sha1-name.con commit midx: use midx in abbreviation calculations (8aac67a)
   1#include "cache.h"
   2#include "config.h"
   3#include "tag.h"
   4#include "commit.h"
   5#include "tree.h"
   6#include "blob.h"
   7#include "tree-walk.h"
   8#include "refs.h"
   9#include "remote.h"
  10#include "dir.h"
  11#include "sha1-array.h"
  12#include "packfile.h"
  13#include "object-store.h"
  14#include "repository.h"
  15#include "midx.h"
  16
  17static int get_oid_oneline(const char *, struct object_id *, struct commit_list *);
  18
  19typedef int (*disambiguate_hint_fn)(const struct object_id *, void *);
  20
  21struct disambiguate_state {
  22        int len; /* length of prefix in hex chars */
  23        char hex_pfx[GIT_MAX_HEXSZ + 1];
  24        struct object_id bin_pfx;
  25
  26        disambiguate_hint_fn fn;
  27        void *cb_data;
  28        struct object_id candidate;
  29        unsigned candidate_exists:1;
  30        unsigned candidate_checked:1;
  31        unsigned candidate_ok:1;
  32        unsigned disambiguate_fn_used:1;
  33        unsigned ambiguous:1;
  34        unsigned always_call_fn:1;
  35};
  36
  37static void update_candidates(struct disambiguate_state *ds, const struct object_id *current)
  38{
  39        if (ds->always_call_fn) {
  40                ds->ambiguous = ds->fn(current, ds->cb_data) ? 1 : 0;
  41                return;
  42        }
  43        if (!ds->candidate_exists) {
  44                /* this is the first candidate */
  45                oidcpy(&ds->candidate, current);
  46                ds->candidate_exists = 1;
  47                return;
  48        } else if (!oidcmp(&ds->candidate, current)) {
  49                /* the same as what we already have seen */
  50                return;
  51        }
  52
  53        if (!ds->fn) {
  54                /* cannot disambiguate between ds->candidate and current */
  55                ds->ambiguous = 1;
  56                return;
  57        }
  58
  59        if (!ds->candidate_checked) {
  60                ds->candidate_ok = ds->fn(&ds->candidate, ds->cb_data);
  61                ds->disambiguate_fn_used = 1;
  62                ds->candidate_checked = 1;
  63        }
  64
  65        if (!ds->candidate_ok) {
  66                /* discard the candidate; we know it does not satisfy fn */
  67                oidcpy(&ds->candidate, current);
  68                ds->candidate_checked = 0;
  69                return;
  70        }
  71
  72        /* if we reach this point, we know ds->candidate satisfies fn */
  73        if (ds->fn(current, ds->cb_data)) {
  74                /*
  75                 * if both current and candidate satisfy fn, we cannot
  76                 * disambiguate.
  77                 */
  78                ds->candidate_ok = 0;
  79                ds->ambiguous = 1;
  80        }
  81
  82        /* otherwise, current can be discarded and candidate is still good */
  83}
  84
  85static int append_loose_object(const struct object_id *oid, const char *path,
  86                               void *data)
  87{
  88        oid_array_append(data, oid);
  89        return 0;
  90}
  91
  92static int match_sha(unsigned, const unsigned char *, const unsigned char *);
  93
  94static void find_short_object_filename(struct disambiguate_state *ds)
  95{
  96        int subdir_nr = ds->bin_pfx.hash[0];
  97        struct alternate_object_database *alt;
  98        static struct alternate_object_database *fakeent;
  99
 100        if (!fakeent) {
 101                /*
 102                 * Create a "fake" alternate object database that
 103                 * points to our own object database, to make it
 104                 * easier to get a temporary working space in
 105                 * alt->name/alt->base while iterating over the
 106                 * object databases including our own.
 107                 */
 108                fakeent = alloc_alt_odb(get_object_directory());
 109        }
 110        fakeent->next = the_repository->objects->alt_odb_list;
 111
 112        for (alt = fakeent; alt && !ds->ambiguous; alt = alt->next) {
 113                int pos;
 114
 115                if (!alt->loose_objects_subdir_seen[subdir_nr]) {
 116                        struct strbuf *buf = alt_scratch_buf(alt);
 117                        for_each_file_in_obj_subdir(subdir_nr, buf,
 118                                                    append_loose_object,
 119                                                    NULL, NULL,
 120                                                    &alt->loose_objects_cache);
 121                        alt->loose_objects_subdir_seen[subdir_nr] = 1;
 122                }
 123
 124                pos = oid_array_lookup(&alt->loose_objects_cache, &ds->bin_pfx);
 125                if (pos < 0)
 126                        pos = -1 - pos;
 127                while (!ds->ambiguous && pos < alt->loose_objects_cache.nr) {
 128                        const struct object_id *oid;
 129                        oid = alt->loose_objects_cache.oid + pos;
 130                        if (!match_sha(ds->len, ds->bin_pfx.hash, oid->hash))
 131                                break;
 132                        update_candidates(ds, oid);
 133                        pos++;
 134                }
 135        }
 136}
 137
 138static int match_sha(unsigned len, const unsigned char *a, const unsigned char *b)
 139{
 140        do {
 141                if (*a != *b)
 142                        return 0;
 143                a++;
 144                b++;
 145                len -= 2;
 146        } while (len > 1);
 147        if (len)
 148                if ((*a ^ *b) & 0xf0)
 149                        return 0;
 150        return 1;
 151}
 152
 153static void unique_in_midx(struct multi_pack_index *m,
 154                           struct disambiguate_state *ds)
 155{
 156        uint32_t num, i, first = 0;
 157        const struct object_id *current = NULL;
 158        num = m->num_objects;
 159
 160        if (!num)
 161                return;
 162
 163        bsearch_midx(&ds->bin_pfx, m, &first);
 164
 165        /*
 166         * At this point, "first" is the location of the lowest object
 167         * with an object name that could match "bin_pfx".  See if we have
 168         * 0, 1 or more objects that actually match(es).
 169         */
 170        for (i = first; i < num && !ds->ambiguous; i++) {
 171                struct object_id oid;
 172                current = nth_midxed_object_oid(&oid, m, i);
 173                if (!match_sha(ds->len, ds->bin_pfx.hash, current->hash))
 174                        break;
 175                update_candidates(ds, current);
 176        }
 177}
 178
 179static void unique_in_pack(struct packed_git *p,
 180                           struct disambiguate_state *ds)
 181{
 182        uint32_t num, i, first = 0;
 183        const struct object_id *current = NULL;
 184
 185        if (open_pack_index(p) || !p->num_objects)
 186                return;
 187
 188        num = p->num_objects;
 189        bsearch_pack(&ds->bin_pfx, p, &first);
 190
 191        /*
 192         * At this point, "first" is the location of the lowest object
 193         * with an object name that could match "bin_pfx".  See if we have
 194         * 0, 1 or more objects that actually match(es).
 195         */
 196        for (i = first; i < num && !ds->ambiguous; i++) {
 197                struct object_id oid;
 198                current = nth_packed_object_oid(&oid, p, i);
 199                if (!match_sha(ds->len, ds->bin_pfx.hash, current->hash))
 200                        break;
 201                update_candidates(ds, current);
 202        }
 203}
 204
 205static void find_short_packed_object(struct disambiguate_state *ds)
 206{
 207        struct multi_pack_index *m;
 208        struct packed_git *p;
 209
 210        for (m = get_multi_pack_index(the_repository); m && !ds->ambiguous;
 211             m = m->next)
 212                unique_in_midx(m, ds);
 213        for (p = get_packed_git(the_repository); p && !ds->ambiguous;
 214             p = p->next)
 215                unique_in_pack(p, ds);
 216}
 217
 218#define SHORT_NAME_NOT_FOUND (-1)
 219#define SHORT_NAME_AMBIGUOUS (-2)
 220
 221static int finish_object_disambiguation(struct disambiguate_state *ds,
 222                                        struct object_id *oid)
 223{
 224        if (ds->ambiguous)
 225                return SHORT_NAME_AMBIGUOUS;
 226
 227        if (!ds->candidate_exists)
 228                return SHORT_NAME_NOT_FOUND;
 229
 230        if (!ds->candidate_checked)
 231                /*
 232                 * If this is the only candidate, there is no point
 233                 * calling the disambiguation hint callback.
 234                 *
 235                 * On the other hand, if the current candidate
 236                 * replaced an earlier candidate that did _not_ pass
 237                 * the disambiguation hint callback, then we do have
 238                 * more than one objects that match the short name
 239                 * given, so we should make sure this one matches;
 240                 * otherwise, if we discovered this one and the one
 241                 * that we previously discarded in the reverse order,
 242                 * we would end up showing different results in the
 243                 * same repository!
 244                 */
 245                ds->candidate_ok = (!ds->disambiguate_fn_used ||
 246                                    ds->fn(&ds->candidate, ds->cb_data));
 247
 248        if (!ds->candidate_ok)
 249                return SHORT_NAME_AMBIGUOUS;
 250
 251        oidcpy(oid, &ds->candidate);
 252        return 0;
 253}
 254
 255static int disambiguate_commit_only(const struct object_id *oid, void *cb_data_unused)
 256{
 257        int kind = oid_object_info(the_repository, oid, NULL);
 258        return kind == OBJ_COMMIT;
 259}
 260
 261static int disambiguate_committish_only(const struct object_id *oid, void *cb_data_unused)
 262{
 263        struct object *obj;
 264        int kind;
 265
 266        kind = oid_object_info(the_repository, oid, NULL);
 267        if (kind == OBJ_COMMIT)
 268                return 1;
 269        if (kind != OBJ_TAG)
 270                return 0;
 271
 272        /* We need to do this the hard way... */
 273        obj = deref_tag(parse_object(oid), NULL, 0);
 274        if (obj && obj->type == OBJ_COMMIT)
 275                return 1;
 276        return 0;
 277}
 278
 279static int disambiguate_tree_only(const struct object_id *oid, void *cb_data_unused)
 280{
 281        int kind = oid_object_info(the_repository, oid, NULL);
 282        return kind == OBJ_TREE;
 283}
 284
 285static int disambiguate_treeish_only(const struct object_id *oid, void *cb_data_unused)
 286{
 287        struct object *obj;
 288        int kind;
 289
 290        kind = oid_object_info(the_repository, oid, NULL);
 291        if (kind == OBJ_TREE || kind == OBJ_COMMIT)
 292                return 1;
 293        if (kind != OBJ_TAG)
 294                return 0;
 295
 296        /* We need to do this the hard way... */
 297        obj = deref_tag(parse_object(oid), NULL, 0);
 298        if (obj && (obj->type == OBJ_TREE || obj->type == OBJ_COMMIT))
 299                return 1;
 300        return 0;
 301}
 302
 303static int disambiguate_blob_only(const struct object_id *oid, void *cb_data_unused)
 304{
 305        int kind = oid_object_info(the_repository, oid, NULL);
 306        return kind == OBJ_BLOB;
 307}
 308
 309static disambiguate_hint_fn default_disambiguate_hint;
 310
 311int set_disambiguate_hint_config(const char *var, const char *value)
 312{
 313        static const struct {
 314                const char *name;
 315                disambiguate_hint_fn fn;
 316        } hints[] = {
 317                { "none", NULL },
 318                { "commit", disambiguate_commit_only },
 319                { "committish", disambiguate_committish_only },
 320                { "tree", disambiguate_tree_only },
 321                { "treeish", disambiguate_treeish_only },
 322                { "blob", disambiguate_blob_only }
 323        };
 324        int i;
 325
 326        if (!value)
 327                return config_error_nonbool(var);
 328
 329        for (i = 0; i < ARRAY_SIZE(hints); i++) {
 330                if (!strcasecmp(value, hints[i].name)) {
 331                        default_disambiguate_hint = hints[i].fn;
 332                        return 0;
 333                }
 334        }
 335
 336        return error("unknown hint type for '%s': %s", var, value);
 337}
 338
 339static int init_object_disambiguation(const char *name, int len,
 340                                      struct disambiguate_state *ds)
 341{
 342        int i;
 343
 344        if (len < MINIMUM_ABBREV || len > GIT_SHA1_HEXSZ)
 345                return -1;
 346
 347        memset(ds, 0, sizeof(*ds));
 348
 349        for (i = 0; i < len ;i++) {
 350                unsigned char c = name[i];
 351                unsigned char val;
 352                if (c >= '0' && c <= '9')
 353                        val = c - '0';
 354                else if (c >= 'a' && c <= 'f')
 355                        val = c - 'a' + 10;
 356                else if (c >= 'A' && c <='F') {
 357                        val = c - 'A' + 10;
 358                        c -= 'A' - 'a';
 359                }
 360                else
 361                        return -1;
 362                ds->hex_pfx[i] = c;
 363                if (!(i & 1))
 364                        val <<= 4;
 365                ds->bin_pfx.hash[i >> 1] |= val;
 366        }
 367
 368        ds->len = len;
 369        ds->hex_pfx[len] = '\0';
 370        prepare_alt_odb(the_repository);
 371        return 0;
 372}
 373
 374static int show_ambiguous_object(const struct object_id *oid, void *data)
 375{
 376        const struct disambiguate_state *ds = data;
 377        struct strbuf desc = STRBUF_INIT;
 378        int type;
 379
 380        if (ds->fn && !ds->fn(oid, ds->cb_data))
 381                return 0;
 382
 383        type = oid_object_info(the_repository, oid, NULL);
 384        if (type == OBJ_COMMIT) {
 385                struct commit *commit = lookup_commit(oid);
 386                if (commit) {
 387                        struct pretty_print_context pp = {0};
 388                        pp.date_mode.type = DATE_SHORT;
 389                        format_commit_message(commit, " %ad - %s", &desc, &pp);
 390                }
 391        } else if (type == OBJ_TAG) {
 392                struct tag *tag = lookup_tag(oid);
 393                if (!parse_tag(tag) && tag->tag)
 394                        strbuf_addf(&desc, " %s", tag->tag);
 395        }
 396
 397        advise("  %s %s%s",
 398               find_unique_abbrev(oid, DEFAULT_ABBREV),
 399               type_name(type) ? type_name(type) : "unknown type",
 400               desc.buf);
 401
 402        strbuf_release(&desc);
 403        return 0;
 404}
 405
 406static int collect_ambiguous(const struct object_id *oid, void *data)
 407{
 408        oid_array_append(data, oid);
 409        return 0;
 410}
 411
 412static int sort_ambiguous(const void *a, const void *b)
 413{
 414        int a_type = oid_object_info(the_repository, a, NULL);
 415        int b_type = oid_object_info(the_repository, b, NULL);
 416        int a_type_sort;
 417        int b_type_sort;
 418
 419        /*
 420         * Sorts by hash within the same object type, just as
 421         * oid_array_for_each_unique() would do.
 422         */
 423        if (a_type == b_type)
 424                return oidcmp(a, b);
 425
 426        /*
 427         * Between object types show tags, then commits, and finally
 428         * trees and blobs.
 429         *
 430         * The object_type enum is commit, tree, blob, tag, but we
 431         * want tag, commit, tree blob. Cleverly (perhaps too
 432         * cleverly) do that with modulus, since the enum assigns 1 to
 433         * commit, so tag becomes 0.
 434         */
 435        a_type_sort = a_type % 4;
 436        b_type_sort = b_type % 4;
 437        return a_type_sort > b_type_sort ? 1 : -1;
 438}
 439
 440static int get_short_oid(const char *name, int len, struct object_id *oid,
 441                          unsigned flags)
 442{
 443        int status;
 444        struct disambiguate_state ds;
 445        int quietly = !!(flags & GET_OID_QUIETLY);
 446
 447        if (init_object_disambiguation(name, len, &ds) < 0)
 448                return -1;
 449
 450        if (HAS_MULTI_BITS(flags & GET_OID_DISAMBIGUATORS))
 451                BUG("multiple get_short_oid disambiguator flags");
 452
 453        if (flags & GET_OID_COMMIT)
 454                ds.fn = disambiguate_commit_only;
 455        else if (flags & GET_OID_COMMITTISH)
 456                ds.fn = disambiguate_committish_only;
 457        else if (flags & GET_OID_TREE)
 458                ds.fn = disambiguate_tree_only;
 459        else if (flags & GET_OID_TREEISH)
 460                ds.fn = disambiguate_treeish_only;
 461        else if (flags & GET_OID_BLOB)
 462                ds.fn = disambiguate_blob_only;
 463        else
 464                ds.fn = default_disambiguate_hint;
 465
 466        find_short_object_filename(&ds);
 467        find_short_packed_object(&ds);
 468        status = finish_object_disambiguation(&ds, oid);
 469
 470        if (!quietly && (status == SHORT_NAME_AMBIGUOUS)) {
 471                struct oid_array collect = OID_ARRAY_INIT;
 472
 473                error(_("short SHA1 %s is ambiguous"), ds.hex_pfx);
 474
 475                /*
 476                 * We may still have ambiguity if we simply saw a series of
 477                 * candidates that did not satisfy our hint function. In
 478                 * that case, we still want to show them, so disable the hint
 479                 * function entirely.
 480                 */
 481                if (!ds.ambiguous)
 482                        ds.fn = NULL;
 483
 484                advise(_("The candidates are:"));
 485                for_each_abbrev(ds.hex_pfx, collect_ambiguous, &collect);
 486                QSORT(collect.oid, collect.nr, sort_ambiguous);
 487
 488                if (oid_array_for_each(&collect, show_ambiguous_object, &ds))
 489                        BUG("show_ambiguous_object shouldn't return non-zero");
 490                oid_array_clear(&collect);
 491        }
 492
 493        return status;
 494}
 495
 496int for_each_abbrev(const char *prefix, each_abbrev_fn fn, void *cb_data)
 497{
 498        struct oid_array collect = OID_ARRAY_INIT;
 499        struct disambiguate_state ds;
 500        int ret;
 501
 502        if (init_object_disambiguation(prefix, strlen(prefix), &ds) < 0)
 503                return -1;
 504
 505        ds.always_call_fn = 1;
 506        ds.fn = collect_ambiguous;
 507        ds.cb_data = &collect;
 508        find_short_object_filename(&ds);
 509        find_short_packed_object(&ds);
 510
 511        ret = oid_array_for_each_unique(&collect, fn, cb_data);
 512        oid_array_clear(&collect);
 513        return ret;
 514}
 515
 516/*
 517 * Return the slot of the most-significant bit set in "val". There are various
 518 * ways to do this quickly with fls() or __builtin_clzl(), but speed is
 519 * probably not a big deal here.
 520 */
 521static unsigned msb(unsigned long val)
 522{
 523        unsigned r = 0;
 524        while (val >>= 1)
 525                r++;
 526        return r;
 527}
 528
 529struct min_abbrev_data {
 530        unsigned int init_len;
 531        unsigned int cur_len;
 532        char *hex;
 533        const struct object_id *oid;
 534};
 535
 536static inline char get_hex_char_from_oid(const struct object_id *oid,
 537                                         unsigned int pos)
 538{
 539        static const char hex[] = "0123456789abcdef";
 540
 541        if ((pos & 1) == 0)
 542                return hex[oid->hash[pos >> 1] >> 4];
 543        else
 544                return hex[oid->hash[pos >> 1] & 0xf];
 545}
 546
 547static int extend_abbrev_len(const struct object_id *oid, void *cb_data)
 548{
 549        struct min_abbrev_data *mad = cb_data;
 550
 551        unsigned int i = mad->init_len;
 552        while (mad->hex[i] && mad->hex[i] == get_hex_char_from_oid(oid, i))
 553                i++;
 554
 555        if (i < GIT_MAX_RAWSZ && i >= mad->cur_len)
 556                mad->cur_len = i + 1;
 557
 558        return 0;
 559}
 560
 561static void find_abbrev_len_for_midx(struct multi_pack_index *m,
 562                                     struct min_abbrev_data *mad)
 563{
 564        int match = 0;
 565        uint32_t num, first = 0;
 566        struct object_id oid;
 567        const struct object_id *mad_oid;
 568
 569        if (!m->num_objects)
 570                return;
 571
 572        num = m->num_objects;
 573        mad_oid = mad->oid;
 574        match = bsearch_midx(mad_oid, m, &first);
 575
 576        /*
 577         * first is now the position in the packfile where we would insert
 578         * mad->hash if it does not exist (or the position of mad->hash if
 579         * it does exist). Hence, we consider a maximum of two objects
 580         * nearby for the abbreviation length.
 581         */
 582        mad->init_len = 0;
 583        if (!match) {
 584                if (nth_midxed_object_oid(&oid, m, first))
 585                        extend_abbrev_len(&oid, mad);
 586        } else if (first < num - 1) {
 587                if (nth_midxed_object_oid(&oid, m, first + 1))
 588                        extend_abbrev_len(&oid, mad);
 589        }
 590        if (first > 0) {
 591                if (nth_midxed_object_oid(&oid, m, first - 1))
 592                        extend_abbrev_len(&oid, mad);
 593        }
 594        mad->init_len = mad->cur_len;
 595}
 596
 597static void find_abbrev_len_for_pack(struct packed_git *p,
 598                                     struct min_abbrev_data *mad)
 599{
 600        int match = 0;
 601        uint32_t num, first = 0;
 602        struct object_id oid;
 603        const struct object_id *mad_oid;
 604
 605        if (open_pack_index(p) || !p->num_objects)
 606                return;
 607
 608        num = p->num_objects;
 609        mad_oid = mad->oid;
 610        match = bsearch_pack(mad_oid, p, &first);
 611
 612        /*
 613         * first is now the position in the packfile where we would insert
 614         * mad->hash if it does not exist (or the position of mad->hash if
 615         * it does exist). Hence, we consider a maximum of two objects
 616         * nearby for the abbreviation length.
 617         */
 618        mad->init_len = 0;
 619        if (!match) {
 620                if (nth_packed_object_oid(&oid, p, first))
 621                        extend_abbrev_len(&oid, mad);
 622        } else if (first < num - 1) {
 623                if (nth_packed_object_oid(&oid, p, first + 1))
 624                        extend_abbrev_len(&oid, mad);
 625        }
 626        if (first > 0) {
 627                if (nth_packed_object_oid(&oid, p, first - 1))
 628                        extend_abbrev_len(&oid, mad);
 629        }
 630        mad->init_len = mad->cur_len;
 631}
 632
 633static void find_abbrev_len_packed(struct min_abbrev_data *mad)
 634{
 635        struct multi_pack_index *m;
 636        struct packed_git *p;
 637
 638        for (m = get_multi_pack_index(the_repository); m; m = m->next)
 639                find_abbrev_len_for_midx(m, mad);
 640        for (p = get_packed_git(the_repository); p; p = p->next)
 641                find_abbrev_len_for_pack(p, mad);
 642}
 643
 644int find_unique_abbrev_r(char *hex, const struct object_id *oid, int len)
 645{
 646        struct disambiguate_state ds;
 647        struct min_abbrev_data mad;
 648        struct object_id oid_ret;
 649        if (len < 0) {
 650                unsigned long count = approximate_object_count();
 651                /*
 652                 * Add one because the MSB only tells us the highest bit set,
 653                 * not including the value of all the _other_ bits (so "15"
 654                 * is only one off of 2^4, but the MSB is the 3rd bit.
 655                 */
 656                len = msb(count) + 1;
 657                /*
 658                 * We now know we have on the order of 2^len objects, which
 659                 * expects a collision at 2^(len/2). But we also care about hex
 660                 * chars, not bits, and there are 4 bits per hex. So all
 661                 * together we need to divide by 2 and round up.
 662                 */
 663                len = DIV_ROUND_UP(len, 2);
 664                /*
 665                 * For very small repos, we stick with our regular fallback.
 666                 */
 667                if (len < FALLBACK_DEFAULT_ABBREV)
 668                        len = FALLBACK_DEFAULT_ABBREV;
 669        }
 670
 671        oid_to_hex_r(hex, oid);
 672        if (len == GIT_SHA1_HEXSZ || !len)
 673                return GIT_SHA1_HEXSZ;
 674
 675        mad.init_len = len;
 676        mad.cur_len = len;
 677        mad.hex = hex;
 678        mad.oid = oid;
 679
 680        find_abbrev_len_packed(&mad);
 681
 682        if (init_object_disambiguation(hex, mad.cur_len, &ds) < 0)
 683                return -1;
 684
 685        ds.fn = extend_abbrev_len;
 686        ds.always_call_fn = 1;
 687        ds.cb_data = (void *)&mad;
 688
 689        find_short_object_filename(&ds);
 690        (void)finish_object_disambiguation(&ds, &oid_ret);
 691
 692        hex[mad.cur_len] = 0;
 693        return mad.cur_len;
 694}
 695
 696const char *find_unique_abbrev(const struct object_id *oid, int len)
 697{
 698        static int bufno;
 699        static char hexbuffer[4][GIT_MAX_HEXSZ + 1];
 700        char *hex = hexbuffer[bufno];
 701        bufno = (bufno + 1) % ARRAY_SIZE(hexbuffer);
 702        find_unique_abbrev_r(hex, oid, len);
 703        return hex;
 704}
 705
 706static int ambiguous_path(const char *path, int len)
 707{
 708        int slash = 1;
 709        int cnt;
 710
 711        for (cnt = 0; cnt < len; cnt++) {
 712                switch (*path++) {
 713                case '\0':
 714                        break;
 715                case '/':
 716                        if (slash)
 717                                break;
 718                        slash = 1;
 719                        continue;
 720                case '.':
 721                        continue;
 722                default:
 723                        slash = 0;
 724                        continue;
 725                }
 726                break;
 727        }
 728        return slash;
 729}
 730
 731static inline int at_mark(const char *string, int len,
 732                          const char **suffix, int nr)
 733{
 734        int i;
 735
 736        for (i = 0; i < nr; i++) {
 737                int suffix_len = strlen(suffix[i]);
 738                if (suffix_len <= len
 739                    && !strncasecmp(string, suffix[i], suffix_len))
 740                        return suffix_len;
 741        }
 742        return 0;
 743}
 744
 745static inline int upstream_mark(const char *string, int len)
 746{
 747        const char *suffix[] = { "@{upstream}", "@{u}" };
 748        return at_mark(string, len, suffix, ARRAY_SIZE(suffix));
 749}
 750
 751static inline int push_mark(const char *string, int len)
 752{
 753        const char *suffix[] = { "@{push}" };
 754        return at_mark(string, len, suffix, ARRAY_SIZE(suffix));
 755}
 756
 757static int get_oid_1(const char *name, int len, struct object_id *oid, unsigned lookup_flags);
 758static int interpret_nth_prior_checkout(const char *name, int namelen, struct strbuf *buf);
 759
 760static int get_oid_basic(const char *str, int len, struct object_id *oid,
 761                          unsigned int flags)
 762{
 763        static const char *warn_msg = "refname '%.*s' is ambiguous.";
 764        static const char *object_name_msg = N_(
 765        "Git normally never creates a ref that ends with 40 hex characters\n"
 766        "because it will be ignored when you just specify 40-hex. These refs\n"
 767        "may be created by mistake. For example,\n"
 768        "\n"
 769        "  git checkout -b $br $(git rev-parse ...)\n"
 770        "\n"
 771        "where \"$br\" is somehow empty and a 40-hex ref is created. Please\n"
 772        "examine these refs and maybe delete them. Turn this message off by\n"
 773        "running \"git config advice.objectNameWarning false\"");
 774        struct object_id tmp_oid;
 775        char *real_ref = NULL;
 776        int refs_found = 0;
 777        int at, reflog_len, nth_prior = 0;
 778
 779        if (len == GIT_SHA1_HEXSZ && !get_oid_hex(str, oid)) {
 780                if (warn_ambiguous_refs && warn_on_object_refname_ambiguity) {
 781                        refs_found = dwim_ref(str, len, &tmp_oid, &real_ref);
 782                        if (refs_found > 0) {
 783                                warning(warn_msg, len, str);
 784                                if (advice_object_name_warning)
 785                                        fprintf(stderr, "%s\n", _(object_name_msg));
 786                        }
 787                        free(real_ref);
 788                }
 789                return 0;
 790        }
 791
 792        /* basic@{time or number or -number} format to query ref-log */
 793        reflog_len = at = 0;
 794        if (len && str[len-1] == '}') {
 795                for (at = len-4; at >= 0; at--) {
 796                        if (str[at] == '@' && str[at+1] == '{') {
 797                                if (str[at+2] == '-') {
 798                                        if (at != 0)
 799                                                /* @{-N} not at start */
 800                                                return -1;
 801                                        nth_prior = 1;
 802                                        continue;
 803                                }
 804                                if (!upstream_mark(str + at, len - at) &&
 805                                    !push_mark(str + at, len - at)) {
 806                                        reflog_len = (len-1) - (at+2);
 807                                        len = at;
 808                                }
 809                                break;
 810                        }
 811                }
 812        }
 813
 814        /* Accept only unambiguous ref paths. */
 815        if (len && ambiguous_path(str, len))
 816                return -1;
 817
 818        if (nth_prior) {
 819                struct strbuf buf = STRBUF_INIT;
 820                int detached;
 821
 822                if (interpret_nth_prior_checkout(str, len, &buf) > 0) {
 823                        detached = (buf.len == GIT_SHA1_HEXSZ && !get_oid_hex(buf.buf, oid));
 824                        strbuf_release(&buf);
 825                        if (detached)
 826                                return 0;
 827                }
 828        }
 829
 830        if (!len && reflog_len)
 831                /* allow "@{...}" to mean the current branch reflog */
 832                refs_found = dwim_ref("HEAD", 4, oid, &real_ref);
 833        else if (reflog_len)
 834                refs_found = dwim_log(str, len, oid, &real_ref);
 835        else
 836                refs_found = dwim_ref(str, len, oid, &real_ref);
 837
 838        if (!refs_found)
 839                return -1;
 840
 841        if (warn_ambiguous_refs && !(flags & GET_OID_QUIETLY) &&
 842            (refs_found > 1 ||
 843             !get_short_oid(str, len, &tmp_oid, GET_OID_QUIETLY)))
 844                warning(warn_msg, len, str);
 845
 846        if (reflog_len) {
 847                int nth, i;
 848                timestamp_t at_time;
 849                timestamp_t co_time;
 850                int co_tz, co_cnt;
 851
 852                /* Is it asking for N-th entry, or approxidate? */
 853                for (i = nth = 0; 0 <= nth && i < reflog_len; i++) {
 854                        char ch = str[at+2+i];
 855                        if ('0' <= ch && ch <= '9')
 856                                nth = nth * 10 + ch - '0';
 857                        else
 858                                nth = -1;
 859                }
 860                if (100000000 <= nth) {
 861                        at_time = nth;
 862                        nth = -1;
 863                } else if (0 <= nth)
 864                        at_time = 0;
 865                else {
 866                        int errors = 0;
 867                        char *tmp = xstrndup(str + at + 2, reflog_len);
 868                        at_time = approxidate_careful(tmp, &errors);
 869                        free(tmp);
 870                        if (errors) {
 871                                free(real_ref);
 872                                return -1;
 873                        }
 874                }
 875                if (read_ref_at(real_ref, flags, at_time, nth, oid, NULL,
 876                                &co_time, &co_tz, &co_cnt)) {
 877                        if (!len) {
 878                                if (starts_with(real_ref, "refs/heads/")) {
 879                                        str = real_ref + 11;
 880                                        len = strlen(real_ref + 11);
 881                                } else {
 882                                        /* detached HEAD */
 883                                        str = "HEAD";
 884                                        len = 4;
 885                                }
 886                        }
 887                        if (at_time) {
 888                                if (!(flags & GET_OID_QUIETLY)) {
 889                                        warning("Log for '%.*s' only goes "
 890                                                "back to %s.", len, str,
 891                                                show_date(co_time, co_tz, DATE_MODE(RFC2822)));
 892                                }
 893                        } else {
 894                                if (flags & GET_OID_QUIETLY) {
 895                                        exit(128);
 896                                }
 897                                die("Log for '%.*s' only has %d entries.",
 898                                    len, str, co_cnt);
 899                        }
 900                }
 901        }
 902
 903        free(real_ref);
 904        return 0;
 905}
 906
 907static int get_parent(const char *name, int len,
 908                      struct object_id *result, int idx)
 909{
 910        struct object_id oid;
 911        int ret = get_oid_1(name, len, &oid, GET_OID_COMMITTISH);
 912        struct commit *commit;
 913        struct commit_list *p;
 914
 915        if (ret)
 916                return ret;
 917        commit = lookup_commit_reference(&oid);
 918        if (parse_commit(commit))
 919                return -1;
 920        if (!idx) {
 921                oidcpy(result, &commit->object.oid);
 922                return 0;
 923        }
 924        p = commit->parents;
 925        while (p) {
 926                if (!--idx) {
 927                        oidcpy(result, &p->item->object.oid);
 928                        return 0;
 929                }
 930                p = p->next;
 931        }
 932        return -1;
 933}
 934
 935static int get_nth_ancestor(const char *name, int len,
 936                            struct object_id *result, int generation)
 937{
 938        struct object_id oid;
 939        struct commit *commit;
 940        int ret;
 941
 942        ret = get_oid_1(name, len, &oid, GET_OID_COMMITTISH);
 943        if (ret)
 944                return ret;
 945        commit = lookup_commit_reference(&oid);
 946        if (!commit)
 947                return -1;
 948
 949        while (generation--) {
 950                if (parse_commit(commit) || !commit->parents)
 951                        return -1;
 952                commit = commit->parents->item;
 953        }
 954        oidcpy(result, &commit->object.oid);
 955        return 0;
 956}
 957
 958struct object *peel_to_type(const char *name, int namelen,
 959                            struct object *o, enum object_type expected_type)
 960{
 961        if (name && !namelen)
 962                namelen = strlen(name);
 963        while (1) {
 964                if (!o || (!o->parsed && !parse_object(&o->oid)))
 965                        return NULL;
 966                if (expected_type == OBJ_ANY || o->type == expected_type)
 967                        return o;
 968                if (o->type == OBJ_TAG)
 969                        o = ((struct tag*) o)->tagged;
 970                else if (o->type == OBJ_COMMIT)
 971                        o = &(get_commit_tree(((struct commit *)o))->object);
 972                else {
 973                        if (name)
 974                                error("%.*s: expected %s type, but the object "
 975                                      "dereferences to %s type",
 976                                      namelen, name, type_name(expected_type),
 977                                      type_name(o->type));
 978                        return NULL;
 979                }
 980        }
 981}
 982
 983static int peel_onion(const char *name, int len, struct object_id *oid,
 984                      unsigned lookup_flags)
 985{
 986        struct object_id outer;
 987        const char *sp;
 988        unsigned int expected_type = 0;
 989        struct object *o;
 990
 991        /*
 992         * "ref^{type}" dereferences ref repeatedly until you cannot
 993         * dereference anymore, or you get an object of given type,
 994         * whichever comes first.  "ref^{}" means just dereference
 995         * tags until you get a non-tag.  "ref^0" is a shorthand for
 996         * "ref^{commit}".  "commit^{tree}" could be used to find the
 997         * top-level tree of the given commit.
 998         */
 999        if (len < 4 || name[len-1] != '}')
1000                return -1;
1001
1002        for (sp = name + len - 1; name <= sp; sp--) {
1003                int ch = *sp;
1004                if (ch == '{' && name < sp && sp[-1] == '^')
1005                        break;
1006        }
1007        if (sp <= name)
1008                return -1;
1009
1010        sp++; /* beginning of type name, or closing brace for empty */
1011        if (starts_with(sp, "commit}"))
1012                expected_type = OBJ_COMMIT;
1013        else if (starts_with(sp, "tag}"))
1014                expected_type = OBJ_TAG;
1015        else if (starts_with(sp, "tree}"))
1016                expected_type = OBJ_TREE;
1017        else if (starts_with(sp, "blob}"))
1018                expected_type = OBJ_BLOB;
1019        else if (starts_with(sp, "object}"))
1020                expected_type = OBJ_ANY;
1021        else if (sp[0] == '}')
1022                expected_type = OBJ_NONE;
1023        else if (sp[0] == '/')
1024                expected_type = OBJ_COMMIT;
1025        else
1026                return -1;
1027
1028        lookup_flags &= ~GET_OID_DISAMBIGUATORS;
1029        if (expected_type == OBJ_COMMIT)
1030                lookup_flags |= GET_OID_COMMITTISH;
1031        else if (expected_type == OBJ_TREE)
1032                lookup_flags |= GET_OID_TREEISH;
1033
1034        if (get_oid_1(name, sp - name - 2, &outer, lookup_flags))
1035                return -1;
1036
1037        o = parse_object(&outer);
1038        if (!o)
1039                return -1;
1040        if (!expected_type) {
1041                o = deref_tag(o, name, sp - name - 2);
1042                if (!o || (!o->parsed && !parse_object(&o->oid)))
1043                        return -1;
1044                oidcpy(oid, &o->oid);
1045                return 0;
1046        }
1047
1048        /*
1049         * At this point, the syntax look correct, so
1050         * if we do not get the needed object, we should
1051         * barf.
1052         */
1053        o = peel_to_type(name, len, o, expected_type);
1054        if (!o)
1055                return -1;
1056
1057        oidcpy(oid, &o->oid);
1058        if (sp[0] == '/') {
1059                /* "$commit^{/foo}" */
1060                char *prefix;
1061                int ret;
1062                struct commit_list *list = NULL;
1063
1064                /*
1065                 * $commit^{/}. Some regex implementation may reject.
1066                 * We don't need regex anyway. '' pattern always matches.
1067                 */
1068                if (sp[1] == '}')
1069                        return 0;
1070
1071                prefix = xstrndup(sp + 1, name + len - 1 - (sp + 1));
1072                commit_list_insert((struct commit *)o, &list);
1073                ret = get_oid_oneline(prefix, oid, list);
1074                free(prefix);
1075                return ret;
1076        }
1077        return 0;
1078}
1079
1080static int get_describe_name(const char *name, int len, struct object_id *oid)
1081{
1082        const char *cp;
1083        unsigned flags = GET_OID_QUIETLY | GET_OID_COMMIT;
1084
1085        for (cp = name + len - 1; name + 2 <= cp; cp--) {
1086                char ch = *cp;
1087                if (!isxdigit(ch)) {
1088                        /* We must be looking at g in "SOMETHING-g"
1089                         * for it to be describe output.
1090                         */
1091                        if (ch == 'g' && cp[-1] == '-') {
1092                                cp++;
1093                                len -= cp - name;
1094                                return get_short_oid(cp, len, oid, flags);
1095                        }
1096                }
1097        }
1098        return -1;
1099}
1100
1101static int get_oid_1(const char *name, int len, struct object_id *oid, unsigned lookup_flags)
1102{
1103        int ret, has_suffix;
1104        const char *cp;
1105
1106        /*
1107         * "name~3" is "name^^^", "name~" is "name~1", and "name^" is "name^1".
1108         */
1109        has_suffix = 0;
1110        for (cp = name + len - 1; name <= cp; cp--) {
1111                int ch = *cp;
1112                if ('0' <= ch && ch <= '9')
1113                        continue;
1114                if (ch == '~' || ch == '^')
1115                        has_suffix = ch;
1116                break;
1117        }
1118
1119        if (has_suffix) {
1120                int num = 0;
1121                int len1 = cp - name;
1122                cp++;
1123                while (cp < name + len)
1124                        num = num * 10 + *cp++ - '0';
1125                if (!num && len1 == len - 1)
1126                        num = 1;
1127                if (has_suffix == '^')
1128                        return get_parent(name, len1, oid, num);
1129                /* else if (has_suffix == '~') -- goes without saying */
1130                return get_nth_ancestor(name, len1, oid, num);
1131        }
1132
1133        ret = peel_onion(name, len, oid, lookup_flags);
1134        if (!ret)
1135                return 0;
1136
1137        ret = get_oid_basic(name, len, oid, lookup_flags);
1138        if (!ret)
1139                return 0;
1140
1141        /* It could be describe output that is "SOMETHING-gXXXX" */
1142        ret = get_describe_name(name, len, oid);
1143        if (!ret)
1144                return 0;
1145
1146        return get_short_oid(name, len, oid, lookup_flags);
1147}
1148
1149/*
1150 * This interprets names like ':/Initial revision of "git"' by searching
1151 * through history and returning the first commit whose message starts
1152 * the given regular expression.
1153 *
1154 * For negative-matching, prefix the pattern-part with '!-', like: ':/!-WIP'.
1155 *
1156 * For a literal '!' character at the beginning of a pattern, you have to repeat
1157 * that, like: ':/!!foo'
1158 *
1159 * For future extension, all other sequences beginning with ':/!' are reserved.
1160 */
1161
1162/* Remember to update object flag allocation in object.h */
1163#define ONELINE_SEEN (1u<<20)
1164
1165static int handle_one_ref(const char *path, const struct object_id *oid,
1166                          int flag, void *cb_data)
1167{
1168        struct commit_list **list = cb_data;
1169        struct object *object = parse_object(oid);
1170        if (!object)
1171                return 0;
1172        if (object->type == OBJ_TAG) {
1173                object = deref_tag(object, path, strlen(path));
1174                if (!object)
1175                        return 0;
1176        }
1177        if (object->type != OBJ_COMMIT)
1178                return 0;
1179        commit_list_insert((struct commit *)object, list);
1180        return 0;
1181}
1182
1183static int get_oid_oneline(const char *prefix, struct object_id *oid,
1184                            struct commit_list *list)
1185{
1186        struct commit_list *backup = NULL, *l;
1187        int found = 0;
1188        int negative = 0;
1189        regex_t regex;
1190
1191        if (prefix[0] == '!') {
1192                prefix++;
1193
1194                if (prefix[0] == '-') {
1195                        prefix++;
1196                        negative = 1;
1197                } else if (prefix[0] != '!') {
1198                        return -1;
1199                }
1200        }
1201
1202        if (regcomp(&regex, prefix, REG_EXTENDED))
1203                return -1;
1204
1205        for (l = list; l; l = l->next) {
1206                l->item->object.flags |= ONELINE_SEEN;
1207                commit_list_insert(l->item, &backup);
1208        }
1209        while (list) {
1210                const char *p, *buf;
1211                struct commit *commit;
1212                int matches;
1213
1214                commit = pop_most_recent_commit(&list, ONELINE_SEEN);
1215                if (!parse_object(&commit->object.oid))
1216                        continue;
1217                buf = get_commit_buffer(commit, NULL);
1218                p = strstr(buf, "\n\n");
1219                matches = negative ^ (p && !regexec(&regex, p + 2, 0, NULL, 0));
1220                unuse_commit_buffer(commit, buf);
1221
1222                if (matches) {
1223                        oidcpy(oid, &commit->object.oid);
1224                        found = 1;
1225                        break;
1226                }
1227        }
1228        regfree(&regex);
1229        free_commit_list(list);
1230        for (l = backup; l; l = l->next)
1231                clear_commit_marks(l->item, ONELINE_SEEN);
1232        free_commit_list(backup);
1233        return found ? 0 : -1;
1234}
1235
1236struct grab_nth_branch_switch_cbdata {
1237        int remaining;
1238        struct strbuf buf;
1239};
1240
1241static int grab_nth_branch_switch(struct object_id *ooid, struct object_id *noid,
1242                                  const char *email, timestamp_t timestamp, int tz,
1243                                  const char *message, void *cb_data)
1244{
1245        struct grab_nth_branch_switch_cbdata *cb = cb_data;
1246        const char *match = NULL, *target = NULL;
1247        size_t len;
1248
1249        if (skip_prefix(message, "checkout: moving from ", &match))
1250                target = strstr(match, " to ");
1251
1252        if (!match || !target)
1253                return 0;
1254        if (--(cb->remaining) == 0) {
1255                len = target - match;
1256                strbuf_reset(&cb->buf);
1257                strbuf_add(&cb->buf, match, len);
1258                return 1; /* we are done */
1259        }
1260        return 0;
1261}
1262
1263/*
1264 * Parse @{-N} syntax, return the number of characters parsed
1265 * if successful; otherwise signal an error with negative value.
1266 */
1267static int interpret_nth_prior_checkout(const char *name, int namelen,
1268                                        struct strbuf *buf)
1269{
1270        long nth;
1271        int retval;
1272        struct grab_nth_branch_switch_cbdata cb;
1273        const char *brace;
1274        char *num_end;
1275
1276        if (namelen < 4)
1277                return -1;
1278        if (name[0] != '@' || name[1] != '{' || name[2] != '-')
1279                return -1;
1280        brace = memchr(name, '}', namelen);
1281        if (!brace)
1282                return -1;
1283        nth = strtol(name + 3, &num_end, 10);
1284        if (num_end != brace)
1285                return -1;
1286        if (nth <= 0)
1287                return -1;
1288        cb.remaining = nth;
1289        strbuf_init(&cb.buf, 20);
1290
1291        retval = 0;
1292        if (0 < for_each_reflog_ent_reverse("HEAD", grab_nth_branch_switch, &cb)) {
1293                strbuf_reset(buf);
1294                strbuf_addbuf(buf, &cb.buf);
1295                retval = brace - name + 1;
1296        }
1297
1298        strbuf_release(&cb.buf);
1299        return retval;
1300}
1301
1302int get_oid_mb(const char *name, struct object_id *oid)
1303{
1304        struct commit *one, *two;
1305        struct commit_list *mbs;
1306        struct object_id oid_tmp;
1307        const char *dots;
1308        int st;
1309
1310        dots = strstr(name, "...");
1311        if (!dots)
1312                return get_oid(name, oid);
1313        if (dots == name)
1314                st = get_oid("HEAD", &oid_tmp);
1315        else {
1316                struct strbuf sb;
1317                strbuf_init(&sb, dots - name);
1318                strbuf_add(&sb, name, dots - name);
1319                st = get_oid_committish(sb.buf, &oid_tmp);
1320                strbuf_release(&sb);
1321        }
1322        if (st)
1323                return st;
1324        one = lookup_commit_reference_gently(&oid_tmp, 0);
1325        if (!one)
1326                return -1;
1327
1328        if (get_oid_committish(dots[3] ? (dots + 3) : "HEAD", &oid_tmp))
1329                return -1;
1330        two = lookup_commit_reference_gently(&oid_tmp, 0);
1331        if (!two)
1332                return -1;
1333        mbs = get_merge_bases(one, two);
1334        if (!mbs || mbs->next)
1335                st = -1;
1336        else {
1337                st = 0;
1338                oidcpy(oid, &mbs->item->object.oid);
1339        }
1340        free_commit_list(mbs);
1341        return st;
1342}
1343
1344/* parse @something syntax, when 'something' is not {.*} */
1345static int interpret_empty_at(const char *name, int namelen, int len, struct strbuf *buf)
1346{
1347        const char *next;
1348
1349        if (len || name[1] == '{')
1350                return -1;
1351
1352        /* make sure it's a single @, or @@{.*}, not @foo */
1353        next = memchr(name + len + 1, '@', namelen - len - 1);
1354        if (next && next[1] != '{')
1355                return -1;
1356        if (!next)
1357                next = name + namelen;
1358        if (next != name + 1)
1359                return -1;
1360
1361        strbuf_reset(buf);
1362        strbuf_add(buf, "HEAD", 4);
1363        return 1;
1364}
1365
1366static int reinterpret(const char *name, int namelen, int len,
1367                       struct strbuf *buf, unsigned allowed)
1368{
1369        /* we have extra data, which might need further processing */
1370        struct strbuf tmp = STRBUF_INIT;
1371        int used = buf->len;
1372        int ret;
1373
1374        strbuf_add(buf, name + len, namelen - len);
1375        ret = interpret_branch_name(buf->buf, buf->len, &tmp, allowed);
1376        /* that data was not interpreted, remove our cruft */
1377        if (ret < 0) {
1378                strbuf_setlen(buf, used);
1379                return len;
1380        }
1381        strbuf_reset(buf);
1382        strbuf_addbuf(buf, &tmp);
1383        strbuf_release(&tmp);
1384        /* tweak for size of {-N} versus expanded ref name */
1385        return ret - used + len;
1386}
1387
1388static void set_shortened_ref(struct strbuf *buf, const char *ref)
1389{
1390        char *s = shorten_unambiguous_ref(ref, 0);
1391        strbuf_reset(buf);
1392        strbuf_addstr(buf, s);
1393        free(s);
1394}
1395
1396static int branch_interpret_allowed(const char *refname, unsigned allowed)
1397{
1398        if (!allowed)
1399                return 1;
1400
1401        if ((allowed & INTERPRET_BRANCH_LOCAL) &&
1402            starts_with(refname, "refs/heads/"))
1403                return 1;
1404        if ((allowed & INTERPRET_BRANCH_REMOTE) &&
1405            starts_with(refname, "refs/remotes/"))
1406                return 1;
1407
1408        return 0;
1409}
1410
1411static int interpret_branch_mark(const char *name, int namelen,
1412                                 int at, struct strbuf *buf,
1413                                 int (*get_mark)(const char *, int),
1414                                 const char *(*get_data)(struct branch *,
1415                                                         struct strbuf *),
1416                                 unsigned allowed)
1417{
1418        int len;
1419        struct branch *branch;
1420        struct strbuf err = STRBUF_INIT;
1421        const char *value;
1422
1423        len = get_mark(name + at, namelen - at);
1424        if (!len)
1425                return -1;
1426
1427        if (memchr(name, ':', at))
1428                return -1;
1429
1430        if (at) {
1431                char *name_str = xmemdupz(name, at);
1432                branch = branch_get(name_str);
1433                free(name_str);
1434        } else
1435                branch = branch_get(NULL);
1436
1437        value = get_data(branch, &err);
1438        if (!value)
1439                die("%s", err.buf);
1440
1441        if (!branch_interpret_allowed(value, allowed))
1442                return -1;
1443
1444        set_shortened_ref(buf, value);
1445        return len + at;
1446}
1447
1448int interpret_branch_name(const char *name, int namelen, struct strbuf *buf,
1449                          unsigned allowed)
1450{
1451        char *at;
1452        const char *start;
1453        int len;
1454
1455        if (!namelen)
1456                namelen = strlen(name);
1457
1458        if (!allowed || (allowed & INTERPRET_BRANCH_LOCAL)) {
1459                len = interpret_nth_prior_checkout(name, namelen, buf);
1460                if (!len) {
1461                        return len; /* syntax Ok, not enough switches */
1462                } else if (len > 0) {
1463                        if (len == namelen)
1464                                return len; /* consumed all */
1465                        else
1466                                return reinterpret(name, namelen, len, buf, allowed);
1467                }
1468        }
1469
1470        for (start = name;
1471             (at = memchr(start, '@', namelen - (start - name)));
1472             start = at + 1) {
1473
1474                if (!allowed || (allowed & INTERPRET_BRANCH_HEAD)) {
1475                        len = interpret_empty_at(name, namelen, at - name, buf);
1476                        if (len > 0)
1477                                return reinterpret(name, namelen, len, buf,
1478                                                   allowed);
1479                }
1480
1481                len = interpret_branch_mark(name, namelen, at - name, buf,
1482                                            upstream_mark, branch_get_upstream,
1483                                            allowed);
1484                if (len > 0)
1485                        return len;
1486
1487                len = interpret_branch_mark(name, namelen, at - name, buf,
1488                                            push_mark, branch_get_push,
1489                                            allowed);
1490                if (len > 0)
1491                        return len;
1492        }
1493
1494        return -1;
1495}
1496
1497void strbuf_branchname(struct strbuf *sb, const char *name, unsigned allowed)
1498{
1499        int len = strlen(name);
1500        int used = interpret_branch_name(name, len, sb, allowed);
1501
1502        if (used < 0)
1503                used = 0;
1504        strbuf_add(sb, name + used, len - used);
1505}
1506
1507int strbuf_check_branch_ref(struct strbuf *sb, const char *name)
1508{
1509        if (startup_info->have_repository)
1510                strbuf_branchname(sb, name, INTERPRET_BRANCH_LOCAL);
1511        else
1512                strbuf_addstr(sb, name);
1513
1514        /*
1515         * This splice must be done even if we end up rejecting the
1516         * name; builtin/branch.c::copy_or_rename_branch() still wants
1517         * to see what the name expanded to so that "branch -m" can be
1518         * used as a tool to correct earlier mistakes.
1519         */
1520        strbuf_splice(sb, 0, 0, "refs/heads/", 11);
1521
1522        if (*name == '-' ||
1523            !strcmp(sb->buf, "refs/heads/HEAD"))
1524                return -1;
1525
1526        return check_refname_format(sb->buf, 0);
1527}
1528
1529/*
1530 * This is like "get_oid_basic()", except it allows "object ID expressions",
1531 * notably "xyz^" for "parent of xyz"
1532 */
1533int get_oid(const char *name, struct object_id *oid)
1534{
1535        struct object_context unused;
1536        return get_oid_with_context(name, 0, oid, &unused);
1537}
1538
1539
1540/*
1541 * Many callers know that the user meant to name a commit-ish by
1542 * syntactical positions where the object name appears.  Calling this
1543 * function allows the machinery to disambiguate shorter-than-unique
1544 * abbreviated object names between commit-ish and others.
1545 *
1546 * Note that this does NOT error out when the named object is not a
1547 * commit-ish. It is merely to give a hint to the disambiguation
1548 * machinery.
1549 */
1550int get_oid_committish(const char *name, struct object_id *oid)
1551{
1552        struct object_context unused;
1553        return get_oid_with_context(name, GET_OID_COMMITTISH,
1554                                    oid, &unused);
1555}
1556
1557int get_oid_treeish(const char *name, struct object_id *oid)
1558{
1559        struct object_context unused;
1560        return get_oid_with_context(name, GET_OID_TREEISH,
1561                                    oid, &unused);
1562}
1563
1564int get_oid_commit(const char *name, struct object_id *oid)
1565{
1566        struct object_context unused;
1567        return get_oid_with_context(name, GET_OID_COMMIT,
1568                                    oid, &unused);
1569}
1570
1571int get_oid_tree(const char *name, struct object_id *oid)
1572{
1573        struct object_context unused;
1574        return get_oid_with_context(name, GET_OID_TREE,
1575                                    oid, &unused);
1576}
1577
1578int get_oid_blob(const char *name, struct object_id *oid)
1579{
1580        struct object_context unused;
1581        return get_oid_with_context(name, GET_OID_BLOB,
1582                                    oid, &unused);
1583}
1584
1585/* Must be called only when object_name:filename doesn't exist. */
1586static void diagnose_invalid_oid_path(const char *prefix,
1587                                      const char *filename,
1588                                      const struct object_id *tree_oid,
1589                                      const char *object_name,
1590                                      int object_name_len)
1591{
1592        struct object_id oid;
1593        unsigned mode;
1594
1595        if (!prefix)
1596                prefix = "";
1597
1598        if (file_exists(filename))
1599                die("Path '%s' exists on disk, but not in '%.*s'.",
1600                    filename, object_name_len, object_name);
1601        if (is_missing_file_error(errno)) {
1602                char *fullname = xstrfmt("%s%s", prefix, filename);
1603
1604                if (!get_tree_entry(tree_oid, fullname, &oid, &mode)) {
1605                        die("Path '%s' exists, but not '%s'.\n"
1606                            "Did you mean '%.*s:%s' aka '%.*s:./%s'?",
1607                            fullname,
1608                            filename,
1609                            object_name_len, object_name,
1610                            fullname,
1611                            object_name_len, object_name,
1612                            filename);
1613                }
1614                die("Path '%s' does not exist in '%.*s'",
1615                    filename, object_name_len, object_name);
1616        }
1617}
1618
1619/* Must be called only when :stage:filename doesn't exist. */
1620static void diagnose_invalid_index_path(int stage,
1621                                        const char *prefix,
1622                                        const char *filename)
1623{
1624        const struct cache_entry *ce;
1625        int pos;
1626        unsigned namelen = strlen(filename);
1627        struct strbuf fullname = STRBUF_INIT;
1628
1629        if (!prefix)
1630                prefix = "";
1631
1632        /* Wrong stage number? */
1633        pos = cache_name_pos(filename, namelen);
1634        if (pos < 0)
1635                pos = -pos - 1;
1636        if (pos < active_nr) {
1637                ce = active_cache[pos];
1638                if (ce_namelen(ce) == namelen &&
1639                    !memcmp(ce->name, filename, namelen))
1640                        die("Path '%s' is in the index, but not at stage %d.\n"
1641                            "Did you mean ':%d:%s'?",
1642                            filename, stage,
1643                            ce_stage(ce), filename);
1644        }
1645
1646        /* Confusion between relative and absolute filenames? */
1647        strbuf_addstr(&fullname, prefix);
1648        strbuf_addstr(&fullname, filename);
1649        pos = cache_name_pos(fullname.buf, fullname.len);
1650        if (pos < 0)
1651                pos = -pos - 1;
1652        if (pos < active_nr) {
1653                ce = active_cache[pos];
1654                if (ce_namelen(ce) == fullname.len &&
1655                    !memcmp(ce->name, fullname.buf, fullname.len))
1656                        die("Path '%s' is in the index, but not '%s'.\n"
1657                            "Did you mean ':%d:%s' aka ':%d:./%s'?",
1658                            fullname.buf, filename,
1659                            ce_stage(ce), fullname.buf,
1660                            ce_stage(ce), filename);
1661        }
1662
1663        if (file_exists(filename))
1664                die("Path '%s' exists on disk, but not in the index.", filename);
1665        if (is_missing_file_error(errno))
1666                die("Path '%s' does not exist (neither on disk nor in the index).",
1667                    filename);
1668
1669        strbuf_release(&fullname);
1670}
1671
1672
1673static char *resolve_relative_path(const char *rel)
1674{
1675        if (!starts_with(rel, "./") && !starts_with(rel, "../"))
1676                return NULL;
1677
1678        if (!is_inside_work_tree())
1679                die("relative path syntax can't be used outside working tree.");
1680
1681        /* die() inside prefix_path() if resolved path is outside worktree */
1682        return prefix_path(startup_info->prefix,
1683                           startup_info->prefix ? strlen(startup_info->prefix) : 0,
1684                           rel);
1685}
1686
1687static int get_oid_with_context_1(const char *name,
1688                                  unsigned flags,
1689                                  const char *prefix,
1690                                  struct object_id *oid,
1691                                  struct object_context *oc)
1692{
1693        int ret, bracket_depth;
1694        int namelen = strlen(name);
1695        const char *cp;
1696        int only_to_die = flags & GET_OID_ONLY_TO_DIE;
1697
1698        if (only_to_die)
1699                flags |= GET_OID_QUIETLY;
1700
1701        memset(oc, 0, sizeof(*oc));
1702        oc->mode = S_IFINVALID;
1703        strbuf_init(&oc->symlink_path, 0);
1704        ret = get_oid_1(name, namelen, oid, flags);
1705        if (!ret)
1706                return ret;
1707        /*
1708         * sha1:path --> object name of path in ent sha1
1709         * :path -> object name of absolute path in index
1710         * :./path -> object name of path relative to cwd in index
1711         * :[0-3]:path -> object name of path in index at stage
1712         * :/foo -> recent commit matching foo
1713         */
1714        if (name[0] == ':') {
1715                int stage = 0;
1716                const struct cache_entry *ce;
1717                char *new_path = NULL;
1718                int pos;
1719                if (!only_to_die && namelen > 2 && name[1] == '/') {
1720                        struct commit_list *list = NULL;
1721
1722                        for_each_ref(handle_one_ref, &list);
1723                        commit_list_sort_by_date(&list);
1724                        return get_oid_oneline(name + 2, oid, list);
1725                }
1726                if (namelen < 3 ||
1727                    name[2] != ':' ||
1728                    name[1] < '0' || '3' < name[1])
1729                        cp = name + 1;
1730                else {
1731                        stage = name[1] - '0';
1732                        cp = name + 3;
1733                }
1734                new_path = resolve_relative_path(cp);
1735                if (!new_path) {
1736                        namelen = namelen - (cp - name);
1737                } else {
1738                        cp = new_path;
1739                        namelen = strlen(cp);
1740                }
1741
1742                if (flags & GET_OID_RECORD_PATH)
1743                        oc->path = xstrdup(cp);
1744
1745                if (!active_cache)
1746                        read_cache();
1747                pos = cache_name_pos(cp, namelen);
1748                if (pos < 0)
1749                        pos = -pos - 1;
1750                while (pos < active_nr) {
1751                        ce = active_cache[pos];
1752                        if (ce_namelen(ce) != namelen ||
1753                            memcmp(ce->name, cp, namelen))
1754                                break;
1755                        if (ce_stage(ce) == stage) {
1756                                oidcpy(oid, &ce->oid);
1757                                oc->mode = ce->ce_mode;
1758                                free(new_path);
1759                                return 0;
1760                        }
1761                        pos++;
1762                }
1763                if (only_to_die && name[1] && name[1] != '/')
1764                        diagnose_invalid_index_path(stage, prefix, cp);
1765                free(new_path);
1766                return -1;
1767        }
1768        for (cp = name, bracket_depth = 0; *cp; cp++) {
1769                if (*cp == '{')
1770                        bracket_depth++;
1771                else if (bracket_depth && *cp == '}')
1772                        bracket_depth--;
1773                else if (!bracket_depth && *cp == ':')
1774                        break;
1775        }
1776        if (*cp == ':') {
1777                struct object_id tree_oid;
1778                int len = cp - name;
1779                unsigned sub_flags = flags;
1780
1781                sub_flags &= ~GET_OID_DISAMBIGUATORS;
1782                sub_flags |= GET_OID_TREEISH;
1783
1784                if (!get_oid_1(name, len, &tree_oid, sub_flags)) {
1785                        const char *filename = cp+1;
1786                        char *new_filename = NULL;
1787
1788                        new_filename = resolve_relative_path(filename);
1789                        if (new_filename)
1790                                filename = new_filename;
1791                        if (flags & GET_OID_FOLLOW_SYMLINKS) {
1792                                ret = get_tree_entry_follow_symlinks(&tree_oid,
1793                                        filename, oid, &oc->symlink_path,
1794                                        &oc->mode);
1795                        } else {
1796                                ret = get_tree_entry(&tree_oid, filename, oid,
1797                                                     &oc->mode);
1798                                if (ret && only_to_die) {
1799                                        diagnose_invalid_oid_path(prefix,
1800                                                                   filename,
1801                                                                   &tree_oid,
1802                                                                   name, len);
1803                                }
1804                        }
1805                        if (flags & GET_OID_RECORD_PATH)
1806                                oc->path = xstrdup(filename);
1807
1808                        free(new_filename);
1809                        return ret;
1810                } else {
1811                        if (only_to_die)
1812                                die("Invalid object name '%.*s'.", len, name);
1813                }
1814        }
1815        return ret;
1816}
1817
1818/*
1819 * Call this function when you know "name" given by the end user must
1820 * name an object but it doesn't; the function _may_ die with a better
1821 * diagnostic message than "no such object 'name'", e.g. "Path 'doc' does not
1822 * exist in 'HEAD'" when given "HEAD:doc", or it may return in which case
1823 * you have a chance to diagnose the error further.
1824 */
1825void maybe_die_on_misspelt_object_name(const char *name, const char *prefix)
1826{
1827        struct object_context oc;
1828        struct object_id oid;
1829        get_oid_with_context_1(name, GET_OID_ONLY_TO_DIE, prefix, &oid, &oc);
1830}
1831
1832int get_oid_with_context(const char *str, unsigned flags, struct object_id *oid, struct object_context *oc)
1833{
1834        if (flags & GET_OID_FOLLOW_SYMLINKS && flags & GET_OID_ONLY_TO_DIE)
1835                BUG("incompatible flags for get_sha1_with_context");
1836        return get_oid_with_context_1(str, flags, NULL, oid, oc);
1837}