##// 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 bdiff.c - efficient binary diff extension for Mercurial
2 bdiff.c - efficient binary diff extension for Mercurial
3
3
4 Copyright 2005, 2006 Matt Mackall <mpm@selenic.com>
4 Copyright 2005, 2006 Matt Mackall <mpm@selenic.com>
5
5
6 This software may be used and distributed according to the terms of
6 This software may be used and distributed according to the terms of
7 the GNU General Public License, incorporated herein by reference.
7 the GNU General Public License, incorporated herein by reference.
8
8
9 Based roughly on Python difflib
9 Based roughly on Python difflib
10 */
10 */
11
11
12 #include <Python.h>
12 #include <Python.h>
13 #include <stdlib.h>
13 #include <stdlib.h>
14 #include <string.h>
14 #include <string.h>
15
15
16 #ifdef __hpux
16 #if defined __hpux || defined __SUNPRO_C || defined _AIX
17 #define inline
18 #endif
19
20 #ifdef __SUNPRO_C
21 # define inline
17 # define inline
22 #endif
18 #endif
23
19
24 #ifdef _WIN32
20 #ifdef _WIN32
25 #ifdef _MSC_VER
21 #ifdef _MSC_VER
26 #define inline __inline
22 #define inline __inline
27 typedef unsigned long uint32_t;
23 typedef unsigned long uint32_t;
28 #else
24 #else
29 #include <stdint.h>
25 #include <stdint.h>
30 #endif
26 #endif
31 static uint32_t htonl(uint32_t x)
27 static uint32_t htonl(uint32_t x)
32 {
28 {
33 return ((x & 0x000000ffUL) << 24) |
29 return ((x & 0x000000ffUL) << 24) |
34 ((x & 0x0000ff00UL) << 8) |
30 ((x & 0x0000ff00UL) << 8) |
35 ((x & 0x00ff0000UL) >> 8) |
31 ((x & 0x00ff0000UL) >> 8) |
36 ((x & 0xff000000UL) >> 24);
32 ((x & 0xff000000UL) >> 24);
37 }
33 }
38 #else
34 #else
39 #include <sys/types.h>
35 #include <sys/types.h>
40 #include <arpa/inet.h>
36 #include <arpa/inet.h>
41 #include <inttypes.h>
37 #include <inttypes.h>
42 #endif
38 #endif
43
39
44 struct line {
40 struct line {
45 int h, len, n, e;
41 int h, len, n, e;
46 const char *l;
42 const char *l;
47 };
43 };
48
44
49 struct pos {
45 struct pos {
50 int pos, len;
46 int pos, len;
51 };
47 };
52
48
53 struct hunk {
49 struct hunk {
54 int a1, a2, b1, b2;
50 int a1, a2, b1, b2;
55 };
51 };
56
52
57 struct hunklist {
53 struct hunklist {
58 struct hunk *base, *head;
54 struct hunk *base, *head;
59 };
55 };
60
56
61 static inline uint32_t rol32(uint32_t word, unsigned int shift)
57 static inline uint32_t rol32(uint32_t word, unsigned int shift)
62 {
58 {
63 return (word << shift) | (word >> (32 - shift));
59 return (word << shift) | (word >> (32 - shift));
64 }
60 }
65
61
66 int splitlines(const char *a, int len, struct line **lr)
62 int splitlines(const char *a, int len, struct line **lr)
67 {
63 {
68 int g, h, i;
64 int g, h, i;
69 const char *p, *b = a;
65 const char *p, *b = a;
70 struct line *l;
66 struct line *l;
71
67
72 /* count the lines */
68 /* count the lines */
73 i = 1; /* extra line for sentinel */
69 i = 1; /* extra line for sentinel */
74 for (p = a; p < a + len; p++)
70 for (p = a; p < a + len; p++)
75 if (*p == '\n' || p == a + len - 1)
71 if (*p == '\n' || p == a + len - 1)
76 i++;
72 i++;
77
73
78 *lr = l = (struct line *)malloc(sizeof(struct line) * i);
74 *lr = l = (struct line *)malloc(sizeof(struct line) * i);
79 if (!l)
75 if (!l)
80 return -1;
76 return -1;
81
77
82 /* build the line array and calculate hashes */
78 /* build the line array and calculate hashes */
83 h = 0;
79 h = 0;
84 for (p = a; p < a + len; p++) {
80 for (p = a; p < a + len; p++) {
85 /*
81 /*
86 * a simple hash from GNU diff, with better collision
82 * a simple hash from GNU diff, with better collision
87 * resistance from hashpjw. this slows down common
83 * resistance from hashpjw. this slows down common
88 * case by 10%, but speeds up worst case by 100x.
84 * case by 10%, but speeds up worst case by 100x.
89 */
85 */
90 h = *p + rol32(h, 7);
86 h = *p + rol32(h, 7);
91 if ((g = h & 0xf0000000)) {
87 if ((g = h & 0xf0000000)) {
92 h ^= g >> 24;
88 h ^= g >> 24;
93 h ^= g;
89 h ^= g;
94 }
90 }
95 if (*p == '\n' || p == a + len - 1) {
91 if (*p == '\n' || p == a + len - 1) {
96 l->len = p - b + 1;
92 l->len = p - b + 1;
97 l->h = h * l->len;
93 l->h = h * l->len;
98 l->l = b;
94 l->l = b;
99 l->n = -1;
95 l->n = -1;
100 l++;
96 l++;
101 b = p + 1;
97 b = p + 1;
102 h = 0;
98 h = 0;
103 }
99 }
104 }
100 }
105
101
106 /* set up a sentinel */
102 /* set up a sentinel */
107 l->h = l->len = 0;
103 l->h = l->len = 0;
108 l->l = a + len;
104 l->l = a + len;
109 return i - 1;
105 return i - 1;
110 }
106 }
111
107
112 int inline cmp(struct line *a, struct line *b)
108 int inline cmp(struct line *a, struct line *b)
113 {
109 {
114 return a->h != b->h || a->len != b->len || memcmp(a->l, b->l, a->len);
110 return a->h != b->h || a->len != b->len || memcmp(a->l, b->l, a->len);
115 }
111 }
116
112
117 static int equatelines(struct line *a, int an, struct line *b, int bn)
113 static int equatelines(struct line *a, int an, struct line *b, int bn)
118 {
114 {
119 int i, j, buckets = 1, t;
115 int i, j, buckets = 1, t;
120 struct pos *h;
116 struct pos *h;
121
117
122 /* build a hash table of the next highest power of 2 */
118 /* build a hash table of the next highest power of 2 */
123 while (buckets < bn + 1)
119 while (buckets < bn + 1)
124 buckets *= 2;
120 buckets *= 2;
125
121
126 h = (struct pos *)malloc(buckets * sizeof(struct pos));
122 h = (struct pos *)malloc(buckets * sizeof(struct pos));
127 buckets = buckets - 1;
123 buckets = buckets - 1;
128 if (!h)
124 if (!h)
129 return 0;
125 return 0;
130
126
131 /* clear the hash table */
127 /* clear the hash table */
132 for (i = 0; i <= buckets; i++) {
128 for (i = 0; i <= buckets; i++) {
133 h[i].pos = -1;
129 h[i].pos = -1;
134 h[i].len = 0;
130 h[i].len = 0;
135 }
131 }
136
132
137 /* add lines to the hash table chains */
133 /* add lines to the hash table chains */
138 for (i = bn - 1; i >= 0; i--) {
134 for (i = bn - 1; i >= 0; i--) {
139 /* find the equivalence class */
135 /* find the equivalence class */
140 for (j = b[i].h & buckets; h[j].pos != -1;
136 for (j = b[i].h & buckets; h[j].pos != -1;
141 j = (j + 1) & buckets)
137 j = (j + 1) & buckets)
142 if (!cmp(b + i, b + h[j].pos))
138 if (!cmp(b + i, b + h[j].pos))
143 break;
139 break;
144
140
145 /* add to the head of the equivalence class */
141 /* add to the head of the equivalence class */
146 b[i].n = h[j].pos;
142 b[i].n = h[j].pos;
147 b[i].e = j;
143 b[i].e = j;
148 h[j].pos = i;
144 h[j].pos = i;
149 h[j].len++; /* keep track of popularity */
145 h[j].len++; /* keep track of popularity */
150 }
146 }
151
147
152 /* compute popularity threshold */
148 /* compute popularity threshold */
153 t = (bn >= 200) ? bn / 100 : bn + 1;
149 t = (bn >= 200) ? bn / 100 : bn + 1;
154
150
155 /* match items in a to their equivalence class in b */
151 /* match items in a to their equivalence class in b */
156 for (i = 0; i < an; i++) {
152 for (i = 0; i < an; i++) {
157 /* find the equivalence class */
153 /* find the equivalence class */
158 for (j = a[i].h & buckets; h[j].pos != -1;
154 for (j = a[i].h & buckets; h[j].pos != -1;
159 j = (j + 1) & buckets)
155 j = (j + 1) & buckets)
160 if (!cmp(a + i, b + h[j].pos))
156 if (!cmp(a + i, b + h[j].pos))
161 break;
157 break;
162
158
163 a[i].e = j; /* use equivalence class for quick compare */
159 a[i].e = j; /* use equivalence class for quick compare */
164 if (h[j].len <= t)
160 if (h[j].len <= t)
165 a[i].n = h[j].pos; /* point to head of match list */
161 a[i].n = h[j].pos; /* point to head of match list */
166 else
162 else
167 a[i].n = -1; /* too popular */
163 a[i].n = -1; /* too popular */
168 }
164 }
169
165
170 /* discard hash tables */
166 /* discard hash tables */
171 free(h);
167 free(h);
172 return 1;
168 return 1;
173 }
169 }
174
170
175 static int longest_match(struct line *a, struct line *b, struct pos *pos,
171 static int longest_match(struct line *a, struct line *b, struct pos *pos,
176 int a1, int a2, int b1, int b2, int *omi, int *omj)
172 int a1, int a2, int b1, int b2, int *omi, int *omj)
177 {
173 {
178 int mi = a1, mj = b1, mk = 0, mb = 0, i, j, k;
174 int mi = a1, mj = b1, mk = 0, mb = 0, i, j, k;
179
175
180 for (i = a1; i < a2; i++) {
176 for (i = a1; i < a2; i++) {
181 /* skip things before the current block */
177 /* skip things before the current block */
182 for (j = a[i].n; j != -1 && j < b1; j = b[j].n)
178 for (j = a[i].n; j != -1 && j < b1; j = b[j].n)
183 ;
179 ;
184
180
185 /* loop through all lines match a[i] in b */
181 /* loop through all lines match a[i] in b */
186 for (; j != -1 && j < b2; j = b[j].n) {
182 for (; j != -1 && j < b2; j = b[j].n) {
187 /* does this extend an earlier match? */
183 /* does this extend an earlier match? */
188 if (i > a1 && j > b1 && pos[j - 1].pos == i - 1)
184 if (i > a1 && j > b1 && pos[j - 1].pos == i - 1)
189 k = pos[j - 1].len + 1;
185 k = pos[j - 1].len + 1;
190 else
186 else
191 k = 1;
187 k = 1;
192 pos[j].pos = i;
188 pos[j].pos = i;
193 pos[j].len = k;
189 pos[j].len = k;
194
190
195 /* best match so far? */
191 /* best match so far? */
196 if (k > mk) {
192 if (k > mk) {
197 mi = i;
193 mi = i;
198 mj = j;
194 mj = j;
199 mk = k;
195 mk = k;
200 }
196 }
201 }
197 }
202 }
198 }
203
199
204 if (mk) {
200 if (mk) {
205 mi = mi - mk + 1;
201 mi = mi - mk + 1;
206 mj = mj - mk + 1;
202 mj = mj - mk + 1;
207 }
203 }
208
204
209 /* expand match to include neighboring popular lines */
205 /* expand match to include neighboring popular lines */
210 while (mi - mb > a1 && mj - mb > b1 &&
206 while (mi - mb > a1 && mj - mb > b1 &&
211 a[mi - mb - 1].e == b[mj - mb - 1].e)
207 a[mi - mb - 1].e == b[mj - mb - 1].e)
212 mb++;
208 mb++;
213 while (mi + mk < a2 && mj + mk < b2 &&
209 while (mi + mk < a2 && mj + mk < b2 &&
214 a[mi + mk].e == b[mj + mk].e)
210 a[mi + mk].e == b[mj + mk].e)
215 mk++;
211 mk++;
216
212
217 *omi = mi - mb;
213 *omi = mi - mb;
218 *omj = mj - mb;
214 *omj = mj - mb;
219 return mk + mb;
215 return mk + mb;
220 }
216 }
221
217
222 static void recurse(struct line *a, struct line *b, struct pos *pos,
218 static void recurse(struct line *a, struct line *b, struct pos *pos,
223 int a1, int a2, int b1, int b2, struct hunklist *l)
219 int a1, int a2, int b1, int b2, struct hunklist *l)
224 {
220 {
225 int i, j, k;
221 int i, j, k;
226
222
227 /* find the longest match in this chunk */
223 /* find the longest match in this chunk */
228 k = longest_match(a, b, pos, a1, a2, b1, b2, &i, &j);
224 k = longest_match(a, b, pos, a1, a2, b1, b2, &i, &j);
229 if (!k)
225 if (!k)
230 return;
226 return;
231
227
232 /* and recurse on the remaining chunks on either side */
228 /* and recurse on the remaining chunks on either side */
233 recurse(a, b, pos, a1, i, b1, j, l);
229 recurse(a, b, pos, a1, i, b1, j, l);
234 l->head->a1 = i;
230 l->head->a1 = i;
235 l->head->a2 = i + k;
231 l->head->a2 = i + k;
236 l->head->b1 = j;
232 l->head->b1 = j;
237 l->head->b2 = j + k;
233 l->head->b2 = j + k;
238 l->head++;
234 l->head++;
239 recurse(a, b, pos, i + k, a2, j + k, b2, l);
235 recurse(a, b, pos, i + k, a2, j + k, b2, l);
240 }
236 }
241
237
242 static struct hunklist diff(struct line *a, int an, struct line *b, int bn)
238 static struct hunklist diff(struct line *a, int an, struct line *b, int bn)
243 {
239 {
244 struct hunklist l;
240 struct hunklist l;
245 struct pos *pos;
241 struct pos *pos;
246 int t;
242 int t;
247
243
248 /* allocate and fill arrays */
244 /* allocate and fill arrays */
249 t = equatelines(a, an, b, bn);
245 t = equatelines(a, an, b, bn);
250 pos = (struct pos *)calloc(bn, sizeof(struct pos));
246 pos = (struct pos *)calloc(bn, sizeof(struct pos));
251 /* we can't have more matches than lines in the shorter file */
247 /* we can't have more matches than lines in the shorter file */
252 l.head = l.base = (struct hunk *)malloc(sizeof(struct hunk) *
248 l.head = l.base = (struct hunk *)malloc(sizeof(struct hunk) *
253 ((an<bn ? an:bn) + 1));
249 ((an<bn ? an:bn) + 1));
254
250
255 if (pos && l.base && t) {
251 if (pos && l.base && t) {
256 /* generate the matching block list */
252 /* generate the matching block list */
257 recurse(a, b, pos, 0, an, 0, bn, &l);
253 recurse(a, b, pos, 0, an, 0, bn, &l);
258 l.head->a1 = an;
254 l.head->a1 = an;
259 l.head->b1 = bn;
255 l.head->b1 = bn;
260 l.head++;
256 l.head++;
261 }
257 }
262
258
263 free(pos);
259 free(pos);
264 return l;
260 return l;
265 }
261 }
266
262
267 static PyObject *blocks(PyObject *self, PyObject *args)
263 static PyObject *blocks(PyObject *self, PyObject *args)
268 {
264 {
269 PyObject *sa, *sb, *rl = NULL, *m;
265 PyObject *sa, *sb, *rl = NULL, *m;
270 struct line *a, *b;
266 struct line *a, *b;
271 struct hunklist l = {NULL, NULL};
267 struct hunklist l = {NULL, NULL};
272 struct hunk *h;
268 struct hunk *h;
273 int an, bn, pos = 0;
269 int an, bn, pos = 0;
274
270
275 if (!PyArg_ParseTuple(args, "SS:bdiff", &sa, &sb))
271 if (!PyArg_ParseTuple(args, "SS:bdiff", &sa, &sb))
276 return NULL;
272 return NULL;
277
273
278 an = splitlines(PyString_AsString(sa), PyString_Size(sa), &a);
274 an = splitlines(PyString_AsString(sa), PyString_Size(sa), &a);
279 bn = splitlines(PyString_AsString(sb), PyString_Size(sb), &b);
275 bn = splitlines(PyString_AsString(sb), PyString_Size(sb), &b);
280 if (!a || !b)
276 if (!a || !b)
281 goto nomem;
277 goto nomem;
282
278
283 l = diff(a, an, b, bn);
279 l = diff(a, an, b, bn);
284 rl = PyList_New(l.head - l.base);
280 rl = PyList_New(l.head - l.base);
285 if (!l.head || !rl)
281 if (!l.head || !rl)
286 goto nomem;
282 goto nomem;
287
283
288 for (h = l.base; h != l.head; h++) {
284 for (h = l.base; h != l.head; h++) {
289 m = Py_BuildValue("iiii", h->a1, h->a2, h->b1, h->b2);
285 m = Py_BuildValue("iiii", h->a1, h->a2, h->b1, h->b2);
290 PyList_SetItem(rl, pos, m);
286 PyList_SetItem(rl, pos, m);
291 pos++;
287 pos++;
292 }
288 }
293
289
294 nomem:
290 nomem:
295 free(a);
291 free(a);
296 free(b);
292 free(b);
297 free(l.base);
293 free(l.base);
298 return rl ? rl : PyErr_NoMemory();
294 return rl ? rl : PyErr_NoMemory();
299 }
295 }
300
296
301 static PyObject *bdiff(PyObject *self, PyObject *args)
297 static PyObject *bdiff(PyObject *self, PyObject *args)
302 {
298 {
303 char *sa, *sb;
299 char *sa, *sb;
304 PyObject *result = NULL;
300 PyObject *result = NULL;
305 struct line *al, *bl;
301 struct line *al, *bl;
306 struct hunklist l = {NULL, NULL};
302 struct hunklist l = {NULL, NULL};
307 struct hunk *h;
303 struct hunk *h;
308 char encode[12], *rb;
304 char encode[12], *rb;
309 int an, bn, len = 0, la, lb;
305 int an, bn, len = 0, la, lb;
310
306
311 if (!PyArg_ParseTuple(args, "s#s#:bdiff", &sa, &la, &sb, &lb))
307 if (!PyArg_ParseTuple(args, "s#s#:bdiff", &sa, &la, &sb, &lb))
312 return NULL;
308 return NULL;
313
309
314 an = splitlines(sa, la, &al);
310 an = splitlines(sa, la, &al);
315 bn = splitlines(sb, lb, &bl);
311 bn = splitlines(sb, lb, &bl);
316 if (!al || !bl)
312 if (!al || !bl)
317 goto nomem;
313 goto nomem;
318
314
319 l = diff(al, an, bl, bn);
315 l = diff(al, an, bl, bn);
320 if (!l.head)
316 if (!l.head)
321 goto nomem;
317 goto nomem;
322
318
323 /* calculate length of output */
319 /* calculate length of output */
324 la = lb = 0;
320 la = lb = 0;
325 for (h = l.base; h != l.head; h++) {
321 for (h = l.base; h != l.head; h++) {
326 if (h->a1 != la || h->b1 != lb)
322 if (h->a1 != la || h->b1 != lb)
327 len += 12 + bl[h->b1].l - bl[lb].l;
323 len += 12 + bl[h->b1].l - bl[lb].l;
328 la = h->a2;
324 la = h->a2;
329 lb = h->b2;
325 lb = h->b2;
330 }
326 }
331
327
332 result = PyString_FromStringAndSize(NULL, len);
328 result = PyString_FromStringAndSize(NULL, len);
333 if (!result)
329 if (!result)
334 goto nomem;
330 goto nomem;
335
331
336 /* build binary patch */
332 /* build binary patch */
337 rb = PyString_AsString(result);
333 rb = PyString_AsString(result);
338 la = lb = 0;
334 la = lb = 0;
339
335
340 for (h = l.base; h != l.head; h++) {
336 for (h = l.base; h != l.head; h++) {
341 if (h->a1 != la || h->b1 != lb) {
337 if (h->a1 != la || h->b1 != lb) {
342 len = bl[h->b1].l - bl[lb].l;
338 len = bl[h->b1].l - bl[lb].l;
343 *(uint32_t *)(encode) = htonl(al[la].l - al->l);
339 *(uint32_t *)(encode) = htonl(al[la].l - al->l);
344 *(uint32_t *)(encode + 4) = htonl(al[h->a1].l - al->l);
340 *(uint32_t *)(encode + 4) = htonl(al[h->a1].l - al->l);
345 *(uint32_t *)(encode + 8) = htonl(len);
341 *(uint32_t *)(encode + 8) = htonl(len);
346 memcpy(rb, encode, 12);
342 memcpy(rb, encode, 12);
347 memcpy(rb + 12, bl[lb].l, len);
343 memcpy(rb + 12, bl[lb].l, len);
348 rb += 12 + len;
344 rb += 12 + len;
349 }
345 }
350 la = h->a2;
346 la = h->a2;
351 lb = h->b2;
347 lb = h->b2;
352 }
348 }
353
349
354 nomem:
350 nomem:
355 free(al);
351 free(al);
356 free(bl);
352 free(bl);
357 free(l.base);
353 free(l.base);
358 return result ? result : PyErr_NoMemory();
354 return result ? result : PyErr_NoMemory();
359 }
355 }
360
356
361 static char mdiff_doc[] = "Efficient binary diff.";
357 static char mdiff_doc[] = "Efficient binary diff.";
362
358
363 static PyMethodDef methods[] = {
359 static PyMethodDef methods[] = {
364 {"bdiff", bdiff, METH_VARARGS, "calculate a binary diff\n"},
360 {"bdiff", bdiff, METH_VARARGS, "calculate a binary diff\n"},
365 {"blocks", blocks, METH_VARARGS, "find a list of matching lines\n"},
361 {"blocks", blocks, METH_VARARGS, "find a list of matching lines\n"},
366 {NULL, NULL}
362 {NULL, NULL}
367 };
363 };
368
364
369 PyMODINIT_FUNC initbdiff(void)
365 PyMODINIT_FUNC initbdiff(void)
370 {
366 {
371 Py_InitModule3("bdiff", methods, mdiff_doc);
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