While recognizing the points in A and in B, I think both miss a few points...
In either A or B, you also assume someone is setting all the possible ARs. While some of us are really this obsessive (and I admit to being one such), most ARs I see only have a few done. So the same track on six releases might have 3 with no ARs, one with a producer AR, one with some performer ARs, and one with some more (and partially overlapping) performer and mixer ARs. B also assumes that any given track has the "earliest release of" AR created to point it back to the earlier track. We also know this isn't really the case for 99.9999% (perhaps higher) of the database. A has the negative of error-creep. It also has the advantage of "multiple eyes on multiple places". The earliest release's liner may give very few credits, while later releases include more info - this I find to often be the case with, for example, classical and soundtracks (the latter especially), where the initial release gives no performer info, while later releases, positively identified as being the exact same release, give at least some credits as to performers, rather than just basic composer info. So, even if the info isn't fully present, between different releases of the exact same track, we may gain better info on that track. B has negatives too. First, it does assume that the earliest track AR is pointed to. Second, as hinted at in Frederic's note, it assumes we really do have the earliest release that all the ARs are being applied to. If a new earliest release is later found, a strict interpretation of B would mean our then removing a bunch of perfectly good ARs, only to reapply them to the (now) earliest track instance. Personally, I would prefer not to codify any of this. I would prefer to instead spend some time discussing quite which release level ARs do or don't inherit, as well as working to fill in the holes in our current AR system (missing "libretto" AR for track-artist, missing "is the same track as" track-track AR, etc). I'd also prefer to spend some time discussing quite what we define as being "the same track". I'd prefer this, because hopefully soon, we will have "track masters" in the system. At that point, both A and B become irrelevant, as they as essentially what is solved by track masters. However, at that same point, we will have to have the AR inheritance worked out, so we know what ARs should or should not inherit from releases to tracks to track masters, as well as from tracks to track masters. We also will have to have a definition of just what is "the same track" so we can know what tracks to merge together to create track masters (where the master has more than just a single instance). At this later track master point, the only part of A or B I see which will really have potential to cause headaches is where track 1 has an AR for one person, and track 2 has the same AR for a different person. But this too we can handle - it may be that both are correct, and each track only was half attributed, or it may be that one track had an incorrect AR. But I see this as a positive benefit of approach A - redundant info later being cross-checked to later find potential errors in the redundancy. Brian
_______________________________________________ Musicbrainz-style mailing list Musicbrainz-style@lists.musicbrainz.org http://lists.musicbrainz.org/mailman/listinfo/musicbrainz-style