2 * Copyright (C) 2005 Junio C Hamano
8 #include "count-delta.h"
10 /* Table of rename/copy destinations */
12 static struct diff_rename_dst {
13 struct diff_filespec *two;
14 struct diff_filepair *pair;
16 static int rename_dst_nr, rename_dst_alloc;
18 static struct diff_rename_dst *locate_rename_dst(struct diff_filespec *two,
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);
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));
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]);
56 /* Table of rename/copy src files */
57 static struct diff_rename_src {
58 struct diff_filespec *one;
59 unsigned src_path_left : 1;
61 static int rename_src_nr, rename_src_alloc;
63 static struct diff_rename_src *register_rename_src(struct diff_filespec *one,
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);
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));
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]);
98 static int is_exact_match(struct diff_filespec *src, struct diff_filespec *dst)
100 if (src->sha1_valid && dst->sha1_valid &&
101 !memcmp(src->sha1, dst->sha1, 20))
103 if (diff_populate_filespec(src, 1) || diff_populate_filespec(dst, 1))
105 if (src->size != dst->size)
107 if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0))
109 if (src->size == dst->size &&
110 !memcmp(src->data, dst->data, src->size))
116 int src; /* index in rename_src */
117 int dst; /* index in rename_dst */
121 static int estimate_similarity(struct diff_filespec *src,
122 struct diff_filespec *dst,
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.
131 * Compare them and return how similar they are, representing
132 * the score as an integer between 0 and MAX_SCORE.
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.
139 unsigned long delta_size, base_size, src_copied, literal_added;
140 unsigned long delta_limit;
143 /* We deal only with regular files. Symlink renames are handled
144 * only when they are exact matches --- in other words, no edits
147 if (!S_ISREG(src->mode) || !S_ISREG(dst->mode))
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);
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).
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.
162 if (base_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE)
165 if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0))
166 return 0; /* error but caught downstream */
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);
173 /* If delta_limit is exceeded, we have too much differences */
176 /* A delta that has a lot of literal additions would have
177 * big delta_size no matter what else it does.
179 if (base_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE)
182 /* Estimate the edit size by interpreting delta. */
183 if (count_delta(delta, delta_size, &src_copied, &literal_added)) {
189 /* Extent of damage */
190 if (src->size + literal_added < src_copied)
193 delta_size = (src->size - src_copied) + literal_added;
196 * Now we will give some score to it. 100% edit gets 0 points
197 * and 0% edit gets MAX_SCORE points.
199 score = MAX_SCORE - (MAX_SCORE * delta_size / base_size);
200 if (score < 0) return 0;
201 if (MAX_SCORE < score) return MAX_SCORE;
205 static void record_rename_pair(int dst_index, int src_index, int score)
207 struct diff_filespec *one, *two, *src, *dst;
208 struct diff_filepair *dp;
210 if (rename_dst[dst_index].pair)
211 die("internal error: dst already matched.");
213 src = rename_src[src_index].one;
214 one = alloc_filespec(src->path);
215 fill_filespec(one, src->sha1, src->mode);
217 dst = rename_dst[dst_index].two;
218 two = alloc_filespec(dst->path);
219 fill_filespec(two, dst->sha1, dst->mode);
221 dp = diff_queue(NULL, one, two);
223 dp->source_stays = rename_src[src_index].src_path_left;
224 rename_dst[dst_index].pair = dp;
228 * We sort the rename similarity matrix with the score, in descending
229 * order (the most similar first).
231 static int score_compare(const void *a_, const void *b_)
233 const struct diff_score *a = a_, *b = b_;
234 return b->score - a->score;
237 static int compute_stays(struct diff_queue_struct *q,
238 struct diff_filespec *one)
241 for (i = 0; i < q->nr; i++) {
242 struct diff_filepair *p = q->queue[i];
243 if (strcmp(one->path, p->two->path))
245 if (DIFF_PAIR_RENAME(p)) {
246 return 0; /* something else is renamed into this */
252 void diffcore_rename(struct diff_options *options)
254 int detect_rename = options->detect_rename;
255 int minimum_score = options->rename_score;
256 int rename_limit = options->rename_limit;
257 struct diff_queue_struct *q = &diff_queued_diff;
258 struct diff_queue_struct outq;
259 struct diff_score *mx;
260 int i, j, rename_count;
261 int num_create, num_src, dst_cnt;
264 minimum_score = DEFAULT_RENAME_SCORE;
267 for (i = 0; i < q->nr; i++) {
268 struct diff_filepair *p = q->queue[i];
269 if (!DIFF_FILE_VALID(p->one))
270 if (!DIFF_FILE_VALID(p->two))
271 continue; /* unmerged */
273 locate_rename_dst(p->two, 1);
274 else if (!DIFF_FILE_VALID(p->two)) {
275 /* If the source is a broken "delete", and
276 * they did not really want to get broken,
277 * that means the source actually stays.
279 int stays = (p->broken_pair && !p->score);
280 register_rename_src(p->one, stays);
282 else if (detect_rename == DIFF_DETECT_COPY)
283 register_rename_src(p->one, 1);
285 if (rename_dst_nr == 0 ||
286 (0 <= rename_limit && rename_limit < rename_dst_nr))
287 goto cleanup; /* nothing to do */
289 /* We really want to cull the candidates list early
290 * with cheap tests in order to avoid doing deltas.
292 for (i = 0; i < rename_dst_nr; i++) {
293 struct diff_filespec *two = rename_dst[i].two;
294 for (j = 0; j < rename_src_nr; j++) {
295 struct diff_filespec *one = rename_src[j].one;
296 if (!is_exact_match(one, two))
298 record_rename_pair(i, j, MAX_SCORE);
300 break; /* we are done with this entry */
304 /* Have we run out the created file pool? If so we can avoid
305 * doing the delta matrix altogether.
307 if (rename_count == rename_dst_nr)
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];
323 m->score = estimate_similarity(one, two,
328 /* cost matrix sorted by most to least similar pair */
329 qsort(mx, num_create * num_src, sizeof(*mx), score_compare);
330 for (i = 0; i < num_create * num_src; i++) {
331 struct diff_rename_dst *dst = &rename_dst[mx[i].dst];
333 continue; /* already done, either exact or fuzzy. */
334 if (mx[i].score < minimum_score)
335 break; /* there is no more usable pair. */
336 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
342 /* At this point, we have found some renames and copies and they
343 * are recorded in rename_dst. The original list is still in *q.
346 outq.nr = outq.alloc = 0;
347 for (i = 0; i < q->nr; i++) {
348 struct diff_filepair *p = q->queue[i];
349 struct diff_filepair *pair_to_free = NULL;
351 if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
355 * We would output this create record if it has
356 * not been turned into a rename/copy already.
358 struct diff_rename_dst *dst =
359 locate_rename_dst(p->two, 0);
360 if (dst && dst->pair) {
361 diff_q(&outq, dst->pair);
365 /* no matching rename/copy source, so
366 * record this as a creation.
370 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) {
374 * We would output this delete record if:
376 * (1) this is a broken delete and the counterpart
377 * broken create remains in the output; or
378 * (2) this is not a broken delete, and rename_dst
379 * does not have a rename/copy to move p->one->path
382 * Otherwise, the counterpart broken create
383 * has been turned into a rename-edit; or
384 * delete did not have a matching create to
387 if (DIFF_PAIR_BROKEN(p)) {
389 struct diff_rename_dst *dst =
390 locate_rename_dst(p->one, 0);
391 if (dst && dst->pair)
392 /* counterpart is now rename/copy */
396 for (j = 0; j < rename_dst_nr; j++) {
397 if (!rename_dst[j].pair)
399 if (strcmp(rename_dst[j].pair->
405 if (j < rename_dst_nr)
406 /* this path remains */
415 else if (!diff_unmodified_pair(p))
416 /* all the usual ones need to be kept */
419 /* no need to keep unmodified pairs */
423 diff_free_filepair(pair_to_free);
425 diff_debug_queue("done copying original", &outq);
429 diff_debug_queue("done collapsing", q);
431 /* We need to see which rename source really stays here;
432 * earlier we only checked if the path is left in the result,
433 * but even if a path remains in the result, if that is coming
434 * from copying something else on top of it, then the original
435 * source is lost and does not stay.
437 for (i = 0; i < q->nr; i++) {
438 struct diff_filepair *p = q->queue[i];
439 if (DIFF_PAIR_RENAME(p) && p->source_stays) {
440 /* If one appears as the target of a rename-copy,
441 * then mark p->source_stays = 0; otherwise
444 p->source_stays = compute_stays(q, p->one);
448 for (i = 0; i < rename_dst_nr; i++) {
449 diff_free_filespec_data(rename_dst[i].two);
450 free(rename_dst[i].two);
455 rename_dst_nr = rename_dst_alloc = 0;
458 rename_src_nr = rename_src_alloc = 0;