bug#23092: 25.0.92; Minibuffer completion fails to resize completion window if reused during same command

[martin rudalics]

>> Obviously, the second branch is based on the assumption that a user will
>> "refine" her completions in the sense that she starts with a large
>> number of possible completions and, by typing characters in the
>> minibuffer, reduces the number of possible completions until she found
>> the right one.
>
> This is a valid assumption and basis of design when the completion is of
> something like a command or a function name, because then the completion
> list is gradually narrowed.
>
> But in the case of completion of a file path, this assumption is not
> valid. (And completing each directory name down a file path is a
> perfectly normal use case.) The difference is that one is not doing a
> single completion but several discrete completions.

I'm not sure whether we should by default change something in this case.
Juri has designed the present concept and I would rather leave it to him
how to proceed.

> For example, I was doing a find-file to find something in my Emacs init.
> I started with `~/.em TAB' expecting a single match of ~/.emacs.d/ but
> instead (of course) I got the initial completions window with .emacs and
> .emacs.d in it -- a very small completions window. Then I needed a
> subdirectory in ~/.emacs.d/ but couldn't remember its name at all, so I
> hit TAB again and got the entire contents of emacs.d/ which is a very
> long list [it seems to be jammed up full of session files for some
> reason] all displayed in a buffer sized to show one line of
> completion candidates.

Does enabling ‘temp-buffer-resize-mode’ handle that case sufficiently
well?

>> Apparently, the OP works in the opposite direction - he starts with
>> few suggestions and removes characters from the minibuffer ending up
>> with more and more suggestions.
>
> No, not normally. When I was trying to come up with a simple
> reproducible recipe I wanted an example in the emacs tree, and the
> example I was able to up with was a bit contrived. (Although I don't
> think it's an unlikely use case.)

Let's wait for Juri to chime in.  Meanwhile please have a look at the
manual changes I proposed.  (I don't even know if the example works as
intended.)

martin