]>
Commit | Line | Data |
---|---|---|
b75a7d8f A |
1 | /* |
2 | ******************************************************************************* | |
4388f060 | 3 | * Copyright (C) 2001-2011, International Business Machines |
b75a7d8f A |
4 | * Corporation and others. All Rights Reserved. |
5 | ******************************************************************************* | |
4388f060 | 6 | * file name: bocsu.cpp |
b75a7d8f A |
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 | ||
4388f060 A |
22 | #include "unicode/bytestream.h" |
23 | #include "unicode/utf16.h" | |
b75a7d8f A |
24 | #include "bocsu.h" |
25 | ||
26 | /* | |
27 | * encode one difference value -0x10ffff..+0x10ffff in 1..3 bytes, | |
28 | * preserving lexical order | |
29 | */ | |
30 | U_CFUNC uint8_t * | |
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 | ||
b75a7d8f A |
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 | */ | |
4388f060 A |
98 | U_CFUNC void |
99 | u_writeIdenticalLevelRun(const UChar *s, int32_t length, icu::ByteSink &sink) { | |
100 | char scratch[64]; | |
101 | int32_t capacity; | |
b75a7d8f | 102 | |
4388f060 A |
103 | UChar32 prev=0; |
104 | int32_t i=0; | |
b75a7d8f | 105 | while(i<length) { |
4388f060 A |
106 | char *buffer=sink.GetAppendBuffer(1, length*2, scratch, (int32_t)sizeof(scratch), &capacity); |
107 | uint8_t *p; | |
108 | // We must have capacity>=SLOPE_MAX_BYTES in case u_writeDiff() writes that much, | |
109 | // but we do not want to force the sink.GetAppendBuffer() to allocate | |
110 | // for a large min_capacity because we might actually only write one byte. | |
111 | if(capacity<16) { | |
112 | buffer=scratch; | |
113 | capacity=(int32_t)sizeof(scratch); | |
b75a7d8f | 114 | } |
4388f060 A |
115 | p=reinterpret_cast<uint8_t *>(buffer); |
116 | uint8_t *lastSafe=p+capacity-SLOPE_MAX_BYTES; | |
117 | while(i<length && p<=lastSafe) { | |
118 | if(prev<0x4e00 || prev>=0xa000) { | |
119 | prev=(prev&~0x7f)-SLOPE_REACH_NEG_1; | |
120 | } else { | |
121 | /* | |
122 | * Unihan U+4e00..U+9fa5: | |
123 | * double-bytes down from the upper end | |
124 | */ | |
125 | prev=0x9fff-SLOPE_REACH_POS_2; | |
126 | } | |
127 | ||
128 | UChar32 c; | |
129 | U16_NEXT(s, i, length, c); | |
130 | p=u_writeDiff(c-prev, p); | |
131 | prev=c; | |
132 | } | |
133 | sink.Append(buffer, (int32_t)(p-reinterpret_cast<uint8_t *>(buffer))); | |
b75a7d8f | 134 | } |
b75a7d8f A |
135 | } |
136 | ||
137 | U_CFUNC int32_t | |
138 | u_writeIdenticalLevelRunTwoChars(UChar32 first, UChar32 second, uint8_t *p) { | |
139 | uint8_t *p0 = p; | |
140 | if(first<0x4e00 || first>=0xa000) { | |
141 | first=(first&~0x7f)-SLOPE_REACH_NEG_1; | |
142 | } else { | |
143 | /* | |
144 | * Unihan U+4e00..U+9fa5: | |
145 | * double-bytes down from the upper end | |
146 | */ | |
147 | first=0x9fff-SLOPE_REACH_POS_2; | |
148 | } | |
149 | ||
150 | p=u_writeDiff(second-first, p); | |
151 | return (int32_t)(p-p0); | |
152 | } | |
153 | ||
b75a7d8f | 154 | #endif /* #if !UCONFIG_NO_COLLATION */ |