##// END OF EJS Templates
bdiff: re-wrap lines per clang-format...
Augie Fackler -
r34631:38463bd9 default
parent child Browse files
Show More
@@ -1,321 +1,322 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 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 return a->hash != b->hash || a->len != b->len || memcmp(a->l, b->l, a->len);
82 return a->hash != b->hash || a->len != b->len ||
83 memcmp(a->l, b->l, a->len);
83 84 }
84 85
85 86 static int equatelines(struct bdiff_line *a, int an, struct bdiff_line *b,
86 87 int bn)
87 88 {
88 89 int i, j, buckets = 1, t, scale;
89 90 struct pos *h = NULL;
90 91
91 92 /* build a hash table of the next highest power of 2 */
92 93 while (buckets < bn + 1)
93 94 buckets *= 2;
94 95
95 96 /* try to allocate a large hash table to avoid collisions */
96 97 for (scale = 4; scale; scale /= 2) {
97 98 h = (struct pos *)malloc(scale * buckets * sizeof(struct pos));
98 99 if (h)
99 100 break;
100 101 }
101 102
102 103 if (!h)
103 104 return 0;
104 105
105 106 buckets = buckets * scale - 1;
106 107
107 108 /* clear the hash table */
108 109 for (i = 0; i <= buckets; i++) {
109 110 h[i].pos = -1;
110 111 h[i].len = 0;
111 112 }
112 113
113 114 /* add lines to the hash table chains */
114 115 for (i = 0; i < bn; i++) {
115 116 /* find the equivalence class */
116 117 for (j = b[i].hash & buckets; h[j].pos != -1;
117 118 j = (j + 1) & buckets)
118 119 if (!cmp(b + i, b + h[j].pos))
119 120 break;
120 121
121 122 /* add to the head of the equivalence class */
122 123 b[i].n = h[j].pos;
123 124 b[i].e = j;
124 125 h[j].pos = i;
125 126 h[j].len++; /* keep track of popularity */
126 127 }
127 128
128 129 /* compute popularity threshold */
129 130 t = (bn >= 31000) ? bn / 1000 : 1000000 / (bn + 1);
130 131
131 132 /* match items in a to their equivalence class in b */
132 133 for (i = 0; i < an; i++) {
133 134 /* find the equivalence class */
134 135 for (j = a[i].hash & buckets; h[j].pos != -1;
135 136 j = (j + 1) & buckets)
136 137 if (!cmp(a + i, b + h[j].pos))
137 138 break;
138 139
139 140 a[i].e = j; /* use equivalence class for quick compare */
140 141 if (h[j].len <= t)
141 142 a[i].n = h[j].pos; /* point to head of match list */
142 143 else
143 144 a[i].n = -1; /* too popular */
144 145 }
145 146
146 147 /* discard hash tables */
147 148 free(h);
148 149 return 1;
149 150 }
150 151
151 152 static int longest_match(struct bdiff_line *a, struct bdiff_line *b,
152 153 struct pos *pos,
153 154 int a1, int a2, int b1, int b2, int *omi, int *omj)
154 155 {
155 156 int mi = a1, mj = b1, mk = 0, i, j, k, half, bhalf;
156 157
157 158 /* window our search on large regions to better bound
158 159 worst-case performance. by choosing a window at the end, we
159 160 reduce skipping overhead on the b chains. */
160 161 if (a2 - a1 > 30000)
161 162 a1 = a2 - 30000;
162 163
163 164 half = (a1 + a2 - 1) / 2;
164 165 bhalf = (b1 + b2 - 1) / 2;
165 166
166 167 for (i = a1; i < a2; i++) {
167 168 /* skip all lines in b after the current block */
168 169 for (j = a[i].n; j >= b2; j = b[j].n)
169 170 ;
170 171
171 172 /* loop through all lines match a[i] in b */
172 173 for (; j >= b1; j = b[j].n) {
173 174 /* does this extend an earlier match? */
174 175 for (k = 1; j - k >= b1 && i - k >= a1; k++) {
175 176 /* reached an earlier match? */
176 177 if (pos[j - k].pos == i - k) {
177 178 k += pos[j - k].len;
178 179 break;
179 180 }
180 181 /* previous line mismatch? */
181 182 if (a[i - k].e != b[j - k].e)
182 183 break;
183 184 }
184 185
185 186 pos[j].pos = i;
186 187 pos[j].len = k;
187 188
188 189 /* best match so far? we prefer matches closer
189 190 to the middle to balance recursion */
190 191 if (k > mk) {
191 192 /* a longer match */
192 193 mi = i;
193 194 mj = j;
194 195 mk = k;
195 196 } else if (k == mk) {
196 197 if (i > mi && i <= half && j > b1) {
197 198 /* same match but closer to half */
198 199 mi = i;
199 200 mj = j;
200 201 } else if (i == mi && (mj > bhalf || i == a1)) {
201 202 /* same i but best earlier j */
202 203 mj = j;
203 204 }
204 205 }
205 206 }
206 207 }
207 208
208 209 if (mk) {
209 210 mi = mi - mk + 1;
210 211 mj = mj - mk + 1;
211 212 }
212 213
213 214 /* expand match to include subsequent popular lines */
214 while (mi + mk < a2 && mj + mk < b2 &&
215 a[mi + mk].e == b[mj + mk].e)
215 while (mi + mk < a2 && mj + mk < b2 && 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 l->next = (struct bdiff_hunk *)malloc(sizeof(struct bdiff_hunk));
241 l->next =
242 (struct bdiff_hunk *)malloc(sizeof(struct bdiff_hunk));
242 243 if (!l->next)
243 244 return NULL;
244 245
245 246 l = l->next;
246 247 l->a1 = i;
247 248 l->a2 = i + k;
248 249 l->b1 = j;
249 250 l->b2 = j + k;
250 251 l->next = NULL;
251 252
252 253 /* tail-recursion didn't happen, so do equivalent iteration */
253 254 a1 = i + k;
254 255 b1 = j + k;
255 256 }
256 257 }
257 258
258 259 int bdiff_diff(struct bdiff_line *a, int an, struct bdiff_line *b,
259 260 int bn, struct bdiff_hunk *base)
260 261 {
261 262 struct bdiff_hunk *curr;
262 263 struct pos *pos;
263 264 int t, count = 0;
264 265
265 266 /* allocate and fill arrays */
266 267 t = equatelines(a, an, b, bn);
267 268 pos = (struct pos *)calloc(bn ? bn : 1, sizeof(struct pos));
268 269
269 270 if (pos && t) {
270 271 /* generate the matching block list */
271 272
272 273 curr = recurse(a, b, pos, 0, an, 0, bn, base);
273 274 if (!curr)
274 275 return -1;
275 276
276 277 /* sentinel end hunk */
277 curr->next = (struct bdiff_hunk *)malloc(sizeof(struct bdiff_hunk));
278 curr->next =
279 (struct bdiff_hunk *)malloc(sizeof(struct bdiff_hunk));
278 280 if (!curr->next)
279 281 return -1;
280 282 curr = curr->next;
281 283 curr->a1 = curr->a2 = an;
282 284 curr->b1 = curr->b2 = bn;
283 285 curr->next = NULL;
284 286 }
285 287
286 288 free(pos);
287 289
288 290 /* normalize the hunk list, try to push each hunk towards the end */
289 291 for (curr = base->next; curr; curr = curr->next) {
290 292 struct bdiff_hunk *next = curr->next;
291 293
292 294 if (!next)
293 295 break;
294 296
295 297 if (curr->a2 == next->a1 || curr->b2 == next->b1)
296 while (curr->a2 < an && curr->b2 < bn
297 && next->a1 < next->a2
298 && next->b1 < next->b2
299 && !cmp(a + curr->a2, b + curr->b2)) {
298 while (curr->a2 < an && curr->b2 < bn &&
299 next->a1 < next->a2 && next->b1 < next->b2 &&
300 !cmp(a + curr->a2, b + curr->b2)) {
300 301 curr->a2++;
301 302 next->a1++;
302 303 curr->b2++;
303 304 next->b1++;
304 305 }
305 306 }
306 307
307 308 for (curr = base->next; curr; curr = curr->next)
308 309 count++;
309 310 return count;
310 311 }
311 312
312 313 void bdiff_freehunks(struct bdiff_hunk *l)
313 314 {
314 315 struct bdiff_hunk *n;
315 316 for (; l; l = n) {
316 317 n = l->next;
317 318 free(l);
318 319 }
319 320 }
320 321
321 322
General Comments 0
You need to be logged in to leave comments. Login now