Comment #4 on issue 2584 by [email protected]: please make partcombine merge slurs
http://code.google.com/p/lilypond/issues/detail?id=2584
I don't have a proper recollection of the detailed reasons I picked that particular behavior. I think it had to do with the point of time where decisions had to be made, meaning that doing better an/or more symmetrical behavior would have required completely changing the typesetting order (not likely good for circular calculation dependencies either).
I'll go over the code again today and will try to give an estimate of how much effort dealing with the single-start/multi-end case would entail. As I said: this behavior was chosen consciously in the expectation that it would be a better balance between effort and results than before. And part of the reason for a reasonably easy to do fix was curing a critical regression.
But I also think that I was of the opinion that this particular constellation would not easily get produced by the part combiner, and at least that assumption, if I made it, was apparently not accurate.
