diffcore-rename.c

   1 /*
   2  *
   3  * Copyright (C) 2005 Junio C Hamano
   4  */
   5 #include "cache.h"
   6 #include "diff.h"
   7 #include "diffcore.h"
   8 #include "object-store.h"
   9 #include "hashmap.h"
  10 #include "progress.h"
  11 #include "promisor-remote.h"
  12 #include "strmap.h"
  13
  14 /* Table of rename/copy destinations */
  15
  16 static struct diff_rename_dst {
  17         struct diff_filepair *p;
  18         struct diff_filespec *filespec_to_free;
  19         int is_rename; /* false -> just a create; true -> rename or copy */
  20 } *rename_dst;
  21 static int rename_dst_nr, rename_dst_alloc;
  22 /* Mapping from break source pathname to break destination index */
  23 static struct strintmap *break_idx = NULL;
  24
  25 static struct diff_rename_dst *locate_rename_dst(struct diff_filepair *p)
  26 {
  27         /* Lookup by p->ONE->path */
  28         int idx = break_idx ? strintmap_get(break_idx, p->one->path) : -1;
  29         return (idx == -1) ? NULL : &rename_dst[idx];
  30 }
  31
  32 /*
  33  * Returns 0 on success, -1 if we found a duplicate.
  34  */
  35 static int add_rename_dst(struct diff_filepair *p)
  36 {
  37         ALLOC_GROW(rename_dst, rename_dst_nr + 1, rename_dst_alloc);
  38         rename_dst[rename_dst_nr].p = p;
  39         rename_dst[rename_dst_nr].filespec_to_free = NULL;
  40         rename_dst[rename_dst_nr].is_rename = 0;
  41         rename_dst_nr++;
  42         return 0;
  43 }
  44
  45 /* Table of rename/copy src files */
  46 static struct diff_rename_src {
  47         struct diff_filepair *p;
  48         unsigned short score; /* to remember the break score */
  49 } *rename_src;
  50 static int rename_src_nr, rename_src_alloc;
  51
  52 static void register_rename_src(struct diff_filepair *p)
  53 {
  54         if (p->broken_pair) {
  55                 if (!break_idx) {
  56                         break_idx = xmalloc(sizeof(*break_idx));
  57                         strintmap_init(break_idx, -1);
  58                 }
  59                 strintmap_set(break_idx, p->one->path, rename_dst_nr);
  60         }
  61
  62         ALLOC_GROW(rename_src, rename_src_nr + 1, rename_src_alloc);
  63         rename_src[rename_src_nr].p = p;
  64         rename_src[rename_src_nr].score = p->score;
  65         rename_src_nr++;
  66 }
  67
  68 static int basename_same(struct diff_filespec *src, struct diff_filespec *dst)
  69 {
  70         int src_len = strlen(src->path), dst_len = strlen(dst->path);
  71         while (src_len && dst_len) {
  72                 char c1 = src->path[--src_len];
  73                 char c2 = dst->path[--dst_len];
  74                 if (c1 != c2)
  75                         return 0;
  76                 if (c1 == '/')
  77                         return 1;
  78         }
  79         return (!src_len || src->path[src_len - 1] == '/') &&
  80                 (!dst_len || dst->path[dst_len - 1] == '/');
  81 }
  82
  83 struct diff_score {
  84         int src; /* index in rename_src */
  85         int dst; /* index in rename_dst */
  86         unsigned short score;
  87         short name_score;
  88 };
  89
  90 struct prefetch_options {
  91         struct repository *repo;
  92         int skip_unmodified;
  93 };
  94 static void prefetch(void *prefetch_options)
  95 {
  96         struct prefetch_options *options = prefetch_options;
  97         int i;
  98         struct oid_array to_fetch = OID_ARRAY_INIT;
  99
 100         for (i = 0; i < rename_dst_nr; i++) {
 101                 if (rename_dst[i].p->renamed_pair)
 102                         /*
 103                          * The loop in diffcore_rename() will not need these
 104                          * blobs, so skip prefetching.
 105                          */
 106                         continue; /* already found exact match */
 107                 diff_add_if_missing(options->repo, &to_fetch,
 108                                     rename_dst[i].p->two);
 109         }
 110         for (i = 0; i < rename_src_nr; i++) {
 111                 if (options->skip_unmodified &&
 112                     diff_unmodified_pair(rename_src[i].p))
 113                         /*
 114                          * The loop in diffcore_rename() will not need these
 115                          * blobs, so skip prefetching.
 116                          */
 117                         continue;
 118                 diff_add_if_missing(options->repo, &to_fetch,
 119                                     rename_src[i].p->one);
 120         }
 121         promisor_remote_get_direct(options->repo, to_fetch.oid, to_fetch.nr);
 122         oid_array_clear(&to_fetch);
 123 }
 124
 125 static int estimate_similarity(struct repository *r,
 126                                struct diff_filespec *src,
 127                                struct diff_filespec *dst,
 128                                int minimum_score,
 129                                int skip_unmodified)
 130 {
 131         /* src points at a file that existed in the original tree (or
 132          * optionally a file in the destination tree) and dst points
 133          * at a newly created file.  They may be quite similar, in which
 134          * case we want to say src is renamed to dst or src is copied into
 135          * dst, and then some edit has been applied to dst.
 136          *
 137          * Compare them and return how similar they are, representing
 138          * the score as an integer between 0 and MAX_SCORE.
 139          *
 140          * When there is an exact match, it is considered a better
 141          * match than anything else; the destination does not even
 142          * call into this function in that case.
 143          */
 144         unsigned long max_size, delta_size, base_size, src_copied, literal_added;
 145         int score;
 146         struct diff_populate_filespec_options dpf_options = {
 147                 .check_size_only = 1
 148         };
 149         struct prefetch_options prefetch_options = {r, skip_unmodified};
 150
 151         if (r == the_repository && has_promisor_remote()) {
 152                 dpf_options.missing_object_cb = prefetch;
 153                 dpf_options.missing_object_data = &prefetch_options;
 154         }
 155
 156         /* We deal only with regular files.  Symlink renames are handled
 157          * only when they are exact matches --- in other words, no edits
 158          * after renaming.
 159          */
 160         if (!S_ISREG(src->mode) || !S_ISREG(dst->mode))
 161                 return 0;
 162
 163         /*
 164          * Need to check that source and destination sizes are
 165          * filled in before comparing them.
 166          *
 167          * If we already have "cnt_data" filled in, we know it's
 168          * all good (avoid checking the size for zero, as that
 169          * is a possible size - we really should have a flag to
 170          * say whether the size is valid or not!)
 171          */
 172         if (!src->cnt_data &&
 173             diff_populate_filespec(r, src, &dpf_options))
 174                 return 0;
 175         if (!dst->cnt_data &&
 176             diff_populate_filespec(r, dst, &dpf_options))
 177                 return 0;
 178
 179         max_size = ((src->size > dst->size) ? src->size : dst->size);
 180         base_size = ((src->size < dst->size) ? src->size : dst->size);
 181         delta_size = max_size - base_size;
 182
 183         /* We would not consider edits that change the file size so
 184          * drastically.  delta_size must be smaller than
 185          * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size).
 186          *
 187          * Note that base_size == 0 case is handled here already
 188          * and the final score computation below would not have a
 189          * divide-by-zero issue.
 190          */
 191         if (max_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE)
 192                 return 0;
 193
 194         dpf_options.check_size_only = 0;
 195
 196         if (!src->cnt_data && diff_populate_filespec(r, src, &dpf_options))
 197                 return 0;
 198         if (!dst->cnt_data && diff_populate_filespec(r, dst, &dpf_options))
 199                 return 0;
 200
 201         if (diffcore_count_changes(r, src, dst,
 202                                    &src->cnt_data, &dst->cnt_data,
 203                                    &src_copied, &literal_added))
 204                 return 0;
 205
 206         /* How similar are they?
 207          * what percentage of material in dst are from source?
 208          */
 209         if (!dst->size)
 210                 score = 0; /* should not happen */
 211         else
 212                 score = (int)(src_copied * MAX_SCORE / max_size);
 213         return score;
 214 }
 215
 216 static void record_rename_pair(int dst_index, int src_index, int score)
 217 {
 218         struct diff_filepair *src = rename_src[src_index].p;
 219         struct diff_filepair *dst = rename_dst[dst_index].p;
 220
 221         if (dst->renamed_pair)
 222                 die("internal error: dst already matched.");
 223
 224         src->one->rename_used++;
 225         src->one->count++;
 226
 227         rename_dst[dst_index].filespec_to_free = dst->one;
 228         rename_dst[dst_index].is_rename = 1;
 229
 230         dst->one = src->one;
 231         dst->renamed_pair = 1;
 232         if (!strcmp(dst->one->path, dst->two->path))
 233                 dst->score = rename_src[src_index].score;
 234         else
 235                 dst->score = score;
 236 }
 237
 238 /*
 239  * We sort the rename similarity matrix with the score, in descending
 240  * order (the most similar first).
 241  */
 242 static int score_compare(const void *a_, const void *b_)
 243 {
 244         const struct diff_score *a = a_, *b = b_;
 245
 246         /* sink the unused ones to the bottom */
 247         if (a->dst < 0)
 248                 return (0 <= b->dst);
 249         else if (b->dst < 0)
 250                 return -1;
 251
 252         if (a->score == b->score)
 253                 return b->name_score - a->name_score;
 254
 255         return b->score - a->score;
 256 }
 257
 258 struct file_similarity {
 259         struct hashmap_entry entry;
 260         int index;
 261         struct diff_filespec *filespec;
 262 };
 263
 264 static unsigned int hash_filespec(struct repository *r,
 265                                   struct diff_filespec *filespec)
 266 {
 267         if (!filespec->oid_valid) {
 268                 if (diff_populate_filespec(r, filespec, NULL))
 269                         return 0;
 270                 hash_object_file(r->hash_algo, filespec->data, filespec->size,
 271                                  "blob", &filespec->oid);
 272         }
 273         return oidhash(&filespec->oid);
 274 }
 275
 276 static int find_identical_files(struct hashmap *srcs,
 277                                 int dst_index,
 278                                 struct diff_options *options)
 279 {
 280         int renames = 0;
 281         struct diff_filespec *target = rename_dst[dst_index].p->two;
 282         struct file_similarity *p, *best = NULL;
 283         int i = 100, best_score = -1;
 284         unsigned int hash = hash_filespec(options->repo, target);
 285
 286         /*
 287          * Find the best source match for specified destination.
 288          */
 289         p = hashmap_get_entry_from_hash(srcs, hash, NULL,
 290                                         struct file_similarity, entry);
 291         hashmap_for_each_entry_from(srcs, p, entry) {
 292                 int score;
 293                 struct diff_filespec *source = p->filespec;
 294
 295                 /* False hash collision? */
 296                 if (!oideq(&source->oid, &target->oid))
 297                         continue;
 298                 /* Non-regular files? If so, the modes must match! */
 299                 if (!S_ISREG(source->mode) || !S_ISREG(target->mode)) {
 300                         if (source->mode != target->mode)
 301                                 continue;
 302                 }
 303                 /* Give higher scores to sources that haven't been used already */
 304                 score = !source->rename_used;
 305                 if (source->rename_used && options->detect_rename != DIFF_DETECT_COPY)
 306                         continue;
 307                 score += basename_same(source, target);
 308                 if (score > best_score) {
 309                         best = p;
 310                         best_score = score;
 311                         if (score == 2)
 312                                 break;
 313                 }
 314
 315                 /* Too many identical alternatives? Pick one */
 316                 if (!--i)
 317                         break;
 318         }
 319         if (best) {
 320                 record_rename_pair(dst_index, best->index, MAX_SCORE);
 321                 renames++;
 322         }
 323         return renames;
 324 }
 325
 326 static void insert_file_table(struct repository *r,
 327                               struct hashmap *table, int index,
 328                               struct diff_filespec *filespec)
 329 {
 330         struct file_similarity *entry = xmalloc(sizeof(*entry));
 331
 332         entry->index = index;
 333         entry->filespec = filespec;
 334
 335         hashmap_entry_init(&entry->entry, hash_filespec(r, filespec));
 336         hashmap_add(table, &entry->entry);
 337 }
 338
 339 /*
 340  * Find exact renames first.
 341  *
 342  * The first round matches up the up-to-date entries,
 343  * and then during the second round we try to match
 344  * cache-dirty entries as well.
 345  */
 346 static int find_exact_renames(struct diff_options *options)
 347 {
 348         int i, renames = 0;
 349         struct hashmap file_table;
 350
 351         /* Add all sources to the hash table in reverse order, because
 352          * later on they will be retrieved in LIFO order.
 353          */
 354         hashmap_init(&file_table, NULL, NULL, rename_src_nr);
 355         for (i = rename_src_nr-1; i >= 0; i--)
 356                 insert_file_table(options->repo,
 357                                   &file_table, i,
 358                                   rename_src[i].p->one);
 359
 360         /* Walk the destinations and find best source match */
 361         for (i = 0; i < rename_dst_nr; i++)
 362                 renames += find_identical_files(&file_table, i, options);
 363
 364         /* Free the hash data structure and entries */
 365         hashmap_clear_and_free(&file_table, struct file_similarity, entry);
 366
 367         return renames;
 368 }
 369
 370 static const char *get_basename(const char *filename)
 371 {
 372         /*
 373          * gitbasename() has to worry about special drives, multiple
 374          * directory separator characters, trailing slashes, NULL or
 375          * empty strings, etc.  We only work on filenames as stored in
 376          * git, and thus get to ignore all those complications.
 377          */
 378         const char *base = strrchr(filename, '/');
 379         return base ? base + 1 : filename;
 380 }
 381
 382 static int find_basename_matches(struct diff_options *options,
 383                                  int minimum_score)
 384 {
 385         /*
 386          * When I checked in early 2020, over 76% of file renames in linux
 387          * just moved files to a different directory but kept the same
 388          * basename.  gcc did that with over 64% of renames, gecko did it
 389          * with over 79%, and WebKit did it with over 89%.
 390          *
 391          * Therefore we can bypass the normal exhaustive NxM matrix
 392          * comparison of similarities between all potential rename sources
 393          * and destinations by instead using file basename as a hint (i.e.
 394          * the portion of the filename after the last '/'), checking for
 395          * similarity between files with the same basename, and if we find
 396          * a pair that are sufficiently similar, record the rename pair and
 397          * exclude those two from the NxM matrix.
 398          *
 399          * This *might* cause us to find a less than optimal pairing (if
 400          * there is another file that we are even more similar to but has a
 401          * different basename).  Given the huge performance advantage
 402          * basename matching provides, and given the frequency with which
 403          * people use the same basename in real world projects, that's a
 404          * trade-off we are willing to accept when doing just rename
 405          * detection.
 406          *
 407          * If someone wants copy detection that implies they are willing to
 408          * spend more cycles to find similarities between files, so it may
 409          * be less likely that this heuristic is wanted.  If someone is
 410          * doing break detection, that means they do not want filename
 411          * similarity to imply any form of content similiarity, and thus
 412          * this heuristic would definitely be incompatible.
 413          */
 414
 415         int i, renames = 0;
 416         struct strintmap sources;
 417         struct strintmap dests;
 418         struct hashmap_iter iter;
 419         struct strmap_entry *entry;
 420
 421         /*
 422          * The prefeteching stuff wants to know if it can skip prefetching
 423          * blobs that are unmodified...and will then do a little extra work
 424          * to verify that the oids are indeed different before prefetching.
 425          * Unmodified blobs are only relevant when doing copy detection;
 426          * when limiting to rename detection, diffcore_rename[_extended]()
 427          * will never be called with unmodified source paths fed to us, so
 428          * the extra work necessary to check if rename_src entries are
 429          * unmodified would be a small waste.
 430          */
 431         int skip_unmodified = 0;
 432
 433         /*
 434          * Create maps of basename -> fullname(s) for remaining sources and
 435          * dests.
 436          */
 437         strintmap_init_with_options(&sources, -1, NULL, 0);
 438         strintmap_init_with_options(&dests, -1, NULL, 0);
 439         for (i = 0; i < rename_src_nr; ++i) {
 440                 char *filename = rename_src[i].p->one->path;
 441                 const char *base;
 442
 443                 /* exact renames removed in remove_unneeded_paths_from_src() */
 444                 assert(!rename_src[i].p->one->rename_used);
 445
 446                 /* Record index within rename_src (i) if basename is unique */
 447                 base = get_basename(filename);
 448                 if (strintmap_contains(&sources, base))
 449                         strintmap_set(&sources, base, -1);
 450                 else
 451                         strintmap_set(&sources, base, i);
 452         }
 453         for (i = 0; i < rename_dst_nr; ++i) {
 454                 char *filename = rename_dst[i].p->two->path;
 455                 const char *base;
 456
 457                 if (rename_dst[i].is_rename)
 458                         continue; /* involved in exact match already. */
 459
 460                 /* Record index within rename_dst (i) if basename is unique */
 461                 base = get_basename(filename);
 462                 if (strintmap_contains(&dests, base))
 463                         strintmap_set(&dests, base, -1);
 464                 else
 465                         strintmap_set(&dests, base, i);
 466         }
 467
 468         /* Now look for basename matchups and do similarity estimation */
 469         strintmap_for_each_entry(&sources, &iter, entry) {
 470                 const char *base = entry->key;
 471                 intptr_t src_index = (intptr_t)entry->value;
 472                 intptr_t dst_index;
 473                 if (src_index == -1)
 474                         continue;
 475
 476                 if (0 <= (dst_index = strintmap_get(&dests, base))) {
 477                         struct diff_filespec *one, *two;
 478                         int score;
 479
 480                         /* Estimate the similarity */
 481                         one = rename_src[src_index].p->one;
 482                         two = rename_dst[dst_index].p->two;
 483                         score = estimate_similarity(options->repo, one, two,
 484                                                     minimum_score, skip_unmodified);
 485
 486                         /* If sufficiently similar, record as rename pair */
 487                         if (score < minimum_score)
 488                                 continue;
 489                         record_rename_pair(dst_index, src_index, score);
 490                         renames++;
 491
 492                         /*
 493                          * Found a rename so don't need text anymore; if we
 494                          * didn't find a rename, the filespec_blob would get
 495                          * re-used when doing the matrix of comparisons.
 496                          */
 497                         diff_free_filespec_blob(one);
 498                         diff_free_filespec_blob(two);
 499                 }
 500         }
 501
 502         strintmap_clear(&sources);
 503         strintmap_clear(&dests);
 504
 505         return renames;
 506 }
 507
 508 #define NUM_CANDIDATE_PER_DST 4
 509 static void record_if_better(struct diff_score m[], struct diff_score *o)
 510 {
 511         int i, worst;
 512
 513         /* find the worst one */
 514         worst = 0;
 515         for (i = 1; i < NUM_CANDIDATE_PER_DST; i++)
 516                 if (score_compare(&m[i], &m[worst]) > 0)
 517                         worst = i;
 518
 519         /* is it better than the worst one? */
 520         if (score_compare(&m[worst], o) > 0)
 521                 m[worst] = *o;
 522 }
 523
 524 /*
 525  * Returns:
 526  * 0 if we are under the limit;
 527  * 1 if we need to disable inexact rename detection;
 528  * 2 if we would be under the limit if we were given -C instead of -C -C.
 529  */
 530 static int too_many_rename_candidates(int num_destinations, int num_sources,
 531                                       struct diff_options *options)
 532 {
 533         int rename_limit = options->rename_limit;
 534         int i, limited_sources;
 535
 536         options->needed_rename_limit = 0;
 537
 538         /*
 539          * This basically does a test for the rename matrix not
 540          * growing larger than a "rename_limit" square matrix, ie:
 541          *
 542          *    num_destinations * num_sources > rename_limit * rename_limit
 543          *
 544          * We use st_mult() to check overflow conditions; in the
 545          * exceptional circumstance that size_t isn't large enough to hold
 546          * the multiplication, the system won't be able to allocate enough
 547          * memory for the matrix anyway.
 548          */
 549         if (rename_limit <= 0)
 550                 rename_limit = 32767;
 551         if (st_mult(num_destinations, num_sources)
 552             <= st_mult(rename_limit, rename_limit))
 553                 return 0;
 554
 555         options->needed_rename_limit =
 556                 num_sources > num_destinations ? num_sources : num_destinations;
 557
 558         /* Are we running under -C -C? */
 559         if (!options->flags.find_copies_harder)
 560                 return 1;
 561
 562         /* Would we bust the limit if we were running under -C? */
 563         for (limited_sources = i = 0; i < num_sources; i++) {
 564                 if (diff_unmodified_pair(rename_src[i].p))
 565                         continue;
 566                 limited_sources++;
 567         }
 568         if (st_mult(num_destinations, limited_sources)
 569             <= st_mult(rename_limit, rename_limit))
 570                 return 2;
 571         return 1;
 572 }
 573
 574 static int find_renames(struct diff_score *mx, int dst_cnt, int minimum_score, int copies)
 575 {
 576         int count = 0, i;
 577
 578         for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) {
 579                 struct diff_rename_dst *dst;
 580
 581                 if ((mx[i].dst < 0) ||
 582                     (mx[i].score < minimum_score))
 583                         break; /* there is no more usable pair. */
 584                 dst = &rename_dst[mx[i].dst];
 585                 if (dst->is_rename)
 586                         continue; /* already done, either exact or fuzzy. */
 587                 if (!copies && rename_src[mx[i].src].p->one->rename_used)
 588                         continue;
 589                 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
 590                 count++;
 591         }
 592         return count;
 593 }
 594
 595 static void remove_unneeded_paths_from_src(int detecting_copies)
 596 {
 597         int i, new_num_src;
 598
 599         if (detecting_copies)
 600                 return; /* nothing to remove */
 601         if (break_idx)
 602                 return; /* culling incompatible with break detection */
 603
 604         /*
 605          * Note on reasons why we cull unneeded sources but not destinations:
 606          *   1) Pairings are stored in rename_dst (not rename_src), which we
 607          *      need to keep around.  So, we just can't cull rename_dst even
 608          *      if we wanted to.  But doing so wouldn't help because...
 609          *
 610          *   2) There is a matrix pairwise comparison that follows the
 611          *      "Performing inexact rename detection" progress message.
 612          *      Iterating over the destinations is done in the outer loop,
 613          *      hence we only iterate over each of those once and we can
 614          *      easily skip the outer loop early if the destination isn't
 615          *      relevant.  That's only one check per destination path to
 616          *      skip.
 617          *
 618          *      By contrast, the sources are iterated in the inner loop; if
 619          *      we check whether a source can be skipped, then we'll be
 620          *      checking it N separate times, once for each destination.
 621          *      We don't want to have to iterate over known-not-needed
 622          *      sources N times each, so avoid that by removing the sources
 623          *      from rename_src here.
 624          */
 625         for (i = 0, new_num_src = 0; i < rename_src_nr; i++) {
 626                 /*
 627                  * renames are stored in rename_dst, so if a rename has
 628                  * already been detected using this source, we can just
 629                  * remove the source knowing rename_dst has its info.
 630                  */
 631                 if (rename_src[i].p->one->rename_used)
 632                         continue;
 633
 634                 if (new_num_src < i)
 635                         memcpy(&rename_src[new_num_src], &rename_src[i],
 636                                sizeof(struct diff_rename_src));
 637                 new_num_src++;
 638         }
 639
 640         rename_src_nr = new_num_src;
 641 }
 642
 643 void diffcore_rename(struct diff_options *options)
 644 {
 645         int detect_rename = options->detect_rename;
 646         int minimum_score = options->rename_score;
 647         struct diff_queue_struct *q = &diff_queued_diff;
 648         struct diff_queue_struct outq;
 649         struct diff_score *mx;
 650         int i, j, rename_count, skip_unmodified = 0;
 651         int num_destinations, dst_cnt;
 652         int num_sources, want_copies;
 653         struct progress *progress = NULL;
 654
 655         trace2_region_enter("diff", "setup", options->repo);
 656         want_copies = (detect_rename == DIFF_DETECT_COPY);
 657         if (!minimum_score)
 658                 minimum_score = DEFAULT_RENAME_SCORE;
 659
 660         for (i = 0; i < q->nr; i++) {
 661                 struct diff_filepair *p = q->queue[i];
 662                 if (!DIFF_FILE_VALID(p->one)) {
 663                         if (!DIFF_FILE_VALID(p->two))
 664                                 continue; /* unmerged */
 665                         else if (options->single_follow &&
 666                                  strcmp(options->single_follow, p->two->path))
 667                                 continue; /* not interested */
 668                         else if (!options->flags.rename_empty &&
 669                                  is_empty_blob_oid(&p->two->oid))
 670                                 continue;
 671                         else if (add_rename_dst(p) < 0) {
 672                                 warning("skipping rename detection, detected"
 673                                         " duplicate destination '%s'",
 674                                         p->two->path);
 675                                 goto cleanup;
 676                         }
 677                 }
 678                 else if (!options->flags.rename_empty &&
 679                          is_empty_blob_oid(&p->one->oid))
 680                         continue;
 681                 else if (!DIFF_PAIR_UNMERGED(p) && !DIFF_FILE_VALID(p->two)) {
 682                         /*
 683                          * If the source is a broken "delete", and
 684                          * they did not really want to get broken,
 685                          * that means the source actually stays.
 686                          * So we increment the "rename_used" score
 687                          * by one, to indicate ourselves as a user
 688                          */
 689                         if (p->broken_pair && !p->score)
 690                                 p->one->rename_used++;
 691                         register_rename_src(p);
 692                 }
 693                 else if (want_copies) {
 694                         /*
 695                          * Increment the "rename_used" score by
 696                          * one, to indicate ourselves as a user.
 697                          */
 698                         p->one->rename_used++;
 699                         register_rename_src(p);
 700                 }
 701         }
 702         trace2_region_leave("diff", "setup", options->repo);
 703         if (rename_dst_nr == 0 || rename_src_nr == 0)
 704                 goto cleanup; /* nothing to do */
 705
 706         trace2_region_enter("diff", "exact renames", options->repo);
 707         /*
 708          * We really want to cull the candidates list early
 709          * with cheap tests in order to avoid doing deltas.
 710          */
 711         rename_count = find_exact_renames(options);
 712         trace2_region_leave("diff", "exact renames", options->repo);
 713
 714         /* Did we only want exact renames? */
 715         if (minimum_score == MAX_SCORE)
 716                 goto cleanup;
 717
 718         num_sources = rename_src_nr;
 719
 720         if (want_copies || break_idx) {
 721                 /*
 722                  * Cull sources:
 723                  *   - remove ones corresponding to exact renames
 724                  */
 725                 trace2_region_enter("diff", "cull after exact", options->repo);
 726                 remove_unneeded_paths_from_src(want_copies);
 727                 trace2_region_leave("diff", "cull after exact", options->repo);
 728         } else {
 729                 /* Determine minimum score to match basenames */
 730                 double factor = 0.5;
 731                 char *basename_factor = getenv("GIT_BASENAME_FACTOR");
 732                 int min_basename_score;
 733
 734                 if (basename_factor)
 735                         factor = strtol(basename_factor, NULL, 10)/100.0;
 736                 assert(factor >= 0.0 && factor <= 1.0);
 737                 min_basename_score = minimum_score +
 738                         (int)(factor * (MAX_SCORE - minimum_score));
 739
 740                 /*
 741                  * Cull sources:
 742                  *   - remove ones involved in renames (found via exact match)
 743                  */
 744                 trace2_region_enter("diff", "cull after exact", options->repo);
 745                 remove_unneeded_paths_from_src(want_copies);
 746                 trace2_region_leave("diff", "cull after exact", options->repo);
 747
 748                 /* Utilize file basenames to quickly find renames. */
 749                 trace2_region_enter("diff", "basename matches", options->repo);
 750                 rename_count += find_basename_matches(options,
 751                                                       min_basename_score);
 752                 trace2_region_leave("diff", "basename matches", options->repo);
 753
 754                 /*
 755                  * Cull sources, again:
 756                  *   - remove ones involved in renames (found via basenames)
 757                  */
 758                 trace2_region_enter("diff", "cull basename", options->repo);
 759                 remove_unneeded_paths_from_src(want_copies);
 760                 trace2_region_leave("diff", "cull basename", options->repo);
 761         }
 762
 763         /* Calculate how many rename destinations are left */
 764         num_destinations = (rename_dst_nr - rename_count);
 765         num_sources = rename_src_nr; /* rename_src_nr reflects lower number */
 766
 767         /* All done? */
 768         if (!num_destinations || !num_sources)
 769                 goto cleanup;
 770
 771         switch (too_many_rename_candidates(num_destinations, num_sources,
 772                                            options)) {
 773         case 1:
 774                 goto cleanup;
 775         case 2:
 776                 options->degraded_cc_to_c = 1;
 777                 skip_unmodified = 1;
 778                 break;
 779         default:
 780                 break;
 781         }
 782
 783         trace2_region_enter("diff", "inexact renames", options->repo);
 784         if (options->show_rename_progress) {
 785                 progress = start_delayed_progress(
 786                                 _("Performing inexact rename detection"),
 787                                 (uint64_t)num_destinations * (uint64_t)num_sources);
 788         }
 789
 790         CALLOC_ARRAY(mx, st_mult(NUM_CANDIDATE_PER_DST, num_destinations));
 791         for (dst_cnt = i = 0; i < rename_dst_nr; i++) {
 792                 struct diff_filespec *two = rename_dst[i].p->two;
 793                 struct diff_score *m;
 794
 795                 if (rename_dst[i].is_rename)
 796                         continue; /* exact or basename match already handled */
 797
 798                 m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST];
 799                 for (j = 0; j < NUM_CANDIDATE_PER_DST; j++)
 800                         m[j].dst = -1;
 801
 802                 for (j = 0; j < rename_src_nr; j++) {
 803                         struct diff_filespec *one = rename_src[j].p->one;
 804                         struct diff_score this_src;
 805
 806                         assert(!one->rename_used || want_copies || break_idx);
 807
 808                         if (skip_unmodified &&
 809                             diff_unmodified_pair(rename_src[j].p))
 810                                 continue;
 811
 812                         this_src.score = estimate_similarity(options->repo,
 813                                                              one, two,
 814                                                              minimum_score,
 815                                                              skip_unmodified);
 816                         this_src.name_score = basename_same(one, two);
 817                         this_src.dst = i;
 818                         this_src.src = j;
 819                         record_if_better(m, &this_src);
 820                         /*
 821                          * Once we run estimate_similarity,
 822                          * We do not need the text anymore.
 823                          */
 824                         diff_free_filespec_blob(one);
 825                         diff_free_filespec_blob(two);
 826                 }
 827                 dst_cnt++;
 828                 display_progress(progress,
 829                                  (uint64_t)dst_cnt * (uint64_t)num_sources);
 830         }
 831         stop_progress(&progress);
 832
 833         /* cost matrix sorted by most to least similar pair */
 834         STABLE_QSORT(mx, dst_cnt * NUM_CANDIDATE_PER_DST, score_compare);
 835
 836         rename_count += find_renames(mx, dst_cnt, minimum_score, 0);
 837         if (want_copies)
 838                 rename_count += find_renames(mx, dst_cnt, minimum_score, 1);
 839         free(mx);
 840         trace2_region_leave("diff", "inexact renames", options->repo);
 841
 842  cleanup:
 843         /* At this point, we have found some renames and copies and they
 844          * are recorded in rename_dst.  The original list is still in *q.
 845          */
 846         trace2_region_enter("diff", "write back to queue", options->repo);
 847         DIFF_QUEUE_CLEAR(&outq);
 848         for (i = 0; i < q->nr; i++) {
 849                 struct diff_filepair *p = q->queue[i];
 850                 struct diff_filepair *pair_to_free = NULL;
 851
 852                 if (DIFF_PAIR_UNMERGED(p)) {
 853                         diff_q(&outq, p);
 854                 }
 855                 else if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
 856                         /* Creation */
 857                         diff_q(&outq, p);
 858                 }
 859                 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) {
 860                         /*
 861                          * Deletion
 862                          *
 863                          * We would output this delete record if:
 864                          *
 865                          * (1) this is a broken delete and the counterpart
 866                          *     broken create remains in the output; or
 867                          * (2) this is not a broken delete, and rename_dst
 868                          *     does not have a rename/copy to move p->one->path
 869                          *     out of existence.
 870                          *
 871                          * Otherwise, the counterpart broken create
 872                          * has been turned into a rename-edit; or
 873                          * delete did not have a matching create to
 874                          * begin with.
 875                          */
 876                         if (DIFF_PAIR_BROKEN(p)) {
 877                                 /* broken delete */
 878                                 struct diff_rename_dst *dst = locate_rename_dst(p);
 879                                 if (!dst)
 880                                         BUG("tracking failed somehow; failed to find associated dst for broken pair");
 881                                 if (dst->is_rename)
 882                                         /* counterpart is now rename/copy */
 883                                         pair_to_free = p;
 884                         }
 885                         else {
 886                                 if (p->one->rename_used)
 887                                         /* this path remains */
 888                                         pair_to_free = p;
 889                         }
 890
 891                         if (!pair_to_free)
 892                                 diff_q(&outq, p);
 893                 }
 894                 else if (!diff_unmodified_pair(p))
 895                         /* all the usual ones need to be kept */
 896                         diff_q(&outq, p);
 897                 else
 898                         /* no need to keep unmodified pairs; FIXME: remove earlier? */
 899                         pair_to_free = p;
 900
 901                 if (pair_to_free)
 902                         diff_free_filepair(pair_to_free);
 903         }
 904         diff_debug_queue("done copying original", &outq);
 905
 906         free(q->queue);
 907         *q = outq;
 908         diff_debug_queue("done collapsing", q);
 909
 910         for (i = 0; i < rename_dst_nr; i++)
 911                 if (rename_dst[i].filespec_to_free)
 912                         free_filespec(rename_dst[i].filespec_to_free);
 913
 914         FREE_AND_NULL(rename_dst);
 915         rename_dst_nr = rename_dst_alloc = 0;
 916         FREE_AND_NULL(rename_src);
 917         rename_src_nr = rename_src_alloc = 0;
 918         if (break_idx) {
 919                 strintmap_clear(break_idx);
 920                 FREE_AND_NULL(break_idx);
 921         }
 922         trace2_region_leave("diff", "write back to queue", options->repo);
 923         return;
 924 }