1/* 2 * Copyright (C) 2005 Junio C Hamano 3 */ 4#include "cache.h" 5#include "diff.h" 6#include "diffcore.h" 7#include "hashmap.h" 8#include "progress.h" 9 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; 17 18static int find_rename_dst(struct diff_filespec *two) 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 next; 30 if (cmp < 0) { 31 last = next; 32 continue; 33 } 34 first = next+1; 35 } 36 return -first - 1; 37} 38 39static struct diff_rename_dst *locate_rename_dst(struct diff_filespec *two) 40{ 41 int ofs = find_rename_dst(two); 42 return ofs < 0 ? NULL : &rename_dst[ofs]; 43} 44 45/* 46 * Returns 0 on success, -1 if we found a duplicate. 47 */ 48static int add_rename_dst(struct diff_filespec *two) 49{ 50 int first = find_rename_dst(two); 51 52 if (first >= 0) 53 return -1; 54 first = -first - 1; 55 56 /* insert to make it at "first" */ 57 ALLOC_GROW(rename_dst, rename_dst_nr + 1, rename_dst_alloc); 58 rename_dst_nr++; 59 if (first < rename_dst_nr) 60 memmove(rename_dst + first + 1, rename_dst + first, 61 (rename_dst_nr - first - 1) * sizeof(*rename_dst)); 62 rename_dst[first].two = alloc_filespec(two->path); 63 fill_filespec(rename_dst[first].two, two->oid.hash, two->oid_valid, 64 two->mode); 65 rename_dst[first].pair = NULL; 66 return 0; 67} 68 69/* Table of rename/copy src files */ 70static struct diff_rename_src { 71 struct diff_filepair *p; 72 unsigned short score; /* to remember the break score */ 73} *rename_src; 74static int rename_src_nr, rename_src_alloc; 75 76static struct diff_rename_src *register_rename_src(struct diff_filepair *p) 77{ 78 int first, last; 79 struct diff_filespec *one = p->one; 80 unsigned short score = p->score; 81 82 first = 0; 83 last = rename_src_nr; 84 while (last > first) { 85 int next = (last + first) >> 1; 86 struct diff_rename_src *src = &(rename_src[next]); 87 int cmp = strcmp(one->path, src->p->one->path); 88 if (!cmp) 89 return src; 90 if (cmp < 0) { 91 last = next; 92 continue; 93 } 94 first = next+1; 95 } 96 97 /* insert to make it at "first" */ 98 ALLOC_GROW(rename_src, rename_src_nr + 1, rename_src_alloc); 99 rename_src_nr++; 100 if (first < rename_src_nr) 101 memmove(rename_src + first + 1, rename_src + first, 102 (rename_src_nr - first - 1) * sizeof(*rename_src)); 103 rename_src[first].p = p; 104 rename_src[first].score = score; 105 return &(rename_src[first]); 106} 107 108static int basename_same(struct diff_filespec *src, struct diff_filespec *dst) 109{ 110 int src_len = strlen(src->path), dst_len = strlen(dst->path); 111 while (src_len && dst_len) { 112 char c1 = src->path[--src_len]; 113 char c2 = dst->path[--dst_len]; 114 if (c1 != c2) 115 return 0; 116 if (c1 == '/') 117 return 1; 118 } 119 return (!src_len || src->path[src_len - 1] == '/') && 120 (!dst_len || dst->path[dst_len - 1] == '/'); 121} 122 123struct diff_score { 124 int src; /* index in rename_src */ 125 int dst; /* index in rename_dst */ 126 unsigned short score; 127 short name_score; 128}; 129 130static int estimate_similarity(struct diff_filespec *src, 131 struct diff_filespec *dst, 132 int minimum_score) 133{ 134 /* src points at a file that existed in the original tree (or 135 * optionally a file in the destination tree) and dst points 136 * at a newly created file. They may be quite similar, in which 137 * case we want to say src is renamed to dst or src is copied into 138 * dst, and then some edit has been applied to dst. 139 * 140 * Compare them and return how similar they are, representing 141 * the score as an integer between 0 and MAX_SCORE. 142 * 143 * When there is an exact match, it is considered a better 144 * match than anything else; the destination does not even 145 * call into this function in that case. 146 */ 147 unsigned long max_size, delta_size, base_size, src_copied, literal_added; 148 unsigned long delta_limit; 149 int score; 150 151 /* We deal only with regular files. Symlink renames are handled 152 * only when they are exact matches --- in other words, no edits 153 * after renaming. 154 */ 155 if (!S_ISREG(src->mode) || !S_ISREG(dst->mode)) 156 return 0; 157 158 /* 159 * Need to check that source and destination sizes are 160 * filled in before comparing them. 161 * 162 * If we already have "cnt_data" filled in, we know it's 163 * all good (avoid checking the size for zero, as that 164 * is a possible size - we really should have a flag to 165 * say whether the size is valid or not!) 166 */ 167 if (!src->cnt_data && 168 diff_populate_filespec(src, CHECK_SIZE_ONLY)) 169 return 0; 170 if (!dst->cnt_data && 171 diff_populate_filespec(dst, CHECK_SIZE_ONLY)) 172 return 0; 173 174 max_size = ((src->size > dst->size) ? src->size : dst->size); 175 base_size = ((src->size < dst->size) ? src->size : dst->size); 176 delta_size = max_size - base_size; 177 178 /* We would not consider edits that change the file size so 179 * drastically. delta_size must be smaller than 180 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size). 181 * 182 * Note that base_size == 0 case is handled here already 183 * and the final score computation below would not have a 184 * divide-by-zero issue. 185 */ 186 if (max_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE) 187 return 0; 188 189 if (!src->cnt_data && diff_populate_filespec(src, 0)) 190 return 0; 191 if (!dst->cnt_data && diff_populate_filespec(dst, 0)) 192 return 0; 193 194 delta_limit = (unsigned long) 195 (base_size * (MAX_SCORE-minimum_score) / MAX_SCORE); 196 if (diffcore_count_changes(src, dst, 197 &src->cnt_data, &dst->cnt_data, 198 delta_limit, 199 &src_copied, &literal_added)) 200 return 0; 201 202 /* How similar are they? 203 * what percentage of material in dst are from source? 204 */ 205 if (!dst->size) 206 score = 0; /* should not happen */ 207 else 208 score = (int)(src_copied * MAX_SCORE / max_size); 209 return score; 210} 211 212static void record_rename_pair(int dst_index, int src_index, int score) 213{ 214 struct diff_filespec *src, *dst; 215 struct diff_filepair *dp; 216 217 if (rename_dst[dst_index].pair) 218 die("internal error: dst already matched."); 219 220 src = rename_src[src_index].p->one; 221 src->rename_used++; 222 src->count++; 223 224 dst = rename_dst[dst_index].two; 225 dst->count++; 226 227 dp = diff_queue(NULL, src, dst); 228 dp->renamed_pair = 1; 229 if (!strcmp(src->path, dst->path)) 230 dp->score = rename_src[src_index].score; 231 else 232 dp->score = score; 233 rename_dst[dst_index].pair = dp; 234} 235 236/* 237 * We sort the rename similarity matrix with the score, in descending 238 * order (the most similar first). 239 */ 240static int score_compare(const void *a_, const void *b_) 241{ 242 const struct diff_score *a = a_, *b = b_; 243 244 /* sink the unused ones to the bottom */ 245 if (a->dst < 0) 246 return (0 <= b->dst); 247 else if (b->dst < 0) 248 return -1; 249 250 if (a->score == b->score) 251 return b->name_score - a->name_score; 252 253 return b->score - a->score; 254} 255 256struct file_similarity { 257 struct hashmap_entry entry; 258 int index; 259 struct diff_filespec *filespec; 260}; 261 262static unsigned int hash_filespec(struct diff_filespec *filespec) 263{ 264 if (!filespec->oid_valid) { 265 if (diff_populate_filespec(filespec, 0)) 266 return 0; 267 hash_sha1_file(filespec->data, filespec->size, "blob", 268 filespec->oid.hash); 269 } 270 return sha1hash(filespec->oid.hash); 271} 272 273static int find_identical_files(struct hashmap *srcs, 274 int dst_index, 275 struct diff_options *options) 276{ 277 int renames = 0; 278 279 struct diff_filespec *target = rename_dst[dst_index].two; 280 struct file_similarity *p, *best = NULL; 281 int i = 100, best_score = -1; 282 283 /* 284 * Find the best source match for specified destination. 285 */ 286 p = hashmap_get_from_hash(srcs, hash_filespec(target), NULL); 287 for (; p; p = hashmap_get_next(srcs, p)) { 288 int score; 289 struct diff_filespec *source = p->filespec; 290 291 /* False hash collision? */ 292 if (oidcmp(&source->oid, &target->oid)) 293 continue; 294 /* Non-regular files? If so, the modes must match! */ 295 if (!S_ISREG(source->mode) || !S_ISREG(target->mode)) { 296 if (source->mode != target->mode) 297 continue; 298 } 299 /* Give higher scores to sources that haven't been used already */ 300 score = !source->rename_used; 301 if (source->rename_used && options->detect_rename != DIFF_DETECT_COPY) 302 continue; 303 score += basename_same(source, target); 304 if (score > best_score) { 305 best = p; 306 best_score = score; 307 if (score == 2) 308 break; 309 } 310 311 /* Too many identical alternatives? Pick one */ 312 if (!--i) 313 break; 314 } 315 if (best) { 316 record_rename_pair(dst_index, best->index, MAX_SCORE); 317 renames++; 318 } 319 return renames; 320} 321 322static void insert_file_table(struct hashmap *table, int index, struct diff_filespec *filespec) 323{ 324 struct file_similarity *entry = xmalloc(sizeof(*entry)); 325 326 entry->index = index; 327 entry->filespec = filespec; 328 329 hashmap_entry_init(entry, hash_filespec(filespec)); 330 hashmap_add(table, entry); 331} 332 333/* 334 * Find exact renames first. 335 * 336 * The first round matches up the up-to-date entries, 337 * and then during the second round we try to match 338 * cache-dirty entries as well. 339 */ 340static int find_exact_renames(struct diff_options *options) 341{ 342 int i, renames = 0; 343 struct hashmap file_table; 344 345 /* Add all sources to the hash table in reverse order, because 346 * later on they will be retrieved in LIFO order. 347 */ 348 hashmap_init(&file_table, NULL, rename_src_nr); 349 for (i = rename_src_nr-1; i >= 0; i--) 350 insert_file_table(&file_table, i, rename_src[i].p->one); 351 352 /* Walk the destinations and find best source match */ 353 for (i = 0; i < rename_dst_nr; i++) 354 renames += find_identical_files(&file_table, i, options); 355 356 /* Free the hash data structure and entries */ 357 hashmap_free(&file_table, 1); 358 359 return renames; 360} 361 362#define NUM_CANDIDATE_PER_DST 4 363static void record_if_better(struct diff_score m[], struct diff_score *o) 364{ 365 int i, worst; 366 367 /* find the worst one */ 368 worst = 0; 369 for (i = 1; i < NUM_CANDIDATE_PER_DST; i++) 370 if (score_compare(&m[i], &m[worst]) > 0) 371 worst = i; 372 373 /* is it better than the worst one? */ 374 if (score_compare(&m[worst], o) > 0) 375 m[worst] = *o; 376} 377 378/* 379 * Returns: 380 * 0 if we are under the limit; 381 * 1 if we need to disable inexact rename detection; 382 * 2 if we would be under the limit if we were given -C instead of -C -C. 383 */ 384static int too_many_rename_candidates(int num_create, 385 struct diff_options *options) 386{ 387 int rename_limit = options->rename_limit; 388 int num_src = rename_src_nr; 389 int i; 390 391 options->needed_rename_limit = 0; 392 393 /* 394 * This basically does a test for the rename matrix not 395 * growing larger than a "rename_limit" square matrix, ie: 396 * 397 * num_create * num_src > rename_limit * rename_limit 398 * 399 * but handles the potential overflow case specially (and we 400 * assume at least 32-bit integers) 401 */ 402 if (rename_limit <= 0 || rename_limit > 32767) 403 rename_limit = 32767; 404 if ((num_create <= rename_limit || num_src <= rename_limit) && 405 (num_create * num_src <= rename_limit * rename_limit)) 406 return 0; 407 408 options->needed_rename_limit = 409 num_src > num_create ? num_src : num_create; 410 411 /* Are we running under -C -C? */ 412 if (!DIFF_OPT_TST(options, FIND_COPIES_HARDER)) 413 return 1; 414 415 /* Would we bust the limit if we were running under -C? */ 416 for (num_src = i = 0; i < rename_src_nr; i++) { 417 if (diff_unmodified_pair(rename_src[i].p)) 418 continue; 419 num_src++; 420 } 421 if ((num_create <= rename_limit || num_src <= rename_limit) && 422 (num_create * num_src <= rename_limit * rename_limit)) 423 return 2; 424 return 1; 425} 426 427static int find_renames(struct diff_score *mx, int dst_cnt, int minimum_score, int copies) 428{ 429 int count = 0, i; 430 431 for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) { 432 struct diff_rename_dst *dst; 433 434 if ((mx[i].dst < 0) || 435 (mx[i].score < minimum_score)) 436 break; /* there is no more usable pair. */ 437 dst = &rename_dst[mx[i].dst]; 438 if (dst->pair) 439 continue; /* already done, either exact or fuzzy. */ 440 if (!copies && rename_src[mx[i].src].p->one->rename_used) 441 continue; 442 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score); 443 count++; 444 } 445 return count; 446} 447 448void diffcore_rename(struct diff_options *options) 449{ 450 int detect_rename = options->detect_rename; 451 int minimum_score = options->rename_score; 452 struct diff_queue_struct *q = &diff_queued_diff; 453 struct diff_queue_struct outq; 454 struct diff_score *mx; 455 int i, j, rename_count, skip_unmodified = 0; 456 int num_create, dst_cnt; 457 struct progress *progress = NULL; 458 459 if (!minimum_score) 460 minimum_score = DEFAULT_RENAME_SCORE; 461 462 for (i = 0; i < q->nr; i++) { 463 struct diff_filepair *p = q->queue[i]; 464 if (!DIFF_FILE_VALID(p->one)) { 465 if (!DIFF_FILE_VALID(p->two)) 466 continue; /* unmerged */ 467 else if (options->single_follow && 468 strcmp(options->single_follow, p->two->path)) 469 continue; /* not interested */ 470 else if (!DIFF_OPT_TST(options, RENAME_EMPTY) && 471 is_empty_blob_sha1(p->two->oid.hash)) 472 continue; 473 else if (add_rename_dst(p->two) < 0) { 474 warning("skipping rename detection, detected" 475 " duplicate destination '%s'", 476 p->two->path); 477 goto cleanup; 478 } 479 } 480 else if (!DIFF_OPT_TST(options, RENAME_EMPTY) && 481 is_empty_blob_sha1(p->one->oid.hash)) 482 continue; 483 else if (!DIFF_PAIR_UNMERGED(p) && !DIFF_FILE_VALID(p->two)) { 484 /* 485 * If the source is a broken "delete", and 486 * they did not really want to get broken, 487 * that means the source actually stays. 488 * So we increment the "rename_used" score 489 * by one, to indicate ourselves as a user 490 */ 491 if (p->broken_pair && !p->score) 492 p->one->rename_used++; 493 register_rename_src(p); 494 } 495 else if (detect_rename == DIFF_DETECT_COPY) { 496 /* 497 * Increment the "rename_used" score by 498 * one, to indicate ourselves as a user. 499 */ 500 p->one->rename_used++; 501 register_rename_src(p); 502 } 503 } 504 if (rename_dst_nr == 0 || rename_src_nr == 0) 505 goto cleanup; /* nothing to do */ 506 507 /* 508 * We really want to cull the candidates list early 509 * with cheap tests in order to avoid doing deltas. 510 */ 511 rename_count = find_exact_renames(options); 512 513 /* Did we only want exact renames? */ 514 if (minimum_score == MAX_SCORE) 515 goto cleanup; 516 517 /* 518 * Calculate how many renames are left (but all the source 519 * files still remain as options for rename/copies!) 520 */ 521 num_create = (rename_dst_nr - rename_count); 522 523 /* All done? */ 524 if (!num_create) 525 goto cleanup; 526 527 switch (too_many_rename_candidates(num_create, options)) { 528 case 1: 529 goto cleanup; 530 case 2: 531 options->degraded_cc_to_c = 1; 532 skip_unmodified = 1; 533 break; 534 default: 535 break; 536 } 537 538 if (options->show_rename_progress) { 539 progress = start_progress_delay( 540 _("Performing inexact rename detection"), 541 rename_dst_nr * rename_src_nr, 50, 1); 542 } 543 544 mx = xcalloc(st_mult(num_create, NUM_CANDIDATE_PER_DST), sizeof(*mx)); 545 for (dst_cnt = i = 0; i < rename_dst_nr; i++) { 546 struct diff_filespec *two = rename_dst[i].two; 547 struct diff_score *m; 548 549 if (rename_dst[i].pair) 550 continue; /* dealt with exact match already. */ 551 552 m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST]; 553 for (j = 0; j < NUM_CANDIDATE_PER_DST; j++) 554 m[j].dst = -1; 555 556 for (j = 0; j < rename_src_nr; j++) { 557 struct diff_filespec *one = rename_src[j].p->one; 558 struct diff_score this_src; 559 560 if (skip_unmodified && 561 diff_unmodified_pair(rename_src[j].p)) 562 continue; 563 564 this_src.score = estimate_similarity(one, two, 565 minimum_score); 566 this_src.name_score = basename_same(one, two); 567 this_src.dst = i; 568 this_src.src = j; 569 record_if_better(m, &this_src); 570 /* 571 * Once we run estimate_similarity, 572 * We do not need the text anymore. 573 */ 574 diff_free_filespec_blob(one); 575 diff_free_filespec_blob(two); 576 } 577 dst_cnt++; 578 display_progress(progress, (i+1)*rename_src_nr); 579 } 580 stop_progress(&progress); 581 582 /* cost matrix sorted by most to least similar pair */ 583 qsort(mx, dst_cnt * NUM_CANDIDATE_PER_DST, sizeof(*mx), score_compare); 584 585 rename_count += find_renames(mx, dst_cnt, minimum_score, 0); 586 if (detect_rename == DIFF_DETECT_COPY) 587 rename_count += find_renames(mx, dst_cnt, minimum_score, 1); 588 free(mx); 589 590 cleanup: 591 /* At this point, we have found some renames and copies and they 592 * are recorded in rename_dst. The original list is still in *q. 593 */ 594 DIFF_QUEUE_CLEAR(&outq); 595 for (i = 0; i < q->nr; i++) { 596 struct diff_filepair *p = q->queue[i]; 597 struct diff_filepair *pair_to_free = NULL; 598 599 if (DIFF_PAIR_UNMERGED(p)) { 600 diff_q(&outq, p); 601 } 602 else if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) { 603 /* 604 * Creation 605 * 606 * We would output this create record if it has 607 * not been turned into a rename/copy already. 608 */ 609 struct diff_rename_dst *dst = locate_rename_dst(p->two); 610 if (dst && dst->pair) { 611 diff_q(&outq, dst->pair); 612 pair_to_free = p; 613 } 614 else 615 /* no matching rename/copy source, so 616 * record this as a creation. 617 */ 618 diff_q(&outq, p); 619 } 620 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) { 621 /* 622 * Deletion 623 * 624 * We would output this delete record if: 625 * 626 * (1) this is a broken delete and the counterpart 627 * broken create remains in the output; or 628 * (2) this is not a broken delete, and rename_dst 629 * does not have a rename/copy to move p->one->path 630 * out of existence. 631 * 632 * Otherwise, the counterpart broken create 633 * has been turned into a rename-edit; or 634 * delete did not have a matching create to 635 * begin with. 636 */ 637 if (DIFF_PAIR_BROKEN(p)) { 638 /* broken delete */ 639 struct diff_rename_dst *dst = locate_rename_dst(p->one); 640 if (dst && dst->pair) 641 /* counterpart is now rename/copy */ 642 pair_to_free = p; 643 } 644 else { 645 if (p->one->rename_used) 646 /* this path remains */ 647 pair_to_free = p; 648 } 649 650 if (pair_to_free) 651 ; 652 else 653 diff_q(&outq, p); 654 } 655 else if (!diff_unmodified_pair(p)) 656 /* all the usual ones need to be kept */ 657 diff_q(&outq, p); 658 else 659 /* no need to keep unmodified pairs */ 660 pair_to_free = p; 661 662 if (pair_to_free) 663 diff_free_filepair(pair_to_free); 664 } 665 diff_debug_queue("done copying original", &outq); 666 667 free(q->queue); 668 *q = outq; 669 diff_debug_queue("done collapsing", q); 670 671 for (i = 0; i < rename_dst_nr; i++) 672 free_filespec(rename_dst[i].two); 673 674 free(rename_dst); 675 rename_dst = NULL; 676 rename_dst_nr = rename_dst_alloc = 0; 677 free(rename_src); 678 rename_src = NULL; 679 rename_src_nr = rename_src_alloc = 0; 680 return; 681}