As I already admitted last is really an unexpected problem... 8(

(Hmm ... maybe there's a way by trying to catch the destruction of the iterator variable... )

Talking about positional subs and leaking I already found a nice workaround by using code-blocks!¹ 8)

sub noleak (&$) {
  my $cr=shift;
  $_[0]=$cr
}
 
my @a;
for (0..9) {
  noleak {0,5} my $a;
  push @a,$a;
  print $a
}
[download]

The code-reference seems - as expected - to be precalculated at compile time and is fix!

 
/usr/bin/perl -w /home/lanx/B/PL/PM/iter_leak.pl 
CODE(0xa099b48)CODE(0xa099b48)CODE(0xa099b48)CODE(0xa099b48)CODE(0xa09
+9b48)CODE(0xa099b48)CODE(0xa099b48)CODE(0xa099b48)CODE(0xa099b48)CODE
+(0xa099b48)
[download]

Or do you have an idea to break this approach?

UPDATE:

ARGHH it only works when using fixed values...

sub noleak (&$) {
  my $cr=shift;
  $_[0]=$cr
}
 
my ($start,$end)=(1,20);
my @a;
for (0..9) {
  $start++;
  noleak {$start,$end} my $a;
  push @a,$a;
  print $a
}
[download]

output

CODE(0x8e29760)CODE(0x8e47ca8)CODE(0x8e47d08)CODE(0x8e47d68)CODE(0x8e4
+7dc8)CODE(0x8e47e28)CODE(0x8e47e88)CODE(0x8e47ee8)CODE(0x8e47f48)CODE
+(0x8e47fa8)
[download]

... but commenting out the push produces:

CODE(0x81f7760)CODE(0x81f7760)CODE(0x81f7760)CODE(0x81f7760)CODE(0x81f
+7760)CODE(0x81f7760)CODE(0x81f7760)CODE(0x81f7760)CODE(0x81f7760)CODE
+(0x81f7760)
[download]

and there is no need to propagate the coderef outside the sub...

UPDATE

¹) I ran some tests, better forget it...it really only seems to work reliably with constant values in the code-block ...seems to be an optimization issue.

UPDATE I'm going to run some tests with state.