They're also declared after INIT, and you can see them from there ;-) Also, moving dump_lex after the declarations like this:

Still gives you this (under 5.8):

when init top-level pad is
    HASH(0x22e150) -> $%bar = \{};
    HASH(0xb5b50) -> $%foo = \{};

when setting Attr in Foo for HASH(0xb5b50) top-level pad is

when setting Attr in Bar for HASH(0x22e150) top-level pad is
    HASH(0x22e150) -> $%bar = \{};

when runtime top-level pad is
[download]

Actually I get different results. On 5.6.1 (win32) lexicals declared after INIT are not visible from INIT.

#!/usr/bin/perl

use warnings;
use strict;

INIT {
    use Data::Dumper;
    use PadWalker qw (peek_my peek_sub);    
    print Dumper(peek_my(0));

}

my $foo = 'lee';
my $bar = "Bar";


{
    my @foo = (1..3);
}

{
    my @bar = (9..12);
}

print Dumper(peek_my(0))
__END___
$VAR1 = {};
$VAR1 = {
          '$foo
          '$bar
        };
[download]

my has both runtime and compile time effects, I wouldn't be suprised if they are on the same pad, but are removed from the padlist when the block is exited.

-Lee

"To be civilized is to deny one's nature."

Actually I get different results. On 5.6.1 (win32) lexicals declared after INIT are not visible from INIT.

My results on 5.8 match yours if I run your code. Not surprising since your code is different from mine :-)

dump_lex climbs up the callstack to the enclosing file lexical scope. So it would be the same sort of thing as doing this:

use warnings;
use strict;

INIT {
   use Data::Dumper;
   use PadWalker qw (peek_my peek_sub);    
   print Dumper(peek_my(1));
}

my $foo = 'lee';
my $bar = "Bar";

print Dumper(peek_my(0));

{ my @foo = (1..3) };
{ my @bar = (9..12) };

print Dumper(peek_my(0));
[download]

Which gives (on 5.8)

$VAR1 = {
          '$foo' => \undef,
          '@bar' => [],
          '@foo' => [],
          '$bar' => \undef
        };
$VAR1 = {
          '$foo' => \'lee',
          '$bar' => \'Bar'
        };
$VAR1 = {
          '$foo' => \'lee',
          '$bar' => \'Bar'
        };
[download]

Again, INIT can see everything, but nothing can be seen before/after the blocks are executed at the top level. I still don't understand what's happening here :-)