summaryrefslogtreecommitdiff
path: root/src/bidi.c
blob: b067c8134e35db6434064864e0349123d024700d (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
/* Low-level bidirectional buffer/string-scanning functions for GNU Emacs.
   Copyright (C) 2000-2001, 2004-2005, 2009-2013 Free Software
   Foundation, Inc.

This file is part of GNU Emacs.

GNU Emacs is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.

GNU Emacs is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with GNU Emacs.  If not, see <http://www.gnu.org/licenses/>.  */

/* Written by Eli Zaretskii <eliz@gnu.org>.

   A sequential implementation of the Unicode Bidirectional algorithm,
   (UBA) as per UAX#9, a part of the Unicode Standard.

   Unlike the reference and most other implementations, this one is
   designed to be called once for every character in the buffer or
   string.

   The main entry point is bidi_move_to_visually_next.  Each time it
   is called, it finds the next character in the visual order, and
   returns its information in a special structure.  The caller is then
   expected to process this character for display or any other
   purposes, and call bidi_move_to_visually_next for the next
   character.  See the comments in bidi_move_to_visually_next for more
   details about its algorithm that finds the next visual-order
   character by resolving their levels on the fly.

   Two other entry points are bidi_paragraph_init and
   bidi_mirror_char.  The first determines the base direction of a
   paragraph, while the second returns the mirrored version of its
   argument character.

   A few auxiliary entry points are used to initialize the bidi
   iterator for iterating an object (buffer or string), push and pop
   the bidi iterator state, and save and restore the state of the bidi
   cache.

   If you want to understand the code, you will have to read it
   together with the relevant portions of UAX#9.  The comments include
   references to UAX#9 rules, for that very reason.

   A note about references to UAX#9 rules: if the reference says
   something like "X9/Retaining", it means that you need to refer to
   rule X9 and to its modifications described in the "Implementation
   Notes" section of UAX#9, under "Retaining Format Codes".  */

#include <config.h>
#include <stdio.h>

#include "lisp.h"
#include "character.h"
#include "buffer.h"
#include "dispextern.h"

static bool bidi_initialized = 0;

static Lisp_Object bidi_type_table, bidi_mirror_table;

#define LRM_CHAR   0x200E
#define RLM_CHAR   0x200F
#define BIDI_EOB   -1

/* Data type for describing the bidirectional character categories.  */
typedef enum {
  UNKNOWN_BC,
  NEUTRAL,
  WEAK,
  STRONG
} bidi_category_t;

/* UAX#9 says to search only for L, AL, or R types of characters, and
   ignore RLE, RLO, LRE, and LRO, when determining the base paragraph
   level.  Yudit indeed ignores them.  This variable is therefore set
   by default to ignore them, but clearing it will take them into
   account.  */
extern bool bidi_ignore_explicit_marks_for_paragraph_level EXTERNALLY_VISIBLE;
bool bidi_ignore_explicit_marks_for_paragraph_level = 1;

static Lisp_Object paragraph_start_re, paragraph_separate_re;
static Lisp_Object Qparagraph_start, Qparagraph_separate;


/***********************************************************************
			Utilities
 ***********************************************************************/

/* Return the bidi type of a character CH, subject to the current
   directional OVERRIDE.  */
static bidi_type_t
bidi_get_type (int ch, bidi_dir_t override)
{
  bidi_type_t default_type;

  if (ch == BIDI_EOB)
    return NEUTRAL_B;
  if (ch < 0 || ch > MAX_CHAR)
    emacs_abort ();

  default_type = (bidi_type_t) XINT (CHAR_TABLE_REF (bidi_type_table, ch));
  /* Every valid character code, even those that are unassigned by the
     UCD, have some bidi-class property, according to
     DerivedBidiClass.txt file.  Therefore, if we ever get UNKNOWN_BT
     (= zero) code from CHAR_TABLE_REF, that's a bug.  */
  if (default_type == UNKNOWN_BT)
    emacs_abort ();

  if (override == NEUTRAL_DIR)
    return default_type;

  switch (default_type)
    {
      /* Although UAX#9 does not tell, it doesn't make sense to
	 override NEUTRAL_B and LRM/RLM characters.  */
      case NEUTRAL_B:
      case LRE:
      case LRO:
      case RLE:
      case RLO:
      case PDF:
	return default_type;
      default:
	switch (ch)
	  {
	    case LRM_CHAR:
	    case RLM_CHAR:
	      return default_type;
	    default:
	      if (override == L2R) /* X6 */
		return STRONG_L;
	      else if (override == R2L)
		return STRONG_R;
	      else
		emacs_abort ();	/* can't happen: handled above */
	  }
    }
}

static void
bidi_check_type (bidi_type_t type)
{
  eassert (UNKNOWN_BT <= type && type <= NEUTRAL_ON);
}

/* Given a bidi TYPE of a character, return its category.  */
static bidi_category_t
bidi_get_category (bidi_type_t type)
{
  switch (type)
    {
      case UNKNOWN_BT:
	return UNKNOWN_BC;
      case STRONG_L:
      case STRONG_R:
      case STRONG_AL:
      case LRE:
      case LRO:
      case RLE:
      case RLO:
	return STRONG;
      case PDF:		/* ??? really?? */
      case WEAK_EN:
      case WEAK_ES:
      case WEAK_ET:
      case WEAK_AN:
      case WEAK_CS:
      case WEAK_NSM:
      case WEAK_BN:
	return WEAK;
      case NEUTRAL_B:
      case NEUTRAL_S:
      case NEUTRAL_WS:
      case NEUTRAL_ON:
	return NEUTRAL;
      default:
	emacs_abort ();
    }
}

/* Return the mirrored character of C, if it has one.  If C has no
   mirrored counterpart, return C.
   Note: The conditions in UAX#9 clause L4 regarding the surrounding
   context must be tested by the caller.  */
int
bidi_mirror_char (int c)
{
  Lisp_Object val;

  if (c == BIDI_EOB)
    return c;
  if (c < 0 || c > MAX_CHAR)
    emacs_abort ();

  val = CHAR_TABLE_REF (bidi_mirror_table, c);
  if (INTEGERP (val))
    {
      int v;

      /* When debugging, check before assigning to V, so that the check
	 isn't broken by undefined behavior due to int overflow.  */
      eassert (CHAR_VALID_P (XINT (val)));

      v = XINT (val);

      /* Minimal test we must do in optimized builds, to prevent weird
	 crashes further down the road.  */
      if (v < 0 || v > MAX_CHAR)
	emacs_abort ();

      return v;
    }

  return c;
}

/* Determine the start-of-run (sor) directional type given the two
   embedding levels on either side of the run boundary.  Also, update
   the saved info about previously seen characters, since that info is
   generally valid for a single level run.  */
static void
bidi_set_sor_type (struct bidi_it *bidi_it, int level_before, int level_after)
{
  int higher_level = (level_before > level_after ? level_before : level_after);

  /* The prev_was_pdf gork is required for when we have several PDFs
     in a row.  In that case, we want to compute the sor type for the
     next level run only once: when we see the first PDF.  That's
     because the sor type depends only on the higher of the two levels
     that we find on the two sides of the level boundary (see UAX#9,
     clause X10), and so we don't need to know the final embedding
     level to which we descend after processing all the PDFs.  */
  if (!bidi_it->prev_was_pdf || level_before < level_after)
    /* FIXME: should the default sor direction be user selectable?  */
    bidi_it->sor = ((higher_level & 1) != 0 ? R2L : L2R);
  if (level_before > level_after)
    bidi_it->prev_was_pdf = 1;

  bidi_it->prev.type = UNKNOWN_BT;
  bidi_it->last_strong.type = bidi_it->last_strong.type_after_w1
    = bidi_it->last_strong.orig_type = UNKNOWN_BT;
  bidi_it->prev_for_neutral.type = (bidi_it->sor == R2L ? STRONG_R : STRONG_L);
  bidi_it->prev_for_neutral.charpos = bidi_it->charpos;
  bidi_it->prev_for_neutral.bytepos = bidi_it->bytepos;
  bidi_it->next_for_neutral.type = bidi_it->next_for_neutral.type_after_w1
    = bidi_it->next_for_neutral.orig_type = UNKNOWN_BT;
  bidi_it->ignore_bn_limit = -1; /* meaning it's unknown */
}

/* Push the current embedding level and override status; reset the
   current level to LEVEL and the current override status to OVERRIDE.  */
static void
bidi_push_embedding_level (struct bidi_it *bidi_it,
			   int level, bidi_dir_t override)
{
  bidi_it->stack_idx++;
  eassert (bidi_it->stack_idx < BIDI_MAXLEVEL);
  bidi_it->level_stack[bidi_it->stack_idx].level = level;
  bidi_it->level_stack[bidi_it->stack_idx].override = override;
}

/* Pop the embedding level and directional override status from the
   stack, and return the new level.  */
static int
bidi_pop_embedding_level (struct bidi_it *bidi_it)
{
  /* UAX#9 says to ignore invalid PDFs.  */
  if (bidi_it->stack_idx > 0)
    bidi_it->stack_idx--;
  return bidi_it->level_stack[bidi_it->stack_idx].level;
}

/* Record in SAVED_INFO the information about the current character.  */
static void
bidi_remember_char (struct bidi_saved_info *saved_info,
		    struct bidi_it *bidi_it)
{
  saved_info->charpos = bidi_it->charpos;
  saved_info->bytepos = bidi_it->bytepos;
  saved_info->type = bidi_it->type;
  bidi_check_type (bidi_it->type);
  saved_info->type_after_w1 = bidi_it->type_after_w1;
  bidi_check_type (bidi_it->type_after_w1);
  saved_info->orig_type = bidi_it->orig_type;
  bidi_check_type (bidi_it->orig_type);
}

/* Copy the bidi iterator from FROM to TO.  To save cycles, this only
   copies the part of the level stack that is actually in use.  */
static void
bidi_copy_it (struct bidi_it *to, struct bidi_it *from)
{
  /* Copy everything from the start through the active part of
     the level stack.  */
  memcpy (to, from,
	  (offsetof (struct bidi_it, level_stack[1])
	   + from->stack_idx * sizeof from->level_stack[0]));
}


/***********************************************************************
			Caching the bidi iterator states
 ***********************************************************************/

#define BIDI_CACHE_CHUNK 200
static struct bidi_it *bidi_cache;
static ptrdiff_t bidi_cache_size = 0;
enum { elsz = sizeof (struct bidi_it) };
static ptrdiff_t bidi_cache_idx;	/* next unused cache slot */
static ptrdiff_t bidi_cache_last_idx;	/* slot of last cache hit */
static ptrdiff_t bidi_cache_start = 0;	/* start of cache for this
					   "stack" level */

/* 5-slot stack for saving the start of the previous level of the
   cache.  xdisp.c maintains a 5-slot stack for its iterator state,
   and we need the same size of our stack.  */
static ptrdiff_t bidi_cache_start_stack[IT_STACK_SIZE];
static int bidi_cache_sp;

/* Size of header used by bidi_shelve_cache.  */
enum
  {
    bidi_shelve_header_size
      = (sizeof (bidi_cache_idx) + sizeof (bidi_cache_start_stack)
	 + sizeof (bidi_cache_sp) + sizeof (bidi_cache_start)
	 + sizeof (bidi_cache_last_idx))
  };

/* Reset the cache state to the empty state.  We only reset the part
   of the cache relevant to iteration of the current object.  Previous
   objects, which are pushed on the display iterator's stack, are left
   intact.  This is called when the cached information is no more
   useful for the current iteration, e.g. when we were reseated to a
   new position on the same object.  */
static void
bidi_cache_reset (void)
{
  bidi_cache_idx = bidi_cache_start;
  bidi_cache_last_idx = -1;
}

/* Shrink the cache to its minimal size.  Called when we init the bidi
   iterator for reordering a buffer or a string that does not come
   from display properties, because that means all the previously
   cached info is of no further use.  */
static void
bidi_cache_shrink (void)
{
  if (bidi_cache_size > BIDI_CACHE_CHUNK)
    {
      bidi_cache = xrealloc (bidi_cache, BIDI_CACHE_CHUNK * elsz);
      bidi_cache_size = BIDI_CACHE_CHUNK;
    }
  bidi_cache_reset ();
}

static void
bidi_cache_fetch_state (ptrdiff_t idx, struct bidi_it *bidi_it)
{
  int current_scan_dir = bidi_it->scan_dir;

  if (idx < bidi_cache_start || idx >= bidi_cache_idx)
    emacs_abort ();

  bidi_copy_it (bidi_it, &bidi_cache[idx]);
  bidi_it->scan_dir = current_scan_dir;
  bidi_cache_last_idx = idx;
}

/* Find a cached state with a given CHARPOS and resolved embedding
   level less or equal to LEVEL.  if LEVEL is -1, disregard the
   resolved levels in cached states.  DIR, if non-zero, means search
   in that direction from the last cache hit.  */
static ptrdiff_t
bidi_cache_search (ptrdiff_t charpos, int level, int dir)
{
  ptrdiff_t i, i_start;

  if (bidi_cache_idx > bidi_cache_start)
    {
      if (bidi_cache_last_idx == -1)
	bidi_cache_last_idx = bidi_cache_idx - 1;
      if (charpos < bidi_cache[bidi_cache_last_idx].charpos)
	{
	  dir = -1;
	  i_start = bidi_cache_last_idx - 1;
	}
      else if (charpos > (bidi_cache[bidi_cache_last_idx].charpos
			  + bidi_cache[bidi_cache_last_idx].nchars - 1))
	{
	  dir = 1;
	  i_start = bidi_cache_last_idx + 1;
	}
      else if (dir)
	i_start = bidi_cache_last_idx;
      else
	{
	  dir = -1;
	  i_start = bidi_cache_idx - 1;
	}

      if (dir < 0)
	{
	  /* Linear search for now; FIXME!  */
	  for (i = i_start; i >= bidi_cache_start; i--)
	    if (bidi_cache[i].charpos <= charpos
		&& charpos < bidi_cache[i].charpos + bidi_cache[i].nchars
		&& (level == -1 || bidi_cache[i].resolved_level <= level))
	      return i;
	}
      else
	{
	  for (i = i_start; i < bidi_cache_idx; i++)
	    if (bidi_cache[i].charpos <= charpos
		&& charpos < bidi_cache[i].charpos + bidi_cache[i].nchars
		&& (level == -1 || bidi_cache[i].resolved_level <= level))
	      return i;
	}
    }

  return -1;
}

/* Find a cached state where the resolved level changes to a value
   that is lower than LEVEL, and return its cache slot index.  DIR is
   the direction to search, starting with the last used cache slot.
   If DIR is zero, we search backwards from the last occupied cache
   slot.  BEFORE means return the index of the slot that
   is ``before'' the level change in the search direction.  That is,
   given the cached levels like this:

	 1122333442211
	  AB        C

   and assuming we are at the position cached at the slot marked with
   C, searching backwards (DIR = -1) for LEVEL = 2 will return the
   index of slot B or A, depending whether BEFORE is, respectively,
   true or false.  */
static ptrdiff_t
bidi_cache_find_level_change (int level, int dir, bool before)
{
  if (bidi_cache_idx)
    {
      ptrdiff_t i = dir ? bidi_cache_last_idx : bidi_cache_idx - 1;
      int incr = before ? 1 : 0;

      eassert (!dir || bidi_cache_last_idx >= 0);

      if (!dir)
	dir = -1;
      else if (!incr)
	i += dir;

      if (dir < 0)
	{
	  while (i >= bidi_cache_start + incr)
	    {
	      if (bidi_cache[i - incr].resolved_level >= 0
		  && bidi_cache[i - incr].resolved_level < level)
		return i;
	      i--;
	    }
	}
      else
	{
	  while (i < bidi_cache_idx - incr)
	    {
	      if (bidi_cache[i + incr].resolved_level >= 0
		  && bidi_cache[i + incr].resolved_level < level)
		return i;
	      i++;
	    }
	}
    }

  return -1;
}

static void
bidi_cache_ensure_space (ptrdiff_t idx)
{
  /* Enlarge the cache as needed.  */
  if (idx >= bidi_cache_size)
    {
      /* The bidi cache cannot be larger than the largest Lisp string
	 or buffer.  */
      ptrdiff_t string_or_buffer_bound
	= max (BUF_BYTES_MAX, STRING_BYTES_BOUND);

      /* Also, it cannot be larger than what C can represent.  */
      ptrdiff_t c_bound
	= (min (PTRDIFF_MAX, SIZE_MAX) - bidi_shelve_header_size) / elsz;

      bidi_cache
	= xpalloc (bidi_cache, &bidi_cache_size,
		   max (BIDI_CACHE_CHUNK, idx - bidi_cache_size + 1),
		   min (string_or_buffer_bound, c_bound), elsz);
    }
}

static void
bidi_cache_iterator_state (struct bidi_it *bidi_it, bool resolved)
{
  ptrdiff_t idx;

  /* We should never cache on backward scans.  */
  if (bidi_it->scan_dir == -1)
    emacs_abort ();
  idx = bidi_cache_search (bidi_it->charpos, -1, 1);

  if (idx < 0)
    {
      idx = bidi_cache_idx;
      bidi_cache_ensure_space (idx);
      /* Character positions should correspond to cache positions 1:1.
	 If we are outside the range of cached positions, the cache is
	 useless and must be reset.  */
      if (idx > bidi_cache_start &&
	  (bidi_it->charpos > (bidi_cache[idx - 1].charpos
			       + bidi_cache[idx - 1].nchars)
	   || bidi_it->charpos < bidi_cache[bidi_cache_start].charpos))
	{
	  bidi_cache_reset ();
	  idx = bidi_cache_start;
	}
      if (bidi_it->nchars <= 0)
	emacs_abort ();
      bidi_copy_it (&bidi_cache[idx], bidi_it);
      if (!resolved)
	bidi_cache[idx].resolved_level = -1;
    }
  else
    {
      /* Copy only the members which could have changed, to avoid
	 costly copying of the entire struct.  */
      bidi_cache[idx].type = bidi_it->type;
      bidi_check_type (bidi_it->type);
      bidi_cache[idx].type_after_w1 = bidi_it->type_after_w1;
      bidi_check_type (bidi_it->type_after_w1);
      if (resolved)
	bidi_cache[idx].resolved_level = bidi_it->resolved_level;
      else
	bidi_cache[idx].resolved_level = -1;
      bidi_cache[idx].invalid_levels = bidi_it->invalid_levels;
      bidi_cache[idx].invalid_rl_levels = bidi_it->invalid_rl_levels;
      bidi_cache[idx].next_for_neutral = bidi_it->next_for_neutral;
      bidi_cache[idx].next_for_ws = bidi_it->next_for_ws;
      bidi_cache[idx].ignore_bn_limit = bidi_it->ignore_bn_limit;
      bidi_cache[idx].disp_pos = bidi_it->disp_pos;
      bidi_cache[idx].disp_prop = bidi_it->disp_prop;
    }

  bidi_cache_last_idx = idx;
  if (idx >= bidi_cache_idx)
    bidi_cache_idx = idx + 1;
}

static bidi_type_t
bidi_cache_find (ptrdiff_t charpos, int level, struct bidi_it *bidi_it)
{
  ptrdiff_t i = bidi_cache_search (charpos, level, bidi_it->scan_dir);

  if (i >= bidi_cache_start)
    {
      bidi_dir_t current_scan_dir = bidi_it->scan_dir;

      bidi_copy_it (bidi_it, &bidi_cache[i]);
      bidi_cache_last_idx = i;
      /* Don't let scan direction from the cached state override
	 the current scan direction.  */
      bidi_it->scan_dir = current_scan_dir;
      return bidi_it->type;
    }

  return UNKNOWN_BT;
}

static int
bidi_peek_at_next_level (struct bidi_it *bidi_it)
{
  if (bidi_cache_idx == bidi_cache_start || bidi_cache_last_idx == -1)
    emacs_abort ();
  return bidi_cache[bidi_cache_last_idx + bidi_it->scan_dir].resolved_level;
}


/***********************************************************************
	     Pushing and popping the bidi iterator state
 ***********************************************************************/

/* Push the bidi iterator state in preparation for reordering a
   different object, e.g. display string found at certain buffer
   position.  Pushing the bidi iterator boils down to saving its
   entire state on the cache and starting a new cache "stacked" on top
   of the current cache.  */
void
bidi_push_it (struct bidi_it *bidi_it)
{
  /* Save the current iterator state in its entirety after the last
     used cache slot.  */
  bidi_cache_ensure_space (bidi_cache_idx);
  bidi_cache[bidi_cache_idx++] = *bidi_it;

  /* Push the current cache start onto the stack.  */
  eassert (bidi_cache_sp < IT_STACK_SIZE);
  bidi_cache_start_stack[bidi_cache_sp++] = bidi_cache_start;

  /* Start a new level of cache, and make it empty.  */
  bidi_cache_start = bidi_cache_idx;
  bidi_cache_last_idx = -1;
}

/* Restore the iterator state saved by bidi_push_it and return the
   cache to the corresponding state.  */
void
bidi_pop_it (struct bidi_it *bidi_it)
{
  if (bidi_cache_start <= 0)
    emacs_abort ();

  /* Reset the next free cache slot index to what it was before the
     call to bidi_push_it.  */
  bidi_cache_idx = bidi_cache_start - 1;

  /* Restore the bidi iterator state saved in the cache.  */
  *bidi_it = bidi_cache[bidi_cache_idx];

  /* Pop the previous cache start from the stack.  */
  if (bidi_cache_sp <= 0)
    emacs_abort ();
  bidi_cache_start = bidi_cache_start_stack[--bidi_cache_sp];

  /* Invalidate the last-used cache slot data.  */
  bidi_cache_last_idx = -1;
}

static ptrdiff_t bidi_cache_total_alloc;

/* Stash away a copy of the cache and its control variables.  */
void *
bidi_shelve_cache (void)
{
  unsigned char *databuf;
  ptrdiff_t alloc;

  /* Empty cache.  */
  if (bidi_cache_idx == 0)
    return NULL;

  alloc = (bidi_shelve_header_size
	   + bidi_cache_idx * sizeof (struct bidi_it));
  databuf = xmalloc (alloc);
  bidi_cache_total_alloc += alloc;

  memcpy (databuf, &bidi_cache_idx, sizeof (bidi_cache_idx));
  memcpy (databuf + sizeof (bidi_cache_idx),
	  bidi_cache, bidi_cache_idx * sizeof (struct bidi_it));
  memcpy (databuf + sizeof (bidi_cache_idx)
	  + bidi_cache_idx * sizeof (struct bidi_it),
	  bidi_cache_start_stack, sizeof (bidi_cache_start_stack));
  memcpy (databuf + sizeof (bidi_cache_idx)
	  + bidi_cache_idx * sizeof (struct bidi_it)
	  + sizeof (bidi_cache_start_stack),
	  &bidi_cache_sp, sizeof (bidi_cache_sp));
  memcpy (databuf + sizeof (bidi_cache_idx)
	  + bidi_cache_idx * sizeof (struct bidi_it)
	  + sizeof (bidi_cache_start_stack) + sizeof (bidi_cache_sp),
	  &bidi_cache_start, sizeof (bidi_cache_start));
  memcpy (databuf + sizeof (bidi_cache_idx)
	  + bidi_cache_idx * sizeof (struct bidi_it)
	  + sizeof (bidi_cache_start_stack) + sizeof (bidi_cache_sp)
	  + sizeof (bidi_cache_start),
	  &bidi_cache_last_idx, sizeof (bidi_cache_last_idx));

  return databuf;
}

/* Restore the cache state from a copy stashed away by
   bidi_shelve_cache, and free the buffer used to stash that copy.
   JUST_FREE means free the buffer, but don't restore the
   cache; used when the corresponding iterator is discarded instead of
   being restored.  */
void
bidi_unshelve_cache (void *databuf, bool just_free)
{
  unsigned char *p = databuf;

  if (!p)
    {
      if (!just_free)
	{
	  /* A NULL pointer means an empty cache.  */
	  bidi_cache_start = 0;
	  bidi_cache_sp = 0;
	  bidi_cache_reset ();
	}
    }
  else
    {
      if (just_free)
	{
	  ptrdiff_t idx;

	  memcpy (&idx, p, sizeof (bidi_cache_idx));
	  bidi_cache_total_alloc
	    -= bidi_shelve_header_size + idx * sizeof (struct bidi_it);
	}
      else
	{
	  memcpy (&bidi_cache_idx, p, sizeof (bidi_cache_idx));
	  bidi_cache_ensure_space (bidi_cache_idx);
	  memcpy (bidi_cache, p + sizeof (bidi_cache_idx),
		  bidi_cache_idx * sizeof (struct bidi_it));
	  memcpy (bidi_cache_start_stack,
		  p + sizeof (bidi_cache_idx)
		  + bidi_cache_idx * sizeof (struct bidi_it),
		  sizeof (bidi_cache_start_stack));
	  memcpy (&bidi_cache_sp,
		  p + sizeof (bidi_cache_idx)
		  + bidi_cache_idx * sizeof (struct bidi_it)
		  + sizeof (bidi_cache_start_stack),
		  sizeof (bidi_cache_sp));
	  memcpy (&bidi_cache_start,
		  p + sizeof (bidi_cache_idx)
		  + bidi_cache_idx * sizeof (struct bidi_it)
		  + sizeof (bidi_cache_start_stack) + sizeof (bidi_cache_sp),
		  sizeof (bidi_cache_start));
	  memcpy (&bidi_cache_last_idx,
		  p + sizeof (bidi_cache_idx)
		  + bidi_cache_idx * sizeof (struct bidi_it)
		  + sizeof (bidi_cache_start_stack) + sizeof (bidi_cache_sp)
		  + sizeof (bidi_cache_start),
		  sizeof (bidi_cache_last_idx));
	  bidi_cache_total_alloc
	    -= (bidi_shelve_header_size
		+ bidi_cache_idx * sizeof (struct bidi_it));
	}

      xfree (p);
    }
}


/***********************************************************************
			Initialization
 ***********************************************************************/
static void
bidi_initialize (void)
{
  bidi_type_table = uniprop_table (intern ("bidi-class"));
  if (NILP (bidi_type_table))
    emacs_abort ();
  staticpro (&bidi_type_table);

  bidi_mirror_table = uniprop_table (intern ("mirroring"));
  if (NILP (bidi_mirror_table))
    emacs_abort ();
  staticpro (&bidi_mirror_table);

  Qparagraph_start = intern ("paragraph-start");
  staticpro (&Qparagraph_start);
  paragraph_start_re = Fsymbol_value (Qparagraph_start);
  if (!STRINGP (paragraph_start_re))
    paragraph_start_re = build_string ("\f\\|[ \t]*$");
  staticpro (&paragraph_start_re);
  Qparagraph_separate = intern ("paragraph-separate");
  staticpro (&Qparagraph_separate);
  paragraph_separate_re = Fsymbol_value (Qparagraph_separate);
  if (!STRINGP (paragraph_separate_re))
    paragraph_separate_re = build_string ("[ \t\f]*$");
  staticpro (&paragraph_separate_re);

  bidi_cache_sp = 0;
  bidi_cache_total_alloc = 0;

  bidi_initialized = 1;
}

/* Do whatever UAX#9 clause X8 says should be done at paragraph's
   end.  */
static void
bidi_set_paragraph_end (struct bidi_it *bidi_it)
{
  bidi_it->invalid_levels = 0;
  bidi_it->invalid_rl_levels = -1;
  bidi_it->stack_idx = 0;
  bidi_it->resolved_level = bidi_it->level_stack[0].level;
}

/* Initialize the bidi iterator from buffer/string position CHARPOS.  */
void
bidi_init_it (ptrdiff_t charpos, ptrdiff_t bytepos, bool frame_window_p,
	      struct bidi_it *bidi_it)
{
  if (! bidi_initialized)
    bidi_initialize ();
  if (charpos >= 0)
    bidi_it->charpos = charpos;
  if (bytepos >= 0)
    bidi_it->bytepos = bytepos;
  bidi_it->frame_window_p = frame_window_p;
  bidi_it->nchars = -1;	/* to be computed in bidi_resolve_explicit_1 */
  bidi_it->first_elt = 1;
  bidi_set_paragraph_end (bidi_it);
  bidi_it->new_paragraph = 1;
  bidi_it->separator_limit = -1;
  bidi_it->type = NEUTRAL_B;
  bidi_it->type_after_w1 = NEUTRAL_B;
  bidi_it->orig_type = NEUTRAL_B;
  bidi_it->prev_was_pdf = 0;
  bidi_it->prev.type = bidi_it->prev.type_after_w1
    = bidi_it->prev.orig_type = UNKNOWN_BT;
  bidi_it->last_strong.type = bidi_it->last_strong.type_after_w1
    = bidi_it->last_strong.orig_type = UNKNOWN_BT;
  bidi_it->next_for_neutral.charpos = -1;
  bidi_it->next_for_neutral.type
    = bidi_it->next_for_neutral.type_after_w1
    = bidi_it->next_for_neutral.orig_type = UNKNOWN_BT;
  bidi_it->prev_for_neutral.charpos = -1;
  bidi_it->prev_for_neutral.type
    = bidi_it->prev_for_neutral.type_after_w1
    = bidi_it->prev_for_neutral.orig_type = UNKNOWN_BT;
  bidi_it->sor = L2R;	 /* FIXME: should it be user-selectable? */
  bidi_it->disp_pos = -1;	/* invalid/unknown */
  bidi_it->disp_prop = 0;
  /* We can only shrink the cache if we are at the bottom level of its
     "stack".  */
  if (bidi_cache_start == 0)
    bidi_cache_shrink ();
  else
    bidi_cache_reset ();
}

/* Perform initializations for reordering a new line of bidi text.  */
static void
bidi_line_init (struct bidi_it *bidi_it)
{
  bidi_it->scan_dir = 1; /* FIXME: do we need to have control on this? */
  bidi_it->resolved_level = bidi_it->level_stack[0].level;
  bidi_it->level_stack[0].override = NEUTRAL_DIR; /* X1 */
  bidi_it->invalid_levels = 0;
  bidi_it->invalid_rl_levels = -1;
  /* Setting this to zero will force its recomputation the first time
     we need it for W5.  */
  bidi_it->next_en_pos = 0;
  bidi_it->next_en_type = UNKNOWN_BT;
  bidi_it->next_for_ws.type = UNKNOWN_BT;
  bidi_set_sor_type (bidi_it,
		     (bidi_it->paragraph_dir == R2L ? 1 : 0),
		     bidi_it->level_stack[0].level); /* X10 */

  bidi_cache_reset ();
}


/***********************************************************************
			Fetching characters
 ***********************************************************************/

/* Count bytes in string S between BEG/BEGBYTE and END.  BEG and END
   are zero-based character positions in S, BEGBYTE is byte position
   corresponding to BEG.  UNIBYTE means S is a unibyte string.  */
static ptrdiff_t
bidi_count_bytes (const unsigned char *s, const ptrdiff_t beg,
		  const ptrdiff_t begbyte, const ptrdiff_t end, bool unibyte)
{
  ptrdiff_t pos = beg;
  const unsigned char *p = s + begbyte, *start = p;

  if (unibyte)
    p = s + end;
  else
    {
      if (!CHAR_HEAD_P (*p))
	emacs_abort ();

      while (pos < end)
	{
	  p += BYTES_BY_CHAR_HEAD (*p);
	  pos++;
	}
    }

  return p - start;
}

/* Fetch and return the character at byte position BYTEPOS.  If S is
   non-NULL, fetch the character from string S; otherwise fetch the
   character from the current buffer.  UNIBYTE means S is a
   unibyte string.  */
static int
bidi_char_at_pos (ptrdiff_t bytepos, const unsigned char *s, bool unibyte)
{
  if (s)
    {
      s += bytepos;
      if (unibyte)
	return *s;
    }
  else
    s = BYTE_POS_ADDR (bytepos);
  return STRING_CHAR (s);
}

/* Fetch and return the character at BYTEPOS/CHARPOS.  If that
   character is covered by a display string, treat the entire run of
   covered characters as a single character, either u+2029 or u+FFFC,
   and return their combined length in CH_LEN and NCHARS.  DISP_POS
   specifies the character position of the next display string, or -1
   if not yet computed.  When the next character is at or beyond that
   position, the function updates DISP_POS with the position of the
   next display string.  *DISP_PROP non-zero means that there's really
   a display string at DISP_POS, as opposed to when we searched till
   DISP_POS without finding one.  If *DISP_PROP is 2, it means the
   display spec is of the form `(space ...)', which is replaced with
   u+2029 to handle it as a paragraph separator.  STRING->s is the C
   string to iterate, or NULL if iterating over a buffer or a Lisp
   string; in the latter case, STRING->lstring is the Lisp string.  */
static int
bidi_fetch_char (ptrdiff_t bytepos, ptrdiff_t charpos, ptrdiff_t *disp_pos,
		 int *disp_prop, struct bidi_string_data *string,
		 bool frame_window_p, ptrdiff_t *ch_len, ptrdiff_t *nchars)
{
  int ch;
  ptrdiff_t endpos
    = (string->s || STRINGP (string->lstring)) ? string->schars : ZV;
  struct text_pos pos;
  int len;

  /* If we got past the last known position of display string, compute
     the position of the next one.  That position could be at CHARPOS.  */
  if (charpos < endpos && charpos > *disp_pos)
    {
      SET_TEXT_POS (pos, charpos, bytepos);
      *disp_pos = compute_display_string_pos (&pos, string, frame_window_p,
					      disp_prop);
    }

  /* Fetch the character at BYTEPOS.  */
  if (charpos >= endpos)
    {
      ch = BIDI_EOB;
      *ch_len = 1;
      *nchars = 1;
      *disp_pos = endpos;
      *disp_prop = 0;
    }
  else if (charpos >= *disp_pos && *disp_prop)
    {
      ptrdiff_t disp_end_pos;

      /* We don't expect to find ourselves in the middle of a display
	 property.  Hopefully, it will never be needed.  */
      if (charpos > *disp_pos)
	emacs_abort ();
      /* Text covered by `display' properties and overlays with
	 display properties or display strings is handled as a single
	 character that represents the entire run of characters
	 covered by the display property.  */
      if (*disp_prop == 2)
	{
	  /* `(space ...)' display specs are handled as paragraph
	     separators for the purposes of the reordering; see UAX#9
	     section 3 and clause HL1 in section 4.3 there.  */
	  ch = 0x2029;
	}
      else
	{
	  /* All other display specs are handled as the Unicode Object
	     Replacement Character.  */
	  ch = 0xFFFC;
	}
      disp_end_pos = compute_display_string_end (*disp_pos, string);
      if (disp_end_pos < 0)
	{
	  /* Somebody removed the display string from the buffer
	     behind our back.  Recover by processing this buffer
	     position as if no display property were present there to
	     begin with.  */
	  *disp_prop = 0;
	  goto normal_char;
	}
      *nchars = disp_end_pos - *disp_pos;
      if (*nchars <= 0)
	emacs_abort ();
      if (string->s)
	*ch_len = bidi_count_bytes (string->s, *disp_pos, bytepos,
				    disp_end_pos, string->unibyte);
      else if (STRINGP (string->lstring))
	*ch_len = bidi_count_bytes (SDATA (string->lstring), *disp_pos,
				    bytepos, disp_end_pos, string->unibyte);
      else
	*ch_len = CHAR_TO_BYTE (disp_end_pos) - bytepos;
    }
  else
    {
    normal_char:
      if (string->s)
	{

	  if (!string->unibyte)
	    {
	      ch = STRING_CHAR_AND_LENGTH (string->s + bytepos, len);
	      *ch_len = len;
	    }
	  else
	    {
	      ch = UNIBYTE_TO_CHAR (string->s[bytepos]);
	      *ch_len = 1;
	    }
	}
      else if (STRINGP (string->lstring))
	{
	  if (!string->unibyte)
	    {
	      ch = STRING_CHAR_AND_LENGTH (SDATA (string->lstring) + bytepos,
					   len);
	      *ch_len = len;
	    }
	  else
	    {
	      ch = UNIBYTE_TO_CHAR (SREF (string->lstring, bytepos));
	      *ch_len = 1;
	    }
	}
      else
	{
	  ch = STRING_CHAR_AND_LENGTH (BYTE_POS_ADDR (bytepos), len);
	  *ch_len = len;
	}
      *nchars = 1;
    }

  /* If we just entered a run of characters covered by a display
     string, compute the position of the next display string.  */
  if (charpos + *nchars <= endpos && charpos + *nchars > *disp_pos
      && *disp_prop)
    {
      SET_TEXT_POS (pos, charpos + *nchars, bytepos + *ch_len);
      *disp_pos = compute_display_string_pos (&pos, string, frame_window_p,
					      disp_prop);
    }

  return ch;
}


/***********************************************************************
			Determining paragraph direction
 ***********************************************************************/

/* Check if buffer position CHARPOS/BYTEPOS is the end of a paragraph.
   Value is the non-negative length of the paragraph separator
   following the buffer position, -1 if position is at the beginning
   of a new paragraph, or -2 if position is neither at beginning nor
   at end of a paragraph.  */
static ptrdiff_t
bidi_at_paragraph_end (ptrdiff_t charpos, ptrdiff_t bytepos)
{
  Lisp_Object sep_re;
  Lisp_Object start_re;
  ptrdiff_t val;

  sep_re = paragraph_separate_re;
  start_re = paragraph_start_re;

  val = fast_looking_at (sep_re, charpos, bytepos, ZV, ZV_BYTE, Qnil);
  if (val < 0)
    {
      if (fast_looking_at (start_re, charpos, bytepos, ZV, ZV_BYTE, Qnil) >= 0)
	val = -1;
      else
	val = -2;
    }

  return val;
}

/* On my 2005-vintage machine, searching back for paragraph start
   takes ~1 ms per line.  And bidi_paragraph_init is called 4 times
   when user types C-p.  The number below limits each call to
   bidi_paragraph_init to about 10 ms.  */
#define MAX_PARAGRAPH_SEARCH 7500

/* Find the beginning of this paragraph by looking back in the buffer.
   Value is the byte position of the paragraph's beginning, or
   BEGV_BYTE if paragraph_start_re is still not found after looking
   back MAX_PARAGRAPH_SEARCH lines in the buffer.  */
static ptrdiff_t
bidi_find_paragraph_start (ptrdiff_t pos, ptrdiff_t pos_byte)
{
  Lisp_Object re = paragraph_start_re;
  ptrdiff_t limit = ZV, limit_byte = ZV_BYTE;
  ptrdiff_t n = 0;

  while (pos_byte > BEGV_BYTE
	 && n++ < MAX_PARAGRAPH_SEARCH
	 && fast_looking_at (re, pos, pos_byte, limit, limit_byte, Qnil) < 0)
    {
      /* FIXME: What if the paragraph beginning is covered by a
	 display string?  And what if a display string covering some
	 of the text over which we scan back includes
	 paragraph_start_re?  */
      pos = find_next_newline_no_quit (pos - 1, -1);
      pos_byte = CHAR_TO_BYTE (pos);
    }
  if (n >= MAX_PARAGRAPH_SEARCH)
    pos_byte = BEGV_BYTE;
  return pos_byte;
}

/* On a 3.4 GHz machine, searching forward for a strong directional
   character in a long paragraph full of weaks or neutrals takes about
   1 ms for each 20K characters.  The number below limits each call to
   bidi_paragraph_init to less than 10 ms even on slow machines.  */
#define MAX_STRONG_CHAR_SEARCH 100000

/* Determine the base direction, a.k.a. base embedding level, of the
   paragraph we are about to iterate through.  If DIR is either L2R or
   R2L, just use that.  Otherwise, determine the paragraph direction
   from the first strong directional character of the paragraph.

   NO_DEFAULT_P means don't default to L2R if the paragraph
   has no strong directional characters and both DIR and
   bidi_it->paragraph_dir are NEUTRAL_DIR.  In that case, search back
   in the buffer until a paragraph is found with a strong character,
   or until hitting BEGV.  In the latter case, fall back to L2R.  This
   flag is used in current-bidi-paragraph-direction.

   Note that this function gives the paragraph separator the same
   direction as the preceding paragraph, even though Emacs generally
   views the separator as not belonging to any paragraph.  */
void
bidi_paragraph_init (bidi_dir_t dir, struct bidi_it *bidi_it, bool no_default_p)
{
  ptrdiff_t bytepos = bidi_it->bytepos;
  bool string_p = bidi_it->string.s || STRINGP (bidi_it->string.lstring);
  ptrdiff_t pstartbyte;
  /* Note that begbyte is a byte position, while end is a character
     position.  Yes, this is ugly, but we are trying to avoid costly
     calls to BYTE_TO_CHAR and its ilk.  */
  ptrdiff_t begbyte = string_p ? 0 : BEGV_BYTE;
  ptrdiff_t end = string_p ? bidi_it->string.schars : ZV;

  /* Special case for an empty buffer. */
  if (bytepos == begbyte && bidi_it->charpos == end)
    dir = L2R;
  /* We should never be called at EOB or before BEGV.  */
  else if (bidi_it->charpos >= end || bytepos < begbyte)
    emacs_abort ();

  if (dir == L2R)
    {
      bidi_it->paragraph_dir = L2R;
      bidi_it->new_paragraph = 0;
    }
  else if (dir == R2L)
    {
      bidi_it->paragraph_dir = R2L;
      bidi_it->new_paragraph = 0;
    }
  else if (dir == NEUTRAL_DIR)	/* P2 */
    {
      int ch;
      ptrdiff_t ch_len, nchars;
      ptrdiff_t pos, disp_pos = -1;
      int disp_prop = 0;
      bidi_type_t type;
      const unsigned char *s;

      if (!bidi_initialized)
	bidi_initialize ();

      /* If we are inside a paragraph separator, we are just waiting
	 for the separator to be exhausted; use the previous paragraph
	 direction.  But don't do that if we have been just reseated,
	 because we need to reinitialize below in that case.  */
      if (!bidi_it->first_elt
	  && bidi_it->charpos < bidi_it->separator_limit)
	return;

      /* If we are on a newline, get past it to where the next
	 paragraph might start.  But don't do that at BEGV since then
	 we are potentially in a new paragraph that doesn't yet
	 exist.  */
      pos = bidi_it->charpos;
      s = (STRINGP (bidi_it->string.lstring)
	   ? SDATA (bidi_it->string.lstring)
	   : bidi_it->string.s);
      if (bytepos > begbyte
	  && bidi_char_at_pos (bytepos, s, bidi_it->string.unibyte) == '\n')
	{
	  bytepos++;
	  pos++;
	}

      /* We are either at the beginning of a paragraph or in the
	 middle of it.  Find where this paragraph starts.  */
      if (string_p)
	{
	  /* We don't support changes of paragraph direction inside a
	     string.  It is treated as a single paragraph.  */
	  pstartbyte = 0;
	}
      else
	pstartbyte = bidi_find_paragraph_start (pos, bytepos);
      bidi_it->separator_limit = -1;
      bidi_it->new_paragraph = 0;

      /* The following loop is run more than once only if NO_DEFAULT_P,
	 and only if we are iterating on a buffer.  */
      do {
	ptrdiff_t pos1;

	bytepos = pstartbyte;
	if (!string_p)
	  pos = BYTE_TO_CHAR (bytepos);
	ch = bidi_fetch_char (bytepos, pos, &disp_pos, &disp_prop,
			      &bidi_it->string,
			      bidi_it->frame_window_p, &ch_len, &nchars);
	type = bidi_get_type (ch, NEUTRAL_DIR);

	pos1 = pos;
	for (pos += nchars, bytepos += ch_len;
	     ((bidi_get_category (type) != STRONG)
	      || (bidi_ignore_explicit_marks_for_paragraph_level
		  && (type == RLE || type == RLO
		      || type == LRE || type == LRO)))
	       /* Stop when searched too far into an abnormally large
		  paragraph full of weak or neutral characters.  */
	       && pos - pos1 < MAX_STRONG_CHAR_SEARCH;
	     type = bidi_get_type (ch, NEUTRAL_DIR))
	  {
	    if (pos >= end)
	      {
		/* Pretend there's a paragraph separator at end of
		   buffer/string.  */
		type = NEUTRAL_B;
		break;
	      }
	    if (!string_p
		&& type == NEUTRAL_B
		&& bidi_at_paragraph_end (pos, bytepos) >= -1)
	      break;
	    /* Fetch next character and advance to get past it.  */
	    ch = bidi_fetch_char (bytepos, pos, &disp_pos,
				  &disp_prop, &bidi_it->string,
				  bidi_it->frame_window_p, &ch_len, &nchars);
	    pos += nchars;
	    bytepos += ch_len;
	  }
	if ((type == STRONG_R || type == STRONG_AL) /* P3 */
	    || (!bidi_ignore_explicit_marks_for_paragraph_level
		&& (type == RLO || type == RLE)))
	  bidi_it->paragraph_dir = R2L;
	else if (type == STRONG_L
		 || (!bidi_ignore_explicit_marks_for_paragraph_level
		     && (type == LRO || type == LRE)))
	  bidi_it->paragraph_dir = L2R;
	if (!string_p
	    && no_default_p && bidi_it->paragraph_dir == NEUTRAL_DIR)
	  {
	    /* If this paragraph is at BEGV, default to L2R.  */
	    if (pstartbyte == BEGV_BYTE)
	      bidi_it->paragraph_dir = L2R; /* P3 and HL1 */
	    else
	      {
		ptrdiff_t prevpbyte = pstartbyte;
		ptrdiff_t p = BYTE_TO_CHAR (pstartbyte), pbyte = pstartbyte;

		/* Find the beginning of the previous paragraph, if any.  */
		while (pbyte > BEGV_BYTE && prevpbyte >= pstartbyte)
		  {
		    /* FXIME: What if p is covered by a display
		       string?  See also a FIXME inside
		       bidi_find_paragraph_start.  */
		    p--;
		    pbyte = CHAR_TO_BYTE (p);
		    prevpbyte = bidi_find_paragraph_start (p, pbyte);
		  }
		pstartbyte = prevpbyte;
	      }
	  }
      } while (!string_p
	       && no_default_p && bidi_it->paragraph_dir == NEUTRAL_DIR);
    }
  else
    emacs_abort ();

  /* Contrary to UAX#9 clause P3, we only default the paragraph
     direction to L2R if we have no previous usable paragraph
     direction.  This is allowed by the HL1 clause.  */
  if (bidi_it->paragraph_dir != L2R && bidi_it->paragraph_dir != R2L)
    bidi_it->paragraph_dir = L2R; /* P3 and HL1 ``higher-level protocols'' */
  if (bidi_it->paragraph_dir == R2L)
    bidi_it->level_stack[0].level = 1;
  else
    bidi_it->level_stack[0].level = 0;

  bidi_line_init (bidi_it);
}


/***********************************************************************
		 Resolving explicit and implicit levels.
  The rest of this file constitutes the core of the UBA implementation.
 ***********************************************************************/

static bool
bidi_explicit_dir_char (int ch)
{
  bidi_type_t ch_type;

  if (!bidi_initialized)
    emacs_abort ();
  ch_type = (bidi_type_t) XINT (CHAR_TABLE_REF (bidi_type_table, ch));
  return (ch_type == LRE || ch_type == LRO
	  || ch_type == RLE || ch_type == RLO
	  || ch_type == PDF);
}

/* A helper function for bidi_resolve_explicit.  It advances to the
   next character in logical order and determines the new embedding
   level and directional override, but does not take into account
   empty embeddings.  */
static int
bidi_resolve_explicit_1 (struct bidi_it *bidi_it)
{
  int curchar;
  bidi_type_t type;
  int current_level;
  int new_level;
  bidi_dir_t override;
  bool string_p = bidi_it->string.s || STRINGP (bidi_it->string.lstring);

  /* If reseat()'ed, don't advance, so as to start iteration from the
     position where we were reseated.  bidi_it->bytepos can be less
     than BEGV_BYTE after reseat to BEGV.  */
  if (bidi_it->bytepos < (string_p ? 0 : BEGV_BYTE)
      || bidi_it->first_elt)
    {
      bidi_it->first_elt = 0;
      if (string_p)
	{
	  const unsigned char *p
	    = (STRINGP (bidi_it->string.lstring)
	       ? SDATA (bidi_it->string.lstring)
	       : bidi_it->string.s);

	  if (bidi_it->charpos < 0)
	    bidi_it->charpos = 0;
	  bidi_it->bytepos = bidi_count_bytes (p, 0, 0, bidi_it->charpos,
					       bidi_it->string.unibyte);
	}
      else
	{
	  if (bidi_it->charpos < BEGV)
	    bidi_it->charpos = BEGV;
	  bidi_it->bytepos = CHAR_TO_BYTE (bidi_it->charpos);
	}
    }
  /* Don't move at end of buffer/string.  */
  else if (bidi_it->charpos < (string_p ? bidi_it->string.schars : ZV))
    {
      /* Advance to the next character, skipping characters covered by
	 display strings (nchars > 1).  */
      if (bidi_it->nchars <= 0)
	emacs_abort ();
      bidi_it->charpos += bidi_it->nchars;
      if (bidi_it->ch_len == 0)
	emacs_abort ();
      bidi_it->bytepos += bidi_it->ch_len;
    }

  current_level = bidi_it->level_stack[bidi_it->stack_idx].level; /* X1 */
  override = bidi_it->level_stack[bidi_it->stack_idx].override;
  new_level = current_level;

  if (bidi_it->charpos >= (string_p ? bidi_it->string.schars : ZV))
    {
      curchar = BIDI_EOB;
      bidi_it->ch_len = 1;
      bidi_it->nchars = 1;
      bidi_it->disp_pos = (string_p ? bidi_it->string.schars : ZV);
      bidi_it->disp_prop = 0;
    }
  else
    {
      /* Fetch the character at BYTEPOS.  If it is covered by a
	 display string, treat the entire run of covered characters as
	 a single character u+FFFC.  */
      curchar = bidi_fetch_char (bidi_it->bytepos, bidi_it->charpos,
				 &bidi_it->disp_pos, &bidi_it->disp_prop,
				 &bidi_it->string, bidi_it->frame_window_p,
				 &bidi_it->ch_len, &bidi_it->nchars);
    }
  bidi_it->ch = curchar;

  /* Don't apply directional override here, as all the types we handle
     below will not be affected by the override anyway, and we need
     the original type unaltered.  The override will be applied in
     bidi_resolve_weak.  */
  type = bidi_get_type (curchar, NEUTRAL_DIR);
  bidi_it->orig_type = type;
  bidi_check_type (bidi_it->orig_type);

  if (type != PDF)
    bidi_it->prev_was_pdf = 0;

  bidi_it->type_after_w1 = UNKNOWN_BT;

  switch (type)
    {
      case RLE:	/* X2 */
      case RLO:	/* X4 */
	bidi_it->type_after_w1 = type;
	bidi_check_type (bidi_it->type_after_w1);
	type = WEAK_BN; /* X9/Retaining */
	if (bidi_it->ignore_bn_limit <= -1)
	  {
	    if (current_level <= BIDI_MAXLEVEL - 4)
	      {
		/* Compute the least odd embedding level greater than
		   the current level.  */
		new_level = ((current_level + 1) & ~1) + 1;
		if (bidi_it->type_after_w1 == RLE)
		  override = NEUTRAL_DIR;
		else
		  override = R2L;
		if (current_level == BIDI_MAXLEVEL - 4)
		  bidi_it->invalid_rl_levels = 0;
		bidi_push_embedding_level (bidi_it, new_level, override);
	      }
	    else
	      {
		bidi_it->invalid_levels++;
		/* See the commentary about invalid_rl_levels below.  */
		if (bidi_it->invalid_rl_levels < 0)
		  bidi_it->invalid_rl_levels = 0;
		bidi_it->invalid_rl_levels++;
	      }
	  }
	else if (bidi_it->prev.type_after_w1 == WEAK_EN /* W5/Retaining */
		 || (bidi_it->next_en_pos > bidi_it->charpos
		     && bidi_it->next_en_type == WEAK_EN))
	  type = WEAK_EN;
	break;
      case LRE:	/* X3 */
      case LRO:	/* X5 */
	bidi_it->type_after_w1 = type;
	bidi_check_type (bidi_it->type_after_w1);
	type = WEAK_BN; /* X9/Retaining */
	if (bidi_it->ignore_bn_limit <= -1)
	  {
	    if (current_level <= BIDI_MAXLEVEL - 5)
	      {
		/* Compute the least even embedding level greater than
		   the current level.  */
		new_level = ((current_level + 2) & ~1);
		if (bidi_it->type_after_w1 == LRE)
		  override = NEUTRAL_DIR;
		else
		  override = L2R;
		bidi_push_embedding_level (bidi_it, new_level, override);
	      }
	    else
	      {
		bidi_it->invalid_levels++;
		/* invalid_rl_levels counts invalid levels encountered
		   while the embedding level was already too high for
		   LRE/LRO, but not for RLE/RLO.  That is because
		   there may be exactly one PDF which we should not
		   ignore even though invalid_levels is non-zero.
		   invalid_rl_levels helps to know what PDF is
		   that.  */
		if (bidi_it->invalid_rl_levels >= 0)
		  bidi_it->invalid_rl_levels++;
	      }
	  }
	else if (bidi_it->prev.type_after_w1 == WEAK_EN /* W5/Retaining */
		 || (bidi_it->next_en_pos > bidi_it->charpos
		     && bidi_it->next_en_type == WEAK_EN))
	  type = WEAK_EN;
	break;
      case PDF:	/* X7 */
	bidi_it->type_after_w1 = type;
	bidi_check_type (bidi_it->type_after_w1);
	type = WEAK_BN; /* X9/Retaining */
	if (bidi_it->ignore_bn_limit <= -1)
	  {
	    if (!bidi_it->invalid_rl_levels)
	      {
		new_level = bidi_pop_embedding_level (bidi_it);
		bidi_it->invalid_rl_levels = -1;
		if (bidi_it->invalid_levels)
		  bidi_it->invalid_levels--;
		/* else nothing: UAX#9 says to ignore invalid PDFs */
	      }
	    if (!bidi_it->invalid_levels)
	      new_level = bidi_pop_embedding_level (bidi_it);
	    else
	      {
		bidi_it->invalid_levels--;
		bidi_it->invalid_rl_levels--;
	      }
	  }
	else if (bidi_it->prev.type_after_w1 == WEAK_EN /* W5/Retaining */
		 || (bidi_it->next_en_pos > bidi_it->charpos
		     && bidi_it->next_en_type == WEAK_EN))
	  type = WEAK_EN;
	break;
      default:
	/* Nothing.  */
	break;
    }

  bidi_it->type = type;
  bidi_check_type (bidi_it->type);

  return new_level;
}

/* Given an iterator state in BIDI_IT, advance one character position
   in the buffer/string to the next character (in the logical order),
   resolve any explicit embeddings and directional overrides, and
   return the embedding level of the character after resolving
   explicit directives and ignoring empty embeddings.  */
static int
bidi_resolve_explicit (struct bidi_it *bidi_it)
{
  int prev_level = bidi_it->level_stack[bidi_it->stack_idx].level;
  int new_level  = bidi_resolve_explicit_1 (bidi_it);
  ptrdiff_t eob = bidi_it->string.s ? bidi_it->string.schars : ZV;
  const unsigned char *s
    = (STRINGP (bidi_it->string.lstring)
       ? SDATA (bidi_it->string.lstring)
       : bidi_it->string.s);

  if (prev_level < new_level
      && bidi_it->type == WEAK_BN
      && bidi_it->ignore_bn_limit == -1 /* only if not already known */
      && bidi_it->charpos < eob		/* not already at EOB */
      && bidi_explicit_dir_char (bidi_char_at_pos (bidi_it->bytepos
						   + bidi_it->ch_len, s,
						   bidi_it->string.unibyte)))
    {
      /* Avoid pushing and popping embedding levels if the level run
	 is empty, as this breaks level runs where it shouldn't.
	 UAX#9 removes all the explicit embedding and override codes,
	 so empty embeddings disappear without a trace.  We need to
	 behave as if we did the same.  */
      struct bidi_it saved_it;
      int level = prev_level;

      bidi_copy_it (&saved_it, bidi_it);

      while (bidi_explicit_dir_char (bidi_char_at_pos (bidi_it->bytepos
						       + bidi_it->ch_len, s,
						       bidi_it->string.unibyte)))
	{
	  /* This advances to the next character, skipping any
	     characters covered by display strings.  */
	  level = bidi_resolve_explicit_1 (bidi_it);
	  /* If string.lstring was relocated inside bidi_resolve_explicit_1,
	     a pointer to its data is no longer valid.  */
	  if (STRINGP (bidi_it->string.lstring))
	    s = SDATA (bidi_it->string.lstring);
	}

      if (bidi_it->nchars <= 0)
	emacs_abort ();
      if (level == prev_level)	/* empty embedding */
	saved_it.ignore_bn_limit = bidi_it->charpos + bidi_it->nchars;
      else			/* this embedding is non-empty */
	saved_it.ignore_bn_limit = -2;

      bidi_copy_it (bidi_it, &saved_it);
      if (bidi_it->ignore_bn_limit > -1)
	{
	  /* We pushed a level, but we shouldn't have.  Undo that. */
	  if (!bidi_it->invalid_rl_levels)
	    {
	      new_level = bidi_pop_embedding_level (bidi_it);
	      bidi_it->invalid_rl_levels = -1;
	      if (bidi_it->invalid_levels)
		bidi_it->invalid_levels--;
	    }
	  if (!bidi_it->invalid_levels)
	    new_level = bidi_pop_embedding_level (bidi_it);
	  else
	    {
	      bidi_it->invalid_levels--;
	      bidi_it->invalid_rl_levels--;
	    }
	}
    }

  if (bidi_it->type == NEUTRAL_B)	/* X8 */
    {
      bidi_set_paragraph_end (bidi_it);
      /* This is needed by bidi_resolve_weak below, and in L1.  */
      bidi_it->type_after_w1 = bidi_it->type;
      bidi_check_type (bidi_it->type_after_w1);
    }

  return new_level;
}

/* Advance in the buffer/string, resolve weak types and return the
   type of the next character after weak type resolution.  */
static bidi_type_t
bidi_resolve_weak (struct bidi_it *bidi_it)
{
  bidi_type_t type;
  bidi_dir_t override;
  int prev_level = bidi_it->level_stack[bidi_it->stack_idx].level;
  int new_level  = bidi_resolve_explicit (bidi_it);
  int next_char;
  bidi_type_t type_of_next;
  struct bidi_it saved_it;
  ptrdiff_t eob
    = ((STRINGP (bidi_it->string.lstring) || bidi_it->string.s)
       ? bidi_it->string.schars : ZV);

  type = bidi_it->type;
  override = bidi_it->level_stack[bidi_it->stack_idx].override;

  if (type == UNKNOWN_BT
      || type == LRE
      || type == LRO
      || type == RLE
      || type == RLO
      || type == PDF)
    emacs_abort ();

  if (new_level != prev_level
      || bidi_it->type == NEUTRAL_B)
    {
      /* We've got a new embedding level run, compute the directional
         type of sor and initialize per-run variables (UAX#9, clause
         X10).  */
      bidi_set_sor_type (bidi_it, prev_level, new_level);
    }
  else if (type == NEUTRAL_S || type == NEUTRAL_WS
	   || type == WEAK_BN || type == STRONG_AL)
    bidi_it->type_after_w1 = type;	/* needed in L1 */
  bidi_check_type (bidi_it->type_after_w1);

  /* Level and directional override status are already recorded in
     bidi_it, and do not need any change; see X6.  */
  if (override == R2L)		/* X6 */
    type = STRONG_R;
  else if (override == L2R)
    type = STRONG_L;
  else
    {
      if (type == WEAK_NSM)	/* W1 */
	{
	  /* Note that we don't need to consider the case where the
	     prev character has its type overridden by an RLO or LRO,
	     because then either the type of this NSM would have been
	     also overridden, or the previous character is outside the
	     current level run, and thus not relevant to this NSM.
	     This is why NSM gets the type_after_w1 of the previous
	     character.  */
	  if (bidi_it->prev.type_after_w1 != UNKNOWN_BT
	      /* if type_after_w1 is NEUTRAL_B, this NSM is at sor */
	      && bidi_it->prev.type_after_w1 != NEUTRAL_B)
	    type = bidi_it->prev.type_after_w1;
	  else if (bidi_it->sor == R2L)
	    type = STRONG_R;
	  else if (bidi_it->sor == L2R)
	    type = STRONG_L;
	  else /* shouldn't happen! */
	    emacs_abort ();
	}
      if (type == WEAK_EN	/* W2 */
	  && bidi_it->last_strong.type_after_w1 == STRONG_AL)
	type = WEAK_AN;
      else if (type == STRONG_AL) /* W3 */
	type = STRONG_R;
      else if ((type == WEAK_ES	/* W4 */
		&& bidi_it->prev.type_after_w1 == WEAK_EN
		&& bidi_it->prev.orig_type == WEAK_EN)
	       || (type == WEAK_CS
		   && ((bidi_it->prev.type_after_w1 == WEAK_EN
			&& bidi_it->prev.orig_type == WEAK_EN)
		       || bidi_it->prev.type_after_w1 == WEAK_AN)))
	{
	  const unsigned char *s
	    = (STRINGP (bidi_it->string.lstring)
	       ? SDATA (bidi_it->string.lstring)
	       : bidi_it->string.s);

	  next_char = (bidi_it->charpos + bidi_it->nchars >= eob
		       ? BIDI_EOB
		       : bidi_char_at_pos (bidi_it->bytepos + bidi_it->ch_len,
					   s, bidi_it->string.unibyte));
	  type_of_next = bidi_get_type (next_char, override);

	  if (type_of_next == WEAK_BN
	      || bidi_explicit_dir_char (next_char))
	    {
	      bidi_copy_it (&saved_it, bidi_it);
	      while (bidi_resolve_explicit (bidi_it) == new_level
		     && bidi_it->type == WEAK_BN)
		;
	      type_of_next = bidi_it->type;
	      bidi_copy_it (bidi_it, &saved_it);
	    }

	  /* If the next character is EN, but the last strong-type
	     character is AL, that next EN will be changed to AN when
	     we process it in W2 above.  So in that case, this ES
	     should not be changed into EN.  */
	  if (type == WEAK_ES
	      && type_of_next == WEAK_EN
	      && bidi_it->last_strong.type_after_w1 != STRONG_AL)
	    type = WEAK_EN;
	  else if (type == WEAK_CS)
	    {
	      if (bidi_it->prev.type_after_w1 == WEAK_AN
		  && (type_of_next == WEAK_AN
		      /* If the next character is EN, but the last
			 strong-type character is AL, EN will be later
			 changed to AN when we process it in W2 above.
			 So in that case, this ES should not be
			 changed into EN.  */
		      || (type_of_next == WEAK_EN
			  && bidi_it->last_strong.type_after_w1 == STRONG_AL)))
		type = WEAK_AN;
	      else if (bidi_it->prev.type_after_w1 == WEAK_EN
		       && type_of_next == WEAK_EN
		       && bidi_it->last_strong.type_after_w1 != STRONG_AL)
		type = WEAK_EN;
	    }
	}
      else if (type == WEAK_ET	/* W5: ET with EN before or after it */
	       || type == WEAK_BN)	/* W5/Retaining */
	{
	  if (bidi_it->prev.type_after_w1 == WEAK_EN) /* ET/BN w/EN before it */
	    type = WEAK_EN;
	  else if (bidi_it->next_en_pos > bidi_it->charpos
		   && bidi_it->next_en_type != WEAK_BN)
	    {
	      if (bidi_it->next_en_type == WEAK_EN) /* ET/BN with EN after it */
		type = WEAK_EN;
	    }
	  else if (bidi_it->next_en_pos >=0)
	    {
	      ptrdiff_t en_pos = bidi_it->charpos + bidi_it->nchars;
	      const unsigned char *s = (STRINGP (bidi_it->string.lstring)
					? SDATA (bidi_it->string.lstring)
					: bidi_it->string.s);

	      if (bidi_it->nchars <= 0)
		emacs_abort ();
	      next_char
		= (bidi_it->charpos + bidi_it->nchars >= eob
		   ? BIDI_EOB
		   : bidi_char_at_pos (bidi_it->bytepos + bidi_it->ch_len, s,
				       bidi_it->string.unibyte));
	      type_of_next = bidi_get_type (next_char, override);

	      if (type_of_next == WEAK_ET
		  || type_of_next == WEAK_BN
		  || bidi_explicit_dir_char (next_char))
		{
		  bidi_copy_it (&saved_it, bidi_it);
		  while (bidi_resolve_explicit (bidi_it) == new_level
			 && (bidi_it->type == WEAK_BN
			     || bidi_it->type == WEAK_ET))
		    ;
		  type_of_next = bidi_it->type;
		  en_pos = bidi_it->charpos;
		  bidi_copy_it (bidi_it, &saved_it);
		}
	      /* Remember this position, to speed up processing of the
		 next ETs.  */
	      bidi_it->next_en_pos = en_pos;
	      if (type_of_next == WEAK_EN)
		{
		  /* If the last strong character is AL, the EN we've
		     found will become AN when we get to it (W2). */
		  if (bidi_it->last_strong.type_after_w1 == STRONG_AL)
		    type_of_next = WEAK_AN;
		  else if (type == WEAK_BN)
		    type = NEUTRAL_ON; /* W6/Retaining */
		  else
		    type = WEAK_EN;
		}
	      else if (type_of_next == NEUTRAL_B)
		/* Record the fact that there are no more ENs from
		   here to the end of paragraph, to avoid entering the
		   loop above ever again in this paragraph.  */
		bidi_it->next_en_pos = -1;
	      /* Record the type of the character where we ended our search.  */
	      bidi_it->next_en_type = type_of_next;
	    }
	}
    }

  if (type == WEAK_ES || type == WEAK_ET || type == WEAK_CS /* W6 */
      || (type == WEAK_BN
	  && (bidi_it->prev.type_after_w1 == WEAK_CS	    /* W6/Retaining */
	      || bidi_it->prev.type_after_w1 == WEAK_ES
	      || bidi_it->prev.type_after_w1 == WEAK_ET)))
    type = NEUTRAL_ON;

  /* Store the type we've got so far, before we clobber it with strong
     types in W7 and while resolving neutral types.  But leave alone
     the original types that were recorded above, because we will need
     them for the L1 clause.  */
  if (bidi_it->type_after_w1 == UNKNOWN_BT)
    bidi_it->type_after_w1 = type;
  bidi_check_type (bidi_it->type_after_w1);

  if (type == WEAK_EN)	/* W7 */
    {
      if ((bidi_it->last_strong.type_after_w1 == STRONG_L)
	  || (bidi_it->last_strong.type == UNKNOWN_BT && bidi_it->sor == L2R))
	type = STRONG_L;
    }

  bidi_it->type = type;
  bidi_check_type (bidi_it->type);
  return type;
}

/* Resolve the type of a neutral character according to the type of
   surrounding strong text and the current embedding level.  */
static bidi_type_t
bidi_resolve_neutral_1 (bidi_type_t prev_type, bidi_type_t next_type, int lev)
{
  /* N1: European and Arabic numbers are treated as though they were R.  */
  if (next_type == WEAK_EN || next_type == WEAK_AN)
    next_type = STRONG_R;
  if (prev_type == WEAK_EN || prev_type == WEAK_AN)
    prev_type = STRONG_R;

  if (next_type == prev_type)	/* N1 */
    return next_type;
  else if ((lev & 1) == 0)	/* N2 */
    return STRONG_L;
  else
    return STRONG_R;
}

static bidi_type_t
bidi_resolve_neutral (struct bidi_it *bidi_it)
{
  int prev_level = bidi_it->level_stack[bidi_it->stack_idx].level;
  bidi_type_t type = bidi_resolve_weak (bidi_it);
  int current_level = bidi_it->level_stack[bidi_it->stack_idx].level;

  if (!(type == STRONG_R
	|| type == STRONG_L
	|| type == WEAK_BN
	|| type == WEAK_EN
	|| type == WEAK_AN
	|| type == NEUTRAL_B
	|| type == NEUTRAL_S
	|| type == NEUTRAL_WS
	|| type == NEUTRAL_ON))
    emacs_abort ();

  if ((type != NEUTRAL_B /* Don't risk entering the long loop below if
			    we are already at paragraph end.  */
       && bidi_get_category (type) == NEUTRAL)
      || (type == WEAK_BN && prev_level == current_level))
    {
      if (bidi_it->next_for_neutral.type != UNKNOWN_BT)
	type = bidi_resolve_neutral_1 (bidi_it->prev_for_neutral.type,
				       bidi_it->next_for_neutral.type,
				       current_level);
      /* The next two "else if" clauses are shortcuts for the
	 important special case when we have a long sequence of
	 neutral or WEAK_BN characters, such as whitespace or nulls or
	 other control characters, on the base embedding level of the
	 paragraph, and that sequence goes all the way to the end of
	 the paragraph and follows a character whose resolved
	 directionality is identical to the base embedding level.
	 (This is what happens in a buffer with plain L2R text that
	 happens to include long sequences of control characters.)  By
	 virtue of N1, the result of examining this long sequence will
	 always be either STRONG_L or STRONG_R, depending on the base
	 embedding level.  So we use this fact directly instead of
	 entering the expensive loop in the "else" clause.  */
      else if (current_level == 0
	       && bidi_it->prev_for_neutral.type == STRONG_L
	       && !bidi_explicit_dir_char (bidi_it->ch))
	type = bidi_resolve_neutral_1 (bidi_it->prev_for_neutral.type,
				       STRONG_L, current_level);
      else if (/* current level is 1 */
	       current_level == 1
	       /* base embedding level is also 1 */
	       && bidi_it->level_stack[0].level == 1
	       /* previous character is one of those considered R for
		  the purposes of W5 */
	       && (bidi_it->prev_for_neutral.type == STRONG_R
		   || bidi_it->prev_for_neutral.type == WEAK_EN
		   || bidi_it->prev_for_neutral.type == WEAK_AN)
	       && !bidi_explicit_dir_char (bidi_it->ch))
	type = bidi_resolve_neutral_1 (bidi_it->prev_for_neutral.type,
				       STRONG_R, current_level);
      else
	{
	  /* Arrrgh!!  The UAX#9 algorithm is too deeply entrenched in
	     the assumption of batch-style processing; see clauses W4,
	     W5, and especially N1, which require to look far forward
	     (as well as back) in the buffer/string.  May the fleas of
	     a thousand camels infest the armpits of those who design
	     supposedly general-purpose algorithms by looking at their
	     own implementations, and fail to consider other possible
	     implementations!  */
	  struct bidi_it saved_it;
	  bidi_type_t next_type;

	  if (bidi_it->scan_dir == -1)
	    emacs_abort ();

	  bidi_copy_it (&saved_it, bidi_it);
	  /* Scan the text forward until we find the first non-neutral
	     character, and then use that to resolve the neutral we
	     are dealing with now.  We also cache the scanned iterator
	     states, to salvage some of the effort later.  */
	  bidi_cache_iterator_state (bidi_it, 0);
	  do {
	    /* Record the info about the previous character, so that
	       it will be cached below with this state.  */
	    if (bidi_it->type_after_w1 != WEAK_BN /* W1/Retaining */
		&& bidi_it->type != WEAK_BN)
	      bidi_remember_char (&bidi_it->prev, bidi_it);
	    type = bidi_resolve_weak (bidi_it);
	    /* Paragraph separators have their levels fully resolved
	       at this point, so cache them as resolved.  */
	    bidi_cache_iterator_state (bidi_it, type == NEUTRAL_B);
	    /* FIXME: implement L1 here, by testing for a newline and
	       resetting the level for any sequence of whitespace
	       characters adjacent to it.  */
	  } while (!(type == NEUTRAL_B
		     || (type != WEAK_BN
			 && bidi_get_category (type) != NEUTRAL)
		     /* This is all per level run, so stop when we
			reach the end of this level run.  */
		     || (bidi_it->level_stack[bidi_it->stack_idx].level
			 != current_level)));

	  bidi_remember_char (&saved_it.next_for_neutral, bidi_it);

	  switch (type)
	    {
	      case STRONG_L:
	      case STRONG_R:
	      case STRONG_AL:
		/* Actually, STRONG_AL cannot happen here, because
		   bidi_resolve_weak converts it to STRONG_R, per W3.  */
		eassert (type != STRONG_AL);
		next_type = type;
		break;
	      case WEAK_EN:
	      case WEAK_AN:
		/* N1: ``European and Arabic numbers are treated as
		   though they were R.''  */
		next_type = STRONG_R;
		break;
	      case WEAK_BN:
		if (!bidi_explicit_dir_char (bidi_it->ch))
		  emacs_abort (); /* can't happen: BNs are skipped */
		/* FALLTHROUGH */
	      case NEUTRAL_B:
		/* Marched all the way to the end of this level run.
		   We need to use the eor type, whose information is
		   stored by bidi_set_sor_type in the prev_for_neutral
		   member.  */
		if (saved_it.type != WEAK_BN
		    || bidi_get_category (bidi_it->prev.type_after_w1) == NEUTRAL)
		  next_type = bidi_it->prev_for_neutral.type;
		else
		  {
		    /* This is a BN which does not adjoin neutrals.
		       Leave its type alone.  */
		    bidi_copy_it (bidi_it, &saved_it);
		    return bidi_it->type;
		  }
		break;
	      default:
		emacs_abort ();
	    }
	  type = bidi_resolve_neutral_1 (saved_it.prev_for_neutral.type,
					 next_type, current_level);
	  saved_it.next_for_neutral.type = next_type;
	  saved_it.type = type;
	  bidi_check_type (next_type);
	  bidi_check_type (type);
	  bidi_copy_it (bidi_it, &saved_it);
	}
    }
  return type;
}

/* Given an iterator state in BIDI_IT, advance one character position
   in the buffer/string to the next character (in the logical order),
   resolve the bidi type of that next character, and return that
   type.  */
static bidi_type_t
bidi_type_of_next_char (struct bidi_it *bidi_it)
{
  bidi_type_t type;

  /* This should always be called during a forward scan.  */
  if (bidi_it->scan_dir != 1)
    emacs_abort ();

  /* Reset the limit until which to ignore BNs if we step out of the
     area where we found only empty levels.  */
  if ((bidi_it->ignore_bn_limit > -1
       && bidi_it->ignore_bn_limit <= bidi_it->charpos)
      || (bidi_it->ignore_bn_limit == -2
	  && !bidi_explicit_dir_char (bidi_it->ch)))
    bidi_it->ignore_bn_limit = -1;

  type = bidi_resolve_neutral (bidi_it);

  return type;
}

/* Given an iterator state BIDI_IT, advance one character position in
   the buffer/string to the next character (in the current scan
   direction), resolve the embedding and implicit levels of that next
   character, and return the resulting level.  */
static int
bidi_level_of_next_char (struct bidi_it *bidi_it)
{
  bidi_type_t type;
  int level, prev_level = -1;
  struct bidi_saved_info next_for_neutral;
  ptrdiff_t next_char_pos = -2;

  if (bidi_it->scan_dir == 1)
    {
      ptrdiff_t eob
	= ((bidi_it->string.s || STRINGP (bidi_it->string.lstring))
	   ? bidi_it->string.schars : ZV);

      /* There's no sense in trying to advance if we hit end of text.  */
      if (bidi_it->charpos >= eob)
	return bidi_it->resolved_level;

      /* Record the info about the previous character.  */
      if (bidi_it->type_after_w1 != WEAK_BN /* W1/Retaining */
	  && bidi_it->type != WEAK_BN)
	bidi_remember_char (&bidi_it->prev, bidi_it);
      if (bidi_it->type_after_w1 == STRONG_R
	  || bidi_it->type_after_w1 == STRONG_L
	  || bidi_it->type_after_w1 == STRONG_AL)
	bidi_remember_char (&bidi_it->last_strong, bidi_it);
      /* FIXME: it sounds like we don't need both prev and
	 prev_for_neutral members, but I'm leaving them both for now.  */
      if (bidi_it->type == STRONG_R || bidi_it->type == STRONG_L
	  || bidi_it->type == WEAK_EN || bidi_it->type == WEAK_AN)
	bidi_remember_char (&bidi_it->prev_for_neutral, bidi_it);

      /* If we overstepped the characters used for resolving neutrals
	 and whitespace, invalidate their info in the iterator.  */
      if (bidi_it->charpos >= bidi_it->next_for_neutral.charpos)
	bidi_it->next_for_neutral.type = UNKNOWN_BT;
      if (bidi_it->next_en_pos >= 0
	  && bidi_it->charpos >= bidi_it->next_en_pos)
	{
	  bidi_it->next_en_pos = 0;
	  bidi_it->next_en_type = UNKNOWN_BT;
	}
      if (bidi_it->next_for_ws.type != UNKNOWN_BT
	  && bidi_it->charpos >= bidi_it->next_for_ws.charpos)
	bidi_it->next_for_ws.type = UNKNOWN_BT;

      /* This must be taken before we fill the iterator with the info
	 about the next char.  If we scan backwards, the iterator
	 state must be already cached, so there's no need to know the
	 embedding level of the previous character, since we will be
	 returning to our caller shortly.  */
      prev_level = bidi_it->level_stack[bidi_it->stack_idx].level;
    }
  next_for_neutral = bidi_it->next_for_neutral;

  /* Perhaps the character we want is already cached.  If it is, the
     call to bidi_cache_find below will return a type other than
     UNKNOWN_BT.  */
  if (bidi_cache_idx > bidi_cache_start && !bidi_it->first_elt)
    {
      int bob = ((bidi_it->string.s || STRINGP (bidi_it->string.lstring))
		 ? 0 : 1);
      if (bidi_it->scan_dir > 0)
	{
	  if (bidi_it->nchars <= 0)
	    emacs_abort ();
	  next_char_pos = bidi_it->charpos + bidi_it->nchars;
	}
      else if (bidi_it->charpos >= bob)
	/* Implementation note: we allow next_char_pos to be as low as
	   0 for buffers or -1 for strings, and that is okay because
	   that's the "position" of the sentinel iterator state we
	   cached at the beginning of the iteration.  */
	next_char_pos = bidi_it->charpos - 1;
      if (next_char_pos >= bob - 1)
	type = bidi_cache_find (next_char_pos, -1, bidi_it);
      else
	type = UNKNOWN_BT;
    }
  else
    type = UNKNOWN_BT;
  if (type != UNKNOWN_BT)
    {
      /* Don't lose the information for resolving neutrals!  The
	 cached states could have been cached before their
	 next_for_neutral member was computed.  If we are on our way
	 forward, we can simply take the info from the previous
	 state.  */
      if (bidi_it->scan_dir == 1
	  && bidi_it->next_for_neutral.type == UNKNOWN_BT)
	bidi_it->next_for_neutral = next_for_neutral;

      /* If resolved_level is -1, it means this state was cached
	 before it was completely resolved, so we cannot return
	 it.  */
      if (bidi_it->resolved_level != -1)
	return bidi_it->resolved_level;
    }
  if (bidi_it->scan_dir == -1)
    /* If we are going backwards, the iterator state is already cached
       from previous scans, and should be fully resolved.  */
    emacs_abort ();

  if (type == UNKNOWN_BT)
    type = bidi_type_of_next_char (bidi_it);

  if (type == NEUTRAL_B)
    return bidi_it->resolved_level;

  level = bidi_it->level_stack[bidi_it->stack_idx].level;
  if ((bidi_get_category (type) == NEUTRAL /* && type != NEUTRAL_B */)
      || (type == WEAK_BN && prev_level == level))
    {
      if (bidi_it->next_for_neutral.type == UNKNOWN_BT)
	emacs_abort ();

      /* If the cached state shows a neutral character, it was not
	 resolved by bidi_resolve_neutral, so do it now.  */
      type = bidi_resolve_neutral_1 (bidi_it->prev_for_neutral.type,
				     bidi_it->next_for_neutral.type,
				     level);
    }

  if (!(type == STRONG_R
	|| type == STRONG_L
	|| type == WEAK_BN
	|| type == WEAK_EN
	|| type == WEAK_AN))
    emacs_abort ();
  bidi_it->type = type;
  bidi_check_type (bidi_it->type);

  /* For L1 below, we need to know, for each WS character, whether
     it belongs to a sequence of WS characters preceding a newline
     or a TAB or a paragraph separator.  */
  if (bidi_it->orig_type == NEUTRAL_WS
      && bidi_it->next_for_ws.type == UNKNOWN_BT)
    {
      int ch;
      ptrdiff_t clen = bidi_it->ch_len;
      ptrdiff_t bpos = bidi_it->bytepos;
      ptrdiff_t cpos = bidi_it->charpos;
      ptrdiff_t disp_pos = bidi_it->disp_pos;
      ptrdiff_t nc = bidi_it->nchars;
      struct bidi_string_data bs = bidi_it->string;
      bidi_type_t chtype;
      bool fwp = bidi_it->frame_window_p;
      int dpp = bidi_it->disp_prop;

      if (bidi_it->nchars <= 0)
	emacs_abort ();
      do {
	ch = bidi_fetch_char (bpos += clen, cpos += nc, &disp_pos, &dpp, &bs,
			      fwp, &clen, &nc);
	if (ch == '\n' || ch == BIDI_EOB)
	  chtype = NEUTRAL_B;
	else
	  chtype = bidi_get_type (ch, NEUTRAL_DIR);
      } while (chtype == NEUTRAL_WS || chtype == WEAK_BN
	       || bidi_explicit_dir_char (ch)); /* L1/Retaining */
      bidi_it->next_for_ws.type = chtype;
      bidi_check_type (bidi_it->next_for_ws.type);
      bidi_it->next_for_ws.charpos = cpos;
      bidi_it->next_for_ws.bytepos = bpos;
    }

  /* Resolve implicit levels, with a twist: PDFs get the embedding
     level of the embedding they terminate.  See below for the
     reason.  */
  if (bidi_it->orig_type == PDF
      /* Don't do this if this formatting code didn't change the
	 embedding level due to invalid or empty embeddings.  */
      && prev_level != level)
    {
      /* Don't look in UAX#9 for the reason for this: it's our own
	 private quirk.  The reason is that we want the formatting
	 codes to be delivered so that they bracket the text of their
	 embedding.  For example, given the text

	     {RLO}teST{PDF}

	 we want it to be displayed as

	     {PDF}STet{RLO}

	 not as

	     STet{RLO}{PDF}

	 which will result because we bump up the embedding level as
	 soon as we see the RLO and pop it as soon as we see the PDF,
	 so RLO itself has the same embedding level as "teST", and
	 thus would be normally delivered last, just before the PDF.
	 The switch below fiddles with the level of PDF so that this
	 ugly side effect does not happen.

	 (This is, of course, only important if the formatting codes
	 are actually displayed, but Emacs does need to display them
	 if the user wants to.)  */
      level = prev_level;
    }
  else if (bidi_it->orig_type == NEUTRAL_B /* L1 */
	   || bidi_it->orig_type == NEUTRAL_S
	   || bidi_it->ch == '\n' || bidi_it->ch == BIDI_EOB
	   || (bidi_it->orig_type == NEUTRAL_WS
	       && (bidi_it->next_for_ws.type == NEUTRAL_B
		   || bidi_it->next_for_ws.type == NEUTRAL_S)))
    level = bidi_it->level_stack[0].level;
  else if ((level & 1) == 0) /* I1 */
    {
      if (type == STRONG_R)
	level++;
      else if (type == WEAK_EN || type == WEAK_AN)
	level += 2;
    }
  else			/* I2 */
    {
      if (type == STRONG_L || type == WEAK_EN || type == WEAK_AN)
	level++;
    }

  bidi_it->resolved_level = level;
  return level;
}

/* Move to the other edge of a level given by LEVEL.  If END_FLAG,
   we are at the end of a level, and we need to prepare to
   resume the scan of the lower level.

   If this level's other edge is cached, we simply jump to it, filling
   the iterator structure with the iterator state on the other edge.
   Otherwise, we walk the buffer or string until we come back to the
   same level as LEVEL.

   Note: we are not talking here about a ``level run'' in the UAX#9
   sense of the term, but rather about a ``level'' which includes
   all the levels higher than it.  In other words, given the levels
   like this:

         11111112222222333333334443343222222111111112223322111
                A      B                    C

   and assuming we are at point A scanning left to right, this
   function moves to point C, whereas the UAX#9 ``level 2 run'' ends
   at point B.  */
static void
bidi_find_other_level_edge (struct bidi_it *bidi_it, int level, bool end_flag)
{
  int dir = end_flag ? -bidi_it->scan_dir : bidi_it->scan_dir;
  ptrdiff_t idx;

  /* Try the cache first.  */
  if ((idx = bidi_cache_find_level_change (level, dir, end_flag))
      >= bidi_cache_start)
    bidi_cache_fetch_state (idx, bidi_it);
  else
    {
      int new_level;

      /* If we are at end of level, its edges must be cached.  */
      if (end_flag)
	emacs_abort ();

      bidi_cache_iterator_state (bidi_it, 1);
      do {
	new_level = bidi_level_of_next_char (bidi_it);
	bidi_cache_iterator_state (bidi_it, 1);
      } while (new_level >= level);
    }
}

void
bidi_move_to_visually_next (struct bidi_it *bidi_it)
{
  int old_level, new_level, next_level;
  struct bidi_it sentinel;
  struct gcpro gcpro1;

  if (bidi_it->charpos < 0 || bidi_it->bytepos < 0)
    emacs_abort ();

  if (bidi_it->scan_dir == 0)
    {
      bidi_it->scan_dir = 1;	/* default to logical order */
    }

  /* The code below can call eval, and thus cause GC.  If we are
     iterating a Lisp string, make sure it won't be GCed.  */
  if (STRINGP (bidi_it->string.lstring))
    GCPRO1 (bidi_it->string.lstring);

  /* If we just passed a newline, initialize for the next line.  */
  if (!bidi_it->first_elt
      && (bidi_it->ch == '\n' || bidi_it->ch == BIDI_EOB))
    bidi_line_init (bidi_it);

  /* Prepare the sentinel iterator state, and cache it.  When we bump
     into it, scanning backwards, we'll know that the last non-base
     level is exhausted.  */
  if (bidi_cache_idx == bidi_cache_start)
    {
      bidi_copy_it (&sentinel, bidi_it);
      if (bidi_it->first_elt)
	{
	  sentinel.charpos--;	/* cached charpos needs to be monotonic */
	  sentinel.bytepos--;
	  sentinel.ch = '\n';	/* doesn't matter, but why not? */
	  sentinel.ch_len = 1;
	  sentinel.nchars = 1;
	}
      bidi_cache_iterator_state (&sentinel, 1);
    }

  old_level = bidi_it->resolved_level;
  new_level = bidi_level_of_next_char (bidi_it);

  /* Reordering of resolved levels (clause L2) is implemented by
     jumping to the other edge of the level and flipping direction of
     scanning the text whenever we find a level change.  */
  if (new_level != old_level)
    {
      bool ascending = new_level > old_level;
      int level_to_search = ascending ? old_level + 1 : old_level;
      int incr = ascending ? 1 : -1;
      int expected_next_level = old_level + incr;

      /* Jump (or walk) to the other edge of this level.  */
      bidi_find_other_level_edge (bidi_it, level_to_search, !ascending);
      /* Switch scan direction and peek at the next character in the
	 new direction.  */
      bidi_it->scan_dir = -bidi_it->scan_dir;

      /* The following loop handles the case where the resolved level
	 jumps by more than one.  This is typical for numbers inside a
	 run of text with left-to-right embedding direction, but can
	 also happen in other situations.  In those cases the decision
	 where to continue after a level change, and in what direction,
	 is tricky.  For example, given a text like below:

	          abcdefgh
	          11336622

	 (where the numbers below the text show the resolved levels),
	 the result of reordering according to UAX#9 should be this:

		  efdcghba

	 This is implemented by the loop below which flips direction
	 and jumps to the other edge of the level each time it finds
	 the new level not to be the expected one.  The expected level
	 is always one more or one less than the previous one.  */
      next_level = bidi_peek_at_next_level (bidi_it);
      while (next_level != expected_next_level)
	{
	  expected_next_level += incr;
	  level_to_search += incr;
	  bidi_find_other_level_edge (bidi_it, level_to_search, !ascending);
	  bidi_it->scan_dir = -bidi_it->scan_dir;
	  next_level = bidi_peek_at_next_level (bidi_it);
	}

      /* Finally, deliver the next character in the new direction.  */
      next_level = bidi_level_of_next_char (bidi_it);
    }

  /* Take note when we have just processed the newline that precedes
     the end of the paragraph.  The next time we are about to be
     called, set_iterator_to_next will automatically reinit the
     paragraph direction, if needed.  We do this at the newline before
     the paragraph separator, because the next character might not be
     the first character of the next paragraph, due to the bidi
     reordering, whereas we _must_ know the paragraph base direction
     _before_ we process the paragraph's text, since the base
     direction affects the reordering.  */
  if (bidi_it->scan_dir == 1
      && (bidi_it->ch == '\n' || bidi_it->ch == BIDI_EOB))
    {
      /* The paragraph direction of the entire string, once
	 determined, is in effect for the entire string.  Setting the
	 separator limit to the end of the string prevents
	 bidi_paragraph_init from being called automatically on this
	 string.  */
      if (bidi_it->string.s || STRINGP (bidi_it->string.lstring))
	bidi_it->separator_limit = bidi_it->string.schars;
      else if (bidi_it->bytepos < ZV_BYTE)
	{
	  ptrdiff_t sep_len
	    = bidi_at_paragraph_end (bidi_it->charpos + bidi_it->nchars,
				     bidi_it->bytepos + bidi_it->ch_len);
	  if (bidi_it->nchars <= 0)
	    emacs_abort ();
	  if (sep_len >= 0)
	    {
	      bidi_it->new_paragraph = 1;
	      /* Record the buffer position of the last character of the
		 paragraph separator.  */
	      bidi_it->separator_limit
		= bidi_it->charpos + bidi_it->nchars + sep_len;
	    }
	}
    }

  if (bidi_it->scan_dir == 1 && bidi_cache_idx > bidi_cache_start)
    {
      /* If we are at paragraph's base embedding level and beyond the
	 last cached position, the cache's job is done and we can
	 discard it.  */
      if (bidi_it->resolved_level == bidi_it->level_stack[0].level
	  && bidi_it->charpos > (bidi_cache[bidi_cache_idx - 1].charpos
				 + bidi_cache[bidi_cache_idx - 1].nchars - 1))
	bidi_cache_reset ();
	/* But as long as we are caching during forward scan, we must
	   cache each state, or else the cache integrity will be
	   compromised: it assumes cached states correspond to buffer
	   positions 1:1.  */
      else
	bidi_cache_iterator_state (bidi_it, 1);
    }

  if (STRINGP (bidi_it->string.lstring))
    UNGCPRO;
}

/* This is meant to be called from within the debugger, whenever you
   wish to examine the cache contents.  */
void bidi_dump_cached_states (void) EXTERNALLY_VISIBLE;
void
bidi_dump_cached_states (void)
{
  ptrdiff_t i;
  int ndigits = 1;

  if (bidi_cache_idx == 0)
    {
      fprintf (stderr, "The cache is empty.\n");
      return;
    }
  fprintf (stderr, "Total of  %"pD"d state%s in cache:\n",
	   bidi_cache_idx, bidi_cache_idx == 1 ? "" : "s");

  for (i = bidi_cache[bidi_cache_idx - 1].charpos; i > 0; i /= 10)
    ndigits++;
  fputs ("ch  ", stderr);
  for (i = 0; i < bidi_cache_idx; i++)
    fprintf (stderr, "%*c", ndigits, bidi_cache[i].ch);
  fputs ("\n", stderr);
  fputs ("lvl ", stderr);
  for (i = 0; i < bidi_cache_idx; i++)
    fprintf (stderr, "%*d", ndigits, bidi_cache[i].resolved_level);
  fputs ("\n", stderr);
  fputs ("pos ", stderr);
  for (i = 0; i < bidi_cache_idx; i++)
    fprintf (stderr, "%*"pD"d", ndigits, bidi_cache[i].charpos);
  fputs ("\n", stderr);
}