1 /*
2 *******************************************************************************
3 * Copyright (C) 2001-2014, International Business Machines
4 * Corporation and others. All Rights Reserved.
5 *******************************************************************************
6 * file name: bocsu.cpp
7 * encoding: US-ASCII
8 * tab size: 8 (not used)
9 * indentation:4
10 *
11 * Author: Markus W. Scherer
12 *
13 * Modification history:
14 * 05/18/2001 weiv Made into separate module
15 */
16
17
18 #include "unicode/utypes.h"
19
20 #if !UCONFIG_NO_COLLATION
21
22 #include "unicode/bytestream.h"
23 #include "unicode/utf16.h"
24 #include "bocsu.h"
25
26 /*
27 * encode one difference value -0x10ffff..+0x10ffff in 1..4 bytes,
28 * preserving lexical order
29 */
30 static uint8_t *
u_writeDiff(int32_t diff,uint8_t * p)31 u_writeDiff(int32_t diff, uint8_t *p) {
32 if(diff>=SLOPE_REACH_NEG_1) {
33 if(diff<=SLOPE_REACH_POS_1) {
34 *p++=(uint8_t)(SLOPE_MIDDLE+diff);
35 } else if(diff<=SLOPE_REACH_POS_2) {
36 *p++=(uint8_t)(SLOPE_START_POS_2+(diff/SLOPE_TAIL_COUNT));
37 *p++=(uint8_t)(SLOPE_MIN+diff%SLOPE_TAIL_COUNT);
38 } else if(diff<=SLOPE_REACH_POS_3) {
39 p[2]=(uint8_t)(SLOPE_MIN+diff%SLOPE_TAIL_COUNT);
40 diff/=SLOPE_TAIL_COUNT;
41 p[1]=(uint8_t)(SLOPE_MIN+diff%SLOPE_TAIL_COUNT);
42 *p=(uint8_t)(SLOPE_START_POS_3+(diff/SLOPE_TAIL_COUNT));
43 p+=3;
44 } else {
45 p[3]=(uint8_t)(SLOPE_MIN+diff%SLOPE_TAIL_COUNT);
46 diff/=SLOPE_TAIL_COUNT;
47 p[2]=(uint8_t)(SLOPE_MIN+diff%SLOPE_TAIL_COUNT);
48 diff/=SLOPE_TAIL_COUNT;
49 p[1]=(uint8_t)(SLOPE_MIN+diff%SLOPE_TAIL_COUNT);
50 *p=SLOPE_MAX;
51 p+=4;
52 }
53 } else {
54 int32_t m;
55
56 if(diff>=SLOPE_REACH_NEG_2) {
57 NEGDIVMOD(diff, SLOPE_TAIL_COUNT, m);
58 *p++=(uint8_t)(SLOPE_START_NEG_2+diff);
59 *p++=(uint8_t)(SLOPE_MIN+m);
60 } else if(diff>=SLOPE_REACH_NEG_3) {
61 NEGDIVMOD(diff, SLOPE_TAIL_COUNT, m);
62 p[2]=(uint8_t)(SLOPE_MIN+m);
63 NEGDIVMOD(diff, SLOPE_TAIL_COUNT, m);
64 p[1]=(uint8_t)(SLOPE_MIN+m);
65 *p=(uint8_t)(SLOPE_START_NEG_3+diff);
66 p+=3;
67 } else {
68 NEGDIVMOD(diff, SLOPE_TAIL_COUNT, m);
69 p[3]=(uint8_t)(SLOPE_MIN+m);
70 NEGDIVMOD(diff, SLOPE_TAIL_COUNT, m);
71 p[2]=(uint8_t)(SLOPE_MIN+m);
72 NEGDIVMOD(diff, SLOPE_TAIL_COUNT, m);
73 p[1]=(uint8_t)(SLOPE_MIN+m);
74 *p=SLOPE_MIN;
75 p+=4;
76 }
77 }
78 return p;
79 }
80
81 /*
82 * Encode the code points of a string as
83 * a sequence of byte-encoded differences (slope detection),
84 * preserving lexical order.
85 *
86 * Optimize the difference-taking for runs of Unicode text within
87 * small scripts:
88 *
89 * Most small scripts are allocated within aligned 128-blocks of Unicode
90 * code points. Lexical order is preserved if "prev" is always moved
91 * into the middle of such a block.
92 *
93 * Additionally, "prev" is moved from anywhere in the Unihan
94 * area into the middle of that area.
95 * Note that the identical-level run in a sort key is generated from
96 * NFD text - there are never Hangul characters included.
97 */
98 U_CFUNC UChar32
u_writeIdenticalLevelRun(UChar32 prev,const UChar * s,int32_t length,icu::ByteSink & sink)99 u_writeIdenticalLevelRun(UChar32 prev, const UChar *s, int32_t length, icu::ByteSink &sink) {
100 char scratch[64];
101 int32_t capacity;
102
103 int32_t i=0;
104 while(i<length) {
105 char *buffer=sink.GetAppendBuffer(1, length*2, scratch, (int32_t)sizeof(scratch), &capacity);
106 uint8_t *p;
107 // We must have capacity>=SLOPE_MAX_BYTES in case u_writeDiff() writes that much,
108 // but we do not want to force the sink.GetAppendBuffer() to allocate
109 // for a large min_capacity because we might actually only write one byte.
110 if(capacity<16) {
111 buffer=scratch;
112 capacity=(int32_t)sizeof(scratch);
113 }
114 p=reinterpret_cast<uint8_t *>(buffer);
115 uint8_t *lastSafe=p+capacity-SLOPE_MAX_BYTES;
116 while(i<length && p<=lastSafe) {
117 if(prev<0x4e00 || prev>=0xa000) {
118 prev=(prev&~0x7f)-SLOPE_REACH_NEG_1;
119 } else {
120 /*
121 * Unihan U+4e00..U+9fa5:
122 * double-bytes down from the upper end
123 */
124 prev=0x9fff-SLOPE_REACH_POS_2;
125 }
126
127 UChar32 c;
128 U16_NEXT(s, i, length, c);
129 if(c==0xfffe) {
130 *p++=2; // merge separator
131 prev=0;
132 } else {
133 p=u_writeDiff(c-prev, p);
134 prev=c;
135 }
136 }
137 sink.Append(buffer, (int32_t)(p-reinterpret_cast<uint8_t *>(buffer)));
138 }
139 return prev;
140 }
141
142 #endif /* #if !UCONFIG_NO_COLLATION */
143