> We could fix this within comment-dwim.

I don't think we could because `comment-indent` is also a command.
Maybe we could make `comment-dwim` do something more useful than calling
`comment-indent` in that case, but that's "another problem".

> However, I propose instead to change comment-indent (or possibly
> comment-search-forward).

+1

> -    (let* ((eolpos (line-end-position))
> -	   (begpos (comment-search-forward eolpos t))
> +    (let* ((startpos (comment-search-forward (line-end-position) t))
> +           (begpos (or startpos
> +                       ;; If we couldn't find a comment *starting* on
> +                       ;; this line, see if we are already within a
> +                       ;; multiline comment at BOL (bug#78003).
> +                       (and (not continue)
> +                            comment-use-syntax comment-use-global-state
> +                            (let ((st (syntax-ppss (line-beginning-position))))
> +                              (and (nth 4 st) (< (nth 8 st) (point))
> +                                   (point))))))
> +           (multilinep (and (not startpos) begpos))

AFAICT we don't need to check (< (nth 8 st) (point)).
Am I missing something?

Also, I think the bug will still bite when
`comment-insert-comment-function` is set.

BTW, the patch looks simple in terms of the diff, but it makes the
control flow (which is already not great) even more complicated.

How 'bout the one below?


        Stefan