Oops, I forgot it. I always localize or lexicalize variables proper to a subroutine That's why I never noticed that $_ is not implicetely localized at the entry of a subroutine contrary to what I thought.

Sadly, localizing *_ or $_ doesn't play well with reference shuffling of strings with positions.

 sub lexer {  (*_) = @_;    print $1 if m/\G(A)/gc || m/\G(B)/gc ; }  
 my $a = "AB"; lexer \$a; ; lexer \$a;
[download]

This prints "A" then "B"; If I add a local *_ or a local $_, at the entry of the lexer routine, that does not work anymore. So much for a cool trick.