I have a fairly extensive collection of music. As most of it was acquired in the "bad old days", I find myself with singles, full albums and compilations all containing the same song, either identical or different mixes, live versions etc.

Now, identical versions I could prune from my collection, either by rating dupes zero, deleting the files or some other such measure. I'm somewhat hesitant to do so with different versions of the same song, however.

This means that short of manual pruning, it's impossible for me to generate playlists where any given song appears only once. This got me thinking about how you would go about enabling something like this.

Clearly, what you would need to implement something like this is some form of clustering of tracks.

Either explicit - i.e., define the cluster "La Isla Bonita" as a list of tracks in the database, e.g., "Madonna/True Blue/La Isla Bonita", "Madonna/The Immaculate Collection/La Isla Bonita", "David Hasselhoff/Sings America/La Isla Bonita" and "Alizée/Psychédélices/La Isla Bonita", and then create a playlist of such clusters, which, when played will select a random entry from the list of each cluster.

Or implicit - i.e., allow tracks to be related to other tracks, e.g., let the user specify in the database that the track "David Hasselhoff/Sings America/La Isla Bonita" is a cover (or mix, instrumental version etc) of "Madonna/True Blue/Like a virgin". Then the current duplicate removal only needed to be extended to consider related tracks to be the same.

As I've been led to believe that Amarok is built on top of MySQL, it doesn't seem to me like doing something like this should be too difficult, but I do all my programming on whiteboards these days...