Spencer Baugh <sbaugh@janestreet.com> writes:
> 1. touch 'foo bxx' 'foo byy'
> 2. emacs -Q
> 3. M-x shell
> 4. Type "echo foo" and hit TAB; it will complete to "echo foo\ b"
> 5. Hit ? to bring up *Completions*
> 6. Note that the highlighting is incorrect; the next character should
>    be highlighted in pink as the next difference, but it's not.
>
> If you modify minibuffer-completion-help to not bind
> completion-lazy-hilit to t (which I recently added), the highlighting
> will be correct.
>
> This is because completion-lazy-hilit=t makes the highlighting function
> run on the *quoted* string (with the \), whereas
> completion-lazy-hilit=nil runs highlighting on the *unquoted* string,
> which is then highlighted correctly.
>
> IMO, we should fix this by changing the API exposed for
> completion-table-with-quoting.  As mentioned in a comment in
> completion--twq-all:
>
>    ;; The better solution is to not quote the *Completions* display,
>    ;; which nicely circumvents the problem.
>
> If we just stopped quoting *Completions*, we would avoid this problem.
>
> This would necessitate adding some new way to quote the completion
> before inserting it.  We could add some new piece of completion metadata
> which provides a function which gets called when a completion candidate
> is about to be inserted, and returns a string that should be inserted
> instead - it can perform the quoting.  (Or it could even be a new
> completion table action?)

Fortunately it actually turns out to be much, much simpler than this.
We continue to return the quoted completions from
completion-all-completions, and just use completion--unquoted for
display instead of the quoted completion.

Completion UIs should start calling the new completion-for-display
function, but even if they don't, nothing new breaks.