##// END OF EJS Templates
bdiff: remove extra space after * per clang-format...
Augie Fackler -
r34630:2c79a5a0 default
parent child Browse files
Show More
@@ -1,321 +1,321 b''
1 1 /*
2 2 bdiff.c - efficient binary diff extension for Mercurial
3 3
4 4 Copyright 2005, 2006 Matt Mackall <mpm@selenic.com>
5 5
6 6 This software may be used and distributed according to the terms of
7 7 the GNU General Public License, incorporated herein by reference.
8 8
9 9 Based roughly on Python difflib
10 10 */
11 11
12 12 #include <limits.h>
13 13 #include <stdlib.h>
14 14 #include <string.h>
15 15
16 16 #include "bdiff.h"
17 17 #include "bitmanipulation.h"
18 18 #include "compat.h"
19 19
20 20 /* Hash implementation from diffutils */
21 21 #define ROL(v, n) ((v) << (n) | (v) >> (sizeof(v) * CHAR_BIT - (n)))
22 22 #define HASH(h, c) ((c) + ROL(h, 7))
23 23
24 24 struct pos {
25 25 int pos, len;
26 26 };
27 27
28 28 int bdiff_splitlines(const char *a, ssize_t len, struct bdiff_line **lr)
29 29 {
30 30 unsigned hash;
31 31 int i;
32 32 const char *p, *b = a;
33 const char * const plast = a + len - 1;
33 const char *const plast = a + len - 1;
34 34 struct bdiff_line *l;
35 35
36 36 /* count the lines */
37 37 i = 1; /* extra line for sentinel */
38 38 for (p = a; p < plast; p++)
39 39 if (*p == '\n')
40 40 i++;
41 41 if (p == plast)
42 42 i++;
43 43
44 44 *lr = l = (struct bdiff_line *)malloc(sizeof(struct bdiff_line) * i);
45 45 if (!l)
46 46 return -1;
47 47
48 48 /* build the line array and calculate hashes */
49 49 hash = 0;
50 50 for (p = a; p < plast; p++) {
51 51 hash = HASH(hash, *p);
52 52
53 53 if (*p == '\n') {
54 54 l->hash = hash;
55 55 hash = 0;
56 56 l->len = p - b + 1;
57 57 l->l = b;
58 58 l->n = INT_MAX;
59 59 l++;
60 60 b = p + 1;
61 61 }
62 62 }
63 63
64 64 if (p == plast) {
65 65 hash = HASH(hash, *p);
66 66 l->hash = hash;
67 67 l->len = p - b + 1;
68 68 l->l = b;
69 69 l->n = INT_MAX;
70 70 l++;
71 71 }
72 72
73 73 /* set up a sentinel */
74 74 l->hash = 0;
75 75 l->len = 0;
76 76 l->l = a + len;
77 77 return i - 1;
78 78 }
79 79
80 80 static inline int cmp(struct bdiff_line *a, struct bdiff_line *b)
81 81 {
82 82 return a->hash != b->hash || a->len != b->len || memcmp(a->l, b->l, a->len);
83 83 }
84 84
85 85 static int equatelines(struct bdiff_line *a, int an, struct bdiff_line *b,
86 86 int bn)
87 87 {
88 88 int i, j, buckets = 1, t, scale;
89 89 struct pos *h = NULL;
90 90
91 91 /* build a hash table of the next highest power of 2 */
92 92 while (buckets < bn + 1)
93 93 buckets *= 2;
94 94
95 95 /* try to allocate a large hash table to avoid collisions */
96 96 for (scale = 4; scale; scale /= 2) {
97 97 h = (struct pos *)malloc(scale * buckets * sizeof(struct pos));
98 98 if (h)
99 99 break;
100 100 }
101 101
102 102 if (!h)
103 103 return 0;
104 104
105 105 buckets = buckets * scale - 1;
106 106
107 107 /* clear the hash table */
108 108 for (i = 0; i <= buckets; i++) {
109 109 h[i].pos = -1;
110 110 h[i].len = 0;
111 111 }
112 112
113 113 /* add lines to the hash table chains */
114 114 for (i = 0; i < bn; i++) {
115 115 /* find the equivalence class */
116 116 for (j = b[i].hash & buckets; h[j].pos != -1;
117 117 j = (j + 1) & buckets)
118 118 if (!cmp(b + i, b + h[j].pos))
119 119 break;
120 120
121 121 /* add to the head of the equivalence class */
122 122 b[i].n = h[j].pos;
123 123 b[i].e = j;
124 124 h[j].pos = i;
125 125 h[j].len++; /* keep track of popularity */
126 126 }
127 127
128 128 /* compute popularity threshold */
129 129 t = (bn >= 31000) ? bn / 1000 : 1000000 / (bn + 1);
130 130
131 131 /* match items in a to their equivalence class in b */
132 132 for (i = 0; i < an; i++) {
133 133 /* find the equivalence class */
134 134 for (j = a[i].hash & buckets; h[j].pos != -1;
135 135 j = (j + 1) & buckets)
136 136 if (!cmp(a + i, b + h[j].pos))
137 137 break;
138 138
139 139 a[i].e = j; /* use equivalence class for quick compare */
140 140 if (h[j].len <= t)
141 141 a[i].n = h[j].pos; /* point to head of match list */
142 142 else
143 143 a[i].n = -1; /* too popular */
144 144 }
145 145
146 146 /* discard hash tables */
147 147 free(h);
148 148 return 1;
149 149 }
150 150
151 151 static int longest_match(struct bdiff_line *a, struct bdiff_line *b,
152 152 struct pos *pos,
153 153 int a1, int a2, int b1, int b2, int *omi, int *omj)
154 154 {
155 155 int mi = a1, mj = b1, mk = 0, i, j, k, half, bhalf;
156 156
157 157 /* window our search on large regions to better bound
158 158 worst-case performance. by choosing a window at the end, we
159 159 reduce skipping overhead on the b chains. */
160 160 if (a2 - a1 > 30000)
161 161 a1 = a2 - 30000;
162 162
163 163 half = (a1 + a2 - 1) / 2;
164 164 bhalf = (b1 + b2 - 1) / 2;
165 165
166 166 for (i = a1; i < a2; i++) {
167 167 /* skip all lines in b after the current block */
168 168 for (j = a[i].n; j >= b2; j = b[j].n)
169 169 ;
170 170
171 171 /* loop through all lines match a[i] in b */
172 172 for (; j >= b1; j = b[j].n) {
173 173 /* does this extend an earlier match? */
174 174 for (k = 1; j - k >= b1 && i - k >= a1; k++) {
175 175 /* reached an earlier match? */
176 176 if (pos[j - k].pos == i - k) {
177 177 k += pos[j - k].len;
178 178 break;
179 179 }
180 180 /* previous line mismatch? */
181 181 if (a[i - k].e != b[j - k].e)
182 182 break;
183 183 }
184 184
185 185 pos[j].pos = i;
186 186 pos[j].len = k;
187 187
188 188 /* best match so far? we prefer matches closer
189 189 to the middle to balance recursion */
190 190 if (k > mk) {
191 191 /* a longer match */
192 192 mi = i;
193 193 mj = j;
194 194 mk = k;
195 195 } else if (k == mk) {
196 196 if (i > mi && i <= half && j > b1) {
197 197 /* same match but closer to half */
198 198 mi = i;
199 199 mj = j;
200 200 } else if (i == mi && (mj > bhalf || i == a1)) {
201 201 /* same i but best earlier j */
202 202 mj = j;
203 203 }
204 204 }
205 205 }
206 206 }
207 207
208 208 if (mk) {
209 209 mi = mi - mk + 1;
210 210 mj = mj - mk + 1;
211 211 }
212 212
213 213 /* expand match to include subsequent popular lines */
214 214 while (mi + mk < a2 && mj + mk < b2 &&
215 215 a[mi + mk].e == b[mj + mk].e)
216 216 mk++;
217 217
218 218 *omi = mi;
219 219 *omj = mj;
220 220
221 221 return mk;
222 222 }
223 223
224 224 static struct bdiff_hunk *recurse(struct bdiff_line *a, struct bdiff_line *b,
225 225 struct pos *pos,
226 226 int a1, int a2, int b1, int b2, struct bdiff_hunk *l)
227 227 {
228 228 int i, j, k;
229 229
230 230 while (1) {
231 231 /* find the longest match in this chunk */
232 232 k = longest_match(a, b, pos, a1, a2, b1, b2, &i, &j);
233 233 if (!k)
234 234 return l;
235 235
236 236 /* and recurse on the remaining chunks on either side */
237 237 l = recurse(a, b, pos, a1, i, b1, j, l);
238 238 if (!l)
239 239 return NULL;
240 240
241 241 l->next = (struct bdiff_hunk *)malloc(sizeof(struct bdiff_hunk));
242 242 if (!l->next)
243 243 return NULL;
244 244
245 245 l = l->next;
246 246 l->a1 = i;
247 247 l->a2 = i + k;
248 248 l->b1 = j;
249 249 l->b2 = j + k;
250 250 l->next = NULL;
251 251
252 252 /* tail-recursion didn't happen, so do equivalent iteration */
253 253 a1 = i + k;
254 254 b1 = j + k;
255 255 }
256 256 }
257 257
258 258 int bdiff_diff(struct bdiff_line *a, int an, struct bdiff_line *b,
259 259 int bn, struct bdiff_hunk *base)
260 260 {
261 261 struct bdiff_hunk *curr;
262 262 struct pos *pos;
263 263 int t, count = 0;
264 264
265 265 /* allocate and fill arrays */
266 266 t = equatelines(a, an, b, bn);
267 267 pos = (struct pos *)calloc(bn ? bn : 1, sizeof(struct pos));
268 268
269 269 if (pos && t) {
270 270 /* generate the matching block list */
271 271
272 272 curr = recurse(a, b, pos, 0, an, 0, bn, base);
273 273 if (!curr)
274 274 return -1;
275 275
276 276 /* sentinel end hunk */
277 277 curr->next = (struct bdiff_hunk *)malloc(sizeof(struct bdiff_hunk));
278 278 if (!curr->next)
279 279 return -1;
280 280 curr = curr->next;
281 281 curr->a1 = curr->a2 = an;
282 282 curr->b1 = curr->b2 = bn;
283 283 curr->next = NULL;
284 284 }
285 285
286 286 free(pos);
287 287
288 288 /* normalize the hunk list, try to push each hunk towards the end */
289 289 for (curr = base->next; curr; curr = curr->next) {
290 290 struct bdiff_hunk *next = curr->next;
291 291
292 292 if (!next)
293 293 break;
294 294
295 295 if (curr->a2 == next->a1 || curr->b2 == next->b1)
296 296 while (curr->a2 < an && curr->b2 < bn
297 297 && next->a1 < next->a2
298 298 && next->b1 < next->b2
299 299 && !cmp(a + curr->a2, b + curr->b2)) {
300 300 curr->a2++;
301 301 next->a1++;
302 302 curr->b2++;
303 303 next->b1++;
304 304 }
305 305 }
306 306
307 307 for (curr = base->next; curr; curr = curr->next)
308 308 count++;
309 309 return count;
310 310 }
311 311
312 312 void bdiff_freehunks(struct bdiff_hunk *l)
313 313 {
314 314 struct bdiff_hunk *n;
315 315 for (; l; l = n) {
316 316 n = l->next;
317 317 free(l);
318 318 }
319 319 }
320 320
321 321
General Comments 0
You need to be logged in to leave comments. Login now