]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * | |
3 | * Copyright (C) 2005 Junio C Hamano | |
4 | */ | |
5 | #include "git-compat-util.h" | |
6 | #include "alloc.h" | |
7 | #include "diff.h" | |
8 | #include "diffcore.h" | |
9 | #include "object-store.h" | |
10 | #include "hashmap.h" | |
11 | #include "mem-pool.h" | |
12 | #include "oid-array.h" | |
13 | #include "progress.h" | |
14 | #include "promisor-remote.h" | |
15 | #include "string-list.h" | |
16 | #include "strmap.h" | |
17 | #include "trace2.h" | |
18 | ||
19 | /* Table of rename/copy destinations */ | |
20 | ||
21 | static struct diff_rename_dst { | |
22 | struct diff_filepair *p; | |
23 | struct diff_filespec *filespec_to_free; | |
24 | int is_rename; /* false -> just a create; true -> rename or copy */ | |
25 | } *rename_dst; | |
26 | static int rename_dst_nr, rename_dst_alloc; | |
27 | /* Mapping from break source pathname to break destination index */ | |
28 | static struct strintmap *break_idx = NULL; | |
29 | ||
30 | static struct diff_rename_dst *locate_rename_dst(struct diff_filepair *p) | |
31 | { | |
32 | /* Lookup by p->ONE->path */ | |
33 | int idx = break_idx ? strintmap_get(break_idx, p->one->path) : -1; | |
34 | return (idx == -1) ? NULL : &rename_dst[idx]; | |
35 | } | |
36 | ||
37 | /* | |
38 | * Returns 0 on success, -1 if we found a duplicate. | |
39 | */ | |
40 | static int add_rename_dst(struct diff_filepair *p) | |
41 | { | |
42 | ALLOC_GROW(rename_dst, rename_dst_nr + 1, rename_dst_alloc); | |
43 | rename_dst[rename_dst_nr].p = p; | |
44 | rename_dst[rename_dst_nr].filespec_to_free = NULL; | |
45 | rename_dst[rename_dst_nr].is_rename = 0; | |
46 | rename_dst_nr++; | |
47 | return 0; | |
48 | } | |
49 | ||
50 | /* Table of rename/copy src files */ | |
51 | static struct diff_rename_src { | |
52 | struct diff_filepair *p; | |
53 | unsigned short score; /* to remember the break score */ | |
54 | } *rename_src; | |
55 | static int rename_src_nr, rename_src_alloc; | |
56 | ||
57 | static void register_rename_src(struct diff_filepair *p) | |
58 | { | |
59 | if (p->broken_pair) { | |
60 | if (!break_idx) { | |
61 | break_idx = xmalloc(sizeof(*break_idx)); | |
62 | strintmap_init_with_options(break_idx, -1, NULL, 0); | |
63 | } | |
64 | strintmap_set(break_idx, p->one->path, rename_dst_nr); | |
65 | } | |
66 | ||
67 | ALLOC_GROW(rename_src, rename_src_nr + 1, rename_src_alloc); | |
68 | rename_src[rename_src_nr].p = p; | |
69 | rename_src[rename_src_nr].score = p->score; | |
70 | rename_src_nr++; | |
71 | } | |
72 | ||
73 | static int basename_same(struct diff_filespec *src, struct diff_filespec *dst) | |
74 | { | |
75 | int src_len = strlen(src->path), dst_len = strlen(dst->path); | |
76 | while (src_len && dst_len) { | |
77 | char c1 = src->path[--src_len]; | |
78 | char c2 = dst->path[--dst_len]; | |
79 | if (c1 != c2) | |
80 | return 0; | |
81 | if (c1 == '/') | |
82 | return 1; | |
83 | } | |
84 | return (!src_len || src->path[src_len - 1] == '/') && | |
85 | (!dst_len || dst->path[dst_len - 1] == '/'); | |
86 | } | |
87 | ||
88 | struct diff_score { | |
89 | int src; /* index in rename_src */ | |
90 | int dst; /* index in rename_dst */ | |
91 | unsigned short score; | |
92 | short name_score; | |
93 | }; | |
94 | ||
95 | struct inexact_prefetch_options { | |
96 | struct repository *repo; | |
97 | int skip_unmodified; | |
98 | }; | |
99 | static void inexact_prefetch(void *prefetch_options) | |
100 | { | |
101 | struct inexact_prefetch_options *options = prefetch_options; | |
102 | int i; | |
103 | struct oid_array to_fetch = OID_ARRAY_INIT; | |
104 | ||
105 | for (i = 0; i < rename_dst_nr; i++) { | |
106 | if (rename_dst[i].p->renamed_pair) | |
107 | /* | |
108 | * The loop in diffcore_rename() will not need these | |
109 | * blobs, so skip prefetching. | |
110 | */ | |
111 | continue; /* already found exact match */ | |
112 | diff_add_if_missing(options->repo, &to_fetch, | |
113 | rename_dst[i].p->two); | |
114 | } | |
115 | for (i = 0; i < rename_src_nr; i++) { | |
116 | if (options->skip_unmodified && | |
117 | diff_unmodified_pair(rename_src[i].p)) | |
118 | /* | |
119 | * The loop in diffcore_rename() will not need these | |
120 | * blobs, so skip prefetching. | |
121 | */ | |
122 | continue; | |
123 | diff_add_if_missing(options->repo, &to_fetch, | |
124 | rename_src[i].p->one); | |
125 | } | |
126 | promisor_remote_get_direct(options->repo, to_fetch.oid, to_fetch.nr); | |
127 | oid_array_clear(&to_fetch); | |
128 | } | |
129 | ||
130 | static int estimate_similarity(struct repository *r, | |
131 | struct diff_filespec *src, | |
132 | struct diff_filespec *dst, | |
133 | int minimum_score, | |
134 | struct diff_populate_filespec_options *dpf_opt) | |
135 | { | |
136 | /* src points at a file that existed in the original tree (or | |
137 | * optionally a file in the destination tree) and dst points | |
138 | * at a newly created file. They may be quite similar, in which | |
139 | * case we want to say src is renamed to dst or src is copied into | |
140 | * dst, and then some edit has been applied to dst. | |
141 | * | |
142 | * Compare them and return how similar they are, representing | |
143 | * the score as an integer between 0 and MAX_SCORE. | |
144 | * | |
145 | * When there is an exact match, it is considered a better | |
146 | * match than anything else; the destination does not even | |
147 | * call into this function in that case. | |
148 | */ | |
149 | unsigned long max_size, delta_size, base_size, src_copied, literal_added; | |
150 | int score; | |
151 | ||
152 | /* We deal only with regular files. Symlink renames are handled | |
153 | * only when they are exact matches --- in other words, no edits | |
154 | * after renaming. | |
155 | */ | |
156 | if (!S_ISREG(src->mode) || !S_ISREG(dst->mode)) | |
157 | return 0; | |
158 | ||
159 | /* | |
160 | * Need to check that source and destination sizes are | |
161 | * filled in before comparing them. | |
162 | * | |
163 | * If we already have "cnt_data" filled in, we know it's | |
164 | * all good (avoid checking the size for zero, as that | |
165 | * is a possible size - we really should have a flag to | |
166 | * say whether the size is valid or not!) | |
167 | */ | |
168 | dpf_opt->check_size_only = 1; | |
169 | ||
170 | if (!src->cnt_data && | |
171 | diff_populate_filespec(r, src, dpf_opt)) | |
172 | return 0; | |
173 | if (!dst->cnt_data && | |
174 | diff_populate_filespec(r, dst, dpf_opt)) | |
175 | return 0; | |
176 | ||
177 | max_size = ((src->size > dst->size) ? src->size : dst->size); | |
178 | base_size = ((src->size < dst->size) ? src->size : dst->size); | |
179 | delta_size = max_size - base_size; | |
180 | ||
181 | /* We would not consider edits that change the file size so | |
182 | * drastically. delta_size must be smaller than | |
183 | * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size). | |
184 | * | |
185 | * Note that base_size == 0 case is handled here already | |
186 | * and the final score computation below would not have a | |
187 | * divide-by-zero issue. | |
188 | */ | |
189 | if (max_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE) | |
190 | return 0; | |
191 | ||
192 | dpf_opt->check_size_only = 0; | |
193 | ||
194 | if (!src->cnt_data && diff_populate_filespec(r, src, dpf_opt)) | |
195 | return 0; | |
196 | if (!dst->cnt_data && diff_populate_filespec(r, dst, dpf_opt)) | |
197 | return 0; | |
198 | ||
199 | if (diffcore_count_changes(r, src, dst, | |
200 | &src->cnt_data, &dst->cnt_data, | |
201 | &src_copied, &literal_added)) | |
202 | return 0; | |
203 | ||
204 | /* How similar are they? | |
205 | * what percentage of material in dst are from source? | |
206 | */ | |
207 | if (!dst->size) | |
208 | score = 0; /* should not happen */ | |
209 | else | |
210 | score = (int)(src_copied * MAX_SCORE / max_size); | |
211 | return score; | |
212 | } | |
213 | ||
214 | static void record_rename_pair(int dst_index, int src_index, int score) | |
215 | { | |
216 | struct diff_filepair *src = rename_src[src_index].p; | |
217 | struct diff_filepair *dst = rename_dst[dst_index].p; | |
218 | ||
219 | if (dst->renamed_pair) | |
220 | die("internal error: dst already matched."); | |
221 | ||
222 | src->one->rename_used++; | |
223 | src->one->count++; | |
224 | ||
225 | rename_dst[dst_index].filespec_to_free = dst->one; | |
226 | rename_dst[dst_index].is_rename = 1; | |
227 | ||
228 | dst->one = src->one; | |
229 | dst->renamed_pair = 1; | |
230 | if (!strcmp(dst->one->path, dst->two->path)) | |
231 | dst->score = rename_src[src_index].score; | |
232 | else | |
233 | dst->score = score; | |
234 | } | |
235 | ||
236 | /* | |
237 | * We sort the rename similarity matrix with the score, in descending | |
238 | * order (the most similar first). | |
239 | */ | |
240 | static int score_compare(const void *a_, const void *b_) | |
241 | { | |
242 | const struct diff_score *a = a_, *b = b_; | |
243 | ||
244 | /* sink the unused ones to the bottom */ | |
245 | if (a->dst < 0) | |
246 | return (0 <= b->dst); | |
247 | else if (b->dst < 0) | |
248 | return -1; | |
249 | ||
250 | if (a->score == b->score) | |
251 | return b->name_score - a->name_score; | |
252 | ||
253 | return b->score - a->score; | |
254 | } | |
255 | ||
256 | struct file_similarity { | |
257 | struct hashmap_entry entry; | |
258 | int index; | |
259 | struct diff_filespec *filespec; | |
260 | }; | |
261 | ||
262 | static unsigned int hash_filespec(struct repository *r, | |
263 | struct diff_filespec *filespec) | |
264 | { | |
265 | if (!filespec->oid_valid) { | |
266 | if (diff_populate_filespec(r, filespec, NULL)) | |
267 | return 0; | |
268 | hash_object_file(r->hash_algo, filespec->data, filespec->size, | |
269 | OBJ_BLOB, &filespec->oid); | |
270 | } | |
271 | return oidhash(&filespec->oid); | |
272 | } | |
273 | ||
274 | static int find_identical_files(struct hashmap *srcs, | |
275 | int dst_index, | |
276 | struct diff_options *options) | |
277 | { | |
278 | int renames = 0; | |
279 | struct diff_filespec *target = rename_dst[dst_index].p->two; | |
280 | struct file_similarity *p, *best = NULL; | |
281 | int i = 100, best_score = -1; | |
282 | unsigned int hash = hash_filespec(options->repo, target); | |
283 | ||
284 | /* | |
285 | * Find the best source match for specified destination. | |
286 | */ | |
287 | p = hashmap_get_entry_from_hash(srcs, hash, NULL, | |
288 | struct file_similarity, entry); | |
289 | hashmap_for_each_entry_from(srcs, p, entry) { | |
290 | int score; | |
291 | struct diff_filespec *source = p->filespec; | |
292 | ||
293 | /* False hash collision? */ | |
294 | if (!oideq(&source->oid, &target->oid)) | |
295 | continue; | |
296 | /* Non-regular files? If so, the modes must match! */ | |
297 | if (!S_ISREG(source->mode) || !S_ISREG(target->mode)) { | |
298 | if (source->mode != target->mode) | |
299 | continue; | |
300 | } | |
301 | /* Give higher scores to sources that haven't been used already */ | |
302 | score = !source->rename_used; | |
303 | if (source->rename_used && options->detect_rename != DIFF_DETECT_COPY) | |
304 | continue; | |
305 | score += basename_same(source, target); | |
306 | if (score > best_score) { | |
307 | best = p; | |
308 | best_score = score; | |
309 | if (score == 2) | |
310 | break; | |
311 | } | |
312 | ||
313 | /* Too many identical alternatives? Pick one */ | |
314 | if (!--i) | |
315 | break; | |
316 | } | |
317 | if (best) { | |
318 | record_rename_pair(dst_index, best->index, MAX_SCORE); | |
319 | renames++; | |
320 | } | |
321 | return renames; | |
322 | } | |
323 | ||
324 | static void insert_file_table(struct repository *r, | |
325 | struct mem_pool *pool, | |
326 | struct hashmap *table, int index, | |
327 | struct diff_filespec *filespec) | |
328 | { | |
329 | struct file_similarity *entry = mem_pool_alloc(pool, sizeof(*entry)); | |
330 | ||
331 | entry->index = index; | |
332 | entry->filespec = filespec; | |
333 | ||
334 | hashmap_entry_init(&entry->entry, hash_filespec(r, filespec)); | |
335 | hashmap_add(table, &entry->entry); | |
336 | } | |
337 | ||
338 | /* | |
339 | * Find exact renames first. | |
340 | * | |
341 | * The first round matches up the up-to-date entries, | |
342 | * and then during the second round we try to match | |
343 | * cache-dirty entries as well. | |
344 | */ | |
345 | static int find_exact_renames(struct diff_options *options, | |
346 | struct mem_pool *pool) | |
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, pool, | |
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 (entries will be freed with the pool) */ | |
365 | hashmap_clear(&file_table); | |
366 | ||
367 | return renames; | |
368 | } | |
369 | ||
370 | struct dir_rename_info { | |
371 | struct strintmap idx_map; | |
372 | struct strmap dir_rename_guess; | |
373 | struct strmap *dir_rename_count; | |
374 | struct strintmap *relevant_source_dirs; | |
375 | unsigned setup; | |
376 | }; | |
377 | ||
378 | static char *get_dirname(const char *filename) | |
379 | { | |
380 | char *slash = strrchr(filename, '/'); | |
381 | return slash ? xstrndup(filename, slash - filename) : xstrdup(""); | |
382 | } | |
383 | ||
384 | static void dirname_munge(char *filename) | |
385 | { | |
386 | char *slash = strrchr(filename, '/'); | |
387 | if (!slash) | |
388 | slash = filename; | |
389 | *slash = '\0'; | |
390 | } | |
391 | ||
392 | static const char *get_highest_rename_path(struct strintmap *counts) | |
393 | { | |
394 | int highest_count = 0; | |
395 | const char *highest_destination_dir = NULL; | |
396 | struct hashmap_iter iter; | |
397 | struct strmap_entry *entry; | |
398 | ||
399 | strintmap_for_each_entry(counts, &iter, entry) { | |
400 | const char *destination_dir = entry->key; | |
401 | intptr_t count = (intptr_t)entry->value; | |
402 | if (count > highest_count) { | |
403 | highest_count = count; | |
404 | highest_destination_dir = destination_dir; | |
405 | } | |
406 | } | |
407 | return highest_destination_dir; | |
408 | } | |
409 | ||
410 | static char *UNKNOWN_DIR = "/"; /* placeholder -- short, illegal directory */ | |
411 | ||
412 | static int dir_rename_already_determinable(struct strintmap *counts) | |
413 | { | |
414 | struct hashmap_iter iter; | |
415 | struct strmap_entry *entry; | |
416 | int first = 0, second = 0, unknown = 0; | |
417 | strintmap_for_each_entry(counts, &iter, entry) { | |
418 | const char *destination_dir = entry->key; | |
419 | intptr_t count = (intptr_t)entry->value; | |
420 | if (!strcmp(destination_dir, UNKNOWN_DIR)) { | |
421 | unknown = count; | |
422 | } else if (count >= first) { | |
423 | second = first; | |
424 | first = count; | |
425 | } else if (count >= second) { | |
426 | second = count; | |
427 | } | |
428 | } | |
429 | return first > second + unknown; | |
430 | } | |
431 | ||
432 | static void increment_count(struct dir_rename_info *info, | |
433 | char *old_dir, | |
434 | char *new_dir) | |
435 | { | |
436 | struct strintmap *counts; | |
437 | struct strmap_entry *e; | |
438 | ||
439 | /* Get the {new_dirs -> counts} mapping using old_dir */ | |
440 | e = strmap_get_entry(info->dir_rename_count, old_dir); | |
441 | if (e) { | |
442 | counts = e->value; | |
443 | } else { | |
444 | counts = xmalloc(sizeof(*counts)); | |
445 | strintmap_init_with_options(counts, 0, NULL, 1); | |
446 | strmap_put(info->dir_rename_count, old_dir, counts); | |
447 | } | |
448 | ||
449 | /* Increment the count for new_dir */ | |
450 | strintmap_incr(counts, new_dir, 1); | |
451 | } | |
452 | ||
453 | static void update_dir_rename_counts(struct dir_rename_info *info, | |
454 | struct strintmap *dirs_removed, | |
455 | const char *oldname, | |
456 | const char *newname) | |
457 | { | |
458 | char *old_dir; | |
459 | char *new_dir; | |
460 | const char new_dir_first_char = newname[0]; | |
461 | int first_time_in_loop = 1; | |
462 | ||
463 | if (!info->setup) | |
464 | /* | |
465 | * info->setup is 0 here in two cases: (1) all auxiliary | |
466 | * vars (like dirs_removed) were NULL so | |
467 | * initialize_dir_rename_info() returned early, or (2) | |
468 | * either break detection or copy detection are active so | |
469 | * that we never called initialize_dir_rename_info(). In | |
470 | * the former case, we don't have enough info to know if | |
471 | * directories were renamed (because dirs_removed lets us | |
472 | * know about a necessary prerequisite, namely if they were | |
473 | * removed), and in the latter, we don't care about | |
474 | * directory renames or find_basename_matches. | |
475 | * | |
476 | * This matters because both basename and inexact matching | |
477 | * will also call update_dir_rename_counts(). In either of | |
478 | * the above two cases info->dir_rename_counts will not | |
479 | * have been properly initialized which prevents us from | |
480 | * updating it, but in these two cases we don't care about | |
481 | * dir_rename_counts anyway, so we can just exit early. | |
482 | */ | |
483 | return; | |
484 | ||
485 | ||
486 | old_dir = xstrdup(oldname); | |
487 | new_dir = xstrdup(newname); | |
488 | ||
489 | while (1) { | |
490 | int drd_flag = NOT_RELEVANT; | |
491 | ||
492 | /* Get old_dir, skip if its directory isn't relevant. */ | |
493 | dirname_munge(old_dir); | |
494 | if (info->relevant_source_dirs && | |
495 | !strintmap_contains(info->relevant_source_dirs, old_dir)) | |
496 | break; | |
497 | ||
498 | /* Get new_dir */ | |
499 | dirname_munge(new_dir); | |
500 | ||
501 | /* | |
502 | * When renaming | |
503 | * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c" | |
504 | * then this suggests that both | |
505 | * a/b/c/d/e/ => a/b/some/thing/else/e/ | |
506 | * a/b/c/d/ => a/b/some/thing/else/ | |
507 | * so we want to increment counters for both. We do NOT, | |
508 | * however, also want to suggest that there was the following | |
509 | * rename: | |
510 | * a/b/c/ => a/b/some/thing/ | |
511 | * so we need to quit at that point. | |
512 | * | |
513 | * Note the when first_time_in_loop, we only strip off the | |
514 | * basename, and we don't care if that's different. | |
515 | */ | |
516 | if (!first_time_in_loop) { | |
517 | char *old_sub_dir = strchr(old_dir, '\0')+1; | |
518 | char *new_sub_dir = strchr(new_dir, '\0')+1; | |
519 | if (!*new_dir) { | |
520 | /* | |
521 | * Special case when renaming to root directory, | |
522 | * i.e. when new_dir == "". In this case, we had | |
523 | * something like | |
524 | * a/b/subdir => subdir | |
525 | * and so dirname_munge() sets things up so that | |
526 | * old_dir = "a/b\0subdir\0" | |
527 | * new_dir = "\0ubdir\0" | |
528 | * We didn't have a '/' to overwrite a '\0' onto | |
529 | * in new_dir, so we have to compare differently. | |
530 | */ | |
531 | if (new_dir_first_char != old_sub_dir[0] || | |
532 | strcmp(old_sub_dir+1, new_sub_dir)) | |
533 | break; | |
534 | } else { | |
535 | if (strcmp(old_sub_dir, new_sub_dir)) | |
536 | break; | |
537 | } | |
538 | } | |
539 | ||
540 | /* | |
541 | * Above we suggested that we'd keep recording renames for | |
542 | * all ancestor directories where the trailing directories | |
543 | * matched, i.e. for | |
544 | * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c" | |
545 | * we'd increment rename counts for each of | |
546 | * a/b/c/d/e/ => a/b/some/thing/else/e/ | |
547 | * a/b/c/d/ => a/b/some/thing/else/ | |
548 | * However, we only need the rename counts for directories | |
549 | * in dirs_removed whose value is RELEVANT_FOR_SELF. | |
550 | * However, we add one special case of also recording it for | |
551 | * first_time_in_loop because find_basename_matches() can | |
552 | * use that as a hint to find a good pairing. | |
553 | */ | |
554 | if (dirs_removed) | |
555 | drd_flag = strintmap_get(dirs_removed, old_dir); | |
556 | if (drd_flag == RELEVANT_FOR_SELF || first_time_in_loop) | |
557 | increment_count(info, old_dir, new_dir); | |
558 | ||
559 | first_time_in_loop = 0; | |
560 | if (drd_flag == NOT_RELEVANT) | |
561 | break; | |
562 | /* If we hit toplevel directory ("") for old or new dir, quit */ | |
563 | if (!*old_dir || !*new_dir) | |
564 | break; | |
565 | } | |
566 | ||
567 | /* Free resources we don't need anymore */ | |
568 | free(old_dir); | |
569 | free(new_dir); | |
570 | } | |
571 | ||
572 | static void initialize_dir_rename_info(struct dir_rename_info *info, | |
573 | struct strintmap *relevant_sources, | |
574 | struct strintmap *dirs_removed, | |
575 | struct strmap *dir_rename_count, | |
576 | struct strmap *cached_pairs) | |
577 | { | |
578 | struct hashmap_iter iter; | |
579 | struct strmap_entry *entry; | |
580 | int i; | |
581 | ||
582 | if (!dirs_removed && !relevant_sources) { | |
583 | info->setup = 0; | |
584 | return; | |
585 | } | |
586 | info->setup = 1; | |
587 | ||
588 | info->dir_rename_count = dir_rename_count; | |
589 | if (!info->dir_rename_count) { | |
590 | info->dir_rename_count = xmalloc(sizeof(*dir_rename_count)); | |
591 | strmap_init(info->dir_rename_count); | |
592 | } | |
593 | strintmap_init_with_options(&info->idx_map, -1, NULL, 0); | |
594 | strmap_init_with_options(&info->dir_rename_guess, NULL, 0); | |
595 | ||
596 | /* Setup info->relevant_source_dirs */ | |
597 | info->relevant_source_dirs = NULL; | |
598 | if (dirs_removed || !relevant_sources) { | |
599 | info->relevant_source_dirs = dirs_removed; /* might be NULL */ | |
600 | } else { | |
601 | info->relevant_source_dirs = xmalloc(sizeof(struct strintmap)); | |
602 | strintmap_init(info->relevant_source_dirs, 0 /* unused */); | |
603 | strintmap_for_each_entry(relevant_sources, &iter, entry) { | |
604 | char *dirname = get_dirname(entry->key); | |
605 | if (!dirs_removed || | |
606 | strintmap_contains(dirs_removed, dirname)) | |
607 | strintmap_set(info->relevant_source_dirs, | |
608 | dirname, 0 /* value irrelevant */); | |
609 | free(dirname); | |
610 | } | |
611 | } | |
612 | ||
613 | /* | |
614 | * Loop setting up both info->idx_map, and doing setup of | |
615 | * info->dir_rename_count. | |
616 | */ | |
617 | for (i = 0; i < rename_dst_nr; ++i) { | |
618 | /* | |
619 | * For non-renamed files, make idx_map contain mapping of | |
620 | * filename -> index (index within rename_dst, that is) | |
621 | */ | |
622 | if (!rename_dst[i].is_rename) { | |
623 | char *filename = rename_dst[i].p->two->path; | |
624 | strintmap_set(&info->idx_map, filename, i); | |
625 | continue; | |
626 | } | |
627 | ||
628 | /* | |
629 | * For everything else (i.e. renamed files), make | |
630 | * dir_rename_count contain a map of a map: | |
631 | * old_directory -> {new_directory -> count} | |
632 | * In other words, for every pair look at the directories for | |
633 | * the old filename and the new filename and count how many | |
634 | * times that pairing occurs. | |
635 | */ | |
636 | update_dir_rename_counts(info, dirs_removed, | |
637 | rename_dst[i].p->one->path, | |
638 | rename_dst[i].p->two->path); | |
639 | } | |
640 | ||
641 | /* Add cached_pairs to counts */ | |
642 | strmap_for_each_entry(cached_pairs, &iter, entry) { | |
643 | const char *old_name = entry->key; | |
644 | const char *new_name = entry->value; | |
645 | if (!new_name) | |
646 | /* known delete; ignore it */ | |
647 | continue; | |
648 | ||
649 | update_dir_rename_counts(info, dirs_removed, old_name, new_name); | |
650 | } | |
651 | ||
652 | /* | |
653 | * Now we collapse | |
654 | * dir_rename_count: old_directory -> {new_directory -> count} | |
655 | * down to | |
656 | * dir_rename_guess: old_directory -> best_new_directory | |
657 | * where best_new_directory is the one with the highest count. | |
658 | */ | |
659 | strmap_for_each_entry(info->dir_rename_count, &iter, entry) { | |
660 | /* entry->key is source_dir */ | |
661 | struct strintmap *counts = entry->value; | |
662 | char *best_newdir; | |
663 | ||
664 | best_newdir = xstrdup(get_highest_rename_path(counts)); | |
665 | strmap_put(&info->dir_rename_guess, entry->key, | |
666 | best_newdir); | |
667 | } | |
668 | } | |
669 | ||
670 | void partial_clear_dir_rename_count(struct strmap *dir_rename_count) | |
671 | { | |
672 | struct hashmap_iter iter; | |
673 | struct strmap_entry *entry; | |
674 | ||
675 | strmap_for_each_entry(dir_rename_count, &iter, entry) { | |
676 | struct strintmap *counts = entry->value; | |
677 | strintmap_clear(counts); | |
678 | } | |
679 | strmap_partial_clear(dir_rename_count, 1); | |
680 | } | |
681 | ||
682 | static void cleanup_dir_rename_info(struct dir_rename_info *info, | |
683 | struct strintmap *dirs_removed, | |
684 | int keep_dir_rename_count) | |
685 | { | |
686 | struct hashmap_iter iter; | |
687 | struct strmap_entry *entry; | |
688 | struct string_list to_remove = STRING_LIST_INIT_NODUP; | |
689 | int i; | |
690 | ||
691 | if (!info->setup) | |
692 | return; | |
693 | ||
694 | /* idx_map */ | |
695 | strintmap_clear(&info->idx_map); | |
696 | ||
697 | /* dir_rename_guess */ | |
698 | strmap_clear(&info->dir_rename_guess, 1); | |
699 | ||
700 | /* relevant_source_dirs */ | |
701 | if (info->relevant_source_dirs && | |
702 | info->relevant_source_dirs != dirs_removed) { | |
703 | strintmap_clear(info->relevant_source_dirs); | |
704 | FREE_AND_NULL(info->relevant_source_dirs); | |
705 | } | |
706 | ||
707 | /* dir_rename_count */ | |
708 | if (!keep_dir_rename_count) { | |
709 | partial_clear_dir_rename_count(info->dir_rename_count); | |
710 | strmap_clear(info->dir_rename_count, 1); | |
711 | FREE_AND_NULL(info->dir_rename_count); | |
712 | return; | |
713 | } | |
714 | ||
715 | /* | |
716 | * Although dir_rename_count was passed in | |
717 | * diffcore_rename_extended() and we want to keep it around and | |
718 | * return it to that caller, we first want to remove any counts in | |
719 | * the maps associated with UNKNOWN_DIR entries and any data | |
720 | * associated with directories that weren't renamed. | |
721 | */ | |
722 | strmap_for_each_entry(info->dir_rename_count, &iter, entry) { | |
723 | const char *source_dir = entry->key; | |
724 | struct strintmap *counts = entry->value; | |
725 | ||
726 | if (!strintmap_get(dirs_removed, source_dir)) { | |
727 | string_list_append(&to_remove, source_dir); | |
728 | strintmap_clear(counts); | |
729 | continue; | |
730 | } | |
731 | ||
732 | if (strintmap_contains(counts, UNKNOWN_DIR)) | |
733 | strintmap_remove(counts, UNKNOWN_DIR); | |
734 | } | |
735 | for (i = 0; i < to_remove.nr; ++i) | |
736 | strmap_remove(info->dir_rename_count, | |
737 | to_remove.items[i].string, 1); | |
738 | string_list_clear(&to_remove, 0); | |
739 | } | |
740 | ||
741 | static const char *get_basename(const char *filename) | |
742 | { | |
743 | /* | |
744 | * gitbasename() has to worry about special drives, multiple | |
745 | * directory separator characters, trailing slashes, NULL or | |
746 | * empty strings, etc. We only work on filenames as stored in | |
747 | * git, and thus get to ignore all those complications. | |
748 | */ | |
749 | const char *base = strrchr(filename, '/'); | |
750 | return base ? base + 1 : filename; | |
751 | } | |
752 | ||
753 | static int idx_possible_rename(char *filename, struct dir_rename_info *info) | |
754 | { | |
755 | /* | |
756 | * Our comparison of files with the same basename (see | |
757 | * find_basename_matches() below), is only helpful when after exact | |
758 | * rename detection we have exactly one file with a given basename | |
759 | * among the rename sources and also only exactly one file with | |
760 | * that basename among the rename destinations. When we have | |
761 | * multiple files with the same basename in either set, we do not | |
762 | * know which to compare against. However, there are some | |
763 | * filenames that occur in large numbers (particularly | |
764 | * build-related filenames such as 'Makefile', '.gitignore', or | |
765 | * 'build.gradle' that potentially exist within every single | |
766 | * subdirectory), and for performance we want to be able to quickly | |
767 | * find renames for these files too. | |
768 | * | |
769 | * The reason basename comparisons are a useful heuristic was that it | |
770 | * is common for people to move files across directories while keeping | |
771 | * their filename the same. If we had a way of determining or even | |
772 | * making a good educated guess about which directory these non-unique | |
773 | * basename files had moved the file to, we could check it. | |
774 | * Luckily... | |
775 | * | |
776 | * When an entire directory is in fact renamed, we have two factors | |
777 | * helping us out: | |
778 | * (a) the original directory disappeared giving us a hint | |
779 | * about when we can apply an extra heuristic. | |
780 | * (a) we often have several files within that directory and | |
781 | * subdirectories that are renamed without changes | |
782 | * So, rules for a heuristic: | |
783 | * (0) If there basename matches are non-unique (the condition under | |
784 | * which this function is called) AND | |
785 | * (1) the directory in which the file was found has disappeared | |
786 | * (i.e. dirs_removed is non-NULL and has a relevant entry) THEN | |
787 | * (2) use exact renames of files within the directory to determine | |
788 | * where the directory is likely to have been renamed to. IF | |
789 | * there is at least one exact rename from within that | |
790 | * directory, we can proceed. | |
791 | * (3) If there are multiple places the directory could have been | |
792 | * renamed to based on exact renames, ignore all but one of them. | |
793 | * Just use the destination with the most renames going to it. | |
794 | * (4) Check if applying that directory rename to the original file | |
795 | * would result in a destination filename that is in the | |
796 | * potential rename set. If so, return the index of the | |
797 | * destination file (the index within rename_dst). | |
798 | * (5) Compare the original file and returned destination for | |
799 | * similarity, and if they are sufficiently similar, record the | |
800 | * rename. | |
801 | * | |
802 | * This function, idx_possible_rename(), is only responsible for (4). | |
803 | * The conditions/steps in (1)-(3) are handled via setting up | |
804 | * dir_rename_count and dir_rename_guess in | |
805 | * initialize_dir_rename_info(). Steps (0) and (5) are handled by | |
806 | * the caller of this function. | |
807 | */ | |
808 | char *old_dir, *new_dir; | |
809 | struct strbuf new_path = STRBUF_INIT; | |
810 | int idx; | |
811 | ||
812 | if (!info->setup) | |
813 | return -1; | |
814 | ||
815 | old_dir = get_dirname(filename); | |
816 | new_dir = strmap_get(&info->dir_rename_guess, old_dir); | |
817 | free(old_dir); | |
818 | if (!new_dir) | |
819 | return -1; | |
820 | ||
821 | strbuf_addstr(&new_path, new_dir); | |
822 | strbuf_addch(&new_path, '/'); | |
823 | strbuf_addstr(&new_path, get_basename(filename)); | |
824 | ||
825 | idx = strintmap_get(&info->idx_map, new_path.buf); | |
826 | strbuf_release(&new_path); | |
827 | return idx; | |
828 | } | |
829 | ||
830 | struct basename_prefetch_options { | |
831 | struct repository *repo; | |
832 | struct strintmap *relevant_sources; | |
833 | struct strintmap *sources; | |
834 | struct strintmap *dests; | |
835 | struct dir_rename_info *info; | |
836 | }; | |
837 | static void basename_prefetch(void *prefetch_options) | |
838 | { | |
839 | struct basename_prefetch_options *options = prefetch_options; | |
840 | struct strintmap *relevant_sources = options->relevant_sources; | |
841 | struct strintmap *sources = options->sources; | |
842 | struct strintmap *dests = options->dests; | |
843 | struct dir_rename_info *info = options->info; | |
844 | int i; | |
845 | struct oid_array to_fetch = OID_ARRAY_INIT; | |
846 | ||
847 | /* | |
848 | * TODO: The following loops mirror the code/logic from | |
849 | * find_basename_matches(), though not quite exactly. Maybe | |
850 | * abstract the iteration logic out somehow? | |
851 | */ | |
852 | for (i = 0; i < rename_src_nr; ++i) { | |
853 | char *filename = rename_src[i].p->one->path; | |
854 | const char *base = NULL; | |
855 | intptr_t src_index; | |
856 | intptr_t dst_index; | |
857 | ||
858 | /* Skip irrelevant sources */ | |
859 | if (relevant_sources && | |
860 | !strintmap_contains(relevant_sources, filename)) | |
861 | continue; | |
862 | ||
863 | /* | |
864 | * If the basename is unique among remaining sources, then | |
865 | * src_index will equal 'i' and we can attempt to match it | |
866 | * to a unique basename in the destinations. Otherwise, | |
867 | * use directory rename heuristics, if possible. | |
868 | */ | |
869 | base = get_basename(filename); | |
870 | src_index = strintmap_get(sources, base); | |
871 | assert(src_index == -1 || src_index == i); | |
872 | ||
873 | if (strintmap_contains(dests, base)) { | |
874 | struct diff_filespec *one, *two; | |
875 | ||
876 | /* Find a matching destination, if possible */ | |
877 | dst_index = strintmap_get(dests, base); | |
878 | if (src_index == -1 || dst_index == -1) { | |
879 | src_index = i; | |
880 | dst_index = idx_possible_rename(filename, info); | |
881 | } | |
882 | if (dst_index == -1) | |
883 | continue; | |
884 | ||
885 | /* Ignore this dest if already used in a rename */ | |
886 | if (rename_dst[dst_index].is_rename) | |
887 | continue; /* already used previously */ | |
888 | ||
889 | one = rename_src[src_index].p->one; | |
890 | two = rename_dst[dst_index].p->two; | |
891 | ||
892 | /* Add the pairs */ | |
893 | diff_add_if_missing(options->repo, &to_fetch, two); | |
894 | diff_add_if_missing(options->repo, &to_fetch, one); | |
895 | } | |
896 | } | |
897 | ||
898 | promisor_remote_get_direct(options->repo, to_fetch.oid, to_fetch.nr); | |
899 | oid_array_clear(&to_fetch); | |
900 | } | |
901 | ||
902 | static int find_basename_matches(struct diff_options *options, | |
903 | int minimum_score, | |
904 | struct dir_rename_info *info, | |
905 | struct strintmap *relevant_sources, | |
906 | struct strintmap *dirs_removed) | |
907 | { | |
908 | /* | |
909 | * When I checked in early 2020, over 76% of file renames in linux | |
910 | * just moved files to a different directory but kept the same | |
911 | * basename. gcc did that with over 64% of renames, gecko did it | |
912 | * with over 79%, and WebKit did it with over 89%. | |
913 | * | |
914 | * Therefore we can bypass the normal exhaustive NxM matrix | |
915 | * comparison of similarities between all potential rename sources | |
916 | * and destinations by instead using file basename as a hint (i.e. | |
917 | * the portion of the filename after the last '/'), checking for | |
918 | * similarity between files with the same basename, and if we find | |
919 | * a pair that are sufficiently similar, record the rename pair and | |
920 | * exclude those two from the NxM matrix. | |
921 | * | |
922 | * This *might* cause us to find a less than optimal pairing (if | |
923 | * there is another file that we are even more similar to but has a | |
924 | * different basename). Given the huge performance advantage | |
925 | * basename matching provides, and given the frequency with which | |
926 | * people use the same basename in real world projects, that's a | |
927 | * trade-off we are willing to accept when doing just rename | |
928 | * detection. | |
929 | * | |
930 | * If someone wants copy detection that implies they are willing to | |
931 | * spend more cycles to find similarities between files, so it may | |
932 | * be less likely that this heuristic is wanted. If someone is | |
933 | * doing break detection, that means they do not want filename | |
934 | * similarity to imply any form of content similiarity, and thus | |
935 | * this heuristic would definitely be incompatible. | |
936 | */ | |
937 | ||
938 | int i, renames = 0; | |
939 | struct strintmap sources; | |
940 | struct strintmap dests; | |
941 | struct diff_populate_filespec_options dpf_options = { | |
942 | .check_binary = 0, | |
943 | .missing_object_cb = NULL, | |
944 | .missing_object_data = NULL | |
945 | }; | |
946 | struct basename_prefetch_options prefetch_options = { | |
947 | .repo = options->repo, | |
948 | .relevant_sources = relevant_sources, | |
949 | .sources = &sources, | |
950 | .dests = &dests, | |
951 | .info = info | |
952 | }; | |
953 | ||
954 | /* | |
955 | * Create maps of basename -> fullname(s) for remaining sources and | |
956 | * dests. | |
957 | */ | |
958 | strintmap_init_with_options(&sources, -1, NULL, 0); | |
959 | strintmap_init_with_options(&dests, -1, NULL, 0); | |
960 | for (i = 0; i < rename_src_nr; ++i) { | |
961 | char *filename = rename_src[i].p->one->path; | |
962 | const char *base; | |
963 | ||
964 | /* exact renames removed in remove_unneeded_paths_from_src() */ | |
965 | assert(!rename_src[i].p->one->rename_used); | |
966 | ||
967 | /* Record index within rename_src (i) if basename is unique */ | |
968 | base = get_basename(filename); | |
969 | if (strintmap_contains(&sources, base)) | |
970 | strintmap_set(&sources, base, -1); | |
971 | else | |
972 | strintmap_set(&sources, base, i); | |
973 | } | |
974 | for (i = 0; i < rename_dst_nr; ++i) { | |
975 | char *filename = rename_dst[i].p->two->path; | |
976 | const char *base; | |
977 | ||
978 | if (rename_dst[i].is_rename) | |
979 | continue; /* involved in exact match already. */ | |
980 | ||
981 | /* Record index within rename_dst (i) if basename is unique */ | |
982 | base = get_basename(filename); | |
983 | if (strintmap_contains(&dests, base)) | |
984 | strintmap_set(&dests, base, -1); | |
985 | else | |
986 | strintmap_set(&dests, base, i); | |
987 | } | |
988 | ||
989 | if (options->repo == the_repository && repo_has_promisor_remote(the_repository)) { | |
990 | dpf_options.missing_object_cb = basename_prefetch; | |
991 | dpf_options.missing_object_data = &prefetch_options; | |
992 | } | |
993 | ||
994 | /* Now look for basename matchups and do similarity estimation */ | |
995 | for (i = 0; i < rename_src_nr; ++i) { | |
996 | char *filename = rename_src[i].p->one->path; | |
997 | const char *base = NULL; | |
998 | intptr_t src_index; | |
999 | intptr_t dst_index; | |
1000 | ||
1001 | /* Skip irrelevant sources */ | |
1002 | if (relevant_sources && | |
1003 | !strintmap_contains(relevant_sources, filename)) | |
1004 | continue; | |
1005 | ||
1006 | /* | |
1007 | * If the basename is unique among remaining sources, then | |
1008 | * src_index will equal 'i' and we can attempt to match it | |
1009 | * to a unique basename in the destinations. Otherwise, | |
1010 | * use directory rename heuristics, if possible. | |
1011 | */ | |
1012 | base = get_basename(filename); | |
1013 | src_index = strintmap_get(&sources, base); | |
1014 | assert(src_index == -1 || src_index == i); | |
1015 | ||
1016 | if (strintmap_contains(&dests, base)) { | |
1017 | struct diff_filespec *one, *two; | |
1018 | int score; | |
1019 | ||
1020 | /* Find a matching destination, if possible */ | |
1021 | dst_index = strintmap_get(&dests, base); | |
1022 | if (src_index == -1 || dst_index == -1) { | |
1023 | src_index = i; | |
1024 | dst_index = idx_possible_rename(filename, info); | |
1025 | } | |
1026 | if (dst_index == -1) | |
1027 | continue; | |
1028 | ||
1029 | /* Ignore this dest if already used in a rename */ | |
1030 | if (rename_dst[dst_index].is_rename) | |
1031 | continue; /* already used previously */ | |
1032 | ||
1033 | /* Estimate the similarity */ | |
1034 | one = rename_src[src_index].p->one; | |
1035 | two = rename_dst[dst_index].p->two; | |
1036 | score = estimate_similarity(options->repo, one, two, | |
1037 | minimum_score, &dpf_options); | |
1038 | ||
1039 | /* If sufficiently similar, record as rename pair */ | |
1040 | if (score < minimum_score) | |
1041 | continue; | |
1042 | record_rename_pair(dst_index, src_index, score); | |
1043 | renames++; | |
1044 | update_dir_rename_counts(info, dirs_removed, | |
1045 | one->path, two->path); | |
1046 | ||
1047 | /* | |
1048 | * Found a rename so don't need text anymore; if we | |
1049 | * didn't find a rename, the filespec_blob would get | |
1050 | * re-used when doing the matrix of comparisons. | |
1051 | */ | |
1052 | diff_free_filespec_blob(one); | |
1053 | diff_free_filespec_blob(two); | |
1054 | } | |
1055 | } | |
1056 | ||
1057 | strintmap_clear(&sources); | |
1058 | strintmap_clear(&dests); | |
1059 | ||
1060 | return renames; | |
1061 | } | |
1062 | ||
1063 | #define NUM_CANDIDATE_PER_DST 4 | |
1064 | static void record_if_better(struct diff_score m[], struct diff_score *o) | |
1065 | { | |
1066 | int i, worst; | |
1067 | ||
1068 | /* find the worst one */ | |
1069 | worst = 0; | |
1070 | for (i = 1; i < NUM_CANDIDATE_PER_DST; i++) | |
1071 | if (score_compare(&m[i], &m[worst]) > 0) | |
1072 | worst = i; | |
1073 | ||
1074 | /* is it better than the worst one? */ | |
1075 | if (score_compare(&m[worst], o) > 0) | |
1076 | m[worst] = *o; | |
1077 | } | |
1078 | ||
1079 | /* | |
1080 | * Returns: | |
1081 | * 0 if we are under the limit; | |
1082 | * 1 if we need to disable inexact rename detection; | |
1083 | * 2 if we would be under the limit if we were given -C instead of -C -C. | |
1084 | */ | |
1085 | static int too_many_rename_candidates(int num_destinations, int num_sources, | |
1086 | struct diff_options *options) | |
1087 | { | |
1088 | int rename_limit = options->rename_limit; | |
1089 | int i, limited_sources; | |
1090 | ||
1091 | options->needed_rename_limit = 0; | |
1092 | ||
1093 | /* | |
1094 | * This basically does a test for the rename matrix not | |
1095 | * growing larger than a "rename_limit" square matrix, ie: | |
1096 | * | |
1097 | * num_destinations * num_sources > rename_limit * rename_limit | |
1098 | * | |
1099 | * We use st_mult() to check overflow conditions; in the | |
1100 | * exceptional circumstance that size_t isn't large enough to hold | |
1101 | * the multiplication, the system won't be able to allocate enough | |
1102 | * memory for the matrix anyway. | |
1103 | */ | |
1104 | if (rename_limit <= 0) | |
1105 | return 0; /* treat as unlimited */ | |
1106 | if (st_mult(num_destinations, num_sources) | |
1107 | <= st_mult(rename_limit, rename_limit)) | |
1108 | return 0; | |
1109 | ||
1110 | options->needed_rename_limit = | |
1111 | num_sources > num_destinations ? num_sources : num_destinations; | |
1112 | ||
1113 | /* Are we running under -C -C? */ | |
1114 | if (!options->flags.find_copies_harder) | |
1115 | return 1; | |
1116 | ||
1117 | /* Would we bust the limit if we were running under -C? */ | |
1118 | for (limited_sources = i = 0; i < num_sources; i++) { | |
1119 | if (diff_unmodified_pair(rename_src[i].p)) | |
1120 | continue; | |
1121 | limited_sources++; | |
1122 | } | |
1123 | if (st_mult(num_destinations, limited_sources) | |
1124 | <= st_mult(rename_limit, rename_limit)) | |
1125 | return 2; | |
1126 | return 1; | |
1127 | } | |
1128 | ||
1129 | static int find_renames(struct diff_score *mx, | |
1130 | int dst_cnt, | |
1131 | int minimum_score, | |
1132 | int copies, | |
1133 | struct dir_rename_info *info, | |
1134 | struct strintmap *dirs_removed) | |
1135 | { | |
1136 | int count = 0, i; | |
1137 | ||
1138 | for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) { | |
1139 | struct diff_rename_dst *dst; | |
1140 | ||
1141 | if ((mx[i].dst < 0) || | |
1142 | (mx[i].score < minimum_score)) | |
1143 | break; /* there is no more usable pair. */ | |
1144 | dst = &rename_dst[mx[i].dst]; | |
1145 | if (dst->is_rename) | |
1146 | continue; /* already done, either exact or fuzzy. */ | |
1147 | if (!copies && rename_src[mx[i].src].p->one->rename_used) | |
1148 | continue; | |
1149 | record_rename_pair(mx[i].dst, mx[i].src, mx[i].score); | |
1150 | count++; | |
1151 | update_dir_rename_counts(info, dirs_removed, | |
1152 | rename_src[mx[i].src].p->one->path, | |
1153 | rename_dst[mx[i].dst].p->two->path); | |
1154 | } | |
1155 | return count; | |
1156 | } | |
1157 | ||
1158 | static void remove_unneeded_paths_from_src(int detecting_copies, | |
1159 | struct strintmap *interesting) | |
1160 | { | |
1161 | int i, new_num_src; | |
1162 | ||
1163 | if (detecting_copies && !interesting) | |
1164 | return; /* nothing to remove */ | |
1165 | if (break_idx) | |
1166 | return; /* culling incompatible with break detection */ | |
1167 | ||
1168 | /* | |
1169 | * Note on reasons why we cull unneeded sources but not destinations: | |
1170 | * 1) Pairings are stored in rename_dst (not rename_src), which we | |
1171 | * need to keep around. So, we just can't cull rename_dst even | |
1172 | * if we wanted to. But doing so wouldn't help because... | |
1173 | * | |
1174 | * 2) There is a matrix pairwise comparison that follows the | |
1175 | * "Performing inexact rename detection" progress message. | |
1176 | * Iterating over the destinations is done in the outer loop, | |
1177 | * hence we only iterate over each of those once and we can | |
1178 | * easily skip the outer loop early if the destination isn't | |
1179 | * relevant. That's only one check per destination path to | |
1180 | * skip. | |
1181 | * | |
1182 | * By contrast, the sources are iterated in the inner loop; if | |
1183 | * we check whether a source can be skipped, then we'll be | |
1184 | * checking it N separate times, once for each destination. | |
1185 | * We don't want to have to iterate over known-not-needed | |
1186 | * sources N times each, so avoid that by removing the sources | |
1187 | * from rename_src here. | |
1188 | */ | |
1189 | for (i = 0, new_num_src = 0; i < rename_src_nr; i++) { | |
1190 | struct diff_filespec *one = rename_src[i].p->one; | |
1191 | ||
1192 | /* | |
1193 | * renames are stored in rename_dst, so if a rename has | |
1194 | * already been detected using this source, we can just | |
1195 | * remove the source knowing rename_dst has its info. | |
1196 | */ | |
1197 | if (!detecting_copies && one->rename_used) | |
1198 | continue; | |
1199 | ||
1200 | /* If we don't care about the source path, skip it */ | |
1201 | if (interesting && !strintmap_contains(interesting, one->path)) | |
1202 | continue; | |
1203 | ||
1204 | if (new_num_src < i) | |
1205 | memcpy(&rename_src[new_num_src], &rename_src[i], | |
1206 | sizeof(struct diff_rename_src)); | |
1207 | new_num_src++; | |
1208 | } | |
1209 | ||
1210 | rename_src_nr = new_num_src; | |
1211 | } | |
1212 | ||
1213 | static void handle_early_known_dir_renames(struct dir_rename_info *info, | |
1214 | struct strintmap *relevant_sources, | |
1215 | struct strintmap *dirs_removed) | |
1216 | { | |
1217 | /* | |
1218 | * Directory renames are determined via an aggregate of all renames | |
1219 | * under them and using a "majority wins" rule. The fact that | |
1220 | * "majority wins", though, means we don't need all the renames | |
1221 | * under the given directory, we only need enough to ensure we have | |
1222 | * a majority. | |
1223 | */ | |
1224 | ||
1225 | int i, new_num_src; | |
1226 | struct hashmap_iter iter; | |
1227 | struct strmap_entry *entry; | |
1228 | ||
1229 | if (!dirs_removed || !relevant_sources) | |
1230 | return; /* nothing to cull */ | |
1231 | if (break_idx) | |
1232 | return; /* culling incompatbile with break detection */ | |
1233 | ||
1234 | /* | |
1235 | * Supplement dir_rename_count with number of potential renames, | |
1236 | * marking all potential rename sources as mapping to UNKNOWN_DIR. | |
1237 | */ | |
1238 | for (i = 0; i < rename_src_nr; i++) { | |
1239 | char *old_dir; | |
1240 | struct diff_filespec *one = rename_src[i].p->one; | |
1241 | ||
1242 | /* | |
1243 | * sources that are part of a rename will have already been | |
1244 | * removed by a prior call to remove_unneeded_paths_from_src() | |
1245 | */ | |
1246 | assert(!one->rename_used); | |
1247 | ||
1248 | old_dir = get_dirname(one->path); | |
1249 | while (*old_dir != '\0' && | |
1250 | NOT_RELEVANT != strintmap_get(dirs_removed, old_dir)) { | |
1251 | char *freeme = old_dir; | |
1252 | ||
1253 | increment_count(info, old_dir, UNKNOWN_DIR); | |
1254 | old_dir = get_dirname(old_dir); | |
1255 | ||
1256 | /* Free resources we don't need anymore */ | |
1257 | free(freeme); | |
1258 | } | |
1259 | /* | |
1260 | * old_dir and new_dir free'd in increment_count, but | |
1261 | * get_dirname() gives us a new pointer we need to free for | |
1262 | * old_dir. Also, if the loop runs 0 times we need old_dir | |
1263 | * to be freed. | |
1264 | */ | |
1265 | free(old_dir); | |
1266 | } | |
1267 | ||
1268 | /* | |
1269 | * For any directory which we need a potential rename detected for | |
1270 | * (i.e. those marked as RELEVANT_FOR_SELF in dirs_removed), check | |
1271 | * whether we have enough renames to satisfy the "majority rules" | |
1272 | * requirement such that detecting any more renames of files under | |
1273 | * it won't change the result. For any such directory, mark that | |
1274 | * we no longer need to detect a rename for it. However, since we | |
1275 | * might need to still detect renames for an ancestor of that | |
1276 | * directory, use RELEVANT_FOR_ANCESTOR. | |
1277 | */ | |
1278 | strmap_for_each_entry(info->dir_rename_count, &iter, entry) { | |
1279 | /* entry->key is source_dir */ | |
1280 | struct strintmap *counts = entry->value; | |
1281 | ||
1282 | if (strintmap_get(dirs_removed, entry->key) == | |
1283 | RELEVANT_FOR_SELF && | |
1284 | dir_rename_already_determinable(counts)) { | |
1285 | strintmap_set(dirs_removed, entry->key, | |
1286 | RELEVANT_FOR_ANCESTOR); | |
1287 | } | |
1288 | } | |
1289 | ||
1290 | for (i = 0, new_num_src = 0; i < rename_src_nr; i++) { | |
1291 | struct diff_filespec *one = rename_src[i].p->one; | |
1292 | int val; | |
1293 | ||
1294 | val = strintmap_get(relevant_sources, one->path); | |
1295 | ||
1296 | /* | |
1297 | * sources that were not found in relevant_sources should | |
1298 | * have already been removed by a prior call to | |
1299 | * remove_unneeded_paths_from_src() | |
1300 | */ | |
1301 | assert(val != -1); | |
1302 | ||
1303 | if (val == RELEVANT_LOCATION) { | |
1304 | int removable = 1; | |
1305 | char *dir = get_dirname(one->path); | |
1306 | while (1) { | |
1307 | char *freeme = dir; | |
1308 | int res = strintmap_get(dirs_removed, dir); | |
1309 | ||
1310 | /* Quit if not found or irrelevant */ | |
1311 | if (res == NOT_RELEVANT) | |
1312 | break; | |
1313 | /* If RELEVANT_FOR_SELF, can't remove */ | |
1314 | if (res == RELEVANT_FOR_SELF) { | |
1315 | removable = 0; | |
1316 | break; | |
1317 | } | |
1318 | /* Else continue searching upwards */ | |
1319 | assert(res == RELEVANT_FOR_ANCESTOR); | |
1320 | dir = get_dirname(dir); | |
1321 | free(freeme); | |
1322 | } | |
1323 | free(dir); | |
1324 | if (removable) { | |
1325 | strintmap_set(relevant_sources, one->path, | |
1326 | RELEVANT_NO_MORE); | |
1327 | continue; | |
1328 | } | |
1329 | } | |
1330 | ||
1331 | if (new_num_src < i) | |
1332 | memcpy(&rename_src[new_num_src], &rename_src[i], | |
1333 | sizeof(struct diff_rename_src)); | |
1334 | new_num_src++; | |
1335 | } | |
1336 | ||
1337 | rename_src_nr = new_num_src; | |
1338 | } | |
1339 | ||
1340 | static void free_filespec_data(struct diff_filespec *spec) | |
1341 | { | |
1342 | if (!--spec->count) | |
1343 | diff_free_filespec_data(spec); | |
1344 | } | |
1345 | ||
1346 | static void pool_free_filespec(struct mem_pool *pool, | |
1347 | struct diff_filespec *spec) | |
1348 | { | |
1349 | if (!pool) { | |
1350 | free_filespec(spec); | |
1351 | return; | |
1352 | } | |
1353 | ||
1354 | /* | |
1355 | * Similar to free_filespec(), but only frees the data. The spec | |
1356 | * itself was allocated in the pool and should not be individually | |
1357 | * freed. | |
1358 | */ | |
1359 | free_filespec_data(spec); | |
1360 | } | |
1361 | ||
1362 | void pool_diff_free_filepair(struct mem_pool *pool, | |
1363 | struct diff_filepair *p) | |
1364 | { | |
1365 | if (!pool) { | |
1366 | diff_free_filepair(p); | |
1367 | return; | |
1368 | } | |
1369 | ||
1370 | /* | |
1371 | * Similar to diff_free_filepair() but only frees the data from the | |
1372 | * filespecs; not the filespecs or the filepair which were | |
1373 | * allocated from the pool. | |
1374 | */ | |
1375 | free_filespec_data(p->one); | |
1376 | free_filespec_data(p->two); | |
1377 | } | |
1378 | ||
1379 | void diffcore_rename_extended(struct diff_options *options, | |
1380 | struct mem_pool *pool, | |
1381 | struct strintmap *relevant_sources, | |
1382 | struct strintmap *dirs_removed, | |
1383 | struct strmap *dir_rename_count, | |
1384 | struct strmap *cached_pairs) | |
1385 | { | |
1386 | int detect_rename = options->detect_rename; | |
1387 | int minimum_score = options->rename_score; | |
1388 | struct diff_queue_struct *q = &diff_queued_diff; | |
1389 | struct diff_queue_struct outq; | |
1390 | struct diff_score *mx; | |
1391 | int i, j, rename_count, skip_unmodified = 0; | |
1392 | int num_destinations, dst_cnt; | |
1393 | int num_sources, want_copies; | |
1394 | struct progress *progress = NULL; | |
1395 | struct mem_pool local_pool; | |
1396 | struct dir_rename_info info; | |
1397 | struct diff_populate_filespec_options dpf_options = { | |
1398 | .check_binary = 0, | |
1399 | .missing_object_cb = NULL, | |
1400 | .missing_object_data = NULL | |
1401 | }; | |
1402 | struct inexact_prefetch_options prefetch_options = { | |
1403 | .repo = options->repo | |
1404 | }; | |
1405 | ||
1406 | trace2_region_enter("diff", "setup", options->repo); | |
1407 | info.setup = 0; | |
1408 | assert(!dir_rename_count || strmap_empty(dir_rename_count)); | |
1409 | want_copies = (detect_rename == DIFF_DETECT_COPY); | |
1410 | if (dirs_removed && (break_idx || want_copies)) | |
1411 | BUG("dirs_removed incompatible with break/copy detection"); | |
1412 | if (break_idx && relevant_sources) | |
1413 | BUG("break detection incompatible with source specification"); | |
1414 | if (!minimum_score) | |
1415 | minimum_score = DEFAULT_RENAME_SCORE; | |
1416 | ||
1417 | for (i = 0; i < q->nr; i++) { | |
1418 | struct diff_filepair *p = q->queue[i]; | |
1419 | if (!DIFF_FILE_VALID(p->one)) { | |
1420 | if (!DIFF_FILE_VALID(p->two)) | |
1421 | continue; /* unmerged */ | |
1422 | else if (options->single_follow && | |
1423 | strcmp(options->single_follow, p->two->path)) | |
1424 | continue; /* not interested */ | |
1425 | else if (!options->flags.rename_empty && | |
1426 | is_empty_blob_oid(&p->two->oid)) | |
1427 | continue; | |
1428 | else if (add_rename_dst(p) < 0) { | |
1429 | warning("skipping rename detection, detected" | |
1430 | " duplicate destination '%s'", | |
1431 | p->two->path); | |
1432 | goto cleanup; | |
1433 | } | |
1434 | } | |
1435 | else if (!options->flags.rename_empty && | |
1436 | is_empty_blob_oid(&p->one->oid)) | |
1437 | continue; | |
1438 | else if (!DIFF_PAIR_UNMERGED(p) && !DIFF_FILE_VALID(p->two)) { | |
1439 | /* | |
1440 | * If the source is a broken "delete", and | |
1441 | * they did not really want to get broken, | |
1442 | * that means the source actually stays. | |
1443 | * So we increment the "rename_used" score | |
1444 | * by one, to indicate ourselves as a user | |
1445 | */ | |
1446 | if (p->broken_pair && !p->score) | |
1447 | p->one->rename_used++; | |
1448 | register_rename_src(p); | |
1449 | } | |
1450 | else if (want_copies) { | |
1451 | /* | |
1452 | * Increment the "rename_used" score by | |
1453 | * one, to indicate ourselves as a user. | |
1454 | */ | |
1455 | p->one->rename_used++; | |
1456 | register_rename_src(p); | |
1457 | } | |
1458 | } | |
1459 | trace2_region_leave("diff", "setup", options->repo); | |
1460 | if (rename_dst_nr == 0 || rename_src_nr == 0) | |
1461 | goto cleanup; /* nothing to do */ | |
1462 | ||
1463 | trace2_region_enter("diff", "exact renames", options->repo); | |
1464 | mem_pool_init(&local_pool, 32*1024); | |
1465 | /* | |
1466 | * We really want to cull the candidates list early | |
1467 | * with cheap tests in order to avoid doing deltas. | |
1468 | */ | |
1469 | rename_count = find_exact_renames(options, &local_pool); | |
1470 | /* | |
1471 | * Discard local_pool immediately instead of at "cleanup:" in order | |
1472 | * to reduce maximum memory usage; inexact rename detection uses up | |
1473 | * a fair amount of memory, and mem_pools can too. | |
1474 | */ | |
1475 | mem_pool_discard(&local_pool, 0); | |
1476 | trace2_region_leave("diff", "exact renames", options->repo); | |
1477 | ||
1478 | /* Did we only want exact renames? */ | |
1479 | if (minimum_score == MAX_SCORE) | |
1480 | goto cleanup; | |
1481 | ||
1482 | num_sources = rename_src_nr; | |
1483 | ||
1484 | if (want_copies || break_idx) { | |
1485 | /* | |
1486 | * Cull sources: | |
1487 | * - remove ones corresponding to exact renames | |
1488 | * - remove ones not found in relevant_sources | |
1489 | */ | |
1490 | trace2_region_enter("diff", "cull after exact", options->repo); | |
1491 | remove_unneeded_paths_from_src(want_copies, relevant_sources); | |
1492 | trace2_region_leave("diff", "cull after exact", options->repo); | |
1493 | } else { | |
1494 | /* Determine minimum score to match basenames */ | |
1495 | double factor = 0.5; | |
1496 | char *basename_factor = getenv("GIT_BASENAME_FACTOR"); | |
1497 | int min_basename_score; | |
1498 | ||
1499 | if (basename_factor) | |
1500 | factor = strtol(basename_factor, NULL, 10)/100.0; | |
1501 | assert(factor >= 0.0 && factor <= 1.0); | |
1502 | min_basename_score = minimum_score + | |
1503 | (int)(factor * (MAX_SCORE - minimum_score)); | |
1504 | ||
1505 | /* | |
1506 | * Cull sources: | |
1507 | * - remove ones involved in renames (found via exact match) | |
1508 | */ | |
1509 | trace2_region_enter("diff", "cull after exact", options->repo); | |
1510 | remove_unneeded_paths_from_src(want_copies, NULL); | |
1511 | trace2_region_leave("diff", "cull after exact", options->repo); | |
1512 | ||
1513 | /* Preparation for basename-driven matching. */ | |
1514 | trace2_region_enter("diff", "dir rename setup", options->repo); | |
1515 | initialize_dir_rename_info(&info, relevant_sources, | |
1516 | dirs_removed, dir_rename_count, | |
1517 | cached_pairs); | |
1518 | trace2_region_leave("diff", "dir rename setup", options->repo); | |
1519 | ||
1520 | /* Utilize file basenames to quickly find renames. */ | |
1521 | trace2_region_enter("diff", "basename matches", options->repo); | |
1522 | rename_count += find_basename_matches(options, | |
1523 | min_basename_score, | |
1524 | &info, | |
1525 | relevant_sources, | |
1526 | dirs_removed); | |
1527 | trace2_region_leave("diff", "basename matches", options->repo); | |
1528 | ||
1529 | /* | |
1530 | * Cull sources, again: | |
1531 | * - remove ones involved in renames (found via basenames) | |
1532 | * - remove ones not found in relevant_sources | |
1533 | * and | |
1534 | * - remove ones in relevant_sources which are needed only | |
1535 | * for directory renames IF no ancestory directory | |
1536 | * actually needs to know any more individual path | |
1537 | * renames under them | |
1538 | */ | |
1539 | trace2_region_enter("diff", "cull basename", options->repo); | |
1540 | remove_unneeded_paths_from_src(want_copies, relevant_sources); | |
1541 | handle_early_known_dir_renames(&info, relevant_sources, | |
1542 | dirs_removed); | |
1543 | trace2_region_leave("diff", "cull basename", options->repo); | |
1544 | } | |
1545 | ||
1546 | /* Calculate how many rename destinations are left */ | |
1547 | num_destinations = (rename_dst_nr - rename_count); | |
1548 | num_sources = rename_src_nr; /* rename_src_nr reflects lower number */ | |
1549 | ||
1550 | /* All done? */ | |
1551 | if (!num_destinations || !num_sources) | |
1552 | goto cleanup; | |
1553 | ||
1554 | switch (too_many_rename_candidates(num_destinations, num_sources, | |
1555 | options)) { | |
1556 | case 1: | |
1557 | goto cleanup; | |
1558 | case 2: | |
1559 | options->degraded_cc_to_c = 1; | |
1560 | skip_unmodified = 1; | |
1561 | break; | |
1562 | default: | |
1563 | break; | |
1564 | } | |
1565 | ||
1566 | trace2_region_enter("diff", "inexact renames", options->repo); | |
1567 | if (options->show_rename_progress) { | |
1568 | progress = start_delayed_progress( | |
1569 | _("Performing inexact rename detection"), | |
1570 | (uint64_t)num_destinations * (uint64_t)num_sources); | |
1571 | } | |
1572 | ||
1573 | /* Finish setting up dpf_options */ | |
1574 | prefetch_options.skip_unmodified = skip_unmodified; | |
1575 | if (options->repo == the_repository && repo_has_promisor_remote(the_repository)) { | |
1576 | dpf_options.missing_object_cb = inexact_prefetch; | |
1577 | dpf_options.missing_object_data = &prefetch_options; | |
1578 | } | |
1579 | ||
1580 | CALLOC_ARRAY(mx, st_mult(NUM_CANDIDATE_PER_DST, num_destinations)); | |
1581 | for (dst_cnt = i = 0; i < rename_dst_nr; i++) { | |
1582 | struct diff_filespec *two = rename_dst[i].p->two; | |
1583 | struct diff_score *m; | |
1584 | ||
1585 | if (rename_dst[i].is_rename) | |
1586 | continue; /* exact or basename match already handled */ | |
1587 | ||
1588 | m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST]; | |
1589 | for (j = 0; j < NUM_CANDIDATE_PER_DST; j++) | |
1590 | m[j].dst = -1; | |
1591 | ||
1592 | for (j = 0; j < rename_src_nr; j++) { | |
1593 | struct diff_filespec *one = rename_src[j].p->one; | |
1594 | struct diff_score this_src; | |
1595 | ||
1596 | assert(!one->rename_used || want_copies || break_idx); | |
1597 | ||
1598 | if (skip_unmodified && | |
1599 | diff_unmodified_pair(rename_src[j].p)) | |
1600 | continue; | |
1601 | ||
1602 | this_src.score = estimate_similarity(options->repo, | |
1603 | one, two, | |
1604 | minimum_score, | |
1605 | &dpf_options); | |
1606 | this_src.name_score = basename_same(one, two); | |
1607 | this_src.dst = i; | |
1608 | this_src.src = j; | |
1609 | record_if_better(m, &this_src); | |
1610 | /* | |
1611 | * Once we run estimate_similarity, | |
1612 | * We do not need the text anymore. | |
1613 | */ | |
1614 | diff_free_filespec_blob(one); | |
1615 | diff_free_filespec_blob(two); | |
1616 | } | |
1617 | dst_cnt++; | |
1618 | display_progress(progress, | |
1619 | (uint64_t)dst_cnt * (uint64_t)num_sources); | |
1620 | } | |
1621 | stop_progress(&progress); | |
1622 | ||
1623 | /* cost matrix sorted by most to least similar pair */ | |
1624 | STABLE_QSORT(mx, dst_cnt * NUM_CANDIDATE_PER_DST, score_compare); | |
1625 | ||
1626 | rename_count += find_renames(mx, dst_cnt, minimum_score, 0, | |
1627 | &info, dirs_removed); | |
1628 | if (want_copies) | |
1629 | rename_count += find_renames(mx, dst_cnt, minimum_score, 1, | |
1630 | &info, dirs_removed); | |
1631 | free(mx); | |
1632 | trace2_region_leave("diff", "inexact renames", options->repo); | |
1633 | ||
1634 | cleanup: | |
1635 | /* At this point, we have found some renames and copies and they | |
1636 | * are recorded in rename_dst. The original list is still in *q. | |
1637 | */ | |
1638 | trace2_region_enter("diff", "write back to queue", options->repo); | |
1639 | DIFF_QUEUE_CLEAR(&outq); | |
1640 | for (i = 0; i < q->nr; i++) { | |
1641 | struct diff_filepair *p = q->queue[i]; | |
1642 | struct diff_filepair *pair_to_free = NULL; | |
1643 | ||
1644 | if (DIFF_PAIR_UNMERGED(p)) { | |
1645 | diff_q(&outq, p); | |
1646 | } | |
1647 | else if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) { | |
1648 | /* Creation */ | |
1649 | diff_q(&outq, p); | |
1650 | } | |
1651 | else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) { | |
1652 | /* | |
1653 | * Deletion | |
1654 | * | |
1655 | * We would output this delete record if: | |
1656 | * | |
1657 | * (1) this is a broken delete and the counterpart | |
1658 | * broken create remains in the output; or | |
1659 | * (2) this is not a broken delete, and rename_dst | |
1660 | * does not have a rename/copy to move p->one->path | |
1661 | * out of existence. | |
1662 | * | |
1663 | * Otherwise, the counterpart broken create | |
1664 | * has been turned into a rename-edit; or | |
1665 | * delete did not have a matching create to | |
1666 | * begin with. | |
1667 | */ | |
1668 | if (DIFF_PAIR_BROKEN(p)) { | |
1669 | /* broken delete */ | |
1670 | struct diff_rename_dst *dst = locate_rename_dst(p); | |
1671 | if (!dst) | |
1672 | BUG("tracking failed somehow; failed to find associated dst for broken pair"); | |
1673 | if (dst->is_rename) | |
1674 | /* counterpart is now rename/copy */ | |
1675 | pair_to_free = p; | |
1676 | } | |
1677 | else { | |
1678 | if (p->one->rename_used) | |
1679 | /* this path remains */ | |
1680 | pair_to_free = p; | |
1681 | } | |
1682 | ||
1683 | if (!pair_to_free) | |
1684 | diff_q(&outq, p); | |
1685 | } | |
1686 | else if (!diff_unmodified_pair(p)) | |
1687 | /* all the usual ones need to be kept */ | |
1688 | diff_q(&outq, p); | |
1689 | else | |
1690 | /* no need to keep unmodified pairs */ | |
1691 | pair_to_free = p; | |
1692 | ||
1693 | if (pair_to_free) | |
1694 | pool_diff_free_filepair(pool, pair_to_free); | |
1695 | } | |
1696 | diff_debug_queue("done copying original", &outq); | |
1697 | ||
1698 | free(q->queue); | |
1699 | *q = outq; | |
1700 | diff_debug_queue("done collapsing", q); | |
1701 | ||
1702 | for (i = 0; i < rename_dst_nr; i++) | |
1703 | if (rename_dst[i].filespec_to_free) | |
1704 | pool_free_filespec(pool, rename_dst[i].filespec_to_free); | |
1705 | ||
1706 | cleanup_dir_rename_info(&info, dirs_removed, dir_rename_count != NULL); | |
1707 | FREE_AND_NULL(rename_dst); | |
1708 | rename_dst_nr = rename_dst_alloc = 0; | |
1709 | FREE_AND_NULL(rename_src); | |
1710 | rename_src_nr = rename_src_alloc = 0; | |
1711 | if (break_idx) { | |
1712 | strintmap_clear(break_idx); | |
1713 | FREE_AND_NULL(break_idx); | |
1714 | } | |
1715 | trace2_region_leave("diff", "write back to queue", options->repo); | |
1716 | return; | |
1717 | } | |
1718 | ||
1719 | void diffcore_rename(struct diff_options *options) | |
1720 | { | |
1721 | diffcore_rename_extended(options, NULL, NULL, NULL, NULL, NULL); | |
1722 | } |