From @Grimy

I have some (horrifyingly golfed) code to recursively build a binary search
tree, based on an array containing the depth of the elements (as described
in RFC 1951 <http​://www.ietf.org/rfc/rfc1951.txt>, pp.6–8). It works fine
under Perl 5.8 and 5.14, but it breaks loudly under 5.16, with 15 “Attempt
to free unreferenced scalar” messages.

Here is the simplest version I could come up with that still triggers the
bug​:

sub rec {
$depth++;
return @​_ if $depth == 16;
@​_ = ((grep { $_[$_] ==

$depth } 0..$

rec(1, 1)

From @jkeenan

On Thu May 09 09​:49​:24 2013, victor.adam@​derpymail.org wrote​:

I have some (horrifyingly golfed) code to recursively build a binary
search
tree, based on an array containing the depth of the elements (as described
in RFC 1951 <http​://www.ietf.org/rfc/rfc1951.txt>, pp.6–8). It works fine
under Perl 5.8 and 5.14, but it breaks loudly under 5.16, with 15 “Attempt
to free unreferenced scalar” messages.

Here is the simplest version I could come up with that still triggers the
bug​:

sub rec {
$depth++;
return @​_ if $depth == 16;
@​_ = ((grep { $_[$_] ==

Extra close brace or missing open brace

$depth } 0..$

#_), &rec);
map { [shift, shift] } @​_
}

rec(1, 1)

I can confirm that I get this in Perl 5.16.0 but not in 5.14.2. But I
can't understand your program well enough (e.g., why are you assigning
to @​_ ?) to help further.

Thank you very much.
Jim Keenan

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

From @iabyn

On Thu, May 09, 2013 at 04​:49​:58PM -0700, James E Keenan via RT wrote​:

On Thu May 09 09​:49​:24 2013, victor.adam@​derpymail.org wrote​:

I have some (horrifyingly golfed) code to recursively build a binary
search
tree, based on an array containing the depth of the elements (as described
in RFC 1951 <http​://www.ietf.org/rfc/rfc1951.txt>, pp.6–8). It works fine
under Perl 5.8 and 5.14, but it breaks loudly under 5.16, with 15 “Attempt
to free unreferenced scalar” messages.

Here is the simplest version I could come up with that still triggers the
bug​:

sub rec {
$depth++;
return @​_ if $depth == 16;
@​_ = ((grep { $_[$_] ==

Extra close brace or missing open brace

$depth } 0..$

#_), &rec);
map { [shift, shift] } @​_
}

rec(1, 1)

The code can be simplified to​:

  @​a = (1,1,1,1);
  map { [shift @​a, shift @​a] } @​a;

i.e. elements of @​a are being shifted off, copied and then freed, while
the elements of the array are still being iterated over.

It's another example of the perl stack not being reference counted.

--
Any [programming] language that doesn't occasionally surprise the
novice will pay for it by continually surprising the expert.
  -- Larry Wall

From @Grimy

Hello,

As I said, I originally discovered the bug while golfing. Assigning to @​_
saved a few keystrokes. I just experemented further, and it turns out the
bug can be reproduced without this assignment​:

sub rec {
$depth++;
return @​_ if $depth == 16;
@​tree = ((grep { $_[$_] ==

$depth } 0..$

rec(1, 1);

Note that the block passed to map completely ignores $_. Since each
iteration consumes 2 elements of @​tree, the sensible thing to do would be
to iterate over 1..@​tree/2 —*way *too many keystrokes. However, the problem
only occurs when iterating over @​tree. I'm aware that this kind of hack is
unlikely to occur in production code.

Thank you,

Victor Adam

2013/5/10 James E Keenan via RT <perlbug-followup@​perl.org>

On Thu May 09 09​:49​:24 2013, victor.adam@​derpymail.org wrote​:

I have some (horrifyingly golfed) code to recursively build a binary
search
tree, based on an array containing the depth of the elements (as
described
in RFC 1951 <http​://www.ietf.org/rfc/rfc1951.txt>, pp.6–8). It works
fine
under Perl 5.8 and 5.14, but it breaks loudly under 5.16, with 15
“Attempt
to free unreferenced scalar” messages.

Here is the simplest version I could come up with that still triggers the
bug​:

sub rec {
$depth++;
return @​_ if $depth == 16;
@​_ = ((grep { $_[$_] ==

Extra close brace or missing open brace

$depth } 0..$

#_), &rec);
map { [shift, shift] } @​_
}

rec(1, 1)

I can confirm that I get this in Perl 5.16.0 but not in 5.14.2. But I
can't understand your program well enough (e.g., why are you assigning
to @​_ ?) to help further.

Thank you very much.
Jim Keenan

From @jkeenan

On Sat May 11 02​:02​:53 2013, victor.adam@​derpymail.org wrote​:

Hello,

As I said, I originally discovered the bug while golfing. Assigning to
@​_
saved a few keystrokes. I just experemented further, and it turns out
the
bug can be reproduced without this assignment​:

sub rec {
$depth++;
return @​_ if $depth == 16;
@​tree = ((grep { $_[$_] ==

Extra close brace or missing open brace

$depth } 0..$

#_), &rec);
map { [shift @​tree, shift @​tree] } @​tree;
}

rec(1, 1);

Note that the block passed to map completely ignores $_. Since each
iteration consumes 2 elements of @​tree, the sensible thing to do would
be
to iterate over 1..@​tree/2 —*way *too many keystrokes. However, the
problem
only occurs when iterating over @​tree. I'm aware that this kind of
hack is
unlikely to occur in production code.

Thank you,

Victor Adam

Thanks for the follow-up. As Dave Mitchell noted, this falls into a
class of known problems that do not have an easy solution. I have
marked the ticket as such.

Thank you very much.
Jim Keenan