| blob | d4c7d2451d7abd1928b6c94a9ad339dc92ffe71d |
1 /* Low-level bidirectional buffer-scanning functions for GNU Emacs.
2 Copyright (C) 2000, 2001, 2004, 2005, 2009, 2010
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 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
27 string.
29 The main entry point is bidi_get_next_char_visually. 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_get_next_char_visually for the next
34 character. See the comments in bidi_get_next_char_visually for
35 more details about its algorithm that finds the next visual-order
36 character by resolving their levels on the fly.
38 The two other entry points are bidi_paragraph_init and
39 bidi_mirror_char. The first determines the base direction of a
40 paragraph, while the second returns the mirrored version of its
41 argument character.
43 If you want to understand the code, you will have to read it
44 together with the relevant portions of UAX#9. The comments include
45 references to UAX#9 rules, for that very reason.
47 A note about references to UAX#9 rules: if the reference says
48 something like "X9/Retaining", it means that you need to refer to
49 rule X9 and to its modifications decribed in the "Implementation
50 Notes" section of UAX#9, under "Retaining Format Codes". */
52 #ifdef HAVE_CONFIG_H
53 #include <config.h>
54 #endif
56 #include <stdio.h>
58 #ifdef HAVE_STRING_H
59 #include <string.h>
60 #endif
62 #include <setjmp.h>
73 /* FIXME: Remove these when bidi_explicit_dir_char uses a lookup table. */
74 #define LRM_CHAR 0x200E
75 #define RLM_CHAR 0x200F
76 #define LRE_CHAR 0x202A
77 #define RLE_CHAR 0x202B
78 #define PDF_CHAR 0x202C
79 #define LRO_CHAR 0x202D
80 #define RLO_CHAR 0x202E
82 #define BIDI_EOB -1
85 /* Local data structures. (Look in dispextern.h for the rest.) */
87 /* What we need to know about the current paragraph. */
93 };
95 /* Data type for describing the bidirectional character categories. */
97 UNKNOWN_BC,
98 NEUTRAL,
99 WEAK,
100 STRONG
108 static void
110 {
111 /* FIXME: This should come from the Unicode Database. */
114 bidi_type_t type;
415 }
417 /* Return the bidi type of a character CH, subject to the current
418 directional OVERRIDE. */
419 static INLINE bidi_type_t
421 {
422 bidi_type_t default_type;
435 {
436 /* Although UAX#9 does not tell, it doesn't make sense to
437 override NEUTRAL_B and LRM/RLM characters. */
447 {
456 else
458 }
459 }
460 }
462 void
464 {
467 }
469 /* Given a bidi TYPE of a character, return its category. */
470 static INLINE bidi_category_t
472 {
474 {
501 }
502 }
504 /* Return the mirrored character of C, if any.
506 Note: The conditions in UAX#9 clause L4 must be tested by the
507 caller. */
508 /* FIXME: exceedingly temporary! Should consult the Unicode database
509 of character properties. */
510 int
512 {
517 {
521 }
523 }
525 /* Copy the bidi iterator from FROM to TO. To save cycles, this only
526 copies the part of the level stack that is actually in use. */
529 {
532 /* Copy everything except the level stack and beyond. */
535 /* Copy the active part of the level stack. */
539 }
541 /* Caching the bidi iterator states. */
549 {
552 }
556 {
565 }
567 /* Find a cached state with a given CHARPOS and resolved embedding
568 level less or equal to LEVEL. if LEVEL is -1, disregard the
569 resolved levels in cached states. DIR, if non-zero, means search
570 in that direction from the last cache hit. */
573 {
577 {
584 else
585 {
588 }
591 {
592 /* Linear search for now; FIXME! */
597 }
598 else
599 {
604 }
605 }
608 }
610 /* Find a cached state where the resolved level changes to a value
611 that is lower than LEVEL, and return its cache slot index. DIR is
612 the direction to search, starting with the last used cache slot.
613 BEFORE, if non-zero, means return the index of the slot that is
614 ``before'' the level change in the search direction. That is,
615 given the cached levels like this:
617 1122333442211
618 AB C
620 and assuming we are at the position cached at the slot marked with
621 C, searching backwards (DIR = -1) for LEVEL = 2 will return the
622 index of slot B or A, depending whether BEFORE is, respectively,
623 non-zero or zero. */
624 static int
626 {
628 {
638 {
640 {
644 i--;
645 }
646 }
647 else
648 {
650 {
654 i++;
655 }
656 }
657 }
660 }
664 {
667 /* We should never cache on backward scans. */
673 {
675 /* Don't overrun the cache limit. */
678 /* Character positions should correspond to cache positions 1:1.
679 If we are outside the range of cached positions, the cache is
680 useless and must be reset. */
684 {
687 }
692 }
693 else
694 {
695 /* Copy only the members which could have changed, to avoid
696 costly copying of the entire struct. */
703 else
710 }
715 }
717 static INLINE bidi_type_t
719 {
723 {
728 /* Don't let scan direction from from the cached state override
729 the current scan direction. */
732 }
735 }
739 {
743 }
745 /* Check if buffer position CHARPOS/BYTEPOS is the end of a paragraph.
746 Value is the non-negative length of the paragraph separator
747 following the buffer position, -1 if position is at the beginning
748 of a new paragraph, or -2 if position is neither at beginning nor
749 at end of a paragraph. */
750 static EMACS_INT
752 {
753 /* FIXME: Why Fbuffer_local_value rather than just Fsymbol_value? */
754 Lisp_Object sep_re;
755 Lisp_Object start_re;
756 EMACS_INT val;
763 {
766 else
768 }
771 }
773 /* Determine the start-of-run (sor) directional type given the two
774 embedding levels on either side of the run boundary. Also, update
775 the saved info about previously seen characters, since that info is
776 generally valid for a single level run. */
779 {
782 /* The prev_was_pdf gork is required for when we have several PDFs
783 in a row. In that case, we want to compute the sor type for the
784 next level run only once: when we see the first PDF. That's
785 because the sor type depends only on the higher of the two levels
786 that we find on the two sides of the level boundary (see UAX#9,
787 clause X10), and so we don't need to know the final embedding
788 level to which we descend after processing all the PDFs. */
790 /* FIXME: should the default sor direction be user selectable? */
804 }
806 static void
808 {
821 }
823 /* Find the beginning of this paragraph by looking back in the buffer.
824 Value is the byte position of the paragraph's beginning. */
825 static EMACS_INT
827 {
833 {
836 }
838 }
840 /* Determine the direction, a.k.a. base embedding level, of the
841 paragraph we are about to iterate through. If DIR is either L2R or
842 R2L, just use that. Otherwise, determine the paragraph direction
843 from the first strong character of the paragraph.
845 Note that this gives the paragraph separator the same direction as
846 the preceding paragraph, even though Emacs generally views the
847 separartor as not belonging to any paragraph. */
848 void
850 {
853 /* Special case for an empty buffer. */
856 /* We should never be called at EOB or before BEGV. */
861 {
864 }
866 {
869 }
871 {
873 EMACS_INT pos;
874 bidi_type_t type;
876 /* If we are inside a paragraph separator, we are just waiting
877 for the separator to be exhausted; use the previous paragraph
878 direction. But don't do that if we have been just reseated,
879 because we need to reinitialize below in that case. */
884 /* If we are on a newline, get past it to where the next
885 paragraph might start. But don't do that at BEGV since then
886 we are potentially in a new paragraph that doesn't yet
887 exist. */
890 {
891 bytepos++;
892 pos++;
893 }
895 /* We are either at the beginning of a paragraph or in the
896 middle of it. Find where this paragraph starts. */
899 /* We should always be at the beginning of a new line at this
900 point. */
912 /* NOTE: UAX#9 says to search only for L, AL, or R types of
913 characters, and ignore RLE, RLO, LRE, and LRO. However,
914 I'm not sure it makes sense to omit those 4; should try
915 with and without that to see the effect. */
917 || (bidi_ignore_explicit_marks_for_paragraph_level
921 {
925 {
926 /* Pretend there's a paragraph separator at end of buffer. */
929 }
931 }
936 }
937 else
940 /* Contrary to UAX#9 clause P3, we only default the paragraph
941 direction to L2R if we have no previous usable paragraph
942 direction. */
947 else
951 }
953 /* Do whatever UAX#9 clause X8 says should be done at paragraph's
954 end. */
957 {
962 }
964 /* Initialize the bidi iterator from buffer position CHARPOS. */
965 void
967 {
993 }
995 /* Push the current embedding level and override status; reset the
996 current level to LEVEL and the current override status to OVERRIDE. */
1000 {
1006 }
1008 /* Pop the embedding level and directional override status from the
1009 stack, and return the new level. */
1012 {
1013 /* UAX#9 says to ignore invalid PDFs. */
1017 }
1019 /* Record in SAVED_INFO the information about the current character. */
1023 {
1032 }
1034 /* Resolve the type of a neutral character according to the type of
1035 surrounding strong text and the current embedding level. */
1036 static INLINE bidi_type_t
1038 {
1039 /* N1: European and Arabic numbers are treated as though they were R. */
1049 else
1051 }
1055 {
1056 /* FIXME: this should be replaced with a lookup table with suitable
1057 bits set, like standard C ctype macros do. */
1060 }
1062 /* A helper function for bidi_resolve_explicit. It advances to the
1063 next character in logical order and determines the new embedding
1064 level and directional override, but does not take into account
1065 empty embeddings. */
1066 static int
1068 {
1070 bidi_type_t type;
1073 bidi_dir_t override;
1077 {
1082 }
1084 {
1089 }
1095 /* in case it is a unibyte character (not yet implemented) */
1096 /* _fetch_multibyte_char_len = 1; */
1098 {
1101 }
1102 else
1103 {
1106 }
1109 /* Don't apply directional override here, as all the types we handle
1110 below will not be affected by the override anyway, and we need
1111 the original type unaltered. The override will be applied in
1112 bidi_resolve_weak. */
1123 {
1130 {
1132 {
1133 /* Compute the least odd embedding level greater than
1134 the current level. */
1138 else
1143 }
1144 else
1145 {
1147 /* See the commentary about invalid_rl_levels below. */
1151 }
1152 }
1163 {
1165 {
1166 /* Compute the least even embedding level greater than
1167 the current level. */
1171 else
1174 }
1175 else
1176 {
1178 /* invalid_rl_levels counts invalid levels encountered
1179 while the embedding level was already too high for
1180 LRE/LRO, but not for RLE/RLO. That is because
1181 there may be exactly one PDF which we should not
1182 ignore even though invalid_levels is non-zero.
1183 invalid_rl_levels helps to know what PDF is
1184 that. */
1187 }
1188 }
1198 {
1200 {
1205 /* else nothing: UAX#9 says to ignore invalid PDFs */
1206 }
1209 else
1210 {
1213 }
1214 }
1220 /* Nothing. */
1222 }
1228 }
1230 /* Given an iterator state in BIDI_IT, advance one character position
1231 in the buffer to the next character (in the logical order), resolve
1232 any explicit embeddings and directional overrides, and return the
1233 embedding level of the character after resolving explicit
1234 directives and ignoring empty embeddings. */
1235 static int
1237 {
1247 {
1248 /* Avoid pushing and popping embedding levels if the level run
1249 is empty, as this breaks level runs where it shouldn't.
1250 UAX#9 removes all the explicit embedding and override codes,
1251 so empty embeddings disappear without a trace. We need to
1252 behave as if we did the same. */
1260 {
1262 }
1271 {
1272 /* We pushed a level, but we shouldn't have. Undo that. */
1274 {
1279 }
1282 else
1283 {
1286 }
1287 }
1288 }
1291 {
1293 /* This is needed by bidi_resolve_weak below, and in L1. */
1296 }
1299 }
1301 /* Advance in the buffer, resolve weak types and return the type of
1302 the next character after weak type resolution. */
1303 static bidi_type_t
1305 {
1306 bidi_type_t type;
1307 bidi_dir_t override;
1311 bidi_type_t type_of_next;
1327 {
1328 /* We've got a new embedding level run, compute the directional
1329 type of sor and initialize per-run variables (UAX#9, clause
1330 X10). */
1332 }
1338 /* Level and directional override status are already recorded in
1339 bidi_it, and do not need any change; see X6. */
1344 else
1345 {
1347 {
1348 /* Note that we don't need to consider the case where the
1349 prev character has its type overridden by an RLO or LRO,
1350 because then either the type of this NSM would have been
1351 also overridden, or the previous character is outside the
1352 current level run, and thus not relevant to this NSM.
1353 This is why NSM gets the type_after_w1 of the previous
1354 character. */
1356 /* if type_after_w1 is NEUTRAL_B, this NSM is at sor */
1365 }
1378 {
1379 next_char =
1386 {
1390 ;
1393 }
1395 /* If the next character is EN, but the last strong-type
1396 character is AL, that next EN will be changed to AN when
1397 we process it in W2 above. So in that case, this ES
1398 should not be changed into EN. */
1404 {
1407 /* If the next character is EN, but the last
1408 strong-type character is AL, EN will be later
1409 changed to AN when we process it in W2 above.
1410 So in that case, this ES should not be
1411 changed into EN. */
1419 }
1420 }
1423 {
1428 {
1431 next_char =
1439 {
1444 ;
1448 }
1450 {
1451 /* If the last strong character is AL, the EN we've
1452 found will become AN when we get to it (W2). */
1454 {
1456 /* Remember this EN position, to speed up processing
1457 of the next ETs. */
1459 }
1462 }
1463 }
1464 }
1465 }
1474 /* Store the type we've got so far, before we clobber it with strong
1475 types in W7 and while resolving neutral types. But leave alone
1476 the original types that were recorded above, because we will need
1477 them for the L1 clause. */
1483 {
1487 }
1492 }
1494 static bidi_type_t
1496 {
1514 {
1518 current_level);
1519 else
1520 {
1521 /* Arrrgh!! The UAX#9 algorithm is too deeply entrenched in
1522 the assumption of batch-style processing; see clauses W4,
1523 W5, and especially N1, which require to look far forward
1524 (as well as back) in the buffer. May the fleas of a
1525 thousand camels infest the armpits of those who design
1526 supposedly general-purpose algorithms by looking at their
1527 own implementations, and fail to consider other possible
1528 implementations! */
1530 bidi_type_t next_type;
1536 /* Scan the text forward until we find the first non-neutral
1537 character, and then use that to resolve the neutral we
1538 are dealing with now. We also cache the scanned iterator
1539 states, to salvage some of the effort later. */
1542 /* Record the info about the previous character, so that
1543 it will be cached below with this state. */
1548 /* Paragraph separators have their levels fully resolved
1549 at this point, so cache them as resolved. */
1551 /* FIXME: implement L1 here, by testing for a newline and
1552 resetting the level for any sequence of whitespace
1553 characters adjacent to it. */
1557 /* This is all per level run, so stop when we
1558 reach the end of this level run. */
1560 current_level));
1565 {
1573 /* N1: ``European and Arabic numbers are treated as
1574 though they were R.'' */
1581 /* FALLTHROUGH */
1583 /* Marched all the way to the end of this level run.
1584 We need to use the eor type, whose information is
1585 stored by bidi_set_sor_type in the prev_for_neutral
1586 member. */
1589 {
1593 }
1594 else
1595 {
1596 /* This is a BN which does not adjoin neutrals.
1597 Leave its type alone. */
1600 }
1604 }
1610 }
1611 }
1613 }
1615 /* Given an iterator state in BIDI_IT, advance one character position
1616 in the buffer to the next character (in the logical order), resolve
1617 the bidi type of that next character, and return that type. */
1618 static bidi_type_t
1620 {
1621 bidi_type_t type;
1623 /* This should always be called during a forward scan. */
1627 /* Reset the limit until which to ignore BNs if we step out of the
1628 area where we found only empty levels. */
1638 }
1640 /* Given an iterator state BIDI_IT, advance one character position in
1641 the buffer to the next character (in the logical order), resolve
1642 the embedding and implicit levels of that next character, and
1643 return the resulting level. */
1644 static int
1646 {
1647 bidi_type_t type;
1652 {
1653 /* There's no sense in trying to advance if we hit end of text. */
1657 /* Record the info about the previous character. */
1665 /* FIXME: it sounds like we don't need both prev and
1666 prev_for_neutral members, but I'm leaving them both for now. */
1671 /* If we overstepped the characters used for resolving neutrals
1672 and whitespace, invalidate their info in the iterator. */
1682 /* This must be taken before we fill the iterator with the info
1683 about the next char. If we scan backwards, the iterator
1684 state must be already cached, so there's no need to know the
1685 embedding level of the previous character, since we will be
1686 returning to our caller shortly. */
1688 }
1691 /* Perhaps it is already cached. */
1694 {
1695 /* Don't lose the information for resolving neutrals! The
1696 cached states could have been cached before their
1697 next_for_neutral member was computed. If we are on our way
1698 forward, we can simply take the info from the previous
1699 state. */
1704 /* If resolved_level is -1, it means this state was cached
1705 before it was completely resolved, so we cannot return
1706 it. */
1709 }
1711 /* If we are going backwards, the iterator state is already cached
1712 from previous scans, and should be fully resolved. */
1724 {
1728 /* If the cached state shows a neutral character, it was not
1729 resolved by bidi_resolve_neutral, so do it now. */
1732 level);
1733 }
1744 /* For L1 below, we need to know, for each WS character, whether
1745 it belongs to a sequence of WS characters preceeding a newline
1746 or a TAB or a paragraph separator. */
1749 {
1754 bidi_type_t chtype;
1757 /*_fetch_multibyte_char_len = 1;*/
1760 cpos++;
1764 else
1772 }
1774 /* Resolve implicit levels, with a twist: PDFs get the embedding
1775 level of the enbedding they terminate. See below for the
1776 reason. */
1778 /* Don't do this if this formatting code didn't change the
1779 embedding level due to invalid or empty embeddings. */
1781 {
1782 /* Don't look in UAX#9 for the reason for this: it's our own
1783 private quirk. The reason is that we want the formatting
1784 codes to be delivered so that they bracket the text of their
1785 embedding. For example, given the text
1787 {RLO}teST{PDF}
1789 we want it to be displayed as
1791 {RLO}STet{PDF}
1793 not as
1795 STet{RLO}{PDF}
1797 which will result because we bump up the embedding level as
1798 soon as we see the RLO and pop it as soon as we see the PDF,
1799 so RLO itself has the same embedding level as "teST", and
1800 thus would be normally delivered last, just before the PDF.
1801 The switch below fiddles with the level of PDF so that this
1802 ugly side effect does not happen.
1804 (This is, of course, only important if the formatting codes
1805 are actually displayed, but Emacs does need to display them
1806 if the user wants to.) */
1808 }
1812 /* || bidi_it->ch == LINESEP_CHAR */
1818 {
1820 level++;
1823 }
1825 {
1827 level++;
1828 }
1832 }
1834 /* Move to the other edge of a level given by LEVEL. If END_FLAG is
1835 non-zero, we are at the end of a level, and we need to prepare to
1836 resume the scan of the lower level.
1838 If this level's other edge is cached, we simply jump to it, filling
1839 the iterator structure with the iterator state on the other edge.
1840 Otherwise, we walk the buffer until we come back to the same level
1841 as LEVEL.
1843 Note: we are not talking here about a ``level run'' in the UAX#9
1844 sense of the term, but rather about a ``level'' which includes
1845 all the levels higher than it. In other words, given the levels
1846 like this:
1848 11111112222222333333334443343222222111111112223322111
1849 A B C
1851 and assuming we are at point A scanning left to right, this
1852 function moves to point C, whereas the UAX#9 ``level 2 run'' ends
1853 at point B. */
1854 static void
1856 {
1860 /* Try the cache first. */
1863 else
1864 {
1875 }
1876 }
1878 void
1880 {
1885 {
1887 }
1889 /* If we just passed a newline, initialize for the next line. */
1893 /* Prepare the sentinel iterator state. */
1895 {
1898 {
1903 }
1904 }
1909 /* Reordering of resolved levels (clause L2) is implemented by
1910 jumping to the other edge of the level and flipping direction of
1911 scanning the text whenever we find a level change. */
1913 {
1919 /* If we don't have anything cached yet, we need to cache the
1920 sentinel state, since we'll need it to record where to jump
1921 when the last non-base level is exhausted. */
1924 /* Jump (or walk) to the other edge of this level. */
1926 /* Switch scan direction and peek at the next character in the
1927 new direction. */
1930 /* The following loop handles the case where the resolved level
1931 jumps by more than one. This is typical for numbers inside a
1932 run of text with left-to-right embedding direction, but can
1933 also happen in other situations. In those cases the decision
1934 where to continue after a level change, and in what direction,
1935 is tricky. For example, given a text like below:
1937 abcdefgh
1938 11336622
1940 (where the numbers below the text show the resolved levels),
1941 the result of reordering according to UAX#9 should be this:
1943 efdcghba
1945 This is implemented by the loop below which flips direction
1946 and jumps to the other edge of the level each time it finds
1947 the new level not to be the expected one. The expected level
1948 is always one more or one less than the previous one. */
1951 {
1957 }
1959 /* Finally, deliver the next character in the new direction. */
1961 }
1963 /* Take note when we have just processed the newline that precedes
1964 the end of the paragraph. The next time we are about to be
1965 called, set_iterator_to_next will automatically reinit the
1966 paragraph direction, if needed. We do this at the newline before
1967 the paragraph separator, because the next character might not be
1968 the first character of the next paragraph, due to the bidi
1969 reordering, whereas we _must_ know the paragraph base direction
1970 _before_ we process the paragraph's text, since the base
1971 direction affects the reordering. */
1975 {
1976 EMACS_INT sep_len =
1980 {
1982 /* Record the buffer position of the last character of the
1983 paragraph separator. */
1985 }
1986 }
1989 {
1990 /* If we are at paragraph's base embedding level and beyond the
1991 last cached position, the cache's job is done and we can
1992 discard it. */
1996 /* But as long as we are caching during forward scan, we must
1997 cache each state, or else the cache integrity will be
1998 compromised: it assumes cached states correspond to buffer
1999 positions 1:1. */
2000 else
2002 }
2003 }
2005 /* This is meant to be called from within the debugger, whenever you
2006 wish to examine the cache contents. */
2007 void
2009 {
2014 {
2017 }
2022 ndigits++;
2035 }