hb-ot-shape-normalize.cc revision af913c5788e600e36d29f44fe4e77db84cf8c442
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 30 31/* 32 * HIGHLEVEL DESIGN: 33 * 34 * This file exports one main function: _hb_ot_shape_normalize(). 35 * 36 * This function closely reflects the Unicode Normalization Algorithm, 37 * yet it's different. The shaper an either prefer decomposed (NFD) or 38 * composed (NFC). 39 * 40 * In general what happens is that: each grapheme is decomposed in a chain 41 * of 1:2 decompositions, marks reordered, and then recomposed if desired, 42 * so far it's like Unicode Normalization. However, the decomposition and 43 * recomposition only happens if the font supports the resulting characters. 44 * 45 * The goals are: 46 * 47 * - Try to render all canonically equivalent strings similarly. To really 48 * achieve this we have to always do the full decomposition and then 49 * selectively recompose from there. It's kinda too expensive though, so 50 * we skip some cases. For example, if composed is desired, we simply 51 * don't touch 1-character clusters that are supported by the font, even 52 * though their NFC may be different. 53 * 54 * - When a font has a precomposed character for a sequence but the 'ccmp' 55 * feature in the font is not adequate, use the precomposed character 56 * which typically has better mark positioning. 57 * 58 * - When a font does not support a combining mark, but supports it precomposed 59 * with previous base. This needs the itemizer to have this knowledge too. 60 * We need ot provide assistance to the itemizer. 61 * 62 * - When a font does not support a character but supports its decomposition, 63 * well, use the decomposition. 64 * 65 * - The Indic shaper requests decomposed output. This will handle splitting 66 * matra for the Indic shaper. 67 */ 68 69static void 70output_glyph (hb_ot_shape_context_t *c, 71 hb_codepoint_t glyph) 72{ 73 hb_buffer_t *buffer = c->buffer; 74 75 buffer->output_glyph (glyph); 76 hb_glyph_info_set_unicode_props (&buffer->out_info[buffer->out_len - 1], buffer->unicode); 77} 78 79static bool 80decompose (hb_ot_shape_context_t *c, 81 bool shortest, 82 hb_codepoint_t ab) 83{ 84 hb_codepoint_t a, b, glyph; 85 86 if (!hb_unicode_decompose (c->buffer->unicode, ab, &a, &b) || 87 (b && !hb_font_get_glyph (c->font, b, 0, &glyph))) 88 return FALSE; 89 90 bool has_a = hb_font_get_glyph (c->font, a, 0, &glyph); 91 if (shortest && has_a) { 92 /* Output a and b */ 93 output_glyph (c, a); 94 if (b) 95 output_glyph (c, b); 96 return TRUE; 97 } 98 99 if (decompose (c, shortest, a)) { 100 if (b) 101 output_glyph (c, b); 102 return TRUE; 103 } 104 105 if (has_a) { 106 output_glyph (c, a); 107 if (b) 108 output_glyph (c, b); 109 return TRUE; 110 } 111 112 return FALSE; 113} 114 115static void 116decompose_current_glyph (hb_ot_shape_context_t *c, 117 bool shortest) 118{ 119 if (decompose (c, shortest, c->buffer->info[c->buffer->idx].codepoint)) 120 c->buffer->skip_glyph (); 121 else 122 c->buffer->next_glyph (); 123} 124 125static void 126decompose_single_char_cluster (hb_ot_shape_context_t *c, 127 bool will_recompose) 128{ 129 hb_codepoint_t glyph; 130 131 /* If recomposing and font supports this, we're good to go */ 132 if (will_recompose && hb_font_get_glyph (c->font, c->buffer->info[c->buffer->idx].codepoint, 0, &glyph)) { 133 c->buffer->next_glyph (); 134 return; 135 } 136 137 decompose_current_glyph (c, will_recompose); 138} 139 140static void 141decompose_multi_char_cluster (hb_ot_shape_context_t *c, 142 unsigned int end) 143{ 144 /* TODO Currently if there's a variation-selector we give-up, it's just too hard. */ 145 for (unsigned int i = c->buffer->idx; i < end; i++) 146 if (unlikely (is_variation_selector (c->buffer->info[i].codepoint))) { 147 while (c->buffer->idx < end) 148 c->buffer->next_glyph (); 149 return; 150 } 151 152 while (c->buffer->idx < end) 153 decompose_current_glyph (c, FALSE); 154} 155 156static int 157compare_combining_class (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb) 158{ 159 unsigned int a = pa->combining_class(); 160 unsigned int b = pb->combining_class(); 161 162 return a < b ? -1 : a == b ? 0 : +1; 163} 164 165void 166_hb_ot_shape_normalize (hb_ot_shape_context_t *c) 167{ 168 hb_buffer_t *buffer = c->buffer; 169 bool recompose = !hb_ot_shape_complex_prefer_decomposed (c->plan->shaper); 170 bool has_multichar_clusters = FALSE; 171 unsigned int count; 172 173 /* We do a fairly straightforward yet custom normalization process in three 174 * separate rounds: decompose, reorder, recompose (if desired). Currently 175 * this makes two buffer swaps. We can make it faster by moving the last 176 * two rounds into the inner loop for the first round, but it's more readable 177 * this way. */ 178 179 180 /* First round, decompose */ 181 182 buffer->clear_output (); 183 count = buffer->len; 184 for (buffer->idx = 0; buffer->idx < count;) 185 { 186 unsigned int end; 187 for (end = buffer->idx + 1; end < count; end++) 188 if (buffer->info[buffer->idx].cluster != buffer->info[end].cluster) 189 break; 190 191 if (buffer->idx + 1 == end) 192 decompose_single_char_cluster (c, recompose); 193 else { 194 decompose_multi_char_cluster (c, end); 195 has_multichar_clusters = TRUE; 196 } 197 } 198 buffer->swap_buffers (); 199 200 201 /* Technically speaking, two characters with ccc=0 may combine. But all 202 * those cases are in languages that the indic module handles (which expects 203 * decomposed), or in Hangul jamo, which again, we want decomposed anyway. 204 * So we don't bother combining across cluster boundaries. This is a huge 205 * performance saver if the compose() callback is slow. 206 * 207 * TODO: Am I right about Hangul? If I am, we should add a Hangul module 208 * that requests decomposed. If for Hangul we end up wanting composed, we 209 * can do that in the Hangul module. 210 */ 211 212 if (!has_multichar_clusters) 213 return; /* Done! */ 214 215 216 /* Second round, reorder (inplace) */ 217 218 count = buffer->len; 219 for (unsigned int i = 0; i < count; i++) 220 { 221 if (buffer->info[i].combining_class() == 0) 222 continue; 223 224 unsigned int end; 225 for (end = i + 1; end < count; end++) 226 if (buffer->info[end].combining_class() == 0) 227 break; 228 229 /* We are going to do a bubble-sort. Only do this if the 230 * sequence is short. Doing it on long sequences can result 231 * in an O(n^2) DoS. */ 232 if (end - i > 10) { 233 i = end; 234 continue; 235 } 236 237 hb_bubble_sort (buffer->info + i, end - i, compare_combining_class); 238 239 i = end; 240 } 241 242 243 if (!recompose) 244 return; 245 246 /* Third round, recompose */ 247 248 /* As noted in the comment earlier, we don't try to combine 249 * ccc=0 chars with their previous Starter. */ 250 251 buffer->clear_output (); 252 count = buffer->len; 253 unsigned int starter = 0; 254 buffer->next_glyph (); 255 while (buffer->idx < count) 256 { 257 if (buffer->info[buffer->idx].combining_class() == 0) { 258 starter = buffer->out_len; 259 buffer->next_glyph (); 260 continue; 261 } 262 263 hb_codepoint_t composed, glyph; 264 if ((buffer->out_info[buffer->out_len - 1].combining_class() >= 265 buffer->info[buffer->idx].combining_class()) || 266 !hb_unicode_compose (c->buffer->unicode, 267 buffer->out_info[starter].codepoint, 268 buffer->info[buffer->idx].codepoint, 269 &composed) || 270 !hb_font_get_glyph (c->font, composed, 0, &glyph)) 271 { 272 /* Blocked, or doesn't compose. */ 273 buffer->next_glyph (); 274 continue; 275 } 276 277 /* Composes. Modify starter and carry on. */ 278 buffer->out_info[starter].codepoint = composed; 279 hb_glyph_info_set_unicode_props (&buffer->out_info[starter], buffer->unicode); 280 281 buffer->skip_glyph (); 282 } 283 buffer->swap_buffers (); 284 285} 286 287