]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * | |
3 | * Copyright (C) 2005 Junio C Hamano | |
4 | */ | |
5 | #include "cache.h" | |
6 | #include "diff.h" | |
7 | #include "diffcore.h" | |
8 | #include "object-store.h" | |
9 | #include "hashmap.h" | |
10 | #include "progress.h" | |
11 | #include "promisor-remote.h" | |
12 | #include "strmap.h" | |
13 | ||
14 | /* Table of rename/copy destinations */ | |
15 | ||
16 | static struct diff_rename_dst { | |
17 | struct diff_filepair *p; | |
18 | struct diff_filespec *filespec_to_free; | |
19 | int is_rename; /* false -> just a create; true -> rename or copy */ | |
20 | } *rename_dst; | |
21 | static int rename_dst_nr, rename_dst_alloc; | |
22 | /* Mapping from break source pathname to break destination index */ | |
23 | static struct strintmap *break_idx = NULL; | |
24 | ||
25 | static struct diff_rename_dst *locate_rename_dst(struct diff_filepair *p) | |
26 | { | |
27 | /* Lookup by p->ONE->path */ | |
28 | int idx = break_idx ? strintmap_get(break_idx, p->one->path) : -1; | |
29 | return (idx == -1) ? NULL : &rename_dst[idx]; | |
30 | } | |
31 | ||
32 | /* | |
33 | * Returns 0 on success, -1 if we found a duplicate. | |
34 | */ | |
35 | static int add_rename_dst(struct diff_filepair *p) | |
36 | { | |
37 | ALLOC_GROW(rename_dst, rename_dst_nr + 1, rename_dst_alloc); | |
38 | rename_dst[rename_dst_nr].p = p; | |
39 | rename_dst[rename_dst_nr].filespec_to_free = NULL; | |
40 | rename_dst[rename_dst_nr].is_rename = 0; | |
41 | rename_dst_nr++; | |
42 | return 0; | |
43 | } | |
44 | ||
45 | /* Table of rename/copy src files */ | |
46 | static struct diff_rename_src { | |
47 | struct diff_filepair *p; | |
48 | unsigned short score; /* to remember the break score */ | |
49 | } *rename_src; | |
50 | static int rename_src_nr, rename_src_alloc; | |
51 | ||
52 | static void register_rename_src(struct diff_filepair *p) | |
53 | { | |
54 | if (p->broken_pair) { | |
55 | if (!break_idx) { | |
56 | break_idx = xmalloc(sizeof(*break_idx)); | |
57 | strintmap_init(break_idx, -1); | |
58 | } | |
59 | strintmap_set(break_idx, p->one->path, rename_dst_nr); | |
60 | } | |
61 | ||
62 | ALLOC_GROW(rename_src, rename_src_nr + 1, rename_src_alloc); | |
63 | rename_src[rename_src_nr].p = p; | |
64 | rename_src[rename_src_nr].score = p->score; | |
65 | rename_src_nr++; | |
66 | } | |
67 | ||
68 | static int basename_same(struct diff_filespec *src, struct diff_filespec *dst) | |
69 | { | |
70 | int src_len = strlen(src->path), dst_len = strlen(dst->path); | |
71 | while (src_len && dst_len) { | |
72 | char c1 = src->path[--src_len]; | |
73 | char c2 = dst->path[--dst_len]; | |
74 | if (c1 != c2) | |
75 | return 0; | |
76 | if (c1 == '/') | |
77 | return 1; | |
78 | } | |
79 | return (!src_len || src->path[src_len - 1] == '/') && | |
80 | (!dst_len || dst->path[dst_len - 1] == '/'); | |
81 | } | |
82 | ||
83 | struct diff_score { | |
84 | int src; /* index in rename_src */ | |
85 | int dst; /* index in rename_dst */ | |
86 | unsigned short score; | |
87 | short name_score; | |
88 | }; | |
89 | ||
90 | struct prefetch_options { | |
91 | struct repository *repo; | |
92 | int skip_unmodified; | |
93 | }; | |
94 | static void prefetch(void *prefetch_options) | |
95 | { | |
96 | struct prefetch_options *options = prefetch_options; | |
97 | int i; | |
98 | struct oid_array to_fetch = OID_ARRAY_INIT; | |
99 | ||
100 | for (i = 0; i < rename_dst_nr; i++) { | |
101 | if (rename_dst[i].p->renamed_pair) | |
102 | /* | |
103 | * The loop in diffcore_rename() will not need these | |
104 | * blobs, so skip prefetching. | |
105 | */ | |
106 | continue; /* already found exact match */ | |
107 | diff_add_if_missing(options->repo, &to_fetch, | |
108 | rename_dst[i].p->two); | |
109 | } | |
110 | for (i = 0; i < rename_src_nr; i++) { | |
111 | if (options->skip_unmodified && | |
112 | diff_unmodified_pair(rename_src[i].p)) | |
113 | /* | |
114 | * The loop in diffcore_rename() will not need these | |
115 | * blobs, so skip prefetching. | |
116 | */ | |
117 | continue; | |
118 | diff_add_if_missing(options->repo, &to_fetch, | |
119 | rename_src[i].p->one); | |
120 | } | |
121 | promisor_remote_get_direct(options->repo, to_fetch.oid, to_fetch.nr); | |
122 | oid_array_clear(&to_fetch); | |
123 | } | |
124 | ||
125 | static int estimate_similarity(struct repository *r, | |
126 | struct diff_filespec *src, | |
127 | struct diff_filespec *dst, | |
128 | int minimum_score, | |
129 | int skip_unmodified) | |
130 | { | |
131 | /* src points at a file that existed in the original tree (or | |
132 | * optionally a file in the destination tree) and dst points | |
133 | * at a newly created file. They may be quite similar, in which | |
134 | * case we want to say src is renamed to dst or src is copied into | |
135 | * dst, and then some edit has been applied to dst. | |
136 | * | |
137 | * Compare them and return how similar they are, representing | |
138 | * the score as an integer between 0 and MAX_SCORE. | |
139 | * | |
140 | * When there is an exact match, it is considered a better | |
141 | * match than anything else; the destination does not even | |
142 | * call into this function in that case. | |
143 | */ | |
144 | unsigned long max_size, delta_size, base_size, src_copied, literal_added; | |
145 | int score; | |
146 | struct diff_populate_filespec_options dpf_options = { | |
147 | .check_size_only = 1 | |
148 | }; | |
149 | struct prefetch_options prefetch_options = {r, skip_unmodified}; | |
150 | ||
151 | if (r == the_repository && has_promisor_remote()) { | |
152 | dpf_options.missing_object_cb = prefetch; | |
153 | dpf_options.missing_object_data = &prefetch_options; | |
154 | } | |
155 | ||
156 | /* We deal only with regular files. Symlink renames are handled | |
157 | * only when they are exact matches --- in other words, no edits | |
158 | * after renaming. | |
159 | */ | |
160 | if (!S_ISREG(src->mode) || !S_ISREG(dst->mode)) | |
161 | return 0; | |
162 | ||
163 | /* | |
164 | * Need to check that source and destination sizes are | |
165 | * filled in before comparing them. | |
166 | * | |
167 | * If we already have "cnt_data" filled in, we know it's | |
168 | * all good (avoid checking the size for zero, as that | |
169 | * is a possible size - we really should have a flag to | |
170 | * say whether the size is valid or not!) | |
171 | */ | |
172 | if (!src->cnt_data && | |
173 | diff_populate_filespec(r, src, &dpf_options)) | |
174 | return 0; | |
175 | if (!dst->cnt_data && | |
176 | diff_populate_filespec(r, dst, &dpf_options)) | |
177 | return 0; | |
178 | ||
179 | max_size = ((src->size > dst->size) ? src->size : dst->size); | |
180 | base_size = ((src->size < dst->size) ? src->size : dst->size); | |
181 | delta_size = max_size - base_size; | |
182 | ||
183 | /* We would not consider edits that change the file size so | |
184 | * drastically. delta_size must be smaller than | |
185 | * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size). | |
186 | * | |
187 | * Note that base_size == 0 case is handled here already | |
188 | * and the final score computation below would not have a | |
189 | * divide-by-zero issue. | |
190 | */ | |
191 | if (max_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE) | |
192 | return 0; | |
193 | ||
194 | dpf_options.check_size_only = 0; | |
195 | ||
196 | if (!src->cnt_data && diff_populate_filespec(r, src, &dpf_options)) | |
197 | return 0; | |
198 | if (!dst->cnt_data && diff_populate_filespec(r, dst, &dpf_options)) | |
199 | return 0; | |
200 | ||
201 | if (diffcore_count_changes(r, src, dst, | |
202 | &src->cnt_data, &dst->cnt_data, | |
203 | &src_copied, &literal_added)) | |
204 | return 0; | |
205 | ||
206 | /* How similar are they? | |
207 | * what percentage of material in dst are from source? | |
208 | */ | |
209 | if (!dst->size) | |
210 | score = 0; /* should not happen */ | |
211 | else | |
212 | score = (int)(src_copied * MAX_SCORE / max_size); | |
213 | return score; | |
214 | } | |
215 | ||
216 | static void record_rename_pair(int dst_index, int src_index, int score) | |
217 | { | |
218 | struct diff_filepair *src = rename_src[src_index].p; | |
219 | struct diff_filepair *dst = rename_dst[dst_index].p; | |
220 | ||
221 | if (dst->renamed_pair) | |
222 | die("internal error: dst already matched."); | |
223 | ||
224 | src->one->rename_used++; | |
225 | src->one->count++; | |
226 | ||
227 | rename_dst[dst_index].filespec_to_free = dst->one; | |
228 | rename_dst[dst_index].is_rename = 1; | |
229 | ||
230 | dst->one = src->one; | |
231 | dst->renamed_pair = 1; | |
232 | if (!strcmp(dst->one->path, dst->two->path)) | |
233 | dst->score = rename_src[src_index].score; | |
234 | else | |
235 | dst->score = score; | |
236 | } | |
237 | ||
238 | /* | |
239 | * We sort the rename similarity matrix with the score, in descending | |
240 | * order (the most similar first). | |
241 | */ | |
242 | static int score_compare(const void *a_, const void *b_) | |
243 | { | |
244 | const struct diff_score *a = a_, *b = b_; | |
245 | ||
246 | /* sink the unused ones to the bottom */ | |
247 | if (a->dst < 0) | |
248 | return (0 <= b->dst); | |
249 | else if (b->dst < 0) | |
250 | return -1; | |
251 | ||
252 | if (a->score == b->score) | |
253 | return b->name_score - a->name_score; | |
254 | ||
255 | return b->score - a->score; | |
256 | } | |
257 | ||
258 | struct file_similarity { | |
259 | struct hashmap_entry entry; | |
260 | int index; | |
261 | struct diff_filespec *filespec; | |
262 | }; | |
263 | ||
264 | static unsigned int hash_filespec(struct repository *r, | |
265 | struct diff_filespec *filespec) | |
266 | { | |
267 | if (!filespec->oid_valid) { | |
268 | if (diff_populate_filespec(r, filespec, NULL)) | |
269 | return 0; | |
270 | hash_object_file(r->hash_algo, filespec->data, filespec->size, | |
271 | "blob", &filespec->oid); | |
272 | } | |
273 | return oidhash(&filespec->oid); | |
274 | } | |
275 | ||
276 | static int find_identical_files(struct hashmap *srcs, | |
277 | int dst_index, | |
278 | struct diff_options *options) | |
279 | { | |
280 | int renames = 0; | |
281 | struct diff_filespec *target = rename_dst[dst_index].p->two; | |
282 | struct file_similarity *p, *best = NULL; | |
283 | int i = 100, best_score = -1; | |
284 | unsigned int hash = hash_filespec(options->repo, target); | |
285 | ||
286 | /* | |
287 | * Find the best source match for specified destination. | |
288 | */ | |
289 | p = hashmap_get_entry_from_hash(srcs, hash, NULL, | |
290 | struct file_similarity, entry); | |
291 | hashmap_for_each_entry_from(srcs, p, entry) { | |
292 | int score; | |
293 | struct diff_filespec *source = p->filespec; | |
294 | ||
295 | /* False hash collision? */ | |
296 | if (!oideq(&source->oid, &target->oid)) | |
297 | continue; | |
298 | /* Non-regular files? If so, the modes must match! */ | |
299 | if (!S_ISREG(source->mode) || !S_ISREG(target->mode)) { | |
300 | if (source->mode != target->mode) | |
301 | continue; | |
302 | } | |
303 | /* Give higher scores to sources that haven't been used already */ | |
304 | score = !source->rename_used; | |
305 | if (source->rename_used && options->detect_rename != DIFF_DETECT_COPY) | |
306 | continue; | |
307 | score += basename_same(source, target); | |
308 | if (score > best_score) { | |
309 | best = p; | |
310 | best_score = score; | |
311 | if (score == 2) | |
312 | break; | |
313 | } | |
314 | ||
315 | /* Too many identical alternatives? Pick one */ | |
316 | if (!--i) | |
317 | break; | |
318 | } | |
319 | if (best) { | |
320 | record_rename_pair(dst_index, best->index, MAX_SCORE); | |
321 | renames++; | |
322 | } | |
323 | return renames; | |
324 | } | |
325 | ||
326 | static void insert_file_table(struct repository *r, | |
327 | struct hashmap *table, int index, | |
328 | struct diff_filespec *filespec) | |
329 | { | |
330 | struct file_similarity *entry = xmalloc(sizeof(*entry)); | |
331 | ||
332 | entry->index = index; | |
333 | entry->filespec = filespec; | |
334 | ||
335 | hashmap_entry_init(&entry->entry, hash_filespec(r, filespec)); | |
336 | hashmap_add(table, &entry->entry); | |
337 | } | |
338 | ||
339 | /* | |
340 | * Find exact renames first. | |
341 | * | |
342 | * The first round matches up the up-to-date entries, | |
343 | * and then during the second round we try to match | |
344 | * cache-dirty entries as well. | |
345 | */ | |
346 | static int find_exact_renames(struct diff_options *options) | |
347 | { | |
348 | int i, renames = 0; | |
349 | struct hashmap file_table; | |
350 | ||
351 | /* Add all sources to the hash table in reverse order, because | |
352 | * later on they will be retrieved in LIFO order. | |
353 | */ | |
354 | hashmap_init(&file_table, NULL, NULL, rename_src_nr); | |
355 | for (i = rename_src_nr-1; i >= 0; i--) | |
356 | insert_file_table(options->repo, | |
357 | &file_table, i, | |
358 | rename_src[i].p->one); | |
359 | ||
360 | /* Walk the destinations and find best source match */ | |
361 | for (i = 0; i < rename_dst_nr; i++) | |
362 | renames += find_identical_files(&file_table, i, options); | |
363 | ||
364 | /* Free the hash data structure and entries */ | |
365 | hashmap_clear_and_free(&file_table, struct file_similarity, entry); | |
366 | ||
367 | return renames; | |
368 | } | |
369 | ||
370 | #define NUM_CANDIDATE_PER_DST 4 | |
371 | static void record_if_better(struct diff_score m[], struct diff_score *o) | |
372 | { | |
373 | int i, worst; | |
374 | ||
375 | /* find the worst one */ | |
376 | worst = 0; | |
377 | for (i = 1; i < NUM_CANDIDATE_PER_DST; i++) | |
378 | if (score_compare(&m[i], &m[worst]) > 0) | |
379 | worst = i; | |
380 | ||
381 | /* is it better than the worst one? */ | |
382 | if (score_compare(&m[worst], o) > 0) | |
383 | m[worst] = *o; | |
384 | } | |
385 | ||
386 | /* | |
387 | * Returns: | |
388 | * 0 if we are under the limit; | |
389 | * 1 if we need to disable inexact rename detection; | |
390 | * 2 if we would be under the limit if we were given -C instead of -C -C. | |
391 | */ | |
392 | static int too_many_rename_candidates(int num_destinations, int num_sources, | |
393 | struct diff_options *options) | |
394 | { | |
395 | int rename_limit = options->rename_limit; | |
396 | int i, limited_sources; | |
397 | ||
398 | options->needed_rename_limit = 0; | |
399 | ||
400 | /* | |
401 | * This basically does a test for the rename matrix not | |
402 | * growing larger than a "rename_limit" square matrix, ie: | |
403 | * | |
404 | * num_destinations * num_sources > rename_limit * rename_limit | |
405 | * | |
406 | * We use st_mult() to check overflow conditions; in the | |
407 | * exceptional circumstance that size_t isn't large enough to hold | |
408 | * the multiplication, the system won't be able to allocate enough | |
409 | * memory for the matrix anyway. | |
410 | */ | |
411 | if (rename_limit <= 0) | |
412 | rename_limit = 32767; | |
413 | if (st_mult(num_destinations, num_sources) | |
414 | <= st_mult(rename_limit, rename_limit)) | |
415 | return 0; | |
416 | ||
417 | options->needed_rename_limit = | |
418 | num_sources > num_destinations ? num_sources : num_destinations; | |
419 | ||
420 | /* Are we running under -C -C? */ | |
421 | if (!options->flags.find_copies_harder) | |
422 | return 1; | |
423 | ||
424 | /* Would we bust the limit if we were running under -C? */ | |
425 | for (limited_sources = i = 0; i < num_sources; i++) { | |
426 | if (diff_unmodified_pair(rename_src[i].p)) | |
427 | continue; | |
428 | limited_sources++; | |
429 | } | |
430 | if (st_mult(num_destinations, limited_sources) | |
431 | <= st_mult(rename_limit, rename_limit)) | |
432 | return 2; | |
433 | return 1; | |
434 | } | |
435 | ||
436 | static int find_renames(struct diff_score *mx, int dst_cnt, int minimum_score, int copies) | |
437 | { | |
438 | int count = 0, i; | |
439 | ||
440 | for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) { | |
441 | struct diff_rename_dst *dst; | |
442 | ||
443 | if ((mx[i].dst < 0) || | |
444 | (mx[i].score < minimum_score)) | |
445 | break; /* there is no more usable pair. */ | |
446 | dst = &rename_dst[mx[i].dst]; | |
447 | if (dst->is_rename) | |
448 | continue; /* already done, either exact or fuzzy. */ | |
449 | if (!copies && rename_src[mx[i].src].p->one->rename_used) | |
450 | continue; | |
451 | record_rename_pair(mx[i].dst, mx[i].src, mx[i].score); | |
452 | count++; | |
453 | } | |
454 | return count; | |
455 | } | |
456 | ||
457 | void diffcore_rename(struct diff_options *options) | |
458 | { | |
459 | int detect_rename = options->detect_rename; | |
460 | int minimum_score = options->rename_score; | |
461 | struct diff_queue_struct *q = &diff_queued_diff; | |
462 | struct diff_queue_struct outq; | |
463 | struct diff_score *mx; | |
464 | int i, j, rename_count, skip_unmodified = 0; | |
465 | int num_destinations, dst_cnt; | |
466 | struct progress *progress = NULL; | |
467 | ||
468 | trace2_region_enter("diff", "setup", options->repo); | |
469 | if (!minimum_score) | |
470 | minimum_score = DEFAULT_RENAME_SCORE; | |
471 | ||
472 | for (i = 0; i < q->nr; i++) { | |
473 | struct diff_filepair *p = q->queue[i]; | |
474 | if (!DIFF_FILE_VALID(p->one)) { | |
475 | if (!DIFF_FILE_VALID(p->two)) | |
476 | continue; /* unmerged */ | |
477 | else if (options->single_follow && | |
478 | strcmp(options->single_follow, p->two->path)) | |
479 | continue; /* not interested */ | |
480 | else if (!options->flags.rename_empty && | |
481 | is_empty_blob_oid(&p->two->oid)) | |
482 | continue; | |
483 | else if (add_rename_dst(p) < 0) { | |
484 | warning("skipping rename detection, detected" | |
485 | " duplicate destination '%s'", | |
486 | p->two->path); | |
487 | goto cleanup; | |
488 | } | |
489 | } | |
490 | else if (!options->flags.rename_empty && | |
491 | is_empty_blob_oid(&p->one->oid)) | |
492 | continue; | |
493 | else if (!DIFF_PAIR_UNMERGED(p) && !DIFF_FILE_VALID(p->two)) { | |
494 | /* | |
495 | * If the source is a broken "delete", and | |
496 | * they did not really want to get broken, | |
497 | * that means the source actually stays. | |
498 | * So we increment the "rename_used" score | |
499 | * by one, to indicate ourselves as a user | |
500 | */ | |
501 | if (p->broken_pair && !p->score) | |
502 | p->one->rename_used++; | |
503 | register_rename_src(p); | |
504 | } | |
505 | else if (detect_rename == DIFF_DETECT_COPY) { | |
506 | /* | |
507 | * Increment the "rename_used" score by | |
508 | * one, to indicate ourselves as a user. | |
509 | */ | |
510 | p->one->rename_used++; | |
511 | register_rename_src(p); | |
512 | } | |
513 | } | |
514 | trace2_region_leave("diff", "setup", options->repo); | |
515 | if (rename_dst_nr == 0 || rename_src_nr == 0) | |
516 | goto cleanup; /* nothing to do */ | |
517 | ||
518 | trace2_region_enter("diff", "exact renames", options->repo); | |
519 | /* | |
520 | * We really want to cull the candidates list early | |
521 | * with cheap tests in order to avoid doing deltas. | |
522 | */ | |
523 | rename_count = find_exact_renames(options); | |
524 | trace2_region_leave("diff", "exact renames", options->repo); | |
525 | ||
526 | /* Did we only want exact renames? */ | |
527 | if (minimum_score == MAX_SCORE) | |
528 | goto cleanup; | |
529 | ||
530 | /* | |
531 | * Calculate how many renames are left (but all the source | |
532 | * files still remain as options for rename/copies!) | |
533 | */ | |
534 | num_destinations = (rename_dst_nr - rename_count); | |
535 | ||
536 | /* All done? */ | |
537 | if (!num_destinations) | |
538 | goto cleanup; | |
539 | ||
540 | switch (too_many_rename_candidates(num_destinations, rename_src_nr, | |
541 | options)) { | |
542 | case 1: | |
543 | goto cleanup; | |
544 | case 2: | |
545 | options->degraded_cc_to_c = 1; | |
546 | skip_unmodified = 1; | |
547 | break; | |
548 | default: | |
549 | break; | |
550 | } | |
551 | ||
552 | trace2_region_enter("diff", "inexact renames", options->repo); | |
553 | if (options->show_rename_progress) { | |
554 | progress = start_delayed_progress( | |
555 | _("Performing inexact rename detection"), | |
556 | (uint64_t)num_destinations * (uint64_t)rename_src_nr); | |
557 | } | |
558 | ||
559 | mx = xcalloc(st_mult(NUM_CANDIDATE_PER_DST, num_destinations), | |
560 | sizeof(*mx)); | |
561 | for (dst_cnt = i = 0; i < rename_dst_nr; i++) { | |
562 | struct diff_filespec *two = rename_dst[i].p->two; | |
563 | struct diff_score *m; | |
564 | ||
565 | if (rename_dst[i].is_rename) | |
566 | continue; /* dealt with exact match already. */ | |
567 | ||
568 | m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST]; | |
569 | for (j = 0; j < NUM_CANDIDATE_PER_DST; j++) | |
570 | m[j].dst = -1; | |
571 | ||
572 | for (j = 0; j < rename_src_nr; j++) { | |
573 | struct diff_filespec *one = rename_src[j].p->one; | |
574 | struct diff_score this_src; | |
575 | ||
576 | if (skip_unmodified && | |
577 | diff_unmodified_pair(rename_src[j].p)) | |
578 | continue; | |
579 | ||
580 | this_src.score = estimate_similarity(options->repo, | |
581 | one, two, | |
582 | minimum_score, | |
583 | skip_unmodified); | |
584 | this_src.name_score = basename_same(one, two); | |
585 | this_src.dst = i; | |
586 | this_src.src = j; | |
587 | record_if_better(m, &this_src); | |
588 | /* | |
589 | * Once we run estimate_similarity, | |
590 | * We do not need the text anymore. | |
591 | */ | |
592 | diff_free_filespec_blob(one); | |
593 | diff_free_filespec_blob(two); | |
594 | } | |
595 | dst_cnt++; | |
596 | display_progress(progress, | |
597 | (uint64_t)dst_cnt * (uint64_t)rename_src_nr); | |
598 | } | |
599 | stop_progress(&progress); | |
600 | ||
601 | /* cost matrix sorted by most to least similar pair */ | |
602 | STABLE_QSORT(mx, dst_cnt * NUM_CANDIDATE_PER_DST, score_compare); | |
603 | ||
604 | rename_count += find_renames(mx, dst_cnt, minimum_score, 0); | |
605 | if (detect_rename == DIFF_DETECT_COPY) | |
606 | rename_count += find_renames(mx, dst_cnt, minimum_score, 1); | |
607 | free(mx); | |
608 | trace2_region_leave("diff", "inexact renames", options->repo); | |
609 | ||
610 | cleanup: | |
611 | /* At this point, we have found some renames and copies and they | |
612 | * are recorded in rename_dst. The original list is still in *q. | |
613 | */ | |
614 | trace2_region_enter("diff", "write back to queue", options->repo); | |
615 | DIFF_QUEUE_CLEAR(&outq); | |
616 | for (i = 0; i < q->nr; i++) { | |
617 | struct diff_filepair *p = q->queue[i]; | |
618 | struct diff_filepair *pair_to_free = NULL; | |
619 | ||
620 | if (DIFF_PAIR_UNMERGED(p)) { | |
621 | diff_q(&outq, p); | |
622 | } | |
623 | else if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) { | |
624 | /* Creation */ | |
625 | diff_q(&outq, p); | |
626 | } | |
627 | else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) { | |
628 | /* | |
629 | * Deletion | |
630 | * | |
631 | * We would output this delete record if: | |
632 | * | |
633 | * (1) this is a broken delete and the counterpart | |
634 | * broken create remains in the output; or | |
635 | * (2) this is not a broken delete, and rename_dst | |
636 | * does not have a rename/copy to move p->one->path | |
637 | * out of existence. | |
638 | * | |
639 | * Otherwise, the counterpart broken create | |
640 | * has been turned into a rename-edit; or | |
641 | * delete did not have a matching create to | |
642 | * begin with. | |
643 | */ | |
644 | if (DIFF_PAIR_BROKEN(p)) { | |
645 | /* broken delete */ | |
646 | struct diff_rename_dst *dst = locate_rename_dst(p); | |
647 | if (!dst) | |
648 | BUG("tracking failed somehow; failed to find associated dst for broken pair"); | |
649 | if (dst->is_rename) | |
650 | /* counterpart is now rename/copy */ | |
651 | pair_to_free = p; | |
652 | } | |
653 | else { | |
654 | if (p->one->rename_used) | |
655 | /* this path remains */ | |
656 | pair_to_free = p; | |
657 | } | |
658 | ||
659 | if (!pair_to_free) | |
660 | diff_q(&outq, p); | |
661 | } | |
662 | else if (!diff_unmodified_pair(p)) | |
663 | /* all the usual ones need to be kept */ | |
664 | diff_q(&outq, p); | |
665 | else | |
666 | /* no need to keep unmodified pairs; FIXME: remove earlier? */ | |
667 | pair_to_free = p; | |
668 | ||
669 | if (pair_to_free) | |
670 | diff_free_filepair(pair_to_free); | |
671 | } | |
672 | diff_debug_queue("done copying original", &outq); | |
673 | ||
674 | free(q->queue); | |
675 | *q = outq; | |
676 | diff_debug_queue("done collapsing", q); | |
677 | ||
678 | for (i = 0; i < rename_dst_nr; i++) | |
679 | if (rename_dst[i].filespec_to_free) | |
680 | free_filespec(rename_dst[i].filespec_to_free); | |
681 | ||
682 | FREE_AND_NULL(rename_dst); | |
683 | rename_dst_nr = rename_dst_alloc = 0; | |
684 | FREE_AND_NULL(rename_src); | |
685 | rename_src_nr = rename_src_alloc = 0; | |
686 | if (break_idx) { | |
687 | strintmap_clear(break_idx); | |
688 | FREE_AND_NULL(break_idx); | |
689 | } | |
690 | trace2_region_leave("diff", "write back to queue", options->repo); | |
691 | return; | |
692 | } |