2 * Copyright (C) 2005 Junio C Hamano
7 #include "object-store.h"
11 /* Table of rename/copy destinations */
13 static struct diff_rename_dst {
14 struct diff_filespec *two;
15 struct diff_filepair *pair;
17 static int rename_dst_nr, rename_dst_alloc;
19 static int find_rename_dst(struct diff_filespec *two)
25 while (last > first) {
26 int next = first + ((last - first) >> 1);
27 struct diff_rename_dst *dst = &(rename_dst[next]);
28 int cmp = strcmp(two->path, dst->two->path);
40 static struct diff_rename_dst *locate_rename_dst(struct diff_filespec *two)
42 int ofs = find_rename_dst(two);
43 return ofs < 0 ? NULL : &rename_dst[ofs];
47 * Returns 0 on success, -1 if we found a duplicate.
49 static int add_rename_dst(struct diff_filespec *two)
51 int first = find_rename_dst(two);
57 /* insert to make it at "first" */
58 ALLOC_GROW(rename_dst, rename_dst_nr + 1, rename_dst_alloc);
60 if (first < rename_dst_nr)
61 MOVE_ARRAY(rename_dst + first + 1, rename_dst + first,
62 rename_dst_nr - first - 1);
63 rename_dst[first].two = alloc_filespec(two->path);
64 fill_filespec(rename_dst[first].two, &two->oid, two->oid_valid,
66 rename_dst[first].pair = NULL;
70 /* Table of rename/copy src files */
71 static struct diff_rename_src {
72 struct diff_filepair *p;
73 unsigned short score; /* to remember the break score */
75 static int rename_src_nr, rename_src_alloc;
77 static struct diff_rename_src *register_rename_src(struct diff_filepair *p)
80 struct diff_filespec *one = p->one;
81 unsigned short score = p->score;
85 while (last > first) {
86 int next = first + ((last - first) >> 1);
87 struct diff_rename_src *src = &(rename_src[next]);
88 int cmp = strcmp(one->path, src->p->one->path);
98 /* insert to make it at "first" */
99 ALLOC_GROW(rename_src, rename_src_nr + 1, rename_src_alloc);
101 if (first < rename_src_nr)
102 MOVE_ARRAY(rename_src + first + 1, rename_src + first,
103 rename_src_nr - first - 1);
104 rename_src[first].p = p;
105 rename_src[first].score = score;
106 return &(rename_src[first]);
109 static int basename_same(struct diff_filespec *src, struct diff_filespec *dst)
111 int src_len = strlen(src->path), dst_len = strlen(dst->path);
112 while (src_len && dst_len) {
113 char c1 = src->path[--src_len];
114 char c2 = dst->path[--dst_len];
120 return (!src_len || src->path[src_len - 1] == '/') &&
121 (!dst_len || dst->path[dst_len - 1] == '/');
125 int src; /* index in rename_src */
126 int dst; /* index in rename_dst */
127 unsigned short score;
131 static int estimate_similarity(struct repository *r,
132 struct diff_filespec *src,
133 struct diff_filespec *dst,
136 /* src points at a file that existed in the original tree (or
137 * optionally a file in the destination tree) and dst points
138 * at a newly created file. They may be quite similar, in which
139 * case we want to say src is renamed to dst or src is copied into
140 * dst, and then some edit has been applied to dst.
142 * Compare them and return how similar they are, representing
143 * the score as an integer between 0 and MAX_SCORE.
145 * When there is an exact match, it is considered a better
146 * match than anything else; the destination does not even
147 * call into this function in that case.
149 unsigned long max_size, delta_size, base_size, src_copied, literal_added;
152 /* We deal only with regular files. Symlink renames are handled
153 * only when they are exact matches --- in other words, no edits
156 if (!S_ISREG(src->mode) || !S_ISREG(dst->mode))
160 * Need to check that source and destination sizes are
161 * filled in before comparing them.
163 * If we already have "cnt_data" filled in, we know it's
164 * all good (avoid checking the size for zero, as that
165 * is a possible size - we really should have a flag to
166 * say whether the size is valid or not!)
168 if (!src->cnt_data &&
169 diff_populate_filespec(r, src, CHECK_SIZE_ONLY))
171 if (!dst->cnt_data &&
172 diff_populate_filespec(r, dst, CHECK_SIZE_ONLY))
175 max_size = ((src->size > dst->size) ? src->size : dst->size);
176 base_size = ((src->size < dst->size) ? src->size : dst->size);
177 delta_size = max_size - base_size;
179 /* We would not consider edits that change the file size so
180 * drastically. delta_size must be smaller than
181 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
183 * Note that base_size == 0 case is handled here already
184 * and the final score computation below would not have a
185 * divide-by-zero issue.
187 if (max_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE)
190 if (!src->cnt_data && diff_populate_filespec(r, src, 0))
192 if (!dst->cnt_data && diff_populate_filespec(r, dst, 0))
195 if (diffcore_count_changes(r, src, dst,
196 &src->cnt_data, &dst->cnt_data,
197 &src_copied, &literal_added))
200 /* How similar are they?
201 * what percentage of material in dst are from source?
204 score = 0; /* should not happen */
206 score = (int)(src_copied * MAX_SCORE / max_size);
210 static void record_rename_pair(int dst_index, int src_index, int score)
212 struct diff_filespec *src, *dst;
213 struct diff_filepair *dp;
215 if (rename_dst[dst_index].pair)
216 die("internal error: dst already matched.");
218 src = rename_src[src_index].p->one;
222 dst = rename_dst[dst_index].two;
225 dp = diff_queue(NULL, src, dst);
226 dp->renamed_pair = 1;
227 if (!strcmp(src->path, dst->path))
228 dp->score = rename_src[src_index].score;
231 rename_dst[dst_index].pair = dp;
235 * We sort the rename similarity matrix with the score, in descending
236 * order (the most similar first).
238 static int score_compare(const void *a_, const void *b_)
240 const struct diff_score *a = a_, *b = b_;
242 /* sink the unused ones to the bottom */
244 return (0 <= b->dst);
248 if (a->score == b->score)
249 return b->name_score - a->name_score;
251 return b->score - a->score;
254 struct file_similarity {
255 struct hashmap_entry entry;
257 struct diff_filespec *filespec;
260 static unsigned int hash_filespec(struct repository *r,
261 struct diff_filespec *filespec)
263 if (!filespec->oid_valid) {
264 if (diff_populate_filespec(r, filespec, 0))
266 hash_object_file(r->hash_algo, filespec->data, filespec->size,
267 "blob", &filespec->oid);
269 return oidhash(&filespec->oid);
272 static int find_identical_files(struct hashmap *srcs,
274 struct diff_options *options)
277 struct diff_filespec *target = rename_dst[dst_index].two;
278 struct file_similarity *p, *best = NULL;
279 int i = 100, best_score = -1;
280 unsigned int hash = hash_filespec(options->repo, target);
283 * Find the best source match for specified destination.
285 p = hashmap_get_entry_from_hash(srcs, hash, NULL,
286 struct file_similarity, entry);
287 hashmap_for_each_entry_from(srcs, p, entry) {
289 struct diff_filespec *source = p->filespec;
291 /* False hash collision? */
292 if (!oideq(&source->oid, &target->oid))
294 /* Non-regular files? If so, the modes must match! */
295 if (!S_ISREG(source->mode) || !S_ISREG(target->mode)) {
296 if (source->mode != target->mode)
299 /* Give higher scores to sources that haven't been used already */
300 score = !source->rename_used;
301 if (source->rename_used && options->detect_rename != DIFF_DETECT_COPY)
303 score += basename_same(source, target);
304 if (score > best_score) {
311 /* Too many identical alternatives? Pick one */
316 record_rename_pair(dst_index, best->index, MAX_SCORE);
322 static void insert_file_table(struct repository *r,
323 struct hashmap *table, int index,
324 struct diff_filespec *filespec)
326 struct file_similarity *entry = xmalloc(sizeof(*entry));
328 entry->index = index;
329 entry->filespec = filespec;
331 hashmap_entry_init(&entry->entry, hash_filespec(r, filespec));
332 hashmap_add(table, &entry->entry);
336 * Find exact renames first.
338 * The first round matches up the up-to-date entries,
339 * and then during the second round we try to match
340 * cache-dirty entries as well.
342 static int find_exact_renames(struct diff_options *options)
345 struct hashmap file_table;
347 /* Add all sources to the hash table in reverse order, because
348 * later on they will be retrieved in LIFO order.
350 hashmap_init(&file_table, NULL, NULL, rename_src_nr);
351 for (i = rename_src_nr-1; i >= 0; i--)
352 insert_file_table(options->repo,
354 rename_src[i].p->one);
356 /* Walk the destinations and find best source match */
357 for (i = 0; i < rename_dst_nr; i++)
358 renames += find_identical_files(&file_table, i, options);
360 /* Free the hash data structure and entries */
361 hashmap_free_entries(&file_table, struct file_similarity, entry);
366 #define NUM_CANDIDATE_PER_DST 4
367 static void record_if_better(struct diff_score m[], struct diff_score *o)
371 /* find the worst one */
373 for (i = 1; i < NUM_CANDIDATE_PER_DST; i++)
374 if (score_compare(&m[i], &m[worst]) > 0)
377 /* is it better than the worst one? */
378 if (score_compare(&m[worst], o) > 0)
384 * 0 if we are under the limit;
385 * 1 if we need to disable inexact rename detection;
386 * 2 if we would be under the limit if we were given -C instead of -C -C.
388 static int too_many_rename_candidates(int num_create,
389 struct diff_options *options)
391 int rename_limit = options->rename_limit;
392 int num_src = rename_src_nr;
395 options->needed_rename_limit = 0;
398 * This basically does a test for the rename matrix not
399 * growing larger than a "rename_limit" square matrix, ie:
401 * num_create * num_src > rename_limit * rename_limit
403 if (rename_limit <= 0)
404 rename_limit = 32767;
405 if ((num_create <= rename_limit || num_src <= rename_limit) &&
406 ((uint64_t)num_create * (uint64_t)num_src
407 <= (uint64_t)rename_limit * (uint64_t)rename_limit))
410 options->needed_rename_limit =
411 num_src > num_create ? num_src : num_create;
413 /* Are we running under -C -C? */
414 if (!options->flags.find_copies_harder)
417 /* Would we bust the limit if we were running under -C? */
418 for (num_src = i = 0; i < rename_src_nr; i++) {
419 if (diff_unmodified_pair(rename_src[i].p))
423 if ((num_create <= rename_limit || num_src <= rename_limit) &&
424 ((uint64_t)num_create * (uint64_t)num_src
425 <= (uint64_t)rename_limit * (uint64_t)rename_limit))
430 static int find_renames(struct diff_score *mx, int dst_cnt, int minimum_score, int copies)
434 for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) {
435 struct diff_rename_dst *dst;
437 if ((mx[i].dst < 0) ||
438 (mx[i].score < minimum_score))
439 break; /* there is no more usable pair. */
440 dst = &rename_dst[mx[i].dst];
442 continue; /* already done, either exact or fuzzy. */
443 if (!copies && rename_src[mx[i].src].p->one->rename_used)
445 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
451 void diffcore_rename(struct diff_options *options)
453 int detect_rename = options->detect_rename;
454 int minimum_score = options->rename_score;
455 struct diff_queue_struct *q = &diff_queued_diff;
456 struct diff_queue_struct outq;
457 struct diff_score *mx;
458 int i, j, rename_count, skip_unmodified = 0;
459 int num_create, dst_cnt;
460 struct progress *progress = NULL;
463 minimum_score = DEFAULT_RENAME_SCORE;
465 for (i = 0; i < q->nr; i++) {
466 struct diff_filepair *p = q->queue[i];
467 if (!DIFF_FILE_VALID(p->one)) {
468 if (!DIFF_FILE_VALID(p->two))
469 continue; /* unmerged */
470 else if (options->single_follow &&
471 strcmp(options->single_follow, p->two->path))
472 continue; /* not interested */
473 else if (!options->flags.rename_empty &&
474 is_empty_blob_oid(&p->two->oid))
476 else if (add_rename_dst(p->two) < 0) {
477 warning("skipping rename detection, detected"
478 " duplicate destination '%s'",
483 else if (!options->flags.rename_empty &&
484 is_empty_blob_oid(&p->one->oid))
486 else if (!DIFF_PAIR_UNMERGED(p) && !DIFF_FILE_VALID(p->two)) {
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
494 if (p->broken_pair && !p->score)
495 p->one->rename_used++;
496 register_rename_src(p);
498 else if (detect_rename == DIFF_DETECT_COPY) {
500 * Increment the "rename_used" score by
501 * one, to indicate ourselves as a user.
503 p->one->rename_used++;
504 register_rename_src(p);
507 if (rename_dst_nr == 0 || rename_src_nr == 0)
508 goto cleanup; /* nothing to do */
511 * We really want to cull the candidates list early
512 * with cheap tests in order to avoid doing deltas.
514 rename_count = find_exact_renames(options);
516 /* Did we only want exact renames? */
517 if (minimum_score == MAX_SCORE)
521 * Calculate how many renames are left (but all the source
522 * files still remain as options for rename/copies!)
524 num_create = (rename_dst_nr - rename_count);
530 switch (too_many_rename_candidates(num_create, options)) {
534 options->degraded_cc_to_c = 1;
541 if (options->show_rename_progress) {
542 progress = start_delayed_progress(
543 _("Performing inexact rename detection"),
544 (uint64_t)rename_dst_nr * (uint64_t)rename_src_nr);
547 mx = xcalloc(st_mult(NUM_CANDIDATE_PER_DST, num_create), sizeof(*mx));
548 for (dst_cnt = i = 0; i < rename_dst_nr; i++) {
549 struct diff_filespec *two = rename_dst[i].two;
550 struct diff_score *m;
552 if (rename_dst[i].pair)
553 continue; /* dealt with exact match already. */
555 m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST];
556 for (j = 0; j < NUM_CANDIDATE_PER_DST; j++)
559 for (j = 0; j < rename_src_nr; j++) {
560 struct diff_filespec *one = rename_src[j].p->one;
561 struct diff_score this_src;
563 if (skip_unmodified &&
564 diff_unmodified_pair(rename_src[j].p))
567 this_src.score = estimate_similarity(options->repo,
570 this_src.name_score = basename_same(one, two);
573 record_if_better(m, &this_src);
575 * Once we run estimate_similarity,
576 * We do not need the text anymore.
578 diff_free_filespec_blob(one);
579 diff_free_filespec_blob(two);
582 display_progress(progress, (uint64_t)(i+1)*(uint64_t)rename_src_nr);
584 stop_progress(&progress);
586 /* cost matrix sorted by most to least similar pair */
587 STABLE_QSORT(mx, dst_cnt * NUM_CANDIDATE_PER_DST, score_compare);
589 rename_count += find_renames(mx, dst_cnt, minimum_score, 0);
590 if (detect_rename == DIFF_DETECT_COPY)
591 rename_count += find_renames(mx, dst_cnt, minimum_score, 1);
595 /* At this point, we have found some renames and copies and they
596 * are recorded in rename_dst. The original list is still in *q.
598 DIFF_QUEUE_CLEAR(&outq);
599 for (i = 0; i < q->nr; i++) {
600 struct diff_filepair *p = q->queue[i];
601 struct diff_filepair *pair_to_free = NULL;
603 if (DIFF_PAIR_UNMERGED(p)) {
606 else if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
610 * We would output this create record if it has
611 * not been turned into a rename/copy already.
613 struct diff_rename_dst *dst = locate_rename_dst(p->two);
614 if (dst && dst->pair) {
615 diff_q(&outq, dst->pair);
619 /* no matching rename/copy source, so
620 * record this as a creation.
624 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) {
628 * We would output this delete record if:
630 * (1) this is a broken delete and the counterpart
631 * broken create remains in the output; or
632 * (2) this is not a broken delete, and rename_dst
633 * does not have a rename/copy to move p->one->path
636 * Otherwise, the counterpart broken create
637 * has been turned into a rename-edit; or
638 * delete did not have a matching create to
641 if (DIFF_PAIR_BROKEN(p)) {
643 struct diff_rename_dst *dst = locate_rename_dst(p->one);
644 if (dst && dst->pair)
645 /* counterpart is now rename/copy */
649 if (p->one->rename_used)
650 /* this path remains */
659 else if (!diff_unmodified_pair(p))
660 /* all the usual ones need to be kept */
663 /* no need to keep unmodified pairs */
667 diff_free_filepair(pair_to_free);
669 diff_debug_queue("done copying original", &outq);
673 diff_debug_queue("done collapsing", q);
675 for (i = 0; i < rename_dst_nr; i++)
676 free_filespec(rename_dst[i].two);
678 FREE_AND_NULL(rename_dst);
679 rename_dst_nr = rename_dst_alloc = 0;
680 FREE_AND_NULL(rename_src);
681 rename_src_nr = rename_src_alloc = 0;