git-whatchanged: make default output format be pretty.
[git/git.git] / diffcore-rename.c
CommitLineData
427dcb4b
JH
1/*
2 * Copyright (C) 2005 Junio C Hamano
3 */
4#include "cache.h"
5#include "diff.h"
6#include "diffcore.h"
7#include "delta.h"
85976974 8#include "count-delta.h"
427dcb4b 9
25d5ea41
JH
10/* Table of rename/copy destinations */
11
12static struct diff_rename_dst {
13 struct diff_filespec *two;
14 struct diff_filepair *pair;
15} *rename_dst;
16static int rename_dst_nr, rename_dst_alloc;
427dcb4b 17
25d5ea41
JH
18static struct diff_rename_dst *locate_rename_dst(struct diff_filespec *two,
19 int insert_ok)
427dcb4b 20{
25d5ea41
JH
21 int first, last;
22
23 first = 0;
24 last = rename_dst_nr;
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);
29 if (!cmp)
30 return dst;
31 if (cmp < 0) {
32 last = next;
33 continue;
34 }
35 first = next+1;
36 }
37 /* not found */
38 if (!insert_ok)
39 return NULL;
40 /* insert to make it at "first" */
41 if (rename_dst_alloc <= rename_dst_nr) {
42 rename_dst_alloc = alloc_nr(rename_dst_alloc);
43 rename_dst = xrealloc(rename_dst,
44 rename_dst_alloc * sizeof(*rename_dst));
45 }
46 rename_dst_nr++;
47 if (first < rename_dst_nr)
48 memmove(rename_dst + first + 1, rename_dst + first,
49 (rename_dst_nr - first - 1) * sizeof(*rename_dst));
50 rename_dst[first].two = two;
51 rename_dst[first].pair = NULL;
52 return &(rename_dst[first]);
427dcb4b
JH
53}
54
15d061b4 55/* Table of rename/copy src files */
25d5ea41
JH
56static struct diff_rename_src {
57 struct diff_filespec *one;
15d061b4 58 unsigned src_stays : 1;
25d5ea41
JH
59} *rename_src;
60static int rename_src_nr, rename_src_alloc;
427dcb4b 61
15d061b4
JH
62static struct diff_rename_src *register_rename_src(struct diff_filespec *one,
63 int src_stays)
25d5ea41
JH
64{
65 int first, last;
66
67 first = 0;
68 last = rename_src_nr;
69 while (last > first) {
70 int next = (last + first) >> 1;
71 struct diff_rename_src *src = &(rename_src[next]);
72 int cmp = strcmp(one->path, src->one->path);
73 if (!cmp)
74 return src;
75 if (cmp < 0) {
76 last = next;
77 continue;
78 }
79 first = next+1;
80 }
15d061b4 81
25d5ea41
JH
82 /* insert to make it at "first" */
83 if (rename_src_alloc <= rename_src_nr) {
84 rename_src_alloc = alloc_nr(rename_src_alloc);
85 rename_src = xrealloc(rename_src,
86 rename_src_alloc * sizeof(*rename_src));
427dcb4b 87 }
25d5ea41
JH
88 rename_src_nr++;
89 if (first < rename_src_nr)
90 memmove(rename_src + first + 1, rename_src + first,
91 (rename_src_nr - first - 1) * sizeof(*rename_src));
92 rename_src[first].one = one;
15d061b4 93 rename_src[first].src_stays = src_stays;
25d5ea41 94 return &(rename_src[first]);
427dcb4b
JH
95}
96
97static int is_exact_match(struct diff_filespec *src, struct diff_filespec *dst)
98{
99 if (src->sha1_valid && dst->sha1_valid &&
100 !memcmp(src->sha1, dst->sha1, 20))
101 return 1;
f0c6b2a2
JH
102 if (diff_populate_filespec(src, 1) || diff_populate_filespec(dst, 1))
103 return 0;
104 if (src->size != dst->size)
105 return 0;
106 if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0))
427dcb4b
JH
107 return 0;
108 if (src->size == dst->size &&
109 !memcmp(src->data, dst->data, src->size))
110 return 1;
111 return 0;
112}
113
114struct diff_score {
25d5ea41
JH
115 int src; /* index in rename_src */
116 int dst; /* index in rename_dst */
427dcb4b 117 int score;
427dcb4b
JH
118};
119
120static int estimate_similarity(struct diff_filespec *src,
121 struct diff_filespec *dst,
122 int minimum_score)
123{
124 /* src points at a file that existed in the original tree (or
125 * optionally a file in the destination tree) and dst points
126 * at a newly created file. They may be quite similar, in which
127 * case we want to say src is renamed to dst or src is copied into
128 * dst, and then some edit has been applied to dst.
129 *
130 * Compare them and return how similar they are, representing
f0c6b2a2
JH
131 * the score as an integer between 0 and MAX_SCORE.
132 *
133 * When there is an exact match, it is considered a better
134 * match than anything else; the destination does not even
135 * call into this function in that case.
427dcb4b
JH
136 */
137 void *delta;
355e76a4 138 unsigned long delta_size, base_size, src_copied, literal_added;
427dcb4b
JH
139 int score;
140
60896c7b
JH
141 /* We deal only with regular files. Symlink renames are handled
142 * only when they are exact matches --- in other words, no edits
143 * after renaming.
144 */
145 if (!S_ISREG(src->mode) || !S_ISREG(dst->mode))
146 return 0;
147
427dcb4b
JH
148 delta_size = ((src->size < dst->size) ?
149 (dst->size - src->size) : (src->size - dst->size));
58b103f5 150 base_size = ((src->size < dst->size) ? src->size : dst->size);
427dcb4b 151
58b103f5
JH
152 /* We would not consider edits that change the file size so
153 * drastically. delta_size must be smaller than
cd1870ed 154 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
f0c6b2a2 155 *
58b103f5
JH
156 * Note that base_size == 0 case is handled here already
157 * and the final score computation below would not have a
158 * divide-by-zero issue.
427dcb4b 159 */
cd1870ed 160 if (base_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE)
427dcb4b
JH
161 return 0;
162
f0c6b2a2
JH
163 if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0))
164 return 0; /* error but caught downstream */
165
427dcb4b
JH
166 delta = diff_delta(src->data, src->size,
167 dst->data, dst->size,
168 &delta_size);
85976974
JH
169
170 /* A delta that has a lot of literal additions would have
171 * big delta_size no matter what else it does.
58b103f5 172 */
a00d7d10 173 if (base_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE)
85976974
JH
174 return 0;
175
176 /* Estimate the edit size by interpreting delta. */
355e76a4
JH
177 if (count_delta(delta, delta_size, &src_copied, &literal_added)) {
178 free(delta);
85976974 179 return 0;
355e76a4
JH
180 }
181 free(delta);
182
183 /* Extent of damage */
184 if (src->size + literal_added < src_copied)
185 delta_size = 0;
186 else
187 delta_size = (src->size - src_copied) + literal_added;
427dcb4b 188
58b103f5
JH
189 /*
190 * Now we will give some score to it. 100% edit gets 0 points
191 * and 0% edit gets MAX_SCORE points.
427dcb4b 192 */
58b103f5 193 score = MAX_SCORE - (MAX_SCORE * delta_size / base_size);
427dcb4b
JH
194 if (score < 0) return 0;
195 if (MAX_SCORE < score) return MAX_SCORE;
196 return score;
197}
198
25d5ea41
JH
199static void record_rename_pair(struct diff_queue_struct *renq,
200 int dst_index, int src_index, int score)
427dcb4b 201{
25d5ea41
JH
202 struct diff_filespec *one, *two, *src, *dst;
203 struct diff_filepair *dp;
f7c1512a 204
25d5ea41
JH
205 if (rename_dst[dst_index].pair)
206 die("internal error: dst already matched.");
427dcb4b 207
25d5ea41
JH
208 src = rename_src[src_index].one;
209 one = alloc_filespec(src->path);
210 fill_filespec(one, src->sha1, src->mode);
427dcb4b 211
25d5ea41
JH
212 dst = rename_dst[dst_index].two;
213 two = alloc_filespec(dst->path);
214 fill_filespec(two, dst->sha1, dst->mode);
427dcb4b 215
25d5ea41 216 dp = diff_queue(renq, one, two);
01c4e70f 217 dp->score = score;
15d061b4 218 dp->source_stays = rename_src[src_index].src_stays;
25d5ea41 219 rename_dst[dst_index].pair = dp;
427dcb4b
JH
220}
221
222/*
223 * We sort the rename similarity matrix with the score, in descending
15d061b4 224 * order (the most similar first).
427dcb4b
JH
225 */
226static int score_compare(const void *a_, const void *b_)
227{
228 const struct diff_score *a = a_, *b = b_;
229 return b->score - a->score;
230}
231
6b14d7fa 232void diffcore_rename(int detect_rename, int minimum_score)
427dcb4b 233{
38c6f780 234 struct diff_queue_struct *q = &diff_queued_diff;
25d5ea41 235 struct diff_queue_struct renq, outq;
427dcb4b 236 struct diff_score *mx;
25d5ea41
JH
237 int i, j;
238 int num_create, num_src, dst_cnt;
427dcb4b 239
26dee0ad 240 if (!minimum_score)
f345b0a0 241 minimum_score = DEFAULT_RENAME_SCORE;
25d5ea41
JH
242 renq.queue = NULL;
243 renq.nr = renq.alloc = 0;
427dcb4b 244
427dcb4b 245 for (i = 0; i < q->nr; i++) {
52e95789 246 struct diff_filepair *p = q->queue[i];
81e50eab
JH
247 if (!DIFF_FILE_VALID(p->one))
248 if (!DIFF_FILE_VALID(p->two))
60896c7b 249 continue; /* unmerged */
427dcb4b 250 else
25d5ea41 251 locate_rename_dst(p->two, 1);
81e50eab 252 else if (!DIFF_FILE_VALID(p->two))
15d061b4
JH
253 register_rename_src(p->one, 0);
254 else if (detect_rename == DIFF_DETECT_COPY)
255 register_rename_src(p->one, 1);
427dcb4b 256 }
25d5ea41 257 if (rename_dst_nr == 0)
427dcb4b
JH
258 goto cleanup; /* nothing to do */
259
260 /* We really want to cull the candidates list early
261 * with cheap tests in order to avoid doing deltas.
427dcb4b 262 */
25d5ea41
JH
263 for (i = 0; i < rename_dst_nr; i++) {
264 struct diff_filespec *two = rename_dst[i].two;
265 for (j = 0; j < rename_src_nr; j++) {
266 struct diff_filespec *one = rename_src[j].one;
267 if (!is_exact_match(one, two))
268 continue;
269 record_rename_pair(&renq, i, j, MAX_SCORE);
270 break; /* we are done with this entry */
427dcb4b
JH
271 }
272 }
25d5ea41 273 diff_debug_queue("done detecting exact", &renq);
427dcb4b
JH
274
275 /* Have we run out the created file pool? If so we can avoid
276 * doing the delta matrix altogether.
277 */
25d5ea41 278 if (renq.nr == rename_dst_nr)
15d061b4 279 goto cleanup;
427dcb4b 280
25d5ea41
JH
281 num_create = (rename_dst_nr - renq.nr);
282 num_src = rename_src_nr;
427dcb4b 283 mx = xmalloc(sizeof(*mx) * num_create * num_src);
25d5ea41 284 for (dst_cnt = i = 0; i < rename_dst_nr; i++) {
427dcb4b 285 int base = dst_cnt * num_src;
25d5ea41
JH
286 struct diff_filespec *two = rename_dst[i].two;
287 if (rename_dst[i].pair)
427dcb4b 288 continue; /* dealt with exact match already. */
25d5ea41
JH
289 for (j = 0; j < rename_src_nr; j++) {
290 struct diff_filespec *one = rename_src[j].one;
291 struct diff_score *m = &mx[base+j];
292 m->src = j;
293 m->dst = i;
294 m->score = estimate_similarity(one, two,
295 minimum_score);
427dcb4b
JH
296 }
297 dst_cnt++;
298 }
299 /* cost matrix sorted by most to least similar pair */
300 qsort(mx, num_create * num_src, sizeof(*mx), score_compare);
301 for (i = 0; i < num_create * num_src; i++) {
25d5ea41
JH
302 struct diff_rename_dst *dst = &rename_dst[mx[i].dst];
303 if (dst->pair)
304 continue; /* already done, either exact or fuzzy. */
427dcb4b 305 if (mx[i].score < minimum_score)
15d061b4 306 break; /* there is no more usable pair. */
25d5ea41 307 record_rename_pair(&renq, mx[i].dst, mx[i].src, mx[i].score);
427dcb4b
JH
308 }
309 free(mx);
25d5ea41 310 diff_debug_queue("done detecting fuzzy", &renq);
427dcb4b 311
15d061b4 312 cleanup:
427dcb4b 313 /* At this point, we have found some renames and copies and they
25d5ea41 314 * are kept in renq. The original list is still in *q.
427dcb4b 315 */
25d5ea41
JH
316 outq.queue = NULL;
317 outq.nr = outq.alloc = 0;
427dcb4b 318 for (i = 0; i < q->nr; i++) {
25d5ea41 319 struct diff_filepair *p = q->queue[i];
25d5ea41
JH
320 struct diff_filepair *pair_to_free = NULL;
321
2cd68882
JH
322 if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
323 /*
324 * Creation
325 *
326 * We would output this create record if it has
327 * not been turned into a rename/copy already.
328 */
329 struct diff_rename_dst *dst =
330 locate_rename_dst(p->two, 0);
331 if (dst && dst->pair) {
25d5ea41
JH
332 diff_q(&outq, dst->pair);
333 pair_to_free = p;
334 }
335 else
2cd68882
JH
336 /* no matching rename/copy source, so
337 * record this as a creation.
25d5ea41
JH
338 */
339 diff_q(&outq, p);
427dcb4b 340 }
2cd68882
JH
341 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) {
342 /*
343 * Deletion
344 *
f345b0a0
JH
345 * We would output this delete record if:
346 *
347 * (1) this is a broken delete and the counterpart
348 * broken create remains in the output; or
349 * (2) this is not a broken delete, and renq does
350 * not have a rename/copy to move p->one->path
351 * out.
352 *
353 * Otherwise, the counterpart broken create
354 * has been turned into a rename-edit; or
355 * delete did not have a matching create to
356 * begin with.
2cd68882 357 */
f345b0a0
JH
358 if (DIFF_PAIR_BROKEN(p)) {
359 /* broken delete */
360 struct diff_rename_dst *dst =
361 locate_rename_dst(p->one, 0);
362 if (dst && dst->pair)
363 /* counterpart is now rename/copy */
364 pair_to_free = p;
365 }
366 else {
367 for (j = 0; j < renq.nr; j++)
368 if (!strcmp(renq.queue[j]->one->path,
369 p->one->path))
370 break;
371 if (j < renq.nr)
372 /* this path remains */
373 pair_to_free = p;
374 }
2cd68882
JH
375
376 if (pair_to_free)
377 ;
378 else
379 diff_q(&outq, p);
380 }
25d5ea41 381 else if (!diff_unmodified_pair(p))
15d061b4 382 /* all the usual ones need to be kept */
25d5ea41 383 diff_q(&outq, p);
15d061b4
JH
384 else
385 /* no need to keep unmodified pairs */
386 pair_to_free = p;
387
226406f6
JH
388 if (pair_to_free)
389 diff_free_filepair(pair_to_free);
427dcb4b 390 }
25d5ea41 391 diff_debug_queue("done copying original", &outq);
427dcb4b 392
25d5ea41
JH
393 free(renq.queue);
394 free(q->queue);
395 *q = outq;
396 diff_debug_queue("done collapsing", q);
427dcb4b 397
25d5ea41
JH
398 free(rename_dst);
399 rename_dst = NULL;
400 rename_dst_nr = rename_dst_alloc = 0;
401 free(rename_src);
402 rename_src = NULL;
403 rename_src_nr = rename_src_alloc = 0;
427dcb4b
JH
404 return;
405}