Seems so. Effectively, the loop variable has no relationship with any pre-existing variable with the same name:

my $i; 
for $i ( 1..10 ){ 
    1; 
}; 
print $i };;
Use of uninitialized value $i in print at (eval 9) line 1, <STDIN> lin
+e 1.
[download]

I believe it is a throw-over from pre-lexical Perl.

I've often wished (and occasionally argued in favour of), that for didn't localise pre-existing, lexical loop variables.

It makes for messy workarounds to the situation where a counting loop can exit early conditionally, and you want to know how far it counted:

my $i;
for $i ( 1 .. $N ) {
    last if <SOMECOND>;
    ...
}
## Here we have no way of knowing if we took an early exit;
## thus necessitating additional, unnecessary complexity to retain or 
+discover that.
[download]

Effectively, the loop variable has no relationship with any pre-existing variable with the same name:

Same goes for local, fyi.

>perl -E"say 0+\$x; local $x; say 0+\$x;"
4885852
2403948
[download]

I've often wished (and occasionally argued in favour of), that for didn't localise pre-existing, lexical loop variables.

It would be useful and intuitive. It would drastically change the behaviour of the following, though

for $lex (...) {
   push @f, sub { ... $lex ... };
}
[download]

As it is now, I never use a pre-existing var.

So loop "localisation" is different to the "localisation" done by local?

No, the difference you are seeing is because subs capture lexical vars but not package vars.

yes and no, since (unfortunately) you can't use local with lexicals!

But foreach tries to mimic this behaviour (a tribute to DWIM)

As you can see localization is as predicted with package vars, but seems to rebind the var to the new pad for lexical vars.

lanx@ubuntu:~$ perl
our $i;
for $i (1..9) {
  tst();
}
sub tst {
  print $i;
}
123456789
lanx@ubuntu:~$ perl
my $i=42;
for $i (1..9) {
  tst();
}
sub tst {
  print $i;
}
424242424242424242
[download]

we already had similar discussions.