Dmitry Gutov <dmitry@gutov.dev> writes:

> On 27/02/2025 20:33, Juri Linkov wrote:
>> Maybe Dmitry could confirm if this is the right fix
>> for completing-read-multiple recently fixed in
>> bug#48356 and bug#76010.
>
> It seems okay, thanks. Though it depends on the prompt always being a
> field, and the only field in the minibuffer. I don't know if that will
> hold for ever, but it seems to be the case currently.

Is it possible for a user to setup a minibuffer prompt with no properties at
all? Don't think so, but just asking.

> Here's a slightly different fix which seems to behave the same in the
> test scenario. I'm happy with whichever of the two you prefer.

Yours seems better, thanks.

> diff --git a/lisp/emacs-lisp/crm.el b/lisp/emacs-lisp/crm.el
> index 676252ae126..a75ccd46f50 100644
> --- a/lisp/emacs-lisp/crm.el
> +++ b/lisp/emacs-lisp/crm.el
> @@ -269,7 +269,9 @@ completing-read-multiple
>            (setq-local completion-list-insert-choice-function
>                        (lambda (_start _end choice)
>                          (let* ((beg (save-excursion
> -                                      (if (search-backward-regexp
>                                        crm-separator nil t)
> +                                      (if (search-backward-regexp
> crm-separator
> + (field-beginning)
> +                                                                  t)
>                                            (1+ (point))
>                                          (minibuffer-prompt-end))))
>                                 (end (save-excursion

-- 
Thierry