+
+(defun decompose-string (string &optional (kind :canonical))
+ (declare (type (member :canonical :compatibility) kind))
+ (flet ((canonical (char)
+ (= 1 (char-decomposition-info char)))
+ (compat (char)
+ (/= 0 (char-decomposition-info char))))
+ (let (result
+ (fun (ecase kind
+ (:canonical #'canonical)
+ (:compatibility #'compat))))
+ (do* ((start 0 (1+ end))
+ (end (position-if fun string :start start)
+ (position-if fun string :start start)))
+ ((null end) (push (subseq string start end) result))
+ (unless (= start end)
+ (push (subseq string start end) result))
+ ;; FIXME: this recursive call to DECOMPOSE-STRING is necessary
+ ;; for correctness given our direct encoding of the
+ ;; decomposition data in UnicodeData.txt. It would, however,
+ ;; be straightforward enough to perform the recursion in table
+ ;; construction, and then have this simply revert to a single
+ ;; lookup. (Wait for tests to be hooked in, then implement).
+ (push (decompose-string (decompose-char (char string end)) kind)
+ result))
+ (apply 'concatenate 'string (nreverse result)))))
+
+(defun sort-combiners (string)
+ (let (result (start 0) first-cc first-non-cc)
+ (tagbody
+ again
+ (setf first-cc (position 0 string :key #'ucd-ccc :test #'/= :start start))
+ (when first-cc
+ (setf first-non-cc (position 0 string :key #'ucd-ccc :test #'= :start first-cc)))
+ (push (subseq string start first-cc) result)
+ (when first-cc
+ (push (stable-sort (subseq string first-cc first-non-cc) #'< :key #'ucd-ccc) result))
+ (when first-non-cc
+ (setf start first-non-cc first-cc nil first-non-cc nil)
+ (go again)))
+ (apply 'concatenate 'string (nreverse result))))
+
+(defun primary-composition (char1 char2)
+ (let ((c1 (char-code char1))
+ (c2 (char-code char2)))
+ (cond
+ ((gethash (dpb (char-code char1) (byte 21 21) (char-code char2))
+ **character-primary-compositions**))
+ ((and (<= #x1100 c1) (<= c1 #x1112)
+ (<= #x1161 c2) (<= c2 #x1175))
+ (let ((lindex (- c1 #x1100))
+ (vindex (- c2 #x1161)))
+ (code-char (+ #xac00 (* lindex 588) (* vindex 28)))))
+ ((and (<= #xac00 c1) (<= c1 #.(+ #xac00 11171))
+ (<= #x11a8 c2) (<= c2 #x11c2)
+ (= 0 (rem (- c1 #xac00) 28)))
+ (code-char (+ c1 (- c2 #x11a7)))))))
+
+;;; This implements a sequence data structure, specialized for
+;;; efficient deletion of characters at an index, along with tolerable
+;;; random access. The purpose is to support the canonical
+;;; composition algorithm from Unicode, which involves replacing (not
+;;; necessarily consecutive) pairs of code points with a single code
+;;; point (e.g. [#\e #\combining_acute_accent] with
+;;; #\latin_small_letter_e_with_acute). The data structure is a list
+;;; of three-element lists, each denoting a chunk of string data
+;;; starting at the first index and ending at the second.
+;;;
+;;; Actually, the implementation isn't particularly efficient, and
+;;; would probably benefit from being rewritten in terms of displaced
+;;; arrays, which would substantially reduce copying.
+;;;
+;;; (also, generic sequences. *sigh*.)
+(defun lref (lstring index)
+ (dolist (l lstring)
+ (when (and (<= (first l) index)
+ (< index (second l)))
+ (return (aref (third l) (- index (first l)))))))
+(defun (setf lref) (newchar lstring index)
+ (dolist (l lstring)
+ (when (and (<= (first l) index)
+ (< index (second l)))
+ (return (setf (aref (third l) (- index (first l))) newchar)))))
+(defun llength (lstring)
+ (second (first (last lstring))))
+(defun lstring (lstring)
+ (let ((result (make-string (llength lstring))))
+ (dolist (l lstring result)
+ (replace result (third l) :start1 (first l) :end1 (second l)))))
+(defun ldelete (lstring index)
+ (do* ((ls lstring (cdr ls))
+ (l (car ls) (car ls))
+ so-fars)
+ ((and (<= (first l) index)
+ (< index (second l)))
+ (append
+ (nreverse so-fars)
+ (cond
+ ((= (first l) index)
+ (list (list (first l) (1- (second l)) (subseq (third l) 1))))
+ ((= index (1- (second l)))
+ (list (list (first l) (1- (second l)) (subseq (third l) 0 (1- (length (third l)))))))
+ (t
+ (list
+ (list (first l) index
+ (subseq (third l) 0 (- index (first l))))
+ (list index (1- (second l))
+ (subseq (third l) (1+ (- index (first l))))))))
+ (mapcar (lambda (x) (list (1- (first x)) (1- (second x)) (third x)))
+ (cdr ls))))
+ (push l so-fars)))
+
+(defun canonically-compose (string)
+ (labels ()
+ (let* ((result (list (list 0 (length string) string)))
+ (previous-starter-index (position 0 string :key #'ucd-ccc))
+ (i (and previous-starter-index (1+ previous-starter-index))))
+ (when (or (not i) (= i (length string)))
+ (return-from canonically-compose string))
+ (tagbody
+ again
+ (when (and (>= (- i previous-starter-index) 2)
+ ;; test for Blocked (Unicode 3.11 para. D115)
+ ;;
+ ;; (assumes here that string has sorted combiners,
+ ;; so can look back just one step)
+ (>= (ucd-ccc (lref result (1- i)))
+ (ucd-ccc (lref result i))))
+ (when (= (ucd-ccc (lref result i)) 0)
+ (setf previous-starter-index i))
+ (incf i)
+ (go next))
+
+ (let ((comp (primary-composition (lref result previous-starter-index)
+ (lref result i))))
+ (cond
+ (comp
+ (setf (lref result previous-starter-index) comp)
+ (setf result (ldelete result i)))
+ (t
+ (when (= (ucd-ccc (lref result i)) 0)
+ (setf previous-starter-index i))
+ (incf i))))
+ next
+ (unless (= i (llength result))
+ (go again)))
+ (if (= i (length string))
+ string
+ (lstring result)))))
+
+(defun normalize-string (string &optional (form :nfd))
+ (declare (type (member :nfd :nfkd :nfc :nfkc) form))
+ #!-sb-unicode
+ (etypecase string
+ ((array nil (*)) string)
+ (string
+ (ecase form
+ ((:nfc :nfkc) string)
+ ((:nfd :nfkd) (error "Cannot normalize to ~A form in #-SB-UNICODE builds" form)))))
+ #!+sb-unicode
+ (etypecase string
+ (base-string string)
+ ((array character (*))
+ (ecase form
+ ((:nfc)
+ (canonically-compose (sort-combiners (decompose-string string))))
+ ((:nfd)
+ (sort-combiners (decompose-string string)))
+ ((:nfkc)
+ (canonically-compose (sort-combiners (decompose-string string :compatibility))))
+ ((:nfkd)
+ (sort-combiners (decompose-string string :compatibility)))))
+ ((array nil (*)) string)))