Code

Fix git-diff A...B
[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"
8 /* Table of rename/copy destinations */
10 static struct diff_rename_dst {
11         struct diff_filespec *two;
12         struct diff_filepair *pair;
13 } *rename_dst;
14 static int rename_dst_nr, rename_dst_alloc;
16 static struct diff_rename_dst *locate_rename_dst(struct diff_filespec *two,
17                                                  int insert_ok)
18 {
19         int first, last;
21         first = 0;
22         last = rename_dst_nr;
23         while (last > first) {
24                 int next = (last + first) >> 1;
25                 struct diff_rename_dst *dst = &(rename_dst[next]);
26                 int cmp = strcmp(two->path, dst->two->path);
27                 if (!cmp)
28                         return dst;
29                 if (cmp < 0) {
30                         last = next;
31                         continue;
32                 }
33                 first = next+1;
34         }
35         /* not found */
36         if (!insert_ok)
37                 return NULL;
38         /* insert to make it at "first" */
39         if (rename_dst_alloc <= rename_dst_nr) {
40                 rename_dst_alloc = alloc_nr(rename_dst_alloc);
41                 rename_dst = xrealloc(rename_dst,
42                                       rename_dst_alloc * sizeof(*rename_dst));
43         }
44         rename_dst_nr++;
45         if (first < rename_dst_nr)
46                 memmove(rename_dst + first + 1, rename_dst + first,
47                         (rename_dst_nr - first - 1) * sizeof(*rename_dst));
48         rename_dst[first].two = alloc_filespec(two->path);
49         fill_filespec(rename_dst[first].two, two->sha1, two->mode);
50         rename_dst[first].pair = NULL;
51         return &(rename_dst[first]);
52 }
54 /* Table of rename/copy src files */
55 static struct diff_rename_src {
56         struct diff_filespec *one;
57         unsigned short score; /* to remember the break score */
58         unsigned src_path_left : 1;
59 } *rename_src;
60 static int rename_src_nr, rename_src_alloc;
62 static struct diff_rename_src *register_rename_src(struct diff_filespec *one,
63                                                    int src_path_left,
64                                                    unsigned short score)
65 {
66         int first, last;
68         first = 0;
69         last = rename_src_nr;
70         while (last > first) {
71                 int next = (last + first) >> 1;
72                 struct diff_rename_src *src = &(rename_src[next]);
73                 int cmp = strcmp(one->path, src->one->path);
74                 if (!cmp)
75                         return src;
76                 if (cmp < 0) {
77                         last = next;
78                         continue;
79                 }
80                 first = next+1;
81         }
83         /* insert to make it at "first" */
84         if (rename_src_alloc <= rename_src_nr) {
85                 rename_src_alloc = alloc_nr(rename_src_alloc);
86                 rename_src = xrealloc(rename_src,
87                                       rename_src_alloc * sizeof(*rename_src));
88         }
89         rename_src_nr++;
90         if (first < rename_src_nr)
91                 memmove(rename_src + first + 1, rename_src + first,
92                         (rename_src_nr - first - 1) * sizeof(*rename_src));
93         rename_src[first].one = one;
94         rename_src[first].score = score;
95         rename_src[first].src_path_left = src_path_left;
96         return &(rename_src[first]);
97 }
99 static int is_exact_match(struct diff_filespec *src,
100                           struct diff_filespec *dst,
101                           int contents_too)
103         if (src->sha1_valid && dst->sha1_valid &&
104             !memcmp(src->sha1, dst->sha1, 20))
105                 return 1;
106         if (!contents_too)
107                 return 0;
108         if (diff_populate_filespec(src, 1) || diff_populate_filespec(dst, 1))
109                 return 0;
110         if (src->size != dst->size)
111                 return 0;
112         if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0))
113                 return 0;
114         if (src->size == dst->size &&
115             !memcmp(src->data, dst->data, src->size))
116                 return 1;
117         return 0;
120 struct diff_score {
121         int src; /* index in rename_src */
122         int dst; /* index in rename_dst */
123         int score;
124 };
126 static int estimate_similarity(struct diff_filespec *src,
127                                struct diff_filespec *dst,
128                                int minimum_score)
130         /* src points at a file that existed in the original tree (or
131          * optionally a file in the destination tree) and dst points
132          * at a newly created file.  They may be quite similar, in which
133          * case we want to say src is renamed to dst or src is copied into
134          * dst, and then some edit has been applied to dst.
135          *
136          * Compare them and return how similar they are, representing
137          * the score as an integer between 0 and MAX_SCORE.
138          *
139          * When there is an exact match, it is considered a better
140          * match than anything else; the destination does not even
141          * call into this function in that case.
142          */
143         unsigned long max_size, delta_size, base_size, src_copied, literal_added;
144         unsigned long delta_limit;
145         int score;
147         /* We deal only with regular files.  Symlink renames are handled
148          * only when they are exact matches --- in other words, no edits
149          * after renaming.
150          */
151         if (!S_ISREG(src->mode) || !S_ISREG(dst->mode))
152                 return 0;
154         max_size = ((src->size > dst->size) ? src->size : dst->size);
155         base_size = ((src->size < dst->size) ? src->size : dst->size);
156         delta_size = max_size - base_size;
158         /* We would not consider edits that change the file size so
159          * drastically.  delta_size must be smaller than
160          * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
161          *
162          * Note that base_size == 0 case is handled here already
163          * and the final score computation below would not have a
164          * divide-by-zero issue.
165          */
166         if (base_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE)
167                 return 0;
169         if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0))
170                 return 0; /* error but caught downstream */
173         delta_limit = base_size * (MAX_SCORE-minimum_score) / MAX_SCORE;
174         if (diffcore_count_changes(src->data, src->size,
175                                    dst->data, dst->size,
176                                    &src->cnt_data, &dst->cnt_data,
177                                    delta_limit,
178                                    &src_copied, &literal_added))
179                 return 0;
181         /* How similar are they?
182          * what percentage of material in dst are from source?
183          */
184         if (!dst->size)
185                 score = 0; /* should not happen */
186         else
187                 score = src_copied * MAX_SCORE / max_size;
188         return score;
191 static void record_rename_pair(int dst_index, int src_index, int score)
193         struct diff_filespec *one, *two, *src, *dst;
194         struct diff_filepair *dp;
196         if (rename_dst[dst_index].pair)
197                 die("internal error: dst already matched.");
199         src = rename_src[src_index].one;
200         one = alloc_filespec(src->path);
201         fill_filespec(one, src->sha1, src->mode);
203         dst = rename_dst[dst_index].two;
204         two = alloc_filespec(dst->path);
205         fill_filespec(two, dst->sha1, dst->mode);
207         dp = diff_queue(NULL, one, two);
208         dp->renamed_pair = 1;
209         if (!strcmp(src->path, dst->path))
210                 dp->score = rename_src[src_index].score;
211         else
212                 dp->score = score;
213         dp->source_stays = rename_src[src_index].src_path_left;
214         rename_dst[dst_index].pair = dp;
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_)
223         const struct diff_score *a = a_, *b = b_;
224         return b->score - a->score;
227 static int compute_stays(struct diff_queue_struct *q,
228                          struct diff_filespec *one)
230         int i;
231         for (i = 0; i < q->nr; i++) {
232                 struct diff_filepair *p = q->queue[i];
233                 if (strcmp(one->path, p->two->path))
234                         continue;
235                 if (DIFF_PAIR_RENAME(p)) {
236                         return 0; /* something else is renamed into this */
237                 }
238         }
239         return 1;
242 void diffcore_rename(struct diff_options *options)
244         int detect_rename = options->detect_rename;
245         int minimum_score = options->rename_score;
246         int rename_limit = options->rename_limit;
247         struct diff_queue_struct *q = &diff_queued_diff;
248         struct diff_queue_struct outq;
249         struct diff_score *mx;
250         int i, j, rename_count, contents_too;
251         int num_create, num_src, dst_cnt;
253         if (!minimum_score)
254                 minimum_score = DEFAULT_RENAME_SCORE;
255         rename_count = 0;
257         for (i = 0; i < q->nr; i++) {
258                 struct diff_filepair *p = q->queue[i];
259                 if (!DIFF_FILE_VALID(p->one))
260                         if (!DIFF_FILE_VALID(p->two))
261                                 continue; /* unmerged */
262                         else
263                                 locate_rename_dst(p->two, 1);
264                 else if (!DIFF_FILE_VALID(p->two)) {
265                         /* If the source is a broken "delete", and
266                          * they did not really want to get broken,
267                          * that means the source actually stays.
268                          */
269                         int stays = (p->broken_pair && !p->score);
270                         register_rename_src(p->one, stays, p->score);
271                 }
272                 else if (detect_rename == DIFF_DETECT_COPY)
273                         register_rename_src(p->one, 1, p->score);
274         }
275         if (rename_dst_nr == 0 || rename_src_nr == 0 ||
276             (0 < rename_limit && rename_limit < rename_dst_nr))
277                 goto cleanup; /* nothing to do */
279         /* We really want to cull the candidates list early
280          * with cheap tests in order to avoid doing deltas.
281          * The first round matches up the up-to-date entries,
282          * and then during the second round we try to match
283          * cache-dirty entries as well.
284          */
285         for (contents_too = 0; contents_too < 2; contents_too++) {
286                 for (i = 0; i < rename_dst_nr; i++) {
287                         struct diff_filespec *two = rename_dst[i].two;
288                         if (rename_dst[i].pair)
289                                 continue; /* dealt with an earlier round */
290                         for (j = 0; j < rename_src_nr; j++) {
291                                 struct diff_filespec *one = rename_src[j].one;
292                                 if (!is_exact_match(one, two, contents_too))
293                                         continue;
294                                 record_rename_pair(i, j, MAX_SCORE);
295                                 rename_count++;
296                                 break; /* we are done with this entry */
297                         }
298                 }
299         }
301         /* Have we run out the created file pool?  If so we can avoid
302          * doing the delta matrix altogether.
303          */
304         if (rename_count == rename_dst_nr)
305                 goto cleanup;
307         if (minimum_score == MAX_SCORE)
308                 goto cleanup;
310         num_create = (rename_dst_nr - rename_count);
311         num_src = rename_src_nr;
312         mx = xmalloc(sizeof(*mx) * num_create * num_src);
313         for (dst_cnt = i = 0; i < rename_dst_nr; i++) {
314                 int base = dst_cnt * num_src;
315                 struct diff_filespec *two = rename_dst[i].two;
316                 if (rename_dst[i].pair)
317                         continue; /* dealt with exact match already. */
318                 for (j = 0; j < rename_src_nr; j++) {
319                         struct diff_filespec *one = rename_src[j].one;
320                         struct diff_score *m = &mx[base+j];
321                         m->src = j;
322                         m->dst = i;
323                         m->score = estimate_similarity(one, two,
324                                                        minimum_score);
325                 }
326                 /* We do not need the text anymore */
327                 diff_free_filespec_data(two);
328                 dst_cnt++;
329         }
330         /* cost matrix sorted by most to least similar pair */
331         qsort(mx, num_create * num_src, sizeof(*mx), score_compare);
332         for (i = 0; i < num_create * num_src; i++) {
333                 struct diff_rename_dst *dst = &rename_dst[mx[i].dst];
334                 if (dst->pair)
335                         continue; /* already done, either exact or fuzzy. */
336                 if (mx[i].score < minimum_score)
337                         break; /* there is no more usable pair. */
338                 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
339                 rename_count++;
340         }
341         free(mx);
343  cleanup:
344         /* At this point, we have found some renames and copies and they
345          * are recorded in rename_dst.  The original list is still in *q.
346          */
347         outq.queue = NULL;
348         outq.nr = outq.alloc = 0;
349         for (i = 0; i < q->nr; i++) {
350                 struct diff_filepair *p = q->queue[i];
351                 struct diff_filepair *pair_to_free = NULL;
353                 if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
354                         /*
355                          * Creation
356                          *
357                          * We would output this create record if it has
358                          * not been turned into a rename/copy already.
359                          */
360                         struct diff_rename_dst *dst =
361                                 locate_rename_dst(p->two, 0);
362                         if (dst && dst->pair) {
363                                 diff_q(&outq, dst->pair);
364                                 pair_to_free = p;
365                         }
366                         else
367                                 /* no matching rename/copy source, so
368                                  * record this as a creation.
369                                  */
370                                 diff_q(&outq, p);
371                 }
372                 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) {
373                         /*
374                          * Deletion
375                          *
376                          * We would output this delete record if:
377                          *
378                          * (1) this is a broken delete and the counterpart
379                          *     broken create remains in the output; or
380                          * (2) this is not a broken delete, and rename_dst
381                          *     does not have a rename/copy to move p->one->path
382                          *     out of existence.
383                          *
384                          * Otherwise, the counterpart broken create
385                          * has been turned into a rename-edit; or
386                          * delete did not have a matching create to
387                          * begin with.
388                          */
389                         if (DIFF_PAIR_BROKEN(p)) {
390                                 /* broken delete */
391                                 struct diff_rename_dst *dst =
392                                         locate_rename_dst(p->one, 0);
393                                 if (dst && dst->pair)
394                                         /* counterpart is now rename/copy */
395                                         pair_to_free = p;
396                         }
397                         else {
398                                 for (j = 0; j < rename_dst_nr; j++) {
399                                         if (!rename_dst[j].pair)
400                                                 continue;
401                                         if (strcmp(rename_dst[j].pair->
402                                                    one->path,
403                                                    p->one->path))
404                                                 continue;
405                                         break;
406                                 }
407                                 if (j < rename_dst_nr)
408                                         /* this path remains */
409                                         pair_to_free = p;
410                         }
412                         if (pair_to_free)
413                                 ;
414                         else
415                                 diff_q(&outq, p);
416                 }
417                 else if (!diff_unmodified_pair(p))
418                         /* all the usual ones need to be kept */
419                         diff_q(&outq, p);
420                 else
421                         /* no need to keep unmodified pairs */
422                         pair_to_free = p;
424                 if (pair_to_free)
425                         diff_free_filepair(pair_to_free);
426         }
427         diff_debug_queue("done copying original", &outq);
429         free(q->queue);
430         *q = outq;
431         diff_debug_queue("done collapsing", q);
433         /* We need to see which rename source really stays here;
434          * earlier we only checked if the path is left in the result,
435          * but even if a path remains in the result, if that is coming
436          * from copying something else on top of it, then the original
437          * source is lost and does not stay.
438          */
439         for (i = 0; i < q->nr; i++) {
440                 struct diff_filepair *p = q->queue[i];
441                 if (DIFF_PAIR_RENAME(p) && p->source_stays) {
442                         /* If one appears as the target of a rename-copy,
443                          * then mark p->source_stays = 0; otherwise
444                          * leave it as is.
445                          */
446                         p->source_stays = compute_stays(q, p->one);
447                 }
448         }
450         for (i = 0; i < rename_dst_nr; i++) {
451                 diff_free_filespec_data(rename_dst[i].two);
452                 free(rename_dst[i].two);
453         }
455         free(rename_dst);
456         rename_dst = NULL;
457         rename_dst_nr = rename_dst_alloc = 0;
458         free(rename_src);
459         rename_src = NULL;
460         rename_src_nr = rename_src_alloc = 0;
461         return;