1 /*
2 * Copyright (C) 2005 Junio C Hamano
3 */
4 #include "cache.h"
5 #include "diff.h"
6 #include "diffcore.h"
7 #include "hash.h"
9 /* Table of rename/copy destinations */
11 static struct diff_rename_dst {
12 struct diff_filespec *two;
13 struct diff_filepair *pair;
14 } *rename_dst;
15 static int rename_dst_nr, rename_dst_alloc;
17 static struct diff_rename_dst *locate_rename_dst(struct diff_filespec *two,
18 int insert_ok)
19 {
20 int first, last;
22 first = 0;
23 last = rename_dst_nr;
24 while (last > first) {
25 int next = (last + first) >> 1;
26 struct diff_rename_dst *dst = &(rename_dst[next]);
27 int cmp = strcmp(two->path, dst->two->path);
28 if (!cmp)
29 return dst;
30 if (cmp < 0) {
31 last = next;
32 continue;
33 }
34 first = next+1;
35 }
36 /* not found */
37 if (!insert_ok)
38 return NULL;
39 /* insert to make it at "first" */
40 if (rename_dst_alloc <= rename_dst_nr) {
41 rename_dst_alloc = alloc_nr(rename_dst_alloc);
42 rename_dst = xrealloc(rename_dst,
43 rename_dst_alloc * sizeof(*rename_dst));
44 }
45 rename_dst_nr++;
46 if (first < rename_dst_nr)
47 memmove(rename_dst + first + 1, rename_dst + first,
48 (rename_dst_nr - first - 1) * sizeof(*rename_dst));
49 rename_dst[first].two = alloc_filespec(two->path);
50 fill_filespec(rename_dst[first].two, two->sha1, two->mode);
51 rename_dst[first].pair = NULL;
52 return &(rename_dst[first]);
53 }
55 /* Table of rename/copy src files */
56 static struct diff_rename_src {
57 struct diff_filepair *p;
58 unsigned short score; /* to remember the break score */
59 } *rename_src;
60 static int rename_src_nr, rename_src_alloc;
62 static struct diff_rename_src *register_rename_src(struct diff_filepair *p)
63 {
64 int first, last;
65 struct diff_filespec *one = p->one;
66 unsigned short score = p->score;
68 first = 0;
69 last = rename_src_nr;
70 while (last > first) {
71 int next = (last + first) >> 1;
72 struct diff_rename_src *src = &(rename_src[next]);
73 int cmp = strcmp(one->path, src->p->one->path);
74 if (!cmp)
75 return src;
76 if (cmp < 0) {
77 last = next;
78 continue;
79 }
80 first = next+1;
81 }
83 /* insert to make it at "first" */
84 if (rename_src_alloc <= rename_src_nr) {
85 rename_src_alloc = alloc_nr(rename_src_alloc);
86 rename_src = xrealloc(rename_src,
87 rename_src_alloc * sizeof(*rename_src));
88 }
89 rename_src_nr++;
90 if (first < rename_src_nr)
91 memmove(rename_src + first + 1, rename_src + first,
92 (rename_src_nr - first - 1) * sizeof(*rename_src));
93 rename_src[first].p = p;
94 rename_src[first].score = score;
95 return &(rename_src[first]);
96 }
98 static int basename_same(struct diff_filespec *src, struct diff_filespec *dst)
99 {
100 int src_len = strlen(src->path), dst_len = strlen(dst->path);
101 while (src_len && dst_len) {
102 char c1 = src->path[--src_len];
103 char c2 = dst->path[--dst_len];
104 if (c1 != c2)
105 return 0;
106 if (c1 == '/')
107 return 1;
108 }
109 return (!src_len || src->path[src_len - 1] == '/') &&
110 (!dst_len || dst->path[dst_len - 1] == '/');
111 }
113 struct diff_score {
114 int src; /* index in rename_src */
115 int dst; /* index in rename_dst */
116 unsigned short score;
117 short name_score;
118 };
120 static int estimate_similarity(struct diff_filespec *src,
121 struct diff_filespec *dst,
122 int minimum_score)
123 {
124 /* src points at a file that existed in the original tree (or
125 * optionally a file in the destination tree) and dst points
126 * at a newly created file. They may be quite similar, in which
127 * case we want to say src is renamed to dst or src is copied into
128 * dst, and then some edit has been applied to dst.
129 *
130 * Compare them and return how similar they are, representing
131 * the score as an integer between 0 and MAX_SCORE.
132 *
133 * When there is an exact match, it is considered a better
134 * match than anything else; the destination does not even
135 * call into this function in that case.
136 */
137 unsigned long max_size, delta_size, base_size, src_copied, literal_added;
138 unsigned long delta_limit;
139 int score;
141 /* We deal only with regular files. Symlink renames are handled
142 * only when they are exact matches --- in other words, no edits
143 * after renaming.
144 */
145 if (!S_ISREG(src->mode) || !S_ISREG(dst->mode))
146 return 0;
148 /*
149 * Need to check that source and destination sizes are
150 * filled in before comparing them.
151 *
152 * If we already have "cnt_data" filled in, we know it's
153 * all good (avoid checking the size for zero, as that
154 * is a possible size - we really should have a flag to
155 * say whether the size is valid or not!)
156 */
157 if (!src->cnt_data && diff_populate_filespec(src, 1))
158 return 0;
159 if (!dst->cnt_data && diff_populate_filespec(dst, 1))
160 return 0;
162 max_size = ((src->size > dst->size) ? src->size : dst->size);
163 base_size = ((src->size < dst->size) ? src->size : dst->size);
164 delta_size = max_size - base_size;
166 /* We would not consider edits that change the file size so
167 * drastically. delta_size must be smaller than
168 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
169 *
170 * Note that base_size == 0 case is handled here already
171 * and the final score computation below would not have a
172 * divide-by-zero issue.
173 */
174 if (base_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE)
175 return 0;
177 if (!src->cnt_data && diff_populate_filespec(src, 0))
178 return 0;
179 if (!dst->cnt_data && diff_populate_filespec(dst, 0))
180 return 0;
182 delta_limit = (unsigned long)
183 (base_size * (MAX_SCORE-minimum_score) / MAX_SCORE);
184 if (diffcore_count_changes(src, dst,
185 &src->cnt_data, &dst->cnt_data,
186 delta_limit,
187 &src_copied, &literal_added))
188 return 0;
190 /* How similar are they?
191 * what percentage of material in dst are from source?
192 */
193 if (!dst->size)
194 score = 0; /* should not happen */
195 else
196 score = (int)(src_copied * MAX_SCORE / max_size);
197 return score;
198 }
200 static void record_rename_pair(int dst_index, int src_index, int score)
201 {
202 struct diff_filespec *src, *dst;
203 struct diff_filepair *dp;
205 if (rename_dst[dst_index].pair)
206 die("internal error: dst already matched.");
208 src = rename_src[src_index].p->one;
209 src->rename_used++;
210 src->count++;
212 dst = rename_dst[dst_index].two;
213 dst->count++;
215 dp = diff_queue(NULL, src, dst);
216 dp->renamed_pair = 1;
217 if (!strcmp(src->path, dst->path))
218 dp->score = rename_src[src_index].score;
219 else
220 dp->score = score;
221 rename_dst[dst_index].pair = dp;
222 }
224 /*
225 * We sort the rename similarity matrix with the score, in descending
226 * order (the most similar first).
227 */
228 static int score_compare(const void *a_, const void *b_)
229 {
230 const struct diff_score *a = a_, *b = b_;
232 /* sink the unused ones to the bottom */
233 if (a->dst < 0)
234 return (0 <= b->dst);
235 else if (b->dst < 0)
236 return -1;
238 if (a->score == b->score)
239 return b->name_score - a->name_score;
241 return b->score - a->score;
242 }
244 struct file_similarity {
245 int src_dst, index;
246 struct diff_filespec *filespec;
247 struct file_similarity *next;
248 };
250 static int find_identical_files(struct file_similarity *src,
251 struct file_similarity *dst)
252 {
253 int renames = 0;
255 /*
256 * Walk over all the destinations ...
257 */
258 do {
259 struct diff_filespec *target = dst->filespec;
260 struct file_similarity *p, *best;
261 int i = 100, best_score = -1;
263 /*
264 * .. to find the best source match
265 */
266 best = NULL;
267 for (p = src; p; p = p->next) {
268 int score;
269 struct diff_filespec *source = p->filespec;
271 /* False hash collision? */
272 if (hashcmp(source->sha1, target->sha1))
273 continue;
274 /* Non-regular files? If so, the modes must match! */
275 if (!S_ISREG(source->mode) || !S_ISREG(target->mode)) {
276 if (source->mode != target->mode)
277 continue;
278 }
279 /* Give higher scores to sources that haven't been used already */
280 score = !source->rename_used;
281 score += basename_same(source, target);
282 if (score > best_score) {
283 best = p;
284 best_score = score;
285 if (score == 2)
286 break;
287 }
289 /* Too many identical alternatives? Pick one */
290 if (!--i)
291 break;
292 }
293 if (best) {
294 record_rename_pair(dst->index, best->index, MAX_SCORE);
295 renames++;
296 }
297 } while ((dst = dst->next) != NULL);
298 return renames;
299 }
301 static void free_similarity_list(struct file_similarity *p)
302 {
303 while (p) {
304 struct file_similarity *entry = p;
305 p = p->next;
306 free(entry);
307 }
308 }
310 static int find_same_files(void *ptr)
311 {
312 int ret;
313 struct file_similarity *p = ptr;
314 struct file_similarity *src = NULL, *dst = NULL;
316 /* Split the hash list up into sources and destinations */
317 do {
318 struct file_similarity *entry = p;
319 p = p->next;
320 if (entry->src_dst < 0) {
321 entry->next = src;
322 src = entry;
323 } else {
324 entry->next = dst;
325 dst = entry;
326 }
327 } while (p);
329 /*
330 * If we have both sources *and* destinations, see if
331 * we can match them up
332 */
333 ret = (src && dst) ? find_identical_files(src, dst) : 0;
335 /* Free the hashes and return the number of renames found */
336 free_similarity_list(src);
337 free_similarity_list(dst);
338 return ret;
339 }
341 static unsigned int hash_filespec(struct diff_filespec *filespec)
342 {
343 unsigned int hash;
344 if (!filespec->sha1_valid) {
345 if (diff_populate_filespec(filespec, 0))
346 return 0;
347 hash_sha1_file(filespec->data, filespec->size, "blob", filespec->sha1);
348 }
349 memcpy(&hash, filespec->sha1, sizeof(hash));
350 return hash;
351 }
353 static void insert_file_table(struct hash_table *table, int src_dst, int index, struct diff_filespec *filespec)
354 {
355 void **pos;
356 unsigned int hash;
357 struct file_similarity *entry = xmalloc(sizeof(*entry));
359 entry->src_dst = src_dst;
360 entry->index = index;
361 entry->filespec = filespec;
362 entry->next = NULL;
364 hash = hash_filespec(filespec);
365 pos = insert_hash(hash, entry, table);
367 /* We already had an entry there? */
368 if (pos) {
369 entry->next = *pos;
370 *pos = entry;
371 }
372 }
374 /*
375 * Find exact renames first.
376 *
377 * The first round matches up the up-to-date entries,
378 * and then during the second round we try to match
379 * cache-dirty entries as well.
380 */
381 static int find_exact_renames(void)
382 {
383 int i;
384 struct hash_table file_table;
386 init_hash(&file_table);
387 for (i = 0; i < rename_src_nr; i++)
388 insert_file_table(&file_table, -1, i, rename_src[i].p->one);
390 for (i = 0; i < rename_dst_nr; i++)
391 insert_file_table(&file_table, 1, i, rename_dst[i].two);
393 /* Find the renames */
394 i = for_each_hash(&file_table, find_same_files);
396 /* .. and free the hash data structure */
397 free_hash(&file_table);
399 return i;
400 }
402 #define NUM_CANDIDATE_PER_DST 4
403 static void record_if_better(struct diff_score m[], struct diff_score *o)
404 {
405 int i, worst;
407 /* find the worst one */
408 worst = 0;
409 for (i = 1; i < NUM_CANDIDATE_PER_DST; i++)
410 if (score_compare(&m[i], &m[worst]) > 0)
411 worst = i;
413 /* is it better than the worst one? */
414 if (score_compare(&m[worst], o) > 0)
415 m[worst] = *o;
416 }
418 /*
419 * Returns:
420 * 0 if we are under the limit;
421 * 1 if we need to disable inexact rename detection;
422 * 2 if we would be under the limit if we were given -C instead of -C -C.
423 */
424 static int too_many_rename_candidates(int num_create,
425 struct diff_options *options)
426 {
427 int rename_limit = options->rename_limit;
428 int num_src = rename_src_nr;
429 int i;
431 /*
432 * This basically does a test for the rename matrix not
433 * growing larger than a "rename_limit" square matrix, ie:
434 *
435 * num_create * num_src > rename_limit * rename_limit
436 *
437 * but handles the potential overflow case specially (and we
438 * assume at least 32-bit integers)
439 */
440 if (rename_limit <= 0 || rename_limit > 32767)
441 rename_limit = 32767;
442 if ((num_create <= rename_limit || num_src <= rename_limit) &&
443 (num_create * num_src <= rename_limit * rename_limit))
444 return 0;
446 /* Are we running under -C -C? */
447 if (!DIFF_OPT_TST(options, FIND_COPIES_HARDER))
448 return 1;
450 /* Would we bust the limit if we were running under -C? */
451 for (num_src = i = 0; i < rename_src_nr; i++) {
452 if (diff_unmodified_pair(rename_src[i].p))
453 continue;
454 num_src++;
455 }
456 if ((num_create <= rename_limit || num_src <= rename_limit) &&
457 (num_create * num_src <= rename_limit * rename_limit))
458 return 2;
459 return 1;
460 }
462 void diffcore_rename(struct diff_options *options)
463 {
464 int detect_rename = options->detect_rename;
465 int minimum_score = options->rename_score;
466 struct diff_queue_struct *q = &diff_queued_diff;
467 struct diff_queue_struct outq;
468 struct diff_score *mx;
469 int i, j, rename_count, skip_unmodified = 0;
470 int num_create, num_src, dst_cnt;
472 if (!minimum_score)
473 minimum_score = DEFAULT_RENAME_SCORE;
475 for (i = 0; i < q->nr; i++) {
476 struct diff_filepair *p = q->queue[i];
477 if (!DIFF_FILE_VALID(p->one)) {
478 if (!DIFF_FILE_VALID(p->two))
479 continue; /* unmerged */
480 else if (options->single_follow &&
481 strcmp(options->single_follow, p->two->path))
482 continue; /* not interested */
483 else
484 locate_rename_dst(p->two, 1);
485 }
486 else if (!DIFF_FILE_VALID(p->two)) {
487 /*
488 * If the source is a broken "delete", and
489 * they did not really want to get broken,
490 * that means the source actually stays.
491 * So we increment the "rename_used" score
492 * by one, to indicate ourselves as a user
493 */
494 if (p->broken_pair && !p->score)
495 p->one->rename_used++;
496 register_rename_src(p);
497 }
498 else if (detect_rename == DIFF_DETECT_COPY) {
499 /*
500 * Increment the "rename_used" score by
501 * one, to indicate ourselves as a user.
502 */
503 p->one->rename_used++;
504 register_rename_src(p);
505 }
506 }
507 if (rename_dst_nr == 0 || rename_src_nr == 0)
508 goto cleanup; /* nothing to do */
510 /*
511 * We really want to cull the candidates list early
512 * with cheap tests in order to avoid doing deltas.
513 */
514 rename_count = find_exact_renames();
516 /* Did we only want exact renames? */
517 if (minimum_score == MAX_SCORE)
518 goto cleanup;
520 /*
521 * Calculate how many renames are left (but all the source
522 * files still remain as options for rename/copies!)
523 */
524 num_create = (rename_dst_nr - rename_count);
525 num_src = rename_src_nr;
527 /* All done? */
528 if (!num_create)
529 goto cleanup;
531 switch (too_many_rename_candidates(num_create, options)) {
532 case 1:
533 if (options->warn_on_too_large_rename)
534 warning("too many files (created: %d deleted: %d), skipping inexact rename detection", num_create, num_src);
535 goto cleanup;
536 case 2:
537 if (options->warn_on_too_large_rename)
538 warning("too many files, falling back to -C");
539 skip_unmodified = 1;
540 break;
541 default:
542 break;
543 }
545 mx = xcalloc(num_create * NUM_CANDIDATE_PER_DST, sizeof(*mx));
546 for (dst_cnt = i = 0; i < rename_dst_nr; i++) {
547 struct diff_filespec *two = rename_dst[i].two;
548 struct diff_score *m;
550 if (rename_dst[i].pair)
551 continue; /* dealt with exact match already. */
553 m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST];
554 for (j = 0; j < NUM_CANDIDATE_PER_DST; j++)
555 m[j].dst = -1;
557 for (j = 0; j < rename_src_nr; j++) {
558 struct diff_filespec *one = rename_src[j].p->one;
559 struct diff_score this_src;
561 if (skip_unmodified &&
562 diff_unmodified_pair(rename_src[j].p))
563 continue;
565 this_src.score = estimate_similarity(one, two,
566 minimum_score);
567 this_src.name_score = basename_same(one, two);
568 this_src.dst = i;
569 this_src.src = j;
570 record_if_better(m, &this_src);
571 /*
572 * Once we run estimate_similarity,
573 * We do not need the text anymore.
574 */
575 diff_free_filespec_blob(one);
576 diff_free_filespec_blob(two);
577 }
578 dst_cnt++;
579 }
581 /* cost matrix sorted by most to least similar pair */
582 qsort(mx, dst_cnt * NUM_CANDIDATE_PER_DST, sizeof(*mx), score_compare);
584 for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) {
585 struct diff_rename_dst *dst;
587 if ((mx[i].dst < 0) ||
588 (mx[i].score < minimum_score))
589 break; /* there is no more usable pair. */
590 dst = &rename_dst[mx[i].dst];
591 if (dst->pair)
592 continue; /* already done, either exact or fuzzy. */
593 if (rename_src[mx[i].src].p->one->rename_used)
594 continue;
595 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
596 rename_count++;
597 }
599 for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) {
600 struct diff_rename_dst *dst;
602 if ((mx[i].dst < 0) ||
603 (mx[i].score < minimum_score))
604 break; /* there is no more usable pair. */
605 dst = &rename_dst[mx[i].dst];
606 if (dst->pair)
607 continue; /* already done, either exact or fuzzy. */
608 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
609 rename_count++;
610 }
611 free(mx);
613 cleanup:
614 /* At this point, we have found some renames and copies and they
615 * are recorded in rename_dst. The original list is still in *q.
616 */
617 DIFF_QUEUE_CLEAR(&outq);
618 for (i = 0; i < q->nr; i++) {
619 struct diff_filepair *p = q->queue[i];
620 struct diff_filepair *pair_to_free = NULL;
622 if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
623 /*
624 * Creation
625 *
626 * We would output this create record if it has
627 * not been turned into a rename/copy already.
628 */
629 struct diff_rename_dst *dst =
630 locate_rename_dst(p->two, 0);
631 if (dst && dst->pair) {
632 diff_q(&outq, dst->pair);
633 pair_to_free = p;
634 }
635 else
636 /* no matching rename/copy source, so
637 * record this as a creation.
638 */
639 diff_q(&outq, p);
640 }
641 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) {
642 /*
643 * Deletion
644 *
645 * We would output this delete record if:
646 *
647 * (1) this is a broken delete and the counterpart
648 * broken create remains in the output; or
649 * (2) this is not a broken delete, and rename_dst
650 * does not have a rename/copy to move p->one->path
651 * out of existence.
652 *
653 * Otherwise, the counterpart broken create
654 * has been turned into a rename-edit; or
655 * delete did not have a matching create to
656 * begin with.
657 */
658 if (DIFF_PAIR_BROKEN(p)) {
659 /* broken delete */
660 struct diff_rename_dst *dst =
661 locate_rename_dst(p->one, 0);
662 if (dst && dst->pair)
663 /* counterpart is now rename/copy */
664 pair_to_free = p;
665 }
666 else {
667 if (p->one->rename_used)
668 /* this path remains */
669 pair_to_free = p;
670 }
672 if (pair_to_free)
673 ;
674 else
675 diff_q(&outq, p);
676 }
677 else if (!diff_unmodified_pair(p))
678 /* all the usual ones need to be kept */
679 diff_q(&outq, p);
680 else
681 /* no need to keep unmodified pairs */
682 pair_to_free = p;
684 if (pair_to_free)
685 diff_free_filepair(pair_to_free);
686 }
687 diff_debug_queue("done copying original", &outq);
689 free(q->queue);
690 *q = outq;
691 diff_debug_queue("done collapsing", q);
693 for (i = 0; i < rename_dst_nr; i++)
694 free_filespec(rename_dst[i].two);
696 free(rename_dst);
697 rename_dst = NULL;
698 rename_dst_nr = rename_dst_alloc = 0;
699 free(rename_src);
700 rename_src = NULL;
701 rename_src_nr = rename_src_alloc = 0;
702 return;
703 }