2 * Copyright © 2011,2012 Google, Inc.
4 * This is part of HarfBuzz, a text shaping library.
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.
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
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.
24 * Google Author(s): Behdad Esfahbod
27 #include "hb-ot-shape-normalize-private.hh"
28 #include "hb-ot-shape-complex-private.hh"
29 #include "hb-ot-shape-private.hh"
35 * This file exports one main function: _hb_ot_shape_normalize().
37 * This function closely reflects the Unicode Normalization Algorithm,
40 * Each shaper specifies whether it prefers decomposed (NFD) or composed (NFC).
41 * The logic however tries to use whatever the font can support.
43 * In general what happens is that: each grapheme is decomposed in a chain
44 * of 1:2 decompositions, marks reordered, and then recomposed if desired,
45 * so far it's like Unicode Normalization. However, the decomposition and
46 * recomposition only happens if the font supports the resulting characters.
50 * - Try to render all canonically equivalent strings similarly. To really
51 * achieve this we have to always do the full decomposition and then
52 * selectively recompose from there. It's kinda too expensive though, so
53 * we skip some cases. For example, if composed is desired, we simply
54 * don't touch 1-character clusters that are supported by the font, even
55 * though their NFC may be different.
57 * - When a font has a precomposed character for a sequence but the 'ccmp'
58 * feature in the font is not adequate, use the precomposed character
59 * which typically has better mark positioning.
61 * - When a font does not support a combining mark, but supports it precomposed
62 * with previous base, use that. This needs the itemizer to have this
63 * knowledge too. We need to provide assistance to the itemizer.
65 * - When a font does not support a character but supports its canonical
66 * decomposition, well, use the decomposition.
68 * - The complex shapers can customize the compose and decompose functions to
69 * offload some of their requirements to the normalizer. For example, the
70 * Indic shaper may want to disallow recomposing of two matras.
74 decompose_unicode (const hb_ot_shape_normalize_context_t *c,
79 return (bool) c->unicode->decompose (ab, a, b);
83 compose_unicode (const hb_ot_shape_normalize_context_t *c,
88 return (bool) c->unicode->compose (a, b, ab);
92 set_glyph (hb_glyph_info_t &info, hb_font_t *font)
94 (void) font->get_nominal_glyph (info.codepoint, &info.glyph_index());
98 output_char (hb_buffer_t *buffer, hb_codepoint_t unichar, hb_codepoint_t glyph)
100 buffer->cur().glyph_index() = glyph;
101 buffer->output_glyph (unichar); /* This is very confusing indeed. */
102 _hb_glyph_info_set_unicode_props (&buffer->prev(), buffer);
106 next_char (hb_buffer_t *buffer, hb_codepoint_t glyph)
108 buffer->cur().glyph_index() = glyph;
109 buffer->next_glyph ();
113 skip_char (hb_buffer_t *buffer)
115 buffer->skip_glyph ();
118 /* Returns 0 if didn't decompose, number of resulting characters otherwise. */
119 static inline unsigned int
120 decompose (const hb_ot_shape_normalize_context_t *c, bool shortest, hb_codepoint_t ab)
122 hb_codepoint_t a, b, a_glyph, b_glyph;
123 hb_buffer_t * const buffer = c->buffer;
124 hb_font_t * const font = c->font;
126 if (!c->decompose (c, ab, &a, &b) ||
127 (b && !font->get_nominal_glyph (b, &b_glyph)))
130 bool has_a = (bool) font->get_nominal_glyph (a, &a_glyph);
131 if (shortest && has_a) {
133 output_char (buffer, a, a_glyph);
135 output_char (buffer, b, b_glyph);
142 if ((ret = decompose (c, shortest, a))) {
144 output_char (buffer, b, b_glyph);
151 output_char (buffer, a, a_glyph);
153 output_char (buffer, b, b_glyph);
163 decompose_current_character (const hb_ot_shape_normalize_context_t *c, bool shortest)
165 hb_buffer_t * const buffer = c->buffer;
166 hb_codepoint_t u = buffer->cur().codepoint;
167 hb_codepoint_t glyph;
169 if (shortest && c->font->get_nominal_glyph (u, &glyph))
171 next_char (buffer, glyph);
175 if (decompose (c, shortest, u))
181 if (!shortest && c->font->get_nominal_glyph (u, &glyph))
183 next_char (buffer, glyph);
187 if (_hb_glyph_info_is_unicode_space (&buffer->cur()))
189 hb_codepoint_t space_glyph;
190 hb_unicode_funcs_t::space_t space_type = buffer->unicode->space_fallback_type (u);
191 if (space_type != hb_unicode_funcs_t::NOT_SPACE && c->font->get_nominal_glyph (0x0020u, &space_glyph))
193 _hb_glyph_info_set_unicode_space_fallback_type (&buffer->cur(), space_type);
194 next_char (buffer, space_glyph);
195 buffer->scratch_flags |= HB_BUFFER_SCRATCH_FLAG_HAS_SPACE_FALLBACK;
202 /* U+2011 is the only sensible character that is a no-break version of another character
203 * and not a space. The space ones are handled already. Handle this lone one. */
204 hb_codepoint_t other_glyph;
205 if (c->font->get_nominal_glyph (0x2010u, &other_glyph))
207 next_char (buffer, other_glyph);
212 next_char (buffer, glyph); /* glyph is initialized in earlier branches. */
216 handle_variation_selector_cluster (const hb_ot_shape_normalize_context_t *c, unsigned int end, bool short_circuit)
218 /* TODO Currently if there's a variation-selector we give-up, it's just too hard. */
219 hb_buffer_t * const buffer = c->buffer;
220 hb_font_t * const font = c->font;
221 for (; buffer->idx < end - 1 && !buffer->in_error;) {
222 if (unlikely (buffer->unicode->is_variation_selector (buffer->cur(+1).codepoint))) {
223 /* The next two lines are some ugly lines... But work. */
224 if (font->get_variation_glyph (buffer->cur().codepoint, buffer->cur(+1).codepoint, &buffer->cur().glyph_index()))
226 buffer->replace_glyphs (2, 1, &buffer->cur().codepoint);
230 /* Just pass on the two characters separately, let GSUB do its magic. */
231 set_glyph (buffer->cur(), font);
232 buffer->next_glyph ();
233 set_glyph (buffer->cur(), font);
234 buffer->next_glyph ();
236 /* Skip any further variation selectors. */
237 while (buffer->idx < end && unlikely (buffer->unicode->is_variation_selector (buffer->cur().codepoint)))
239 set_glyph (buffer->cur(), font);
240 buffer->next_glyph ();
243 set_glyph (buffer->cur(), font);
244 buffer->next_glyph ();
247 if (likely (buffer->idx < end)) {
248 set_glyph (buffer->cur(), font);
249 buffer->next_glyph ();
254 decompose_multi_char_cluster (const hb_ot_shape_normalize_context_t *c, unsigned int end, bool short_circuit)
256 hb_buffer_t * const buffer = c->buffer;
257 for (unsigned int i = buffer->idx; i < end && !buffer->in_error; i++)
258 if (unlikely (buffer->unicode->is_variation_selector (buffer->info[i].codepoint))) {
259 handle_variation_selector_cluster (c, end, short_circuit);
263 while (buffer->idx < end && !buffer->in_error)
264 decompose_current_character (c, short_circuit);
268 decompose_cluster (const hb_ot_shape_normalize_context_t *c, unsigned int end, bool might_short_circuit, bool always_short_circuit)
270 if (likely (c->buffer->idx + 1 == end))
271 decompose_current_character (c, might_short_circuit);
273 decompose_multi_char_cluster (c, end, always_short_circuit);
278 compare_combining_class (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb)
280 unsigned int a = _hb_glyph_info_get_modified_combining_class (pa);
281 unsigned int b = _hb_glyph_info_get_modified_combining_class (pb);
283 return a < b ? -1 : a == b ? 0 : +1;
288 _hb_ot_shape_normalize (const hb_ot_shape_plan_t *plan,
292 if (unlikely (!buffer->len)) return;
294 _hb_buffer_assert_unicode_vars (buffer);
296 hb_ot_shape_normalization_mode_t mode = plan->shaper->normalization_preference;
297 const hb_ot_shape_normalize_context_t c = {
302 plan->shaper->decompose ? plan->shaper->decompose : decompose_unicode,
303 plan->shaper->compose ? plan->shaper->compose : compose_unicode
306 bool always_short_circuit = mode == HB_OT_SHAPE_NORMALIZATION_MODE_NONE;
307 bool might_short_circuit = always_short_circuit ||
308 (mode != HB_OT_SHAPE_NORMALIZATION_MODE_DECOMPOSED &&
309 mode != HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT);
312 /* We do a fairly straightforward yet custom normalization process in three
313 * separate rounds: decompose, reorder, recompose (if desired). Currently
314 * this makes two buffer swaps. We can make it faster by moving the last
315 * two rounds into the inner loop for the first round, but it's more readable
319 /* First round, decompose */
321 buffer->clear_output ();
323 for (buffer->idx = 0; buffer->idx < count && !buffer->in_error;)
326 for (end = buffer->idx + 1; end < count; end++)
327 if (likely (!HB_UNICODE_GENERAL_CATEGORY_IS_MARK (_hb_glyph_info_get_general_category (&buffer->info[end]))))
330 decompose_cluster (&c, end, might_short_circuit, always_short_circuit);
332 buffer->swap_buffers ();
335 /* Second round, reorder (inplace) */
338 for (unsigned int i = 0; i < count; i++)
340 if (_hb_glyph_info_get_modified_combining_class (&buffer->info[i]) == 0)
344 for (end = i + 1; end < count; end++)
345 if (_hb_glyph_info_get_modified_combining_class (&buffer->info[end]) == 0)
348 /* We are going to do a O(n^2). Only do this if the sequence is short. */
349 if (end - i > HB_OT_SHAPE_COMPLEX_MAX_COMBINING_MARKS) {
354 buffer->sort (i, end, compare_combining_class);
356 if (plan->shaper->reorder_marks)
357 plan->shaper->reorder_marks (plan, buffer, i, end);
363 if (mode == HB_OT_SHAPE_NORMALIZATION_MODE_NONE ||
364 mode == HB_OT_SHAPE_NORMALIZATION_MODE_DECOMPOSED)
367 /* Third round, recompose */
369 /* As noted in the comment earlier, we don't try to combine
370 * ccc=0 chars with their previous Starter. */
372 buffer->clear_output ();
374 unsigned int starter = 0;
375 buffer->next_glyph ();
376 while (buffer->idx < count && !buffer->in_error)
378 hb_codepoint_t composed, glyph;
379 if (/* We don't try to compose a non-mark character with it's preceding starter.
380 * This is both an optimization to avoid trying to compose every two neighboring
381 * glyphs in most scripts AND a desired feature for Hangul. Apparently Hangul
382 * fonts are not designed to mix-and-match pre-composed syllables and Jamo. */
383 HB_UNICODE_GENERAL_CATEGORY_IS_MARK (_hb_glyph_info_get_general_category (&buffer->cur())))
385 if (/* If there's anything between the starter and this char, they should have CCC
386 * smaller than this character's. */
387 (starter == buffer->out_len - 1 ||
388 info_cc (buffer->prev()) < info_cc (buffer->cur())) &&
391 buffer->out_info[starter].codepoint,
392 buffer->cur().codepoint,
394 /* And the font has glyph for the composite. */
395 font->get_nominal_glyph (composed, &glyph))
398 buffer->next_glyph (); /* Copy to out-buffer. */
399 if (unlikely (buffer->in_error))
401 buffer->merge_out_clusters (starter, buffer->out_len);
402 buffer->out_len--; /* Remove the second composable. */
403 /* Modify starter and carry on. */
404 buffer->out_info[starter].codepoint = composed;
405 buffer->out_info[starter].glyph_index() = glyph;
406 _hb_glyph_info_set_unicode_props (&buffer->out_info[starter], buffer);
412 /* Blocked, or doesn't compose. */
413 buffer->next_glyph ();
415 if (info_cc (buffer->prev()) == 0)
416 starter = buffer->out_len - 1;
418 buffer->swap_buffers ();
420 if (buffer->scratch_flags & HB_BUFFER_SCRATCH_FLAG_HAS_CGJ)
422 /* For all CGJ, check if it prevented any reordering at all.
423 * If it did NOT, then make it skippable.
424 * https://github.com/harfbuzz/harfbuzz/issues/554
426 for (unsigned int i = 1; i + 1 < buffer->len; i++)
427 if (buffer->info[i].codepoint == 0x034Fu/*CGJ*/ &&
428 info_cc(buffer->info[i-1]) <= info_cc(buffer->info[i+1]))
430 _hb_glyph_info_unhide (&buffer->info[i]);