Actually I have a Logitech Slimserver / Squeezeserver (or whatever is its present name). Everytime I play a song and don't fast-forward it to the next one, the server increases that song's "popularity". If I want to listen to the "most popular songs" or even "most popular songs not recently played" (or any other combination of parameters attached to each song), the server goes out to the database and finds them by using some standard SQL.

For perl, there is Array::Heap, which ought to perform adequately.
This following does 1M counts + 1000 toplists in a second or two.
(Update: performance comparisons with a database are welcome.)

#! /usr/bin/perl -l

use Array::Heap;
print "starting";

my @h = map [-rand, "item $_"], 'aa000' .. 'zz999';
print int(@h), " elements";
make_heap @h;
print "heap made";

sub hits { --$h[$_][0], adjust_heap @h,$_ for @_; }
sub batch { hits map {int @h*rand()**4} 1..$_[0]; }

sub rotate {
  $_->[0] *= rand, $_->[1] .= '.',
    push_heap @h,$_ for map {pop_heap @h} 1..$_[0];
}
sub top {
  map sprintf("%8d  %s", -$_->[0], $_->[1]),
    map {push_heap @h,$_;$_} map {pop_heap @h} 1..$_[0];
}

for (;;) { 
  batch(1000), top(10) for 1..1000;
  print for "", top(10);
  rotate(3);
}
[download]

Is a heap the right structure? The "adjust_heap" operation must have a time complexity of some significance. I too thought of a heap at first, but can't get around the thought that adjusting the priority of potentially every single song, once per second, seems like it would mean re-heaping potentially the entire structure once per second.

I hope I'm wrong. ;)




Dave