| 1 | // © 2016 and later: Unicode, Inc. and others. |
| 2 | // License & terms of use: http://www.unicode.org/copyright.html |
| 3 | /* |
| 4 | ******************************************************************************* |
| 5 | * Copyright (C) 2012-2014, International Business Machines |
| 6 | * Corporation and others. All Rights Reserved. |
| 7 | ******************************************************************************* |
| 8 | * collationfcd.h |
| 9 | * |
| 10 | * created on: 2012aug18 |
| 11 | * created by: Markus W. Scherer |
| 12 | */ |
| 13 | |
| 14 | #ifndef __COLLATIONFCD_H__ |
| 15 | #define __COLLATIONFCD_H__ |
| 16 | |
| 17 | #include "unicode/utypes.h" |
| 18 | |
| 19 | #if !UCONFIG_NO_COLLATION |
| 20 | |
| 21 | #include "unicode/utf16.h" |
| 22 | |
| 23 | U_NAMESPACE_BEGIN |
| 24 | |
| 25 | /** |
| 26 | * Data and functions for the FCD check fast path. |
| 27 | * |
| 28 | * The fast path looks at a pair of 16-bit code units and checks |
| 29 | * whether there is an FCD boundary between them; |
| 30 | * there is if the first unit has a trailing ccc=0 (!hasTccc(first)) |
| 31 | * or the second unit has a leading ccc=0 (!hasLccc(second)), |
| 32 | * or both. |
| 33 | * When the fast path finds a possible non-boundary, |
| 34 | * then the FCD check slow path looks at the actual sequence of FCD values. |
| 35 | * |
| 36 | * This is a pure optimization. |
| 37 | * The fast path must at least find all possible non-boundaries. |
| 38 | * If the fast path is too pessimistic, it costs performance. |
| 39 | * |
| 40 | * For a pair of BMP characters, the fast path tests are precise (1 bit per character). |
| 41 | * |
| 42 | * For a supplementary code point, the two units are its lead and trail surrogates. |
| 43 | * We set hasTccc(lead)=true if any of its 1024 associated supplementary code points |
| 44 | * has lccc!=0 or tccc!=0. |
| 45 | * We set hasLccc(trail)=true for all trail surrogates. |
| 46 | * As a result, we leave the fast path if the lead surrogate might start a |
| 47 | * supplementary code point that is not FCD-inert. |
| 48 | * (So the fast path need not detect that there is a surrogate pair, |
| 49 | * nor look ahead to the next full code point.) |
| 50 | * |
| 51 | * hasLccc(lead)=true if any of its 1024 associated supplementary code points |
| 52 | * has lccc!=0, for fast boundary checking between BMP & supplementary. |
| 53 | * |
| 54 | * hasTccc(trail)=false: |
| 55 | * It should only be tested for unpaired trail surrogates which are FCD-inert. |
| 56 | */ |
| 57 | class U_I18N_API CollationFCD { |
| 58 | public: |
| 59 | static inline UBool hasLccc(UChar32 c) { |
| 60 | // assert c <= 0xffff |
| 61 | // c can be negative, e.g., U_SENTINEL from UCharIterator; |
| 62 | // that is handled in the first test. |
| 63 | int32_t i; |
| 64 | return |
| 65 | // U+0300 is the first character with lccc!=0. |
| 66 | c >= 0x300 && |
| 67 | (i = lcccIndex[c >> 5]) != 0 && |
| 68 | (lcccBits[i] & ((uint32_t)1 << (c & 0x1f))) != 0; |
| 69 | } |
| 70 | |
| 71 | static inline UBool hasTccc(UChar32 c) { |
| 72 | // assert c <= 0xffff |
| 73 | // c can be negative, e.g., U_SENTINEL from UCharIterator; |
| 74 | // that is handled in the first test. |
| 75 | int32_t i; |
| 76 | return |
| 77 | // U+00C0 is the first character with tccc!=0. |
| 78 | c >= 0xc0 && |
| 79 | (i = tcccIndex[c >> 5]) != 0 && |
| 80 | (tcccBits[i] & ((uint32_t)1 << (c & 0x1f))) != 0; |
| 81 | } |
| 82 | |
| 83 | static inline UBool mayHaveLccc(UChar32 c) { |
| 84 | // Handles all of Unicode 0..10FFFF. |
| 85 | // c can be negative, e.g., U_SENTINEL. |
| 86 | // U+0300 is the first character with lccc!=0. |
| 87 | if(c < 0x300) { return FALSE; } |
| 88 | if(c > 0xffff) { c = U16_LEAD(c); } |
| 89 | int32_t i; |
| 90 | return |
| 91 | (i = lcccIndex[c >> 5]) != 0 && |
| 92 | (lcccBits[i] & ((uint32_t)1 << (c & 0x1f))) != 0; |
| 93 | } |
| 94 | |
| 95 | /** |
| 96 | * Tibetan composite vowel signs (U+0F73, U+0F75, U+0F81) |
| 97 | * must be decomposed before reaching the core collation code, |
| 98 | * or else some sequences including them, even ones passing the FCD check, |
| 99 | * do not yield canonically equivalent results. |
| 100 | * |
| 101 | * This is a fast and imprecise test. |
| 102 | * |
| 103 | * @param c a code point |
| 104 | * @return TRUE if c is U+0F73, U+0F75 or U+0F81 or one of several other Tibetan characters |
| 105 | */ |
| 106 | static inline UBool maybeTibetanCompositeVowel(UChar32 c) { |
| 107 | return (c & 0x1fff01) == 0xf01; |
| 108 | } |
| 109 | |
| 110 | /** |
| 111 | * Tibetan composite vowel signs (U+0F73, U+0F75, U+0F81) |
| 112 | * must be decomposed before reaching the core collation code, |
| 113 | * or else some sequences including them, even ones passing the FCD check, |
| 114 | * do not yield canonically equivalent results. |
| 115 | * |
| 116 | * They have distinct lccc/tccc combinations: 129/130 or 129/132. |
| 117 | * |
| 118 | * @param fcd16 the FCD value (lccc/tccc combination) of a code point |
| 119 | * @return TRUE if fcd16 is from U+0F73, U+0F75 or U+0F81 |
| 120 | */ |
| 121 | static inline UBool isFCD16OfTibetanCompositeVowel(uint16_t fcd16) { |
| 122 | return fcd16 == 0x8182 || fcd16 == 0x8184; |
| 123 | } |
| 124 | |
| 125 | private: |
| 126 | CollationFCD(); // No instantiation. |
| 127 | |
| 128 | static const uint8_t lcccIndex[2048]; |
| 129 | static const uint8_t tcccIndex[2048]; |
| 130 | static const uint32_t lcccBits[]; |
| 131 | static const uint32_t tcccBits[]; |
| 132 | }; |
| 133 | |
| 134 | U_NAMESPACE_END |
| 135 | |
| 136 | #endif // !UCONFIG_NO_COLLATION |
| 137 | #endif // __COLLATIONFCD_H__ |
| 138 | |