michael@0: /* michael@0: ****************************************************************************** michael@0: * michael@0: * Copyright (C) 2003-2013, International Business Machines michael@0: * Corporation and others. All Rights Reserved. michael@0: * michael@0: ****************************************************************************** michael@0: * file name: ucnv_ext.h michael@0: * encoding: US-ASCII michael@0: * tab size: 8 (not used) michael@0: * indentation:4 michael@0: * michael@0: * created on: 2003jun13 michael@0: * created by: Markus W. Scherer michael@0: * michael@0: * Conversion extensions michael@0: */ michael@0: michael@0: #ifndef __UCNV_EXT_H__ michael@0: #define __UCNV_EXT_H__ michael@0: michael@0: #include "unicode/utypes.h" michael@0: michael@0: #if !UCONFIG_NO_CONVERSION michael@0: michael@0: #include "unicode/ucnv.h" michael@0: #include "ucnv_cnv.h" michael@0: michael@0: /* michael@0: * See icuhtml/design/conversion/conversion_extensions.html michael@0: * michael@0: * Conversion extensions serve three purposes: michael@0: * 1. They support m:n mappings. michael@0: * 2. They support extension-only conversion files that are used together michael@0: * with the regular conversion data in base files. michael@0: * 3. They support mappings with more complicated meta data, michael@0: * for example "good one-way" mappings (|4). michael@0: * michael@0: * A base file may contain an extension table (explicitly requested or michael@0: * implicitly generated for m:n mappings), but its extension table is not michael@0: * used when an extension-only file is used. michael@0: * michael@0: * It is an error if a base file contains any regular (not extension) mapping michael@0: * from the same sequence as a mapping in the extension file michael@0: * because the base mapping would hide the extension mapping. michael@0: * michael@0: * michael@0: * Data for conversion extensions: michael@0: * michael@0: * One set of data structures per conversion direction (to/from Unicode). michael@0: * The data structures are sorted by input units to allow for binary search. michael@0: * Input sequences of more than one unit are handled like contraction tables michael@0: * in collation: michael@0: * The lookup value of a unit points to another table that is to be searched michael@0: * for the next unit, recursively. michael@0: * michael@0: * For conversion from Unicode, the initial code point is looked up in michael@0: * a 3-stage trie for speed, michael@0: * with an additional table of unique results to save space. michael@0: * michael@0: * Long output strings are stored in separate arrays, with length and index michael@0: * in the lookup tables. michael@0: * Output results also include a flag distinguishing roundtrip from michael@0: * (reverse) fallback mappings. michael@0: * michael@0: * Input Unicode strings must not begin or end with unpaired surrogates michael@0: * to avoid problems with matches on parts of surrogate pairs. michael@0: * michael@0: * Mappings from multiple characters (code points or codepage state michael@0: * table sequences) must be searched preferring the longest match. michael@0: * For this to work and be efficient, the variable-width table must contain michael@0: * all mappings that contain prefixes of the multiple characters. michael@0: * If an extension table is built on top of a base table in another file michael@0: * and a base table entry is a prefix of a multi-character mapping, then michael@0: * this is an error. michael@0: * michael@0: * michael@0: * Implementation note: michael@0: * michael@0: * Currently, the parser and several checks in the code limit the number michael@0: * of UChars or bytes in a mapping to michael@0: * UCNV_EXT_MAX_UCHARS and UCNV_EXT_MAX_BYTES, respectively, michael@0: * which are output value limits in the data structure. michael@0: * michael@0: * For input, this is not strictly necessary - it is a hard limit only for the michael@0: * buffers in UConverter that are used to store partial matches. michael@0: * michael@0: * Input sequences could otherwise be arbitrarily long if partial matches michael@0: * need not be stored (i.e., if a sequence does not span several buffers with too michael@0: * many units before the last buffer), although then results would differ michael@0: * depending on whether partial matches exceed the limits or not, michael@0: * which depends on the pattern of buffer sizes. michael@0: * michael@0: * michael@0: * Data structure: michael@0: * michael@0: * int32_t indexes[>=32]; michael@0: * michael@0: * Array of indexes and lengths etc. The length of the array is at least 32. michael@0: * The actual length is stored in indexes[0] to be forward compatible. michael@0: * michael@0: * Each index to another array is the number of bytes from indexes[]. michael@0: * Each length of an array is the number of array base units in that array. michael@0: * michael@0: * Some of the structures may not be present, in which case their indexes michael@0: * and lengths are 0. michael@0: * michael@0: * Usage of indexes[i]: michael@0: * [0] length of indexes[] michael@0: * michael@0: * // to Unicode table michael@0: * [1] index of toUTable[] (array of uint32_t) michael@0: * [2] length of toUTable[] michael@0: * [3] index of toUUChars[] (array of UChar) michael@0: * [4] length of toUUChars[] michael@0: * michael@0: * // from Unicode table, not for the initial code point michael@0: * [5] index of fromUTableUChars[] (array of UChar) michael@0: * [6] index of fromUTableValues[] (array of uint32_t) michael@0: * [7] length of fromUTableUChars[] and fromUTableValues[] michael@0: * [8] index of fromUBytes[] (array of char) michael@0: * [9] length of fromUBytes[] michael@0: * michael@0: * // from Unicode trie for initial-code point lookup michael@0: * [10] index of fromUStage12[] (combined array of uint16_t for stages 1 & 2) michael@0: * [11] length of stage 1 portion of fromUStage12[] michael@0: * [12] length of fromUStage12[] michael@0: * [13] index of fromUStage3[] (array of uint16_t indexes into fromUStage3b[]) michael@0: * [14] length of fromUStage3[] michael@0: * [15] index of fromUStage3b[] (array of uint32_t like fromUTableValues[]) michael@0: * [16] length of fromUStage3b[] michael@0: * michael@0: * [17] Bit field containing numbers of bytes: michael@0: * 31..24 reserved, 0 michael@0: * 23..16 maximum input bytes michael@0: * 15.. 8 maximum output bytes michael@0: * 7.. 0 maximum bytes per UChar michael@0: * michael@0: * [18] Bit field containing numbers of UChars: michael@0: * 31..24 reserved, 0 michael@0: * 23..16 maximum input UChars michael@0: * 15.. 8 maximum output UChars michael@0: * 7.. 0 maximum UChars per byte michael@0: * michael@0: * [19] Bit field containing flags: michael@0: * (extension table unicodeMask) michael@0: * 1 UCNV_HAS_SURROGATES flag for the extension table michael@0: * 0 UCNV_HAS_SUPPLEMENTARY flag for the extension table michael@0: * michael@0: * [20]..[30] reserved, 0 michael@0: * [31] number of bytes for the entire extension structure michael@0: * [>31] reserved; there are indexes[0] indexes michael@0: * michael@0: * michael@0: * uint32_t toUTable[]; michael@0: * michael@0: * Array of byte/value pairs for lookups for toUnicode conversion. michael@0: * The array is partitioned into sections like collation contraction tables. michael@0: * Each section contains one word with the number of following words and michael@0: * a default value for when the lookup in this section yields no match. michael@0: * michael@0: * A section is sorted in ascending order of input bytes, michael@0: * allowing for fast linear or binary searches. michael@0: * The builder may store entries for a contiguous range of byte values michael@0: * (compare difference between the first and last one with count), michael@0: * which then allows for direct array access. michael@0: * The builder should always do this for the initial table section. michael@0: * michael@0: * Entries may have 0 values, see below. michael@0: * No two entries in a section have the same byte values. michael@0: * michael@0: * Each uint32_t contains an input byte value in bits 31..24 and the michael@0: * corresponding lookup value in bits 23..0. michael@0: * Interpret the value as follows: michael@0: * if(value==0) { michael@0: * no match, see below michael@0: * } else if(value<0x1f0000) { michael@0: * partial match - use value as index to the next toUTable section michael@0: * and match the next unit; (value indexes toUTable[value]) michael@0: * } else { michael@0: * if(bit 23 set) { michael@0: * roundtrip; michael@0: * } else { michael@0: * fallback; michael@0: * } michael@0: * unset value bit 23; michael@0: * if(value<=0x2fffff) { michael@0: * (value-0x1f0000) is a code point; (BMP: value<=0x1fffff) michael@0: * } else { michael@0: * bits 17..0 (value&0x3ffff) is an index to michael@0: * the result UChars in toUUChars[]; (0 indexes toUUChars[0]) michael@0: * length of the result=((value>>18)-12); (length=0..19) michael@0: * } michael@0: * } michael@0: * michael@0: * The first word in a section contains the number of following words in the michael@0: * input byte position (bits 31..24, number=1..0xff). michael@0: * The value of the initial word is used when the current byte is not found michael@0: * in this section. michael@0: * If the value is not 0, then it represents a result as above. michael@0: * If the value is 0, then the search has to return a shorter match with an michael@0: * earlier default value as the result, or result in "unmappable" even for the michael@0: * initial bytes. michael@0: * If the value is 0 for the initial toUTable entry, then the initial byte michael@0: * does not start any mapping input. michael@0: * michael@0: * michael@0: * UChar toUUChars[]; michael@0: * michael@0: * Contains toUnicode mapping results, stored as sequences of UChars. michael@0: * Indexes and lengths stored in the toUTable[]. michael@0: * michael@0: * michael@0: * UChar fromUTableUChars[]; michael@0: * uint32_t fromUTableValues[]; michael@0: * michael@0: * The fromUTable is split into two arrays, but works otherwise much like michael@0: * the toUTable. The array is partitioned into sections like collation michael@0: * contraction tables and toUTable. michael@0: * A row in the table consists of same-index entries in fromUTableUChars[] michael@0: * and fromUTableValues[]. michael@0: * michael@0: * Interpret a value as follows: michael@0: * if(value==0) { michael@0: * no match, see below michael@0: * } else if(value<=0xffffff) { (bits 31..24 are 0) michael@0: * partial match - use value as index to the next fromUTable section michael@0: * and match the next unit; (value indexes fromUTable[value]) michael@0: * } else { michael@0: * if(value==0x80000001) { michael@0: * return no mapping, but request for ; michael@0: * } michael@0: * if(bit 31 set) { michael@0: * roundtrip (|0); michael@0: * } else if(bit 30 set) { michael@0: * "good one-way" mapping (|4); -- new in ICU4C 51, _MBCSHeader.version 5.4/4.4 michael@0: * } else { michael@0: * normal fallback (|1); michael@0: * } michael@0: * // bit 29 reserved, 0 michael@0: * length=(value>>24)&0x1f; (bits 28..24) michael@0: * if(length==1..3) { michael@0: * bits 23..0 contain 1..3 bytes, padded with 00s on the left; michael@0: * } else { michael@0: * bits 23..0 (value&0xffffff) is an index to michael@0: * the result bytes in fromUBytes[]; (0 indexes fromUBytes[0]) michael@0: * } michael@0: * } michael@0: * michael@0: * The first pair in a section contains the number of following pairs in the michael@0: * UChar position (16 bits, number=1..0xffff). michael@0: * The value of the initial pair is used when the current UChar is not found michael@0: * in this section. michael@0: * If the value is not 0, then it represents a result as above. michael@0: * If the value is 0, then the search has to return a shorter match with an michael@0: * earlier default value as the result, or result in "unmappable" even for the michael@0: * initial UChars. michael@0: * michael@0: * If the from Unicode trie is present, then the from Unicode search tables michael@0: * are not used for initial code points. michael@0: * In this case, the first entries (index 0) in the tables are not used michael@0: * (reserved, set to 0) because a value of 0 is used in trie results michael@0: * to indicate no mapping. michael@0: * michael@0: * michael@0: * uint16_t fromUStage12[]; michael@0: * michael@0: * Stages 1 & 2 of a trie that maps an initial code point. michael@0: * Indexes in stage 1 are all offset by the length of stage 1 so that the michael@0: * same array pointer can be used for both stages. michael@0: * If (c>>10)>=(length of stage 1) then c does not start any mapping. michael@0: * Same bit distribution as for regular conversion tries. michael@0: * michael@0: * michael@0: * uint16_t fromUStage3[]; michael@0: * uint32_t fromUStage3b[]; michael@0: * michael@0: * Stage 3 of the trie. The first array simply contains indexes to the second, michael@0: * which contains words in the same format as fromUTableValues[]. michael@0: * Use a stage 3 granularity of 4, which allows for 256k stage 3 entries, michael@0: * and 16-bit entries in stage 3 allow for 64k stage 3b entries. michael@0: * The stage 3 granularity means that the stage 2 entry needs to be left-shifted. michael@0: * michael@0: * Two arrays are used because it is expected that more than half of the stage 3 michael@0: * entries will be zero. The 16-bit index stage 3 array saves space even michael@0: * considering storing a total of 6 bytes per non-zero entry in both arrays michael@0: * together. michael@0: * Using a stage 3 granularity of >1 diminishes the compactability in that stage michael@0: * but provides a larger effective addressing space in stage 2. michael@0: * All but the final result stage use 16-bit entries to save space. michael@0: * michael@0: * fromUStage3b[] contains a zero for "no mapping" at its index 0, michael@0: * and may contain UCNV_EXT_FROM_U_SUBCHAR1 at index 1 for " SUB mapping" michael@0: * (i.e., "no mapping" with preference for rather than ), michael@0: * and all other items are unique non-zero results. michael@0: * michael@0: * The default value of a fromUTableValues[] section that is referenced michael@0: * _directly_ from a fromUStage3b[] item may also be UCNV_EXT_FROM_U_SUBCHAR1, michael@0: * but this value must not occur anywhere else in fromUTableValues[] michael@0: * because "no mapping" is always a property of a single code point, michael@0: * never of multiple. michael@0: * michael@0: * michael@0: * char fromUBytes[]; michael@0: * michael@0: * Contains fromUnicode mapping results, stored as sequences of chars. michael@0: * Indexes and lengths stored in the fromUTableValues[]. michael@0: */ michael@0: enum { michael@0: UCNV_EXT_INDEXES_LENGTH, /* 0 */ michael@0: michael@0: UCNV_EXT_TO_U_INDEX, /* 1 */ michael@0: UCNV_EXT_TO_U_LENGTH, michael@0: UCNV_EXT_TO_U_UCHARS_INDEX, michael@0: UCNV_EXT_TO_U_UCHARS_LENGTH, michael@0: michael@0: UCNV_EXT_FROM_U_UCHARS_INDEX, /* 5 */ michael@0: UCNV_EXT_FROM_U_VALUES_INDEX, michael@0: UCNV_EXT_FROM_U_LENGTH, michael@0: UCNV_EXT_FROM_U_BYTES_INDEX, michael@0: UCNV_EXT_FROM_U_BYTES_LENGTH, michael@0: michael@0: UCNV_EXT_FROM_U_STAGE_12_INDEX, /* 10 */ michael@0: UCNV_EXT_FROM_U_STAGE_1_LENGTH, michael@0: UCNV_EXT_FROM_U_STAGE_12_LENGTH, michael@0: UCNV_EXT_FROM_U_STAGE_3_INDEX, michael@0: UCNV_EXT_FROM_U_STAGE_3_LENGTH, michael@0: UCNV_EXT_FROM_U_STAGE_3B_INDEX, michael@0: UCNV_EXT_FROM_U_STAGE_3B_LENGTH, michael@0: michael@0: UCNV_EXT_COUNT_BYTES, /* 17 */ michael@0: UCNV_EXT_COUNT_UCHARS, michael@0: UCNV_EXT_FLAGS, michael@0: michael@0: UCNV_EXT_RESERVED_INDEX, /* 20, moves with additional indexes */ michael@0: michael@0: UCNV_EXT_SIZE=31, michael@0: UCNV_EXT_INDEXES_MIN_LENGTH=32 michael@0: }; michael@0: michael@0: /* get the pointer to an extension array from indexes[index] */ michael@0: #define UCNV_EXT_ARRAY(indexes, index, itemType) \ michael@0: ((const itemType *)((const char *)(indexes)+(indexes)[index])) michael@0: michael@0: #define UCNV_GET_MAX_BYTES_PER_UCHAR(indexes) \ michael@0: ((indexes)[UCNV_EXT_COUNT_BYTES]&0xff) michael@0: michael@0: /* internal API ------------------------------------------------------------- */ michael@0: michael@0: U_CFUNC UBool michael@0: ucnv_extInitialMatchToU(UConverter *cnv, const int32_t *cx, michael@0: int32_t firstLength, michael@0: const char **src, const char *srcLimit, michael@0: UChar **target, const UChar *targetLimit, michael@0: int32_t **offsets, int32_t srcIndex, michael@0: UBool flush, michael@0: UErrorCode *pErrorCode); michael@0: michael@0: U_CFUNC UChar32 michael@0: ucnv_extSimpleMatchToU(const int32_t *cx, michael@0: const char *source, int32_t length, michael@0: UBool useFallback); michael@0: michael@0: U_CFUNC void michael@0: ucnv_extContinueMatchToU(UConverter *cnv, michael@0: UConverterToUnicodeArgs *pArgs, int32_t srcIndex, michael@0: UErrorCode *pErrorCode); michael@0: michael@0: michael@0: U_CFUNC UBool michael@0: ucnv_extInitialMatchFromU(UConverter *cnv, const int32_t *cx, michael@0: UChar32 cp, michael@0: const UChar **src, const UChar *srcLimit, michael@0: char **target, const char *targetLimit, michael@0: int32_t **offsets, int32_t srcIndex, michael@0: UBool flush, michael@0: UErrorCode *pErrorCode); michael@0: michael@0: U_CFUNC int32_t michael@0: ucnv_extSimpleMatchFromU(const int32_t *cx, michael@0: UChar32 cp, uint32_t *pValue, michael@0: UBool useFallback); michael@0: michael@0: U_CFUNC void michael@0: ucnv_extContinueMatchFromU(UConverter *cnv, michael@0: UConverterFromUnicodeArgs *pArgs, int32_t srcIndex, michael@0: UErrorCode *pErrorCode); michael@0: michael@0: /* michael@0: * Add code points and strings to the set according to the extension mappings. michael@0: * Limitation on the UConverterSetFilter: michael@0: * The filters currently assume that they are used with 1:1 mappings. michael@0: * They only apply to single input code points, and then they pass through michael@0: * only mappings with single-charset-code results. michael@0: * For example, the Shift-JIS filter only works for 2-byte results and tests michael@0: * that those 2 bytes are in the JIS X 0208 range of Shift-JIS. michael@0: */ michael@0: U_CFUNC void michael@0: ucnv_extGetUnicodeSet(const UConverterSharedData *sharedData, michael@0: const USetAdder *sa, michael@0: UConverterUnicodeSet which, michael@0: UConverterSetFilter filter, michael@0: UErrorCode *pErrorCode); michael@0: michael@0: /* toUnicode helpers -------------------------------------------------------- */ michael@0: michael@0: #define UCNV_EXT_TO_U_BYTE_SHIFT 24 michael@0: #define UCNV_EXT_TO_U_VALUE_MASK 0xffffff michael@0: #define UCNV_EXT_TO_U_MIN_CODE_POINT 0x1f0000 michael@0: #define UCNV_EXT_TO_U_MAX_CODE_POINT 0x2fffff michael@0: #define UCNV_EXT_TO_U_ROUNDTRIP_FLAG ((uint32_t)1<<23) michael@0: #define UCNV_EXT_TO_U_INDEX_MASK 0x3ffff michael@0: #define UCNV_EXT_TO_U_LENGTH_SHIFT 18 michael@0: #define UCNV_EXT_TO_U_LENGTH_OFFSET 12 michael@0: michael@0: /* maximum number of indexed UChars */ michael@0: #define UCNV_EXT_MAX_UCHARS 19 michael@0: michael@0: #define UCNV_EXT_TO_U_MAKE_WORD(byte, value) (((uint32_t)(byte)<>UCNV_EXT_TO_U_BYTE_SHIFT) michael@0: #define UCNV_EXT_TO_U_GET_VALUE(word) ((word)&UCNV_EXT_TO_U_VALUE_MASK) michael@0: michael@0: #define UCNV_EXT_TO_U_IS_PARTIAL(value) ((value)>UCNV_EXT_TO_U_LENGTH_SHIFT)-UCNV_EXT_TO_U_LENGTH_OFFSET) michael@0: michael@0: /* fromUnicode helpers ------------------------------------------------------ */ michael@0: michael@0: /* most trie constants are shared with ucnvmbcs.h */ michael@0: michael@0: /* see similar utrie.h UTRIE_INDEX_SHIFT and UTRIE_DATA_GRANULARITY */ michael@0: #define UCNV_EXT_STAGE_2_LEFT_SHIFT 2 michael@0: #define UCNV_EXT_STAGE_3_GRANULARITY 4 michael@0: michael@0: /* trie access, returns the stage 3 value=index to stage 3b; s1Index=c>>10 */ michael@0: #define UCNV_EXT_FROM_U(stage12, stage3, s1Index, c) \ michael@0: (stage3)[ ((int32_t)(stage12)[ (stage12)[s1Index] +(((c)>>4)&0x3f) ]< (impossible roundtrip to 0 bytes, value 01) */ michael@0: #define UCNV_EXT_FROM_U_SUBCHAR1 0x80000001 michael@0: michael@0: /* at most 3 bytes in the lower part of the value */ michael@0: #define UCNV_EXT_FROM_U_MAX_DIRECT_LENGTH 3 michael@0: michael@0: /* maximum number of indexed bytes */ michael@0: #define UCNV_EXT_MAX_BYTES 0x1f michael@0: michael@0: #define UCNV_EXT_FROM_U_IS_PARTIAL(value) (((value)>>UCNV_EXT_FROM_U_LENGTH_SHIFT)==0) michael@0: #define UCNV_EXT_FROM_U_GET_PARTIAL_INDEX(value) (value) michael@0: michael@0: #define UCNV_EXT_FROM_U_IS_ROUNDTRIP(value) (((value)&UCNV_EXT_FROM_U_ROUNDTRIP_FLAG)!=0) michael@0: #define UCNV_EXT_FROM_U_MASK_ROUNDTRIP(value) ((value)&~UCNV_EXT_FROM_U_ROUNDTRIP_FLAG) michael@0: michael@0: /* get length; masks away all other bits */ michael@0: #define UCNV_EXT_FROM_U_GET_LENGTH(value) (int32_t)(((value)>>UCNV_EXT_FROM_U_LENGTH_SHIFT)&UCNV_EXT_MAX_BYTES) michael@0: michael@0: /* get bytes or bytes index */ michael@0: #define UCNV_EXT_FROM_U_GET_DATA(value) ((value)&UCNV_EXT_FROM_U_DATA_MASK) michael@0: michael@0: #endif michael@0: michael@0: #endif