]> git.ipfire.org Git - thirdparty/git.git/blob - diffcore-rename.c
t5318: avoid unnecessary command substitutions
[thirdparty/git.git] / diffcore-rename.c
1 /*
2 * Copyright (C) 2005 Junio C Hamano
3 */
4 #include "cache.h"
5 #include "diff.h"
6 #include "diffcore.h"
7 #include "object-store.h"
8 #include "hashmap.h"
9 #include "progress.h"
10
11 /* Table of rename/copy destinations */
12
13 static struct diff_rename_dst {
14 struct diff_filespec *two;
15 struct diff_filepair *pair;
16 } *rename_dst;
17 static int rename_dst_nr, rename_dst_alloc;
18
19 static int find_rename_dst(struct diff_filespec *two)
20 {
21 int first, last;
22
23 first = 0;
24 last = rename_dst_nr;
25 while (last > first) {
26 int next = (last + first) >> 1;
27 struct diff_rename_dst *dst = &(rename_dst[next]);
28 int cmp = strcmp(two->path, dst->two->path);
29 if (!cmp)
30 return next;
31 if (cmp < 0) {
32 last = next;
33 continue;
34 }
35 first = next+1;
36 }
37 return -first - 1;
38 }
39
40 static struct diff_rename_dst *locate_rename_dst(struct diff_filespec *two)
41 {
42 int ofs = find_rename_dst(two);
43 return ofs < 0 ? NULL : &rename_dst[ofs];
44 }
45
46 /*
47 * Returns 0 on success, -1 if we found a duplicate.
48 */
49 static int add_rename_dst(struct diff_filespec *two)
50 {
51 int first = find_rename_dst(two);
52
53 if (first >= 0)
54 return -1;
55 first = -first - 1;
56
57 /* insert to make it at "first" */
58 ALLOC_GROW(rename_dst, rename_dst_nr + 1, rename_dst_alloc);
59 rename_dst_nr++;
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,
65 two->mode);
66 rename_dst[first].pair = NULL;
67 return 0;
68 }
69
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 */
74 } *rename_src;
75 static int rename_src_nr, rename_src_alloc;
76
77 static struct diff_rename_src *register_rename_src(struct diff_filepair *p)
78 {
79 int first, last;
80 struct diff_filespec *one = p->one;
81 unsigned short score = p->score;
82
83 first = 0;
84 last = rename_src_nr;
85 while (last > first) {
86 int next = (last + first) >> 1;
87 struct diff_rename_src *src = &(rename_src[next]);
88 int cmp = strcmp(one->path, src->p->one->path);
89 if (!cmp)
90 return src;
91 if (cmp < 0) {
92 last = next;
93 continue;
94 }
95 first = next+1;
96 }
97
98 /* insert to make it at "first" */
99 ALLOC_GROW(rename_src, rename_src_nr + 1, rename_src_alloc);
100 rename_src_nr++;
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]);
107 }
108
109 static int basename_same(struct diff_filespec *src, struct diff_filespec *dst)
110 {
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];
115 if (c1 != c2)
116 return 0;
117 if (c1 == '/')
118 return 1;
119 }
120 return (!src_len || src->path[src_len - 1] == '/') &&
121 (!dst_len || dst->path[dst_len - 1] == '/');
122 }
123
124 struct diff_score {
125 int src; /* index in rename_src */
126 int dst; /* index in rename_dst */
127 unsigned short score;
128 short name_score;
129 };
130
131 static int estimate_similarity(struct diff_filespec *src,
132 struct diff_filespec *dst,
133 int minimum_score)
134 {
135 /* src points at a file that existed in the original tree (or
136 * optionally a file in the destination tree) and dst points
137 * at a newly created file. They may be quite similar, in which
138 * case we want to say src is renamed to dst or src is copied into
139 * dst, and then some edit has been applied to dst.
140 *
141 * Compare them and return how similar they are, representing
142 * the score as an integer between 0 and MAX_SCORE.
143 *
144 * When there is an exact match, it is considered a better
145 * match than anything else; the destination does not even
146 * call into this function in that case.
147 */
148 unsigned long max_size, delta_size, base_size, src_copied, literal_added;
149 int score;
150
151 /* We deal only with regular files. Symlink renames are handled
152 * only when they are exact matches --- in other words, no edits
153 * after renaming.
154 */
155 if (!S_ISREG(src->mode) || !S_ISREG(dst->mode))
156 return 0;
157
158 /*
159 * Need to check that source and destination sizes are
160 * filled in before comparing them.
161 *
162 * If we already have "cnt_data" filled in, we know it's
163 * all good (avoid checking the size for zero, as that
164 * is a possible size - we really should have a flag to
165 * say whether the size is valid or not!)
166 */
167 if (!src->cnt_data &&
168 diff_populate_filespec(src, CHECK_SIZE_ONLY))
169 return 0;
170 if (!dst->cnt_data &&
171 diff_populate_filespec(dst, CHECK_SIZE_ONLY))
172 return 0;
173
174 max_size = ((src->size > dst->size) ? src->size : dst->size);
175 base_size = ((src->size < dst->size) ? src->size : dst->size);
176 delta_size = max_size - base_size;
177
178 /* We would not consider edits that change the file size so
179 * drastically. delta_size must be smaller than
180 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
181 *
182 * Note that base_size == 0 case is handled here already
183 * and the final score computation below would not have a
184 * divide-by-zero issue.
185 */
186 if (max_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE)
187 return 0;
188
189 if (!src->cnt_data && diff_populate_filespec(src, 0))
190 return 0;
191 if (!dst->cnt_data && diff_populate_filespec(dst, 0))
192 return 0;
193
194 if (diffcore_count_changes(src, dst,
195 &src->cnt_data, &dst->cnt_data,
196 &src_copied, &literal_added))
197 return 0;
198
199 /* How similar are they?
200 * what percentage of material in dst are from source?
201 */
202 if (!dst->size)
203 score = 0; /* should not happen */
204 else
205 score = (int)(src_copied * MAX_SCORE / max_size);
206 return score;
207 }
208
209 static void record_rename_pair(int dst_index, int src_index, int score)
210 {
211 struct diff_filespec *src, *dst;
212 struct diff_filepair *dp;
213
214 if (rename_dst[dst_index].pair)
215 die("internal error: dst already matched.");
216
217 src = rename_src[src_index].p->one;
218 src->rename_used++;
219 src->count++;
220
221 dst = rename_dst[dst_index].two;
222 dst->count++;
223
224 dp = diff_queue(NULL, src, dst);
225 dp->renamed_pair = 1;
226 if (!strcmp(src->path, dst->path))
227 dp->score = rename_src[src_index].score;
228 else
229 dp->score = score;
230 rename_dst[dst_index].pair = dp;
231 }
232
233 /*
234 * We sort the rename similarity matrix with the score, in descending
235 * order (the most similar first).
236 */
237 static int score_compare(const void *a_, const void *b_)
238 {
239 const struct diff_score *a = a_, *b = b_;
240
241 /* sink the unused ones to the bottom */
242 if (a->dst < 0)
243 return (0 <= b->dst);
244 else if (b->dst < 0)
245 return -1;
246
247 if (a->score == b->score)
248 return b->name_score - a->name_score;
249
250 return b->score - a->score;
251 }
252
253 struct file_similarity {
254 struct hashmap_entry entry;
255 int index;
256 struct diff_filespec *filespec;
257 };
258
259 static unsigned int hash_filespec(struct diff_filespec *filespec)
260 {
261 if (!filespec->oid_valid) {
262 if (diff_populate_filespec(filespec, 0))
263 return 0;
264 hash_object_file(filespec->data, filespec->size, "blob",
265 &filespec->oid);
266 }
267 return sha1hash(filespec->oid.hash);
268 }
269
270 static int find_identical_files(struct hashmap *srcs,
271 int dst_index,
272 struct diff_options *options)
273 {
274 int renames = 0;
275
276 struct diff_filespec *target = rename_dst[dst_index].two;
277 struct file_similarity *p, *best = NULL;
278 int i = 100, best_score = -1;
279
280 /*
281 * Find the best source match for specified destination.
282 */
283 p = hashmap_get_from_hash(srcs, hash_filespec(target), NULL);
284 for (; p; p = hashmap_get_next(srcs, p)) {
285 int score;
286 struct diff_filespec *source = p->filespec;
287
288 /* False hash collision? */
289 if (oidcmp(&source->oid, &target->oid))
290 continue;
291 /* Non-regular files? If so, the modes must match! */
292 if (!S_ISREG(source->mode) || !S_ISREG(target->mode)) {
293 if (source->mode != target->mode)
294 continue;
295 }
296 /* Give higher scores to sources that haven't been used already */
297 score = !source->rename_used;
298 if (source->rename_used && options->detect_rename != DIFF_DETECT_COPY)
299 continue;
300 score += basename_same(source, target);
301 if (score > best_score) {
302 best = p;
303 best_score = score;
304 if (score == 2)
305 break;
306 }
307
308 /* Too many identical alternatives? Pick one */
309 if (!--i)
310 break;
311 }
312 if (best) {
313 record_rename_pair(dst_index, best->index, MAX_SCORE);
314 renames++;
315 }
316 return renames;
317 }
318
319 static void insert_file_table(struct hashmap *table, int index, struct diff_filespec *filespec)
320 {
321 struct file_similarity *entry = xmalloc(sizeof(*entry));
322
323 entry->index = index;
324 entry->filespec = filespec;
325
326 hashmap_entry_init(entry, hash_filespec(filespec));
327 hashmap_add(table, entry);
328 }
329
330 /*
331 * Find exact renames first.
332 *
333 * The first round matches up the up-to-date entries,
334 * and then during the second round we try to match
335 * cache-dirty entries as well.
336 */
337 static int find_exact_renames(struct diff_options *options)
338 {
339 int i, renames = 0;
340 struct hashmap file_table;
341
342 /* Add all sources to the hash table in reverse order, because
343 * later on they will be retrieved in LIFO order.
344 */
345 hashmap_init(&file_table, NULL, NULL, rename_src_nr);
346 for (i = rename_src_nr-1; i >= 0; i--)
347 insert_file_table(&file_table, i, rename_src[i].p->one);
348
349 /* Walk the destinations and find best source match */
350 for (i = 0; i < rename_dst_nr; i++)
351 renames += find_identical_files(&file_table, i, options);
352
353 /* Free the hash data structure and entries */
354 hashmap_free(&file_table, 1);
355
356 return renames;
357 }
358
359 #define NUM_CANDIDATE_PER_DST 4
360 static void record_if_better(struct diff_score m[], struct diff_score *o)
361 {
362 int i, worst;
363
364 /* find the worst one */
365 worst = 0;
366 for (i = 1; i < NUM_CANDIDATE_PER_DST; i++)
367 if (score_compare(&m[i], &m[worst]) > 0)
368 worst = i;
369
370 /* is it better than the worst one? */
371 if (score_compare(&m[worst], o) > 0)
372 m[worst] = *o;
373 }
374
375 /*
376 * Returns:
377 * 0 if we are under the limit;
378 * 1 if we need to disable inexact rename detection;
379 * 2 if we would be under the limit if we were given -C instead of -C -C.
380 */
381 static int too_many_rename_candidates(int num_create,
382 struct diff_options *options)
383 {
384 int rename_limit = options->rename_limit;
385 int num_src = rename_src_nr;
386 int i;
387
388 options->needed_rename_limit = 0;
389
390 /*
391 * This basically does a test for the rename matrix not
392 * growing larger than a "rename_limit" square matrix, ie:
393 *
394 * num_create * num_src > rename_limit * rename_limit
395 */
396 if (rename_limit <= 0)
397 rename_limit = 32767;
398 if ((num_create <= rename_limit || num_src <= rename_limit) &&
399 ((uint64_t)num_create * (uint64_t)num_src
400 <= (uint64_t)rename_limit * (uint64_t)rename_limit))
401 return 0;
402
403 options->needed_rename_limit =
404 num_src > num_create ? num_src : num_create;
405
406 /* Are we running under -C -C? */
407 if (!options->flags.find_copies_harder)
408 return 1;
409
410 /* Would we bust the limit if we were running under -C? */
411 for (num_src = i = 0; i < rename_src_nr; i++) {
412 if (diff_unmodified_pair(rename_src[i].p))
413 continue;
414 num_src++;
415 }
416 if ((num_create <= rename_limit || num_src <= rename_limit) &&
417 ((uint64_t)num_create * (uint64_t)num_src
418 <= (uint64_t)rename_limit * (uint64_t)rename_limit))
419 return 2;
420 return 1;
421 }
422
423 static int find_renames(struct diff_score *mx, int dst_cnt, int minimum_score, int copies)
424 {
425 int count = 0, i;
426
427 for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) {
428 struct diff_rename_dst *dst;
429
430 if ((mx[i].dst < 0) ||
431 (mx[i].score < minimum_score))
432 break; /* there is no more usable pair. */
433 dst = &rename_dst[mx[i].dst];
434 if (dst->pair)
435 continue; /* already done, either exact or fuzzy. */
436 if (!copies && rename_src[mx[i].src].p->one->rename_used)
437 continue;
438 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
439 count++;
440 }
441 return count;
442 }
443
444 void diffcore_rename(struct diff_options *options)
445 {
446 int detect_rename = options->detect_rename;
447 int minimum_score = options->rename_score;
448 struct diff_queue_struct *q = &diff_queued_diff;
449 struct diff_queue_struct outq;
450 struct diff_score *mx;
451 int i, j, rename_count, skip_unmodified = 0;
452 int num_create, dst_cnt;
453 struct progress *progress = NULL;
454
455 if (!minimum_score)
456 minimum_score = DEFAULT_RENAME_SCORE;
457
458 for (i = 0; i < q->nr; i++) {
459 struct diff_filepair *p = q->queue[i];
460 if (!DIFF_FILE_VALID(p->one)) {
461 if (!DIFF_FILE_VALID(p->two))
462 continue; /* unmerged */
463 else if (options->single_follow &&
464 strcmp(options->single_follow, p->two->path))
465 continue; /* not interested */
466 else if (!options->flags.rename_empty &&
467 is_empty_blob_oid(&p->two->oid))
468 continue;
469 else if (add_rename_dst(p->two) < 0) {
470 warning("skipping rename detection, detected"
471 " duplicate destination '%s'",
472 p->two->path);
473 goto cleanup;
474 }
475 }
476 else if (!options->flags.rename_empty &&
477 is_empty_blob_oid(&p->one->oid))
478 continue;
479 else if (!DIFF_PAIR_UNMERGED(p) && !DIFF_FILE_VALID(p->two)) {
480 /*
481 * If the source is a broken "delete", and
482 * they did not really want to get broken,
483 * that means the source actually stays.
484 * So we increment the "rename_used" score
485 * by one, to indicate ourselves as a user
486 */
487 if (p->broken_pair && !p->score)
488 p->one->rename_used++;
489 register_rename_src(p);
490 }
491 else if (detect_rename == DIFF_DETECT_COPY) {
492 /*
493 * Increment the "rename_used" score by
494 * one, to indicate ourselves as a user.
495 */
496 p->one->rename_used++;
497 register_rename_src(p);
498 }
499 }
500 if (rename_dst_nr == 0 || rename_src_nr == 0)
501 goto cleanup; /* nothing to do */
502
503 /*
504 * We really want to cull the candidates list early
505 * with cheap tests in order to avoid doing deltas.
506 */
507 rename_count = find_exact_renames(options);
508
509 /* Did we only want exact renames? */
510 if (minimum_score == MAX_SCORE)
511 goto cleanup;
512
513 /*
514 * Calculate how many renames are left (but all the source
515 * files still remain as options for rename/copies!)
516 */
517 num_create = (rename_dst_nr - rename_count);
518
519 /* All done? */
520 if (!num_create)
521 goto cleanup;
522
523 switch (too_many_rename_candidates(num_create, options)) {
524 case 1:
525 goto cleanup;
526 case 2:
527 options->degraded_cc_to_c = 1;
528 skip_unmodified = 1;
529 break;
530 default:
531 break;
532 }
533
534 if (options->show_rename_progress) {
535 progress = start_delayed_progress(
536 _("Performing inexact rename detection"),
537 (uint64_t)rename_dst_nr * (uint64_t)rename_src_nr);
538 }
539
540 mx = xcalloc(st_mult(NUM_CANDIDATE_PER_DST, num_create), sizeof(*mx));
541 for (dst_cnt = i = 0; i < rename_dst_nr; i++) {
542 struct diff_filespec *two = rename_dst[i].two;
543 struct diff_score *m;
544
545 if (rename_dst[i].pair)
546 continue; /* dealt with exact match already. */
547
548 m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST];
549 for (j = 0; j < NUM_CANDIDATE_PER_DST; j++)
550 m[j].dst = -1;
551
552 for (j = 0; j < rename_src_nr; j++) {
553 struct diff_filespec *one = rename_src[j].p->one;
554 struct diff_score this_src;
555
556 if (skip_unmodified &&
557 diff_unmodified_pair(rename_src[j].p))
558 continue;
559
560 this_src.score = estimate_similarity(one, two,
561 minimum_score);
562 this_src.name_score = basename_same(one, two);
563 this_src.dst = i;
564 this_src.src = j;
565 record_if_better(m, &this_src);
566 /*
567 * Once we run estimate_similarity,
568 * We do not need the text anymore.
569 */
570 diff_free_filespec_blob(one);
571 diff_free_filespec_blob(two);
572 }
573 dst_cnt++;
574 display_progress(progress, (uint64_t)(i+1)*(uint64_t)rename_src_nr);
575 }
576 stop_progress(&progress);
577
578 /* cost matrix sorted by most to least similar pair */
579 QSORT(mx, dst_cnt * NUM_CANDIDATE_PER_DST, score_compare);
580
581 rename_count += find_renames(mx, dst_cnt, minimum_score, 0);
582 if (detect_rename == DIFF_DETECT_COPY)
583 rename_count += find_renames(mx, dst_cnt, minimum_score, 1);
584 free(mx);
585
586 cleanup:
587 /* At this point, we have found some renames and copies and they
588 * are recorded in rename_dst. The original list is still in *q.
589 */
590 DIFF_QUEUE_CLEAR(&outq);
591 for (i = 0; i < q->nr; i++) {
592 struct diff_filepair *p = q->queue[i];
593 struct diff_filepair *pair_to_free = NULL;
594
595 if (DIFF_PAIR_UNMERGED(p)) {
596 diff_q(&outq, p);
597 }
598 else if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
599 /*
600 * Creation
601 *
602 * We would output this create record if it has
603 * not been turned into a rename/copy already.
604 */
605 struct diff_rename_dst *dst = locate_rename_dst(p->two);
606 if (dst && dst->pair) {
607 diff_q(&outq, dst->pair);
608 pair_to_free = p;
609 }
610 else
611 /* no matching rename/copy source, so
612 * record this as a creation.
613 */
614 diff_q(&outq, p);
615 }
616 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) {
617 /*
618 * Deletion
619 *
620 * We would output this delete record if:
621 *
622 * (1) this is a broken delete and the counterpart
623 * broken create remains in the output; or
624 * (2) this is not a broken delete, and rename_dst
625 * does not have a rename/copy to move p->one->path
626 * out of existence.
627 *
628 * Otherwise, the counterpart broken create
629 * has been turned into a rename-edit; or
630 * delete did not have a matching create to
631 * begin with.
632 */
633 if (DIFF_PAIR_BROKEN(p)) {
634 /* broken delete */
635 struct diff_rename_dst *dst = locate_rename_dst(p->one);
636 if (dst && dst->pair)
637 /* counterpart is now rename/copy */
638 pair_to_free = p;
639 }
640 else {
641 if (p->one->rename_used)
642 /* this path remains */
643 pair_to_free = p;
644 }
645
646 if (pair_to_free)
647 ;
648 else
649 diff_q(&outq, p);
650 }
651 else if (!diff_unmodified_pair(p))
652 /* all the usual ones need to be kept */
653 diff_q(&outq, p);
654 else
655 /* no need to keep unmodified pairs */
656 pair_to_free = p;
657
658 if (pair_to_free)
659 diff_free_filepair(pair_to_free);
660 }
661 diff_debug_queue("done copying original", &outq);
662
663 free(q->queue);
664 *q = outq;
665 diff_debug_queue("done collapsing", q);
666
667 for (i = 0; i < rename_dst_nr; i++)
668 free_filespec(rename_dst[i].two);
669
670 FREE_AND_NULL(rename_dst);
671 rename_dst_nr = rename_dst_alloc = 0;
672 FREE_AND_NULL(rename_src);
673 rename_src_nr = rename_src_alloc = 0;
674 return;
675 }