]> git.ipfire.org Git - thirdparty/git.git/blob - diffcore-rename.c
Merge branch 'jt/t5500-unflake'
[thirdparty/git.git] / diffcore-rename.c
1 /*
2 *
3 * Copyright (C) 2005 Junio C Hamano
4 */
5 #include "cache.h"
6 #include "diff.h"
7 #include "diffcore.h"
8 #include "object-store.h"
9 #include "hashmap.h"
10 #include "progress.h"
11 #include "promisor-remote.h"
12
13 /* Table of rename/copy destinations */
14
15 static struct diff_rename_dst {
16 struct diff_filespec *two;
17 struct diff_filepair *pair;
18 } *rename_dst;
19 static int rename_dst_nr, rename_dst_alloc;
20
21 static int find_rename_dst(struct diff_filespec *two)
22 {
23 int first, last;
24
25 first = 0;
26 last = rename_dst_nr;
27 while (last > first) {
28 int next = first + ((last - first) >> 1);
29 struct diff_rename_dst *dst = &(rename_dst[next]);
30 int cmp = strcmp(two->path, dst->two->path);
31 if (!cmp)
32 return next;
33 if (cmp < 0) {
34 last = next;
35 continue;
36 }
37 first = next+1;
38 }
39 return -first - 1;
40 }
41
42 static struct diff_rename_dst *locate_rename_dst(struct diff_filespec *two)
43 {
44 int ofs = find_rename_dst(two);
45 return ofs < 0 ? NULL : &rename_dst[ofs];
46 }
47
48 /*
49 * Returns 0 on success, -1 if we found a duplicate.
50 */
51 static int add_rename_dst(struct diff_filespec *two)
52 {
53 int first = find_rename_dst(two);
54
55 if (first >= 0)
56 return -1;
57 first = -first - 1;
58
59 /* insert to make it at "first" */
60 ALLOC_GROW(rename_dst, rename_dst_nr + 1, rename_dst_alloc);
61 rename_dst_nr++;
62 if (first < rename_dst_nr)
63 MOVE_ARRAY(rename_dst + first + 1, rename_dst + first,
64 rename_dst_nr - first - 1);
65 rename_dst[first].two = alloc_filespec(two->path);
66 fill_filespec(rename_dst[first].two, &two->oid, two->oid_valid,
67 two->mode);
68 rename_dst[first].pair = NULL;
69 return 0;
70 }
71
72 /* Table of rename/copy src files */
73 static struct diff_rename_src {
74 struct diff_filepair *p;
75 unsigned short score; /* to remember the break score */
76 } *rename_src;
77 static int rename_src_nr, rename_src_alloc;
78
79 static struct diff_rename_src *register_rename_src(struct diff_filepair *p)
80 {
81 int first, last;
82 struct diff_filespec *one = p->one;
83 unsigned short score = p->score;
84
85 first = 0;
86 last = rename_src_nr;
87 while (last > first) {
88 int next = first + ((last - first) >> 1);
89 struct diff_rename_src *src = &(rename_src[next]);
90 int cmp = strcmp(one->path, src->p->one->path);
91 if (!cmp)
92 return src;
93 if (cmp < 0) {
94 last = next;
95 continue;
96 }
97 first = next+1;
98 }
99
100 /* insert to make it at "first" */
101 ALLOC_GROW(rename_src, rename_src_nr + 1, rename_src_alloc);
102 rename_src_nr++;
103 if (first < rename_src_nr)
104 MOVE_ARRAY(rename_src + first + 1, rename_src + first,
105 rename_src_nr - first - 1);
106 rename_src[first].p = p;
107 rename_src[first].score = score;
108 return &(rename_src[first]);
109 }
110
111 static int basename_same(struct diff_filespec *src, struct diff_filespec *dst)
112 {
113 int src_len = strlen(src->path), dst_len = strlen(dst->path);
114 while (src_len && dst_len) {
115 char c1 = src->path[--src_len];
116 char c2 = dst->path[--dst_len];
117 if (c1 != c2)
118 return 0;
119 if (c1 == '/')
120 return 1;
121 }
122 return (!src_len || src->path[src_len - 1] == '/') &&
123 (!dst_len || dst->path[dst_len - 1] == '/');
124 }
125
126 struct diff_score {
127 int src; /* index in rename_src */
128 int dst; /* index in rename_dst */
129 unsigned short score;
130 short name_score;
131 };
132
133 struct prefetch_options {
134 struct repository *repo;
135 int skip_unmodified;
136 };
137 static void prefetch(void *prefetch_options)
138 {
139 struct prefetch_options *options = prefetch_options;
140 int i;
141 struct oid_array to_fetch = OID_ARRAY_INIT;
142
143 for (i = 0; i < rename_dst_nr; i++) {
144 if (rename_dst[i].pair)
145 /*
146 * The loop in diffcore_rename() will not need these
147 * blobs, so skip prefetching.
148 */
149 continue; /* already found exact match */
150 diff_add_if_missing(options->repo, &to_fetch,
151 rename_dst[i].two);
152 }
153 for (i = 0; i < rename_src_nr; i++) {
154 if (options->skip_unmodified &&
155 diff_unmodified_pair(rename_src[i].p))
156 /*
157 * The loop in diffcore_rename() will not need these
158 * blobs, so skip prefetching.
159 */
160 continue;
161 diff_add_if_missing(options->repo, &to_fetch,
162 rename_src[i].p->one);
163 }
164 promisor_remote_get_direct(options->repo, to_fetch.oid, to_fetch.nr);
165 oid_array_clear(&to_fetch);
166 }
167
168 static int estimate_similarity(struct repository *r,
169 struct diff_filespec *src,
170 struct diff_filespec *dst,
171 int minimum_score,
172 int skip_unmodified)
173 {
174 /* src points at a file that existed in the original tree (or
175 * optionally a file in the destination tree) and dst points
176 * at a newly created file. They may be quite similar, in which
177 * case we want to say src is renamed to dst or src is copied into
178 * dst, and then some edit has been applied to dst.
179 *
180 * Compare them and return how similar they are, representing
181 * the score as an integer between 0 and MAX_SCORE.
182 *
183 * When there is an exact match, it is considered a better
184 * match than anything else; the destination does not even
185 * call into this function in that case.
186 */
187 unsigned long max_size, delta_size, base_size, src_copied, literal_added;
188 int score;
189 struct diff_populate_filespec_options dpf_options = {
190 .check_size_only = 1
191 };
192 struct prefetch_options prefetch_options = {r, skip_unmodified};
193
194 if (r == the_repository && has_promisor_remote()) {
195 dpf_options.missing_object_cb = prefetch;
196 dpf_options.missing_object_data = &prefetch_options;
197 }
198
199 /* We deal only with regular files. Symlink renames are handled
200 * only when they are exact matches --- in other words, no edits
201 * after renaming.
202 */
203 if (!S_ISREG(src->mode) || !S_ISREG(dst->mode))
204 return 0;
205
206 /*
207 * Need to check that source and destination sizes are
208 * filled in before comparing them.
209 *
210 * If we already have "cnt_data" filled in, we know it's
211 * all good (avoid checking the size for zero, as that
212 * is a possible size - we really should have a flag to
213 * say whether the size is valid or not!)
214 */
215 if (!src->cnt_data &&
216 diff_populate_filespec(r, src, &dpf_options))
217 return 0;
218 if (!dst->cnt_data &&
219 diff_populate_filespec(r, dst, &dpf_options))
220 return 0;
221
222 max_size = ((src->size > dst->size) ? src->size : dst->size);
223 base_size = ((src->size < dst->size) ? src->size : dst->size);
224 delta_size = max_size - base_size;
225
226 /* We would not consider edits that change the file size so
227 * drastically. delta_size must be smaller than
228 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
229 *
230 * Note that base_size == 0 case is handled here already
231 * and the final score computation below would not have a
232 * divide-by-zero issue.
233 */
234 if (max_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE)
235 return 0;
236
237 dpf_options.check_size_only = 0;
238
239 if (!src->cnt_data && diff_populate_filespec(r, src, &dpf_options))
240 return 0;
241 if (!dst->cnt_data && diff_populate_filespec(r, dst, &dpf_options))
242 return 0;
243
244 if (diffcore_count_changes(r, src, dst,
245 &src->cnt_data, &dst->cnt_data,
246 &src_copied, &literal_added))
247 return 0;
248
249 /* How similar are they?
250 * what percentage of material in dst are from source?
251 */
252 if (!dst->size)
253 score = 0; /* should not happen */
254 else
255 score = (int)(src_copied * MAX_SCORE / max_size);
256 return score;
257 }
258
259 static void record_rename_pair(int dst_index, int src_index, int score)
260 {
261 struct diff_filespec *src, *dst;
262 struct diff_filepair *dp;
263
264 if (rename_dst[dst_index].pair)
265 die("internal error: dst already matched.");
266
267 src = rename_src[src_index].p->one;
268 src->rename_used++;
269 src->count++;
270
271 dst = rename_dst[dst_index].two;
272 dst->count++;
273
274 dp = diff_queue(NULL, src, dst);
275 dp->renamed_pair = 1;
276 if (!strcmp(src->path, dst->path))
277 dp->score = rename_src[src_index].score;
278 else
279 dp->score = score;
280 rename_dst[dst_index].pair = dp;
281 }
282
283 /*
284 * We sort the rename similarity matrix with the score, in descending
285 * order (the most similar first).
286 */
287 static int score_compare(const void *a_, const void *b_)
288 {
289 const struct diff_score *a = a_, *b = b_;
290
291 /* sink the unused ones to the bottom */
292 if (a->dst < 0)
293 return (0 <= b->dst);
294 else if (b->dst < 0)
295 return -1;
296
297 if (a->score == b->score)
298 return b->name_score - a->name_score;
299
300 return b->score - a->score;
301 }
302
303 struct file_similarity {
304 struct hashmap_entry entry;
305 int index;
306 struct diff_filespec *filespec;
307 };
308
309 static unsigned int hash_filespec(struct repository *r,
310 struct diff_filespec *filespec)
311 {
312 if (!filespec->oid_valid) {
313 if (diff_populate_filespec(r, filespec, NULL))
314 return 0;
315 hash_object_file(r->hash_algo, filespec->data, filespec->size,
316 "blob", &filespec->oid);
317 }
318 return oidhash(&filespec->oid);
319 }
320
321 static int find_identical_files(struct hashmap *srcs,
322 int dst_index,
323 struct diff_options *options)
324 {
325 int renames = 0;
326 struct diff_filespec *target = rename_dst[dst_index].two;
327 struct file_similarity *p, *best = NULL;
328 int i = 100, best_score = -1;
329 unsigned int hash = hash_filespec(options->repo, target);
330
331 /*
332 * Find the best source match for specified destination.
333 */
334 p = hashmap_get_entry_from_hash(srcs, hash, NULL,
335 struct file_similarity, entry);
336 hashmap_for_each_entry_from(srcs, p, entry) {
337 int score;
338 struct diff_filespec *source = p->filespec;
339
340 /* False hash collision? */
341 if (!oideq(&source->oid, &target->oid))
342 continue;
343 /* Non-regular files? If so, the modes must match! */
344 if (!S_ISREG(source->mode) || !S_ISREG(target->mode)) {
345 if (source->mode != target->mode)
346 continue;
347 }
348 /* Give higher scores to sources that haven't been used already */
349 score = !source->rename_used;
350 if (source->rename_used && options->detect_rename != DIFF_DETECT_COPY)
351 continue;
352 score += basename_same(source, target);
353 if (score > best_score) {
354 best = p;
355 best_score = score;
356 if (score == 2)
357 break;
358 }
359
360 /* Too many identical alternatives? Pick one */
361 if (!--i)
362 break;
363 }
364 if (best) {
365 record_rename_pair(dst_index, best->index, MAX_SCORE);
366 renames++;
367 }
368 return renames;
369 }
370
371 static void insert_file_table(struct repository *r,
372 struct hashmap *table, int index,
373 struct diff_filespec *filespec)
374 {
375 struct file_similarity *entry = xmalloc(sizeof(*entry));
376
377 entry->index = index;
378 entry->filespec = filespec;
379
380 hashmap_entry_init(&entry->entry, hash_filespec(r, filespec));
381 hashmap_add(table, &entry->entry);
382 }
383
384 /*
385 * Find exact renames first.
386 *
387 * The first round matches up the up-to-date entries,
388 * and then during the second round we try to match
389 * cache-dirty entries as well.
390 */
391 static int find_exact_renames(struct diff_options *options)
392 {
393 int i, renames = 0;
394 struct hashmap file_table;
395
396 /* Add all sources to the hash table in reverse order, because
397 * later on they will be retrieved in LIFO order.
398 */
399 hashmap_init(&file_table, NULL, NULL, rename_src_nr);
400 for (i = rename_src_nr-1; i >= 0; i--)
401 insert_file_table(options->repo,
402 &file_table, i,
403 rename_src[i].p->one);
404
405 /* Walk the destinations and find best source match */
406 for (i = 0; i < rename_dst_nr; i++)
407 renames += find_identical_files(&file_table, i, options);
408
409 /* Free the hash data structure and entries */
410 hashmap_free_entries(&file_table, struct file_similarity, entry);
411
412 return renames;
413 }
414
415 #define NUM_CANDIDATE_PER_DST 4
416 static void record_if_better(struct diff_score m[], struct diff_score *o)
417 {
418 int i, worst;
419
420 /* find the worst one */
421 worst = 0;
422 for (i = 1; i < NUM_CANDIDATE_PER_DST; i++)
423 if (score_compare(&m[i], &m[worst]) > 0)
424 worst = i;
425
426 /* is it better than the worst one? */
427 if (score_compare(&m[worst], o) > 0)
428 m[worst] = *o;
429 }
430
431 /*
432 * Returns:
433 * 0 if we are under the limit;
434 * 1 if we need to disable inexact rename detection;
435 * 2 if we would be under the limit if we were given -C instead of -C -C.
436 */
437 static int too_many_rename_candidates(int num_create,
438 struct diff_options *options)
439 {
440 int rename_limit = options->rename_limit;
441 int num_src = rename_src_nr;
442 int i;
443
444 options->needed_rename_limit = 0;
445
446 /*
447 * This basically does a test for the rename matrix not
448 * growing larger than a "rename_limit" square matrix, ie:
449 *
450 * num_create * num_src > rename_limit * rename_limit
451 */
452 if (rename_limit <= 0)
453 rename_limit = 32767;
454 if ((num_create <= rename_limit || num_src <= rename_limit) &&
455 ((uint64_t)num_create * (uint64_t)num_src
456 <= (uint64_t)rename_limit * (uint64_t)rename_limit))
457 return 0;
458
459 options->needed_rename_limit =
460 num_src > num_create ? num_src : num_create;
461
462 /* Are we running under -C -C? */
463 if (!options->flags.find_copies_harder)
464 return 1;
465
466 /* Would we bust the limit if we were running under -C? */
467 for (num_src = i = 0; i < rename_src_nr; i++) {
468 if (diff_unmodified_pair(rename_src[i].p))
469 continue;
470 num_src++;
471 }
472 if ((num_create <= rename_limit || num_src <= rename_limit) &&
473 ((uint64_t)num_create * (uint64_t)num_src
474 <= (uint64_t)rename_limit * (uint64_t)rename_limit))
475 return 2;
476 return 1;
477 }
478
479 static int find_renames(struct diff_score *mx, int dst_cnt, int minimum_score, int copies)
480 {
481 int count = 0, i;
482
483 for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) {
484 struct diff_rename_dst *dst;
485
486 if ((mx[i].dst < 0) ||
487 (mx[i].score < minimum_score))
488 break; /* there is no more usable pair. */
489 dst = &rename_dst[mx[i].dst];
490 if (dst->pair)
491 continue; /* already done, either exact or fuzzy. */
492 if (!copies && rename_src[mx[i].src].p->one->rename_used)
493 continue;
494 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
495 count++;
496 }
497 return count;
498 }
499
500 void diffcore_rename(struct diff_options *options)
501 {
502 int detect_rename = options->detect_rename;
503 int minimum_score = options->rename_score;
504 struct diff_queue_struct *q = &diff_queued_diff;
505 struct diff_queue_struct outq;
506 struct diff_score *mx;
507 int i, j, rename_count, skip_unmodified = 0;
508 int num_create, dst_cnt;
509 struct progress *progress = NULL;
510
511 if (!minimum_score)
512 minimum_score = DEFAULT_RENAME_SCORE;
513
514 for (i = 0; i < q->nr; i++) {
515 struct diff_filepair *p = q->queue[i];
516 if (!DIFF_FILE_VALID(p->one)) {
517 if (!DIFF_FILE_VALID(p->two))
518 continue; /* unmerged */
519 else if (options->single_follow &&
520 strcmp(options->single_follow, p->two->path))
521 continue; /* not interested */
522 else if (!options->flags.rename_empty &&
523 is_empty_blob_oid(&p->two->oid))
524 continue;
525 else if (add_rename_dst(p->two) < 0) {
526 warning("skipping rename detection, detected"
527 " duplicate destination '%s'",
528 p->two->path);
529 goto cleanup;
530 }
531 }
532 else if (!options->flags.rename_empty &&
533 is_empty_blob_oid(&p->one->oid))
534 continue;
535 else if (!DIFF_PAIR_UNMERGED(p) && !DIFF_FILE_VALID(p->two)) {
536 /*
537 * If the source is a broken "delete", and
538 * they did not really want to get broken,
539 * that means the source actually stays.
540 * So we increment the "rename_used" score
541 * by one, to indicate ourselves as a user
542 */
543 if (p->broken_pair && !p->score)
544 p->one->rename_used++;
545 register_rename_src(p);
546 }
547 else if (detect_rename == DIFF_DETECT_COPY) {
548 /*
549 * Increment the "rename_used" score by
550 * one, to indicate ourselves as a user.
551 */
552 p->one->rename_used++;
553 register_rename_src(p);
554 }
555 }
556 if (rename_dst_nr == 0 || rename_src_nr == 0)
557 goto cleanup; /* nothing to do */
558
559 /*
560 * We really want to cull the candidates list early
561 * with cheap tests in order to avoid doing deltas.
562 */
563 rename_count = find_exact_renames(options);
564
565 /* Did we only want exact renames? */
566 if (minimum_score == MAX_SCORE)
567 goto cleanup;
568
569 /*
570 * Calculate how many renames are left (but all the source
571 * files still remain as options for rename/copies!)
572 */
573 num_create = (rename_dst_nr - rename_count);
574
575 /* All done? */
576 if (!num_create)
577 goto cleanup;
578
579 switch (too_many_rename_candidates(num_create, options)) {
580 case 1:
581 goto cleanup;
582 case 2:
583 options->degraded_cc_to_c = 1;
584 skip_unmodified = 1;
585 break;
586 default:
587 break;
588 }
589
590 if (options->show_rename_progress) {
591 progress = start_delayed_progress(
592 _("Performing inexact rename detection"),
593 (uint64_t)rename_dst_nr * (uint64_t)rename_src_nr);
594 }
595
596 mx = xcalloc(st_mult(NUM_CANDIDATE_PER_DST, num_create), sizeof(*mx));
597 for (dst_cnt = i = 0; i < rename_dst_nr; i++) {
598 struct diff_filespec *two = rename_dst[i].two;
599 struct diff_score *m;
600
601 if (rename_dst[i].pair)
602 continue; /* dealt with exact match already. */
603
604 m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST];
605 for (j = 0; j < NUM_CANDIDATE_PER_DST; j++)
606 m[j].dst = -1;
607
608 for (j = 0; j < rename_src_nr; j++) {
609 struct diff_filespec *one = rename_src[j].p->one;
610 struct diff_score this_src;
611
612 if (skip_unmodified &&
613 diff_unmodified_pair(rename_src[j].p))
614 continue;
615
616 this_src.score = estimate_similarity(options->repo,
617 one, two,
618 minimum_score,
619 skip_unmodified);
620 this_src.name_score = basename_same(one, two);
621 this_src.dst = i;
622 this_src.src = j;
623 record_if_better(m, &this_src);
624 /*
625 * Once we run estimate_similarity,
626 * We do not need the text anymore.
627 */
628 diff_free_filespec_blob(one);
629 diff_free_filespec_blob(two);
630 }
631 dst_cnt++;
632 display_progress(progress, (uint64_t)(i+1)*(uint64_t)rename_src_nr);
633 }
634 stop_progress(&progress);
635
636 /* cost matrix sorted by most to least similar pair */
637 STABLE_QSORT(mx, dst_cnt * NUM_CANDIDATE_PER_DST, score_compare);
638
639 rename_count += find_renames(mx, dst_cnt, minimum_score, 0);
640 if (detect_rename == DIFF_DETECT_COPY)
641 rename_count += find_renames(mx, dst_cnt, minimum_score, 1);
642 free(mx);
643
644 cleanup:
645 /* At this point, we have found some renames and copies and they
646 * are recorded in rename_dst. The original list is still in *q.
647 */
648 DIFF_QUEUE_CLEAR(&outq);
649 for (i = 0; i < q->nr; i++) {
650 struct diff_filepair *p = q->queue[i];
651 struct diff_filepair *pair_to_free = NULL;
652
653 if (DIFF_PAIR_UNMERGED(p)) {
654 diff_q(&outq, p);
655 }
656 else if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
657 /*
658 * Creation
659 *
660 * We would output this create record if it has
661 * not been turned into a rename/copy already.
662 */
663 struct diff_rename_dst *dst = locate_rename_dst(p->two);
664 if (dst && dst->pair) {
665 diff_q(&outq, dst->pair);
666 pair_to_free = p;
667 }
668 else
669 /* no matching rename/copy source, so
670 * record this as a creation.
671 */
672 diff_q(&outq, p);
673 }
674 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) {
675 /*
676 * Deletion
677 *
678 * We would output this delete record if:
679 *
680 * (1) this is a broken delete and the counterpart
681 * broken create remains in the output; or
682 * (2) this is not a broken delete, and rename_dst
683 * does not have a rename/copy to move p->one->path
684 * out of existence.
685 *
686 * Otherwise, the counterpart broken create
687 * has been turned into a rename-edit; or
688 * delete did not have a matching create to
689 * begin with.
690 */
691 if (DIFF_PAIR_BROKEN(p)) {
692 /* broken delete */
693 struct diff_rename_dst *dst = locate_rename_dst(p->one);
694 if (dst && dst->pair)
695 /* counterpart is now rename/copy */
696 pair_to_free = p;
697 }
698 else {
699 if (p->one->rename_used)
700 /* this path remains */
701 pair_to_free = p;
702 }
703
704 if (pair_to_free)
705 ;
706 else
707 diff_q(&outq, p);
708 }
709 else if (!diff_unmodified_pair(p))
710 /* all the usual ones need to be kept */
711 diff_q(&outq, p);
712 else
713 /* no need to keep unmodified pairs */
714 pair_to_free = p;
715
716 if (pair_to_free)
717 diff_free_filepair(pair_to_free);
718 }
719 diff_debug_queue("done copying original", &outq);
720
721 free(q->queue);
722 *q = outq;
723 diff_debug_queue("done collapsing", q);
724
725 for (i = 0; i < rename_dst_nr; i++)
726 free_filespec(rename_dst[i].two);
727
728 FREE_AND_NULL(rename_dst);
729 rename_dst_nr = rename_dst_alloc = 0;
730 FREE_AND_NULL(rename_src);
731 rename_src_nr = rename_src_alloc = 0;
732 return;
733 }