refs / ref-cache.con commit cache_ref_iterator_begin(): avoid priming unneeded directories (f23092f)
   1#include "../cache.h"
   2#include "../refs.h"
   3#include "refs-internal.h"
   4#include "ref-cache.h"
   5#include "../iterator.h"
   6
   7void add_entry_to_dir(struct ref_dir *dir, struct ref_entry *entry)
   8{
   9        ALLOC_GROW(dir->entries, dir->nr + 1, dir->alloc);
  10        dir->entries[dir->nr++] = entry;
  11        /* optimize for the case that entries are added in order */
  12        if (dir->nr == 1 ||
  13            (dir->nr == dir->sorted + 1 &&
  14             strcmp(dir->entries[dir->nr - 2]->name,
  15                    dir->entries[dir->nr - 1]->name) < 0))
  16                dir->sorted = dir->nr;
  17}
  18
  19struct ref_dir *get_ref_dir(struct ref_entry *entry)
  20{
  21        struct ref_dir *dir;
  22        assert(entry->flag & REF_DIR);
  23        dir = &entry->u.subdir;
  24        if (entry->flag & REF_INCOMPLETE) {
  25                if (!dir->cache->fill_ref_dir)
  26                        die("BUG: incomplete ref_store without fill_ref_dir function");
  27
  28                dir->cache->fill_ref_dir(dir->cache->ref_store, dir, entry->name);
  29                entry->flag &= ~REF_INCOMPLETE;
  30        }
  31        return dir;
  32}
  33
  34struct ref_entry *create_ref_entry(const char *refname,
  35                                   const struct object_id *oid, int flag)
  36{
  37        struct ref_entry *ref;
  38
  39        FLEX_ALLOC_STR(ref, name, refname);
  40        oidcpy(&ref->u.value.oid, oid);
  41        oidclr(&ref->u.value.peeled);
  42        ref->flag = flag;
  43        return ref;
  44}
  45
  46struct ref_cache *create_ref_cache(struct ref_store *refs,
  47                                   fill_ref_dir_fn *fill_ref_dir)
  48{
  49        struct ref_cache *ret = xcalloc(1, sizeof(*ret));
  50
  51        ret->ref_store = refs;
  52        ret->fill_ref_dir = fill_ref_dir;
  53        ret->root = create_dir_entry(ret, "", 0, 1);
  54        return ret;
  55}
  56
  57static void clear_ref_dir(struct ref_dir *dir);
  58
  59static void free_ref_entry(struct ref_entry *entry)
  60{
  61        if (entry->flag & REF_DIR) {
  62                /*
  63                 * Do not use get_ref_dir() here, as that might
  64                 * trigger the reading of loose refs.
  65                 */
  66                clear_ref_dir(&entry->u.subdir);
  67        }
  68        free(entry);
  69}
  70
  71void free_ref_cache(struct ref_cache *cache)
  72{
  73        free_ref_entry(cache->root);
  74        free(cache);
  75}
  76
  77/*
  78 * Clear and free all entries in dir, recursively.
  79 */
  80static void clear_ref_dir(struct ref_dir *dir)
  81{
  82        int i;
  83        for (i = 0; i < dir->nr; i++)
  84                free_ref_entry(dir->entries[i]);
  85        free(dir->entries);
  86        dir->sorted = dir->nr = dir->alloc = 0;
  87        dir->entries = NULL;
  88}
  89
  90struct ref_entry *create_dir_entry(struct ref_cache *cache,
  91                                   const char *dirname, size_t len,
  92                                   int incomplete)
  93{
  94        struct ref_entry *direntry;
  95
  96        FLEX_ALLOC_MEM(direntry, name, dirname, len);
  97        direntry->u.subdir.cache = cache;
  98        direntry->flag = REF_DIR | (incomplete ? REF_INCOMPLETE : 0);
  99        return direntry;
 100}
 101
 102static int ref_entry_cmp(const void *a, const void *b)
 103{
 104        struct ref_entry *one = *(struct ref_entry **)a;
 105        struct ref_entry *two = *(struct ref_entry **)b;
 106        return strcmp(one->name, two->name);
 107}
 108
 109static void sort_ref_dir(struct ref_dir *dir);
 110
 111struct string_slice {
 112        size_t len;
 113        const char *str;
 114};
 115
 116static int ref_entry_cmp_sslice(const void *key_, const void *ent_)
 117{
 118        const struct string_slice *key = key_;
 119        const struct ref_entry *ent = *(const struct ref_entry * const *)ent_;
 120        int cmp = strncmp(key->str, ent->name, key->len);
 121        if (cmp)
 122                return cmp;
 123        return '\0' - (unsigned char)ent->name[key->len];
 124}
 125
 126int search_ref_dir(struct ref_dir *dir, const char *refname, size_t len)
 127{
 128        struct ref_entry **r;
 129        struct string_slice key;
 130
 131        if (refname == NULL || !dir->nr)
 132                return -1;
 133
 134        sort_ref_dir(dir);
 135        key.len = len;
 136        key.str = refname;
 137        r = bsearch(&key, dir->entries, dir->nr, sizeof(*dir->entries),
 138                    ref_entry_cmp_sslice);
 139
 140        if (r == NULL)
 141                return -1;
 142
 143        return r - dir->entries;
 144}
 145
 146/*
 147 * Search for a directory entry directly within dir (without
 148 * recursing).  Sort dir if necessary.  subdirname must be a directory
 149 * name (i.e., end in '/').  If mkdir is set, then create the
 150 * directory if it is missing; otherwise, return NULL if the desired
 151 * directory cannot be found.  dir must already be complete.
 152 */
 153static struct ref_dir *search_for_subdir(struct ref_dir *dir,
 154                                         const char *subdirname, size_t len,
 155                                         int mkdir)
 156{
 157        int entry_index = search_ref_dir(dir, subdirname, len);
 158        struct ref_entry *entry;
 159        if (entry_index == -1) {
 160                if (!mkdir)
 161                        return NULL;
 162                /*
 163                 * Since dir is complete, the absence of a subdir
 164                 * means that the subdir really doesn't exist;
 165                 * therefore, create an empty record for it but mark
 166                 * the record complete.
 167                 */
 168                entry = create_dir_entry(dir->cache, subdirname, len, 0);
 169                add_entry_to_dir(dir, entry);
 170        } else {
 171                entry = dir->entries[entry_index];
 172        }
 173        return get_ref_dir(entry);
 174}
 175
 176/*
 177 * If refname is a reference name, find the ref_dir within the dir
 178 * tree that should hold refname. If refname is a directory name
 179 * (i.e., it ends in '/'), then return that ref_dir itself. dir must
 180 * represent the top-level directory and must already be complete.
 181 * Sort ref_dirs and recurse into subdirectories as necessary. If
 182 * mkdir is set, then create any missing directories; otherwise,
 183 * return NULL if the desired directory cannot be found.
 184 */
 185static struct ref_dir *find_containing_dir(struct ref_dir *dir,
 186                                           const char *refname, int mkdir)
 187{
 188        const char *slash;
 189        for (slash = strchr(refname, '/'); slash; slash = strchr(slash + 1, '/')) {
 190                size_t dirnamelen = slash - refname + 1;
 191                struct ref_dir *subdir;
 192                subdir = search_for_subdir(dir, refname, dirnamelen, mkdir);
 193                if (!subdir) {
 194                        dir = NULL;
 195                        break;
 196                }
 197                dir = subdir;
 198        }
 199
 200        return dir;
 201}
 202
 203struct ref_entry *find_ref_entry(struct ref_dir *dir, const char *refname)
 204{
 205        int entry_index;
 206        struct ref_entry *entry;
 207        dir = find_containing_dir(dir, refname, 0);
 208        if (!dir)
 209                return NULL;
 210        entry_index = search_ref_dir(dir, refname, strlen(refname));
 211        if (entry_index == -1)
 212                return NULL;
 213        entry = dir->entries[entry_index];
 214        return (entry->flag & REF_DIR) ? NULL : entry;
 215}
 216
 217int remove_entry_from_dir(struct ref_dir *dir, const char *refname)
 218{
 219        int refname_len = strlen(refname);
 220        int entry_index;
 221        struct ref_entry *entry;
 222        int is_dir = refname[refname_len - 1] == '/';
 223        if (is_dir) {
 224                /*
 225                 * refname represents a reference directory.  Remove
 226                 * the trailing slash; otherwise we will get the
 227                 * directory *representing* refname rather than the
 228                 * one *containing* it.
 229                 */
 230                char *dirname = xmemdupz(refname, refname_len - 1);
 231                dir = find_containing_dir(dir, dirname, 0);
 232                free(dirname);
 233        } else {
 234                dir = find_containing_dir(dir, refname, 0);
 235        }
 236        if (!dir)
 237                return -1;
 238        entry_index = search_ref_dir(dir, refname, refname_len);
 239        if (entry_index == -1)
 240                return -1;
 241        entry = dir->entries[entry_index];
 242
 243        memmove(&dir->entries[entry_index],
 244                &dir->entries[entry_index + 1],
 245                (dir->nr - entry_index - 1) * sizeof(*dir->entries)
 246                );
 247        dir->nr--;
 248        if (dir->sorted > entry_index)
 249                dir->sorted--;
 250        free_ref_entry(entry);
 251        return dir->nr;
 252}
 253
 254int add_ref_entry(struct ref_dir *dir, struct ref_entry *ref)
 255{
 256        dir = find_containing_dir(dir, ref->name, 1);
 257        if (!dir)
 258                return -1;
 259        add_entry_to_dir(dir, ref);
 260        return 0;
 261}
 262
 263/*
 264 * Emit a warning and return true iff ref1 and ref2 have the same name
 265 * and the same sha1.  Die if they have the same name but different
 266 * sha1s.
 267 */
 268static int is_dup_ref(const struct ref_entry *ref1, const struct ref_entry *ref2)
 269{
 270        if (strcmp(ref1->name, ref2->name))
 271                return 0;
 272
 273        /* Duplicate name; make sure that they don't conflict: */
 274
 275        if ((ref1->flag & REF_DIR) || (ref2->flag & REF_DIR))
 276                /* This is impossible by construction */
 277                die("Reference directory conflict: %s", ref1->name);
 278
 279        if (oidcmp(&ref1->u.value.oid, &ref2->u.value.oid))
 280                die("Duplicated ref, and SHA1s don't match: %s", ref1->name);
 281
 282        warning("Duplicated ref: %s", ref1->name);
 283        return 1;
 284}
 285
 286/*
 287 * Sort the entries in dir non-recursively (if they are not already
 288 * sorted) and remove any duplicate entries.
 289 */
 290static void sort_ref_dir(struct ref_dir *dir)
 291{
 292        int i, j;
 293        struct ref_entry *last = NULL;
 294
 295        /*
 296         * This check also prevents passing a zero-length array to qsort(),
 297         * which is a problem on some platforms.
 298         */
 299        if (dir->sorted == dir->nr)
 300                return;
 301
 302        QSORT(dir->entries, dir->nr, ref_entry_cmp);
 303
 304        /* Remove any duplicates: */
 305        for (i = 0, j = 0; j < dir->nr; j++) {
 306                struct ref_entry *entry = dir->entries[j];
 307                if (last && is_dup_ref(last, entry))
 308                        free_ref_entry(entry);
 309                else
 310                        last = dir->entries[i++] = entry;
 311        }
 312        dir->sorted = dir->nr = i;
 313}
 314
 315enum prefix_state {
 316        /* All refs within the directory would match prefix: */
 317        PREFIX_CONTAINS_DIR,
 318
 319        /* Some, but not all, refs within the directory might match prefix: */
 320        PREFIX_WITHIN_DIR,
 321
 322        /* No refs within the directory could possibly match prefix: */
 323        PREFIX_EXCLUDES_DIR
 324};
 325
 326/*
 327 * Return a `prefix_state` constant describing the relationship
 328 * between the directory with the specified `dirname` and `prefix`.
 329 */
 330static enum prefix_state overlaps_prefix(const char *dirname,
 331                                         const char *prefix)
 332{
 333        while (*prefix && *dirname == *prefix) {
 334                dirname++;
 335                prefix++;
 336        }
 337        if (!*prefix)
 338                return PREFIX_CONTAINS_DIR;
 339        else if (!*dirname)
 340                return PREFIX_WITHIN_DIR;
 341        else
 342                return PREFIX_EXCLUDES_DIR;
 343}
 344
 345/*
 346 * Load all of the refs from `dir` (recursively) that could possibly
 347 * contain references matching `prefix` into our in-memory cache. If
 348 * `prefix` is NULL, prime unconditionally.
 349 */
 350static void prime_ref_dir(struct ref_dir *dir, const char *prefix)
 351{
 352        /*
 353         * The hard work of loading loose refs is done by get_ref_dir(), so we
 354         * just need to recurse through all of the sub-directories. We do not
 355         * even need to care about sorting, as traversal order does not matter
 356         * to us.
 357         */
 358        int i;
 359        for (i = 0; i < dir->nr; i++) {
 360                struct ref_entry *entry = dir->entries[i];
 361                if (!(entry->flag & REF_DIR)) {
 362                        /* Not a directory; no need to recurse. */
 363                } else if (!prefix) {
 364                        /* Recurse in any case: */
 365                        prime_ref_dir(get_ref_dir(entry), NULL);
 366                } else {
 367                        switch (overlaps_prefix(entry->name, prefix)) {
 368                        case PREFIX_CONTAINS_DIR:
 369                                /*
 370                                 * Recurse, and from here down we
 371                                 * don't have to check the prefix
 372                                 * anymore:
 373                                 */
 374                                prime_ref_dir(get_ref_dir(entry), NULL);
 375                                break;
 376                        case PREFIX_WITHIN_DIR:
 377                                prime_ref_dir(get_ref_dir(entry), prefix);
 378                                break;
 379                        case PREFIX_EXCLUDES_DIR:
 380                                /* No need to prime this directory. */
 381                                break;
 382                        }
 383                }
 384        }
 385}
 386
 387/*
 388 * A level in the reference hierarchy that is currently being iterated
 389 * through.
 390 */
 391struct cache_ref_iterator_level {
 392        /*
 393         * The ref_dir being iterated over at this level. The ref_dir
 394         * is sorted before being stored here.
 395         */
 396        struct ref_dir *dir;
 397
 398        enum prefix_state prefix_state;
 399
 400        /*
 401         * The index of the current entry within dir (which might
 402         * itself be a directory). If index == -1, then the iteration
 403         * hasn't yet begun. If index == dir->nr, then the iteration
 404         * through this level is over.
 405         */
 406        int index;
 407};
 408
 409/*
 410 * Represent an iteration through a ref_dir in the memory cache. The
 411 * iteration recurses through subdirectories.
 412 */
 413struct cache_ref_iterator {
 414        struct ref_iterator base;
 415
 416        /*
 417         * The number of levels currently on the stack. This is always
 418         * at least 1, because when it becomes zero the iteration is
 419         * ended and this struct is freed.
 420         */
 421        size_t levels_nr;
 422
 423        /* The number of levels that have been allocated on the stack */
 424        size_t levels_alloc;
 425
 426        /*
 427         * Only include references with this prefix in the iteration.
 428         * The prefix is matched textually, without regard for path
 429         * component boundaries.
 430         */
 431        const char *prefix;
 432
 433        /*
 434         * A stack of levels. levels[0] is the uppermost level that is
 435         * being iterated over in this iteration. (This is not
 436         * necessary the top level in the references hierarchy. If we
 437         * are iterating through a subtree, then levels[0] will hold
 438         * the ref_dir for that subtree, and subsequent levels will go
 439         * on from there.)
 440         */
 441        struct cache_ref_iterator_level *levels;
 442};
 443
 444static int cache_ref_iterator_advance(struct ref_iterator *ref_iterator)
 445{
 446        struct cache_ref_iterator *iter =
 447                (struct cache_ref_iterator *)ref_iterator;
 448
 449        while (1) {
 450                struct cache_ref_iterator_level *level =
 451                        &iter->levels[iter->levels_nr - 1];
 452                struct ref_dir *dir = level->dir;
 453                struct ref_entry *entry;
 454                enum prefix_state entry_prefix_state;
 455
 456                if (level->index == -1)
 457                        sort_ref_dir(dir);
 458
 459                if (++level->index == level->dir->nr) {
 460                        /* This level is exhausted; pop up a level */
 461                        if (--iter->levels_nr == 0)
 462                                return ref_iterator_abort(ref_iterator);
 463
 464                        continue;
 465                }
 466
 467                entry = dir->entries[level->index];
 468
 469                if (level->prefix_state == PREFIX_WITHIN_DIR) {
 470                        entry_prefix_state = overlaps_prefix(entry->name, iter->prefix);
 471                        if (entry_prefix_state == PREFIX_EXCLUDES_DIR)
 472                                continue;
 473                } else {
 474                        entry_prefix_state = level->prefix_state;
 475                }
 476
 477                if (entry->flag & REF_DIR) {
 478                        /* push down a level */
 479                        ALLOC_GROW(iter->levels, iter->levels_nr + 1,
 480                                   iter->levels_alloc);
 481
 482                        level = &iter->levels[iter->levels_nr++];
 483                        level->dir = get_ref_dir(entry);
 484                        level->prefix_state = entry_prefix_state;
 485                        level->index = -1;
 486                } else {
 487                        iter->base.refname = entry->name;
 488                        iter->base.oid = &entry->u.value.oid;
 489                        iter->base.flags = entry->flag;
 490                        return ITER_OK;
 491                }
 492        }
 493}
 494
 495enum peel_status peel_entry(struct ref_entry *entry, int repeel)
 496{
 497        enum peel_status status;
 498
 499        if (entry->flag & REF_KNOWS_PEELED) {
 500                if (repeel) {
 501                        entry->flag &= ~REF_KNOWS_PEELED;
 502                        oidclr(&entry->u.value.peeled);
 503                } else {
 504                        return is_null_oid(&entry->u.value.peeled) ?
 505                                PEEL_NON_TAG : PEEL_PEELED;
 506                }
 507        }
 508        if (entry->flag & REF_ISBROKEN)
 509                return PEEL_BROKEN;
 510        if (entry->flag & REF_ISSYMREF)
 511                return PEEL_IS_SYMREF;
 512
 513        status = peel_object(entry->u.value.oid.hash, entry->u.value.peeled.hash);
 514        if (status == PEEL_PEELED || status == PEEL_NON_TAG)
 515                entry->flag |= REF_KNOWS_PEELED;
 516        return status;
 517}
 518
 519static int cache_ref_iterator_peel(struct ref_iterator *ref_iterator,
 520                                   struct object_id *peeled)
 521{
 522        struct cache_ref_iterator *iter =
 523                (struct cache_ref_iterator *)ref_iterator;
 524        struct cache_ref_iterator_level *level;
 525        struct ref_entry *entry;
 526
 527        level = &iter->levels[iter->levels_nr - 1];
 528
 529        if (level->index == -1)
 530                die("BUG: peel called before advance for cache iterator");
 531
 532        entry = level->dir->entries[level->index];
 533
 534        if (peel_entry(entry, 0))
 535                return -1;
 536        oidcpy(peeled, &entry->u.value.peeled);
 537        return 0;
 538}
 539
 540static int cache_ref_iterator_abort(struct ref_iterator *ref_iterator)
 541{
 542        struct cache_ref_iterator *iter =
 543                (struct cache_ref_iterator *)ref_iterator;
 544
 545        free((char *)iter->prefix);
 546        free(iter->levels);
 547        base_ref_iterator_free(ref_iterator);
 548        return ITER_DONE;
 549}
 550
 551static struct ref_iterator_vtable cache_ref_iterator_vtable = {
 552        cache_ref_iterator_advance,
 553        cache_ref_iterator_peel,
 554        cache_ref_iterator_abort
 555};
 556
 557struct ref_iterator *cache_ref_iterator_begin(struct ref_cache *cache,
 558                                              const char *prefix,
 559                                              int prime_dir)
 560{
 561        struct ref_dir *dir;
 562        struct cache_ref_iterator *iter;
 563        struct ref_iterator *ref_iterator;
 564        struct cache_ref_iterator_level *level;
 565
 566        dir = get_ref_dir(cache->root);
 567        if (prefix && *prefix)
 568                dir = find_containing_dir(dir, prefix, 0);
 569        if (!dir)
 570                /* There's nothing to iterate over. */
 571                return empty_ref_iterator_begin();
 572
 573        if (prime_dir)
 574                prime_ref_dir(dir, prefix);
 575
 576        iter = xcalloc(1, sizeof(*iter));
 577        ref_iterator = &iter->base;
 578        base_ref_iterator_init(ref_iterator, &cache_ref_iterator_vtable);
 579        ALLOC_GROW(iter->levels, 10, iter->levels_alloc);
 580
 581        iter->levels_nr = 1;
 582        level = &iter->levels[0];
 583        level->index = -1;
 584        level->dir = dir;
 585
 586        if (prefix && *prefix) {
 587                iter->prefix = xstrdup(prefix);
 588                level->prefix_state = PREFIX_WITHIN_DIR;
 589        } else {
 590                level->prefix_state = PREFIX_CONTAINS_DIR;
 591        }
 592
 593        return ref_iterator;
 594}