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->sha1, two->sha1_valid, two->mode); 64 rename_dst[first].pair = NULL; 65 return 0; 66} 67 68/* Table of rename/copy src files */ 69static struct diff_rename_src { 70 struct diff_filepair *p; 71 unsigned short score; /* to remember the break score */ 72} *rename_src; 73static int rename_src_nr, rename_src_alloc; 74 75static struct diff_rename_src *register_rename_src(struct diff_filepair *p) 76{ 77 int first, last; 78 struct diff_filespec *one = p->one; 79 unsigned short score = p->score; 80 81 first = 0; 82 last = rename_src_nr; 83 while (last > first) { 84 int next = (last + first) >> 1; 85 struct diff_rename_src *src = &(rename_src[next]); 86 int cmp = strcmp(one->path, src->p->one->path); 87 if (!cmp) 88 return src; 89 if (cmp < 0) { 90 last = next; 91 continue; 92 } 93 first = next+1; 94 } 95 96 /* insert to make it at "first" */ 97 ALLOC_GROW(rename_src, rename_src_nr + 1, rename_src_alloc); 98 rename_src_nr++; 99 if (first < rename_src_nr) 100 memmove(rename_src + first + 1, rename_src + first, 101 (rename_src_nr - first - 1) * sizeof(*rename_src)); 102 rename_src[first].p = p; 103 rename_src[first].score = score; 104 return &(rename_src[first]); 105} 106 107static int basename_same(struct diff_filespec *src, struct diff_filespec *dst) 108{ 109 int src_len = strlen(src->path), dst_len = strlen(dst->path); 110 while (src_len && dst_len) { 111 char c1 = src->path[--src_len]; 112 char c2 = dst->path[--dst_len]; 113 if (c1 != c2) 114 return 0; 115 if (c1 == '/') 116 return 1; 117 } 118 return (!src_len || src->path[src_len - 1] == '/') && 119 (!dst_len || dst->path[dst_len - 1] == '/'); 120} 121 122struct diff_score { 123 int src; /* index in rename_src */ 124 int dst; /* index in rename_dst */ 125 unsigned short score; 126 short name_score; 127}; 128 129static int estimate_similarity(struct diff_filespec *src, 130 struct diff_filespec *dst, 131 int minimum_score) 132{ 133 /* src points at a file that existed in the original tree (or 134 * optionally a file in the destination tree) and dst points 135 * at a newly created file. They may be quite similar, in which 136 * case we want to say src is renamed to dst or src is copied into 137 * dst, and then some edit has been applied to dst. 138 * 139 * Compare them and return how similar they are, representing 140 * the score as an integer between 0 and MAX_SCORE. 141 * 142 * When there is an exact match, it is considered a better 143 * match than anything else; the destination does not even 144 * call into this function in that case. 145 */ 146 unsigned long max_size, delta_size, base_size, src_copied, literal_added; 147 unsigned long delta_limit; 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 delta_limit = (unsigned long) 194 (base_size * (MAX_SCORE-minimum_score) / MAX_SCORE); 195 if (diffcore_count_changes(src, dst, 196 &src->cnt_data, &dst->cnt_data, 197 delta_limit, 198 &src_copied, &literal_added)) 199 return 0; 200 201 /* How similar are they? 202 * what percentage of material in dst are from source? 203 */ 204 if (!dst->size) 205 score = 0; /* should not happen */ 206 else 207 score = (int)(src_copied * MAX_SCORE / max_size); 208 return score; 209} 210 211static void record_rename_pair(int dst_index, int src_index, int score) 212{ 213 struct diff_filespec *src, *dst; 214 struct diff_filepair *dp; 215 216 if (rename_dst[dst_index].pair) 217 die("internal error: dst already matched."); 218 219 src = rename_src[src_index].p->one; 220 src->rename_used++; 221 src->count++; 222 223 dst = rename_dst[dst_index].two; 224 dst->count++; 225 226 dp = diff_queue(NULL, src, dst); 227 dp->renamed_pair = 1; 228 if (!strcmp(src->path, dst->path)) 229 dp->score = rename_src[src_index].score; 230 else 231 dp->score = score; 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 /* sink the unused ones to the bottom */ 244 if (a->dst < 0) 245 return (0 <= b->dst); 246 else if (b->dst < 0) 247 return -1; 248 249 if (a->score == b->score) 250 return b->name_score - a->name_score; 251 252 return b->score - a->score; 253} 254 255struct file_similarity { 256 struct hashmap_entry entry; 257 int index; 258 struct diff_filespec *filespec; 259}; 260 261static unsigned int hash_filespec(struct diff_filespec *filespec) 262{ 263 if (!filespec->sha1_valid) { 264 if (diff_populate_filespec(filespec, 0)) 265 return 0; 266 hash_sha1_file(filespec->data, filespec->size, "blob", filespec->sha1); 267 } 268 return sha1hash(filespec->sha1); 269} 270 271static int find_identical_files(struct hashmap *srcs, 272 int dst_index, 273 struct diff_options *options) 274{ 275 int renames = 0; 276 277 struct diff_filespec *target = rename_dst[dst_index].two; 278 struct file_similarity *p, *best = NULL; 279 int i = 100, best_score = -1; 280 281 /* 282 * Find the best source match for specified destination. 283 */ 284 p = hashmap_get_from_hash(srcs, hash_filespec(target), NULL); 285 for (; p; p = hashmap_get_next(srcs, p)) { 286 int score; 287 struct diff_filespec *source = p->filespec; 288 289 /* False hash collision? */ 290 if (hashcmp(source->sha1, target->sha1)) 291 continue; 292 /* Non-regular files? If so, the modes must match! */ 293 if (!S_ISREG(source->mode) || !S_ISREG(target->mode)) { 294 if (source->mode != target->mode) 295 continue; 296 } 297 /* Give higher scores to sources that haven't been used already */ 298 score = !source->rename_used; 299 if (source->rename_used && options->detect_rename != DIFF_DETECT_COPY) 300 continue; 301 score += basename_same(source, target); 302 if (score > best_score) { 303 best = p; 304 best_score = score; 305 if (score == 2) 306 break; 307 } 308 309 /* Too many identical alternatives? Pick one */ 310 if (!--i) 311 break; 312 } 313 if (best) { 314 record_rename_pair(dst_index, best->index, MAX_SCORE); 315 renames++; 316 } 317 return renames; 318} 319 320static void insert_file_table(struct hashmap *table, int index, struct diff_filespec *filespec) 321{ 322 struct file_similarity *entry = xmalloc(sizeof(*entry)); 323 324 entry->index = index; 325 entry->filespec = filespec; 326 327 hashmap_entry_init(entry, hash_filespec(filespec)); 328 hashmap_add(table, entry); 329} 330 331/* 332 * Find exact renames first. 333 * 334 * The first round matches up the up-to-date entries, 335 * and then during the second round we try to match 336 * cache-dirty entries as well. 337 */ 338static int find_exact_renames(struct diff_options *options) 339{ 340 int i, renames = 0; 341 struct hashmap file_table; 342 343 /* Add all sources to the hash table in reverse order, because 344 * later on they will be retrieved in LIFO order. 345 */ 346 hashmap_init(&file_table, NULL, rename_src_nr); 347 for (i = rename_src_nr-1; i >= 0; i--) 348 insert_file_table(&file_table, i, rename_src[i].p->one); 349 350 /* Walk the destinations and find best source match */ 351 for (i = 0; i < rename_dst_nr; i++) 352 renames += find_identical_files(&file_table, i, options); 353 354 /* Free the hash data structure and entries */ 355 hashmap_free(&file_table, 1); 356 357 return renames; 358} 359 360#define NUM_CANDIDATE_PER_DST 4 361static void record_if_better(struct diff_score m[], struct diff_score *o) 362{ 363 int i, worst; 364 365 /* find the worst one */ 366 worst = 0; 367 for (i = 1; i < NUM_CANDIDATE_PER_DST; i++) 368 if (score_compare(&m[i], &m[worst]) > 0) 369 worst = i; 370 371 /* is it better than the worst one? */ 372 if (score_compare(&m[worst], o) > 0) 373 m[worst] = *o; 374} 375 376/* 377 * Returns: 378 * 0 if we are under the limit; 379 * 1 if we need to disable inexact rename detection; 380 * 2 if we would be under the limit if we were given -C instead of -C -C. 381 */ 382static int too_many_rename_candidates(int num_create, 383 struct diff_options *options) 384{ 385 int rename_limit = options->rename_limit; 386 int num_src = rename_src_nr; 387 int i; 388 389 options->needed_rename_limit = 0; 390 391 /* 392 * This basically does a test for the rename matrix not 393 * growing larger than a "rename_limit" square matrix, ie: 394 * 395 * num_create * num_src > rename_limit * rename_limit 396 * 397 * but handles the potential overflow case specially (and we 398 * assume at least 32-bit integers) 399 */ 400 if (rename_limit <= 0 || rename_limit > 32767) 401 rename_limit = 32767; 402 if ((num_create <= rename_limit || num_src <= rename_limit) && 403 (num_create * num_src <= rename_limit * rename_limit)) 404 return 0; 405 406 options->needed_rename_limit = 407 num_src > num_create ? num_src : num_create; 408 409 /* Are we running under -C -C? */ 410 if (!DIFF_OPT_TST(options, FIND_COPIES_HARDER)) 411 return 1; 412 413 /* Would we bust the limit if we were running under -C? */ 414 for (num_src = i = 0; i < rename_src_nr; i++) { 415 if (diff_unmodified_pair(rename_src[i].p)) 416 continue; 417 num_src++; 418 } 419 if ((num_create <= rename_limit || num_src <= rename_limit) && 420 (num_create * num_src <= rename_limit * rename_limit)) 421 return 2; 422 return 1; 423} 424 425static int find_renames(struct diff_score *mx, int dst_cnt, int minimum_score, int copies) 426{ 427 int count = 0, i; 428 429 for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) { 430 struct diff_rename_dst *dst; 431 432 if ((mx[i].dst < 0) || 433 (mx[i].score < minimum_score)) 434 break; /* there is no more usable pair. */ 435 dst = &rename_dst[mx[i].dst]; 436 if (dst->pair) 437 continue; /* already done, either exact or fuzzy. */ 438 if (!copies && rename_src[mx[i].src].p->one->rename_used) 439 continue; 440 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score); 441 count++; 442 } 443 return count; 444} 445 446void diffcore_rename(struct diff_options *options) 447{ 448 int detect_rename = options->detect_rename; 449 int minimum_score = options->rename_score; 450 struct diff_queue_struct *q = &diff_queued_diff; 451 struct diff_queue_struct outq; 452 struct diff_score *mx; 453 int i, j, rename_count, skip_unmodified = 0; 454 int num_create, dst_cnt; 455 struct progress *progress = NULL; 456 457 if (!minimum_score) 458 minimum_score = DEFAULT_RENAME_SCORE; 459 460 for (i = 0; i < q->nr; i++) { 461 struct diff_filepair *p = q->queue[i]; 462 if (!DIFF_FILE_VALID(p->one)) { 463 if (!DIFF_FILE_VALID(p->two)) 464 continue; /* unmerged */ 465 else if (options->single_follow && 466 strcmp(options->single_follow, p->two->path)) 467 continue; /* not interested */ 468 else if (!DIFF_OPT_TST(options, RENAME_EMPTY) && 469 is_empty_blob_sha1(p->two->sha1)) 470 continue; 471 else if (add_rename_dst(p->two) < 0) { 472 warning("skipping rename detection, detected" 473 " duplicate destination '%s'", 474 p->two->path); 475 goto cleanup; 476 } 477 } 478 else if (!DIFF_OPT_TST(options, RENAME_EMPTY) && 479 is_empty_blob_sha1(p->one->sha1)) 480 continue; 481 else if (!DIFF_PAIR_UNMERGED(p) && !DIFF_FILE_VALID(p->two)) { 482 /* 483 * If the source is a broken "delete", and 484 * they did not really want to get broken, 485 * that means the source actually stays. 486 * So we increment the "rename_used" score 487 * by one, to indicate ourselves as a user 488 */ 489 if (p->broken_pair && !p->score) 490 p->one->rename_used++; 491 register_rename_src(p); 492 } 493 else if (detect_rename == DIFF_DETECT_COPY) { 494 /* 495 * Increment the "rename_used" score by 496 * one, to indicate ourselves as a user. 497 */ 498 p->one->rename_used++; 499 register_rename_src(p); 500 } 501 } 502 if (rename_dst_nr == 0 || rename_src_nr == 0) 503 goto cleanup; /* nothing to do */ 504 505 /* 506 * We really want to cull the candidates list early 507 * with cheap tests in order to avoid doing deltas. 508 */ 509 rename_count = find_exact_renames(options); 510 511 /* Did we only want exact renames? */ 512 if (minimum_score == MAX_SCORE) 513 goto cleanup; 514 515 /* 516 * Calculate how many renames are left (but all the source 517 * files still remain as options for rename/copies!) 518 */ 519 num_create = (rename_dst_nr - rename_count); 520 521 /* All done? */ 522 if (!num_create) 523 goto cleanup; 524 525 switch (too_many_rename_candidates(num_create, options)) { 526 case 1: 527 goto cleanup; 528 case 2: 529 options->degraded_cc_to_c = 1; 530 skip_unmodified = 1; 531 break; 532 default: 533 break; 534 } 535 536 if (options->show_rename_progress) { 537 progress = start_progress_delay( 538 _("Performing inexact rename detection"), 539 rename_dst_nr * rename_src_nr, 50, 1); 540 } 541 542 mx = xcalloc(st_mult(num_create, NUM_CANDIDATE_PER_DST), sizeof(*mx)); 543 for (dst_cnt = i = 0; i < rename_dst_nr; i++) { 544 struct diff_filespec *two = rename_dst[i].two; 545 struct diff_score *m; 546 547 if (rename_dst[i].pair) 548 continue; /* dealt with exact match already. */ 549 550 m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST]; 551 for (j = 0; j < NUM_CANDIDATE_PER_DST; j++) 552 m[j].dst = -1; 553 554 for (j = 0; j < rename_src_nr; j++) { 555 struct diff_filespec *one = rename_src[j].p->one; 556 struct diff_score this_src; 557 558 if (skip_unmodified && 559 diff_unmodified_pair(rename_src[j].p)) 560 continue; 561 562 this_src.score = estimate_similarity(one, two, 563 minimum_score); 564 this_src.name_score = basename_same(one, two); 565 this_src.dst = i; 566 this_src.src = j; 567 record_if_better(m, &this_src); 568 /* 569 * Once we run estimate_similarity, 570 * We do not need the text anymore. 571 */ 572 diff_free_filespec_blob(one); 573 diff_free_filespec_blob(two); 574 } 575 dst_cnt++; 576 display_progress(progress, (i+1)*rename_src_nr); 577 } 578 stop_progress(&progress); 579 580 /* cost matrix sorted by most to least similar pair */ 581 qsort(mx, dst_cnt * NUM_CANDIDATE_PER_DST, sizeof(*mx), score_compare); 582 583 rename_count += find_renames(mx, dst_cnt, minimum_score, 0); 584 if (detect_rename == DIFF_DETECT_COPY) 585 rename_count += find_renames(mx, dst_cnt, minimum_score, 1); 586 free(mx); 587 588 cleanup: 589 /* At this point, we have found some renames and copies and they 590 * are recorded in rename_dst. The original list is still in *q. 591 */ 592 DIFF_QUEUE_CLEAR(&outq); 593 for (i = 0; i < q->nr; i++) { 594 struct diff_filepair *p = q->queue[i]; 595 struct diff_filepair *pair_to_free = NULL; 596 597 if (DIFF_PAIR_UNMERGED(p)) { 598 diff_q(&outq, p); 599 } 600 else if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) { 601 /* 602 * Creation 603 * 604 * We would output this create record if it has 605 * not been turned into a rename/copy already. 606 */ 607 struct diff_rename_dst *dst = locate_rename_dst(p->two); 608 if (dst && dst->pair) { 609 diff_q(&outq, dst->pair); 610 pair_to_free = p; 611 } 612 else 613 /* no matching rename/copy source, so 614 * record this as a creation. 615 */ 616 diff_q(&outq, p); 617 } 618 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) { 619 /* 620 * Deletion 621 * 622 * We would output this delete record if: 623 * 624 * (1) this is a broken delete and the counterpart 625 * broken create remains in the output; or 626 * (2) this is not a broken delete, and rename_dst 627 * does not have a rename/copy to move p->one->path 628 * out of existence. 629 * 630 * Otherwise, the counterpart broken create 631 * has been turned into a rename-edit; or 632 * delete did not have a matching create to 633 * begin with. 634 */ 635 if (DIFF_PAIR_BROKEN(p)) { 636 /* broken delete */ 637 struct diff_rename_dst *dst = locate_rename_dst(p->one); 638 if (dst && dst->pair) 639 /* counterpart is now rename/copy */ 640 pair_to_free = p; 641 } 642 else { 643 if (p->one->rename_used) 644 /* this path remains */ 645 pair_to_free = p; 646 } 647 648 if (pair_to_free) 649 ; 650 else 651 diff_q(&outq, p); 652 } 653 else if (!diff_unmodified_pair(p)) 654 /* all the usual ones need to be kept */ 655 diff_q(&outq, p); 656 else 657 /* no need to keep unmodified pairs */ 658 pair_to_free = p; 659 660 if (pair_to_free) 661 diff_free_filepair(pair_to_free); 662 } 663 diff_debug_queue("done copying original", &outq); 664 665 free(q->queue); 666 *q = outq; 667 diff_debug_queue("done collapsing", q); 668 669 for (i = 0; i < rename_dst_nr; i++) 670 free_filespec(rename_dst[i].two); 671 672 free(rename_dst); 673 rename_dst = NULL; 674 rename_dst_nr = rename_dst_alloc = 0; 675 free(rename_src); 676 rename_src = NULL; 677 rename_src_nr = rename_src_alloc = 0; 678 return; 679}