As of now, perl 5.10.0, Tk 804.028, even this code leaks:

perl -e 'use Tk;use Tk::Button;my $top = new MainWindow;for (1..100) {
  for (1..1000) {
    $top->Button(-text=>"foo")->destroy;
  }
  sleep 1;
}
'
[download]

There's no function in there, so it supports what I guessed: It's got nothing to do with closures.

Another danger of the non-list approach is that the function sees the lexical context of its definition and can accidentally use something from there

No, they don't accidentally capture anything. Closures only capture the variables they need.

The following shows that $x wasn't captured even though it was visible to the capture.

use strict;
use warnings;

printf "x=%s\n",
   do {
      my $x = 2;
      sub { $x }
   }->() // 'undef';

printf "x=%s\n",
   do {
      my $x = 2;
      sub { eval '$x' }
   }->() // 'undef';
[download]

x=2
Variable "$x" is not available at (eval 1) line 2.
x=undef
[download]

The warning is new, but the behaviour is ancient.

I perhaps do not use English conjunctions (or other words) in the right way, for which I apologize. But by accidental leaking I meant an error made by the programmer (like missing "my" inside of the annonymous sub), not by perl. And the example of the behaviour of Button was given to show why I am no longer able to reproduce the tests we did.

Hi,

maybe you could check this using the latest devel release Tk-804.028_502. Your code should not leak - and I do not see it leaking using 804.028_501.

Cheers, Christoph