> Which means that read-command is still returning something that
> matches [[:INCOMPLETE:]]* on that very first call, which ought to
> be impossible as I understand it.  And indeed, if I step across
> getrestchar() with a debugger, it's failing on any character that
> is more than two bytes wide (returning only the first two bytes),
> which probably leaves mbrtowc() in an indeterminate state.  (This
> is reading from the "zle -U" buffer so key timeout does not matter.)

getrestchar()  only changed to the extent of passing back extra info.
Itʼs being called at a different point but as long as the first byte  is
correct that shouldn't matter. Is that happening with 5.1.1?

pws