Merge branch 'kb/hashmap-updates'
[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 "hashmap.h"
8 #include "progress.h"
9
10 /* Table of rename/copy destinations */
11
12 static struct diff_rename_dst {
13         struct diff_filespec *two;
14         struct diff_filepair *pair;
15 } *rename_dst;
16 static int rename_dst_nr, rename_dst_alloc;
17
18 static struct diff_rename_dst *locate_rename_dst(struct diff_filespec *two,
19                                                  int insert_ok)
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 dst;
31                 if (cmp < 0) {
32                         last = next;
33                         continue;
34                 }
35                 first = next+1;
36         }
37         /* not found */
38         if (!insert_ok)
39                 return NULL;
40         /* insert to make it at "first" */
41         ALLOC_GROW(rename_dst, rename_dst_nr + 1, rename_dst_alloc);
42         rename_dst_nr++;
43         if (first < rename_dst_nr)
44                 memmove(rename_dst + first + 1, rename_dst + first,
45                         (rename_dst_nr - first - 1) * sizeof(*rename_dst));
46         rename_dst[first].two = alloc_filespec(two->path);
47         fill_filespec(rename_dst[first].two, two->sha1, two->sha1_valid, two->mode);
48         rename_dst[first].pair = NULL;
49         return &(rename_dst[first]);
50 }
51
52 /* Table of rename/copy src files */
53 static struct diff_rename_src {
54         struct diff_filepair *p;
55         unsigned short score; /* to remember the break score */
56 } *rename_src;
57 static int rename_src_nr, rename_src_alloc;
58
59 static struct diff_rename_src *register_rename_src(struct diff_filepair *p)
60 {
61         int first, last;
62         struct diff_filespec *one = p->one;
63         unsigned short score = p->score;
64
65         first = 0;
66         last = rename_src_nr;
67         while (last > first) {
68                 int next = (last + first) >> 1;
69                 struct diff_rename_src *src = &(rename_src[next]);
70                 int cmp = strcmp(one->path, src->p->one->path);
71                 if (!cmp)
72                         return src;
73                 if (cmp < 0) {
74                         last = next;
75                         continue;
76                 }
77                 first = next+1;
78         }
79
80         /* insert to make it at "first" */
81         ALLOC_GROW(rename_src, rename_src_nr + 1, rename_src_alloc);
82         rename_src_nr++;
83         if (first < rename_src_nr)
84                 memmove(rename_src + first + 1, rename_src + first,
85                         (rename_src_nr - first - 1) * sizeof(*rename_src));
86         rename_src[first].p = p;
87         rename_src[first].score = score;
88         return &(rename_src[first]);
89 }
90
91 static int basename_same(struct diff_filespec *src, struct diff_filespec *dst)
92 {
93         int src_len = strlen(src->path), dst_len = strlen(dst->path);
94         while (src_len && dst_len) {
95                 char c1 = src->path[--src_len];
96                 char c2 = dst->path[--dst_len];
97                 if (c1 != c2)
98                         return 0;
99                 if (c1 == '/')
100                         return 1;
101         }
102         return (!src_len || src->path[src_len - 1] == '/') &&
103                 (!dst_len || dst->path[dst_len - 1] == '/');
104 }
105
106 struct diff_score {
107         int src; /* index in rename_src */
108         int dst; /* index in rename_dst */
109         unsigned short score;
110         short name_score;
111 };
112
113 static int estimate_similarity(struct diff_filespec *src,
114                                struct diff_filespec *dst,
115                                int minimum_score)
116 {
117         /* src points at a file that existed in the original tree (or
118          * optionally a file in the destination tree) and dst points
119          * at a newly created file.  They may be quite similar, in which
120          * case we want to say src is renamed to dst or src is copied into
121          * dst, and then some edit has been applied to dst.
122          *
123          * Compare them and return how similar they are, representing
124          * the score as an integer between 0 and MAX_SCORE.
125          *
126          * When there is an exact match, it is considered a better
127          * match than anything else; the destination does not even
128          * call into this function in that case.
129          */
130         unsigned long max_size, delta_size, base_size, src_copied, literal_added;
131         unsigned long delta_limit;
132         int score;
133
134         /* We deal only with regular files.  Symlink renames are handled
135          * only when they are exact matches --- in other words, no edits
136          * after renaming.
137          */
138         if (!S_ISREG(src->mode) || !S_ISREG(dst->mode))
139                 return 0;
140
141         /*
142          * Need to check that source and destination sizes are
143          * filled in before comparing them.
144          *
145          * If we already have "cnt_data" filled in, we know it's
146          * all good (avoid checking the size for zero, as that
147          * is a possible size - we really should have a flag to
148          * say whether the size is valid or not!)
149          */
150         if (!src->cnt_data && diff_populate_filespec(src, 1))
151                 return 0;
152         if (!dst->cnt_data && diff_populate_filespec(dst, 1))
153                 return 0;
154
155         max_size = ((src->size > dst->size) ? src->size : dst->size);
156         base_size = ((src->size < dst->size) ? src->size : dst->size);
157         delta_size = max_size - base_size;
158
159         /* We would not consider edits that change the file size so
160          * drastically.  delta_size must be smaller than
161          * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
162          *
163          * Note that base_size == 0 case is handled here already
164          * and the final score computation below would not have a
165          * divide-by-zero issue.
166          */
167         if (max_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE)
168                 return 0;
169
170         if (!src->cnt_data && diff_populate_filespec(src, 0))
171                 return 0;
172         if (!dst->cnt_data && diff_populate_filespec(dst, 0))
173                 return 0;
174
175         delta_limit = (unsigned long)
176                 (base_size * (MAX_SCORE-minimum_score) / MAX_SCORE);
177         if (diffcore_count_changes(src, dst,
178                                    &src->cnt_data, &dst->cnt_data,
179                                    delta_limit,
180                                    &src_copied, &literal_added))
181                 return 0;
182
183         /* How similar are they?
184          * what percentage of material in dst are from source?
185          */
186         if (!dst->size)
187                 score = 0; /* should not happen */
188         else
189                 score = (int)(src_copied * MAX_SCORE / max_size);
190         return score;
191 }
192
193 static void record_rename_pair(int dst_index, int src_index, int score)
194 {
195         struct diff_filespec *src, *dst;
196         struct diff_filepair *dp;
197
198         if (rename_dst[dst_index].pair)
199                 die("internal error: dst already matched.");
200
201         src = rename_src[src_index].p->one;
202         src->rename_used++;
203         src->count++;
204
205         dst = rename_dst[dst_index].two;
206         dst->count++;
207
208         dp = diff_queue(NULL, src, dst);
209         dp->renamed_pair = 1;
210         if (!strcmp(src->path, dst->path))
211                 dp->score = rename_src[src_index].score;
212         else
213                 dp->score = score;
214         rename_dst[dst_index].pair = dp;
215 }
216
217 /*
218  * We sort the rename similarity matrix with the score, in descending
219  * order (the most similar first).
220  */
221 static int score_compare(const void *a_, const void *b_)
222 {
223         const struct diff_score *a = a_, *b = b_;
224
225         /* sink the unused ones to the bottom */
226         if (a->dst < 0)
227                 return (0 <= b->dst);
228         else if (b->dst < 0)
229                 return -1;
230
231         if (a->score == b->score)
232                 return b->name_score - a->name_score;
233
234         return b->score - a->score;
235 }
236
237 struct file_similarity {
238         struct hashmap_entry entry;
239         int index;
240         struct diff_filespec *filespec;
241 };
242
243 static unsigned int hash_filespec(struct diff_filespec *filespec)
244 {
245         if (!filespec->sha1_valid) {
246                 if (diff_populate_filespec(filespec, 0))
247                         return 0;
248                 hash_sha1_file(filespec->data, filespec->size, "blob", filespec->sha1);
249         }
250         return sha1hash(filespec->sha1);
251 }
252
253 static int find_identical_files(struct hashmap *srcs,
254                                 int dst_index,
255                                 struct diff_options *options)
256 {
257         int renames = 0;
258
259         struct diff_filespec *target = rename_dst[dst_index].two;
260         struct file_similarity *p, *best = NULL;
261         int i = 100, best_score = -1;
262
263         /*
264          * Find the best source match for specified destination.
265          */
266         p = hashmap_get_from_hash(srcs, hash_filespec(target), NULL);
267         for (; p; p = hashmap_get_next(srcs, p)) {
268                 int score;
269                 struct diff_filespec *source = p->filespec;
270
271                 /* False hash collision? */
272                 if (hashcmp(source->sha1, target->sha1))
273                         continue;
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)
277                                 continue;
278                 }
279                 /* Give higher scores to sources that haven't been used already */
280                 score = !source->rename_used;
281                 if (source->rename_used && options->detect_rename != DIFF_DETECT_COPY)
282                         continue;
283                 score += basename_same(source, target);
284                 if (score > best_score) {
285                         best = p;
286                         best_score = score;
287                         if (score == 2)
288                                 break;
289                 }
290
291                 /* Too many identical alternatives? Pick one */
292                 if (!--i)
293                         break;
294         }
295         if (best) {
296                 record_rename_pair(dst_index, best->index, MAX_SCORE);
297                 renames++;
298         }
299         return renames;
300 }
301
302 static void insert_file_table(struct hashmap *table, int index, struct diff_filespec *filespec)
303 {
304         struct file_similarity *entry = xmalloc(sizeof(*entry));
305
306         entry->index = index;
307         entry->filespec = filespec;
308
309         hashmap_entry_init(entry, hash_filespec(filespec));
310         hashmap_add(table, entry);
311 }
312
313 /*
314  * Find exact renames first.
315  *
316  * The first round matches up the up-to-date entries,
317  * and then during the second round we try to match
318  * cache-dirty entries as well.
319  */
320 static int find_exact_renames(struct diff_options *options)
321 {
322         int i, renames = 0;
323         struct hashmap file_table;
324
325         /* Add all sources to the hash table */
326         hashmap_init(&file_table, NULL, rename_src_nr);
327         for (i = 0; i < rename_src_nr; i++)
328                 insert_file_table(&file_table, i, rename_src[i].p->one);
329
330         /* Walk the destinations and find best source match */
331         for (i = 0; i < rename_dst_nr; i++)
332                 renames += find_identical_files(&file_table, i, options);
333
334         /* Free the hash data structure and entries */
335         hashmap_free(&file_table, 1);
336
337         return renames;
338 }
339
340 #define NUM_CANDIDATE_PER_DST 4
341 static void record_if_better(struct diff_score m[], struct diff_score *o)
342 {
343         int i, worst;
344
345         /* find the worst one */
346         worst = 0;
347         for (i = 1; i < NUM_CANDIDATE_PER_DST; i++)
348                 if (score_compare(&m[i], &m[worst]) > 0)
349                         worst = i;
350
351         /* is it better than the worst one? */
352         if (score_compare(&m[worst], o) > 0)
353                 m[worst] = *o;
354 }
355
356 /*
357  * Returns:
358  * 0 if we are under the limit;
359  * 1 if we need to disable inexact rename detection;
360  * 2 if we would be under the limit if we were given -C instead of -C -C.
361  */
362 static int too_many_rename_candidates(int num_create,
363                                       struct diff_options *options)
364 {
365         int rename_limit = options->rename_limit;
366         int num_src = rename_src_nr;
367         int i;
368
369         options->needed_rename_limit = 0;
370
371         /*
372          * This basically does a test for the rename matrix not
373          * growing larger than a "rename_limit" square matrix, ie:
374          *
375          *    num_create * num_src > rename_limit * rename_limit
376          *
377          * but handles the potential overflow case specially (and we
378          * assume at least 32-bit integers)
379          */
380         if (rename_limit <= 0 || rename_limit > 32767)
381                 rename_limit = 32767;
382         if ((num_create <= rename_limit || num_src <= rename_limit) &&
383             (num_create * num_src <= rename_limit * rename_limit))
384                 return 0;
385
386         options->needed_rename_limit =
387                 num_src > num_create ? num_src : num_create;
388
389         /* Are we running under -C -C? */
390         if (!DIFF_OPT_TST(options, FIND_COPIES_HARDER))
391                 return 1;
392
393         /* Would we bust the limit if we were running under -C? */
394         for (num_src = i = 0; i < rename_src_nr; i++) {
395                 if (diff_unmodified_pair(rename_src[i].p))
396                         continue;
397                 num_src++;
398         }
399         if ((num_create <= rename_limit || num_src <= rename_limit) &&
400             (num_create * num_src <= rename_limit * rename_limit))
401                 return 2;
402         return 1;
403 }
404
405 static int find_renames(struct diff_score *mx, int dst_cnt, int minimum_score, int copies)
406 {
407         int count = 0, i;
408
409         for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) {
410                 struct diff_rename_dst *dst;
411
412                 if ((mx[i].dst < 0) ||
413                     (mx[i].score < minimum_score))
414                         break; /* there is no more usable pair. */
415                 dst = &rename_dst[mx[i].dst];
416                 if (dst->pair)
417                         continue; /* already done, either exact or fuzzy. */
418                 if (!copies && rename_src[mx[i].src].p->one->rename_used)
419                         continue;
420                 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
421                 count++;
422         }
423         return count;
424 }
425
426 void diffcore_rename(struct diff_options *options)
427 {
428         int detect_rename = options->detect_rename;
429         int minimum_score = options->rename_score;
430         struct diff_queue_struct *q = &diff_queued_diff;
431         struct diff_queue_struct outq;
432         struct diff_score *mx;
433         int i, j, rename_count, skip_unmodified = 0;
434         int num_create, dst_cnt;
435         struct progress *progress = NULL;
436
437         if (!minimum_score)
438                 minimum_score = DEFAULT_RENAME_SCORE;
439
440         for (i = 0; i < q->nr; i++) {
441                 struct diff_filepair *p = q->queue[i];
442                 if (!DIFF_FILE_VALID(p->one)) {
443                         if (!DIFF_FILE_VALID(p->two))
444                                 continue; /* unmerged */
445                         else if (options->single_follow &&
446                                  strcmp(options->single_follow, p->two->path))
447                                 continue; /* not interested */
448                         else if (!DIFF_OPT_TST(options, RENAME_EMPTY) &&
449                                  is_empty_blob_sha1(p->two->sha1))
450                                 continue;
451                         else
452                                 locate_rename_dst(p->two, 1);
453                 }
454                 else if (!DIFF_OPT_TST(options, RENAME_EMPTY) &&
455                          is_empty_blob_sha1(p->one->sha1))
456                         continue;
457                 else if (!DIFF_PAIR_UNMERGED(p) && !DIFF_FILE_VALID(p->two)) {
458                         /*
459                          * If the source is a broken "delete", and
460                          * they did not really want to get broken,
461                          * that means the source actually stays.
462                          * So we increment the "rename_used" score
463                          * by one, to indicate ourselves as a user
464                          */
465                         if (p->broken_pair && !p->score)
466                                 p->one->rename_used++;
467                         register_rename_src(p);
468                 }
469                 else if (detect_rename == DIFF_DETECT_COPY) {
470                         /*
471                          * Increment the "rename_used" score by
472                          * one, to indicate ourselves as a user.
473                          */
474                         p->one->rename_used++;
475                         register_rename_src(p);
476                 }
477         }
478         if (rename_dst_nr == 0 || rename_src_nr == 0)
479                 goto cleanup; /* nothing to do */
480
481         /*
482          * We really want to cull the candidates list early
483          * with cheap tests in order to avoid doing deltas.
484          */
485         rename_count = find_exact_renames(options);
486
487         /* Did we only want exact renames? */
488         if (minimum_score == MAX_SCORE)
489                 goto cleanup;
490
491         /*
492          * Calculate how many renames are left (but all the source
493          * files still remain as options for rename/copies!)
494          */
495         num_create = (rename_dst_nr - rename_count);
496
497         /* All done? */
498         if (!num_create)
499                 goto cleanup;
500
501         switch (too_many_rename_candidates(num_create, options)) {
502         case 1:
503                 goto cleanup;
504         case 2:
505                 options->degraded_cc_to_c = 1;
506                 skip_unmodified = 1;
507                 break;
508         default:
509                 break;
510         }
511
512         if (options->show_rename_progress) {
513                 progress = start_progress_delay(
514                                 _("Performing inexact rename detection"),
515                                 rename_dst_nr * rename_src_nr, 50, 1);
516         }
517
518         mx = xcalloc(num_create * NUM_CANDIDATE_PER_DST, sizeof(*mx));
519         for (dst_cnt = i = 0; i < rename_dst_nr; i++) {
520                 struct diff_filespec *two = rename_dst[i].two;
521                 struct diff_score *m;
522
523                 if (rename_dst[i].pair)
524                         continue; /* dealt with exact match already. */
525
526                 m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST];
527                 for (j = 0; j < NUM_CANDIDATE_PER_DST; j++)
528                         m[j].dst = -1;
529
530                 for (j = 0; j < rename_src_nr; j++) {
531                         struct diff_filespec *one = rename_src[j].p->one;
532                         struct diff_score this_src;
533
534                         if (skip_unmodified &&
535                             diff_unmodified_pair(rename_src[j].p))
536                                 continue;
537
538                         this_src.score = estimate_similarity(one, two,
539                                                              minimum_score);
540                         this_src.name_score = basename_same(one, two);
541                         this_src.dst = i;
542                         this_src.src = j;
543                         record_if_better(m, &this_src);
544                         /*
545                          * Once we run estimate_similarity,
546                          * We do not need the text anymore.
547                          */
548                         diff_free_filespec_blob(one);
549                         diff_free_filespec_blob(two);
550                 }
551                 dst_cnt++;
552                 display_progress(progress, (i+1)*rename_src_nr);
553         }
554         stop_progress(&progress);
555
556         /* cost matrix sorted by most to least similar pair */
557         qsort(mx, dst_cnt * NUM_CANDIDATE_PER_DST, sizeof(*mx), score_compare);
558
559         rename_count += find_renames(mx, dst_cnt, minimum_score, 0);
560         if (detect_rename == DIFF_DETECT_COPY)
561                 rename_count += find_renames(mx, dst_cnt, minimum_score, 1);
562         free(mx);
563
564  cleanup:
565         /* At this point, we have found some renames and copies and they
566          * are recorded in rename_dst.  The original list is still in *q.
567          */
568         DIFF_QUEUE_CLEAR(&outq);
569         for (i = 0; i < q->nr; i++) {
570                 struct diff_filepair *p = q->queue[i];
571                 struct diff_filepair *pair_to_free = NULL;
572
573                 if (DIFF_PAIR_UNMERGED(p)) {
574                         diff_q(&outq, p);
575                 }
576                 else if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
577                         /*
578                          * Creation
579                          *
580                          * We would output this create record if it has
581                          * not been turned into a rename/copy already.
582                          */
583                         struct diff_rename_dst *dst =
584                                 locate_rename_dst(p->two, 0);
585                         if (dst && dst->pair) {
586                                 diff_q(&outq, dst->pair);
587                                 pair_to_free = p;
588                         }
589                         else
590                                 /* no matching rename/copy source, so
591                                  * record this as a creation.
592                                  */
593                                 diff_q(&outq, p);
594                 }
595                 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) {
596                         /*
597                          * Deletion
598                          *
599                          * We would output this delete record if:
600                          *
601                          * (1) this is a broken delete and the counterpart
602                          *     broken create remains in the output; or
603                          * (2) this is not a broken delete, and rename_dst
604                          *     does not have a rename/copy to move p->one->path
605                          *     out of existence.
606                          *
607                          * Otherwise, the counterpart broken create
608                          * has been turned into a rename-edit; or
609                          * delete did not have a matching create to
610                          * begin with.
611                          */
612                         if (DIFF_PAIR_BROKEN(p)) {
613                                 /* broken delete */
614                                 struct diff_rename_dst *dst =
615                                         locate_rename_dst(p->one, 0);
616                                 if (dst && dst->pair)
617                                         /* counterpart is now rename/copy */
618                                         pair_to_free = p;
619                         }
620                         else {
621                                 if (p->one->rename_used)
622                                         /* this path remains */
623                                         pair_to_free = p;
624                         }
625
626                         if (pair_to_free)
627                                 ;
628                         else
629                                 diff_q(&outq, p);
630                 }
631                 else if (!diff_unmodified_pair(p))
632                         /* all the usual ones need to be kept */
633                         diff_q(&outq, p);
634                 else
635                         /* no need to keep unmodified pairs */
636                         pair_to_free = p;
637
638                 if (pair_to_free)
639                         diff_free_filepair(pair_to_free);
640         }
641         diff_debug_queue("done copying original", &outq);
642
643         free(q->queue);
644         *q = outq;
645         diff_debug_queue("done collapsing", q);
646
647         for (i = 0; i < rename_dst_nr; i++)
648                 free_filespec(rename_dst[i].two);
649
650         free(rename_dst);
651         rename_dst = NULL;
652         rename_dst_nr = rename_dst_alloc = 0;
653         free(rename_src);
654         rename_src = NULL;
655         rename_src_nr = rename_src_alloc = 0;
656         return;
657 }