Re: [:upper:] inconsistency

[Francis Wright]

From: "Stefan Monnier" <address@hidden>
To: "Francis Wright" <address@hidden>
Cc: <address@hidden>
Sent: Sunday, January 07, 2007 7:32 PM
Subject: Re: [:upper:] inconsistency


> > I would expect the regexp [[:upper:]] to behave the same as [A-Z] (in
> > English text).  However, by default [[:upper:]] matches all letters,
> > whereas [A-Z] matches only upper case letters, as I expect.
>
> Hmm... I can't see to get A-Z to only match upper-case letters ...
> Oh I see what you mean now.  Does the patch below fix it for you?
> (I guess we could also allows the use of [:UPPER:] so you could workaround
> this problem ;-)
>
>
>        Stefan
>
>
> --- orig/lisp/isearch.el
> +++ mod/lisp/isearch.el
> @@ -2297,7 +2297,15 @@
>        (setq found t))
>    (setq quote-flag nil)))
>       (setq i (1+ i)))
> -    (not found)))
> +    (or (not found)
> +        ;; Even if there's no uppercase char, we want to detect the use 
> of
> +        ;; [:upper:] char-class.
> +        (and regexp-flag (string-match "\\[:upper:]" string)
> +             (condition-case err
> +                 (progn (string-match (substring string 0 (match-string 
> 0)) "")
> +                        nil)
> +               (invalid-regexp
> +                (equal "Unmatched [ or [^" (cadr err))))))))
>
> ;; Portability functions to support various Emacs versions.


Thanks for the patch.  However, it doesn't fix the problem.  The following 
version of the function that you patched does seem to work.  I have moved 
the `not' outside the `or' at the start of your patch.  The code from your 
patch that I have commented out does not work but I don't understand it well 
enough to suggest a fix.  The sexpr `(match-string 0)' returns a string, 
whereas `substring' wants an integer argument.

(defun isearch-no-upper-case-p (string regexp-flag)
 "Return t if there are no upper case chars in STRING.
If REGEXP-FLAG is non-nil, disregard letters preceded by `\\' (but not 
`\\\\')
since they have special meaning in a regexp."
 (let (quote-flag (i 0) (len (length string)) found)
   (while (and (not found) (< i len))
     (let ((char (aref string i)))
       (if (and regexp-flag (eq char ?\\))
           (setq quote-flag (not quote-flag))
         (if (and (not quote-flag) (not (eq char (downcase char))))
             (setq found t))
         (setq quote-flag nil)))
     (setq i (1+ i)))
   (not
    (or found
        ;; Even if there's no uppercase char, we want to detect the
        ;; use of [:upper:] char-class.
        (and regexp-flag (string-match "\\[:upper:]" string)
;;            (condition-case err
;;                (progn (string-match (substring string 0 (match-string 0)) 
"")
;;                       nil)
;;              (invalid-regexp
;;               (equal "Unmatched [ or [^" (cadr err))))
             )))))

I think this problem really needs fixing at a lower level.  When the regexp 
matcher is matching character classes such as [:upper:] and [:lower:] it 
should ignore case-fold-search.

Francis