2 * Copyright (C) 2005 Junio C Hamano
7 #include "object-store.h"
11 /* Table of rename/copy destinations */
13 static struct diff_rename_dst
{
14 struct diff_filespec
*two
;
15 struct diff_filepair
*pair
;
17 static int rename_dst_nr
, rename_dst_alloc
;
19 static int find_rename_dst(struct diff_filespec
*two
)
25 while (last
> first
) {
26 int next
= (last
+ first
) >> 1;
27 struct diff_rename_dst
*dst
= &(rename_dst
[next
]);
28 int cmp
= strcmp(two
->path
, dst
->two
->path
);
40 static struct diff_rename_dst
*locate_rename_dst(struct diff_filespec
*two
)
42 int ofs
= find_rename_dst(two
);
43 return ofs
< 0 ? NULL
: &rename_dst
[ofs
];
47 * Returns 0 on success, -1 if we found a duplicate.
49 static int add_rename_dst(struct diff_filespec
*two
)
51 int first
= find_rename_dst(two
);
57 /* insert to make it at "first" */
58 ALLOC_GROW(rename_dst
, rename_dst_nr
+ 1, rename_dst_alloc
);
60 if (first
< rename_dst_nr
)
61 MOVE_ARRAY(rename_dst
+ first
+ 1, rename_dst
+ first
,
62 rename_dst_nr
- first
- 1);
63 rename_dst
[first
].two
= alloc_filespec(two
->path
);
64 fill_filespec(rename_dst
[first
].two
, &two
->oid
, two
->oid_valid
,
66 rename_dst
[first
].pair
= NULL
;
70 /* Table of rename/copy src files */
71 static struct diff_rename_src
{
72 struct diff_filepair
*p
;
73 unsigned short score
; /* to remember the break score */
75 static int rename_src_nr
, rename_src_alloc
;
77 static struct diff_rename_src
*register_rename_src(struct diff_filepair
*p
)
80 struct diff_filespec
*one
= p
->one
;
81 unsigned short score
= p
->score
;
85 while (last
> first
) {
86 int next
= (last
+ first
) >> 1;
87 struct diff_rename_src
*src
= &(rename_src
[next
]);
88 int cmp
= strcmp(one
->path
, src
->p
->one
->path
);
98 /* insert to make it at "first" */
99 ALLOC_GROW(rename_src
, rename_src_nr
+ 1, rename_src_alloc
);
101 if (first
< rename_src_nr
)
102 MOVE_ARRAY(rename_src
+ first
+ 1, rename_src
+ first
,
103 rename_src_nr
- first
- 1);
104 rename_src
[first
].p
= p
;
105 rename_src
[first
].score
= score
;
106 return &(rename_src
[first
]);
109 static int basename_same(struct diff_filespec
*src
, struct diff_filespec
*dst
)
111 int src_len
= strlen(src
->path
), dst_len
= strlen(dst
->path
);
112 while (src_len
&& dst_len
) {
113 char c1
= src
->path
[--src_len
];
114 char c2
= dst
->path
[--dst_len
];
120 return (!src_len
|| src
->path
[src_len
- 1] == '/') &&
121 (!dst_len
|| dst
->path
[dst_len
- 1] == '/');
125 int src
; /* index in rename_src */
126 int dst
; /* index in rename_dst */
127 unsigned short score
;
131 static int estimate_similarity(struct diff_filespec
*src
,
132 struct diff_filespec
*dst
,
135 /* src points at a file that existed in the original tree (or
136 * optionally a file in the destination tree) and dst points
137 * at a newly created file. They may be quite similar, in which
138 * case we want to say src is renamed to dst or src is copied into
139 * dst, and then some edit has been applied to dst.
141 * Compare them and return how similar they are, representing
142 * the score as an integer between 0 and MAX_SCORE.
144 * When there is an exact match, it is considered a better
145 * match than anything else; the destination does not even
146 * call into this function in that case.
148 unsigned long max_size
, delta_size
, base_size
, src_copied
, literal_added
;
151 /* We deal only with regular files. Symlink renames are handled
152 * only when they are exact matches --- in other words, no edits
155 if (!S_ISREG(src
->mode
) || !S_ISREG(dst
->mode
))
159 * Need to check that source and destination sizes are
160 * filled in before comparing them.
162 * If we already have "cnt_data" filled in, we know it's
163 * all good (avoid checking the size for zero, as that
164 * is a possible size - we really should have a flag to
165 * say whether the size is valid or not!)
167 if (!src
->cnt_data
&&
168 diff_populate_filespec(src
, CHECK_SIZE_ONLY
))
170 if (!dst
->cnt_data
&&
171 diff_populate_filespec(dst
, CHECK_SIZE_ONLY
))
174 max_size
= ((src
->size
> dst
->size
) ? src
->size
: dst
->size
);
175 base_size
= ((src
->size
< dst
->size
) ? src
->size
: dst
->size
);
176 delta_size
= max_size
- base_size
;
178 /* We would not consider edits that change the file size so
179 * drastically. delta_size must be smaller than
180 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
182 * Note that base_size == 0 case is handled here already
183 * and the final score computation below would not have a
184 * divide-by-zero issue.
186 if (max_size
* (MAX_SCORE
-minimum_score
) < delta_size
* MAX_SCORE
)
189 if (!src
->cnt_data
&& diff_populate_filespec(src
, 0))
191 if (!dst
->cnt_data
&& diff_populate_filespec(dst
, 0))
194 if (diffcore_count_changes(src
, dst
,
195 &src
->cnt_data
, &dst
->cnt_data
,
196 &src_copied
, &literal_added
))
199 /* How similar are they?
200 * what percentage of material in dst are from source?
203 score
= 0; /* should not happen */
205 score
= (int)(src_copied
* MAX_SCORE
/ max_size
);
209 static void record_rename_pair(int dst_index
, int src_index
, int score
)
211 struct diff_filespec
*src
, *dst
;
212 struct diff_filepair
*dp
;
214 if (rename_dst
[dst_index
].pair
)
215 die("internal error: dst already matched.");
217 src
= rename_src
[src_index
].p
->one
;
221 dst
= rename_dst
[dst_index
].two
;
224 dp
= diff_queue(NULL
, src
, dst
);
225 dp
->renamed_pair
= 1;
226 if (!strcmp(src
->path
, dst
->path
))
227 dp
->score
= rename_src
[src_index
].score
;
230 rename_dst
[dst_index
].pair
= dp
;
234 * We sort the rename similarity matrix with the score, in descending
235 * order (the most similar first).
237 static int score_compare(const void *a_
, const void *b_
)
239 const struct diff_score
*a
= a_
, *b
= b_
;
241 /* sink the unused ones to the bottom */
243 return (0 <= b
->dst
);
247 if (a
->score
== b
->score
)
248 return b
->name_score
- a
->name_score
;
250 return b
->score
- a
->score
;
253 struct file_similarity
{
254 struct hashmap_entry entry
;
256 struct diff_filespec
*filespec
;
259 static unsigned int hash_filespec(struct diff_filespec
*filespec
)
261 if (!filespec
->oid_valid
) {
262 if (diff_populate_filespec(filespec
, 0))
264 hash_object_file(filespec
->data
, filespec
->size
, "blob",
267 return sha1hash(filespec
->oid
.hash
);
270 static int find_identical_files(struct hashmap
*srcs
,
272 struct diff_options
*options
)
276 struct diff_filespec
*target
= rename_dst
[dst_index
].two
;
277 struct file_similarity
*p
, *best
= NULL
;
278 int i
= 100, best_score
= -1;
281 * Find the best source match for specified destination.
283 p
= hashmap_get_from_hash(srcs
, hash_filespec(target
), NULL
);
284 for (; p
; p
= hashmap_get_next(srcs
, p
)) {
286 struct diff_filespec
*source
= p
->filespec
;
288 /* False hash collision? */
289 if (oidcmp(&source
->oid
, &target
->oid
))
291 /* Non-regular files? If so, the modes must match! */
292 if (!S_ISREG(source
->mode
) || !S_ISREG(target
->mode
)) {
293 if (source
->mode
!= target
->mode
)
296 /* Give higher scores to sources that haven't been used already */
297 score
= !source
->rename_used
;
298 if (source
->rename_used
&& options
->detect_rename
!= DIFF_DETECT_COPY
)
300 score
+= basename_same(source
, target
);
301 if (score
> best_score
) {
308 /* Too many identical alternatives? Pick one */
313 record_rename_pair(dst_index
, best
->index
, MAX_SCORE
);
319 static void insert_file_table(struct hashmap
*table
, int index
, struct diff_filespec
*filespec
)
321 struct file_similarity
*entry
= xmalloc(sizeof(*entry
));
323 entry
->index
= index
;
324 entry
->filespec
= filespec
;
326 hashmap_entry_init(entry
, hash_filespec(filespec
));
327 hashmap_add(table
, entry
);
331 * Find exact renames first.
333 * The first round matches up the up-to-date entries,
334 * and then during the second round we try to match
335 * cache-dirty entries as well.
337 static int find_exact_renames(struct diff_options
*options
)
340 struct hashmap file_table
;
342 /* Add all sources to the hash table in reverse order, because
343 * later on they will be retrieved in LIFO order.
345 hashmap_init(&file_table
, NULL
, NULL
, rename_src_nr
);
346 for (i
= rename_src_nr
-1; i
>= 0; i
--)
347 insert_file_table(&file_table
, i
, rename_src
[i
].p
->one
);
349 /* Walk the destinations and find best source match */
350 for (i
= 0; i
< rename_dst_nr
; i
++)
351 renames
+= find_identical_files(&file_table
, i
, options
);
353 /* Free the hash data structure and entries */
354 hashmap_free(&file_table
, 1);
359 #define NUM_CANDIDATE_PER_DST 4
360 static void record_if_better(struct diff_score m
[], struct diff_score
*o
)
364 /* find the worst one */
366 for (i
= 1; i
< NUM_CANDIDATE_PER_DST
; i
++)
367 if (score_compare(&m
[i
], &m
[worst
]) > 0)
370 /* is it better than the worst one? */
371 if (score_compare(&m
[worst
], o
) > 0)
377 * 0 if we are under the limit;
378 * 1 if we need to disable inexact rename detection;
379 * 2 if we would be under the limit if we were given -C instead of -C -C.
381 static int too_many_rename_candidates(int num_create
,
382 struct diff_options
*options
)
384 int rename_limit
= options
->rename_limit
;
385 int num_src
= rename_src_nr
;
388 options
->needed_rename_limit
= 0;
391 * This basically does a test for the rename matrix not
392 * growing larger than a "rename_limit" square matrix, ie:
394 * num_create * num_src > rename_limit * rename_limit
396 if (rename_limit
<= 0)
397 rename_limit
= 32767;
398 if ((num_create
<= rename_limit
|| num_src
<= rename_limit
) &&
399 ((uint64_t)num_create
* (uint64_t)num_src
400 <= (uint64_t)rename_limit
* (uint64_t)rename_limit
))
403 options
->needed_rename_limit
=
404 num_src
> num_create
? num_src
: num_create
;
406 /* Are we running under -C -C? */
407 if (!options
->flags
.find_copies_harder
)
410 /* Would we bust the limit if we were running under -C? */
411 for (num_src
= i
= 0; i
< rename_src_nr
; i
++) {
412 if (diff_unmodified_pair(rename_src
[i
].p
))
416 if ((num_create
<= rename_limit
|| num_src
<= rename_limit
) &&
417 ((uint64_t)num_create
* (uint64_t)num_src
418 <= (uint64_t)rename_limit
* (uint64_t)rename_limit
))
423 static int find_renames(struct diff_score
*mx
, int dst_cnt
, int minimum_score
, int copies
)
427 for (i
= 0; i
< dst_cnt
* NUM_CANDIDATE_PER_DST
; i
++) {
428 struct diff_rename_dst
*dst
;
430 if ((mx
[i
].dst
< 0) ||
431 (mx
[i
].score
< minimum_score
))
432 break; /* there is no more usable pair. */
433 dst
= &rename_dst
[mx
[i
].dst
];
435 continue; /* already done, either exact or fuzzy. */
436 if (!copies
&& rename_src
[mx
[i
].src
].p
->one
->rename_used
)
438 record_rename_pair(mx
[i
].dst
, mx
[i
].src
, mx
[i
].score
);
444 void diffcore_rename(struct diff_options
*options
)
446 int detect_rename
= options
->detect_rename
;
447 int minimum_score
= options
->rename_score
;
448 struct diff_queue_struct
*q
= &diff_queued_diff
;
449 struct diff_queue_struct outq
;
450 struct diff_score
*mx
;
451 int i
, j
, rename_count
, skip_unmodified
= 0;
452 int num_create
, dst_cnt
;
453 struct progress
*progress
= NULL
;
456 minimum_score
= DEFAULT_RENAME_SCORE
;
458 for (i
= 0; i
< q
->nr
; i
++) {
459 struct diff_filepair
*p
= q
->queue
[i
];
460 if (!DIFF_FILE_VALID(p
->one
)) {
461 if (!DIFF_FILE_VALID(p
->two
))
462 continue; /* unmerged */
463 else if (options
->single_follow
&&
464 strcmp(options
->single_follow
, p
->two
->path
))
465 continue; /* not interested */
466 else if (!options
->flags
.rename_empty
&&
467 is_empty_blob_oid(&p
->two
->oid
))
469 else if (add_rename_dst(p
->two
) < 0) {
470 warning("skipping rename detection, detected"
471 " duplicate destination '%s'",
476 else if (!options
->flags
.rename_empty
&&
477 is_empty_blob_oid(&p
->one
->oid
))
479 else if (!DIFF_PAIR_UNMERGED(p
) && !DIFF_FILE_VALID(p
->two
)) {
481 * If the source is a broken "delete", and
482 * they did not really want to get broken,
483 * that means the source actually stays.
484 * So we increment the "rename_used" score
485 * by one, to indicate ourselves as a user
487 if (p
->broken_pair
&& !p
->score
)
488 p
->one
->rename_used
++;
489 register_rename_src(p
);
491 else if (detect_rename
== DIFF_DETECT_COPY
) {
493 * Increment the "rename_used" score by
494 * one, to indicate ourselves as a user.
496 p
->one
->rename_used
++;
497 register_rename_src(p
);
500 if (rename_dst_nr
== 0 || rename_src_nr
== 0)
501 goto cleanup
; /* nothing to do */
504 * We really want to cull the candidates list early
505 * with cheap tests in order to avoid doing deltas.
507 rename_count
= find_exact_renames(options
);
509 /* Did we only want exact renames? */
510 if (minimum_score
== MAX_SCORE
)
514 * Calculate how many renames are left (but all the source
515 * files still remain as options for rename/copies!)
517 num_create
= (rename_dst_nr
- rename_count
);
523 switch (too_many_rename_candidates(num_create
, options
)) {
527 options
->degraded_cc_to_c
= 1;
534 if (options
->show_rename_progress
) {
535 progress
= start_delayed_progress(
536 _("Performing inexact rename detection"),
537 (uint64_t)rename_dst_nr
* (uint64_t)rename_src_nr
);
540 mx
= xcalloc(st_mult(NUM_CANDIDATE_PER_DST
, num_create
), sizeof(*mx
));
541 for (dst_cnt
= i
= 0; i
< rename_dst_nr
; i
++) {
542 struct diff_filespec
*two
= rename_dst
[i
].two
;
543 struct diff_score
*m
;
545 if (rename_dst
[i
].pair
)
546 continue; /* dealt with exact match already. */
548 m
= &mx
[dst_cnt
* NUM_CANDIDATE_PER_DST
];
549 for (j
= 0; j
< NUM_CANDIDATE_PER_DST
; j
++)
552 for (j
= 0; j
< rename_src_nr
; j
++) {
553 struct diff_filespec
*one
= rename_src
[j
].p
->one
;
554 struct diff_score this_src
;
556 if (skip_unmodified
&&
557 diff_unmodified_pair(rename_src
[j
].p
))
560 this_src
.score
= estimate_similarity(one
, two
,
562 this_src
.name_score
= basename_same(one
, two
);
565 record_if_better(m
, &this_src
);
567 * Once we run estimate_similarity,
568 * We do not need the text anymore.
570 diff_free_filespec_blob(one
);
571 diff_free_filespec_blob(two
);
574 display_progress(progress
, (uint64_t)(i
+1)*(uint64_t)rename_src_nr
);
576 stop_progress(&progress
);
578 /* cost matrix sorted by most to least similar pair */
579 QSORT(mx
, dst_cnt
* NUM_CANDIDATE_PER_DST
, score_compare
);
581 rename_count
+= find_renames(mx
, dst_cnt
, minimum_score
, 0);
582 if (detect_rename
== DIFF_DETECT_COPY
)
583 rename_count
+= find_renames(mx
, dst_cnt
, minimum_score
, 1);
587 /* At this point, we have found some renames and copies and they
588 * are recorded in rename_dst. The original list is still in *q.
590 DIFF_QUEUE_CLEAR(&outq
);
591 for (i
= 0; i
< q
->nr
; i
++) {
592 struct diff_filepair
*p
= q
->queue
[i
];
593 struct diff_filepair
*pair_to_free
= NULL
;
595 if (DIFF_PAIR_UNMERGED(p
)) {
598 else if (!DIFF_FILE_VALID(p
->one
) && DIFF_FILE_VALID(p
->two
)) {
602 * We would output this create record if it has
603 * not been turned into a rename/copy already.
605 struct diff_rename_dst
*dst
= locate_rename_dst(p
->two
);
606 if (dst
&& dst
->pair
) {
607 diff_q(&outq
, dst
->pair
);
611 /* no matching rename/copy source, so
612 * record this as a creation.
616 else if (DIFF_FILE_VALID(p
->one
) && !DIFF_FILE_VALID(p
->two
)) {
620 * We would output this delete record if:
622 * (1) this is a broken delete and the counterpart
623 * broken create remains in the output; or
624 * (2) this is not a broken delete, and rename_dst
625 * does not have a rename/copy to move p->one->path
628 * Otherwise, the counterpart broken create
629 * has been turned into a rename-edit; or
630 * delete did not have a matching create to
633 if (DIFF_PAIR_BROKEN(p
)) {
635 struct diff_rename_dst
*dst
= locate_rename_dst(p
->one
);
636 if (dst
&& dst
->pair
)
637 /* counterpart is now rename/copy */
641 if (p
->one
->rename_used
)
642 /* this path remains */
651 else if (!diff_unmodified_pair(p
))
652 /* all the usual ones need to be kept */
655 /* no need to keep unmodified pairs */
659 diff_free_filepair(pair_to_free
);
661 diff_debug_queue("done copying original", &outq
);
665 diff_debug_queue("done collapsing", q
);
667 for (i
= 0; i
< rename_dst_nr
; i
++)
668 free_filespec(rename_dst
[i
].two
);
670 FREE_AND_NULL(rename_dst
);
671 rename_dst_nr
= rename_dst_alloc
= 0;
672 FREE_AND_NULL(rename_src
);
673 rename_src_nr
= rename_src_alloc
= 0;