3 * Copyright (C) 2005 Junio C Hamano
8 #include "object-store.h"
11 #include "promisor-remote.h"
13 /* Table of rename/copy destinations */
15 static struct diff_rename_dst
{
16 struct diff_filespec
*two
;
17 struct diff_filepair
*pair
;
19 static int rename_dst_nr
, rename_dst_alloc
;
21 static int find_rename_dst(struct diff_filespec
*two
)
27 while (last
> first
) {
28 int next
= first
+ ((last
- first
) >> 1);
29 struct diff_rename_dst
*dst
= &(rename_dst
[next
]);
30 int cmp
= strcmp(two
->path
, dst
->two
->path
);
42 static struct diff_rename_dst
*locate_rename_dst(struct diff_filespec
*two
)
44 int ofs
= find_rename_dst(two
);
45 return ofs
< 0 ? NULL
: &rename_dst
[ofs
];
49 * Returns 0 on success, -1 if we found a duplicate.
51 static int add_rename_dst(struct diff_filespec
*two
)
53 int first
= find_rename_dst(two
);
59 /* insert to make it at "first" */
60 ALLOC_GROW(rename_dst
, rename_dst_nr
+ 1, rename_dst_alloc
);
62 if (first
< rename_dst_nr
)
63 MOVE_ARRAY(rename_dst
+ first
+ 1, rename_dst
+ first
,
64 rename_dst_nr
- first
- 1);
65 rename_dst
[first
].two
= alloc_filespec(two
->path
);
66 fill_filespec(rename_dst
[first
].two
, &two
->oid
, two
->oid_valid
,
68 rename_dst
[first
].pair
= NULL
;
72 /* Table of rename/copy src files */
73 static struct diff_rename_src
{
74 struct diff_filepair
*p
;
75 unsigned short score
; /* to remember the break score */
77 static int rename_src_nr
, rename_src_alloc
;
79 static struct diff_rename_src
*register_rename_src(struct diff_filepair
*p
)
82 struct diff_filespec
*one
= p
->one
;
83 unsigned short score
= p
->score
;
87 while (last
> first
) {
88 int next
= first
+ ((last
- first
) >> 1);
89 struct diff_rename_src
*src
= &(rename_src
[next
]);
90 int cmp
= strcmp(one
->path
, src
->p
->one
->path
);
100 /* insert to make it at "first" */
101 ALLOC_GROW(rename_src
, rename_src_nr
+ 1, rename_src_alloc
);
103 if (first
< rename_src_nr
)
104 MOVE_ARRAY(rename_src
+ first
+ 1, rename_src
+ first
,
105 rename_src_nr
- first
- 1);
106 rename_src
[first
].p
= p
;
107 rename_src
[first
].score
= score
;
108 return &(rename_src
[first
]);
111 static int basename_same(struct diff_filespec
*src
, struct diff_filespec
*dst
)
113 int src_len
= strlen(src
->path
), dst_len
= strlen(dst
->path
);
114 while (src_len
&& dst_len
) {
115 char c1
= src
->path
[--src_len
];
116 char c2
= dst
->path
[--dst_len
];
122 return (!src_len
|| src
->path
[src_len
- 1] == '/') &&
123 (!dst_len
|| dst
->path
[dst_len
- 1] == '/');
127 int src
; /* index in rename_src */
128 int dst
; /* index in rename_dst */
129 unsigned short score
;
133 struct prefetch_options
{
134 struct repository
*repo
;
137 static void prefetch(void *prefetch_options
)
139 struct prefetch_options
*options
= prefetch_options
;
141 struct oid_array to_fetch
= OID_ARRAY_INIT
;
143 for (i
= 0; i
< rename_dst_nr
; i
++) {
144 if (rename_dst
[i
].pair
)
146 * The loop in diffcore_rename() will not need these
147 * blobs, so skip prefetching.
149 continue; /* already found exact match */
150 diff_add_if_missing(options
->repo
, &to_fetch
,
153 for (i
= 0; i
< rename_src_nr
; i
++) {
154 if (options
->skip_unmodified
&&
155 diff_unmodified_pair(rename_src
[i
].p
))
157 * The loop in diffcore_rename() will not need these
158 * blobs, so skip prefetching.
161 diff_add_if_missing(options
->repo
, &to_fetch
,
162 rename_src
[i
].p
->one
);
164 promisor_remote_get_direct(options
->repo
, to_fetch
.oid
, to_fetch
.nr
);
165 oid_array_clear(&to_fetch
);
168 static int estimate_similarity(struct repository
*r
,
169 struct diff_filespec
*src
,
170 struct diff_filespec
*dst
,
174 /* src points at a file that existed in the original tree (or
175 * optionally a file in the destination tree) and dst points
176 * at a newly created file. They may be quite similar, in which
177 * case we want to say src is renamed to dst or src is copied into
178 * dst, and then some edit has been applied to dst.
180 * Compare them and return how similar they are, representing
181 * the score as an integer between 0 and MAX_SCORE.
183 * When there is an exact match, it is considered a better
184 * match than anything else; the destination does not even
185 * call into this function in that case.
187 unsigned long max_size
, delta_size
, base_size
, src_copied
, literal_added
;
189 struct diff_populate_filespec_options dpf_options
= {
192 struct prefetch_options prefetch_options
= {r
, skip_unmodified
};
194 if (r
== the_repository
&& has_promisor_remote()) {
195 dpf_options
.missing_object_cb
= prefetch
;
196 dpf_options
.missing_object_data
= &prefetch_options
;
199 /* We deal only with regular files. Symlink renames are handled
200 * only when they are exact matches --- in other words, no edits
203 if (!S_ISREG(src
->mode
) || !S_ISREG(dst
->mode
))
207 * Need to check that source and destination sizes are
208 * filled in before comparing them.
210 * If we already have "cnt_data" filled in, we know it's
211 * all good (avoid checking the size for zero, as that
212 * is a possible size - we really should have a flag to
213 * say whether the size is valid or not!)
215 if (!src
->cnt_data
&&
216 diff_populate_filespec(r
, src
, &dpf_options
))
218 if (!dst
->cnt_data
&&
219 diff_populate_filespec(r
, dst
, &dpf_options
))
222 max_size
= ((src
->size
> dst
->size
) ? src
->size
: dst
->size
);
223 base_size
= ((src
->size
< dst
->size
) ? src
->size
: dst
->size
);
224 delta_size
= max_size
- base_size
;
226 /* We would not consider edits that change the file size so
227 * drastically. delta_size must be smaller than
228 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
230 * Note that base_size == 0 case is handled here already
231 * and the final score computation below would not have a
232 * divide-by-zero issue.
234 if (max_size
* (MAX_SCORE
-minimum_score
) < delta_size
* MAX_SCORE
)
237 dpf_options
.check_size_only
= 0;
239 if (!src
->cnt_data
&& diff_populate_filespec(r
, src
, &dpf_options
))
241 if (!dst
->cnt_data
&& diff_populate_filespec(r
, dst
, &dpf_options
))
244 if (diffcore_count_changes(r
, src
, dst
,
245 &src
->cnt_data
, &dst
->cnt_data
,
246 &src_copied
, &literal_added
))
249 /* How similar are they?
250 * what percentage of material in dst are from source?
253 score
= 0; /* should not happen */
255 score
= (int)(src_copied
* MAX_SCORE
/ max_size
);
259 static void record_rename_pair(int dst_index
, int src_index
, int score
)
261 struct diff_filespec
*src
, *dst
;
262 struct diff_filepair
*dp
;
264 if (rename_dst
[dst_index
].pair
)
265 die("internal error: dst already matched.");
267 src
= rename_src
[src_index
].p
->one
;
271 dst
= rename_dst
[dst_index
].two
;
274 dp
= diff_queue(NULL
, src
, dst
);
275 dp
->renamed_pair
= 1;
276 if (!strcmp(src
->path
, dst
->path
))
277 dp
->score
= rename_src
[src_index
].score
;
280 rename_dst
[dst_index
].pair
= dp
;
284 * We sort the rename similarity matrix with the score, in descending
285 * order (the most similar first).
287 static int score_compare(const void *a_
, const void *b_
)
289 const struct diff_score
*a
= a_
, *b
= b_
;
291 /* sink the unused ones to the bottom */
293 return (0 <= b
->dst
);
297 if (a
->score
== b
->score
)
298 return b
->name_score
- a
->name_score
;
300 return b
->score
- a
->score
;
303 struct file_similarity
{
304 struct hashmap_entry entry
;
306 struct diff_filespec
*filespec
;
309 static unsigned int hash_filespec(struct repository
*r
,
310 struct diff_filespec
*filespec
)
312 if (!filespec
->oid_valid
) {
313 if (diff_populate_filespec(r
, filespec
, NULL
))
315 hash_object_file(r
->hash_algo
, filespec
->data
, filespec
->size
,
316 "blob", &filespec
->oid
);
318 return oidhash(&filespec
->oid
);
321 static int find_identical_files(struct hashmap
*srcs
,
323 struct diff_options
*options
)
326 struct diff_filespec
*target
= rename_dst
[dst_index
].two
;
327 struct file_similarity
*p
, *best
= NULL
;
328 int i
= 100, best_score
= -1;
329 unsigned int hash
= hash_filespec(options
->repo
, target
);
332 * Find the best source match for specified destination.
334 p
= hashmap_get_entry_from_hash(srcs
, hash
, NULL
,
335 struct file_similarity
, entry
);
336 hashmap_for_each_entry_from(srcs
, p
, entry
) {
338 struct diff_filespec
*source
= p
->filespec
;
340 /* False hash collision? */
341 if (!oideq(&source
->oid
, &target
->oid
))
343 /* Non-regular files? If so, the modes must match! */
344 if (!S_ISREG(source
->mode
) || !S_ISREG(target
->mode
)) {
345 if (source
->mode
!= target
->mode
)
348 /* Give higher scores to sources that haven't been used already */
349 score
= !source
->rename_used
;
350 if (source
->rename_used
&& options
->detect_rename
!= DIFF_DETECT_COPY
)
352 score
+= basename_same(source
, target
);
353 if (score
> best_score
) {
360 /* Too many identical alternatives? Pick one */
365 record_rename_pair(dst_index
, best
->index
, MAX_SCORE
);
371 static void insert_file_table(struct repository
*r
,
372 struct hashmap
*table
, int index
,
373 struct diff_filespec
*filespec
)
375 struct file_similarity
*entry
= xmalloc(sizeof(*entry
));
377 entry
->index
= index
;
378 entry
->filespec
= filespec
;
380 hashmap_entry_init(&entry
->entry
, hash_filespec(r
, filespec
));
381 hashmap_add(table
, &entry
->entry
);
385 * Find exact renames first.
387 * The first round matches up the up-to-date entries,
388 * and then during the second round we try to match
389 * cache-dirty entries as well.
391 static int find_exact_renames(struct diff_options
*options
)
394 struct hashmap file_table
;
396 /* Add all sources to the hash table in reverse order, because
397 * later on they will be retrieved in LIFO order.
399 hashmap_init(&file_table
, NULL
, NULL
, rename_src_nr
);
400 for (i
= rename_src_nr
-1; i
>= 0; i
--)
401 insert_file_table(options
->repo
,
403 rename_src
[i
].p
->one
);
405 /* Walk the destinations and find best source match */
406 for (i
= 0; i
< rename_dst_nr
; i
++)
407 renames
+= find_identical_files(&file_table
, i
, options
);
409 /* Free the hash data structure and entries */
410 hashmap_free_entries(&file_table
, struct file_similarity
, entry
);
415 #define NUM_CANDIDATE_PER_DST 4
416 static void record_if_better(struct diff_score m
[], struct diff_score
*o
)
420 /* find the worst one */
422 for (i
= 1; i
< NUM_CANDIDATE_PER_DST
; i
++)
423 if (score_compare(&m
[i
], &m
[worst
]) > 0)
426 /* is it better than the worst one? */
427 if (score_compare(&m
[worst
], o
) > 0)
433 * 0 if we are under the limit;
434 * 1 if we need to disable inexact rename detection;
435 * 2 if we would be under the limit if we were given -C instead of -C -C.
437 static int too_many_rename_candidates(int num_create
,
438 struct diff_options
*options
)
440 int rename_limit
= options
->rename_limit
;
441 int num_src
= rename_src_nr
;
444 options
->needed_rename_limit
= 0;
447 * This basically does a test for the rename matrix not
448 * growing larger than a "rename_limit" square matrix, ie:
450 * num_create * num_src > rename_limit * rename_limit
452 if (rename_limit
<= 0)
453 rename_limit
= 32767;
454 if ((num_create
<= rename_limit
|| num_src
<= rename_limit
) &&
455 ((uint64_t)num_create
* (uint64_t)num_src
456 <= (uint64_t)rename_limit
* (uint64_t)rename_limit
))
459 options
->needed_rename_limit
=
460 num_src
> num_create
? num_src
: num_create
;
462 /* Are we running under -C -C? */
463 if (!options
->flags
.find_copies_harder
)
466 /* Would we bust the limit if we were running under -C? */
467 for (num_src
= i
= 0; i
< rename_src_nr
; i
++) {
468 if (diff_unmodified_pair(rename_src
[i
].p
))
472 if ((num_create
<= rename_limit
|| num_src
<= rename_limit
) &&
473 ((uint64_t)num_create
* (uint64_t)num_src
474 <= (uint64_t)rename_limit
* (uint64_t)rename_limit
))
479 static int find_renames(struct diff_score
*mx
, int dst_cnt
, int minimum_score
, int copies
)
483 for (i
= 0; i
< dst_cnt
* NUM_CANDIDATE_PER_DST
; i
++) {
484 struct diff_rename_dst
*dst
;
486 if ((mx
[i
].dst
< 0) ||
487 (mx
[i
].score
< minimum_score
))
488 break; /* there is no more usable pair. */
489 dst
= &rename_dst
[mx
[i
].dst
];
491 continue; /* already done, either exact or fuzzy. */
492 if (!copies
&& rename_src
[mx
[i
].src
].p
->one
->rename_used
)
494 record_rename_pair(mx
[i
].dst
, mx
[i
].src
, mx
[i
].score
);
500 void diffcore_rename(struct diff_options
*options
)
502 int detect_rename
= options
->detect_rename
;
503 int minimum_score
= options
->rename_score
;
504 struct diff_queue_struct
*q
= &diff_queued_diff
;
505 struct diff_queue_struct outq
;
506 struct diff_score
*mx
;
507 int i
, j
, rename_count
, skip_unmodified
= 0;
508 int num_create
, dst_cnt
;
509 struct progress
*progress
= NULL
;
512 minimum_score
= DEFAULT_RENAME_SCORE
;
514 for (i
= 0; i
< q
->nr
; i
++) {
515 struct diff_filepair
*p
= q
->queue
[i
];
516 if (!DIFF_FILE_VALID(p
->one
)) {
517 if (!DIFF_FILE_VALID(p
->two
))
518 continue; /* unmerged */
519 else if (options
->single_follow
&&
520 strcmp(options
->single_follow
, p
->two
->path
))
521 continue; /* not interested */
522 else if (!options
->flags
.rename_empty
&&
523 is_empty_blob_oid(&p
->two
->oid
))
525 else if (add_rename_dst(p
->two
) < 0) {
526 warning("skipping rename detection, detected"
527 " duplicate destination '%s'",
532 else if (!options
->flags
.rename_empty
&&
533 is_empty_blob_oid(&p
->one
->oid
))
535 else if (!DIFF_PAIR_UNMERGED(p
) && !DIFF_FILE_VALID(p
->two
)) {
537 * If the source is a broken "delete", and
538 * they did not really want to get broken,
539 * that means the source actually stays.
540 * So we increment the "rename_used" score
541 * by one, to indicate ourselves as a user
543 if (p
->broken_pair
&& !p
->score
)
544 p
->one
->rename_used
++;
545 register_rename_src(p
);
547 else if (detect_rename
== DIFF_DETECT_COPY
) {
549 * Increment the "rename_used" score by
550 * one, to indicate ourselves as a user.
552 p
->one
->rename_used
++;
553 register_rename_src(p
);
556 if (rename_dst_nr
== 0 || rename_src_nr
== 0)
557 goto cleanup
; /* nothing to do */
560 * We really want to cull the candidates list early
561 * with cheap tests in order to avoid doing deltas.
563 rename_count
= find_exact_renames(options
);
565 /* Did we only want exact renames? */
566 if (minimum_score
== MAX_SCORE
)
570 * Calculate how many renames are left (but all the source
571 * files still remain as options for rename/copies!)
573 num_create
= (rename_dst_nr
- rename_count
);
579 switch (too_many_rename_candidates(num_create
, options
)) {
583 options
->degraded_cc_to_c
= 1;
590 if (options
->show_rename_progress
) {
591 progress
= start_delayed_progress(
592 _("Performing inexact rename detection"),
593 (uint64_t)rename_dst_nr
* (uint64_t)rename_src_nr
);
596 mx
= xcalloc(st_mult(NUM_CANDIDATE_PER_DST
, num_create
), sizeof(*mx
));
597 for (dst_cnt
= i
= 0; i
< rename_dst_nr
; i
++) {
598 struct diff_filespec
*two
= rename_dst
[i
].two
;
599 struct diff_score
*m
;
601 if (rename_dst
[i
].pair
)
602 continue; /* dealt with exact match already. */
604 m
= &mx
[dst_cnt
* NUM_CANDIDATE_PER_DST
];
605 for (j
= 0; j
< NUM_CANDIDATE_PER_DST
; j
++)
608 for (j
= 0; j
< rename_src_nr
; j
++) {
609 struct diff_filespec
*one
= rename_src
[j
].p
->one
;
610 struct diff_score this_src
;
612 if (skip_unmodified
&&
613 diff_unmodified_pair(rename_src
[j
].p
))
616 this_src
.score
= estimate_similarity(options
->repo
,
620 this_src
.name_score
= basename_same(one
, two
);
623 record_if_better(m
, &this_src
);
625 * Once we run estimate_similarity,
626 * We do not need the text anymore.
628 diff_free_filespec_blob(one
);
629 diff_free_filespec_blob(two
);
632 display_progress(progress
, (uint64_t)(i
+1)*(uint64_t)rename_src_nr
);
634 stop_progress(&progress
);
636 /* cost matrix sorted by most to least similar pair */
637 STABLE_QSORT(mx
, dst_cnt
* NUM_CANDIDATE_PER_DST
, score_compare
);
639 rename_count
+= find_renames(mx
, dst_cnt
, minimum_score
, 0);
640 if (detect_rename
== DIFF_DETECT_COPY
)
641 rename_count
+= find_renames(mx
, dst_cnt
, minimum_score
, 1);
645 /* At this point, we have found some renames and copies and they
646 * are recorded in rename_dst. The original list is still in *q.
648 DIFF_QUEUE_CLEAR(&outq
);
649 for (i
= 0; i
< q
->nr
; i
++) {
650 struct diff_filepair
*p
= q
->queue
[i
];
651 struct diff_filepair
*pair_to_free
= NULL
;
653 if (DIFF_PAIR_UNMERGED(p
)) {
656 else if (!DIFF_FILE_VALID(p
->one
) && DIFF_FILE_VALID(p
->two
)) {
660 * We would output this create record if it has
661 * not been turned into a rename/copy already.
663 struct diff_rename_dst
*dst
= locate_rename_dst(p
->two
);
664 if (dst
&& dst
->pair
) {
665 diff_q(&outq
, dst
->pair
);
669 /* no matching rename/copy source, so
670 * record this as a creation.
674 else if (DIFF_FILE_VALID(p
->one
) && !DIFF_FILE_VALID(p
->two
)) {
678 * We would output this delete record if:
680 * (1) this is a broken delete and the counterpart
681 * broken create remains in the output; or
682 * (2) this is not a broken delete, and rename_dst
683 * does not have a rename/copy to move p->one->path
686 * Otherwise, the counterpart broken create
687 * has been turned into a rename-edit; or
688 * delete did not have a matching create to
691 if (DIFF_PAIR_BROKEN(p
)) {
693 struct diff_rename_dst
*dst
= locate_rename_dst(p
->one
);
694 if (dst
&& dst
->pair
)
695 /* counterpart is now rename/copy */
699 if (p
->one
->rename_used
)
700 /* this path remains */
709 else if (!diff_unmodified_pair(p
))
710 /* all the usual ones need to be kept */
713 /* no need to keep unmodified pairs */
717 diff_free_filepair(pair_to_free
);
719 diff_debug_queue("done copying original", &outq
);
723 diff_debug_queue("done collapsing", q
);
725 for (i
= 0; i
< rename_dst_nr
; i
++)
726 free_filespec(rename_dst
[i
].two
);
728 FREE_AND_NULL(rename_dst
);
729 rename_dst_nr
= rename_dst_alloc
= 0;
730 FREE_AND_NULL(rename_src
);
731 rename_src_nr
= rename_src_alloc
= 0;