bug#6187: 23.1; can't save: "Wrong type argument: number-of-marker-p, nil"

[Glenn Morris <rgm@gnu.org>]

Ryo Furue wrote:

>   apply(min nil)
>   select-safe-coding-system-interactively(1 1362 (utf-8
>     adobe-standard-encoding next mac-roman gb18030 utf-7 utf-16
>     utf-16be-with-signature utf-16le-with-signature utf-16be utf-16le
>     x-ctext iso-2022-7bit utf-8-auto utf-8-with-signature emacs-mule
>     raw-text iso-2022-8bit-ss2 utf-7-imap utf-8-emacs no-conversion
>     compound-text-with-extensions ctext-no-compositions
>     iso-2022-7bit-lock iso-2022-7bit-ss2) (japanese-iso-8bit-unix) nil
>     utf-8)

So:

select-safe-coding-system-interactively is called with

UNSAFE = (japanese-iso-8bit-unix)

This is presumably because it does not appear in the list of codings
returned by the call to find-coding-systems-region in
select-safe-coding-system (passed to
select-safe-coding-system-interactively as 3rd argument above).

Yet when select-safe-coding-system-interactively calls
unencodable-char-position with coding == japanese-iso-8bit-unix, it
returns nil, meaning it can encode the region.

So apparently it both can and cannot encode the region.

One could paper over this by excluding non-numeric elements in

    (goto-char (apply 'min (mapcar #'(lambda (x) (car (cadr x)))
                                  unsafe))))

But I guess the real bug is that find-coding-systems-region and
unencodable-char-position somehow manage to disagree.


An example of a file that triggers the problem would probably be
very helpful.