]>
Commit | Line | Data |
---|---|---|
427dcb4b | 1 | /* |
95acf11a | 2 | * |
427dcb4b JH |
3 | * Copyright (C) 2005 Junio C Hamano |
4 | */ | |
5 | #include "cache.h" | |
6 | #include "diff.h" | |
7 | #include "diffcore.h" | |
cbd53a21 | 8 | #include "object-store.h" |
f79d9c58 | 9 | #include "hashmap.h" |
3ac942d4 | 10 | #include "progress.h" |
95acf11a | 11 | #include "promisor-remote.h" |
9db2ac56 | 12 | #include "strmap.h" |
427dcb4b | 13 | |
25d5ea41 JH |
14 | /* Table of rename/copy destinations */ |
15 | ||
16 | static struct diff_rename_dst { | |
9db2ac56 EN |
17 | struct diff_filepair *p; |
18 | struct diff_filespec *filespec_to_free; | |
19 | int is_rename; /* false -> just a create; true -> rename or copy */ | |
25d5ea41 JH |
20 | } *rename_dst; |
21 | static int rename_dst_nr, rename_dst_alloc; | |
9db2ac56 EN |
22 | /* Mapping from break source pathname to break destination index */ |
23 | static struct strintmap *break_idx = NULL; | |
427dcb4b | 24 | |
9db2ac56 | 25 | static struct diff_rename_dst *locate_rename_dst(struct diff_filepair *p) |
f98c2f7e | 26 | { |
9db2ac56 EN |
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]; | |
f98c2f7e JK |
30 | } |
31 | ||
32 | /* | |
33 | * Returns 0 on success, -1 if we found a duplicate. | |
34 | */ | |
9db2ac56 | 35 | static int add_rename_dst(struct diff_filepair *p) |
f98c2f7e | 36 | { |
337ce247 | 37 | ALLOC_GROW(rename_dst, rename_dst_nr + 1, rename_dst_alloc); |
9db2ac56 EN |
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; | |
25d5ea41 | 41 | rename_dst_nr++; |
f98c2f7e | 42 | return 0; |
427dcb4b JH |
43 | } |
44 | ||
15d061b4 | 45 | /* Table of rename/copy src files */ |
25d5ea41 | 46 | static struct diff_rename_src { |
e88d6bc6 | 47 | struct diff_filepair *p; |
fc580719 | 48 | unsigned short score; /* to remember the break score */ |
25d5ea41 JH |
49 | } *rename_src; |
50 | static int rename_src_nr, rename_src_alloc; | |
427dcb4b | 51 | |
b970b4ef | 52 | static void register_rename_src(struct diff_filepair *p) |
25d5ea41 | 53 | { |
9db2ac56 EN |
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 | ||
337ce247 | 62 | ALLOC_GROW(rename_src, rename_src_nr + 1, rename_src_alloc); |
b970b4ef EN |
63 | rename_src[rename_src_nr].p = p; |
64 | rename_src[rename_src_nr].score = p->score; | |
25d5ea41 | 65 | rename_src_nr++; |
427dcb4b JH |
66 | } |
67 | ||
0ce39643 JS |
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 | ||
427dcb4b | 83 | struct diff_score { |
25d5ea41 JH |
84 | int src; /* index in rename_src */ |
85 | int dst; /* index in rename_dst */ | |
6d24ad97 JH |
86 | unsigned short score; |
87 | short name_score; | |
427dcb4b JH |
88 | }; |
89 | ||
95acf11a JT |
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++) { | |
9db2ac56 | 101 | if (rename_dst[i].p->renamed_pair) |
95acf11a JT |
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, | |
9db2ac56 | 108 | rename_dst[i].p->two); |
95acf11a JT |
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 | ||
b78ea5fc NTND |
125 | static int estimate_similarity(struct repository *r, |
126 | struct diff_filespec *src, | |
427dcb4b | 127 | struct diff_filespec *dst, |
95acf11a JT |
128 | int minimum_score, |
129 | int skip_unmodified) | |
427dcb4b JH |
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 | |
f0c6b2a2 JH |
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. | |
427dcb4b | 143 | */ |
90bd932c | 144 | unsigned long max_size, delta_size, base_size, src_copied, literal_added; |
427dcb4b | 145 | int score; |
1c37e86a JT |
146 | struct diff_populate_filespec_options dpf_options = { |
147 | .check_size_only = 1 | |
148 | }; | |
95acf11a JT |
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 | } | |
427dcb4b | 155 | |
60896c7b JH |
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 | ||
81ac051d LT |
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 | */ | |
8e5dd3d6 | 172 | if (!src->cnt_data && |
1c37e86a | 173 | diff_populate_filespec(r, src, &dpf_options)) |
81ac051d | 174 | return 0; |
8e5dd3d6 | 175 | if (!dst->cnt_data && |
1c37e86a | 176 | diff_populate_filespec(r, dst, &dpf_options)) |
81ac051d LT |
177 | return 0; |
178 | ||
90bd932c | 179 | max_size = ((src->size > dst->size) ? src->size : dst->size); |
58b103f5 | 180 | base_size = ((src->size < dst->size) ? src->size : dst->size); |
90bd932c | 181 | delta_size = max_size - base_size; |
427dcb4b | 182 | |
58b103f5 JH |
183 | /* We would not consider edits that change the file size so |
184 | * drastically. delta_size must be smaller than | |
cd1870ed | 185 | * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size). |
f0c6b2a2 | 186 | * |
58b103f5 JH |
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. | |
427dcb4b | 190 | */ |
3a4d6769 | 191 | if (max_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE) |
427dcb4b JH |
192 | return 0; |
193 | ||
95acf11a JT |
194 | dpf_options.check_size_only = 0; |
195 | ||
196 | if (!src->cnt_data && diff_populate_filespec(r, src, &dpf_options)) | |
885c716f | 197 | return 0; |
95acf11a | 198 | if (!dst->cnt_data && diff_populate_filespec(r, dst, &dpf_options)) |
885c716f BS |
199 | return 0; |
200 | ||
b78ea5fc | 201 | if (diffcore_count_changes(r, src, dst, |
c06c7966 | 202 | &src->cnt_data, &dst->cnt_data, |
65416758 | 203 | &src_copied, &literal_added)) |
85976974 | 204 | return 0; |
355e76a4 | 205 | |
1706306a JH |
206 | /* How similar are they? |
207 | * what percentage of material in dst are from source? | |
427dcb4b | 208 | */ |
90bd932c | 209 | if (!dst->size) |
1706306a | 210 | score = 0; /* should not happen */ |
cfc0aef1 | 211 | else |
dc49cd76 | 212 | score = (int)(src_copied * MAX_SCORE / max_size); |
427dcb4b JH |
213 | return score; |
214 | } | |
215 | ||
5098bafb | 216 | static void record_rename_pair(int dst_index, int src_index, int score) |
427dcb4b | 217 | { |
9db2ac56 EN |
218 | struct diff_filepair *src = rename_src[src_index].p; |
219 | struct diff_filepair *dst = rename_dst[dst_index].p; | |
f7c1512a | 220 | |
9db2ac56 | 221 | if (dst->renamed_pair) |
25d5ea41 | 222 | die("internal error: dst already matched."); |
427dcb4b | 223 | |
9db2ac56 EN |
224 | src->one->rename_used++; |
225 | src->one->count++; | |
427dcb4b | 226 | |
9db2ac56 EN |
227 | rename_dst[dst_index].filespec_to_free = dst->one; |
228 | rename_dst[dst_index].is_rename = 1; | |
427dcb4b | 229 | |
9db2ac56 EN |
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; | |
fc580719 | 234 | else |
9db2ac56 | 235 | dst->score = score; |
427dcb4b JH |
236 | } |
237 | ||
238 | /* | |
239 | * We sort the rename similarity matrix with the score, in descending | |
15d061b4 | 240 | * order (the most similar first). |
427dcb4b JH |
241 | */ |
242 | static int score_compare(const void *a_, const void *b_) | |
243 | { | |
244 | const struct diff_score *a = a_, *b = b_; | |
cfc0aef1 | 245 | |
6d24ad97 JH |
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 | ||
cfc0aef1 RS |
252 | if (a->score == b->score) |
253 | return b->name_score - a->name_score; | |
254 | ||
427dcb4b JH |
255 | return b->score - a->score; |
256 | } | |
257 | ||
9027f53c | 258 | struct file_similarity { |
f79d9c58 | 259 | struct hashmap_entry entry; |
7c85f8ac | 260 | int index; |
9027f53c | 261 | struct diff_filespec *filespec; |
9027f53c LT |
262 | }; |
263 | ||
b78ea5fc NTND |
264 | static unsigned int hash_filespec(struct repository *r, |
265 | struct diff_filespec *filespec) | |
48f6407f | 266 | { |
41c9560e | 267 | if (!filespec->oid_valid) { |
1c37e86a | 268 | if (diff_populate_filespec(r, filespec, NULL)) |
48f6407f | 269 | return 0; |
2dcde20e MT |
270 | hash_object_file(r->hash_algo, filespec->data, filespec->size, |
271 | "blob", &filespec->oid); | |
48f6407f | 272 | } |
d40abc8e | 273 | return oidhash(&filespec->oid); |
48f6407f KB |
274 | } |
275 | ||
f79d9c58 | 276 | static int find_identical_files(struct hashmap *srcs, |
7c85f8ac | 277 | int dst_index, |
11f944dd | 278 | struct diff_options *options) |
9027f53c LT |
279 | { |
280 | int renames = 0; | |
9db2ac56 | 281 | struct diff_filespec *target = rename_dst[dst_index].p->two; |
ab73a9d1 | 282 | struct file_similarity *p, *best = NULL; |
48f6407f | 283 | int i = 100, best_score = -1; |
f0e63c41 | 284 | unsigned int hash = hash_filespec(options->repo, target); |
48f6407f KB |
285 | |
286 | /* | |
7c85f8ac | 287 | * Find the best source match for specified destination. |
48f6407f | 288 | */ |
f0e63c41 EW |
289 | p = hashmap_get_entry_from_hash(srcs, hash, NULL, |
290 | struct file_similarity, entry); | |
23dee69f | 291 | hashmap_for_each_entry_from(srcs, p, entry) { |
48f6407f KB |
292 | int score; |
293 | struct diff_filespec *source = p->filespec; | |
294 | ||
295 | /* False hash collision? */ | |
9001dc2a | 296 | if (!oideq(&source->oid, &target->oid)) |
48f6407f KB |
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) | |
0940e5f2 | 301 | continue; |
9027f53c | 302 | } |
48f6407f KB |
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; | |
9027f53c | 313 | } |
48f6407f KB |
314 | |
315 | /* Too many identical alternatives? Pick one */ | |
316 | if (!--i) | |
317 | break; | |
318 | } | |
319 | if (best) { | |
7c85f8ac | 320 | record_rename_pair(dst_index, best->index, MAX_SCORE); |
48f6407f KB |
321 | renames++; |
322 | } | |
9027f53c LT |
323 | return renames; |
324 | } | |
325 | ||
b78ea5fc NTND |
326 | static void insert_file_table(struct repository *r, |
327 | struct hashmap *table, int index, | |
328 | struct diff_filespec *filespec) | |
9027f53c | 329 | { |
9027f53c LT |
330 | struct file_similarity *entry = xmalloc(sizeof(*entry)); |
331 | ||
9027f53c LT |
332 | entry->index = index; |
333 | entry->filespec = filespec; | |
9027f53c | 334 | |
d22245a2 | 335 | hashmap_entry_init(&entry->entry, hash_filespec(r, filespec)); |
b94e5c1d | 336 | hashmap_add(table, &entry->entry); |
9027f53c LT |
337 | } |
338 | ||
cb1491b6 LT |
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. | |
cb1491b6 | 345 | */ |
11f944dd | 346 | static int find_exact_renames(struct diff_options *options) |
cb1491b6 | 347 | { |
7c85f8ac | 348 | int i, renames = 0; |
f79d9c58 | 349 | struct hashmap file_table; |
cb1491b6 | 350 | |
ca4e3ca0 SG |
351 | /* Add all sources to the hash table in reverse order, because |
352 | * later on they will be retrieved in LIFO order. | |
353 | */ | |
7663cdc8 | 354 | hashmap_init(&file_table, NULL, NULL, rename_src_nr); |
ca4e3ca0 | 355 | for (i = rename_src_nr-1; i >= 0; i--) |
b78ea5fc NTND |
356 | insert_file_table(options->repo, |
357 | &file_table, i, | |
358 | rename_src[i].p->one); | |
9027f53c | 359 | |
7c85f8ac | 360 | /* Walk the destinations and find best source match */ |
9027f53c | 361 | for (i = 0; i < rename_dst_nr; i++) |
7c85f8ac | 362 | renames += find_identical_files(&file_table, i, options); |
9027f53c | 363 | |
f79d9c58 | 364 | /* Free the hash data structure and entries */ |
6da1a258 | 365 | hashmap_clear_and_free(&file_table, struct file_similarity, entry); |
9027f53c | 366 | |
7c85f8ac | 367 | return renames; |
cb1491b6 LT |
368 | } |
369 | ||
bde8b9f3 EN |
370 | struct dir_rename_info { |
371 | struct strintmap idx_map; | |
372 | struct strmap dir_rename_guess; | |
373 | struct strmap *dir_rename_count; | |
333899e1 | 374 | struct strset *relevant_source_dirs; |
bde8b9f3 EN |
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 | ||
0c4fd732 EN |
384 | static void dirname_munge(char *filename) |
385 | { | |
386 | char *slash = strrchr(filename, '/'); | |
387 | if (!slash) | |
388 | slash = filename; | |
389 | *slash = '\0'; | |
390 | } | |
391 | ||
81afdf7a EN |
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 | ||
b6e3d274 | 410 | static void increment_count(struct dir_rename_info *info, |
0c4fd732 EN |
411 | char *old_dir, |
412 | char *new_dir) | |
413 | { | |
414 | struct strintmap *counts; | |
415 | struct strmap_entry *e; | |
416 | ||
417 | /* Get the {new_dirs -> counts} mapping using old_dir */ | |
b6e3d274 | 418 | e = strmap_get_entry(info->dir_rename_count, old_dir); |
0c4fd732 EN |
419 | if (e) { |
420 | counts = e->value; | |
421 | } else { | |
422 | counts = xmalloc(sizeof(*counts)); | |
423 | strintmap_init_with_options(counts, 0, NULL, 1); | |
b6e3d274 | 424 | strmap_put(info->dir_rename_count, old_dir, counts); |
0c4fd732 EN |
425 | } |
426 | ||
427 | /* Increment the count for new_dir */ | |
428 | strintmap_incr(counts, new_dir, 1); | |
429 | } | |
430 | ||
b6e3d274 | 431 | static void update_dir_rename_counts(struct dir_rename_info *info, |
0c4fd732 EN |
432 | struct strset *dirs_removed, |
433 | const char *oldname, | |
434 | const char *newname) | |
435 | { | |
436 | char *old_dir = xstrdup(oldname); | |
437 | char *new_dir = xstrdup(newname); | |
438 | char new_dir_first_char = new_dir[0]; | |
439 | int first_time_in_loop = 1; | |
440 | ||
1ad69eb0 EN |
441 | if (!info->setup) |
442 | /* | |
443 | * info->setup is 0 here in two cases: (1) all auxiliary | |
444 | * vars (like dirs_removed) were NULL so | |
445 | * initialize_dir_rename_info() returned early, or (2) | |
446 | * either break detection or copy detection are active so | |
447 | * that we never called initialize_dir_rename_info(). In | |
448 | * the former case, we don't have enough info to know if | |
449 | * directories were renamed (because dirs_removed lets us | |
450 | * know about a necessary prerequisite, namely if they were | |
451 | * removed), and in the latter, we don't care about | |
452 | * directory renames or find_basename_matches. | |
453 | * | |
454 | * This matters because both basename and inexact matching | |
455 | * will also call update_dir_rename_counts(). In either of | |
456 | * the above two cases info->dir_rename_counts will not | |
457 | * have been properly initialized which prevents us from | |
458 | * updating it, but in these two cases we don't care about | |
459 | * dir_rename_counts anyway, so we can just exit early. | |
460 | */ | |
461 | return; | |
462 | ||
0c4fd732 | 463 | while (1) { |
333899e1 | 464 | /* Get old_dir, skip if its directory isn't relevant. */ |
0c4fd732 | 465 | dirname_munge(old_dir); |
333899e1 EN |
466 | if (info->relevant_source_dirs && |
467 | !strset_contains(info->relevant_source_dirs, old_dir)) | |
468 | break; | |
469 | ||
470 | /* Get new_dir */ | |
0c4fd732 EN |
471 | dirname_munge(new_dir); |
472 | ||
473 | /* | |
474 | * When renaming | |
475 | * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c" | |
476 | * then this suggests that both | |
477 | * a/b/c/d/e/ => a/b/some/thing/else/e/ | |
478 | * a/b/c/d/ => a/b/some/thing/else/ | |
479 | * so we want to increment counters for both. We do NOT, | |
480 | * however, also want to suggest that there was the following | |
481 | * rename: | |
482 | * a/b/c/ => a/b/some/thing/ | |
483 | * so we need to quit at that point. | |
484 | * | |
485 | * Note the when first_time_in_loop, we only strip off the | |
486 | * basename, and we don't care if that's different. | |
487 | */ | |
488 | if (!first_time_in_loop) { | |
489 | char *old_sub_dir = strchr(old_dir, '\0')+1; | |
490 | char *new_sub_dir = strchr(new_dir, '\0')+1; | |
491 | if (!*new_dir) { | |
492 | /* | |
493 | * Special case when renaming to root directory, | |
494 | * i.e. when new_dir == "". In this case, we had | |
495 | * something like | |
496 | * a/b/subdir => subdir | |
497 | * and so dirname_munge() sets things up so that | |
498 | * old_dir = "a/b\0subdir\0" | |
499 | * new_dir = "\0ubdir\0" | |
500 | * We didn't have a '/' to overwrite a '\0' onto | |
501 | * in new_dir, so we have to compare differently. | |
502 | */ | |
503 | if (new_dir_first_char != old_sub_dir[0] || | |
504 | strcmp(old_sub_dir+1, new_sub_dir)) | |
505 | break; | |
506 | } else { | |
507 | if (strcmp(old_sub_dir, new_sub_dir)) | |
508 | break; | |
509 | } | |
510 | } | |
511 | ||
512 | if (strset_contains(dirs_removed, old_dir)) | |
b6e3d274 | 513 | increment_count(info, old_dir, new_dir); |
0c4fd732 EN |
514 | else |
515 | break; | |
516 | ||
517 | /* If we hit toplevel directory ("") for old or new dir, quit */ | |
518 | if (!*old_dir || !*new_dir) | |
519 | break; | |
520 | ||
521 | first_time_in_loop = 0; | |
522 | } | |
523 | ||
524 | /* Free resources we don't need anymore */ | |
525 | free(old_dir); | |
526 | free(new_dir); | |
527 | } | |
528 | ||
1ad69eb0 | 529 | static void initialize_dir_rename_info(struct dir_rename_info *info, |
e4fd06e7 | 530 | struct strset *relevant_sources, |
1ad69eb0 EN |
531 | struct strset *dirs_removed, |
532 | struct strmap *dir_rename_count) | |
0c4fd732 | 533 | { |
81afdf7a EN |
534 | struct hashmap_iter iter; |
535 | struct strmap_entry *entry; | |
0c4fd732 EN |
536 | int i; |
537 | ||
e4fd06e7 | 538 | if (!dirs_removed && !relevant_sources) { |
1ad69eb0 EN |
539 | info->setup = 0; |
540 | return; | |
0c4fd732 | 541 | } |
ae8cf74d EN |
542 | info->setup = 1; |
543 | ||
1ad69eb0 EN |
544 | info->dir_rename_count = dir_rename_count; |
545 | if (!info->dir_rename_count) { | |
546 | info->dir_rename_count = xmalloc(sizeof(*dir_rename_count)); | |
547 | strmap_init(info->dir_rename_count); | |
548 | } | |
ae8cf74d EN |
549 | strintmap_init_with_options(&info->idx_map, -1, NULL, 0); |
550 | strmap_init_with_options(&info->dir_rename_guess, NULL, 0); | |
ae8cf74d | 551 | |
333899e1 | 552 | /* Setup info->relevant_source_dirs */ |
e4fd06e7 EN |
553 | info->relevant_source_dirs = NULL; |
554 | if (dirs_removed || !relevant_sources) { | |
555 | info->relevant_source_dirs = dirs_removed; /* might be NULL */ | |
556 | } else { | |
557 | info->relevant_source_dirs = xmalloc(sizeof(struct strintmap)); | |
558 | strset_init(info->relevant_source_dirs); | |
559 | strset_for_each_entry(relevant_sources, &iter, entry) { | |
560 | char *dirname = get_dirname(entry->key); | |
561 | if (!dirs_removed || | |
562 | strset_contains(dirs_removed, dirname)) | |
563 | strset_add(info->relevant_source_dirs, dirname); | |
564 | free(dirname); | |
565 | } | |
566 | } | |
333899e1 | 567 | |
ae8cf74d | 568 | /* |
1ad69eb0 EN |
569 | * Loop setting up both info->idx_map, and doing setup of |
570 | * info->dir_rename_count. | |
ae8cf74d EN |
571 | */ |
572 | for (i = 0; i < rename_dst_nr; ++i) { | |
573 | /* | |
574 | * For non-renamed files, make idx_map contain mapping of | |
575 | * filename -> index (index within rename_dst, that is) | |
576 | */ | |
577 | if (!rename_dst[i].is_rename) { | |
578 | char *filename = rename_dst[i].p->two->path; | |
579 | strintmap_set(&info->idx_map, filename, i); | |
1ad69eb0 | 580 | continue; |
ae8cf74d | 581 | } |
1ad69eb0 EN |
582 | |
583 | /* | |
584 | * For everything else (i.e. renamed files), make | |
585 | * dir_rename_count contain a map of a map: | |
586 | * old_directory -> {new_directory -> count} | |
587 | * In other words, for every pair look at the directories for | |
588 | * the old filename and the new filename and count how many | |
589 | * times that pairing occurs. | |
590 | */ | |
591 | update_dir_rename_counts(info, dirs_removed, | |
592 | rename_dst[i].p->one->path, | |
593 | rename_dst[i].p->two->path); | |
ae8cf74d | 594 | } |
81afdf7a EN |
595 | |
596 | /* | |
597 | * Now we collapse | |
598 | * dir_rename_count: old_directory -> {new_directory -> count} | |
599 | * down to | |
600 | * dir_rename_guess: old_directory -> best_new_directory | |
601 | * where best_new_directory is the one with the highest count. | |
602 | */ | |
603 | strmap_for_each_entry(info->dir_rename_count, &iter, entry) { | |
604 | /* entry->key is source_dir */ | |
605 | struct strintmap *counts = entry->value; | |
606 | char *best_newdir; | |
607 | ||
608 | best_newdir = xstrdup(get_highest_rename_path(counts)); | |
609 | strmap_put(&info->dir_rename_guess, entry->key, | |
610 | best_newdir); | |
611 | } | |
ae8cf74d EN |
612 | } |
613 | ||
cd52e005 EN |
614 | void partial_clear_dir_rename_count(struct strmap *dir_rename_count) |
615 | { | |
616 | struct hashmap_iter iter; | |
617 | struct strmap_entry *entry; | |
618 | ||
619 | strmap_for_each_entry(dir_rename_count, &iter, entry) { | |
620 | struct strintmap *counts = entry->value; | |
621 | strintmap_clear(counts); | |
622 | } | |
623 | strmap_partial_clear(dir_rename_count, 1); | |
624 | } | |
625 | ||
b1473019 EN |
626 | static void cleanup_dir_rename_info(struct dir_rename_info *info, |
627 | struct strset *dirs_removed, | |
628 | int keep_dir_rename_count) | |
ae8cf74d | 629 | { |
b1473019 EN |
630 | struct hashmap_iter iter; |
631 | struct strmap_entry *entry; | |
632 | struct string_list to_remove = STRING_LIST_INIT_NODUP; | |
633 | int i; | |
634 | ||
ae8cf74d EN |
635 | if (!info->setup) |
636 | return; | |
637 | ||
638 | /* idx_map */ | |
639 | strintmap_clear(&info->idx_map); | |
640 | ||
641 | /* dir_rename_guess */ | |
642 | strmap_clear(&info->dir_rename_guess, 1); | |
643 | ||
e4fd06e7 EN |
644 | /* relevant_source_dirs */ |
645 | if (info->relevant_source_dirs && | |
646 | info->relevant_source_dirs != dirs_removed) { | |
647 | strset_clear(info->relevant_source_dirs); | |
648 | FREE_AND_NULL(info->relevant_source_dirs); | |
649 | } | |
650 | ||
b6e3d274 | 651 | /* dir_rename_count */ |
b1473019 EN |
652 | if (!keep_dir_rename_count) { |
653 | partial_clear_dir_rename_count(info->dir_rename_count); | |
654 | strmap_clear(info->dir_rename_count, 1); | |
655 | FREE_AND_NULL(info->dir_rename_count); | |
656 | return; | |
657 | } | |
658 | ||
659 | /* | |
660 | * Although dir_rename_count was passed in | |
661 | * diffcore_rename_extended() and we want to keep it around and | |
662 | * return it to that caller, we first want to remove any data | |
663 | * associated with directories that weren't renamed. | |
664 | */ | |
665 | strmap_for_each_entry(info->dir_rename_count, &iter, entry) { | |
666 | const char *source_dir = entry->key; | |
667 | struct strintmap *counts = entry->value; | |
668 | ||
669 | if (!strset_contains(dirs_removed, source_dir)) { | |
670 | string_list_append(&to_remove, source_dir); | |
671 | strintmap_clear(counts); | |
672 | continue; | |
673 | } | |
674 | } | |
675 | for (i = 0; i < to_remove.nr; ++i) | |
676 | strmap_remove(info->dir_rename_count, | |
677 | to_remove.items[i].string, 1); | |
678 | string_list_clear(&to_remove, 0); | |
ae8cf74d EN |
679 | } |
680 | ||
a35df337 EN |
681 | static const char *get_basename(const char *filename) |
682 | { | |
683 | /* | |
684 | * gitbasename() has to worry about special drives, multiple | |
685 | * directory separator characters, trailing slashes, NULL or | |
686 | * empty strings, etc. We only work on filenames as stored in | |
687 | * git, and thus get to ignore all those complications. | |
688 | */ | |
689 | const char *base = strrchr(filename, '/'); | |
690 | return base ? base + 1 : filename; | |
691 | } | |
692 | ||
bde8b9f3 | 693 | static int idx_possible_rename(char *filename, struct dir_rename_info *info) |
37a25143 | 694 | { |
bde8b9f3 EN |
695 | /* |
696 | * Our comparison of files with the same basename (see | |
697 | * find_basename_matches() below), is only helpful when after exact | |
698 | * rename detection we have exactly one file with a given basename | |
699 | * among the rename sources and also only exactly one file with | |
700 | * that basename among the rename destinations. When we have | |
701 | * multiple files with the same basename in either set, we do not | |
702 | * know which to compare against. However, there are some | |
703 | * filenames that occur in large numbers (particularly | |
704 | * build-related filenames such as 'Makefile', '.gitignore', or | |
705 | * 'build.gradle' that potentially exist within every single | |
706 | * subdirectory), and for performance we want to be able to quickly | |
707 | * find renames for these files too. | |
708 | * | |
709 | * The reason basename comparisons are a useful heuristic was that it | |
710 | * is common for people to move files across directories while keeping | |
711 | * their filename the same. If we had a way of determining or even | |
712 | * making a good educated guess about which directory these non-unique | |
713 | * basename files had moved the file to, we could check it. | |
714 | * Luckily... | |
715 | * | |
716 | * When an entire directory is in fact renamed, we have two factors | |
717 | * helping us out: | |
718 | * (a) the original directory disappeared giving us a hint | |
719 | * about when we can apply an extra heuristic. | |
720 | * (a) we often have several files within that directory and | |
721 | * subdirectories that are renamed without changes | |
722 | * So, rules for a heuristic: | |
723 | * (0) If there basename matches are non-unique (the condition under | |
724 | * which this function is called) AND | |
725 | * (1) the directory in which the file was found has disappeared | |
726 | * (i.e. dirs_removed is non-NULL and has a relevant entry) THEN | |
727 | * (2) use exact renames of files within the directory to determine | |
728 | * where the directory is likely to have been renamed to. IF | |
729 | * there is at least one exact rename from within that | |
730 | * directory, we can proceed. | |
731 | * (3) If there are multiple places the directory could have been | |
732 | * renamed to based on exact renames, ignore all but one of them. | |
733 | * Just use the destination with the most renames going to it. | |
734 | * (4) Check if applying that directory rename to the original file | |
735 | * would result in a destination filename that is in the | |
736 | * potential rename set. If so, return the index of the | |
737 | * destination file (the index within rename_dst). | |
738 | * (5) Compare the original file and returned destination for | |
739 | * similarity, and if they are sufficiently similar, record the | |
740 | * rename. | |
741 | * | |
742 | * This function, idx_possible_rename(), is only responsible for (4). | |
81afdf7a EN |
743 | * The conditions/steps in (1)-(3) are handled via setting up |
744 | * dir_rename_count and dir_rename_guess in | |
745 | * initialize_dir_rename_info(). Steps (0) and (5) are handled by | |
746 | * the caller of this function. | |
bde8b9f3 EN |
747 | */ |
748 | char *old_dir, *new_dir; | |
749 | struct strbuf new_path = STRBUF_INIT; | |
750 | int idx; | |
751 | ||
752 | if (!info->setup) | |
753 | return -1; | |
754 | ||
755 | old_dir = get_dirname(filename); | |
756 | new_dir = strmap_get(&info->dir_rename_guess, old_dir); | |
757 | free(old_dir); | |
758 | if (!new_dir) | |
759 | return -1; | |
760 | ||
761 | strbuf_addstr(&new_path, new_dir); | |
762 | strbuf_addch(&new_path, '/'); | |
763 | strbuf_addstr(&new_path, get_basename(filename)); | |
764 | ||
765 | idx = strintmap_get(&info->idx_map, new_path.buf); | |
766 | strbuf_release(&new_path); | |
767 | return idx; | |
37a25143 EN |
768 | } |
769 | ||
a35df337 | 770 | static int find_basename_matches(struct diff_options *options, |
bde8b9f3 | 771 | int minimum_score, |
1ad69eb0 | 772 | struct dir_rename_info *info, |
e4fd06e7 | 773 | struct strset *relevant_sources, |
1ad69eb0 | 774 | struct strset *dirs_removed) |
a35df337 | 775 | { |
da09f651 EN |
776 | /* |
777 | * When I checked in early 2020, over 76% of file renames in linux | |
778 | * just moved files to a different directory but kept the same | |
779 | * basename. gcc did that with over 64% of renames, gecko did it | |
780 | * with over 79%, and WebKit did it with over 89%. | |
781 | * | |
782 | * Therefore we can bypass the normal exhaustive NxM matrix | |
783 | * comparison of similarities between all potential rename sources | |
784 | * and destinations by instead using file basename as a hint (i.e. | |
785 | * the portion of the filename after the last '/'), checking for | |
786 | * similarity between files with the same basename, and if we find | |
787 | * a pair that are sufficiently similar, record the rename pair and | |
788 | * exclude those two from the NxM matrix. | |
789 | * | |
790 | * This *might* cause us to find a less than optimal pairing (if | |
791 | * there is another file that we are even more similar to but has a | |
792 | * different basename). Given the huge performance advantage | |
793 | * basename matching provides, and given the frequency with which | |
794 | * people use the same basename in real world projects, that's a | |
795 | * trade-off we are willing to accept when doing just rename | |
796 | * detection. | |
797 | * | |
798 | * If someone wants copy detection that implies they are willing to | |
799 | * spend more cycles to find similarities between files, so it may | |
800 | * be less likely that this heuristic is wanted. If someone is | |
801 | * doing break detection, that means they do not want filename | |
802 | * similarity to imply any form of content similiarity, and thus | |
803 | * this heuristic would definitely be incompatible. | |
804 | */ | |
805 | ||
806 | int i, renames = 0; | |
a35df337 EN |
807 | struct strintmap sources; |
808 | struct strintmap dests; | |
da09f651 EN |
809 | |
810 | /* | |
811 | * The prefeteching stuff wants to know if it can skip prefetching | |
812 | * blobs that are unmodified...and will then do a little extra work | |
813 | * to verify that the oids are indeed different before prefetching. | |
814 | * Unmodified blobs are only relevant when doing copy detection; | |
815 | * when limiting to rename detection, diffcore_rename[_extended]() | |
816 | * will never be called with unmodified source paths fed to us, so | |
817 | * the extra work necessary to check if rename_src entries are | |
818 | * unmodified would be a small waste. | |
819 | */ | |
820 | int skip_unmodified = 0; | |
a35df337 EN |
821 | |
822 | /* | |
823 | * Create maps of basename -> fullname(s) for remaining sources and | |
824 | * dests. | |
825 | */ | |
826 | strintmap_init_with_options(&sources, -1, NULL, 0); | |
827 | strintmap_init_with_options(&dests, -1, NULL, 0); | |
828 | for (i = 0; i < rename_src_nr; ++i) { | |
829 | char *filename = rename_src[i].p->one->path; | |
830 | const char *base; | |
831 | ||
832 | /* exact renames removed in remove_unneeded_paths_from_src() */ | |
833 | assert(!rename_src[i].p->one->rename_used); | |
834 | ||
835 | /* Record index within rename_src (i) if basename is unique */ | |
836 | base = get_basename(filename); | |
837 | if (strintmap_contains(&sources, base)) | |
838 | strintmap_set(&sources, base, -1); | |
839 | else | |
840 | strintmap_set(&sources, base, i); | |
841 | } | |
842 | for (i = 0; i < rename_dst_nr; ++i) { | |
843 | char *filename = rename_dst[i].p->two->path; | |
844 | const char *base; | |
845 | ||
846 | if (rename_dst[i].is_rename) | |
847 | continue; /* involved in exact match already. */ | |
848 | ||
849 | /* Record index within rename_dst (i) if basename is unique */ | |
850 | base = get_basename(filename); | |
851 | if (strintmap_contains(&dests, base)) | |
852 | strintmap_set(&dests, base, -1); | |
853 | else | |
854 | strintmap_set(&dests, base, i); | |
855 | } | |
856 | ||
da09f651 | 857 | /* Now look for basename matchups and do similarity estimation */ |
37a25143 EN |
858 | for (i = 0; i < rename_src_nr; ++i) { |
859 | char *filename = rename_src[i].p->one->path; | |
860 | const char *base = NULL; | |
861 | intptr_t src_index; | |
da09f651 | 862 | intptr_t dst_index; |
da09f651 | 863 | |
e4fd06e7 EN |
864 | /* Skip irrelevant sources */ |
865 | if (relevant_sources && | |
866 | !strset_contains(relevant_sources, filename)) | |
867 | continue; | |
868 | ||
37a25143 EN |
869 | /* |
870 | * If the basename is unique among remaining sources, then | |
871 | * src_index will equal 'i' and we can attempt to match it | |
872 | * to a unique basename in the destinations. Otherwise, | |
873 | * use directory rename heuristics, if possible. | |
874 | */ | |
875 | base = get_basename(filename); | |
876 | src_index = strintmap_get(&sources, base); | |
877 | assert(src_index == -1 || src_index == i); | |
878 | ||
879 | if (strintmap_contains(&dests, base)) { | |
da09f651 EN |
880 | struct diff_filespec *one, *two; |
881 | int score; | |
882 | ||
37a25143 EN |
883 | /* Find a matching destination, if possible */ |
884 | dst_index = strintmap_get(&dests, base); | |
885 | if (src_index == -1 || dst_index == -1) { | |
886 | src_index = i; | |
bde8b9f3 | 887 | dst_index = idx_possible_rename(filename, info); |
37a25143 EN |
888 | } |
889 | if (dst_index == -1) | |
890 | continue; | |
891 | ||
892 | /* Ignore this dest if already used in a rename */ | |
893 | if (rename_dst[dst_index].is_rename) | |
894 | continue; /* already used previously */ | |
895 | ||
da09f651 EN |
896 | /* Estimate the similarity */ |
897 | one = rename_src[src_index].p->one; | |
898 | two = rename_dst[dst_index].p->two; | |
899 | score = estimate_similarity(options->repo, one, two, | |
900 | minimum_score, skip_unmodified); | |
901 | ||
902 | /* If sufficiently similar, record as rename pair */ | |
903 | if (score < minimum_score) | |
904 | continue; | |
905 | record_rename_pair(dst_index, src_index, score); | |
906 | renames++; | |
1ad69eb0 EN |
907 | update_dir_rename_counts(info, dirs_removed, |
908 | one->path, two->path); | |
da09f651 EN |
909 | |
910 | /* | |
911 | * Found a rename so don't need text anymore; if we | |
912 | * didn't find a rename, the filespec_blob would get | |
913 | * re-used when doing the matrix of comparisons. | |
914 | */ | |
915 | diff_free_filespec_blob(one); | |
916 | diff_free_filespec_blob(two); | |
917 | } | |
918 | } | |
a35df337 EN |
919 | |
920 | strintmap_clear(&sources); | |
921 | strintmap_clear(&dests); | |
922 | ||
da09f651 | 923 | return renames; |
a35df337 EN |
924 | } |
925 | ||
6d24ad97 JH |
926 | #define NUM_CANDIDATE_PER_DST 4 |
927 | static void record_if_better(struct diff_score m[], struct diff_score *o) | |
928 | { | |
929 | int i, worst; | |
930 | ||
931 | /* find the worst one */ | |
932 | worst = 0; | |
933 | for (i = 1; i < NUM_CANDIDATE_PER_DST; i++) | |
934 | if (score_compare(&m[i], &m[worst]) > 0) | |
935 | worst = i; | |
936 | ||
937 | /* is it better than the worst one? */ | |
938 | if (score_compare(&m[worst], o) > 0) | |
939 | m[worst] = *o; | |
940 | } | |
941 | ||
f31027c9 JH |
942 | /* |
943 | * Returns: | |
944 | * 0 if we are under the limit; | |
945 | * 1 if we need to disable inexact rename detection; | |
946 | * 2 if we would be under the limit if we were given -C instead of -C -C. | |
947 | */ | |
00b8cccd | 948 | static int too_many_rename_candidates(int num_destinations, int num_sources, |
9d8a5a50 JH |
949 | struct diff_options *options) |
950 | { | |
951 | int rename_limit = options->rename_limit; | |
00b8cccd | 952 | int i, limited_sources; |
9d8a5a50 JH |
953 | |
954 | options->needed_rename_limit = 0; | |
955 | ||
956 | /* | |
957 | * This basically does a test for the rename matrix not | |
958 | * growing larger than a "rename_limit" square matrix, ie: | |
959 | * | |
00b8cccd | 960 | * num_destinations * num_sources > rename_limit * rename_limit |
ad8a1be5 EN |
961 | * |
962 | * We use st_mult() to check overflow conditions; in the | |
963 | * exceptional circumstance that size_t isn't large enough to hold | |
964 | * the multiplication, the system won't be able to allocate enough | |
965 | * memory for the matrix anyway. | |
9d8a5a50 | 966 | */ |
89973554 JT |
967 | if (rename_limit <= 0) |
968 | rename_limit = 32767; | |
ad8a1be5 EN |
969 | if (st_mult(num_destinations, num_sources) |
970 | <= st_mult(rename_limit, rename_limit)) | |
9d8a5a50 JH |
971 | return 0; |
972 | ||
973 | options->needed_rename_limit = | |
00b8cccd | 974 | num_sources > num_destinations ? num_sources : num_destinations; |
f31027c9 JH |
975 | |
976 | /* Are we running under -C -C? */ | |
0d1e0e78 | 977 | if (!options->flags.find_copies_harder) |
f31027c9 JH |
978 | return 1; |
979 | ||
980 | /* Would we bust the limit if we were running under -C? */ | |
00b8cccd | 981 | for (limited_sources = i = 0; i < num_sources; i++) { |
f31027c9 JH |
982 | if (diff_unmodified_pair(rename_src[i].p)) |
983 | continue; | |
00b8cccd | 984 | limited_sources++; |
f31027c9 | 985 | } |
ad8a1be5 EN |
986 | if (st_mult(num_destinations, limited_sources) |
987 | <= st_mult(rename_limit, rename_limit)) | |
f31027c9 | 988 | return 2; |
9d8a5a50 JH |
989 | return 1; |
990 | } | |
991 | ||
1ad69eb0 EN |
992 | static int find_renames(struct diff_score *mx, |
993 | int dst_cnt, | |
994 | int minimum_score, | |
995 | int copies, | |
996 | struct dir_rename_info *info, | |
997 | struct strset *dirs_removed) | |
0940e5f2 LT |
998 | { |
999 | int count = 0, i; | |
1000 | ||
1001 | for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) { | |
1002 | struct diff_rename_dst *dst; | |
1003 | ||
1004 | if ((mx[i].dst < 0) || | |
1005 | (mx[i].score < minimum_score)) | |
1006 | break; /* there is no more usable pair. */ | |
1007 | dst = &rename_dst[mx[i].dst]; | |
9db2ac56 | 1008 | if (dst->is_rename) |
0940e5f2 | 1009 | continue; /* already done, either exact or fuzzy. */ |
e88d6bc6 | 1010 | if (!copies && rename_src[mx[i].src].p->one->rename_used) |
0940e5f2 LT |
1011 | continue; |
1012 | record_rename_pair(mx[i].dst, mx[i].src, mx[i].score); | |
1013 | count++; | |
1ad69eb0 EN |
1014 | update_dir_rename_counts(info, dirs_removed, |
1015 | rename_src[mx[i].src].p->one->path, | |
1016 | rename_dst[mx[i].dst].p->two->path); | |
0940e5f2 LT |
1017 | } |
1018 | return count; | |
1019 | } | |
1020 | ||
9799889f EN |
1021 | static void remove_unneeded_paths_from_src(int detecting_copies, |
1022 | struct strset *interesting) | |
829514c5 EN |
1023 | { |
1024 | int i, new_num_src; | |
1025 | ||
9799889f | 1026 | if (detecting_copies && !interesting) |
829514c5 EN |
1027 | return; /* nothing to remove */ |
1028 | if (break_idx) | |
1029 | return; /* culling incompatible with break detection */ | |
1030 | ||
1031 | /* | |
1032 | * Note on reasons why we cull unneeded sources but not destinations: | |
1033 | * 1) Pairings are stored in rename_dst (not rename_src), which we | |
1034 | * need to keep around. So, we just can't cull rename_dst even | |
1035 | * if we wanted to. But doing so wouldn't help because... | |
1036 | * | |
1037 | * 2) There is a matrix pairwise comparison that follows the | |
1038 | * "Performing inexact rename detection" progress message. | |
1039 | * Iterating over the destinations is done in the outer loop, | |
1040 | * hence we only iterate over each of those once and we can | |
1041 | * easily skip the outer loop early if the destination isn't | |
1042 | * relevant. That's only one check per destination path to | |
1043 | * skip. | |
1044 | * | |
1045 | * By contrast, the sources are iterated in the inner loop; if | |
1046 | * we check whether a source can be skipped, then we'll be | |
1047 | * checking it N separate times, once for each destination. | |
1048 | * We don't want to have to iterate over known-not-needed | |
1049 | * sources N times each, so avoid that by removing the sources | |
1050 | * from rename_src here. | |
1051 | */ | |
1052 | for (i = 0, new_num_src = 0; i < rename_src_nr; i++) { | |
9799889f EN |
1053 | struct diff_filespec *one = rename_src[i].p->one; |
1054 | ||
829514c5 EN |
1055 | /* |
1056 | * renames are stored in rename_dst, so if a rename has | |
1057 | * already been detected using this source, we can just | |
1058 | * remove the source knowing rename_dst has its info. | |
1059 | */ | |
9799889f EN |
1060 | if (!detecting_copies && one->rename_used) |
1061 | continue; | |
1062 | ||
1063 | /* If we don't care about the source path, skip it */ | |
1064 | if (interesting && !strset_contains(interesting, one->path)) | |
829514c5 EN |
1065 | continue; |
1066 | ||
1067 | if (new_num_src < i) | |
1068 | memcpy(&rename_src[new_num_src], &rename_src[i], | |
1069 | sizeof(struct diff_rename_src)); | |
1070 | new_num_src++; | |
1071 | } | |
1072 | ||
1073 | rename_src_nr = new_num_src; | |
1074 | } | |
1075 | ||
0c4fd732 | 1076 | void diffcore_rename_extended(struct diff_options *options, |
9799889f | 1077 | struct strset *relevant_sources, |
0c4fd732 EN |
1078 | struct strset *dirs_removed, |
1079 | struct strmap *dir_rename_count) | |
427dcb4b | 1080 | { |
8082d8d3 JH |
1081 | int detect_rename = options->detect_rename; |
1082 | int minimum_score = options->rename_score; | |
38c6f780 | 1083 | struct diff_queue_struct *q = &diff_queued_diff; |
5098bafb | 1084 | struct diff_queue_struct outq; |
427dcb4b | 1085 | struct diff_score *mx; |
f31027c9 | 1086 | int i, j, rename_count, skip_unmodified = 0; |
26a66a6b | 1087 | int num_destinations, dst_cnt; |
f15eb7c1 | 1088 | int num_sources, want_copies; |
3ac942d4 | 1089 | struct progress *progress = NULL; |
bde8b9f3 | 1090 | struct dir_rename_info info; |
427dcb4b | 1091 | |
557ac035 | 1092 | trace2_region_enter("diff", "setup", options->repo); |
bde8b9f3 | 1093 | info.setup = 0; |
0c4fd732 | 1094 | assert(!dir_rename_count || strmap_empty(dir_rename_count)); |
f15eb7c1 | 1095 | want_copies = (detect_rename == DIFF_DETECT_COPY); |
1ad69eb0 EN |
1096 | if (dirs_removed && (break_idx || want_copies)) |
1097 | BUG("dirs_removed incompatible with break/copy detection"); | |
9799889f EN |
1098 | if (break_idx && relevant_sources) |
1099 | BUG("break detection incompatible with source specification"); | |
26dee0ad | 1100 | if (!minimum_score) |
f345b0a0 | 1101 | minimum_score = DEFAULT_RENAME_SCORE; |
427dcb4b | 1102 | |
427dcb4b | 1103 | for (i = 0; i < q->nr; i++) { |
52e95789 | 1104 | struct diff_filepair *p = q->queue[i]; |
2f3f8b21 | 1105 | if (!DIFF_FILE_VALID(p->one)) { |
81e50eab | 1106 | if (!DIFF_FILE_VALID(p->two)) |
60896c7b | 1107 | continue; /* unmerged */ |
2f3f8b21 JH |
1108 | else if (options->single_follow && |
1109 | strcmp(options->single_follow, p->two->path)) | |
1110 | continue; /* not interested */ | |
0d1e0e78 | 1111 | else if (!options->flags.rename_empty && |
02491b67 | 1112 | is_empty_blob_oid(&p->two->oid)) |
90d43b07 | 1113 | continue; |
9db2ac56 | 1114 | else if (add_rename_dst(p) < 0) { |
4d6be03b JK |
1115 | warning("skipping rename detection, detected" |
1116 | " duplicate destination '%s'", | |
1117 | p->two->path); | |
1118 | goto cleanup; | |
1119 | } | |
2f3f8b21 | 1120 | } |
0d1e0e78 | 1121 | else if (!options->flags.rename_empty && |
02491b67 | 1122 | is_empty_blob_oid(&p->one->oid)) |
90d43b07 | 1123 | continue; |
d7c9bf22 | 1124 | else if (!DIFF_PAIR_UNMERGED(p) && !DIFF_FILE_VALID(p->two)) { |
64479711 LT |
1125 | /* |
1126 | * If the source is a broken "delete", and | |
2210100a JH |
1127 | * they did not really want to get broken, |
1128 | * that means the source actually stays. | |
64479711 LT |
1129 | * So we increment the "rename_used" score |
1130 | * by one, to indicate ourselves as a user | |
1131 | */ | |
1132 | if (p->broken_pair && !p->score) | |
1133 | p->one->rename_used++; | |
e88d6bc6 | 1134 | register_rename_src(p); |
64479711 | 1135 | } |
f15eb7c1 | 1136 | else if (want_copies) { |
64479711 LT |
1137 | /* |
1138 | * Increment the "rename_used" score by | |
1139 | * one, to indicate ourselves as a user. | |
2210100a | 1140 | */ |
64479711 | 1141 | p->one->rename_used++; |
e88d6bc6 | 1142 | register_rename_src(p); |
2210100a | 1143 | } |
427dcb4b | 1144 | } |
557ac035 | 1145 | trace2_region_leave("diff", "setup", options->repo); |
0024a549 | 1146 | if (rename_dst_nr == 0 || rename_src_nr == 0) |
427dcb4b JH |
1147 | goto cleanup; /* nothing to do */ |
1148 | ||
557ac035 | 1149 | trace2_region_enter("diff", "exact renames", options->repo); |
17559a64 LT |
1150 | /* |
1151 | * We really want to cull the candidates list early | |
1152 | * with cheap tests in order to avoid doing deltas. | |
1153 | */ | |
11f944dd | 1154 | rename_count = find_exact_renames(options); |
557ac035 | 1155 | trace2_region_leave("diff", "exact renames", options->repo); |
17559a64 | 1156 | |
42899ac8 LT |
1157 | /* Did we only want exact renames? */ |
1158 | if (minimum_score == MAX_SCORE) | |
1159 | goto cleanup; | |
1160 | ||
f15eb7c1 | 1161 | num_sources = rename_src_nr; |
42899ac8 | 1162 | |
bd24aa2f EN |
1163 | if (want_copies || break_idx) { |
1164 | /* | |
1165 | * Cull sources: | |
1166 | * - remove ones corresponding to exact renames | |
9799889f | 1167 | * - remove ones not found in relevant_sources |
bd24aa2f EN |
1168 | */ |
1169 | trace2_region_enter("diff", "cull after exact", options->repo); | |
9799889f | 1170 | remove_unneeded_paths_from_src(want_copies, relevant_sources); |
bd24aa2f EN |
1171 | trace2_region_leave("diff", "cull after exact", options->repo); |
1172 | } else { | |
1173 | /* Determine minimum score to match basenames */ | |
1174 | double factor = 0.5; | |
1175 | char *basename_factor = getenv("GIT_BASENAME_FACTOR"); | |
1176 | int min_basename_score; | |
1177 | ||
1178 | if (basename_factor) | |
1179 | factor = strtol(basename_factor, NULL, 10)/100.0; | |
1180 | assert(factor >= 0.0 && factor <= 1.0); | |
1181 | min_basename_score = minimum_score + | |
1182 | (int)(factor * (MAX_SCORE - minimum_score)); | |
1183 | ||
1184 | /* | |
1185 | * Cull sources: | |
1186 | * - remove ones involved in renames (found via exact match) | |
1187 | */ | |
1188 | trace2_region_enter("diff", "cull after exact", options->repo); | |
9799889f | 1189 | remove_unneeded_paths_from_src(want_copies, NULL); |
bd24aa2f EN |
1190 | trace2_region_leave("diff", "cull after exact", options->repo); |
1191 | ||
ae8cf74d EN |
1192 | /* Preparation for basename-driven matching. */ |
1193 | trace2_region_enter("diff", "dir rename setup", options->repo); | |
e4fd06e7 | 1194 | initialize_dir_rename_info(&info, relevant_sources, |
1ad69eb0 | 1195 | dirs_removed, dir_rename_count); |
ae8cf74d EN |
1196 | trace2_region_leave("diff", "dir rename setup", options->repo); |
1197 | ||
bd24aa2f EN |
1198 | /* Utilize file basenames to quickly find renames. */ |
1199 | trace2_region_enter("diff", "basename matches", options->repo); | |
1200 | rename_count += find_basename_matches(options, | |
bde8b9f3 | 1201 | min_basename_score, |
e4fd06e7 EN |
1202 | &info, |
1203 | relevant_sources, | |
1204 | dirs_removed); | |
bd24aa2f EN |
1205 | trace2_region_leave("diff", "basename matches", options->repo); |
1206 | ||
1207 | /* | |
1208 | * Cull sources, again: | |
1209 | * - remove ones involved in renames (found via basenames) | |
9799889f | 1210 | * - remove ones not found in relevant_sources |
bd24aa2f EN |
1211 | */ |
1212 | trace2_region_enter("diff", "cull basename", options->repo); | |
9799889f | 1213 | remove_unneeded_paths_from_src(want_copies, relevant_sources); |
bd24aa2f EN |
1214 | trace2_region_leave("diff", "cull basename", options->repo); |
1215 | } | |
1216 | ||
1217 | /* Calculate how many rename destinations are left */ | |
1218 | num_destinations = (rename_dst_nr - rename_count); | |
1219 | num_sources = rename_src_nr; /* rename_src_nr reflects lower number */ | |
1220 | ||
42899ac8 | 1221 | /* All done? */ |
f15eb7c1 | 1222 | if (!num_destinations || !num_sources) |
42899ac8 LT |
1223 | goto cleanup; |
1224 | ||
f15eb7c1 | 1225 | switch (too_many_rename_candidates(num_destinations, num_sources, |
00b8cccd | 1226 | options)) { |
f31027c9 | 1227 | case 1: |
0024a549 | 1228 | goto cleanup; |
f31027c9 JH |
1229 | case 2: |
1230 | options->degraded_cc_to_c = 1; | |
1231 | skip_unmodified = 1; | |
1232 | break; | |
1233 | default: | |
1234 | break; | |
1235 | } | |
0024a549 | 1236 | |
557ac035 | 1237 | trace2_region_enter("diff", "inexact renames", options->repo); |
3ac942d4 | 1238 | if (options->show_rename_progress) { |
8aade107 | 1239 | progress = start_delayed_progress( |
754dbc43 | 1240 | _("Performing inexact rename detection"), |
f15eb7c1 | 1241 | (uint64_t)num_destinations * (uint64_t)num_sources); |
3ac942d4 JK |
1242 | } |
1243 | ||
ca56dadb | 1244 | CALLOC_ARRAY(mx, st_mult(NUM_CANDIDATE_PER_DST, num_destinations)); |
25d5ea41 | 1245 | for (dst_cnt = i = 0; i < rename_dst_nr; i++) { |
9db2ac56 | 1246 | struct diff_filespec *two = rename_dst[i].p->two; |
6d24ad97 JH |
1247 | struct diff_score *m; |
1248 | ||
9db2ac56 | 1249 | if (rename_dst[i].is_rename) |
bd24aa2f | 1250 | continue; /* exact or basename match already handled */ |
6d24ad97 JH |
1251 | |
1252 | m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST]; | |
1253 | for (j = 0; j < NUM_CANDIDATE_PER_DST; j++) | |
1254 | m[j].dst = -1; | |
1255 | ||
25d5ea41 | 1256 | for (j = 0; j < rename_src_nr; j++) { |
e88d6bc6 | 1257 | struct diff_filespec *one = rename_src[j].p->one; |
6d24ad97 | 1258 | struct diff_score this_src; |
f31027c9 | 1259 | |
829514c5 | 1260 | assert(!one->rename_used || want_copies || break_idx); |
f15eb7c1 | 1261 | |
f31027c9 JH |
1262 | if (skip_unmodified && |
1263 | diff_unmodified_pair(rename_src[j].p)) | |
1264 | continue; | |
1265 | ||
b78ea5fc NTND |
1266 | this_src.score = estimate_similarity(options->repo, |
1267 | one, two, | |
95acf11a JT |
1268 | minimum_score, |
1269 | skip_unmodified); | |
6d24ad97 JH |
1270 | this_src.name_score = basename_same(one, two); |
1271 | this_src.dst = i; | |
1272 | this_src.src = j; | |
1273 | record_if_better(m, &this_src); | |
809809bb JH |
1274 | /* |
1275 | * Once we run estimate_similarity, | |
1276 | * We do not need the text anymore. | |
1277 | */ | |
8ae92e63 | 1278 | diff_free_filespec_blob(one); |
809809bb | 1279 | diff_free_filespec_blob(two); |
427dcb4b JH |
1280 | } |
1281 | dst_cnt++; | |
81c4bf02 | 1282 | display_progress(progress, |
f15eb7c1 | 1283 | (uint64_t)dst_cnt * (uint64_t)num_sources); |
427dcb4b | 1284 | } |
3ac942d4 | 1285 | stop_progress(&progress); |
6d24ad97 | 1286 | |
427dcb4b | 1287 | /* cost matrix sorted by most to least similar pair */ |
2049b8dc | 1288 | STABLE_QSORT(mx, dst_cnt * NUM_CANDIDATE_PER_DST, score_compare); |
6d24ad97 | 1289 | |
1ad69eb0 EN |
1290 | rename_count += find_renames(mx, dst_cnt, minimum_score, 0, |
1291 | &info, dirs_removed); | |
f15eb7c1 | 1292 | if (want_copies) |
1ad69eb0 EN |
1293 | rename_count += find_renames(mx, dst_cnt, minimum_score, 1, |
1294 | &info, dirs_removed); | |
427dcb4b | 1295 | free(mx); |
557ac035 | 1296 | trace2_region_leave("diff", "inexact renames", options->repo); |
427dcb4b | 1297 | |
15d061b4 | 1298 | cleanup: |
427dcb4b | 1299 | /* At this point, we have found some renames and copies and they |
5098bafb | 1300 | * are recorded in rename_dst. The original list is still in *q. |
427dcb4b | 1301 | */ |
557ac035 | 1302 | trace2_region_enter("diff", "write back to queue", options->repo); |
9ca5df90 | 1303 | DIFF_QUEUE_CLEAR(&outq); |
427dcb4b | 1304 | for (i = 0; i < q->nr; i++) { |
25d5ea41 | 1305 | struct diff_filepair *p = q->queue[i]; |
25d5ea41 JH |
1306 | struct diff_filepair *pair_to_free = NULL; |
1307 | ||
d7c9bf22 MZ |
1308 | if (DIFF_PAIR_UNMERGED(p)) { |
1309 | diff_q(&outq, p); | |
1310 | } | |
1311 | else if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) { | |
9db2ac56 EN |
1312 | /* Creation */ |
1313 | diff_q(&outq, p); | |
427dcb4b | 1314 | } |
2cd68882 JH |
1315 | else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) { |
1316 | /* | |
1317 | * Deletion | |
1318 | * | |
f345b0a0 JH |
1319 | * We would output this delete record if: |
1320 | * | |
1321 | * (1) this is a broken delete and the counterpart | |
1322 | * broken create remains in the output; or | |
5098bafb JH |
1323 | * (2) this is not a broken delete, and rename_dst |
1324 | * does not have a rename/copy to move p->one->path | |
1325 | * out of existence. | |
f345b0a0 JH |
1326 | * |
1327 | * Otherwise, the counterpart broken create | |
1328 | * has been turned into a rename-edit; or | |
1329 | * delete did not have a matching create to | |
1330 | * begin with. | |
2cd68882 | 1331 | */ |
f345b0a0 JH |
1332 | if (DIFF_PAIR_BROKEN(p)) { |
1333 | /* broken delete */ | |
9db2ac56 EN |
1334 | struct diff_rename_dst *dst = locate_rename_dst(p); |
1335 | if (!dst) | |
1336 | BUG("tracking failed somehow; failed to find associated dst for broken pair"); | |
1337 | if (dst->is_rename) | |
f345b0a0 JH |
1338 | /* counterpart is now rename/copy */ |
1339 | pair_to_free = p; | |
1340 | } | |
1341 | else { | |
64479711 | 1342 | if (p->one->rename_used) |
f345b0a0 JH |
1343 | /* this path remains */ |
1344 | pair_to_free = p; | |
1345 | } | |
2cd68882 | 1346 | |
9db2ac56 | 1347 | if (!pair_to_free) |
2cd68882 JH |
1348 | diff_q(&outq, p); |
1349 | } | |
25d5ea41 | 1350 | else if (!diff_unmodified_pair(p)) |
15d061b4 | 1351 | /* all the usual ones need to be kept */ |
25d5ea41 | 1352 | diff_q(&outq, p); |
15d061b4 | 1353 | else |
9db2ac56 | 1354 | /* no need to keep unmodified pairs; FIXME: remove earlier? */ |
15d061b4 JH |
1355 | pair_to_free = p; |
1356 | ||
226406f6 JH |
1357 | if (pair_to_free) |
1358 | diff_free_filepair(pair_to_free); | |
427dcb4b | 1359 | } |
25d5ea41 | 1360 | diff_debug_queue("done copying original", &outq); |
427dcb4b | 1361 | |
25d5ea41 JH |
1362 | free(q->queue); |
1363 | *q = outq; | |
1364 | diff_debug_queue("done collapsing", q); | |
427dcb4b | 1365 | |
9fb88419 | 1366 | for (i = 0; i < rename_dst_nr; i++) |
9db2ac56 EN |
1367 | if (rename_dst[i].filespec_to_free) |
1368 | free_filespec(rename_dst[i].filespec_to_free); | |
5098bafb | 1369 | |
b1473019 | 1370 | cleanup_dir_rename_info(&info, dirs_removed, dir_rename_count != NULL); |
6a83d902 | 1371 | FREE_AND_NULL(rename_dst); |
25d5ea41 | 1372 | rename_dst_nr = rename_dst_alloc = 0; |
6a83d902 | 1373 | FREE_AND_NULL(rename_src); |
25d5ea41 | 1374 | rename_src_nr = rename_src_alloc = 0; |
9db2ac56 EN |
1375 | if (break_idx) { |
1376 | strintmap_clear(break_idx); | |
1377 | FREE_AND_NULL(break_idx); | |
1378 | } | |
557ac035 | 1379 | trace2_region_leave("diff", "write back to queue", options->repo); |
427dcb4b JH |
1380 | return; |
1381 | } | |
0c4fd732 EN |
1382 | |
1383 | void diffcore_rename(struct diff_options *options) | |
1384 | { | |
9799889f | 1385 | diffcore_rename_extended(options, NULL, NULL, NULL); |
0c4fd732 | 1386 | } |