icuSources/i18n/collationfcd.h

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