(declaim (maybe-inline digit-char-p digit-weight))
(deftype char-code ()
- `(integer 0 (,char-code-limit)))
+ `(integer 0 (,sb!xc:char-code-limit)))
#!+sb-unicode
(progn
:directory
'(:relative :up :up "output")
:name name :type type)
- sb!xc:*compile-file-truename*)))
- (let ((character-database
- (with-open-file (stream (file "ucd" "dat")
+ sb!xc:*compile-file-truename*))
+ (read-ub8-vector (pathname)
+ (with-open-file (stream pathname
:direction :input
:element-type '(unsigned-byte 8))
(let* ((length (file-length stream))
length :element-type '(unsigned-byte 8))))
(read-sequence array stream)
array))))
+ (let ((character-database (read-ub8-vector (file "ucd" "dat")))
+ (decompositions (read-ub8-vector (file "decomp" "dat")))
+ (long-decompositions (read-ub8-vector (file "ldecomp" "dat")))
+ (primary-compositions (read-ub8-vector (file "comp" "dat"))))
`(progn
- (declaim (type (simple-array (unsigned-byte 8) (*)) **character-database**))
+ (declaim (type (simple-array (unsigned-byte 8) (*))
+ **character-database**
+ **character-decompositions**
+ **character-long-decompositions**))
(defglobal **character-database** ,character-database)
+ (defglobal **character-decompositions** ,decompositions)
+ (defglobal **character-long-decompositions** ,long-decompositions)
+ ;; KLUDGE: temporary value, fixed up in cold-load
+ (defglobal **character-primary-compositions** ,primary-compositions)
(defun !character-database-cold-init ()
- (setf **character-database** ,character-database))
+ (setf **character-database** ,character-database)
+ (setf **character-primary-compositions**
+ (let ((table (make-hash-table))
+ (info ,primary-compositions))
+ (flet ((code (j)
+ (dpb (aref info (* 4 j))
+ (byte 8 24)
+ (dpb (aref info (+ (* 4 j) 1))
+ (byte 8 16)
+ (dpb (aref info (+ (* 4 j) 2))
+ (byte 8 8)
+ (aref info (+ (* 4 j) 3)))))))
+ #!+sb-unicode
+ (dotimes (i (/ (length info) 12))
+ (setf (gethash (dpb (code (* 3 i)) (byte 21 21)
+ (code (1+ (* 3 i))))
+ table)
+ (code-char (code (+ (* 3 i) 2)))))
+ table))))
,(with-open-file (stream (file "ucd-names" "lisp-expr")
:direction :input
:element-type 'character)
\f
;;;; UCD accessor functions
-;;; The first (* 8 215) => 1720 entries in **CHARACTER-DATABASE**
+;;; The first (* 8 396) => 3168 entries in **CHARACTER-DATABASE**
;;; contain entries for the distinct character attributes:
;;; specifically, indexes into the GC kinds, Bidi kinds, CCC kinds,
;;; the decimal digit property, the digit property and the
;;; the next (ash #x110000 -8) entries contain single-byte indexes
;;; into a table of 256-element 4-byte-sized entries. These entries
;;; follow directly on, and are of the form
-;;; {attribute-index[1B],transformed-code-point[3B]}x256, where the
+;;; {attribute-index[11b],transformed-code-point[21b]}x256, where the
;;; attribute index is an index into the miscellaneous information
;;; table, and the transformed code point is the code point of the
;;; simple mapping of the character to its lowercase or uppercase
;;;
;;; To look up information about a character, take the high 13 bits of
;;; its code point, and index the character database with that and a
-;;; base of 1720 (going past the miscellaneous information[*], so
+;;; base of 3168 (going past the miscellaneous information[*], so
;;; treating (a) as the start of the array). This, labelled A, gives
;;; us another index into the detailed pages[-], which we can use to
;;; look up the details for the character in question: we add the low
;;; 8 bits of the character, shifted twice (because we have four-byte
-;;; table entries) to 1024 times the `page' index, with a base of 6072
+;;; table entries) to 1024 times the `page' index, with a base of 7520
;;; to skip over everything else. This gets us to point B. If we're
;;; after a transformed code point (i.e. an upcase or downcase
;;; operation), we can simply read it off now, beginning with an
-;;; offset of 1 byte from point B in some endianness; if we're looking
-;;; for miscellaneous information, we take the value at B, and index
-;;; the character database once more to get to the relevant
-;;; miscellaneous information.
+;;; offset of 11 bits from point B in some endianness; if we're
+;;; looking for miscellaneous information, we take the 11-bit value at
+;;; B, and index the character database once more to get to the
+;;; relevant miscellaneous information.
+;;;
+;;; As an optimization to the common case (pun intended) of looking up
+;;; case information for a character, the entries in C above are
+;;; sorted such that the characters which are UPPER-CASE-P in CL terms
+;;; have index values lower than all others, followed by those which
+;;; are LOWER-CASE-P in CL terms; this permits implementation of
+;;; character case tests without actually going to the trouble of
+;;; looking up the value associated with the index. (Actually, this
+;;; isn't just a speed optimization; the information about whether a
+;;; character is BOTH-CASE-P is used just in the ordering and not
+;;; explicitly recorded in the database).
;;;
;;; The moral of all this? Next time, don't just say "FIXME: document
;;; this"
(defun ucd-index (char)
(let* ((cp (char-code char))
(cp-high (ash cp -8))
- (page (aref **character-database** (+ 1720 cp-high))))
- (+ 6072 (ash page 10) (ash (ldb (byte 8 0) cp) 2))))
+ (page (aref **character-database** (+ 3168 cp-high))))
+ (+ 7520 (ash page 10) (ash (ldb (byte 8 0) cp) 2))))
-(declaim (ftype (sfunction (t) (unsigned-byte 8)) ucd-value-0))
+(declaim (ftype (sfunction (t) (unsigned-byte 11)) ucd-value-0))
(defun ucd-value-0 (char)
- (aref **character-database** (ucd-index char)))
+ (let ((index (ucd-index char))
+ (character-database **character-database**))
+ (dpb (aref character-database index)
+ (byte 8 3)
+ (ldb (byte 3 5) (aref character-database (+ index 1))))))
-(declaim (ftype (sfunction (t) (unsigned-byte 24)) ucd-value-1))
+(declaim (ftype (sfunction (t) (unsigned-byte 21)) ucd-value-1))
(defun ucd-value-1 (char)
(let ((index (ucd-index char))
(character-database **character-database**))
- (dpb (aref character-database (+ index 3))
- (byte 8 16)
+ (dpb (aref character-database (+ index 1))
+ (byte 5 16)
(dpb (aref character-database (+ index 2))
(byte 8 8)
- (aref character-database (1+ index))))))
+ (aref character-database (+ index 3))))))
(declaim (ftype (sfunction (t) (unsigned-byte 8)) ucd-general-category))
(defun ucd-general-category (char)
(+ 3 (* 8 (ucd-value-0 char))))))
(when (< decimal-digit 10)
decimal-digit)))
+(declaim (ftype (sfunction (t) (unsigned-byte 8)) ucd-ccc))
+(defun ucd-ccc (char)
+ (aref **character-database** (+ 2 (* 8 (ucd-value-0 char)))))
(defun char-code (char)
#!+sb-doc
#!+sb-doc
"The argument must be a character object; UPPER-CASE-P returns T if the
argument is an upper-case character, NIL otherwise."
- (= (ucd-value-0 char) 0))
+ (< (ucd-value-0 char) 5))
(defun lower-case-p (char)
#!+sb-doc
"The argument must be a character object; LOWER-CASE-P returns T if the
argument is a lower-case character, NIL otherwise."
- (= (ucd-value-0 char) 1))
+ (< 4 (ucd-value-0 char) 9))
(defun both-case-p (char)
#!+sb-doc
"The argument must be a character object. BOTH-CASE-P returns T if the
argument is an alphabetic character and if the character exists in both upper
and lower case. For ASCII, this is the same as ALPHA-CHAR-P."
- (< (ucd-value-0 char) 2))
+ (< (ucd-value-0 char) 9))
(defun digit-char-p (char &optional (radix 10.))
#!+sb-doc
(defmacro equal-char-code (character)
(let ((ch (gensym)))
`(let ((,ch ,character))
- (if (= (ucd-value-0 ,ch) 0)
+ (if (< (ucd-value-0 ,ch) 5)
(ucd-value-1 ,ch)
(char-code ,ch)))))
(defun two-arg-char-equal (c1 c2)
- (= (equal-char-code c1) (equal-char-code c2)))
+ (flet ((base-char-equal-p ()
+ (let* ((code1 (char-code c1))
+ (code2 (char-code c2))
+ (sum (logxor code1 code2)))
+ (when (eql sum #x20)
+ (let ((sum (+ code1 code2)))
+ (or (and (> sum 161) (< sum 213))
+ (and (> sum 415) (< sum 461))
+ (and (> sum 463) (< sum 477))))))))
+ (declare (inline base-char-equal-p))
+ (or (eq c1 c2)
+ #!-sb-unicode
+ (base-char-equal-p)
+ #!+sb-unicode
+ (typecase c1
+ (base-char
+ (and (base-char-p c2)
+ (base-char-equal-p)))
+ (t
+ (= (equal-char-code c1) (equal-char-code c2)))))))
+
+(defun char-equal-constant (x char reverse-case-char)
+ (declare (type character x))
+ (or (eq char x)
+ (eq reverse-case-char x)))
(defun char-equal (character &rest more-characters)
#!+sb-doc
#!+sb-doc
"Return CHAR converted to upper-case if that is possible. Don't convert
lowercase eszet (U+DF)."
- (if (= (ucd-value-0 char) 1)
+ (if (< 4 (ucd-value-0 char) 9)
(code-char (ucd-value-1 char))
char))
(defun char-downcase (char)
#!+sb-doc
"Return CHAR converted to lower-case if that is possible."
- (if (= (ucd-value-0 char) 0)
+ (if (< (ucd-value-0 char) 5)
(code-char (ucd-value-1 char))
char))
(and (typep weight 'fixnum)
(>= weight 0) (< weight radix) (< weight 36)
(code-char (if (< weight 10) (+ 48 weight) (+ 55 weight)))))
+\f
+(defun char-decomposition-info (char)
+ (aref **character-database** (+ 6 (* 8 (ucd-value-0 char)))))
+
+(defun char-decomposition (char)
+ (let* ((cp (char-code char))
+ (cp-high (ash cp -8))
+ (decompositions **character-decompositions**)
+ (long-decompositions **character-long-decompositions**)
+ (index (+ #x1100
+ (ash (aref decompositions cp-high) 10)
+ (ash (ldb (byte 8 0) cp) 2)))
+ (v0 (aref decompositions index))
+ (v1 (aref decompositions (+ index 1)))
+ (v2 (aref decompositions (+ index 2)))
+ (v3 (aref decompositions (+ index 3)))
+ (length (dpb v0 (byte 8 3) (ldb (byte 3 5) v1)))
+ (entry (dpb (ldb (byte 5 0) v1) (byte 5 16)
+ (dpb v2 (byte 8 8) v3))))
+ (if (= length 1)
+ (string (code-char entry))
+ (if (<= #xac00 cp #xd7a3)
+ ;; see Unicode 6.2, section 3-12
+ (let* ((sbase #xac00)
+ (lbase #x1100)
+ (vbase #x1161)
+ (tbase #x11a7)
+ (lcount 19)
+ (vcount 21)
+ (tcount 28)
+ (ncount (* vcount tcount))
+ (scount (* lcount ncount))
+ (sindex (- cp sbase))
+ (lindex (floor sindex ncount))
+ (vindex (floor (mod sindex ncount) tcount))
+ (tindex (mod sindex tcount))
+ (result (make-string length)))
+ (declare (ignore scount))
+ (setf (char result 0) (code-char (+ lbase lindex)))
+ (setf (char result 1) (code-char (+ vbase vindex)))
+ (when (> tindex 0)
+ (setf (char result 2) (code-char (+ tbase tindex))))
+ result)
+ (let ((result (make-string length))
+ (e (* 4 entry)))
+ (dotimes (i length result)
+ (let ((code (dpb (aref long-decompositions (+ e 1))
+ (byte 8 16)
+ (dpb (aref long-decompositions (+ e 2))
+ (byte 8 8)
+ (aref long-decompositions (+ e 3))))))
+ (setf (char result i) (code-char code)))
+ (incf e 4)))))))
+
+(defun decompose-char (char)
+ (if (= (char-decomposition-info char) 0)
+ (string char)
+ (char-decomposition char)))
+
+(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)))