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