2 * diff-delta.c: generate a delta between two buffers
4 * Many parts of this file have been lifted from LibXDiff version 0.10.
5 * http://www.xmailserver.org/xdiff-lib.html
7 * LibXDiff was written by Davide Libenzi <davidel@xmailserver.org>
8 * Copyright (C) 2003 Davide Libenzi
10 * Many mods for GIT usage by Nicolas Pitre <nico@cam.org>, (C) 2005.
12 * This file is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2.1 of the License, or (at your option) any later version.
17 * Use of this within git automatically means that the LGPL
18 * licensing gets turned into GPLv2 within this project.
27 const unsigned char *ptr;
31 static struct index ** delta_index(const unsigned char *buf,
32 unsigned long bufsize,
33 unsigned long trg_bufsize)
36 unsigned int i, hshift, hlimit, *hash_count;
37 const unsigned char *data;
38 struct index *entry, **hash;
41 /* determine index hash size */
43 for (i = 8; (1 << i) < hsize && i < 24; i += 2);
48 * Allocate lookup index. Note the first hash pointer
49 * is used to store the hash shift value.
51 mem = malloc((1 + hsize) * sizeof(*hash) + bufsize * sizeof(*entry));
55 *hash++ = (void *)hshift;
56 entry = mem + (1 + hsize) * sizeof(*hash);
57 memset(hash, 0, hsize * sizeof(*hash));
59 /* allocate an array to count hash entries */
60 hash_count = calloc(hsize, sizeof(*hash_count));
66 /* then populate the index */
67 data = buf + bufsize - 2;
70 i = data[0] ^ ((data[1] ^ (data[2] << hshift)) << hshift);
71 entry->next = hash[i];
77 * Determine a limit on the number of entries in the same hash
78 * bucket. This guard us against patological data sets causing
79 * really bad hash distribution with most entries in the same hash
80 * bucket that would bring us to O(m*n) computing costs (m and n
81 * corresponding to reference and target buffer sizes).
83 * The more the target buffer is large, the more it is important to
84 * have small entry lists for each hash buckets. With such a limit
85 * the cost is bounded to something more like O(m+n).
87 hlimit = (1 << 26) / trg_bufsize;
92 * Now make sure none of the hash buckets has more entries than
93 * we're willing to test. Otherwise we cull the entry list to
94 * limit identical three byte prefixes to still preserve a good
95 * repartition across the reference buffer.
97 for (i = 0; i < hsize; i++) {
98 struct index **list, *bucket, *remaining;
100 if (hash_count[i] < hlimit)
107 while (cnt < hlimit && remaining) {
108 struct index *this = remaining, *that;
109 remaining = remaining->next;
110 for (that = bucket; that; that = that->next) {
111 if (!memcmp(that->ptr, this->ptr, 3))
115 continue; /* discard */
118 list = &(this->next);
128 /* provide the size of the copy opcode given the block offset and size */
129 #define COPYOP_SIZE(o, s) \
130 (!!(o & 0xff) + !!(o & 0xff00) + !!(o & 0xff0000) + !!(o & 0xff000000) + \
131 !!(s & 0xff) + !!(s & 0xff00) + 1)
133 /* the maximum size for any opcode */
134 #define MAX_OP_SIZE COPYOP_SIZE(0xffffffff, 0xffffffff)
136 void *diff_delta(void *from_buf, unsigned long from_size,
137 void *to_buf, unsigned long to_size,
138 unsigned long *delta_size,
139 unsigned long max_size,
142 unsigned int i, outpos, outsize, inscnt, hash_shift;
143 const unsigned char *ref_data, *ref_top, *data, *top;
145 struct index *entry, **hash;
147 if (!from_size || !to_size)
149 if (from_index && *from_index) {
152 hash = delta_index(from_buf, from_size, to_size);
158 hash_shift = (unsigned int)(*hash++);
162 if (max_size && outsize >= max_size)
163 outsize = max_size + MAX_OP_SIZE + 1;
164 out = malloc(outsize);
172 ref_top = from_buf + from_size;
174 top = to_buf + to_size;
176 /* store reference buffer size */
177 out[outpos++] = from_size;
180 out[outpos - 1] |= 0x80;
181 out[outpos++] = from_size;
185 /* store target buffer size */
186 out[outpos++] = to_size;
189 out[outpos - 1] |= 0x80;
190 out[outpos++] = to_size;
197 unsigned int moff = 0, msize = 0;
198 if (data + 3 <= top) {
199 i = data[0] ^ ((data[1] ^ (data[2] << hash_shift)) << hash_shift);
200 for (entry = hash[i]; entry; entry = entry->next) {
201 const unsigned char *ref = entry->ptr;
202 const unsigned char *src = data;
203 unsigned int ref_size = ref_top - ref;
204 if (ref_size > top - src)
205 ref_size = top - src;
206 if (ref_size > 0x10000)
208 if (ref_size <= msize)
212 while (ref_size-- && *++src == *++ref);
213 if (msize < ref - entry->ptr) {
214 /* this is our best match so far */
215 msize = ref - entry->ptr;
216 moff = entry->ptr - ref_data;
221 if (!msize || msize < COPYOP_SIZE(moff, msize)) {
224 out[outpos++] = *data++;
226 if (inscnt == 0x7f) {
227 out[outpos - inscnt - 1] = inscnt;
234 out[outpos - inscnt - 1] = inscnt;
242 if (moff & 0xff) { out[outpos++] = moff; i |= 0x01; }
244 if (moff & 0xff) { out[outpos++] = moff; i |= 0x02; }
246 if (moff & 0xff) { out[outpos++] = moff; i |= 0x04; }
248 if (moff & 0xff) { out[outpos++] = moff; i |= 0x08; }
250 if (msize & 0xff) { out[outpos++] = msize; i |= 0x10; }
252 if (msize & 0xff) { out[outpos++] = msize; i |= 0x20; }
257 if (outpos >= outsize - MAX_OP_SIZE) {
259 outsize = outsize * 3 / 2;
260 if (max_size && outsize >= max_size)
261 outsize = max_size + MAX_OP_SIZE + 1;
262 if (max_size && outpos > max_size)
265 out = realloc(out, outsize);
276 out[outpos - inscnt - 1] = inscnt;
280 *delta_size = outpos;