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, 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 int score; 149 150 /* We deal only with regular files. Symlink renames are handled 151 * only when they are exact matches --- in other words, no edits 152 * after renaming. 153 */ 154 if (!S_ISREG(src->mode) || !S_ISREG(dst->mode)) 155 return 0; 156 157 /* 158 * Need to check that source and destination sizes are 159 * filled in before comparing them. 160 * 161 * If we already have "cnt_data" filled in, we know it's 162 * all good (avoid checking the size for zero, as that 163 * is a possible size - we really should have a flag to 164 * say whether the size is valid or not!) 165 */ 166 if (!src->cnt_data && 167 diff_populate_filespec(src, CHECK_SIZE_ONLY)) 168 return 0; 169 if (!dst->cnt_data && 170 diff_populate_filespec(dst, CHECK_SIZE_ONLY)) 171 return 0; 172 173 max_size = ((src->size > dst->size) ? src->size : dst->size); 174 base_size = ((src->size < dst->size) ? src->size : dst->size); 175 delta_size = max_size - base_size; 176 177 /* We would not consider edits that change the file size so 178 * drastically. delta_size must be smaller than 179 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size). 180 * 181 * Note that base_size == 0 case is handled here already 182 * and the final score computation below would not have a 183 * divide-by-zero issue. 184 */ 185 if (max_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE) 186 return 0; 187 188 if (!src->cnt_data && diff_populate_filespec(src, 0)) 189 return 0; 190 if (!dst->cnt_data && diff_populate_filespec(dst, 0)) 191 return 0; 192 193 if (diffcore_count_changes(src, dst, 194 &src->cnt_data, &dst->cnt_data, 195 &src_copied, &literal_added)) 196 return 0; 197 198 /* How similar are they? 199 * what percentage of material in dst are from source? 200 */ 201 if (!dst->size) 202 score = 0; /* should not happen */ 203 else 204 score = (int)(src_copied * MAX_SCORE / max_size); 205 return score; 206} 207 208static void record_rename_pair(int dst_index, int src_index, int score) 209{ 210 struct diff_filespec *src, *dst; 211 struct diff_filepair *dp; 212 213 if (rename_dst[dst_index].pair) 214 die("internal error: dst already matched."); 215 216 src = rename_src[src_index].p->one; 217 src->rename_used++; 218 src->count++; 219 220 dst = rename_dst[dst_index].two; 221 dst->count++; 222 223 dp = diff_queue(NULL, src, dst); 224 dp->renamed_pair = 1; 225 if (!strcmp(src->path, dst->path)) 226 dp->score = rename_src[src_index].score; 227 else 228 dp->score = score; 229 rename_dst[dst_index].pair = dp; 230} 231 232/* 233 * We sort the rename similarity matrix with the score, in descending 234 * order (the most similar first). 235 */ 236static int score_compare(const void *a_, const void *b_) 237{ 238 const struct diff_score *a = a_, *b = b_; 239 240 /* sink the unused ones to the bottom */ 241 if (a->dst < 0) 242 return (0 <= b->dst); 243 else if (b->dst < 0) 244 return -1; 245 246 if (a->score == b->score) 247 return b->name_score - a->name_score; 248 249 return b->score - a->score; 250} 251 252struct file_similarity { 253 struct hashmap_entry entry; 254 int index; 255 struct diff_filespec *filespec; 256}; 257 258static unsigned int hash_filespec(struct diff_filespec *filespec) 259{ 260 if (!filespec->oid_valid) { 261 if (diff_populate_filespec(filespec, 0)) 262 return 0; 263 hash_sha1_file(filespec->data, filespec->size, "blob", 264 filespec->oid.hash); 265 } 266 return sha1hash(filespec->oid.hash); 267} 268 269static int find_identical_files(struct hashmap *srcs, 270 int dst_index, 271 struct diff_options *options) 272{ 273 int renames = 0; 274 275 struct diff_filespec *target = rename_dst[dst_index].two; 276 struct file_similarity *p, *best = NULL; 277 int i = 100, best_score = -1; 278 279 /* 280 * Find the best source match for specified destination. 281 */ 282 p = hashmap_get_from_hash(srcs, hash_filespec(target), NULL); 283 for (; p; p = hashmap_get_next(srcs, p)) { 284 int score; 285 struct diff_filespec *source = p->filespec; 286 287 /* False hash collision? */ 288 if (oidcmp(&source->oid, &target->oid)) 289 continue; 290 /* Non-regular files? If so, the modes must match! */ 291 if (!S_ISREG(source->mode) || !S_ISREG(target->mode)) { 292 if (source->mode != target->mode) 293 continue; 294 } 295 /* Give higher scores to sources that haven't been used already */ 296 score = !source->rename_used; 297 if (source->rename_used && options->detect_rename != DIFF_DETECT_COPY) 298 continue; 299 score += basename_same(source, target); 300 if (score > best_score) { 301 best = p; 302 best_score = score; 303 if (score == 2) 304 break; 305 } 306 307 /* Too many identical alternatives? Pick one */ 308 if (!--i) 309 break; 310 } 311 if (best) { 312 record_rename_pair(dst_index, best->index, MAX_SCORE); 313 renames++; 314 } 315 return renames; 316} 317 318static void insert_file_table(struct hashmap *table, int index, struct diff_filespec *filespec) 319{ 320 struct file_similarity *entry = xmalloc(sizeof(*entry)); 321 322 entry->index = index; 323 entry->filespec = filespec; 324 325 hashmap_entry_init(entry, hash_filespec(filespec)); 326 hashmap_add(table, entry); 327} 328 329/* 330 * Find exact renames first. 331 * 332 * The first round matches up the up-to-date entries, 333 * and then during the second round we try to match 334 * cache-dirty entries as well. 335 */ 336static int find_exact_renames(struct diff_options *options) 337{ 338 int i, renames = 0; 339 struct hashmap file_table; 340 341 /* Add all sources to the hash table in reverse order, because 342 * later on they will be retrieved in LIFO order. 343 */ 344 hashmap_init(&file_table, NULL, NULL, rename_src_nr); 345 for (i = rename_src_nr-1; i >= 0; i--) 346 insert_file_table(&file_table, i, rename_src[i].p->one); 347 348 /* Walk the destinations and find best source match */ 349 for (i = 0; i < rename_dst_nr; i++) 350 renames += find_identical_files(&file_table, i, options); 351 352 /* Free the hash data structure and entries */ 353 hashmap_free(&file_table, 1); 354 355 return renames; 356} 357 358#define NUM_CANDIDATE_PER_DST 4 359static void record_if_better(struct diff_score m[], struct diff_score *o) 360{ 361 int i, worst; 362 363 /* find the worst one */ 364 worst = 0; 365 for (i = 1; i < NUM_CANDIDATE_PER_DST; i++) 366 if (score_compare(&m[i], &m[worst]) > 0) 367 worst = i; 368 369 /* is it better than the worst one? */ 370 if (score_compare(&m[worst], o) > 0) 371 m[worst] = *o; 372} 373 374/* 375 * Returns: 376 * 0 if we are under the limit; 377 * 1 if we need to disable inexact rename detection; 378 * 2 if we would be under the limit if we were given -C instead of -C -C. 379 */ 380static int too_many_rename_candidates(int num_create, 381 struct diff_options *options) 382{ 383 int rename_limit = options->rename_limit; 384 int num_src = rename_src_nr; 385 int i; 386 387 options->needed_rename_limit = 0; 388 389 /* 390 * This basically does a test for the rename matrix not 391 * growing larger than a "rename_limit" square matrix, ie: 392 * 393 * num_create * num_src > rename_limit * rename_limit 394 */ 395 if ((num_create <= rename_limit || num_src <= rename_limit) && 396 ((uint64_t)num_create * (uint64_t)num_src 397 <= (uint64_t)rename_limit * (uint64_t)rename_limit)) 398 return 0; 399 400 options->needed_rename_limit = 401 num_src > num_create ? num_src : num_create; 402 403 /* Are we running under -C -C? */ 404 if (!options->flags.find_copies_harder) 405 return 1; 406 407 /* Would we bust the limit if we were running under -C? */ 408 for (num_src = i = 0; i < rename_src_nr; i++) { 409 if (diff_unmodified_pair(rename_src[i].p)) 410 continue; 411 num_src++; 412 } 413 if ((num_create <= rename_limit || num_src <= rename_limit) && 414 ((uint64_t)num_create * (uint64_t)num_src 415 <= (uint64_t)rename_limit * (uint64_t)rename_limit)) 416 return 2; 417 return 1; 418} 419 420static int find_renames(struct diff_score *mx, int dst_cnt, int minimum_score, int copies) 421{ 422 int count = 0, i; 423 424 for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) { 425 struct diff_rename_dst *dst; 426 427 if ((mx[i].dst < 0) || 428 (mx[i].score < minimum_score)) 429 break; /* there is no more usable pair. */ 430 dst = &rename_dst[mx[i].dst]; 431 if (dst->pair) 432 continue; /* already done, either exact or fuzzy. */ 433 if (!copies && rename_src[mx[i].src].p->one->rename_used) 434 continue; 435 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score); 436 count++; 437 } 438 return count; 439} 440 441void diffcore_rename(struct diff_options *options) 442{ 443 int detect_rename = options->detect_rename; 444 int minimum_score = options->rename_score; 445 struct diff_queue_struct *q = &diff_queued_diff; 446 struct diff_queue_struct outq; 447 struct diff_score *mx; 448 int i, j, rename_count, skip_unmodified = 0; 449 int num_create, dst_cnt; 450 struct progress *progress = NULL; 451 452 if (!minimum_score) 453 minimum_score = DEFAULT_RENAME_SCORE; 454 455 for (i = 0; i < q->nr; i++) { 456 struct diff_filepair *p = q->queue[i]; 457 if (!DIFF_FILE_VALID(p->one)) { 458 if (!DIFF_FILE_VALID(p->two)) 459 continue; /* unmerged */ 460 else if (options->single_follow && 461 strcmp(options->single_follow, p->two->path)) 462 continue; /* not interested */ 463 else if (!options->flags.rename_empty && 464 is_empty_blob_oid(&p->two->oid)) 465 continue; 466 else if (add_rename_dst(p->two) < 0) { 467 warning("skipping rename detection, detected" 468 " duplicate destination '%s'", 469 p->two->path); 470 goto cleanup; 471 } 472 } 473 else if (!options->flags.rename_empty && 474 is_empty_blob_oid(&p->one->oid)) 475 continue; 476 else if (!DIFF_PAIR_UNMERGED(p) && !DIFF_FILE_VALID(p->two)) { 477 /* 478 * If the source is a broken "delete", and 479 * they did not really want to get broken, 480 * that means the source actually stays. 481 * So we increment the "rename_used" score 482 * by one, to indicate ourselves as a user 483 */ 484 if (p->broken_pair && !p->score) 485 p->one->rename_used++; 486 register_rename_src(p); 487 } 488 else if (detect_rename == DIFF_DETECT_COPY) { 489 /* 490 * Increment the "rename_used" score by 491 * one, to indicate ourselves as a user. 492 */ 493 p->one->rename_used++; 494 register_rename_src(p); 495 } 496 } 497 if (rename_dst_nr == 0 || rename_src_nr == 0) 498 goto cleanup; /* nothing to do */ 499 500 /* 501 * We really want to cull the candidates list early 502 * with cheap tests in order to avoid doing deltas. 503 */ 504 rename_count = find_exact_renames(options); 505 506 /* Did we only want exact renames? */ 507 if (minimum_score == MAX_SCORE) 508 goto cleanup; 509 510 /* 511 * Calculate how many renames are left (but all the source 512 * files still remain as options for rename/copies!) 513 */ 514 num_create = (rename_dst_nr - rename_count); 515 516 /* All done? */ 517 if (!num_create) 518 goto cleanup; 519 520 switch (too_many_rename_candidates(num_create, options)) { 521 case 1: 522 goto cleanup; 523 case 2: 524 options->degraded_cc_to_c = 1; 525 skip_unmodified = 1; 526 break; 527 default: 528 break; 529 } 530 531 if (options->show_rename_progress) { 532 progress = start_delayed_progress( 533 _("Performing inexact rename detection"), 534 (uint64_t)rename_dst_nr * (uint64_t)rename_src_nr); 535 } 536 537 mx = xcalloc(st_mult(NUM_CANDIDATE_PER_DST, num_create), sizeof(*mx)); 538 for (dst_cnt = i = 0; i < rename_dst_nr; i++) { 539 struct diff_filespec *two = rename_dst[i].two; 540 struct diff_score *m; 541 542 if (rename_dst[i].pair) 543 continue; /* dealt with exact match already. */ 544 545 m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST]; 546 for (j = 0; j < NUM_CANDIDATE_PER_DST; j++) 547 m[j].dst = -1; 548 549 for (j = 0; j < rename_src_nr; j++) { 550 struct diff_filespec *one = rename_src[j].p->one; 551 struct diff_score this_src; 552 553 if (skip_unmodified && 554 diff_unmodified_pair(rename_src[j].p)) 555 continue; 556 557 this_src.score = estimate_similarity(one, two, 558 minimum_score); 559 this_src.name_score = basename_same(one, two); 560 this_src.dst = i; 561 this_src.src = j; 562 record_if_better(m, &this_src); 563 /* 564 * Once we run estimate_similarity, 565 * We do not need the text anymore. 566 */ 567 diff_free_filespec_blob(one); 568 diff_free_filespec_blob(two); 569 } 570 dst_cnt++; 571 display_progress(progress, (uint64_t)(i+1)*(uint64_t)rename_src_nr); 572 } 573 stop_progress(&progress); 574 575 /* cost matrix sorted by most to least similar pair */ 576 QSORT(mx, dst_cnt * NUM_CANDIDATE_PER_DST, score_compare); 577 578 rename_count += find_renames(mx, dst_cnt, minimum_score, 0); 579 if (detect_rename == DIFF_DETECT_COPY) 580 rename_count += find_renames(mx, dst_cnt, minimum_score, 1); 581 free(mx); 582 583 cleanup: 584 /* At this point, we have found some renames and copies and they 585 * are recorded in rename_dst. The original list is still in *q. 586 */ 587 DIFF_QUEUE_CLEAR(&outq); 588 for (i = 0; i < q->nr; i++) { 589 struct diff_filepair *p = q->queue[i]; 590 struct diff_filepair *pair_to_free = NULL; 591 592 if (DIFF_PAIR_UNMERGED(p)) { 593 diff_q(&outq, p); 594 } 595 else if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) { 596 /* 597 * Creation 598 * 599 * We would output this create record if it has 600 * not been turned into a rename/copy already. 601 */ 602 struct diff_rename_dst *dst = locate_rename_dst(p->two); 603 if (dst && dst->pair) { 604 diff_q(&outq, dst->pair); 605 pair_to_free = p; 606 } 607 else 608 /* no matching rename/copy source, so 609 * record this as a creation. 610 */ 611 diff_q(&outq, p); 612 } 613 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) { 614 /* 615 * Deletion 616 * 617 * We would output this delete record if: 618 * 619 * (1) this is a broken delete and the counterpart 620 * broken create remains in the output; or 621 * (2) this is not a broken delete, and rename_dst 622 * does not have a rename/copy to move p->one->path 623 * out of existence. 624 * 625 * Otherwise, the counterpart broken create 626 * has been turned into a rename-edit; or 627 * delete did not have a matching create to 628 * begin with. 629 */ 630 if (DIFF_PAIR_BROKEN(p)) { 631 /* broken delete */ 632 struct diff_rename_dst *dst = locate_rename_dst(p->one); 633 if (dst && dst->pair) 634 /* counterpart is now rename/copy */ 635 pair_to_free = p; 636 } 637 else { 638 if (p->one->rename_used) 639 /* this path remains */ 640 pair_to_free = p; 641 } 642 643 if (pair_to_free) 644 ; 645 else 646 diff_q(&outq, p); 647 } 648 else if (!diff_unmodified_pair(p)) 649 /* all the usual ones need to be kept */ 650 diff_q(&outq, p); 651 else 652 /* no need to keep unmodified pairs */ 653 pair_to_free = p; 654 655 if (pair_to_free) 656 diff_free_filepair(pair_to_free); 657 } 658 diff_debug_queue("done copying original", &outq); 659 660 free(q->queue); 661 *q = outq; 662 diff_debug_queue("done collapsing", q); 663 664 for (i = 0; i < rename_dst_nr; i++) 665 free_filespec(rename_dst[i].two); 666 667 FREE_AND_NULL(rename_dst); 668 rename_dst_nr = rename_dst_alloc = 0; 669 FREE_AND_NULL(rename_src); 670 rename_src_nr = rename_src_alloc = 0; 671 return; 672}