Re: (regexp-opt-depth "[\\(]") => 1 :-(

[Alan Mackenzie]

On Sun, 20 Apr 2003, Alan Mackenzie wrote:

>There Is No Alternative:  regexp-opt-depth MUST analyse its argument
>properly.  The following rewrite of regexp-opt-depth does just that.

>[Well, OK, it would be as well for somebody else to check the formulation
>of regexp-opt-not-groupie*-re.  ;-]

Many thanks to RMS for doing just that and telling me that 
# This is a good idea, but it fails on "[[:alpha:]\\(]".
# I think the value for `class' needs to be more sophisticated.

>The patch below passes the following test cases:
>(regexp-opt-depth "(asdf)")             => 0
>(regexp-opt-depth "\\(asdf\\)")         => 1
>(regexp-opt-depth "\\(\\(asdf\\)\\)")   => 2
>(regexp-opt-depth "\\(?:asdf\\)")       => 0
>(regexp-opt-depth "[\\(]")              => 0

>(regexp-opt-depth "[a]\\(]asd\\)")      => 1
>(regexp-opt-depth "[^a]\\(]asd\\)")     => 1
>(regexp-opt-depth "[]\\(]asd)")         => 0
>(regexp-opt-depth "[^]\\(]asd)")        => 0
>(regexp-opt-depth "\\(? \\)")           signals "invalid regexp".

Here is the amended patch with that more sophisticated regexp for class.
In addition to the above test cases, the newer version passes these:

(regexp-opt-depth "[[:alpha:]\\(]")     => 0
(regexp-opt-depth "[[:alpha]\\(")       signals "invalid regexp"
(regexp-opt-depth "[[:alpha]\\(\\)")    => 1
(regexp-opt-depth "[[:alp$ha:]\\(\\)")  signals "Invalid regexp"
(regexp-opt-depth "[[alpha:]\\(]\\)")   => 1

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
2003-04-23  Alan Mackenzie  <acm@muc.de>

        * regexp-opt.el: In regexp-opt-depth, don't count a "//(" which appears
        inside a character set].  New constant regexp-opt-not-groupie*-re.

*** regexp-opt.1.24.el  Fri Apr 18 18:34:34 2003
--- regexp-opt.acm.1.24.el      Tue Apr 22 20:52:53 2003
***************
*** 110,115 ****
--- 110,133 ----
           (re (regexp-opt-group sorted-strings open)))
        (if words (concat "\\<" re "\\>") re))))
  
+ (defconst regexp-opt-not-groupie*-re
+   (let* ((harmless-ch "[^\\\\[]")
+          (esc-pair-not-lp "\\\\[^(]")
+          (class-harmless-ch "[^][]")
+          (class-lb-harmless "[^]:]")
+          (class-lb-colon-maybe-charclass ":\\([a-z]+:]\\)?")
+          (class-lb (concat "\\[\\(" class-lb-harmless
+                            "\\|" class-lb-colon-maybe-charclass "\\)"))
+          (class
+           (concat "\\[^?]?"
+                   "\\(" class-harmless-ch
+                   "\\|" class-lb "\\)*"
+                   "\\[?]"))         ; special handling for bare [ at end of re
+          (shy-lp "\\\\(\\?:"))
+     (concat "\\(" harmless-ch "\\|" esc-pair-not-lp
+             "\\|" class "\\|" shy-lp "\\)*"))
+   "Matches any part of a regular expression EXCEPT for non-shy \"\\\\(\"s")
+ 
  ;;;###autoload
  (defun regexp-opt-depth (regexp)
    "Return the depth of REGEXP.
***************
*** 120,130 ****
      (string-match regexp "")
      ;; Count the number of open parentheses in REGEXP.
      (let ((count 0) start)
!       (while (string-match "\\(\\`\\|[^\\]\\)\\\\\\(\\\\\\\\\\)*([^?]"
!                          regexp start)
!       (setq count (1+ count)
!             ;; Go back 2 chars (one for [^?] and one for [^\\]).
!             start (- (match-end 0) 2)))
        count)))
  
  ;;; Workhorse functions.
--- 138,149 ----
      (string-match regexp "")
      ;; Count the number of open parentheses in REGEXP.
      (let ((count 0) start)
!       (while
!           (progn
!             (string-match regexp-opt-not-groupie*-re regexp start)
!             (setq start ( + (match-end 0) 2))  ; +2 for "\\(" after match-end.
!             (<= start (length regexp)))
!         (setq count (1+ count)))
        count)))
  
  ;;; Workhorse functions.
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;

-- 
Alan Mackenzie (Munich, Germany)