bug#66187: read-file-name unexpected behavior when MUSTMATCH is a function

[Joseph Turner]

Hello!

For reference, here's the little snippet I'm testing with:

(completing-read "Prompt: " '("a" "b") nil
                 (lambda (input)
                   (string= "a" input)))

where the only expected valid input is "a".

Stefan Monnier <monnier@iro.umontreal.ca> writes:

>> Here's another potential solution.  While the attached patch seems to
>> work, I'm not sure that minibuffer-completion-confirm should be checked
>> in completion--do-completion instead of completion--complete-and-exit.
>
> It's definitely better to check it in `completion--complete-and-exit`
> (which is about exiting and thus related to whether the content is
> acceptable) than in `completion--do-completion` (which is just about
> completion).

I'm still not sure `completion--complete-and-exit' is the best place to
check that input is valid before exiting.

`completion--do-completion' contains the following cond clause, which
prevents the user from exiting with, e.g., "c".

((null comp)
 (minibuffer-hide-completions)
 (unless completion-fail-discreetly
   (ding)
   (completion--message "No match"))
 (minibuffer--bitset nil nil nil))

Perhaps we should also run that same body when REQUIRE-MATCH is a
function which returns nil?

> The patch looks OK, thanks.  We could make it call `completion-function`
> instead of saying "no match" (after all, it has "complete" in its name),
> but it comes with its own set of problems.

If I understand correctly, the attached patch does this.

I think this works inside of `completion-complete-and-exit', but maybe
not inside of `minibuffer-force-complete-and-exit' since the former
calls `completion--do-completion' internally but the latter does not.

Still exploring ideas...  Thanks!

Joseph

0001-Fix-completing-read-functional-REQUIRE-MATCH-behavio-2.patch
Description: Text Data