1// © 2016 and later: Unicode, Inc. and others.
2// License & terms of use: http://www.unicode.org/copyright.html
3/*
4*******************************************************************************
5*
6* Copyright (C) 2001-2011, International Business Machines
7* Corporation and others. All Rights Reserved.
8*
9*******************************************************************************
10* file name: unormimp.h
11* encoding: UTF-8
12* tab size: 8 (not used)
13* indentation:4
14*
15* created on: 2001may25
16* created by: Markus W. Scherer
17*/
18
19#ifndef __UNORMIMP_H__
20#define __UNORMIMP_H__
21
22#include "unicode/utypes.h"
23
24#if !UCONFIG_NO_NORMALIZATION
25
26#include "udataswp.h"
27
28/*
29 * The 2001-2010 implementation of the normalization code loads its data from
30 * unorm.icu, which is generated with the gennorm tool.
31 * The format of that file is described at the end of this file.
32 */
33
34/* norm32 value constants */
35enum {
36 /* quick check flags 0..3 set mean "no" for their forms */
37 _NORM_QC_NFC=0x11, /* no|maybe */
38 _NORM_QC_NFKC=0x22, /* no|maybe */
39 _NORM_QC_NFD=4, /* no */
40 _NORM_QC_NFKD=8, /* no */
41
42 _NORM_QC_ANY_NO=0xf,
43
44 /* quick check flags 4..5 mean "maybe" for their forms; test flags>=_NORM_QC_MAYBE */
45 _NORM_QC_MAYBE=0x10,
46 _NORM_QC_ANY_MAYBE=0x30,
47
48 _NORM_QC_MASK=0x3f,
49
50 _NORM_COMBINES_FWD=0x40,
51 _NORM_COMBINES_BACK=0x80,
52 _NORM_COMBINES_ANY=0xc0,
53
54 _NORM_CC_SHIFT=8, /* UnicodeData.txt combining class in bits 15..8 */
55 _NORM_CC_MASK=0xff00,
56
57 _NORM_EXTRA_SHIFT=16, /* 16 bits for the index to UChars and other extra data */
58 _NORM_EXTRA_INDEX_TOP=0xfc00, /* start of surrogate specials after shift */
59
60 _NORM_EXTRA_SURROGATE_MASK=0x3ff,
61 _NORM_EXTRA_SURROGATE_TOP=0x3f0, /* hangul etc. */
62
63 _NORM_EXTRA_HANGUL=_NORM_EXTRA_SURROGATE_TOP,
64 _NORM_EXTRA_JAMO_L,
65 _NORM_EXTRA_JAMO_V,
66 _NORM_EXTRA_JAMO_T
67};
68
69/* norm32 value constants using >16 bits */
70#define _NORM_MIN_SPECIAL 0xfc000000
71#define _NORM_SURROGATES_TOP 0xfff00000
72#define _NORM_MIN_HANGUL 0xfff00000
73#define _NORM_MIN_JAMO_V 0xfff20000
74#define _NORM_JAMO_V_TOP 0xfff30000
75
76/* value constants for auxTrie */
77enum {
78 _NORM_AUX_COMP_EX_SHIFT=10,
79 _NORM_AUX_UNSAFE_SHIFT=11,
80 _NORM_AUX_NFC_SKIPPABLE_F_SHIFT=12
81};
82
83#define _NORM_AUX_MAX_FNC ((int32_t)1<<_NORM_AUX_COMP_EX_SHIFT)
84
85#define _NORM_AUX_FNC_MASK (uint32_t)(_NORM_AUX_MAX_FNC-1)
86#define _NORM_AUX_COMP_EX_MASK ((uint32_t)1<<_NORM_AUX_COMP_EX_SHIFT)
87#define _NORM_AUX_UNSAFE_MASK ((uint32_t)1<<_NORM_AUX_UNSAFE_SHIFT)
88#define _NORM_AUX_NFC_SKIP_F_MASK ((uint32_t)1<<_NORM_AUX_NFC_SKIPPABLE_F_SHIFT)
89
90/* canonStartSets[0..31] contains indexes for what is in the array */
91enum {
92 _NORM_SET_INDEX_CANON_SETS_LENGTH, /* number of uint16_t in canonical starter sets */
93 _NORM_SET_INDEX_CANON_BMP_TABLE_LENGTH, /* number of uint16_t in the BMP search table (contains pairs) */
94 _NORM_SET_INDEX_CANON_SUPP_TABLE_LENGTH,/* number of uint16_t in the supplementary search table (contains triplets) */
95
96 /* from formatVersion 2.3: */
97 _NORM_SET_INDEX_NX_CJK_COMPAT_OFFSET, /* uint16_t offset from canonStartSets[0] to the
98 exclusion set for CJK compatibility characters */
99 _NORM_SET_INDEX_NX_UNICODE32_OFFSET, /* uint16_t offset from canonStartSets[0] to the
100 exclusion set for Unicode 3.2 characters */
101 _NORM_SET_INDEX_NX_RESERVED_OFFSET, /* uint16_t offset from canonStartSets[0] to the
102 end of the previous exclusion set */
103
104 _NORM_SET_INDEX_TOP=32 /* changing this requires a new formatVersion */
105};
106
107/* more constants for canonical starter sets */
108
109/* 14 bit indexes to canonical USerializedSets */
110#define _NORM_MAX_CANON_SETS 0x4000
111
112/* single-code point BMP sets are encoded directly in the search table except if result=0x4000..0x7fff */
113#define _NORM_CANON_SET_BMP_MASK 0xc000
114#define _NORM_CANON_SET_BMP_IS_INDEX 0x4000
115
116/* indexes[] value names */
117enum {
118 _NORM_INDEX_TRIE_SIZE, /* number of bytes in normalization trie */
119 _NORM_INDEX_UCHAR_COUNT, /* number of UChars in extra data */
120
121 _NORM_INDEX_COMBINE_DATA_COUNT, /* number of uint16_t words for combining data */
122 _NORM_INDEX_COMBINE_FWD_COUNT, /* number of code points that combine forward */
123 _NORM_INDEX_COMBINE_BOTH_COUNT, /* number of code points that combine forward and backward */
124 _NORM_INDEX_COMBINE_BACK_COUNT, /* number of code points that combine backward */
125
126 _NORM_INDEX_MIN_NFC_NO_MAYBE, /* first code point with quick check NFC NO/MAYBE */
127 _NORM_INDEX_MIN_NFKC_NO_MAYBE, /* first code point with quick check NFKC NO/MAYBE */
128 _NORM_INDEX_MIN_NFD_NO_MAYBE, /* first code point with quick check NFD NO/MAYBE */
129 _NORM_INDEX_MIN_NFKD_NO_MAYBE, /* first code point with quick check NFKD NO/MAYBE */
130
131 _NORM_INDEX_FCD_TRIE_SIZE, /* number of bytes in FCD trie */
132
133 _NORM_INDEX_AUX_TRIE_SIZE, /* number of bytes in the auxiliary trie */
134 _NORM_INDEX_CANON_SET_COUNT, /* number of uint16_t in the array of serialized USet */
135
136 _NORM_INDEX_TOP=32 /* changing this requires a new formatVersion */
137};
138
139enum {
140 /* FCD check: everything below this code point is known to have a 0 lead combining class */
141 _NORM_MIN_WITH_LEAD_CC=0x300
142};
143
144enum {
145 /**
146 * Bit 7 of the length byte for a decomposition string in extra data is
147 * a flag indicating whether the decomposition string is
148 * preceded by a 16-bit word with the leading and trailing cc
149 * of the decomposition (like for A-umlaut);
150 * if not, then both cc's are zero (like for compatibility ideographs).
151 */
152 _NORM_DECOMP_FLAG_LENGTH_HAS_CC=0x80,
153 /**
154 * Bits 6..0 of the length byte contain the actual length.
155 */
156 _NORM_DECOMP_LENGTH_MASK=0x7f
157};
158
159/** Constants for options flags for normalization. */
160enum {
161 /** Options bit 0, do not decompose Hangul syllables. */
162 UNORM_NX_HANGUL=1,
163 /** Options bit 1, do not decompose CJK compatibility characters. */
164 UNORM_NX_CJK_COMPAT=2
165};
166
167/**
168 * Description of the format of unorm.icu version 2.3.
169 *
170 * Main change from version 1 to version 2:
171 * Use of new, common UTrie instead of normalization-specific tries.
172 * Change to version 2.1: add third/auxiliary trie with associated data.
173 * Change to version 2.2: add skippable (f) flag data (_NORM_AUX_NFC_SKIP_F_MASK).
174 * Change to version 2.3: add serialized sets for normalization exclusions
175 * stored inside canonStartSets[]
176 *
177 * For more details of how to use the data structures see the code
178 * in unorm.cpp (runtime normalization code) and
179 * in gennorm.c and gennorm/store.c (build-time data generation).
180 *
181 * For the serialized format of UTrie see utrie.c/UTrieHeader.
182 *
183 * - Overall partition
184 *
185 * unorm.dat customarily begins with a UDataInfo structure, see udata.h and .c.
186 * After that there are the following structures:
187 *
188 * int32_t indexes[_NORM_INDEX_TOP]; -- _NORM_INDEX_TOP=32, see enum in this file
189 *
190 * UTrie normTrie; -- size in bytes=indexes[_NORM_INDEX_TRIE_SIZE]
191 *
192 * uint16_t extraData[extraDataTop]; -- extraDataTop=indexes[_NORM_INDEX_UCHAR_COUNT]
193 * extraData[0] contains the number of units for
194 * FC_NFKC_Closure (formatVersion>=2.1)
195 *
196 * uint16_t combiningTable[combiningTableTop]; -- combiningTableTop=indexes[_NORM_INDEX_COMBINE_DATA_COUNT]
197 * combiningTableTop may include one 16-bit padding unit
198 * to make sure that fcdTrie is 32-bit-aligned
199 *
200 * UTrie fcdTrie; -- size in bytes=indexes[_NORM_INDEX_FCD_TRIE_SIZE]
201 *
202 * UTrie auxTrie; -- size in bytes=indexes[_NORM_INDEX_AUX_TRIE_SIZE]
203 *
204 * uint16_t canonStartSets[canonStartSetsTop] -- canonStartSetsTop=indexes[_NORM_INDEX_CANON_SET_COUNT]
205 * serialized USets and binary search tables, see below
206 *
207 *
208 * The indexes array contains lengths and sizes of the following arrays and structures
209 * as well as the following values:
210 * indexes[_NORM_INDEX_COMBINE_FWD_COUNT]=combineFwdTop
211 * -- one more than the highest combining index computed for forward-only-combining characters
212 * indexes[_NORM_INDEX_COMBINE_BOTH_COUNT]=combineBothTop-combineFwdTop
213 * -- number of combining indexes computed for both-ways-combining characters
214 * indexes[_NORM_INDEX_COMBINE_BACK_COUNT]=combineBackTop-combineBothTop
215 * -- number of combining indexes computed for backward-only-combining characters
216 *
217 * indexes[_NORM_INDEX_MIN_NF*_NO_MAYBE] (where *={ C, D, KC, KD })
218 * -- first code point with a quick check NF* value of NO/MAYBE
219 *
220 *
221 * - Tries
222 *
223 * The main structures are two UTrie tables ("compact arrays"),
224 * each with one index array and one data array.
225 * See utrie.h and utrie.c.
226 *
227 *
228 * - Tries in unorm.dat
229 *
230 * The first trie (normTrie above)
231 * provides data for the NF* quick checks and normalization.
232 * The second trie (fcdTrie above) provides data just for FCD checks.
233 *
234 *
235 * - norm32 data words from the first trie
236 *
237 * The norm32Table contains one 32-bit word "norm32" per code point.
238 * It contains the following bit fields:
239 * 31..16 extra data index, _NORM_EXTRA_SHIFT is used to shift this field down
240 * if this index is <_NORM_EXTRA_INDEX_TOP then it is an index into
241 * extraData[] where variable-length normalization data for this
242 * code point is found
243 * if this index is <_NORM_EXTRA_INDEX_TOP+_NORM_EXTRA_SURROGATE_TOP
244 * then this is a norm32 for a leading surrogate, and the index
245 * value is used together with the following trailing surrogate
246 * code unit in the second trie access
247 * if this index is >=_NORM_EXTRA_INDEX_TOP+_NORM_EXTRA_SURROGATE_TOP
248 * then this is a norm32 for a "special" character,
249 * i.e., the character is a Hangul syllable or a Jamo
250 * see _NORM_EXTRA_HANGUL etc.
251 * generally, instead of extracting this index from the norm32 and
252 * comparing it with the above constants,
253 * the normalization code compares the entire norm32 value
254 * with _NORM_MIN_SPECIAL, _NORM_SURROGATES_TOP, _NORM_MIN_HANGUL etc.
255 *
256 * 15..8 combining class (cc) according to UnicodeData.txt
257 *
258 * 7..6 _NORM_COMBINES_ANY flags, used in composition to see if a character
259 * combines with any following or preceding character(s)
260 * at all
261 * 7 _NORM_COMBINES_BACK
262 * 6 _NORM_COMBINES_FWD
263 *
264 * 5..0 quick check flags, set for "no" or "maybe", with separate flags for
265 * each normalization form
266 * the higher bits are "maybe" flags; for NF*D there are no such flags
267 * the lower bits are "no" flags for all forms, in the same order
268 * as the "maybe" flags,
269 * which is (MSB to LSB): NFKD NFD NFKC NFC
270 * 5..4 _NORM_QC_ANY_MAYBE
271 * 3..0 _NORM_QC_ANY_NO
272 * see further related constants
273 *
274 *
275 * - Extra data per code point
276 *
277 * "Extra data" is referenced by the index in norm32.
278 * It is variable-length data. It is only present, and only those parts
279 * of it are, as needed for a given character.
280 * The norm32 extra data index is added to the beginning of extraData[]
281 * to get to a vector of 16-bit words with data at the following offsets:
282 *
283 * [-1] Combining index for composition.
284 * Stored only if norm32&_NORM_COMBINES_ANY .
285 * [0] Lengths of the canonical and compatibility decomposition strings.
286 * Stored only if there are decompositions, i.e.,
287 * if norm32&(_NORM_QC_NFD|_NORM_QC_NFKD)
288 * High byte: length of NFKD, or 0 if none
289 * Low byte: length of NFD, or 0 if none
290 * Each length byte also has another flag:
291 * Bit 7 of a length byte is set if there are non-zero
292 * combining classes (cc's) associated with the respective
293 * decomposition. If this flag is set, then the decomposition
294 * is preceded by a 16-bit word that contains the
295 * leading and trailing cc's.
296 * Bits 6..0 of a length byte are the length of the
297 * decomposition string, not counting the cc word.
298 * [1..n] NFD
299 * [n+1..] NFKD
300 *
301 * Each of the two decompositions consists of up to two parts:
302 * - The 16-bit words with the leading and trailing cc's.
303 * This is only stored if bit 7 of the corresponding length byte
304 * is set. In this case, at least one of the cc's is not zero.
305 * High byte: leading cc==cc of the first code point in the decomposition string
306 * Low byte: trailing cc==cc of the last code point in the decomposition string
307 * - The decomposition string in UTF-16, with length code units.
308 *
309 *
310 * - Combining indexes and combiningTable[]
311 *
312 * Combining indexes are stored at the [-1] offset of the extra data
313 * if the character combines forward or backward with any other characters.
314 * They are used for (re)composition in NF*C.
315 * Values of combining indexes are arranged according to whether a character
316 * combines forward, backward, or both ways:
317 * forward-only < both ways < backward-only
318 *
319 * The index values for forward-only and both-ways combining characters
320 * are indexes into the combiningTable[].
321 * The index values for backward-only combining characters are simply
322 * incremented from the preceding index values to be unique.
323 *
324 * In the combiningTable[], a variable-length list
325 * of variable-length (back-index, code point) pair entries is stored
326 * for each forward-combining character.
327 *
328 * These back-indexes are the combining indexes of both-ways or backward-only
329 * combining characters that the forward-combining character combines with.
330 *
331 * Each list is sorted in ascending order of back-indexes.
332 * Each list is terminated with the last back-index having bit 15 set.
333 *
334 * Each pair (back-index, code point) takes up either 2 or 3
335 * 16-bit words.
336 * The first word of a list entry is the back-index, with its bit 15 set if
337 * this is the last pair in the list.
338 *
339 * The second word contains flags in bits 15..13 that determine
340 * if there is a third word and how the combined character is encoded:
341 * 15 set if there is a third word in this list entry
342 * 14 set if the result is a supplementary character
343 * 13 set if the result itself combines forward
344 *
345 * According to these bits 15..14 of the second word,
346 * the result character is encoded as follows:
347 * 00 or 01 The result is <=0x1fff and stored in bits 12..0 of
348 * the second word.
349 * 10 The result is 0x2000..0xffff and stored in the third word.
350 * Bits 12..0 of the second word are not used.
351 * 11 The result is a supplementary character.
352 * Bits 9..0 of the leading surrogate are in bits 9..0 of
353 * the second word.
354 * Add 0xd800 to these bits to get the complete surrogate.
355 * Bits 12..10 of the second word are not used.
356 * The trailing surrogate is stored in the third word.
357 *
358 *
359 * - FCD trie
360 *
361 * The FCD trie is very simple.
362 * It is a folded trie with 16-bit data words.
363 * In each word, the high byte contains the leading cc of the character,
364 * and the low byte contains the trailing cc of the character.
365 * These cc's are the cc's of the first and last code points in the
366 * canonical decomposition of the character.
367 *
368 * Since all 16 bits are used for cc's, lead surrogates must be tested
369 * by checking the code unit instead of the trie data.
370 * This is done only if the 16-bit data word is not zero.
371 * If the code unit is a leading surrogate and the data word is not zero,
372 * then instead of cc's it contains the offset for the second trie lookup.
373 *
374 *
375 * - Auxiliary trie and data
376 *
377 * The auxiliary 16-bit trie contains data for additional properties.
378 * Bits
379 * 15..13 reserved
380 * 12 not NFC_Skippable (f) (formatVersion>=2.2)
381 * 11 flag: not a safe starter for canonical closure
382 * 10 composition exclusion
383 * 9.. 0 index into extraData[] to FC_NFKC_Closure string
384 * (not for lead surrogate),
385 * or lead surrogate offset (for lead surrogate, if 9..0 not zero)
386 *
387 * - FC_NFKC_Closure strings in extraData[]
388 *
389 * Strings are either stored as a single code unit or as the length
390 * followed by that many units.
391 * const char16_t *s=extraData+(index from auxTrie data bits 9..0);
392 * int32_t length;
393 * if(*s<0xff00) {
394 * // s points to the single-unit string
395 * length=1;
396 * } else {
397 * length=*s&0xff;
398 * ++s;
399 * }
400 *
401 * Conditions for "NF* Skippable" from Mark Davis' com.ibm.text.UCD.NFSkippable:
402 * (used in NormalizerTransliterator)
403 *
404 * A skippable character is
405 * a) unassigned, or ALL of the following:
406 * b) of combining class 0.
407 * c) not decomposed by this normalization form.
408 * AND if NFC or NFKC,
409 * d) can never compose with a previous character.
410 * e) can never compose with a following character.
411 * f) can never change if another character is added.
412 * Example: a-breve might satisfy all but f, but if you
413 * add an ogonek it changes to a-ogonek + breve
414 *
415 * a)..e) must be tested from norm32.
416 * Since f) is more complicated, the (not-)NFC_Skippable flag (f) is built
417 * into the auxiliary trie.
418 * The same bit is used for NFC and NFKC; (c) differs for them.
419 * As usual, we build the "not skippable" flags so that unassigned
420 * code points get a 0 bit.
421 * This bit is only valid after (a)..(e) test false; test NFD_NO before (f) as well.
422 * Test Hangul LV syllables entirely in code.
423 *
424 *
425 * - structure inside canonStartSets[]
426 *
427 * This array maps from code points c to sets of code points (USerializedSet).
428 * The result sets are the code points whose canonical decompositions start
429 * with c.
430 *
431 * canonStartSets[] contains the following sub-arrays:
432 *
433 * indexes[_NORM_SET_INDEX_TOP]
434 * - contains lengths of sub-arrays etc.
435 *
436 * startSets[indexes[_NORM_SET_INDEX_CANON_SETS_LENGTH]-_NORM_SET_INDEX_TOP]
437 * - contains serialized sets (USerializedSet) of canonical starters for
438 * enumerating canonically equivalent strings
439 * indexes[_NORM_SET_INDEX_CANON_SETS_LENGTH] includes _NORM_SET_INDEX_TOP
440 * for details about the structure see uset.c
441 *
442 * bmpTable[indexes[_NORM_SET_INDEX_CANON_BMP_TABLE_LENGTH]]
443 * - a sorted search table for BMP code points whose results are
444 * either indexes to USerializedSets or single code points for
445 * single-code point sets;
446 * each entry is a pair of { code point, result } with result=(binary) yy xxxxxx xxxxxxxx
447 * if yy==01 then there is a USerializedSet at canonStartSets+x
448 * else build a USerializedSet with result as the single code point
449 *
450 * suppTable[indexes[_NORM_SET_INDEX_CANON_SUPP_TABLE_LENGTH]]
451 * - a sorted search table for supplementary code points whose results are
452 * either indexes to USerializedSets or single code points for
453 * single-code point sets;
454 * each entry is a triplet of { high16(cp), low16(cp), result }
455 * each code point's high-word may contain extra data in bits 15..5:
456 * if the high word has bit 15 set, then build a set with a single code point
457 * which is (((high16(cp)&0x1f00)<<8)|result;
458 * else there is a USerializedSet at canonStartSets+result
459 *
460 * FormatVersion 2.3 adds 2 serialized sets for normalization exclusions.
461 * They are stored in the data file so that the runtime normalization code need
462 * not depend on other properties and their data and implementation files.
463 * The _NORM_SET_INDEX_NX_..._OFFSET offsets in the canonStartSets index table
464 * give the location for each set.
465 * There is no set stored for UNORM_NX_HANGUL because it's trivial to create
466 * without using properties.
467 *
468 * Set contents:
469 *
470 * _NORM_SET_INDEX_NX_CJK_COMPAT_OFFSET (for UNORM_NX_CJK_COMPAT)
471 * [[:Ideographic:]&[:NFD_QC=No:]]
472 * =[CJK Ideographs]&[has canonical decomposition]
473 *
474 * _NORM_SET_INDEX_NX_UNICODE32_OFFSET (for UNORM_UNICODE_3_2)
475 * [:^Age=3.2:]
476 * =set with all code points that were not designated by the specified Unicode version
477 *
478 * _NORM_SET_INDEX_NX_RESERVED_OFFSET
479 * This is an offset that points to where the next, future set would start.
480 * Currently it indicates where the previous set ends, and thus its length.
481 * The name for this enum constant may in the future be applied to different
482 * index slots. In order to get the limit of a set, use its index slot and
483 * the immediately following one regardless of that one's enum name.
484 */
485
486#endif /* #if !UCONFIG_NO_NORMALIZATION */
487
488#endif
489