| blob | 7da8cd4a88cb4134cd8587ca79b348cf83088dc2 |
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.
28 The main entry point is bidi_get_next_char_visually. Each time it
29 is called, it finds the next character in the visual order, and
30 returns its information in a special structure. The caller is then
31 expected to process this character for display or any other
32 purposes, and call bidi_get_next_char_visually for the next
33 character. See the comments in bidi_get_next_char_visually for
34 more details about its algorithm that finds the next visual-order
35 character by resolving their levels on the fly.
37 If you want to understand the code, you will have to read it
38 together with the relevant portions of UAX#9. The comments include
39 references to UAX#9 rules, for that very reason.
41 A note about references to UAX#9 rules: if the reference says
42 something like "X9/Retaining", it means that you need to refer to
43 rule X9 and to its modifications decribed in the "Implementation
44 Notes" section of UAX#9, under "Retaining Format Codes". */
46 #ifdef HAVE_CONFIG_H
47 #include <config.h>
48 #endif
50 #include <stdio.h>
52 #ifdef HAVE_STRING_H
53 #include <string.h>
54 #endif
56 #include <setjmp.h>
67 /* FIXME: Remove these when bidi_explicit_dir_char uses a lookup table. */
68 #define LRM_CHAR 0x200E
69 #define RLM_CHAR 0x200F
70 #define LRE_CHAR 0x202A
71 #define RLE_CHAR 0x202B
72 #define PDF_CHAR 0x202C
73 #define LRO_CHAR 0x202D
74 #define RLO_CHAR 0x202E
76 #define BIDI_EOB -1
79 /* Local data structures. (Look in dispextern.h for the rest.) */
81 /* What we need to know about the current paragraph. */
87 };
89 /* Data type for describing the bidirectional character categories. */
91 UNKNOWN_BC,
92 NEUTRAL,
93 WEAK,
94 STRONG
102 static void
104 {
105 /* FIXME: This should come from the Unicode Database. */
108 bidi_type_t type;
396 fallback_paragraph_start_re =
403 fallback_paragraph_separate_re =
411 }
413 /* Return the bidi type of a character CH, subject to the current
414 directional OVERRIDE. */
415 bidi_type_t
417 {
418 bidi_type_t default_type;
431 {
432 /* Although UAX#9 does not tell, it doesn't make sense to
433 override NEUTRAL_B and LRM/RLM characters. */
443 {
452 else
454 }
455 }
456 }
458 void
460 {
463 }
465 /* Given a bidi TYPE of a character, return its category. */
466 bidi_category_t
468 {
470 {
497 }
498 }
500 /* Return the mirrored character of C, if any.
502 Note: The conditions in UAX#9 clause L4 must be tested by the
503 caller. */
504 /* FIXME: exceedingly temporary! Should consult the Unicode database
505 of character properties. */
506 int
508 {
513 {
517 }
519 }
521 /* Copy the bidi iterator from FROM to TO. To save cycles, this only
522 copies the part of the level stack that is actually in use. */
525 {
528 /* Copy everything except the level stack and beyond. */
531 /* Copy the active part of the level stack. */
535 }
537 /* Caching the bidi iterator states. */
545 {
548 }
552 {
561 }
563 /* Find a cached state with a given CHARPOS and resolved embedding
564 level less or equal to LEVEL. if LEVEL is -1, disregard the
565 resolved levels in cached states. DIR, if non-zero, means search
566 in that direction from the last cache hit. */
569 {
573 {
580 else
581 {
584 }
587 {
588 /* Linear search for now; FIXME! */
593 }
594 else
595 {
600 }
601 }
604 }
606 /* Find a cached state where the resolved level changes to a value
607 that is lower than LEVEL, and return its cache slot index. DIR is
608 the direction to search, starting with the last used cache slot.
609 BEFORE, if non-zero, means return the index of the slot that is
610 ``before'' the level change in the search direction. That is,
611 given the cached levels like this:
613 1122333442211
614 AB C
616 and assuming we are at the position cached at the slot marked with
617 C, searching backwards (DIR = -1) for LEVEL = 2 will return the
618 index of slot B or A, depending whether BEFORE is, respectively,
619 non-zero or zero. */
620 static int
622 {
624 {
634 {
636 {
640 i--;
641 }
642 }
643 else
644 {
646 {
650 i++;
651 }
652 }
653 }
656 }
660 {
663 /* We should never cache on backward scans. */
669 {
671 /* Don't overrun the cache limit. */
674 /* Character positions should correspond to cache positions 1:1.
675 If we are outside the range of cached positions, the cache is
676 useless and must be reset. */
680 {
683 }
688 }
689 else
690 {
691 /* Copy only the members which could have changed, to avoid
692 costly copying of the entire struct. */
699 else
706 }
711 }
715 {
719 {
724 /* Don't let scan direction from from the cached state override
725 the current scan direction. */
728 }
731 }
735 {
739 }
741 /* Check if buffer position CHARPOS/BYTEPOS is the end of a paragraph.
742 Value is the non-negative length of the paragraph separator
743 following the buffer position, -1 if position is at the beginning
744 of a new paragraph, or -2 if position is neither at beginning nor
745 at end of a paragraph. */
746 EMACS_INT
748 {
753 EMACS_INT val;
762 {
765 else
767 }
770 }
772 /* Determine the start-of-run (sor) directional type given the two
773 embedding levels on either side of the run boundary. Also, update
774 the saved info about previously seen characters, since that info is
775 generally valid for a single level run. */
778 {
781 /* The prev_was_pdf gork is required for when we have several PDFs
782 in a row. In that case, we want to compute the sor type for the
783 next level run only once: when we see the first PDF. That's
784 because the sor type depends only on the higher of the two levels
785 that we find on the two sides of the level boundary (see UAX#9,
786 clause X10), and so we don't need to know the final embedding
787 level to which we descend after processing all the PDFs. */
789 /* FIXME: should the default sor direction be user selectable? */
803 }
805 static void
807 {
820 }
822 /* Find the beginning of this paragraph by looking back in the buffer.
823 Value is the byte position of the paragraph's beginning. */
824 static EMACS_INT
826 {
834 {
837 }
839 }
841 /* Determine the direction, a.k.a. base embedding level, of the
842 paragraph we are about to iterate through. If DIR is either L2R or
843 R2L, just use that. Otherwise, determine the paragraph direction
844 from the first strong character of the paragraph.
846 Note that this gives the paragraph separator the same direction as
847 the preceding paragraph, even though Emacs generally views the
848 separartor as not belonging to any paragraph. */
849 void
851 {
854 /* Special case for an empty buffer. */
857 /* We should never be called at EOB or before BEGV. */
862 {
865 }
867 {
870 }
872 {
874 EMACS_INT pos;
875 bidi_type_t type;
876 EMACS_INT sep_len;
878 /* If we are inside a paragraph separator, we are just waiting
879 for the separator to be exhausted; use the previous paragraph
880 direction. But don't do that if we have been just reseated,
881 because we need to reinitialize below in that case. */
886 /* If we are on a newline, get past it to where the next
887 paragraph might start. But don't do that at BEGV since then
888 we are potentially in a new paragraph that doesn't yet
889 exist. */
892 {
893 bytepos++;
894 pos++;
895 }
897 /* We are either at the beginning of a paragraph or in the
898 middle of it. Find where this paragraph starts. */
901 /* We should always be at the beginning of a new line at this
902 point. */
914 /* NOTE: UAX#9 says to search only for L, AL, or R types of
915 characters, and ignore RLE, RLO, LRE, and LRO. However,
916 I'm not sure it makes sense to omit those 4; should try
917 with and without that to see the effect. */
919 || (bidi_ignore_explicit_marks_for_paragraph_level
923 {
927 {
928 /* Pretend there's a paragraph separator at end of buffer. */
931 }
933 }
938 }
939 else
942 /* Contrary to UAX#9 clause P3, we only default the paragraph
943 direction to L2R if we have no previous usable paragraph
944 direction. */
949 else
953 }
955 /* Do whatever UAX#9 clause X8 says should be done at paragraph's
956 end. */
959 {
964 }
966 /* Initialize the bidi iterator from buffer position CHARPOS. */
967 void
969 {
994 }
996 /* Push the current embedding level and override status; reset the
997 current level to LEVEL and the current override status to OVERRIDE. */
1001 {
1007 }
1009 /* Pop the embedding level and directional override status from the
1010 stack, and return the new level. */
1013 {
1014 /* UAX#9 says to ignore invalid PDFs. */
1018 }
1020 /* Record in SAVED_INFO the information about the current character. */
1024 {
1033 }
1035 /* Resolve the type of a neutral character according to the type of
1036 surrounding strong text and the current embedding level. */
1039 {
1040 /* N1: European and Arabic numbers are treated as though they were R. */
1050 else
1052 }
1056 {
1057 /* FIXME: this should be replaced with a lookup table with suitable
1058 bits set, like standard C ctype macros do. */
1061 }
1063 /* A helper function for bidi_resolve_explicit. It advances to the
1064 next character in logical order and determines the new embedding
1065 level and directional override, but does not take into account
1066 empty embeddings. */
1067 static int
1069 {
1071 bidi_type_t type;
1074 bidi_dir_t override;
1078 {
1083 }
1085 {
1090 }
1096 /* in case it is a unibyte character (not yet implemented) */
1097 /* _fetch_multibyte_char_len = 1; */
1099 {
1102 }
1103 else
1104 {
1107 }
1110 /* Don't apply directional override here, as all the types we handle
1111 below will not be affected by the override anyway, and we need
1112 the original type unaltered. The override will be applied in
1113 bidi_resolve_weak. */
1124 {
1131 {
1133 {
1134 /* Compute the least odd embedding level greater than
1135 the current level. */
1139 else
1144 }
1145 else
1146 {
1148 /* See the commentary about invalid_rl_levels below. */
1152 }
1153 }
1164 {
1166 {
1167 /* Compute the least even embedding level greater than
1168 the current level. */
1172 else
1175 }
1176 else
1177 {
1179 /* invalid_rl_levels counts invalid levels encountered
1180 while the embedding level was already too high for
1181 LRE/LRO, but not for RLE/RLO. That is because
1182 there may be exactly one PDF which we should not
1183 ignore even though invalid_levels is non-zero.
1184 invalid_rl_levels helps to know what PDF is
1185 that. */
1188 }
1189 }
1199 {
1201 {
1206 /* else nothing: UAX#9 says to ignore invalid PDFs */
1207 }
1210 else
1211 {
1214 }
1215 }
1221 /* Nothing. */
1223 }
1229 }
1231 /* Given an iterator state in BIDI_IT, advance one character position
1232 in the buffer to the next character (in the logical order), resolve
1233 any explicit embeddings and directional overrides, and return the
1234 embedding level of the character after resolving explicit
1235 directives and ignoring empty embeddings. */
1236 static int
1238 {
1248 {
1249 /* Avoid pushing and popping embedding levels if the level run
1250 is empty, as this breaks level runs where it shouldn't.
1251 UAX#9 removes all the explicit embedding and override codes,
1252 so empty embeddings disappear without a trace. We need to
1253 behave as if we did the same. */
1261 {
1263 }
1272 {
1273 /* We pushed a level, but we shouldn't have. Undo that. */
1275 {
1280 }
1283 else
1284 {
1287 }
1288 }
1289 }
1292 {
1294 /* This is needed by bidi_resolve_weak below, and in L1. */
1297 }
1300 }
1302 /* Advance in the buffer, resolve weak types and return the type of
1303 the next character after weak type resolution. */
1304 bidi_type_t
1306 {
1307 bidi_type_t type;
1308 bidi_dir_t override;
1312 bidi_type_t type_of_next;
1328 {
1329 /* We've got a new embedding level run, compute the directional
1330 type of sor and initialize per-run variables (UAX#9, clause
1331 X10). */
1333 }
1339 /* Level and directional override status are already recorded in
1340 bidi_it, and do not need any change; see X6. */
1345 else
1346 {
1348 {
1349 /* Note that we don't need to consider the case where the
1350 prev character has its type overridden by an RLO or LRO:
1351 such characters are outside the current level run, and
1352 thus not relevant to this NSM. Thus, NSM gets the
1353 orig_type of the previous character. */
1362 }
1375 {
1376 next_char =
1383 {
1387 ;
1390 }
1392 /* If the next character is EN, but the last strong-type
1393 character is AL, that next EN will be changed to AN when
1394 we process it in W2 above. So in that case, this ES
1395 should not be changed into EN. */
1401 {
1404 /* If the next character is EN, but the last
1405 strong-type character is AL, EN will be later
1406 changed to AN when we process it in W2 above.
1407 So in that case, this ES should not be
1408 changed into EN. */
1416 }
1417 }
1420 {
1425 {
1428 next_char =
1436 {
1441 ;
1445 }
1447 {
1448 /* If the last strong character is AL, the EN we've
1449 found will become AN when we get to it (W2). */
1451 {
1453 /* Remember this EN position, to speed up processing
1454 of the next ETs. */
1456 }
1459 }
1460 }
1461 }
1462 }
1471 /* Store the type we've got so far, before we clobber it with strong
1472 types in W7 and while resolving neutral types. But leave alone
1473 the original types that were recorded above, because we will need
1474 them for the L1 clause. */
1480 {
1484 }
1489 }
1491 bidi_type_t
1493 {
1511 {
1515 current_level);
1516 else
1517 {
1518 /* Arrrgh!! The UAX#9 algorithm is too deeply entrenched in
1519 the assumption of batch-style processing; see clauses W4,
1520 W5, and especially N1, which require to look far forward
1521 (as well as back) in the buffer. May the fleas of a
1522 thousand camels infest the armpits of those who design
1523 supposedly general-purpose algorithms by looking at their
1524 own implementations, and fail to consider other possible
1525 implementations! */
1527 bidi_type_t next_type;
1533 /* Scan the text forward until we find the first non-neutral
1534 character, and then use that to resolve the neutral we
1535 are dealing with now. We also cache the scanned iterator
1536 states, to salvage some of the effort later. */
1539 /* Record the info about the previous character, so that
1540 it will be cached below with this state. */
1545 /* Paragraph separators have their levels fully resolved
1546 at this point, so cache them as resolved. */
1548 /* FIXME: implement L1 here, by testing for a newline and
1549 resetting the level for any sequence of whitespace
1550 characters adjacent to it. */
1554 /* This is all per level run, so stop when we
1555 reach the end of this level run. */
1557 current_level));
1562 {
1570 /* N1: ``European and Arabic numbers are treated as
1571 though they were R.'' */
1578 /* FALLTHROUGH */
1580 /* Marched all the way to the end of this level run.
1581 We need to use the eor type, whose information is
1582 stored by bidi_set_sor_type in the prev_for_neutral
1583 member. */
1586 {
1590 }
1591 else
1592 {
1593 /* This is a BN which does not adjoin neutrals.
1594 Leave its type alone. */
1597 }
1601 }
1607 }
1608 }
1610 }
1612 /* Given an iterator state in BIDI_IT, advance one character position
1613 in the buffer to the next character (in the logical order), resolve
1614 the bidi type of that next character, and return that type. */
1615 bidi_type_t
1617 {
1618 bidi_type_t type;
1620 /* This should always be called during a forward scan. */
1624 /* Reset the limit until which to ignore BNs if we step out of the
1625 area where we found only empty levels. */
1635 }
1637 /* Given an iterator state BIDI_IT, advance one character position in
1638 the buffer to the next character (in the logical order), resolve
1639 the embedding and implicit levels of that next character, and
1640 return the resulting level. */
1641 int
1643 {
1644 bidi_type_t type;
1649 {
1650 /* There's no sense in trying to advance if we hit end of text. */
1654 /* Record the info about the previous character. */
1662 /* FIXME: it sounds like we don't need both prev and
1663 prev_for_neutral members, but I'm leaving them both for now. */
1668 /* If we overstepped the characters used for resolving neutrals
1669 and whitespace, invalidate their info in the iterator. */
1679 /* This must be taken before we fill the iterator with the info
1680 about the next char. If we scan backwards, the iterator
1681 state must be already cached, so there's no need to know the
1682 embedding level of the previous character, since we will be
1683 returning to our caller shortly. */
1685 }
1688 /* Perhaps it is already cached. */
1691 {
1692 /* Don't lose the information for resolving neutrals! The
1693 cached states could have been cached before their
1694 next_for_neutral member was computed. If we are on our way
1695 forward, we can simply take the info from the previous
1696 state. */
1701 /* If resolved_level is -1, it means this state was cached
1702 before it was completely resolved, so we cannot return
1703 it. */
1706 }
1708 /* If we are going backwards, the iterator state is already cached
1709 from previous scans, and should be fully resolved. */
1721 {
1725 /* If the cached state shows a neutral character, it was not
1726 resolved by bidi_resolve_neutral, so do it now. */
1729 level);
1730 }
1741 /* For L1 below, we need to know, for each WS character, whether
1742 it belongs to a sequence of WS characters preceeding a newline
1743 or a TAB or a paragraph separator. */
1746 {
1751 bidi_type_t chtype;
1754 /*_fetch_multibyte_char_len = 1;*/
1757 cpos++;
1761 else
1769 }
1771 /* Resolve implicit levels, with a twist: PDFs get the embedding
1772 level of the enbedding they terminate. See below for the
1773 reason. */
1775 /* Don't do this if this formatting code didn't change the
1776 embedding level due to invalid or empty embeddings. */
1778 {
1779 /* Don't look in UAX#9 for the reason for this: it's our own
1780 private quirk. The reason is that we want the formatting
1781 codes to be delivered so that they bracket the text of their
1782 embedding. For example, given the text
1784 {RLO}teST{PDF}
1786 we want it to be displayed as
1788 {RLO}STet{PDF}
1790 not as
1792 STet{RLO}{PDF}
1794 which will result because we bump up the embedding level as
1795 soon as we see the RLO and pop it as soon as we see the PDF,
1796 so RLO itself has the same embedding level as "teST", and
1797 thus would be normally delivered last, just before the PDF.
1798 The switch below fiddles with the level of PDF so that this
1799 ugly side effect does not happen.
1801 (This is, of course, only important if the formatting codes
1802 are actually displayed, but Emacs does need to display them
1803 if the user wants to.) */
1805 }
1809 /* || bidi_it->ch == LINESEP_CHAR */
1815 {
1817 level++;
1820 }
1822 {
1824 level++;
1825 }
1829 }
1831 /* Move to the other edge of a level given by LEVEL. If END_FLAG is
1832 non-zero, we are at the end of a level, and we need to prepare to
1833 resume the scan of the lower level.
1835 If this level's other edge is cached, we simply jump to it, filling
1836 the iterator structure with the iterator state on the other edge.
1837 Otherwise, we walk the buffer until we come back to the same level
1838 as LEVEL.
1840 Note: we are not talking here about a ``level run'' in the UAX#9
1841 sense of the term, but rather about a ``level'' which includes
1842 all the levels higher than it. In other words, given the levels
1843 like this:
1845 11111112222222333333334443343222222111111112223322111
1846 A B C
1848 and assuming we are at point A scanning left to right, this
1849 function moves to point C, whereas the UAX#9 ``level 2 run'' ends
1850 at point B. */
1851 static void
1853 {
1857 /* Try the cache first. */
1860 else
1861 {
1872 }
1873 }
1875 void
1877 {
1882 {
1884 }
1886 /* If we just passed a newline, initialize for the next line. */
1890 /* Prepare the sentinel iterator state. */
1892 {
1895 {
1900 }
1901 }
1906 /* Reordering of resolved levels (clause L2) is implemented by
1907 jumping to the other edge of the level and flipping direction of
1908 scanning the text whenever we find a level change. */
1910 {
1916 /* If we don't have anything cached yet, we need to cache the
1917 sentinel state, since we'll need it to record where to jump
1918 when the last non-base level is exhausted. */
1921 /* Jump (or walk) to the other edge of this level. */
1923 /* Switch scan direction and peek at the next character in the
1924 new direction. */
1927 /* The following loop handles the case where the resolved level
1928 jumps by more than one. This is typical for numbers inside a
1929 run of text with left-to-right embedding direction, but can
1930 also happen in other situations. In those cases the decision
1931 where to continue after a level change, and in what direction,
1932 is tricky. For example, given a text like below:
1934 abcdefgh
1935 11336622
1937 (where the numbers below the text show the resolved levels),
1938 the result of reordering according to UAX#9 should be this:
1940 efdcghba
1942 This is implemented by the loop below which flips direction
1943 and jumps to the other edge of the level each time it finds
1944 the new level not to be the expected one. The expected level
1945 is always one more or one less than the previous one. */
1948 {
1954 }
1956 /* Finally, deliver the next character in the new direction. */
1958 }
1960 /* Take note when we have just processed the newline that precedes
1961 the end of the paragraph. The next time we are about to be
1962 called, set_iterator_to_next will automatically reinit the
1963 paragraph direction, if needed. We do this at the newline before
1964 the paragraph separator, because the next character might not be
1965 the first character of the next paragraph, due to the bidi
1966 reordering, whereas we _must_ know the paragraph base direction
1967 _before_ we process the paragraph's text, since the base
1968 direction affects the reordering. */
1972 {
1973 EMACS_INT sep_len =
1977 {
1979 /* Record the buffer position of the last character of the
1980 paragraph separator. */
1982 }
1983 }
1986 {
1987 /* If we are at paragraph's base embedding level and beyond the
1988 last cached position, the cache's job is done and we can
1989 discard it. */
1993 /* But as long as we are caching during forward scan, we must
1994 cache each state, or else the cache integrity will be
1995 compromised: it assumes cached states correspond to buffer
1996 positions 1:1. */
1997 else
1999 }
2000 }
2002 /* This is meant to be called from within the debugger, whenever you
2003 wish to examine the cache contents. */
2004 void
2006 {
2011 {
2014 }
2019 ndigits++;
2032 }