I am attaching a second revision of the earlier patch.  

It is not that patch is final.  It is worthy of further discussion and
experimentation.

Changes are in minibuffer.el and icomplete.el.

1. `minibuffer-summarize-completions' is meant as a replacement for
   `icomplete-exhibit'.  As the name suggests, it is meant to go in to
    minibuffer.el.  It's presence in minibuffer.el proved problematic
   (details in followup mail) and I had to move it to icomplete.el.

   TODO: Handle key binding hints for sole command matches?  Is it
   really necessary.  It seems so `old school'.  May be it made sense
   when Emacs /did not/ provide key binding hints.

2. There is a display overlay that is used in minibuffer (see
   `minibuffer-message').  There is a counterpart in icomplete.el named
   `icomplete-overlay'. 

   I am wondering whether `icomplete-overlay' could be thrown away and
   have it use, yet to be introduced `minibuffer-overlay'.  Should this
   be buffer-local etc etc. I am not sure of.

   I can exchange notes if there is some interest around this area.

3. Apropos `minibuffer-force-complete-and-exit' and `confirm' etc.

    >> +(defun icomplete-this-match ()
    >> +  "Input first of the displayed matches to minibuffer prompt.
    >> +See `icomplete-matches'."
    >> +  (interactive)
    >> +  (delete-region (minibuffer-prompt-end) (point))
    >> +  (when icomplete-matches
    >> +    (insert (car icomplete-matches)))
    >> +  (exit-minibuffer))
    >
    > I think it should still call test-completion and obey
    > minibuffer-completion-confirm if that test fails.

  Can `test-completion' fail if the prompt is filled from valid
  candidates? Remember, `minibuffer-force-and-complete-and-exit' doesn't
  rely on manual-typing at the prompt.


--8<---------------cut here---------------start------------->8---