ffd126af0d2cd3539e0f186e83c7671132758e6b
[git/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 #include "delta.h"
8 #include "count-delta.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->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_filespec *one;
59 unsigned src_path_left : 1;
60 } *rename_src;
61 static int rename_src_nr, rename_src_alloc;
62
63 static struct diff_rename_src *register_rename_src(struct diff_filespec *one,
64 int src_path_left)
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].src_path_left = src_path_left;
95 return &(rename_src[first]);
96 }
97
98 static int is_exact_match(struct diff_filespec *src, struct diff_filespec *dst)
99 {
100 if (src->sha1_valid && dst->sha1_valid &&
101 !memcmp(src->sha1, dst->sha1, 20))
102 return 1;
103 if (diff_populate_filespec(src, 1) || diff_populate_filespec(dst, 1))
104 return 0;
105 if (src->size != dst->size)
106 return 0;
107 if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0))
108 return 0;
109 if (src->size == dst->size &&
110 !memcmp(src->data, dst->data, src->size))
111 return 1;
112 return 0;
113 }
114
115 struct diff_score {
116 int src; /* index in rename_src */
117 int dst; /* index in rename_dst */
118 int 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 void *delta;
139 unsigned long 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 delta_size = ((src->size < dst->size) ?
151 (dst->size - src->size) : (src->size - dst->size));
152 base_size = ((src->size < dst->size) ? src->size : dst->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 delta_limit = base_size * (MAX_SCORE-minimum_score) / MAX_SCORE;
169 delta = diff_delta(src->data, src->size,
170 dst->data, dst->size,
171 &delta_size, delta_limit);
172 if (!delta)
173 /* If delta_limit is exceeded, we have too much differences */
174 return 0;
175
176 /* A delta that has a lot of literal additions would have
177 * big delta_size no matter what else it does.
178 */
179 if (base_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE) {
180 free(delta);
181 return 0;
182 }
183
184 /* Estimate the edit size by interpreting delta. */
185 if (count_delta(delta, delta_size, &src_copied, &literal_added)) {
186 free(delta);
187 return 0;
188 }
189 free(delta);
190
191 /* Extent of damage */
192 if (src->size + literal_added < src_copied)
193 delta_size = 0;
194 else
195 delta_size = (src->size - src_copied) + literal_added;
196
197 /*
198 * Now we will give some score to it. 100% edit gets 0 points
199 * and 0% edit gets MAX_SCORE points.
200 */
201 score = MAX_SCORE - (MAX_SCORE * delta_size / base_size);
202 if (score < 0) return 0;
203 if (MAX_SCORE < score) return MAX_SCORE;
204 return score;
205 }
206
207 static void record_rename_pair(int dst_index, int src_index, int score)
208 {
209 struct diff_filespec *one, *two, *src, *dst;
210 struct diff_filepair *dp;
211
212 if (rename_dst[dst_index].pair)
213 die("internal error: dst already matched.");
214
215 src = rename_src[src_index].one;
216 one = alloc_filespec(src->path);
217 fill_filespec(one, src->sha1, src->mode);
218
219 dst = rename_dst[dst_index].two;
220 two = alloc_filespec(dst->path);
221 fill_filespec(two, dst->sha1, dst->mode);
222
223 dp = diff_queue(NULL, one, two);
224 dp->score = score;
225 dp->source_stays = rename_src[src_index].src_path_left;
226 rename_dst[dst_index].pair = dp;
227 }
228
229 /*
230 * We sort the rename similarity matrix with the score, in descending
231 * order (the most similar first).
232 */
233 static int score_compare(const void *a_, const void *b_)
234 {
235 const struct diff_score *a = a_, *b = b_;
236 return b->score - a->score;
237 }
238
239 static int compute_stays(struct diff_queue_struct *q,
240 struct diff_filespec *one)
241 {
242 int i;
243 for (i = 0; i < q->nr; i++) {
244 struct diff_filepair *p = q->queue[i];
245 if (strcmp(one->path, p->two->path))
246 continue;
247 if (DIFF_PAIR_RENAME(p)) {
248 return 0; /* something else is renamed into this */
249 }
250 }
251 return 1;
252 }
253
254 void diffcore_rename(struct diff_options *options)
255 {
256 int detect_rename = options->detect_rename;
257 int minimum_score = options->rename_score;
258 int rename_limit = options->rename_limit;
259 struct diff_queue_struct *q = &diff_queued_diff;
260 struct diff_queue_struct outq;
261 struct diff_score *mx;
262 int i, j, rename_count;
263 int num_create, num_src, dst_cnt;
264
265 if (!minimum_score)
266 minimum_score = DEFAULT_RENAME_SCORE;
267 rename_count = 0;
268
269 for (i = 0; i < q->nr; i++) {
270 struct diff_filepair *p = q->queue[i];
271 if (!DIFF_FILE_VALID(p->one))
272 if (!DIFF_FILE_VALID(p->two))
273 continue; /* unmerged */
274 else
275 locate_rename_dst(p->two, 1);
276 else if (!DIFF_FILE_VALID(p->two)) {
277 /* If the source is a broken "delete", and
278 * they did not really want to get broken,
279 * that means the source actually stays.
280 */
281 int stays = (p->broken_pair && !p->score);
282 register_rename_src(p->one, stays);
283 }
284 else if (detect_rename == DIFF_DETECT_COPY)
285 register_rename_src(p->one, 1);
286 }
287 if (rename_dst_nr == 0 || rename_src_nr == 0 ||
288 (0 < rename_limit && rename_limit < rename_dst_nr))
289 goto cleanup; /* nothing to do */
290
291 /* We really want to cull the candidates list early
292 * with cheap tests in order to avoid doing deltas.
293 */
294 for (i = 0; i < rename_dst_nr; i++) {
295 struct diff_filespec *two = rename_dst[i].two;
296 for (j = 0; j < rename_src_nr; j++) {
297 struct diff_filespec *one = rename_src[j].one;
298 if (!is_exact_match(one, two))
299 continue;
300 record_rename_pair(i, j, MAX_SCORE);
301 rename_count++;
302 break; /* we are done with this entry */
303 }
304 }
305
306 /* Have we run out the created file pool? If so we can avoid
307 * doing the delta matrix altogether.
308 */
309 if (rename_count == rename_dst_nr)
310 goto cleanup;
311
312 if (minimum_score == MAX_SCORE)
313 goto cleanup;
314
315 num_create = (rename_dst_nr - rename_count);
316 num_src = rename_src_nr;
317 mx = xmalloc(sizeof(*mx) * num_create * num_src);
318 for (dst_cnt = i = 0; i < rename_dst_nr; i++) {
319 int base = dst_cnt * num_src;
320 struct diff_filespec *two = rename_dst[i].two;
321 if (rename_dst[i].pair)
322 continue; /* dealt with exact match already. */
323 for (j = 0; j < rename_src_nr; j++) {
324 struct diff_filespec *one = rename_src[j].one;
325 struct diff_score *m = &mx[base+j];
326 m->src = j;
327 m->dst = i;
328 m->score = estimate_similarity(one, two,
329 minimum_score);
330 }
331 dst_cnt++;
332 }
333 /* cost matrix sorted by most to least similar pair */
334 qsort(mx, num_create * num_src, sizeof(*mx), score_compare);
335 for (i = 0; i < num_create * num_src; i++) {
336 struct diff_rename_dst *dst = &rename_dst[mx[i].dst];
337 if (dst->pair)
338 continue; /* already done, either exact or fuzzy. */
339 if (mx[i].score < minimum_score)
340 break; /* there is no more usable pair. */
341 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
342 rename_count++;
343 }
344 free(mx);
345
346 cleanup:
347 /* At this point, we have found some renames and copies and they
348 * are recorded in rename_dst. The original list is still in *q.
349 */
350 outq.queue = NULL;
351 outq.nr = outq.alloc = 0;
352 for (i = 0; i < q->nr; i++) {
353 struct diff_filepair *p = q->queue[i];
354 struct diff_filepair *pair_to_free = NULL;
355
356 if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
357 /*
358 * Creation
359 *
360 * We would output this create record if it has
361 * not been turned into a rename/copy already.
362 */
363 struct diff_rename_dst *dst =
364 locate_rename_dst(p->two, 0);
365 if (dst && dst->pair) {
366 diff_q(&outq, dst->pair);
367 pair_to_free = p;
368 }
369 else
370 /* no matching rename/copy source, so
371 * record this as a creation.
372 */
373 diff_q(&outq, p);
374 }
375 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) {
376 /*
377 * Deletion
378 *
379 * We would output this delete record if:
380 *
381 * (1) this is a broken delete and the counterpart
382 * broken create remains in the output; or
383 * (2) this is not a broken delete, and rename_dst
384 * does not have a rename/copy to move p->one->path
385 * out of existence.
386 *
387 * Otherwise, the counterpart broken create
388 * has been turned into a rename-edit; or
389 * delete did not have a matching create to
390 * begin with.
391 */
392 if (DIFF_PAIR_BROKEN(p)) {
393 /* broken delete */
394 struct diff_rename_dst *dst =
395 locate_rename_dst(p->one, 0);
396 if (dst && dst->pair)
397 /* counterpart is now rename/copy */
398 pair_to_free = p;
399 }
400 else {
401 for (j = 0; j < rename_dst_nr; j++) {
402 if (!rename_dst[j].pair)
403 continue;
404 if (strcmp(rename_dst[j].pair->
405 one->path,
406 p->one->path))
407 continue;
408 break;
409 }
410 if (j < rename_dst_nr)
411 /* this path remains */
412 pair_to_free = p;
413 }
414
415 if (pair_to_free)
416 ;
417 else
418 diff_q(&outq, p);
419 }
420 else if (!diff_unmodified_pair(p))
421 /* all the usual ones need to be kept */
422 diff_q(&outq, p);
423 else
424 /* no need to keep unmodified pairs */
425 pair_to_free = p;
426
427 if (pair_to_free)
428 diff_free_filepair(pair_to_free);
429 }
430 diff_debug_queue("done copying original", &outq);
431
432 free(q->queue);
433 *q = outq;
434 diff_debug_queue("done collapsing", q);
435
436 /* We need to see which rename source really stays here;
437 * earlier we only checked if the path is left in the result,
438 * but even if a path remains in the result, if that is coming
439 * from copying something else on top of it, then the original
440 * source is lost and does not stay.
441 */
442 for (i = 0; i < q->nr; i++) {
443 struct diff_filepair *p = q->queue[i];
444 if (DIFF_PAIR_RENAME(p) && p->source_stays) {
445 /* If one appears as the target of a rename-copy,
446 * then mark p->source_stays = 0; otherwise
447 * leave it as is.
448 */
449 p->source_stays = compute_stays(q, p->one);
450 }
451 }
452
453 for (i = 0; i < rename_dst_nr; i++) {
454 diff_free_filespec_data(rename_dst[i].two);
455 free(rename_dst[i].two);
456 }
457
458 free(rename_dst);
459 rename_dst = NULL;
460 rename_dst_nr = rename_dst_alloc = 0;
461 free(rename_src);
462 rename_src = NULL;
463 rename_src_nr = rename_src_alloc = 0;
464 return;
465 }