##// END OF EJS Templates
add AIX to the list of compilers that don't have inline keyword
Benoit Boissinot -
r3561:8c617d48 default
parent child Browse files
Show More
@@ -1,373 +1,369 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 <Python.h>
13 13 #include <stdlib.h>
14 14 #include <string.h>
15 15
16 #ifdef __hpux
17 #define inline
18 #endif
19
20 #ifdef __SUNPRO_C
16 #if defined __hpux || defined __SUNPRO_C || defined _AIX
21 17 # define inline
22 18 #endif
23 19
24 20 #ifdef _WIN32
25 21 #ifdef _MSC_VER
26 22 #define inline __inline
27 23 typedef unsigned long uint32_t;
28 24 #else
29 25 #include <stdint.h>
30 26 #endif
31 27 static uint32_t htonl(uint32_t x)
32 28 {
33 29 return ((x & 0x000000ffUL) << 24) |
34 30 ((x & 0x0000ff00UL) << 8) |
35 31 ((x & 0x00ff0000UL) >> 8) |
36 32 ((x & 0xff000000UL) >> 24);
37 33 }
38 34 #else
39 35 #include <sys/types.h>
40 36 #include <arpa/inet.h>
41 37 #include <inttypes.h>
42 38 #endif
43 39
44 40 struct line {
45 41 int h, len, n, e;
46 42 const char *l;
47 43 };
48 44
49 45 struct pos {
50 46 int pos, len;
51 47 };
52 48
53 49 struct hunk {
54 50 int a1, a2, b1, b2;
55 51 };
56 52
57 53 struct hunklist {
58 54 struct hunk *base, *head;
59 55 };
60 56
61 57 static inline uint32_t rol32(uint32_t word, unsigned int shift)
62 58 {
63 59 return (word << shift) | (word >> (32 - shift));
64 60 }
65 61
66 62 int splitlines(const char *a, int len, struct line **lr)
67 63 {
68 64 int g, h, i;
69 65 const char *p, *b = a;
70 66 struct line *l;
71 67
72 68 /* count the lines */
73 69 i = 1; /* extra line for sentinel */
74 70 for (p = a; p < a + len; p++)
75 71 if (*p == '\n' || p == a + len - 1)
76 72 i++;
77 73
78 74 *lr = l = (struct line *)malloc(sizeof(struct line) * i);
79 75 if (!l)
80 76 return -1;
81 77
82 78 /* build the line array and calculate hashes */
83 79 h = 0;
84 80 for (p = a; p < a + len; p++) {
85 81 /*
86 82 * a simple hash from GNU diff, with better collision
87 83 * resistance from hashpjw. this slows down common
88 84 * case by 10%, but speeds up worst case by 100x.
89 85 */
90 86 h = *p + rol32(h, 7);
91 87 if ((g = h & 0xf0000000)) {
92 88 h ^= g >> 24;
93 89 h ^= g;
94 90 }
95 91 if (*p == '\n' || p == a + len - 1) {
96 92 l->len = p - b + 1;
97 93 l->h = h * l->len;
98 94 l->l = b;
99 95 l->n = -1;
100 96 l++;
101 97 b = p + 1;
102 98 h = 0;
103 99 }
104 100 }
105 101
106 102 /* set up a sentinel */
107 103 l->h = l->len = 0;
108 104 l->l = a + len;
109 105 return i - 1;
110 106 }
111 107
112 108 int inline cmp(struct line *a, struct line *b)
113 109 {
114 110 return a->h != b->h || a->len != b->len || memcmp(a->l, b->l, a->len);
115 111 }
116 112
117 113 static int equatelines(struct line *a, int an, struct line *b, int bn)
118 114 {
119 115 int i, j, buckets = 1, t;
120 116 struct pos *h;
121 117
122 118 /* build a hash table of the next highest power of 2 */
123 119 while (buckets < bn + 1)
124 120 buckets *= 2;
125 121
126 122 h = (struct pos *)malloc(buckets * sizeof(struct pos));
127 123 buckets = buckets - 1;
128 124 if (!h)
129 125 return 0;
130 126
131 127 /* clear the hash table */
132 128 for (i = 0; i <= buckets; i++) {
133 129 h[i].pos = -1;
134 130 h[i].len = 0;
135 131 }
136 132
137 133 /* add lines to the hash table chains */
138 134 for (i = bn - 1; i >= 0; i--) {
139 135 /* find the equivalence class */
140 136 for (j = b[i].h & buckets; h[j].pos != -1;
141 137 j = (j + 1) & buckets)
142 138 if (!cmp(b + i, b + h[j].pos))
143 139 break;
144 140
145 141 /* add to the head of the equivalence class */
146 142 b[i].n = h[j].pos;
147 143 b[i].e = j;
148 144 h[j].pos = i;
149 145 h[j].len++; /* keep track of popularity */
150 146 }
151 147
152 148 /* compute popularity threshold */
153 149 t = (bn >= 200) ? bn / 100 : bn + 1;
154 150
155 151 /* match items in a to their equivalence class in b */
156 152 for (i = 0; i < an; i++) {
157 153 /* find the equivalence class */
158 154 for (j = a[i].h & buckets; h[j].pos != -1;
159 155 j = (j + 1) & buckets)
160 156 if (!cmp(a + i, b + h[j].pos))
161 157 break;
162 158
163 159 a[i].e = j; /* use equivalence class for quick compare */
164 160 if (h[j].len <= t)
165 161 a[i].n = h[j].pos; /* point to head of match list */
166 162 else
167 163 a[i].n = -1; /* too popular */
168 164 }
169 165
170 166 /* discard hash tables */
171 167 free(h);
172 168 return 1;
173 169 }
174 170
175 171 static int longest_match(struct line *a, struct line *b, struct pos *pos,
176 172 int a1, int a2, int b1, int b2, int *omi, int *omj)
177 173 {
178 174 int mi = a1, mj = b1, mk = 0, mb = 0, i, j, k;
179 175
180 176 for (i = a1; i < a2; i++) {
181 177 /* skip things before the current block */
182 178 for (j = a[i].n; j != -1 && j < b1; j = b[j].n)
183 179 ;
184 180
185 181 /* loop through all lines match a[i] in b */
186 182 for (; j != -1 && j < b2; j = b[j].n) {
187 183 /* does this extend an earlier match? */
188 184 if (i > a1 && j > b1 && pos[j - 1].pos == i - 1)
189 185 k = pos[j - 1].len + 1;
190 186 else
191 187 k = 1;
192 188 pos[j].pos = i;
193 189 pos[j].len = k;
194 190
195 191 /* best match so far? */
196 192 if (k > mk) {
197 193 mi = i;
198 194 mj = j;
199 195 mk = k;
200 196 }
201 197 }
202 198 }
203 199
204 200 if (mk) {
205 201 mi = mi - mk + 1;
206 202 mj = mj - mk + 1;
207 203 }
208 204
209 205 /* expand match to include neighboring popular lines */
210 206 while (mi - mb > a1 && mj - mb > b1 &&
211 207 a[mi - mb - 1].e == b[mj - mb - 1].e)
212 208 mb++;
213 209 while (mi + mk < a2 && mj + mk < b2 &&
214 210 a[mi + mk].e == b[mj + mk].e)
215 211 mk++;
216 212
217 213 *omi = mi - mb;
218 214 *omj = mj - mb;
219 215 return mk + mb;
220 216 }
221 217
222 218 static void recurse(struct line *a, struct line *b, struct pos *pos,
223 219 int a1, int a2, int b1, int b2, struct hunklist *l)
224 220 {
225 221 int i, j, k;
226 222
227 223 /* find the longest match in this chunk */
228 224 k = longest_match(a, b, pos, a1, a2, b1, b2, &i, &j);
229 225 if (!k)
230 226 return;
231 227
232 228 /* and recurse on the remaining chunks on either side */
233 229 recurse(a, b, pos, a1, i, b1, j, l);
234 230 l->head->a1 = i;
235 231 l->head->a2 = i + k;
236 232 l->head->b1 = j;
237 233 l->head->b2 = j + k;
238 234 l->head++;
239 235 recurse(a, b, pos, i + k, a2, j + k, b2, l);
240 236 }
241 237
242 238 static struct hunklist diff(struct line *a, int an, struct line *b, int bn)
243 239 {
244 240 struct hunklist l;
245 241 struct pos *pos;
246 242 int t;
247 243
248 244 /* allocate and fill arrays */
249 245 t = equatelines(a, an, b, bn);
250 246 pos = (struct pos *)calloc(bn, sizeof(struct pos));
251 247 /* we can't have more matches than lines in the shorter file */
252 248 l.head = l.base = (struct hunk *)malloc(sizeof(struct hunk) *
253 249 ((an<bn ? an:bn) + 1));
254 250
255 251 if (pos && l.base && t) {
256 252 /* generate the matching block list */
257 253 recurse(a, b, pos, 0, an, 0, bn, &l);
258 254 l.head->a1 = an;
259 255 l.head->b1 = bn;
260 256 l.head++;
261 257 }
262 258
263 259 free(pos);
264 260 return l;
265 261 }
266 262
267 263 static PyObject *blocks(PyObject *self, PyObject *args)
268 264 {
269 265 PyObject *sa, *sb, *rl = NULL, *m;
270 266 struct line *a, *b;
271 267 struct hunklist l = {NULL, NULL};
272 268 struct hunk *h;
273 269 int an, bn, pos = 0;
274 270
275 271 if (!PyArg_ParseTuple(args, "SS:bdiff", &sa, &sb))
276 272 return NULL;
277 273
278 274 an = splitlines(PyString_AsString(sa), PyString_Size(sa), &a);
279 275 bn = splitlines(PyString_AsString(sb), PyString_Size(sb), &b);
280 276 if (!a || !b)
281 277 goto nomem;
282 278
283 279 l = diff(a, an, b, bn);
284 280 rl = PyList_New(l.head - l.base);
285 281 if (!l.head || !rl)
286 282 goto nomem;
287 283
288 284 for (h = l.base; h != l.head; h++) {
289 285 m = Py_BuildValue("iiii", h->a1, h->a2, h->b1, h->b2);
290 286 PyList_SetItem(rl, pos, m);
291 287 pos++;
292 288 }
293 289
294 290 nomem:
295 291 free(a);
296 292 free(b);
297 293 free(l.base);
298 294 return rl ? rl : PyErr_NoMemory();
299 295 }
300 296
301 297 static PyObject *bdiff(PyObject *self, PyObject *args)
302 298 {
303 299 char *sa, *sb;
304 300 PyObject *result = NULL;
305 301 struct line *al, *bl;
306 302 struct hunklist l = {NULL, NULL};
307 303 struct hunk *h;
308 304 char encode[12], *rb;
309 305 int an, bn, len = 0, la, lb;
310 306
311 307 if (!PyArg_ParseTuple(args, "s#s#:bdiff", &sa, &la, &sb, &lb))
312 308 return NULL;
313 309
314 310 an = splitlines(sa, la, &al);
315 311 bn = splitlines(sb, lb, &bl);
316 312 if (!al || !bl)
317 313 goto nomem;
318 314
319 315 l = diff(al, an, bl, bn);
320 316 if (!l.head)
321 317 goto nomem;
322 318
323 319 /* calculate length of output */
324 320 la = lb = 0;
325 321 for (h = l.base; h != l.head; h++) {
326 322 if (h->a1 != la || h->b1 != lb)
327 323 len += 12 + bl[h->b1].l - bl[lb].l;
328 324 la = h->a2;
329 325 lb = h->b2;
330 326 }
331 327
332 328 result = PyString_FromStringAndSize(NULL, len);
333 329 if (!result)
334 330 goto nomem;
335 331
336 332 /* build binary patch */
337 333 rb = PyString_AsString(result);
338 334 la = lb = 0;
339 335
340 336 for (h = l.base; h != l.head; h++) {
341 337 if (h->a1 != la || h->b1 != lb) {
342 338 len = bl[h->b1].l - bl[lb].l;
343 339 *(uint32_t *)(encode) = htonl(al[la].l - al->l);
344 340 *(uint32_t *)(encode + 4) = htonl(al[h->a1].l - al->l);
345 341 *(uint32_t *)(encode + 8) = htonl(len);
346 342 memcpy(rb, encode, 12);
347 343 memcpy(rb + 12, bl[lb].l, len);
348 344 rb += 12 + len;
349 345 }
350 346 la = h->a2;
351 347 lb = h->b2;
352 348 }
353 349
354 350 nomem:
355 351 free(al);
356 352 free(bl);
357 353 free(l.base);
358 354 return result ? result : PyErr_NoMemory();
359 355 }
360 356
361 357 static char mdiff_doc[] = "Efficient binary diff.";
362 358
363 359 static PyMethodDef methods[] = {
364 360 {"bdiff", bdiff, METH_VARARGS, "calculate a binary diff\n"},
365 361 {"blocks", blocks, METH_VARARGS, "find a list of matching lines\n"},
366 362 {NULL, NULL}
367 363 };
368 364
369 365 PyMODINIT_FUNC initbdiff(void)
370 366 {
371 367 Py_InitModule3("bdiff", methods, mdiff_doc);
372 368 }
373 369
General Comments 0
You need to be logged in to leave comments. Login now