1 // © 2016 and later: Unicode, Inc. and others.
2 // License & terms of use: http://www.unicode.org/copyright.html#License
3 /**
4 *******************************************************************************
5 * Copyright (C) 1996-2014, International Business Machines Corporation and
6 * others. All Rights Reserved.
7 *******************************************************************************
8 */
9 package com.ibm.icu.impl.coll;
10 
11 import com.ibm.icu.util.ByteArrayWrapper;
12 
13 /**
14  * <p>Binary Ordered Compression Scheme for Unicode</p>
15  *
16  * <p>Users are strongly encouraged to read the ICU paper on
17  * <a href="http://www.icu-project.org/docs/papers/binary_ordered_compression_for_unicode.html">
18  * BOCU</a> before attempting to use this class.</p>
19  *
20  * <p>BOCU is used to compress unicode text into a stream of unsigned
21  * bytes.  For many kinds of text the compression compares favorably
22  * to UTF-8, and for some kinds of text (such as CJK) it does better.
23  * The resulting bytes will compare in the same order as the original
24  * code points.  The byte stream does not contain the values 0, 1, or
25  * 2.</p>
26  *
27  * <p>One example of a use of BOCU is in
28  * com.ibm.icu.text.Collator#getCollationKey(String) for a RuleBasedCollator object with
29  * collation strength IDENTICAL. The result CollationKey will consist of the
30  * collation order of the source string followed by the BOCU result of the
31  * source string.
32  * </p>
33  *
34  * <p>Unlike a UTF encoding, BOCU-compressed text is not suitable for
35  * random access.</p>
36  *
37  * <p>Method: Slope Detection<br> Remember the previous code point
38  * (initial 0).  For each code point in the string, encode the
39  * difference with the previous one.  Similar to a UTF, the length of
40  * the byte sequence is encoded in the lead bytes.  Unlike a UTF, the
41  * trail byte values may overlap with lead/single byte values.  The
42  * signedness of the difference must be encoded as the most
43  * significant part.</p>
44  *
45  * <p>We encode differences with few bytes if their absolute values
46  * are small.  For correct ordering, we must treat the entire value
47  * range -10ffff..+10ffff in ascending order, which forbids encoding
48  * the sign and the absolute value separately. Instead, we split the
49  * lead byte range in the middle and encode non-negative values going
50  * up and negative values going down.</p>
51  *
52  * <p>For very small absolute values, the difference is added to a
53  * middle byte value for single-byte encoded differences.  For
54  * somewhat larger absolute values, the difference is divided by the
55  * number of byte values available, the modulo is used for one trail
56  * byte, and the remainder is added to a lead byte avoiding the
57  * single-byte range.  For large absolute values, the difference is
58  * similarly encoded in three bytes. (Syn Wee, I need examples
59  * here.)</p>
60  *
61  * <p>BOCU does not use byte values 0, 1, or 2, but uses all other
62  * byte values for lead and single bytes, so that the middle range of
63  * single bytes is as large as possible.</p>
64  *
65  * <p>Note that the lead byte ranges overlap some, but that the
66  * sequences as a whole are well ordered. I.e., even if the lead byte
67  * is the same for sequences of different lengths, the trail bytes
68  * establish correct order.  It would be possible to encode slightly
69  * larger ranges for each length (>1) by subtracting the lower bound
70  * of the range. However, that would also slow down the calculation.
71  * (Syn Wee, need an example).</p>
72  *
73  * <p>For the actual string encoding, an optimization moves the
74  * previous code point value to the middle of its Unicode script block
75  * to minimize the differences in same-script text runs.  (Syn Wee,
76  * need an example.)</p>
77  *
78  * @author Syn Wee Quek
79  * @since release 2.2, May 3rd 2002
80  */
81 public class BOCSU
82 {
83     // public methods -------------------------------------------------------
84 
85     /**
86      * Encode the code points of a string as
87      * a sequence of byte-encoded differences (slope detection),
88      * preserving lexical order.
89      *
90      * <p>Optimize the difference-taking for runs of Unicode text within
91      * small scripts:
92      *
93      * <p>Most small scripts are allocated within aligned 128-blocks of Unicode
94      * code points. Lexical order is preserved if "prev" is always moved
95      * into the middle of such a block.
96      *
97      * <p>Additionally, "prev" is moved from anywhere in the Unihan
98      * area into the middle of that area.
99      * Note that the identical-level run in a sort key is generated from
100      * NFD text - there are never Hangul characters included.
101      */
writeIdenticalLevelRun(int prev, CharSequence s, int i, int length, ByteArrayWrapper sink)102     public static int writeIdenticalLevelRun(int prev, CharSequence s, int i, int length, ByteArrayWrapper sink) {
103         while (i < length) {
104             // We must have capacity>=SLOPE_MAX_BYTES in case writeDiff() writes that much,
105             // but we do not want to force the sink to allocate
106             // for a large min_capacity because we might actually only write one byte.
107             ensureAppendCapacity(sink, 16, s.length() * 2);
108             byte[] buffer = sink.bytes;
109             int capacity = buffer.length;
110             int p = sink.size;
111             int lastSafe = capacity - SLOPE_MAX_BYTES_;
112             while (i < length && p <= lastSafe) {
113                 if (prev < 0x4e00 || prev >= 0xa000) {
114                     prev = (prev & ~0x7f) - SLOPE_REACH_NEG_1_;
115                 } else {
116                     // Unihan U+4e00..U+9fa5:
117                     // double-bytes down from the upper end
118                     prev = 0x9fff - SLOPE_REACH_POS_2_;
119                 }
120 
121                 int c = Character.codePointAt(s, i);
122                 i += Character.charCount(c);
123                 if (c == 0xfffe) {
124                     buffer[p++] = 2;  // merge separator
125                     prev = 0;
126                 } else {
127                     p = writeDiff(c - prev, buffer, p);
128                     prev = c;
129                 }
130             }
131             sink.size = p;
132         }
133         return prev;
134     }
135 
ensureAppendCapacity(ByteArrayWrapper sink, int minCapacity, int desiredCapacity)136     private static void ensureAppendCapacity(ByteArrayWrapper sink, int minCapacity, int desiredCapacity) {
137         int remainingCapacity = sink.bytes.length - sink.size;
138         if (remainingCapacity >= minCapacity) { return; }
139         if (desiredCapacity < minCapacity) { desiredCapacity = minCapacity; }
140         sink.ensureCapacity(sink.size + desiredCapacity);
141     }
142 
143     // private data members --------------------------------------------------
144 
145     /**
146      * Do not use byte values 0, 1, 2 because they are separators in sort keys.
147      */
148     private static final int SLOPE_MIN_ = 3;
149     private static final int SLOPE_MAX_ = 0xff;
150     private static final int SLOPE_MIDDLE_ = 0x81;
151     private static final int SLOPE_TAIL_COUNT_ = SLOPE_MAX_ - SLOPE_MIN_ + 1;
152     private static final int SLOPE_MAX_BYTES_ = 4;
153 
154     /**
155      * Number of lead bytes:
156      * 1        middle byte for 0
157      * 2*80=160 single bytes for !=0
158      * 2*42=84  for double-byte values
159      * 2*3=6    for 3-byte values
160      * 2*1=2    for 4-byte values
161      *
162      * The sum must be <=SLOPE_TAIL_COUNT.
163      *
164      * Why these numbers?
165      * - There should be >=128 single-byte values to cover 128-blocks
166      *   with small scripts.
167      * - There should be >=20902 single/double-byte values to cover Unihan.
168      * - It helps CJK Extension B some if there are 3-byte values that cover
169      *   the distance between them and Unihan.
170      *   This also helps to jump among distant places in the BMP.
171      * - Four-byte values are necessary to cover the rest of Unicode.
172      *
173      * Symmetrical lead byte counts are for convenience.
174      * With an equal distribution of even and odd differences there is also
175      * no advantage to asymmetrical lead byte counts.
176      */
177     private static final int SLOPE_SINGLE_ = 80;
178     private static final int SLOPE_LEAD_2_ = 42;
179     private static final int SLOPE_LEAD_3_ = 3;
180     //private static final int SLOPE_LEAD_4_ = 1;
181 
182     /**
183      * The difference value range for single-byters.
184      */
185     private static final int SLOPE_REACH_POS_1_ = SLOPE_SINGLE_;
186     private static final int SLOPE_REACH_NEG_1_ = (-SLOPE_SINGLE_);
187 
188     /**
189      * The difference value range for double-byters.
190      */
191     private static final int SLOPE_REACH_POS_2_ =
192         SLOPE_LEAD_2_ * SLOPE_TAIL_COUNT_ + SLOPE_LEAD_2_ - 1;
193     private static final int SLOPE_REACH_NEG_2_ = (-SLOPE_REACH_POS_2_ - 1);
194 
195     /**
196      * The difference value range for 3-byters.
197      */
198     private static final int SLOPE_REACH_POS_3_ = SLOPE_LEAD_3_
199         * SLOPE_TAIL_COUNT_
200         * SLOPE_TAIL_COUNT_
201         + (SLOPE_LEAD_3_ - 1)
202         * SLOPE_TAIL_COUNT_ +
203         (SLOPE_TAIL_COUNT_ - 1);
204     private static final int SLOPE_REACH_NEG_3_ = (-SLOPE_REACH_POS_3_ - 1);
205 
206     /**
207      * The lead byte start values.
208      */
209     private static final int SLOPE_START_POS_2_ = SLOPE_MIDDLE_
210         + SLOPE_SINGLE_ + 1;
211     private static final int SLOPE_START_POS_3_ = SLOPE_START_POS_2_
212         + SLOPE_LEAD_2_;
213     private static final int SLOPE_START_NEG_2_ = SLOPE_MIDDLE_ +
214         SLOPE_REACH_NEG_1_;
215     private static final int SLOPE_START_NEG_3_ = SLOPE_START_NEG_2_
216         - SLOPE_LEAD_2_;
217 
218     // private constructor ---------------------------------------------------
219 
220     /**
221      * Constructor private to prevent initialization
222      */
223     ///CLOVER:OFF
BOCSU()224     private BOCSU()
225     {
226     }
227     ///CLOVER:ON
228 
229     // private methods -------------------------------------------------------
230 
231     /**
232      * Integer division and modulo with negative numerators
233      * yields negative modulo results and quotients that are one more than
234      * what we need here.
235      * @param number which operations are to be performed on
236      * @param factor the factor to use for division
237      * @return (result of division) << 32 | modulo
238      */
getNegDivMod(int number, int factor)239     private static final long getNegDivMod(int number, int factor)
240     {
241         int modulo = number % factor;
242         long result = number / factor;
243         if (modulo < 0) {
244             -- result;
245             modulo += factor;
246         }
247         return (result << 32) | modulo;
248     }
249 
250     /**
251      * Encode one difference value -0x10ffff..+0x10ffff in 1..4 bytes,
252      * preserving lexical order
253      * @param diff
254      * @param buffer byte buffer to append to
255      * @param offset to the byte buffer to start appending
256      * @return end offset where the appending stops
257      */
writeDiff(int diff, byte buffer[], int offset)258     private static final int writeDiff(int diff, byte buffer[], int offset)
259     {
260         if (diff >= SLOPE_REACH_NEG_1_) {
261             if (diff <= SLOPE_REACH_POS_1_) {
262                 buffer[offset ++] = (byte)(SLOPE_MIDDLE_ + diff);
263             }
264             else if (diff <= SLOPE_REACH_POS_2_) {
265                 buffer[offset ++] = (byte)(SLOPE_START_POS_2_
266                                            + (diff / SLOPE_TAIL_COUNT_));
267                 buffer[offset ++] = (byte)(SLOPE_MIN_ +
268                                            (diff % SLOPE_TAIL_COUNT_));
269             }
270             else if (diff <= SLOPE_REACH_POS_3_) {
271                 buffer[offset + 2] = (byte)(SLOPE_MIN_
272                                             + (diff % SLOPE_TAIL_COUNT_));
273                 diff /= SLOPE_TAIL_COUNT_;
274                 buffer[offset + 1] = (byte)(SLOPE_MIN_
275                                             + (diff % SLOPE_TAIL_COUNT_));
276                 buffer[offset] = (byte)(SLOPE_START_POS_3_
277                                         + (diff / SLOPE_TAIL_COUNT_));
278                 offset += 3;
279             }
280             else {
281                 buffer[offset + 3] = (byte)(SLOPE_MIN_
282                                             + diff % SLOPE_TAIL_COUNT_);
283                 diff /= SLOPE_TAIL_COUNT_;
284                 buffer[offset + 2] = (byte)(SLOPE_MIN_
285                                         + diff % SLOPE_TAIL_COUNT_);
286                 diff /= SLOPE_TAIL_COUNT_;
287                 buffer[offset + 1] = (byte)(SLOPE_MIN_
288                                             + diff % SLOPE_TAIL_COUNT_);
289                 buffer[offset] = (byte)SLOPE_MAX_;
290                 offset += 4;
291             }
292         }
293         else {
294             long division = getNegDivMod(diff, SLOPE_TAIL_COUNT_);
295             int modulo = (int)division;
296             if (diff >= SLOPE_REACH_NEG_2_) {
297                 diff = (int)(division >> 32);
298                 buffer[offset ++] = (byte)(SLOPE_START_NEG_2_ + diff);
299                 buffer[offset ++] = (byte)(SLOPE_MIN_ + modulo);
300             }
301             else if (diff >= SLOPE_REACH_NEG_3_) {
302                 buffer[offset + 2] = (byte)(SLOPE_MIN_ + modulo);
303                 diff = (int)(division >> 32);
304                 division = getNegDivMod(diff, SLOPE_TAIL_COUNT_);
305                 modulo = (int)division;
306                 diff = (int)(division >> 32);
307                 buffer[offset + 1] = (byte)(SLOPE_MIN_ + modulo);
308                 buffer[offset] = (byte)(SLOPE_START_NEG_3_ + diff);
309                 offset += 3;
310             }
311             else {
312                 buffer[offset + 3] = (byte)(SLOPE_MIN_ + modulo);
313                 diff = (int)(division >> 32);
314                 division = getNegDivMod(diff, SLOPE_TAIL_COUNT_);
315                 modulo = (int)division;
316                 diff = (int)(division >> 32);
317                 buffer[offset + 2] = (byte)(SLOPE_MIN_ + modulo);
318                 division = getNegDivMod(diff, SLOPE_TAIL_COUNT_);
319                 modulo = (int)division;
320                 buffer[offset + 1] = (byte)(SLOPE_MIN_ + modulo);
321                 buffer[offset] = SLOPE_MIN_;
322                 offset += 4;
323             }
324         }
325         return offset;
326     }
327 }
328