1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/gfx/harfbuzz/src/hb-ot-shape-normalize.cc Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,413 @@
1.4 +/*
1.5 + * Copyright © 2011,2012 Google, Inc.
1.6 + *
1.7 + * This is part of HarfBuzz, a text shaping library.
1.8 + *
1.9 + * Permission is hereby granted, without written agreement and without
1.10 + * license or royalty fees, to use, copy, modify, and distribute this
1.11 + * software and its documentation for any purpose, provided that the
1.12 + * above copyright notice and the following two paragraphs appear in
1.13 + * all copies of this software.
1.14 + *
1.15 + * IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR
1.16 + * DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES
1.17 + * ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN
1.18 + * IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
1.19 + * DAMAGE.
1.20 + *
1.21 + * THE COPYRIGHT HOLDER SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING,
1.22 + * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
1.23 + * FITNESS FOR A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS
1.24 + * ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO
1.25 + * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
1.26 + *
1.27 + * Google Author(s): Behdad Esfahbod
1.28 + */
1.29 +
1.30 +#include "hb-ot-shape-normalize-private.hh"
1.31 +#include "hb-ot-shape-complex-private.hh"
1.32 +#include "hb-ot-shape-private.hh"
1.33 +
1.34 +
1.35 +/*
1.36 + * HIGHLEVEL DESIGN:
1.37 + *
1.38 + * This file exports one main function: _hb_ot_shape_normalize().
1.39 + *
1.40 + * This function closely reflects the Unicode Normalization Algorithm,
1.41 + * yet it's different.
1.42 + *
1.43 + * Each shaper specifies whether it prefers decomposed (NFD) or composed (NFC).
1.44 + * The logic however tries to use whatever the font can support.
1.45 + *
1.46 + * In general what happens is that: each grapheme is decomposed in a chain
1.47 + * of 1:2 decompositions, marks reordered, and then recomposed if desired,
1.48 + * so far it's like Unicode Normalization. However, the decomposition and
1.49 + * recomposition only happens if the font supports the resulting characters.
1.50 + *
1.51 + * The goals are:
1.52 + *
1.53 + * - Try to render all canonically equivalent strings similarly. To really
1.54 + * achieve this we have to always do the full decomposition and then
1.55 + * selectively recompose from there. It's kinda too expensive though, so
1.56 + * we skip some cases. For example, if composed is desired, we simply
1.57 + * don't touch 1-character clusters that are supported by the font, even
1.58 + * though their NFC may be different.
1.59 + *
1.60 + * - When a font has a precomposed character for a sequence but the 'ccmp'
1.61 + * feature in the font is not adequate, use the precomposed character
1.62 + * which typically has better mark positioning.
1.63 + *
1.64 + * - When a font does not support a combining mark, but supports it precomposed
1.65 + * with previous base, use that. This needs the itemizer to have this
1.66 + * knowledge too. We need to provide assistance to the itemizer.
1.67 + *
1.68 + * - When a font does not support a character but supports its decomposition,
1.69 + * well, use the decomposition (preferring the canonical decomposition, but
1.70 + * falling back to the compatibility decomposition if necessary). The
1.71 + * compatibility decomposition is really nice to have, for characters like
1.72 + * ellipsis, or various-sized space characters.
1.73 + *
1.74 + * - The complex shapers can customize the compose and decompose functions to
1.75 + * offload some of their requirements to the normalizer. For example, the
1.76 + * Indic shaper may want to disallow recomposing of two matras.
1.77 + *
1.78 + * - We try compatibility decomposition if decomposing through canonical
1.79 + * decomposition alone failed to find a sequence that the font supports.
1.80 + * We don't try compatibility decomposition recursively during the canonical
1.81 + * decomposition phase. This has minimal impact. There are only a handful
1.82 + * of Greek letter that have canonical decompositions that include characters
1.83 + * with compatibility decomposition. Those can be found using this command:
1.84 + *
1.85 + * egrep "`echo -n ';('; grep ';<' UnicodeData.txt | cut -d';' -f1 | tr '\n' '|'; echo ') '`" UnicodeData.txt
1.86 + */
1.87 +
1.88 +static bool
1.89 +decompose_unicode (const hb_ot_shape_normalize_context_t *c,
1.90 + hb_codepoint_t ab,
1.91 + hb_codepoint_t *a,
1.92 + hb_codepoint_t *b)
1.93 +{
1.94 + return c->unicode->decompose (ab, a, b);
1.95 +}
1.96 +
1.97 +static bool
1.98 +compose_unicode (const hb_ot_shape_normalize_context_t *c,
1.99 + hb_codepoint_t a,
1.100 + hb_codepoint_t b,
1.101 + hb_codepoint_t *ab)
1.102 +{
1.103 + return c->unicode->compose (a, b, ab);
1.104 +}
1.105 +
1.106 +static inline void
1.107 +set_glyph (hb_glyph_info_t &info, hb_font_t *font)
1.108 +{
1.109 + font->get_glyph (info.codepoint, 0, &info.glyph_index());
1.110 +}
1.111 +
1.112 +static inline void
1.113 +output_char (hb_buffer_t *buffer, hb_codepoint_t unichar, hb_codepoint_t glyph)
1.114 +{
1.115 + buffer->cur().glyph_index() = glyph;
1.116 + buffer->output_glyph (unichar);
1.117 + _hb_glyph_info_set_unicode_props (&buffer->prev(), buffer->unicode);
1.118 +}
1.119 +
1.120 +static inline void
1.121 +next_char (hb_buffer_t *buffer, hb_codepoint_t glyph)
1.122 +{
1.123 + buffer->cur().glyph_index() = glyph;
1.124 + buffer->next_glyph ();
1.125 +}
1.126 +
1.127 +static inline void
1.128 +skip_char (hb_buffer_t *buffer)
1.129 +{
1.130 + buffer->skip_glyph ();
1.131 +}
1.132 +
1.133 +/* Returns 0 if didn't decompose, number of resulting characters otherwise. */
1.134 +static inline unsigned int
1.135 +decompose (const hb_ot_shape_normalize_context_t *c, bool shortest, hb_codepoint_t ab)
1.136 +{
1.137 + hb_codepoint_t a, b, a_glyph, b_glyph;
1.138 + hb_buffer_t * const buffer = c->buffer;
1.139 + hb_font_t * const font = c->font;
1.140 +
1.141 + if (!c->decompose (c, ab, &a, &b) ||
1.142 + (b && !font->get_glyph (b, 0, &b_glyph)))
1.143 + return 0;
1.144 +
1.145 + bool has_a = font->get_glyph (a, 0, &a_glyph);
1.146 + if (shortest && has_a) {
1.147 + /* Output a and b */
1.148 + output_char (buffer, a, a_glyph);
1.149 + if (likely (b)) {
1.150 + output_char (buffer, b, b_glyph);
1.151 + return 2;
1.152 + }
1.153 + return 1;
1.154 + }
1.155 +
1.156 + unsigned int ret;
1.157 + if ((ret = decompose (c, shortest, a))) {
1.158 + if (b) {
1.159 + output_char (buffer, b, b_glyph);
1.160 + return ret + 1;
1.161 + }
1.162 + return ret;
1.163 + }
1.164 +
1.165 + if (has_a) {
1.166 + output_char (buffer, a, a_glyph);
1.167 + if (likely (b)) {
1.168 + output_char (buffer, b, b_glyph);
1.169 + return 2;
1.170 + }
1.171 + return 1;
1.172 + }
1.173 +
1.174 + return 0;
1.175 +}
1.176 +
1.177 +/* Returns 0 if didn't decompose, number of resulting characters otherwise. */
1.178 +static inline unsigned int
1.179 +decompose_compatibility (const hb_ot_shape_normalize_context_t *c, hb_codepoint_t u)
1.180 +{
1.181 + unsigned int len, i;
1.182 + hb_codepoint_t decomposed[HB_UNICODE_MAX_DECOMPOSITION_LEN];
1.183 + hb_codepoint_t glyphs[HB_UNICODE_MAX_DECOMPOSITION_LEN];
1.184 +
1.185 + len = c->buffer->unicode->decompose_compatibility (u, decomposed);
1.186 + if (!len)
1.187 + return 0;
1.188 +
1.189 + for (i = 0; i < len; i++)
1.190 + if (!c->font->get_glyph (decomposed[i], 0, &glyphs[i]))
1.191 + return 0;
1.192 +
1.193 + for (i = 0; i < len; i++)
1.194 + output_char (c->buffer, decomposed[i], glyphs[i]);
1.195 +
1.196 + return len;
1.197 +}
1.198 +
1.199 +static inline void
1.200 +decompose_current_character (const hb_ot_shape_normalize_context_t *c, bool shortest)
1.201 +{
1.202 + hb_buffer_t * const buffer = c->buffer;
1.203 + hb_codepoint_t glyph;
1.204 +
1.205 + /* Kind of a cute waterfall here... */
1.206 + if (shortest && c->font->get_glyph (buffer->cur().codepoint, 0, &glyph))
1.207 + next_char (buffer, glyph);
1.208 + else if (decompose (c, shortest, buffer->cur().codepoint))
1.209 + skip_char (buffer);
1.210 + else if (!shortest && c->font->get_glyph (buffer->cur().codepoint, 0, &glyph))
1.211 + next_char (buffer, glyph);
1.212 + else if (decompose_compatibility (c, buffer->cur().codepoint))
1.213 + skip_char (buffer);
1.214 + else
1.215 + next_char (buffer, glyph); /* glyph is initialized in earlier branches. */
1.216 +}
1.217 +
1.218 +static inline void
1.219 +handle_variation_selector_cluster (const hb_ot_shape_normalize_context_t *c, unsigned int end, bool short_circuit)
1.220 +{
1.221 + /* TODO Currently if there's a variation-selector we give-up, it's just too hard. */
1.222 + hb_buffer_t * const buffer = c->buffer;
1.223 + hb_font_t * const font = c->font;
1.224 + for (; buffer->idx < end - 1;) {
1.225 + if (unlikely (buffer->unicode->is_variation_selector (buffer->cur(+1).codepoint))) {
1.226 + /* The next two lines are some ugly lines... But work. */
1.227 + if (font->get_glyph (buffer->cur().codepoint, buffer->cur(+1).codepoint, &buffer->cur().glyph_index()))
1.228 + {
1.229 + buffer->replace_glyphs (2, 1, &buffer->cur().codepoint);
1.230 + }
1.231 + else
1.232 + {
1.233 + /* Just pass on the two characters separately, let GSUB do its magic. */
1.234 + set_glyph (buffer->cur(), font);
1.235 + buffer->next_glyph ();
1.236 + set_glyph (buffer->cur(), font);
1.237 + buffer->next_glyph ();
1.238 + }
1.239 + /* Skip any further variation selectors. */
1.240 + while (buffer->idx < end && unlikely (buffer->unicode->is_variation_selector (buffer->cur().codepoint)))
1.241 + {
1.242 + set_glyph (buffer->cur(), font);
1.243 + buffer->next_glyph ();
1.244 + }
1.245 + } else {
1.246 + set_glyph (buffer->cur(), font);
1.247 + buffer->next_glyph ();
1.248 + }
1.249 + }
1.250 + if (likely (buffer->idx < end)) {
1.251 + set_glyph (buffer->cur(), font);
1.252 + buffer->next_glyph ();
1.253 + }
1.254 +}
1.255 +
1.256 +static inline void
1.257 +decompose_multi_char_cluster (const hb_ot_shape_normalize_context_t *c, unsigned int end, bool short_circuit)
1.258 +{
1.259 + hb_buffer_t * const buffer = c->buffer;
1.260 + for (unsigned int i = buffer->idx; i < end; i++)
1.261 + if (unlikely (buffer->unicode->is_variation_selector (buffer->info[i].codepoint))) {
1.262 + handle_variation_selector_cluster (c, end, short_circuit);
1.263 + return;
1.264 + }
1.265 +
1.266 + while (buffer->idx < end)
1.267 + decompose_current_character (c, short_circuit);
1.268 +}
1.269 +
1.270 +static inline void
1.271 +decompose_cluster (const hb_ot_shape_normalize_context_t *c, unsigned int end, bool might_short_circuit, bool always_short_circuit)
1.272 +{
1.273 + if (likely (c->buffer->idx + 1 == end))
1.274 + decompose_current_character (c, might_short_circuit);
1.275 + else
1.276 + decompose_multi_char_cluster (c, end, always_short_circuit);
1.277 +}
1.278 +
1.279 +
1.280 +static int
1.281 +compare_combining_class (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb)
1.282 +{
1.283 + unsigned int a = _hb_glyph_info_get_modified_combining_class (pa);
1.284 + unsigned int b = _hb_glyph_info_get_modified_combining_class (pb);
1.285 +
1.286 + return a < b ? -1 : a == b ? 0 : +1;
1.287 +}
1.288 +
1.289 +
1.290 +void
1.291 +_hb_ot_shape_normalize (const hb_ot_shape_plan_t *plan,
1.292 + hb_buffer_t *buffer,
1.293 + hb_font_t *font)
1.294 +{
1.295 + hb_ot_shape_normalization_mode_t mode = plan->shaper->normalization_preference;
1.296 + const hb_ot_shape_normalize_context_t c = {
1.297 + plan,
1.298 + buffer,
1.299 + font,
1.300 + buffer->unicode,
1.301 + plan->shaper->decompose ? plan->shaper->decompose : decompose_unicode,
1.302 + plan->shaper->compose ? plan->shaper->compose : compose_unicode
1.303 + };
1.304 +
1.305 + bool always_short_circuit = mode == HB_OT_SHAPE_NORMALIZATION_MODE_NONE;
1.306 + bool might_short_circuit = always_short_circuit ||
1.307 + (mode != HB_OT_SHAPE_NORMALIZATION_MODE_DECOMPOSED &&
1.308 + mode != HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT);
1.309 + unsigned int count;
1.310 +
1.311 + /* We do a fairly straightforward yet custom normalization process in three
1.312 + * separate rounds: decompose, reorder, recompose (if desired). Currently
1.313 + * this makes two buffer swaps. We can make it faster by moving the last
1.314 + * two rounds into the inner loop for the first round, but it's more readable
1.315 + * this way. */
1.316 +
1.317 +
1.318 + /* First round, decompose */
1.319 +
1.320 + buffer->clear_output ();
1.321 + count = buffer->len;
1.322 + for (buffer->idx = 0; buffer->idx < count;)
1.323 + {
1.324 + unsigned int end;
1.325 + for (end = buffer->idx + 1; end < count; end++)
1.326 + if (buffer->cur().cluster != buffer->info[end].cluster)
1.327 + break;
1.328 +
1.329 + decompose_cluster (&c, end, might_short_circuit, always_short_circuit);
1.330 + }
1.331 + buffer->swap_buffers ();
1.332 +
1.333 +
1.334 + /* Second round, reorder (inplace) */
1.335 +
1.336 + count = buffer->len;
1.337 + for (unsigned int i = 0; i < count; i++)
1.338 + {
1.339 + if (_hb_glyph_info_get_modified_combining_class (&buffer->info[i]) == 0)
1.340 + continue;
1.341 +
1.342 + unsigned int end;
1.343 + for (end = i + 1; end < count; end++)
1.344 + if (_hb_glyph_info_get_modified_combining_class (&buffer->info[end]) == 0)
1.345 + break;
1.346 +
1.347 + /* We are going to do a bubble-sort. Only do this if the
1.348 + * sequence is short. Doing it on long sequences can result
1.349 + * in an O(n^2) DoS. */
1.350 + if (end - i > 10) {
1.351 + i = end;
1.352 + continue;
1.353 + }
1.354 +
1.355 + hb_bubble_sort (buffer->info + i, end - i, compare_combining_class);
1.356 +
1.357 + i = end;
1.358 + }
1.359 +
1.360 +
1.361 + if (mode == HB_OT_SHAPE_NORMALIZATION_MODE_NONE ||
1.362 + mode == HB_OT_SHAPE_NORMALIZATION_MODE_DECOMPOSED)
1.363 + return;
1.364 +
1.365 + /* Third round, recompose */
1.366 +
1.367 + /* As noted in the comment earlier, we don't try to combine
1.368 + * ccc=0 chars with their previous Starter. */
1.369 +
1.370 + buffer->clear_output ();
1.371 + count = buffer->len;
1.372 + unsigned int starter = 0;
1.373 + buffer->next_glyph ();
1.374 + while (buffer->idx < count)
1.375 + {
1.376 + hb_codepoint_t composed, glyph;
1.377 + if (/* We don't try to compose a non-mark character with it's preceding starter.
1.378 + * This is both an optimization to avoid trying to compose every two neighboring
1.379 + * glyphs in most scripts AND a desired feature for Hangul. Apparently Hangul
1.380 + * fonts are not designed to mix-and-match pre-composed syllables and Jamo. */
1.381 + HB_UNICODE_GENERAL_CATEGORY_IS_MARK (_hb_glyph_info_get_general_category (&buffer->cur())) &&
1.382 + /* If there's anything between the starter and this char, they should have CCC
1.383 + * smaller than this character's. */
1.384 + (starter == buffer->out_len - 1 ||
1.385 + _hb_glyph_info_get_modified_combining_class (&buffer->prev()) < _hb_glyph_info_get_modified_combining_class (&buffer->cur())) &&
1.386 + /* And compose. */
1.387 + c.compose (&c,
1.388 + buffer->out_info[starter].codepoint,
1.389 + buffer->cur().codepoint,
1.390 + &composed) &&
1.391 + /* And the font has glyph for the composite. */
1.392 + font->get_glyph (composed, 0, &glyph))
1.393 + {
1.394 + /* Composes. */
1.395 + buffer->next_glyph (); /* Copy to out-buffer. */
1.396 + if (unlikely (buffer->in_error))
1.397 + return;
1.398 + buffer->merge_out_clusters (starter, buffer->out_len);
1.399 + buffer->out_len--; /* Remove the second composable. */
1.400 + /* Modify starter and carry on. */
1.401 + buffer->out_info[starter].codepoint = composed;
1.402 + buffer->out_info[starter].glyph_index() = glyph;
1.403 + _hb_glyph_info_set_unicode_props (&buffer->out_info[starter], buffer->unicode);
1.404 +
1.405 + continue;
1.406 + }
1.407 +
1.408 + /* Blocked, or doesn't compose. */
1.409 + buffer->next_glyph ();
1.410 +
1.411 + if (_hb_glyph_info_get_modified_combining_class (&buffer->prev()) == 0)
1.412 + starter = buffer->out_len - 1;
1.413 + }
1.414 + buffer->swap_buffers ();
1.415 +
1.416 +}
