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 /* Every valid character code, even those that are unassigned by the
112 UCD, have some bidi-class property, according to
113 DerivedBidiClass.txt file. Therefore, if we ever get UNKNOWN_BT
114 (= zero) code from CHAR_TABLE_REF, that's a bug. */
115 if (default_type == UNKNOWN_BT)
116 abort ();
117
118 if (override == NEUTRAL_DIR)
119 return default_type;
120
121 switch (default_type)
122 {
123 /* Although UAX#9 does not tell, it doesn't make sense to
124 override NEUTRAL_B and LRM/RLM characters. */
125 case NEUTRAL_B:
126 case LRE:
127 case LRO:
128 case RLE:
129 case RLO:
130 case PDF:
131 return default_type;
132 default:
133 switch (ch)
134 {
135 case LRM_CHAR:
136 case RLM_CHAR:
137 return default_type;
138 default:
139 if (override == L2R) /* X6 */
140 return STRONG_L;
141 else if (override == R2L)
142 return STRONG_R;
143 else
144 abort (); /* can't happen: handled above */
145 }
146 }
147 }
148
149 static inline void
150 bidi_check_type (bidi_type_t type)
151 {
152 xassert (UNKNOWN_BT <= type && type <= NEUTRAL_ON);
153 }
154
155 /* Given a bidi TYPE of a character, return its category. */
156 static inline bidi_category_t
157 bidi_get_category (bidi_type_t type)
158 {
159 switch (type)
160 {
161 case UNKNOWN_BT:
162 return UNKNOWN_BC;
163 case STRONG_L:
164 case STRONG_R:
165 case STRONG_AL:
166 case LRE:
167 case LRO:
168 case RLE:
169 case RLO:
170 return STRONG;
171 case PDF: /* ??? really?? */
172 case WEAK_EN:
173 case WEAK_ES:
174 case WEAK_ET:
175 case WEAK_AN:
176 case WEAK_CS:
177 case WEAK_NSM:
178 case WEAK_BN:
179 return WEAK;
180 case NEUTRAL_B:
181 case NEUTRAL_S:
182 case NEUTRAL_WS:
183 case NEUTRAL_ON:
184 return NEUTRAL;
185 default:
186 abort ();
187 }
188 }
189
190 /* Return the mirrored character of C, if it has one. If C has no
191 mirrored counterpart, return C.
192 Note: The conditions in UAX#9 clause L4 regarding the surrounding
193 context must be tested by the caller. */
194 int
195 bidi_mirror_char (int c)
196 {
197 Lisp_Object val;
198
199 if (c == BIDI_EOB)
200 return c;
201 if (c < 0 || c > MAX_CHAR)
202 abort ();
203
204 val = CHAR_TABLE_REF (bidi_mirror_table, c);
205 if (INTEGERP (val))
206 {
207 int v = XINT (val);
208
209 if (v < 0 || v > MAX_CHAR)
210 abort ();
211
212 return v;
213 }
214
215 return c;
216 }
217
218 /* Determine the start-of-run (sor) directional type given the two
219 embedding levels on either side of the run boundary. Also, update
220 the saved info about previously seen characters, since that info is
221 generally valid for a single level run. */
222 static inline void
223 bidi_set_sor_type (struct bidi_it *bidi_it, int level_before, int level_after)
224 {
225 int higher_level = level_before > level_after ? level_before : level_after;
226
227 /* The prev_was_pdf gork is required for when we have several PDFs
228 in a row. In that case, we want to compute the sor type for the
229 next level run only once: when we see the first PDF. That's
230 because the sor type depends only on the higher of the two levels
231 that we find on the two sides of the level boundary (see UAX#9,
232 clause X10), and so we don't need to know the final embedding
233 level to which we descend after processing all the PDFs. */
234 if (!bidi_it->prev_was_pdf || level_before < level_after)
235 /* FIXME: should the default sor direction be user selectable? */
236 bidi_it->sor = (higher_level & 1) != 0 ? R2L : L2R;
237 if (level_before > level_after)
238 bidi_it->prev_was_pdf = 1;
239
240 bidi_it->prev.type = UNKNOWN_BT;
241 bidi_it->last_strong.type = bidi_it->last_strong.type_after_w1 =
242 bidi_it->last_strong.orig_type = UNKNOWN_BT;
243 bidi_it->prev_for_neutral.type = bidi_it->sor == R2L ? STRONG_R : STRONG_L;
244 bidi_it->prev_for_neutral.charpos = bidi_it->charpos;
245 bidi_it->prev_for_neutral.bytepos = bidi_it->bytepos;
246 bidi_it->next_for_neutral.type = bidi_it->next_for_neutral.type_after_w1 =
247 bidi_it->next_for_neutral.orig_type = UNKNOWN_BT;
248 bidi_it->ignore_bn_limit = -1; /* meaning it's unknown */
249 }
250
251 /* Push the current embedding level and override status; reset the
252 current level to LEVEL and the current override status to OVERRIDE. */
253 static inline void
254 bidi_push_embedding_level (struct bidi_it *bidi_it,
255 int level, bidi_dir_t override)
256 {
257 bidi_it->stack_idx++;
258 xassert (bidi_it->stack_idx < BIDI_MAXLEVEL);
259 bidi_it->level_stack[bidi_it->stack_idx].level = level;
260 bidi_it->level_stack[bidi_it->stack_idx].override = override;
261 }
262
263 /* Pop the embedding level and directional override status from the
264 stack, and return the new level. */
265 static inline int
266 bidi_pop_embedding_level (struct bidi_it *bidi_it)
267 {
268 /* UAX#9 says to ignore invalid PDFs. */
269 if (bidi_it->stack_idx > 0)
270 bidi_it->stack_idx--;
271 return bidi_it->level_stack[bidi_it->stack_idx].level;
272 }
273
274 /* Record in SAVED_INFO the information about the current character. */
275 static inline void
276 bidi_remember_char (struct bidi_saved_info *saved_info,
277 struct bidi_it *bidi_it)
278 {
279 saved_info->charpos = bidi_it->charpos;
280 saved_info->bytepos = bidi_it->bytepos;
281 saved_info->type = bidi_it->type;
282 bidi_check_type (bidi_it->type);
283 saved_info->type_after_w1 = bidi_it->type_after_w1;
284 bidi_check_type (bidi_it->type_after_w1);
285 saved_info->orig_type = bidi_it->orig_type;
286 bidi_check_type (bidi_it->orig_type);
287 }
288
289 /* Copy the bidi iterator from FROM to TO. To save cycles, this only
290 copies the part of the level stack that is actually in use. */
291 static inline void
292 bidi_copy_it (struct bidi_it *to, struct bidi_it *from)
293 {
294 int i;
295
296 /* Copy everything except the level stack and beyond. */
297 memcpy (to, from, offsetof (struct bidi_it, level_stack[0]));
298
299 /* Copy the active part of the level stack. */
300 to->level_stack[0] = from->level_stack[0]; /* level zero is always in use */
301 for (i = 1; i <= from->stack_idx; i++)
302 to->level_stack[i] = from->level_stack[i];
303 }
304
305 \f
306 /***********************************************************************
307 Caching the bidi iterator states
308 ***********************************************************************/
309
310 #define BIDI_CACHE_CHUNK 200
311 static struct bidi_it *bidi_cache;
312 static ptrdiff_t bidi_cache_size = 0;
313 enum { elsz = sizeof (struct bidi_it) };
314 static ptrdiff_t bidi_cache_idx; /* next unused cache slot */
315 static ptrdiff_t bidi_cache_last_idx; /* slot of last cache hit */
316 static ptrdiff_t bidi_cache_start = 0; /* start of cache for this
317 "stack" level */
318
319 /* 5-slot stack for saving the start of the previous level of the
320 cache. xdisp.c maintains a 5-slot stack for its iterator state,
321 and we need the same size of our stack. */
322 static ptrdiff_t bidi_cache_start_stack[IT_STACK_SIZE];
323 static int bidi_cache_sp;
324
325 /* Size of header used by bidi_shelve_cache. */
326 enum
327 {
328 bidi_shelve_header_size =
329 (sizeof (bidi_cache_idx) + sizeof (bidi_cache_start_stack)
330 + sizeof (bidi_cache_sp) + sizeof (bidi_cache_start)
331 + sizeof (bidi_cache_last_idx))
332 };
333
334 /* Reset the cache state to the empty state. We only reset the part
335 of the cache relevant to iteration of the current object. Previous
336 objects, which are pushed on the display iterator's stack, are left
337 intact. This is called when the cached information is no more
338 useful for the current iteration, e.g. when we were reseated to a
339 new position on the same object. */
340 static inline void
341 bidi_cache_reset (void)
342 {
343 bidi_cache_idx = bidi_cache_start;
344 bidi_cache_last_idx = -1;
345 }
346
347 /* Shrink the cache to its minimal size. Called when we init the bidi
348 iterator for reordering a buffer or a string that does not come
349 from display properties, because that means all the previously
350 cached info is of no further use. */
351 static inline void
352 bidi_cache_shrink (void)
353 {
354 if (bidi_cache_size > BIDI_CACHE_CHUNK)
355 {
356 bidi_cache =
357 (struct bidi_it *) xrealloc (bidi_cache, BIDI_CACHE_CHUNK * elsz);
358 bidi_cache_size = BIDI_CACHE_CHUNK;
359 }
360 bidi_cache_reset ();
361 }
362
363 static inline void
364 bidi_cache_fetch_state (ptrdiff_t idx, struct bidi_it *bidi_it)
365 {
366 int current_scan_dir = bidi_it->scan_dir;
367
368 if (idx < bidi_cache_start || idx >= bidi_cache_idx)
369 abort ();
370
371 bidi_copy_it (bidi_it, &bidi_cache[idx]);
372 bidi_it->scan_dir = current_scan_dir;
373 bidi_cache_last_idx = idx;
374 }
375
376 /* Find a cached state with a given CHARPOS and resolved embedding
377 level less or equal to LEVEL. if LEVEL is -1, disregard the
378 resolved levels in cached states. DIR, if non-zero, means search
379 in that direction from the last cache hit. */
380 static inline ptrdiff_t
381 bidi_cache_search (EMACS_INT charpos, int level, int dir)
382 {
383 ptrdiff_t i, i_start;
384
385 if (bidi_cache_idx > bidi_cache_start)
386 {
387 if (bidi_cache_last_idx == -1)
388 bidi_cache_last_idx = bidi_cache_idx - 1;
389 if (charpos < bidi_cache[bidi_cache_last_idx].charpos)
390 {
391 dir = -1;
392 i_start = bidi_cache_last_idx - 1;
393 }
394 else if (charpos > (bidi_cache[bidi_cache_last_idx].charpos
395 + bidi_cache[bidi_cache_last_idx].nchars - 1))
396 {
397 dir = 1;
398 i_start = bidi_cache_last_idx + 1;
399 }
400 else if (dir)
401 i_start = bidi_cache_last_idx;
402 else
403 {
404 dir = -1;
405 i_start = bidi_cache_idx - 1;
406 }
407
408 if (dir < 0)
409 {
410 /* Linear search for now; FIXME! */
411 for (i = i_start; i >= bidi_cache_start; i--)
412 if (bidi_cache[i].charpos <= charpos
413 && charpos < bidi_cache[i].charpos + bidi_cache[i].nchars
414 && (level == -1 || bidi_cache[i].resolved_level <= level))
415 return i;
416 }
417 else
418 {
419 for (i = i_start; i < bidi_cache_idx; i++)
420 if (bidi_cache[i].charpos <= charpos
421 && charpos < bidi_cache[i].charpos + bidi_cache[i].nchars
422 && (level == -1 || bidi_cache[i].resolved_level <= level))
423 return i;
424 }
425 }
426
427 return -1;
428 }
429
430 /* Find a cached state where the resolved level changes to a value
431 that is lower than LEVEL, and return its cache slot index. DIR is
432 the direction to search, starting with the last used cache slot.
433 If DIR is zero, we search backwards from the last occupied cache
434 slot. BEFORE, if non-zero, means return the index of the slot that
435 is ``before'' the level change in the search direction. That is,
436 given the cached levels like this:
437
438 1122333442211
439 AB C
440
441 and assuming we are at the position cached at the slot marked with
442 C, searching backwards (DIR = -1) for LEVEL = 2 will return the
443 index of slot B or A, depending whether BEFORE is, respectively,
444 non-zero or zero. */
445 static ptrdiff_t
446 bidi_cache_find_level_change (int level, int dir, int before)
447 {
448 if (bidi_cache_idx)
449 {
450 ptrdiff_t i = dir ? bidi_cache_last_idx : bidi_cache_idx - 1;
451 int incr = before ? 1 : 0;
452
453 xassert (!dir || bidi_cache_last_idx >= 0);
454
455 if (!dir)
456 dir = -1;
457 else if (!incr)
458 i += dir;
459
460 if (dir < 0)
461 {
462 while (i >= bidi_cache_start + incr)
463 {
464 if (bidi_cache[i - incr].resolved_level >= 0
465 && bidi_cache[i - incr].resolved_level < level)
466 return i;
467 i--;
468 }
469 }
470 else
471 {
472 while (i < bidi_cache_idx - incr)
473 {
474 if (bidi_cache[i + incr].resolved_level >= 0
475 && bidi_cache[i + incr].resolved_level < level)
476 return i;
477 i++;
478 }
479 }
480 }
481
482 return -1;
483 }
484
485 static inline void
486 bidi_cache_ensure_space (ptrdiff_t idx)
487 {
488 /* Enlarge the cache as needed. */
489 if (idx >= bidi_cache_size)
490 {
491 /* The bidi cache cannot be larger than the largest Lisp string
492 or buffer. */
493 ptrdiff_t string_or_buffer_bound =
494 max (BUF_BYTES_MAX, STRING_BYTES_BOUND);
495
496 /* Also, it cannot be larger than what C can represent. */
497 ptrdiff_t c_bound =
498 (min (PTRDIFF_MAX, SIZE_MAX) - bidi_shelve_header_size) / elsz;
499
500 bidi_cache =
501 xpalloc (bidi_cache, &bidi_cache_size,
502 max (BIDI_CACHE_CHUNK, idx - bidi_cache_size + 1),
503 min (string_or_buffer_bound, c_bound), elsz);
504 }
505 }
506
507 static inline void
508 bidi_cache_iterator_state (struct bidi_it *bidi_it, int resolved)
509 {
510 ptrdiff_t idx;
511
512 /* We should never cache on backward scans. */
513 if (bidi_it->scan_dir == -1)
514 abort ();
515 idx = bidi_cache_search (bidi_it->charpos, -1, 1);
516
517 if (idx < 0)
518 {
519 idx = bidi_cache_idx;
520 bidi_cache_ensure_space (idx);
521 /* Character positions should correspond to cache positions 1:1.
522 If we are outside the range of cached positions, the cache is
523 useless and must be reset. */
524 if (idx > bidi_cache_start &&
525 (bidi_it->charpos > (bidi_cache[idx - 1].charpos
526 + bidi_cache[idx - 1].nchars)
527 || bidi_it->charpos < bidi_cache[bidi_cache_start].charpos))
528 {
529 bidi_cache_reset ();
530 idx = bidi_cache_start;
531 }
532 if (bidi_it->nchars <= 0)
533 abort ();
534 bidi_copy_it (&bidi_cache[idx], bidi_it);
535 if (!resolved)
536 bidi_cache[idx].resolved_level = -1;
537 }
538 else
539 {
540 /* Copy only the members which could have changed, to avoid
541 costly copying of the entire struct. */
542 bidi_cache[idx].type = bidi_it->type;
543 bidi_check_type (bidi_it->type);
544 bidi_cache[idx].type_after_w1 = bidi_it->type_after_w1;
545 bidi_check_type (bidi_it->type_after_w1);
546 if (resolved)
547 bidi_cache[idx].resolved_level = bidi_it->resolved_level;
548 else
549 bidi_cache[idx].resolved_level = -1;
550 bidi_cache[idx].invalid_levels = bidi_it->invalid_levels;
551 bidi_cache[idx].invalid_rl_levels = bidi_it->invalid_rl_levels;
552 bidi_cache[idx].next_for_neutral = bidi_it->next_for_neutral;
553 bidi_cache[idx].next_for_ws = bidi_it->next_for_ws;
554 bidi_cache[idx].ignore_bn_limit = bidi_it->ignore_bn_limit;
555 bidi_cache[idx].disp_pos = bidi_it->disp_pos;
556 bidi_cache[idx].disp_prop_p = bidi_it->disp_prop_p;
557 }
558
559 bidi_cache_last_idx = idx;
560 if (idx >= bidi_cache_idx)
561 bidi_cache_idx = idx + 1;
562 }
563
564 static inline bidi_type_t
565 bidi_cache_find (EMACS_INT charpos, int level, struct bidi_it *bidi_it)
566 {
567 ptrdiff_t i = bidi_cache_search (charpos, level, bidi_it->scan_dir);
568
569 if (i >= bidi_cache_start)
570 {
571 bidi_dir_t current_scan_dir = bidi_it->scan_dir;
572
573 bidi_copy_it (bidi_it, &bidi_cache[i]);
574 bidi_cache_last_idx = i;
575 /* Don't let scan direction from from the cached state override
576 the current scan direction. */
577 bidi_it->scan_dir = current_scan_dir;
578 return bidi_it->type;
579 }
580
581 return UNKNOWN_BT;
582 }
583
584 static inline int
585 bidi_peek_at_next_level (struct bidi_it *bidi_it)
586 {
587 if (bidi_cache_idx == bidi_cache_start || bidi_cache_last_idx == -1)
588 abort ();
589 return bidi_cache[bidi_cache_last_idx + bidi_it->scan_dir].resolved_level;
590 }
591
592 \f
593 /***********************************************************************
594 Pushing and popping the bidi iterator state
595 ***********************************************************************/
596
597 /* Push the bidi iterator state in preparation for reordering a
598 different object, e.g. display string found at certain buffer
599 position. Pushing the bidi iterator boils down to saving its
600 entire state on the cache and starting a new cache "stacked" on top
601 of the current cache. */
602 void
603 bidi_push_it (struct bidi_it *bidi_it)
604 {
605 /* Save the current iterator state in its entirety after the last
606 used cache slot. */
607 bidi_cache_ensure_space (bidi_cache_idx);
608 memcpy (&bidi_cache[bidi_cache_idx++], bidi_it, sizeof (struct bidi_it));
609
610 /* Push the current cache start onto the stack. */
611 xassert (bidi_cache_sp < IT_STACK_SIZE);
612 bidi_cache_start_stack[bidi_cache_sp++] = bidi_cache_start;
613
614 /* Start a new level of cache, and make it empty. */
615 bidi_cache_start = bidi_cache_idx;
616 bidi_cache_last_idx = -1;
617 }
618
619 /* Restore the iterator state saved by bidi_push_it and return the
620 cache to the corresponding state. */
621 void
622 bidi_pop_it (struct bidi_it *bidi_it)
623 {
624 if (bidi_cache_start <= 0)
625 abort ();
626
627 /* Reset the next free cache slot index to what it was before the
628 call to bidi_push_it. */
629 bidi_cache_idx = bidi_cache_start - 1;
630
631 /* Restore the bidi iterator state saved in the cache. */
632 memcpy (bidi_it, &bidi_cache[bidi_cache_idx], sizeof (struct bidi_it));
633
634 /* Pop the previous cache start from the stack. */
635 if (bidi_cache_sp <= 0)
636 abort ();
637 bidi_cache_start = bidi_cache_start_stack[--bidi_cache_sp];
638
639 /* Invalidate the last-used cache slot data. */
640 bidi_cache_last_idx = -1;
641 }
642
643 static ptrdiff_t bidi_cache_total_alloc;
644
645 /* Stash away a copy of the cache and its control variables. */
646 void *
647 bidi_shelve_cache (void)
648 {
649 unsigned char *databuf;
650 ptrdiff_t alloc;
651
652 /* Empty cache. */
653 if (bidi_cache_idx == 0)
654 return NULL;
655
656 alloc = (bidi_shelve_header_size
657 + bidi_cache_idx * sizeof (struct bidi_it));
658 databuf = xmalloc (alloc);
659 bidi_cache_total_alloc += alloc;
660
661 memcpy (databuf, &bidi_cache_idx, sizeof (bidi_cache_idx));
662 memcpy (databuf + sizeof (bidi_cache_idx),
663 bidi_cache, bidi_cache_idx * sizeof (struct bidi_it));
664 memcpy (databuf + sizeof (bidi_cache_idx)
665 + bidi_cache_idx * sizeof (struct bidi_it),
666 bidi_cache_start_stack, sizeof (bidi_cache_start_stack));
667 memcpy (databuf + sizeof (bidi_cache_idx)
668 + bidi_cache_idx * sizeof (struct bidi_it)
669 + sizeof (bidi_cache_start_stack),
670 &bidi_cache_sp, sizeof (bidi_cache_sp));
671 memcpy (databuf + sizeof (bidi_cache_idx)
672 + bidi_cache_idx * sizeof (struct bidi_it)
673 + sizeof (bidi_cache_start_stack) + sizeof (bidi_cache_sp),
674 &bidi_cache_start, sizeof (bidi_cache_start));
675 memcpy (databuf + sizeof (bidi_cache_idx)
676 + bidi_cache_idx * sizeof (struct bidi_it)
677 + sizeof (bidi_cache_start_stack) + sizeof (bidi_cache_sp)
678 + sizeof (bidi_cache_start),
679 &bidi_cache_last_idx, sizeof (bidi_cache_last_idx));
680
681 return databuf;
682 }
683
684 /* Restore the cache state from a copy stashed away by
685 bidi_shelve_cache, and free the buffer used to stash that copy.
686 JUST_FREE non-zero means free the buffer, but don't restore the
687 cache; used when the corresponding iterator is discarded instead of
688 being restored. */
689 void
690 bidi_unshelve_cache (void *databuf, int just_free)
691 {
692 unsigned char *p = databuf;
693
694 if (!p)
695 {
696 if (!just_free)
697 {
698 /* A NULL pointer means an empty cache. */
699 bidi_cache_start = 0;
700 bidi_cache_sp = 0;
701 bidi_cache_reset ();
702 }
703 }
704 else
705 {
706 if (just_free)
707 {
708 ptrdiff_t idx;
709
710 memcpy (&idx, p, sizeof (bidi_cache_idx));
711 bidi_cache_total_alloc -=
712 bidi_shelve_header_size + idx * sizeof (struct bidi_it);
713 }
714 else
715 {
716 memcpy (&bidi_cache_idx, p, sizeof (bidi_cache_idx));
717 bidi_cache_ensure_space (bidi_cache_idx);
718 memcpy (bidi_cache, p + sizeof (bidi_cache_idx),
719 bidi_cache_idx * sizeof (struct bidi_it));
720 memcpy (bidi_cache_start_stack,
721 p + sizeof (bidi_cache_idx)
722 + bidi_cache_idx * sizeof (struct bidi_it),
723 sizeof (bidi_cache_start_stack));
724 memcpy (&bidi_cache_sp,
725 p + sizeof (bidi_cache_idx)
726 + bidi_cache_idx * sizeof (struct bidi_it)
727 + sizeof (bidi_cache_start_stack),
728 sizeof (bidi_cache_sp));
729 memcpy (&bidi_cache_start,
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 memcpy (&bidi_cache_last_idx,
735 p + sizeof (bidi_cache_idx)
736 + bidi_cache_idx * sizeof (struct bidi_it)
737 + sizeof (bidi_cache_start_stack) + sizeof (bidi_cache_sp)
738 + sizeof (bidi_cache_start),
739 sizeof (bidi_cache_last_idx));
740 bidi_cache_total_alloc -=
741 bidi_shelve_header_size + bidi_cache_idx * sizeof (struct bidi_it);
742 }
743
744 xfree (p);
745 }
746 }
747
748 \f
749 /***********************************************************************
750 Initialization
751 ***********************************************************************/
752 static void
753 bidi_initialize (void)
754 {
755 bidi_type_table = uniprop_table (intern ("bidi-class"));
756 if (NILP (bidi_type_table))
757 abort ();
758 staticpro (&bidi_type_table);
759
760 bidi_mirror_table = uniprop_table (intern ("mirroring"));
761 if (NILP (bidi_mirror_table))
762 abort ();
763 staticpro (&bidi_mirror_table);
764
765 Qparagraph_start = intern ("paragraph-start");
766 staticpro (&Qparagraph_start);
767 paragraph_start_re = Fsymbol_value (Qparagraph_start);
768 if (!STRINGP (paragraph_start_re))
769 paragraph_start_re = build_string ("\f\\|[ \t]*$");
770 staticpro (&paragraph_start_re);
771 Qparagraph_separate = intern ("paragraph-separate");
772 staticpro (&Qparagraph_separate);
773 paragraph_separate_re = Fsymbol_value (Qparagraph_separate);
774 if (!STRINGP (paragraph_separate_re))
775 paragraph_separate_re = build_string ("[ \t\f]*$");
776 staticpro (&paragraph_separate_re);
777
778 bidi_cache_sp = 0;
779 bidi_cache_total_alloc = 0;
780
781 bidi_initialized = 1;
782 }
783
784 /* Do whatever UAX#9 clause X8 says should be done at paragraph's
785 end. */
786 static inline void
787 bidi_set_paragraph_end (struct bidi_it *bidi_it)
788 {
789 bidi_it->invalid_levels = 0;
790 bidi_it->invalid_rl_levels = -1;
791 bidi_it->stack_idx = 0;
792 bidi_it->resolved_level = bidi_it->level_stack[0].level;
793 }
794
795 /* Initialize the bidi iterator from buffer/string position CHARPOS. */
796 void
797 bidi_init_it (EMACS_INT charpos, EMACS_INT bytepos, int frame_window_p,
798 struct bidi_it *bidi_it)
799 {
800 if (! bidi_initialized)
801 bidi_initialize ();
802 if (charpos >= 0)
803 bidi_it->charpos = charpos;
804 if (bytepos >= 0)
805 bidi_it->bytepos = bytepos;
806 bidi_it->frame_window_p = frame_window_p;
807 bidi_it->nchars = -1; /* to be computed in bidi_resolve_explicit_1 */
808 bidi_it->first_elt = 1;
809 bidi_set_paragraph_end (bidi_it);
810 bidi_it->new_paragraph = 1;
811 bidi_it->separator_limit = -1;
812 bidi_it->type = NEUTRAL_B;
813 bidi_it->type_after_w1 = NEUTRAL_B;
814 bidi_it->orig_type = NEUTRAL_B;
815 bidi_it->prev_was_pdf = 0;
816 bidi_it->prev.type = bidi_it->prev.type_after_w1 =
817 bidi_it->prev.orig_type = UNKNOWN_BT;
818 bidi_it->last_strong.type = bidi_it->last_strong.type_after_w1 =
819 bidi_it->last_strong.orig_type = UNKNOWN_BT;
820 bidi_it->next_for_neutral.charpos = -1;
821 bidi_it->next_for_neutral.type =
822 bidi_it->next_for_neutral.type_after_w1 =
823 bidi_it->next_for_neutral.orig_type = UNKNOWN_BT;
824 bidi_it->prev_for_neutral.charpos = -1;
825 bidi_it->prev_for_neutral.type =
826 bidi_it->prev_for_neutral.type_after_w1 =
827 bidi_it->prev_for_neutral.orig_type = UNKNOWN_BT;
828 bidi_it->sor = L2R; /* FIXME: should it be user-selectable? */
829 bidi_it->disp_pos = -1; /* invalid/unknown */
830 bidi_it->disp_prop_p = 0;
831 /* We can only shrink the cache if we are at the bottom level of its
832 "stack". */
833 if (bidi_cache_start == 0)
834 bidi_cache_shrink ();
835 else
836 bidi_cache_reset ();
837 }
838
839 /* Perform initializations for reordering a new line of bidi text. */
840 static void
841 bidi_line_init (struct bidi_it *bidi_it)
842 {
843 bidi_it->scan_dir = 1; /* FIXME: do we need to have control on this? */
844 bidi_it->resolved_level = bidi_it->level_stack[0].level;
845 bidi_it->level_stack[0].override = NEUTRAL_DIR; /* X1 */
846 bidi_it->invalid_levels = 0;
847 bidi_it->invalid_rl_levels = -1;
848 bidi_it->next_en_pos = -1;
849 bidi_it->next_for_ws.type = UNKNOWN_BT;
850 bidi_set_sor_type (bidi_it,
851 bidi_it->paragraph_dir == R2L ? 1 : 0,
852 bidi_it->level_stack[0].level); /* X10 */
853
854 bidi_cache_reset ();
855 }
856
857 \f
858 /***********************************************************************
859 Fetching characters
860 ***********************************************************************/
861
862 /* Count bytes in string S between BEG/BEGBYTE and END. BEG and END
863 are zero-based character positions in S, BEGBYTE is byte position
864 corresponding to BEG. UNIBYTE, if non-zero, means S is a unibyte
865 string. */
866 static inline EMACS_INT
867 bidi_count_bytes (const unsigned char *s, const EMACS_INT beg,
868 const EMACS_INT begbyte, const EMACS_INT end, int unibyte)
869 {
870 EMACS_INT pos = beg;
871 const unsigned char *p = s + begbyte, *start = p;
872
873 if (unibyte)
874 p = s + end;
875 else
876 {
877 if (!CHAR_HEAD_P (*p))
878 abort ();
879
880 while (pos < end)
881 {
882 p += BYTES_BY_CHAR_HEAD (*p);
883 pos++;
884 }
885 }
886
887 return p - start;
888 }
889
890 /* Fetch and returns the character at byte position BYTEPOS. If S is
891 non-NULL, fetch the character from string S; otherwise fetch the
892 character from the current buffer. UNIBYTE non-zero means S is a
893 unibyte string. */
894 static inline int
895 bidi_char_at_pos (EMACS_INT bytepos, const unsigned char *s, int unibyte)
896 {
897 if (s)
898 {
899 if (unibyte)
900 return s[bytepos];
901 else
902 return STRING_CHAR (s + bytepos);
903 }
904 else
905 return FETCH_MULTIBYTE_CHAR (bytepos);
906 }
907
908 /* Fetch and return the character at BYTEPOS/CHARPOS. If that
909 character is covered by a display string, treat the entire run of
910 covered characters as a single character u+FFFC, and return their
911 combined length in CH_LEN and NCHARS. DISP_POS specifies the
912 character position of the next display string, or -1 if not yet
913 computed. DISP_PROP_P non-zero means that there's really a display
914 string at DISP_POS, as opposed to when we searched till DISP_POS
915 without findingone. When the next character is at or beyond that
916 position, the function updates DISP_POS with the position of the
917 next display string. STRING->s is the C string to iterate, or NULL
918 if iterating over a buffer or a Lisp string; in the latter case,
919 STRING->lstring is the Lisp string. */
920 static inline int
921 bidi_fetch_char (EMACS_INT bytepos, EMACS_INT charpos, EMACS_INT *disp_pos,
922 int *disp_prop_p, struct bidi_string_data *string,
923 int frame_window_p, EMACS_INT *ch_len, EMACS_INT *nchars)
924 {
925 int ch;
926 EMACS_INT endpos =
927 (string->s || STRINGP (string->lstring)) ? string->schars : ZV;
928 struct text_pos pos;
929
930 /* If we got past the last known position of display string, compute
931 the position of the next one. That position could be at CHARPOS. */
932 if (charpos < endpos && charpos > *disp_pos)
933 {
934 SET_TEXT_POS (pos, charpos, bytepos);
935 *disp_pos = compute_display_string_pos (&pos, string, frame_window_p,
936 disp_prop_p);
937 }
938
939 /* Fetch the character at BYTEPOS. */
940 if (charpos >= endpos)
941 {
942 ch = BIDI_EOB;
943 *ch_len = 1;
944 *nchars = 1;
945 *disp_pos = endpos;
946 *disp_prop_p = 0;
947 }
948 else if (charpos >= *disp_pos && *disp_prop_p)
949 {
950 EMACS_INT disp_end_pos;
951
952 /* We don't expect to find ourselves in the middle of a display
953 property. Hopefully, it will never be needed. */
954 if (charpos > *disp_pos)
955 abort ();
956 /* Return the Unicode Object Replacement Character to represent
957 the entire run of characters covered by the display string. */
958 ch = 0xFFFC;
959 disp_end_pos = compute_display_string_end (*disp_pos, string);
960 *nchars = disp_end_pos - *disp_pos;
961 if (*nchars <= 0)
962 abort ();
963 if (string->s)
964 *ch_len = bidi_count_bytes (string->s, *disp_pos, bytepos,
965 disp_end_pos, string->unibyte);
966 else if (STRINGP (string->lstring))
967 *ch_len = bidi_count_bytes (SDATA (string->lstring), *disp_pos,
968 bytepos, disp_end_pos, string->unibyte);
969 else
970 *ch_len = CHAR_TO_BYTE (disp_end_pos) - bytepos;
971 }
972 else
973 {
974 if (string->s)
975 {
976 int len;
977
978 if (!string->unibyte)
979 {
980 ch = STRING_CHAR_AND_LENGTH (string->s + bytepos, len);
981 *ch_len = len;
982 }
983 else
984 {
985 ch = UNIBYTE_TO_CHAR (string->s[bytepos]);
986 *ch_len = 1;
987 }
988 }
989 else if (STRINGP (string->lstring))
990 {
991 int len;
992
993 if (!string->unibyte)
994 {
995 ch = STRING_CHAR_AND_LENGTH (SDATA (string->lstring) + bytepos,
996 len);
997 *ch_len = len;
998 }
999 else
1000 {
1001 ch = UNIBYTE_TO_CHAR (SREF (string->lstring, bytepos));
1002 *ch_len = 1;
1003 }
1004 }
1005 else
1006 {
1007 ch = FETCH_MULTIBYTE_CHAR (bytepos);
1008 *ch_len = CHAR_BYTES (ch);
1009 }
1010 *nchars = 1;
1011 }
1012
1013 /* If we just entered a run of characters covered by a display
1014 string, compute the position of the next display string. */
1015 if (charpos + *nchars <= endpos && charpos + *nchars > *disp_pos
1016 && *disp_prop_p)
1017 {
1018 SET_TEXT_POS (pos, charpos + *nchars, bytepos + *ch_len);
1019 *disp_pos = compute_display_string_pos (&pos, string, frame_window_p,
1020 disp_prop_p);
1021 }
1022
1023 return ch;
1024 }
1025
1026 \f
1027 /***********************************************************************
1028 Determining paragraph direction
1029 ***********************************************************************/
1030
1031 /* Check if buffer position CHARPOS/BYTEPOS is the end of a paragraph.
1032 Value is the non-negative length of the paragraph separator
1033 following the buffer position, -1 if position is at the beginning
1034 of a new paragraph, or -2 if position is neither at beginning nor
1035 at end of a paragraph. */
1036 static EMACS_INT
1037 bidi_at_paragraph_end (EMACS_INT charpos, EMACS_INT bytepos)
1038 {
1039 Lisp_Object sep_re;
1040 Lisp_Object start_re;
1041 EMACS_INT val;
1042
1043 sep_re = paragraph_separate_re;
1044 start_re = paragraph_start_re;
1045
1046 val = fast_looking_at (sep_re, charpos, bytepos, ZV, ZV_BYTE, Qnil);
1047 if (val < 0)
1048 {
1049 if (fast_looking_at (start_re, charpos, bytepos, ZV, ZV_BYTE, Qnil) >= 0)
1050 val = -1;
1051 else
1052 val = -2;
1053 }
1054
1055 return val;
1056 }
1057
1058 /* Find the beginning of this paragraph by looking back in the buffer.
1059 Value is the byte position of the paragraph's beginning. */
1060 static EMACS_INT
1061 bidi_find_paragraph_start (EMACS_INT pos, EMACS_INT pos_byte)
1062 {
1063 Lisp_Object re = paragraph_start_re;
1064 EMACS_INT limit = ZV, limit_byte = ZV_BYTE;
1065
1066 while (pos_byte > BEGV_BYTE
1067 && fast_looking_at (re, pos, pos_byte, limit, limit_byte, Qnil) < 0)
1068 {
1069 /* FIXME: What if the paragraph beginning is covered by a
1070 display string? And what if a display string covering some
1071 of the text over which we scan back includes
1072 paragraph_start_re? */
1073 pos = find_next_newline_no_quit (pos - 1, -1);
1074 pos_byte = CHAR_TO_BYTE (pos);
1075 }
1076 return pos_byte;
1077 }
1078
1079 /* Determine the base direction, a.k.a. base embedding level, of the
1080 paragraph we are about to iterate through. If DIR is either L2R or
1081 R2L, just use that. Otherwise, determine the paragraph direction
1082 from the first strong directional character of the paragraph.
1083
1084 NO_DEFAULT_P non-zero means don't default to L2R if the paragraph
1085 has no strong directional characters and both DIR and
1086 bidi_it->paragraph_dir are NEUTRAL_DIR. In that case, search back
1087 in the buffer until a paragraph is found with a strong character,
1088 or until hitting BEGV. In the latter case, fall back to L2R. This
1089 flag is used in current-bidi-paragraph-direction.
1090
1091 Note that this function gives the paragraph separator the same
1092 direction as the preceding paragraph, even though Emacs generally
1093 views the separartor as not belonging to any paragraph. */
1094 void
1095 bidi_paragraph_init (bidi_dir_t dir, struct bidi_it *bidi_it, int no_default_p)
1096 {
1097 EMACS_INT bytepos = bidi_it->bytepos;
1098 int string_p = bidi_it->string.s != NULL || STRINGP (bidi_it->string.lstring);
1099 EMACS_INT pstartbyte;
1100 /* Note that begbyte is a byte position, while end is a character
1101 position. Yes, this is ugly, but we are trying to avoid costly
1102 calls to BYTE_TO_CHAR and its ilk. */
1103 EMACS_INT begbyte = string_p ? 0 : BEGV_BYTE;
1104 EMACS_INT end = string_p ? bidi_it->string.schars : ZV;
1105
1106 /* Special case for an empty buffer. */
1107 if (bytepos == begbyte && bidi_it->charpos == end)
1108 dir = L2R;
1109 /* We should never be called at EOB or before BEGV. */
1110 else if (bidi_it->charpos >= end || bytepos < begbyte)
1111 abort ();
1112
1113 if (dir == L2R)
1114 {
1115 bidi_it->paragraph_dir = L2R;
1116 bidi_it->new_paragraph = 0;
1117 }
1118 else if (dir == R2L)
1119 {
1120 bidi_it->paragraph_dir = R2L;
1121 bidi_it->new_paragraph = 0;
1122 }
1123 else if (dir == NEUTRAL_DIR) /* P2 */
1124 {
1125 int ch;
1126 EMACS_INT ch_len, nchars;
1127 EMACS_INT pos, disp_pos = -1;
1128 int disp_prop_p = 0;
1129 bidi_type_t type;
1130 const unsigned char *s;
1131
1132 if (!bidi_initialized)
1133 bidi_initialize ();
1134
1135 /* If we are inside a paragraph separator, we are just waiting
1136 for the separator to be exhausted; use the previous paragraph
1137 direction. But don't do that if we have been just reseated,
1138 because we need to reinitialize below in that case. */
1139 if (!bidi_it->first_elt
1140 && bidi_it->charpos < bidi_it->separator_limit)
1141 return;
1142
1143 /* If we are on a newline, get past it to where the next
1144 paragraph might start. But don't do that at BEGV since then
1145 we are potentially in a new paragraph that doesn't yet
1146 exist. */
1147 pos = bidi_it->charpos;
1148 s = STRINGP (bidi_it->string.lstring) ?
1149 SDATA (bidi_it->string.lstring) : bidi_it->string.s;
1150 if (bytepos > begbyte
1151 && bidi_char_at_pos (bytepos, s, bidi_it->string.unibyte) == '\n')
1152 {
1153 bytepos++;
1154 pos++;
1155 }
1156
1157 /* We are either at the beginning of a paragraph or in the
1158 middle of it. Find where this paragraph starts. */
1159 if (string_p)
1160 {
1161 /* We don't support changes of paragraph direction inside a
1162 string. It is treated as a single paragraph. */
1163 pstartbyte = 0;
1164 }
1165 else
1166 pstartbyte = bidi_find_paragraph_start (pos, bytepos);
1167 bidi_it->separator_limit = -1;
1168 bidi_it->new_paragraph = 0;
1169
1170 /* The following loop is run more than once only if NO_DEFAULT_P
1171 is non-zero, and only if we are iterating on a buffer. */
1172 do {
1173 bytepos = pstartbyte;
1174 if (!string_p)
1175 pos = BYTE_TO_CHAR (bytepos);
1176 ch = bidi_fetch_char (bytepos, pos, &disp_pos, &disp_prop_p,
1177 &bidi_it->string,
1178 bidi_it->frame_window_p, &ch_len, &nchars);
1179 type = bidi_get_type (ch, NEUTRAL_DIR);
1180
1181 for (pos += nchars, bytepos += ch_len;
1182 (bidi_get_category (type) != STRONG)
1183 || (bidi_ignore_explicit_marks_for_paragraph_level
1184 && (type == RLE || type == RLO
1185 || type == LRE || type == LRO));
1186 type = bidi_get_type (ch, NEUTRAL_DIR))
1187 {
1188 if (pos >= end)
1189 {
1190 /* Pretend there's a paragraph separator at end of
1191 buffer/string. */
1192 type = NEUTRAL_B;
1193 break;
1194 }
1195 if (!string_p
1196 && type == NEUTRAL_B
1197 && bidi_at_paragraph_end (pos, bytepos) >= -1)
1198 break;
1199 /* Fetch next character and advance to get past it. */
1200 ch = bidi_fetch_char (bytepos, pos, &disp_pos,
1201 &disp_prop_p, &bidi_it->string,
1202 bidi_it->frame_window_p, &ch_len, &nchars);
1203 pos += nchars;
1204 bytepos += ch_len;
1205 }
1206 if ((type == STRONG_R || type == STRONG_AL) /* P3 */
1207 || (!bidi_ignore_explicit_marks_for_paragraph_level
1208 && (type == RLO || type == RLE)))
1209 bidi_it->paragraph_dir = R2L;
1210 else if (type == STRONG_L
1211 || (!bidi_ignore_explicit_marks_for_paragraph_level
1212 && (type == LRO || type == LRE)))
1213 bidi_it->paragraph_dir = L2R;
1214 if (!string_p
1215 && no_default_p && bidi_it->paragraph_dir == NEUTRAL_DIR)
1216 {
1217 /* If this paragraph is at BEGV, default to L2R. */
1218 if (pstartbyte == BEGV_BYTE)
1219 bidi_it->paragraph_dir = L2R; /* P3 and HL1 */
1220 else
1221 {
1222 EMACS_INT prevpbyte = pstartbyte;
1223 EMACS_INT p = BYTE_TO_CHAR (pstartbyte), pbyte = pstartbyte;
1224
1225 /* Find the beginning of the previous paragraph, if any. */
1226 while (pbyte > BEGV_BYTE && prevpbyte >= pstartbyte)
1227 {
1228 /* FXIME: What if p is covered by a display
1229 string? See also a FIXME inside
1230 bidi_find_paragraph_start. */
1231 p--;
1232 pbyte = CHAR_TO_BYTE (p);
1233 prevpbyte = bidi_find_paragraph_start (p, pbyte);
1234 }
1235 pstartbyte = prevpbyte;
1236 }
1237 }
1238 } while (!string_p
1239 && no_default_p && bidi_it->paragraph_dir == NEUTRAL_DIR);
1240 }
1241 else
1242 abort ();
1243
1244 /* Contrary to UAX#9 clause P3, we only default the paragraph
1245 direction to L2R if we have no previous usable paragraph
1246 direction. This is allowed by the HL1 clause. */
1247 if (bidi_it->paragraph_dir != L2R && bidi_it->paragraph_dir != R2L)
1248 bidi_it->paragraph_dir = L2R; /* P3 and HL1 ``higher-level protocols'' */
1249 if (bidi_it->paragraph_dir == R2L)
1250 bidi_it->level_stack[0].level = 1;
1251 else
1252 bidi_it->level_stack[0].level = 0;
1253
1254 bidi_line_init (bidi_it);
1255 }
1256
1257 \f
1258 /***********************************************************************
1259 Resolving explicit and implicit levels.
1260 The rest of this file constitutes the core of the UBA implementation.
1261 ***********************************************************************/
1262
1263 static inline int
1264 bidi_explicit_dir_char (int ch)
1265 {
1266 bidi_type_t ch_type;
1267
1268 if (!bidi_initialized)
1269 abort ();
1270 ch_type = (bidi_type_t) XINT (CHAR_TABLE_REF (bidi_type_table, ch));
1271 return (ch_type == LRE || ch_type == LRO
1272 || ch_type == RLE || ch_type == RLO
1273 || ch_type == PDF);
1274 }
1275
1276 /* A helper function for bidi_resolve_explicit. It advances to the
1277 next character in logical order and determines the new embedding
1278 level and directional override, but does not take into account
1279 empty embeddings. */
1280 static int
1281 bidi_resolve_explicit_1 (struct bidi_it *bidi_it)
1282 {
1283 int curchar;
1284 bidi_type_t type;
1285 int current_level;
1286 int new_level;
1287 bidi_dir_t override;
1288 int string_p = bidi_it->string.s != NULL || STRINGP (bidi_it->string.lstring);
1289
1290 /* If reseat()'ed, don't advance, so as to start iteration from the
1291 position where we were reseated. bidi_it->bytepos can be less
1292 than BEGV_BYTE after reseat to BEGV. */
1293 if (bidi_it->bytepos < (string_p ? 0 : BEGV_BYTE)
1294 || bidi_it->first_elt)
1295 {
1296 bidi_it->first_elt = 0;
1297 if (string_p)
1298 {
1299 const unsigned char *p =
1300 STRINGP (bidi_it->string.lstring)
1301 ? SDATA (bidi_it->string.lstring) : bidi_it->string.s;
1302
1303 if (bidi_it->charpos < 0)
1304 bidi_it->charpos = 0;
1305 bidi_it->bytepos = bidi_count_bytes (p, 0, 0, bidi_it->charpos,
1306 bidi_it->string.unibyte);
1307 }
1308 else
1309 {
1310 if (bidi_it->charpos < BEGV)
1311 bidi_it->charpos = BEGV;
1312 bidi_it->bytepos = CHAR_TO_BYTE (bidi_it->charpos);
1313 }
1314 }
1315 /* Don't move at end of buffer/string. */
1316 else if (bidi_it->charpos < (string_p ? bidi_it->string.schars : ZV))
1317 {
1318 /* Advance to the next character, skipping characters covered by
1319 display strings (nchars > 1). */
1320 if (bidi_it->nchars <= 0)
1321 abort ();
1322 bidi_it->charpos += bidi_it->nchars;
1323 if (bidi_it->ch_len == 0)
1324 abort ();
1325 bidi_it->bytepos += bidi_it->ch_len;
1326 }
1327
1328 current_level = bidi_it->level_stack[bidi_it->stack_idx].level; /* X1 */
1329 override = bidi_it->level_stack[bidi_it->stack_idx].override;
1330 new_level = current_level;
1331
1332 if (bidi_it->charpos >= (string_p ? bidi_it->string.schars : ZV))
1333 {
1334 curchar = BIDI_EOB;
1335 bidi_it->ch_len = 1;
1336 bidi_it->nchars = 1;
1337 bidi_it->disp_pos = (string_p ? bidi_it->string.schars : ZV);
1338 bidi_it->disp_prop_p = 0;
1339 }
1340 else
1341 {
1342 /* Fetch the character at BYTEPOS. If it is covered by a
1343 display string, treat the entire run of covered characters as
1344 a single character u+FFFC. */
1345 curchar = bidi_fetch_char (bidi_it->bytepos, bidi_it->charpos,
1346 &bidi_it->disp_pos, &bidi_it->disp_prop_p,
1347 &bidi_it->string, bidi_it->frame_window_p,
1348 &bidi_it->ch_len, &bidi_it->nchars);
1349 }
1350 bidi_it->ch = curchar;
1351
1352 /* Don't apply directional override here, as all the types we handle
1353 below will not be affected by the override anyway, and we need
1354 the original type unaltered. The override will be applied in
1355 bidi_resolve_weak. */
1356 type = bidi_get_type (curchar, NEUTRAL_DIR);
1357 bidi_it->orig_type = type;
1358 bidi_check_type (bidi_it->orig_type);
1359
1360 if (type != PDF)
1361 bidi_it->prev_was_pdf = 0;
1362
1363 bidi_it->type_after_w1 = UNKNOWN_BT;
1364
1365 switch (type)
1366 {
1367 case RLE: /* X2 */
1368 case RLO: /* X4 */
1369 bidi_it->type_after_w1 = type;
1370 bidi_check_type (bidi_it->type_after_w1);
1371 type = WEAK_BN; /* X9/Retaining */
1372 if (bidi_it->ignore_bn_limit <= -1)
1373 {
1374 if (current_level <= BIDI_MAXLEVEL - 4)
1375 {
1376 /* Compute the least odd embedding level greater than
1377 the current level. */
1378 new_level = ((current_level + 1) & ~1) + 1;
1379 if (bidi_it->type_after_w1 == RLE)
1380 override = NEUTRAL_DIR;
1381 else
1382 override = R2L;
1383 if (current_level == BIDI_MAXLEVEL - 4)
1384 bidi_it->invalid_rl_levels = 0;
1385 bidi_push_embedding_level (bidi_it, new_level, override);
1386 }
1387 else
1388 {
1389 bidi_it->invalid_levels++;
1390 /* See the commentary about invalid_rl_levels below. */
1391 if (bidi_it->invalid_rl_levels < 0)
1392 bidi_it->invalid_rl_levels = 0;
1393 bidi_it->invalid_rl_levels++;
1394 }
1395 }
1396 else if (bidi_it->prev.type_after_w1 == WEAK_EN /* W5/Retaining */
1397 || bidi_it->next_en_pos > bidi_it->charpos)
1398 type = WEAK_EN;
1399 break;
1400 case LRE: /* X3 */
1401 case LRO: /* X5 */
1402 bidi_it->type_after_w1 = type;
1403 bidi_check_type (bidi_it->type_after_w1);
1404 type = WEAK_BN; /* X9/Retaining */
1405 if (bidi_it->ignore_bn_limit <= -1)
1406 {
1407 if (current_level <= BIDI_MAXLEVEL - 5)
1408 {
1409 /* Compute the least even embedding level greater than
1410 the current level. */
1411 new_level = ((current_level + 2) & ~1);
1412 if (bidi_it->type_after_w1 == LRE)
1413 override = NEUTRAL_DIR;
1414 else
1415 override = L2R;
1416 bidi_push_embedding_level (bidi_it, new_level, override);
1417 }
1418 else
1419 {
1420 bidi_it->invalid_levels++;
1421 /* invalid_rl_levels counts invalid levels encountered
1422 while the embedding level was already too high for
1423 LRE/LRO, but not for RLE/RLO. That is because
1424 there may be exactly one PDF which we should not
1425 ignore even though invalid_levels is non-zero.
1426 invalid_rl_levels helps to know what PDF is
1427 that. */
1428 if (bidi_it->invalid_rl_levels >= 0)
1429 bidi_it->invalid_rl_levels++;
1430 }
1431 }
1432 else if (bidi_it->prev.type_after_w1 == WEAK_EN /* W5/Retaining */
1433 || bidi_it->next_en_pos > bidi_it->charpos)
1434 type = WEAK_EN;
1435 break;
1436 case PDF: /* X7 */
1437 bidi_it->type_after_w1 = type;
1438 bidi_check_type (bidi_it->type_after_w1);
1439 type = WEAK_BN; /* X9/Retaining */
1440 if (bidi_it->ignore_bn_limit <= -1)
1441 {
1442 if (!bidi_it->invalid_rl_levels)
1443 {
1444 new_level = bidi_pop_embedding_level (bidi_it);
1445 bidi_it->invalid_rl_levels = -1;
1446 if (bidi_it->invalid_levels)
1447 bidi_it->invalid_levels--;
1448 /* else nothing: UAX#9 says to ignore invalid PDFs */
1449 }
1450 if (!bidi_it->invalid_levels)
1451 new_level = bidi_pop_embedding_level (bidi_it);
1452 else
1453 {
1454 bidi_it->invalid_levels--;
1455 bidi_it->invalid_rl_levels--;
1456 }
1457 }
1458 else if (bidi_it->prev.type_after_w1 == WEAK_EN /* W5/Retaining */
1459 || bidi_it->next_en_pos > bidi_it->charpos)
1460 type = WEAK_EN;
1461 break;
1462 default:
1463 /* Nothing. */
1464 break;
1465 }
1466
1467 bidi_it->type = type;
1468 bidi_check_type (bidi_it->type);
1469
1470 return new_level;
1471 }
1472
1473 /* Given an iterator state in BIDI_IT, advance one character position
1474 in the buffer/string to the next character (in the logical order),
1475 resolve any explicit embeddings and directional overrides, and
1476 return the embedding level of the character after resolving
1477 explicit directives and ignoring empty embeddings. */
1478 static int
1479 bidi_resolve_explicit (struct bidi_it *bidi_it)
1480 {
1481 int prev_level = bidi_it->level_stack[bidi_it->stack_idx].level;
1482 int new_level = bidi_resolve_explicit_1 (bidi_it);
1483 EMACS_INT eob = bidi_it->string.s ? bidi_it->string.schars : ZV;
1484 const unsigned char *s = STRINGP (bidi_it->string.lstring)
1485 ? SDATA (bidi_it->string.lstring) : bidi_it->string.s;
1486
1487 if (prev_level < new_level
1488 && bidi_it->type == WEAK_BN
1489 && bidi_it->ignore_bn_limit == -1 /* only if not already known */
1490 && bidi_it->charpos < eob /* not already at EOB */
1491 && bidi_explicit_dir_char (bidi_char_at_pos (bidi_it->bytepos
1492 + bidi_it->ch_len, s,
1493 bidi_it->string.unibyte)))
1494 {
1495 /* Avoid pushing and popping embedding levels if the level run
1496 is empty, as this breaks level runs where it shouldn't.
1497 UAX#9 removes all the explicit embedding and override codes,
1498 so empty embeddings disappear without a trace. We need to
1499 behave as if we did the same. */
1500 struct bidi_it saved_it;
1501 int level = prev_level;
1502
1503 bidi_copy_it (&saved_it, bidi_it);
1504
1505 while (bidi_explicit_dir_char (bidi_char_at_pos (bidi_it->bytepos
1506 + bidi_it->ch_len, s,
1507 bidi_it->string.unibyte)))
1508 {
1509 /* This advances to the next character, skipping any
1510 characters covered by display strings. */
1511 level = bidi_resolve_explicit_1 (bidi_it);
1512 /* If string.lstring was relocated inside bidi_resolve_explicit_1,
1513 a pointer to its data is no longer valid. */
1514 if (STRINGP (bidi_it->string.lstring))
1515 s = SDATA (bidi_it->string.lstring);
1516 }
1517
1518 if (bidi_it->nchars <= 0)
1519 abort ();
1520 if (level == prev_level) /* empty embedding */
1521 saved_it.ignore_bn_limit = bidi_it->charpos + bidi_it->nchars;
1522 else /* this embedding is non-empty */
1523 saved_it.ignore_bn_limit = -2;
1524
1525 bidi_copy_it (bidi_it, &saved_it);
1526 if (bidi_it->ignore_bn_limit > -1)
1527 {
1528 /* We pushed a level, but we shouldn't have. Undo that. */
1529 if (!bidi_it->invalid_rl_levels)
1530 {
1531 new_level = bidi_pop_embedding_level (bidi_it);
1532 bidi_it->invalid_rl_levels = -1;
1533 if (bidi_it->invalid_levels)
1534 bidi_it->invalid_levels--;
1535 }
1536 if (!bidi_it->invalid_levels)
1537 new_level = bidi_pop_embedding_level (bidi_it);
1538 else
1539 {
1540 bidi_it->invalid_levels--;
1541 bidi_it->invalid_rl_levels--;
1542 }
1543 }
1544 }
1545
1546 if (bidi_it->type == NEUTRAL_B) /* X8 */
1547 {
1548 bidi_set_paragraph_end (bidi_it);
1549 /* This is needed by bidi_resolve_weak below, and in L1. */
1550 bidi_it->type_after_w1 = bidi_it->type;
1551 bidi_check_type (bidi_it->type_after_w1);
1552 }
1553
1554 return new_level;
1555 }
1556
1557 /* Advance in the buffer/string, resolve weak types and return the
1558 type of the next character after weak type resolution. */
1559 static bidi_type_t
1560 bidi_resolve_weak (struct bidi_it *bidi_it)
1561 {
1562 bidi_type_t type;
1563 bidi_dir_t override;
1564 int prev_level = bidi_it->level_stack[bidi_it->stack_idx].level;
1565 int new_level = bidi_resolve_explicit (bidi_it);
1566 int next_char;
1567 bidi_type_t type_of_next;
1568 struct bidi_it saved_it;
1569 EMACS_INT eob =
1570 (STRINGP (bidi_it->string.lstring) || bidi_it->string.s)
1571 ? bidi_it->string.schars : ZV;
1572
1573 type = bidi_it->type;
1574 override = bidi_it->level_stack[bidi_it->stack_idx].override;
1575
1576 if (type == UNKNOWN_BT
1577 || type == LRE
1578 || type == LRO
1579 || type == RLE
1580 || type == RLO
1581 || type == PDF)
1582 abort ();
1583
1584 if (new_level != prev_level
1585 || bidi_it->type == NEUTRAL_B)
1586 {
1587 /* We've got a new embedding level run, compute the directional
1588 type of sor and initialize per-run variables (UAX#9, clause
1589 X10). */
1590 bidi_set_sor_type (bidi_it, prev_level, new_level);
1591 }
1592 else if (type == NEUTRAL_S || type == NEUTRAL_WS
1593 || type == WEAK_BN || type == STRONG_AL)
1594 bidi_it->type_after_w1 = type; /* needed in L1 */
1595 bidi_check_type (bidi_it->type_after_w1);
1596
1597 /* Level and directional override status are already recorded in
1598 bidi_it, and do not need any change; see X6. */
1599 if (override == R2L) /* X6 */
1600 type = STRONG_R;
1601 else if (override == L2R)
1602 type = STRONG_L;
1603 else
1604 {
1605 if (type == WEAK_NSM) /* W1 */
1606 {
1607 /* Note that we don't need to consider the case where the
1608 prev character has its type overridden by an RLO or LRO,
1609 because then either the type of this NSM would have been
1610 also overridden, or the previous character is outside the
1611 current level run, and thus not relevant to this NSM.
1612 This is why NSM gets the type_after_w1 of the previous
1613 character. */
1614 if (bidi_it->prev.type_after_w1 != UNKNOWN_BT
1615 /* if type_after_w1 is NEUTRAL_B, this NSM is at sor */
1616 && bidi_it->prev.type_after_w1 != NEUTRAL_B)
1617 type = bidi_it->prev.type_after_w1;
1618 else if (bidi_it->sor == R2L)
1619 type = STRONG_R;
1620 else if (bidi_it->sor == L2R)
1621 type = STRONG_L;
1622 else /* shouldn't happen! */
1623 abort ();
1624 }
1625 if (type == WEAK_EN /* W2 */
1626 && bidi_it->last_strong.type_after_w1 == STRONG_AL)
1627 type = WEAK_AN;
1628 else if (type == STRONG_AL) /* W3 */
1629 type = STRONG_R;
1630 else if ((type == WEAK_ES /* W4 */
1631 && bidi_it->prev.type_after_w1 == WEAK_EN
1632 && bidi_it->prev.orig_type == WEAK_EN)
1633 || (type == WEAK_CS
1634 && ((bidi_it->prev.type_after_w1 == WEAK_EN
1635 && bidi_it->prev.orig_type == WEAK_EN)
1636 || bidi_it->prev.type_after_w1 == WEAK_AN)))
1637 {
1638 const unsigned char *s =
1639 STRINGP (bidi_it->string.lstring)
1640 ? SDATA (bidi_it->string.lstring) : bidi_it->string.s;
1641
1642 next_char =
1643 bidi_it->charpos + bidi_it->nchars >= eob
1644 ? BIDI_EOB
1645 : bidi_char_at_pos (bidi_it->bytepos + bidi_it->ch_len, s,
1646 bidi_it->string.unibyte);
1647 type_of_next = bidi_get_type (next_char, override);
1648
1649 if (type_of_next == WEAK_BN
1650 || bidi_explicit_dir_char (next_char))
1651 {
1652 bidi_copy_it (&saved_it, bidi_it);
1653 while (bidi_resolve_explicit (bidi_it) == new_level
1654 && bidi_it->type == WEAK_BN)
1655 ;
1656 type_of_next = bidi_it->type;
1657 bidi_copy_it (bidi_it, &saved_it);
1658 }
1659
1660 /* If the next character is EN, but the last strong-type
1661 character is AL, that next EN will be changed to AN when
1662 we process it in W2 above. So in that case, this ES
1663 should not be changed into EN. */
1664 if (type == WEAK_ES
1665 && type_of_next == WEAK_EN
1666 && bidi_it->last_strong.type_after_w1 != STRONG_AL)
1667 type = WEAK_EN;
1668 else if (type == WEAK_CS)
1669 {
1670 if (bidi_it->prev.type_after_w1 == WEAK_AN
1671 && (type_of_next == WEAK_AN
1672 /* If the next character is EN, but the last
1673 strong-type character is AL, EN will be later
1674 changed to AN when we process it in W2 above.
1675 So in that case, this ES should not be
1676 changed into EN. */
1677 || (type_of_next == WEAK_EN
1678 && bidi_it->last_strong.type_after_w1 == STRONG_AL)))
1679 type = WEAK_AN;
1680 else if (bidi_it->prev.type_after_w1 == WEAK_EN
1681 && type_of_next == WEAK_EN
1682 && bidi_it->last_strong.type_after_w1 != STRONG_AL)
1683 type = WEAK_EN;
1684 }
1685 }
1686 else if (type == WEAK_ET /* W5: ET with EN before or after it */
1687 || type == WEAK_BN) /* W5/Retaining */
1688 {
1689 if (bidi_it->prev.type_after_w1 == WEAK_EN /* ET/BN w/EN before it */
1690 || bidi_it->next_en_pos > bidi_it->charpos)
1691 type = WEAK_EN;
1692 else /* W5: ET/BN with EN after it. */
1693 {
1694 EMACS_INT en_pos = bidi_it->charpos + bidi_it->nchars;
1695 const unsigned char *s =
1696 STRINGP (bidi_it->string.lstring)
1697 ? SDATA (bidi_it->string.lstring) : bidi_it->string.s;
1698
1699 if (bidi_it->nchars <= 0)
1700 abort ();
1701 next_char =
1702 bidi_it->charpos + bidi_it->nchars >= eob
1703 ? BIDI_EOB
1704 : bidi_char_at_pos (bidi_it->bytepos + bidi_it->ch_len, s,
1705 bidi_it->string.unibyte);
1706 type_of_next = bidi_get_type (next_char, override);
1707
1708 if (type_of_next == WEAK_ET
1709 || type_of_next == WEAK_BN
1710 || bidi_explicit_dir_char (next_char))
1711 {
1712 bidi_copy_it (&saved_it, bidi_it);
1713 while (bidi_resolve_explicit (bidi_it) == new_level
1714 && (bidi_it->type == WEAK_BN
1715 || bidi_it->type == WEAK_ET))
1716 ;
1717 type_of_next = bidi_it->type;
1718 en_pos = bidi_it->charpos;
1719 bidi_copy_it (bidi_it, &saved_it);
1720 }
1721 if (type_of_next == WEAK_EN)
1722 {
1723 /* If the last strong character is AL, the EN we've
1724 found will become AN when we get to it (W2). */
1725 if (bidi_it->last_strong.type_after_w1 != STRONG_AL)
1726 {
1727 type = WEAK_EN;
1728 /* Remember this EN position, to speed up processing
1729 of the next ETs. */
1730 bidi_it->next_en_pos = en_pos;
1731 }
1732 else if (type == WEAK_BN)
1733 type = NEUTRAL_ON; /* W6/Retaining */
1734 }
1735 }
1736 }
1737 }
1738
1739 if (type == WEAK_ES || type == WEAK_ET || type == WEAK_CS /* W6 */
1740 || (type == WEAK_BN
1741 && (bidi_it->prev.type_after_w1 == WEAK_CS /* W6/Retaining */
1742 || bidi_it->prev.type_after_w1 == WEAK_ES
1743 || bidi_it->prev.type_after_w1 == WEAK_ET)))
1744 type = NEUTRAL_ON;
1745
1746 /* Store the type we've got so far, before we clobber it with strong
1747 types in W7 and while resolving neutral types. But leave alone
1748 the original types that were recorded above, because we will need
1749 them for the L1 clause. */
1750 if (bidi_it->type_after_w1 == UNKNOWN_BT)
1751 bidi_it->type_after_w1 = type;
1752 bidi_check_type (bidi_it->type_after_w1);
1753
1754 if (type == WEAK_EN) /* W7 */
1755 {
1756 if ((bidi_it->last_strong.type_after_w1 == STRONG_L)
1757 || (bidi_it->last_strong.type == UNKNOWN_BT && bidi_it->sor == L2R))
1758 type = STRONG_L;
1759 }
1760
1761 bidi_it->type = type;
1762 bidi_check_type (bidi_it->type);
1763 return type;
1764 }
1765
1766 /* Resolve the type of a neutral character according to the type of
1767 surrounding strong text and the current embedding level. */
1768 static inline bidi_type_t
1769 bidi_resolve_neutral_1 (bidi_type_t prev_type, bidi_type_t next_type, int lev)
1770 {
1771 /* N1: European and Arabic numbers are treated as though they were R. */
1772 if (next_type == WEAK_EN || next_type == WEAK_AN)
1773 next_type = STRONG_R;
1774 if (prev_type == WEAK_EN || prev_type == WEAK_AN)
1775 prev_type = STRONG_R;
1776
1777 if (next_type == prev_type) /* N1 */
1778 return next_type;
1779 else if ((lev & 1) == 0) /* N2 */
1780 return STRONG_L;
1781 else
1782 return STRONG_R;
1783 }
1784
1785 static bidi_type_t
1786 bidi_resolve_neutral (struct bidi_it *bidi_it)
1787 {
1788 int prev_level = bidi_it->level_stack[bidi_it->stack_idx].level;
1789 bidi_type_t type = bidi_resolve_weak (bidi_it);
1790 int current_level = bidi_it->level_stack[bidi_it->stack_idx].level;
1791
1792 if (!(type == STRONG_R
1793 || type == STRONG_L
1794 || type == WEAK_BN
1795 || type == WEAK_EN
1796 || type == WEAK_AN
1797 || type == NEUTRAL_B
1798 || type == NEUTRAL_S
1799 || type == NEUTRAL_WS
1800 || type == NEUTRAL_ON))
1801 abort ();
1802
1803 if (bidi_get_category (type) == NEUTRAL
1804 || (type == WEAK_BN && prev_level == current_level))
1805 {
1806 if (bidi_it->next_for_neutral.type != UNKNOWN_BT)
1807 type = bidi_resolve_neutral_1 (bidi_it->prev_for_neutral.type,
1808 bidi_it->next_for_neutral.type,
1809 current_level);
1810 else
1811 {
1812 /* Arrrgh!! The UAX#9 algorithm is too deeply entrenched in
1813 the assumption of batch-style processing; see clauses W4,
1814 W5, and especially N1, which require to look far forward
1815 (as well as back) in the buffer/string. May the fleas of
1816 a thousand camels infest the armpits of those who design
1817 supposedly general-purpose algorithms by looking at their
1818 own implementations, and fail to consider other possible
1819 implementations! */
1820 struct bidi_it saved_it;
1821 bidi_type_t next_type;
1822
1823 if (bidi_it->scan_dir == -1)
1824 abort ();
1825
1826 bidi_copy_it (&saved_it, bidi_it);
1827 /* Scan the text forward until we find the first non-neutral
1828 character, and then use that to resolve the neutral we
1829 are dealing with now. We also cache the scanned iterator
1830 states, to salvage some of the effort later. */
1831 bidi_cache_iterator_state (bidi_it, 0);
1832 do {
1833 /* Record the info about the previous character, so that
1834 it will be cached below with this state. */
1835 if (bidi_it->type_after_w1 != WEAK_BN /* W1/Retaining */
1836 && bidi_it->type != WEAK_BN)
1837 bidi_remember_char (&bidi_it->prev, bidi_it);
1838 type = bidi_resolve_weak (bidi_it);
1839 /* Paragraph separators have their levels fully resolved
1840 at this point, so cache them as resolved. */
1841 bidi_cache_iterator_state (bidi_it, type == NEUTRAL_B);
1842 /* FIXME: implement L1 here, by testing for a newline and
1843 resetting the level for any sequence of whitespace
1844 characters adjacent to it. */
1845 } while (!(type == NEUTRAL_B
1846 || (type != WEAK_BN
1847 && bidi_get_category (type) != NEUTRAL)
1848 /* This is all per level run, so stop when we
1849 reach the end of this level run. */
1850 || bidi_it->level_stack[bidi_it->stack_idx].level !=
1851 current_level));
1852
1853 bidi_remember_char (&saved_it.next_for_neutral, bidi_it);
1854
1855 switch (type)
1856 {
1857 case STRONG_L:
1858 case STRONG_R:
1859 case STRONG_AL:
1860 next_type = type;
1861 break;
1862 case WEAK_EN:
1863 case WEAK_AN:
1864 /* N1: ``European and Arabic numbers are treated as
1865 though they were R.'' */
1866 next_type = STRONG_R;
1867 saved_it.next_for_neutral.type = STRONG_R;
1868 break;
1869 case WEAK_BN:
1870 if (!bidi_explicit_dir_char (bidi_it->ch))
1871 abort (); /* can't happen: BNs are skipped */
1872 /* FALLTHROUGH */
1873 case NEUTRAL_B:
1874 /* Marched all the way to the end of this level run.
1875 We need to use the eor type, whose information is
1876 stored by bidi_set_sor_type in the prev_for_neutral
1877 member. */
1878 if (saved_it.type != WEAK_BN
1879 || bidi_get_category (bidi_it->prev.type_after_w1) == NEUTRAL)
1880 {
1881 next_type = bidi_it->prev_for_neutral.type;
1882 saved_it.next_for_neutral.type = next_type;
1883 bidi_check_type (next_type);
1884 }
1885 else
1886 {
1887 /* This is a BN which does not adjoin neutrals.
1888 Leave its type alone. */
1889 bidi_copy_it (bidi_it, &saved_it);
1890 return bidi_it->type;
1891 }
1892 break;
1893 default:
1894 abort ();
1895 }
1896 type = bidi_resolve_neutral_1 (saved_it.prev_for_neutral.type,
1897 next_type, current_level);
1898 saved_it.type = type;
1899 bidi_check_type (type);
1900 bidi_copy_it (bidi_it, &saved_it);
1901 }
1902 }
1903 return type;
1904 }
1905
1906 /* Given an iterator state in BIDI_IT, advance one character position
1907 in the buffer/string to the next character (in the logical order),
1908 resolve the bidi type of that next character, and return that
1909 type. */
1910 static bidi_type_t
1911 bidi_type_of_next_char (struct bidi_it *bidi_it)
1912 {
1913 bidi_type_t type;
1914
1915 /* This should always be called during a forward scan. */
1916 if (bidi_it->scan_dir != 1)
1917 abort ();
1918
1919 /* Reset the limit until which to ignore BNs if we step out of the
1920 area where we found only empty levels. */
1921 if ((bidi_it->ignore_bn_limit > -1
1922 && bidi_it->ignore_bn_limit <= bidi_it->charpos)
1923 || (bidi_it->ignore_bn_limit == -2
1924 && !bidi_explicit_dir_char (bidi_it->ch)))
1925 bidi_it->ignore_bn_limit = -1;
1926
1927 type = bidi_resolve_neutral (bidi_it);
1928
1929 return type;
1930 }
1931
1932 /* Given an iterator state BIDI_IT, advance one character position in
1933 the buffer/string to the next character (in the current scan
1934 direction), resolve the embedding and implicit levels of that next
1935 character, and return the resulting level. */
1936 static int
1937 bidi_level_of_next_char (struct bidi_it *bidi_it)
1938 {
1939 bidi_type_t type;
1940 int level, prev_level = -1;
1941 struct bidi_saved_info next_for_neutral;
1942 EMACS_INT next_char_pos = -2;
1943
1944 if (bidi_it->scan_dir == 1)
1945 {
1946 EMACS_INT eob =
1947 (bidi_it->string.s || STRINGP (bidi_it->string.lstring))
1948 ? bidi_it->string.schars : ZV;
1949
1950 /* There's no sense in trying to advance if we hit end of text. */
1951 if (bidi_it->charpos >= eob)
1952 return bidi_it->resolved_level;
1953
1954 /* Record the info about the previous character. */
1955 if (bidi_it->type_after_w1 != WEAK_BN /* W1/Retaining */
1956 && bidi_it->type != WEAK_BN)
1957 bidi_remember_char (&bidi_it->prev, bidi_it);
1958 if (bidi_it->type_after_w1 == STRONG_R
1959 || bidi_it->type_after_w1 == STRONG_L
1960 || bidi_it->type_after_w1 == STRONG_AL)
1961 bidi_remember_char (&bidi_it->last_strong, bidi_it);
1962 /* FIXME: it sounds like we don't need both prev and
1963 prev_for_neutral members, but I'm leaving them both for now. */
1964 if (bidi_it->type == STRONG_R || bidi_it->type == STRONG_L
1965 || bidi_it->type == WEAK_EN || bidi_it->type == WEAK_AN)
1966 bidi_remember_char (&bidi_it->prev_for_neutral, bidi_it);
1967
1968 /* If we overstepped the characters used for resolving neutrals
1969 and whitespace, invalidate their info in the iterator. */
1970 if (bidi_it->charpos >= bidi_it->next_for_neutral.charpos)
1971 bidi_it->next_for_neutral.type = UNKNOWN_BT;
1972 if (bidi_it->next_en_pos >= 0
1973 && bidi_it->charpos >= bidi_it->next_en_pos)
1974 bidi_it->next_en_pos = -1;
1975 if (bidi_it->next_for_ws.type != UNKNOWN_BT
1976 && bidi_it->charpos >= bidi_it->next_for_ws.charpos)
1977 bidi_it->next_for_ws.type = UNKNOWN_BT;
1978
1979 /* This must be taken before we fill the iterator with the info
1980 about the next char. If we scan backwards, the iterator
1981 state must be already cached, so there's no need to know the
1982 embedding level of the previous character, since we will be
1983 returning to our caller shortly. */
1984 prev_level = bidi_it->level_stack[bidi_it->stack_idx].level;
1985 }
1986 next_for_neutral = bidi_it->next_for_neutral;
1987
1988 /* Perhaps the character we want is already cached. If it is, the
1989 call to bidi_cache_find below will return a type other than
1990 UNKNOWN_BT. */
1991 if (bidi_cache_idx > bidi_cache_start && !bidi_it->first_elt)
1992 {
1993 int bob =
1994 (bidi_it->string.s || STRINGP (bidi_it->string.lstring)) ? 0 : 1;
1995
1996 if (bidi_it->scan_dir > 0)
1997 {
1998 if (bidi_it->nchars <= 0)
1999 abort ();
2000 next_char_pos = bidi_it->charpos + bidi_it->nchars;
2001 }
2002 else if (bidi_it->charpos >= bob)
2003 /* Implementation note: we allow next_char_pos to be as low as
2004 0 for buffers or -1 for strings, and that is okay because
2005 that's the "position" of the sentinel iterator state we
2006 cached at the beginning of the iteration. */
2007 next_char_pos = bidi_it->charpos - 1;
2008 if (next_char_pos >= bob - 1)
2009 type = bidi_cache_find (next_char_pos, -1, bidi_it);
2010 else
2011 type = UNKNOWN_BT;
2012 }
2013 else
2014 type = UNKNOWN_BT;
2015 if (type != UNKNOWN_BT)
2016 {
2017 /* Don't lose the information for resolving neutrals! The
2018 cached states could have been cached before their
2019 next_for_neutral member was computed. If we are on our way
2020 forward, we can simply take the info from the previous
2021 state. */
2022 if (bidi_it->scan_dir == 1
2023 && bidi_it->next_for_neutral.type == UNKNOWN_BT)
2024 bidi_it->next_for_neutral = next_for_neutral;
2025
2026 /* If resolved_level is -1, it means this state was cached
2027 before it was completely resolved, so we cannot return
2028 it. */
2029 if (bidi_it->resolved_level != -1)
2030 return bidi_it->resolved_level;
2031 }
2032 if (bidi_it->scan_dir == -1)
2033 /* If we are going backwards, the iterator state is already cached
2034 from previous scans, and should be fully resolved. */
2035 abort ();
2036
2037 if (type == UNKNOWN_BT)
2038 type = bidi_type_of_next_char (bidi_it);
2039
2040 if (type == NEUTRAL_B)
2041 return bidi_it->resolved_level;
2042
2043 level = bidi_it->level_stack[bidi_it->stack_idx].level;
2044 if ((bidi_get_category (type) == NEUTRAL /* && type != NEUTRAL_B */)
2045 || (type == WEAK_BN && prev_level == level))
2046 {
2047 if (bidi_it->next_for_neutral.type == UNKNOWN_BT)
2048 abort ();
2049
2050 /* If the cached state shows a neutral character, it was not
2051 resolved by bidi_resolve_neutral, so do it now. */
2052 type = bidi_resolve_neutral_1 (bidi_it->prev_for_neutral.type,
2053 bidi_it->next_for_neutral.type,
2054 level);
2055 }
2056
2057 if (!(type == STRONG_R
2058 || type == STRONG_L
2059 || type == WEAK_BN
2060 || type == WEAK_EN
2061 || type == WEAK_AN))
2062 abort ();
2063 bidi_it->type = type;
2064 bidi_check_type (bidi_it->type);
2065
2066 /* For L1 below, we need to know, for each WS character, whether
2067 it belongs to a sequence of WS characters preceding a newline
2068 or a TAB or a paragraph separator. */
2069 if (bidi_it->orig_type == NEUTRAL_WS
2070 && bidi_it->next_for_ws.type == UNKNOWN_BT)
2071 {
2072 int ch;
2073 EMACS_INT clen = bidi_it->ch_len;
2074 EMACS_INT bpos = bidi_it->bytepos;
2075 EMACS_INT cpos = bidi_it->charpos;
2076 EMACS_INT disp_pos = bidi_it->disp_pos;
2077 EMACS_INT nc = bidi_it->nchars;
2078 struct bidi_string_data bs = bidi_it->string;
2079 bidi_type_t chtype;
2080 int fwp = bidi_it->frame_window_p;
2081 int dpp = bidi_it->disp_prop_p;
2082
2083 if (bidi_it->nchars <= 0)
2084 abort ();
2085 do {
2086 ch = bidi_fetch_char (bpos += clen, cpos += nc, &disp_pos, &dpp, &bs,
2087 fwp, &clen, &nc);
2088 if (ch == '\n' || ch == BIDI_EOB /* || ch == LINESEP_CHAR */)
2089 chtype = NEUTRAL_B;
2090 else
2091 chtype = bidi_get_type (ch, NEUTRAL_DIR);
2092 } while (chtype == NEUTRAL_WS || chtype == WEAK_BN
2093 || bidi_explicit_dir_char (ch)); /* L1/Retaining */
2094 bidi_it->next_for_ws.type = chtype;
2095 bidi_check_type (bidi_it->next_for_ws.type);
2096 bidi_it->next_for_ws.charpos = cpos;
2097 bidi_it->next_for_ws.bytepos = bpos;
2098 }
2099
2100 /* Resolve implicit levels, with a twist: PDFs get the embedding
2101 level of the enbedding they terminate. See below for the
2102 reason. */
2103 if (bidi_it->orig_type == PDF
2104 /* Don't do this if this formatting code didn't change the
2105 embedding level due to invalid or empty embeddings. */
2106 && prev_level != level)
2107 {
2108 /* Don't look in UAX#9 for the reason for this: it's our own
2109 private quirk. The reason is that we want the formatting
2110 codes to be delivered so that they bracket the text of their
2111 embedding. For example, given the text
2112
2113 {RLO}teST{PDF}
2114
2115 we want it to be displayed as
2116
2117 {PDF}STet{RLO}
2118
2119 not as
2120
2121 STet{RLO}{PDF}
2122
2123 which will result because we bump up the embedding level as
2124 soon as we see the RLO and pop it as soon as we see the PDF,
2125 so RLO itself has the same embedding level as "teST", and
2126 thus would be normally delivered last, just before the PDF.
2127 The switch below fiddles with the level of PDF so that this
2128 ugly side effect does not happen.
2129
2130 (This is, of course, only important if the formatting codes
2131 are actually displayed, but Emacs does need to display them
2132 if the user wants to.) */
2133 level = prev_level;
2134 }
2135 else if (bidi_it->orig_type == NEUTRAL_B /* L1 */
2136 || bidi_it->orig_type == NEUTRAL_S
2137 || bidi_it->ch == '\n' || bidi_it->ch == BIDI_EOB
2138 /* || bidi_it->ch == LINESEP_CHAR */
2139 || (bidi_it->orig_type == NEUTRAL_WS
2140 && (bidi_it->next_for_ws.type == NEUTRAL_B
2141 || bidi_it->next_for_ws.type == NEUTRAL_S)))
2142 level = bidi_it->level_stack[0].level;
2143 else if ((level & 1) == 0) /* I1 */
2144 {
2145 if (type == STRONG_R)
2146 level++;
2147 else if (type == WEAK_EN || type == WEAK_AN)
2148 level += 2;
2149 }
2150 else /* I2 */
2151 {
2152 if (type == STRONG_L || type == WEAK_EN || type == WEAK_AN)
2153 level++;
2154 }
2155
2156 bidi_it->resolved_level = level;
2157 return level;
2158 }
2159
2160 /* Move to the other edge of a level given by LEVEL. If END_FLAG is
2161 non-zero, we are at the end of a level, and we need to prepare to
2162 resume the scan of the lower level.
2163
2164 If this level's other edge is cached, we simply jump to it, filling
2165 the iterator structure with the iterator state on the other edge.
2166 Otherwise, we walk the buffer or string until we come back to the
2167 same level as LEVEL.
2168
2169 Note: we are not talking here about a ``level run'' in the UAX#9
2170 sense of the term, but rather about a ``level'' which includes
2171 all the levels higher than it. In other words, given the levels
2172 like this:
2173
2174 11111112222222333333334443343222222111111112223322111
2175 A B C
2176
2177 and assuming we are at point A scanning left to right, this
2178 function moves to point C, whereas the UAX#9 ``level 2 run'' ends
2179 at point B. */
2180 static void
2181 bidi_find_other_level_edge (struct bidi_it *bidi_it, int level, int end_flag)
2182 {
2183 int dir = end_flag ? -bidi_it->scan_dir : bidi_it->scan_dir;
2184 ptrdiff_t idx;
2185
2186 /* Try the cache first. */
2187 if ((idx = bidi_cache_find_level_change (level, dir, end_flag))
2188 >= bidi_cache_start)
2189 bidi_cache_fetch_state (idx, bidi_it);
2190 else
2191 {
2192 int new_level;
2193
2194 if (end_flag)
2195 abort (); /* if we are at end of level, its edges must be cached */
2196
2197 bidi_cache_iterator_state (bidi_it, 1);
2198 do {
2199 new_level = bidi_level_of_next_char (bidi_it);
2200 bidi_cache_iterator_state (bidi_it, 1);
2201 } while (new_level >= level);
2202 }
2203 }
2204
2205 void
2206 bidi_move_to_visually_next (struct bidi_it *bidi_it)
2207 {
2208 int old_level, new_level, next_level;
2209 struct bidi_it sentinel;
2210 struct gcpro gcpro1;
2211
2212 if (bidi_it->charpos < 0 || bidi_it->bytepos < 0)
2213 abort ();
2214
2215 if (bidi_it->scan_dir == 0)
2216 {
2217 bidi_it->scan_dir = 1; /* default to logical order */
2218 }
2219
2220 /* The code below can call eval, and thus cause GC. If we are
2221 iterating a Lisp string, make sure it won't be GCed. */
2222 if (STRINGP (bidi_it->string.lstring))
2223 GCPRO1 (bidi_it->string.lstring);
2224
2225 /* If we just passed a newline, initialize for the next line. */
2226 if (!bidi_it->first_elt && bidi_it->orig_type == NEUTRAL_B)
2227 bidi_line_init (bidi_it);
2228
2229 /* Prepare the sentinel iterator state, and cache it. When we bump
2230 into it, scanning backwards, we'll know that the last non-base
2231 level is exhausted. */
2232 if (bidi_cache_idx == bidi_cache_start)
2233 {
2234 bidi_copy_it (&sentinel, bidi_it);
2235 if (bidi_it->first_elt)
2236 {
2237 sentinel.charpos--; /* cached charpos needs to be monotonic */
2238 sentinel.bytepos--;
2239 sentinel.ch = '\n'; /* doesn't matter, but why not? */
2240 sentinel.ch_len = 1;
2241 sentinel.nchars = 1;
2242 }
2243 bidi_cache_iterator_state (&sentinel, 1);
2244 }
2245
2246 old_level = bidi_it->resolved_level;
2247 new_level = bidi_level_of_next_char (bidi_it);
2248
2249 /* Reordering of resolved levels (clause L2) is implemented by
2250 jumping to the other edge of the level and flipping direction of
2251 scanning the text whenever we find a level change. */
2252 if (new_level != old_level)
2253 {
2254 int ascending = new_level > old_level;
2255 int level_to_search = ascending ? old_level + 1 : old_level;
2256 int incr = ascending ? 1 : -1;
2257 int expected_next_level = old_level + incr;
2258
2259 /* Jump (or walk) to the other edge of this level. */
2260 bidi_find_other_level_edge (bidi_it, level_to_search, !ascending);
2261 /* Switch scan direction and peek at the next character in the
2262 new direction. */
2263 bidi_it->scan_dir = -bidi_it->scan_dir;
2264
2265 /* The following loop handles the case where the resolved level
2266 jumps by more than one. This is typical for numbers inside a
2267 run of text with left-to-right embedding direction, but can
2268 also happen in other situations. In those cases the decision
2269 where to continue after a level change, and in what direction,
2270 is tricky. For example, given a text like below:
2271
2272 abcdefgh
2273 11336622
2274
2275 (where the numbers below the text show the resolved levels),
2276 the result of reordering according to UAX#9 should be this:
2277
2278 efdcghba
2279
2280 This is implemented by the loop below which flips direction
2281 and jumps to the other edge of the level each time it finds
2282 the new level not to be the expected one. The expected level
2283 is always one more or one less than the previous one. */
2284 next_level = bidi_peek_at_next_level (bidi_it);
2285 while (next_level != expected_next_level)
2286 {
2287 expected_next_level += incr;
2288 level_to_search += incr;
2289 bidi_find_other_level_edge (bidi_it, level_to_search, !ascending);
2290 bidi_it->scan_dir = -bidi_it->scan_dir;
2291 next_level = bidi_peek_at_next_level (bidi_it);
2292 }
2293
2294 /* Finally, deliver the next character in the new direction. */
2295 next_level = bidi_level_of_next_char (bidi_it);
2296 }
2297
2298 /* Take note when we have just processed the newline that precedes
2299 the end of the paragraph. The next time we are about to be
2300 called, set_iterator_to_next will automatically reinit the
2301 paragraph direction, if needed. We do this at the newline before
2302 the paragraph separator, because the next character might not be
2303 the first character of the next paragraph, due to the bidi
2304 reordering, whereas we _must_ know the paragraph base direction
2305 _before_ we process the paragraph's text, since the base
2306 direction affects the reordering. */
2307 if (bidi_it->scan_dir == 1 && bidi_it->orig_type == NEUTRAL_B)
2308 {
2309 /* The paragraph direction of the entire string, once
2310 determined, is in effect for the entire string. Setting the
2311 separator limit to the end of the string prevents
2312 bidi_paragraph_init from being called automatically on this
2313 string. */
2314 if (bidi_it->string.s || STRINGP (bidi_it->string.lstring))
2315 bidi_it->separator_limit = bidi_it->string.schars;
2316 else if (bidi_it->bytepos < ZV_BYTE)
2317 {
2318 EMACS_INT sep_len =
2319 bidi_at_paragraph_end (bidi_it->charpos + bidi_it->nchars,
2320 bidi_it->bytepos + bidi_it->ch_len);
2321 if (bidi_it->nchars <= 0)
2322 abort ();
2323 if (sep_len >= 0)
2324 {
2325 bidi_it->new_paragraph = 1;
2326 /* Record the buffer position of the last character of the
2327 paragraph separator. */
2328 bidi_it->separator_limit =
2329 bidi_it->charpos + bidi_it->nchars + sep_len;
2330 }
2331 }
2332 }
2333
2334 if (bidi_it->scan_dir == 1 && bidi_cache_idx > bidi_cache_start)
2335 {
2336 /* If we are at paragraph's base embedding level and beyond the
2337 last cached position, the cache's job is done and we can
2338 discard it. */
2339 if (bidi_it->resolved_level == bidi_it->level_stack[0].level
2340 && bidi_it->charpos > (bidi_cache[bidi_cache_idx - 1].charpos
2341 + bidi_cache[bidi_cache_idx - 1].nchars - 1))
2342 bidi_cache_reset ();
2343 /* But as long as we are caching during forward scan, we must
2344 cache each state, or else the cache integrity will be
2345 compromised: it assumes cached states correspond to buffer
2346 positions 1:1. */
2347 else
2348 bidi_cache_iterator_state (bidi_it, 1);
2349 }
2350
2351 if (STRINGP (bidi_it->string.lstring))
2352 UNGCPRO;
2353 }
2354
2355 /* This is meant to be called from within the debugger, whenever you
2356 wish to examine the cache contents. */
2357 void bidi_dump_cached_states (void) EXTERNALLY_VISIBLE;
2358 void
2359 bidi_dump_cached_states (void)
2360 {
2361 ptrdiff_t i;
2362 int ndigits = 1;
2363
2364 if (bidi_cache_idx == 0)
2365 {
2366 fprintf (stderr, "The cache is empty.\n");
2367 return;
2368 }
2369 fprintf (stderr, "Total of %"pD"d state%s in cache:\n",
2370 bidi_cache_idx, bidi_cache_idx == 1 ? "" : "s");
2371
2372 for (i = bidi_cache[bidi_cache_idx - 1].charpos; i > 0; i /= 10)
2373 ndigits++;
2374 fputs ("ch ", stderr);
2375 for (i = 0; i < bidi_cache_idx; i++)
2376 fprintf (stderr, "%*c", ndigits, bidi_cache[i].ch);
2377 fputs ("\n", stderr);
2378 fputs ("lvl ", stderr);
2379 for (i = 0; i < bidi_cache_idx; i++)
2380 fprintf (stderr, "%*d", ndigits, bidi_cache[i].resolved_level);
2381 fputs ("\n", stderr);
2382 fputs ("pos ", stderr);
2383 for (i = 0; i < bidi_cache_idx; i++)
2384 fprintf (stderr, "%*"pI"d", ndigits, bidi_cache[i].charpos);
2385 fputs ("\n", stderr);
2386 }