What I meant is: run 'src/nano +7 NEWS' with my version of the
patch. Then type: port ^] ^] ^] ^] ^]. Now type M-U (which
means: undo), and retype that slowly, and see how each time
the completion is removed and then the previous completion
readded, and so on, until the first completion is removed and
then "port" itself is removed. What I expect to happen when
I type M-U after having typed ^] for a few times, is: that the
completion disappears (as it does now), and then a second M-U
should remove "port" -- the undo system should not retain any
memory of all the tried possibilities. So each strike of ^]
that is not the first in a series, should probably pop the top
two items off the undo stack (or just one item if you use a
memmove instead of backspaces).