 
use strict;
my %cache = ( ctime => <<'RXA', cdtate => <<'RXB', fulldate -> <<'RXC'
+ );

##... monster regex initialisation ellided;

my $refull  = qr[$cache{ fulldate }]x;
my $rectime = qr[$cache{ ctime }]x;
my $recdate = qr[$cache{ cdate }]x;

for (1..100e6) {
    "2010-02-01 01:02:03" =~ $refull;
    "2010-02-01 01:02:03" =~ $rectime;
    "2010-02-01 01:02:03" =~ $recdate;
}

it doesn't leak at all. Not a jot.

So, it's not just the monster regexes, but also how they're are being used, or the results are being used that triggers the leak.

I'm kinda stuck for a direction in which to go now, but I hope that this will help you zero in on the cause. I'll keep looking.

[SBECK]

I think you just became the first person in the world who knows Date::Manip parsing internals better than myself!

Thanks for everything. I think this gives me enough information to track it down, though it's going to take me some time to really digest everything you said, but I think that this is enough to get me a long ways towards fixing the problem.

[SBECK]

I just found out a little bit about the leak.

Using the original Date::Manip code, there's a line in the _parse_datetime_iso8601 function which looks like:

   ($y,$md,$d,...) = @+{qw(y m d ...)};

where I just matched on the regexp from _iso8601_rx. If I comment this line out (and just set $y,$m,$d to some static values), there's no leaking. Note that I STILL match the regexp, I simply never refer to the %+ hash.

Unfortunately, I wasn't able to reproduce this in a simple test script, so I still need to investigate further, but I think this is an interesting result.

[SBECK]

I was able to reproduce the leak in a trivial script, and I think that I'm down to the most basic illustration.

$a   = '(?<a>\d)';
$b   = '(?<b>\d)';

$rx  = qr/(?:${a}${b}|${a}:${b})/;

#$string = "12";
$string = "1:2";

while (1) {
   $string =~ $rx;
   @tmp = @+{qw(a b)};
}

This leaks.

If I modify $rx to include only one of the two choices, it doesn't leak. If I plug in a string which matches the first option (i.e. use the $string = "12" line), it doesn't leak. And if you comment out the @tmp = @+ line so you don't access %+, it doesn't leak.

At this point, I guess I no longer believe that it is a Date::Manip problem. In other words, I don't think the above script is buggy... I think it points out a bug in perl itself. If you agree, I think I'll pass it on as a perl bug.

Final (I hope) comment?

[BrowserUk]

If you agree, I think I'll pass it on as a perl bug.

Yes, I absolutely agree. And your example demos the bug perfectly.

Nice to know my instincts weren't too far off--I always suspect new features first. But try as hard as I might I couldn't arrive at the simple demo that leaked. Congratulations on that.

The downside is you'll have to wait a while for the fix, but at least you now know.

[BrowserUk]

One final thought :)

There is another way of doing "named captures", without using the construct or %+ or %-. It's a bit um ... obscure, and may be slower, but it might allow you to workaround the problem in the interim without radically altering your existing code.

#! perl -slw
use strict;
use re qw[ eval ];
use Data::Dump qw[ pp ];


my $reY =  '(\d{4})(?{ $match{ y }  = $^N })';;
my $reM =  '(\d{2})(?{ $match{ m }  = $^N })';;
my $reD =  '(\d{2})(?{ $match{ d }  = $^N })';;
my $reH =  '(\d{2})(?{ $match{ h }  = $^N })';;
my $reMN = '(\d{2})(?{ $match{ mn } = $^N })';;
my $reS =  '(\d{2})(?{ $match{ s }  = $^N })';;

my $reDT = "$reY-$reM-$reD\\s+$reH:$reMN:$reS";

our %match = ();

'2010-10-06 20:55:31' =~ $reDT;

pp \%match;;

__END__
c:\test>junk57.pl
{ d => "06", h => 20, "m" => 10, mn => 55, "s" => 31, "y" => 2010 }

Or better still, cut out the middleman and put the captures straight into the names variables themselves (I wish named captures worked this way full stop) :

#! perl -slw
use strict;
use re qw[ eval ];
use Data::Dump qw[ pp ];


my $reY =  '(\d{4})(?{ $y  = $^N })';;
my $reM =  '(\d{2})(?{ $m  = $^N })';;
my $reD =  '(\d{2})(?{ $d  = $^N })';;
my $reH =  '(\d{2})(?{ $h  = $^N })';;
my $reMN = '(\d{2})(?{ $mn = $^N })';;
my $reS =  '(\d{2})(?{ $s  = $^N })';;

my $reDT = "$reY-$reM-$reD\\s+$reH:$reMN:$reS";

local our( $y, $m, $d, $h, $mn, $s );

'2010-10-06 20:55:31' =~ $reDT;

print "$y, $m, $d, $h, $mn, $s";

__END__
c:\test>junk57.pl
2010, 10, 06, 20, 55, 31

Note: The variables referenced inside the (?{ code }) blocks have to be global, but judicious use of local and our makes it reasonably convenient. Also, I've had iffy results using qr// with this. Never really understood why.

I realise that it would be considerable work to modify all your regex generators to use this method, but hey!

You can always knock up a few regex to do it for you ;)