1 /* Low-level bidirectional buffer/string-scanning functions for GNU Emacs.
2 Copyright (C) 2000-2001, 2004-2005, 2009-2013 Free Software
5 This file is part of GNU Emacs.
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.
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.
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/>. */
20 /* Written by Eli Zaretskii <eliz@gnu.org>.
22 A sequential implementation of the Unicode Bidirectional algorithm,
23 (UBA) as per UAX#9, a part of the Unicode Standard.
25 Unlike the reference and most other implementations, this one is
26 designed to be called once for every character in the buffer or
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.
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
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
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.
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 described in the "Implementation
55 Notes" section of UAX#9, under "Retaining Format Codes". */
61 #include "character.h"
63 #include "dispextern.h"
65 static bool bidi_initialized
= 0;
67 static Lisp_Object bidi_type_table
, bidi_mirror_table
;
69 #define LRM_CHAR 0x200E
70 #define RLM_CHAR 0x200F
73 /* Data type for describing the bidirectional character categories. */
81 /* UAX#9 says to search only for L, AL, or R types of characters, and
82 ignore RLE, RLO, LRE, and LRO, when determining the base paragraph
83 level. Yudit indeed ignores them. This variable is therefore set
84 by default to ignore them, but clearing it will take them into
86 extern bool bidi_ignore_explicit_marks_for_paragraph_level EXTERNALLY_VISIBLE
;
87 bool bidi_ignore_explicit_marks_for_paragraph_level
= 1;
89 static Lisp_Object paragraph_start_re
, paragraph_separate_re
;
90 static Lisp_Object Qparagraph_start
, Qparagraph_separate
;
93 /***********************************************************************
95 ***********************************************************************/
97 /* Return the bidi type of a character CH, subject to the current
98 directional OVERRIDE. */
100 bidi_get_type (int ch
, bidi_dir_t override
)
102 bidi_type_t default_type
;
106 if (ch
< 0 || ch
> MAX_CHAR
)
109 default_type
= (bidi_type_t
) XINT (CHAR_TABLE_REF (bidi_type_table
, ch
));
110 /* Every valid character code, even those that are unassigned by the
111 UCD, have some bidi-class property, according to
112 DerivedBidiClass.txt file. Therefore, if we ever get UNKNOWN_BT
113 (= zero) code from CHAR_TABLE_REF, that's a bug. */
114 if (default_type
== UNKNOWN_BT
)
117 if (override
== NEUTRAL_DIR
)
120 switch (default_type
)
122 /* Although UAX#9 does not tell, it doesn't make sense to
123 override NEUTRAL_B and LRM/RLM characters. */
138 if (override
== L2R
) /* X6 */
140 else if (override
== R2L
)
143 emacs_abort (); /* can't happen: handled above */
149 bidi_check_type (bidi_type_t type
)
151 eassert (UNKNOWN_BT
<= type
&& type
<= NEUTRAL_ON
);
154 /* Given a bidi TYPE of a character, return its category. */
155 static bidi_category_t
156 bidi_get_category (bidi_type_t type
)
170 case PDF
: /* ??? really?? */
189 /* Return the mirrored character of C, if it has one. If C has no
190 mirrored counterpart, return C.
191 Note: The conditions in UAX#9 clause L4 regarding the surrounding
192 context must be tested by the caller. */
194 bidi_mirror_char (int c
)
200 if (c
< 0 || c
> MAX_CHAR
)
203 val
= CHAR_TABLE_REF (bidi_mirror_table
, c
);
208 /* When debugging, check before assigning to V, so that the check
209 isn't broken by undefined behavior due to int overflow. */
210 eassert (CHAR_VALID_P (XINT (val
)));
214 /* Minimal test we must do in optimized builds, to prevent weird
215 crashes further down the road. */
216 if (v
< 0 || v
> MAX_CHAR
)
225 /* Determine the start-of-run (sor) directional type given the two
226 embedding levels on either side of the run boundary. Also, update
227 the saved info about previously seen characters, since that info is
228 generally valid for a single level run. */
230 bidi_set_sor_type (struct bidi_it
*bidi_it
, int level_before
, int level_after
)
232 int higher_level
= (level_before
> level_after
? level_before
: level_after
);
234 /* The prev_was_pdf gork is required for when we have several PDFs
235 in a row. In that case, we want to compute the sor type for the
236 next level run only once: when we see the first PDF. That's
237 because the sor type depends only on the higher of the two levels
238 that we find on the two sides of the level boundary (see UAX#9,
239 clause X10), and so we don't need to know the final embedding
240 level to which we descend after processing all the PDFs. */
241 if (!bidi_it
->prev_was_pdf
|| level_before
< level_after
)
242 /* FIXME: should the default sor direction be user selectable? */
243 bidi_it
->sor
= ((higher_level
& 1) != 0 ? R2L
: L2R
);
244 if (level_before
> level_after
)
245 bidi_it
->prev_was_pdf
= 1;
247 bidi_it
->prev
.type
= UNKNOWN_BT
;
248 bidi_it
->last_strong
.type
= bidi_it
->last_strong
.type_after_w1
249 = bidi_it
->last_strong
.orig_type
= UNKNOWN_BT
;
250 bidi_it
->prev_for_neutral
.type
= (bidi_it
->sor
== R2L
? STRONG_R
: STRONG_L
);
251 bidi_it
->prev_for_neutral
.charpos
= bidi_it
->charpos
;
252 bidi_it
->prev_for_neutral
.bytepos
= bidi_it
->bytepos
;
253 bidi_it
->next_for_neutral
.type
= bidi_it
->next_for_neutral
.type_after_w1
254 = bidi_it
->next_for_neutral
.orig_type
= UNKNOWN_BT
;
255 bidi_it
->ignore_bn_limit
= -1; /* meaning it's unknown */
258 /* Push the current embedding level and override status; reset the
259 current level to LEVEL and the current override status to OVERRIDE. */
261 bidi_push_embedding_level (struct bidi_it
*bidi_it
,
262 int level
, bidi_dir_t override
)
264 bidi_it
->stack_idx
++;
265 eassert (bidi_it
->stack_idx
< BIDI_MAXLEVEL
);
266 bidi_it
->level_stack
[bidi_it
->stack_idx
].level
= level
;
267 bidi_it
->level_stack
[bidi_it
->stack_idx
].override
= override
;
270 /* Pop the embedding level and directional override status from the
271 stack, and return the new level. */
273 bidi_pop_embedding_level (struct bidi_it
*bidi_it
)
275 /* UAX#9 says to ignore invalid PDFs. */
276 if (bidi_it
->stack_idx
> 0)
277 bidi_it
->stack_idx
--;
278 return bidi_it
->level_stack
[bidi_it
->stack_idx
].level
;
281 /* Record in SAVED_INFO the information about the current character. */
283 bidi_remember_char (struct bidi_saved_info
*saved_info
,
284 struct bidi_it
*bidi_it
)
286 saved_info
->charpos
= bidi_it
->charpos
;
287 saved_info
->bytepos
= bidi_it
->bytepos
;
288 saved_info
->type
= bidi_it
->type
;
289 bidi_check_type (bidi_it
->type
);
290 saved_info
->type_after_w1
= bidi_it
->type_after_w1
;
291 bidi_check_type (bidi_it
->type_after_w1
);
292 saved_info
->orig_type
= bidi_it
->orig_type
;
293 bidi_check_type (bidi_it
->orig_type
);
296 /* Copy the bidi iterator from FROM to TO. To save cycles, this only
297 copies the part of the level stack that is actually in use. */
299 bidi_copy_it (struct bidi_it
*to
, struct bidi_it
*from
)
301 /* Copy everything from the start through the active part of
304 (offsetof (struct bidi_it
, level_stack
[1])
305 + from
->stack_idx
* sizeof from
->level_stack
[0]));
309 /***********************************************************************
310 Caching the bidi iterator states
311 ***********************************************************************/
313 #define BIDI_CACHE_CHUNK 200
314 static struct bidi_it
*bidi_cache
;
315 static ptrdiff_t bidi_cache_size
= 0;
316 enum { elsz
= sizeof (struct bidi_it
) };
317 static ptrdiff_t bidi_cache_idx
; /* next unused cache slot */
318 static ptrdiff_t bidi_cache_last_idx
; /* slot of last cache hit */
319 static ptrdiff_t bidi_cache_start
= 0; /* start of cache for this
322 /* 5-slot stack for saving the start of the previous level of the
323 cache. xdisp.c maintains a 5-slot stack for its iterator state,
324 and we need the same size of our stack. */
325 static ptrdiff_t bidi_cache_start_stack
[IT_STACK_SIZE
];
326 static int bidi_cache_sp
;
328 /* Size of header used by bidi_shelve_cache. */
331 bidi_shelve_header_size
332 = (sizeof (bidi_cache_idx
) + sizeof (bidi_cache_start_stack
)
333 + sizeof (bidi_cache_sp
) + sizeof (bidi_cache_start
)
334 + sizeof (bidi_cache_last_idx
))
337 /* Reset the cache state to the empty state. We only reset the part
338 of the cache relevant to iteration of the current object. Previous
339 objects, which are pushed on the display iterator's stack, are left
340 intact. This is called when the cached information is no more
341 useful for the current iteration, e.g. when we were reseated to a
342 new position on the same object. */
344 bidi_cache_reset (void)
346 bidi_cache_idx
= bidi_cache_start
;
347 bidi_cache_last_idx
= -1;
350 /* Shrink the cache to its minimal size. Called when we init the bidi
351 iterator for reordering a buffer or a string that does not come
352 from display properties, because that means all the previously
353 cached info is of no further use. */
355 bidi_cache_shrink (void)
357 if (bidi_cache_size
> BIDI_CACHE_CHUNK
)
359 bidi_cache
= xrealloc (bidi_cache
, BIDI_CACHE_CHUNK
* elsz
);
360 bidi_cache_size
= BIDI_CACHE_CHUNK
;
366 bidi_cache_fetch_state (ptrdiff_t idx
, struct bidi_it
*bidi_it
)
368 int current_scan_dir
= bidi_it
->scan_dir
;
370 if (idx
< bidi_cache_start
|| idx
>= bidi_cache_idx
)
373 bidi_copy_it (bidi_it
, &bidi_cache
[idx
]);
374 bidi_it
->scan_dir
= current_scan_dir
;
375 bidi_cache_last_idx
= idx
;
378 /* Find a cached state with a given CHARPOS and resolved embedding
379 level less or equal to LEVEL. if LEVEL is -1, disregard the
380 resolved levels in cached states. DIR, if non-zero, means search
381 in that direction from the last cache hit. */
383 bidi_cache_search (ptrdiff_t charpos
, int level
, int dir
)
385 ptrdiff_t i
, i_start
;
387 if (bidi_cache_idx
> bidi_cache_start
)
389 if (bidi_cache_last_idx
== -1)
390 bidi_cache_last_idx
= bidi_cache_idx
- 1;
391 if (charpos
< bidi_cache
[bidi_cache_last_idx
].charpos
)
394 i_start
= bidi_cache_last_idx
- 1;
396 else if (charpos
> (bidi_cache
[bidi_cache_last_idx
].charpos
397 + bidi_cache
[bidi_cache_last_idx
].nchars
- 1))
400 i_start
= bidi_cache_last_idx
+ 1;
403 i_start
= bidi_cache_last_idx
;
407 i_start
= bidi_cache_idx
- 1;
412 /* Linear search for now; FIXME! */
413 for (i
= i_start
; i
>= bidi_cache_start
; i
--)
414 if (bidi_cache
[i
].charpos
<= charpos
415 && charpos
< bidi_cache
[i
].charpos
+ bidi_cache
[i
].nchars
416 && (level
== -1 || bidi_cache
[i
].resolved_level
<= level
))
421 for (i
= i_start
; i
< bidi_cache_idx
; i
++)
422 if (bidi_cache
[i
].charpos
<= charpos
423 && charpos
< bidi_cache
[i
].charpos
+ bidi_cache
[i
].nchars
424 && (level
== -1 || bidi_cache
[i
].resolved_level
<= level
))
432 /* Find a cached state where the resolved level changes to a value
433 that is lower than LEVEL, and return its cache slot index. DIR is
434 the direction to search, starting with the last used cache slot.
435 If DIR is zero, we search backwards from the last occupied cache
436 slot. BEFORE means return the index of the slot that
437 is ``before'' the level change in the search direction. That is,
438 given the cached levels like this:
443 and assuming we are at the position cached at the slot marked with
444 C, searching backwards (DIR = -1) for LEVEL = 2 will return the
445 index of slot B or A, depending whether BEFORE is, respectively,
448 bidi_cache_find_level_change (int level
, int dir
, bool before
)
452 ptrdiff_t i
= dir
? bidi_cache_last_idx
: bidi_cache_idx
- 1;
453 int incr
= before
? 1 : 0;
455 eassert (!dir
|| bidi_cache_last_idx
>= 0);
464 while (i
>= bidi_cache_start
+ incr
)
466 if (bidi_cache
[i
- incr
].resolved_level
>= 0
467 && bidi_cache
[i
- incr
].resolved_level
< level
)
474 while (i
< bidi_cache_idx
- incr
)
476 if (bidi_cache
[i
+ incr
].resolved_level
>= 0
477 && bidi_cache
[i
+ incr
].resolved_level
< level
)
488 bidi_cache_ensure_space (ptrdiff_t idx
)
490 /* Enlarge the cache as needed. */
491 if (idx
>= bidi_cache_size
)
493 /* The bidi cache cannot be larger than the largest Lisp string
495 ptrdiff_t string_or_buffer_bound
496 = max (BUF_BYTES_MAX
, STRING_BYTES_BOUND
);
498 /* Also, it cannot be larger than what C can represent. */
500 = (min (PTRDIFF_MAX
, SIZE_MAX
) - bidi_shelve_header_size
) / elsz
;
503 = xpalloc (bidi_cache
, &bidi_cache_size
,
504 max (BIDI_CACHE_CHUNK
, idx
- bidi_cache_size
+ 1),
505 min (string_or_buffer_bound
, c_bound
), elsz
);
510 bidi_cache_iterator_state (struct bidi_it
*bidi_it
, bool resolved
)
514 /* We should never cache on backward scans. */
515 if (bidi_it
->scan_dir
== -1)
517 idx
= bidi_cache_search (bidi_it
->charpos
, -1, 1);
521 idx
= bidi_cache_idx
;
522 bidi_cache_ensure_space (idx
);
523 /* Character positions should correspond to cache positions 1:1.
524 If we are outside the range of cached positions, the cache is
525 useless and must be reset. */
526 if (idx
> bidi_cache_start
&&
527 (bidi_it
->charpos
> (bidi_cache
[idx
- 1].charpos
528 + bidi_cache
[idx
- 1].nchars
)
529 || bidi_it
->charpos
< bidi_cache
[bidi_cache_start
].charpos
))
532 idx
= bidi_cache_start
;
534 if (bidi_it
->nchars
<= 0)
536 bidi_copy_it (&bidi_cache
[idx
], bidi_it
);
538 bidi_cache
[idx
].resolved_level
= -1;
542 /* Copy only the members which could have changed, to avoid
543 costly copying of the entire struct. */
544 bidi_cache
[idx
].type
= bidi_it
->type
;
545 bidi_check_type (bidi_it
->type
);
546 bidi_cache
[idx
].type_after_w1
= bidi_it
->type_after_w1
;
547 bidi_check_type (bidi_it
->type_after_w1
);
549 bidi_cache
[idx
].resolved_level
= bidi_it
->resolved_level
;
551 bidi_cache
[idx
].resolved_level
= -1;
552 bidi_cache
[idx
].invalid_levels
= bidi_it
->invalid_levels
;
553 bidi_cache
[idx
].invalid_rl_levels
= bidi_it
->invalid_rl_levels
;
554 bidi_cache
[idx
].next_for_neutral
= bidi_it
->next_for_neutral
;
555 bidi_cache
[idx
].next_for_ws
= bidi_it
->next_for_ws
;
556 bidi_cache
[idx
].ignore_bn_limit
= bidi_it
->ignore_bn_limit
;
557 bidi_cache
[idx
].disp_pos
= bidi_it
->disp_pos
;
558 bidi_cache
[idx
].disp_prop
= bidi_it
->disp_prop
;
561 bidi_cache_last_idx
= idx
;
562 if (idx
>= bidi_cache_idx
)
563 bidi_cache_idx
= idx
+ 1;
567 bidi_cache_find (ptrdiff_t charpos
, int level
, struct bidi_it
*bidi_it
)
569 ptrdiff_t i
= bidi_cache_search (charpos
, level
, bidi_it
->scan_dir
);
571 if (i
>= bidi_cache_start
)
573 bidi_dir_t current_scan_dir
= bidi_it
->scan_dir
;
575 bidi_copy_it (bidi_it
, &bidi_cache
[i
]);
576 bidi_cache_last_idx
= i
;
577 /* Don't let scan direction from the cached state override
578 the current scan direction. */
579 bidi_it
->scan_dir
= current_scan_dir
;
580 return bidi_it
->type
;
587 bidi_peek_at_next_level (struct bidi_it
*bidi_it
)
589 if (bidi_cache_idx
== bidi_cache_start
|| bidi_cache_last_idx
== -1)
591 return bidi_cache
[bidi_cache_last_idx
+ bidi_it
->scan_dir
].resolved_level
;
595 /***********************************************************************
596 Pushing and popping the bidi iterator state
597 ***********************************************************************/
599 /* Push the bidi iterator state in preparation for reordering a
600 different object, e.g. display string found at certain buffer
601 position. Pushing the bidi iterator boils down to saving its
602 entire state on the cache and starting a new cache "stacked" on top
603 of the current cache. */
605 bidi_push_it (struct bidi_it
*bidi_it
)
607 /* Save the current iterator state in its entirety after the last
609 bidi_cache_ensure_space (bidi_cache_idx
);
610 bidi_cache
[bidi_cache_idx
++] = *bidi_it
;
612 /* Push the current cache start onto the stack. */
613 eassert (bidi_cache_sp
< IT_STACK_SIZE
);
614 bidi_cache_start_stack
[bidi_cache_sp
++] = bidi_cache_start
;
616 /* Start a new level of cache, and make it empty. */
617 bidi_cache_start
= bidi_cache_idx
;
618 bidi_cache_last_idx
= -1;
621 /* Restore the iterator state saved by bidi_push_it and return the
622 cache to the corresponding state. */
624 bidi_pop_it (struct bidi_it
*bidi_it
)
626 if (bidi_cache_start
<= 0)
629 /* Reset the next free cache slot index to what it was before the
630 call to bidi_push_it. */
631 bidi_cache_idx
= bidi_cache_start
- 1;
633 /* Restore the bidi iterator state saved in the cache. */
634 *bidi_it
= bidi_cache
[bidi_cache_idx
];
636 /* Pop the previous cache start from the stack. */
637 if (bidi_cache_sp
<= 0)
639 bidi_cache_start
= bidi_cache_start_stack
[--bidi_cache_sp
];
641 /* Invalidate the last-used cache slot data. */
642 bidi_cache_last_idx
= -1;
645 static ptrdiff_t bidi_cache_total_alloc
;
647 /* Stash away a copy of the cache and its control variables. */
649 bidi_shelve_cache (void)
651 unsigned char *databuf
;
655 if (bidi_cache_idx
== 0)
658 alloc
= (bidi_shelve_header_size
659 + bidi_cache_idx
* sizeof (struct bidi_it
));
660 databuf
= xmalloc (alloc
);
661 bidi_cache_total_alloc
+= alloc
;
663 memcpy (databuf
, &bidi_cache_idx
, sizeof (bidi_cache_idx
));
664 memcpy (databuf
+ sizeof (bidi_cache_idx
),
665 bidi_cache
, bidi_cache_idx
* sizeof (struct bidi_it
));
666 memcpy (databuf
+ sizeof (bidi_cache_idx
)
667 + bidi_cache_idx
* sizeof (struct bidi_it
),
668 bidi_cache_start_stack
, sizeof (bidi_cache_start_stack
));
669 memcpy (databuf
+ sizeof (bidi_cache_idx
)
670 + bidi_cache_idx
* sizeof (struct bidi_it
)
671 + sizeof (bidi_cache_start_stack
),
672 &bidi_cache_sp
, sizeof (bidi_cache_sp
));
673 memcpy (databuf
+ sizeof (bidi_cache_idx
)
674 + bidi_cache_idx
* sizeof (struct bidi_it
)
675 + sizeof (bidi_cache_start_stack
) + sizeof (bidi_cache_sp
),
676 &bidi_cache_start
, sizeof (bidi_cache_start
));
677 memcpy (databuf
+ sizeof (bidi_cache_idx
)
678 + bidi_cache_idx
* sizeof (struct bidi_it
)
679 + sizeof (bidi_cache_start_stack
) + sizeof (bidi_cache_sp
)
680 + sizeof (bidi_cache_start
),
681 &bidi_cache_last_idx
, sizeof (bidi_cache_last_idx
));
686 /* Restore the cache state from a copy stashed away by
687 bidi_shelve_cache, and free the buffer used to stash that copy.
688 JUST_FREE means free the buffer, but don't restore the
689 cache; used when the corresponding iterator is discarded instead of
692 bidi_unshelve_cache (void *databuf
, bool just_free
)
694 unsigned char *p
= databuf
;
700 /* A NULL pointer means an empty cache. */
701 bidi_cache_start
= 0;
712 memcpy (&idx
, p
, sizeof (bidi_cache_idx
));
713 bidi_cache_total_alloc
714 -= bidi_shelve_header_size
+ idx
* sizeof (struct bidi_it
);
718 memcpy (&bidi_cache_idx
, p
, sizeof (bidi_cache_idx
));
719 bidi_cache_ensure_space (bidi_cache_idx
);
720 memcpy (bidi_cache
, p
+ sizeof (bidi_cache_idx
),
721 bidi_cache_idx
* sizeof (struct bidi_it
));
722 memcpy (bidi_cache_start_stack
,
723 p
+ sizeof (bidi_cache_idx
)
724 + bidi_cache_idx
* sizeof (struct bidi_it
),
725 sizeof (bidi_cache_start_stack
));
726 memcpy (&bidi_cache_sp
,
727 p
+ sizeof (bidi_cache_idx
)
728 + bidi_cache_idx
* sizeof (struct bidi_it
)
729 + sizeof (bidi_cache_start_stack
),
730 sizeof (bidi_cache_sp
));
731 memcpy (&bidi_cache_start
,
732 p
+ sizeof (bidi_cache_idx
)
733 + bidi_cache_idx
* sizeof (struct bidi_it
)
734 + sizeof (bidi_cache_start_stack
) + sizeof (bidi_cache_sp
),
735 sizeof (bidi_cache_start
));
736 memcpy (&bidi_cache_last_idx
,
737 p
+ sizeof (bidi_cache_idx
)
738 + bidi_cache_idx
* sizeof (struct bidi_it
)
739 + sizeof (bidi_cache_start_stack
) + sizeof (bidi_cache_sp
)
740 + sizeof (bidi_cache_start
),
741 sizeof (bidi_cache_last_idx
));
742 bidi_cache_total_alloc
743 -= (bidi_shelve_header_size
744 + bidi_cache_idx
* sizeof (struct bidi_it
));
752 /***********************************************************************
754 ***********************************************************************/
756 bidi_initialize (void)
758 bidi_type_table
= uniprop_table (intern ("bidi-class"));
759 if (NILP (bidi_type_table
))
761 staticpro (&bidi_type_table
);
763 bidi_mirror_table
= uniprop_table (intern ("mirroring"));
764 if (NILP (bidi_mirror_table
))
766 staticpro (&bidi_mirror_table
);
768 Qparagraph_start
= intern ("paragraph-start");
769 staticpro (&Qparagraph_start
);
770 paragraph_start_re
= Fsymbol_value (Qparagraph_start
);
771 if (!STRINGP (paragraph_start_re
))
772 paragraph_start_re
= build_string ("\f\\|[ \t]*$");
773 staticpro (¶graph_start_re
);
774 Qparagraph_separate
= intern ("paragraph-separate");
775 staticpro (&Qparagraph_separate
);
776 paragraph_separate_re
= Fsymbol_value (Qparagraph_separate
);
777 if (!STRINGP (paragraph_separate_re
))
778 paragraph_separate_re
= build_string ("[ \t\f]*$");
779 staticpro (¶graph_separate_re
);
782 bidi_cache_total_alloc
= 0;
784 bidi_initialized
= 1;
787 /* Do whatever UAX#9 clause X8 says should be done at paragraph's
790 bidi_set_paragraph_end (struct bidi_it
*bidi_it
)
792 bidi_it
->invalid_levels
= 0;
793 bidi_it
->invalid_rl_levels
= -1;
794 bidi_it
->stack_idx
= 0;
795 bidi_it
->resolved_level
= bidi_it
->level_stack
[0].level
;
798 /* Initialize the bidi iterator from buffer/string position CHARPOS. */
800 bidi_init_it (ptrdiff_t charpos
, ptrdiff_t bytepos
, bool frame_window_p
,
801 struct bidi_it
*bidi_it
)
803 if (! bidi_initialized
)
806 bidi_it
->charpos
= charpos
;
808 bidi_it
->bytepos
= bytepos
;
809 bidi_it
->frame_window_p
= frame_window_p
;
810 bidi_it
->nchars
= -1; /* to be computed in bidi_resolve_explicit_1 */
811 bidi_it
->first_elt
= 1;
812 bidi_set_paragraph_end (bidi_it
);
813 bidi_it
->new_paragraph
= 1;
814 bidi_it
->separator_limit
= -1;
815 bidi_it
->type
= NEUTRAL_B
;
816 bidi_it
->type_after_w1
= NEUTRAL_B
;
817 bidi_it
->orig_type
= NEUTRAL_B
;
818 bidi_it
->prev_was_pdf
= 0;
819 bidi_it
->prev
.type
= bidi_it
->prev
.type_after_w1
820 = bidi_it
->prev
.orig_type
= UNKNOWN_BT
;
821 bidi_it
->last_strong
.type
= bidi_it
->last_strong
.type_after_w1
822 = bidi_it
->last_strong
.orig_type
= UNKNOWN_BT
;
823 bidi_it
->next_for_neutral
.charpos
= -1;
824 bidi_it
->next_for_neutral
.type
825 = bidi_it
->next_for_neutral
.type_after_w1
826 = bidi_it
->next_for_neutral
.orig_type
= UNKNOWN_BT
;
827 bidi_it
->prev_for_neutral
.charpos
= -1;
828 bidi_it
->prev_for_neutral
.type
829 = bidi_it
->prev_for_neutral
.type_after_w1
830 = bidi_it
->prev_for_neutral
.orig_type
= UNKNOWN_BT
;
831 bidi_it
->sor
= L2R
; /* FIXME: should it be user-selectable? */
832 bidi_it
->disp_pos
= -1; /* invalid/unknown */
833 bidi_it
->disp_prop
= 0;
834 /* We can only shrink the cache if we are at the bottom level of its
836 if (bidi_cache_start
== 0)
837 bidi_cache_shrink ();
842 /* Perform initializations for reordering a new line of bidi text. */
844 bidi_line_init (struct bidi_it
*bidi_it
)
846 bidi_it
->scan_dir
= 1; /* FIXME: do we need to have control on this? */
847 bidi_it
->resolved_level
= bidi_it
->level_stack
[0].level
;
848 bidi_it
->level_stack
[0].override
= NEUTRAL_DIR
; /* X1 */
849 bidi_it
->invalid_levels
= 0;
850 bidi_it
->invalid_rl_levels
= -1;
851 /* Setting this to zero will force its recomputation the first time
852 we need it for W5. */
853 bidi_it
->next_en_pos
= 0;
854 bidi_it
->next_en_type
= UNKNOWN_BT
;
855 bidi_it
->next_for_ws
.type
= UNKNOWN_BT
;
856 bidi_set_sor_type (bidi_it
,
857 (bidi_it
->paragraph_dir
== R2L
? 1 : 0),
858 bidi_it
->level_stack
[0].level
); /* X10 */
864 /***********************************************************************
866 ***********************************************************************/
868 /* Count bytes in string S between BEG/BEGBYTE and END. BEG and END
869 are zero-based character positions in S, BEGBYTE is byte position
870 corresponding to BEG. UNIBYTE means S is a unibyte string. */
872 bidi_count_bytes (const unsigned char *s
, const ptrdiff_t beg
,
873 const ptrdiff_t begbyte
, const ptrdiff_t end
, bool unibyte
)
876 const unsigned char *p
= s
+ begbyte
, *start
= p
;
882 if (!CHAR_HEAD_P (*p
))
887 p
+= BYTES_BY_CHAR_HEAD (*p
);
895 /* Fetch and return the character at byte position BYTEPOS. If S is
896 non-NULL, fetch the character from string S; otherwise fetch the
897 character from the current buffer. UNIBYTE means S is a
900 bidi_char_at_pos (ptrdiff_t bytepos
, const unsigned char *s
, bool unibyte
)
909 s
= BYTE_POS_ADDR (bytepos
);
910 return STRING_CHAR (s
);
913 /* Fetch and return the character at BYTEPOS/CHARPOS. If that
914 character is covered by a display string, treat the entire run of
915 covered characters as a single character, either u+2029 or u+FFFC,
916 and return their combined length in CH_LEN and NCHARS. DISP_POS
917 specifies the character position of the next display string, or -1
918 if not yet computed. When the next character is at or beyond that
919 position, the function updates DISP_POS with the position of the
920 next display string. *DISP_PROP non-zero means that there's really
921 a display string at DISP_POS, as opposed to when we searched till
922 DISP_POS without finding one. If *DISP_PROP is 2, it means the
923 display spec is of the form `(space ...)', which is replaced with
924 u+2029 to handle it as a paragraph separator. STRING->s is the C
925 string to iterate, or NULL if iterating over a buffer or a Lisp
926 string; in the latter case, STRING->lstring is the Lisp string. */
928 bidi_fetch_char (ptrdiff_t bytepos
, ptrdiff_t charpos
, ptrdiff_t *disp_pos
,
929 int *disp_prop
, struct bidi_string_data
*string
,
930 bool frame_window_p
, ptrdiff_t *ch_len
, ptrdiff_t *nchars
)
934 = (string
->s
|| STRINGP (string
->lstring
)) ? string
->schars
: ZV
;
938 /* If we got past the last known position of display string, compute
939 the position of the next one. That position could be at CHARPOS. */
940 if (charpos
< endpos
&& charpos
> *disp_pos
)
942 SET_TEXT_POS (pos
, charpos
, bytepos
);
943 *disp_pos
= compute_display_string_pos (&pos
, string
, frame_window_p
,
947 /* Fetch the character at BYTEPOS. */
948 if (charpos
>= endpos
)
956 else if (charpos
>= *disp_pos
&& *disp_prop
)
958 ptrdiff_t disp_end_pos
;
960 /* We don't expect to find ourselves in the middle of a display
961 property. Hopefully, it will never be needed. */
962 if (charpos
> *disp_pos
)
964 /* Text covered by `display' properties and overlays with
965 display properties or display strings is handled as a single
966 character that represents the entire run of characters
967 covered by the display property. */
970 /* `(space ...)' display specs are handled as paragraph
971 separators for the purposes of the reordering; see UAX#9
972 section 3 and clause HL1 in section 4.3 there. */
977 /* All other display specs are handled as the Unicode Object
978 Replacement Character. */
981 disp_end_pos
= compute_display_string_end (*disp_pos
, string
);
982 if (disp_end_pos
< 0)
984 /* Somebody removed the display string from the buffer
985 behind our back. Recover by processing this buffer
986 position as if no display property were present there to
991 *nchars
= disp_end_pos
- *disp_pos
;
995 *ch_len
= bidi_count_bytes (string
->s
, *disp_pos
, bytepos
,
996 disp_end_pos
, string
->unibyte
);
997 else if (STRINGP (string
->lstring
))
998 *ch_len
= bidi_count_bytes (SDATA (string
->lstring
), *disp_pos
,
999 bytepos
, disp_end_pos
, string
->unibyte
);
1001 *ch_len
= CHAR_TO_BYTE (disp_end_pos
) - bytepos
;
1009 if (!string
->unibyte
)
1011 ch
= STRING_CHAR_AND_LENGTH (string
->s
+ bytepos
, len
);
1016 ch
= UNIBYTE_TO_CHAR (string
->s
[bytepos
]);
1020 else if (STRINGP (string
->lstring
))
1022 if (!string
->unibyte
)
1024 ch
= STRING_CHAR_AND_LENGTH (SDATA (string
->lstring
) + bytepos
,
1030 ch
= UNIBYTE_TO_CHAR (SREF (string
->lstring
, bytepos
));
1036 ch
= STRING_CHAR_AND_LENGTH (BYTE_POS_ADDR (bytepos
), len
);
1042 /* If we just entered a run of characters covered by a display
1043 string, compute the position of the next display string. */
1044 if (charpos
+ *nchars
<= endpos
&& charpos
+ *nchars
> *disp_pos
1047 SET_TEXT_POS (pos
, charpos
+ *nchars
, bytepos
+ *ch_len
);
1048 *disp_pos
= compute_display_string_pos (&pos
, string
, frame_window_p
,
1056 /***********************************************************************
1057 Determining paragraph direction
1058 ***********************************************************************/
1060 /* Check if buffer position CHARPOS/BYTEPOS is the end of a paragraph.
1061 Value is the non-negative length of the paragraph separator
1062 following the buffer position, -1 if position is at the beginning
1063 of a new paragraph, or -2 if position is neither at beginning nor
1064 at end of a paragraph. */
1066 bidi_at_paragraph_end (ptrdiff_t charpos
, ptrdiff_t bytepos
)
1069 Lisp_Object start_re
;
1072 sep_re
= paragraph_separate_re
;
1073 start_re
= paragraph_start_re
;
1075 val
= fast_looking_at (sep_re
, charpos
, bytepos
, ZV
, ZV_BYTE
, Qnil
);
1078 if (fast_looking_at (start_re
, charpos
, bytepos
, ZV
, ZV_BYTE
, Qnil
) >= 0)
1087 /* On my 2005-vintage machine, searching back for paragraph start
1088 takes ~1 ms per line. And bidi_paragraph_init is called 4 times
1089 when user types C-p. The number below limits each call to
1090 bidi_paragraph_init to about 10 ms. */
1091 #define MAX_PARAGRAPH_SEARCH 7500
1093 /* Find the beginning of this paragraph by looking back in the buffer.
1094 Value is the byte position of the paragraph's beginning, or
1095 BEGV_BYTE if paragraph_start_re is still not found after looking
1096 back MAX_PARAGRAPH_SEARCH lines in the buffer. */
1098 bidi_find_paragraph_start (ptrdiff_t pos
, ptrdiff_t pos_byte
)
1100 Lisp_Object re
= paragraph_start_re
;
1101 ptrdiff_t limit
= ZV
, limit_byte
= ZV_BYTE
;
1104 while (pos_byte
> BEGV_BYTE
1105 && n
++ < MAX_PARAGRAPH_SEARCH
1106 && fast_looking_at (re
, pos
, pos_byte
, limit
, limit_byte
, Qnil
) < 0)
1107 /* FIXME: What if the paragraph beginning is covered by a
1108 display string? And what if a display string covering some
1109 of the text over which we scan back includes
1110 paragraph_start_re? */
1111 pos
= find_next_newline (pos
- 1, -1, &pos_byte
);
1112 if (n
>= MAX_PARAGRAPH_SEARCH
)
1113 pos_byte
= BEGV_BYTE
;
1117 /* On a 3.4 GHz machine, searching forward for a strong directional
1118 character in a long paragraph full of weaks or neutrals takes about
1119 1 ms for each 20K characters. The number below limits each call to
1120 bidi_paragraph_init to less than 10 ms even on slow machines. */
1121 #define MAX_STRONG_CHAR_SEARCH 100000
1123 /* Determine the base direction, a.k.a. base embedding level, of the
1124 paragraph we are about to iterate through. If DIR is either L2R or
1125 R2L, just use that. Otherwise, determine the paragraph direction
1126 from the first strong directional character of the paragraph.
1128 NO_DEFAULT_P means don't default to L2R if the paragraph
1129 has no strong directional characters and both DIR and
1130 bidi_it->paragraph_dir are NEUTRAL_DIR. In that case, search back
1131 in the buffer until a paragraph is found with a strong character,
1132 or until hitting BEGV. In the latter case, fall back to L2R. This
1133 flag is used in current-bidi-paragraph-direction.
1135 Note that this function gives the paragraph separator the same
1136 direction as the preceding paragraph, even though Emacs generally
1137 views the separator as not belonging to any paragraph. */
1139 bidi_paragraph_init (bidi_dir_t dir
, struct bidi_it
*bidi_it
, bool no_default_p
)
1141 ptrdiff_t bytepos
= bidi_it
->bytepos
;
1142 bool string_p
= bidi_it
->string
.s
|| STRINGP (bidi_it
->string
.lstring
);
1143 ptrdiff_t pstartbyte
;
1144 /* Note that begbyte is a byte position, while end is a character
1145 position. Yes, this is ugly, but we are trying to avoid costly
1146 calls to BYTE_TO_CHAR and its ilk. */
1147 ptrdiff_t begbyte
= string_p
? 0 : BEGV_BYTE
;
1148 ptrdiff_t end
= string_p
? bidi_it
->string
.schars
: ZV
;
1150 /* Special case for an empty buffer. */
1151 if (bytepos
== begbyte
&& bidi_it
->charpos
== end
)
1153 /* We should never be called at EOB or before BEGV. */
1154 else if (bidi_it
->charpos
>= end
|| bytepos
< begbyte
)
1159 bidi_it
->paragraph_dir
= L2R
;
1160 bidi_it
->new_paragraph
= 0;
1162 else if (dir
== R2L
)
1164 bidi_it
->paragraph_dir
= R2L
;
1165 bidi_it
->new_paragraph
= 0;
1167 else if (dir
== NEUTRAL_DIR
) /* P2 */
1170 ptrdiff_t ch_len
, nchars
;
1171 ptrdiff_t pos
, disp_pos
= -1;
1174 const unsigned char *s
;
1176 if (!bidi_initialized
)
1179 /* If we are inside a paragraph separator, we are just waiting
1180 for the separator to be exhausted; use the previous paragraph
1181 direction. But don't do that if we have been just reseated,
1182 because we need to reinitialize below in that case. */
1183 if (!bidi_it
->first_elt
1184 && bidi_it
->charpos
< bidi_it
->separator_limit
)
1187 /* If we are on a newline, get past it to where the next
1188 paragraph might start. But don't do that at BEGV since then
1189 we are potentially in a new paragraph that doesn't yet
1191 pos
= bidi_it
->charpos
;
1192 s
= (STRINGP (bidi_it
->string
.lstring
)
1193 ? SDATA (bidi_it
->string
.lstring
)
1194 : bidi_it
->string
.s
);
1195 if (bytepos
> begbyte
1196 && bidi_char_at_pos (bytepos
, s
, bidi_it
->string
.unibyte
) == '\n')
1202 /* We are either at the beginning of a paragraph or in the
1203 middle of it. Find where this paragraph starts. */
1206 /* We don't support changes of paragraph direction inside a
1207 string. It is treated as a single paragraph. */
1211 pstartbyte
= bidi_find_paragraph_start (pos
, bytepos
);
1212 bidi_it
->separator_limit
= -1;
1213 bidi_it
->new_paragraph
= 0;
1215 /* The following loop is run more than once only if NO_DEFAULT_P,
1216 and only if we are iterating on a buffer. */
1220 bytepos
= pstartbyte
;
1222 pos
= BYTE_TO_CHAR (bytepos
);
1223 ch
= bidi_fetch_char (bytepos
, pos
, &disp_pos
, &disp_prop
,
1225 bidi_it
->frame_window_p
, &ch_len
, &nchars
);
1226 type
= bidi_get_type (ch
, NEUTRAL_DIR
);
1229 for (pos
+= nchars
, bytepos
+= ch_len
;
1230 ((bidi_get_category (type
) != STRONG
)
1231 || (bidi_ignore_explicit_marks_for_paragraph_level
1232 && (type
== RLE
|| type
== RLO
1233 || type
== LRE
|| type
== LRO
)))
1234 /* Stop when searched too far into an abnormally large
1235 paragraph full of weak or neutral characters. */
1236 && pos
- pos1
< MAX_STRONG_CHAR_SEARCH
;
1237 type
= bidi_get_type (ch
, NEUTRAL_DIR
))
1241 /* Pretend there's a paragraph separator at end of
1247 && type
== NEUTRAL_B
1248 && bidi_at_paragraph_end (pos
, bytepos
) >= -1)
1250 /* Fetch next character and advance to get past it. */
1251 ch
= bidi_fetch_char (bytepos
, pos
, &disp_pos
,
1252 &disp_prop
, &bidi_it
->string
,
1253 bidi_it
->frame_window_p
, &ch_len
, &nchars
);
1257 if ((type
== STRONG_R
|| type
== STRONG_AL
) /* P3 */
1258 || (!bidi_ignore_explicit_marks_for_paragraph_level
1259 && (type
== RLO
|| type
== RLE
)))
1260 bidi_it
->paragraph_dir
= R2L
;
1261 else if (type
== STRONG_L
1262 || (!bidi_ignore_explicit_marks_for_paragraph_level
1263 && (type
== LRO
|| type
== LRE
)))
1264 bidi_it
->paragraph_dir
= L2R
;
1266 && no_default_p
&& bidi_it
->paragraph_dir
== NEUTRAL_DIR
)
1268 /* If this paragraph is at BEGV, default to L2R. */
1269 if (pstartbyte
== BEGV_BYTE
)
1270 bidi_it
->paragraph_dir
= L2R
; /* P3 and HL1 */
1273 ptrdiff_t prevpbyte
= pstartbyte
;
1274 ptrdiff_t p
= BYTE_TO_CHAR (pstartbyte
), pbyte
= pstartbyte
;
1276 /* Find the beginning of the previous paragraph, if any. */
1277 while (pbyte
> BEGV_BYTE
&& prevpbyte
>= pstartbyte
)
1279 /* FXIME: What if p is covered by a display
1280 string? See also a FIXME inside
1281 bidi_find_paragraph_start. */
1283 pbyte
= CHAR_TO_BYTE (p
);
1284 prevpbyte
= bidi_find_paragraph_start (p
, pbyte
);
1286 pstartbyte
= prevpbyte
;
1290 && no_default_p
&& bidi_it
->paragraph_dir
== NEUTRAL_DIR
);
1295 /* Contrary to UAX#9 clause P3, we only default the paragraph
1296 direction to L2R if we have no previous usable paragraph
1297 direction. This is allowed by the HL1 clause. */
1298 if (bidi_it
->paragraph_dir
!= L2R
&& bidi_it
->paragraph_dir
!= R2L
)
1299 bidi_it
->paragraph_dir
= L2R
; /* P3 and HL1 ``higher-level protocols'' */
1300 if (bidi_it
->paragraph_dir
== R2L
)
1301 bidi_it
->level_stack
[0].level
= 1;
1303 bidi_it
->level_stack
[0].level
= 0;
1305 bidi_line_init (bidi_it
);
1309 /***********************************************************************
1310 Resolving explicit and implicit levels.
1311 The rest of this file constitutes the core of the UBA implementation.
1312 ***********************************************************************/
1315 bidi_explicit_dir_char (int ch
)
1317 bidi_type_t ch_type
;
1319 if (!bidi_initialized
)
1321 ch_type
= (bidi_type_t
) XINT (CHAR_TABLE_REF (bidi_type_table
, ch
));
1322 return (ch_type
== LRE
|| ch_type
== LRO
1323 || ch_type
== RLE
|| ch_type
== RLO
1327 /* A helper function for bidi_resolve_explicit. It advances to the
1328 next character in logical order and determines the new embedding
1329 level and directional override, but does not take into account
1330 empty embeddings. */
1332 bidi_resolve_explicit_1 (struct bidi_it
*bidi_it
)
1338 bidi_dir_t override
;
1339 bool string_p
= bidi_it
->string
.s
|| STRINGP (bidi_it
->string
.lstring
);
1341 /* If reseat()'ed, don't advance, so as to start iteration from the
1342 position where we were reseated. bidi_it->bytepos can be less
1343 than BEGV_BYTE after reseat to BEGV. */
1344 if (bidi_it
->bytepos
< (string_p
? 0 : BEGV_BYTE
)
1345 || bidi_it
->first_elt
)
1347 bidi_it
->first_elt
= 0;
1350 const unsigned char *p
1351 = (STRINGP (bidi_it
->string
.lstring
)
1352 ? SDATA (bidi_it
->string
.lstring
)
1353 : bidi_it
->string
.s
);
1355 if (bidi_it
->charpos
< 0)
1356 bidi_it
->charpos
= bidi_it
->bytepos
= 0;
1357 eassert (bidi_it
->bytepos
== bidi_count_bytes (p
, 0, 0,
1359 bidi_it
->string
.unibyte
));
1363 if (bidi_it
->charpos
< BEGV
)
1365 bidi_it
->charpos
= BEGV
;
1366 bidi_it
->bytepos
= BEGV_BYTE
;
1368 eassert (bidi_it
->bytepos
== CHAR_TO_BYTE (bidi_it
->charpos
));
1371 /* Don't move at end of buffer/string. */
1372 else if (bidi_it
->charpos
< (string_p
? bidi_it
->string
.schars
: ZV
))
1374 /* Advance to the next character, skipping characters covered by
1375 display strings (nchars > 1). */
1376 if (bidi_it
->nchars
<= 0)
1378 bidi_it
->charpos
+= bidi_it
->nchars
;
1379 if (bidi_it
->ch_len
== 0)
1381 bidi_it
->bytepos
+= bidi_it
->ch_len
;
1384 current_level
= bidi_it
->level_stack
[bidi_it
->stack_idx
].level
; /* X1 */
1385 override
= bidi_it
->level_stack
[bidi_it
->stack_idx
].override
;
1386 new_level
= current_level
;
1388 if (bidi_it
->charpos
>= (string_p
? bidi_it
->string
.schars
: ZV
))
1391 bidi_it
->ch_len
= 1;
1392 bidi_it
->nchars
= 1;
1393 bidi_it
->disp_pos
= (string_p
? bidi_it
->string
.schars
: ZV
);
1394 bidi_it
->disp_prop
= 0;
1398 /* Fetch the character at BYTEPOS. If it is covered by a
1399 display string, treat the entire run of covered characters as
1400 a single character u+FFFC. */
1401 curchar
= bidi_fetch_char (bidi_it
->bytepos
, bidi_it
->charpos
,
1402 &bidi_it
->disp_pos
, &bidi_it
->disp_prop
,
1403 &bidi_it
->string
, bidi_it
->frame_window_p
,
1404 &bidi_it
->ch_len
, &bidi_it
->nchars
);
1406 bidi_it
->ch
= curchar
;
1408 /* Don't apply directional override here, as all the types we handle
1409 below will not be affected by the override anyway, and we need
1410 the original type unaltered. The override will be applied in
1411 bidi_resolve_weak. */
1412 type
= bidi_get_type (curchar
, NEUTRAL_DIR
);
1413 bidi_it
->orig_type
= type
;
1414 bidi_check_type (bidi_it
->orig_type
);
1417 bidi_it
->prev_was_pdf
= 0;
1419 bidi_it
->type_after_w1
= UNKNOWN_BT
;
1425 bidi_it
->type_after_w1
= type
;
1426 bidi_check_type (bidi_it
->type_after_w1
);
1427 type
= WEAK_BN
; /* X9/Retaining */
1428 if (bidi_it
->ignore_bn_limit
<= -1)
1430 if (current_level
<= BIDI_MAXLEVEL
- 4)
1432 /* Compute the least odd embedding level greater than
1433 the current level. */
1434 new_level
= ((current_level
+ 1) & ~1) + 1;
1435 if (bidi_it
->type_after_w1
== RLE
)
1436 override
= NEUTRAL_DIR
;
1439 if (current_level
== BIDI_MAXLEVEL
- 4)
1440 bidi_it
->invalid_rl_levels
= 0;
1441 bidi_push_embedding_level (bidi_it
, new_level
, override
);
1445 bidi_it
->invalid_levels
++;
1446 /* See the commentary about invalid_rl_levels below. */
1447 if (bidi_it
->invalid_rl_levels
< 0)
1448 bidi_it
->invalid_rl_levels
= 0;
1449 bidi_it
->invalid_rl_levels
++;
1452 else if (bidi_it
->prev
.type_after_w1
== WEAK_EN
/* W5/Retaining */
1453 || (bidi_it
->next_en_pos
> bidi_it
->charpos
1454 && bidi_it
->next_en_type
== WEAK_EN
))
1459 bidi_it
->type_after_w1
= type
;
1460 bidi_check_type (bidi_it
->type_after_w1
);
1461 type
= WEAK_BN
; /* X9/Retaining */
1462 if (bidi_it
->ignore_bn_limit
<= -1)
1464 if (current_level
<= BIDI_MAXLEVEL
- 5)
1466 /* Compute the least even embedding level greater than
1467 the current level. */
1468 new_level
= ((current_level
+ 2) & ~1);
1469 if (bidi_it
->type_after_w1
== LRE
)
1470 override
= NEUTRAL_DIR
;
1473 bidi_push_embedding_level (bidi_it
, new_level
, override
);
1477 bidi_it
->invalid_levels
++;
1478 /* invalid_rl_levels counts invalid levels encountered
1479 while the embedding level was already too high for
1480 LRE/LRO, but not for RLE/RLO. That is because
1481 there may be exactly one PDF which we should not
1482 ignore even though invalid_levels is non-zero.
1483 invalid_rl_levels helps to know what PDF is
1485 if (bidi_it
->invalid_rl_levels
>= 0)
1486 bidi_it
->invalid_rl_levels
++;
1489 else if (bidi_it
->prev
.type_after_w1
== WEAK_EN
/* W5/Retaining */
1490 || (bidi_it
->next_en_pos
> bidi_it
->charpos
1491 && bidi_it
->next_en_type
== WEAK_EN
))
1495 bidi_it
->type_after_w1
= type
;
1496 bidi_check_type (bidi_it
->type_after_w1
);
1497 type
= WEAK_BN
; /* X9/Retaining */
1498 if (bidi_it
->ignore_bn_limit
<= -1)
1500 if (!bidi_it
->invalid_rl_levels
)
1502 new_level
= bidi_pop_embedding_level (bidi_it
);
1503 bidi_it
->invalid_rl_levels
= -1;
1504 if (bidi_it
->invalid_levels
)
1505 bidi_it
->invalid_levels
--;
1506 /* else nothing: UAX#9 says to ignore invalid PDFs */
1508 if (!bidi_it
->invalid_levels
)
1509 new_level
= bidi_pop_embedding_level (bidi_it
);
1512 bidi_it
->invalid_levels
--;
1513 bidi_it
->invalid_rl_levels
--;
1516 else if (bidi_it
->prev
.type_after_w1
== WEAK_EN
/* W5/Retaining */
1517 || (bidi_it
->next_en_pos
> bidi_it
->charpos
1518 && bidi_it
->next_en_type
== WEAK_EN
))
1526 bidi_it
->type
= type
;
1527 bidi_check_type (bidi_it
->type
);
1532 /* Given an iterator state in BIDI_IT, advance one character position
1533 in the buffer/string to the next character (in the logical order),
1534 resolve any explicit embeddings and directional overrides, and
1535 return the embedding level of the character after resolving
1536 explicit directives and ignoring empty embeddings. */
1538 bidi_resolve_explicit (struct bidi_it
*bidi_it
)
1540 int prev_level
= bidi_it
->level_stack
[bidi_it
->stack_idx
].level
;
1541 int new_level
= bidi_resolve_explicit_1 (bidi_it
);
1542 ptrdiff_t eob
= bidi_it
->string
.s
? bidi_it
->string
.schars
: ZV
;
1543 const unsigned char *s
1544 = (STRINGP (bidi_it
->string
.lstring
)
1545 ? SDATA (bidi_it
->string
.lstring
)
1546 : bidi_it
->string
.s
);
1548 if (prev_level
< new_level
1549 && bidi_it
->type
== WEAK_BN
1550 && bidi_it
->ignore_bn_limit
== -1 /* only if not already known */
1551 && bidi_it
->charpos
< eob
/* not already at EOB */
1552 && bidi_explicit_dir_char (bidi_char_at_pos (bidi_it
->bytepos
1553 + bidi_it
->ch_len
, s
,
1554 bidi_it
->string
.unibyte
)))
1556 /* Avoid pushing and popping embedding levels if the level run
1557 is empty, as this breaks level runs where it shouldn't.
1558 UAX#9 removes all the explicit embedding and override codes,
1559 so empty embeddings disappear without a trace. We need to
1560 behave as if we did the same. */
1561 struct bidi_it saved_it
;
1562 int level
= prev_level
;
1564 bidi_copy_it (&saved_it
, bidi_it
);
1566 while (bidi_explicit_dir_char (bidi_char_at_pos (bidi_it
->bytepos
1567 + bidi_it
->ch_len
, s
,
1568 bidi_it
->string
.unibyte
)))
1570 /* This advances to the next character, skipping any
1571 characters covered by display strings. */
1572 level
= bidi_resolve_explicit_1 (bidi_it
);
1573 /* If string.lstring was relocated inside bidi_resolve_explicit_1,
1574 a pointer to its data is no longer valid. */
1575 if (STRINGP (bidi_it
->string
.lstring
))
1576 s
= SDATA (bidi_it
->string
.lstring
);
1579 if (bidi_it
->nchars
<= 0)
1581 if (level
== prev_level
) /* empty embedding */
1582 saved_it
.ignore_bn_limit
= bidi_it
->charpos
+ bidi_it
->nchars
;
1583 else /* this embedding is non-empty */
1584 saved_it
.ignore_bn_limit
= -2;
1586 bidi_copy_it (bidi_it
, &saved_it
);
1587 if (bidi_it
->ignore_bn_limit
> -1)
1589 /* We pushed a level, but we shouldn't have. Undo that. */
1590 if (!bidi_it
->invalid_rl_levels
)
1592 new_level
= bidi_pop_embedding_level (bidi_it
);
1593 bidi_it
->invalid_rl_levels
= -1;
1594 if (bidi_it
->invalid_levels
)
1595 bidi_it
->invalid_levels
--;
1597 if (!bidi_it
->invalid_levels
)
1598 new_level
= bidi_pop_embedding_level (bidi_it
);
1601 bidi_it
->invalid_levels
--;
1602 bidi_it
->invalid_rl_levels
--;
1607 if (bidi_it
->type
== NEUTRAL_B
) /* X8 */
1609 bidi_set_paragraph_end (bidi_it
);
1610 /* This is needed by bidi_resolve_weak below, and in L1. */
1611 bidi_it
->type_after_w1
= bidi_it
->type
;
1612 bidi_check_type (bidi_it
->type_after_w1
);
1618 /* Advance in the buffer/string, resolve weak types and return the
1619 type of the next character after weak type resolution. */
1621 bidi_resolve_weak (struct bidi_it
*bidi_it
)
1624 bidi_dir_t override
;
1625 int prev_level
= bidi_it
->level_stack
[bidi_it
->stack_idx
].level
;
1626 int new_level
= bidi_resolve_explicit (bidi_it
);
1628 bidi_type_t type_of_next
;
1629 struct bidi_it saved_it
;
1631 = ((STRINGP (bidi_it
->string
.lstring
) || bidi_it
->string
.s
)
1632 ? bidi_it
->string
.schars
: ZV
);
1634 type
= bidi_it
->type
;
1635 override
= bidi_it
->level_stack
[bidi_it
->stack_idx
].override
;
1637 if (type
== UNKNOWN_BT
1645 if (new_level
!= prev_level
1646 || bidi_it
->type
== NEUTRAL_B
)
1648 /* We've got a new embedding level run, compute the directional
1649 type of sor and initialize per-run variables (UAX#9, clause
1651 bidi_set_sor_type (bidi_it
, prev_level
, new_level
);
1653 else if (type
== NEUTRAL_S
|| type
== NEUTRAL_WS
1654 || type
== WEAK_BN
|| type
== STRONG_AL
)
1655 bidi_it
->type_after_w1
= type
; /* needed in L1 */
1656 bidi_check_type (bidi_it
->type_after_w1
);
1658 /* Level and directional override status are already recorded in
1659 bidi_it, and do not need any change; see X6. */
1660 if (override
== R2L
) /* X6 */
1662 else if (override
== L2R
)
1666 if (type
== WEAK_NSM
) /* W1 */
1668 /* Note that we don't need to consider the case where the
1669 prev character has its type overridden by an RLO or LRO,
1670 because then either the type of this NSM would have been
1671 also overridden, or the previous character is outside the
1672 current level run, and thus not relevant to this NSM.
1673 This is why NSM gets the type_after_w1 of the previous
1675 if (bidi_it
->prev
.type_after_w1
!= UNKNOWN_BT
1676 /* if type_after_w1 is NEUTRAL_B, this NSM is at sor */
1677 && bidi_it
->prev
.type_after_w1
!= NEUTRAL_B
)
1678 type
= bidi_it
->prev
.type_after_w1
;
1679 else if (bidi_it
->sor
== R2L
)
1681 else if (bidi_it
->sor
== L2R
)
1683 else /* shouldn't happen! */
1686 if (type
== WEAK_EN
/* W2 */
1687 && bidi_it
->last_strong
.type_after_w1
== STRONG_AL
)
1689 else if (type
== STRONG_AL
) /* W3 */
1691 else if ((type
== WEAK_ES
/* W4 */
1692 && bidi_it
->prev
.type_after_w1
== WEAK_EN
1693 && bidi_it
->prev
.orig_type
== WEAK_EN
)
1695 && ((bidi_it
->prev
.type_after_w1
== WEAK_EN
1696 && bidi_it
->prev
.orig_type
== WEAK_EN
)
1697 || bidi_it
->prev
.type_after_w1
== WEAK_AN
)))
1699 const unsigned char *s
1700 = (STRINGP (bidi_it
->string
.lstring
)
1701 ? SDATA (bidi_it
->string
.lstring
)
1702 : bidi_it
->string
.s
);
1704 next_char
= (bidi_it
->charpos
+ bidi_it
->nchars
>= eob
1706 : bidi_char_at_pos (bidi_it
->bytepos
+ bidi_it
->ch_len
,
1707 s
, bidi_it
->string
.unibyte
));
1708 type_of_next
= bidi_get_type (next_char
, override
);
1710 if (type_of_next
== WEAK_BN
1711 || bidi_explicit_dir_char (next_char
))
1713 bidi_copy_it (&saved_it
, bidi_it
);
1714 while (bidi_resolve_explicit (bidi_it
) == new_level
1715 && bidi_it
->type
== WEAK_BN
)
1717 type_of_next
= bidi_it
->type
;
1718 bidi_copy_it (bidi_it
, &saved_it
);
1721 /* If the next character is EN, but the last strong-type
1722 character is AL, that next EN will be changed to AN when
1723 we process it in W2 above. So in that case, this ES
1724 should not be changed into EN. */
1726 && type_of_next
== WEAK_EN
1727 && bidi_it
->last_strong
.type_after_w1
!= STRONG_AL
)
1729 else if (type
== WEAK_CS
)
1731 if (bidi_it
->prev
.type_after_w1
== WEAK_AN
1732 && (type_of_next
== WEAK_AN
1733 /* If the next character is EN, but the last
1734 strong-type character is AL, EN will be later
1735 changed to AN when we process it in W2 above.
1736 So in that case, this ES should not be
1738 || (type_of_next
== WEAK_EN
1739 && bidi_it
->last_strong
.type_after_w1
== STRONG_AL
)))
1741 else if (bidi_it
->prev
.type_after_w1
== WEAK_EN
1742 && type_of_next
== WEAK_EN
1743 && bidi_it
->last_strong
.type_after_w1
!= STRONG_AL
)
1747 else if (type
== WEAK_ET
/* W5: ET with EN before or after it */
1748 || type
== WEAK_BN
) /* W5/Retaining */
1750 if (bidi_it
->prev
.type_after_w1
== WEAK_EN
) /* ET/BN w/EN before it */
1752 else if (bidi_it
->next_en_pos
> bidi_it
->charpos
1753 && bidi_it
->next_en_type
!= WEAK_BN
)
1755 if (bidi_it
->next_en_type
== WEAK_EN
) /* ET/BN with EN after it */
1758 else if (bidi_it
->next_en_pos
>=0)
1760 ptrdiff_t en_pos
= bidi_it
->charpos
+ bidi_it
->nchars
;
1761 const unsigned char *s
= (STRINGP (bidi_it
->string
.lstring
)
1762 ? SDATA (bidi_it
->string
.lstring
)
1763 : bidi_it
->string
.s
);
1765 if (bidi_it
->nchars
<= 0)
1768 = (bidi_it
->charpos
+ bidi_it
->nchars
>= eob
1770 : bidi_char_at_pos (bidi_it
->bytepos
+ bidi_it
->ch_len
, s
,
1771 bidi_it
->string
.unibyte
));
1772 type_of_next
= bidi_get_type (next_char
, override
);
1774 if (type_of_next
== WEAK_ET
1775 || type_of_next
== WEAK_BN
1776 || bidi_explicit_dir_char (next_char
))
1778 bidi_copy_it (&saved_it
, bidi_it
);
1779 while (bidi_resolve_explicit (bidi_it
) == new_level
1780 && (bidi_it
->type
== WEAK_BN
1781 || bidi_it
->type
== WEAK_ET
))
1783 type_of_next
= bidi_it
->type
;
1784 en_pos
= bidi_it
->charpos
;
1785 bidi_copy_it (bidi_it
, &saved_it
);
1787 /* Remember this position, to speed up processing of the
1789 bidi_it
->next_en_pos
= en_pos
;
1790 if (type_of_next
== WEAK_EN
)
1792 /* If the last strong character is AL, the EN we've
1793 found will become AN when we get to it (W2). */
1794 if (bidi_it
->last_strong
.type_after_w1
== STRONG_AL
)
1795 type_of_next
= WEAK_AN
;
1796 else if (type
== WEAK_BN
)
1797 type
= NEUTRAL_ON
; /* W6/Retaining */
1801 else if (type_of_next
== NEUTRAL_B
)
1802 /* Record the fact that there are no more ENs from
1803 here to the end of paragraph, to avoid entering the
1804 loop above ever again in this paragraph. */
1805 bidi_it
->next_en_pos
= -1;
1806 /* Record the type of the character where we ended our search. */
1807 bidi_it
->next_en_type
= type_of_next
;
1812 if (type
== WEAK_ES
|| type
== WEAK_ET
|| type
== WEAK_CS
/* W6 */
1814 && (bidi_it
->prev
.type_after_w1
== WEAK_CS
/* W6/Retaining */
1815 || bidi_it
->prev
.type_after_w1
== WEAK_ES
1816 || bidi_it
->prev
.type_after_w1
== WEAK_ET
)))
1819 /* Store the type we've got so far, before we clobber it with strong
1820 types in W7 and while resolving neutral types. But leave alone
1821 the original types that were recorded above, because we will need
1822 them for the L1 clause. */
1823 if (bidi_it
->type_after_w1
== UNKNOWN_BT
)
1824 bidi_it
->type_after_w1
= type
;
1825 bidi_check_type (bidi_it
->type_after_w1
);
1827 if (type
== WEAK_EN
) /* W7 */
1829 if ((bidi_it
->last_strong
.type_after_w1
== STRONG_L
)
1830 || (bidi_it
->last_strong
.type
== UNKNOWN_BT
&& bidi_it
->sor
== L2R
))
1834 bidi_it
->type
= type
;
1835 bidi_check_type (bidi_it
->type
);
1839 /* Resolve the type of a neutral character according to the type of
1840 surrounding strong text and the current embedding level. */
1842 bidi_resolve_neutral_1 (bidi_type_t prev_type
, bidi_type_t next_type
, int lev
)
1844 /* N1: European and Arabic numbers are treated as though they were R. */
1845 if (next_type
== WEAK_EN
|| next_type
== WEAK_AN
)
1846 next_type
= STRONG_R
;
1847 if (prev_type
== WEAK_EN
|| prev_type
== WEAK_AN
)
1848 prev_type
= STRONG_R
;
1850 if (next_type
== prev_type
) /* N1 */
1852 else if ((lev
& 1) == 0) /* N2 */
1859 bidi_resolve_neutral (struct bidi_it
*bidi_it
)
1861 int prev_level
= bidi_it
->level_stack
[bidi_it
->stack_idx
].level
;
1862 bidi_type_t type
= bidi_resolve_weak (bidi_it
);
1863 int current_level
= bidi_it
->level_stack
[bidi_it
->stack_idx
].level
;
1865 if (!(type
== STRONG_R
1870 || type
== NEUTRAL_B
1871 || type
== NEUTRAL_S
1872 || type
== NEUTRAL_WS
1873 || type
== NEUTRAL_ON
))
1876 if ((type
!= NEUTRAL_B
/* Don't risk entering the long loop below if
1877 we are already at paragraph end. */
1878 && bidi_get_category (type
) == NEUTRAL
)
1879 || (type
== WEAK_BN
&& prev_level
== current_level
))
1881 if (bidi_it
->next_for_neutral
.type
!= UNKNOWN_BT
)
1882 type
= bidi_resolve_neutral_1 (bidi_it
->prev_for_neutral
.type
,
1883 bidi_it
->next_for_neutral
.type
,
1885 /* The next two "else if" clauses are shortcuts for the
1886 important special case when we have a long sequence of
1887 neutral or WEAK_BN characters, such as whitespace or nulls or
1888 other control characters, on the base embedding level of the
1889 paragraph, and that sequence goes all the way to the end of
1890 the paragraph and follows a character whose resolved
1891 directionality is identical to the base embedding level.
1892 (This is what happens in a buffer with plain L2R text that
1893 happens to include long sequences of control characters.) By
1894 virtue of N1, the result of examining this long sequence will
1895 always be either STRONG_L or STRONG_R, depending on the base
1896 embedding level. So we use this fact directly instead of
1897 entering the expensive loop in the "else" clause. */
1898 else if (current_level
== 0
1899 && bidi_it
->prev_for_neutral
.type
== STRONG_L
1900 && !bidi_explicit_dir_char (bidi_it
->ch
))
1901 type
= bidi_resolve_neutral_1 (bidi_it
->prev_for_neutral
.type
,
1902 STRONG_L
, current_level
);
1903 else if (/* current level is 1 */
1905 /* base embedding level is also 1 */
1906 && bidi_it
->level_stack
[0].level
== 1
1907 /* previous character is one of those considered R for
1908 the purposes of W5 */
1909 && (bidi_it
->prev_for_neutral
.type
== STRONG_R
1910 || bidi_it
->prev_for_neutral
.type
== WEAK_EN
1911 || bidi_it
->prev_for_neutral
.type
== WEAK_AN
)
1912 && !bidi_explicit_dir_char (bidi_it
->ch
))
1913 type
= bidi_resolve_neutral_1 (bidi_it
->prev_for_neutral
.type
,
1914 STRONG_R
, current_level
);
1917 /* Arrrgh!! The UAX#9 algorithm is too deeply entrenched in
1918 the assumption of batch-style processing; see clauses W4,
1919 W5, and especially N1, which require to look far forward
1920 (as well as back) in the buffer/string. May the fleas of
1921 a thousand camels infest the armpits of those who design
1922 supposedly general-purpose algorithms by looking at their
1923 own implementations, and fail to consider other possible
1925 struct bidi_it saved_it
;
1926 bidi_type_t next_type
;
1928 if (bidi_it
->scan_dir
== -1)
1931 bidi_copy_it (&saved_it
, bidi_it
);
1932 /* Scan the text forward until we find the first non-neutral
1933 character, and then use that to resolve the neutral we
1934 are dealing with now. We also cache the scanned iterator
1935 states, to salvage some of the effort later. */
1936 bidi_cache_iterator_state (bidi_it
, 0);
1938 /* Record the info about the previous character, so that
1939 it will be cached below with this state. */
1940 if (bidi_it
->type_after_w1
!= WEAK_BN
/* W1/Retaining */
1941 && bidi_it
->type
!= WEAK_BN
)
1942 bidi_remember_char (&bidi_it
->prev
, bidi_it
);
1943 type
= bidi_resolve_weak (bidi_it
);
1944 /* Paragraph separators have their levels fully resolved
1945 at this point, so cache them as resolved. */
1946 bidi_cache_iterator_state (bidi_it
, type
== NEUTRAL_B
);
1947 /* FIXME: implement L1 here, by testing for a newline and
1948 resetting the level for any sequence of whitespace
1949 characters adjacent to it. */
1950 } while (!(type
== NEUTRAL_B
1952 && bidi_get_category (type
) != NEUTRAL
)
1953 /* This is all per level run, so stop when we
1954 reach the end of this level run. */
1955 || (bidi_it
->level_stack
[bidi_it
->stack_idx
].level
1956 != current_level
)));
1958 bidi_remember_char (&saved_it
.next_for_neutral
, bidi_it
);
1965 /* Actually, STRONG_AL cannot happen here, because
1966 bidi_resolve_weak converts it to STRONG_R, per W3. */
1967 eassert (type
!= STRONG_AL
);
1972 /* N1: ``European and Arabic numbers are treated as
1973 though they were R.'' */
1974 next_type
= STRONG_R
;
1977 case NEUTRAL_ON
: /* W6/Retaining */
1978 if (!bidi_explicit_dir_char (bidi_it
->ch
))
1979 emacs_abort (); /* can't happen: BNs are skipped */
1982 /* Marched all the way to the end of this level run.
1983 We need to use the eor type, whose information is
1984 stored by bidi_set_sor_type in the prev_for_neutral
1986 if (saved_it
.type
!= WEAK_BN
1987 || bidi_get_category (bidi_it
->prev
.type_after_w1
) == NEUTRAL
)
1988 next_type
= bidi_it
->prev_for_neutral
.type
;
1991 /* This is a BN which does not adjoin neutrals.
1992 Leave its type alone. */
1993 bidi_copy_it (bidi_it
, &saved_it
);
1994 return bidi_it
->type
;
2000 type
= bidi_resolve_neutral_1 (saved_it
.prev_for_neutral
.type
,
2001 next_type
, current_level
);
2002 saved_it
.next_for_neutral
.type
= next_type
;
2003 saved_it
.type
= type
;
2004 bidi_check_type (next_type
);
2005 bidi_check_type (type
);
2006 bidi_copy_it (bidi_it
, &saved_it
);
2012 /* Given an iterator state in BIDI_IT, advance one character position
2013 in the buffer/string to the next character (in the logical order),
2014 resolve the bidi type of that next character, and return that
2017 bidi_type_of_next_char (struct bidi_it
*bidi_it
)
2021 /* This should always be called during a forward scan. */
2022 if (bidi_it
->scan_dir
!= 1)
2025 /* Reset the limit until which to ignore BNs if we step out of the
2026 area where we found only empty levels. */
2027 if ((bidi_it
->ignore_bn_limit
> -1
2028 && bidi_it
->ignore_bn_limit
<= bidi_it
->charpos
)
2029 || (bidi_it
->ignore_bn_limit
== -2
2030 && !bidi_explicit_dir_char (bidi_it
->ch
)))
2031 bidi_it
->ignore_bn_limit
= -1;
2033 type
= bidi_resolve_neutral (bidi_it
);
2038 /* Given an iterator state BIDI_IT, advance one character position in
2039 the buffer/string to the next character (in the current scan
2040 direction), resolve the embedding and implicit levels of that next
2041 character, and return the resulting level. */
2043 bidi_level_of_next_char (struct bidi_it
*bidi_it
)
2046 int level
, prev_level
= -1;
2047 struct bidi_saved_info next_for_neutral
;
2048 ptrdiff_t next_char_pos
= -2;
2050 if (bidi_it
->scan_dir
== 1)
2053 = ((bidi_it
->string
.s
|| STRINGP (bidi_it
->string
.lstring
))
2054 ? bidi_it
->string
.schars
: ZV
);
2056 /* There's no sense in trying to advance if we hit end of text. */
2057 if (bidi_it
->charpos
>= eob
)
2058 return bidi_it
->resolved_level
;
2060 /* Record the info about the previous character. */
2061 if (bidi_it
->type_after_w1
!= WEAK_BN
/* W1/Retaining */
2062 && bidi_it
->type
!= WEAK_BN
)
2063 bidi_remember_char (&bidi_it
->prev
, bidi_it
);
2064 if (bidi_it
->type_after_w1
== STRONG_R
2065 || bidi_it
->type_after_w1
== STRONG_L
2066 || bidi_it
->type_after_w1
== STRONG_AL
)
2067 bidi_remember_char (&bidi_it
->last_strong
, bidi_it
);
2068 /* FIXME: it sounds like we don't need both prev and
2069 prev_for_neutral members, but I'm leaving them both for now. */
2070 if (bidi_it
->type
== STRONG_R
|| bidi_it
->type
== STRONG_L
2071 || bidi_it
->type
== WEAK_EN
|| bidi_it
->type
== WEAK_AN
)
2072 bidi_remember_char (&bidi_it
->prev_for_neutral
, bidi_it
);
2074 /* If we overstepped the characters used for resolving neutrals
2075 and whitespace, invalidate their info in the iterator. */
2076 if (bidi_it
->charpos
>= bidi_it
->next_for_neutral
.charpos
)
2077 bidi_it
->next_for_neutral
.type
= UNKNOWN_BT
;
2078 if (bidi_it
->next_en_pos
>= 0
2079 && bidi_it
->charpos
>= bidi_it
->next_en_pos
)
2081 bidi_it
->next_en_pos
= 0;
2082 bidi_it
->next_en_type
= UNKNOWN_BT
;
2084 if (bidi_it
->next_for_ws
.type
!= UNKNOWN_BT
2085 && bidi_it
->charpos
>= bidi_it
->next_for_ws
.charpos
)
2086 bidi_it
->next_for_ws
.type
= UNKNOWN_BT
;
2088 /* This must be taken before we fill the iterator with the info
2089 about the next char. If we scan backwards, the iterator
2090 state must be already cached, so there's no need to know the
2091 embedding level of the previous character, since we will be
2092 returning to our caller shortly. */
2093 prev_level
= bidi_it
->level_stack
[bidi_it
->stack_idx
].level
;
2095 next_for_neutral
= bidi_it
->next_for_neutral
;
2097 /* Perhaps the character we want is already cached. If it is, the
2098 call to bidi_cache_find below will return a type other than
2100 if (bidi_cache_idx
> bidi_cache_start
&& !bidi_it
->first_elt
)
2102 int bob
= ((bidi_it
->string
.s
|| STRINGP (bidi_it
->string
.lstring
))
2104 if (bidi_it
->scan_dir
> 0)
2106 if (bidi_it
->nchars
<= 0)
2108 next_char_pos
= bidi_it
->charpos
+ bidi_it
->nchars
;
2110 else if (bidi_it
->charpos
>= bob
)
2111 /* Implementation note: we allow next_char_pos to be as low as
2112 0 for buffers or -1 for strings, and that is okay because
2113 that's the "position" of the sentinel iterator state we
2114 cached at the beginning of the iteration. */
2115 next_char_pos
= bidi_it
->charpos
- 1;
2116 if (next_char_pos
>= bob
- 1)
2117 type
= bidi_cache_find (next_char_pos
, -1, bidi_it
);
2123 if (type
!= UNKNOWN_BT
)
2125 /* Don't lose the information for resolving neutrals! The
2126 cached states could have been cached before their
2127 next_for_neutral member was computed. If we are on our way
2128 forward, we can simply take the info from the previous
2130 if (bidi_it
->scan_dir
== 1
2131 && bidi_it
->next_for_neutral
.type
== UNKNOWN_BT
)
2132 bidi_it
->next_for_neutral
= next_for_neutral
;
2134 /* If resolved_level is -1, it means this state was cached
2135 before it was completely resolved, so we cannot return
2137 if (bidi_it
->resolved_level
!= -1)
2138 return bidi_it
->resolved_level
;
2140 if (bidi_it
->scan_dir
== -1)
2141 /* If we are going backwards, the iterator state is already cached
2142 from previous scans, and should be fully resolved. */
2145 if (type
== UNKNOWN_BT
)
2146 type
= bidi_type_of_next_char (bidi_it
);
2148 if (type
== NEUTRAL_B
)
2149 return bidi_it
->resolved_level
;
2151 level
= bidi_it
->level_stack
[bidi_it
->stack_idx
].level
;
2152 if ((bidi_get_category (type
) == NEUTRAL
/* && type != NEUTRAL_B */)
2153 || (type
== WEAK_BN
&& prev_level
== level
))
2155 if (bidi_it
->next_for_neutral
.type
== UNKNOWN_BT
)
2158 /* If the cached state shows a neutral character, it was not
2159 resolved by bidi_resolve_neutral, so do it now. */
2160 type
= bidi_resolve_neutral_1 (bidi_it
->prev_for_neutral
.type
,
2161 bidi_it
->next_for_neutral
.type
,
2165 if (!(type
== STRONG_R
2169 || type
== WEAK_AN
))
2171 bidi_it
->type
= type
;
2172 bidi_check_type (bidi_it
->type
);
2174 /* For L1 below, we need to know, for each WS character, whether
2175 it belongs to a sequence of WS characters preceding a newline
2176 or a TAB or a paragraph separator. */
2177 if (bidi_it
->orig_type
== NEUTRAL_WS
2178 && bidi_it
->next_for_ws
.type
== UNKNOWN_BT
)
2181 ptrdiff_t clen
= bidi_it
->ch_len
;
2182 ptrdiff_t bpos
= bidi_it
->bytepos
;
2183 ptrdiff_t cpos
= bidi_it
->charpos
;
2184 ptrdiff_t disp_pos
= bidi_it
->disp_pos
;
2185 ptrdiff_t nc
= bidi_it
->nchars
;
2186 struct bidi_string_data bs
= bidi_it
->string
;
2188 bool fwp
= bidi_it
->frame_window_p
;
2189 int dpp
= bidi_it
->disp_prop
;
2191 if (bidi_it
->nchars
<= 0)
2194 ch
= bidi_fetch_char (bpos
+= clen
, cpos
+= nc
, &disp_pos
, &dpp
, &bs
,
2196 if (ch
== '\n' || ch
== BIDI_EOB
)
2199 chtype
= bidi_get_type (ch
, NEUTRAL_DIR
);
2200 } while (chtype
== NEUTRAL_WS
|| chtype
== WEAK_BN
2201 || bidi_explicit_dir_char (ch
)); /* L1/Retaining */
2202 bidi_it
->next_for_ws
.type
= chtype
;
2203 bidi_check_type (bidi_it
->next_for_ws
.type
);
2204 bidi_it
->next_for_ws
.charpos
= cpos
;
2205 bidi_it
->next_for_ws
.bytepos
= bpos
;
2208 /* Resolve implicit levels, with a twist: PDFs get the embedding
2209 level of the embedding they terminate. See below for the
2211 if (bidi_it
->orig_type
== PDF
2212 /* Don't do this if this formatting code didn't change the
2213 embedding level due to invalid or empty embeddings. */
2214 && prev_level
!= level
)
2216 /* Don't look in UAX#9 for the reason for this: it's our own
2217 private quirk. The reason is that we want the formatting
2218 codes to be delivered so that they bracket the text of their
2219 embedding. For example, given the text
2223 we want it to be displayed as
2231 which will result because we bump up the embedding level as
2232 soon as we see the RLO and pop it as soon as we see the PDF,
2233 so RLO itself has the same embedding level as "teST", and
2234 thus would be normally delivered last, just before the PDF.
2235 The switch below fiddles with the level of PDF so that this
2236 ugly side effect does not happen.
2238 (This is, of course, only important if the formatting codes
2239 are actually displayed, but Emacs does need to display them
2240 if the user wants to.) */
2243 else if (bidi_it
->orig_type
== NEUTRAL_B
/* L1 */
2244 || bidi_it
->orig_type
== NEUTRAL_S
2245 || bidi_it
->ch
== '\n' || bidi_it
->ch
== BIDI_EOB
2246 || (bidi_it
->orig_type
== NEUTRAL_WS
2247 && (bidi_it
->next_for_ws
.type
== NEUTRAL_B
2248 || bidi_it
->next_for_ws
.type
== NEUTRAL_S
)))
2249 level
= bidi_it
->level_stack
[0].level
;
2250 else if ((level
& 1) == 0) /* I1 */
2252 if (type
== STRONG_R
)
2254 else if (type
== WEAK_EN
|| type
== WEAK_AN
)
2259 if (type
== STRONG_L
|| type
== WEAK_EN
|| type
== WEAK_AN
)
2263 bidi_it
->resolved_level
= level
;
2267 /* Move to the other edge of a level given by LEVEL. If END_FLAG,
2268 we are at the end of a level, and we need to prepare to
2269 resume the scan of the lower level.
2271 If this level's other edge is cached, we simply jump to it, filling
2272 the iterator structure with the iterator state on the other edge.
2273 Otherwise, we walk the buffer or string until we come back to the
2274 same level as LEVEL.
2276 Note: we are not talking here about a ``level run'' in the UAX#9
2277 sense of the term, but rather about a ``level'' which includes
2278 all the levels higher than it. In other words, given the levels
2281 11111112222222333333334443343222222111111112223322111
2284 and assuming we are at point A scanning left to right, this
2285 function moves to point C, whereas the UAX#9 ``level 2 run'' ends
2288 bidi_find_other_level_edge (struct bidi_it
*bidi_it
, int level
, bool end_flag
)
2290 int dir
= end_flag
? -bidi_it
->scan_dir
: bidi_it
->scan_dir
;
2293 /* Try the cache first. */
2294 if ((idx
= bidi_cache_find_level_change (level
, dir
, end_flag
))
2295 >= bidi_cache_start
)
2296 bidi_cache_fetch_state (idx
, bidi_it
);
2301 /* If we are at end of level, its edges must be cached. */
2305 bidi_cache_iterator_state (bidi_it
, 1);
2307 new_level
= bidi_level_of_next_char (bidi_it
);
2308 bidi_cache_iterator_state (bidi_it
, 1);
2309 } while (new_level
>= level
);
2314 bidi_move_to_visually_next (struct bidi_it
*bidi_it
)
2316 int old_level
, new_level
, next_level
;
2317 struct bidi_it sentinel
;
2318 struct gcpro gcpro1
;
2320 if (bidi_it
->charpos
< 0 || bidi_it
->bytepos
< 0)
2323 if (bidi_it
->scan_dir
== 0)
2325 bidi_it
->scan_dir
= 1; /* default to logical order */
2328 /* The code below can call eval, and thus cause GC. If we are
2329 iterating a Lisp string, make sure it won't be GCed. */
2330 if (STRINGP (bidi_it
->string
.lstring
))
2331 GCPRO1 (bidi_it
->string
.lstring
);
2333 /* If we just passed a newline, initialize for the next line. */
2334 if (!bidi_it
->first_elt
2335 && (bidi_it
->ch
== '\n' || bidi_it
->ch
== BIDI_EOB
))
2336 bidi_line_init (bidi_it
);
2338 /* Prepare the sentinel iterator state, and cache it. When we bump
2339 into it, scanning backwards, we'll know that the last non-base
2340 level is exhausted. */
2341 if (bidi_cache_idx
== bidi_cache_start
)
2343 bidi_copy_it (&sentinel
, bidi_it
);
2344 if (bidi_it
->first_elt
)
2346 sentinel
.charpos
--; /* cached charpos needs to be monotonic */
2348 sentinel
.ch
= '\n'; /* doesn't matter, but why not? */
2349 sentinel
.ch_len
= 1;
2350 sentinel
.nchars
= 1;
2352 bidi_cache_iterator_state (&sentinel
, 1);
2355 old_level
= bidi_it
->resolved_level
;
2356 new_level
= bidi_level_of_next_char (bidi_it
);
2358 /* Reordering of resolved levels (clause L2) is implemented by
2359 jumping to the other edge of the level and flipping direction of
2360 scanning the text whenever we find a level change. */
2361 if (new_level
!= old_level
)
2363 bool ascending
= new_level
> old_level
;
2364 int level_to_search
= ascending
? old_level
+ 1 : old_level
;
2365 int incr
= ascending
? 1 : -1;
2366 int expected_next_level
= old_level
+ incr
;
2368 /* Jump (or walk) to the other edge of this level. */
2369 bidi_find_other_level_edge (bidi_it
, level_to_search
, !ascending
);
2370 /* Switch scan direction and peek at the next character in the
2372 bidi_it
->scan_dir
= -bidi_it
->scan_dir
;
2374 /* The following loop handles the case where the resolved level
2375 jumps by more than one. This is typical for numbers inside a
2376 run of text with left-to-right embedding direction, but can
2377 also happen in other situations. In those cases the decision
2378 where to continue after a level change, and in what direction,
2379 is tricky. For example, given a text like below:
2384 (where the numbers below the text show the resolved levels),
2385 the result of reordering according to UAX#9 should be this:
2389 This is implemented by the loop below which flips direction
2390 and jumps to the other edge of the level each time it finds
2391 the new level not to be the expected one. The expected level
2392 is always one more or one less than the previous one. */
2393 next_level
= bidi_peek_at_next_level (bidi_it
);
2394 while (next_level
!= expected_next_level
)
2396 /* If next_level is -1, it means we have an unresolved level
2397 in the cache, which at this point should not happen. If
2398 it does, we will infloop. */
2399 eassert (next_level
>= 0);
2400 expected_next_level
+= incr
;
2401 level_to_search
+= incr
;
2402 bidi_find_other_level_edge (bidi_it
, level_to_search
, !ascending
);
2403 bidi_it
->scan_dir
= -bidi_it
->scan_dir
;
2404 next_level
= bidi_peek_at_next_level (bidi_it
);
2407 /* Finally, deliver the next character in the new direction. */
2408 next_level
= bidi_level_of_next_char (bidi_it
);
2411 /* Take note when we have just processed the newline that precedes
2412 the end of the paragraph. The next time we are about to be
2413 called, set_iterator_to_next will automatically reinit the
2414 paragraph direction, if needed. We do this at the newline before
2415 the paragraph separator, because the next character might not be
2416 the first character of the next paragraph, due to the bidi
2417 reordering, whereas we _must_ know the paragraph base direction
2418 _before_ we process the paragraph's text, since the base
2419 direction affects the reordering. */
2420 if (bidi_it
->scan_dir
== 1
2421 && (bidi_it
->ch
== '\n' || bidi_it
->ch
== BIDI_EOB
))
2423 /* The paragraph direction of the entire string, once
2424 determined, is in effect for the entire string. Setting the
2425 separator limit to the end of the string prevents
2426 bidi_paragraph_init from being called automatically on this
2428 if (bidi_it
->string
.s
|| STRINGP (bidi_it
->string
.lstring
))
2429 bidi_it
->separator_limit
= bidi_it
->string
.schars
;
2430 else if (bidi_it
->bytepos
< ZV_BYTE
)
2433 = bidi_at_paragraph_end (bidi_it
->charpos
+ bidi_it
->nchars
,
2434 bidi_it
->bytepos
+ bidi_it
->ch_len
);
2435 if (bidi_it
->nchars
<= 0)
2439 bidi_it
->new_paragraph
= 1;
2440 /* Record the buffer position of the last character of the
2441 paragraph separator. */
2442 bidi_it
->separator_limit
2443 = bidi_it
->charpos
+ bidi_it
->nchars
+ sep_len
;
2448 if (bidi_it
->scan_dir
== 1 && bidi_cache_idx
> bidi_cache_start
)
2450 /* If we are at paragraph's base embedding level and beyond the
2451 last cached position, the cache's job is done and we can
2453 if (bidi_it
->resolved_level
== bidi_it
->level_stack
[0].level
2454 && bidi_it
->charpos
> (bidi_cache
[bidi_cache_idx
- 1].charpos
2455 + bidi_cache
[bidi_cache_idx
- 1].nchars
- 1))
2456 bidi_cache_reset ();
2457 /* But as long as we are caching during forward scan, we must
2458 cache each state, or else the cache integrity will be
2459 compromised: it assumes cached states correspond to buffer
2462 bidi_cache_iterator_state (bidi_it
, 1);
2465 if (STRINGP (bidi_it
->string
.lstring
))
2469 /* This is meant to be called from within the debugger, whenever you
2470 wish to examine the cache contents. */
2471 void bidi_dump_cached_states (void) EXTERNALLY_VISIBLE
;
2473 bidi_dump_cached_states (void)
2478 if (bidi_cache_idx
== 0)
2480 fprintf (stderr
, "The cache is empty.\n");
2483 fprintf (stderr
, "Total of %"pD
"d state%s in cache:\n",
2484 bidi_cache_idx
, bidi_cache_idx
== 1 ? "" : "s");
2486 for (i
= bidi_cache
[bidi_cache_idx
- 1].charpos
; i
> 0; i
/= 10)
2488 fputs ("ch ", stderr
);
2489 for (i
= 0; i
< bidi_cache_idx
; i
++)
2490 fprintf (stderr
, "%*c", ndigits
, bidi_cache
[i
].ch
);
2491 fputs ("\n", stderr
);
2492 fputs ("lvl ", stderr
);
2493 for (i
= 0; i
< bidi_cache_idx
; i
++)
2494 fprintf (stderr
, "%*d", ndigits
, bidi_cache
[i
].resolved_level
);
2495 fputs ("\n", stderr
);
2496 fputs ("pos ", stderr
);
2497 for (i
= 0; i
< bidi_cache_idx
; i
++)
2498 fprintf (stderr
, "%*"pD
"d", ndigits
, bidi_cache
[i
].charpos
);
2499 fputs ("\n", stderr
);