2 * Copyright (C) 2005 Junio C Hamano
10 /* Table of rename/copy destinations */
12 static struct diff_rename_dst
{
13 struct diff_filespec
*two
;
14 struct diff_filepair
*pair
;
16 static int rename_dst_nr
, rename_dst_alloc
;
18 static struct diff_rename_dst
*locate_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 /* insert to make it at "first" */
41 ALLOC_GROW(rename_dst
, rename_dst_nr
+ 1, rename_dst_alloc
);
43 if (first
< rename_dst_nr
)
44 memmove(rename_dst
+ first
+ 1, rename_dst
+ first
,
45 (rename_dst_nr
- first
- 1) * sizeof(*rename_dst
));
46 rename_dst
[first
].two
= alloc_filespec(two
->path
);
47 fill_filespec(rename_dst
[first
].two
, two
->sha1
, two
->sha1_valid
, two
->mode
);
48 rename_dst
[first
].pair
= NULL
;
49 return &(rename_dst
[first
]);
52 /* Table of rename/copy src files */
53 static struct diff_rename_src
{
54 struct diff_filepair
*p
;
55 unsigned short score
; /* to remember the break score */
57 static int rename_src_nr
, rename_src_alloc
;
59 static struct diff_rename_src
*register_rename_src(struct diff_filepair
*p
)
62 struct diff_filespec
*one
= p
->one
;
63 unsigned short score
= p
->score
;
67 while (last
> first
) {
68 int next
= (last
+ first
) >> 1;
69 struct diff_rename_src
*src
= &(rename_src
[next
]);
70 int cmp
= strcmp(one
->path
, src
->p
->one
->path
);
80 /* insert to make it at "first" */
81 ALLOC_GROW(rename_src
, rename_src_nr
+ 1, rename_src_alloc
);
83 if (first
< rename_src_nr
)
84 memmove(rename_src
+ first
+ 1, rename_src
+ first
,
85 (rename_src_nr
- first
- 1) * sizeof(*rename_src
));
86 rename_src
[first
].p
= p
;
87 rename_src
[first
].score
= score
;
88 return &(rename_src
[first
]);
91 static int basename_same(struct diff_filespec
*src
, struct diff_filespec
*dst
)
93 int src_len
= strlen(src
->path
), dst_len
= strlen(dst
->path
);
94 while (src_len
&& dst_len
) {
95 char c1
= src
->path
[--src_len
];
96 char c2
= dst
->path
[--dst_len
];
102 return (!src_len
|| src
->path
[src_len
- 1] == '/') &&
103 (!dst_len
|| dst
->path
[dst_len
- 1] == '/');
107 int src
; /* index in rename_src */
108 int dst
; /* index in rename_dst */
109 unsigned short score
;
113 static int estimate_similarity(struct diff_filespec
*src
,
114 struct diff_filespec
*dst
,
117 /* src points at a file that existed in the original tree (or
118 * optionally a file in the destination tree) and dst points
119 * at a newly created file. They may be quite similar, in which
120 * case we want to say src is renamed to dst or src is copied into
121 * dst, and then some edit has been applied to dst.
123 * Compare them and return how similar they are, representing
124 * the score as an integer between 0 and MAX_SCORE.
126 * When there is an exact match, it is considered a better
127 * match than anything else; the destination does not even
128 * call into this function in that case.
130 unsigned long max_size
, delta_size
, base_size
, src_copied
, literal_added
;
131 unsigned long delta_limit
;
134 /* We deal only with regular files. Symlink renames are handled
135 * only when they are exact matches --- in other words, no edits
138 if (!S_ISREG(src
->mode
) || !S_ISREG(dst
->mode
))
142 * Need to check that source and destination sizes are
143 * filled in before comparing them.
145 * If we already have "cnt_data" filled in, we know it's
146 * all good (avoid checking the size for zero, as that
147 * is a possible size - we really should have a flag to
148 * say whether the size is valid or not!)
150 if (!src
->cnt_data
&& diff_populate_filespec(src
, 1))
152 if (!dst
->cnt_data
&& diff_populate_filespec(dst
, 1))
155 max_size
= ((src
->size
> dst
->size
) ? src
->size
: dst
->size
);
156 base_size
= ((src
->size
< dst
->size
) ? src
->size
: dst
->size
);
157 delta_size
= max_size
- base_size
;
159 /* We would not consider edits that change the file size so
160 * drastically. delta_size must be smaller than
161 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
163 * Note that base_size == 0 case is handled here already
164 * and the final score computation below would not have a
165 * divide-by-zero issue.
167 if (max_size
* (MAX_SCORE
-minimum_score
) < delta_size
* MAX_SCORE
)
170 if (!src
->cnt_data
&& diff_populate_filespec(src
, 0))
172 if (!dst
->cnt_data
&& diff_populate_filespec(dst
, 0))
175 delta_limit
= (unsigned long)
176 (base_size
* (MAX_SCORE
-minimum_score
) / MAX_SCORE
);
177 if (diffcore_count_changes(src
, dst
,
178 &src
->cnt_data
, &dst
->cnt_data
,
180 &src_copied
, &literal_added
))
183 /* How similar are they?
184 * what percentage of material in dst are from source?
187 score
= 0; /* should not happen */
189 score
= (int)(src_copied
* MAX_SCORE
/ max_size
);
193 static void record_rename_pair(int dst_index
, int src_index
, int score
)
195 struct diff_filespec
*src
, *dst
;
196 struct diff_filepair
*dp
;
198 if (rename_dst
[dst_index
].pair
)
199 die("internal error: dst already matched.");
201 src
= rename_src
[src_index
].p
->one
;
205 dst
= rename_dst
[dst_index
].two
;
208 dp
= diff_queue(NULL
, src
, dst
);
209 dp
->renamed_pair
= 1;
210 if (!strcmp(src
->path
, dst
->path
))
211 dp
->score
= rename_src
[src_index
].score
;
214 rename_dst
[dst_index
].pair
= dp
;
218 * We sort the rename similarity matrix with the score, in descending
219 * order (the most similar first).
221 static int score_compare(const void *a_
, const void *b_
)
223 const struct diff_score
*a
= a_
, *b
= b_
;
225 /* sink the unused ones to the bottom */
227 return (0 <= b
->dst
);
231 if (a
->score
== b
->score
)
232 return b
->name_score
- a
->name_score
;
234 return b
->score
- a
->score
;
237 struct file_similarity
{
238 struct hashmap_entry entry
;
240 struct diff_filespec
*filespec
;
243 static unsigned int hash_filespec(struct diff_filespec
*filespec
)
246 if (!filespec
->sha1_valid
) {
247 if (diff_populate_filespec(filespec
, 0))
249 hash_sha1_file(filespec
->data
, filespec
->size
, "blob", filespec
->sha1
);
251 memcpy(&hash
, filespec
->sha1
, sizeof(hash
));
255 static int find_identical_files(struct hashmap
*srcs
,
257 struct diff_options
*options
)
261 struct diff_filespec
*target
= rename_dst
[dst_index
].two
;
262 struct file_similarity
*p
, *best
, dst
;
263 int i
= 100, best_score
= -1;
266 * Find the best source match for specified destination.
269 hashmap_entry_init(&dst
, hash_filespec(target
));
270 for (p
= hashmap_get(srcs
, &dst
, NULL
); p
; p
= hashmap_get_next(srcs
, p
)) {
272 struct diff_filespec
*source
= p
->filespec
;
274 /* False hash collision? */
275 if (hashcmp(source
->sha1
, target
->sha1
))
277 /* Non-regular files? If so, the modes must match! */
278 if (!S_ISREG(source
->mode
) || !S_ISREG(target
->mode
)) {
279 if (source
->mode
!= target
->mode
)
282 /* Give higher scores to sources that haven't been used already */
283 score
= !source
->rename_used
;
284 if (source
->rename_used
&& options
->detect_rename
!= DIFF_DETECT_COPY
)
286 score
+= basename_same(source
, target
);
287 if (score
> best_score
) {
294 /* Too many identical alternatives? Pick one */
299 record_rename_pair(dst_index
, best
->index
, MAX_SCORE
);
305 static void insert_file_table(struct hashmap
*table
, int index
, struct diff_filespec
*filespec
)
307 struct file_similarity
*entry
= xmalloc(sizeof(*entry
));
309 entry
->index
= index
;
310 entry
->filespec
= filespec
;
312 hashmap_entry_init(entry
, hash_filespec(filespec
));
313 hashmap_add(table
, entry
);
317 * Find exact renames first.
319 * The first round matches up the up-to-date entries,
320 * and then during the second round we try to match
321 * cache-dirty entries as well.
323 static int find_exact_renames(struct diff_options
*options
)
326 struct hashmap file_table
;
328 /* Add all sources to the hash table */
329 hashmap_init(&file_table
, NULL
, rename_src_nr
);
330 for (i
= 0; i
< rename_src_nr
; i
++)
331 insert_file_table(&file_table
, i
, rename_src
[i
].p
->one
);
333 /* Walk the destinations and find best source match */
334 for (i
= 0; i
< rename_dst_nr
; i
++)
335 renames
+= find_identical_files(&file_table
, i
, options
);
337 /* Free the hash data structure and entries */
338 hashmap_free(&file_table
, 1);
343 #define NUM_CANDIDATE_PER_DST 4
344 static void record_if_better(struct diff_score m
[], struct diff_score
*o
)
348 /* find the worst one */
350 for (i
= 1; i
< NUM_CANDIDATE_PER_DST
; i
++)
351 if (score_compare(&m
[i
], &m
[worst
]) > 0)
354 /* is it better than the worst one? */
355 if (score_compare(&m
[worst
], o
) > 0)
361 * 0 if we are under the limit;
362 * 1 if we need to disable inexact rename detection;
363 * 2 if we would be under the limit if we were given -C instead of -C -C.
365 static int too_many_rename_candidates(int num_create
,
366 struct diff_options
*options
)
368 int rename_limit
= options
->rename_limit
;
369 int num_src
= rename_src_nr
;
372 options
->needed_rename_limit
= 0;
375 * This basically does a test for the rename matrix not
376 * growing larger than a "rename_limit" square matrix, ie:
378 * num_create * num_src > rename_limit * rename_limit
380 * but handles the potential overflow case specially (and we
381 * assume at least 32-bit integers)
383 if (rename_limit
<= 0 || rename_limit
> 32767)
384 rename_limit
= 32767;
385 if ((num_create
<= rename_limit
|| num_src
<= rename_limit
) &&
386 (num_create
* num_src
<= rename_limit
* rename_limit
))
389 options
->needed_rename_limit
=
390 num_src
> num_create
? num_src
: num_create
;
392 /* Are we running under -C -C? */
393 if (!DIFF_OPT_TST(options
, FIND_COPIES_HARDER
))
396 /* Would we bust the limit if we were running under -C? */
397 for (num_src
= i
= 0; i
< rename_src_nr
; i
++) {
398 if (diff_unmodified_pair(rename_src
[i
].p
))
402 if ((num_create
<= rename_limit
|| num_src
<= rename_limit
) &&
403 (num_create
* num_src
<= rename_limit
* rename_limit
))
408 static int find_renames(struct diff_score
*mx
, int dst_cnt
, int minimum_score
, int copies
)
412 for (i
= 0; i
< dst_cnt
* NUM_CANDIDATE_PER_DST
; i
++) {
413 struct diff_rename_dst
*dst
;
415 if ((mx
[i
].dst
< 0) ||
416 (mx
[i
].score
< minimum_score
))
417 break; /* there is no more usable pair. */
418 dst
= &rename_dst
[mx
[i
].dst
];
420 continue; /* already done, either exact or fuzzy. */
421 if (!copies
&& rename_src
[mx
[i
].src
].p
->one
->rename_used
)
423 record_rename_pair(mx
[i
].dst
, mx
[i
].src
, mx
[i
].score
);
429 void diffcore_rename(struct diff_options
*options
)
431 int detect_rename
= options
->detect_rename
;
432 int minimum_score
= options
->rename_score
;
433 struct diff_queue_struct
*q
= &diff_queued_diff
;
434 struct diff_queue_struct outq
;
435 struct diff_score
*mx
;
436 int i
, j
, rename_count
, skip_unmodified
= 0;
437 int num_create
, dst_cnt
;
438 struct progress
*progress
= NULL
;
441 minimum_score
= DEFAULT_RENAME_SCORE
;
443 for (i
= 0; i
< q
->nr
; i
++) {
444 struct diff_filepair
*p
= q
->queue
[i
];
445 if (!DIFF_FILE_VALID(p
->one
)) {
446 if (!DIFF_FILE_VALID(p
->two
))
447 continue; /* unmerged */
448 else if (options
->single_follow
&&
449 strcmp(options
->single_follow
, p
->two
->path
))
450 continue; /* not interested */
451 else if (!DIFF_OPT_TST(options
, RENAME_EMPTY
) &&
452 is_empty_blob_sha1(p
->two
->sha1
))
455 locate_rename_dst(p
->two
, 1);
457 else if (!DIFF_OPT_TST(options
, RENAME_EMPTY
) &&
458 is_empty_blob_sha1(p
->one
->sha1
))
460 else if (!DIFF_PAIR_UNMERGED(p
) && !DIFF_FILE_VALID(p
->two
)) {
462 * If the source is a broken "delete", and
463 * they did not really want to get broken,
464 * that means the source actually stays.
465 * So we increment the "rename_used" score
466 * by one, to indicate ourselves as a user
468 if (p
->broken_pair
&& !p
->score
)
469 p
->one
->rename_used
++;
470 register_rename_src(p
);
472 else if (detect_rename
== DIFF_DETECT_COPY
) {
474 * Increment the "rename_used" score by
475 * one, to indicate ourselves as a user.
477 p
->one
->rename_used
++;
478 register_rename_src(p
);
481 if (rename_dst_nr
== 0 || rename_src_nr
== 0)
482 goto cleanup
; /* nothing to do */
485 * We really want to cull the candidates list early
486 * with cheap tests in order to avoid doing deltas.
488 rename_count
= find_exact_renames(options
);
490 /* Did we only want exact renames? */
491 if (minimum_score
== MAX_SCORE
)
495 * Calculate how many renames are left (but all the source
496 * files still remain as options for rename/copies!)
498 num_create
= (rename_dst_nr
- rename_count
);
504 switch (too_many_rename_candidates(num_create
, options
)) {
508 options
->degraded_cc_to_c
= 1;
515 if (options
->show_rename_progress
) {
516 progress
= start_progress_delay(
517 _("Performing inexact rename detection"),
518 rename_dst_nr
* rename_src_nr
, 50, 1);
521 mx
= xcalloc(num_create
* NUM_CANDIDATE_PER_DST
, sizeof(*mx
));
522 for (dst_cnt
= i
= 0; i
< rename_dst_nr
; i
++) {
523 struct diff_filespec
*two
= rename_dst
[i
].two
;
524 struct diff_score
*m
;
526 if (rename_dst
[i
].pair
)
527 continue; /* dealt with exact match already. */
529 m
= &mx
[dst_cnt
* NUM_CANDIDATE_PER_DST
];
530 for (j
= 0; j
< NUM_CANDIDATE_PER_DST
; j
++)
533 for (j
= 0; j
< rename_src_nr
; j
++) {
534 struct diff_filespec
*one
= rename_src
[j
].p
->one
;
535 struct diff_score this_src
;
537 if (skip_unmodified
&&
538 diff_unmodified_pair(rename_src
[j
].p
))
541 this_src
.score
= estimate_similarity(one
, two
,
543 this_src
.name_score
= basename_same(one
, two
);
546 record_if_better(m
, &this_src
);
548 * Once we run estimate_similarity,
549 * We do not need the text anymore.
551 diff_free_filespec_blob(one
);
552 diff_free_filespec_blob(two
);
555 display_progress(progress
, (i
+1)*rename_src_nr
);
557 stop_progress(&progress
);
559 /* cost matrix sorted by most to least similar pair */
560 qsort(mx
, dst_cnt
* NUM_CANDIDATE_PER_DST
, sizeof(*mx
), score_compare
);
562 rename_count
+= find_renames(mx
, dst_cnt
, minimum_score
, 0);
563 if (detect_rename
== DIFF_DETECT_COPY
)
564 rename_count
+= find_renames(mx
, dst_cnt
, minimum_score
, 1);
568 /* At this point, we have found some renames and copies and they
569 * are recorded in rename_dst. The original list is still in *q.
571 DIFF_QUEUE_CLEAR(&outq
);
572 for (i
= 0; i
< q
->nr
; i
++) {
573 struct diff_filepair
*p
= q
->queue
[i
];
574 struct diff_filepair
*pair_to_free
= NULL
;
576 if (DIFF_PAIR_UNMERGED(p
)) {
579 else if (!DIFF_FILE_VALID(p
->one
) && DIFF_FILE_VALID(p
->two
)) {
583 * We would output this create record if it has
584 * not been turned into a rename/copy already.
586 struct diff_rename_dst
*dst
=
587 locate_rename_dst(p
->two
, 0);
588 if (dst
&& dst
->pair
) {
589 diff_q(&outq
, dst
->pair
);
593 /* no matching rename/copy source, so
594 * record this as a creation.
598 else if (DIFF_FILE_VALID(p
->one
) && !DIFF_FILE_VALID(p
->two
)) {
602 * We would output this delete record if:
604 * (1) this is a broken delete and the counterpart
605 * broken create remains in the output; or
606 * (2) this is not a broken delete, and rename_dst
607 * does not have a rename/copy to move p->one->path
610 * Otherwise, the counterpart broken create
611 * has been turned into a rename-edit; or
612 * delete did not have a matching create to
615 if (DIFF_PAIR_BROKEN(p
)) {
617 struct diff_rename_dst
*dst
=
618 locate_rename_dst(p
->one
, 0);
619 if (dst
&& dst
->pair
)
620 /* counterpart is now rename/copy */
624 if (p
->one
->rename_used
)
625 /* this path remains */
634 else if (!diff_unmodified_pair(p
))
635 /* all the usual ones need to be kept */
638 /* no need to keep unmodified pairs */
642 diff_free_filepair(pair_to_free
);
644 diff_debug_queue("done copying original", &outq
);
648 diff_debug_queue("done collapsing", q
);
650 for (i
= 0; i
< rename_dst_nr
; i
++)
651 free_filespec(rename_dst
[i
].two
);
655 rename_dst_nr
= rename_dst_alloc
= 0;
658 rename_src_nr
= rename_src_alloc
= 0;