bug#45474: Icomplete exhibiting in recursive minibuffer when it shouldn’t

[Juri Linkov]

> @@ -2812,7 +2832,7 @@ read-number
> -       (let ((str (read-from-minibuffer
> +       (let ((str (read-from-minibuffer-simple

Thanks for handling 'read-number', I use commands from the 'M-g'
prefix map that read numbers in the completion minibuffer.

> @@ -3052,8 +3072,8 @@ read-char-from-minibuffer
> -          (read-from-minibuffer prompt nil map nil
> -                                (or history 'empty-history)))
> +          (read-from-minibuffer-simple prompt nil map nil
> +                                       (or history 'empty-history)))
>…
> @@ -3247,7 +3267,7 @@ y-or-n-p
> -             (str (read-from-minibuffer
> +             (str (read-from-minibuffer-simple

I wonder do you really intend to replace all hundreds of
read-from-minibuffer calls with read-from-minibuffer-simple?
For example, I use 'query-replace' a lot in the completion minibuffer,
but it's not fixed.

To be able to narrow the fix to icomplete.el only, it's
possible to check the value returned from (minibuffer-depth)
before icomplete kicks in, and disable icomplete completions
when the value is greater than it was when entered the first
minibuffer initially, thus handling recursive minibuffers
that don't provide completions.  Or maybe simply disable
previous icomplete when recursive minibuffer doesn't use
completions.