Dumping a filehandle leaks memory [rt.cpan.org #41199]

[toddr]

Migrated from rt.cpan.org#41199 (status was 'stalled')

Requestors:

• [email protected]

From [email protected] on 2008-11-25 03:21:22:

This program leaks memory at a high rate:

perl -le 'use YAML::Syck; while (){ open my $fh, "/etc/hosts" or die;
YAML::Syck::Dump($fh)}' > /dev/null

Let me know if you need further details.

Thanks!

From [email protected] on 2008-11-25 09:30:10:

On Mon Nov 24 22:21:22 2008, ANDK wrote:
> This program leaks memory at a high rate:
> 
> perl -le 'use YAML::Syck; while (){ open my $fh, "/etc/hosts" or die;
> YAML::Syck::Dump($fh)}' > /dev/null
> 
> Let me know if you need further details.
> 
> Thanks!

Devel::Leak shows that it's leaking one SV per iteration. The SV dump
looks like this:

new 0x823d6f0 : SV = PVIV(0x823f0c0) at 0x823d6f0
  REFCNT = 1
  FLAGS = (POK,FAKE,READONLY,pPOK)
  IV = 1529493059
  PV = 0x8239a78 "="
  CUR = 1
  LEN = 0

Checking the process memory also suggests that it's only this SV, per
iteration it's leaking about 24 to 28 bytes, depending on perl/OS.

Regards,
    Slaven

From [email protected] on 2010-05-20 11:19:43:

On Mon Nov 24 22:21:22 2008, ANDK wrote:
> This program leaks memory at a high rate:
> 
> perl -le 'use YAML::Syck; while (){ open my $fh, "/etc/hosts" or die;
> YAML::Syck::Dump($fh)}' > /dev/null
> 
> Let me know if you need further details.

A failing test would be nice, you could e.g. run it on linux only and 
check ps -o rss $$, then see if it's e.g. 2x what it used to be after a 
few thousand iterations.

See also this boring form letter I'm pasting around:

(This is a form-reply that isn't specific to your particular report)

YAML::Syck has just acquired one new maintainer (me), it still doesn't
have anyone that *cares* about it. But I'm willing to help solve your
report & release a new version with the fix if it's easy for me.

It now has a Git repository at:

    http://github.com/avar/YAML-Syck

If your report is a patch that fixes a problem, great. Please remake
the patch against Git by forking that repo and sending me a pull
request on GitHub (or an update to this bug if you prefer
git-format-patch(1) or some other repo provider..). Make sure to
include a test for what you fixed.

If your report is some code that fails (and you have a testcase for
it) a patch against the test suite to demonstrate that failure would
be very useful. It's OK if the test crashes and burns, see
Test::More's docs for how to make TODO tests that fail now, but
shouldn't. Even if it segfaults perl C<system $^X => qw/ -Mblib
-MYAML::Syck .../> or something like that and checking the return
value will do.