Merge branch 'jc/show-merge'
[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
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 !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;
118 }
119
120 struct diff_score {
121 int src; /* index in rename_src */
122 int dst; /* index in rename_dst */
123 int score;
124 };
125
126 static int estimate_similarity(struct diff_filespec *src,
127 struct diff_filespec *dst,
128 int minimum_score)
129 {
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;
146
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;
153
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;
157
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;
168
169 if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0))
170 return 0; /* error but caught downstream */
171
172
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;
180
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;
189 }
190
191 static void record_rename_pair(int dst_index, int src_index, int score)
192 {
193 struct diff_filespec *one, *two, *src, *dst;
194 struct diff_filepair *dp;
195
196 if (rename_dst[dst_index].pair)
197 die("internal error: dst already matched.");
198
199 src = rename_src[src_index].one;
200 one = alloc_filespec(src->path);
201 fill_filespec(one, src->sha1, src->mode);
202
203 dst = rename_dst[dst_index].two;
204 two = alloc_filespec(dst->path);
205 fill_filespec(two, dst->sha1, dst->mode);
206
207 dp = diff_queue(NULL, one, two);
208 if (!strcmp(src->path, dst->path))
209 dp->score = rename_src[src_index].score;
210 else
211 dp->score = score;
212 dp->source_stays = rename_src[src_index].src_path_left;
213 rename_dst[dst_index].pair = dp;
214 }
215
216 /*
217 * We sort the rename similarity matrix with the score, in descending
218 * order (the most similar first).
219 */
220 static int score_compare(const void *a_, const void *b_)
221 {
222 const struct diff_score *a = a_, *b = b_;
223 return b->score - a->score;
224 }
225
226 static int compute_stays(struct diff_queue_struct *q,
227 struct diff_filespec *one)
228 {
229 int i;
230 for (i = 0; i < q->nr; i++) {
231 struct diff_filepair *p = q->queue[i];
232 if (strcmp(one->path, p->two->path))
233 continue;
234 if (DIFF_PAIR_RENAME(p)) {
235 return 0; /* something else is renamed into this */
236 }
237 }
238 return 1;
239 }
240
241 void diffcore_rename(struct diff_options *options)
242 {
243 int detect_rename = options->detect_rename;
244 int minimum_score = options->rename_score;
245 int rename_limit = options->rename_limit;
246 struct diff_queue_struct *q = &diff_queued_diff;
247 struct diff_queue_struct outq;
248 struct diff_score *mx;
249 int i, j, rename_count, contents_too;
250 int num_create, num_src, dst_cnt;
251
252 if (!minimum_score)
253 minimum_score = DEFAULT_RENAME_SCORE;
254 rename_count = 0;
255
256 for (i = 0; i < q->nr; i++) {
257 struct diff_filepair *p = q->queue[i];
258 if (!DIFF_FILE_VALID(p->one))
259 if (!DIFF_FILE_VALID(p->two))
260 continue; /* unmerged */
261 else
262 locate_rename_dst(p->two, 1);
263 else if (!DIFF_FILE_VALID(p->two)) {
264 /* If the source is a broken "delete", and
265 * they did not really want to get broken,
266 * that means the source actually stays.
267 */
268 int stays = (p->broken_pair && !p->score);
269 register_rename_src(p->one, stays, p->score);
270 }
271 else if (detect_rename == DIFF_DETECT_COPY)
272 register_rename_src(p->one, 1, p->score);
273 }
274 if (rename_dst_nr == 0 || rename_src_nr == 0 ||
275 (0 < rename_limit && rename_limit < rename_dst_nr))
276 goto cleanup; /* nothing to do */
277
278 /* We really want to cull the candidates list early
279 * with cheap tests in order to avoid doing deltas.
280 * The first round matches up the up-to-date entries,
281 * and then during the second round we try to match
282 * cache-dirty entries as well.
283 */
284 for (contents_too = 0; contents_too < 2; contents_too++) {
285 for (i = 0; i < rename_dst_nr; i++) {
286 struct diff_filespec *two = rename_dst[i].two;
287 if (rename_dst[i].pair)
288 continue; /* dealt with an earlier round */
289 for (j = 0; j < rename_src_nr; j++) {
290 struct diff_filespec *one = rename_src[j].one;
291 if (!is_exact_match(one, two, contents_too))
292 continue;
293 record_rename_pair(i, j, MAX_SCORE);
294 rename_count++;
295 break; /* we are done with this entry */
296 }
297 }
298 }
299
300 /* Have we run out the created file pool? If so we can avoid
301 * doing the delta matrix altogether.
302 */
303 if (rename_count == rename_dst_nr)
304 goto cleanup;
305
306 if (minimum_score == MAX_SCORE)
307 goto cleanup;
308
309 num_create = (rename_dst_nr - rename_count);
310 num_src = rename_src_nr;
311 mx = xmalloc(sizeof(*mx) * num_create * num_src);
312 for (dst_cnt = i = 0; i < rename_dst_nr; i++) {
313 int base = dst_cnt * num_src;
314 struct diff_filespec *two = rename_dst[i].two;
315 if (rename_dst[i].pair)
316 continue; /* dealt with exact match already. */
317 for (j = 0; j < rename_src_nr; j++) {
318 struct diff_filespec *one = rename_src[j].one;
319 struct diff_score *m = &mx[base+j];
320 m->src = j;
321 m->dst = i;
322 m->score = estimate_similarity(one, two,
323 minimum_score);
324 }
325 /* We do not need the text anymore */
326 diff_free_filespec_data(two);
327 dst_cnt++;
328 }
329 /* cost matrix sorted by most to least similar pair */
330 qsort(mx, num_create * num_src, sizeof(*mx), score_compare);
331 for (i = 0; i < num_create * num_src; i++) {
332 struct diff_rename_dst *dst = &rename_dst[mx[i].dst];
333 if (dst->pair)
334 continue; /* already done, either exact or fuzzy. */
335 if (mx[i].score < minimum_score)
336 break; /* there is no more usable pair. */
337 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
338 rename_count++;
339 }
340 free(mx);
341
342 cleanup:
343 /* At this point, we have found some renames and copies and they
344 * are recorded in rename_dst. The original list is still in *q.
345 */
346 outq.queue = NULL;
347 outq.nr = outq.alloc = 0;
348 for (i = 0; i < q->nr; i++) {
349 struct diff_filepair *p = q->queue[i];
350 struct diff_filepair *pair_to_free = NULL;
351
352 if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
353 /*
354 * Creation
355 *
356 * We would output this create record if it has
357 * not been turned into a rename/copy already.
358 */
359 struct diff_rename_dst *dst =
360 locate_rename_dst(p->two, 0);
361 if (dst && dst->pair) {
362 diff_q(&outq, dst->pair);
363 pair_to_free = p;
364 }
365 else
366 /* no matching rename/copy source, so
367 * record this as a creation.
368 */
369 diff_q(&outq, p);
370 }
371 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) {
372 /*
373 * Deletion
374 *
375 * We would output this delete record if:
376 *
377 * (1) this is a broken delete and the counterpart
378 * broken create remains in the output; or
379 * (2) this is not a broken delete, and rename_dst
380 * does not have a rename/copy to move p->one->path
381 * out of existence.
382 *
383 * Otherwise, the counterpart broken create
384 * has been turned into a rename-edit; or
385 * delete did not have a matching create to
386 * begin with.
387 */
388 if (DIFF_PAIR_BROKEN(p)) {
389 /* broken delete */
390 struct diff_rename_dst *dst =
391 locate_rename_dst(p->one, 0);
392 if (dst && dst->pair)
393 /* counterpart is now rename/copy */
394 pair_to_free = p;
395 }
396 else {
397 for (j = 0; j < rename_dst_nr; j++) {
398 if (!rename_dst[j].pair)
399 continue;
400 if (strcmp(rename_dst[j].pair->
401 one->path,
402 p->one->path))
403 continue;
404 break;
405 }
406 if (j < rename_dst_nr)
407 /* this path remains */
408 pair_to_free = p;
409 }
410
411 if (pair_to_free)
412 ;
413 else
414 diff_q(&outq, p);
415 }
416 else if (!diff_unmodified_pair(p))
417 /* all the usual ones need to be kept */
418 diff_q(&outq, p);
419 else
420 /* no need to keep unmodified pairs */
421 pair_to_free = p;
422
423 if (pair_to_free)
424 diff_free_filepair(pair_to_free);
425 }
426 diff_debug_queue("done copying original", &outq);
427
428 free(q->queue);
429 *q = outq;
430 diff_debug_queue("done collapsing", q);
431
432 /* We need to see which rename source really stays here;
433 * earlier we only checked if the path is left in the result,
434 * but even if a path remains in the result, if that is coming
435 * from copying something else on top of it, then the original
436 * source is lost and does not stay.
437 */
438 for (i = 0; i < q->nr; i++) {
439 struct diff_filepair *p = q->queue[i];
440 if (DIFF_PAIR_RENAME(p) && p->source_stays) {
441 /* If one appears as the target of a rename-copy,
442 * then mark p->source_stays = 0; otherwise
443 * leave it as is.
444 */
445 p->source_stays = compute_stays(q, p->one);
446 }
447 }
448
449 for (i = 0; i < rename_dst_nr; i++) {
450 diff_free_filespec_data(rename_dst[i].two);
451 free(rename_dst[i].two);
452 }
453
454 free(rename_dst);
455 rename_dst = NULL;
456 rename_dst_nr = rename_dst_alloc = 0;
457 free(rename_src);
458 rename_src = NULL;
459 rename_src_nr = rename_src_alloc = 0;
460 return;
461 }