Here is quite an artificial example (which is wrong, by the way, though it may hopefully illustrate my point, at least to those reasonably imaginative among the readers.) The result is not what I'd intend/expect:
test = \relative c'' {
c4 a b c ~ \repeat volta 2 { c a\( b c ~ }
\alternative { { c2\) r4 c ~ } { c2\) r } }
}
What I want for the tie in alternative 1 is "clearly" to bind to the first tone of the volta repeat. Now it binds ("incorrectly", but without any diagnostic) to the c2 note in alternative 2. Instead, however, a warning is generated, again "incorrectly", for the ending of the phrasing slur in alternative 2.
I put the word incorrectly in citation marks, to suggest my _expression_ of what is merely an opinion of mine, not necessarily the truth (in a Lilypond sense). I don't think, however, that this problem is in any way fictitional or uncommon. (I've hopefully not missed an FAQ on this!) Slurs, ties, dynamics, and so on, could obviously live past branching points. I may well be that Lilypond doesn't have the necessary contextual information, at least not at da capo/dal segno/al coda jumps to do the right thing, but as I don't use any MIDI backend, I'd be happy with any suggested workaround for generating at least the "correct" score.