Okay... now here's the next quick shot after some trying. Have to look at the failed attempts. This really a fun challenge, zerohero++ ;-)

Since $_ is localized for map and grep, it seems the most natural thing to localize @_ for that purpose, and it seems to work... (update: heh, but it hasn't to be localized - perl does that for us)

#!perl -l
#

sub hmap (&%) {
    my $code = shift;
    @_ % 2 and die "odd number of variables for hmap";
    my @i = @_;
    my @rv;
    push @rv, $code->(@_=(shift @i,shift @i)) while @i;
    @rv;
}
sub hgrep (&%) {
    my $code = shift;
    @_ % 2 and die "odd number of variables for hgrep";
    my @i = @_;
    my @rv;
    for(my $i = 0; $i<= $#_; $i+=2) {
        local @_ = ($_[$i],$_[$i+1]);
        $code->(@_) and push @rv, @_;
    }
    @rv;
}

%h = (foo => 1, bar => 2, baz => 3);

%new =  hmap { $_[0]."l",++$_[1] } %h;

print "hmap result:";
print "$_ => $new{$_}" for keys %new;
print "original \%h:";
print "$_ => $h{$_}" for keys %h;

%new = hgrep {
    $_[0] =~ /f/
        and ($_[0].="l") 
            and $_[1] =~ /1/ 
                and $_[1] += 41;
} %h;

print "hgrep result:";
print "$_ => $new{$_}" for keys %new;
print "original \%h:";
print "$_ => $h{$_}" for keys %h;
__END__
hmap result:
bazl => 4
barl => 3
fool => 2
original %h:
bar => 2
baz => 3
foo => 1
hgrep result:
fool => 42
original %h:
bar => 2
baz => 3
foo => 1
[download]

No package vars introduced. But it looks like in hgrep localizing @_ is necessary, however in hmap it isn't. Why?