This isn't surprising, since the part combiner keeps shuffling
different notes and rests to different Voice contexts, so it's
not easy for LilyPond to realize which voice the \setTextCresc
setting should be applied to.
The easiest solution is to modify the definition of this macro, so it
applies
In the following example, the crescendo has airpins instead of text.
Everything is fine if \partcombine is replaced by polyphony << ... \\
... >>.
\version "2.6.4"
\include "english.ly"
\score {
\partcombine
{
\setTextCresc
r4 c'\< r d'
