2 * Copyright (C) 2005 Junio C Hamano
9 /* Table of rename/copy destinations */
11 static struct diff_rename_dst {
12 struct diff_filespec *two;
13 struct diff_filepair *pair;
15 static int rename_dst_nr, rename_dst_alloc;
17 static struct diff_rename_dst *locate_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 /* 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));
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]);
55 /* Table of rename/copy src files */
56 static struct diff_rename_src {
57 struct diff_filespec *one;
58 unsigned short score; /* to remember the break score */
60 static int rename_src_nr, rename_src_alloc;
62 static struct diff_rename_src *register_rename_src(struct diff_filespec *one,
69 while (last > first) {
70 int next = (last + first) >> 1;
71 struct diff_rename_src *src = &(rename_src[next]);
72 int cmp = strcmp(one->path, src->one->path);
82 /* insert to make it at "first" */
83 if (rename_src_alloc <= rename_src_nr) {
84 rename_src_alloc = alloc_nr(rename_src_alloc);
85 rename_src = xrealloc(rename_src,
86 rename_src_alloc * sizeof(*rename_src));
89 if (first < rename_src_nr)
90 memmove(rename_src + first + 1, rename_src + first,
91 (rename_src_nr - first - 1) * sizeof(*rename_src));
92 rename_src[first].one = one;
93 rename_src[first].score = score;
94 return &(rename_src[first]);
97 static int basename_same(struct diff_filespec *src, struct diff_filespec *dst)
99 int src_len = strlen(src->path), dst_len = strlen(dst->path);
100 while (src_len && dst_len) {
101 char c1 = src->path[--src_len];
102 char c2 = dst->path[--dst_len];
108 return (!src_len || src->path[src_len - 1] == '/') &&
109 (!dst_len || dst->path[dst_len - 1] == '/');
113 int src; /* index in rename_src */
114 int dst; /* index in rename_dst */
115 unsigned short score;
119 static int estimate_similarity(struct diff_filespec *src,
120 struct diff_filespec *dst,
123 /* src points at a file that existed in the original tree (or
124 * optionally a file in the destination tree) and dst points
125 * at a newly created file. They may be quite similar, in which
126 * case we want to say src is renamed to dst or src is copied into
127 * dst, and then some edit has been applied to dst.
129 * Compare them and return how similar they are, representing
130 * the score as an integer between 0 and MAX_SCORE.
132 * When there is an exact match, it is considered a better
133 * match than anything else; the destination does not even
134 * call into this function in that case.
136 unsigned long max_size, delta_size, base_size, src_copied, literal_added;
137 unsigned long delta_limit;
140 /* We deal only with regular files. Symlink renames are handled
141 * only when they are exact matches --- in other words, no edits
144 if (!S_ISREG(src->mode) || !S_ISREG(dst->mode))
148 * Need to check that source and destination sizes are
149 * filled in before comparing them.
151 * If we already have "cnt_data" filled in, we know it's
152 * all good (avoid checking the size for zero, as that
153 * is a possible size - we really should have a flag to
154 * say whether the size is valid or not!)
156 if (!src->cnt_data && diff_populate_filespec(src, 1))
158 if (!dst->cnt_data && diff_populate_filespec(dst, 1))
161 max_size = ((src->size > dst->size) ? src->size : dst->size);
162 base_size = ((src->size < dst->size) ? src->size : dst->size);
163 delta_size = max_size - base_size;
165 /* We would not consider edits that change the file size so
166 * drastically. delta_size must be smaller than
167 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
169 * Note that base_size == 0 case is handled here already
170 * and the final score computation below would not have a
171 * divide-by-zero issue.
173 if (base_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE)
176 if (!src->cnt_data && diff_populate_filespec(src, 0))
178 if (!dst->cnt_data && diff_populate_filespec(dst, 0))
181 delta_limit = (unsigned long)
182 (base_size * (MAX_SCORE-minimum_score) / MAX_SCORE);
183 if (diffcore_count_changes(src, dst,
184 &src->cnt_data, &dst->cnt_data,
186 &src_copied, &literal_added))
189 /* How similar are they?
190 * what percentage of material in dst are from source?
193 score = 0; /* should not happen */
195 score = (int)(src_copied * MAX_SCORE / max_size);
199 static void record_rename_pair(int dst_index, int src_index, int score)
201 struct diff_filespec *src, *dst;
202 struct diff_filepair *dp;
204 if (rename_dst[dst_index].pair)
205 die("internal error: dst already matched.");
207 src = rename_src[src_index].one;
211 dst = rename_dst[dst_index].two;
214 dp = diff_queue(NULL, src, dst);
215 dp->renamed_pair = 1;
216 if (!strcmp(src->path, dst->path))
217 dp->score = rename_src[src_index].score;
220 rename_dst[dst_index].pair = dp;
224 * We sort the rename similarity matrix with the score, in descending
225 * order (the most similar first).
227 static int score_compare(const void *a_, const void *b_)
229 const struct diff_score *a = a_, *b = b_;
231 /* sink the unused ones to the bottom */
233 return (0 <= b->dst);
237 if (a->score == b->score)
238 return b->name_score - a->name_score;
240 return b->score - a->score;
243 struct file_similarity {
245 struct diff_filespec *filespec;
246 struct file_similarity *next;
249 static int find_identical_files(struct file_similarity *src,
250 struct file_similarity *dst,
251 struct diff_options *options)
256 * Walk over all the destinations ...
259 struct diff_filespec *target = dst->filespec;
260 struct file_similarity *p, *best;
261 int i = 100, best_score = -1;
264 * .. to find the best source match
267 for (p = src; p; p = p->next) {
269 struct diff_filespec *source = p->filespec;
271 /* False hash collision? */
272 if (hashcmp(source->sha1, target->sha1))
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)
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) {
289 /* Too many identical alternatives? Pick one */
294 record_rename_pair(dst->index, best->index, MAX_SCORE);
297 } while ((dst = dst->next) != NULL);
301 static void free_similarity_list(struct file_similarity *p)
304 struct file_similarity *entry = p;
310 static int find_same_files(void *ptr, void *data)
313 struct file_similarity *p = ptr;
314 struct file_similarity *src = NULL, *dst = NULL;
315 struct diff_options *options = data;
317 /* Split the hash list up into sources and destinations */
319 struct file_similarity *entry = p;
321 if (entry->src_dst < 0) {
331 * If we have both sources *and* destinations, see if
332 * we can match them up
334 ret = (src && dst) ? find_identical_files(src, dst, options) : 0;
336 /* Free the hashes and return the number of renames found */
337 free_similarity_list(src);
338 free_similarity_list(dst);
342 static unsigned int hash_filespec(struct diff_filespec *filespec)
345 if (!filespec->sha1_valid) {
346 if (diff_populate_filespec(filespec, 0))
348 hash_sha1_file(filespec->data, filespec->size, "blob", filespec->sha1);
350 memcpy(&hash, filespec->sha1, sizeof(hash));
354 static void insert_file_table(struct hash_table *table, int src_dst, int index, struct diff_filespec *filespec)
358 struct file_similarity *entry = xmalloc(sizeof(*entry));
360 entry->src_dst = src_dst;
361 entry->index = index;
362 entry->filespec = filespec;
365 hash = hash_filespec(filespec);
366 pos = insert_hash(hash, entry, table);
368 /* We already had an entry there? */
376 * Find exact renames first.
378 * The first round matches up the up-to-date entries,
379 * and then during the second round we try to match
380 * cache-dirty entries as well.
382 static int find_exact_renames(struct diff_options *options)
385 struct hash_table file_table;
387 init_hash(&file_table);
388 for (i = 0; i < rename_src_nr; i++)
389 insert_file_table(&file_table, -1, i, rename_src[i].one);
391 for (i = 0; i < rename_dst_nr; i++)
392 insert_file_table(&file_table, 1, i, rename_dst[i].two);
394 /* Find the renames */
395 i = for_each_hash(&file_table, find_same_files, options);
397 /* .. and free the hash data structure */
398 free_hash(&file_table);
403 #define NUM_CANDIDATE_PER_DST 4
404 static void record_if_better(struct diff_score m[], struct diff_score *o)
408 /* find the worst one */
410 for (i = 1; i < NUM_CANDIDATE_PER_DST; i++)
411 if (score_compare(&m[i], &m[worst]) > 0)
414 /* is it better than the worst one? */
415 if (score_compare(&m[worst], o) > 0)
419 void diffcore_rename(struct diff_options *options)
421 int detect_rename = options->detect_rename;
422 int minimum_score = options->rename_score;
423 int rename_limit = options->rename_limit;
424 struct diff_queue_struct *q = &diff_queued_diff;
425 struct diff_queue_struct outq;
426 struct diff_score *mx;
427 int i, j, rename_count;
428 int num_create, num_src, dst_cnt;
431 minimum_score = DEFAULT_RENAME_SCORE;
433 for (i = 0; i < q->nr; i++) {
434 struct diff_filepair *p = q->queue[i];
435 if (!DIFF_FILE_VALID(p->one)) {
436 if (!DIFF_FILE_VALID(p->two))
437 continue; /* unmerged */
438 else if (options->single_follow &&
439 strcmp(options->single_follow, p->two->path))
440 continue; /* not interested */
442 locate_rename_dst(p->two, 1);
444 else if (!DIFF_FILE_VALID(p->two)) {
446 * If the source is a broken "delete", and
447 * they did not really want to get broken,
448 * that means the source actually stays.
449 * So we increment the "rename_used" score
450 * by one, to indicate ourselves as a user
452 if (p->broken_pair && !p->score)
453 p->one->rename_used++;
454 register_rename_src(p->one, p->score);
456 else if (detect_rename == DIFF_DETECT_COPY) {
458 * Increment the "rename_used" score by
459 * one, to indicate ourselves as a user.
461 p->one->rename_used++;
462 register_rename_src(p->one, p->score);
465 if (rename_dst_nr == 0 || rename_src_nr == 0)
466 goto cleanup; /* nothing to do */
469 * We really want to cull the candidates list early
470 * with cheap tests in order to avoid doing deltas.
472 rename_count = find_exact_renames(options);
474 /* Did we only want exact renames? */
475 if (minimum_score == MAX_SCORE)
479 * Calculate how many renames are left (but all the source
480 * files still remain as options for rename/copies!)
482 num_create = (rename_dst_nr - rename_count);
483 num_src = rename_src_nr;
490 * This basically does a test for the rename matrix not
491 * growing larger than a "rename_limit" square matrix, ie:
493 * num_create * num_src > rename_limit * rename_limit
495 * but handles the potential overflow case specially (and we
496 * assume at least 32-bit integers)
498 if (rename_limit <= 0 || rename_limit > 32767)
499 rename_limit = 32767;
500 if ((num_create > rename_limit && num_src > rename_limit) ||
501 (num_create * num_src > rename_limit * rename_limit)) {
502 if (options->warn_on_too_large_rename)
503 warning("too many files (created: %d deleted: %d), skipping inexact rename detection", num_create, num_src);
507 mx = xcalloc(num_create * NUM_CANDIDATE_PER_DST, sizeof(*mx));
508 for (dst_cnt = i = 0; i < rename_dst_nr; i++) {
509 struct diff_filespec *two = rename_dst[i].two;
510 struct diff_score *m;
512 if (rename_dst[i].pair)
513 continue; /* dealt with exact match already. */
515 m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST];
516 for (j = 0; j < NUM_CANDIDATE_PER_DST; j++)
519 for (j = 0; j < rename_src_nr; j++) {
520 struct diff_filespec *one = rename_src[j].one;
521 struct diff_score this_src;
522 this_src.score = estimate_similarity(one, two,
524 this_src.name_score = basename_same(one, two);
527 record_if_better(m, &this_src);
529 * Once we run estimate_similarity,
530 * We do not need the text anymore.
532 diff_free_filespec_blob(one);
533 diff_free_filespec_blob(two);
538 /* cost matrix sorted by most to least similar pair */
539 qsort(mx, dst_cnt * NUM_CANDIDATE_PER_DST, sizeof(*mx), score_compare);
541 for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) {
542 struct diff_rename_dst *dst;
544 if ((mx[i].dst < 0) ||
545 (mx[i].score < minimum_score))
546 break; /* there is no more usable pair. */
547 dst = &rename_dst[mx[i].dst];
549 continue; /* already done, either exact or fuzzy. */
550 if (rename_src[mx[i].src].one->rename_used)
552 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
556 for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) {
557 struct diff_rename_dst *dst;
559 if ((mx[i].dst < 0) ||
560 (mx[i].score < minimum_score))
561 break; /* there is no more usable pair. */
562 dst = &rename_dst[mx[i].dst];
564 continue; /* already done, either exact or fuzzy. */
565 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
571 /* At this point, we have found some renames and copies and they
572 * are recorded in rename_dst. The original list is still in *q.
574 DIFF_QUEUE_CLEAR(&outq);
575 for (i = 0; i < q->nr; i++) {
576 struct diff_filepair *p = q->queue[i];
577 struct diff_filepair *pair_to_free = NULL;
579 if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
583 * We would output this create record if it has
584 * not been turned into a rename/copy already.
586 struct diff_rename_dst *dst =
587 locate_rename_dst(p->two, 0);
588 if (dst && dst->pair) {
589 diff_q(&outq, dst->pair);
593 /* no matching rename/copy source, so
594 * record this as a creation.
598 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) {
602 * We would output this delete record if:
604 * (1) this is a broken delete and the counterpart
605 * broken create remains in the output; or
606 * (2) this is not a broken delete, and rename_dst
607 * does not have a rename/copy to move p->one->path
610 * Otherwise, the counterpart broken create
611 * has been turned into a rename-edit; or
612 * delete did not have a matching create to
615 if (DIFF_PAIR_BROKEN(p)) {
617 struct diff_rename_dst *dst =
618 locate_rename_dst(p->one, 0);
619 if (dst && dst->pair)
620 /* counterpart is now rename/copy */
624 if (p->one->rename_used)
625 /* this path remains */
634 else if (!diff_unmodified_pair(p))
635 /* all the usual ones need to be kept */
638 /* no need to keep unmodified pairs */
642 diff_free_filepair(pair_to_free);
644 diff_debug_queue("done copying original", &outq);
648 diff_debug_queue("done collapsing", q);
650 for (i = 0; i < rename_dst_nr; i++)
651 free_filespec(rename_dst[i].two);
655 rename_dst_nr = rename_dst_alloc = 0;
658 rename_src_nr = rename_src_alloc = 0;