diffcore-delta.c: update the comment on the algorithm.
[git/git.git] / diffcore-delta.c
1 #include "cache.h"
2 #include "diff.h"
3 #include "diffcore.h"
4
5 /*
6 * Idea here is very simple.
7 *
8 * Almost all data we are interested in are text, but sometimes we have
9 * to deal with binary data. So we cut them into chunks delimited by
10 * LF byte, or 64-byte sequence, whichever comes first, and hash them.
11 *
12 * For those chunks, if the source buffer has more instances of it
13 * than the destination buffer, that means the difference are the
14 * number of bytes not copied from source to destination. If the
15 * counts are the same, everything was copied from source to
16 * destination. If the destination has more, everything was copied,
17 * and destination added more.
18 *
19 * We are doing an approximation so we do not really have to waste
20 * memory by actually storing the sequence. We just hash them into
21 * somewhere around 2^16 hashbuckets and count the occurrences.
22 */
23
24 /* Wild guess at the initial hash size */
25 #define INITIAL_HASH_SIZE 9
26
27 /* We leave more room in smaller hash but do not let it
28 * grow to have unused hole too much.
29 */
30 #define INITIAL_FREE(sz_log2) ((1<<(sz_log2))*(sz_log2-3)/(sz_log2))
31
32 /* A prime rather carefully chosen between 2^16..2^17, so that
33 * HASHBASE < INITIAL_FREE(17). We want to keep the maximum hashtable
34 * size under the current 2<<17 maximum, which can hold this many
35 * different values before overflowing to hashtable of size 2<<18.
36 */
37 #define HASHBASE 107927
38
39 struct spanhash {
40 unsigned int hashval;
41 unsigned int cnt;
42 };
43 struct spanhash_top {
44 int alloc_log2;
45 int free;
46 struct spanhash data[FLEX_ARRAY];
47 };
48
49 static struct spanhash *spanhash_find(struct spanhash_top *top,
50 unsigned int hashval)
51 {
52 int sz = 1 << top->alloc_log2;
53 int bucket = hashval & (sz - 1);
54 while (1) {
55 struct spanhash *h = &(top->data[bucket++]);
56 if (!h->cnt)
57 return NULL;
58 if (h->hashval == hashval)
59 return h;
60 if (sz <= bucket)
61 bucket = 0;
62 }
63 }
64
65 static struct spanhash_top *spanhash_rehash(struct spanhash_top *orig)
66 {
67 struct spanhash_top *new;
68 int i;
69 int osz = 1 << orig->alloc_log2;
70 int sz = osz << 1;
71
72 new = xmalloc(sizeof(*orig) + sizeof(struct spanhash) * sz);
73 new->alloc_log2 = orig->alloc_log2 + 1;
74 new->free = INITIAL_FREE(new->alloc_log2);
75 memset(new->data, 0, sizeof(struct spanhash) * sz);
76 for (i = 0; i < osz; i++) {
77 struct spanhash *o = &(orig->data[i]);
78 int bucket;
79 if (!o->cnt)
80 continue;
81 bucket = o->hashval & (sz - 1);
82 while (1) {
83 struct spanhash *h = &(new->data[bucket++]);
84 if (!h->cnt) {
85 h->hashval = o->hashval;
86 h->cnt = o->cnt;
87 new->free--;
88 break;
89 }
90 if (sz <= bucket)
91 bucket = 0;
92 }
93 }
94 free(orig);
95 return new;
96 }
97
98 static struct spanhash_top *add_spanhash(struct spanhash_top *top,
99 unsigned int hashval, int cnt)
100 {
101 int bucket, lim;
102 struct spanhash *h;
103
104 lim = (1 << top->alloc_log2);
105 bucket = hashval & (lim - 1);
106 while (1) {
107 h = &(top->data[bucket++]);
108 if (!h->cnt) {
109 h->hashval = hashval;
110 h->cnt = cnt;
111 top->free--;
112 if (top->free < 0)
113 return spanhash_rehash(top);
114 return top;
115 }
116 if (h->hashval == hashval) {
117 h->cnt += cnt;
118 return top;
119 }
120 if (lim <= bucket)
121 bucket = 0;
122 }
123 }
124
125 static struct spanhash_top *hash_chars(unsigned char *buf, unsigned int sz)
126 {
127 int i, n;
128 unsigned int accum1, accum2, hashval;
129 struct spanhash_top *hash;
130
131 i = INITIAL_HASH_SIZE;
132 hash = xmalloc(sizeof(*hash) + sizeof(struct spanhash) * (1<<i));
133 hash->alloc_log2 = i;
134 hash->free = INITIAL_FREE(i);
135 memset(hash->data, 0, sizeof(struct spanhash) * (1<<i));
136
137 n = 0;
138 accum1 = accum2 = 0;
139 while (sz) {
140 unsigned int c = *buf++;
141 unsigned int old_1 = accum1;
142 sz--;
143 accum1 = (accum1 << 7) ^ (accum2 >> 25);
144 accum2 = (accum2 << 7) ^ (old_1 >> 25);
145 accum1 += c;
146 if (++n < 64 && c != '\n')
147 continue;
148 hashval = (accum1 + accum2 * 0x61) % HASHBASE;
149 hash = add_spanhash(hash, hashval, n);
150 n = 0;
151 accum1 = accum2 = 0;
152 }
153 return hash;
154 }
155
156 int diffcore_count_changes(struct diff_filespec *src,
157 struct diff_filespec *dst,
158 void **src_count_p,
159 void **dst_count_p,
160 unsigned long delta_limit,
161 unsigned long *src_copied,
162 unsigned long *literal_added)
163 {
164 int i, ssz;
165 struct spanhash_top *src_count, *dst_count;
166 unsigned long sc, la;
167
168 src_count = dst_count = NULL;
169 if (src_count_p)
170 src_count = *src_count_p;
171 if (!src_count) {
172 src_count = hash_chars(src->data, src->size);
173 if (src_count_p)
174 *src_count_p = src_count;
175 }
176 if (dst_count_p)
177 dst_count = *dst_count_p;
178 if (!dst_count) {
179 dst_count = hash_chars(dst->data, dst->size);
180 if (dst_count_p)
181 *dst_count_p = dst_count;
182 }
183 sc = la = 0;
184
185 ssz = 1 << src_count->alloc_log2;
186 for (i = 0; i < ssz; i++) {
187 struct spanhash *s = &(src_count->data[i]);
188 struct spanhash *d;
189 unsigned dst_cnt, src_cnt;
190 if (!s->cnt)
191 continue;
192 src_cnt = s->cnt;
193 d = spanhash_find(dst_count, s->hashval);
194 dst_cnt = d ? d->cnt : 0;
195 if (src_cnt < dst_cnt) {
196 la += dst_cnt - src_cnt;
197 sc += src_cnt;
198 }
199 else
200 sc += dst_cnt;
201 }
202
203 if (!src_count_p)
204 free(src_count);
205 if (!dst_count_p)
206 free(dst_count);
207 *src_copied = sc;
208 *literal_added = la;
209 return 0;
210 }