1/*
2 * Copyright (C) 2005 Junio C Hamano
3 */
4#include "cache.h"
5#include "diff.h"
6#include "diffcore.h"
7#include "hashmap.h"
8#include "progress.h"
9
10/* Table of rename/copy destinations */
11
12static struct diff_rename_dst {
13 struct diff_filespec *two;
14 struct diff_filepair *pair;
15} *rename_dst;
16static int rename_dst_nr, rename_dst_alloc;
17
18static int find_rename_dst(struct diff_filespec *two)
19{
20 int first, last;
21
22 first = 0;
23 last = rename_dst_nr;
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);
28 if (!cmp)
29 return next;
30 if (cmp < 0) {
31 last = next;
32 continue;
33 }
34 first = next+1;
35 }
36 return -first - 1;
37}
38
39static struct diff_rename_dst *locate_rename_dst(struct diff_filespec *two)
40{
41 int ofs = find_rename_dst(two);
42 return ofs < 0 ? NULL : &rename_dst[ofs];
43}
44
45/*
46 * Returns 0 on success, -1 if we found a duplicate.
47 */
48static int add_rename_dst(struct diff_filespec *two)
49{
50 int first = find_rename_dst(two);
51
52 if (first >= 0)
53 return -1;
54 first = -first - 1;
55
56 /* insert to make it at "first" */
57 ALLOC_GROW(rename_dst, rename_dst_nr + 1, rename_dst_alloc);
58 rename_dst_nr++;
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->oid.hash, two->oid_valid,
64 two->mode);
65 rename_dst[first].pair = NULL;
66 return 0;
67}
68
69/* Table of rename/copy src files */
70static struct diff_rename_src {
71 struct diff_filepair *p;
72 unsigned short score; /* to remember the break score */
73} *rename_src;
74static int rename_src_nr, rename_src_alloc;
75
76static struct diff_rename_src *register_rename_src(struct diff_filepair *p)
77{
78 int first, last;
79 struct diff_filespec *one = p->one;
80 unsigned short score = p->score;
81
82 first = 0;
83 last = rename_src_nr;
84 while (last > first) {
85 int next = (last + first) >> 1;
86 struct diff_rename_src *src = &(rename_src[next]);
87 int cmp = strcmp(one->path, src->p->one->path);
88 if (!cmp)
89 return src;
90 if (cmp < 0) {
91 last = next;
92 continue;
93 }
94 first = next+1;
95 }
96
97 /* insert to make it at "first" */
98 ALLOC_GROW(rename_src, rename_src_nr + 1, rename_src_alloc);
99 rename_src_nr++;
100 if (first < rename_src_nr)
101 memmove(rename_src + first + 1, rename_src + first,
102 (rename_src_nr - first - 1) * sizeof(*rename_src));
103 rename_src[first].p = p;
104 rename_src[first].score = score;
105 return &(rename_src[first]);
106}
107
108static int basename_same(struct diff_filespec *src, struct diff_filespec *dst)
109{
110 int src_len = strlen(src->path), dst_len = strlen(dst->path);
111 while (src_len && dst_len) {
112 char c1 = src->path[--src_len];
113 char c2 = dst->path[--dst_len];
114 if (c1 != c2)
115 return 0;
116 if (c1 == '/')
117 return 1;
118 }
119 return (!src_len || src->path[src_len - 1] == '/') &&
120 (!dst_len || dst->path[dst_len - 1] == '/');
121}
122
123struct diff_score {
124 int src; /* index in rename_src */
125 int dst; /* index in rename_dst */
126 unsigned short score;
127 short name_score;
128};
129
130static int estimate_similarity(struct diff_filespec *src,
131 struct diff_filespec *dst,
132 int minimum_score)
133{
134 /* src points at a file that existed in the original tree (or
135 * optionally a file in the destination tree) and dst points
136 * at a newly created file. They may be quite similar, in which
137 * case we want to say src is renamed to dst or src is copied into
138 * dst, and then some edit has been applied to dst.
139 *
140 * Compare them and return how similar they are, representing
141 * the score as an integer between 0 and MAX_SCORE.
142 *
143 * When there is an exact match, it is considered a better
144 * match than anything else; the destination does not even
145 * call into this function in that case.
146 */
147 unsigned long max_size, delta_size, base_size, src_copied, literal_added;
148 int score;
149
150 /* We deal only with regular files. Symlink renames are handled
151 * only when they are exact matches --- in other words, no edits
152 * after renaming.
153 */
154 if (!S_ISREG(src->mode) || !S_ISREG(dst->mode))
155 return 0;
156
157 /*
158 * Need to check that source and destination sizes are
159 * filled in before comparing them.
160 *
161 * If we already have "cnt_data" filled in, we know it's
162 * all good (avoid checking the size for zero, as that
163 * is a possible size - we really should have a flag to
164 * say whether the size is valid or not!)
165 */
166 if (!src->cnt_data &&
167 diff_populate_filespec(src, CHECK_SIZE_ONLY))
168 return 0;
169 if (!dst->cnt_data &&
170 diff_populate_filespec(dst, CHECK_SIZE_ONLY))
171 return 0;
172
173 max_size = ((src->size > dst->size) ? src->size : dst->size);
174 base_size = ((src->size < dst->size) ? src->size : dst->size);
175 delta_size = max_size - base_size;
176
177 /* We would not consider edits that change the file size so
178 * drastically. delta_size must be smaller than
179 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
180 *
181 * Note that base_size == 0 case is handled here already
182 * and the final score computation below would not have a
183 * divide-by-zero issue.
184 */
185 if (max_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE)
186 return 0;
187
188 if (!src->cnt_data && diff_populate_filespec(src, 0))
189 return 0;
190 if (!dst->cnt_data && diff_populate_filespec(dst, 0))
191 return 0;
192
193 if (diffcore_count_changes(src, dst,
194 &src->cnt_data, &dst->cnt_data,
195 &src_copied, &literal_added))
196 return 0;
197
198 /* How similar are they?
199 * what percentage of material in dst are from source?
200 */
201 if (!dst->size)
202 score = 0; /* should not happen */
203 else
204 score = (int)(src_copied * MAX_SCORE / max_size);
205 return score;
206}
207
208static void record_rename_pair(int dst_index, int src_index, int score)
209{
210 struct diff_filespec *src, *dst;
211 struct diff_filepair *dp;
212
213 if (rename_dst[dst_index].pair)
214 die("internal error: dst already matched.");
215
216 src = rename_src[src_index].p->one;
217 src->rename_used++;
218 src->count++;
219
220 dst = rename_dst[dst_index].two;
221 dst->count++;
222
223 dp = diff_queue(NULL, src, dst);
224 dp->renamed_pair = 1;
225 if (!strcmp(src->path, dst->path))
226 dp->score = rename_src[src_index].score;
227 else
228 dp->score = score;
229 rename_dst[dst_index].pair = dp;
230}
231
232/*
233 * We sort the rename similarity matrix with the score, in descending
234 * order (the most similar first).
235 */
236static int score_compare(const void *a_, const void *b_)
237{
238 const struct diff_score *a = a_, *b = b_;
239
240 /* sink the unused ones to the bottom */
241 if (a->dst < 0)
242 return (0 <= b->dst);
243 else if (b->dst < 0)
244 return -1;
245
246 if (a->score == b->score)
247 return b->name_score - a->name_score;
248
249 return b->score - a->score;
250}
251
252struct file_similarity {
253 struct hashmap_entry entry;
254 int index;
255 struct diff_filespec *filespec;
256};
257
258static unsigned int hash_filespec(struct diff_filespec *filespec)
259{
260 if (!filespec->oid_valid) {
261 if (diff_populate_filespec(filespec, 0))
262 return 0;
263 hash_sha1_file(filespec->data, filespec->size, "blob",
264 filespec->oid.hash);
265 }
266 return sha1hash(filespec->oid.hash);
267}
268
269static int find_identical_files(struct hashmap *srcs,
270 int dst_index,
271 struct diff_options *options)
272{
273 int renames = 0;
274
275 struct diff_filespec *target = rename_dst[dst_index].two;
276 struct file_similarity *p, *best = NULL;
277 int i = 100, best_score = -1;
278
279 /*
280 * Find the best source match for specified destination.
281 */
282 p = hashmap_get_from_hash(srcs, hash_filespec(target), NULL);
283 for (; p; p = hashmap_get_next(srcs, p)) {
284 int score;
285 struct diff_filespec *source = p->filespec;
286
287 /* False hash collision? */
288 if (oidcmp(&source->oid, &target->oid))
289 continue;
290 /* Non-regular files? If so, the modes must match! */
291 if (!S_ISREG(source->mode) || !S_ISREG(target->mode)) {
292 if (source->mode != target->mode)
293 continue;
294 }
295 /* Give higher scores to sources that haven't been used already */
296 score = !source->rename_used;
297 if (source->rename_used && options->detect_rename != DIFF_DETECT_COPY)
298 continue;
299 score += basename_same(source, target);
300 if (score > best_score) {
301 best = p;
302 best_score = score;
303 if (score == 2)
304 break;
305 }
306
307 /* Too many identical alternatives? Pick one */
308 if (!--i)
309 break;
310 }
311 if (best) {
312 record_rename_pair(dst_index, best->index, MAX_SCORE);
313 renames++;
314 }
315 return renames;
316}
317
318static void insert_file_table(struct hashmap *table, int index, struct diff_filespec *filespec)
319{
320 struct file_similarity *entry = xmalloc(sizeof(*entry));
321
322 entry->index = index;
323 entry->filespec = filespec;
324
325 hashmap_entry_init(entry, hash_filespec(filespec));
326 hashmap_add(table, entry);
327}
328
329/*
330 * Find exact renames first.
331 *
332 * The first round matches up the up-to-date entries,
333 * and then during the second round we try to match
334 * cache-dirty entries as well.
335 */
336static int find_exact_renames(struct diff_options *options)
337{
338 int i, renames = 0;
339 struct hashmap file_table;
340
341 /* Add all sources to the hash table in reverse order, because
342 * later on they will be retrieved in LIFO order.
343 */
344 hashmap_init(&file_table, NULL, rename_src_nr);
345 for (i = rename_src_nr-1; i >= 0; i--)
346 insert_file_table(&file_table, i, rename_src[i].p->one);
347
348 /* Walk the destinations and find best source match */
349 for (i = 0; i < rename_dst_nr; i++)
350 renames += find_identical_files(&file_table, i, options);
351
352 /* Free the hash data structure and entries */
353 hashmap_free(&file_table, 1);
354
355 return renames;
356}
357
358#define NUM_CANDIDATE_PER_DST 4
359static void record_if_better(struct diff_score m[], struct diff_score *o)
360{
361 int i, worst;
362
363 /* find the worst one */
364 worst = 0;
365 for (i = 1; i < NUM_CANDIDATE_PER_DST; i++)
366 if (score_compare(&m[i], &m[worst]) > 0)
367 worst = i;
368
369 /* is it better than the worst one? */
370 if (score_compare(&m[worst], o) > 0)
371 m[worst] = *o;
372}
373
374/*
375 * Returns:
376 * 0 if we are under the limit;
377 * 1 if we need to disable inexact rename detection;
378 * 2 if we would be under the limit if we were given -C instead of -C -C.
379 */
380static int too_many_rename_candidates(int num_create,
381 struct diff_options *options)
382{
383 int rename_limit = options->rename_limit;
384 int num_src = rename_src_nr;
385 int i;
386
387 options->needed_rename_limit = 0;
388
389 /*
390 * This basically does a test for the rename matrix not
391 * growing larger than a "rename_limit" square matrix, ie:
392 *
393 * num_create * num_src > rename_limit * rename_limit
394 *
395 * but handles the potential overflow case specially (and we
396 * assume at least 32-bit integers)
397 */
398 if (rename_limit <= 0 || rename_limit > 32767)
399 rename_limit = 32767;
400 if ((num_create <= rename_limit || num_src <= rename_limit) &&
401 (num_create * num_src <= rename_limit * rename_limit))
402 return 0;
403
404 options->needed_rename_limit =
405 num_src > num_create ? num_src : num_create;
406
407 /* Are we running under -C -C? */
408 if (!DIFF_OPT_TST(options, FIND_COPIES_HARDER))
409 return 1;
410
411 /* Would we bust the limit if we were running under -C? */
412 for (num_src = i = 0; i < rename_src_nr; i++) {
413 if (diff_unmodified_pair(rename_src[i].p))
414 continue;
415 num_src++;
416 }
417 if ((num_create <= rename_limit || num_src <= rename_limit) &&
418 (num_create * num_src <= rename_limit * rename_limit))
419 return 2;
420 return 1;
421}
422
423static 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
444void 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 (!DIFF_OPT_TST(options, RENAME_EMPTY) &&
467 is_empty_blob_sha1(p->two->oid.hash))
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 (!DIFF_OPT_TST(options, RENAME_EMPTY) &&
477 is_empty_blob_sha1(p->one->oid.hash))
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_progress_delay(
536 _("Performing inexact rename detection"),
537 rename_dst_nr * rename_src_nr, 50, 1);
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, (i+1)*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(rename_dst);
671 rename_dst = NULL;
672 rename_dst_nr = rename_dst_alloc = 0;
673 free(rename_src);
674 rename_src = NULL;
675 rename_src_nr = rename_src_alloc = 0;
676 return;
677}