Re: improving the Scalars leaked debugging technique?

Stas Bekman Fri, 04 Mar 2005 15:23:33 -0800

Stas Bekman wrote:
[...]

Well Devel::Leak (in my CPAN directory) is what did when doing Tk. It works by recording all the SVs in use at one point and then comparing them at a later point. It can do sv_dump()s of the new ones.

Thanks for the reminder Nick, I forgot about this tool. but as you have you need to have a clue of where it comes from :( I'll keep on looking.

It doesn't seem to help in case where the leak happens in the ithread. You can't bracket with Devel::Leak::NoteSV and Devel::Leak::CheckSV around a thread, since both need to run inside that thread, but it's too late to bracket at the start and too late to bracket on exit.

You can for example take the ticket [perl #34341], where we already know that $0 leaks when the thread exists. Let's try to bracket around the thread:

#!/usr/bin/perl -T
use Devel::Leak;
use threads;

local $0 = "test"; # <== XXX: leaks scalar
Devel::Leak::NoteSV($handle);
my $thr = threads->new(sub { 1 });
$thr->join;        # <== XXX: triggers the leak
Devel::Leak::CheckSV($handle);


% perl-5.8.7-ithread -T /tmp/leak
new 0x80e1598 : SV = IV(0x80de84c) at 0x80e1598
  REFCNT = 1
  FLAGS = (IOK,pIOK)
  IV = 134659032
new 0x80e15a4 : SV = PVMG(0x808a1d0) at 0x80e15a4
  REFCNT = 1
  FLAGS = (OBJECT,GMG,READONLY,pIOK)
  IV = 135194992
  NV = 0
  PV = 0
  MAGIC = 0x813dbe0
    MG_VIRTUAL = 0x4038ab54
    MG_TYPE = PERL_MAGIC_shared_scalar(n)
    MG_FLAGS = 0x10
    MG_PTR = 0x80ee970 ""
  STASH = 0x80664a8     "threads"
old (1):
 0 SV = UNKNOWN(0xff) (0x804cd0c) at 0x80e1490
  REFCNT = 0
  FLAGS = ()
Scalars leaked: 1
leaked: sv=0x8147b04 flags=0x084046007 refcnt=0, Perl interpreter: 0x80ee9b8

No help whatsover, since the leak happens inside the thread.

Also what does the following means? (output of CheckSV)
old (1):
 0 SV = UNKNOWN(0xff) (0x804cd0c) at 0x80e1490
  REFCNT = 0
  FLAGS = ()

does it mean that it has found some bogus scalar, or it just couldn't figure out what's the scalar?

Should there be a mechanism allowing us to register at the beginning of the thread and its exit? I suppose CLONE may help with the former (in case it's run before $0 is cloned), but I'm not sure if something similar can be done on exit. I've tried something like the following, trying to get as close as possible to the start/end of the thread.

#!/usr/bin/perl -T
use Devel::Leak;
use threads;
use threads::shared;

my $handle : shared;

package Foo;
sub CLONE { Devel::Leak::NoteSV($handle) }
sub new { bless {}, __PACKAGE__ }
sub DESTROY { Devel::Leak::CheckSV($handle) }

package main;
local $0 = 'mememe'; # <== XXX: leaks scalar
my $thr = threads->new(sub { Foo->new; 1 });
$thr->join;        # <== XXX: triggers the leak

but it's not good enough, Devel::Leak::NoteSV runs after the leaked $0 was cloned. So at the moment there are no tools that I can use to trace bugs, which aren't in my code.

--
__________________________________________________________________
Stas Bekman            JAm_pH ------> Just Another mod_perl Hacker
http://stason.org/     mod_perl Guide ---> http://perl.apache.org
mailto:[EMAIL PROTECTED] http://use.perl.org http://apacheweek.com
http://modperlbook.org http://apache.org   http://ticketmaster.com

Re: improving the Scalars leaked debugging technique?

Reply via email to