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
&&
151 diff_populate_filespec(src
, CHECK_SIZE_ONLY
))
153 if (!dst
->cnt_data
&&
154 diff_populate_filespec(dst
, CHECK_SIZE_ONLY
))
157 max_size
= ((src
->size
> dst
->size
) ? src
->size
: dst
->size
);
158 base_size
= ((src
->size
< dst
->size
) ? src
->size
: dst
->size
);
159 delta_size
= max_size
- base_size
;
161 /* We would not consider edits that change the file size so
162 * drastically. delta_size must be smaller than
163 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
165 * Note that base_size == 0 case is handled here already
166 * and the final score computation below would not have a
167 * divide-by-zero issue.
169 if (max_size
* (MAX_SCORE
-minimum_score
) < delta_size
* MAX_SCORE
)
172 if (!src
->cnt_data
&& diff_populate_filespec(src
, 0))
174 if (!dst
->cnt_data
&& diff_populate_filespec(dst
, 0))
177 delta_limit
= (unsigned long)
178 (base_size
* (MAX_SCORE
-minimum_score
) / MAX_SCORE
);
179 if (diffcore_count_changes(src
, dst
,
180 &src
->cnt_data
, &dst
->cnt_data
,
182 &src_copied
, &literal_added
))
185 /* How similar are they?
186 * what percentage of material in dst are from source?
189 score
= 0; /* should not happen */
191 score
= (int)(src_copied
* MAX_SCORE
/ max_size
);
195 static void record_rename_pair(int dst_index
, int src_index
, int score
)
197 struct diff_filespec
*src
, *dst
;
198 struct diff_filepair
*dp
;
200 if (rename_dst
[dst_index
].pair
)
201 die("internal error: dst already matched.");
203 src
= rename_src
[src_index
].p
->one
;
207 dst
= rename_dst
[dst_index
].two
;
210 dp
= diff_queue(NULL
, src
, dst
);
211 dp
->renamed_pair
= 1;
212 if (!strcmp(src
->path
, dst
->path
))
213 dp
->score
= rename_src
[src_index
].score
;
216 rename_dst
[dst_index
].pair
= dp
;
220 * We sort the rename similarity matrix with the score, in descending
221 * order (the most similar first).
223 static int score_compare(const void *a_
, const void *b_
)
225 const struct diff_score
*a
= a_
, *b
= b_
;
227 /* sink the unused ones to the bottom */
229 return (0 <= b
->dst
);
233 if (a
->score
== b
->score
)
234 return b
->name_score
- a
->name_score
;
236 return b
->score
- a
->score
;
239 struct file_similarity
{
240 struct hashmap_entry entry
;
242 struct diff_filespec
*filespec
;
245 static unsigned int hash_filespec(struct diff_filespec
*filespec
)
247 if (!filespec
->sha1_valid
) {
248 if (diff_populate_filespec(filespec
, 0))
250 hash_sha1_file(filespec
->data
, filespec
->size
, "blob", filespec
->sha1
);
252 return sha1hash(filespec
->sha1
);
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
= NULL
;
263 int i
= 100, best_score
= -1;
266 * Find the best source match for specified destination.
268 p
= hashmap_get_from_hash(srcs
, hash_filespec(target
), NULL
);
269 for (; p
; p
= hashmap_get_next(srcs
, p
)) {
271 struct diff_filespec
*source
= p
->filespec
;
273 /* False hash collision? */
274 if (hashcmp(source
->sha1
, target
->sha1
))
276 /* Non-regular files? If so, the modes must match! */
277 if (!S_ISREG(source
->mode
) || !S_ISREG(target
->mode
)) {
278 if (source
->mode
!= target
->mode
)
281 /* Give higher scores to sources that haven't been used already */
282 score
= !source
->rename_used
;
283 if (source
->rename_used
&& options
->detect_rename
!= DIFF_DETECT_COPY
)
285 score
+= basename_same(source
, target
);
286 if (score
> best_score
) {
293 /* Too many identical alternatives? Pick one */
298 record_rename_pair(dst_index
, best
->index
, MAX_SCORE
);
304 static void insert_file_table(struct hashmap
*table
, int index
, struct diff_filespec
*filespec
)
306 struct file_similarity
*entry
= xmalloc(sizeof(*entry
));
308 entry
->index
= index
;
309 entry
->filespec
= filespec
;
311 hashmap_entry_init(entry
, hash_filespec(filespec
));
312 hashmap_add(table
, entry
);
316 * Find exact renames first.
318 * The first round matches up the up-to-date entries,
319 * and then during the second round we try to match
320 * cache-dirty entries as well.
322 static int find_exact_renames(struct diff_options
*options
)
325 struct hashmap file_table
;
327 /* Add all sources to the hash table */
328 hashmap_init(&file_table
, NULL
, rename_src_nr
);
329 for (i
= 0; i
< rename_src_nr
; i
++)
330 insert_file_table(&file_table
, i
, rename_src
[i
].p
->one
);
332 /* Walk the destinations and find best source match */
333 for (i
= 0; i
< rename_dst_nr
; i
++)
334 renames
+= find_identical_files(&file_table
, i
, options
);
336 /* Free the hash data structure and entries */
337 hashmap_free(&file_table
, 1);
342 #define NUM_CANDIDATE_PER_DST 4
343 static void record_if_better(struct diff_score m
[], struct diff_score
*o
)
347 /* find the worst one */
349 for (i
= 1; i
< NUM_CANDIDATE_PER_DST
; i
++)
350 if (score_compare(&m
[i
], &m
[worst
]) > 0)
353 /* is it better than the worst one? */
354 if (score_compare(&m
[worst
], o
) > 0)
360 * 0 if we are under the limit;
361 * 1 if we need to disable inexact rename detection;
362 * 2 if we would be under the limit if we were given -C instead of -C -C.
364 static int too_many_rename_candidates(int num_create
,
365 struct diff_options
*options
)
367 int rename_limit
= options
->rename_limit
;
368 int num_src
= rename_src_nr
;
371 options
->needed_rename_limit
= 0;
374 * This basically does a test for the rename matrix not
375 * growing larger than a "rename_limit" square matrix, ie:
377 * num_create * num_src > rename_limit * rename_limit
379 * but handles the potential overflow case specially (and we
380 * assume at least 32-bit integers)
382 if (rename_limit
<= 0 || rename_limit
> 32767)
383 rename_limit
= 32767;
384 if ((num_create
<= rename_limit
|| num_src
<= rename_limit
) &&
385 (num_create
* num_src
<= rename_limit
* rename_limit
))
388 options
->needed_rename_limit
=
389 num_src
> num_create
? num_src
: num_create
;
391 /* Are we running under -C -C? */
392 if (!DIFF_OPT_TST(options
, FIND_COPIES_HARDER
))
395 /* Would we bust the limit if we were running under -C? */
396 for (num_src
= i
= 0; i
< rename_src_nr
; i
++) {
397 if (diff_unmodified_pair(rename_src
[i
].p
))
401 if ((num_create
<= rename_limit
|| num_src
<= rename_limit
) &&
402 (num_create
* num_src
<= rename_limit
* rename_limit
))
407 static int find_renames(struct diff_score
*mx
, int dst_cnt
, int minimum_score
, int copies
)
411 for (i
= 0; i
< dst_cnt
* NUM_CANDIDATE_PER_DST
; i
++) {
412 struct diff_rename_dst
*dst
;
414 if ((mx
[i
].dst
< 0) ||
415 (mx
[i
].score
< minimum_score
))
416 break; /* there is no more usable pair. */
417 dst
= &rename_dst
[mx
[i
].dst
];
419 continue; /* already done, either exact or fuzzy. */
420 if (!copies
&& rename_src
[mx
[i
].src
].p
->one
->rename_used
)
422 record_rename_pair(mx
[i
].dst
, mx
[i
].src
, mx
[i
].score
);
428 void diffcore_rename(struct diff_options
*options
)
430 int detect_rename
= options
->detect_rename
;
431 int minimum_score
= options
->rename_score
;
432 struct diff_queue_struct
*q
= &diff_queued_diff
;
433 struct diff_queue_struct outq
;
434 struct diff_score
*mx
;
435 int i
, j
, rename_count
, skip_unmodified
= 0;
436 int num_create
, dst_cnt
;
437 struct progress
*progress
= NULL
;
440 minimum_score
= DEFAULT_RENAME_SCORE
;
442 for (i
= 0; i
< q
->nr
; i
++) {
443 struct diff_filepair
*p
= q
->queue
[i
];
444 if (!DIFF_FILE_VALID(p
->one
)) {
445 if (!DIFF_FILE_VALID(p
->two
))
446 continue; /* unmerged */
447 else if (options
->single_follow
&&
448 strcmp(options
->single_follow
, p
->two
->path
))
449 continue; /* not interested */
450 else if (!DIFF_OPT_TST(options
, RENAME_EMPTY
) &&
451 is_empty_blob_sha1(p
->two
->sha1
))
454 locate_rename_dst(p
->two
, 1);
456 else if (!DIFF_OPT_TST(options
, RENAME_EMPTY
) &&
457 is_empty_blob_sha1(p
->one
->sha1
))
459 else if (!DIFF_PAIR_UNMERGED(p
) && !DIFF_FILE_VALID(p
->two
)) {
461 * If the source is a broken "delete", and
462 * they did not really want to get broken,
463 * that means the source actually stays.
464 * So we increment the "rename_used" score
465 * by one, to indicate ourselves as a user
467 if (p
->broken_pair
&& !p
->score
)
468 p
->one
->rename_used
++;
469 register_rename_src(p
);
471 else if (detect_rename
== DIFF_DETECT_COPY
) {
473 * Increment the "rename_used" score by
474 * one, to indicate ourselves as a user.
476 p
->one
->rename_used
++;
477 register_rename_src(p
);
480 if (rename_dst_nr
== 0 || rename_src_nr
== 0)
481 goto cleanup
; /* nothing to do */
484 * We really want to cull the candidates list early
485 * with cheap tests in order to avoid doing deltas.
487 rename_count
= find_exact_renames(options
);
489 /* Did we only want exact renames? */
490 if (minimum_score
== MAX_SCORE
)
494 * Calculate how many renames are left (but all the source
495 * files still remain as options for rename/copies!)
497 num_create
= (rename_dst_nr
- rename_count
);
503 switch (too_many_rename_candidates(num_create
, options
)) {
507 options
->degraded_cc_to_c
= 1;
514 if (options
->show_rename_progress
) {
515 progress
= start_progress_delay(
516 _("Performing inexact rename detection"),
517 rename_dst_nr
* rename_src_nr
, 50, 1);
520 mx
= xcalloc(num_create
* NUM_CANDIDATE_PER_DST
, sizeof(*mx
));
521 for (dst_cnt
= i
= 0; i
< rename_dst_nr
; i
++) {
522 struct diff_filespec
*two
= rename_dst
[i
].two
;
523 struct diff_score
*m
;
525 if (rename_dst
[i
].pair
)
526 continue; /* dealt with exact match already. */
528 m
= &mx
[dst_cnt
* NUM_CANDIDATE_PER_DST
];
529 for (j
= 0; j
< NUM_CANDIDATE_PER_DST
; j
++)
532 for (j
= 0; j
< rename_src_nr
; j
++) {
533 struct diff_filespec
*one
= rename_src
[j
].p
->one
;
534 struct diff_score this_src
;
536 if (skip_unmodified
&&
537 diff_unmodified_pair(rename_src
[j
].p
))
540 this_src
.score
= estimate_similarity(one
, two
,
542 this_src
.name_score
= basename_same(one
, two
);
545 record_if_better(m
, &this_src
);
547 * Once we run estimate_similarity,
548 * We do not need the text anymore.
550 diff_free_filespec_blob(one
);
551 diff_free_filespec_blob(two
);
554 display_progress(progress
, (i
+1)*rename_src_nr
);
556 stop_progress(&progress
);
558 /* cost matrix sorted by most to least similar pair */
559 qsort(mx
, dst_cnt
* NUM_CANDIDATE_PER_DST
, sizeof(*mx
), score_compare
);
561 rename_count
+= find_renames(mx
, dst_cnt
, minimum_score
, 0);
562 if (detect_rename
== DIFF_DETECT_COPY
)
563 rename_count
+= find_renames(mx
, dst_cnt
, minimum_score
, 1);
567 /* At this point, we have found some renames and copies and they
568 * are recorded in rename_dst. The original list is still in *q.
570 DIFF_QUEUE_CLEAR(&outq
);
571 for (i
= 0; i
< q
->nr
; i
++) {
572 struct diff_filepair
*p
= q
->queue
[i
];
573 struct diff_filepair
*pair_to_free
= NULL
;
575 if (DIFF_PAIR_UNMERGED(p
)) {
578 else if (!DIFF_FILE_VALID(p
->one
) && DIFF_FILE_VALID(p
->two
)) {
582 * We would output this create record if it has
583 * not been turned into a rename/copy already.
585 struct diff_rename_dst
*dst
=
586 locate_rename_dst(p
->two
, 0);
587 if (dst
&& dst
->pair
) {
588 diff_q(&outq
, dst
->pair
);
592 /* no matching rename/copy source, so
593 * record this as a creation.
597 else if (DIFF_FILE_VALID(p
->one
) && !DIFF_FILE_VALID(p
->two
)) {
601 * We would output this delete record if:
603 * (1) this is a broken delete and the counterpart
604 * broken create remains in the output; or
605 * (2) this is not a broken delete, and rename_dst
606 * does not have a rename/copy to move p->one->path
609 * Otherwise, the counterpart broken create
610 * has been turned into a rename-edit; or
611 * delete did not have a matching create to
614 if (DIFF_PAIR_BROKEN(p
)) {
616 struct diff_rename_dst
*dst
=
617 locate_rename_dst(p
->one
, 0);
618 if (dst
&& dst
->pair
)
619 /* counterpart is now rename/copy */
623 if (p
->one
->rename_used
)
624 /* this path remains */
633 else if (!diff_unmodified_pair(p
))
634 /* all the usual ones need to be kept */
637 /* no need to keep unmodified pairs */
641 diff_free_filepair(pair_to_free
);
643 diff_debug_queue("done copying original", &outq
);
647 diff_debug_queue("done collapsing", q
);
649 for (i
= 0; i
< rename_dst_nr
; i
++)
650 free_filespec(rename_dst
[i
].two
);
654 rename_dst_nr
= rename_dst_alloc
= 0;
657 rename_src_nr
= rename_src_alloc
= 0;