1/* 2 * Copyright (C) 2005 Junio C Hamano 3 */ 4#include "cache.h" 5#include "diff.h" 6#include "diffcore.h" 7#include "hash.h" 8 9/* Table of rename/copy destinations */ 10 11static struct diff_rename_dst { 12 struct diff_filespec *two; 13 struct diff_filepair *pair; 14} *rename_dst; 15static int rename_dst_nr, rename_dst_alloc; 16 17static struct diff_rename_dst *locate_rename_dst(struct diff_filespec *two, 18 int insert_ok) 19{ 20 int first, last; 21 22 first = 0; 23 last = rename_dst_nr; 24 while (last > first) { 25 int next = (last + first) >> 1; 26 struct diff_rename_dst *dst = &(rename_dst[next]); 27 int cmp = strcmp(two->path, dst->two->path); 28 if (!cmp) 29 return dst; 30 if (cmp < 0) { 31 last = next; 32 continue; 33 } 34 first = next+1; 35 } 36 /* not found */ 37 if (!insert_ok) 38 return NULL; 39 /* insert to make it at "first" */ 40 if (rename_dst_alloc <= rename_dst_nr) { 41 rename_dst_alloc = alloc_nr(rename_dst_alloc); 42 rename_dst = xrealloc(rename_dst, 43 rename_dst_alloc * sizeof(*rename_dst)); 44 } 45 rename_dst_nr++; 46 if (first < rename_dst_nr) 47 memmove(rename_dst + first + 1, rename_dst + first, 48 (rename_dst_nr - first - 1) * sizeof(*rename_dst)); 49 rename_dst[first].two = alloc_filespec(two->path); 50 fill_filespec(rename_dst[first].two, two->sha1, two->mode); 51 rename_dst[first].pair = NULL; 52 return &(rename_dst[first]); 53} 54 55/* Table of rename/copy src files */ 56static struct diff_rename_src { 57 struct diff_filespec *one; 58 unsigned short score; /* to remember the break score */ 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 unsigned short score) 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 } 81 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)); 87 } 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; 93 rename_src[first].score = score; 94 return &(rename_src[first]); 95} 96 97static int basename_same(struct diff_filespec *src, struct diff_filespec *dst) 98{ 99 int src_len = strlen(src->path), dst_len = strlen(dst->path); 100 while (src_len && dst_len) { 101 char c1 = src->path[--src_len]; 102 char c2 = dst->path[--dst_len]; 103 if (c1 != c2) 104 return 0; 105 if (c1 == '/') 106 return 1; 107 } 108 return (!src_len || src->path[src_len - 1] == '/') && 109 (!dst_len || dst->path[dst_len - 1] == '/'); 110} 111 112struct diff_score { 113 int src; /* index in rename_src */ 114 int dst; /* index in rename_dst */ 115 unsigned short score; 116 short name_score; 117}; 118 119static int estimate_similarity(struct diff_filespec *src, 120 struct diff_filespec *dst, 121 int minimum_score) 122{ 123 /* src points at a file that existed in the original tree (or 124 * optionally a file in the destination tree) and dst points 125 * at a newly created file. They may be quite similar, in which 126 * case we want to say src is renamed to dst or src is copied into 127 * dst, and then some edit has been applied to dst. 128 * 129 * Compare them and return how similar they are, representing 130 * the score as an integer between 0 and MAX_SCORE. 131 * 132 * When there is an exact match, it is considered a better 133 * match than anything else; the destination does not even 134 * call into this function in that case. 135 */ 136 unsigned long max_size, delta_size, base_size, src_copied, literal_added; 137 unsigned long delta_limit; 138 int score; 139 140 /* We deal only with regular files. Symlink renames are handled 141 * only when they are exact matches --- in other words, no edits 142 * after renaming. 143 */ 144 if (!S_ISREG(src->mode) || !S_ISREG(dst->mode)) 145 return 0; 146 147 /* 148 * Need to check that source and destination sizes are 149 * filled in before comparing them. 150 * 151 * If we already have "cnt_data" filled in, we know it's 152 * all good (avoid checking the size for zero, as that 153 * is a possible size - we really should have a flag to 154 * say whether the size is valid or not!) 155 */ 156 if (!src->cnt_data && diff_populate_filespec(src, 0)) 157 return 0; 158 if (!dst->cnt_data && diff_populate_filespec(dst, 0)) 159 return 0; 160 161 max_size = ((src->size > dst->size) ? src->size : dst->size); 162 base_size = ((src->size < dst->size) ? src->size : dst->size); 163 delta_size = max_size - base_size; 164 165 /* We would not consider edits that change the file size so 166 * drastically. delta_size must be smaller than 167 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size). 168 * 169 * Note that base_size == 0 case is handled here already 170 * and the final score computation below would not have a 171 * divide-by-zero issue. 172 */ 173 if (base_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE) 174 return 0; 175 176 delta_limit = (unsigned long) 177 (base_size * (MAX_SCORE-minimum_score) / MAX_SCORE); 178 if (diffcore_count_changes(src, dst, 179 &src->cnt_data, &dst->cnt_data, 180 delta_limit, 181 &src_copied, &literal_added)) 182 return 0; 183 184 /* How similar are they? 185 * what percentage of material in dst are from source? 186 */ 187 if (!dst->size) 188 score = 0; /* should not happen */ 189 else 190 score = (int)(src_copied * MAX_SCORE / max_size); 191 return score; 192} 193 194static void record_rename_pair(int dst_index, int src_index, int score) 195{ 196 struct diff_filespec *src, *dst; 197 struct diff_filepair *dp; 198 199 if (rename_dst[dst_index].pair) 200 die("internal error: dst already matched."); 201 202 src = rename_src[src_index].one; 203 src->rename_used++; 204 src->count++; 205 206 dst = rename_dst[dst_index].two; 207 dst->count++; 208 209 dp = diff_queue(NULL, src, dst); 210 dp->renamed_pair = 1; 211 if (!strcmp(src->path, dst->path)) 212 dp->score = rename_src[src_index].score; 213 else 214 dp->score = score; 215 rename_dst[dst_index].pair = dp; 216} 217 218/* 219 * We sort the rename similarity matrix with the score, in descending 220 * order (the most similar first). 221 */ 222static int score_compare(const void *a_, const void *b_) 223{ 224 const struct diff_score *a = a_, *b = b_; 225 226 /* sink the unused ones to the bottom */ 227 if (a->dst < 0) 228 return (0 <= b->dst); 229 else if (b->dst < 0) 230 return -1; 231 232 if (a->score == b->score) 233 return b->name_score - a->name_score; 234 235 return b->score - a->score; 236} 237 238struct file_similarity { 239 int src_dst, index; 240 struct diff_filespec *filespec; 241 struct file_similarity *next; 242}; 243 244static int find_identical_files(struct file_similarity *src, 245 struct file_similarity *dst) 246{ 247 int renames = 0; 248 249 /* 250 * Walk over all the destinations ... 251 */ 252 do { 253 struct diff_filespec *target = dst->filespec; 254 struct file_similarity *p, *best; 255 int i = 100, best_score = -1; 256 257 /* 258 * .. to find the best source match 259 */ 260 best = NULL; 261 for (p = src; p; p = p->next) { 262 int score; 263 struct diff_filespec *source = p->filespec; 264 265 /* False hash collission? */ 266 if (hashcmp(source->sha1, target->sha1)) 267 continue; 268 /* Non-regular files? If so, the modes must match! */ 269 if (!S_ISREG(source->mode) || !S_ISREG(target->mode)) { 270 if (source->mode != target->mode) 271 continue; 272 } 273 /* Give higher scores to sources that haven't been used already */ 274 score = !source->rename_used; 275 score += basename_same(source, target); 276 if (score > best_score) { 277 best = p; 278 best_score = score; 279 if (score == 2) 280 break; 281 } 282 283 /* Too many identical alternatives? Pick one */ 284 if (!--i) 285 break; 286 } 287 if (best) { 288 record_rename_pair(dst->index, best->index, MAX_SCORE); 289 renames++; 290 } 291 } while ((dst = dst->next) != NULL); 292 return renames; 293} 294 295static void free_similarity_list(struct file_similarity *p) 296{ 297 while (p) { 298 struct file_similarity *entry = p; 299 p = p->next; 300 free(entry); 301 } 302} 303 304static int find_same_files(void *ptr) 305{ 306 int ret; 307 struct file_similarity *p = ptr; 308 struct file_similarity *src = NULL, *dst = NULL; 309 310 /* Split the hash list up into sources and destinations */ 311 do { 312 struct file_similarity *entry = p; 313 p = p->next; 314 if (entry->src_dst < 0) { 315 entry->next = src; 316 src = entry; 317 } else { 318 entry->next = dst; 319 dst = entry; 320 } 321 } while (p); 322 323 /* 324 * If we have both sources *and* destinations, see if 325 * we can match them up 326 */ 327 ret = (src && dst) ? find_identical_files(src, dst) : 0; 328 329 /* Free the hashes and return the number of renames found */ 330 free_similarity_list(src); 331 free_similarity_list(dst); 332 return ret; 333} 334 335static unsigned int hash_filespec(struct diff_filespec *filespec) 336{ 337 unsigned int hash; 338 if (!filespec->sha1_valid) { 339 if (diff_populate_filespec(filespec, 0)) 340 return 0; 341 hash_sha1_file(filespec->data, filespec->size, "blob", filespec->sha1); 342 } 343 memcpy(&hash, filespec->sha1, sizeof(hash)); 344 return hash; 345} 346 347static void insert_file_table(struct hash_table *table, int src_dst, int index, struct diff_filespec *filespec) 348{ 349 void **pos; 350 unsigned int hash; 351 struct file_similarity *entry = xmalloc(sizeof(*entry)); 352 353 entry->src_dst = src_dst; 354 entry->index = index; 355 entry->filespec = filespec; 356 entry->next = NULL; 357 358 hash = hash_filespec(filespec); 359 pos = insert_hash(hash, entry, table); 360 361 /* We already had an entry there? */ 362 if (pos) { 363 entry->next = *pos; 364 *pos = entry; 365 } 366} 367 368/* 369 * Find exact renames first. 370 * 371 * The first round matches up the up-to-date entries, 372 * and then during the second round we try to match 373 * cache-dirty entries as well. 374 */ 375static int find_exact_renames(void) 376{ 377 int i; 378 struct hash_table file_table; 379 380 init_hash(&file_table); 381 for (i = 0; i < rename_src_nr; i++) 382 insert_file_table(&file_table, -1, i, rename_src[i].one); 383 384 for (i = 0; i < rename_dst_nr; i++) 385 insert_file_table(&file_table, 1, i, rename_dst[i].two); 386 387 /* Find the renames */ 388 i = for_each_hash(&file_table, find_same_files); 389 390 /* .. and free the hash data structure */ 391 free_hash(&file_table); 392 393 return i; 394} 395 396#define NUM_CANDIDATE_PER_DST 4 397static void record_if_better(struct diff_score m[], struct diff_score *o) 398{ 399 int i, worst; 400 401 /* find the worst one */ 402 worst = 0; 403 for (i = 1; i < NUM_CANDIDATE_PER_DST; i++) 404 if (score_compare(&m[i], &m[worst]) > 0) 405 worst = i; 406 407 /* is it better than the worst one? */ 408 if (score_compare(&m[worst], o) > 0) 409 m[worst] = *o; 410} 411 412void diffcore_rename(struct diff_options *options) 413{ 414 int detect_rename = options->detect_rename; 415 int minimum_score = options->rename_score; 416 int rename_limit = options->rename_limit; 417 struct diff_queue_struct *q = &diff_queued_diff; 418 struct diff_queue_struct outq; 419 struct diff_score *mx; 420 int i, j, rename_count; 421 int num_create, num_src, dst_cnt; 422 423 if (!minimum_score) 424 minimum_score = DEFAULT_RENAME_SCORE; 425 426 for (i = 0; i < q->nr; i++) { 427 struct diff_filepair *p = q->queue[i]; 428 if (!DIFF_FILE_VALID(p->one)) { 429 if (!DIFF_FILE_VALID(p->two)) 430 continue; /* unmerged */ 431 else if (options->single_follow && 432 strcmp(options->single_follow, p->two->path)) 433 continue; /* not interested */ 434 else 435 locate_rename_dst(p->two, 1); 436 } 437 else if (!DIFF_FILE_VALID(p->two)) { 438 /* 439 * If the source is a broken "delete", and 440 * they did not really want to get broken, 441 * that means the source actually stays. 442 * So we increment the "rename_used" score 443 * by one, to indicate ourselves as a user 444 */ 445 if (p->broken_pair && !p->score) 446 p->one->rename_used++; 447 register_rename_src(p->one, p->score); 448 } 449 else if (detect_rename == DIFF_DETECT_COPY) { 450 /* 451 * Increment the "rename_used" score by 452 * one, to indicate ourselves as a user. 453 */ 454 p->one->rename_used++; 455 register_rename_src(p->one, p->score); 456 } 457 } 458 if (rename_dst_nr == 0 || rename_src_nr == 0) 459 goto cleanup; /* nothing to do */ 460 461 /* 462 * We really want to cull the candidates list early 463 * with cheap tests in order to avoid doing deltas. 464 */ 465 rename_count = find_exact_renames(); 466 467 /* Did we only want exact renames? */ 468 if (minimum_score == MAX_SCORE) 469 goto cleanup; 470 471 /* 472 * Calculate how many renames are left (but all the source 473 * files still remain as options for rename/copies!) 474 */ 475 num_create = (rename_dst_nr - rename_count); 476 num_src = rename_src_nr; 477 478 /* All done? */ 479 if (!num_create) 480 goto cleanup; 481 482 /* 483 * This basically does a test for the rename matrix not 484 * growing larger than a "rename_limit" square matrix, ie: 485 * 486 * num_create * num_src > rename_limit * rename_limit 487 * 488 * but handles the potential overflow case specially (and we 489 * assume at least 32-bit integers) 490 */ 491 if (rename_limit <= 0 || rename_limit > 32767) 492 rename_limit = 32767; 493 if ((num_create > rename_limit && num_src > rename_limit) || 494 (num_create * num_src > rename_limit * rename_limit)) { 495 warning("too many files, skipping inexact rename detection"); 496 goto cleanup; 497 } 498 499 mx = xcalloc(num_create * NUM_CANDIDATE_PER_DST, sizeof(*mx)); 500 for (dst_cnt = i = 0; i < rename_dst_nr; i++) { 501 struct diff_filespec *two = rename_dst[i].two; 502 struct diff_score *m; 503 504 if (rename_dst[i].pair) 505 continue; /* dealt with exact match already. */ 506 507 m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST]; 508 for (j = 0; j < NUM_CANDIDATE_PER_DST; j++) 509 m[j].dst = -1; 510 511 for (j = 0; j < rename_src_nr; j++) { 512 struct diff_filespec *one = rename_src[j].one; 513 struct diff_score this_src; 514 this_src.score = estimate_similarity(one, two, 515 minimum_score); 516 this_src.name_score = basename_same(one, two); 517 this_src.dst = i; 518 this_src.src = j; 519 record_if_better(m, &this_src); 520 diff_free_filespec_blob(one); 521 } 522 /* We do not need the text anymore */ 523 diff_free_filespec_blob(two); 524 dst_cnt++; 525 } 526 527 /* cost matrix sorted by most to least similar pair */ 528 qsort(mx, dst_cnt * NUM_CANDIDATE_PER_DST, sizeof(*mx), score_compare); 529 530 for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) { 531 struct diff_rename_dst *dst; 532 533 if ((mx[i].dst < 0) || 534 (mx[i].score < minimum_score)) 535 break; /* there is no more usable pair. */ 536 dst = &rename_dst[mx[i].dst]; 537 if (dst->pair) 538 continue; /* already done, either exact or fuzzy. */ 539 if (rename_src[mx[i].src].one->rename_used) 540 continue; 541 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score); 542 rename_count++; 543 } 544 545 for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) { 546 struct diff_rename_dst *dst; 547 548 if ((mx[i].dst < 0) || 549 (mx[i].score < minimum_score)) 550 break; /* there is no more usable pair. */ 551 dst = &rename_dst[mx[i].dst]; 552 if (dst->pair) 553 continue; /* already done, either exact or fuzzy. */ 554 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score); 555 rename_count++; 556 } 557 free(mx); 558 559 cleanup: 560 /* At this point, we have found some renames and copies and they 561 * are recorded in rename_dst. The original list is still in *q. 562 */ 563 outq.queue = NULL; 564 outq.nr = outq.alloc = 0; 565 for (i = 0; i < q->nr; i++) { 566 struct diff_filepair *p = q->queue[i]; 567 struct diff_filepair *pair_to_free = NULL; 568 569 if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) { 570 /* 571 * Creation 572 * 573 * We would output this create record if it has 574 * not been turned into a rename/copy already. 575 */ 576 struct diff_rename_dst *dst = 577 locate_rename_dst(p->two, 0); 578 if (dst && dst->pair) { 579 diff_q(&outq, dst->pair); 580 pair_to_free = p; 581 } 582 else 583 /* no matching rename/copy source, so 584 * record this as a creation. 585 */ 586 diff_q(&outq, p); 587 } 588 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) { 589 /* 590 * Deletion 591 * 592 * We would output this delete record if: 593 * 594 * (1) this is a broken delete and the counterpart 595 * broken create remains in the output; or 596 * (2) this is not a broken delete, and rename_dst 597 * does not have a rename/copy to move p->one->path 598 * out of existence. 599 * 600 * Otherwise, the counterpart broken create 601 * has been turned into a rename-edit; or 602 * delete did not have a matching create to 603 * begin with. 604 */ 605 if (DIFF_PAIR_BROKEN(p)) { 606 /* broken delete */ 607 struct diff_rename_dst *dst = 608 locate_rename_dst(p->one, 0); 609 if (dst && dst->pair) 610 /* counterpart is now rename/copy */ 611 pair_to_free = p; 612 } 613 else { 614 if (p->one->rename_used) 615 /* this path remains */ 616 pair_to_free = p; 617 } 618 619 if (pair_to_free) 620 ; 621 else 622 diff_q(&outq, p); 623 } 624 else if (!diff_unmodified_pair(p)) 625 /* all the usual ones need to be kept */ 626 diff_q(&outq, p); 627 else 628 /* no need to keep unmodified pairs */ 629 pair_to_free = p; 630 631 if (pair_to_free) 632 diff_free_filepair(pair_to_free); 633 } 634 diff_debug_queue("done copying original", &outq); 635 636 free(q->queue); 637 *q = outq; 638 diff_debug_queue("done collapsing", q); 639 640 for (i = 0; i < rename_dst_nr; i++) 641 free_filespec(rename_dst[i].two); 642 643 free(rename_dst); 644 rename_dst = NULL; 645 rename_dst_nr = rename_dst_alloc = 0; 646 free(rename_src); 647 rename_src = NULL; 648 rename_src_nr = rename_src_alloc = 0; 649 return; 650}