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.
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 decribed in the "Implementation
55 Notes" section of UAX#9, under "Retaining Format Codes". */
63 #include "character.h"
64 #include "dispextern.h"
66 static int bidi_initialized
= 0;
68 static Lisp_Object bidi_type_table
, bidi_mirror_table
;
70 #define LRM_CHAR 0x200E
71 #define RLM_CHAR 0x200F
74 /* Data type for describing the bidirectional character categories. */
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
87 extern int bidi_ignore_explicit_marks_for_paragraph_level EXTERNALLY_VISIBLE
;
88 int bidi_ignore_explicit_marks_for_paragraph_level
= 1;
90 static Lisp_Object paragraph_start_re
, paragraph_separate_re
;
91 static Lisp_Object Qparagraph_start
, Qparagraph_separate
;
94 /***********************************************************************
96 ***********************************************************************/
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
)
103 bidi_type_t default_type
;
107 if (ch
< 0 || ch
> MAX_CHAR
)
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
)
118 if (override
== NEUTRAL_DIR
)
121 switch (default_type
)
123 /* Although UAX#9 does not tell, it doesn't make sense to
124 override NEUTRAL_B and LRM/RLM characters. */
139 if (override
== L2R
) /* X6 */
141 else if (override
== R2L
)
144 abort (); /* can't happen: handled above */
150 bidi_check_type (bidi_type_t type
)
152 xassert (UNKNOWN_BT
<= type
&& type
<= NEUTRAL_ON
);
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
)
171 case PDF
: /* ??? really?? */
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. */
195 bidi_mirror_char (int c
)
201 if (c
< 0 || c
> MAX_CHAR
)
204 val
= CHAR_TABLE_REF (bidi_mirror_table
, c
);
209 if (v
< 0 || v
> MAX_CHAR
)
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. */
223 bidi_set_sor_type (struct bidi_it
*bidi_it
, int level_before
, int level_after
)
225 int higher_level
= (level_before
> level_after
? level_before
: level_after
);
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;
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 */
251 /* Push the current embedding level and override status; reset the
252 current level to LEVEL and the current override status to OVERRIDE. */
254 bidi_push_embedding_level (struct bidi_it
*bidi_it
,
255 int level
, bidi_dir_t override
)
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
;
263 /* Pop the embedding level and directional override status from the
264 stack, and return the new level. */
266 bidi_pop_embedding_level (struct bidi_it
*bidi_it
)
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
;
274 /* Record in SAVED_INFO the information about the current character. */
276 bidi_remember_char (struct bidi_saved_info
*saved_info
,
277 struct bidi_it
*bidi_it
)
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
);
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. */
292 bidi_copy_it (struct bidi_it
*to
, struct bidi_it
*from
)
296 /* Copy everything except the level stack and beyond. */
297 memcpy (to
, from
, offsetof (struct bidi_it
, level_stack
[0]));
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
];
306 /***********************************************************************
307 Caching the bidi iterator states
308 ***********************************************************************/
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
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
;
325 /* Size of header used by bidi_shelve_cache. */
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
))
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. */
341 bidi_cache_reset (void)
343 bidi_cache_idx
= bidi_cache_start
;
344 bidi_cache_last_idx
= -1;
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. */
352 bidi_cache_shrink (void)
354 if (bidi_cache_size
> BIDI_CACHE_CHUNK
)
357 = (struct bidi_it
*) xrealloc (bidi_cache
, BIDI_CACHE_CHUNK
* elsz
);
358 bidi_cache_size
= BIDI_CACHE_CHUNK
;
364 bidi_cache_fetch_state (ptrdiff_t idx
, struct bidi_it
*bidi_it
)
366 int current_scan_dir
= bidi_it
->scan_dir
;
368 if (idx
< bidi_cache_start
|| idx
>= bidi_cache_idx
)
371 bidi_copy_it (bidi_it
, &bidi_cache
[idx
]);
372 bidi_it
->scan_dir
= current_scan_dir
;
373 bidi_cache_last_idx
= idx
;
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
)
383 ptrdiff_t i
, i_start
;
385 if (bidi_cache_idx
> bidi_cache_start
)
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
)
392 i_start
= bidi_cache_last_idx
- 1;
394 else if (charpos
> (bidi_cache
[bidi_cache_last_idx
].charpos
395 + bidi_cache
[bidi_cache_last_idx
].nchars
- 1))
398 i_start
= bidi_cache_last_idx
+ 1;
401 i_start
= bidi_cache_last_idx
;
405 i_start
= bidi_cache_idx
- 1;
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
))
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
))
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:
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,
446 bidi_cache_find_level_change (int level
, int dir
, int before
)
450 ptrdiff_t i
= dir
? bidi_cache_last_idx
: bidi_cache_idx
- 1;
451 int incr
= before
? 1 : 0;
453 xassert (!dir
|| bidi_cache_last_idx
>= 0);
462 while (i
>= bidi_cache_start
+ incr
)
464 if (bidi_cache
[i
- incr
].resolved_level
>= 0
465 && bidi_cache
[i
- incr
].resolved_level
< level
)
472 while (i
< bidi_cache_idx
- incr
)
474 if (bidi_cache
[i
+ incr
].resolved_level
>= 0
475 && bidi_cache
[i
+ incr
].resolved_level
< level
)
486 bidi_cache_ensure_space (ptrdiff_t idx
)
488 /* Enlarge the cache as needed. */
489 if (idx
>= bidi_cache_size
)
491 /* The bidi cache cannot be larger than the largest Lisp string
493 ptrdiff_t string_or_buffer_bound
494 = max (BUF_BYTES_MAX
, STRING_BYTES_BOUND
);
496 /* Also, it cannot be larger than what C can represent. */
498 = (min (PTRDIFF_MAX
, SIZE_MAX
) - bidi_shelve_header_size
) / elsz
;
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
);
508 bidi_cache_iterator_state (struct bidi_it
*bidi_it
, int resolved
)
512 /* We should never cache on backward scans. */
513 if (bidi_it
->scan_dir
== -1)
515 idx
= bidi_cache_search (bidi_it
->charpos
, -1, 1);
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
))
530 idx
= bidi_cache_start
;
532 if (bidi_it
->nchars
<= 0)
534 bidi_copy_it (&bidi_cache
[idx
], bidi_it
);
536 bidi_cache
[idx
].resolved_level
= -1;
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
);
547 bidi_cache
[idx
].resolved_level
= bidi_it
->resolved_level
;
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
= bidi_it
->disp_prop
;
559 bidi_cache_last_idx
= idx
;
560 if (idx
>= bidi_cache_idx
)
561 bidi_cache_idx
= idx
+ 1;
564 static inline bidi_type_t
565 bidi_cache_find (EMACS_INT charpos
, int level
, struct bidi_it
*bidi_it
)
567 ptrdiff_t i
= bidi_cache_search (charpos
, level
, bidi_it
->scan_dir
);
569 if (i
>= bidi_cache_start
)
571 bidi_dir_t current_scan_dir
= bidi_it
->scan_dir
;
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
;
585 bidi_peek_at_next_level (struct bidi_it
*bidi_it
)
587 if (bidi_cache_idx
== bidi_cache_start
|| bidi_cache_last_idx
== -1)
589 return bidi_cache
[bidi_cache_last_idx
+ bidi_it
->scan_dir
].resolved_level
;
593 /***********************************************************************
594 Pushing and popping the bidi iterator state
595 ***********************************************************************/
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. */
603 bidi_push_it (struct bidi_it
*bidi_it
)
605 /* Save the current iterator state in its entirety after the last
607 bidi_cache_ensure_space (bidi_cache_idx
);
608 memcpy (&bidi_cache
[bidi_cache_idx
++], bidi_it
, sizeof (struct bidi_it
));
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
;
614 /* Start a new level of cache, and make it empty. */
615 bidi_cache_start
= bidi_cache_idx
;
616 bidi_cache_last_idx
= -1;
619 /* Restore the iterator state saved by bidi_push_it and return the
620 cache to the corresponding state. */
622 bidi_pop_it (struct bidi_it
*bidi_it
)
624 if (bidi_cache_start
<= 0)
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;
631 /* Restore the bidi iterator state saved in the cache. */
632 memcpy (bidi_it
, &bidi_cache
[bidi_cache_idx
], sizeof (struct bidi_it
));
634 /* Pop the previous cache start from the stack. */
635 if (bidi_cache_sp
<= 0)
637 bidi_cache_start
= bidi_cache_start_stack
[--bidi_cache_sp
];
639 /* Invalidate the last-used cache slot data. */
640 bidi_cache_last_idx
= -1;
643 static ptrdiff_t bidi_cache_total_alloc
;
645 /* Stash away a copy of the cache and its control variables. */
647 bidi_shelve_cache (void)
649 unsigned char *databuf
;
653 if (bidi_cache_idx
== 0)
656 alloc
= (bidi_shelve_header_size
657 + bidi_cache_idx
* sizeof (struct bidi_it
));
658 databuf
= xmalloc (alloc
);
659 bidi_cache_total_alloc
+= alloc
;
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
));
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
690 bidi_unshelve_cache (void *databuf
, int just_free
)
692 unsigned char *p
= databuf
;
698 /* A NULL pointer means an empty cache. */
699 bidi_cache_start
= 0;
710 memcpy (&idx
, p
, sizeof (bidi_cache_idx
));
711 bidi_cache_total_alloc
712 -= bidi_shelve_header_size
+ idx
* sizeof (struct bidi_it
);
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
742 + bidi_cache_idx
* sizeof (struct bidi_it
));
750 /***********************************************************************
752 ***********************************************************************/
754 bidi_initialize (void)
756 bidi_type_table
= uniprop_table (intern ("bidi-class"));
757 if (NILP (bidi_type_table
))
759 staticpro (&bidi_type_table
);
761 bidi_mirror_table
= uniprop_table (intern ("mirroring"));
762 if (NILP (bidi_mirror_table
))
764 staticpro (&bidi_mirror_table
);
766 Qparagraph_start
= intern ("paragraph-start");
767 staticpro (&Qparagraph_start
);
768 paragraph_start_re
= Fsymbol_value (Qparagraph_start
);
769 if (!STRINGP (paragraph_start_re
))
770 paragraph_start_re
= build_string ("\f\\|[ \t]*$");
771 staticpro (¶graph_start_re
);
772 Qparagraph_separate
= intern ("paragraph-separate");
773 staticpro (&Qparagraph_separate
);
774 paragraph_separate_re
= Fsymbol_value (Qparagraph_separate
);
775 if (!STRINGP (paragraph_separate_re
))
776 paragraph_separate_re
= build_string ("[ \t\f]*$");
777 staticpro (¶graph_separate_re
);
780 bidi_cache_total_alloc
= 0;
782 bidi_initialized
= 1;
785 /* Do whatever UAX#9 clause X8 says should be done at paragraph's
788 bidi_set_paragraph_end (struct bidi_it
*bidi_it
)
790 bidi_it
->invalid_levels
= 0;
791 bidi_it
->invalid_rl_levels
= -1;
792 bidi_it
->stack_idx
= 0;
793 bidi_it
->resolved_level
= bidi_it
->level_stack
[0].level
;
796 /* Initialize the bidi iterator from buffer/string position CHARPOS. */
798 bidi_init_it (EMACS_INT charpos
, EMACS_INT bytepos
, int frame_window_p
,
799 struct bidi_it
*bidi_it
)
801 if (! bidi_initialized
)
804 bidi_it
->charpos
= charpos
;
806 bidi_it
->bytepos
= bytepos
;
807 bidi_it
->frame_window_p
= frame_window_p
;
808 bidi_it
->nchars
= -1; /* to be computed in bidi_resolve_explicit_1 */
809 bidi_it
->first_elt
= 1;
810 bidi_set_paragraph_end (bidi_it
);
811 bidi_it
->paragraph_dir
= NEUTRAL_DIR
;
812 bidi_it
->new_paragraph
= 1;
813 bidi_it
->separator_limit
= -1;
814 bidi_it
->type
= NEUTRAL_B
;
815 bidi_it
->type_after_w1
= NEUTRAL_B
;
816 bidi_it
->orig_type
= NEUTRAL_B
;
817 bidi_it
->prev_was_pdf
= 0;
818 bidi_it
->prev
.type
= bidi_it
->prev
.type_after_w1
819 = bidi_it
->prev
.orig_type
= UNKNOWN_BT
;
820 bidi_it
->last_strong
.type
= bidi_it
->last_strong
.type_after_w1
821 = bidi_it
->last_strong
.orig_type
= UNKNOWN_BT
;
822 bidi_it
->next_for_neutral
.charpos
= -1;
823 bidi_it
->next_for_neutral
.type
824 = bidi_it
->next_for_neutral
.type_after_w1
825 = bidi_it
->next_for_neutral
.orig_type
= UNKNOWN_BT
;
826 bidi_it
->prev_for_neutral
.charpos
= -1;
827 bidi_it
->prev_for_neutral
.type
828 = bidi_it
->prev_for_neutral
.type_after_w1
829 = bidi_it
->prev_for_neutral
.orig_type
= UNKNOWN_BT
;
830 bidi_it
->sor
= L2R
; /* FIXME: should it be user-selectable? */
831 bidi_it
->disp_pos
= -1; /* invalid/unknown */
832 bidi_it
->disp_prop
= 0;
833 /* We can only shrink the cache if we are at the bottom level of its
835 if (bidi_cache_start
== 0)
836 bidi_cache_shrink ();
841 /* Perform initializations for reordering a new line of bidi text. */
843 bidi_line_init (struct bidi_it
*bidi_it
)
845 bidi_it
->scan_dir
= 1; /* FIXME: do we need to have control on this? */
846 bidi_it
->resolved_level
= bidi_it
->level_stack
[0].level
;
847 bidi_it
->level_stack
[0].override
= NEUTRAL_DIR
; /* X1 */
848 bidi_it
->invalid_levels
= 0;
849 bidi_it
->invalid_rl_levels
= -1;
850 bidi_it
->next_en_pos
= -1;
851 bidi_it
->next_for_ws
.type
= UNKNOWN_BT
;
852 bidi_set_sor_type (bidi_it
,
853 (bidi_it
->paragraph_dir
== R2L
? 1 : 0),
854 bidi_it
->level_stack
[0].level
); /* X10 */
860 /***********************************************************************
862 ***********************************************************************/
864 /* Count bytes in string S between BEG/BEGBYTE and END. BEG and END
865 are zero-based character positions in S, BEGBYTE is byte position
866 corresponding to BEG. UNIBYTE, if non-zero, means S is a unibyte
868 static inline EMACS_INT
869 bidi_count_bytes (const unsigned char *s
, const EMACS_INT beg
,
870 const EMACS_INT begbyte
, const EMACS_INT end
, int unibyte
)
873 const unsigned char *p
= s
+ begbyte
, *start
= p
;
879 if (!CHAR_HEAD_P (*p
))
884 p
+= BYTES_BY_CHAR_HEAD (*p
);
892 /* Fetch and returns the character at byte position BYTEPOS. If S is
893 non-NULL, fetch the character from string S; otherwise fetch the
894 character from the current buffer. UNIBYTE non-zero means S is a
897 bidi_char_at_pos (EMACS_INT bytepos
, const unsigned char *s
, int unibyte
)
904 return STRING_CHAR (s
+ bytepos
);
907 return FETCH_MULTIBYTE_CHAR (bytepos
);
910 /* Fetch and return the character at BYTEPOS/CHARPOS. If that
911 character is covered by a display string, treat the entire run of
912 covered characters as a single character, either u+2029 or u+FFFC,
913 and return their combined length in CH_LEN and NCHARS. DISP_POS
914 specifies the character position of the next display string, or -1
915 if not yet computed. 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. DISP_PROP non-zero means that there's really
918 a display string at DISP_POS, as opposed to when we searched till
919 DISP_POS without finding one. If DISP_PROP is 2, it means the
920 display spec is of the form `(space ...)', which is replaced with
921 u+2029 to handle it as a paragraph separator. STRING->s is the C
922 string to iterate, or NULL if iterating over a buffer or a Lisp
923 string; in the latter case, STRING->lstring is the Lisp string. */
925 bidi_fetch_char (EMACS_INT bytepos
, EMACS_INT charpos
, EMACS_INT
*disp_pos
,
926 int *disp_prop
, struct bidi_string_data
*string
,
927 int frame_window_p
, EMACS_INT
*ch_len
, EMACS_INT
*nchars
)
931 = (string
->s
|| STRINGP (string
->lstring
)) ? string
->schars
: ZV
;
934 /* If we got past the last known position of display string, compute
935 the position of the next one. That position could be at CHARPOS. */
936 if (charpos
< endpos
&& charpos
> *disp_pos
)
938 SET_TEXT_POS (pos
, charpos
, bytepos
);
939 *disp_pos
= compute_display_string_pos (&pos
, string
, frame_window_p
,
943 /* Fetch the character at BYTEPOS. */
944 if (charpos
>= endpos
)
952 else if (charpos
>= *disp_pos
&& *disp_prop
)
954 EMACS_INT disp_end_pos
;
956 /* We don't expect to find ourselves in the middle of a display
957 property. Hopefully, it will never be needed. */
958 if (charpos
> *disp_pos
)
960 /* Text covered by `display' properties and overlays with
961 display properties or display strings is handled as a single
962 character that represents the entire run of characters
963 covered by the display property. */
966 /* `(space ...)' display specs are handled as paragraph
967 separators for the purposes of the reordering; see UAX#9
968 section 3 and clause HL1 in section 4.3 there. */
973 /* All other display specs are handled as the Unicode Object
974 Replacement Character. */
977 disp_end_pos
= compute_display_string_end (*disp_pos
, string
);
978 if (disp_end_pos
< 0)
980 /* Somebody removed the display string from the buffer
981 behind our back. Recover by processing this buffer
982 position as if no display property were present there to
987 *nchars
= disp_end_pos
- *disp_pos
;
991 *ch_len
= bidi_count_bytes (string
->s
, *disp_pos
, bytepos
,
992 disp_end_pos
, string
->unibyte
);
993 else if (STRINGP (string
->lstring
))
994 *ch_len
= bidi_count_bytes (SDATA (string
->lstring
), *disp_pos
,
995 bytepos
, disp_end_pos
, string
->unibyte
);
997 *ch_len
= CHAR_TO_BYTE (disp_end_pos
) - bytepos
;
1006 if (!string
->unibyte
)
1008 ch
= STRING_CHAR_AND_LENGTH (string
->s
+ bytepos
, len
);
1013 ch
= UNIBYTE_TO_CHAR (string
->s
[bytepos
]);
1017 else if (STRINGP (string
->lstring
))
1021 if (!string
->unibyte
)
1023 ch
= STRING_CHAR_AND_LENGTH (SDATA (string
->lstring
) + bytepos
,
1029 ch
= UNIBYTE_TO_CHAR (SREF (string
->lstring
, bytepos
));
1035 ch
= FETCH_MULTIBYTE_CHAR (bytepos
);
1036 *ch_len
= CHAR_BYTES (ch
);
1041 /* If we just entered a run of characters covered by a display
1042 string, compute the position of the next display string. */
1043 if (charpos
+ *nchars
<= endpos
&& charpos
+ *nchars
> *disp_pos
1046 SET_TEXT_POS (pos
, charpos
+ *nchars
, bytepos
+ *ch_len
);
1047 *disp_pos
= compute_display_string_pos (&pos
, string
, frame_window_p
,
1055 /***********************************************************************
1056 Determining paragraph direction
1057 ***********************************************************************/
1059 /* Check if buffer position CHARPOS/BYTEPOS is the end of a paragraph.
1060 Value is the non-negative length of the paragraph separator
1061 following the buffer position, -1 if position is at the beginning
1062 of a new paragraph, or -2 if position is neither at beginning nor
1063 at end of a paragraph. */
1065 bidi_at_paragraph_end (EMACS_INT charpos
, EMACS_INT bytepos
)
1068 Lisp_Object start_re
;
1071 sep_re
= paragraph_separate_re
;
1072 start_re
= paragraph_start_re
;
1074 val
= fast_looking_at (sep_re
, charpos
, bytepos
, ZV
, ZV_BYTE
, Qnil
);
1077 if (fast_looking_at (start_re
, charpos
, bytepos
, ZV
, ZV_BYTE
, Qnil
) >= 0)
1086 /* On my 2005-vintage machine, searching back for paragraph start
1087 takes ~1 ms per line. And bidi_paragraph_init is called 4 times
1088 when user types C-p. The number below limits each call to
1089 bidi_paragraph_init to about 10 ms. */
1090 #define MAX_PARAGRAPH_SEARCH 7500
1092 /* Find the beginning of this paragraph by looking back in the buffer.
1093 Value is the byte position of the paragraph's beginning, or
1094 BEGV_BYTE if paragraph_start_re is still not found after looking
1095 back MAX_PARAGRAPH_SEARCH lines in the buffer. */
1097 bidi_find_paragraph_start (EMACS_INT pos
, EMACS_INT pos_byte
)
1099 Lisp_Object re
= paragraph_start_re
;
1100 EMACS_INT limit
= ZV
, limit_byte
= ZV_BYTE
;
1103 while (pos_byte
> BEGV_BYTE
1104 && n
++ < MAX_PARAGRAPH_SEARCH
1105 && 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_no_quit (pos
- 1, -1);
1112 pos_byte
= CHAR_TO_BYTE (pos
);
1114 if (n
>= MAX_PARAGRAPH_SEARCH
)
1115 pos_byte
= BEGV_BYTE
;
1119 /* Determine the base direction, a.k.a. base embedding level, of the
1120 paragraph we are about to iterate through. If DIR is either L2R or
1121 R2L, just use that. Otherwise, determine the paragraph direction
1122 from the first strong directional character of the paragraph.
1124 NO_DEFAULT_P non-zero means don't default to L2R if the paragraph
1125 has no strong directional characters and both DIR and
1126 bidi_it->paragraph_dir are NEUTRAL_DIR. In that case, search back
1127 in the buffer until a paragraph is found with a strong character,
1128 or until hitting BEGV. In the latter case, fall back to L2R. This
1129 flag is used in current-bidi-paragraph-direction.
1131 Note that this function gives the paragraph separator the same
1132 direction as the preceding paragraph, even though Emacs generally
1133 views the separartor as not belonging to any paragraph. */
1135 bidi_paragraph_init (bidi_dir_t dir
, struct bidi_it
*bidi_it
, int no_default_p
)
1137 EMACS_INT bytepos
= bidi_it
->bytepos
;
1138 int string_p
= bidi_it
->string
.s
!= NULL
|| STRINGP (bidi_it
->string
.lstring
);
1139 EMACS_INT pstartbyte
;
1140 /* Note that begbyte is a byte position, while end is a character
1141 position. Yes, this is ugly, but we are trying to avoid costly
1142 calls to BYTE_TO_CHAR and its ilk. */
1143 EMACS_INT begbyte
= string_p
? 0 : BEGV_BYTE
;
1144 EMACS_INT end
= string_p
? bidi_it
->string
.schars
: ZV
;
1146 /* Special case for an empty buffer. */
1147 if (bytepos
== begbyte
&& bidi_it
->charpos
== end
)
1149 /* We should never be called at EOB or before BEGV. */
1150 else if (bidi_it
->charpos
>= end
|| bytepos
< begbyte
)
1155 bidi_it
->paragraph_dir
= L2R
;
1156 bidi_it
->new_paragraph
= 0;
1158 else if (dir
== R2L
)
1160 bidi_it
->paragraph_dir
= R2L
;
1161 bidi_it
->new_paragraph
= 0;
1163 else if (dir
== NEUTRAL_DIR
) /* P2 */
1166 EMACS_INT ch_len
, nchars
;
1167 EMACS_INT pos
, disp_pos
= -1;
1170 const unsigned char *s
;
1172 if (!bidi_initialized
)
1175 /* If we are inside a paragraph separator, we are just waiting
1176 for the separator to be exhausted; use the previous paragraph
1177 direction. But don't do that if we have been just reseated,
1178 because we need to reinitialize below in that case. */
1179 if (!bidi_it
->first_elt
1180 && bidi_it
->charpos
< bidi_it
->separator_limit
)
1183 /* If we are on a newline, get past it to where the next
1184 paragraph might start. But don't do that at BEGV since then
1185 we are potentially in a new paragraph that doesn't yet
1187 pos
= bidi_it
->charpos
;
1188 s
= (STRINGP (bidi_it
->string
.lstring
)
1189 ? SDATA (bidi_it
->string
.lstring
)
1190 : bidi_it
->string
.s
);
1191 if (bytepos
> begbyte
1192 && bidi_char_at_pos (bytepos
, s
, bidi_it
->string
.unibyte
) == '\n')
1198 /* We are either at the beginning of a paragraph or in the
1199 middle of it. Find where this paragraph starts. */
1202 /* We don't support changes of paragraph direction inside a
1203 string. It is treated as a single paragraph. */
1207 pstartbyte
= bidi_find_paragraph_start (pos
, bytepos
);
1208 bidi_it
->separator_limit
= -1;
1209 bidi_it
->new_paragraph
= 0;
1211 /* The following loop is run more than once only if NO_DEFAULT_P
1212 is non-zero, and only if we are iterating on a buffer. */
1214 bytepos
= pstartbyte
;
1216 pos
= BYTE_TO_CHAR (bytepos
);
1217 ch
= bidi_fetch_char (bytepos
, pos
, &disp_pos
, &disp_prop
,
1219 bidi_it
->frame_window_p
, &ch_len
, &nchars
);
1220 type
= bidi_get_type (ch
, NEUTRAL_DIR
);
1222 for (pos
+= nchars
, bytepos
+= ch_len
;
1223 (bidi_get_category (type
) != STRONG
)
1224 || (bidi_ignore_explicit_marks_for_paragraph_level
1225 && (type
== RLE
|| type
== RLO
1226 || type
== LRE
|| type
== LRO
));
1227 type
= bidi_get_type (ch
, NEUTRAL_DIR
))
1231 /* Pretend there's a paragraph separator at end of
1237 && type
== NEUTRAL_B
1238 && bidi_at_paragraph_end (pos
, bytepos
) >= -1)
1240 /* Fetch next character and advance to get past it. */
1241 ch
= bidi_fetch_char (bytepos
, pos
, &disp_pos
,
1242 &disp_prop
, &bidi_it
->string
,
1243 bidi_it
->frame_window_p
, &ch_len
, &nchars
);
1247 if ((type
== STRONG_R
|| type
== STRONG_AL
) /* P3 */
1248 || (!bidi_ignore_explicit_marks_for_paragraph_level
1249 && (type
== RLO
|| type
== RLE
)))
1250 bidi_it
->paragraph_dir
= R2L
;
1251 else if (type
== STRONG_L
1252 || (!bidi_ignore_explicit_marks_for_paragraph_level
1253 && (type
== LRO
|| type
== LRE
)))
1254 bidi_it
->paragraph_dir
= L2R
;
1256 && no_default_p
&& bidi_it
->paragraph_dir
== NEUTRAL_DIR
)
1258 /* If this paragraph is at BEGV, default to L2R. */
1259 if (pstartbyte
== BEGV_BYTE
)
1260 bidi_it
->paragraph_dir
= L2R
; /* P3 and HL1 */
1263 EMACS_INT prevpbyte
= pstartbyte
;
1264 EMACS_INT p
= BYTE_TO_CHAR (pstartbyte
), pbyte
= pstartbyte
;
1266 /* Find the beginning of the previous paragraph, if any. */
1267 while (pbyte
> BEGV_BYTE
&& prevpbyte
>= pstartbyte
)
1269 /* FXIME: What if p is covered by a display
1270 string? See also a FIXME inside
1271 bidi_find_paragraph_start. */
1273 pbyte
= CHAR_TO_BYTE (p
);
1274 prevpbyte
= bidi_find_paragraph_start (p
, pbyte
);
1276 pstartbyte
= prevpbyte
;
1280 && no_default_p
&& bidi_it
->paragraph_dir
== NEUTRAL_DIR
);
1285 /* Contrary to UAX#9 clause P3, we only default the paragraph
1286 direction to L2R if we have no previous usable paragraph
1287 direction. This is allowed by the HL1 clause. */
1288 if (bidi_it
->paragraph_dir
!= L2R
&& bidi_it
->paragraph_dir
!= R2L
)
1289 bidi_it
->paragraph_dir
= L2R
; /* P3 and HL1 ``higher-level protocols'' */
1290 if (bidi_it
->paragraph_dir
== R2L
)
1291 bidi_it
->level_stack
[0].level
= 1;
1293 bidi_it
->level_stack
[0].level
= 0;
1295 bidi_line_init (bidi_it
);
1299 /***********************************************************************
1300 Resolving explicit and implicit levels.
1301 The rest of this file constitutes the core of the UBA implementation.
1302 ***********************************************************************/
1305 bidi_explicit_dir_char (int ch
)
1307 bidi_type_t ch_type
;
1309 if (!bidi_initialized
)
1311 ch_type
= (bidi_type_t
) XINT (CHAR_TABLE_REF (bidi_type_table
, ch
));
1312 return (ch_type
== LRE
|| ch_type
== LRO
1313 || ch_type
== RLE
|| ch_type
== RLO
1317 /* A helper function for bidi_resolve_explicit. It advances to the
1318 next character in logical order and determines the new embedding
1319 level and directional override, but does not take into account
1320 empty embeddings. */
1322 bidi_resolve_explicit_1 (struct bidi_it
*bidi_it
)
1328 bidi_dir_t override
;
1329 int string_p
= bidi_it
->string
.s
!= NULL
|| STRINGP (bidi_it
->string
.lstring
);
1331 /* If reseat()'ed, don't advance, so as to start iteration from the
1332 position where we were reseated. bidi_it->bytepos can be less
1333 than BEGV_BYTE after reseat to BEGV. */
1334 if (bidi_it
->bytepos
< (string_p
? 0 : BEGV_BYTE
)
1335 || bidi_it
->first_elt
)
1337 bidi_it
->first_elt
= 0;
1340 const unsigned char *p
1341 = (STRINGP (bidi_it
->string
.lstring
)
1342 ? SDATA (bidi_it
->string
.lstring
)
1343 : bidi_it
->string
.s
);
1345 if (bidi_it
->charpos
< 0)
1346 bidi_it
->charpos
= 0;
1347 bidi_it
->bytepos
= bidi_count_bytes (p
, 0, 0, bidi_it
->charpos
,
1348 bidi_it
->string
.unibyte
);
1352 if (bidi_it
->charpos
< BEGV
)
1353 bidi_it
->charpos
= BEGV
;
1354 bidi_it
->bytepos
= CHAR_TO_BYTE (bidi_it
->charpos
);
1357 /* Don't move at end of buffer/string. */
1358 else if (bidi_it
->charpos
< (string_p
? bidi_it
->string
.schars
: ZV
))
1360 /* Advance to the next character, skipping characters covered by
1361 display strings (nchars > 1). */
1362 if (bidi_it
->nchars
<= 0)
1364 bidi_it
->charpos
+= bidi_it
->nchars
;
1365 if (bidi_it
->ch_len
== 0)
1367 bidi_it
->bytepos
+= bidi_it
->ch_len
;
1370 current_level
= bidi_it
->level_stack
[bidi_it
->stack_idx
].level
; /* X1 */
1371 override
= bidi_it
->level_stack
[bidi_it
->stack_idx
].override
;
1372 new_level
= current_level
;
1374 if (bidi_it
->charpos
>= (string_p
? bidi_it
->string
.schars
: ZV
))
1377 bidi_it
->ch_len
= 1;
1378 bidi_it
->nchars
= 1;
1379 bidi_it
->disp_pos
= (string_p
? bidi_it
->string
.schars
: ZV
);
1380 bidi_it
->disp_prop
= 0;
1384 /* Fetch the character at BYTEPOS. If it is covered by a
1385 display string, treat the entire run of covered characters as
1386 a single character u+FFFC. */
1387 curchar
= bidi_fetch_char (bidi_it
->bytepos
, bidi_it
->charpos
,
1388 &bidi_it
->disp_pos
, &bidi_it
->disp_prop
,
1389 &bidi_it
->string
, bidi_it
->frame_window_p
,
1390 &bidi_it
->ch_len
, &bidi_it
->nchars
);
1392 bidi_it
->ch
= curchar
;
1394 /* Don't apply directional override here, as all the types we handle
1395 below will not be affected by the override anyway, and we need
1396 the original type unaltered. The override will be applied in
1397 bidi_resolve_weak. */
1398 type
= bidi_get_type (curchar
, NEUTRAL_DIR
);
1399 bidi_it
->orig_type
= type
;
1400 bidi_check_type (bidi_it
->orig_type
);
1403 bidi_it
->prev_was_pdf
= 0;
1405 bidi_it
->type_after_w1
= UNKNOWN_BT
;
1411 bidi_it
->type_after_w1
= type
;
1412 bidi_check_type (bidi_it
->type_after_w1
);
1413 type
= WEAK_BN
; /* X9/Retaining */
1414 if (bidi_it
->ignore_bn_limit
<= -1)
1416 if (current_level
<= BIDI_MAXLEVEL
- 4)
1418 /* Compute the least odd embedding level greater than
1419 the current level. */
1420 new_level
= ((current_level
+ 1) & ~1) + 1;
1421 if (bidi_it
->type_after_w1
== RLE
)
1422 override
= NEUTRAL_DIR
;
1425 if (current_level
== BIDI_MAXLEVEL
- 4)
1426 bidi_it
->invalid_rl_levels
= 0;
1427 bidi_push_embedding_level (bidi_it
, new_level
, override
);
1431 bidi_it
->invalid_levels
++;
1432 /* See the commentary about invalid_rl_levels below. */
1433 if (bidi_it
->invalid_rl_levels
< 0)
1434 bidi_it
->invalid_rl_levels
= 0;
1435 bidi_it
->invalid_rl_levels
++;
1438 else if (bidi_it
->prev
.type_after_w1
== WEAK_EN
/* W5/Retaining */
1439 || bidi_it
->next_en_pos
> bidi_it
->charpos
)
1444 bidi_it
->type_after_w1
= type
;
1445 bidi_check_type (bidi_it
->type_after_w1
);
1446 type
= WEAK_BN
; /* X9/Retaining */
1447 if (bidi_it
->ignore_bn_limit
<= -1)
1449 if (current_level
<= BIDI_MAXLEVEL
- 5)
1451 /* Compute the least even embedding level greater than
1452 the current level. */
1453 new_level
= ((current_level
+ 2) & ~1);
1454 if (bidi_it
->type_after_w1
== LRE
)
1455 override
= NEUTRAL_DIR
;
1458 bidi_push_embedding_level (bidi_it
, new_level
, override
);
1462 bidi_it
->invalid_levels
++;
1463 /* invalid_rl_levels counts invalid levels encountered
1464 while the embedding level was already too high for
1465 LRE/LRO, but not for RLE/RLO. That is because
1466 there may be exactly one PDF which we should not
1467 ignore even though invalid_levels is non-zero.
1468 invalid_rl_levels helps to know what PDF is
1470 if (bidi_it
->invalid_rl_levels
>= 0)
1471 bidi_it
->invalid_rl_levels
++;
1474 else if (bidi_it
->prev
.type_after_w1
== WEAK_EN
/* W5/Retaining */
1475 || bidi_it
->next_en_pos
> bidi_it
->charpos
)
1479 bidi_it
->type_after_w1
= type
;
1480 bidi_check_type (bidi_it
->type_after_w1
);
1481 type
= WEAK_BN
; /* X9/Retaining */
1482 if (bidi_it
->ignore_bn_limit
<= -1)
1484 if (!bidi_it
->invalid_rl_levels
)
1486 new_level
= bidi_pop_embedding_level (bidi_it
);
1487 bidi_it
->invalid_rl_levels
= -1;
1488 if (bidi_it
->invalid_levels
)
1489 bidi_it
->invalid_levels
--;
1490 /* else nothing: UAX#9 says to ignore invalid PDFs */
1492 if (!bidi_it
->invalid_levels
)
1493 new_level
= bidi_pop_embedding_level (bidi_it
);
1496 bidi_it
->invalid_levels
--;
1497 bidi_it
->invalid_rl_levels
--;
1500 else if (bidi_it
->prev
.type_after_w1
== WEAK_EN
/* W5/Retaining */
1501 || bidi_it
->next_en_pos
> bidi_it
->charpos
)
1509 bidi_it
->type
= type
;
1510 bidi_check_type (bidi_it
->type
);
1515 /* Given an iterator state in BIDI_IT, advance one character position
1516 in the buffer/string to the next character (in the logical order),
1517 resolve any explicit embeddings and directional overrides, and
1518 return the embedding level of the character after resolving
1519 explicit directives and ignoring empty embeddings. */
1521 bidi_resolve_explicit (struct bidi_it
*bidi_it
)
1523 int prev_level
= bidi_it
->level_stack
[bidi_it
->stack_idx
].level
;
1524 int new_level
= bidi_resolve_explicit_1 (bidi_it
);
1525 EMACS_INT eob
= bidi_it
->string
.s
? bidi_it
->string
.schars
: ZV
;
1526 const unsigned char *s
1527 = (STRINGP (bidi_it
->string
.lstring
)
1528 ? SDATA (bidi_it
->string
.lstring
)
1529 : bidi_it
->string
.s
);
1531 if (prev_level
< new_level
1532 && bidi_it
->type
== WEAK_BN
1533 && bidi_it
->ignore_bn_limit
== -1 /* only if not already known */
1534 && bidi_it
->charpos
< eob
/* not already at EOB */
1535 && bidi_explicit_dir_char (bidi_char_at_pos (bidi_it
->bytepos
1536 + bidi_it
->ch_len
, s
,
1537 bidi_it
->string
.unibyte
)))
1539 /* Avoid pushing and popping embedding levels if the level run
1540 is empty, as this breaks level runs where it shouldn't.
1541 UAX#9 removes all the explicit embedding and override codes,
1542 so empty embeddings disappear without a trace. We need to
1543 behave as if we did the same. */
1544 struct bidi_it saved_it
;
1545 int level
= prev_level
;
1547 bidi_copy_it (&saved_it
, bidi_it
);
1549 while (bidi_explicit_dir_char (bidi_char_at_pos (bidi_it
->bytepos
1550 + bidi_it
->ch_len
, s
,
1551 bidi_it
->string
.unibyte
)))
1553 /* This advances to the next character, skipping any
1554 characters covered by display strings. */
1555 level
= bidi_resolve_explicit_1 (bidi_it
);
1556 /* If string.lstring was relocated inside bidi_resolve_explicit_1,
1557 a pointer to its data is no longer valid. */
1558 if (STRINGP (bidi_it
->string
.lstring
))
1559 s
= SDATA (bidi_it
->string
.lstring
);
1562 if (bidi_it
->nchars
<= 0)
1564 if (level
== prev_level
) /* empty embedding */
1565 saved_it
.ignore_bn_limit
= bidi_it
->charpos
+ bidi_it
->nchars
;
1566 else /* this embedding is non-empty */
1567 saved_it
.ignore_bn_limit
= -2;
1569 bidi_copy_it (bidi_it
, &saved_it
);
1570 if (bidi_it
->ignore_bn_limit
> -1)
1572 /* We pushed a level, but we shouldn't have. Undo that. */
1573 if (!bidi_it
->invalid_rl_levels
)
1575 new_level
= bidi_pop_embedding_level (bidi_it
);
1576 bidi_it
->invalid_rl_levels
= -1;
1577 if (bidi_it
->invalid_levels
)
1578 bidi_it
->invalid_levels
--;
1580 if (!bidi_it
->invalid_levels
)
1581 new_level
= bidi_pop_embedding_level (bidi_it
);
1584 bidi_it
->invalid_levels
--;
1585 bidi_it
->invalid_rl_levels
--;
1590 if (bidi_it
->type
== NEUTRAL_B
) /* X8 */
1592 bidi_set_paragraph_end (bidi_it
);
1593 /* This is needed by bidi_resolve_weak below, and in L1. */
1594 bidi_it
->type_after_w1
= bidi_it
->type
;
1595 bidi_check_type (bidi_it
->type_after_w1
);
1601 /* Advance in the buffer/string, resolve weak types and return the
1602 type of the next character after weak type resolution. */
1604 bidi_resolve_weak (struct bidi_it
*bidi_it
)
1607 bidi_dir_t override
;
1608 int prev_level
= bidi_it
->level_stack
[bidi_it
->stack_idx
].level
;
1609 int new_level
= bidi_resolve_explicit (bidi_it
);
1611 bidi_type_t type_of_next
;
1612 struct bidi_it saved_it
;
1614 = ((STRINGP (bidi_it
->string
.lstring
) || bidi_it
->string
.s
)
1615 ? bidi_it
->string
.schars
: ZV
);
1617 type
= bidi_it
->type
;
1618 override
= bidi_it
->level_stack
[bidi_it
->stack_idx
].override
;
1620 if (type
== UNKNOWN_BT
1628 if (new_level
!= prev_level
1629 || bidi_it
->type
== NEUTRAL_B
)
1631 /* We've got a new embedding level run, compute the directional
1632 type of sor and initialize per-run variables (UAX#9, clause
1634 bidi_set_sor_type (bidi_it
, prev_level
, new_level
);
1636 else if (type
== NEUTRAL_S
|| type
== NEUTRAL_WS
1637 || type
== WEAK_BN
|| type
== STRONG_AL
)
1638 bidi_it
->type_after_w1
= type
; /* needed in L1 */
1639 bidi_check_type (bidi_it
->type_after_w1
);
1641 /* Level and directional override status are already recorded in
1642 bidi_it, and do not need any change; see X6. */
1643 if (override
== R2L
) /* X6 */
1645 else if (override
== L2R
)
1649 if (type
== WEAK_NSM
) /* W1 */
1651 /* Note that we don't need to consider the case where the
1652 prev character has its type overridden by an RLO or LRO,
1653 because then either the type of this NSM would have been
1654 also overridden, or the previous character is outside the
1655 current level run, and thus not relevant to this NSM.
1656 This is why NSM gets the type_after_w1 of the previous
1658 if (bidi_it
->prev
.type_after_w1
!= UNKNOWN_BT
1659 /* if type_after_w1 is NEUTRAL_B, this NSM is at sor */
1660 && bidi_it
->prev
.type_after_w1
!= NEUTRAL_B
)
1661 type
= bidi_it
->prev
.type_after_w1
;
1662 else if (bidi_it
->sor
== R2L
)
1664 else if (bidi_it
->sor
== L2R
)
1666 else /* shouldn't happen! */
1669 if (type
== WEAK_EN
/* W2 */
1670 && bidi_it
->last_strong
.type_after_w1
== STRONG_AL
)
1672 else if (type
== STRONG_AL
) /* W3 */
1674 else if ((type
== WEAK_ES
/* W4 */
1675 && bidi_it
->prev
.type_after_w1
== WEAK_EN
1676 && bidi_it
->prev
.orig_type
== WEAK_EN
)
1678 && ((bidi_it
->prev
.type_after_w1
== WEAK_EN
1679 && bidi_it
->prev
.orig_type
== WEAK_EN
)
1680 || bidi_it
->prev
.type_after_w1
== WEAK_AN
)))
1682 const unsigned char *s
1683 = (STRINGP (bidi_it
->string
.lstring
)
1684 ? SDATA (bidi_it
->string
.lstring
)
1685 : bidi_it
->string
.s
);
1687 next_char
= (bidi_it
->charpos
+ bidi_it
->nchars
>= eob
1689 : bidi_char_at_pos (bidi_it
->bytepos
+ bidi_it
->ch_len
,
1690 s
, bidi_it
->string
.unibyte
));
1691 type_of_next
= bidi_get_type (next_char
, override
);
1693 if (type_of_next
== WEAK_BN
1694 || bidi_explicit_dir_char (next_char
))
1696 bidi_copy_it (&saved_it
, bidi_it
);
1697 while (bidi_resolve_explicit (bidi_it
) == new_level
1698 && bidi_it
->type
== WEAK_BN
)
1700 type_of_next
= bidi_it
->type
;
1701 bidi_copy_it (bidi_it
, &saved_it
);
1704 /* If the next character is EN, but the last strong-type
1705 character is AL, that next EN will be changed to AN when
1706 we process it in W2 above. So in that case, this ES
1707 should not be changed into EN. */
1709 && type_of_next
== WEAK_EN
1710 && bidi_it
->last_strong
.type_after_w1
!= STRONG_AL
)
1712 else if (type
== WEAK_CS
)
1714 if (bidi_it
->prev
.type_after_w1
== WEAK_AN
1715 && (type_of_next
== WEAK_AN
1716 /* If the next character is EN, but the last
1717 strong-type character is AL, EN will be later
1718 changed to AN when we process it in W2 above.
1719 So in that case, this ES should not be
1721 || (type_of_next
== WEAK_EN
1722 && bidi_it
->last_strong
.type_after_w1
== STRONG_AL
)))
1724 else if (bidi_it
->prev
.type_after_w1
== WEAK_EN
1725 && type_of_next
== WEAK_EN
1726 && bidi_it
->last_strong
.type_after_w1
!= STRONG_AL
)
1730 else if (type
== WEAK_ET
/* W5: ET with EN before or after it */
1731 || type
== WEAK_BN
) /* W5/Retaining */
1733 if (bidi_it
->prev
.type_after_w1
== WEAK_EN
/* ET/BN w/EN before it */
1734 || bidi_it
->next_en_pos
> bidi_it
->charpos
)
1736 else /* W5: ET/BN with EN after it. */
1738 EMACS_INT en_pos
= bidi_it
->charpos
+ bidi_it
->nchars
;
1739 const unsigned char *s
= (STRINGP (bidi_it
->string
.lstring
)
1740 ? SDATA (bidi_it
->string
.lstring
)
1741 : bidi_it
->string
.s
);
1743 if (bidi_it
->nchars
<= 0)
1746 = (bidi_it
->charpos
+ bidi_it
->nchars
>= eob
1748 : bidi_char_at_pos (bidi_it
->bytepos
+ bidi_it
->ch_len
, s
,
1749 bidi_it
->string
.unibyte
));
1750 type_of_next
= bidi_get_type (next_char
, override
);
1752 if (type_of_next
== WEAK_ET
1753 || type_of_next
== WEAK_BN
1754 || bidi_explicit_dir_char (next_char
))
1756 bidi_copy_it (&saved_it
, bidi_it
);
1757 while (bidi_resolve_explicit (bidi_it
) == new_level
1758 && (bidi_it
->type
== WEAK_BN
1759 || bidi_it
->type
== WEAK_ET
))
1761 type_of_next
= bidi_it
->type
;
1762 en_pos
= bidi_it
->charpos
;
1763 bidi_copy_it (bidi_it
, &saved_it
);
1765 if (type_of_next
== WEAK_EN
)
1767 /* If the last strong character is AL, the EN we've
1768 found will become AN when we get to it (W2). */
1769 if (bidi_it
->last_strong
.type_after_w1
!= STRONG_AL
)
1772 /* Remember this EN position, to speed up processing
1774 bidi_it
->next_en_pos
= en_pos
;
1776 else if (type
== WEAK_BN
)
1777 type
= NEUTRAL_ON
; /* W6/Retaining */
1783 if (type
== WEAK_ES
|| type
== WEAK_ET
|| type
== WEAK_CS
/* W6 */
1785 && (bidi_it
->prev
.type_after_w1
== WEAK_CS
/* W6/Retaining */
1786 || bidi_it
->prev
.type_after_w1
== WEAK_ES
1787 || bidi_it
->prev
.type_after_w1
== WEAK_ET
)))
1790 /* Store the type we've got so far, before we clobber it with strong
1791 types in W7 and while resolving neutral types. But leave alone
1792 the original types that were recorded above, because we will need
1793 them for the L1 clause. */
1794 if (bidi_it
->type_after_w1
== UNKNOWN_BT
)
1795 bidi_it
->type_after_w1
= type
;
1796 bidi_check_type (bidi_it
->type_after_w1
);
1798 if (type
== WEAK_EN
) /* W7 */
1800 if ((bidi_it
->last_strong
.type_after_w1
== STRONG_L
)
1801 || (bidi_it
->last_strong
.type
== UNKNOWN_BT
&& bidi_it
->sor
== L2R
))
1805 bidi_it
->type
= type
;
1806 bidi_check_type (bidi_it
->type
);
1810 /* Resolve the type of a neutral character according to the type of
1811 surrounding strong text and the current embedding level. */
1812 static inline bidi_type_t
1813 bidi_resolve_neutral_1 (bidi_type_t prev_type
, bidi_type_t next_type
, int lev
)
1815 /* N1: European and Arabic numbers are treated as though they were R. */
1816 if (next_type
== WEAK_EN
|| next_type
== WEAK_AN
)
1817 next_type
= STRONG_R
;
1818 if (prev_type
== WEAK_EN
|| prev_type
== WEAK_AN
)
1819 prev_type
= STRONG_R
;
1821 if (next_type
== prev_type
) /* N1 */
1823 else if ((lev
& 1) == 0) /* N2 */
1830 bidi_resolve_neutral (struct bidi_it
*bidi_it
)
1832 int prev_level
= bidi_it
->level_stack
[bidi_it
->stack_idx
].level
;
1833 bidi_type_t type
= bidi_resolve_weak (bidi_it
);
1834 int current_level
= bidi_it
->level_stack
[bidi_it
->stack_idx
].level
;
1836 if (!(type
== STRONG_R
1841 || type
== NEUTRAL_B
1842 || type
== NEUTRAL_S
1843 || type
== NEUTRAL_WS
1844 || type
== NEUTRAL_ON
))
1847 if (bidi_get_category (type
) == NEUTRAL
1848 || (type
== WEAK_BN
&& prev_level
== current_level
))
1850 if (bidi_it
->next_for_neutral
.type
!= UNKNOWN_BT
)
1851 type
= bidi_resolve_neutral_1 (bidi_it
->prev_for_neutral
.type
,
1852 bidi_it
->next_for_neutral
.type
,
1856 /* Arrrgh!! The UAX#9 algorithm is too deeply entrenched in
1857 the assumption of batch-style processing; see clauses W4,
1858 W5, and especially N1, which require to look far forward
1859 (as well as back) in the buffer/string. May the fleas of
1860 a thousand camels infest the armpits of those who design
1861 supposedly general-purpose algorithms by looking at their
1862 own implementations, and fail to consider other possible
1864 struct bidi_it saved_it
;
1865 bidi_type_t next_type
;
1867 if (bidi_it
->scan_dir
== -1)
1870 bidi_copy_it (&saved_it
, bidi_it
);
1871 /* Scan the text forward until we find the first non-neutral
1872 character, and then use that to resolve the neutral we
1873 are dealing with now. We also cache the scanned iterator
1874 states, to salvage some of the effort later. */
1875 bidi_cache_iterator_state (bidi_it
, 0);
1877 /* Record the info about the previous character, so that
1878 it will be cached below with this state. */
1879 if (bidi_it
->type_after_w1
!= WEAK_BN
/* W1/Retaining */
1880 && bidi_it
->type
!= WEAK_BN
)
1881 bidi_remember_char (&bidi_it
->prev
, bidi_it
);
1882 type
= bidi_resolve_weak (bidi_it
);
1883 /* Paragraph separators have their levels fully resolved
1884 at this point, so cache them as resolved. */
1885 bidi_cache_iterator_state (bidi_it
, type
== NEUTRAL_B
);
1886 /* FIXME: implement L1 here, by testing for a newline and
1887 resetting the level for any sequence of whitespace
1888 characters adjacent to it. */
1889 } while (!(type
== NEUTRAL_B
1891 && bidi_get_category (type
) != NEUTRAL
)
1892 /* This is all per level run, so stop when we
1893 reach the end of this level run. */
1894 || (bidi_it
->level_stack
[bidi_it
->stack_idx
].level
1895 != current_level
)));
1897 bidi_remember_char (&saved_it
.next_for_neutral
, bidi_it
);
1908 /* N1: ``European and Arabic numbers are treated as
1909 though they were R.'' */
1910 next_type
= STRONG_R
;
1911 saved_it
.next_for_neutral
.type
= STRONG_R
;
1914 if (!bidi_explicit_dir_char (bidi_it
->ch
))
1915 abort (); /* can't happen: BNs are skipped */
1918 /* Marched all the way to the end of this level run.
1919 We need to use the eor type, whose information is
1920 stored by bidi_set_sor_type in the prev_for_neutral
1922 if (saved_it
.type
!= WEAK_BN
1923 || bidi_get_category (bidi_it
->prev
.type_after_w1
) == NEUTRAL
)
1925 next_type
= bidi_it
->prev_for_neutral
.type
;
1926 saved_it
.next_for_neutral
.type
= next_type
;
1927 bidi_check_type (next_type
);
1931 /* This is a BN which does not adjoin neutrals.
1932 Leave its type alone. */
1933 bidi_copy_it (bidi_it
, &saved_it
);
1934 return bidi_it
->type
;
1940 type
= bidi_resolve_neutral_1 (saved_it
.prev_for_neutral
.type
,
1941 next_type
, current_level
);
1942 saved_it
.type
= type
;
1943 bidi_check_type (type
);
1944 bidi_copy_it (bidi_it
, &saved_it
);
1950 /* Given an iterator state in BIDI_IT, advance one character position
1951 in the buffer/string to the next character (in the logical order),
1952 resolve the bidi type of that next character, and return that
1955 bidi_type_of_next_char (struct bidi_it
*bidi_it
)
1959 /* This should always be called during a forward scan. */
1960 if (bidi_it
->scan_dir
!= 1)
1963 /* Reset the limit until which to ignore BNs if we step out of the
1964 area where we found only empty levels. */
1965 if ((bidi_it
->ignore_bn_limit
> -1
1966 && bidi_it
->ignore_bn_limit
<= bidi_it
->charpos
)
1967 || (bidi_it
->ignore_bn_limit
== -2
1968 && !bidi_explicit_dir_char (bidi_it
->ch
)))
1969 bidi_it
->ignore_bn_limit
= -1;
1971 type
= bidi_resolve_neutral (bidi_it
);
1976 /* Given an iterator state BIDI_IT, advance one character position in
1977 the buffer/string to the next character (in the current scan
1978 direction), resolve the embedding and implicit levels of that next
1979 character, and return the resulting level. */
1981 bidi_level_of_next_char (struct bidi_it
*bidi_it
)
1984 int level
, prev_level
= -1;
1985 struct bidi_saved_info next_for_neutral
;
1986 EMACS_INT next_char_pos
= -2;
1988 if (bidi_it
->scan_dir
== 1)
1991 = ((bidi_it
->string
.s
|| STRINGP (bidi_it
->string
.lstring
))
1992 ? bidi_it
->string
.schars
: ZV
);
1994 /* There's no sense in trying to advance if we hit end of text. */
1995 if (bidi_it
->charpos
>= eob
)
1996 return bidi_it
->resolved_level
;
1998 /* Record the info about the previous character. */
1999 if (bidi_it
->type_after_w1
!= WEAK_BN
/* W1/Retaining */
2000 && bidi_it
->type
!= WEAK_BN
)
2001 bidi_remember_char (&bidi_it
->prev
, bidi_it
);
2002 if (bidi_it
->type_after_w1
== STRONG_R
2003 || bidi_it
->type_after_w1
== STRONG_L
2004 || bidi_it
->type_after_w1
== STRONG_AL
)
2005 bidi_remember_char (&bidi_it
->last_strong
, bidi_it
);
2006 /* FIXME: it sounds like we don't need both prev and
2007 prev_for_neutral members, but I'm leaving them both for now. */
2008 if (bidi_it
->type
== STRONG_R
|| bidi_it
->type
== STRONG_L
2009 || bidi_it
->type
== WEAK_EN
|| bidi_it
->type
== WEAK_AN
)
2010 bidi_remember_char (&bidi_it
->prev_for_neutral
, bidi_it
);
2012 /* If we overstepped the characters used for resolving neutrals
2013 and whitespace, invalidate their info in the iterator. */
2014 if (bidi_it
->charpos
>= bidi_it
->next_for_neutral
.charpos
)
2015 bidi_it
->next_for_neutral
.type
= UNKNOWN_BT
;
2016 if (bidi_it
->next_en_pos
>= 0
2017 && bidi_it
->charpos
>= bidi_it
->next_en_pos
)
2018 bidi_it
->next_en_pos
= -1;
2019 if (bidi_it
->next_for_ws
.type
!= UNKNOWN_BT
2020 && bidi_it
->charpos
>= bidi_it
->next_for_ws
.charpos
)
2021 bidi_it
->next_for_ws
.type
= UNKNOWN_BT
;
2023 /* This must be taken before we fill the iterator with the info
2024 about the next char. If we scan backwards, the iterator
2025 state must be already cached, so there's no need to know the
2026 embedding level of the previous character, since we will be
2027 returning to our caller shortly. */
2028 prev_level
= bidi_it
->level_stack
[bidi_it
->stack_idx
].level
;
2030 next_for_neutral
= bidi_it
->next_for_neutral
;
2032 /* Perhaps the character we want is already cached. If it is, the
2033 call to bidi_cache_find below will return a type other than
2035 if (bidi_cache_idx
> bidi_cache_start
&& !bidi_it
->first_elt
)
2037 int bob
= ((bidi_it
->string
.s
|| STRINGP (bidi_it
->string
.lstring
))
2039 if (bidi_it
->scan_dir
> 0)
2041 if (bidi_it
->nchars
<= 0)
2043 next_char_pos
= bidi_it
->charpos
+ bidi_it
->nchars
;
2045 else if (bidi_it
->charpos
>= bob
)
2046 /* Implementation note: we allow next_char_pos to be as low as
2047 0 for buffers or -1 for strings, and that is okay because
2048 that's the "position" of the sentinel iterator state we
2049 cached at the beginning of the iteration. */
2050 next_char_pos
= bidi_it
->charpos
- 1;
2051 if (next_char_pos
>= bob
- 1)
2052 type
= bidi_cache_find (next_char_pos
, -1, bidi_it
);
2058 if (type
!= UNKNOWN_BT
)
2060 /* Don't lose the information for resolving neutrals! The
2061 cached states could have been cached before their
2062 next_for_neutral member was computed. If we are on our way
2063 forward, we can simply take the info from the previous
2065 if (bidi_it
->scan_dir
== 1
2066 && bidi_it
->next_for_neutral
.type
== UNKNOWN_BT
)
2067 bidi_it
->next_for_neutral
= next_for_neutral
;
2069 /* If resolved_level is -1, it means this state was cached
2070 before it was completely resolved, so we cannot return
2072 if (bidi_it
->resolved_level
!= -1)
2073 return bidi_it
->resolved_level
;
2075 if (bidi_it
->scan_dir
== -1)
2076 /* If we are going backwards, the iterator state is already cached
2077 from previous scans, and should be fully resolved. */
2080 if (type
== UNKNOWN_BT
)
2081 type
= bidi_type_of_next_char (bidi_it
);
2083 if (type
== NEUTRAL_B
)
2084 return bidi_it
->resolved_level
;
2086 level
= bidi_it
->level_stack
[bidi_it
->stack_idx
].level
;
2087 if ((bidi_get_category (type
) == NEUTRAL
/* && type != NEUTRAL_B */)
2088 || (type
== WEAK_BN
&& prev_level
== level
))
2090 if (bidi_it
->next_for_neutral
.type
== UNKNOWN_BT
)
2093 /* If the cached state shows a neutral character, it was not
2094 resolved by bidi_resolve_neutral, so do it now. */
2095 type
= bidi_resolve_neutral_1 (bidi_it
->prev_for_neutral
.type
,
2096 bidi_it
->next_for_neutral
.type
,
2100 if (!(type
== STRONG_R
2104 || type
== WEAK_AN
))
2106 bidi_it
->type
= type
;
2107 bidi_check_type (bidi_it
->type
);
2109 /* For L1 below, we need to know, for each WS character, whether
2110 it belongs to a sequence of WS characters preceding a newline
2111 or a TAB or a paragraph separator. */
2112 if (bidi_it
->orig_type
== NEUTRAL_WS
2113 && bidi_it
->next_for_ws
.type
== UNKNOWN_BT
)
2116 EMACS_INT clen
= bidi_it
->ch_len
;
2117 EMACS_INT bpos
= bidi_it
->bytepos
;
2118 EMACS_INT cpos
= bidi_it
->charpos
;
2119 EMACS_INT disp_pos
= bidi_it
->disp_pos
;
2120 EMACS_INT nc
= bidi_it
->nchars
;
2121 struct bidi_string_data bs
= bidi_it
->string
;
2123 int fwp
= bidi_it
->frame_window_p
;
2124 int dpp
= bidi_it
->disp_prop
;
2126 if (bidi_it
->nchars
<= 0)
2129 ch
= bidi_fetch_char (bpos
+= clen
, cpos
+= nc
, &disp_pos
, &dpp
, &bs
,
2131 if (ch
== '\n' || ch
== BIDI_EOB
/* || ch == LINESEP_CHAR */)
2134 chtype
= bidi_get_type (ch
, NEUTRAL_DIR
);
2135 } while (chtype
== NEUTRAL_WS
|| chtype
== WEAK_BN
2136 || bidi_explicit_dir_char (ch
)); /* L1/Retaining */
2137 bidi_it
->next_for_ws
.type
= chtype
;
2138 bidi_check_type (bidi_it
->next_for_ws
.type
);
2139 bidi_it
->next_for_ws
.charpos
= cpos
;
2140 bidi_it
->next_for_ws
.bytepos
= bpos
;
2143 /* Resolve implicit levels, with a twist: PDFs get the embedding
2144 level of the enbedding they terminate. See below for the
2146 if (bidi_it
->orig_type
== PDF
2147 /* Don't do this if this formatting code didn't change the
2148 embedding level due to invalid or empty embeddings. */
2149 && prev_level
!= level
)
2151 /* Don't look in UAX#9 for the reason for this: it's our own
2152 private quirk. The reason is that we want the formatting
2153 codes to be delivered so that they bracket the text of their
2154 embedding. For example, given the text
2158 we want it to be displayed as
2166 which will result because we bump up the embedding level as
2167 soon as we see the RLO and pop it as soon as we see the PDF,
2168 so RLO itself has the same embedding level as "teST", and
2169 thus would be normally delivered last, just before the PDF.
2170 The switch below fiddles with the level of PDF so that this
2171 ugly side effect does not happen.
2173 (This is, of course, only important if the formatting codes
2174 are actually displayed, but Emacs does need to display them
2175 if the user wants to.) */
2178 else if (bidi_it
->orig_type
== NEUTRAL_B
/* L1 */
2179 || bidi_it
->orig_type
== NEUTRAL_S
2180 || bidi_it
->ch
== '\n' || bidi_it
->ch
== BIDI_EOB
2181 /* || bidi_it->ch == LINESEP_CHAR */
2182 || (bidi_it
->orig_type
== NEUTRAL_WS
2183 && (bidi_it
->next_for_ws
.type
== NEUTRAL_B
2184 || bidi_it
->next_for_ws
.type
== NEUTRAL_S
)))
2185 level
= bidi_it
->level_stack
[0].level
;
2186 else if ((level
& 1) == 0) /* I1 */
2188 if (type
== STRONG_R
)
2190 else if (type
== WEAK_EN
|| type
== WEAK_AN
)
2195 if (type
== STRONG_L
|| type
== WEAK_EN
|| type
== WEAK_AN
)
2199 bidi_it
->resolved_level
= level
;
2203 /* Move to the other edge of a level given by LEVEL. If END_FLAG is
2204 non-zero, we are at the end of a level, and we need to prepare to
2205 resume the scan of the lower level.
2207 If this level's other edge is cached, we simply jump to it, filling
2208 the iterator structure with the iterator state on the other edge.
2209 Otherwise, we walk the buffer or string until we come back to the
2210 same level as LEVEL.
2212 Note: we are not talking here about a ``level run'' in the UAX#9
2213 sense of the term, but rather about a ``level'' which includes
2214 all the levels higher than it. In other words, given the levels
2217 11111112222222333333334443343222222111111112223322111
2220 and assuming we are at point A scanning left to right, this
2221 function moves to point C, whereas the UAX#9 ``level 2 run'' ends
2224 bidi_find_other_level_edge (struct bidi_it
*bidi_it
, int level
, int end_flag
)
2226 int dir
= end_flag
? -bidi_it
->scan_dir
: bidi_it
->scan_dir
;
2229 /* Try the cache first. */
2230 if ((idx
= bidi_cache_find_level_change (level
, dir
, end_flag
))
2231 >= bidi_cache_start
)
2232 bidi_cache_fetch_state (idx
, bidi_it
);
2238 abort (); /* if we are at end of level, its edges must be cached */
2240 bidi_cache_iterator_state (bidi_it
, 1);
2242 new_level
= bidi_level_of_next_char (bidi_it
);
2243 bidi_cache_iterator_state (bidi_it
, 1);
2244 } while (new_level
>= level
);
2249 bidi_move_to_visually_next (struct bidi_it
*bidi_it
)
2251 int old_level
, new_level
, next_level
;
2252 struct bidi_it sentinel
;
2253 struct gcpro gcpro1
;
2255 if (bidi_it
->charpos
< 0 || bidi_it
->bytepos
< 0)
2258 if (bidi_it
->scan_dir
== 0)
2260 bidi_it
->scan_dir
= 1; /* default to logical order */
2263 /* The code below can call eval, and thus cause GC. If we are
2264 iterating a Lisp string, make sure it won't be GCed. */
2265 if (STRINGP (bidi_it
->string
.lstring
))
2266 GCPRO1 (bidi_it
->string
.lstring
);
2268 /* If we just passed a newline, initialize for the next line. */
2269 if (!bidi_it
->first_elt
2270 && (bidi_it
->ch
== '\n' || bidi_it
->ch
== BIDI_EOB
))
2271 bidi_line_init (bidi_it
);
2273 /* Prepare the sentinel iterator state, and cache it. When we bump
2274 into it, scanning backwards, we'll know that the last non-base
2275 level is exhausted. */
2276 if (bidi_cache_idx
== bidi_cache_start
)
2278 bidi_copy_it (&sentinel
, bidi_it
);
2279 if (bidi_it
->first_elt
)
2281 sentinel
.charpos
--; /* cached charpos needs to be monotonic */
2283 sentinel
.ch
= '\n'; /* doesn't matter, but why not? */
2284 sentinel
.ch_len
= 1;
2285 sentinel
.nchars
= 1;
2287 bidi_cache_iterator_state (&sentinel
, 1);
2290 old_level
= bidi_it
->resolved_level
;
2291 new_level
= bidi_level_of_next_char (bidi_it
);
2293 /* Reordering of resolved levels (clause L2) is implemented by
2294 jumping to the other edge of the level and flipping direction of
2295 scanning the text whenever we find a level change. */
2296 if (new_level
!= old_level
)
2298 int ascending
= new_level
> old_level
;
2299 int level_to_search
= ascending
? old_level
+ 1 : old_level
;
2300 int incr
= ascending
? 1 : -1;
2301 int expected_next_level
= old_level
+ incr
;
2303 /* Jump (or walk) to the other edge of this level. */
2304 bidi_find_other_level_edge (bidi_it
, level_to_search
, !ascending
);
2305 /* Switch scan direction and peek at the next character in the
2307 bidi_it
->scan_dir
= -bidi_it
->scan_dir
;
2309 /* The following loop handles the case where the resolved level
2310 jumps by more than one. This is typical for numbers inside a
2311 run of text with left-to-right embedding direction, but can
2312 also happen in other situations. In those cases the decision
2313 where to continue after a level change, and in what direction,
2314 is tricky. For example, given a text like below:
2319 (where the numbers below the text show the resolved levels),
2320 the result of reordering according to UAX#9 should be this:
2324 This is implemented by the loop below which flips direction
2325 and jumps to the other edge of the level each time it finds
2326 the new level not to be the expected one. The expected level
2327 is always one more or one less than the previous one. */
2328 next_level
= bidi_peek_at_next_level (bidi_it
);
2329 while (next_level
!= expected_next_level
)
2331 expected_next_level
+= incr
;
2332 level_to_search
+= incr
;
2333 bidi_find_other_level_edge (bidi_it
, level_to_search
, !ascending
);
2334 bidi_it
->scan_dir
= -bidi_it
->scan_dir
;
2335 next_level
= bidi_peek_at_next_level (bidi_it
);
2338 /* Finally, deliver the next character in the new direction. */
2339 next_level
= bidi_level_of_next_char (bidi_it
);
2342 /* Take note when we have just processed the newline that precedes
2343 the end of the paragraph. The next time we are about to be
2344 called, set_iterator_to_next will automatically reinit the
2345 paragraph direction, if needed. We do this at the newline before
2346 the paragraph separator, because the next character might not be
2347 the first character of the next paragraph, due to the bidi
2348 reordering, whereas we _must_ know the paragraph base direction
2349 _before_ we process the paragraph's text, since the base
2350 direction affects the reordering. */
2351 if (bidi_it
->scan_dir
== 1
2352 && (bidi_it
->ch
== '\n' || bidi_it
->ch
== BIDI_EOB
))
2354 /* The paragraph direction of the entire string, once
2355 determined, is in effect for the entire string. Setting the
2356 separator limit to the end of the string prevents
2357 bidi_paragraph_init from being called automatically on this
2359 if (bidi_it
->string
.s
|| STRINGP (bidi_it
->string
.lstring
))
2360 bidi_it
->separator_limit
= bidi_it
->string
.schars
;
2361 else if (bidi_it
->bytepos
< ZV_BYTE
)
2364 = bidi_at_paragraph_end (bidi_it
->charpos
+ bidi_it
->nchars
,
2365 bidi_it
->bytepos
+ bidi_it
->ch_len
);
2366 if (bidi_it
->nchars
<= 0)
2370 bidi_it
->new_paragraph
= 1;
2371 /* Record the buffer position of the last character of the
2372 paragraph separator. */
2373 bidi_it
->separator_limit
2374 = bidi_it
->charpos
+ bidi_it
->nchars
+ sep_len
;
2379 if (bidi_it
->scan_dir
== 1 && bidi_cache_idx
> bidi_cache_start
)
2381 /* If we are at paragraph's base embedding level and beyond the
2382 last cached position, the cache's job is done and we can
2384 if (bidi_it
->resolved_level
== bidi_it
->level_stack
[0].level
2385 && bidi_it
->charpos
> (bidi_cache
[bidi_cache_idx
- 1].charpos
2386 + bidi_cache
[bidi_cache_idx
- 1].nchars
- 1))
2387 bidi_cache_reset ();
2388 /* But as long as we are caching during forward scan, we must
2389 cache each state, or else the cache integrity will be
2390 compromised: it assumes cached states correspond to buffer
2393 bidi_cache_iterator_state (bidi_it
, 1);
2396 if (STRINGP (bidi_it
->string
.lstring
))
2400 /* This is meant to be called from within the debugger, whenever you
2401 wish to examine the cache contents. */
2402 void bidi_dump_cached_states (void) EXTERNALLY_VISIBLE
;
2404 bidi_dump_cached_states (void)
2409 if (bidi_cache_idx
== 0)
2411 fprintf (stderr
, "The cache is empty.\n");
2414 fprintf (stderr
, "Total of %"pD
"d state%s in cache:\n",
2415 bidi_cache_idx
, bidi_cache_idx
== 1 ? "" : "s");
2417 for (i
= bidi_cache
[bidi_cache_idx
- 1].charpos
; i
> 0; i
/= 10)
2419 fputs ("ch ", stderr
);
2420 for (i
= 0; i
< bidi_cache_idx
; i
++)
2421 fprintf (stderr
, "%*c", ndigits
, bidi_cache
[i
].ch
);
2422 fputs ("\n", stderr
);
2423 fputs ("lvl ", stderr
);
2424 for (i
= 0; i
< bidi_cache_idx
; i
++)
2425 fprintf (stderr
, "%*d", ndigits
, bidi_cache
[i
].resolved_level
);
2426 fputs ("\n", stderr
);
2427 fputs ("pos ", stderr
);
2428 for (i
= 0; i
< bidi_cache_idx
; i
++)
2429 fprintf (stderr
, "%*"pI
"d", ndigits
, bidi_cache
[i
].charpos
);
2430 fputs ("\n", stderr
);