Merge from trunk.
[bpt/emacs.git] / src / bidi.c
1 /* Low-level bidirectional buffer/string-scanning functions for GNU Emacs.
2 Copyright (C) 2000-2001, 2004-2005, 2009-2011
3 Free Software Foundation, Inc.
4
5 This file is part of GNU Emacs.
6
7 GNU Emacs is free software: you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation, either version 3 of the License, or
10 (at your option) any later version.
11
12 GNU Emacs is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
16
17 You should have received a copy of the GNU General Public License
18 along with GNU Emacs. If not, see <http://www.gnu.org/licenses/>. */
19
20 /* Written by Eli Zaretskii <eliz@gnu.org>.
21
22 A sequential implementation of the Unicode Bidirectional algorithm,
23 (UBA) as per UAX#9, a part of the Unicode Standard.
24
25 Unlike the reference and most other implementations, this one is
26 designed to be called once for every character in the buffer or
27 string.
28
29 The main entry point is bidi_move_to_visually_next. Each time it
30 is called, it finds the next character in the visual order, and
31 returns its information in a special structure. The caller is then
32 expected to process this character for display or any other
33 purposes, and call bidi_move_to_visually_next for the next
34 character. See the comments in bidi_move_to_visually_next for more
35 details about its algorithm that finds the next visual-order
36 character by resolving their levels on the fly.
37
38 Two other entry points are bidi_paragraph_init and
39 bidi_mirror_char. The first determines the base direction of a
40 paragraph, while the second returns the mirrored version of its
41 argument character.
42
43 A few auxiliary entry points are used to initialize the bidi
44 iterator for iterating an object (buffer or string), push and pop
45 the bidi iterator state, and save and restore the state of the bidi
46 cache.
47
48 If you want to understand the code, you will have to read it
49 together with the relevant portions of UAX#9. The comments include
50 references to UAX#9 rules, for that very reason.
51
52 A note about references to UAX#9 rules: if the reference says
53 something like "X9/Retaining", it means that you need to refer to
54 rule X9 and to its modifications decribed in the "Implementation
55 Notes" section of UAX#9, under "Retaining Format Codes". */
56
57 #include <config.h>
58 #include <stdio.h>
59 #include <setjmp.h>
60
61 #include "lisp.h"
62 #include "buffer.h"
63 #include "character.h"
64 #include "dispextern.h"
65
66 static int bidi_initialized = 0;
67
68 static Lisp_Object bidi_type_table, bidi_mirror_table;
69
70 #define LRM_CHAR 0x200E
71 #define RLM_CHAR 0x200F
72 #define BIDI_EOB -1
73
74 /* Data type for describing the bidirectional character categories. */
75 typedef enum {
76 UNKNOWN_BC,
77 NEUTRAL,
78 WEAK,
79 STRONG
80 } bidi_category_t;
81
82 /* UAX#9 says to search only for L, AL, or R types of characters, and
83 ignore RLE, RLO, LRE, and LRO, when determining the base paragraph
84 level. Yudit indeed ignores them. This variable is therefore set
85 by default to ignore them, but setting it to zero will take them
86 into account. */
87 extern int bidi_ignore_explicit_marks_for_paragraph_level EXTERNALLY_VISIBLE;
88 int bidi_ignore_explicit_marks_for_paragraph_level = 1;
89
90 static Lisp_Object paragraph_start_re, paragraph_separate_re;
91 static Lisp_Object Qparagraph_start, Qparagraph_separate;
92
93 \f
94 /***********************************************************************
95 Utilities
96 ***********************************************************************/
97
98 /* Return the bidi type of a character CH, subject to the current
99 directional OVERRIDE. */
100 static inline bidi_type_t
101 bidi_get_type (int ch, bidi_dir_t override)
102 {
103 bidi_type_t default_type;
104
105 if (ch == BIDI_EOB)
106 return NEUTRAL_B;
107 if (ch < 0 || ch > MAX_CHAR)
108 abort ();
109
110 default_type = (bidi_type_t) XINT (CHAR_TABLE_REF (bidi_type_table, ch));
111 if (default_type == 0)
112 default_type = STRONG_L;
113
114 if (override == NEUTRAL_DIR)
115 return default_type;
116
117 switch (default_type)
118 {
119 /* Although UAX#9 does not tell, it doesn't make sense to
120 override NEUTRAL_B and LRM/RLM characters. */
121 case NEUTRAL_B:
122 case LRE:
123 case LRO:
124 case RLE:
125 case RLO:
126 case PDF:
127 return default_type;
128 default:
129 switch (ch)
130 {
131 case LRM_CHAR:
132 case RLM_CHAR:
133 return default_type;
134 default:
135 if (override == L2R) /* X6 */
136 return STRONG_L;
137 else if (override == R2L)
138 return STRONG_R;
139 else
140 abort (); /* can't happen: handled above */
141 }
142 }
143 }
144
145 static void
146 bidi_check_type (bidi_type_t type)
147 {
148 if (type < UNKNOWN_BT || type > NEUTRAL_ON)
149 abort ();
150 }
151
152 /* Given a bidi TYPE of a character, return its category. */
153 static inline bidi_category_t
154 bidi_get_category (bidi_type_t type)
155 {
156 switch (type)
157 {
158 case UNKNOWN_BT:
159 return UNKNOWN_BC;
160 case STRONG_L:
161 case STRONG_R:
162 case STRONG_AL:
163 case LRE:
164 case LRO:
165 case RLE:
166 case RLO:
167 return STRONG;
168 case PDF: /* ??? really?? */
169 case WEAK_EN:
170 case WEAK_ES:
171 case WEAK_ET:
172 case WEAK_AN:
173 case WEAK_CS:
174 case WEAK_NSM:
175 case WEAK_BN:
176 return WEAK;
177 case NEUTRAL_B:
178 case NEUTRAL_S:
179 case NEUTRAL_WS:
180 case NEUTRAL_ON:
181 return NEUTRAL;
182 default:
183 abort ();
184 }
185 }
186
187 /* Return the mirrored character of C, if it has one. If C has no
188 mirrored counterpart, return C.
189 Note: The conditions in UAX#9 clause L4 regarding the surrounding
190 context must be tested by the caller. */
191 int
192 bidi_mirror_char (int c)
193 {
194 Lisp_Object val;
195
196 if (c == BIDI_EOB)
197 return c;
198 if (c < 0 || c > MAX_CHAR)
199 abort ();
200
201 val = CHAR_TABLE_REF (bidi_mirror_table, c);
202 if (INTEGERP (val))
203 {
204 int v = XINT (val);
205
206 if (v < 0 || v > MAX_CHAR)
207 abort ();
208
209 return v;
210 }
211
212 return c;
213 }
214
215 /* Determine the start-of-run (sor) directional type given the two
216 embedding levels on either side of the run boundary. Also, update
217 the saved info about previously seen characters, since that info is
218 generally valid for a single level run. */
219 static inline void
220 bidi_set_sor_type (struct bidi_it *bidi_it, int level_before, int level_after)
221 {
222 int higher_level = level_before > level_after ? level_before : level_after;
223
224 /* The prev_was_pdf gork is required for when we have several PDFs
225 in a row. In that case, we want to compute the sor type for the
226 next level run only once: when we see the first PDF. That's
227 because the sor type depends only on the higher of the two levels
228 that we find on the two sides of the level boundary (see UAX#9,
229 clause X10), and so we don't need to know the final embedding
230 level to which we descend after processing all the PDFs. */
231 if (!bidi_it->prev_was_pdf || level_before < level_after)
232 /* FIXME: should the default sor direction be user selectable? */
233 bidi_it->sor = (higher_level & 1) != 0 ? R2L : L2R;
234 if (level_before > level_after)
235 bidi_it->prev_was_pdf = 1;
236
237 bidi_it->prev.type = UNKNOWN_BT;
238 bidi_it->last_strong.type = bidi_it->last_strong.type_after_w1 =
239 bidi_it->last_strong.orig_type = UNKNOWN_BT;
240 bidi_it->prev_for_neutral.type = bidi_it->sor == R2L ? STRONG_R : STRONG_L;
241 bidi_it->prev_for_neutral.charpos = bidi_it->charpos;
242 bidi_it->prev_for_neutral.bytepos = bidi_it->bytepos;
243 bidi_it->next_for_neutral.type = bidi_it->next_for_neutral.type_after_w1 =
244 bidi_it->next_for_neutral.orig_type = UNKNOWN_BT;
245 bidi_it->ignore_bn_limit = -1; /* meaning it's unknown */
246 }
247
248 /* Push the current embedding level and override status; reset the
249 current level to LEVEL and the current override status to OVERRIDE. */
250 static inline void
251 bidi_push_embedding_level (struct bidi_it *bidi_it,
252 int level, bidi_dir_t override)
253 {
254 bidi_it->stack_idx++;
255 xassert (bidi_it->stack_idx < BIDI_MAXLEVEL);
256 bidi_it->level_stack[bidi_it->stack_idx].level = level;
257 bidi_it->level_stack[bidi_it->stack_idx].override = override;
258 }
259
260 /* Pop the embedding level and directional override status from the
261 stack, and return the new level. */
262 static inline int
263 bidi_pop_embedding_level (struct bidi_it *bidi_it)
264 {
265 /* UAX#9 says to ignore invalid PDFs. */
266 if (bidi_it->stack_idx > 0)
267 bidi_it->stack_idx--;
268 return bidi_it->level_stack[bidi_it->stack_idx].level;
269 }
270
271 /* Record in SAVED_INFO the information about the current character. */
272 static inline void
273 bidi_remember_char (struct bidi_saved_info *saved_info,
274 struct bidi_it *bidi_it)
275 {
276 saved_info->charpos = bidi_it->charpos;
277 saved_info->bytepos = bidi_it->bytepos;
278 saved_info->type = bidi_it->type;
279 bidi_check_type (bidi_it->type);
280 saved_info->type_after_w1 = bidi_it->type_after_w1;
281 bidi_check_type (bidi_it->type_after_w1);
282 saved_info->orig_type = bidi_it->orig_type;
283 bidi_check_type (bidi_it->orig_type);
284 }
285
286 /* Copy the bidi iterator from FROM to TO. To save cycles, this only
287 copies the part of the level stack that is actually in use. */
288 static inline void
289 bidi_copy_it (struct bidi_it *to, struct bidi_it *from)
290 {
291 int i;
292
293 /* Copy everything except the level stack and beyond. */
294 memcpy (to, from, offsetof (struct bidi_it, level_stack[0]));
295
296 /* Copy the active part of the level stack. */
297 to->level_stack[0] = from->level_stack[0]; /* level zero is always in use */
298 for (i = 1; i <= from->stack_idx; i++)
299 to->level_stack[i] = from->level_stack[i];
300 }
301
302 \f
303 /***********************************************************************
304 Caching the bidi iterator states
305 ***********************************************************************/
306
307 #define BIDI_CACHE_CHUNK 200
308 static struct bidi_it *bidi_cache;
309 static ptrdiff_t bidi_cache_size = 0;
310 enum { elsz = sizeof (struct bidi_it) };
311 static ptrdiff_t bidi_cache_idx; /* next unused cache slot */
312 static ptrdiff_t bidi_cache_last_idx; /* slot of last cache hit */
313 static ptrdiff_t bidi_cache_start = 0; /* start of cache for this
314 "stack" level */
315
316 /* 5-slot stack for saving the start of the previous level of the
317 cache. xdisp.c maintains a 5-slot stack for its iterator state,
318 and we need the same size of our stack. */
319 static ptrdiff_t bidi_cache_start_stack[IT_STACK_SIZE];
320 static int bidi_cache_sp;
321
322 /* Size of header used by bidi_shelve_cache. */
323 enum
324 {
325 bidi_shelve_header_size =
326 (sizeof (bidi_cache_idx) + sizeof (bidi_cache_start_stack)
327 + sizeof (bidi_cache_sp) + sizeof (bidi_cache_start)
328 + sizeof (bidi_cache_last_idx))
329 };
330
331 /* Reset the cache state to the empty state. We only reset the part
332 of the cache relevant to iteration of the current object. Previous
333 objects, which are pushed on the display iterator's stack, are left
334 intact. This is called when the cached information is no more
335 useful for the current iteration, e.g. when we were reseated to a
336 new position on the same object. */
337 static inline void
338 bidi_cache_reset (void)
339 {
340 bidi_cache_idx = bidi_cache_start;
341 bidi_cache_last_idx = -1;
342 }
343
344 /* Shrink the cache to its minimal size. Called when we init the bidi
345 iterator for reordering a buffer or a string that does not come
346 from display properties, because that means all the previously
347 cached info is of no further use. */
348 static inline void
349 bidi_cache_shrink (void)
350 {
351 if (bidi_cache_size > BIDI_CACHE_CHUNK)
352 {
353 bidi_cache_size = BIDI_CACHE_CHUNK;
354 bidi_cache =
355 (struct bidi_it *) xrealloc (bidi_cache, bidi_cache_size * elsz);
356 }
357 bidi_cache_reset ();
358 }
359
360 static inline void
361 bidi_cache_fetch_state (ptrdiff_t idx, struct bidi_it *bidi_it)
362 {
363 int current_scan_dir = bidi_it->scan_dir;
364
365 if (idx < bidi_cache_start || idx >= bidi_cache_idx)
366 abort ();
367
368 bidi_copy_it (bidi_it, &bidi_cache[idx]);
369 bidi_it->scan_dir = current_scan_dir;
370 bidi_cache_last_idx = idx;
371 }
372
373 /* Find a cached state with a given CHARPOS and resolved embedding
374 level less or equal to LEVEL. if LEVEL is -1, disregard the
375 resolved levels in cached states. DIR, if non-zero, means search
376 in that direction from the last cache hit. */
377 static inline ptrdiff_t
378 bidi_cache_search (EMACS_INT charpos, int level, int dir)
379 {
380 ptrdiff_t i, i_start;
381
382 if (bidi_cache_idx > bidi_cache_start)
383 {
384 if (bidi_cache_last_idx == -1)
385 bidi_cache_last_idx = bidi_cache_idx - 1;
386 if (charpos < bidi_cache[bidi_cache_last_idx].charpos)
387 {
388 dir = -1;
389 i_start = bidi_cache_last_idx - 1;
390 }
391 else if (charpos > (bidi_cache[bidi_cache_last_idx].charpos
392 + bidi_cache[bidi_cache_last_idx].nchars - 1))
393 {
394 dir = 1;
395 i_start = bidi_cache_last_idx + 1;
396 }
397 else if (dir)
398 i_start = bidi_cache_last_idx;
399 else
400 {
401 dir = -1;
402 i_start = bidi_cache_idx - 1;
403 }
404
405 if (dir < 0)
406 {
407 /* Linear search for now; FIXME! */
408 for (i = i_start; i >= bidi_cache_start; i--)
409 if (bidi_cache[i].charpos <= charpos
410 && charpos < bidi_cache[i].charpos + bidi_cache[i].nchars
411 && (level == -1 || bidi_cache[i].resolved_level <= level))
412 return i;
413 }
414 else
415 {
416 for (i = i_start; i < bidi_cache_idx; i++)
417 if (bidi_cache[i].charpos <= charpos
418 && charpos < bidi_cache[i].charpos + bidi_cache[i].nchars
419 && (level == -1 || bidi_cache[i].resolved_level <= level))
420 return i;
421 }
422 }
423
424 return -1;
425 }
426
427 /* Find a cached state where the resolved level changes to a value
428 that is lower than LEVEL, and return its cache slot index. DIR is
429 the direction to search, starting with the last used cache slot.
430 If DIR is zero, we search backwards from the last occupied cache
431 slot. BEFORE, if non-zero, means return the index of the slot that
432 is ``before'' the level change in the search direction. That is,
433 given the cached levels like this:
434
435 1122333442211
436 AB C
437
438 and assuming we are at the position cached at the slot marked with
439 C, searching backwards (DIR = -1) for LEVEL = 2 will return the
440 index of slot B or A, depending whether BEFORE is, respectively,
441 non-zero or zero. */
442 static ptrdiff_t
443 bidi_cache_find_level_change (int level, int dir, int before)
444 {
445 if (bidi_cache_idx)
446 {
447 ptrdiff_t i = dir ? bidi_cache_last_idx : bidi_cache_idx - 1;
448 int incr = before ? 1 : 0;
449
450 xassert (!dir || bidi_cache_last_idx >= 0);
451
452 if (!dir)
453 dir = -1;
454 else if (!incr)
455 i += dir;
456
457 if (dir < 0)
458 {
459 while (i >= bidi_cache_start + incr)
460 {
461 if (bidi_cache[i - incr].resolved_level >= 0
462 && bidi_cache[i - incr].resolved_level < level)
463 return i;
464 i--;
465 }
466 }
467 else
468 {
469 while (i < bidi_cache_idx - incr)
470 {
471 if (bidi_cache[i + incr].resolved_level >= 0
472 && bidi_cache[i + incr].resolved_level < level)
473 return i;
474 i++;
475 }
476 }
477 }
478
479 return -1;
480 }
481
482 static inline void
483 bidi_cache_ensure_space (ptrdiff_t idx)
484 {
485 /* Enlarge the cache as needed. */
486 if (idx >= bidi_cache_size)
487 {
488 /* The bidi cache cannot be larger than the largest Lisp string
489 or buffer. */
490 ptrdiff_t string_or_buffer_bound =
491 max (BUF_BYTES_MAX, STRING_BYTES_BOUND);
492
493 /* Also, it cannot be larger than what C can represent. */
494 ptrdiff_t c_bound =
495 (min (PTRDIFF_MAX, SIZE_MAX) - bidi_shelve_header_size) / elsz;
496
497 bidi_cache =
498 xpalloc (bidi_cache, &bidi_cache_size,
499 max (BIDI_CACHE_CHUNK, idx - bidi_cache_size + 1),
500 min (string_or_buffer_bound, c_bound), elsz);
501 }
502 }
503
504 static inline void
505 bidi_cache_iterator_state (struct bidi_it *bidi_it, int resolved)
506 {
507 ptrdiff_t idx;
508
509 /* We should never cache on backward scans. */
510 if (bidi_it->scan_dir == -1)
511 abort ();
512 idx = bidi_cache_search (bidi_it->charpos, -1, 1);
513
514 if (idx < 0)
515 {
516 idx = bidi_cache_idx;
517 bidi_cache_ensure_space (idx);
518 /* Character positions should correspond to cache positions 1:1.
519 If we are outside the range of cached positions, the cache is
520 useless and must be reset. */
521 if (idx > bidi_cache_start &&
522 (bidi_it->charpos > (bidi_cache[idx - 1].charpos
523 + bidi_cache[idx - 1].nchars)
524 || bidi_it->charpos < bidi_cache[bidi_cache_start].charpos))
525 {
526 bidi_cache_reset ();
527 idx = bidi_cache_start;
528 }
529 if (bidi_it->nchars <= 0)
530 abort ();
531 bidi_copy_it (&bidi_cache[idx], bidi_it);
532 if (!resolved)
533 bidi_cache[idx].resolved_level = -1;
534 }
535 else
536 {
537 /* Copy only the members which could have changed, to avoid
538 costly copying of the entire struct. */
539 bidi_cache[idx].type = bidi_it->type;
540 bidi_check_type (bidi_it->type);
541 bidi_cache[idx].type_after_w1 = bidi_it->type_after_w1;
542 bidi_check_type (bidi_it->type_after_w1);
543 if (resolved)
544 bidi_cache[idx].resolved_level = bidi_it->resolved_level;
545 else
546 bidi_cache[idx].resolved_level = -1;
547 bidi_cache[idx].invalid_levels = bidi_it->invalid_levels;
548 bidi_cache[idx].invalid_rl_levels = bidi_it->invalid_rl_levels;
549 bidi_cache[idx].next_for_neutral = bidi_it->next_for_neutral;
550 bidi_cache[idx].next_for_ws = bidi_it->next_for_ws;
551 bidi_cache[idx].ignore_bn_limit = bidi_it->ignore_bn_limit;
552 }
553
554 bidi_cache_last_idx = idx;
555 if (idx >= bidi_cache_idx)
556 bidi_cache_idx = idx + 1;
557 }
558
559 static inline bidi_type_t
560 bidi_cache_find (EMACS_INT charpos, int level, struct bidi_it *bidi_it)
561 {
562 ptrdiff_t i = bidi_cache_search (charpos, level, bidi_it->scan_dir);
563
564 if (i >= bidi_cache_start)
565 {
566 bidi_dir_t current_scan_dir = bidi_it->scan_dir;
567
568 bidi_copy_it (bidi_it, &bidi_cache[i]);
569 bidi_cache_last_idx = i;
570 /* Don't let scan direction from from the cached state override
571 the current scan direction. */
572 bidi_it->scan_dir = current_scan_dir;
573 return bidi_it->type;
574 }
575
576 return UNKNOWN_BT;
577 }
578
579 static inline int
580 bidi_peek_at_next_level (struct bidi_it *bidi_it)
581 {
582 if (bidi_cache_idx == bidi_cache_start || bidi_cache_last_idx == -1)
583 abort ();
584 return bidi_cache[bidi_cache_last_idx + bidi_it->scan_dir].resolved_level;
585 }
586
587 \f
588 /***********************************************************************
589 Pushing and popping the bidi iterator state
590 ***********************************************************************/
591
592 /* Push the bidi iterator state in preparation for reordering a
593 different object, e.g. display string found at certain buffer
594 position. Pushing the bidi iterator boils down to saving its
595 entire state on the cache and starting a new cache "stacked" on top
596 of the current cache. */
597 void
598 bidi_push_it (struct bidi_it *bidi_it)
599 {
600 /* Save the current iterator state in its entirety after the last
601 used cache slot. */
602 bidi_cache_ensure_space (bidi_cache_idx);
603 memcpy (&bidi_cache[bidi_cache_idx++], bidi_it, sizeof (struct bidi_it));
604
605 /* Push the current cache start onto the stack. */
606 xassert (bidi_cache_sp < IT_STACK_SIZE);
607 bidi_cache_start_stack[bidi_cache_sp++] = bidi_cache_start;
608
609 /* Start a new level of cache, and make it empty. */
610 bidi_cache_start = bidi_cache_idx;
611 bidi_cache_last_idx = -1;
612 }
613
614 /* Restore the iterator state saved by bidi_push_it and return the
615 cache to the corresponding state. */
616 void
617 bidi_pop_it (struct bidi_it *bidi_it)
618 {
619 if (bidi_cache_start <= 0)
620 abort ();
621
622 /* Reset the next free cache slot index to what it was before the
623 call to bidi_push_it. */
624 bidi_cache_idx = bidi_cache_start - 1;
625
626 /* Restore the bidi iterator state saved in the cache. */
627 memcpy (bidi_it, &bidi_cache[bidi_cache_idx], sizeof (struct bidi_it));
628
629 /* Pop the previous cache start from the stack. */
630 if (bidi_cache_sp <= 0)
631 abort ();
632 bidi_cache_start = bidi_cache_start_stack[--bidi_cache_sp];
633
634 /* Invalidate the last-used cache slot data. */
635 bidi_cache_last_idx = -1;
636 }
637
638 static ptrdiff_t bidi_cache_total_alloc;
639
640 /* Stash away a copy of the cache and its control variables. */
641 void *
642 bidi_shelve_cache (void)
643 {
644 unsigned char *databuf;
645 ptrdiff_t alloc;
646
647 /* Empty cache. */
648 if (bidi_cache_idx == 0)
649 return NULL;
650
651 alloc = (bidi_shelve_header_size
652 + bidi_cache_idx * sizeof (struct bidi_it));
653 databuf = xmalloc (alloc);
654 bidi_cache_total_alloc += alloc;
655
656 memcpy (databuf, &bidi_cache_idx, sizeof (bidi_cache_idx));
657 memcpy (databuf + sizeof (bidi_cache_idx),
658 bidi_cache, bidi_cache_idx * sizeof (struct bidi_it));
659 memcpy (databuf + sizeof (bidi_cache_idx)
660 + bidi_cache_idx * sizeof (struct bidi_it),
661 bidi_cache_start_stack, sizeof (bidi_cache_start_stack));
662 memcpy (databuf + sizeof (bidi_cache_idx)
663 + bidi_cache_idx * sizeof (struct bidi_it)
664 + sizeof (bidi_cache_start_stack),
665 &bidi_cache_sp, sizeof (bidi_cache_sp));
666 memcpy (databuf + sizeof (bidi_cache_idx)
667 + bidi_cache_idx * sizeof (struct bidi_it)
668 + sizeof (bidi_cache_start_stack) + sizeof (bidi_cache_sp),
669 &bidi_cache_start, sizeof (bidi_cache_start));
670 memcpy (databuf + sizeof (bidi_cache_idx)
671 + bidi_cache_idx * sizeof (struct bidi_it)
672 + sizeof (bidi_cache_start_stack) + sizeof (bidi_cache_sp)
673 + sizeof (bidi_cache_start),
674 &bidi_cache_last_idx, sizeof (bidi_cache_last_idx));
675
676 return databuf;
677 }
678
679 /* Restore the cache state from a copy stashed away by
680 bidi_shelve_cache, and free the buffer used to stash that copy.
681 JUST_FREE non-zero means free the buffer, but don't restore the
682 cache; used when the corresponding iterator is discarded instead of
683 being restored. */
684 void
685 bidi_unshelve_cache (void *databuf, int just_free)
686 {
687 unsigned char *p = databuf;
688
689 if (!p)
690 {
691 if (!just_free)
692 {
693 /* A NULL pointer means an empty cache. */
694 bidi_cache_start = 0;
695 bidi_cache_sp = 0;
696 bidi_cache_reset ();
697 }
698 }
699 else
700 {
701 if (just_free)
702 {
703 ptrdiff_t idx;
704
705 memcpy (&idx, p, sizeof (bidi_cache_idx));
706 bidi_cache_total_alloc -=
707 bidi_shelve_header_size + idx * sizeof (struct bidi_it);
708 }
709 else
710 {
711 memcpy (&bidi_cache_idx, p, sizeof (bidi_cache_idx));
712 bidi_cache_ensure_space (bidi_cache_idx);
713 memcpy (bidi_cache, p + sizeof (bidi_cache_idx),
714 bidi_cache_idx * sizeof (struct bidi_it));
715 memcpy (bidi_cache_start_stack,
716 p + sizeof (bidi_cache_idx)
717 + bidi_cache_idx * sizeof (struct bidi_it),
718 sizeof (bidi_cache_start_stack));
719 memcpy (&bidi_cache_sp,
720 p + sizeof (bidi_cache_idx)
721 + bidi_cache_idx * sizeof (struct bidi_it)
722 + sizeof (bidi_cache_start_stack),
723 sizeof (bidi_cache_sp));
724 memcpy (&bidi_cache_start,
725 p + sizeof (bidi_cache_idx)
726 + bidi_cache_idx * sizeof (struct bidi_it)
727 + sizeof (bidi_cache_start_stack) + sizeof (bidi_cache_sp),
728 sizeof (bidi_cache_start));
729 memcpy (&bidi_cache_last_idx,
730 p + sizeof (bidi_cache_idx)
731 + bidi_cache_idx * sizeof (struct bidi_it)
732 + sizeof (bidi_cache_start_stack) + sizeof (bidi_cache_sp)
733 + sizeof (bidi_cache_start),
734 sizeof (bidi_cache_last_idx));
735 bidi_cache_total_alloc -=
736 bidi_shelve_header_size + bidi_cache_idx * sizeof (struct bidi_it);
737 }
738
739 xfree (p);
740 }
741 }
742
743 \f
744 /***********************************************************************
745 Initialization
746 ***********************************************************************/
747 static void
748 bidi_initialize (void)
749 {
750 bidi_type_table = uniprop_table (intern ("bidi-class"));
751 if (NILP (bidi_type_table))
752 abort ();
753 staticpro (&bidi_type_table);
754
755 bidi_mirror_table = uniprop_table (intern ("mirroring"));
756 if (NILP (bidi_mirror_table))
757 abort ();
758 staticpro (&bidi_mirror_table);
759
760 Qparagraph_start = intern ("paragraph-start");
761 staticpro (&Qparagraph_start);
762 paragraph_start_re = Fsymbol_value (Qparagraph_start);
763 if (!STRINGP (paragraph_start_re))
764 paragraph_start_re = build_string ("\f\\|[ \t]*$");
765 staticpro (&paragraph_start_re);
766 Qparagraph_separate = intern ("paragraph-separate");
767 staticpro (&Qparagraph_separate);
768 paragraph_separate_re = Fsymbol_value (Qparagraph_separate);
769 if (!STRINGP (paragraph_separate_re))
770 paragraph_separate_re = build_string ("[ \t\f]*$");
771 staticpro (&paragraph_separate_re);
772
773 bidi_cache_sp = 0;
774 bidi_cache_total_alloc = 0;
775
776 bidi_initialized = 1;
777 }
778
779 /* Do whatever UAX#9 clause X8 says should be done at paragraph's
780 end. */
781 static inline void
782 bidi_set_paragraph_end (struct bidi_it *bidi_it)
783 {
784 bidi_it->invalid_levels = 0;
785 bidi_it->invalid_rl_levels = -1;
786 bidi_it->stack_idx = 0;
787 bidi_it->resolved_level = bidi_it->level_stack[0].level;
788 }
789
790 /* Initialize the bidi iterator from buffer/string position CHARPOS. */
791 void
792 bidi_init_it (EMACS_INT charpos, EMACS_INT bytepos, int frame_window_p,
793 struct bidi_it *bidi_it)
794 {
795 if (! bidi_initialized)
796 bidi_initialize ();
797 if (charpos >= 0)
798 bidi_it->charpos = charpos;
799 if (bytepos >= 0)
800 bidi_it->bytepos = bytepos;
801 bidi_it->frame_window_p = frame_window_p;
802 bidi_it->nchars = -1; /* to be computed in bidi_resolve_explicit_1 */
803 bidi_it->first_elt = 1;
804 bidi_set_paragraph_end (bidi_it);
805 bidi_it->new_paragraph = 1;
806 bidi_it->separator_limit = -1;
807 bidi_it->type = NEUTRAL_B;
808 bidi_it->type_after_w1 = NEUTRAL_B;
809 bidi_it->orig_type = NEUTRAL_B;
810 bidi_it->prev_was_pdf = 0;
811 bidi_it->prev.type = bidi_it->prev.type_after_w1 =
812 bidi_it->prev.orig_type = UNKNOWN_BT;
813 bidi_it->last_strong.type = bidi_it->last_strong.type_after_w1 =
814 bidi_it->last_strong.orig_type = UNKNOWN_BT;
815 bidi_it->next_for_neutral.charpos = -1;
816 bidi_it->next_for_neutral.type =
817 bidi_it->next_for_neutral.type_after_w1 =
818 bidi_it->next_for_neutral.orig_type = UNKNOWN_BT;
819 bidi_it->prev_for_neutral.charpos = -1;
820 bidi_it->prev_for_neutral.type =
821 bidi_it->prev_for_neutral.type_after_w1 =
822 bidi_it->prev_for_neutral.orig_type = UNKNOWN_BT;
823 bidi_it->sor = L2R; /* FIXME: should it be user-selectable? */
824 bidi_it->disp_pos = -1; /* invalid/unknown */
825 bidi_it->disp_prop_p = 0;
826 /* We can only shrink the cache if we are at the bottom level of its
827 "stack". */
828 if (bidi_cache_start == 0)
829 bidi_cache_shrink ();
830 else
831 bidi_cache_reset ();
832 }
833
834 /* Perform initializations for reordering a new line of bidi text. */
835 static void
836 bidi_line_init (struct bidi_it *bidi_it)
837 {
838 bidi_it->scan_dir = 1; /* FIXME: do we need to have control on this? */
839 bidi_it->resolved_level = bidi_it->level_stack[0].level;
840 bidi_it->level_stack[0].override = NEUTRAL_DIR; /* X1 */
841 bidi_it->invalid_levels = 0;
842 bidi_it->invalid_rl_levels = -1;
843 bidi_it->next_en_pos = -1;
844 bidi_it->next_for_ws.type = UNKNOWN_BT;
845 bidi_set_sor_type (bidi_it,
846 bidi_it->paragraph_dir == R2L ? 1 : 0,
847 bidi_it->level_stack[0].level); /* X10 */
848
849 bidi_cache_reset ();
850 }
851
852 \f
853 /***********************************************************************
854 Fetching characters
855 ***********************************************************************/
856
857 /* Count bytes in string S between BEG/BEGBYTE and END. BEG and END
858 are zero-based character positions in S, BEGBYTE is byte position
859 corresponding to BEG. UNIBYTE, if non-zero, means S is a unibyte
860 string. */
861 static inline EMACS_INT
862 bidi_count_bytes (const unsigned char *s, const EMACS_INT beg,
863 const EMACS_INT begbyte, const EMACS_INT end, int unibyte)
864 {
865 EMACS_INT pos = beg;
866 const unsigned char *p = s + begbyte, *start = p;
867
868 if (unibyte)
869 p = s + end;
870 else
871 {
872 if (!CHAR_HEAD_P (*p))
873 abort ();
874
875 while (pos < end)
876 {
877 p += BYTES_BY_CHAR_HEAD (*p);
878 pos++;
879 }
880 }
881
882 return p - start;
883 }
884
885 /* Fetch and returns the character at byte position BYTEPOS. If S is
886 non-NULL, fetch the character from string S; otherwise fetch the
887 character from the current buffer. UNIBYTE non-zero means S is a
888 unibyte string. */
889 static inline int
890 bidi_char_at_pos (EMACS_INT bytepos, const unsigned char *s, int unibyte)
891 {
892 if (s)
893 {
894 if (unibyte)
895 return s[bytepos];
896 else
897 return STRING_CHAR (s + bytepos);
898 }
899 else
900 return FETCH_MULTIBYTE_CHAR (bytepos);
901 }
902
903 /* Fetch and return the character at BYTEPOS/CHARPOS. If that
904 character is covered by a display string, treat the entire run of
905 covered characters as a single character u+FFFC, and return their
906 combined length in CH_LEN and NCHARS. DISP_POS specifies the
907 character position of the next display string, or -1 if not yet
908 computed. DISP_PROP_P non-zero means that there's really a display
909 string at DISP_POS, as opposed to when we searched till DISP_POS
910 without findingone. When the next character is at or beyond that
911 position, the function updates DISP_POS with the position of the
912 next display string. STRING->s is the C string to iterate, or NULL
913 if iterating over a buffer or a Lisp string; in the latter case,
914 STRING->lstring is the Lisp string. */
915 static inline int
916 bidi_fetch_char (EMACS_INT bytepos, EMACS_INT charpos, EMACS_INT *disp_pos,
917 int *disp_prop_p, struct bidi_string_data *string,
918 int frame_window_p, EMACS_INT *ch_len, EMACS_INT *nchars)
919 {
920 int ch;
921 EMACS_INT endpos =
922 (string->s || STRINGP (string->lstring)) ? string->schars : ZV;
923 struct text_pos pos;
924
925 /* If we got past the last known position of display string, compute
926 the position of the next one. That position could be at CHARPOS. */
927 if (charpos < endpos && charpos > *disp_pos)
928 {
929 SET_TEXT_POS (pos, charpos, bytepos);
930 *disp_pos = compute_display_string_pos (&pos, string, frame_window_p,
931 disp_prop_p);
932 }
933
934 /* Fetch the character at BYTEPOS. */
935 if (charpos >= endpos)
936 {
937 ch = BIDI_EOB;
938 *ch_len = 1;
939 *nchars = 1;
940 *disp_pos = endpos;
941 *disp_prop_p = 0;
942 }
943 else if (charpos >= *disp_pos && *disp_prop_p)
944 {
945 EMACS_INT disp_end_pos;
946
947 /* We don't expect to find ourselves in the middle of a display
948 property. Hopefully, it will never be needed. */
949 if (charpos > *disp_pos)
950 abort ();
951 /* Return the Unicode Object Replacement Character to represent
952 the entire run of characters covered by the display string. */
953 ch = 0xFFFC;
954 disp_end_pos = compute_display_string_end (*disp_pos, string);
955 *nchars = disp_end_pos - *disp_pos;
956 if (*nchars <= 0)
957 abort ();
958 if (string->s)
959 *ch_len = bidi_count_bytes (string->s, *disp_pos, bytepos,
960 disp_end_pos, string->unibyte);
961 else if (STRINGP (string->lstring))
962 *ch_len = bidi_count_bytes (SDATA (string->lstring), *disp_pos,
963 bytepos, disp_end_pos, string->unibyte);
964 else
965 *ch_len = CHAR_TO_BYTE (disp_end_pos) - bytepos;
966 }
967 else
968 {
969 if (string->s)
970 {
971 int len;
972
973 if (!string->unibyte)
974 {
975 ch = STRING_CHAR_AND_LENGTH (string->s + bytepos, len);
976 *ch_len = len;
977 }
978 else
979 {
980 ch = UNIBYTE_TO_CHAR (string->s[bytepos]);
981 *ch_len = 1;
982 }
983 }
984 else if (STRINGP (string->lstring))
985 {
986 int len;
987
988 if (!string->unibyte)
989 {
990 ch = STRING_CHAR_AND_LENGTH (SDATA (string->lstring) + bytepos,
991 len);
992 *ch_len = len;
993 }
994 else
995 {
996 ch = UNIBYTE_TO_CHAR (SREF (string->lstring, bytepos));
997 *ch_len = 1;
998 }
999 }
1000 else
1001 {
1002 ch = FETCH_MULTIBYTE_CHAR (bytepos);
1003 *ch_len = CHAR_BYTES (ch);
1004 }
1005 *nchars = 1;
1006 }
1007
1008 /* If we just entered a run of characters covered by a display
1009 string, compute the position of the next display string. */
1010 if (charpos + *nchars <= endpos && charpos + *nchars > *disp_pos
1011 && *disp_prop_p)
1012 {
1013 SET_TEXT_POS (pos, charpos + *nchars, bytepos + *ch_len);
1014 *disp_pos = compute_display_string_pos (&pos, string, frame_window_p,
1015 disp_prop_p);
1016 }
1017
1018 return ch;
1019 }
1020
1021 \f
1022 /***********************************************************************
1023 Determining paragraph direction
1024 ***********************************************************************/
1025
1026 /* Check if buffer position CHARPOS/BYTEPOS is the end of a paragraph.
1027 Value is the non-negative length of the paragraph separator
1028 following the buffer position, -1 if position is at the beginning
1029 of a new paragraph, or -2 if position is neither at beginning nor
1030 at end of a paragraph. */
1031 static EMACS_INT
1032 bidi_at_paragraph_end (EMACS_INT charpos, EMACS_INT bytepos)
1033 {
1034 Lisp_Object sep_re;
1035 Lisp_Object start_re;
1036 EMACS_INT val;
1037
1038 sep_re = paragraph_separate_re;
1039 start_re = paragraph_start_re;
1040
1041 val = fast_looking_at (sep_re, charpos, bytepos, ZV, ZV_BYTE, Qnil);
1042 if (val < 0)
1043 {
1044 if (fast_looking_at (start_re, charpos, bytepos, ZV, ZV_BYTE, Qnil) >= 0)
1045 val = -1;
1046 else
1047 val = -2;
1048 }
1049
1050 return val;
1051 }
1052
1053 /* Find the beginning of this paragraph by looking back in the buffer.
1054 Value is the byte position of the paragraph's beginning. */
1055 static EMACS_INT
1056 bidi_find_paragraph_start (EMACS_INT pos, EMACS_INT pos_byte)
1057 {
1058 Lisp_Object re = paragraph_start_re;
1059 EMACS_INT limit = ZV, limit_byte = ZV_BYTE;
1060
1061 while (pos_byte > BEGV_BYTE
1062 && fast_looking_at (re, pos, pos_byte, limit, limit_byte, Qnil) < 0)
1063 {
1064 /* FIXME: What if the paragraph beginning is covered by a
1065 display string? And what if a display string covering some
1066 of the text over which we scan back includes
1067 paragraph_start_re? */
1068 pos = find_next_newline_no_quit (pos - 1, -1);
1069 pos_byte = CHAR_TO_BYTE (pos);
1070 }
1071 return pos_byte;
1072 }
1073
1074 /* Determine the base direction, a.k.a. base embedding level, of the
1075 paragraph we are about to iterate through. If DIR is either L2R or
1076 R2L, just use that. Otherwise, determine the paragraph direction
1077 from the first strong directional character of the paragraph.
1078
1079 NO_DEFAULT_P non-zero means don't default to L2R if the paragraph
1080 has no strong directional characters and both DIR and
1081 bidi_it->paragraph_dir are NEUTRAL_DIR. In that case, search back
1082 in the buffer until a paragraph is found with a strong character,
1083 or until hitting BEGV. In the latter case, fall back to L2R. This
1084 flag is used in current-bidi-paragraph-direction.
1085
1086 Note that this function gives the paragraph separator the same
1087 direction as the preceding paragraph, even though Emacs generally
1088 views the separartor as not belonging to any paragraph. */
1089 void
1090 bidi_paragraph_init (bidi_dir_t dir, struct bidi_it *bidi_it, int no_default_p)
1091 {
1092 EMACS_INT bytepos = bidi_it->bytepos;
1093 int string_p = bidi_it->string.s != NULL || STRINGP (bidi_it->string.lstring);
1094 EMACS_INT pstartbyte;
1095 /* Note that begbyte is a byte position, while end is a character
1096 position. Yes, this is ugly, but we are trying to avoid costly
1097 calls to BYTE_TO_CHAR and its ilk. */
1098 EMACS_INT begbyte = string_p ? 0 : BEGV_BYTE;
1099 EMACS_INT end = string_p ? bidi_it->string.schars : ZV;
1100
1101 /* Special case for an empty buffer. */
1102 if (bytepos == begbyte && bidi_it->charpos == end)
1103 dir = L2R;
1104 /* We should never be called at EOB or before BEGV. */
1105 else if (bidi_it->charpos >= end || bytepos < begbyte)
1106 abort ();
1107
1108 if (dir == L2R)
1109 {
1110 bidi_it->paragraph_dir = L2R;
1111 bidi_it->new_paragraph = 0;
1112 }
1113 else if (dir == R2L)
1114 {
1115 bidi_it->paragraph_dir = R2L;
1116 bidi_it->new_paragraph = 0;
1117 }
1118 else if (dir == NEUTRAL_DIR) /* P2 */
1119 {
1120 int ch;
1121 EMACS_INT ch_len, nchars;
1122 EMACS_INT pos, disp_pos = -1;
1123 int disp_prop_p = 0;
1124 bidi_type_t type;
1125 const unsigned char *s;
1126
1127 if (!bidi_initialized)
1128 bidi_initialize ();
1129
1130 /* If we are inside a paragraph separator, we are just waiting
1131 for the separator to be exhausted; use the previous paragraph
1132 direction. But don't do that if we have been just reseated,
1133 because we need to reinitialize below in that case. */
1134 if (!bidi_it->first_elt
1135 && bidi_it->charpos < bidi_it->separator_limit)
1136 return;
1137
1138 /* If we are on a newline, get past it to where the next
1139 paragraph might start. But don't do that at BEGV since then
1140 we are potentially in a new paragraph that doesn't yet
1141 exist. */
1142 pos = bidi_it->charpos;
1143 s = STRINGP (bidi_it->string.lstring) ?
1144 SDATA (bidi_it->string.lstring) : bidi_it->string.s;
1145 if (bytepos > begbyte
1146 && bidi_char_at_pos (bytepos, s, bidi_it->string.unibyte) == '\n')
1147 {
1148 bytepos++;
1149 pos++;
1150 }
1151
1152 /* We are either at the beginning of a paragraph or in the
1153 middle of it. Find where this paragraph starts. */
1154 if (string_p)
1155 {
1156 /* We don't support changes of paragraph direction inside a
1157 string. It is treated as a single paragraph. */
1158 pstartbyte = 0;
1159 }
1160 else
1161 pstartbyte = bidi_find_paragraph_start (pos, bytepos);
1162 bidi_it->separator_limit = -1;
1163 bidi_it->new_paragraph = 0;
1164
1165 /* The following loop is run more than once only if NO_DEFAULT_P
1166 is non-zero, and only if we are iterating on a buffer. */
1167 do {
1168 bytepos = pstartbyte;
1169 if (!string_p)
1170 pos = BYTE_TO_CHAR (bytepos);
1171 ch = bidi_fetch_char (bytepos, pos, &disp_pos, &disp_prop_p,
1172 &bidi_it->string,
1173 bidi_it->frame_window_p, &ch_len, &nchars);
1174 type = bidi_get_type (ch, NEUTRAL_DIR);
1175
1176 for (pos += nchars, bytepos += ch_len;
1177 (bidi_get_category (type) != STRONG)
1178 || (bidi_ignore_explicit_marks_for_paragraph_level
1179 && (type == RLE || type == RLO
1180 || type == LRE || type == LRO));
1181 type = bidi_get_type (ch, NEUTRAL_DIR))
1182 {
1183 if (pos >= end)
1184 {
1185 /* Pretend there's a paragraph separator at end of
1186 buffer/string. */
1187 type = NEUTRAL_B;
1188 break;
1189 }
1190 if (!string_p
1191 && type == NEUTRAL_B
1192 && bidi_at_paragraph_end (pos, bytepos) >= -1)
1193 break;
1194 /* Fetch next character and advance to get past it. */
1195 ch = bidi_fetch_char (bytepos, pos, &disp_pos,
1196 &disp_prop_p, &bidi_it->string,
1197 bidi_it->frame_window_p, &ch_len, &nchars);
1198 pos += nchars;
1199 bytepos += ch_len;
1200 }
1201 if ((type == STRONG_R || type == STRONG_AL) /* P3 */
1202 || (!bidi_ignore_explicit_marks_for_paragraph_level
1203 && (type == RLO || type == RLE)))
1204 bidi_it->paragraph_dir = R2L;
1205 else if (type == STRONG_L
1206 || (!bidi_ignore_explicit_marks_for_paragraph_level
1207 && (type == LRO || type == LRE)))
1208 bidi_it->paragraph_dir = L2R;
1209 if (!string_p
1210 && no_default_p && bidi_it->paragraph_dir == NEUTRAL_DIR)
1211 {
1212 /* If this paragraph is at BEGV, default to L2R. */
1213 if (pstartbyte == BEGV_BYTE)
1214 bidi_it->paragraph_dir = L2R; /* P3 and HL1 */
1215 else
1216 {
1217 EMACS_INT prevpbyte = pstartbyte;
1218 EMACS_INT p = BYTE_TO_CHAR (pstartbyte), pbyte = pstartbyte;
1219
1220 /* Find the beginning of the previous paragraph, if any. */
1221 while (pbyte > BEGV_BYTE && prevpbyte >= pstartbyte)
1222 {
1223 /* FXIME: What if p is covered by a display
1224 string? See also a FIXME inside
1225 bidi_find_paragraph_start. */
1226 p--;
1227 pbyte = CHAR_TO_BYTE (p);
1228 prevpbyte = bidi_find_paragraph_start (p, pbyte);
1229 }
1230 pstartbyte = prevpbyte;
1231 }
1232 }
1233 } while (!string_p
1234 && no_default_p && bidi_it->paragraph_dir == NEUTRAL_DIR);
1235 }
1236 else
1237 abort ();
1238
1239 /* Contrary to UAX#9 clause P3, we only default the paragraph
1240 direction to L2R if we have no previous usable paragraph
1241 direction. This is allowed by the HL1 clause. */
1242 if (bidi_it->paragraph_dir != L2R && bidi_it->paragraph_dir != R2L)
1243 bidi_it->paragraph_dir = L2R; /* P3 and HL1 ``higher-level protocols'' */
1244 if (bidi_it->paragraph_dir == R2L)
1245 bidi_it->level_stack[0].level = 1;
1246 else
1247 bidi_it->level_stack[0].level = 0;
1248
1249 bidi_line_init (bidi_it);
1250 }
1251
1252 \f
1253 /***********************************************************************
1254 Resolving explicit and implicit levels.
1255 The rest of this file constitutes the core of the UBA implementation.
1256 ***********************************************************************/
1257
1258 static inline int
1259 bidi_explicit_dir_char (int ch)
1260 {
1261 bidi_type_t ch_type;
1262
1263 if (!bidi_initialized)
1264 abort ();
1265 ch_type = (bidi_type_t) XINT (CHAR_TABLE_REF (bidi_type_table, ch));
1266 return (ch_type == LRE || ch_type == LRO
1267 || ch_type == RLE || ch_type == RLO
1268 || ch_type == PDF);
1269 }
1270
1271 /* A helper function for bidi_resolve_explicit. It advances to the
1272 next character in logical order and determines the new embedding
1273 level and directional override, but does not take into account
1274 empty embeddings. */
1275 static int
1276 bidi_resolve_explicit_1 (struct bidi_it *bidi_it)
1277 {
1278 int curchar;
1279 bidi_type_t type;
1280 int current_level;
1281 int new_level;
1282 bidi_dir_t override;
1283 int string_p = bidi_it->string.s != NULL || STRINGP (bidi_it->string.lstring);
1284
1285 /* If reseat()'ed, don't advance, so as to start iteration from the
1286 position where we were reseated. bidi_it->bytepos can be less
1287 than BEGV_BYTE after reseat to BEGV. */
1288 if (bidi_it->bytepos < (string_p ? 0 : BEGV_BYTE)
1289 || bidi_it->first_elt)
1290 {
1291 bidi_it->first_elt = 0;
1292 if (string_p)
1293 {
1294 const unsigned char *p =
1295 STRINGP (bidi_it->string.lstring)
1296 ? SDATA (bidi_it->string.lstring) : bidi_it->string.s;
1297
1298 if (bidi_it->charpos < 0)
1299 bidi_it->charpos = 0;
1300 bidi_it->bytepos = bidi_count_bytes (p, 0, 0, bidi_it->charpos,
1301 bidi_it->string.unibyte);
1302 }
1303 else
1304 {
1305 if (bidi_it->charpos < BEGV)
1306 bidi_it->charpos = BEGV;
1307 bidi_it->bytepos = CHAR_TO_BYTE (bidi_it->charpos);
1308 }
1309 }
1310 /* Don't move at end of buffer/string. */
1311 else if (bidi_it->charpos < (string_p ? bidi_it->string.schars : ZV))
1312 {
1313 /* Advance to the next character, skipping characters covered by
1314 display strings (nchars > 1). */
1315 if (bidi_it->nchars <= 0)
1316 abort ();
1317 bidi_it->charpos += bidi_it->nchars;
1318 if (bidi_it->ch_len == 0)
1319 abort ();
1320 bidi_it->bytepos += bidi_it->ch_len;
1321 }
1322
1323 current_level = bidi_it->level_stack[bidi_it->stack_idx].level; /* X1 */
1324 override = bidi_it->level_stack[bidi_it->stack_idx].override;
1325 new_level = current_level;
1326
1327 if (bidi_it->charpos >= (string_p ? bidi_it->string.schars : ZV))
1328 {
1329 curchar = BIDI_EOB;
1330 bidi_it->ch_len = 1;
1331 bidi_it->nchars = 1;
1332 bidi_it->disp_pos = (string_p ? bidi_it->string.schars : ZV);
1333 bidi_it->disp_prop_p = 0;
1334 }
1335 else
1336 {
1337 /* Fetch the character at BYTEPOS. If it is covered by a
1338 display string, treat the entire run of covered characters as
1339 a single character u+FFFC. */
1340 curchar = bidi_fetch_char (bidi_it->bytepos, bidi_it->charpos,
1341 &bidi_it->disp_pos, &bidi_it->disp_prop_p,
1342 &bidi_it->string, bidi_it->frame_window_p,
1343 &bidi_it->ch_len, &bidi_it->nchars);
1344 }
1345 bidi_it->ch = curchar;
1346
1347 /* Don't apply directional override here, as all the types we handle
1348 below will not be affected by the override anyway, and we need
1349 the original type unaltered. The override will be applied in
1350 bidi_resolve_weak. */
1351 type = bidi_get_type (curchar, NEUTRAL_DIR);
1352 bidi_it->orig_type = type;
1353 bidi_check_type (bidi_it->orig_type);
1354
1355 if (type != PDF)
1356 bidi_it->prev_was_pdf = 0;
1357
1358 bidi_it->type_after_w1 = UNKNOWN_BT;
1359
1360 switch (type)
1361 {
1362 case RLE: /* X2 */
1363 case RLO: /* X4 */
1364 bidi_it->type_after_w1 = type;
1365 bidi_check_type (bidi_it->type_after_w1);
1366 type = WEAK_BN; /* X9/Retaining */
1367 if (bidi_it->ignore_bn_limit <= -1)
1368 {
1369 if (current_level <= BIDI_MAXLEVEL - 4)
1370 {
1371 /* Compute the least odd embedding level greater than
1372 the current level. */
1373 new_level = ((current_level + 1) & ~1) + 1;
1374 if (bidi_it->type_after_w1 == RLE)
1375 override = NEUTRAL_DIR;
1376 else
1377 override = R2L;
1378 if (current_level == BIDI_MAXLEVEL - 4)
1379 bidi_it->invalid_rl_levels = 0;
1380 bidi_push_embedding_level (bidi_it, new_level, override);
1381 }
1382 else
1383 {
1384 bidi_it->invalid_levels++;
1385 /* See the commentary about invalid_rl_levels below. */
1386 if (bidi_it->invalid_rl_levels < 0)
1387 bidi_it->invalid_rl_levels = 0;
1388 bidi_it->invalid_rl_levels++;
1389 }
1390 }
1391 else if (bidi_it->prev.type_after_w1 == WEAK_EN /* W5/Retaining */
1392 || bidi_it->next_en_pos > bidi_it->charpos)
1393 type = WEAK_EN;
1394 break;
1395 case LRE: /* X3 */
1396 case LRO: /* X5 */
1397 bidi_it->type_after_w1 = type;
1398 bidi_check_type (bidi_it->type_after_w1);
1399 type = WEAK_BN; /* X9/Retaining */
1400 if (bidi_it->ignore_bn_limit <= -1)
1401 {
1402 if (current_level <= BIDI_MAXLEVEL - 5)
1403 {
1404 /* Compute the least even embedding level greater than
1405 the current level. */
1406 new_level = ((current_level + 2) & ~1);
1407 if (bidi_it->type_after_w1 == LRE)
1408 override = NEUTRAL_DIR;
1409 else
1410 override = L2R;
1411 bidi_push_embedding_level (bidi_it, new_level, override);
1412 }
1413 else
1414 {
1415 bidi_it->invalid_levels++;
1416 /* invalid_rl_levels counts invalid levels encountered
1417 while the embedding level was already too high for
1418 LRE/LRO, but not for RLE/RLO. That is because
1419 there may be exactly one PDF which we should not
1420 ignore even though invalid_levels is non-zero.
1421 invalid_rl_levels helps to know what PDF is
1422 that. */
1423 if (bidi_it->invalid_rl_levels >= 0)
1424 bidi_it->invalid_rl_levels++;
1425 }
1426 }
1427 else if (bidi_it->prev.type_after_w1 == WEAK_EN /* W5/Retaining */
1428 || bidi_it->next_en_pos > bidi_it->charpos)
1429 type = WEAK_EN;
1430 break;
1431 case PDF: /* X7 */
1432 bidi_it->type_after_w1 = type;
1433 bidi_check_type (bidi_it->type_after_w1);
1434 type = WEAK_BN; /* X9/Retaining */
1435 if (bidi_it->ignore_bn_limit <= -1)
1436 {
1437 if (!bidi_it->invalid_rl_levels)
1438 {
1439 new_level = bidi_pop_embedding_level (bidi_it);
1440 bidi_it->invalid_rl_levels = -1;
1441 if (bidi_it->invalid_levels)
1442 bidi_it->invalid_levels--;
1443 /* else nothing: UAX#9 says to ignore invalid PDFs */
1444 }
1445 if (!bidi_it->invalid_levels)
1446 new_level = bidi_pop_embedding_level (bidi_it);
1447 else
1448 {
1449 bidi_it->invalid_levels--;
1450 bidi_it->invalid_rl_levels--;
1451 }
1452 }
1453 else if (bidi_it->prev.type_after_w1 == WEAK_EN /* W5/Retaining */
1454 || bidi_it->next_en_pos > bidi_it->charpos)
1455 type = WEAK_EN;
1456 break;
1457 default:
1458 /* Nothing. */
1459 break;
1460 }
1461
1462 bidi_it->type = type;
1463 bidi_check_type (bidi_it->type);
1464
1465 return new_level;
1466 }
1467
1468 /* Given an iterator state in BIDI_IT, advance one character position
1469 in the buffer/string to the next character (in the logical order),
1470 resolve any explicit embeddings and directional overrides, and
1471 return the embedding level of the character after resolving
1472 explicit directives and ignoring empty embeddings. */
1473 static int
1474 bidi_resolve_explicit (struct bidi_it *bidi_it)
1475 {
1476 int prev_level = bidi_it->level_stack[bidi_it->stack_idx].level;
1477 int new_level = bidi_resolve_explicit_1 (bidi_it);
1478 EMACS_INT eob = bidi_it->string.s ? bidi_it->string.schars : ZV;
1479 const unsigned char *s = STRINGP (bidi_it->string.lstring)
1480 ? SDATA (bidi_it->string.lstring) : bidi_it->string.s;
1481
1482 if (prev_level < new_level
1483 && bidi_it->type == WEAK_BN
1484 && bidi_it->ignore_bn_limit == -1 /* only if not already known */
1485 && bidi_it->charpos < eob /* not already at EOB */
1486 && bidi_explicit_dir_char (bidi_char_at_pos (bidi_it->bytepos
1487 + bidi_it->ch_len, s,
1488 bidi_it->string.unibyte)))
1489 {
1490 /* Avoid pushing and popping embedding levels if the level run
1491 is empty, as this breaks level runs where it shouldn't.
1492 UAX#9 removes all the explicit embedding and override codes,
1493 so empty embeddings disappear without a trace. We need to
1494 behave as if we did the same. */
1495 struct bidi_it saved_it;
1496 int level = prev_level;
1497
1498 bidi_copy_it (&saved_it, bidi_it);
1499
1500 while (bidi_explicit_dir_char (bidi_char_at_pos (bidi_it->bytepos
1501 + bidi_it->ch_len, s,
1502 bidi_it->string.unibyte)))
1503 {
1504 /* This advances to the next character, skipping any
1505 characters covered by display strings. */
1506 level = bidi_resolve_explicit_1 (bidi_it);
1507 /* If string.lstring was relocated inside bidi_resolve_explicit_1,
1508 a pointer to its data is no longer valid. */
1509 if (STRINGP (bidi_it->string.lstring))
1510 s = SDATA (bidi_it->string.lstring);
1511 }
1512
1513 if (bidi_it->nchars <= 0)
1514 abort ();
1515 if (level == prev_level) /* empty embedding */
1516 saved_it.ignore_bn_limit = bidi_it->charpos + bidi_it->nchars;
1517 else /* this embedding is non-empty */
1518 saved_it.ignore_bn_limit = -2;
1519
1520 bidi_copy_it (bidi_it, &saved_it);
1521 if (bidi_it->ignore_bn_limit > -1)
1522 {
1523 /* We pushed a level, but we shouldn't have. Undo that. */
1524 if (!bidi_it->invalid_rl_levels)
1525 {
1526 new_level = bidi_pop_embedding_level (bidi_it);
1527 bidi_it->invalid_rl_levels = -1;
1528 if (bidi_it->invalid_levels)
1529 bidi_it->invalid_levels--;
1530 }
1531 if (!bidi_it->invalid_levels)
1532 new_level = bidi_pop_embedding_level (bidi_it);
1533 else
1534 {
1535 bidi_it->invalid_levels--;
1536 bidi_it->invalid_rl_levels--;
1537 }
1538 }
1539 }
1540
1541 if (bidi_it->type == NEUTRAL_B) /* X8 */
1542 {
1543 bidi_set_paragraph_end (bidi_it);
1544 /* This is needed by bidi_resolve_weak below, and in L1. */
1545 bidi_it->type_after_w1 = bidi_it->type;
1546 bidi_check_type (bidi_it->type_after_w1);
1547 }
1548
1549 return new_level;
1550 }
1551
1552 /* Advance in the buffer/string, resolve weak types and return the
1553 type of the next character after weak type resolution. */
1554 static bidi_type_t
1555 bidi_resolve_weak (struct bidi_it *bidi_it)
1556 {
1557 bidi_type_t type;
1558 bidi_dir_t override;
1559 int prev_level = bidi_it->level_stack[bidi_it->stack_idx].level;
1560 int new_level = bidi_resolve_explicit (bidi_it);
1561 int next_char;
1562 bidi_type_t type_of_next;
1563 struct bidi_it saved_it;
1564 EMACS_INT eob =
1565 (STRINGP (bidi_it->string.lstring) || bidi_it->string.s)
1566 ? bidi_it->string.schars : ZV;
1567
1568 type = bidi_it->type;
1569 override = bidi_it->level_stack[bidi_it->stack_idx].override;
1570
1571 if (type == UNKNOWN_BT
1572 || type == LRE
1573 || type == LRO
1574 || type == RLE
1575 || type == RLO
1576 || type == PDF)
1577 abort ();
1578
1579 if (new_level != prev_level
1580 || bidi_it->type == NEUTRAL_B)
1581 {
1582 /* We've got a new embedding level run, compute the directional
1583 type of sor and initialize per-run variables (UAX#9, clause
1584 X10). */
1585 bidi_set_sor_type (bidi_it, prev_level, new_level);
1586 }
1587 else if (type == NEUTRAL_S || type == NEUTRAL_WS
1588 || type == WEAK_BN || type == STRONG_AL)
1589 bidi_it->type_after_w1 = type; /* needed in L1 */
1590 bidi_check_type (bidi_it->type_after_w1);
1591
1592 /* Level and directional override status are already recorded in
1593 bidi_it, and do not need any change; see X6. */
1594 if (override == R2L) /* X6 */
1595 type = STRONG_R;
1596 else if (override == L2R)
1597 type = STRONG_L;
1598 else
1599 {
1600 if (type == WEAK_NSM) /* W1 */
1601 {
1602 /* Note that we don't need to consider the case where the
1603 prev character has its type overridden by an RLO or LRO,
1604 because then either the type of this NSM would have been
1605 also overridden, or the previous character is outside the
1606 current level run, and thus not relevant to this NSM.
1607 This is why NSM gets the type_after_w1 of the previous
1608 character. */
1609 if (bidi_it->prev.type_after_w1 != UNKNOWN_BT
1610 /* if type_after_w1 is NEUTRAL_B, this NSM is at sor */
1611 && bidi_it->prev.type_after_w1 != NEUTRAL_B)
1612 type = bidi_it->prev.type_after_w1;
1613 else if (bidi_it->sor == R2L)
1614 type = STRONG_R;
1615 else if (bidi_it->sor == L2R)
1616 type = STRONG_L;
1617 else /* shouldn't happen! */
1618 abort ();
1619 }
1620 if (type == WEAK_EN /* W2 */
1621 && bidi_it->last_strong.type_after_w1 == STRONG_AL)
1622 type = WEAK_AN;
1623 else if (type == STRONG_AL) /* W3 */
1624 type = STRONG_R;
1625 else if ((type == WEAK_ES /* W4 */
1626 && bidi_it->prev.type_after_w1 == WEAK_EN
1627 && bidi_it->prev.orig_type == WEAK_EN)
1628 || (type == WEAK_CS
1629 && ((bidi_it->prev.type_after_w1 == WEAK_EN
1630 && bidi_it->prev.orig_type == WEAK_EN)
1631 || bidi_it->prev.type_after_w1 == WEAK_AN)))
1632 {
1633 const unsigned char *s =
1634 STRINGP (bidi_it->string.lstring)
1635 ? SDATA (bidi_it->string.lstring) : bidi_it->string.s;
1636
1637 next_char =
1638 bidi_it->charpos + bidi_it->nchars >= eob
1639 ? BIDI_EOB
1640 : bidi_char_at_pos (bidi_it->bytepos + bidi_it->ch_len, s,
1641 bidi_it->string.unibyte);
1642 type_of_next = bidi_get_type (next_char, override);
1643
1644 if (type_of_next == WEAK_BN
1645 || bidi_explicit_dir_char (next_char))
1646 {
1647 bidi_copy_it (&saved_it, bidi_it);
1648 while (bidi_resolve_explicit (bidi_it) == new_level
1649 && bidi_it->type == WEAK_BN)
1650 ;
1651 type_of_next = bidi_it->type;
1652 bidi_copy_it (bidi_it, &saved_it);
1653 }
1654
1655 /* If the next character is EN, but the last strong-type
1656 character is AL, that next EN will be changed to AN when
1657 we process it in W2 above. So in that case, this ES
1658 should not be changed into EN. */
1659 if (type == WEAK_ES
1660 && type_of_next == WEAK_EN
1661 && bidi_it->last_strong.type_after_w1 != STRONG_AL)
1662 type = WEAK_EN;
1663 else if (type == WEAK_CS)
1664 {
1665 if (bidi_it->prev.type_after_w1 == WEAK_AN
1666 && (type_of_next == WEAK_AN
1667 /* If the next character is EN, but the last
1668 strong-type character is AL, EN will be later
1669 changed to AN when we process it in W2 above.
1670 So in that case, this ES should not be
1671 changed into EN. */
1672 || (type_of_next == WEAK_EN
1673 && bidi_it->last_strong.type_after_w1 == STRONG_AL)))
1674 type = WEAK_AN;
1675 else if (bidi_it->prev.type_after_w1 == WEAK_EN
1676 && type_of_next == WEAK_EN
1677 && bidi_it->last_strong.type_after_w1 != STRONG_AL)
1678 type = WEAK_EN;
1679 }
1680 }
1681 else if (type == WEAK_ET /* W5: ET with EN before or after it */
1682 || type == WEAK_BN) /* W5/Retaining */
1683 {
1684 if (bidi_it->prev.type_after_w1 == WEAK_EN /* ET/BN w/EN before it */
1685 || bidi_it->next_en_pos > bidi_it->charpos)
1686 type = WEAK_EN;
1687 else /* W5: ET/BN with EN after it. */
1688 {
1689 EMACS_INT en_pos = bidi_it->charpos + bidi_it->nchars;
1690 const unsigned char *s =
1691 STRINGP (bidi_it->string.lstring)
1692 ? SDATA (bidi_it->string.lstring) : bidi_it->string.s;
1693
1694 if (bidi_it->nchars <= 0)
1695 abort ();
1696 next_char =
1697 bidi_it->charpos + bidi_it->nchars >= eob
1698 ? BIDI_EOB
1699 : bidi_char_at_pos (bidi_it->bytepos + bidi_it->ch_len, s,
1700 bidi_it->string.unibyte);
1701 type_of_next = bidi_get_type (next_char, override);
1702
1703 if (type_of_next == WEAK_ET
1704 || type_of_next == WEAK_BN
1705 || bidi_explicit_dir_char (next_char))
1706 {
1707 bidi_copy_it (&saved_it, bidi_it);
1708 while (bidi_resolve_explicit (bidi_it) == new_level
1709 && (bidi_it->type == WEAK_BN
1710 || bidi_it->type == WEAK_ET))
1711 ;
1712 type_of_next = bidi_it->type;
1713 en_pos = bidi_it->charpos;
1714 bidi_copy_it (bidi_it, &saved_it);
1715 }
1716 if (type_of_next == WEAK_EN)
1717 {
1718 /* If the last strong character is AL, the EN we've
1719 found will become AN when we get to it (W2). */
1720 if (bidi_it->last_strong.type_after_w1 != STRONG_AL)
1721 {
1722 type = WEAK_EN;
1723 /* Remember this EN position, to speed up processing
1724 of the next ETs. */
1725 bidi_it->next_en_pos = en_pos;
1726 }
1727 else if (type == WEAK_BN)
1728 type = NEUTRAL_ON; /* W6/Retaining */
1729 }
1730 }
1731 }
1732 }
1733
1734 if (type == WEAK_ES || type == WEAK_ET || type == WEAK_CS /* W6 */
1735 || (type == WEAK_BN
1736 && (bidi_it->prev.type_after_w1 == WEAK_CS /* W6/Retaining */
1737 || bidi_it->prev.type_after_w1 == WEAK_ES
1738 || bidi_it->prev.type_after_w1 == WEAK_ET)))
1739 type = NEUTRAL_ON;
1740
1741 /* Store the type we've got so far, before we clobber it with strong
1742 types in W7 and while resolving neutral types. But leave alone
1743 the original types that were recorded above, because we will need
1744 them for the L1 clause. */
1745 if (bidi_it->type_after_w1 == UNKNOWN_BT)
1746 bidi_it->type_after_w1 = type;
1747 bidi_check_type (bidi_it->type_after_w1);
1748
1749 if (type == WEAK_EN) /* W7 */
1750 {
1751 if ((bidi_it->last_strong.type_after_w1 == STRONG_L)
1752 || (bidi_it->last_strong.type == UNKNOWN_BT && bidi_it->sor == L2R))
1753 type = STRONG_L;
1754 }
1755
1756 bidi_it->type = type;
1757 bidi_check_type (bidi_it->type);
1758 return type;
1759 }
1760
1761 /* Resolve the type of a neutral character according to the type of
1762 surrounding strong text and the current embedding level. */
1763 static inline bidi_type_t
1764 bidi_resolve_neutral_1 (bidi_type_t prev_type, bidi_type_t next_type, int lev)
1765 {
1766 /* N1: European and Arabic numbers are treated as though they were R. */
1767 if (next_type == WEAK_EN || next_type == WEAK_AN)
1768 next_type = STRONG_R;
1769 if (prev_type == WEAK_EN || prev_type == WEAK_AN)
1770 prev_type = STRONG_R;
1771
1772 if (next_type == prev_type) /* N1 */
1773 return next_type;
1774 else if ((lev & 1) == 0) /* N2 */
1775 return STRONG_L;
1776 else
1777 return STRONG_R;
1778 }
1779
1780 static bidi_type_t
1781 bidi_resolve_neutral (struct bidi_it *bidi_it)
1782 {
1783 int prev_level = bidi_it->level_stack[bidi_it->stack_idx].level;
1784 bidi_type_t type = bidi_resolve_weak (bidi_it);
1785 int current_level = bidi_it->level_stack[bidi_it->stack_idx].level;
1786
1787 if (!(type == STRONG_R
1788 || type == STRONG_L
1789 || type == WEAK_BN
1790 || type == WEAK_EN
1791 || type == WEAK_AN
1792 || type == NEUTRAL_B
1793 || type == NEUTRAL_S
1794 || type == NEUTRAL_WS
1795 || type == NEUTRAL_ON))
1796 abort ();
1797
1798 if (bidi_get_category (type) == NEUTRAL
1799 || (type == WEAK_BN && prev_level == current_level))
1800 {
1801 if (bidi_it->next_for_neutral.type != UNKNOWN_BT)
1802 type = bidi_resolve_neutral_1 (bidi_it->prev_for_neutral.type,
1803 bidi_it->next_for_neutral.type,
1804 current_level);
1805 else
1806 {
1807 /* Arrrgh!! The UAX#9 algorithm is too deeply entrenched in
1808 the assumption of batch-style processing; see clauses W4,
1809 W5, and especially N1, which require to look far forward
1810 (as well as back) in the buffer/string. May the fleas of
1811 a thousand camels infest the armpits of those who design
1812 supposedly general-purpose algorithms by looking at their
1813 own implementations, and fail to consider other possible
1814 implementations! */
1815 struct bidi_it saved_it;
1816 bidi_type_t next_type;
1817
1818 if (bidi_it->scan_dir == -1)
1819 abort ();
1820
1821 bidi_copy_it (&saved_it, bidi_it);
1822 /* Scan the text forward until we find the first non-neutral
1823 character, and then use that to resolve the neutral we
1824 are dealing with now. We also cache the scanned iterator
1825 states, to salvage some of the effort later. */
1826 bidi_cache_iterator_state (bidi_it, 0);
1827 do {
1828 /* Record the info about the previous character, so that
1829 it will be cached below with this state. */
1830 if (bidi_it->type_after_w1 != WEAK_BN /* W1/Retaining */
1831 && bidi_it->type != WEAK_BN)
1832 bidi_remember_char (&bidi_it->prev, bidi_it);
1833 type = bidi_resolve_weak (bidi_it);
1834 /* Paragraph separators have their levels fully resolved
1835 at this point, so cache them as resolved. */
1836 bidi_cache_iterator_state (bidi_it, type == NEUTRAL_B);
1837 /* FIXME: implement L1 here, by testing for a newline and
1838 resetting the level for any sequence of whitespace
1839 characters adjacent to it. */
1840 } while (!(type == NEUTRAL_B
1841 || (type != WEAK_BN
1842 && bidi_get_category (type) != NEUTRAL)
1843 /* This is all per level run, so stop when we
1844 reach the end of this level run. */
1845 || bidi_it->level_stack[bidi_it->stack_idx].level !=
1846 current_level));
1847
1848 bidi_remember_char (&saved_it.next_for_neutral, bidi_it);
1849
1850 switch (type)
1851 {
1852 case STRONG_L:
1853 case STRONG_R:
1854 case STRONG_AL:
1855 next_type = type;
1856 break;
1857 case WEAK_EN:
1858 case WEAK_AN:
1859 /* N1: ``European and Arabic numbers are treated as
1860 though they were R.'' */
1861 next_type = STRONG_R;
1862 saved_it.next_for_neutral.type = STRONG_R;
1863 break;
1864 case WEAK_BN:
1865 if (!bidi_explicit_dir_char (bidi_it->ch))
1866 abort (); /* can't happen: BNs are skipped */
1867 /* FALLTHROUGH */
1868 case NEUTRAL_B:
1869 /* Marched all the way to the end of this level run.
1870 We need to use the eor type, whose information is
1871 stored by bidi_set_sor_type in the prev_for_neutral
1872 member. */
1873 if (saved_it.type != WEAK_BN
1874 || bidi_get_category (bidi_it->prev.type_after_w1) == NEUTRAL)
1875 {
1876 next_type = bidi_it->prev_for_neutral.type;
1877 saved_it.next_for_neutral.type = next_type;
1878 bidi_check_type (next_type);
1879 }
1880 else
1881 {
1882 /* This is a BN which does not adjoin neutrals.
1883 Leave its type alone. */
1884 bidi_copy_it (bidi_it, &saved_it);
1885 return bidi_it->type;
1886 }
1887 break;
1888 default:
1889 abort ();
1890 }
1891 type = bidi_resolve_neutral_1 (saved_it.prev_for_neutral.type,
1892 next_type, current_level);
1893 saved_it.type = type;
1894 bidi_check_type (type);
1895 bidi_copy_it (bidi_it, &saved_it);
1896 }
1897 }
1898 return type;
1899 }
1900
1901 /* Given an iterator state in BIDI_IT, advance one character position
1902 in the buffer/string to the next character (in the logical order),
1903 resolve the bidi type of that next character, and return that
1904 type. */
1905 static bidi_type_t
1906 bidi_type_of_next_char (struct bidi_it *bidi_it)
1907 {
1908 bidi_type_t type;
1909
1910 /* This should always be called during a forward scan. */
1911 if (bidi_it->scan_dir != 1)
1912 abort ();
1913
1914 /* Reset the limit until which to ignore BNs if we step out of the
1915 area where we found only empty levels. */
1916 if ((bidi_it->ignore_bn_limit > -1
1917 && bidi_it->ignore_bn_limit <= bidi_it->charpos)
1918 || (bidi_it->ignore_bn_limit == -2
1919 && !bidi_explicit_dir_char (bidi_it->ch)))
1920 bidi_it->ignore_bn_limit = -1;
1921
1922 type = bidi_resolve_neutral (bidi_it);
1923
1924 return type;
1925 }
1926
1927 /* Given an iterator state BIDI_IT, advance one character position in
1928 the buffer/string to the next character (in the current scan
1929 direction), resolve the embedding and implicit levels of that next
1930 character, and return the resulting level. */
1931 static int
1932 bidi_level_of_next_char (struct bidi_it *bidi_it)
1933 {
1934 bidi_type_t type;
1935 int level, prev_level = -1;
1936 struct bidi_saved_info next_for_neutral;
1937 EMACS_INT next_char_pos = -2;
1938
1939 if (bidi_it->scan_dir == 1)
1940 {
1941 EMACS_INT eob =
1942 (bidi_it->string.s || STRINGP (bidi_it->string.lstring))
1943 ? bidi_it->string.schars : ZV;
1944
1945 /* There's no sense in trying to advance if we hit end of text. */
1946 if (bidi_it->charpos >= eob)
1947 return bidi_it->resolved_level;
1948
1949 /* Record the info about the previous character. */
1950 if (bidi_it->type_after_w1 != WEAK_BN /* W1/Retaining */
1951 && bidi_it->type != WEAK_BN)
1952 bidi_remember_char (&bidi_it->prev, bidi_it);
1953 if (bidi_it->type_after_w1 == STRONG_R
1954 || bidi_it->type_after_w1 == STRONG_L
1955 || bidi_it->type_after_w1 == STRONG_AL)
1956 bidi_remember_char (&bidi_it->last_strong, bidi_it);
1957 /* FIXME: it sounds like we don't need both prev and
1958 prev_for_neutral members, but I'm leaving them both for now. */
1959 if (bidi_it->type == STRONG_R || bidi_it->type == STRONG_L
1960 || bidi_it->type == WEAK_EN || bidi_it->type == WEAK_AN)
1961 bidi_remember_char (&bidi_it->prev_for_neutral, bidi_it);
1962
1963 /* If we overstepped the characters used for resolving neutrals
1964 and whitespace, invalidate their info in the iterator. */
1965 if (bidi_it->charpos >= bidi_it->next_for_neutral.charpos)
1966 bidi_it->next_for_neutral.type = UNKNOWN_BT;
1967 if (bidi_it->next_en_pos >= 0
1968 && bidi_it->charpos >= bidi_it->next_en_pos)
1969 bidi_it->next_en_pos = -1;
1970 if (bidi_it->next_for_ws.type != UNKNOWN_BT
1971 && bidi_it->charpos >= bidi_it->next_for_ws.charpos)
1972 bidi_it->next_for_ws.type = UNKNOWN_BT;
1973
1974 /* This must be taken before we fill the iterator with the info
1975 about the next char. If we scan backwards, the iterator
1976 state must be already cached, so there's no need to know the
1977 embedding level of the previous character, since we will be
1978 returning to our caller shortly. */
1979 prev_level = bidi_it->level_stack[bidi_it->stack_idx].level;
1980 }
1981 next_for_neutral = bidi_it->next_for_neutral;
1982
1983 /* Perhaps the character we want is already cached. If it is, the
1984 call to bidi_cache_find below will return a type other than
1985 UNKNOWN_BT. */
1986 if (bidi_cache_idx > bidi_cache_start && !bidi_it->first_elt)
1987 {
1988 int bob =
1989 (bidi_it->string.s || STRINGP (bidi_it->string.lstring)) ? 0 : 1;
1990
1991 if (bidi_it->scan_dir > 0)
1992 {
1993 if (bidi_it->nchars <= 0)
1994 abort ();
1995 next_char_pos = bidi_it->charpos + bidi_it->nchars;
1996 }
1997 else if (bidi_it->charpos >= bob)
1998 /* Implementation note: we allow next_char_pos to be as low as
1999 0 for buffers or -1 for strings, and that is okay because
2000 that's the "position" of the sentinel iterator state we
2001 cached at the beginning of the iteration. */
2002 next_char_pos = bidi_it->charpos - 1;
2003 if (next_char_pos >= bob - 1)
2004 type = bidi_cache_find (next_char_pos, -1, bidi_it);
2005 else
2006 type = UNKNOWN_BT;
2007 }
2008 else
2009 type = UNKNOWN_BT;
2010 if (type != UNKNOWN_BT)
2011 {
2012 /* Don't lose the information for resolving neutrals! The
2013 cached states could have been cached before their
2014 next_for_neutral member was computed. If we are on our way
2015 forward, we can simply take the info from the previous
2016 state. */
2017 if (bidi_it->scan_dir == 1
2018 && bidi_it->next_for_neutral.type == UNKNOWN_BT)
2019 bidi_it->next_for_neutral = next_for_neutral;
2020
2021 /* If resolved_level is -1, it means this state was cached
2022 before it was completely resolved, so we cannot return
2023 it. */
2024 if (bidi_it->resolved_level != -1)
2025 return bidi_it->resolved_level;
2026 }
2027 if (bidi_it->scan_dir == -1)
2028 /* If we are going backwards, the iterator state is already cached
2029 from previous scans, and should be fully resolved. */
2030 abort ();
2031
2032 if (type == UNKNOWN_BT)
2033 type = bidi_type_of_next_char (bidi_it);
2034
2035 if (type == NEUTRAL_B)
2036 return bidi_it->resolved_level;
2037
2038 level = bidi_it->level_stack[bidi_it->stack_idx].level;
2039 if ((bidi_get_category (type) == NEUTRAL /* && type != NEUTRAL_B */)
2040 || (type == WEAK_BN && prev_level == level))
2041 {
2042 if (bidi_it->next_for_neutral.type == UNKNOWN_BT)
2043 abort ();
2044
2045 /* If the cached state shows a neutral character, it was not
2046 resolved by bidi_resolve_neutral, so do it now. */
2047 type = bidi_resolve_neutral_1 (bidi_it->prev_for_neutral.type,
2048 bidi_it->next_for_neutral.type,
2049 level);
2050 }
2051
2052 if (!(type == STRONG_R
2053 || type == STRONG_L
2054 || type == WEAK_BN
2055 || type == WEAK_EN
2056 || type == WEAK_AN))
2057 abort ();
2058 bidi_it->type = type;
2059 bidi_check_type (bidi_it->type);
2060
2061 /* For L1 below, we need to know, for each WS character, whether
2062 it belongs to a sequence of WS characters preceding a newline
2063 or a TAB or a paragraph separator. */
2064 if (bidi_it->orig_type == NEUTRAL_WS
2065 && bidi_it->next_for_ws.type == UNKNOWN_BT)
2066 {
2067 int ch;
2068 EMACS_INT clen = bidi_it->ch_len;
2069 EMACS_INT bpos = bidi_it->bytepos;
2070 EMACS_INT cpos = bidi_it->charpos;
2071 EMACS_INT disp_pos = bidi_it->disp_pos;
2072 EMACS_INT nc = bidi_it->nchars;
2073 struct bidi_string_data bs = bidi_it->string;
2074 bidi_type_t chtype;
2075 int fwp = bidi_it->frame_window_p;
2076 int dpp = bidi_it->disp_prop_p;
2077
2078 if (bidi_it->nchars <= 0)
2079 abort ();
2080 do {
2081 ch = bidi_fetch_char (bpos += clen, cpos += nc, &disp_pos, &dpp, &bs,
2082 fwp, &clen, &nc);
2083 if (ch == '\n' || ch == BIDI_EOB /* || ch == LINESEP_CHAR */)
2084 chtype = NEUTRAL_B;
2085 else
2086 chtype = bidi_get_type (ch, NEUTRAL_DIR);
2087 } while (chtype == NEUTRAL_WS || chtype == WEAK_BN
2088 || bidi_explicit_dir_char (ch)); /* L1/Retaining */
2089 bidi_it->next_for_ws.type = chtype;
2090 bidi_check_type (bidi_it->next_for_ws.type);
2091 bidi_it->next_for_ws.charpos = cpos;
2092 bidi_it->next_for_ws.bytepos = bpos;
2093 }
2094
2095 /* Resolve implicit levels, with a twist: PDFs get the embedding
2096 level of the enbedding they terminate. See below for the
2097 reason. */
2098 if (bidi_it->orig_type == PDF
2099 /* Don't do this if this formatting code didn't change the
2100 embedding level due to invalid or empty embeddings. */
2101 && prev_level != level)
2102 {
2103 /* Don't look in UAX#9 for the reason for this: it's our own
2104 private quirk. The reason is that we want the formatting
2105 codes to be delivered so that they bracket the text of their
2106 embedding. For example, given the text
2107
2108 {RLO}teST{PDF}
2109
2110 we want it to be displayed as
2111
2112 {PDF}STet{RLO}
2113
2114 not as
2115
2116 STet{RLO}{PDF}
2117
2118 which will result because we bump up the embedding level as
2119 soon as we see the RLO and pop it as soon as we see the PDF,
2120 so RLO itself has the same embedding level as "teST", and
2121 thus would be normally delivered last, just before the PDF.
2122 The switch below fiddles with the level of PDF so that this
2123 ugly side effect does not happen.
2124
2125 (This is, of course, only important if the formatting codes
2126 are actually displayed, but Emacs does need to display them
2127 if the user wants to.) */
2128 level = prev_level;
2129 }
2130 else if (bidi_it->orig_type == NEUTRAL_B /* L1 */
2131 || bidi_it->orig_type == NEUTRAL_S
2132 || bidi_it->ch == '\n' || bidi_it->ch == BIDI_EOB
2133 /* || bidi_it->ch == LINESEP_CHAR */
2134 || (bidi_it->orig_type == NEUTRAL_WS
2135 && (bidi_it->next_for_ws.type == NEUTRAL_B
2136 || bidi_it->next_for_ws.type == NEUTRAL_S)))
2137 level = bidi_it->level_stack[0].level;
2138 else if ((level & 1) == 0) /* I1 */
2139 {
2140 if (type == STRONG_R)
2141 level++;
2142 else if (type == WEAK_EN || type == WEAK_AN)
2143 level += 2;
2144 }
2145 else /* I2 */
2146 {
2147 if (type == STRONG_L || type == WEAK_EN || type == WEAK_AN)
2148 level++;
2149 }
2150
2151 bidi_it->resolved_level = level;
2152 return level;
2153 }
2154
2155 /* Move to the other edge of a level given by LEVEL. If END_FLAG is
2156 non-zero, we are at the end of a level, and we need to prepare to
2157 resume the scan of the lower level.
2158
2159 If this level's other edge is cached, we simply jump to it, filling
2160 the iterator structure with the iterator state on the other edge.
2161 Otherwise, we walk the buffer or string until we come back to the
2162 same level as LEVEL.
2163
2164 Note: we are not talking here about a ``level run'' in the UAX#9
2165 sense of the term, but rather about a ``level'' which includes
2166 all the levels higher than it. In other words, given the levels
2167 like this:
2168
2169 11111112222222333333334443343222222111111112223322111
2170 A B C
2171
2172 and assuming we are at point A scanning left to right, this
2173 function moves to point C, whereas the UAX#9 ``level 2 run'' ends
2174 at point B. */
2175 static void
2176 bidi_find_other_level_edge (struct bidi_it *bidi_it, int level, int end_flag)
2177 {
2178 int dir = end_flag ? -bidi_it->scan_dir : bidi_it->scan_dir;
2179 ptrdiff_t idx;
2180
2181 /* Try the cache first. */
2182 if ((idx = bidi_cache_find_level_change (level, dir, end_flag))
2183 >= bidi_cache_start)
2184 bidi_cache_fetch_state (idx, bidi_it);
2185 else
2186 {
2187 int new_level;
2188
2189 if (end_flag)
2190 abort (); /* if we are at end of level, its edges must be cached */
2191
2192 bidi_cache_iterator_state (bidi_it, 1);
2193 do {
2194 new_level = bidi_level_of_next_char (bidi_it);
2195 bidi_cache_iterator_state (bidi_it, 1);
2196 } while (new_level >= level);
2197 }
2198 }
2199
2200 void
2201 bidi_move_to_visually_next (struct bidi_it *bidi_it)
2202 {
2203 int old_level, new_level, next_level;
2204 struct bidi_it sentinel;
2205 struct gcpro gcpro1;
2206
2207 if (bidi_it->charpos < 0 || bidi_it->bytepos < 0)
2208 abort ();
2209
2210 if (bidi_it->scan_dir == 0)
2211 {
2212 bidi_it->scan_dir = 1; /* default to logical order */
2213 }
2214
2215 /* The code below can call eval, and thus cause GC. If we are
2216 iterating a Lisp string, make sure it won't be GCed. */
2217 if (STRINGP (bidi_it->string.lstring))
2218 GCPRO1 (bidi_it->string.lstring);
2219
2220 /* If we just passed a newline, initialize for the next line. */
2221 if (!bidi_it->first_elt && bidi_it->orig_type == NEUTRAL_B)
2222 bidi_line_init (bidi_it);
2223
2224 /* Prepare the sentinel iterator state, and cache it. When we bump
2225 into it, scanning backwards, we'll know that the last non-base
2226 level is exhausted. */
2227 if (bidi_cache_idx == bidi_cache_start)
2228 {
2229 bidi_copy_it (&sentinel, bidi_it);
2230 if (bidi_it->first_elt)
2231 {
2232 sentinel.charpos--; /* cached charpos needs to be monotonic */
2233 sentinel.bytepos--;
2234 sentinel.ch = '\n'; /* doesn't matter, but why not? */
2235 sentinel.ch_len = 1;
2236 sentinel.nchars = 1;
2237 }
2238 bidi_cache_iterator_state (&sentinel, 1);
2239 }
2240
2241 old_level = bidi_it->resolved_level;
2242 new_level = bidi_level_of_next_char (bidi_it);
2243
2244 /* Reordering of resolved levels (clause L2) is implemented by
2245 jumping to the other edge of the level and flipping direction of
2246 scanning the text whenever we find a level change. */
2247 if (new_level != old_level)
2248 {
2249 int ascending = new_level > old_level;
2250 int level_to_search = ascending ? old_level + 1 : old_level;
2251 int incr = ascending ? 1 : -1;
2252 int expected_next_level = old_level + incr;
2253
2254 /* Jump (or walk) to the other edge of this level. */
2255 bidi_find_other_level_edge (bidi_it, level_to_search, !ascending);
2256 /* Switch scan direction and peek at the next character in the
2257 new direction. */
2258 bidi_it->scan_dir = -bidi_it->scan_dir;
2259
2260 /* The following loop handles the case where the resolved level
2261 jumps by more than one. This is typical for numbers inside a
2262 run of text with left-to-right embedding direction, but can
2263 also happen in other situations. In those cases the decision
2264 where to continue after a level change, and in what direction,
2265 is tricky. For example, given a text like below:
2266
2267 abcdefgh
2268 11336622
2269
2270 (where the numbers below the text show the resolved levels),
2271 the result of reordering according to UAX#9 should be this:
2272
2273 efdcghba
2274
2275 This is implemented by the loop below which flips direction
2276 and jumps to the other edge of the level each time it finds
2277 the new level not to be the expected one. The expected level
2278 is always one more or one less than the previous one. */
2279 next_level = bidi_peek_at_next_level (bidi_it);
2280 while (next_level != expected_next_level)
2281 {
2282 expected_next_level += incr;
2283 level_to_search += incr;
2284 bidi_find_other_level_edge (bidi_it, level_to_search, !ascending);
2285 bidi_it->scan_dir = -bidi_it->scan_dir;
2286 next_level = bidi_peek_at_next_level (bidi_it);
2287 }
2288
2289 /* Finally, deliver the next character in the new direction. */
2290 next_level = bidi_level_of_next_char (bidi_it);
2291 }
2292
2293 /* Take note when we have just processed the newline that precedes
2294 the end of the paragraph. The next time we are about to be
2295 called, set_iterator_to_next will automatically reinit the
2296 paragraph direction, if needed. We do this at the newline before
2297 the paragraph separator, because the next character might not be
2298 the first character of the next paragraph, due to the bidi
2299 reordering, whereas we _must_ know the paragraph base direction
2300 _before_ we process the paragraph's text, since the base
2301 direction affects the reordering. */
2302 if (bidi_it->scan_dir == 1 && bidi_it->orig_type == NEUTRAL_B)
2303 {
2304 /* The paragraph direction of the entire string, once
2305 determined, is in effect for the entire string. Setting the
2306 separator limit to the end of the string prevents
2307 bidi_paragraph_init from being called automatically on this
2308 string. */
2309 if (bidi_it->string.s || STRINGP (bidi_it->string.lstring))
2310 bidi_it->separator_limit = bidi_it->string.schars;
2311 else if (bidi_it->bytepos < ZV_BYTE)
2312 {
2313 EMACS_INT sep_len =
2314 bidi_at_paragraph_end (bidi_it->charpos + bidi_it->nchars,
2315 bidi_it->bytepos + bidi_it->ch_len);
2316 if (bidi_it->nchars <= 0)
2317 abort ();
2318 if (sep_len >= 0)
2319 {
2320 bidi_it->new_paragraph = 1;
2321 /* Record the buffer position of the last character of the
2322 paragraph separator. */
2323 bidi_it->separator_limit =
2324 bidi_it->charpos + bidi_it->nchars + sep_len;
2325 }
2326 }
2327 }
2328
2329 if (bidi_it->scan_dir == 1 && bidi_cache_idx > bidi_cache_start)
2330 {
2331 /* If we are at paragraph's base embedding level and beyond the
2332 last cached position, the cache's job is done and we can
2333 discard it. */
2334 if (bidi_it->resolved_level == bidi_it->level_stack[0].level
2335 && bidi_it->charpos > (bidi_cache[bidi_cache_idx - 1].charpos
2336 + bidi_cache[bidi_cache_idx - 1].nchars - 1))
2337 bidi_cache_reset ();
2338 /* But as long as we are caching during forward scan, we must
2339 cache each state, or else the cache integrity will be
2340 compromised: it assumes cached states correspond to buffer
2341 positions 1:1. */
2342 else
2343 bidi_cache_iterator_state (bidi_it, 1);
2344 }
2345
2346 if (STRINGP (bidi_it->string.lstring))
2347 UNGCPRO;
2348 }
2349
2350 /* This is meant to be called from within the debugger, whenever you
2351 wish to examine the cache contents. */
2352 void bidi_dump_cached_states (void) EXTERNALLY_VISIBLE;
2353 void
2354 bidi_dump_cached_states (void)
2355 {
2356 ptrdiff_t i;
2357 int ndigits = 1;
2358
2359 if (bidi_cache_idx == 0)
2360 {
2361 fprintf (stderr, "The cache is empty.\n");
2362 return;
2363 }
2364 fprintf (stderr, "Total of %"pD"d state%s in cache:\n",
2365 bidi_cache_idx, bidi_cache_idx == 1 ? "" : "s");
2366
2367 for (i = bidi_cache[bidi_cache_idx - 1].charpos; i > 0; i /= 10)
2368 ndigits++;
2369 fputs ("ch ", stderr);
2370 for (i = 0; i < bidi_cache_idx; i++)
2371 fprintf (stderr, "%*c", ndigits, bidi_cache[i].ch);
2372 fputs ("\n", stderr);
2373 fputs ("lvl ", stderr);
2374 for (i = 0; i < bidi_cache_idx; i++)
2375 fprintf (stderr, "%*d", ndigits, bidi_cache[i].resolved_level);
2376 fputs ("\n", stderr);
2377 fputs ("pos ", stderr);
2378 for (i = 0; i < bidi_cache_idx; i++)
2379 fprintf (stderr, "%*"pI"d", ndigits, bidi_cache[i].charpos);
2380 fputs ("\n", stderr);
2381 }