Merge branch 'maint' to sync with 1.5.2.5
[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
8 /* Table of rename/copy destinations */
9
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;
15
16 static struct diff_rename_dst *locate_rename_dst(struct diff_filespec *two,
17                                                  int insert_ok)
18 {
19         int first, last;
20
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 }
53
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;
61
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;
67
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         }
82
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 }
98
99 static int is_exact_match(struct diff_filespec *src,
100                           struct diff_filespec *dst,
101                           int contents_too)
102 {
103         if (src->sha1_valid && dst->sha1_valid &&
104             !hashcmp(src->sha1, dst->sha1))
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 (src->sha1_valid && dst->sha1_valid)
113             return !hashcmp(src->sha1, dst->sha1);
114         if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0))
115                 return 0;
116         if (src->size == dst->size &&
117             !memcmp(src->data, dst->data, src->size))
118                 return 1;
119         return 0;
120 }
121
122 static int basename_same(struct diff_filespec *src, struct diff_filespec *dst)
123 {
124         int src_len = strlen(src->path), dst_len = strlen(dst->path);
125         while (src_len && dst_len) {
126                 char c1 = src->path[--src_len];
127                 char c2 = dst->path[--dst_len];
128                 if (c1 != c2)
129                         return 0;
130                 if (c1 == '/')
131                         return 1;
132         }
133         return (!src_len || src->path[src_len - 1] == '/') &&
134                 (!dst_len || dst->path[dst_len - 1] == '/');
135 }
136
137 struct diff_score {
138         int src; /* index in rename_src */
139         int dst; /* index in rename_dst */
140         int score;
141         int name_score;
142 };
143
144 static int estimate_similarity(struct diff_filespec *src,
145                                struct diff_filespec *dst,
146                                int minimum_score)
147 {
148         /* src points at a file that existed in the original tree (or
149          * optionally a file in the destination tree) and dst points
150          * at a newly created file.  They may be quite similar, in which
151          * case we want to say src is renamed to dst or src is copied into
152          * dst, and then some edit has been applied to dst.
153          *
154          * Compare them and return how similar they are, representing
155          * the score as an integer between 0 and MAX_SCORE.
156          *
157          * When there is an exact match, it is considered a better
158          * match than anything else; the destination does not even
159          * call into this function in that case.
160          */
161         unsigned long max_size, delta_size, base_size, src_copied, literal_added;
162         unsigned long delta_limit;
163         int score;
164
165         /* We deal only with regular files.  Symlink renames are handled
166          * only when they are exact matches --- in other words, no edits
167          * after renaming.
168          */
169         if (!S_ISREG(src->mode) || !S_ISREG(dst->mode))
170                 return 0;
171
172         max_size = ((src->size > dst->size) ? src->size : dst->size);
173         base_size = ((src->size < dst->size) ? src->size : dst->size);
174         delta_size = max_size - base_size;
175
176         /* We would not consider edits that change the file size so
177          * drastically.  delta_size must be smaller than
178          * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
179          *
180          * Note that base_size == 0 case is handled here already
181          * and the final score computation below would not have a
182          * divide-by-zero issue.
183          */
184         if (base_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE)
185                 return 0;
186
187         if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0))
188                 return 0; /* error but caught downstream */
189
190
191         delta_limit = (unsigned long)
192                 (base_size * (MAX_SCORE-minimum_score) / MAX_SCORE);
193         if (diffcore_count_changes(src, dst,
194                                    &src->cnt_data, &dst->cnt_data,
195                                    delta_limit,
196                                    &src_copied, &literal_added))
197                 return 0;
198
199         /* How similar are they?
200          * what percentage of material in dst are from source?
201          */
202         if (!dst->size)
203                 score = 0; /* should not happen */
204         else
205                 score = (int)(src_copied * MAX_SCORE / max_size);
206         return score;
207 }
208
209 static void record_rename_pair(int dst_index, int src_index, int score)
210 {
211         struct diff_filespec *one, *two, *src, *dst;
212         struct diff_filepair *dp;
213
214         if (rename_dst[dst_index].pair)
215                 die("internal error: dst already matched.");
216
217         src = rename_src[src_index].one;
218         one = alloc_filespec(src->path);
219         fill_filespec(one, src->sha1, src->mode);
220
221         dst = rename_dst[dst_index].two;
222         two = alloc_filespec(dst->path);
223         fill_filespec(two, dst->sha1, dst->mode);
224
225         dp = diff_queue(NULL, one, two);
226         dp->renamed_pair = 1;
227         if (!strcmp(src->path, dst->path))
228                 dp->score = rename_src[src_index].score;
229         else
230                 dp->score = score;
231         dp->source_stays = rename_src[src_index].src_path_left;
232         rename_dst[dst_index].pair = dp;
233 }
234
235 /*
236  * We sort the rename similarity matrix with the score, in descending
237  * order (the most similar first).
238  */
239 static int score_compare(const void *a_, const void *b_)
240 {
241         const struct diff_score *a = a_, *b = b_;
242
243         if (a->score == b->score)
244                 return b->name_score - a->name_score;
245
246         return b->score - a->score;
247 }
248
249 static int compute_stays(struct diff_queue_struct *q,
250                          struct diff_filespec *one)
251 {
252         int i;
253         for (i = 0; i < q->nr; i++) {
254                 struct diff_filepair *p = q->queue[i];
255                 if (strcmp(one->path, p->two->path))
256                         continue;
257                 if (DIFF_PAIR_RENAME(p)) {
258                         return 0; /* something else is renamed into this */
259                 }
260         }
261         return 1;
262 }
263
264 void diffcore_rename(struct diff_options *options)
265 {
266         int detect_rename = options->detect_rename;
267         int minimum_score = options->rename_score;
268         int rename_limit = options->rename_limit;
269         struct diff_queue_struct *q = &diff_queued_diff;
270         struct diff_queue_struct outq;
271         struct diff_score *mx;
272         int i, j, rename_count, contents_too;
273         int num_create, num_src, dst_cnt;
274
275         if (!minimum_score)
276                 minimum_score = DEFAULT_RENAME_SCORE;
277         rename_count = 0;
278
279         for (i = 0; i < q->nr; i++) {
280                 struct diff_filepair *p = q->queue[i];
281                 if (!DIFF_FILE_VALID(p->one)) {
282                         if (!DIFF_FILE_VALID(p->two))
283                                 continue; /* unmerged */
284                         else if (options->single_follow &&
285                                  strcmp(options->single_follow, p->two->path))
286                                 continue; /* not interested */
287                         else
288                                 locate_rename_dst(p->two, 1);
289                 }
290                 else if (!DIFF_FILE_VALID(p->two)) {
291                         /* If the source is a broken "delete", and
292                          * they did not really want to get broken,
293                          * that means the source actually stays.
294                          */
295                         int stays = (p->broken_pair && !p->score);
296                         register_rename_src(p->one, stays, p->score);
297                 }
298                 else if (detect_rename == DIFF_DETECT_COPY)
299                         register_rename_src(p->one, 1, p->score);
300         }
301         if (rename_dst_nr == 0 || rename_src_nr == 0 ||
302             (0 < rename_limit && rename_limit < rename_dst_nr))
303                 goto cleanup; /* nothing to do */
304
305         /* We really want to cull the candidates list early
306          * with cheap tests in order to avoid doing deltas.
307          * The first round matches up the up-to-date entries,
308          * and then during the second round we try to match
309          * cache-dirty entries as well.
310          */
311         for (contents_too = 0; contents_too < 2; contents_too++) {
312                 for (i = 0; i < rename_dst_nr; i++) {
313                         struct diff_filespec *two = rename_dst[i].two;
314                         if (rename_dst[i].pair)
315                                 continue; /* dealt with an earlier round */
316                         for (j = 0; j < rename_src_nr; j++) {
317                                 int k;
318                                 struct diff_filespec *one = rename_src[j].one;
319                                 if (!is_exact_match(one, two, contents_too))
320                                         continue;
321
322                                 /* see if there is a basename match, too */
323                                 for (k = j; k < rename_src_nr; k++) {
324                                         one = rename_src[k].one;
325                                         if (basename_same(one, two) &&
326                                                 is_exact_match(one, two,
327                                                         contents_too)) {
328                                                 j = k;
329                                                 break;
330                                         }
331                                 }
332
333                                 record_rename_pair(i, j, (int)MAX_SCORE);
334                                 rename_count++;
335                                 break; /* we are done with this entry */
336                         }
337                 }
338         }
339
340         /* Have we run out the created file pool?  If so we can avoid
341          * doing the delta matrix altogether.
342          */
343         if (rename_count == rename_dst_nr)
344                 goto cleanup;
345
346         if (minimum_score == MAX_SCORE)
347                 goto cleanup;
348
349         num_create = (rename_dst_nr - rename_count);
350         num_src = rename_src_nr;
351         mx = xmalloc(sizeof(*mx) * num_create * num_src);
352         for (dst_cnt = i = 0; i < rename_dst_nr; i++) {
353                 int base = dst_cnt * num_src;
354                 struct diff_filespec *two = rename_dst[i].two;
355                 if (rename_dst[i].pair)
356                         continue; /* dealt with exact match already. */
357                 for (j = 0; j < rename_src_nr; j++) {
358                         struct diff_filespec *one = rename_src[j].one;
359                         struct diff_score *m = &mx[base+j];
360                         m->src = j;
361                         m->dst = i;
362                         m->score = estimate_similarity(one, two,
363                                                        minimum_score);
364                         m->name_score = basename_same(one, two);
365                         diff_free_filespec_data(one);
366                 }
367                 /* We do not need the text anymore */
368                 diff_free_filespec_data(two);
369                 dst_cnt++;
370         }
371         /* cost matrix sorted by most to least similar pair */
372         qsort(mx, num_create * num_src, sizeof(*mx), score_compare);
373         for (i = 0; i < num_create * num_src; i++) {
374                 struct diff_rename_dst *dst = &rename_dst[mx[i].dst];
375                 if (dst->pair)
376                         continue; /* already done, either exact or fuzzy. */
377                 if (mx[i].score < minimum_score)
378                         break; /* there is no more usable pair. */
379                 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
380                 rename_count++;
381         }
382         free(mx);
383
384  cleanup:
385         /* At this point, we have found some renames and copies and they
386          * are recorded in rename_dst.  The original list is still in *q.
387          */
388         outq.queue = NULL;
389         outq.nr = outq.alloc = 0;
390         for (i = 0; i < q->nr; i++) {
391                 struct diff_filepair *p = q->queue[i];
392                 struct diff_filepair *pair_to_free = NULL;
393
394                 if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
395                         /*
396                          * Creation
397                          *
398                          * We would output this create record if it has
399                          * not been turned into a rename/copy already.
400                          */
401                         struct diff_rename_dst *dst =
402                                 locate_rename_dst(p->two, 0);
403                         if (dst && dst->pair) {
404                                 diff_q(&outq, dst->pair);
405                                 pair_to_free = p;
406                         }
407                         else
408                                 /* no matching rename/copy source, so
409                                  * record this as a creation.
410                                  */
411                                 diff_q(&outq, p);
412                 }
413                 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) {
414                         /*
415                          * Deletion
416                          *
417                          * We would output this delete record if:
418                          *
419                          * (1) this is a broken delete and the counterpart
420                          *     broken create remains in the output; or
421                          * (2) this is not a broken delete, and rename_dst
422                          *     does not have a rename/copy to move p->one->path
423                          *     out of existence.
424                          *
425                          * Otherwise, the counterpart broken create
426                          * has been turned into a rename-edit; or
427                          * delete did not have a matching create to
428                          * begin with.
429                          */
430                         if (DIFF_PAIR_BROKEN(p)) {
431                                 /* broken delete */
432                                 struct diff_rename_dst *dst =
433                                         locate_rename_dst(p->one, 0);
434                                 if (dst && dst->pair)
435                                         /* counterpart is now rename/copy */
436                                         pair_to_free = p;
437                         }
438                         else {
439                                 for (j = 0; j < rename_dst_nr; j++) {
440                                         if (!rename_dst[j].pair)
441                                                 continue;
442                                         if (strcmp(rename_dst[j].pair->
443                                                    one->path,
444                                                    p->one->path))
445                                                 continue;
446                                         break;
447                                 }
448                                 if (j < rename_dst_nr)
449                                         /* this path remains */
450                                         pair_to_free = p;
451                         }
452
453                         if (pair_to_free)
454                                 ;
455                         else
456                                 diff_q(&outq, p);
457                 }
458                 else if (!diff_unmodified_pair(p))
459                         /* all the usual ones need to be kept */
460                         diff_q(&outq, p);
461                 else
462                         /* no need to keep unmodified pairs */
463                         pair_to_free = p;
464
465                 if (pair_to_free)
466                         diff_free_filepair(pair_to_free);
467         }
468         diff_debug_queue("done copying original", &outq);
469
470         free(q->queue);
471         *q = outq;
472         diff_debug_queue("done collapsing", q);
473
474         /* We need to see which rename source really stays here;
475          * earlier we only checked if the path is left in the result,
476          * but even if a path remains in the result, if that is coming
477          * from copying something else on top of it, then the original
478          * source is lost and does not stay.
479          */
480         for (i = 0; i < q->nr; i++) {
481                 struct diff_filepair *p = q->queue[i];
482                 if (DIFF_PAIR_RENAME(p) && p->source_stays) {
483                         /* If one appears as the target of a rename-copy,
484                          * then mark p->source_stays = 0; otherwise
485                          * leave it as is.
486                          */
487                         p->source_stays = compute_stays(q, p->one);
488                 }
489         }
490
491         for (i = 0; i < rename_dst_nr; i++) {
492                 diff_free_filespec_data(rename_dst[i].two);
493                 free(rename_dst[i].two);
494         }
495
496         free(rename_dst);
497         rename_dst = NULL;
498         rename_dst_nr = rename_dst_alloc = 0;
499         free(rename_src);
500         rename_src = NULL;
501         rename_src_nr = rename_src_alloc = 0;
502         return;
503 }