name-hash.c

   1 /*
   2  * name-hash.c
   3  *
   4  * Hashing names in the index state
   5  *
   6  * Copyright (C) 2008 Linus Torvalds
   7  */
   8
   9 #define USE_THE_REPOSITORY_VARIABLE
  10
  11 #include "git-compat-util.h"
  12 #include "environment.h"
  13 #include "gettext.h"
  14 #include "name-hash.h"
  15 #include "object.h"
  16 #include "read-cache-ll.h"
  17 #include "thread-utils.h"
  18 #include "trace.h"
  19 #include "trace2.h"
  20 #include "sparse-index.h"
  21
  22 struct dir_entry {
  23         struct hashmap_entry ent;
  24         struct dir_entry *parent;
  25         int nr;
  26         unsigned int namelen;
  27         char name[FLEX_ARRAY];
  28 };
  29
  30 static int dir_entry_cmp(const void *cmp_data UNUSED,
  31                          const struct hashmap_entry *eptr,
  32                          const struct hashmap_entry *entry_or_key,
  33                          const void *keydata)
  34 {
  35         const struct dir_entry *e1, *e2;
  36         const char *name = keydata;
  37
  38         e1 = container_of(eptr, const struct dir_entry, ent);
  39         e2 = container_of(entry_or_key, const struct dir_entry, ent);
  40
  41         return e1->namelen != e2->namelen || strncasecmp(e1->name,
  42                         name ? name : e2->name, e1->namelen);
  43 }
  44
  45 static struct dir_entry *find_dir_entry__hash(struct index_state *istate,
  46                 const char *name, unsigned int namelen, unsigned int hash)
  47 {
  48         struct dir_entry key;
  49         hashmap_entry_init(&key.ent, hash);
  50         key.namelen = namelen;
  51         return hashmap_get_entry(&istate->dir_hash, &key, ent, name);
  52 }
  53
  54 static struct dir_entry *find_dir_entry(struct index_state *istate,
  55                 const char *name, unsigned int namelen)
  56 {
  57         return find_dir_entry__hash(istate, name, namelen, memihash(name, namelen));
  58 }
  59
  60 static struct dir_entry *hash_dir_entry(struct index_state *istate,
  61                 struct cache_entry *ce, int namelen)
  62 {
  63         /*
  64          * Throw each directory component in the hash for quick lookup
  65          * during a git status. Directory components are stored without their
  66          * closing slash.  Despite submodules being a directory, they never
  67          * reach this point, because they are stored
  68          * in index_state.name_hash (as ordinary cache_entries).
  69          */
  70         struct dir_entry *dir;
  71
  72         /* get length of parent directory */
  73         while (namelen > 0 && !is_dir_sep(ce->name[namelen - 1]))
  74                 namelen--;
  75         if (namelen <= 0)
  76                 return NULL;
  77         namelen--;
  78
  79         /* lookup existing entry for that directory */
  80         dir = find_dir_entry(istate, ce->name, namelen);
  81         if (!dir) {
  82                 /* not found, create it and add to hash table */
  83                 FLEX_ALLOC_MEM(dir, name, ce->name, namelen);
  84                 hashmap_entry_init(&dir->ent, memihash(ce->name, namelen));
  85                 dir->namelen = namelen;
  86                 hashmap_add(&istate->dir_hash, &dir->ent);
  87
  88                 /* recursively add missing parent directories */
  89                 dir->parent = hash_dir_entry(istate, ce, namelen);
  90         }
  91         return dir;
  92 }
  93
  94 static void add_dir_entry(struct index_state *istate, struct cache_entry *ce)
  95 {
  96         /* Add reference to the directory entry (and parents if 0). */
  97         struct dir_entry *dir = hash_dir_entry(istate, ce, ce_namelen(ce));
  98         while (dir && !(dir->nr++))
  99                 dir = dir->parent;
 100 }
 101
 102 static void remove_dir_entry(struct index_state *istate, struct cache_entry *ce)
 103 {
 104         /*
 105          * Release reference to the directory entry. If 0, remove and continue
 106          * with parent directory.
 107          */
 108         struct dir_entry *dir = hash_dir_entry(istate, ce, ce_namelen(ce));
 109         while (dir && !(--dir->nr)) {
 110                 struct dir_entry *parent = dir->parent;
 111                 hashmap_remove(&istate->dir_hash, &dir->ent, NULL);
 112                 free(dir);
 113                 dir = parent;
 114         }
 115 }
 116
 117 static void hash_index_entry(struct index_state *istate, struct cache_entry *ce)
 118 {
 119         if (ce->ce_flags & CE_HASHED)
 120                 return;
 121         ce->ce_flags |= CE_HASHED;
 122
 123         if (!S_ISSPARSEDIR(ce->ce_mode)) {
 124                 hashmap_entry_init(&ce->ent, memihash(ce->name, ce_namelen(ce)));
 125                 hashmap_add(&istate->name_hash, &ce->ent);
 126         }
 127
 128         if (ignore_case)
 129                 add_dir_entry(istate, ce);
 130 }
 131
 132 static int cache_entry_cmp(const void *cmp_data UNUSED,
 133                            const struct hashmap_entry *eptr,
 134                            const struct hashmap_entry *entry_or_key,
 135                            const void *remove)
 136 {
 137         const struct cache_entry *ce1, *ce2;
 138
 139         ce1 = container_of(eptr, const struct cache_entry, ent);
 140         ce2 = container_of(entry_or_key, const struct cache_entry, ent);
 141
 142         /*
 143          * For remove_name_hash, find the exact entry (pointer equality); for
 144          * index_file_exists, find all entries with matching hash code and
 145          * decide whether the entry matches in same_name.
 146          */
 147         return remove ? !(ce1 == ce2) : 0;
 148 }
 149
 150 static int lazy_try_threaded = 1;
 151 static int lazy_nr_dir_threads;
 152
 153 /*
 154  * Set a minimum number of cache_entries that we will handle per
 155  * thread and use that to decide how many threads to run (up to
 156  * the number on the system).
 157  *
 158  * For guidance setting the lower per-thread bound, see:
 159  *     t/helper/test-lazy-init-name-hash --analyze
 160  */
 161 #define LAZY_THREAD_COST (2000)
 162
 163 /*
 164  * We use n mutexes to guard n partitions of the "istate->dir_hash"
 165  * hashtable.  Since "find" and "insert" operations will hash to a
 166  * particular bucket and modify/search a single chain, we can say
 167  * that "all chains mod n" are guarded by the same mutex -- rather
 168  * than having a single mutex to guard the entire table.  (This does
 169  * require that we disable "rehashing" on the hashtable.)
 170  *
 171  * So, a larger value here decreases the probability of a collision
 172  * and the time that each thread must wait for the mutex.
 173  */
 174 #define LAZY_MAX_MUTEX   (32)
 175
 176 static pthread_mutex_t *lazy_dir_mutex_array;
 177
 178 /*
 179  * An array of lazy_entry items is used by the n threads in
 180  * the directory parse (first) phase to (lock-free) store the
 181  * intermediate results.  These values are then referenced by
 182  * the 2 threads in the second phase.
 183  */
 184 struct lazy_entry {
 185         struct dir_entry *dir;
 186         unsigned int hash_dir;
 187         unsigned int hash_name;
 188 };
 189
 190 /*
 191  * Decide if we want to use threads (if available) to load
 192  * the hash tables.  We set "lazy_nr_dir_threads" to zero when
 193  * it is not worth it.
 194  */
 195 static int lookup_lazy_params(struct index_state *istate)
 196 {
 197         int nr_cpus;
 198
 199         lazy_nr_dir_threads = 0;
 200
 201         if (!lazy_try_threaded)
 202                 return 0;
 203
 204         /*
 205          * If we are respecting case, just use the original
 206          * code to build the "istate->name_hash".  We don't
 207          * need the complexity here.
 208          */
 209         if (!ignore_case)
 210                 return 0;
 211
 212         nr_cpus = online_cpus();
 213         if (nr_cpus < 2)
 214                 return 0;
 215
 216         if (istate->cache_nr < 2 * LAZY_THREAD_COST)
 217                 return 0;
 218
 219         if (istate->cache_nr < nr_cpus * LAZY_THREAD_COST)
 220                 nr_cpus = istate->cache_nr / LAZY_THREAD_COST;
 221         lazy_nr_dir_threads = nr_cpus;
 222         return lazy_nr_dir_threads;
 223 }
 224
 225 /*
 226  * Initialize n mutexes for use when searching and inserting
 227  * into "istate->dir_hash".  All "dir" threads are trying
 228  * to insert partial pathnames into the hash as they iterate
 229  * over their portions of the index, so lock contention is
 230  * high.
 231  *
 232  * However, the hashmap is going to put items into bucket
 233  * chains based on their hash values.  Use that to create n
 234  * mutexes and lock on mutex[bucket(hash) % n].  This will
 235  * decrease the collision rate by (hopefully) a factor of n.
 236  */
 237 static void init_dir_mutex(void)
 238 {
 239         int j;
 240
 241         CALLOC_ARRAY(lazy_dir_mutex_array, LAZY_MAX_MUTEX);
 242
 243         for (j = 0; j < LAZY_MAX_MUTEX; j++)
 244                 init_recursive_mutex(&lazy_dir_mutex_array[j]);
 245 }
 246
 247 static void cleanup_dir_mutex(void)
 248 {
 249         int j;
 250
 251         for (j = 0; j < LAZY_MAX_MUTEX; j++)
 252                 pthread_mutex_destroy(&lazy_dir_mutex_array[j]);
 253
 254         free(lazy_dir_mutex_array);
 255 }
 256
 257 static void lock_dir_mutex(int j)
 258 {
 259         pthread_mutex_lock(&lazy_dir_mutex_array[j]);
 260 }
 261
 262 static void unlock_dir_mutex(int j)
 263 {
 264         pthread_mutex_unlock(&lazy_dir_mutex_array[j]);
 265 }
 266
 267 static inline int compute_dir_lock_nr(
 268         const struct hashmap *map,
 269         unsigned int hash)
 270 {
 271         return hashmap_bucket(map, hash) % LAZY_MAX_MUTEX;
 272 }
 273
 274 static struct dir_entry *hash_dir_entry_with_parent_and_prefix(
 275         struct index_state *istate,
 276         struct dir_entry *parent,
 277         struct strbuf *prefix)
 278 {
 279         struct dir_entry *dir;
 280         unsigned int hash;
 281         int lock_nr;
 282
 283         /*
 284          * Either we have a parent directory and path with slash(es)
 285          * or the directory is an immediate child of the root directory.
 286          */
 287         assert((parent != NULL) ^ (strchr(prefix->buf, '/') == NULL));
 288
 289         if (parent)
 290                 hash = memihash_cont(parent->ent.hash,
 291                         prefix->buf + parent->namelen,
 292                         prefix->len - parent->namelen);
 293         else
 294                 hash = memihash(prefix->buf, prefix->len);
 295
 296         lock_nr = compute_dir_lock_nr(&istate->dir_hash, hash);
 297         lock_dir_mutex(lock_nr);
 298
 299         dir = find_dir_entry__hash(istate, prefix->buf, prefix->len, hash);
 300         if (!dir) {
 301                 FLEX_ALLOC_MEM(dir, name, prefix->buf, prefix->len);
 302                 hashmap_entry_init(&dir->ent, hash);
 303                 dir->namelen = prefix->len;
 304                 dir->parent = parent;
 305                 hashmap_add(&istate->dir_hash, &dir->ent);
 306
 307                 if (parent) {
 308                         unlock_dir_mutex(lock_nr);
 309
 310                         /* All I really need here is an InterlockedIncrement(&(parent->nr)) */
 311                         lock_nr = compute_dir_lock_nr(&istate->dir_hash, parent->ent.hash);
 312                         lock_dir_mutex(lock_nr);
 313                         parent->nr++;
 314                 }
 315         }
 316
 317         unlock_dir_mutex(lock_nr);
 318
 319         return dir;
 320 }
 321
 322 /*
 323  * handle_range_1() and handle_range_dir() are derived from
 324  * clear_ce_flags_1() and clear_ce_flags_dir() in unpack-trees.c
 325  * and handle the iteration over the entire array of index entries.
 326  * They use recursion for adjacent entries in the same parent
 327  * directory.
 328  */
 329 static int handle_range_1(
 330         struct index_state *istate,
 331         int k_start,
 332         int k_end,
 333         struct dir_entry *parent,
 334         struct strbuf *prefix,
 335         struct lazy_entry *lazy_entries);
 336
 337 static int handle_range_dir(
 338         struct index_state *istate,
 339         int k_start,
 340         int k_end,
 341         struct dir_entry *parent,
 342         struct strbuf *prefix,
 343         struct lazy_entry *lazy_entries,
 344         struct dir_entry **dir_new_out)
 345 {
 346         int rc, k;
 347         int input_prefix_len = prefix->len;
 348         struct dir_entry *dir_new;
 349
 350         dir_new = hash_dir_entry_with_parent_and_prefix(istate, parent, prefix);
 351
 352         strbuf_addch(prefix, '/');
 353
 354         /*
 355          * Scan forward in the index array for index entries having the same
 356          * path prefix (that are also in this directory).
 357          */
 358         if (k_start + 1 >= k_end)
 359                 k = k_end;
 360         else if (strncmp(istate->cache[k_start + 1]->name, prefix->buf, prefix->len) > 0)
 361                 k = k_start + 1;
 362         else if (strncmp(istate->cache[k_end - 1]->name, prefix->buf, prefix->len) == 0)
 363                 k = k_end;
 364         else {
 365                 int begin = k_start;
 366                 int end = k_end;
 367                 assert(begin >= 0);
 368                 while (begin < end) {
 369                         int mid = begin + ((end - begin) >> 1);
 370                         int cmp = strncmp(istate->cache[mid]->name, prefix->buf, prefix->len);
 371                         if (cmp == 0) /* mid has same prefix; look in second part */
 372                                 begin = mid + 1;
 373                         else if (cmp > 0) /* mid is past group; look in first part */
 374                                 end = mid;
 375                         else
 376                                 die("cache entry out of order");
 377                 }
 378                 k = begin;
 379         }
 380
 381         /*
 382          * Recurse and process what we can of this subset [k_start, k).
 383          */
 384         rc = handle_range_1(istate, k_start, k, dir_new, prefix, lazy_entries);
 385
 386         strbuf_setlen(prefix, input_prefix_len);
 387
 388         *dir_new_out = dir_new;
 389         return rc;
 390 }
 391
 392 static int handle_range_1(
 393         struct index_state *istate,
 394         int k_start,
 395         int k_end,
 396         struct dir_entry *parent,
 397         struct strbuf *prefix,
 398         struct lazy_entry *lazy_entries)
 399 {
 400         int input_prefix_len = prefix->len;
 401         int k = k_start;
 402
 403         while (k < k_end) {
 404                 struct cache_entry *ce_k = istate->cache[k];
 405                 const char *name, *slash;
 406
 407                 if (prefix->len && strncmp(ce_k->name, prefix->buf, prefix->len))
 408                         break;
 409
 410                 name = ce_k->name + prefix->len;
 411                 slash = strchr(name, '/');
 412
 413                 if (slash) {
 414                         int len = slash - name;
 415                         int processed;
 416                         struct dir_entry *dir_new;
 417
 418                         strbuf_add(prefix, name, len);
 419                         processed = handle_range_dir(istate, k, k_end, parent, prefix, lazy_entries, &dir_new);
 420                         if (processed) {
 421                                 k += processed;
 422                                 strbuf_setlen(prefix, input_prefix_len);
 423                                 continue;
 424                         }
 425
 426                         strbuf_addch(prefix, '/');
 427                         processed = handle_range_1(istate, k, k_end, dir_new, prefix, lazy_entries);
 428                         k += processed;
 429                         strbuf_setlen(prefix, input_prefix_len);
 430                         continue;
 431                 }
 432
 433                 /*
 434                  * It is too expensive to take a lock to insert "ce_k"
 435                  * into "istate->name_hash" and increment the ref-count
 436                  * on the "parent" dir.  So we defer actually updating
 437                  * permanent data structures until phase 2 (where we
 438                  * can change the locking requirements) and simply
 439                  * accumulate our current results into the lazy_entries
 440                  * data array).
 441                  *
 442                  * We do not need to lock the lazy_entries array because
 443                  * we have exclusive access to the cells in the range
 444                  * [k_start,k_end) that this thread was given.
 445                  */
 446                 lazy_entries[k].dir = parent;
 447                 if (parent) {
 448                         lazy_entries[k].hash_name = memihash_cont(
 449                                 parent->ent.hash,
 450                                 ce_k->name + parent->namelen,
 451                                 ce_namelen(ce_k) - parent->namelen);
 452                         lazy_entries[k].hash_dir = parent->ent.hash;
 453                 } else {
 454                         lazy_entries[k].hash_name = memihash(ce_k->name, ce_namelen(ce_k));
 455                 }
 456
 457                 k++;
 458         }
 459
 460         return k - k_start;
 461 }
 462
 463 struct lazy_dir_thread_data {
 464         pthread_t pthread;
 465         struct index_state *istate;
 466         struct lazy_entry *lazy_entries;
 467         int k_start;
 468         int k_end;
 469 };
 470
 471 static void *lazy_dir_thread_proc(void *_data)
 472 {
 473         struct lazy_dir_thread_data *d = _data;
 474         struct strbuf prefix = STRBUF_INIT;
 475         handle_range_1(d->istate, d->k_start, d->k_end, NULL, &prefix, d->lazy_entries);
 476         strbuf_release(&prefix);
 477         return NULL;
 478 }
 479
 480 struct lazy_name_thread_data {
 481         pthread_t pthread;
 482         struct index_state *istate;
 483         struct lazy_entry *lazy_entries;
 484 };
 485
 486 static void *lazy_name_thread_proc(void *_data)
 487 {
 488         struct lazy_name_thread_data *d = _data;
 489         int k;
 490
 491         for (k = 0; k < d->istate->cache_nr; k++) {
 492                 struct cache_entry *ce_k = d->istate->cache[k];
 493                 ce_k->ce_flags |= CE_HASHED;
 494                 hashmap_entry_init(&ce_k->ent, d->lazy_entries[k].hash_name);
 495                 hashmap_add(&d->istate->name_hash, &ce_k->ent);
 496         }
 497
 498         return NULL;
 499 }
 500
 501 static inline void lazy_update_dir_ref_counts(
 502         struct index_state *istate,
 503         struct lazy_entry *lazy_entries)
 504 {
 505         int k;
 506
 507         for (k = 0; k < istate->cache_nr; k++) {
 508                 if (lazy_entries[k].dir)
 509                         lazy_entries[k].dir->nr++;
 510         }
 511 }
 512
 513 static void threaded_lazy_init_name_hash(
 514         struct index_state *istate)
 515 {
 516         int err;
 517         int nr_each;
 518         int k_start;
 519         int t;
 520         struct lazy_entry *lazy_entries;
 521         struct lazy_dir_thread_data *td_dir;
 522         struct lazy_name_thread_data *td_name;
 523
 524         if (!HAVE_THREADS)
 525                 return;
 526
 527         k_start = 0;
 528         nr_each = DIV_ROUND_UP(istate->cache_nr, lazy_nr_dir_threads);
 529
 530         CALLOC_ARRAY(lazy_entries, istate->cache_nr);
 531         CALLOC_ARRAY(td_dir, lazy_nr_dir_threads);
 532         CALLOC_ARRAY(td_name, 1);
 533
 534         init_dir_mutex();
 535
 536         /*
 537          * Phase 1:
 538          * Build "istate->dir_hash" using n "dir" threads (and a read-only index).
 539          */
 540         for (t = 0; t < lazy_nr_dir_threads; t++) {
 541                 struct lazy_dir_thread_data *td_dir_t = td_dir + t;
 542                 td_dir_t->istate = istate;
 543                 td_dir_t->lazy_entries = lazy_entries;
 544                 td_dir_t->k_start = k_start;
 545                 k_start += nr_each;
 546                 if (k_start > istate->cache_nr)
 547                         k_start = istate->cache_nr;
 548                 td_dir_t->k_end = k_start;
 549                 err = pthread_create(&td_dir_t->pthread, NULL, lazy_dir_thread_proc, td_dir_t);
 550                 if (err)
 551                         die(_("unable to create lazy_dir thread: %s"), strerror(err));
 552         }
 553         for (t = 0; t < lazy_nr_dir_threads; t++) {
 554                 struct lazy_dir_thread_data *td_dir_t = td_dir + t;
 555                 if (pthread_join(td_dir_t->pthread, NULL))
 556                         die("unable to join lazy_dir_thread");
 557         }
 558
 559         /*
 560          * Phase 2:
 561          * Iterate over all index entries and add them to the "istate->name_hash"
 562          * using a single "name" background thread.
 563          * (Testing showed it wasn't worth running more than 1 thread for this.)
 564          *
 565          * Meanwhile, finish updating the parent directory ref-counts for each
 566          * index entry using the current thread.  (This step is very fast and
 567          * doesn't need threading.)
 568          */
 569         td_name->istate = istate;
 570         td_name->lazy_entries = lazy_entries;
 571         err = pthread_create(&td_name->pthread, NULL, lazy_name_thread_proc, td_name);
 572         if (err)
 573                 die(_("unable to create lazy_name thread: %s"), strerror(err));
 574
 575         lazy_update_dir_ref_counts(istate, lazy_entries);
 576
 577         err = pthread_join(td_name->pthread, NULL);
 578         if (err)
 579                 die(_("unable to join lazy_name thread: %s"), strerror(err));
 580
 581         cleanup_dir_mutex();
 582
 583         free(td_name);
 584         free(td_dir);
 585         free(lazy_entries);
 586 }
 587
 588 static void lazy_init_name_hash(struct index_state *istate)
 589 {
 590
 591         if (istate->name_hash_initialized)
 592                 return;
 593         trace_performance_enter();
 594         trace2_region_enter("index", "name-hash-init", istate->repo);
 595         hashmap_init(&istate->name_hash, cache_entry_cmp, NULL, istate->cache_nr);
 596         hashmap_init(&istate->dir_hash, dir_entry_cmp, NULL, istate->cache_nr);
 597
 598         if (lookup_lazy_params(istate)) {
 599                 /*
 600                  * Disable item counting and automatic rehashing because
 601                  * we do per-chain (mod n) locking rather than whole hashmap
 602                  * locking and we need to prevent the table-size from changing
 603                  * and bucket items from being redistributed.
 604                  */
 605                 hashmap_disable_item_counting(&istate->dir_hash);
 606                 threaded_lazy_init_name_hash(istate);
 607                 hashmap_enable_item_counting(&istate->dir_hash);
 608         } else {
 609                 int nr;
 610                 for (nr = 0; nr < istate->cache_nr; nr++)
 611                         hash_index_entry(istate, istate->cache[nr]);
 612         }
 613
 614         istate->name_hash_initialized = 1;
 615         trace2_region_leave("index", "name-hash-init", istate->repo);
 616         trace_performance_leave("initialize name hash");
 617 }
 618
 619 /*
 620  * A test routine for t/helper/ sources.
 621  *
 622  * Returns the number of threads used or 0 when
 623  * the non-threaded code path was used.
 624  *
 625  * Requesting threading WILL NOT override guards
 626  * in lookup_lazy_params().
 627  */
 628 int test_lazy_init_name_hash(struct index_state *istate, int try_threaded)
 629 {
 630         lazy_nr_dir_threads = 0;
 631         lazy_try_threaded = try_threaded;
 632
 633         lazy_init_name_hash(istate);
 634
 635         return lazy_nr_dir_threads;
 636 }
 637
 638 void add_name_hash(struct index_state *istate, struct cache_entry *ce)
 639 {
 640         if (istate->name_hash_initialized)
 641                 hash_index_entry(istate, ce);
 642 }
 643
 644 void remove_name_hash(struct index_state *istate, struct cache_entry *ce)
 645 {
 646         if (!istate->name_hash_initialized || !(ce->ce_flags & CE_HASHED))
 647                 return;
 648         ce->ce_flags &= ~CE_HASHED;
 649         hashmap_remove(&istate->name_hash, &ce->ent, ce);
 650
 651         if (ignore_case)
 652                 remove_dir_entry(istate, ce);
 653 }
 654
 655 static int slow_same_name(const char *name1, int len1, const char *name2, int len2)
 656 {
 657         if (len1 != len2)
 658                 return 0;
 659
 660         while (len1) {
 661                 unsigned char c1 = *name1++;
 662                 unsigned char c2 = *name2++;
 663                 len1--;
 664                 if (c1 != c2) {
 665                         c1 = toupper(c1);
 666                         c2 = toupper(c2);
 667                         if (c1 != c2)
 668                                 return 0;
 669                 }
 670         }
 671         return 1;
 672 }
 673
 674 static int same_name(const struct cache_entry *ce, const char *name, int namelen, int icase)
 675 {
 676         int len = ce_namelen(ce);
 677
 678         /*
 679          * Always do exact compare, even if we want a case-ignoring comparison;
 680          * we do the quick exact one first, because it will be the common case.
 681          */
 682         if (len == namelen && !memcmp(name, ce->name, len))
 683                 return 1;
 684
 685         if (!icase)
 686                 return 0;
 687
 688         return slow_same_name(name, namelen, ce->name, len);
 689 }
 690
 691 int index_dir_find(struct index_state *istate, const char *name, int namelen,
 692                    struct strbuf *canonical_path)
 693 {
 694         struct dir_entry *dir;
 695
 696         lazy_init_name_hash(istate);
 697         expand_to_path(istate, name, namelen, 0);
 698         dir = find_dir_entry(istate, name, namelen);
 699
 700         if (canonical_path && dir && dir->nr) {
 701                 strbuf_reset(canonical_path);
 702                 strbuf_add(canonical_path, dir->name, dir->namelen);
 703         }
 704
 705         return dir && dir->nr;
 706 }
 707
 708 void adjust_dirname_case(struct index_state *istate, char *name)
 709 {
 710         const char *startPtr = name;
 711         const char *ptr = startPtr;
 712
 713         lazy_init_name_hash(istate);
 714         expand_to_path(istate, name, strlen(name), 0);
 715         while (*ptr) {
 716                 while (*ptr && *ptr != '/')
 717                         ptr++;
 718
 719                 if (*ptr == '/') {
 720                         struct dir_entry *dir;
 721
 722                         dir = find_dir_entry(istate, name, ptr - name);
 723                         if (dir) {
 724                                 memcpy((void *)startPtr, dir->name + (startPtr - name), ptr - startPtr);
 725                                 startPtr = ptr + 1;
 726                         }
 727                         ptr++;
 728                 }
 729         }
 730 }
 731
 732 struct cache_entry *index_file_exists(struct index_state *istate, const char *name, int namelen, int icase)
 733 {
 734         struct cache_entry *ce;
 735         unsigned int hash = memihash(name, namelen);
 736
 737         lazy_init_name_hash(istate);
 738         expand_to_path(istate, name, namelen, icase);
 739
 740         ce = hashmap_get_entry_from_hash(&istate->name_hash, hash, NULL,
 741                                          struct cache_entry, ent);
 742         hashmap_for_each_entry_from(&istate->name_hash, ce, ent) {
 743                 if (same_name(ce, name, namelen, icase))
 744                         return ce;
 745         }
 746         return NULL;
 747 }
 748
 749 void free_name_hash(struct index_state *istate)
 750 {
 751         if (!istate->name_hash_initialized)
 752                 return;
 753         istate->name_hash_initialized = 0;
 754
 755         hashmap_clear(&istate->name_hash);
 756         hashmap_clear_and_free(&istate->dir_hash, struct dir_entry, ent);
 757 }