grep: honor --textconv for the case rev:path
[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 "hash.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         if (rename_dst_alloc <= rename_dst_nr) {
42                 rename_dst_alloc = alloc_nr(rename_dst_alloc);
43                 rename_dst = xrealloc(rename_dst,
44                                       rename_dst_alloc * sizeof(*rename_dst));
45         }
46         rename_dst_nr++;
47         if (first < rename_dst_nr)
48                 memmove(rename_dst + first + 1, rename_dst + first,
49                         (rename_dst_nr - first - 1) * sizeof(*rename_dst));
50         rename_dst[first].two = alloc_filespec(two->path);
51         fill_filespec(rename_dst[first].two, two->sha1, two->sha1_valid, two->mode);
52         rename_dst[first].pair = NULL;
53         return &(rename_dst[first]);
54 }
55
56 /* Table of rename/copy src files */
57 static struct diff_rename_src {
58         struct diff_filepair *p;
59         unsigned short score; /* to remember the break score */
60 } *rename_src;
61 static int rename_src_nr, rename_src_alloc;
62
63 static struct diff_rename_src *register_rename_src(struct diff_filepair *p)
64 {
65         int first, last;
66         struct diff_filespec *one = p->one;
67         unsigned short score = p->score;
68
69         first = 0;
70         last = rename_src_nr;
71         while (last > first) {
72                 int next = (last + first) >> 1;
73                 struct diff_rename_src *src = &(rename_src[next]);
74                 int cmp = strcmp(one->path, src->p->one->path);
75                 if (!cmp)
76                         return src;
77                 if (cmp < 0) {
78                         last = next;
79                         continue;
80                 }
81                 first = next+1;
82         }
83
84         /* insert to make it at "first" */
85         if (rename_src_alloc <= rename_src_nr) {
86                 rename_src_alloc = alloc_nr(rename_src_alloc);
87                 rename_src = xrealloc(rename_src,
88                                       rename_src_alloc * sizeof(*rename_src));
89         }
90         rename_src_nr++;
91         if (first < rename_src_nr)
92                 memmove(rename_src + first + 1, rename_src + first,
93                         (rename_src_nr - first - 1) * sizeof(*rename_src));
94         rename_src[first].p = p;
95         rename_src[first].score = score;
96         return &(rename_src[first]);
97 }
98
99 static int basename_same(struct diff_filespec *src, struct diff_filespec *dst)
100 {
101         int src_len = strlen(src->path), dst_len = strlen(dst->path);
102         while (src_len && dst_len) {
103                 char c1 = src->path[--src_len];
104                 char c2 = dst->path[--dst_len];
105                 if (c1 != c2)
106                         return 0;
107                 if (c1 == '/')
108                         return 1;
109         }
110         return (!src_len || src->path[src_len - 1] == '/') &&
111                 (!dst_len || dst->path[dst_len - 1] == '/');
112 }
113
114 struct diff_score {
115         int src; /* index in rename_src */
116         int dst; /* index in rename_dst */
117         unsigned short score;
118         short name_score;
119 };
120
121 static int estimate_similarity(struct diff_filespec *src,
122                                struct diff_filespec *dst,
123                                int minimum_score)
124 {
125         /* src points at a file that existed in the original tree (or
126          * optionally a file in the destination tree) and dst points
127          * at a newly created file.  They may be quite similar, in which
128          * case we want to say src is renamed to dst or src is copied into
129          * dst, and then some edit has been applied to dst.
130          *
131          * Compare them and return how similar they are, representing
132          * the score as an integer between 0 and MAX_SCORE.
133          *
134          * When there is an exact match, it is considered a better
135          * match than anything else; the destination does not even
136          * call into this function in that case.
137          */
138         unsigned long max_size, delta_size, base_size, src_copied, literal_added;
139         unsigned long delta_limit;
140         int score;
141
142         /* We deal only with regular files.  Symlink renames are handled
143          * only when they are exact matches --- in other words, no edits
144          * after renaming.
145          */
146         if (!S_ISREG(src->mode) || !S_ISREG(dst->mode))
147                 return 0;
148
149         /*
150          * Need to check that source and destination sizes are
151          * filled in before comparing them.
152          *
153          * If we already have "cnt_data" filled in, we know it's
154          * all good (avoid checking the size for zero, as that
155          * is a possible size - we really should have a flag to
156          * say whether the size is valid or not!)
157          */
158         if (!src->cnt_data && diff_populate_filespec(src, 1))
159                 return 0;
160         if (!dst->cnt_data && diff_populate_filespec(dst, 1))
161                 return 0;
162
163         max_size = ((src->size > dst->size) ? src->size : dst->size);
164         base_size = ((src->size < dst->size) ? src->size : dst->size);
165         delta_size = max_size - base_size;
166
167         /* We would not consider edits that change the file size so
168          * drastically.  delta_size must be smaller than
169          * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
170          *
171          * Note that base_size == 0 case is handled here already
172          * and the final score computation below would not have a
173          * divide-by-zero issue.
174          */
175         if (max_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE)
176                 return 0;
177
178         if (!src->cnt_data && diff_populate_filespec(src, 0))
179                 return 0;
180         if (!dst->cnt_data && diff_populate_filespec(dst, 0))
181                 return 0;
182
183         delta_limit = (unsigned long)
184                 (base_size * (MAX_SCORE-minimum_score) / MAX_SCORE);
185         if (diffcore_count_changes(src, dst,
186                                    &src->cnt_data, &dst->cnt_data,
187                                    delta_limit,
188                                    &src_copied, &literal_added))
189                 return 0;
190
191         /* How similar are they?
192          * what percentage of material in dst are from source?
193          */
194         if (!dst->size)
195                 score = 0; /* should not happen */
196         else
197                 score = (int)(src_copied * MAX_SCORE / max_size);
198         return score;
199 }
200
201 static void record_rename_pair(int dst_index, int src_index, int score)
202 {
203         struct diff_filespec *src, *dst;
204         struct diff_filepair *dp;
205
206         if (rename_dst[dst_index].pair)
207                 die("internal error: dst already matched.");
208
209         src = rename_src[src_index].p->one;
210         src->rename_used++;
211         src->count++;
212
213         dst = rename_dst[dst_index].two;
214         dst->count++;
215
216         dp = diff_queue(NULL, src, dst);
217         dp->renamed_pair = 1;
218         if (!strcmp(src->path, dst->path))
219                 dp->score = rename_src[src_index].score;
220         else
221                 dp->score = score;
222         rename_dst[dst_index].pair = dp;
223 }
224
225 /*
226  * We sort the rename similarity matrix with the score, in descending
227  * order (the most similar first).
228  */
229 static int score_compare(const void *a_, const void *b_)
230 {
231         const struct diff_score *a = a_, *b = b_;
232
233         /* sink the unused ones to the bottom */
234         if (a->dst < 0)
235                 return (0 <= b->dst);
236         else if (b->dst < 0)
237                 return -1;
238
239         if (a->score == b->score)
240                 return b->name_score - a->name_score;
241
242         return b->score - a->score;
243 }
244
245 struct file_similarity {
246         int src_dst, index;
247         struct diff_filespec *filespec;
248         struct file_similarity *next;
249 };
250
251 static int find_identical_files(struct file_similarity *src,
252                                 struct file_similarity *dst,
253                                 struct diff_options *options)
254 {
255         int renames = 0;
256
257         /*
258          * Walk over all the destinations ...
259          */
260         do {
261                 struct diff_filespec *target = dst->filespec;
262                 struct file_similarity *p, *best;
263                 int i = 100, best_score = -1;
264
265                 /*
266                  * .. to find the best source match
267                  */
268                 best = NULL;
269                 for (p = src; p; p = p->next) {
270                         int score;
271                         struct diff_filespec *source = p->filespec;
272
273                         /* False hash collision? */
274                         if (hashcmp(source->sha1, target->sha1))
275                                 continue;
276                         /* Non-regular files? If so, the modes must match! */
277                         if (!S_ISREG(source->mode) || !S_ISREG(target->mode)) {
278                                 if (source->mode != target->mode)
279                                         continue;
280                         }
281                         /* Give higher scores to sources that haven't been used already */
282                         score = !source->rename_used;
283                         if (source->rename_used && options->detect_rename != DIFF_DETECT_COPY)
284                                 continue;
285                         score += basename_same(source, target);
286                         if (score > best_score) {
287                                 best = p;
288                                 best_score = score;
289                                 if (score == 2)
290                                         break;
291                         }
292
293                         /* Too many identical alternatives? Pick one */
294                         if (!--i)
295                                 break;
296                 }
297                 if (best) {
298                         record_rename_pair(dst->index, best->index, MAX_SCORE);
299                         renames++;
300                 }
301         } while ((dst = dst->next) != NULL);
302         return renames;
303 }
304
305 static void free_similarity_list(struct file_similarity *p)
306 {
307         while (p) {
308                 struct file_similarity *entry = p;
309                 p = p->next;
310                 free(entry);
311         }
312 }
313
314 static int find_same_files(void *ptr, void *data)
315 {
316         int ret;
317         struct file_similarity *p = ptr;
318         struct file_similarity *src = NULL, *dst = NULL;
319         struct diff_options *options = data;
320
321         /* Split the hash list up into sources and destinations */
322         do {
323                 struct file_similarity *entry = p;
324                 p = p->next;
325                 if (entry->src_dst < 0) {
326                         entry->next = src;
327                         src = entry;
328                 } else {
329                         entry->next = dst;
330                         dst = entry;
331                 }
332         } while (p);
333
334         /*
335          * If we have both sources *and* destinations, see if
336          * we can match them up
337          */
338         ret = (src && dst) ? find_identical_files(src, dst, options) : 0;
339
340         /* Free the hashes and return the number of renames found */
341         free_similarity_list(src);
342         free_similarity_list(dst);
343         return ret;
344 }
345
346 static unsigned int hash_filespec(struct diff_filespec *filespec)
347 {
348         unsigned int hash;
349         if (!filespec->sha1_valid) {
350                 if (diff_populate_filespec(filespec, 0))
351                         return 0;
352                 hash_sha1_file(filespec->data, filespec->size, "blob", filespec->sha1);
353         }
354         memcpy(&hash, filespec->sha1, sizeof(hash));
355         return hash;
356 }
357
358 static void insert_file_table(struct hash_table *table, int src_dst, int index, struct diff_filespec *filespec)
359 {
360         void **pos;
361         unsigned int hash;
362         struct file_similarity *entry = xmalloc(sizeof(*entry));
363
364         entry->src_dst = src_dst;
365         entry->index = index;
366         entry->filespec = filespec;
367         entry->next = NULL;
368
369         hash = hash_filespec(filespec);
370         pos = insert_hash(hash, entry, table);
371
372         /* We already had an entry there? */
373         if (pos) {
374                 entry->next = *pos;
375                 *pos = entry;
376         }
377 }
378
379 /*
380  * Find exact renames first.
381  *
382  * The first round matches up the up-to-date entries,
383  * and then during the second round we try to match
384  * cache-dirty entries as well.
385  */
386 static int find_exact_renames(struct diff_options *options)
387 {
388         int i;
389         struct hash_table file_table;
390
391         init_hash(&file_table);
392         preallocate_hash(&file_table, rename_src_nr + rename_dst_nr);
393         for (i = 0; i < rename_src_nr; i++)
394                 insert_file_table(&file_table, -1, i, rename_src[i].p->one);
395
396         for (i = 0; i < rename_dst_nr; i++)
397                 insert_file_table(&file_table, 1, i, rename_dst[i].two);
398
399         /* Find the renames */
400         i = for_each_hash(&file_table, find_same_files, options);
401
402         /* .. and free the hash data structure */
403         free_hash(&file_table);
404
405         return i;
406 }
407
408 #define NUM_CANDIDATE_PER_DST 4
409 static void record_if_better(struct diff_score m[], struct diff_score *o)
410 {
411         int i, worst;
412
413         /* find the worst one */
414         worst = 0;
415         for (i = 1; i < NUM_CANDIDATE_PER_DST; i++)
416                 if (score_compare(&m[i], &m[worst]) > 0)
417                         worst = i;
418
419         /* is it better than the worst one? */
420         if (score_compare(&m[worst], o) > 0)
421                 m[worst] = *o;
422 }
423
424 /*
425  * Returns:
426  * 0 if we are under the limit;
427  * 1 if we need to disable inexact rename detection;
428  * 2 if we would be under the limit if we were given -C instead of -C -C.
429  */
430 static int too_many_rename_candidates(int num_create,
431                                       struct diff_options *options)
432 {
433         int rename_limit = options->rename_limit;
434         int num_src = rename_src_nr;
435         int i;
436
437         options->needed_rename_limit = 0;
438
439         /*
440          * This basically does a test for the rename matrix not
441          * growing larger than a "rename_limit" square matrix, ie:
442          *
443          *    num_create * num_src > rename_limit * rename_limit
444          *
445          * but handles the potential overflow case specially (and we
446          * assume at least 32-bit integers)
447          */
448         if (rename_limit <= 0 || rename_limit > 32767)
449                 rename_limit = 32767;
450         if ((num_create <= rename_limit || num_src <= rename_limit) &&
451             (num_create * num_src <= rename_limit * rename_limit))
452                 return 0;
453
454         options->needed_rename_limit =
455                 num_src > num_create ? num_src : num_create;
456
457         /* Are we running under -C -C? */
458         if (!DIFF_OPT_TST(options, FIND_COPIES_HARDER))
459                 return 1;
460
461         /* Would we bust the limit if we were running under -C? */
462         for (num_src = i = 0; i < rename_src_nr; i++) {
463                 if (diff_unmodified_pair(rename_src[i].p))
464                         continue;
465                 num_src++;
466         }
467         if ((num_create <= rename_limit || num_src <= rename_limit) &&
468             (num_create * num_src <= rename_limit * rename_limit))
469                 return 2;
470         return 1;
471 }
472
473 static int find_renames(struct diff_score *mx, int dst_cnt, int minimum_score, int copies)
474 {
475         int count = 0, i;
476
477         for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) {
478                 struct diff_rename_dst *dst;
479
480                 if ((mx[i].dst < 0) ||
481                     (mx[i].score < minimum_score))
482                         break; /* there is no more usable pair. */
483                 dst = &rename_dst[mx[i].dst];
484                 if (dst->pair)
485                         continue; /* already done, either exact or fuzzy. */
486                 if (!copies && rename_src[mx[i].src].p->one->rename_used)
487                         continue;
488                 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
489                 count++;
490         }
491         return count;
492 }
493
494 void diffcore_rename(struct diff_options *options)
495 {
496         int detect_rename = options->detect_rename;
497         int minimum_score = options->rename_score;
498         struct diff_queue_struct *q = &diff_queued_diff;
499         struct diff_queue_struct outq;
500         struct diff_score *mx;
501         int i, j, rename_count, skip_unmodified = 0;
502         int num_create, dst_cnt;
503         struct progress *progress = NULL;
504
505         if (!minimum_score)
506                 minimum_score = DEFAULT_RENAME_SCORE;
507
508         for (i = 0; i < q->nr; i++) {
509                 struct diff_filepair *p = q->queue[i];
510                 if (!DIFF_FILE_VALID(p->one)) {
511                         if (!DIFF_FILE_VALID(p->two))
512                                 continue; /* unmerged */
513                         else if (options->single_follow &&
514                                  strcmp(options->single_follow, p->two->path))
515                                 continue; /* not interested */
516                         else if (!DIFF_OPT_TST(options, RENAME_EMPTY) &&
517                                  is_empty_blob_sha1(p->two->sha1))
518                                 continue;
519                         else
520                                 locate_rename_dst(p->two, 1);
521                 }
522                 else if (!DIFF_OPT_TST(options, RENAME_EMPTY) &&
523                          is_empty_blob_sha1(p->one->sha1))
524                         continue;
525                 else if (!DIFF_PAIR_UNMERGED(p) && !DIFF_FILE_VALID(p->two)) {
526                         /*
527                          * If the source is a broken "delete", and
528                          * they did not really want to get broken,
529                          * that means the source actually stays.
530                          * So we increment the "rename_used" score
531                          * by one, to indicate ourselves as a user
532                          */
533                         if (p->broken_pair && !p->score)
534                                 p->one->rename_used++;
535                         register_rename_src(p);
536                 }
537                 else if (detect_rename == DIFF_DETECT_COPY) {
538                         /*
539                          * Increment the "rename_used" score by
540                          * one, to indicate ourselves as a user.
541                          */
542                         p->one->rename_used++;
543                         register_rename_src(p);
544                 }
545         }
546         if (rename_dst_nr == 0 || rename_src_nr == 0)
547                 goto cleanup; /* nothing to do */
548
549         /*
550          * We really want to cull the candidates list early
551          * with cheap tests in order to avoid doing deltas.
552          */
553         rename_count = find_exact_renames(options);
554
555         /* Did we only want exact renames? */
556         if (minimum_score == MAX_SCORE)
557                 goto cleanup;
558
559         /*
560          * Calculate how many renames are left (but all the source
561          * files still remain as options for rename/copies!)
562          */
563         num_create = (rename_dst_nr - rename_count);
564
565         /* All done? */
566         if (!num_create)
567                 goto cleanup;
568
569         switch (too_many_rename_candidates(num_create, options)) {
570         case 1:
571                 goto cleanup;
572         case 2:
573                 options->degraded_cc_to_c = 1;
574                 skip_unmodified = 1;
575                 break;
576         default:
577                 break;
578         }
579
580         if (options->show_rename_progress) {
581                 progress = start_progress_delay(
582                                 "Performing inexact rename detection",
583                                 rename_dst_nr * rename_src_nr, 50, 1);
584         }
585
586         mx = xcalloc(num_create * NUM_CANDIDATE_PER_DST, sizeof(*mx));
587         for (dst_cnt = i = 0; i < rename_dst_nr; i++) {
588                 struct diff_filespec *two = rename_dst[i].two;
589                 struct diff_score *m;
590
591                 if (rename_dst[i].pair)
592                         continue; /* dealt with exact match already. */
593
594                 m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST];
595                 for (j = 0; j < NUM_CANDIDATE_PER_DST; j++)
596                         m[j].dst = -1;
597
598                 for (j = 0; j < rename_src_nr; j++) {
599                         struct diff_filespec *one = rename_src[j].p->one;
600                         struct diff_score this_src;
601
602                         if (skip_unmodified &&
603                             diff_unmodified_pair(rename_src[j].p))
604                                 continue;
605
606                         this_src.score = estimate_similarity(one, two,
607                                                              minimum_score);
608                         this_src.name_score = basename_same(one, two);
609                         this_src.dst = i;
610                         this_src.src = j;
611                         record_if_better(m, &this_src);
612                         /*
613                          * Once we run estimate_similarity,
614                          * We do not need the text anymore.
615                          */
616                         diff_free_filespec_blob(one);
617                         diff_free_filespec_blob(two);
618                 }
619                 dst_cnt++;
620                 display_progress(progress, (i+1)*rename_src_nr);
621         }
622         stop_progress(&progress);
623
624         /* cost matrix sorted by most to least similar pair */
625         qsort(mx, dst_cnt * NUM_CANDIDATE_PER_DST, sizeof(*mx), score_compare);
626
627         rename_count += find_renames(mx, dst_cnt, minimum_score, 0);
628         if (detect_rename == DIFF_DETECT_COPY)
629                 rename_count += find_renames(mx, dst_cnt, minimum_score, 1);
630         free(mx);
631
632  cleanup:
633         /* At this point, we have found some renames and copies and they
634          * are recorded in rename_dst.  The original list is still in *q.
635          */
636         DIFF_QUEUE_CLEAR(&outq);
637         for (i = 0; i < q->nr; i++) {
638                 struct diff_filepair *p = q->queue[i];
639                 struct diff_filepair *pair_to_free = NULL;
640
641                 if (DIFF_PAIR_UNMERGED(p)) {
642                         diff_q(&outq, p);
643                 }
644                 else if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
645                         /*
646                          * Creation
647                          *
648                          * We would output this create record if it has
649                          * not been turned into a rename/copy already.
650                          */
651                         struct diff_rename_dst *dst =
652                                 locate_rename_dst(p->two, 0);
653                         if (dst && dst->pair) {
654                                 diff_q(&outq, dst->pair);
655                                 pair_to_free = p;
656                         }
657                         else
658                                 /* no matching rename/copy source, so
659                                  * record this as a creation.
660                                  */
661                                 diff_q(&outq, p);
662                 }
663                 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) {
664                         /*
665                          * Deletion
666                          *
667                          * We would output this delete record if:
668                          *
669                          * (1) this is a broken delete and the counterpart
670                          *     broken create remains in the output; or
671                          * (2) this is not a broken delete, and rename_dst
672                          *     does not have a rename/copy to move p->one->path
673                          *     out of existence.
674                          *
675                          * Otherwise, the counterpart broken create
676                          * has been turned into a rename-edit; or
677                          * delete did not have a matching create to
678                          * begin with.
679                          */
680                         if (DIFF_PAIR_BROKEN(p)) {
681                                 /* broken delete */
682                                 struct diff_rename_dst *dst =
683                                         locate_rename_dst(p->one, 0);
684                                 if (dst && dst->pair)
685                                         /* counterpart is now rename/copy */
686                                         pair_to_free = p;
687                         }
688                         else {
689                                 if (p->one->rename_used)
690                                         /* this path remains */
691                                         pair_to_free = p;
692                         }
693
694                         if (pair_to_free)
695                                 ;
696                         else
697                                 diff_q(&outq, p);
698                 }
699                 else if (!diff_unmodified_pair(p))
700                         /* all the usual ones need to be kept */
701                         diff_q(&outq, p);
702                 else
703                         /* no need to keep unmodified pairs */
704                         pair_to_free = p;
705
706                 if (pair_to_free)
707                         diff_free_filepair(pair_to_free);
708         }
709         diff_debug_queue("done copying original", &outq);
710
711         free(q->queue);
712         *q = outq;
713         diff_debug_queue("done collapsing", q);
714
715         for (i = 0; i < rename_dst_nr; i++)
716                 free_filespec(rename_dst[i].two);
717
718         free(rename_dst);
719         rename_dst = NULL;
720         rename_dst_nr = rename_dst_alloc = 0;
721         free(rename_src);
722         rename_src = NULL;
723         rename_src_nr = rename_src_alloc = 0;
724         return;
725 }