Dear Knute,
Here is the best that I could do...but I'm not sure that this is the
most correct solution or that it will look good in context...
This solution looks the closest to what most of the editions I'm looking at do. None of them use the split-stem distributed notes. Thank you for working it out for me.
%%% BEGIN %%%
[solution snipped]
Yes, that looks good, thanks. It's weird, though, since this is a not uncommon thing that LilyPond has so much trouble with it. I'm not sure what could be done, maybe a "do not merge" option or even knowing that two notes with different accidentals shouldn't be merged. (I am a little surprised that d-natural and d-sharp get merged.)
Yes, earlier in the thread Paul made the point that Elaine Gould does specifically deal with this case (on altered unisons, p 91), and I just glanced at a copy and he's right.
I don't know how to reason through LilyPond's logic on this well enough to understand what's going on. But it looks like Lily collects all accidentalled notes that occur together at a given moment -- across all voices in a single staff -- then collects these into a bundle, and then engraves the bundle of accidentals in a single column to the left of all simultaneously occurrings notes (or chords). Even if the simultaneity created between voices is a type of unison. I thought that perhaps removing the Accidental_engraver from the Staff context and adding it to both Voice contexts might change the behavior to what Gould recommends, but it doesn't.
Gould's examples suggest that there should be "columns" of accidentals in front of each voice on a staff ... but only when the interval we're engraving an altered unison. When polyphonic voices combine to create any other type of interval, her accidental positioning looks very much like LilyPond's.