]>
Commit | Line | Data |
---|---|---|
427dcb4b JH |
1 | /* |
2 | * Copyright (C) 2005 Junio C Hamano | |
3 | */ | |
4 | #include "cache.h" | |
5 | #include "diff.h" | |
6 | #include "diffcore.h" | |
7 | #include "delta.h" | |
8 | ||
9 | struct diff_rename_pool { | |
10 | struct diff_filespec **s; | |
11 | int nr, alloc; | |
12 | }; | |
13 | ||
14 | static void diff_rename_pool_clear(struct diff_rename_pool *pool) | |
15 | { | |
16 | pool->s = NULL; pool->nr = pool->alloc = 0; | |
17 | } | |
18 | ||
19 | static void diff_rename_pool_add(struct diff_rename_pool *pool, | |
20 | struct diff_filespec *s) | |
21 | { | |
22 | if (S_ISDIR(s->mode)) | |
23 | return; /* rename/copy patch for tree does not make sense. */ | |
24 | ||
25 | if (pool->alloc <= pool->nr) { | |
26 | pool->alloc = alloc_nr(pool->alloc); | |
27 | pool->s = xrealloc(pool->s, | |
28 | sizeof(*(pool->s)) * pool->alloc); | |
29 | } | |
30 | pool->s[pool->nr] = s; | |
31 | pool->nr++; | |
32 | } | |
33 | ||
34 | static int is_exact_match(struct diff_filespec *src, struct diff_filespec *dst) | |
35 | { | |
36 | if (src->sha1_valid && dst->sha1_valid && | |
37 | !memcmp(src->sha1, dst->sha1, 20)) | |
38 | return 1; | |
39 | if (diff_populate_filespec(src) || diff_populate_filespec(dst)) | |
40 | /* this is an error but will be caught downstream */ | |
41 | return 0; | |
42 | if (src->size == dst->size && | |
43 | !memcmp(src->data, dst->data, src->size)) | |
44 | return 1; | |
45 | return 0; | |
46 | } | |
47 | ||
48 | struct diff_score { | |
49 | struct diff_filespec *src; | |
50 | struct diff_filespec *dst; | |
51 | int score; | |
52 | int rank; | |
53 | }; | |
54 | ||
55 | static int estimate_similarity(struct diff_filespec *src, | |
56 | struct diff_filespec *dst, | |
57 | int minimum_score) | |
58 | { | |
59 | /* src points at a file that existed in the original tree (or | |
60 | * optionally a file in the destination tree) and dst points | |
61 | * at a newly created file. They may be quite similar, in which | |
62 | * case we want to say src is renamed to dst or src is copied into | |
63 | * dst, and then some edit has been applied to dst. | |
64 | * | |
65 | * Compare them and return how similar they are, representing | |
66 | * the score as an integer between 0 and 10000, except | |
67 | * where they match exactly it is considered better than anything | |
68 | * else. | |
69 | */ | |
70 | void *delta; | |
71 | unsigned long delta_size; | |
72 | int score; | |
73 | ||
74 | delta_size = ((src->size < dst->size) ? | |
75 | (dst->size - src->size) : (src->size - dst->size)); | |
76 | ||
77 | /* We would not consider rename followed by more than | |
78 | * minimum_score/MAX_SCORE edits; that is, delta_size must be smaller | |
79 | * than (src->size + dst->size)/2 * minimum_score/MAX_SCORE, | |
80 | * which means... | |
81 | */ | |
82 | ||
83 | if ((src->size+dst->size)*minimum_score < delta_size*MAX_SCORE*2) | |
84 | return 0; | |
85 | ||
86 | delta = diff_delta(src->data, src->size, | |
87 | dst->data, dst->size, | |
88 | &delta_size); | |
89 | free(delta); | |
90 | ||
91 | /* This "delta" is really xdiff with adler32 and all the | |
92 | * overheads but it is a quick and dirty approximation. | |
93 | * | |
94 | * Now we will give some score to it. 100% edit gets | |
95 | * 0 points and 0% edit gets MAX_SCORE points. That is, every | |
96 | * 1/MAX_SCORE edit gets 1 point penalty. The amount of penalty is: | |
97 | * | |
98 | * (delta_size * 2 / (src->size + dst->size)) * MAX_SCORE | |
99 | * | |
100 | */ | |
101 | score = MAX_SCORE-(MAX_SCORE*2*delta_size/(src->size+dst->size)); | |
102 | if (score < 0) return 0; | |
103 | if (MAX_SCORE < score) return MAX_SCORE; | |
104 | return score; | |
105 | } | |
106 | ||
107 | static void record_rename_pair(struct diff_queue_struct *outq, | |
108 | struct diff_filespec *src, | |
109 | struct diff_filespec *dst, | |
110 | int rank, | |
111 | int score) | |
112 | { | |
113 | /* The rank is used to sort the final output, because there | |
114 | * are certain dependencies. | |
115 | * | |
116 | * - rank #0 depends on deleted ones. | |
117 | * - rank #1 depends on kept files before they are modified. | |
118 | * - rank #2 depends on kept files after they are modified; | |
119 | * currently not used. | |
120 | * | |
121 | * Therefore, the final output order should be: | |
122 | * | |
123 | * 1. rank #0 rename/copy diffs. | |
124 | * 2. deletions in the original. | |
125 | * 3. rank #1 rename/copy diffs. | |
126 | * 4. additions and modifications in the original. | |
127 | * 5. rank #2 rename/copy diffs; currently not used. | |
128 | * | |
129 | * To achieve this sort order, we give xform_work the number | |
130 | * above. | |
131 | */ | |
52e95789 | 132 | struct diff_filepair *dp = diff_queue(outq, src, dst); |
427dcb4b JH |
133 | dp->xfrm_work = (rank * 2 + 1) | (score<<RENAME_SCORE_SHIFT); |
134 | dst->xfrm_flags |= RENAME_DST_MATCHED; | |
135 | } | |
136 | ||
137 | #if 0 | |
138 | static void debug_filespec(struct diff_filespec *s, int x, const char *one) | |
139 | { | |
140 | fprintf(stderr, "queue[%d] %s (%s) %s %06o %s\n", | |
141 | x, one, | |
142 | s->path, | |
143 | s->file_valid ? "valid" : "invalid", | |
144 | s->mode, | |
145 | s->sha1_valid ? sha1_to_hex(s->sha1) : ""); | |
146 | fprintf(stderr, "queue[%d] %s size %lu flags %d\n", | |
147 | x, one, | |
148 | s->size, s->xfrm_flags); | |
149 | } | |
150 | ||
52e95789 | 151 | static void debug_filepair(const struct diff_filepair *p, int i) |
427dcb4b JH |
152 | { |
153 | debug_filespec(p->one, i, "one"); | |
154 | debug_filespec(p->two, i, "two"); | |
155 | fprintf(stderr, "pair flags %d, orig order %d, score %d\n", | |
156 | (p->xfrm_work & ((1<<RENAME_SCORE_SHIFT) - 1)), | |
157 | p->orig_order, | |
158 | (p->xfrm_work >> RENAME_SCORE_SHIFT)); | |
159 | } | |
160 | ||
161 | static void debug_queue(const char *msg, struct diff_queue_struct *q) | |
162 | { | |
163 | int i; | |
164 | if (msg) | |
165 | fprintf(stderr, "%s\n", msg); | |
166 | fprintf(stderr, "q->nr = %d\n", q->nr); | |
167 | for (i = 0; i < q->nr; i++) { | |
52e95789 | 168 | struct diff_filepair *p = q->queue[i]; |
427dcb4b JH |
169 | debug_filepair(p, i); |
170 | } | |
171 | } | |
172 | #else | |
173 | #define debug_queue(a,b) do { ; /*nothing*/ } while(0) | |
174 | #endif | |
175 | ||
176 | /* | |
177 | * We sort the outstanding diff entries according to the rank (see | |
178 | * comment at the beginning of record_rename_pair) and tiebreak with | |
179 | * the order in the original input. | |
180 | */ | |
181 | static int rank_compare(const void *a_, const void *b_) | |
182 | { | |
52e95789 JH |
183 | const struct diff_filepair *a = *(const struct diff_filepair **)a_; |
184 | const struct diff_filepair *b = *(const struct diff_filepair **)b_; | |
427dcb4b JH |
185 | int a_rank = a->xfrm_work & ((1<<RENAME_SCORE_SHIFT) - 1); |
186 | int b_rank = b->xfrm_work & ((1<<RENAME_SCORE_SHIFT) - 1); | |
187 | ||
188 | if (a_rank != b_rank) | |
189 | return a_rank - b_rank; | |
190 | return a->orig_order - b->orig_order; | |
191 | } | |
192 | ||
193 | /* | |
194 | * We sort the rename similarity matrix with the score, in descending | |
195 | * order (more similar first). | |
196 | */ | |
197 | static int score_compare(const void *a_, const void *b_) | |
198 | { | |
199 | const struct diff_score *a = a_, *b = b_; | |
200 | return b->score - a->score; | |
201 | } | |
202 | ||
203 | static int needs_to_stay(struct diff_queue_struct *q, int i, | |
204 | struct diff_filespec *it) | |
205 | { | |
206 | /* If it will be used in later entry (either stay or used | |
207 | * as the source of rename/copy), we need to copy, not rename. | |
208 | */ | |
209 | while (i < q->nr) { | |
52e95789 | 210 | struct diff_filepair *p = q->queue[i++]; |
427dcb4b JH |
211 | if (!p->two->file_valid) |
212 | continue; /* removed is fine */ | |
213 | if (strcmp(p->one->path, it->path)) | |
214 | continue; /* not relevant */ | |
215 | ||
216 | /* p has its src set to *it and it is not a delete; | |
217 | * it will be used for in-place change or rename/copy, | |
218 | * so we cannot rename it out. | |
219 | */ | |
220 | return 1; | |
221 | } | |
222 | return 0; | |
223 | } | |
224 | ||
225 | void diff_detect_rename(struct diff_queue_struct *q, | |
226 | int detect_rename, | |
227 | int minimum_score) | |
228 | { | |
229 | struct diff_queue_struct outq; | |
230 | struct diff_rename_pool created, deleted, stay; | |
231 | struct diff_rename_pool *(srcs[2]); | |
232 | struct diff_score *mx; | |
233 | int h, i, j; | |
234 | int num_create, num_src, dst_cnt, src_cnt; | |
235 | ||
236 | outq.queue = NULL; | |
237 | outq.nr = outq.alloc = 0; | |
238 | ||
239 | diff_rename_pool_clear(&created); | |
240 | diff_rename_pool_clear(&deleted); | |
241 | diff_rename_pool_clear(&stay); | |
242 | ||
243 | srcs[0] = &deleted; | |
244 | srcs[1] = &stay; | |
245 | ||
427dcb4b | 246 | for (i = 0; i < q->nr; i++) { |
52e95789 | 247 | struct diff_filepair *p = q->queue[i]; |
427dcb4b JH |
248 | if (!p->one->file_valid) |
249 | if (!p->two->file_valid) | |
250 | continue; /* ignore nonsense */ | |
251 | else | |
252 | diff_rename_pool_add(&created, p->two); | |
253 | else if (!p->two->file_valid) | |
254 | diff_rename_pool_add(&deleted, p->one); | |
255 | else if (1 < detect_rename) /* find copy, too */ | |
256 | diff_rename_pool_add(&stay, p->one); | |
257 | } | |
258 | if (created.nr == 0) | |
259 | goto cleanup; /* nothing to do */ | |
260 | ||
261 | /* We really want to cull the candidates list early | |
262 | * with cheap tests in order to avoid doing deltas. | |
263 | * | |
264 | * With the current callers, we should not have already | |
265 | * matched entries at this point, but it is nonetheless | |
266 | * checked for sanity. | |
267 | */ | |
268 | for (i = 0; i < created.nr; i++) { | |
269 | if (created.s[i]->xfrm_flags & RENAME_DST_MATCHED) | |
270 | continue; /* we have matched exactly already */ | |
271 | for (h = 0; h < sizeof(srcs)/sizeof(srcs[0]); h++) { | |
272 | struct diff_rename_pool *p = srcs[h]; | |
273 | for (j = 0; j < p->nr; j++) { | |
274 | if (!is_exact_match(p->s[j], created.s[i])) | |
275 | continue; | |
276 | record_rename_pair(&outq, | |
277 | p->s[j], created.s[i], h, | |
278 | MAX_SCORE); | |
279 | break; /* we are done with this entry */ | |
280 | } | |
281 | } | |
282 | } | |
283 | debug_queue("done detecting exact", &outq); | |
284 | ||
285 | /* Have we run out the created file pool? If so we can avoid | |
286 | * doing the delta matrix altogether. | |
287 | */ | |
288 | if (outq.nr == created.nr) | |
289 | goto flush_rest; | |
290 | ||
291 | num_create = (created.nr - outq.nr); | |
292 | num_src = deleted.nr + stay.nr; | |
293 | mx = xmalloc(sizeof(*mx) * num_create * num_src); | |
294 | for (dst_cnt = i = 0; i < created.nr; i++) { | |
295 | int base = dst_cnt * num_src; | |
296 | if (created.s[i]->xfrm_flags & RENAME_DST_MATCHED) | |
297 | continue; /* dealt with exact match already. */ | |
298 | for (src_cnt = h = 0; h < sizeof(srcs)/sizeof(srcs[0]); h++) { | |
299 | struct diff_rename_pool *p = srcs[h]; | |
300 | for (j = 0; j < p->nr; j++, src_cnt++) { | |
301 | struct diff_score *m = &mx[base + src_cnt]; | |
302 | m->src = p->s[j]; | |
303 | m->dst = created.s[i]; | |
304 | m->score = estimate_similarity(m->src, m->dst, | |
305 | minimum_score); | |
306 | m->rank = h; | |
307 | } | |
308 | } | |
309 | dst_cnt++; | |
310 | } | |
311 | /* cost matrix sorted by most to least similar pair */ | |
312 | qsort(mx, num_create * num_src, sizeof(*mx), score_compare); | |
313 | for (i = 0; i < num_create * num_src; i++) { | |
314 | if (mx[i].dst->xfrm_flags & RENAME_DST_MATCHED) | |
315 | continue; /* alreayd done, either exact or fuzzy. */ | |
316 | if (mx[i].score < minimum_score) | |
317 | continue; | |
318 | record_rename_pair(&outq, | |
319 | mx[i].src, mx[i].dst, mx[i].rank, | |
320 | mx[i].score); | |
321 | } | |
322 | free(mx); | |
323 | debug_queue("done detecting fuzzy", &outq); | |
324 | ||
325 | flush_rest: | |
326 | /* At this point, we have found some renames and copies and they | |
327 | * are kept in outq. The original list is still in *q. | |
328 | * | |
329 | * Scan the original list and move them into the outq; we will sort | |
330 | * outq and swap it into the queue supplied to pass that to | |
331 | * downstream, so we assign the sort keys in this loop. | |
332 | * | |
333 | * See comments at the top of record_rename_pair for numbers used | |
334 | * to assign xfrm_work. | |
335 | * | |
52e95789 | 336 | * Note that we have not annotated the diff_filepair with any comment |
427dcb4b JH |
337 | * so there is nothing other than p to free. |
338 | */ | |
339 | for (i = 0; i < q->nr; i++) { | |
52e95789 | 340 | struct diff_filepair *dp, *p = q->queue[i]; |
427dcb4b JH |
341 | if (!p->one->file_valid) { |
342 | if (p->two->file_valid) { | |
343 | /* creation */ | |
344 | dp = diff_queue(&outq, p->one, p->two); | |
345 | dp->xfrm_work = 4; | |
346 | } | |
347 | /* otherwise it is a nonsense; just ignore it */ | |
348 | } | |
349 | else if (!p->two->file_valid) { | |
350 | /* deletion */ | |
351 | dp = diff_queue(&outq, p->one, p->two); | |
352 | dp->xfrm_work = 2; | |
353 | } | |
354 | else { | |
355 | /* modification, or stay as is */ | |
356 | dp = diff_queue(&outq, p->one, p->two); | |
357 | dp->xfrm_work = 4; | |
358 | } | |
359 | free(p); | |
360 | } | |
361 | debug_queue("done copying original", &outq); | |
362 | ||
363 | /* Sort outq */ | |
364 | qsort(outq.queue, outq.nr, sizeof(outq.queue[0]), rank_compare); | |
365 | ||
366 | debug_queue("done sorting", &outq); | |
367 | ||
368 | free(q->queue); | |
369 | q->nr = q->alloc = 0; | |
370 | q->queue = NULL; | |
371 | ||
372 | /* Copy it out to q, removing duplicates. */ | |
373 | for (i = 0; i < outq.nr; i++) { | |
52e95789 | 374 | struct diff_filepair *p = outq.queue[i]; |
427dcb4b JH |
375 | if (!p->one->file_valid) { |
376 | /* created */ | |
377 | if (p->two->xfrm_flags & RENAME_DST_MATCHED) | |
378 | ; /* rename/copy created it already */ | |
379 | else | |
380 | diff_queue(q, p->one, p->two); | |
381 | } | |
382 | else if (!p->two->file_valid) { | |
383 | /* deleted */ | |
384 | if (p->one->xfrm_flags & RENAME_SRC_GONE) | |
385 | ; /* rename/copy deleted it already */ | |
386 | else | |
387 | diff_queue(q, p->one, p->two); | |
388 | } | |
389 | else if (strcmp(p->one->path, p->two->path)) { | |
390 | /* rename or copy */ | |
52e95789 | 391 | struct diff_filepair *dp = |
427dcb4b JH |
392 | diff_queue(q, p->one, p->two); |
393 | int msglen = (strlen(p->one->path) + | |
394 | strlen(p->two->path) + 100); | |
395 | int score = (p->xfrm_work >> RENAME_SCORE_SHIFT); | |
396 | dp->xfrm_msg = xmalloc(msglen); | |
397 | ||
398 | /* if we have a later entry that is a rename/copy | |
399 | * that depends on p->one, then we copy here. | |
400 | * otherwise we rename it. | |
401 | */ | |
402 | if (needs_to_stay(&outq, i+1, p->one)) { | |
403 | /* copy it */ | |
404 | sprintf(dp->xfrm_msg, | |
405 | "similarity index %d%%\n" | |
406 | "copy from %s\n" | |
407 | "copy to %s\n", | |
408 | (int)(0.5 + score * 100 / MAX_SCORE), | |
409 | p->one->path, p->two->path); | |
410 | } | |
411 | else { | |
412 | /* rename it, and mark it as gone. */ | |
413 | p->one->xfrm_flags |= RENAME_SRC_GONE; | |
414 | sprintf(dp->xfrm_msg, | |
415 | "similarity index %d%%\n" | |
416 | "rename old %s\n" | |
417 | "rename new %s\n", | |
418 | (int)(0.5 + score * 100 / MAX_SCORE), | |
419 | p->one->path, p->two->path); | |
420 | } | |
421 | } | |
422 | else | |
423 | /* otherwise it is a modified (or stayed) entry */ | |
424 | diff_queue(q, p->one, p->two); | |
425 | free(p); | |
426 | } | |
427 | ||
428 | free(outq.queue); | |
429 | debug_queue("done collapsing", q); | |
430 | ||
431 | cleanup: | |
432 | free(created.s); | |
433 | free(deleted.s); | |
434 | free(stay.s); | |
435 | return; | |
436 | } |