[PATCH] Prepare diffcore interface for diff-tree header supression.
[git/git.git] / diffcore-rename.c
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"
8
9 struct diff_rename_pool {
10 struct diff_filespec **s;
11 int nr, alloc;
12 };
13
14 static void diff_rename_pool_clear(struct diff_rename_pool *pool)
15 {
16 pool->s = NULL; pool->nr = pool->alloc = 0;
17 }
18
19 static void diff_rename_pool_add(struct diff_rename_pool *pool,
20 struct diff_filespec *s)
21 {
22 if (S_ISDIR(s->mode))
23 return; /* rename/copy patch for tree does not make sense. */
24
25 if (pool->alloc <= pool->nr) {
26 pool->alloc = alloc_nr(pool->alloc);
27 pool->s = xrealloc(pool->s,
28 sizeof(*(pool->s)) * pool->alloc);
29 }
30 pool->s[pool->nr] = s;
31 pool->nr++;
32 }
33
34 static int is_exact_match(struct diff_filespec *src, struct diff_filespec *dst)
35 {
36 if (src->sha1_valid && dst->sha1_valid &&
37 !memcmp(src->sha1, dst->sha1, 20))
38 return 1;
39 if (diff_populate_filespec(src) || diff_populate_filespec(dst))
40 /* this is an error but will be caught downstream */
41 return 0;
42 if (src->size == dst->size &&
43 !memcmp(src->data, dst->data, src->size))
44 return 1;
45 return 0;
46 }
47
48 struct diff_score {
49 struct diff_filespec *src;
50 struct diff_filespec *dst;
51 int score;
52 int rank;
53 };
54
55 static int estimate_similarity(struct diff_filespec *src,
56 struct diff_filespec *dst,
57 int minimum_score)
58 {
59 /* src points at a file that existed in the original tree (or
60 * optionally a file in the destination tree) and dst points
61 * at a newly created file. They may be quite similar, in which
62 * case we want to say src is renamed to dst or src is copied into
63 * dst, and then some edit has been applied to dst.
64 *
65 * Compare them and return how similar they are, representing
66 * the score as an integer between 0 and 10000, except
67 * where they match exactly it is considered better than anything
68 * else.
69 */
70 void *delta;
71 unsigned long delta_size, base_size;
72 int score;
73
74 delta_size = ((src->size < dst->size) ?
75 (dst->size - src->size) : (src->size - dst->size));
76 base_size = ((src->size < dst->size) ? src->size : dst->size);
77
78 /* We would not consider edits that change the file size so
79 * drastically. delta_size must be smaller than
80 * minimum_score/MAX_SCORE * min(src->size, dst->size).
81 * Note that base_size == 0 case is handled here already
82 * and the final score computation below would not have a
83 * divide-by-zero issue.
84 */
85 if (base_size * minimum_score < delta_size * MAX_SCORE)
86 return 0;
87
88 delta = diff_delta(src->data, src->size,
89 dst->data, dst->size,
90 &delta_size);
91 /*
92 * We currently punt here, but we may later end up parsing the
93 * delta to really assess the extent of damage. A big consecutive
94 * remove would produce small delta_size that affects quite a
95 * big portion of the file.
96 */
97 free(delta);
98
99 /*
100 * Now we will give some score to it. 100% edit gets 0 points
101 * and 0% edit gets MAX_SCORE points.
102 */
103 score = MAX_SCORE - (MAX_SCORE * delta_size / base_size);
104 if (score < 0) return 0;
105 if (MAX_SCORE < score) return MAX_SCORE;
106 return score;
107 }
108
109 static void record_rename_pair(struct diff_queue_struct *outq,
110 struct diff_filespec *src,
111 struct diff_filespec *dst,
112 int rank,
113 int score)
114 {
115 /* The rank is used to sort the final output, because there
116 * are certain dependencies.
117 *
118 * - rank #0 depends on deleted ones.
119 * - rank #1 depends on kept files before they are modified.
120 * - rank #2 depends on kept files after they are modified;
121 * currently not used.
122 *
123 * Therefore, the final output order should be:
124 *
125 * 1. rank #0 rename/copy diffs.
126 * 2. deletions in the original.
127 * 3. rank #1 rename/copy diffs.
128 * 4. additions and modifications in the original.
129 * 5. rank #2 rename/copy diffs; currently not used.
130 *
131 * To achieve this sort order, we give xform_work the number
132 * above.
133 */
134 struct diff_filepair *dp = diff_queue(outq, src, dst);
135 dp->xfrm_work = (rank * 2 + 1) | (score<<RENAME_SCORE_SHIFT);
136 dst->xfrm_flags |= RENAME_DST_MATCHED;
137 }
138
139 #if 0
140 static void debug_filespec(struct diff_filespec *s, int x, const char *one)
141 {
142 fprintf(stderr, "queue[%d] %s (%s) %s %06o %s\n",
143 x, one,
144 s->path,
145 s->file_valid ? "valid" : "invalid",
146 s->mode,
147 s->sha1_valid ? sha1_to_hex(s->sha1) : "");
148 fprintf(stderr, "queue[%d] %s size %lu flags %d\n",
149 x, one,
150 s->size, s->xfrm_flags);
151 }
152
153 static void debug_filepair(const struct diff_filepair *p, int i)
154 {
155 debug_filespec(p->one, i, "one");
156 debug_filespec(p->two, i, "two");
157 fprintf(stderr, "pair flags %d, orig order %d, score %d\n",
158 (p->xfrm_work & ((1<<RENAME_SCORE_SHIFT) - 1)),
159 p->orig_order,
160 (p->xfrm_work >> RENAME_SCORE_SHIFT));
161 }
162
163 static void debug_queue(const char *msg, struct diff_queue_struct *q)
164 {
165 int i;
166 if (msg)
167 fprintf(stderr, "%s\n", msg);
168 fprintf(stderr, "q->nr = %d\n", q->nr);
169 for (i = 0; i < q->nr; i++) {
170 struct diff_filepair *p = q->queue[i];
171 debug_filepair(p, i);
172 }
173 }
174 #else
175 #define debug_queue(a,b) do { ; /*nothing*/ } while(0)
176 #endif
177
178 /*
179 * We sort the outstanding diff entries according to the rank (see
180 * comment at the beginning of record_rename_pair) and tiebreak with
181 * the order in the original input.
182 */
183 static int rank_compare(const void *a_, const void *b_)
184 {
185 const struct diff_filepair *a = *(const struct diff_filepair **)a_;
186 const struct diff_filepair *b = *(const struct diff_filepair **)b_;
187 int a_rank = a->xfrm_work & ((1<<RENAME_SCORE_SHIFT) - 1);
188 int b_rank = b->xfrm_work & ((1<<RENAME_SCORE_SHIFT) - 1);
189
190 if (a_rank != b_rank)
191 return a_rank - b_rank;
192 return a->orig_order - b->orig_order;
193 }
194
195 /*
196 * We sort the rename similarity matrix with the score, in descending
197 * order (more similar first).
198 */
199 static int score_compare(const void *a_, const void *b_)
200 {
201 const struct diff_score *a = a_, *b = b_;
202 return b->score - a->score;
203 }
204
205 static int needs_to_stay(struct diff_queue_struct *q, int i,
206 struct diff_filespec *it)
207 {
208 /* If it will be used in later entry (either stay or used
209 * as the source of rename/copy), we need to copy, not rename.
210 */
211 while (i < q->nr) {
212 struct diff_filepair *p = q->queue[i++];
213 if (!p->two->file_valid)
214 continue; /* removed is fine */
215 if (strcmp(p->one->path, it->path))
216 continue; /* not relevant */
217
218 /* p has its src set to *it and it is not a delete;
219 * it will be used for in-place change or rename/copy,
220 * so we cannot rename it out.
221 */
222 return 1;
223 }
224 return 0;
225 }
226
227 void diff_detect_rename(int detect_rename,
228 int minimum_score)
229 {
230 struct diff_queue_struct *q = &diff_queued_diff;
231 struct diff_queue_struct outq;
232 struct diff_rename_pool created, deleted, stay;
233 struct diff_rename_pool *(srcs[2]);
234 struct diff_score *mx;
235 int h, i, j;
236 int num_create, num_src, dst_cnt, src_cnt;
237
238 outq.queue = NULL;
239 outq.nr = outq.alloc = 0;
240
241 diff_rename_pool_clear(&created);
242 diff_rename_pool_clear(&deleted);
243 diff_rename_pool_clear(&stay);
244
245 srcs[0] = &deleted;
246 srcs[1] = &stay;
247
248 for (i = 0; i < q->nr; i++) {
249 struct diff_filepair *p = q->queue[i];
250 if (!p->one->file_valid)
251 if (!p->two->file_valid)
252 continue; /* ignore nonsense */
253 else
254 diff_rename_pool_add(&created, p->two);
255 else if (!p->two->file_valid)
256 diff_rename_pool_add(&deleted, p->one);
257 else if (1 < detect_rename) /* find copy, too */
258 diff_rename_pool_add(&stay, p->one);
259 }
260 if (created.nr == 0)
261 goto cleanup; /* nothing to do */
262
263 /* We really want to cull the candidates list early
264 * with cheap tests in order to avoid doing deltas.
265 *
266 * With the current callers, we should not have already
267 * matched entries at this point, but it is nonetheless
268 * checked for sanity.
269 */
270 for (i = 0; i < created.nr; i++) {
271 if (created.s[i]->xfrm_flags & RENAME_DST_MATCHED)
272 continue; /* we have matched exactly already */
273 for (h = 0; h < sizeof(srcs)/sizeof(srcs[0]); h++) {
274 struct diff_rename_pool *p = srcs[h];
275 for (j = 0; j < p->nr; j++) {
276 if (!is_exact_match(p->s[j], created.s[i]))
277 continue;
278 record_rename_pair(&outq,
279 p->s[j], created.s[i], h,
280 MAX_SCORE);
281 break; /* we are done with this entry */
282 }
283 }
284 }
285 debug_queue("done detecting exact", &outq);
286
287 /* Have we run out the created file pool? If so we can avoid
288 * doing the delta matrix altogether.
289 */
290 if (outq.nr == created.nr)
291 goto flush_rest;
292
293 num_create = (created.nr - outq.nr);
294 num_src = deleted.nr + stay.nr;
295 mx = xmalloc(sizeof(*mx) * num_create * num_src);
296 for (dst_cnt = i = 0; i < created.nr; i++) {
297 int base = dst_cnt * num_src;
298 if (created.s[i]->xfrm_flags & RENAME_DST_MATCHED)
299 continue; /* dealt with exact match already. */
300 for (src_cnt = h = 0; h < sizeof(srcs)/sizeof(srcs[0]); h++) {
301 struct diff_rename_pool *p = srcs[h];
302 for (j = 0; j < p->nr; j++, src_cnt++) {
303 struct diff_score *m = &mx[base + src_cnt];
304 m->src = p->s[j];
305 m->dst = created.s[i];
306 m->score = estimate_similarity(m->src, m->dst,
307 minimum_score);
308 m->rank = h;
309 }
310 }
311 dst_cnt++;
312 }
313 /* cost matrix sorted by most to least similar pair */
314 qsort(mx, num_create * num_src, sizeof(*mx), score_compare);
315 for (i = 0; i < num_create * num_src; i++) {
316 if (mx[i].dst->xfrm_flags & RENAME_DST_MATCHED)
317 continue; /* alreayd done, either exact or fuzzy. */
318 if (mx[i].score < minimum_score)
319 break; /* there is not any more diffs applicable. */
320 record_rename_pair(&outq,
321 mx[i].src, mx[i].dst, mx[i].rank,
322 mx[i].score);
323 }
324 free(mx);
325 debug_queue("done detecting fuzzy", &outq);
326
327 flush_rest:
328 /* At this point, we have found some renames and copies and they
329 * are kept in outq. The original list is still in *q.
330 *
331 * Scan the original list and move them into the outq; we will sort
332 * outq and swap it into the queue supplied to pass that to
333 * downstream, so we assign the sort keys in this loop.
334 *
335 * See comments at the top of record_rename_pair for numbers used
336 * to assign xfrm_work.
337 *
338 * Note that we have not annotated the diff_filepair with any comment
339 * so there is nothing other than p to free.
340 */
341 for (i = 0; i < q->nr; i++) {
342 struct diff_filepair *dp, *p = q->queue[i];
343 if (!p->one->file_valid) {
344 if (p->two->file_valid) {
345 /* creation */
346 dp = diff_queue(&outq, p->one, p->two);
347 dp->xfrm_work = 4;
348 }
349 /* otherwise it is a nonsense; just ignore it */
350 }
351 else if (!p->two->file_valid) {
352 /* deletion */
353 dp = diff_queue(&outq, p->one, p->two);
354 dp->xfrm_work = 2;
355 }
356 else {
357 /* modification, or stay as is */
358 dp = diff_queue(&outq, p->one, p->two);
359 dp->xfrm_work = 4;
360 }
361 free(p);
362 }
363 debug_queue("done copying original", &outq);
364
365 /* Sort outq */
366 qsort(outq.queue, outq.nr, sizeof(outq.queue[0]), rank_compare);
367
368 debug_queue("done sorting", &outq);
369
370 free(q->queue);
371 q->nr = q->alloc = 0;
372 q->queue = NULL;
373
374 /* Copy it out to q, removing duplicates. */
375 for (i = 0; i < outq.nr; i++) {
376 struct diff_filepair *p = outq.queue[i];
377 if (!p->one->file_valid) {
378 /* created */
379 if (p->two->xfrm_flags & RENAME_DST_MATCHED)
380 ; /* rename/copy created it already */
381 else
382 diff_queue(q, p->one, p->two);
383 }
384 else if (!p->two->file_valid) {
385 /* deleted */
386 if (p->one->xfrm_flags & RENAME_SRC_GONE)
387 ; /* rename/copy deleted it already */
388 else
389 diff_queue(q, p->one, p->two);
390 }
391 else if (strcmp(p->one->path, p->two->path)) {
392 /* rename or copy */
393 struct diff_filepair *dp =
394 diff_queue(q, p->one, p->two);
395 int msglen = (strlen(p->one->path) +
396 strlen(p->two->path) + 100);
397 int score = (p->xfrm_work >> RENAME_SCORE_SHIFT);
398 dp->xfrm_msg = xmalloc(msglen);
399
400 /* if we have a later entry that is a rename/copy
401 * that depends on p->one, then we copy here.
402 * otherwise we rename it.
403 */
404 if (needs_to_stay(&outq, i+1, p->one)) {
405 /* copy it */
406 sprintf(dp->xfrm_msg,
407 "similarity index %d%%\n"
408 "copy from %s\n"
409 "copy to %s\n",
410 (int)(0.5 + score * 100 / MAX_SCORE),
411 p->one->path, p->two->path);
412 }
413 else {
414 /* rename it, and mark it as gone. */
415 p->one->xfrm_flags |= RENAME_SRC_GONE;
416 sprintf(dp->xfrm_msg,
417 "similarity index %d%%\n"
418 "rename old %s\n"
419 "rename new %s\n",
420 (int)(0.5 + score * 100 / MAX_SCORE),
421 p->one->path, p->two->path);
422 }
423 }
424 else
425 /* otherwise it is a modified (or stayed) entry */
426 diff_queue(q, p->one, p->two);
427 free(p);
428 }
429
430 free(outq.queue);
431 debug_queue("done collapsing", q);
432
433 cleanup:
434 free(created.s);
435 free(deleted.s);
436 free(stay.s);
437 return;
438 }