From @atoomic

This clearly looks like a bug to me, but this might be a limitation I'm not aware of, as I'm able to reproduce the same issue with perl 5.14.4 or perl 5.22.0

Running this program shows that $x destruction is not happening when expecting,

Note that removing the INIT block or using an eval quote inside the INIT block avoid the issue
and does not increase the REFCNT.

# test.pl ____________
#!./perl

use feature q/say/;
use B ();

my $ok;

sub X​::DESTROY {
  say 'Destroyed @​'. ${^GLOBAL_PHASE};
  $ok = 1;
}

{
  my $x;
  BEGIN { $x = 42 }
  # no problem when using eval quote or removing the INIT block
  # INIT { eval q/$x = 43/ }
  INIT { $x = 43 }
  $x = bless {}, "X";
  say "x REFCNT​: ", B​::svref_2object( \$x )->REFCNT;
  undef($x) if grep { /force/ } @​ARGV;
}

say "x REFCNT​: ", eval { B​::svref_2object( \$x )->REFCNT };

say $ok ? "ok" : "not ok";

# ________________________________________

perl test.pl
x REFCNT​: 3
x REFCNT​: 2
not ok
Destroyed @​RUN
perl test.pl --force
x REFCNT​: 3
Destroyed @​RUN
x REFCNT​: 2
ok

From @atoomic

test.pl

From @ysth

On Tue, Nov 3, 2015 at 9​:33 AM, Atoomic <perlbug-followup@​perl.org> wrote​:

# New Ticket Created by Atoomic
# Please include the string​: [perl #126556]
# in the subject line of all future correspondence about this issue.
# <URL​: https://rt-archive.perl.org/perl5/Ticket/Display.html?id=126556 >

This clearly looks like a bug to me, but this might be a limitation I'm not aware of, as I'm able to reproduce the same issue with perl 5.14.4 or perl 5.22.0

Running this program shows that $x destruction is not happening when expecting,

Note that removing the INIT block or using an eval quote inside the INIT block avoid the issue
and does not increase the REFCNT.

# test.pl ____________
#!./perl

use feature q/say/;
use B ();

my $ok;

sub X​::DESTROY {
say 'Destroyed @​'. ${^GLOBAL_PHASE};
$ok = 1;
}

{
my $x;
BEGIN { $x = 42 }
# no problem when using eval quote or removing the INIT block
# INIT { eval q/$x = 43/ }
INIT { $x = 43 }
$x = bless {}, "X";
say "x REFCNT​: ", B​::svref_2object( \$x )->REFCNT;
undef($x) if grep { /force/ } @​ARGV;
}

say "x REFCNT​: ", eval { B​::svref_2object( \$x )->REFCNT };

Note that this is a differerent (and non-lexical) $x.

say $ok ? "ok" : "not ok";

The (non-eval) INIT block is a closure on $x, so holds a reference to it.

Note that perlmod specifically says with respect to BEGIN blocks​:
"Once a BEGIN has run, it is immediately undefined and any code it
used is returned to Perl's memory pool." but no such thing happens
with INIT blocks.

So I'd have to say this is a limitation, not a bug. But that doesn't
mean it is a desired one :)

The RT System itself - Status changed from 'new' to 'open'

From @toddr

On Tue Nov 03 09​:49​:28 2015, sthoenna@​gmail.com wrote​:

The (non-eval) INIT block is a closure on $x, so holds a reference to
it.

Note that perlmod specifically says with respect to BEGIN blocks​:
"Once a BEGIN has run, it is immediately undefined and any code it
used is returned to Perl's memory pool." but no such thing happens
with INIT blocks.

So I'd have to say this is a limitation, not a bug. But that doesn't
mean it is a desired one :)

So to rephrase what we're saying here​: We're saying that any variable used in an INIT block like this either needs to be weakened or will never be destroyed until global destruction? I'm inclined to say that's a problem. At the least it should be documented right?