Front page | perl.perl5.porters |
Postings from August 2011
On Wed Aug 31 09:16:37 2011, sprout wrote:
> On Tue Aug 30 09:46:32 2011, nicholas wrote:
> > $ ./perl -Ilib dist/threads-shared/t/clone.t
> > 1..34
> > ok 1 - Loaded
> > ok 2 - number
> > ok 3 - string
> > ok 4 - 1 arg
> > ok 5 - cloned string
> > ok 6 - cloned scalar ref
> > ok 7 - cloned scalar ref in thread
> > ok 8 - Circular ref typ
> > ok 9 - Circular ref
> > ok 10 - Circular ref in thread
> > Attempt to free unreferenced scalar: SV 0x91eeb0, Perl interpreter:
> > 0x973840 at dist/threads-shared/t/clone.t line 86.
> > Attempt to free unreferenced scalar: SV 0x91eec8, Perl interpreter:
> > 0x973840 at dist/threads-shared/t/clone.t line 86.
> > Attempt to free unreferenced scalar: SV 0x8d15b8, Perl interpreter:
> > 0x973840 at dist/threads-shared/t/clone.t line 86.
> > ok 11 - Cloned circular refs from thread
> > ok 12 - Cloned array
> > ok 13 - Clone mod
> > ok 14 - Original array
> > ok 15 - Clone mod in thread
> > Attempt to free unreferenced scalar: SV 0x902d18, Perl interpreter:
> > 0x74ce60 at dist/threads-shared/t/clone.t line 108.
> > Attempt to free unreferenced scalar: SV 0x902d30, Perl interpreter:
> > 0x74ce60 at dist/threads-shared/t/clone.t line 108.
> > Attempt to free unreferenced scalar: SV 0x977fd8, Perl interpreter:
> > 0x74ce60 at dist/threads-shared/t/clone.t line 108.
> > ok 16 - Clone mod from thread
> >
> > [etc, no more warnings]
> >
> > The test case resists most attempts to simplify it - removing
> > seemingly
> > unrelated earlier blocks from the test case removes the warning.
>
> I’ve reduced it to this:
>
> use threads;
> use threads::shared;
>
> use Devel::Peek;
> Dump *DB::args;
>
> {
> my %hsh = ('' => undef);
> eval {
> shared_clone(%hsh);
> };
> }
>
> my $bar :shared = shared_clone($x);
>
> #{
> my $foo :shared;
> $foo = shared_clone(\$foo);
> threads->create(sub {
> Dump *DB::args;
> undef $foo
> })->join();
>
> #}
> __END__
>
> Those Dump statements I’m using for debugging.
>
> So far I’ve ascertained that:
> 1) The first unreferenced scalar is an entry in @DB::args that has
> already been used as an AV by the time the thread is started.
> 2) It gets cloned by sv_dup (not _inc) when the thread starts, so it
> ends up in param->unreferenced, and hence on the temps stack in the
thread.
> 3) When the thread exits, cv_undef ends up freeing it, because it is
> somehow reused as a padlist, even though it’s on the temps stack still
> with a refcount of one. I don’t know which cv it is.
> 4) When the temps are freed, we get ‘Attempt to freed unreferenced
scalar’.
>
> Any ideas where to look now?
>
Aha!
In pad.c:Perl_padlist_dup:
/* look for it in the table first.
I *think* that it shouldn't be possible to find it there.
Well, except for how Perl_sv_compile_2op() "works" :-( */
dstpad = (AV*)ptr_table_fetch(PL_ptr_table, srcpad);
if (dstpad)
return dstpad;
No SvREFCNT_inc!
commit 6de654a5795b6f7915432ff16bcdac0688492a9b
Author: Nicholas Clark <nick@ccl4.org>
Date: Thu Feb 25 14:21:18 2010 +0000
In Perl_padlist_dup() don't duplicate @_ or pads caused by recursion.
CvDEPTH() is 0 in a new thread, so duplicating pads beyond the
always-present
first level is a waste of effort and memory.
Adding SvREFCNT_inc makes one warning go away.
Thread Previous
|
Thread Next