hb-ot-shape-normalize.cc revision dcdc51cdc0ba9d9fb75f84dd5fa7a49aa0b24ea0 
1/*
2 * Copyright © 2011  Google, Inc.
3 *
4 *  This is part of HarfBuzz, a text shaping library.
5 *
6 * Permission is hereby granted, without written agreement and without
7 * license or royalty fees, to use, copy, modify, and distribute this
8 * software and its documentation for any purpose, provided that the
9 * above copyright notice and the following two paragraphs appear in
10 * all copies of this software.
11 *
12 * IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR
13 * DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES
14 * ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN
15 * IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
16 * DAMAGE.
17 *
18 * THE COPYRIGHT HOLDER SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING,
19 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
20 * FITNESS FOR A PARTICULAR PURPOSE.  THE SOFTWARE PROVIDED HEREUNDER IS
21 * ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO
22 * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
23 *
24 * Google Author(s): Behdad Esfahbod
25 */
26
27#include "hb-ot-shape-private.hh"
28#include "hb-ot-shape-complex-private.hh"
29
30HB_BEGIN_DECLS
31
32/*
33 * HIGHLEVEL DESIGN:
34 *
35 * This file exports one main function: _hb_ot_shape_normalize().
36 *
37 * This function closely reflects the Unicode Normalization Algorithm,
38 * yet it's different.  The shaper an either prefer decomposed (NFD) or
39 * composed (NFC).
40 *
41 * In general what happens is that: each grapheme is decomposed in a chain
42 * of 1:2 decompositions, marks reordered, and then recomposed if desires,
43 * so far it's like Unicode Normalization.  However, the decomposition and
44 * recomposition only happens if the font supports the resulting characters.
45 *
46 * The goals are:
47 *
48 *   - Try to render all canonically equivalent strings similarly.  To really
49 *     achieve this we have to always do the full decomposition and then
50 *     selectively recompose from there.  It's kinda too expensive though, so
51 *     we skip some cases.  For example, if composed is desired, we simply
52 *     don't touch 1-character clusters that are supported by the font, even
53 *     though their NFC may be different.
54 *
55 *   - When a font has a precomposed character for a sequence but the 'ccmp'
56 *     feature in the font is not adequate, form use the precomposed character
57 *     which typically has better mark positioning.
58 *
59 *   - When a font does not support a character but supports its decomposition,
60 *     well, use the decomposition.
61 *
62 *   - The Indic shaper requests decomposed output.  This will handle splitting
63 *     matra for the Indic shaper.
64 */
65
66
67static bool
68decompose (hb_ot_shape_context_t *c,
69	   bool shortest,
70	   hb_codepoint_t ab)
71{
72  hb_codepoint_t a, b, glyph;
73
74  if (!hb_unicode_decompose (c->buffer->unicode, ab, &a, &b) ||
75      (b && !hb_font_get_glyph (c->font, b, 0, &glyph)))
76    return FALSE;
77
78  bool has_a = hb_font_get_glyph (c->font, a, 0, &glyph);
79  if (shortest && has_a) {
80    /* Output a and b */
81    c->buffer->output_glyph (a);
82    if (b)
83      c->buffer->output_glyph (b);
84    return TRUE;
85  }
86
87  if (decompose (c, shortest, a)) {
88    if (b)
89      c->buffer->output_glyph (b);
90    return TRUE;
91  }
92
93  if (has_a) {
94    c->buffer->output_glyph (a);
95    if (b)
96      c->buffer->output_glyph (b);
97    return TRUE;
98  }
99
100  return FALSE;
101}
102
103static bool
104decompose_current_glyph (hb_ot_shape_context_t *c,
105			 bool shortest)
106{
107  if (decompose (c, shortest, c->buffer->info[c->buffer->idx].codepoint)) {
108    c->buffer->skip_glyph ();
109    return TRUE;
110  } else {
111    c->buffer->next_glyph ();
112    return FALSE;
113  }
114}
115
116static bool
117decompose_single_char_cluster (hb_ot_shape_context_t *c,
118			       bool will_recompose)
119{
120  hb_codepoint_t glyph;
121
122  /* If recomposing and font supports this, we're good to go */
123  if (will_recompose && hb_font_get_glyph (c->font, c->buffer->info[c->buffer->idx].codepoint, 0, &glyph)) {
124    c->buffer->next_glyph ();
125    return FALSE;
126  }
127
128  return decompose_current_glyph (c, will_recompose);
129}
130
131static bool
132decompose_multi_char_cluster (hb_ot_shape_context_t *c,
133			      unsigned int end)
134{
135  bool changed = FALSE;
136
137  /* TODO Currently if there's a variation-selector we give-up, it's just too hard. */
138  for (unsigned int i = c->buffer->idx; i < end; i++)
139    if (unlikely (is_variation_selector (c->buffer->info[i].codepoint)))
140      return changed;
141
142  while (c->buffer->idx < end)
143    changed |= decompose_current_glyph (c, FALSE);
144
145  return changed;
146}
147
148void
149_hb_ot_shape_normalize (hb_ot_shape_context_t *c)
150{
151  hb_buffer_t *buffer = c->buffer;
152  bool recompose = !hb_ot_shape_complex_prefer_decomposed (c->plan->shaper);
153  bool changed = FALSE;
154  bool has_multichar_clusters = FALSE;
155  unsigned int count;
156
157  buffer->clear_output ();
158
159
160  /* First round, decompose */
161
162  count = buffer->len;
163  for (buffer->idx = 0; buffer->idx < count;)
164  {
165    unsigned int end;
166    for (end = buffer->idx + 1; end < count; end++)
167      if (buffer->info[buffer->idx].cluster != buffer->info[end].cluster)
168        break;
169
170    if (buffer->idx + 1 == end)
171      changed |= decompose_single_char_cluster (c, recompose);
172    else {
173      changed |= decompose_multi_char_cluster (c, end);
174      has_multichar_clusters = TRUE;
175    }
176  }
177  buffer->swap_buffers ();
178
179
180  /* Technically speaking, two characters with ccc=0 may combine.  But all
181   * those cases are in languages that the indic module handles (which expects
182   * decomposed), or in Hangul jamo, which again, we want decomposed anyway.
183   * So we don't bother combining across cluster boundaries. */
184
185  if (!has_multichar_clusters)
186    return; /* Done! */
187
188  if (changed)
189    _hb_set_unicode_props (c->buffer); /* BUFFER: Set general_category and combining_class in var1 */
190
191
192  /* Second round, reorder (inplace) */
193
194  count = buffer->len;
195  for (unsigned int i = 0; i < count; i++)
196  {
197    if (buffer->info[i].combining_class() == 0)
198      continue;
199
200    unsigned int end;
201    for (end = i + 1; end < count; end++)
202      if (buffer->info[end].combining_class() == 0)
203        break;
204
205    /* We are going to do a bubble-sort.  Only do this if the
206     * sequence is short.  Doing it on long sequences can result
207     * in an O(n^2) DoS. */
208    if (end - i > 10) {
209      i = end;
210      continue;
211    }
212
213    unsigned int k = end - i - 1;
214    do {
215      hb_glyph_info_t *pinfo = buffer->info + i;
216      unsigned int new_k = 0;
217
218      for (unsigned int j = 0; j < k; j++)
219	if (pinfo[j].combining_class() > pinfo[j+1].combining_class()) {
220	  hb_glyph_info_t t;
221	  t = pinfo[j];
222	  pinfo[j] = pinfo[j + 1];
223	  pinfo[j + 1] = t;
224
225	  new_k = j;
226	}
227      k = new_k;
228    } while (k);
229
230    i = end;
231  }
232
233
234  /* Third round, recompose */
235
236  if (recompose) {
237
238
239  }
240
241}
242
243HB_END_DECLS
244