No, sorry -- that is not good enough either. Given my test harness,

my @list = (
     a => [ 'b', 'c', 'e' ],
     b => [ 'd' ],
     c => [ 'b', 'd' ],
     f => [ 'a' ],
);
print join (',', map { "'$_'" } f2(@list)),"\n";
[download]

... your code produces 'd','b','c','a','f','e', which fails to reflect that 'a' depends on 'e'.

perl's sort() requires more information. It first compares the "f" and "e" elements -- and you have not produced for sort() any way for it to know which of these comes first. Your sort function returns 0 for $a="f" and $b="e", and perl's sort() arbitrarily places "f" before "e" -- wich is the wrong choice, since "e" must come before "a" must come before "f".

So the sort() fails in the first test, and never recovers. You will have to implement something like this yourself in order to solve this -- perl's sort(), and indeed Perl's sort(), does not.

One way to reveal your delusions to yourself before posting them here, is to add warnings to your sort functions (while testing) -- like so, for example:

sub f {
%r=@_;map{$m{$_}={map{$m{$_}||={};$_=>1}@{$r{$_}}}}keys%r;
reverse sort{my$t=$m{$b}{$a}-$m{$a}{$b};warn"$a,$b:$t";$t}keys%m;
}
[download]

Now with your test harness I see these warnings (plus script name and line number ...):

f,e:0 at ...
a,f:0 at ...
b,a:1 at ...
c,b:1 at ...
d,c:1 at ...
[download]

And so you can see that it is rather undefined how perl sorts "a" and "e" -- which is a poor sign, since "e" depends on "a" (this is with your test harness, remember). It is just dumb luck that it gets it "right".

And yes, your problem has set me pondering old tricks with graph theory. I have yet to produce anything short that actually works.

The Sidhekin
print "Just another Perl ${\(trickster and hacker)},"