2
*******************************************************************************
4
* Copyright (C) 2001-2010, International Business Machines
5
* Corporation and others. All Rights Reserved.
7
*******************************************************************************
8
* file name: unormimp.h
10
* tab size: 8 (not used)
13
* created on: 2001may25
14
* created by: Markus W. Scherer
17
#ifndef __UNORMIMP_H__
18
#define __UNORMIMP_H__
20
#include "unicode/utypes.h"
22
#if !UCONFIG_NO_NORMALIZATION
27
* The 2001-2010 implementation of the normalization code loads its data from
28
* unorm.icu, which is generated with the gennorm tool.
29
* The format of that file is described at the end of this file.
32
/* norm32 value constants */
34
/* quick check flags 0..3 set mean "no" for their forms */
35
_NORM_QC_NFC=0x11, /* no|maybe */
36
_NORM_QC_NFKC=0x22, /* no|maybe */
37
_NORM_QC_NFD=4, /* no */
38
_NORM_QC_NFKD=8, /* no */
42
/* quick check flags 4..5 mean "maybe" for their forms; test flags>=_NORM_QC_MAYBE */
44
_NORM_QC_ANY_MAYBE=0x30,
48
_NORM_COMBINES_FWD=0x40,
49
_NORM_COMBINES_BACK=0x80,
50
_NORM_COMBINES_ANY=0xc0,
52
_NORM_CC_SHIFT=8, /* UnicodeData.txt combining class in bits 15..8 */
55
_NORM_EXTRA_SHIFT=16, /* 16 bits for the index to UChars and other extra data */
56
_NORM_EXTRA_INDEX_TOP=0xfc00, /* start of surrogate specials after shift */
58
_NORM_EXTRA_SURROGATE_MASK=0x3ff,
59
_NORM_EXTRA_SURROGATE_TOP=0x3f0, /* hangul etc. */
61
_NORM_EXTRA_HANGUL=_NORM_EXTRA_SURROGATE_TOP,
67
/* norm32 value constants using >16 bits */
68
#define _NORM_MIN_SPECIAL 0xfc000000
69
#define _NORM_SURROGATES_TOP 0xfff00000
70
#define _NORM_MIN_HANGUL 0xfff00000
71
#define _NORM_MIN_JAMO_V 0xfff20000
72
#define _NORM_JAMO_V_TOP 0xfff30000
74
/* value constants for auxTrie */
76
_NORM_AUX_COMP_EX_SHIFT=10,
77
_NORM_AUX_UNSAFE_SHIFT=11,
78
_NORM_AUX_NFC_SKIPPABLE_F_SHIFT=12
81
#define _NORM_AUX_MAX_FNC ((int32_t)1<<_NORM_AUX_COMP_EX_SHIFT)
83
#define _NORM_AUX_FNC_MASK (uint32_t)(_NORM_AUX_MAX_FNC-1)
84
#define _NORM_AUX_COMP_EX_MASK ((uint32_t)1<<_NORM_AUX_COMP_EX_SHIFT)
85
#define _NORM_AUX_UNSAFE_MASK ((uint32_t)1<<_NORM_AUX_UNSAFE_SHIFT)
86
#define _NORM_AUX_NFC_SKIP_F_MASK ((uint32_t)1<<_NORM_AUX_NFC_SKIPPABLE_F_SHIFT)
88
/* canonStartSets[0..31] contains indexes for what is in the array */
90
_NORM_SET_INDEX_CANON_SETS_LENGTH, /* number of uint16_t in canonical starter sets */
91
_NORM_SET_INDEX_CANON_BMP_TABLE_LENGTH, /* number of uint16_t in the BMP search table (contains pairs) */
92
_NORM_SET_INDEX_CANON_SUPP_TABLE_LENGTH,/* number of uint16_t in the supplementary search table (contains triplets) */
94
/* from formatVersion 2.3: */
95
_NORM_SET_INDEX_NX_CJK_COMPAT_OFFSET, /* uint16_t offset from canonStartSets[0] to the
96
exclusion set for CJK compatibility characters */
97
_NORM_SET_INDEX_NX_UNICODE32_OFFSET, /* uint16_t offset from canonStartSets[0] to the
98
exclusion set for Unicode 3.2 characters */
99
_NORM_SET_INDEX_NX_RESERVED_OFFSET, /* uint16_t offset from canonStartSets[0] to the
100
end of the previous exclusion set */
102
_NORM_SET_INDEX_TOP=32 /* changing this requires a new formatVersion */
105
/* more constants for canonical starter sets */
107
/* 14 bit indexes to canonical USerializedSets */
108
#define _NORM_MAX_CANON_SETS 0x4000
110
/* single-code point BMP sets are encoded directly in the search table except if result=0x4000..0x7fff */
111
#define _NORM_CANON_SET_BMP_MASK 0xc000
112
#define _NORM_CANON_SET_BMP_IS_INDEX 0x4000
114
/* indexes[] value names */
116
_NORM_INDEX_TRIE_SIZE, /* number of bytes in normalization trie */
117
_NORM_INDEX_UCHAR_COUNT, /* number of UChars in extra data */
119
_NORM_INDEX_COMBINE_DATA_COUNT, /* number of uint16_t words for combining data */
120
_NORM_INDEX_COMBINE_FWD_COUNT, /* number of code points that combine forward */
121
_NORM_INDEX_COMBINE_BOTH_COUNT, /* number of code points that combine forward and backward */
122
_NORM_INDEX_COMBINE_BACK_COUNT, /* number of code points that combine backward */
124
_NORM_INDEX_MIN_NFC_NO_MAYBE, /* first code point with quick check NFC NO/MAYBE */
125
_NORM_INDEX_MIN_NFKC_NO_MAYBE, /* first code point with quick check NFKC NO/MAYBE */
126
_NORM_INDEX_MIN_NFD_NO_MAYBE, /* first code point with quick check NFD NO/MAYBE */
127
_NORM_INDEX_MIN_NFKD_NO_MAYBE, /* first code point with quick check NFKD NO/MAYBE */
129
_NORM_INDEX_FCD_TRIE_SIZE, /* number of bytes in FCD trie */
131
_NORM_INDEX_AUX_TRIE_SIZE, /* number of bytes in the auxiliary trie */
132
_NORM_INDEX_CANON_SET_COUNT, /* number of uint16_t in the array of serialized USet */
134
_NORM_INDEX_TOP=32 /* changing this requires a new formatVersion */
138
/* FCD check: everything below this code point is known to have a 0 lead combining class */
139
_NORM_MIN_WITH_LEAD_CC=0x300
144
* Bit 7 of the length byte for a decomposition string in extra data is
145
* a flag indicating whether the decomposition string is
146
* preceded by a 16-bit word with the leading and trailing cc
147
* of the decomposition (like for A-umlaut);
148
* if not, then both cc's are zero (like for compatibility ideographs).
150
_NORM_DECOMP_FLAG_LENGTH_HAS_CC=0x80,
152
* Bits 6..0 of the length byte contain the actual length.
154
_NORM_DECOMP_LENGTH_MASK=0x7f
157
/* Constants for options flags for normalization. @draft ICU 2.6 */
159
/** Options bit 0, do not decompose Hangul syllables. @draft ICU 2.6 */
161
/** Options bit 1, do not decompose CJK compatibility characters. @draft ICU 2.6 */
162
UNORM_NX_CJK_COMPAT=2
166
* Swap unorm.icu. See udataswp.h.
169
U_CAPI int32_t U_EXPORT2
170
unorm_swap(const UDataSwapper *ds,
171
const void *inData, int32_t length, void *outData,
172
UErrorCode *pErrorCode);
175
* Description of the format of unorm.icu version 2.3.
177
* Main change from version 1 to version 2:
178
* Use of new, common UTrie instead of normalization-specific tries.
179
* Change to version 2.1: add third/auxiliary trie with associated data.
180
* Change to version 2.2: add skippable (f) flag data (_NORM_AUX_NFC_SKIP_F_MASK).
181
* Change to version 2.3: add serialized sets for normalization exclusions
182
* stored inside canonStartSets[]
184
* For more details of how to use the data structures see the code
185
* in unorm.cpp (runtime normalization code) and
186
* in gennorm.c and gennorm/store.c (build-time data generation).
188
* For the serialized format of UTrie see utrie.c/UTrieHeader.
190
* - Overall partition
192
* unorm.dat customarily begins with a UDataInfo structure, see udata.h and .c.
193
* After that there are the following structures:
195
* int32_t indexes[_NORM_INDEX_TOP]; -- _NORM_INDEX_TOP=32, see enum in this file
197
* UTrie normTrie; -- size in bytes=indexes[_NORM_INDEX_TRIE_SIZE]
199
* uint16_t extraData[extraDataTop]; -- extraDataTop=indexes[_NORM_INDEX_UCHAR_COUNT]
200
* extraData[0] contains the number of units for
201
* FC_NFKC_Closure (formatVersion>=2.1)
203
* uint16_t combiningTable[combiningTableTop]; -- combiningTableTop=indexes[_NORM_INDEX_COMBINE_DATA_COUNT]
204
* combiningTableTop may include one 16-bit padding unit
205
* to make sure that fcdTrie is 32-bit-aligned
207
* UTrie fcdTrie; -- size in bytes=indexes[_NORM_INDEX_FCD_TRIE_SIZE]
209
* UTrie auxTrie; -- size in bytes=indexes[_NORM_INDEX_AUX_TRIE_SIZE]
211
* uint16_t canonStartSets[canonStartSetsTop] -- canonStartSetsTop=indexes[_NORM_INDEX_CANON_SET_COUNT]
212
* serialized USets and binary search tables, see below
215
* The indexes array contains lengths and sizes of the following arrays and structures
216
* as well as the following values:
217
* indexes[_NORM_INDEX_COMBINE_FWD_COUNT]=combineFwdTop
218
* -- one more than the highest combining index computed for forward-only-combining characters
219
* indexes[_NORM_INDEX_COMBINE_BOTH_COUNT]=combineBothTop-combineFwdTop
220
* -- number of combining indexes computed for both-ways-combining characters
221
* indexes[_NORM_INDEX_COMBINE_BACK_COUNT]=combineBackTop-combineBothTop
222
* -- number of combining indexes computed for backward-only-combining characters
224
* indexes[_NORM_INDEX_MIN_NF*_NO_MAYBE] (where *={ C, D, KC, KD })
225
* -- first code point with a quick check NF* value of NO/MAYBE
230
* The main structures are two UTrie tables ("compact arrays"),
231
* each with one index array and one data array.
232
* See utrie.h and utrie.c.
235
* - Tries in unorm.dat
237
* The first trie (normTrie above)
238
* provides data for the NF* quick checks and normalization.
239
* The second trie (fcdTrie above) provides data just for FCD checks.
242
* - norm32 data words from the first trie
244
* The norm32Table contains one 32-bit word "norm32" per code point.
245
* It contains the following bit fields:
246
* 31..16 extra data index, _NORM_EXTRA_SHIFT is used to shift this field down
247
* if this index is <_NORM_EXTRA_INDEX_TOP then it is an index into
248
* extraData[] where variable-length normalization data for this
249
* code point is found
250
* if this index is <_NORM_EXTRA_INDEX_TOP+_NORM_EXTRA_SURROGATE_TOP
251
* then this is a norm32 for a leading surrogate, and the index
252
* value is used together with the following trailing surrogate
253
* code unit in the second trie access
254
* if this index is >=_NORM_EXTRA_INDEX_TOP+_NORM_EXTRA_SURROGATE_TOP
255
* then this is a norm32 for a "special" character,
256
* i.e., the character is a Hangul syllable or a Jamo
257
* see _NORM_EXTRA_HANGUL etc.
258
* generally, instead of extracting this index from the norm32 and
259
* comparing it with the above constants,
260
* the normalization code compares the entire norm32 value
261
* with _NORM_MIN_SPECIAL, _NORM_SURROGATES_TOP, _NORM_MIN_HANGUL etc.
263
* 15..8 combining class (cc) according to UnicodeData.txt
265
* 7..6 _NORM_COMBINES_ANY flags, used in composition to see if a character
266
* combines with any following or preceding character(s)
268
* 7 _NORM_COMBINES_BACK
269
* 6 _NORM_COMBINES_FWD
271
* 5..0 quick check flags, set for "no" or "maybe", with separate flags for
272
* each normalization form
273
* the higher bits are "maybe" flags; for NF*D there are no such flags
274
* the lower bits are "no" flags for all forms, in the same order
275
* as the "maybe" flags,
276
* which is (MSB to LSB): NFKD NFD NFKC NFC
277
* 5..4 _NORM_QC_ANY_MAYBE
278
* 3..0 _NORM_QC_ANY_NO
279
* see further related constants
282
* - Extra data per code point
284
* "Extra data" is referenced by the index in norm32.
285
* It is variable-length data. It is only present, and only those parts
286
* of it are, as needed for a given character.
287
* The norm32 extra data index is added to the beginning of extraData[]
288
* to get to a vector of 16-bit words with data at the following offsets:
290
* [-1] Combining index for composition.
291
* Stored only if norm32&_NORM_COMBINES_ANY .
292
* [0] Lengths of the canonical and compatibility decomposition strings.
293
* Stored only if there are decompositions, i.e.,
294
* if norm32&(_NORM_QC_NFD|_NORM_QC_NFKD)
295
* High byte: length of NFKD, or 0 if none
296
* Low byte: length of NFD, or 0 if none
297
* Each length byte also has another flag:
298
* Bit 7 of a length byte is set if there are non-zero
299
* combining classes (cc's) associated with the respective
300
* decomposition. If this flag is set, then the decomposition
301
* is preceded by a 16-bit word that contains the
302
* leading and trailing cc's.
303
* Bits 6..0 of a length byte are the length of the
304
* decomposition string, not counting the cc word.
308
* Each of the two decompositions consists of up to two parts:
309
* - The 16-bit words with the leading and trailing cc's.
310
* This is only stored if bit 7 of the corresponding length byte
311
* is set. In this case, at least one of the cc's is not zero.
312
* High byte: leading cc==cc of the first code point in the decomposition string
313
* Low byte: trailing cc==cc of the last code point in the decomposition string
314
* - The decomposition string in UTF-16, with length code units.
317
* - Combining indexes and combiningTable[]
319
* Combining indexes are stored at the [-1] offset of the extra data
320
* if the character combines forward or backward with any other characters.
321
* They are used for (re)composition in NF*C.
322
* Values of combining indexes are arranged according to whether a character
323
* combines forward, backward, or both ways:
324
* forward-only < both ways < backward-only
326
* The index values for forward-only and both-ways combining characters
327
* are indexes into the combiningTable[].
328
* The index values for backward-only combining characters are simply
329
* incremented from the preceding index values to be unique.
331
* In the combiningTable[], a variable-length list
332
* of variable-length (back-index, code point) pair entries is stored
333
* for each forward-combining character.
335
* These back-indexes are the combining indexes of both-ways or backward-only
336
* combining characters that the forward-combining character combines with.
338
* Each list is sorted in ascending order of back-indexes.
339
* Each list is terminated with the last back-index having bit 15 set.
341
* Each pair (back-index, code point) takes up either 2 or 3
343
* The first word of a list entry is the back-index, with its bit 15 set if
344
* this is the last pair in the list.
346
* The second word contains flags in bits 15..13 that determine
347
* if there is a third word and how the combined character is encoded:
348
* 15 set if there is a third word in this list entry
349
* 14 set if the result is a supplementary character
350
* 13 set if the result itself combines forward
352
* According to these bits 15..14 of the second word,
353
* the result character is encoded as follows:
354
* 00 or 01 The result is <=0x1fff and stored in bits 12..0 of
356
* 10 The result is 0x2000..0xffff and stored in the third word.
357
* Bits 12..0 of the second word are not used.
358
* 11 The result is a supplementary character.
359
* Bits 9..0 of the leading surrogate are in bits 9..0 of
361
* Add 0xd800 to these bits to get the complete surrogate.
362
* Bits 12..10 of the second word are not used.
363
* The trailing surrogate is stored in the third word.
368
* The FCD trie is very simple.
369
* It is a folded trie with 16-bit data words.
370
* In each word, the high byte contains the leading cc of the character,
371
* and the low byte contains the trailing cc of the character.
372
* These cc's are the cc's of the first and last code points in the
373
* canonical decomposition of the character.
375
* Since all 16 bits are used for cc's, lead surrogates must be tested
376
* by checking the code unit instead of the trie data.
377
* This is done only if the 16-bit data word is not zero.
378
* If the code unit is a leading surrogate and the data word is not zero,
379
* then instead of cc's it contains the offset for the second trie lookup.
382
* - Auxiliary trie and data
384
* The auxiliary 16-bit trie contains data for additional properties.
387
* 12 not NFC_Skippable (f) (formatVersion>=2.2)
388
* 11 flag: not a safe starter for canonical closure
389
* 10 composition exclusion
390
* 9.. 0 index into extraData[] to FC_NFKC_Closure string
391
* (not for lead surrogate),
392
* or lead surrogate offset (for lead surrogate, if 9..0 not zero)
394
* - FC_NFKC_Closure strings in extraData[]
396
* Strings are either stored as a single code unit or as the length
397
* followed by that many units.
398
* const UChar *s=extraData+(index from auxTrie data bits 9..0);
401
* // s points to the single-unit string
408
* Conditions for "NF* Skippable" from Mark Davis' com.ibm.text.UCD.NFSkippable:
409
* (used in NormalizerTransliterator)
411
* A skippable character is
412
* a) unassigned, or ALL of the following:
413
* b) of combining class 0.
414
* c) not decomposed by this normalization form.
415
* AND if NFC or NFKC,
416
* d) can never compose with a previous character.
417
* e) can never compose with a following character.
418
* f) can never change if another character is added.
419
* Example: a-breve might satisfy all but f, but if you
420
* add an ogonek it changes to a-ogonek + breve
422
* a)..e) must be tested from norm32.
423
* Since f) is more complicated, the (not-)NFC_Skippable flag (f) is built
424
* into the auxiliary trie.
425
* The same bit is used for NFC and NFKC; (c) differs for them.
426
* As usual, we build the "not skippable" flags so that unassigned
427
* code points get a 0 bit.
428
* This bit is only valid after (a)..(e) test FALSE; test NFD_NO before (f) as well.
429
* Test Hangul LV syllables entirely in code.
432
* - structure inside canonStartSets[]
434
* This array maps from code points c to sets of code points (USerializedSet).
435
* The result sets are the code points whose canonical decompositions start
438
* canonStartSets[] contains the following sub-arrays:
440
* indexes[_NORM_SET_INDEX_TOP]
441
* - contains lengths of sub-arrays etc.
443
* startSets[indexes[_NORM_SET_INDEX_CANON_SETS_LENGTH]-_NORM_SET_INDEX_TOP]
444
* - contains serialized sets (USerializedSet) of canonical starters for
445
* enumerating canonically equivalent strings
446
* indexes[_NORM_SET_INDEX_CANON_SETS_LENGTH] includes _NORM_SET_INDEX_TOP
447
* for details about the structure see uset.c
449
* bmpTable[indexes[_NORM_SET_INDEX_CANON_BMP_TABLE_LENGTH]]
450
* - a sorted search table for BMP code points whose results are
451
* either indexes to USerializedSets or single code points for
452
* single-code point sets;
453
* each entry is a pair of { code point, result } with result=(binary) yy xxxxxx xxxxxxxx
454
* if yy==01 then there is a USerializedSet at canonStartSets+x
455
* else build a USerializedSet with result as the single code point
457
* suppTable[indexes[_NORM_SET_INDEX_CANON_SUPP_TABLE_LENGTH]]
458
* - a sorted search table for supplementary code points whose results are
459
* either indexes to USerializedSets or single code points for
460
* single-code point sets;
461
* each entry is a triplet of { high16(cp), low16(cp), result }
462
* each code point's high-word may contain extra data in bits 15..5:
463
* if the high word has bit 15 set, then build a set with a single code point
464
* which is (((high16(cp)&0x1f00)<<8)|result;
465
* else there is a USerializedSet at canonStartSets+result
467
* FormatVersion 2.3 adds 2 serialized sets for normalization exclusions.
468
* They are stored in the data file so that the runtime normalization code need
469
* not depend on other properties and their data and implementation files.
470
* The _NORM_SET_INDEX_NX_..._OFFSET offsets in the canonStartSets index table
471
* give the location for each set.
472
* There is no set stored for UNORM_NX_HANGUL because it's trivial to create
473
* without using properties.
477
* _NORM_SET_INDEX_NX_CJK_COMPAT_OFFSET (for UNORM_NX_CJK_COMPAT)
478
* [[:Ideographic:]&[:NFD_QC=No:]]
479
* =[CJK Ideographs]&[has canonical decomposition]
481
* _NORM_SET_INDEX_NX_UNICODE32_OFFSET (for UNORM_UNICODE_3_2)
483
* =set with all code points that were not designated by the specified Unicode version
485
* _NORM_SET_INDEX_NX_RESERVED_OFFSET
486
* This is an offset that points to where the next, future set would start.
487
* Currently it indicates where the previous set ends, and thus its length.
488
* The name for this enum constant may in the future be applied to different
489
* index slots. In order to get the limit of a set, use its index slot and
490
* the immediately following one regardless of that one's enum name.
493
#endif /* #if !UCONFIG_NO_NORMALIZATION */