2 * Copyright (C) 2005 Junio C Hamano
10 /* Table of rename/copy destinations */
12 static struct diff_rename_dst {
13 struct diff_filespec *two;
14 struct diff_filepair *pair;
16 static int rename_dst_nr, rename_dst_alloc;
18 static int find_rename_dst(struct diff_filespec *two)
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);
39 static struct diff_rename_dst *locate_rename_dst(struct diff_filespec *two)
41 int ofs = find_rename_dst(two);
42 return ofs < 0 ? NULL : &rename_dst[ofs];
46 * Returns 0 on success, -1 if we found a duplicate.
48 static int add_rename_dst(struct diff_filespec *two)
50 int first = find_rename_dst(two);
56 /* insert to make it at "first" */
57 ALLOC_GROW(rename_dst, rename_dst_nr + 1, rename_dst_alloc);
59 if (first < rename_dst_nr)
60 memmove(rename_dst + first + 1, rename_dst + first,
61 (rename_dst_nr - first - 1) * sizeof(*rename_dst));
62 rename_dst[first].two = alloc_filespec(two->path);
63 fill_filespec(rename_dst[first].two, two->sha1, two->sha1_valid, two->mode);
64 rename_dst[first].pair = NULL;
68 /* Table of rename/copy src files */
69 static struct diff_rename_src {
70 struct diff_filepair *p;
71 unsigned short score; /* to remember the break score */
73 static int rename_src_nr, rename_src_alloc;
75 static struct diff_rename_src *register_rename_src(struct diff_filepair *p)
78 struct diff_filespec *one = p->one;
79 unsigned short score = p->score;
83 while (last > first) {
84 int next = (last + first) >> 1;
85 struct diff_rename_src *src = &(rename_src[next]);
86 int cmp = strcmp(one->path, src->p->one->path);
96 /* insert to make it at "first" */
97 ALLOC_GROW(rename_src, rename_src_nr + 1, rename_src_alloc);
99 if (first < rename_src_nr)
100 memmove(rename_src + first + 1, rename_src + first,
101 (rename_src_nr - first - 1) * sizeof(*rename_src));
102 rename_src[first].p = p;
103 rename_src[first].score = score;
104 return &(rename_src[first]);
107 static int basename_same(struct diff_filespec *src, struct diff_filespec *dst)
109 int src_len = strlen(src->path), dst_len = strlen(dst->path);
110 while (src_len && dst_len) {
111 char c1 = src->path[--src_len];
112 char c2 = dst->path[--dst_len];
118 return (!src_len || src->path[src_len - 1] == '/') &&
119 (!dst_len || dst->path[dst_len - 1] == '/');
123 int src; /* index in rename_src */
124 int dst; /* index in rename_dst */
125 unsigned short score;
129 static int estimate_similarity(struct diff_filespec *src,
130 struct diff_filespec *dst,
133 /* src points at a file that existed in the original tree (or
134 * optionally a file in the destination tree) and dst points
135 * at a newly created file. They may be quite similar, in which
136 * case we want to say src is renamed to dst or src is copied into
137 * dst, and then some edit has been applied to dst.
139 * Compare them and return how similar they are, representing
140 * the score as an integer between 0 and MAX_SCORE.
142 * When there is an exact match, it is considered a better
143 * match than anything else; the destination does not even
144 * call into this function in that case.
146 unsigned long max_size, delta_size, base_size, src_copied, literal_added;
149 /* We deal only with regular files. Symlink renames are handled
150 * only when they are exact matches --- in other words, no edits
153 if (!S_ISREG(src->mode) || !S_ISREG(dst->mode))
157 * Need to check that source and destination sizes are
158 * filled in before comparing them.
160 * If we already have "cnt_data" filled in, we know it's
161 * all good (avoid checking the size for zero, as that
162 * is a possible size - we really should have a flag to
163 * say whether the size is valid or not!)
165 if (!src->cnt_data &&
166 diff_populate_filespec(src, CHECK_SIZE_ONLY))
168 if (!dst->cnt_data &&
169 diff_populate_filespec(dst, CHECK_SIZE_ONLY))
172 max_size = ((src->size > dst->size) ? src->size : dst->size);
173 base_size = ((src->size < dst->size) ? src->size : dst->size);
174 delta_size = max_size - base_size;
176 /* We would not consider edits that change the file size so
177 * drastically. delta_size must be smaller than
178 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
180 * Note that base_size == 0 case is handled here already
181 * and the final score computation below would not have a
182 * divide-by-zero issue.
184 if (max_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE)
187 if (!src->cnt_data && diff_populate_filespec(src, 0))
189 if (!dst->cnt_data && diff_populate_filespec(dst, 0))
192 if (diffcore_count_changes(src, dst,
193 &src->cnt_data, &dst->cnt_data,
194 &src_copied, &literal_added))
197 /* How similar are they?
198 * what percentage of material in dst are from source?
201 score = 0; /* should not happen */
203 score = (int)(src_copied * MAX_SCORE / max_size);
207 static void record_rename_pair(int dst_index, int src_index, int score)
209 struct diff_filespec *src, *dst;
210 struct diff_filepair *dp;
212 if (rename_dst[dst_index].pair)
213 die("internal error: dst already matched.");
215 src = rename_src[src_index].p->one;
219 dst = rename_dst[dst_index].two;
222 dp = diff_queue(NULL, src, dst);
223 dp->renamed_pair = 1;
224 if (!strcmp(src->path, dst->path))
225 dp->score = rename_src[src_index].score;
228 rename_dst[dst_index].pair = dp;
232 * We sort the rename similarity matrix with the score, in descending
233 * order (the most similar first).
235 static int score_compare(const void *a_, const void *b_)
237 const struct diff_score *a = a_, *b = b_;
239 /* sink the unused ones to the bottom */
241 return (0 <= b->dst);
245 if (a->score == b->score)
246 return b->name_score - a->name_score;
248 return b->score - a->score;
251 struct file_similarity {
252 struct hashmap_entry entry;
254 struct diff_filespec *filespec;
257 static unsigned int hash_filespec(struct diff_filespec *filespec)
259 if (!filespec->sha1_valid) {
260 if (diff_populate_filespec(filespec, 0))
262 hash_sha1_file(filespec->data, filespec->size, "blob", filespec->sha1);
264 return sha1hash(filespec->sha1);
267 static int find_identical_files(struct hashmap *srcs,
269 struct diff_options *options)
273 struct diff_filespec *target = rename_dst[dst_index].two;
274 struct file_similarity *p, *best = NULL;
275 int i = 100, best_score = -1;
278 * Find the best source match for specified destination.
280 p = hashmap_get_from_hash(srcs, hash_filespec(target), NULL);
281 for (; p; p = hashmap_get_next(srcs, p)) {
283 struct diff_filespec *source = p->filespec;
285 /* False hash collision? */
286 if (hashcmp(source->sha1, target->sha1))
288 /* Non-regular files? If so, the modes must match! */
289 if (!S_ISREG(source->mode) || !S_ISREG(target->mode)) {
290 if (source->mode != target->mode)
293 /* Give higher scores to sources that haven't been used already */
294 score = !source->rename_used;
295 if (source->rename_used && options->detect_rename != DIFF_DETECT_COPY)
297 score += basename_same(source, target);
298 if (score > best_score) {
305 /* Too many identical alternatives? Pick one */
310 record_rename_pair(dst_index, best->index, MAX_SCORE);
316 static void insert_file_table(struct hashmap *table, int index, struct diff_filespec *filespec)
318 struct file_similarity *entry = xmalloc(sizeof(*entry));
320 entry->index = index;
321 entry->filespec = filespec;
323 hashmap_entry_init(entry, hash_filespec(filespec));
324 hashmap_add(table, entry);
328 * Find exact renames first.
330 * The first round matches up the up-to-date entries,
331 * and then during the second round we try to match
332 * cache-dirty entries as well.
334 static int find_exact_renames(struct diff_options *options)
337 struct hashmap file_table;
339 /* Add all sources to the hash table in reverse order, because
340 * later on they will be retrieved in LIFO order.
342 hashmap_init(&file_table, NULL, rename_src_nr);
343 for (i = rename_src_nr-1; i >= 0; i--)
344 insert_file_table(&file_table, i, rename_src[i].p->one);
346 /* Walk the destinations and find best source match */
347 for (i = 0; i < rename_dst_nr; i++)
348 renames += find_identical_files(&file_table, i, options);
350 /* Free the hash data structure and entries */
351 hashmap_free(&file_table, 1);
356 #define NUM_CANDIDATE_PER_DST 4
357 static void record_if_better(struct diff_score m[], struct diff_score *o)
361 /* find the worst one */
363 for (i = 1; i < NUM_CANDIDATE_PER_DST; i++)
364 if (score_compare(&m[i], &m[worst]) > 0)
367 /* is it better than the worst one? */
368 if (score_compare(&m[worst], o) > 0)
374 * 0 if we are under the limit;
375 * 1 if we need to disable inexact rename detection;
376 * 2 if we would be under the limit if we were given -C instead of -C -C.
378 static int too_many_rename_candidates(int num_create,
379 struct diff_options *options)
381 int rename_limit = options->rename_limit;
382 int num_src = rename_src_nr;
385 options->needed_rename_limit = 0;
388 * This basically does a test for the rename matrix not
389 * growing larger than a "rename_limit" square matrix, ie:
391 * num_create * num_src > rename_limit * rename_limit
393 * but handles the potential overflow case specially (and we
394 * assume at least 32-bit integers)
396 if (rename_limit <= 0 || rename_limit > 32767)
397 rename_limit = 32767;
398 if ((num_create <= rename_limit || num_src <= rename_limit) &&
399 (num_create * num_src <= rename_limit * rename_limit))
402 options->needed_rename_limit =
403 num_src > num_create ? num_src : num_create;
405 /* Are we running under -C -C? */
406 if (!DIFF_OPT_TST(options, FIND_COPIES_HARDER))
409 /* Would we bust the limit if we were running under -C? */
410 for (num_src = i = 0; i < rename_src_nr; i++) {
411 if (diff_unmodified_pair(rename_src[i].p))
415 if ((num_create <= rename_limit || num_src <= rename_limit) &&
416 (num_create * num_src <= rename_limit * rename_limit))
421 static int find_renames(struct diff_score *mx, int dst_cnt, int minimum_score, int copies)
425 for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) {
426 struct diff_rename_dst *dst;
428 if ((mx[i].dst < 0) ||
429 (mx[i].score < minimum_score))
430 break; /* there is no more usable pair. */
431 dst = &rename_dst[mx[i].dst];
433 continue; /* already done, either exact or fuzzy. */
434 if (!copies && rename_src[mx[i].src].p->one->rename_used)
436 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
442 void diffcore_rename(struct diff_options *options)
444 int detect_rename = options->detect_rename;
445 int minimum_score = options->rename_score;
446 struct diff_queue_struct *q = &diff_queued_diff;
447 struct diff_queue_struct outq;
448 struct diff_score *mx;
449 int i, j, rename_count, skip_unmodified = 0;
450 int num_create, dst_cnt;
451 struct progress *progress = NULL;
454 minimum_score = DEFAULT_RENAME_SCORE;
456 for (i = 0; i < q->nr; i++) {
457 struct diff_filepair *p = q->queue[i];
458 if (!DIFF_FILE_VALID(p->one)) {
459 if (!DIFF_FILE_VALID(p->two))
460 continue; /* unmerged */
461 else if (options->single_follow &&
462 strcmp(options->single_follow, p->two->path))
463 continue; /* not interested */
464 else if (!DIFF_OPT_TST(options, RENAME_EMPTY) &&
465 is_empty_blob_sha1(p->two->sha1))
467 else if (add_rename_dst(p->two) < 0) {
468 warning("skipping rename detection, detected"
469 " duplicate destination '%s'",
474 else if (!DIFF_OPT_TST(options, RENAME_EMPTY) &&
475 is_empty_blob_sha1(p->one->sha1))
477 else if (!DIFF_PAIR_UNMERGED(p) && !DIFF_FILE_VALID(p->two)) {
479 * If the source is a broken "delete", and
480 * they did not really want to get broken,
481 * that means the source actually stays.
482 * So we increment the "rename_used" score
483 * by one, to indicate ourselves as a user
485 if (p->broken_pair && !p->score)
486 p->one->rename_used++;
487 register_rename_src(p);
489 else if (detect_rename == DIFF_DETECT_COPY) {
491 * Increment the "rename_used" score by
492 * one, to indicate ourselves as a user.
494 p->one->rename_used++;
495 register_rename_src(p);
498 if (rename_dst_nr == 0 || rename_src_nr == 0)
499 goto cleanup; /* nothing to do */
502 * We really want to cull the candidates list early
503 * with cheap tests in order to avoid doing deltas.
505 rename_count = find_exact_renames(options);
507 /* Did we only want exact renames? */
508 if (minimum_score == MAX_SCORE)
512 * Calculate how many renames are left (but all the source
513 * files still remain as options for rename/copies!)
515 num_create = (rename_dst_nr - rename_count);
521 switch (too_many_rename_candidates(num_create, options)) {
525 options->degraded_cc_to_c = 1;
532 if (options->show_rename_progress) {
533 progress = start_progress_delay(
534 _("Performing inexact rename detection"),
535 rename_dst_nr * rename_src_nr, 50, 1);
538 mx = xcalloc(st_mult(num_create, NUM_CANDIDATE_PER_DST), sizeof(*mx));
539 for (dst_cnt = i = 0; i < rename_dst_nr; i++) {
540 struct diff_filespec *two = rename_dst[i].two;
541 struct diff_score *m;
543 if (rename_dst[i].pair)
544 continue; /* dealt with exact match already. */
546 m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST];
547 for (j = 0; j < NUM_CANDIDATE_PER_DST; j++)
550 for (j = 0; j < rename_src_nr; j++) {
551 struct diff_filespec *one = rename_src[j].p->one;
552 struct diff_score this_src;
554 if (skip_unmodified &&
555 diff_unmodified_pair(rename_src[j].p))
558 this_src.score = estimate_similarity(one, two,
560 this_src.name_score = basename_same(one, two);
563 record_if_better(m, &this_src);
565 * Once we run estimate_similarity,
566 * We do not need the text anymore.
568 diff_free_filespec_blob(one);
569 diff_free_filespec_blob(two);
572 display_progress(progress, (i+1)*rename_src_nr);
574 stop_progress(&progress);
576 /* cost matrix sorted by most to least similar pair */
577 qsort(mx, dst_cnt * NUM_CANDIDATE_PER_DST, sizeof(*mx), score_compare);
579 rename_count += find_renames(mx, dst_cnt, minimum_score, 0);
580 if (detect_rename == DIFF_DETECT_COPY)
581 rename_count += find_renames(mx, dst_cnt, minimum_score, 1);
585 /* At this point, we have found some renames and copies and they
586 * are recorded in rename_dst. The original list is still in *q.
588 DIFF_QUEUE_CLEAR(&outq);
589 for (i = 0; i < q->nr; i++) {
590 struct diff_filepair *p = q->queue[i];
591 struct diff_filepair *pair_to_free = NULL;
593 if (DIFF_PAIR_UNMERGED(p)) {
596 else if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
600 * We would output this create record if it has
601 * not been turned into a rename/copy already.
603 struct diff_rename_dst *dst = locate_rename_dst(p->two);
604 if (dst && dst->pair) {
605 diff_q(&outq, dst->pair);
609 /* no matching rename/copy source, so
610 * record this as a creation.
614 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) {
618 * We would output this delete record if:
620 * (1) this is a broken delete and the counterpart
621 * broken create remains in the output; or
622 * (2) this is not a broken delete, and rename_dst
623 * does not have a rename/copy to move p->one->path
626 * Otherwise, the counterpart broken create
627 * has been turned into a rename-edit; or
628 * delete did not have a matching create to
631 if (DIFF_PAIR_BROKEN(p)) {
633 struct diff_rename_dst *dst = locate_rename_dst(p->one);
634 if (dst && dst->pair)
635 /* counterpart is now rename/copy */
639 if (p->one->rename_used)
640 /* this path remains */
649 else if (!diff_unmodified_pair(p))
650 /* all the usual ones need to be kept */
653 /* no need to keep unmodified pairs */
657 diff_free_filepair(pair_to_free);
659 diff_debug_queue("done copying original", &outq);
663 diff_debug_queue("done collapsing", q);
665 for (i = 0; i < rename_dst_nr; i++)
666 free_filespec(rename_dst[i].two);
670 rename_dst_nr = rename_dst_alloc = 0;
673 rename_src_nr = rename_src_alloc = 0;