X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~mdw/git/disorder/blobdiff_plain/18cda350fe80c2226684f9660f2956869552b317..fde67de26e36cf02d5632d934623bbc054a3c1d9:/lib/unicode.c

diff --git a/lib/unicode.c b/lib/unicode.c
index 8fe20a6..675f703 100644
--- a/lib/unicode.c
+++ b/lib/unicode.c
@@ -1,21 +1,19 @@
 /*
  * This file is part of DisOrder
- * Copyright (C) 2007 Richard Kettlewell
+ * Copyright (C) 2007, 2009 Richard Kettlewell
  *
- * This program is free software; you can redistribute it and/or modify
+ * This program is free software: you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation; either version 2 of the License, or
+ * the Free Software Foundation, either version 3 of the License, or
  * (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful, but
- * WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * General Public License for more details.
- *
+ * 
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ * 
  * You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
- * USA
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  */
 /** @file lib/unicode.c
  * @brief Unicode support functions
@@ -30,13 +28,16 @@
  * database code.
  *
  * As the code stands this guarantee is not well met!
+ *
+ * Subpages:
+ * - @ref utf32props
+ * - @ref utftransform
+ * - @ref utf32iterator
+ * - @ref utf32
+ * - @ref utf8
  */
 
-#include <config.h>
-#include "types.h"
-
-#include <string.h>
-#include <stdio.h>		/* TODO */
+#include "common.h"
 
 #include "mem.h"
 #include "vector.h"
@@ -102,8 +103,18 @@ static inline int utf32__combining_class(uint32_t c) {
   return utf32__unidata(c)->ccc;
 }
 
+/** @brief Return the combining class of @p c
+ * @param c Code point
+ * @return Combining class of @p c
+ *
+ * @p c can be any 32-bit value, a sensible value will be returned regardless.
+ */
+int utf32_combining_class(uint32_t c) {
+  return utf32__combining_class(c);
+}
+
 /** @brief Return the General_Category value for @p c
- * @param Code point
+ * @param c Code point
  * @return General_Category property value
  *
  * @p c can be any 32-bit value, a sensible value will be returned regardless.
@@ -151,6 +162,28 @@ static inline int utf32__boundary_ignorable(enum unicode_Word_Break wb) {
           || wb == unicode_Word_Break_Format);
 }
 
+/** @brief Return the canonical decomposition of @p c
+ * @param c Code point
+ * @return 0-terminated canonical decomposition, or 0
+ */
+static inline const uint32_t *utf32__decomposition_canon(uint32_t c) {
+  const struct unidata *const data = utf32__unidata(c);
+  const uint32_t *const decomp = data->decomp;
+
+  if(decomp && !(data->flags & unicode_compatibility_decomposition))
+    return decomp;
+  else
+    return 0;
+}
+
+/** @brief Return the compatibility decomposition of @p c
+ * @param c Code point
+ * @return 0-terminated decomposition, or 0
+ */
+static inline const uint32_t *utf32__decomposition_compat(uint32_t c) {
+  return utf32__unidata(c)->decomp;
+}
+
 /*@}*/
 /** @defgroup utftransform Functions that transform between different Unicode encoding forms */
 /*@{*/
@@ -209,7 +242,7 @@ error:
  * @param s Source string
  * @param ns Length of source string in code points
  * @param ndp Where to store length of destination string (or NULL)
- * @return Newly allocated destination string or NULL
+ * @return Newly allocated destination string or NULL on error
  *
  * The return value is always 0-terminated.  The value returned via @p *ndp
  * does not include the terminator.
@@ -222,69 +255,46 @@ error:
  */
 uint32_t *utf8_to_utf32(const char *s, size_t ns, size_t *ndp) {
   struct dynstr_ucs4 d;
-  uint32_t c32, c;
+  uint32_t c32;
   const uint8_t *ss = (const uint8_t *)s;
+  int n;
 
   dynstr_ucs4_init(&d);
   while(ns > 0) {
-    c = *ss++;
-    --ns;
-    /* Acceptable UTF-8 is that which codes for Unicode Scalar Values
-     * (Unicode 5.0.0 s3.9 D76)
-     *
-     * 0xxxxxxx
-     * 7 data bits gives 0x00 - 0x7F and all are acceptable
-     * 
-     * 110xxxxx 10xxxxxx
-     * 11 data bits gives 0x0000 - 0x07FF but only 0x0080 - 0x07FF acceptable
-     *   
-     * 1110xxxx 10xxxxxx 10xxxxxx
-     * 16 data bits gives 0x0000 - 0xFFFF but only 0x0800 - 0xFFFF acceptable
-     * (and UTF-16 surrogates are not acceptable)
-     *
-     * 11110xxx 10xxxxxx 10xxxxxx 10xxxxxx
-     * 21 data bits gives 0x00000000 - 0x001FFFFF
-     * but only           0x00010000 - 0x0010FFFF are acceptable
-     *
-     * It is NOT always the case that the data bits in the first byte are
-     * always non-0 for the acceptable values, so we do a separate check after
-     * decoding.
-     */
-    if(c < 0x80)
-      c32 = c;
-    else if(c <= 0xDF) {
-      if(ns < 1) goto error;
-      c32 = c & 0x1F;
-      c = *ss++;
-      if((c & 0xC0) != 0x80) goto error;
-      c32 = (c32 << 6) | (c & 0x3F);
-      if(c32 < 0x80) goto error;
-    } else if(c <= 0xEF) {
-      if(ns < 2) goto error;
-      c32 = c & 0x0F;
-      c = *ss++;
-      if((c & 0xC0) != 0x80) goto error;
-      c32 = (c32 << 6) | (c & 0x3F);
-      c = *ss++;
-      if((c & 0xC0) != 0x80) goto error;
-      c32 = (c32 << 6) | (c & 0x3F);
-      if(c32 < 0x0800 || (c32 >= 0xD800 && c32 <= 0xDFFF)) goto error;
-    } else if(c <= 0xF7) {
-      if(ns < 3) goto error;
-      c32 = c & 0x07;
-      c = *ss++;
-      if((c & 0xC0) != 0x80) goto error;
-      c32 = (c32 << 6) | (c & 0x3F);
-      c = *ss++;
-      if((c & 0xC0) != 0x80) goto error;
-      c32 = (c32 << 6) | (c & 0x3F);
-      c = *ss++;
-      if((c & 0xC0) != 0x80) goto error;
-      c32 = (c32 << 6) | (c & 0x3F);
-      if(c32 < 0x00010000 || c32 > 0x0010FFFF) goto error;
+    const struct unicode_utf8_row *const r = &unicode_utf8_valid[*ss];
+    if(r->count <= ns) {
+      switch(r->count) {
+      case 1:
+        c32 = *ss;
+        break;
+      case 2:
+        if(ss[1] < r->min2 || ss[1] > r->max2)
+          goto error;
+        c32 = *ss & 0x1F;
+        break;
+      case 3:
+        if(ss[1] < r->min2 || ss[1] > r->max2)
+          goto error;
+        c32 = *ss & 0x0F;
+        break;
+      case 4:
+        if(ss[1] < r->min2 || ss[1] > r->max2)
+          goto error;
+        c32 = *ss & 0x07;
+        break;
+      default:
+        goto error;
+      }
     } else
       goto error;
+    for(n = 1; n < r->count; ++n) {
+      if(ss[n] < 0x80 || ss[n] > 0xBF)
+        goto error;
+      c32 = (c32 << 6) | (ss[n] & 0x3F);
+    }
     dynstr_ucs4_append(&d, c32);
+    ss += r->count;
+    ns -= r->count;
   }
   dynstr_ucs4_terminate(&d);
   if(ndp)
@@ -366,21 +376,10 @@ struct utf32_iterator_data {
    * the value is (uint32_t)-1.
    */
   uint32_t last[2];
-};
 
-/** @brief Create a new iterator pointing at the start of a string
- * @param s Start of string
- * @param ns Length of string
- * @return New iterator
- */
-utf32_iterator utf32_iterator_new(const uint32_t *s, size_t ns) {
-  utf32_iterator it = xmalloc(sizeof *it);
-  it->s = s;
-  it->ns = ns;
-  it->n = 0;
-  it->last[0] = it->last[1] = -1;
-  return it;
-}
+  /** @brief Tailoring for Word_Break */
+  unicode_property_tailor *word_break;
+};
 
 /** @brief Initialize an internal private iterator
  * @param it Iterator
@@ -394,7 +393,52 @@ static void utf32__iterator_init(utf32_iterator it,
   it->ns = ns;
   it->n = 0;
   it->last[0] = it->last[1] = -1;
-  utf32_iterator_advance(it, n);
+  it->word_break = 0;
+  utf32_iterator_set(it, n);
+}
+
+/** @brief Create a new iterator pointing at the start of a string
+ * @param s Start of string
+ * @param ns Length of string
+ * @return New iterator
+ */
+utf32_iterator utf32_iterator_new(const uint32_t *s, size_t ns) {
+  utf32_iterator it = xmalloc(sizeof *it);
+  utf32__iterator_init(it, s, ns, 0);
+  return it;
+}
+
+/** @brief Tailor this iterator's interpretation of the Word_Break property.
+ * @param it Iterator
+ * @param pt Property tailor function or NULL
+ *
+ * After calling this the iterator will call @p pt to determine the Word_Break
+ * property of each code point.  If it returns -1 the default value will be
+ * used otherwise the returned value will be used.
+ *
+ * @p pt can be NULL to revert to the default value of the property.
+ *
+ * It is safe to call this function at any time; the iterator's internal state
+ * will be reset to suit the new tailoring.
+ */
+void utf32_iterator_tailor_word_break(utf32_iterator it,
+                                      unicode_property_tailor *pt) {
+  it->word_break = pt;
+  utf32_iterator_set(it, it->n);
+}
+
+static inline enum unicode_Word_Break utf32__iterator_word_break(utf32_iterator it,
+                                                                 uint32_t c) {
+  if(!it->word_break)
+    return utf32__word_break(c);
+  else {
+    const int t = it->word_break(c);
+
+    if(t < 0)
+      return utf32__word_break(c);
+    else
+      return t;
+  }
 }
 
 /** @brief Destroy an iterator
@@ -419,19 +463,45 @@ size_t utf32_iterator_where(utf32_iterator it) {
  * It is an error to position the iterator outside the string (but acceptable
  * to point it at the hypothetical post-final character).  If an invalid value
  * of @p n is specified then the iterator is not changed.
+ *
+ * This function works by backing up and then advancing to reconstruct the
+ * iterator's internal state for position @p n.  The worst case will be O(n)
+ * time complexity (with a worse constant factor that utf32_iterator_advance())
+ * but the typical case is essentially constant-time.
  */
 int utf32_iterator_set(utf32_iterator it, size_t n) {
-  /* TODO figure out how far we must back up to be able to re-synchronize; see
-   * UAX #29 s6.4. */
-  if(n > it->ns)
+  /* We can't just jump to position @p n; the @p last[] values will be wrong.
+   * What we need is to jump a bit behind @p n and then advance forward,
+   * updating @p last[] along the way.  How far back?  We need to cross two
+   * non-ignorable code points as we advance forwards, so we'd better pass two
+   * such characters on the way back (if such are available).
+   */
+  size_t m;
+
+  if(n > it->ns)                        /* range check */
     return -1;
-  if(n >= it->n)
-    n -= it->n;
-  else {
-    it->n = 0;
-    it->last[0] = it->last[1] = -1;
+  /* Walk backwards skipping ignorable code points */
+  m = n;
+  while(m > 0
+        && (utf32__boundary_ignorable(utf32__iterator_word_break(it,
+                                                                 it->s[m-1]))))
+    --m;
+  /* Either m=0 or s[m-1] is not ignorable */
+  if(m > 0) {
+    --m;
+    /* s[m] is our first non-ignorable code; look for a second in the same
+       way **/
+    while(m > 0
+          && (utf32__boundary_ignorable(utf32__iterator_word_break(it,
+                                                                   it->s[m-1]))))
+      --m;
+    /* Either m=0 or s[m-1] is not ignorable */
+    if(m > 0)
+      --m;
   }
-  return utf32_iterator_advance(it, n);
+  it->last[0] = it->last[1] = -1;
+  it->n = m;
+  return utf32_iterator_advance(it, n - m);
 }
 
 /** @brief Advance an iterator
@@ -450,7 +520,7 @@ int utf32_iterator_advance(utf32_iterator it, size_t count) {
   if(count <= it->ns - it->n) {
     while(count > 0) {
       const uint32_t c = it->s[it->n];
-      const enum unicode_Word_Break wb = utf32__word_break(c);
+      const enum unicode_Word_Break wb = utf32__iterator_word_break(it, c);
       if(it->last[1] == (uint32_t)-1
          || !utf32__boundary_ignorable(wb)) {
         it->last[0] = it->last[1];
@@ -482,6 +552,11 @@ uint32_t utf32_iterator_code(utf32_iterator it) {
 /** @brief Test for a grapheme boundary
  * @param it Iterator
  * @return Non-0 if pointing just after a grapheme boundary, otherwise 0
+ *
+ * This function identifies default grapheme cluster boundaries as described in
+ * UAX #29 s3.  It returns non-0 if @p it points at the code point just after a
+ * grapheme cluster boundary (including the hypothetical code point just after
+ * the end of the string).
  */
 int utf32_iterator_grapheme_boundary(utf32_iterator it) {
   uint32_t before, after;
@@ -528,6 +603,12 @@ int utf32_iterator_grapheme_boundary(utf32_iterator it) {
   /* GB9 */
   if(gbafter == unicode_Grapheme_Break_Extend)
     return 0;
+  /* GB9a */
+  if(gbafter == unicode_Grapheme_Break_SpacingMark)
+    return 0;
+  /* GB9b */
+  if(gbbefore == unicode_Grapheme_Break_Prepend)
+    return 0;
   /* GB10 */
   return 1;
 
@@ -536,96 +617,119 @@ int utf32_iterator_grapheme_boundary(utf32_iterator it) {
 /** @brief Test for a word boundary
  * @param it Iterator
  * @return Non-0 if pointing just after a word boundary, otherwise 0
+ *
+ * This function identifies default word boundaries as described in UAX #29 s4.
+ * It returns non-0 if @p it points at the code point just after a word
+ * boundary (including the hypothetical code point just after the end of the
+ * string) and 0 otherwise.
  */
 int utf32_iterator_word_boundary(utf32_iterator it) {
-  enum unicode_Word_Break twobefore, before, after, twoafter;
+  uint32_t before, after;
+  enum unicode_Word_Break wbtwobefore, wbbefore, wbafter, wbtwoafter;
   size_t nn;
 
   /* WB1 and WB2 */
   if(it->n == 0 || it->n == it->ns)
     return 1;
+  before = it->s[it->n-1];
+  after = it->s[it->n];
   /* WB3 */
-  if(it->s[it->n-1] == 0x000D && it->s[it->n] == 0x000A)
+  if(before == 0x000D && after == 0x000A)
     return 0;
+  /* WB3a */
+  if(utf32__iterator_word_break(it, before) == unicode_Word_Break_Newline
+     || before == 0x000D
+     || before == 0x000A)
+    return 1;
+  /* WB3b */
+  if(utf32__iterator_word_break(it, after) == unicode_Word_Break_Newline
+     || after == 0x000D
+     || after == 0x000A)
+    return 1;
   /* WB4 */
   /* (!Sep) x (Extend|Format) as in UAX #29 s6.2 */
-  if(utf32__sentence_break(it->s[it->n-1]) != unicode_Sentence_Break_Sep
-     && utf32__boundary_ignorable(utf32__word_break(it->s[it->n])))
+  if(utf32__sentence_break(before) != unicode_Sentence_Break_Sep
+     && utf32__boundary_ignorable(utf32__iterator_word_break(it, after)))
     return 0;
   /* Gather the property values we'll need for the rest of the test taking the
    * s6.2 changes into account */
   /* First we look at the code points after the proposed boundary */
   nn = it->n;                           /* <it->ns */
-  after = utf32__word_break(it->s[nn++]);
-  if(!utf32__boundary_ignorable(after)) {
+  wbafter = utf32__iterator_word_break(it, it->s[nn++]);
+  if(!utf32__boundary_ignorable(wbafter)) {
     /* X (Extend|Format)* -> X */
     while(nn < it->ns
-          && utf32__boundary_ignorable(utf32__word_break(it->s[nn])))
+          && utf32__boundary_ignorable(utf32__iterator_word_break(it,
+                                                                  it->s[nn])))
       ++nn;
   }
   /* It's possible now that nn=ns */
   if(nn < it->ns)
-    twoafter = utf32__word_break(it->s[nn]);
+    wbtwoafter = utf32__iterator_word_break(it, it->s[nn]);
   else
-    twoafter = unicode_Word_Break_Other;
+    wbtwoafter = unicode_Word_Break_Other;
 
   /* We've already recorded the non-ignorable code points before the proposed
    * boundary */
-  before = utf32__word_break(it->last[1]);
-  twobefore = utf32__word_break(it->last[0]);
+  wbbefore = utf32__iterator_word_break(it, it->last[1]);
+  wbtwobefore = utf32__iterator_word_break(it, it->last[0]);
 
   /* WB5 */
-  if(before == unicode_Word_Break_ALetter
-     && after == unicode_Word_Break_ALetter)
+  if(wbbefore == unicode_Word_Break_ALetter
+     && wbafter == unicode_Word_Break_ALetter)
     return 0;
   /* WB6 */
-  if(before == unicode_Word_Break_ALetter
-     && after == unicode_Word_Break_MidLetter
-     && twoafter == unicode_Word_Break_ALetter)
+  if(wbbefore == unicode_Word_Break_ALetter
+     && (wbafter == unicode_Word_Break_MidLetter
+         || wbafter == unicode_Word_Break_MidNumLet)
+     && wbtwoafter == unicode_Word_Break_ALetter)
     return 0;
   /* WB7 */
-  if(twobefore == unicode_Word_Break_ALetter
-     && before == unicode_Word_Break_MidLetter
-     && after == unicode_Word_Break_ALetter)
+  if(wbtwobefore == unicode_Word_Break_ALetter
+     && (wbbefore == unicode_Word_Break_MidLetter
+         || wbbefore == unicode_Word_Break_MidNumLet)
+     && wbafter == unicode_Word_Break_ALetter)
     return 0;
-  /* WB8 */  
-  if(before == unicode_Word_Break_Numeric
-     && after == unicode_Word_Break_Numeric)
+  /* WB8 */
+  if(wbbefore == unicode_Word_Break_Numeric
+     && wbafter == unicode_Word_Break_Numeric)
     return 0;
   /* WB9 */
-  if(before == unicode_Word_Break_ALetter
-     && after == unicode_Word_Break_Numeric)
+  if(wbbefore == unicode_Word_Break_ALetter
+     && wbafter == unicode_Word_Break_Numeric)
     return 0;
   /* WB10 */
-  if(before == unicode_Word_Break_Numeric
-     && after == unicode_Word_Break_ALetter)
+  if(wbbefore == unicode_Word_Break_Numeric
+     && wbafter == unicode_Word_Break_ALetter)
     return 0;
    /* WB11 */
-  if(twobefore == unicode_Word_Break_Numeric
-     && before == unicode_Word_Break_MidNum
-     && after == unicode_Word_Break_Numeric)
+  if(wbtwobefore == unicode_Word_Break_Numeric
+     && (wbbefore == unicode_Word_Break_MidNum
+         || wbbefore == unicode_Word_Break_MidNumLet)
+     && wbafter == unicode_Word_Break_Numeric)
     return 0;
   /* WB12 */
-  if(before == unicode_Word_Break_Numeric
-     && after == unicode_Word_Break_MidNum
-     && twoafter == unicode_Word_Break_Numeric)
+  if(wbbefore == unicode_Word_Break_Numeric
+     && (wbafter == unicode_Word_Break_MidNum
+         || wbafter == unicode_Word_Break_MidNumLet)
+     && wbtwoafter == unicode_Word_Break_Numeric)
     return 0;
   /* WB13 */
-  if(before == unicode_Word_Break_Katakana
-     && after == unicode_Word_Break_Katakana)
+  if(wbbefore == unicode_Word_Break_Katakana
+     && wbafter == unicode_Word_Break_Katakana)
     return 0;
   /* WB13a */
-  if((before == unicode_Word_Break_ALetter
-      || before == unicode_Word_Break_Numeric
-      || before == unicode_Word_Break_Katakana
-      || before == unicode_Word_Break_ExtendNumLet)
-     && after == unicode_Word_Break_ExtendNumLet)
+  if((wbbefore == unicode_Word_Break_ALetter
+      || wbbefore == unicode_Word_Break_Numeric
+      || wbbefore == unicode_Word_Break_Katakana
+      || wbbefore == unicode_Word_Break_ExtendNumLet)
+     && wbafter == unicode_Word_Break_ExtendNumLet)
     return 0;
   /* WB13b */
-  if(before == unicode_Word_Break_ExtendNumLet
-     && (after == unicode_Word_Break_ALetter
-         || after == unicode_Word_Break_Numeric
-         || after == unicode_Word_Break_Katakana))
+  if(wbbefore == unicode_Word_Break_ExtendNumLet
+     && (wbafter == unicode_Word_Break_ALetter
+         || wbafter == unicode_Word_Break_Numeric
+         || wbafter == unicode_Word_Break_Katakana))
     return 0;
   /* WB14 */
   return 1;
@@ -680,7 +784,7 @@ static void utf32__sort_ccc(uint32_t *s, size_t ns, uint32_t *buffer) {
     /* Merge them back into one, via the buffer */
     bp = buffer;
     while(na > 0 && nb > 0) {
-      /* We want descending order of combining class (hence <)
+      /* We want ascending order of combining class (hence <)
        * and we want stability within combining classes (hence <=)
        */
       if(utf32__combining_class(*a) <= utf32__combining_class(*b)) {
@@ -707,13 +811,13 @@ static void utf32__sort_ccc(uint32_t *s, size_t ns, uint32_t *buffer) {
 /** @brief Put combining characters into canonical order
  * @param s Pointer to UTF-32 string
  * @param ns Length of @p s
- * @return 0 on success, -1 on error
+ * @return 0 on success, non-0 on error
  *
  * @p s is modified in-place.  See Unicode 5.0 s3.11 for details of the
  * ordering.
  *
  * Currently we only support a maximum of 1024 combining characters after each
- * base character.  If this limit is exceeded then -1 is returned.
+ * base character.  If this limit is exceeded then a non-0 value is returned.
  */
 static int utf32__canonical_ordering(uint32_t *s, size_t ns) {
   size_t nc;
@@ -755,7 +859,7 @@ static int utf32__canonical_ordering(uint32_t *s, size_t ns) {
 
 /** @brief Guts of the decomposition lookup functions */
 #define utf32__decompose_one_generic(WHICH) do {                        \
-  const uint32_t *dc = utf32__unidata(c)->WHICH;                        \
+  const uint32_t *dc = utf32__decomposition_##WHICH(c);	                \
   if(dc) {                                                              \
     /* Found a canonical decomposition in the table */                  \
     while(*dc)                                                          \
@@ -778,7 +882,7 @@ static int utf32__canonical_ordering(uint32_t *s, size_t ns) {
 /** @brief Recursively compute the canonical decomposition of @p c
  * @param d Dynamic string to store decomposition in
  * @param c Code point to decompose (must be a valid!)
- * @return 0 on success, -1 on error
+ * @return 0 on success, non-0 on error
  */
 static void utf32__decompose_one_canon(struct dynstr_ucs4 *d, uint32_t c) {
   utf32__decompose_one_generic(canon);
@@ -787,14 +891,159 @@ static void utf32__decompose_one_canon(struct dynstr_ucs4 *d, uint32_t c) {
 /** @brief Recursively compute the compatibility decomposition of @p c
  * @param d Dynamic string to store decomposition in
  * @param c Code point to decompose (must be a valid!)
- * @return 0 on success, -1 on error
+ * @return 0 on success, non-0 on error
  */
 static void utf32__decompose_one_compat(struct dynstr_ucs4 *d, uint32_t c) {
   utf32__decompose_one_generic(compat);
 }
 
-/** @brief Guts of the decomposition functions */
-#define utf32__decompose_generic(WHICH) do {            \
+/** @brief Magic utf32__compositions() return value for Hangul Choseong */
+static const uint32_t utf32__hangul_L[1];
+
+/** @brief Return the list of compositions that @p c starts
+ * @param c Starter code point
+ * @return Composition list or NULL
+ *
+ * For Hangul leading (Choseong) jamo we return the special value
+ * utf32__hangul_L.  These code points are not listed as the targets of
+ * canonical decompositions (make-unidata checks) so there is no confusion with
+ * real decompositions here.
+ */
+static const uint32_t *utf32__compositions(uint32_t c) {
+  const uint32_t *compositions = utf32__unidata(c)->composed;
+
+  if(compositions)
+    return compositions;
+  /* Special-casing for Hangul */
+  switch(utf32__grapheme_break(c)) {
+  default:
+    return 0;
+  case unicode_Grapheme_Break_L:
+    return utf32__hangul_L;
+  }
+}
+
+/** @brief Composition step
+ * @param s Start of string
+ * @param ns Length of string
+ * @return New length of string
+ *
+ * This is called from utf32__decompose_generic() to compose the result string
+ * in place.
+ */
+static size_t utf32__compose(uint32_t *s, size_t ns) {
+  const uint32_t *compositions;
+  uint32_t *start = s, *t = s, *tt, cc;
+
+  while(ns > 0) {
+    uint32_t starter = *s++;
+    int block_starters = 0;
+    --ns;
+    /* We don't attempt to compose the following things:
+     * - final characters whatever kind they are
+     * - non-starter characters
+     * - starters that don't take part in a canonical decomposition mapping
+     */
+    if(ns == 0
+       || utf32__combining_class(starter)
+       || !(compositions = utf32__compositions(starter))) {
+      *t++ = starter;
+      continue;
+    }
+    if(compositions != utf32__hangul_L) {
+      /* Where we'll put the eventual starter */
+      tt = t++;
+      do {
+        /* See if we can find composition of starter+*s */
+        const uint32_t cchar = *s, *cp = compositions;
+        while((cc = *cp++)) {
+          const uint32_t *decomp = utf32__decomposition_canon(cc);
+          /* We know decomp[0] == starter */
+          if(decomp[1] == cchar)
+            break;
+        }
+        if(cc) {
+          /* Found a composition: cc decomposes to starter,*s */
+          starter = cc;
+          compositions = utf32__compositions(starter);
+          ++s;
+          --ns;
+        } else {
+          /* No composition found. */
+          const int class = utf32__combining_class(*s);
+          if(class) {
+            /* Transfer the uncomposable combining character to the output */
+            *t++ = *s++;
+            --ns;
+            /* All the combining characters of the same class of the
+             * uncomposable character are blocked by it, but there may be
+             * others of higher class later.  We eat the uncomposable and
+             * blocked characters and go back round the loop for that higher
+             * class. */
+            while(ns > 0 && utf32__combining_class(*s) == class) {
+              *t++ = *s++;
+              --ns;
+            }
+            /* Block any subsequent starters */
+            block_starters = 1;
+          } else {
+            /* The uncombinable character is itself a starter, so we don't
+             * transfer it to the output but instead go back round the main
+             * loop. */
+            break;
+          }
+        }
+        /* Keep going while there are still characters and the starter takes
+         * part in some composition */
+      } while(ns > 0 && compositions
+              && (!block_starters || utf32__combining_class(*s)));
+      /* Store any remaining combining characters */
+      while(ns > 0 && utf32__combining_class(*s)) {
+        *t++ = *s++;
+        --ns;
+      }
+      /* Store the resulting starter */
+      *tt = starter;
+    } else {
+      /* Special-casing for Hangul
+       *
+       * If there are combining characters between the L and the V then they
+       * will block the V and so no composition happens.  Similarly combining
+       * characters between V and T will block the T and so we only get as far
+       * as LV.
+       */
+      if(utf32__grapheme_break(*s) == unicode_Grapheme_Break_V) {
+        const uint32_t V = *s++;
+        const uint32_t LIndex = starter - LBase;
+        const uint32_t VIndex = V - VBase;
+        uint32_t TIndex;
+        --ns;
+        if(ns > 0
+           && utf32__grapheme_break(*s) == unicode_Grapheme_Break_T) {
+          /* We have an L V T sequence */
+          const uint32_t T = *s++;
+          TIndex = T - TBase;
+          --ns;
+        } else
+          /* It's just L V */
+          TIndex = 0;
+        /* Compose to LVT or LV as appropriate */
+        starter = (LIndex * VCount + VIndex) * TCount + TIndex + SBase;
+      } /* else we only have L or LV and no V or T */
+      *t++ = starter;
+      /* There could be some combining characters that belong to the V or T.
+       * These will be treated as non-starter characters at the top of the loop
+       * and thuss transferred to the output. */
+    }
+  }
+  return t - start;
+}
+
+/** @brief Guts of the composition and decomposition functions
+ * @param WHICH @c canon or @c compat to choose decomposition
+ * @param COMPOSE @c 0 or @c 1 to compose
+ */
+#define utf32__decompose_generic(WHICH, COMPOSE) do {   \
   struct dynstr_ucs4 d;                                 \
   uint32_t c;                                           \
                                                         \
@@ -808,6 +1057,8 @@ static void utf32__decompose_one_compat(struct dynstr_ucs4 *d, uint32_t c) {
   }                                                     \
   if(utf32__canonical_ordering(d.vec, d.nvec))          \
     goto error;                                         \
+  if(COMPOSE)                                           \
+    d.nvec = utf32__compose(d.vec, d.nvec);             \
   dynstr_ucs4_terminate(&d);                            \
   if(ndp)                                               \
     *ndp = d.nvec;                                      \
@@ -821,38 +1072,89 @@ error:                                                  \
  * @param s Pointer to string
  * @param ns Length of string
  * @param ndp Where to store length of result
- * @return Pointer to result string, or NULL
+ * @return Pointer to result string, or NULL on error
  *
- * Computes the canonical decomposition of a string and stably sorts combining
- * characters into canonical order.  The result is in Normalization Form D and
- * (at the time of writing!) passes the NFD tests defined in Unicode 5.0's
- * NormalizationTest.txt.
+ * Computes NFD (Normalization Form D) of the string at @p s.  This implies
+ * performing all canonical decompositions and then normalizing the order of
+ * combining characters.
  *
  * Returns NULL if the string is not valid for either of the following reasons:
  * - it codes for a UTF-16 surrogate
  * - it codes for a value outside the unicode code space
+ *
+ * See also:
+ * - utf32_decompose_compat()
+ * - utf32_compose_canon()
  */
 uint32_t *utf32_decompose_canon(const uint32_t *s, size_t ns, size_t *ndp) {
-  utf32__decompose_generic(canon);
+  utf32__decompose_generic(canon, 0);
 }
 
 /** @brief Compatibility decompose @p [s,s+ns)
  * @param s Pointer to string
  * @param ns Length of string
  * @param ndp Where to store length of result
- * @return Pointer to result string, or NULL
+ * @return Pointer to result string, or NULL on error
  *
- * Computes the compatibility decomposition of a string and stably sorts
- * combining characters into canonical order.  The result is in Normalization
- * Form KD and (at the time of writing!) passes the NFKD tests defined in
- * Unicode 5.0's NormalizationTest.txt.
+ * Computes NFKD (Normalization Form KD) of the string at @p s.  This implies
+ * performing all canonical and compatibility decompositions and then
+ * normalizing the order of combining characters.
  *
  * Returns NULL if the string is not valid for either of the following reasons:
  * - it codes for a UTF-16 surrogate
  * - it codes for a value outside the unicode code space
+ *
+ * See also:
+ * - utf32_decompose_canon()
+ * - utf32_compose_compat()
  */
 uint32_t *utf32_decompose_compat(const uint32_t *s, size_t ns, size_t *ndp) {
-  utf32__decompose_generic(compat);
+  utf32__decompose_generic(compat, 0);
+}
+
+/** @brief Canonically compose @p [s,s+ns)
+ * @param s Pointer to string
+ * @param ns Length of string
+ * @param ndp Where to store length of result
+ * @return Pointer to result string, or NULL on error
+ *
+ * Computes NFC (Normalization Form C) of the string at @p s.  This implies
+ * performing all canonical decompositions, normalizing the order of combining
+ * characters and then composing all unblocked primary compositables.
+ *
+ * Returns NULL if the string is not valid for either of the following reasons:
+ * - it codes for a UTF-16 surrogate
+ * - it codes for a value outside the unicode code space
+ *
+ * See also:
+ * - utf32_compose_compat()
+ * - utf32_decompose_canon()
+ */
+uint32_t *utf32_compose_canon(const uint32_t *s, size_t ns, size_t *ndp) {
+  utf32__decompose_generic(canon, 1);
+}
+
+/** @brief Compatibility compose @p [s,s+ns)
+ * @param s Pointer to string
+ * @param ns Length of string
+ * @param ndp Where to store length of result
+ * @return Pointer to result string, or NULL on error
+ *
+ * Computes NFKC (Normalization Form KC) of the string at @p s.  This implies
+ * performing all canonical and compatibility decompositions, normalizing the
+ * order of combining characters and then composing all unblocked primary
+ * compositables.
+ *
+ * Returns NULL if the string is not valid for either of the following reasons:
+ * - it codes for a UTF-16 surrogate
+ * - it codes for a value outside the unicode code space
+ *
+ * See also:
+ * - utf32_compose_canon()
+ * - utf32_decompose_compat()
+ */
+uint32_t *utf32_compose_compat(const uint32_t *s, size_t ns, size_t *ndp) {
+  utf32__decompose_generic(compat, 1);
 }
 
 /** @brief Single-character case-fold and decompose operation */
@@ -870,7 +1172,7 @@ uint32_t *utf32_decompose_compat(const uint32_t *s, size_t ns, size_t *ndp) {
  * @param s Pointer to string
  * @param ns Length of string
  * @param ndp Where to store length of result
- * @return Pointer to result string, or NULL
+ * @return Pointer to result string, or NULL on error
  *
  * Case-fold the string at @p s according to full default case-folding rules
  * (s3.13) for caseless matching.  The result will be in NFD.
@@ -919,11 +1221,11 @@ error:
   return 0;
 }
 
-/** @brief Compatibilit case-fold @p [s,s+ns)
+/** @brief Compatibility case-fold @p [s,s+ns)
  * @param s Pointer to string
  * @param ns Length of string
  * @param ndp Where to store length of result
- * @return Pointer to result string, or NULL
+ * @return Pointer to result string, or NULL on error
  *
  * Case-fold the string at @p s according to full default case-folding rules
  * (s3.13) for compatibility caseless matching.  The result will be in NFKD.
@@ -1001,9 +1303,12 @@ int utf32_cmp(const uint32_t *a, const uint32_t *b) {
  * @return 1 at a grapheme cluster boundary, 0 otherwise
  *
  * This function identifies default grapheme cluster boundaries as described in
- * UAX #29 s3.  It returns 1 if @p n points at the code point just after a
+ * UAX #29 s3.  It returns non-0 if @p n points at the code point just after a
  * grapheme cluster boundary (including the hypothetical code point just after
  * the end of the string).
+ *
+ * This function uses utf32_iterator_set() internally; see that function for
+ * remarks on performance.
  */
 int utf32_is_grapheme_boundary(const uint32_t *s, size_t ns, size_t n) {
   struct utf32_iterator_data it[1];
@@ -1019,8 +1324,11 @@ int utf32_is_grapheme_boundary(const uint32_t *s, size_t ns, size_t n) {
  * @return 1 at a word boundary, 0 otherwise
  *
  * This function identifies default word boundaries as described in UAX #29 s4.
- * It returns 1 if @p n points at the code point just after a word boundary
+ * It returns non-0 if @p n points at the code point just after a word boundary
  * (including the hypothetical code point just after the end of the string).
+ *
+ * This function uses utf32_iterator_set() internally; see that function for
+ * remarks on performance.
  */
 int utf32_is_word_boundary(const uint32_t *s, size_t ns, size_t n) {
   struct utf32_iterator_data it[1];
@@ -1029,6 +1337,64 @@ int utf32_is_word_boundary(const uint32_t *s, size_t ns, size_t n) {
   return utf32_iterator_word_boundary(it);
 }
 
+/** @brief Split [s,ns) into multiple words
+ * @param s Pointer to start of string
+ * @param ns Length of string
+ * @param nwp Where to store word count, or NULL
+ * @param wbreak Word_Break property tailor, or NULL
+ * @return Pointer to array of pointers to words
+ *
+ * The returned array is terminated by a NULL pointer and individual
+ * strings are 0-terminated.
+ */
+uint32_t **utf32_word_split(const uint32_t *s, size_t ns, size_t *nwp,
+                            unicode_property_tailor *wbreak) {
+  struct utf32_iterator_data it[1];
+  size_t b1 = 0, b2 = 0 ,i;
+  int isword;
+  struct vector32 v32[1];
+  uint32_t *w;
+
+  vector32_init(v32);
+  utf32__iterator_init(it, s, ns, 0);
+  it->word_break = wbreak;
+  /* Work our way through the string stopping at each word break. */
+  do {
+    if(utf32_iterator_word_boundary(it)) {
+      /* We've found a new boundary */
+      b1 = b2;
+      b2 = it->n;
+      /*fprintf(stderr, "[%zu, %zu) is a candidate word\n", b1, b2);*/
+      /* Inspect the characters between the boundary and form an opinion as to
+       * whether they are a word or not */
+      isword = 0;
+      for(i = b1; i < b2; ++i) {
+        switch(utf32__iterator_word_break(it, it->s[i])) {
+        case unicode_Word_Break_ALetter:
+        case unicode_Word_Break_Numeric:
+        case unicode_Word_Break_Katakana:
+          isword = 1;
+          break;
+        default:
+          break;
+        }
+      }
+      /* If it's a word add it to the list of results */
+      if(isword) {
+        const size_t len = b2 - b1;
+        w = xcalloc_noptr(len + 1, sizeof(uint32_t));
+        memcpy(w, it->s + b1, len * sizeof (uint32_t));
+        w[len] = 0;
+        vector32_append(v32, w);
+      }
+    }
+  } while(!utf32_iterator_advance(it, 1));
+  vector32_terminate(v32);
+  if(nwp)
+    *nwp = v32->nvec;
+  return v32->vec;
+}
+
 /*@}*/
 /** @defgroup utf8 Functions that operate on UTF-8 strings */
 /*@{*/
@@ -1052,17 +1418,19 @@ error:                                                          \
  * @param s Pointer to string
  * @param ns Length of string
  * @param ndp Where to store length of result
- * @return Pointer to result string, or NULL
+ * @return Pointer to result string, or NULL on error
  *
- * Computes the canonical decomposition of a string and stably sorts combining
- * characters into canonical order.  The result is in Normalization Form D and
- * (at the time of writing!) passes the NFD tests defined in Unicode 5.0's
- * NormalizationTest.txt.
+ * Computes NFD (Normalization Form D) of the string at @p s.  This implies
+ * performing all canonical decompositions and then normalizing the order of
+ * combining characters.
  *
  * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why
  * this might be.
  *
- * See also utf32_decompose_canon().
+ * See also:
+ * - utf32_decompose_canon().
+ * - utf8_decompose_compat()
+ * - utf8_compose_canon()
  */
 char *utf8_decompose_canon(const char *s, size_t ns, size_t *ndp) {
   utf8__transform(utf32_decompose_canon);
@@ -1072,27 +1440,74 @@ char *utf8_decompose_canon(const char *s, size_t ns, size_t *ndp) {
  * @param s Pointer to string
  * @param ns Length of string
  * @param ndp Where to store length of result
- * @return Pointer to result string, or NULL
+ * @return Pointer to result string, or NULL on error
  *
- * Computes the compatibility decomposition of a string and stably sorts
- * combining characters into canonical order.  The result is in Normalization
- * Form KD and (at the time of writing!) passes the NFKD tests defined in
- * Unicode 5.0's NormalizationTest.txt.
+ * Computes NFKD (Normalization Form KD) of the string at @p s.  This implies
+ * performing all canonical and compatibility decompositions and then
+ * normalizing the order of combining characters.
  *
  * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why
  * this might be.
  *
- * See also utf32_decompose_compat().
+ * See also:
+ * - utf32_decompose_compat().
+ * - utf8_decompose_canon()
+ * - utf8_compose_compat()
  */
 char *utf8_decompose_compat(const char *s, size_t ns, size_t *ndp) {
   utf8__transform(utf32_decompose_compat);
 }
 
+/** @brief Canonically compose @p [s,s+ns)
+ * @param s Pointer to string
+ * @param ns Length of string
+ * @param ndp Where to store length of result
+ * @return Pointer to result string, or NULL on error
+ *
+ * Computes NFC (Normalization Form C) of the string at @p s.  This implies
+ * performing all canonical decompositions, normalizing the order of combining
+ * characters and then composing all unblocked primary compositables.
+ *
+ * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why
+ * this might be.
+ *
+ * See also:
+ * - utf32_compose_canon()
+ * - utf8_compose_compat()
+ * - utf8_decompose_canon()
+ */
+char *utf8_compose_canon(const char *s, size_t ns, size_t *ndp) {
+  utf8__transform(utf32_compose_canon);
+}
+
+/** @brief Compatibility compose @p [s,s+ns)
+ * @param s Pointer to string
+ * @param ns Length of string
+ * @param ndp Where to store length of result
+ * @return Pointer to result string, or NULL on error
+ *
+ * Computes NFKC (Normalization Form KC) of the string at @p s.  This implies
+ * performing all canonical and compatibility decompositions, normalizing the
+ * order of combining characters and then composing all unblocked primary
+ * compositables.
+ *
+ * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why
+ * this might be.
+ *
+ * See also:
+ * - utf32_compose_compat()
+ * - utf8_compose_canon()
+ * - utf8_decompose_compat()
+ */
+char *utf8_compose_compat(const char *s, size_t ns, size_t *ndp) {
+  utf8__transform(utf32_compose_compat);
+}
+
 /** @brief Case-fold @p [s,s+ns)
  * @param s Pointer to string
  * @param ns Length of string
  * @param ndp Where to store length of result
- * @return Pointer to result string, or NULL
+ * @return Pointer to result string, or NULL on error
  *
  * Case-fold the string at @p s according to full default case-folding rules
  * (s3.13).  The result will be in NFD.
@@ -1108,7 +1523,7 @@ char *utf8_casefold_canon(const char *s, size_t ns, size_t *ndp) {
  * @param s Pointer to string
  * @param ns Length of string
  * @param ndp Where to store length of result
- * @return Pointer to result string, or NULL
+ * @return Pointer to result string, or NULL on error
  *
  * Case-fold the string at @p s according to full default case-folding rules
  * (s3.13).  The result will be in NFKD.
@@ -1120,6 +1535,47 @@ char *utf8_casefold_compat(const char *s, size_t ns, size_t *ndp) {
   utf8__transform(utf32_casefold_compat);
 }
 
+/** @brief Split [s,ns) into multiple words
+ * @param s Pointer to start of string
+ * @param ns Length of string
+ * @param nwp Where to store word count, or NULL
+ * @param wbreak Word_Break property tailor, or NULL
+ * @return Pointer to array of pointers to words
+ *
+ * The returned array is terminated by a NULL pointer and individual
+ * strings are 0-terminated.
+ */
+char **utf8_word_split(const char *s, size_t ns, size_t *nwp,
+                       unicode_property_tailor *wbreak) {
+  uint32_t *to32 = 0, **v32 = 0;
+  size_t nto32, nv, n;
+  char **v8 = 0, **ret = 0;
+
+  if(!(to32 = utf8_to_utf32(s, ns, &nto32))) goto error;
+  if(!(v32 = utf32_word_split(to32, nto32, &nv, wbreak))) goto error;
+  v8 = xcalloc(sizeof (char *), nv + 1);
+  for(n = 0; n < nv; ++n)
+    if(!(v8[n] = utf32_to_utf8(v32[n], utf32_len(v32[n]), 0)))
+      goto error;
+  ret = v8;
+  *nwp = nv;
+  v8 = 0;                               /* don't free */
+error:
+  if(v8) {
+    for(n = 0; n < nv; ++n)
+      xfree(v8[n]);
+    xfree(v8);
+  }
+  if(v32) {
+    for(n = 0; n < nv; ++n)
+      xfree(v32[n]);
+    xfree(v32);
+  }
+  xfree(to32);
+  return ret;
+}
+
+
 /*@}*/
 
 /*