ICU 56 update step 1

source/i18n/collationfastlatin.h

 /*
 *******************************************************************************
-* Copyright (C) 2013-2014, International Business Machines
+* Copyright (C) 2013-2015, International Business Machines
 * Corporation and others.  All Rights Reserved.
 *******************************************************************************
 * collationfastlatin.h
 *
 * created on: 2013aug09
 * created by: Markus W. Scherer
 */
 
 #ifndef __COLLATIONFASTLATIN_H__
 #define __COLLATIONFASTLATIN_H__
@@ skipping 10 matching lines @@
 class U_I18N_API CollationFastLatin /* all static */ {
 public:
     /**
      * Fast Latin format version (one byte 1..FF).
      * Must be incremented for any runtime-incompatible changes,
      * in particular, for changes to any of the following constants.
      *
      * When the major version number of the main data format changes,
      * we can reset this fast Latin version to 1.
      */
-    static const uint16_t VERSION = 1;
+    static const uint16_t VERSION = 2;
 
     static const int32_t LATIN_MAX = 0x17f;
     static const int32_t LATIN_LIMIT = LATIN_MAX + 1;
 
     static const int32_t LATIN_MAX_UTF8_LEAD = 0xc5;  // UTF-8 lead byte of LATIN_MAX
 
     static const int32_t PUNCT_START = 0x2000;
     static const int32_t PUNCT_LIMIT = 0x2040;
 
     // excludes U+FFFE & U+FFFF
@@ skipping 200 matching lines @@
     static uint32_t getCases(uint32_t variableTop, UBool strengthIsPrimary, uint32_t pair);
     static uint32_t getTertiaries(uint32_t variableTop, UBool withCaseBits, uint32_t pair);
     static uint32_t getQuaternaries(uint32_t variableTop, uint32_t pair);
 
 private:
     CollationFastLatin();  // no constructor
 };
 
 /*
  * Format of the CollationFastLatin data table.
- * CollationFastLatin::VERSION = 1.
+ * CollationFastLatin::VERSION = 2.
  *
  * This table contains data for a Latin-text collation fastpath.
  * The data is stored as an array of uint16_t which contains the following parts.
  *
  * uint16_t  -- version & header length
  *   Bits 15..8: version, must match the VERSION
  *         7..0: length of the header
  *
  * uint16_t varTops[header length - 1]
+ *   Version 2:
+ *   varTops[m] is the highest CollationFastLatin long-primary weight
+ *   of supported maxVariable group m
+ *   (special reorder group space, punct, symbol, currency).
+ *
+ *   Version 1:
  *   Each of these values maps the variable top lead byte of a supported maxVariable group
  *   to the highest CollationFastLatin long-primary weight.
  *   The values are stored in ascending order.
  *   Bits 15..7: max fast-Latin long-primary weight (bits 11..3 shifted left by 4 bits)
  *         6..0: regular primary lead byte
  *
  * uint16_t miniCEs[0x1c0]
  *   A mini collation element for each character U+0000..U+017F and U+2000..U+203F.
  *   Each value encodes one or two mini CEs (two are possible if the first one
  *   has a short mini primary and the second one is a secondary CE, i.e., primary == 0),
@@ skipping 11 matching lines @@
  * uint16_t contractions[variable length];
  *   Contraction mini CEs contain an offset relative to just after the miniCEs table.
  *   It points to a list of tuples which map from a contraction suffix character to a result.
  *   First uint16_t of each tuple:
  *     Bits 10..9: Length of the result (1..3), see comments on CONTR_LENGTH_SHIFT.
  *     Bits  8..0: Contraction character, see comments on CONTR_CHAR_MASK.
  *   This is followed by 0, 1, or 2 uint16_t according to the length.
  *   Each list is terminated by an entry with CONTR_CHAR_MASK.
  *   Each list starts with such an entry which also contains the default result
  *   for when there is no contraction match.
+ *
+ * -----------------
+ * Changes for version 2 (ICU 55)
+ *
+ * Special reorder groups do not necessarily start on whole primary lead bytes any more.
+ * Therefore, the varTops data has a new format:
+ * Version 1 stored the lead bytes of the highest root primaries for
+ * the maxVariable-supported special reorder groups.
+ * Now the top 16 bits would need to be stored,
+ * and it is simpler to store only the fast-Latin weights.
  */
 
 U_NAMESPACE_END
 
 #endif  // !UCONFIG_NO_COLLATION
 #endif  // __COLLATIONFASTLATIN_H__