1/* 2 * Copyright (C) 2005 Junio C Hamano 3 */ 4#include "cache.h" 5#include "diff.h" 6#include "diffcore.h" 7 8/* Table of rename/copy destinations */ 9 10static struct diff_rename_dst { 11 struct diff_filespec *two; 12 struct diff_filepair *pair; 13} *rename_dst; 14static int rename_dst_nr, rename_dst_alloc; 15 16static struct diff_rename_dst *locate_rename_dst(struct diff_filespec *two, 17 int insert_ok) 18{ 19 int first, last; 20 21 first = 0; 22 last = rename_dst_nr; 23 while (last > first) { 24 int next = (last + first) >> 1; 25 struct diff_rename_dst *dst = &(rename_dst[next]); 26 int cmp = strcmp(two->path, dst->two->path); 27 if (!cmp) 28 return dst; 29 if (cmp < 0) { 30 last = next; 31 continue; 32 } 33 first = next+1; 34 } 35 /* not found */ 36 if (!insert_ok) 37 return NULL; 38 /* insert to make it at "first" */ 39 if (rename_dst_alloc <= rename_dst_nr) { 40 rename_dst_alloc = alloc_nr(rename_dst_alloc); 41 rename_dst = xrealloc(rename_dst, 42 rename_dst_alloc * sizeof(*rename_dst)); 43 } 44 rename_dst_nr++; 45 if (first < rename_dst_nr) 46 memmove(rename_dst + first + 1, rename_dst + first, 47 (rename_dst_nr - first - 1) * sizeof(*rename_dst)); 48 rename_dst[first].two = alloc_filespec(two->path); 49 fill_filespec(rename_dst[first].two, two->sha1, two->mode); 50 rename_dst[first].pair = NULL; 51 return &(rename_dst[first]); 52} 53 54/* Table of rename/copy src files */ 55static struct diff_rename_src { 56 struct diff_filespec *one; 57 unsigned short score; /* to remember the break score */ 58 unsigned src_path_left : 1; 59} *rename_src; 60static int rename_src_nr, rename_src_alloc; 61 62static struct diff_rename_src *register_rename_src(struct diff_filespec *one, 63 int src_path_left, 64 unsigned short score) 65{ 66 int first, last; 67 68 first = 0; 69 last = rename_src_nr; 70 while (last > first) { 71 int next = (last + first) >> 1; 72 struct diff_rename_src *src = &(rename_src[next]); 73 int cmp = strcmp(one->path, src->one->path); 74 if (!cmp) 75 return src; 76 if (cmp < 0) { 77 last = next; 78 continue; 79 } 80 first = next+1; 81 } 82 83 /* insert to make it at "first" */ 84 if (rename_src_alloc <= rename_src_nr) { 85 rename_src_alloc = alloc_nr(rename_src_alloc); 86 rename_src = xrealloc(rename_src, 87 rename_src_alloc * sizeof(*rename_src)); 88 } 89 rename_src_nr++; 90 if (first < rename_src_nr) 91 memmove(rename_src + first + 1, rename_src + first, 92 (rename_src_nr - first - 1) * sizeof(*rename_src)); 93 rename_src[first].one = one; 94 rename_src[first].score = score; 95 rename_src[first].src_path_left = src_path_left; 96 return &(rename_src[first]); 97} 98 99static int is_exact_match(struct diff_filespec *src, 100 struct diff_filespec *dst, 101 int contents_too) 102{ 103 if (src->sha1_valid && dst->sha1_valid && 104 !hashcmp(src->sha1, dst->sha1)) 105 return 1; 106 if (!contents_too) 107 return 0; 108 if (diff_populate_filespec(src, 1) || diff_populate_filespec(dst, 1)) 109 return 0; 110 if (src->size != dst->size) 111 return 0; 112 if (src->sha1_valid && dst->sha1_valid) 113 return !hashcmp(src->sha1, dst->sha1); 114 if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0)) 115 return 0; 116 if (src->size == dst->size && 117 !memcmp(src->data, dst->data, src->size)) 118 return 1; 119 return 0; 120} 121 122static int basename_same(struct diff_filespec *src, struct diff_filespec *dst) 123{ 124 int src_len = strlen(src->path), dst_len = strlen(dst->path); 125 while (src_len && dst_len) { 126 char c1 = src->path[--src_len]; 127 char c2 = dst->path[--dst_len]; 128 if (c1 != c2) 129 return 0; 130 if (c1 == '/') 131 return 1; 132 } 133 return (!src_len || src->path[src_len - 1] == '/') && 134 (!dst_len || dst->path[dst_len - 1] == '/'); 135} 136 137struct diff_score { 138 int src; /* index in rename_src */ 139 int dst; /* index in rename_dst */ 140 int score; 141 int name_score; 142}; 143 144static int estimate_similarity(struct diff_filespec *src, 145 struct diff_filespec *dst, 146 int minimum_score) 147{ 148 /* src points at a file that existed in the original tree (or 149 * optionally a file in the destination tree) and dst points 150 * at a newly created file. They may be quite similar, in which 151 * case we want to say src is renamed to dst or src is copied into 152 * dst, and then some edit has been applied to dst. 153 * 154 * Compare them and return how similar they are, representing 155 * the score as an integer between 0 and MAX_SCORE. 156 * 157 * When there is an exact match, it is considered a better 158 * match than anything else; the destination does not even 159 * call into this function in that case. 160 */ 161 unsigned long max_size, delta_size, base_size, src_copied, literal_added; 162 unsigned long delta_limit; 163 int score; 164 165 /* We deal only with regular files. Symlink renames are handled 166 * only when they are exact matches --- in other words, no edits 167 * after renaming. 168 */ 169 if (!S_ISREG(src->mode) || !S_ISREG(dst->mode)) 170 return 0; 171 172 max_size = ((src->size > dst->size) ? src->size : dst->size); 173 base_size = ((src->size < dst->size) ? src->size : dst->size); 174 delta_size = max_size - base_size; 175 176 /* We would not consider edits that change the file size so 177 * drastically. delta_size must be smaller than 178 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size). 179 * 180 * Note that base_size == 0 case is handled here already 181 * and the final score computation below would not have a 182 * divide-by-zero issue. 183 */ 184 if (base_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE) 185 return 0; 186 187 if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0)) 188 return 0; /* error but caught downstream */ 189 190 191 delta_limit = (unsigned long) 192 (base_size * (MAX_SCORE-minimum_score) / MAX_SCORE); 193 if (diffcore_count_changes(src, dst, 194 &src->cnt_data, &dst->cnt_data, 195 delta_limit, 196 &src_copied, &literal_added)) 197 return 0; 198 199 /* How similar are they? 200 * what percentage of material in dst are from source? 201 */ 202 if (!dst->size) 203 score = 0; /* should not happen */ 204 else 205 score = (int)(src_copied * MAX_SCORE / max_size); 206 return score; 207} 208 209static void record_rename_pair(int dst_index, int src_index, int score) 210{ 211 struct diff_filespec *one, *two, *src, *dst; 212 struct diff_filepair *dp; 213 214 if (rename_dst[dst_index].pair) 215 die("internal error: dst already matched."); 216 217 src = rename_src[src_index].one; 218 one = alloc_filespec(src->path); 219 fill_filespec(one, src->sha1, src->mode); 220 221 dst = rename_dst[dst_index].two; 222 two = alloc_filespec(dst->path); 223 fill_filespec(two, dst->sha1, dst->mode); 224 225 dp = diff_queue(NULL, one, two); 226 dp->renamed_pair = 1; 227 if (!strcmp(src->path, dst->path)) 228 dp->score = rename_src[src_index].score; 229 else 230 dp->score = score; 231 dp->source_stays = rename_src[src_index].src_path_left; 232 rename_dst[dst_index].pair = dp; 233} 234 235/* 236 * We sort the rename similarity matrix with the score, in descending 237 * order (the most similar first). 238 */ 239static int score_compare(const void *a_, const void *b_) 240{ 241 const struct diff_score *a = a_, *b = b_; 242 243 if (a->score == b->score) 244 return b->name_score - a->name_score; 245 246 return b->score - a->score; 247} 248 249static int compute_stays(struct diff_queue_struct *q, 250 struct diff_filespec *one) 251{ 252 int i; 253 for (i = 0; i < q->nr; i++) { 254 struct diff_filepair *p = q->queue[i]; 255 if (strcmp(one->path, p->two->path)) 256 continue; 257 if (DIFF_PAIR_RENAME(p)) { 258 return 0; /* something else is renamed into this */ 259 } 260 } 261 return 1; 262} 263 264void diffcore_rename(struct diff_options *options) 265{ 266 int detect_rename = options->detect_rename; 267 int minimum_score = options->rename_score; 268 int rename_limit = options->rename_limit; 269 struct diff_queue_struct *q = &diff_queued_diff; 270 struct diff_queue_struct outq; 271 struct diff_score *mx; 272 int i, j, rename_count, contents_too; 273 int num_create, num_src, dst_cnt; 274 275 if (!minimum_score) 276 minimum_score = DEFAULT_RENAME_SCORE; 277 rename_count = 0; 278 279 for (i = 0; i < q->nr; i++) { 280 struct diff_filepair *p = q->queue[i]; 281 if (!DIFF_FILE_VALID(p->one)) { 282 if (!DIFF_FILE_VALID(p->two)) 283 continue; /* unmerged */ 284 else if (options->single_follow && 285 strcmp(options->single_follow, p->two->path)) 286 continue; /* not interested */ 287 else 288 locate_rename_dst(p->two, 1); 289 } 290 else if (!DIFF_FILE_VALID(p->two)) { 291 /* If the source is a broken "delete", and 292 * they did not really want to get broken, 293 * that means the source actually stays. 294 */ 295 int stays = (p->broken_pair && !p->score); 296 register_rename_src(p->one, stays, p->score); 297 } 298 else if (detect_rename == DIFF_DETECT_COPY) 299 register_rename_src(p->one, 1, p->score); 300 } 301 if (rename_dst_nr == 0 || rename_src_nr == 0) 302 goto cleanup; /* nothing to do */ 303 304 /* 305 * This basically does a test for the rename matrix not 306 * growing larger than a "rename_limit" square matrix, ie: 307 * 308 * rename_dst_nr * rename_src_nr > rename_limit * rename_limit 309 * 310 * but handles the potential overflow case specially (and we 311 * assume at least 32-bit integers) 312 */ 313 if (rename_limit <= 0 || rename_limit > 32767) 314 rename_limit = 32767; 315 if (rename_dst_nr > rename_limit && rename_src_nr > rename_limit) 316 goto cleanup; 317 if (rename_dst_nr * rename_src_nr > rename_limit * rename_limit) 318 goto cleanup; 319 320 /* We really want to cull the candidates list early 321 * with cheap tests in order to avoid doing deltas. 322 * The first round matches up the up-to-date entries, 323 * and then during the second round we try to match 324 * cache-dirty entries as well. 325 */ 326 for (contents_too = 0; contents_too < 2; contents_too++) { 327 for (i = 0; i < rename_dst_nr; i++) { 328 struct diff_filespec *two = rename_dst[i].two; 329 if (rename_dst[i].pair) 330 continue; /* dealt with an earlier round */ 331 for (j = 0; j < rename_src_nr; j++) { 332 int k; 333 struct diff_filespec *one = rename_src[j].one; 334 if (!is_exact_match(one, two, contents_too)) 335 continue; 336 337 /* see if there is a basename match, too */ 338 for (k = j; k < rename_src_nr; k++) { 339 one = rename_src[k].one; 340 if (basename_same(one, two) && 341 is_exact_match(one, two, 342 contents_too)) { 343 j = k; 344 break; 345 } 346 } 347 348 record_rename_pair(i, j, (int)MAX_SCORE); 349 rename_count++; 350 break; /* we are done with this entry */ 351 } 352 } 353 } 354 355 /* Have we run out the created file pool? If so we can avoid 356 * doing the delta matrix altogether. 357 */ 358 if (rename_count == rename_dst_nr) 359 goto cleanup; 360 361 if (minimum_score == MAX_SCORE) 362 goto cleanup; 363 364 num_create = (rename_dst_nr - rename_count); 365 num_src = rename_src_nr; 366 mx = xmalloc(sizeof(*mx) * num_create * num_src); 367 for (dst_cnt = i = 0; i < rename_dst_nr; i++) { 368 int base = dst_cnt * num_src; 369 struct diff_filespec *two = rename_dst[i].two; 370 if (rename_dst[i].pair) 371 continue; /* dealt with exact match already. */ 372 for (j = 0; j < rename_src_nr; j++) { 373 struct diff_filespec *one = rename_src[j].one; 374 struct diff_score *m = &mx[base+j]; 375 m->src = j; 376 m->dst = i; 377 m->score = estimate_similarity(one, two, 378 minimum_score); 379 m->name_score = basename_same(one, two); 380 diff_free_filespec_data(one); 381 } 382 /* We do not need the text anymore */ 383 diff_free_filespec_data(two); 384 dst_cnt++; 385 } 386 /* cost matrix sorted by most to least similar pair */ 387 qsort(mx, num_create * num_src, sizeof(*mx), score_compare); 388 for (i = 0; i < num_create * num_src; i++) { 389 struct diff_rename_dst *dst = &rename_dst[mx[i].dst]; 390 if (dst->pair) 391 continue; /* already done, either exact or fuzzy. */ 392 if (mx[i].score < minimum_score) 393 break; /* there is no more usable pair. */ 394 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score); 395 rename_count++; 396 } 397 free(mx); 398 399 cleanup: 400 /* At this point, we have found some renames and copies and they 401 * are recorded in rename_dst. The original list is still in *q. 402 */ 403 outq.queue = NULL; 404 outq.nr = outq.alloc = 0; 405 for (i = 0; i < q->nr; i++) { 406 struct diff_filepair *p = q->queue[i]; 407 struct diff_filepair *pair_to_free = NULL; 408 409 if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) { 410 /* 411 * Creation 412 * 413 * We would output this create record if it has 414 * not been turned into a rename/copy already. 415 */ 416 struct diff_rename_dst *dst = 417 locate_rename_dst(p->two, 0); 418 if (dst && dst->pair) { 419 diff_q(&outq, dst->pair); 420 pair_to_free = p; 421 } 422 else 423 /* no matching rename/copy source, so 424 * record this as a creation. 425 */ 426 diff_q(&outq, p); 427 } 428 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) { 429 /* 430 * Deletion 431 * 432 * We would output this delete record if: 433 * 434 * (1) this is a broken delete and the counterpart 435 * broken create remains in the output; or 436 * (2) this is not a broken delete, and rename_dst 437 * does not have a rename/copy to move p->one->path 438 * out of existence. 439 * 440 * Otherwise, the counterpart broken create 441 * has been turned into a rename-edit; or 442 * delete did not have a matching create to 443 * begin with. 444 */ 445 if (DIFF_PAIR_BROKEN(p)) { 446 /* broken delete */ 447 struct diff_rename_dst *dst = 448 locate_rename_dst(p->one, 0); 449 if (dst && dst->pair) 450 /* counterpart is now rename/copy */ 451 pair_to_free = p; 452 } 453 else { 454 for (j = 0; j < rename_dst_nr; j++) { 455 if (!rename_dst[j].pair) 456 continue; 457 if (strcmp(rename_dst[j].pair-> 458 one->path, 459 p->one->path)) 460 continue; 461 break; 462 } 463 if (j < rename_dst_nr) 464 /* this path remains */ 465 pair_to_free = p; 466 } 467 468 if (pair_to_free) 469 ; 470 else 471 diff_q(&outq, p); 472 } 473 else if (!diff_unmodified_pair(p)) 474 /* all the usual ones need to be kept */ 475 diff_q(&outq, p); 476 else 477 /* no need to keep unmodified pairs */ 478 pair_to_free = p; 479 480 if (pair_to_free) 481 diff_free_filepair(pair_to_free); 482 } 483 diff_debug_queue("done copying original", &outq); 484 485 free(q->queue); 486 *q = outq; 487 diff_debug_queue("done collapsing", q); 488 489 /* We need to see which rename source really stays here; 490 * earlier we only checked if the path is left in the result, 491 * but even if a path remains in the result, if that is coming 492 * from copying something else on top of it, then the original 493 * source is lost and does not stay. 494 */ 495 for (i = 0; i < q->nr; i++) { 496 struct diff_filepair *p = q->queue[i]; 497 if (DIFF_PAIR_RENAME(p) && p->source_stays) { 498 /* If one appears as the target of a rename-copy, 499 * then mark p->source_stays = 0; otherwise 500 * leave it as is. 501 */ 502 p->source_stays = compute_stays(q, p->one); 503 } 504 } 505 506 for (i = 0; i < rename_dst_nr; i++) { 507 diff_free_filespec_data(rename_dst[i].two); 508 free(rename_dst[i].two); 509 } 510 511 free(rename_dst); 512 rename_dst = NULL; 513 rename_dst_nr = rename_dst_alloc = 0; 514 free(rename_src); 515 rename_src = NULL; 516 rename_src_nr = rename_src_alloc = 0; 517 return; 518}