git.oblomov.eu Git - git/blob - sha1-name.c

   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 "submodule.h"
  16 #include "midx.h"
  17 #include "commit-reach.h"
  18
  19 static int get_oid_oneline(struct repository *r, const char *, struct object_id *, struct commit_list *);
  20
  21 typedef int (*disambiguate_hint_fn)(struct repository *, const struct object_id *, void *);
  22
  23 struct disambiguate_state {
  24         int len; /* length of prefix in hex chars */
  25         char hex_pfx[GIT_MAX_HEXSZ + 1];
  26         struct object_id bin_pfx;
  27
  28         struct repository *repo;
  29         disambiguate_hint_fn fn;
  30         void *cb_data;
  31         struct object_id candidate;
  32         unsigned candidate_exists:1;
  33         unsigned candidate_checked:1;
  34         unsigned candidate_ok:1;
  35         unsigned disambiguate_fn_used:1;
  36         unsigned ambiguous:1;
  37         unsigned always_call_fn:1;
  38 };
  39
  40 static void update_candidates(struct disambiguate_state *ds, const struct object_id *current)
  41 {
  42         if (ds->always_call_fn) {
  43                 ds->ambiguous = ds->fn(ds->repo, current, ds->cb_data) ? 1 : 0;
  44                 return;
  45         }
  46         if (!ds->candidate_exists) {
  47                 /* this is the first candidate */
  48                 oidcpy(&ds->candidate, current);
  49                 ds->candidate_exists = 1;
  50                 return;
  51         } else if (oideq(&ds->candidate, current)) {
  52                 /* the same as what we already have seen */
  53                 return;
  54         }
  55
  56         if (!ds->fn) {
  57                 /* cannot disambiguate between ds->candidate and current */
  58                 ds->ambiguous = 1;
  59                 return;
  60         }
  61
  62         if (!ds->candidate_checked) {
  63                 ds->candidate_ok = ds->fn(ds->repo, &ds->candidate, ds->cb_data);
  64                 ds->disambiguate_fn_used = 1;
  65                 ds->candidate_checked = 1;
  66         }
  67
  68         if (!ds->candidate_ok) {
  69                 /* discard the candidate; we know it does not satisfy fn */
  70                 oidcpy(&ds->candidate, current);
  71                 ds->candidate_checked = 0;
  72                 return;
  73         }
  74
  75         /* if we reach this point, we know ds->candidate satisfies fn */
  76         if (ds->fn(ds->repo, current, ds->cb_data)) {
  77                 /*
  78                  * if both current and candidate satisfy fn, we cannot
  79                  * disambiguate.
  80                  */
  81                 ds->candidate_ok = 0;
  82                 ds->ambiguous = 1;
  83         }
  84
  85         /* otherwise, current can be discarded and candidate is still good */
  86 }
  87
  88 static int match_sha(unsigned, const unsigned char *, const unsigned char *);
  89
  90 static void find_short_object_filename(struct disambiguate_state *ds)
  91 {
  92         struct object_directory *odb;
  93
  94         for (odb = ds->repo->objects->odb; odb && !ds->ambiguous; odb = odb->next) {
  95                 int pos;
  96                 struct oid_array *loose_objects;
  97
  98                 loose_objects = odb_loose_cache(odb, &ds->bin_pfx);
  99                 pos = oid_array_lookup(loose_objects, &ds->bin_pfx);
 100                 if (pos < 0)
 101                         pos = -1 - pos;
 102                 while (!ds->ambiguous && pos < loose_objects->nr) {
 103                         const struct object_id *oid;
 104                         oid = loose_objects->oid + pos;
 105                         if (!match_sha(ds->len, ds->bin_pfx.hash, oid->hash))
 106                                 break;
 107                         update_candidates(ds, oid);
 108                         pos++;
 109                 }
 110         }
 111 }
 112
 113 static int match_sha(unsigned len, const unsigned char *a, const unsigned char *b)
 114 {
 115         do {
 116                 if (*a != *b)
 117                         return 0;
 118                 a++;
 119                 b++;
 120                 len -= 2;
 121         } while (len > 1);
 122         if (len)
 123                 if ((*a ^ *b) & 0xf0)
 124                         return 0;
 125         return 1;
 126 }
 127
 128 static void unique_in_midx(struct multi_pack_index *m,
 129                            struct disambiguate_state *ds)
 130 {
 131         uint32_t num, i, first = 0;
 132         const struct object_id *current = NULL;
 133         num = m->num_objects;
 134
 135         if (!num)
 136                 return;
 137
 138         bsearch_midx(&ds->bin_pfx, m, &first);
 139
 140         /*
 141          * At this point, "first" is the location of the lowest object
 142          * with an object name that could match "bin_pfx".  See if we have
 143          * 0, 1 or more objects that actually match(es).
 144          */
 145         for (i = first; i < num && !ds->ambiguous; i++) {
 146                 struct object_id oid;
 147                 current = nth_midxed_object_oid(&oid, m, i);
 148                 if (!match_sha(ds->len, ds->bin_pfx.hash, current->hash))
 149                         break;
 150                 update_candidates(ds, current);
 151         }
 152 }
 153
 154 static void unique_in_pack(struct packed_git *p,
 155                            struct disambiguate_state *ds)
 156 {
 157         uint32_t num, i, first = 0;
 158         const struct object_id *current = NULL;
 159
 160         if (p->multi_pack_index)
 161                 return;
 162
 163         if (open_pack_index(p) || !p->num_objects)
 164                 return;
 165
 166         num = p->num_objects;
 167         bsearch_pack(&ds->bin_pfx, p, &first);
 168
 169         /*
 170          * At this point, "first" is the location of the lowest object
 171          * with an object name that could match "bin_pfx".  See if we have
 172          * 0, 1 or more objects that actually match(es).
 173          */
 174         for (i = first; i < num && !ds->ambiguous; i++) {
 175                 struct object_id oid;
 176                 current = nth_packed_object_oid(&oid, p, i);
 177                 if (!match_sha(ds->len, ds->bin_pfx.hash, current->hash))
 178                         break;
 179                 update_candidates(ds, current);
 180         }
 181 }
 182
 183 static void find_short_packed_object(struct disambiguate_state *ds)
 184 {
 185         struct multi_pack_index *m;
 186         struct packed_git *p;
 187
 188         for (m = get_multi_pack_index(ds->repo); m && !ds->ambiguous;
 189              m = m->next)
 190                 unique_in_midx(m, ds);
 191         for (p = get_packed_git(ds->repo); p && !ds->ambiguous;
 192              p = p->next)
 193                 unique_in_pack(p, ds);
 194 }
 195
 196 static int finish_object_disambiguation(struct disambiguate_state *ds,
 197                                         struct object_id *oid)
 198 {
 199         if (ds->ambiguous)
 200                 return SHORT_NAME_AMBIGUOUS;
 201
 202         if (!ds->candidate_exists)
 203                 return MISSING_OBJECT;
 204
 205         if (!ds->candidate_checked)
 206                 /*
 207                  * If this is the only candidate, there is no point
 208                  * calling the disambiguation hint callback.
 209                  *
 210                  * On the other hand, if the current candidate
 211                  * replaced an earlier candidate that did _not_ pass
 212                  * the disambiguation hint callback, then we do have
 213                  * more than one objects that match the short name
 214                  * given, so we should make sure this one matches;
 215                  * otherwise, if we discovered this one and the one
 216                  * that we previously discarded in the reverse order,
 217                  * we would end up showing different results in the
 218                  * same repository!
 219                  */
 220                 ds->candidate_ok = (!ds->disambiguate_fn_used ||
 221                                     ds->fn(ds->repo, &ds->candidate, ds->cb_data));
 222
 223         if (!ds->candidate_ok)
 224                 return SHORT_NAME_AMBIGUOUS;
 225
 226         oidcpy(oid, &ds->candidate);
 227         return 0;
 228 }
 229
 230 static int disambiguate_commit_only(struct repository *r,
 231                                     const struct object_id *oid,
 232                                     void *cb_data_unused)
 233 {
 234         int kind = oid_object_info(r, oid, NULL);
 235         return kind == OBJ_COMMIT;
 236 }
 237
 238 static int disambiguate_committish_only(struct repository *r,
 239                                         const struct object_id *oid,
 240                                         void *cb_data_unused)
 241 {
 242         struct object *obj;
 243         int kind;
 244
 245         kind = oid_object_info(r, oid, NULL);
 246         if (kind == OBJ_COMMIT)
 247                 return 1;
 248         if (kind != OBJ_TAG)
 249                 return 0;
 250
 251         /* We need to do this the hard way... */
 252         obj = deref_tag(r, parse_object(r, oid), NULL, 0);
 253         if (obj && obj->type == OBJ_COMMIT)
 254                 return 1;
 255         return 0;
 256 }
 257
 258 static int disambiguate_tree_only(struct repository *r,
 259                                   const struct object_id *oid,
 260                                   void *cb_data_unused)
 261 {
 262         int kind = oid_object_info(r, oid, NULL);
 263         return kind == OBJ_TREE;
 264 }
 265
 266 static int disambiguate_treeish_only(struct repository *r,
 267                                      const struct object_id *oid,
 268                                      void *cb_data_unused)
 269 {
 270         struct object *obj;
 271         int kind;
 272
 273         kind = oid_object_info(r, oid, NULL);
 274         if (kind == OBJ_TREE || kind == OBJ_COMMIT)
 275                 return 1;
 276         if (kind != OBJ_TAG)
 277                 return 0;
 278
 279         /* We need to do this the hard way... */
 280         obj = deref_tag(r, parse_object(r, oid), NULL, 0);
 281         if (obj && (obj->type == OBJ_TREE || obj->type == OBJ_COMMIT))
 282                 return 1;
 283         return 0;
 284 }
 285
 286 static int disambiguate_blob_only(struct repository *r,
 287                                   const struct object_id *oid,
 288                                   void *cb_data_unused)
 289 {
 290         int kind = oid_object_info(r, oid, NULL);
 291         return kind == OBJ_BLOB;
 292 }
 293
 294 static disambiguate_hint_fn default_disambiguate_hint;
 295
 296 int set_disambiguate_hint_config(const char *var, const char *value)
 297 {
 298         static const struct {
 299                 const char *name;
 300                 disambiguate_hint_fn fn;
 301         } hints[] = {
 302                 { "none", NULL },
 303                 { "commit", disambiguate_commit_only },
 304                 { "committish", disambiguate_committish_only },
 305                 { "tree", disambiguate_tree_only },
 306                 { "treeish", disambiguate_treeish_only },
 307                 { "blob", disambiguate_blob_only }
 308         };
 309         int i;
 310
 311         if (!value)
 312                 return config_error_nonbool(var);
 313
 314         for (i = 0; i < ARRAY_SIZE(hints); i++) {
 315                 if (!strcasecmp(value, hints[i].name)) {
 316                         default_disambiguate_hint = hints[i].fn;
 317                         return 0;
 318                 }
 319         }
 320
 321         return error("unknown hint type for '%s': %s", var, value);
 322 }
 323
 324 static int init_object_disambiguation(struct repository *r,
 325                                       const char *name, int len,
 326                                       struct disambiguate_state *ds)
 327 {
 328         int i;
 329
 330         if (len < MINIMUM_ABBREV || len > the_hash_algo->hexsz)
 331                 return -1;
 332
 333         memset(ds, 0, sizeof(*ds));
 334
 335         for (i = 0; i < len ;i++) {
 336                 unsigned char c = name[i];
 337                 unsigned char val;
 338                 if (c >= '0' && c <= '9')
 339                         val = c - '0';
 340                 else if (c >= 'a' && c <= 'f')
 341                         val = c - 'a' + 10;
 342                 else if (c >= 'A' && c <='F') {
 343                         val = c - 'A' + 10;
 344                         c -= 'A' - 'a';
 345                 }
 346                 else
 347                         return -1;
 348                 ds->hex_pfx[i] = c;
 349                 if (!(i & 1))
 350                         val <<= 4;
 351                 ds->bin_pfx.hash[i >> 1] |= val;
 352         }
 353
 354         ds->len = len;
 355         ds->hex_pfx[len] = '\0';
 356         ds->repo = r;
 357         prepare_alt_odb(r);
 358         return 0;
 359 }
 360
 361 static int show_ambiguous_object(const struct object_id *oid, void *data)
 362 {
 363         const struct disambiguate_state *ds = data;
 364         struct strbuf desc = STRBUF_INIT;
 365         int type;
 366
 367         if (ds->fn && !ds->fn(ds->repo, oid, ds->cb_data))
 368                 return 0;
 369
 370         type = oid_object_info(ds->repo, oid, NULL);
 371         if (type == OBJ_COMMIT) {
 372                 struct commit *commit = lookup_commit(ds->repo, oid);
 373                 if (commit) {
 374                         struct pretty_print_context pp = {0};
 375                         pp.date_mode.type = DATE_SHORT;
 376                         format_commit_message(commit, " %ad - %s", &desc, &pp);
 377                 }
 378         } else if (type == OBJ_TAG) {
 379                 struct tag *tag = lookup_tag(ds->repo, oid);
 380                 if (!parse_tag(tag) && tag->tag)
 381                         strbuf_addf(&desc, " %s", tag->tag);
 382         }
 383
 384         advise("  %s %s%s",
 385                repo_find_unique_abbrev(ds->repo, oid, DEFAULT_ABBREV),
 386                type_name(type) ? type_name(type) : "unknown type",
 387                desc.buf);
 388
 389         strbuf_release(&desc);
 390         return 0;
 391 }
 392
 393 static int collect_ambiguous(const struct object_id *oid, void *data)
 394 {
 395         oid_array_append(data, oid);
 396         return 0;
 397 }
 398
 399 static int repo_collect_ambiguous(struct repository *r,
 400                                   const struct object_id *oid,
 401                                   void *data)
 402 {
 403         return collect_ambiguous(oid, data);
 404 }
 405
 406 static int sort_ambiguous(const void *a, const void *b, void *ctx)
 407 {
 408         struct repository *sort_ambiguous_repo = ctx;
 409         int a_type = oid_object_info(sort_ambiguous_repo, a, NULL);
 410         int b_type = oid_object_info(sort_ambiguous_repo, b, NULL);
 411         int a_type_sort;
 412         int b_type_sort;
 413
 414         /*
 415          * Sorts by hash within the same object type, just as
 416          * oid_array_for_each_unique() would do.
 417          */
 418         if (a_type == b_type)
 419                 return oidcmp(a, b);
 420
 421         /*
 422          * Between object types show tags, then commits, and finally
 423          * trees and blobs.
 424          *
 425          * The object_type enum is commit, tree, blob, tag, but we
 426          * want tag, commit, tree blob. Cleverly (perhaps too
 427          * cleverly) do that with modulus, since the enum assigns 1 to
 428          * commit, so tag becomes 0.
 429          */
 430         a_type_sort = a_type % 4;
 431         b_type_sort = b_type % 4;
 432         return a_type_sort > b_type_sort ? 1 : -1;
 433 }
 434
 435 static void sort_ambiguous_oid_array(struct repository *r, struct oid_array *a)
 436 {
 437         QSORT_S(a->oid, a->nr, sort_ambiguous, r);
 438 }
 439
 440 static enum get_oid_result get_short_oid(struct repository *r,
 441                                          const char *name, int len,
 442                                          struct object_id *oid,
 443                                          unsigned flags)
 444 {
 445         int status;
 446         struct disambiguate_state ds;
 447         int quietly = !!(flags & GET_OID_QUIETLY);
 448
 449         if (init_object_disambiguation(r, name, len, &ds) < 0)
 450                 return -1;
 451
 452         if (HAS_MULTI_BITS(flags & GET_OID_DISAMBIGUATORS))
 453                 BUG("multiple get_short_oid disambiguator flags");
 454
 455         if (flags & GET_OID_COMMIT)
 456                 ds.fn = disambiguate_commit_only;
 457         else if (flags & GET_OID_COMMITTISH)
 458                 ds.fn = disambiguate_committish_only;
 459         else if (flags & GET_OID_TREE)
 460                 ds.fn = disambiguate_tree_only;
 461         else if (flags & GET_OID_TREEISH)
 462                 ds.fn = disambiguate_treeish_only;
 463         else if (flags & GET_OID_BLOB)
 464                 ds.fn = disambiguate_blob_only;
 465         else
 466                 ds.fn = default_disambiguate_hint;
 467
 468         find_short_object_filename(&ds);
 469         find_short_packed_object(&ds);
 470         status = finish_object_disambiguation(&ds, oid);
 471
 472         /*
 473          * If we didn't find it, do the usual reprepare() slow-path,
 474          * since the object may have recently been added to the repository
 475          * or migrated from loose to packed.
 476          */
 477         if (status == MISSING_OBJECT) {
 478                 reprepare_packed_git(r);
 479                 find_short_object_filename(&ds);
 480                 find_short_packed_object(&ds);
 481                 status = finish_object_disambiguation(&ds, oid);
 482         }
 483
 484         if (!quietly && (status == SHORT_NAME_AMBIGUOUS)) {
 485                 struct oid_array collect = OID_ARRAY_INIT;
 486
 487                 error(_("short SHA1 %s is ambiguous"), ds.hex_pfx);
 488
 489                 /*
 490                  * We may still have ambiguity if we simply saw a series of
 491                  * candidates that did not satisfy our hint function. In
 492                  * that case, we still want to show them, so disable the hint
 493                  * function entirely.
 494                  */
 495                 if (!ds.ambiguous)
 496                         ds.fn = NULL;
 497
 498                 advise(_("The candidates are:"));
 499                 repo_for_each_abbrev(r, ds.hex_pfx, collect_ambiguous, &collect);
 500                 sort_ambiguous_oid_array(r, &collect);
 501
 502                 if (oid_array_for_each(&collect, show_ambiguous_object, &ds))
 503                         BUG("show_ambiguous_object shouldn't return non-zero");
 504                 oid_array_clear(&collect);
 505         }
 506
 507         return status;
 508 }
 509
 510 int repo_for_each_abbrev(struct repository *r, const char *prefix,
 511                          each_abbrev_fn fn, void *cb_data)
 512 {
 513         struct oid_array collect = OID_ARRAY_INIT;
 514         struct disambiguate_state ds;
 515         int ret;
 516
 517         if (init_object_disambiguation(r, prefix, strlen(prefix), &ds) < 0)
 518                 return -1;
 519
 520         ds.always_call_fn = 1;
 521         ds.fn = repo_collect_ambiguous;
 522         ds.cb_data = &collect;
 523         find_short_object_filename(&ds);
 524         find_short_packed_object(&ds);
 525
 526         ret = oid_array_for_each_unique(&collect, fn, cb_data);
 527         oid_array_clear(&collect);
 528         return ret;
 529 }
 530
 531 /*
 532  * Return the slot of the most-significant bit set in "val". There are various
 533  * ways to do this quickly with fls() or __builtin_clzl(), but speed is
 534  * probably not a big deal here.
 535  */
 536 static unsigned msb(unsigned long val)
 537 {
 538         unsigned r = 0;
 539         while (val >>= 1)
 540                 r++;
 541         return r;
 542 }
 543
 544 struct min_abbrev_data {
 545         unsigned int init_len;
 546         unsigned int cur_len;
 547         char *hex;
 548         struct repository *repo;
 549         const struct object_id *oid;
 550 };
 551
 552 static inline char get_hex_char_from_oid(const struct object_id *oid,
 553                                          unsigned int pos)
 554 {
 555         static const char hex[] = "0123456789abcdef";
 556
 557         if ((pos & 1) == 0)
 558                 return hex[oid->hash[pos >> 1] >> 4];
 559         else
 560                 return hex[oid->hash[pos >> 1] & 0xf];
 561 }
 562
 563 static int extend_abbrev_len(const struct object_id *oid, void *cb_data)
 564 {
 565         struct min_abbrev_data *mad = cb_data;
 566
 567         unsigned int i = mad->init_len;
 568         while (mad->hex[i] && mad->hex[i] == get_hex_char_from_oid(oid, i))
 569                 i++;
 570
 571         if (i < GIT_MAX_RAWSZ && i >= mad->cur_len)
 572                 mad->cur_len = i + 1;
 573
 574         return 0;
 575 }
 576
 577 static int repo_extend_abbrev_len(struct repository *r,
 578                                   const struct object_id *oid,
 579                                   void *cb_data)
 580 {
 581         return extend_abbrev_len(oid, cb_data);
 582 }
 583
 584 static void find_abbrev_len_for_midx(struct multi_pack_index *m,
 585                                      struct min_abbrev_data *mad)
 586 {
 587         int match = 0;
 588         uint32_t num, first = 0;
 589         struct object_id oid;
 590         const struct object_id *mad_oid;
 591
 592         if (!m->num_objects)
 593                 return;
 594
 595         num = m->num_objects;
 596         mad_oid = mad->oid;
 597         match = bsearch_midx(mad_oid, m, &first);
 598
 599         /*
 600          * first is now the position in the packfile where we would insert
 601          * mad->hash if it does not exist (or the position of mad->hash if
 602          * it does exist). Hence, we consider a maximum of two objects
 603          * nearby for the abbreviation length.
 604          */
 605         mad->init_len = 0;
 606         if (!match) {
 607                 if (nth_midxed_object_oid(&oid, m, first))
 608                         extend_abbrev_len(&oid, mad);
 609         } else if (first < num - 1) {
 610                 if (nth_midxed_object_oid(&oid, m, first + 1))
 611                         extend_abbrev_len(&oid, mad);
 612         }
 613         if (first > 0) {
 614                 if (nth_midxed_object_oid(&oid, m, first - 1))
 615                         extend_abbrev_len(&oid, mad);
 616         }
 617         mad->init_len = mad->cur_len;
 618 }
 619
 620 static void find_abbrev_len_for_pack(struct packed_git *p,
 621                                      struct min_abbrev_data *mad)
 622 {
 623         int match = 0;
 624         uint32_t num, first = 0;
 625         struct object_id oid;
 626         const struct object_id *mad_oid;
 627
 628         if (p->multi_pack_index)
 629                 return;
 630
 631         if (open_pack_index(p) || !p->num_objects)
 632                 return;
 633
 634         num = p->num_objects;
 635         mad_oid = mad->oid;
 636         match = bsearch_pack(mad_oid, p, &first);
 637
 638         /*
 639          * first is now the position in the packfile where we would insert
 640          * mad->hash if it does not exist (or the position of mad->hash if
 641          * it does exist). Hence, we consider a maximum of two objects
 642          * nearby for the abbreviation length.
 643          */
 644         mad->init_len = 0;
 645         if (!match) {
 646                 if (nth_packed_object_oid(&oid, p, first))
 647                         extend_abbrev_len(&oid, mad);
 648         } else if (first < num - 1) {
 649                 if (nth_packed_object_oid(&oid, p, first + 1))
 650                         extend_abbrev_len(&oid, mad);
 651         }
 652         if (first > 0) {
 653                 if (nth_packed_object_oid(&oid, p, first - 1))
 654                         extend_abbrev_len(&oid, mad);
 655         }
 656         mad->init_len = mad->cur_len;
 657 }
 658
 659 static void find_abbrev_len_packed(struct min_abbrev_data *mad)
 660 {
 661         struct multi_pack_index *m;
 662         struct packed_git *p;
 663
 664         for (m = get_multi_pack_index(mad->repo); m; m = m->next)
 665                 find_abbrev_len_for_midx(m, mad);
 666         for (p = get_packed_git(mad->repo); p; p = p->next)
 667                 find_abbrev_len_for_pack(p, mad);
 668 }
 669
 670 int repo_find_unique_abbrev_r(struct repository *r, char *hex,
 671                               const struct object_id *oid, int len)
 672 {
 673         struct disambiguate_state ds;
 674         struct min_abbrev_data mad;
 675         struct object_id oid_ret;
 676         const unsigned hexsz = r->hash_algo->hexsz;
 677
 678         if (len < 0) {
 679                 unsigned long count = repo_approximate_object_count(r);
 680                 /*
 681                  * Add one because the MSB only tells us the highest bit set,
 682                  * not including the value of all the _other_ bits (so "15"
 683                  * is only one off of 2^4, but the MSB is the 3rd bit.
 684                  */
 685                 len = msb(count) + 1;
 686                 /*
 687                  * We now know we have on the order of 2^len objects, which
 688                  * expects a collision at 2^(len/2). But we also care about hex
 689                  * chars, not bits, and there are 4 bits per hex. So all
 690                  * together we need to divide by 2 and round up.
 691                  */
 692                 len = DIV_ROUND_UP(len, 2);
 693                 /*
 694                  * For very small repos, we stick with our regular fallback.
 695                  */
 696                 if (len < FALLBACK_DEFAULT_ABBREV)
 697                         len = FALLBACK_DEFAULT_ABBREV;
 698         }
 699
 700         oid_to_hex_r(hex, oid);
 701         if (len == hexsz || !len)
 702                 return hexsz;
 703
 704         mad.repo = r;
 705         mad.init_len = len;
 706         mad.cur_len = len;
 707         mad.hex = hex;
 708         mad.oid = oid;
 709
 710         find_abbrev_len_packed(&mad);
 711
 712         if (init_object_disambiguation(r, hex, mad.cur_len, &ds) < 0)
 713                 return -1;
 714
 715         ds.fn = repo_extend_abbrev_len;
 716         ds.always_call_fn = 1;
 717         ds.cb_data = (void *)&mad;
 718
 719         find_short_object_filename(&ds);
 720         (void)finish_object_disambiguation(&ds, &oid_ret);
 721
 722         hex[mad.cur_len] = 0;
 723         return mad.cur_len;
 724 }
 725
 726 const char *repo_find_unique_abbrev(struct repository *r,
 727                                     const struct object_id *oid,
 728                                     int len)
 729 {
 730         static int bufno;
 731         static char hexbuffer[4][GIT_MAX_HEXSZ + 1];
 732         char *hex = hexbuffer[bufno];
 733         bufno = (bufno + 1) % ARRAY_SIZE(hexbuffer);
 734         repo_find_unique_abbrev_r(r, hex, oid, len);
 735         return hex;
 736 }
 737
 738 static int ambiguous_path(const char *path, int len)
 739 {
 740         int slash = 1;
 741         int cnt;
 742
 743         for (cnt = 0; cnt < len; cnt++) {
 744                 switch (*path++) {
 745                 case '\0':
 746                         break;
 747                 case '/':
 748                         if (slash)
 749                                 break;
 750                         slash = 1;
 751                         continue;
 752                 case '.':
 753                         continue;
 754                 default:
 755                         slash = 0;
 756                         continue;
 757                 }
 758                 break;
 759         }
 760         return slash;
 761 }
 762
 763 static inline int at_mark(const char *string, int len,
 764                           const char **suffix, int nr)
 765 {
 766         int i;
 767
 768         for (i = 0; i < nr; i++) {
 769                 int suffix_len = strlen(suffix[i]);
 770                 if (suffix_len <= len
 771                     && !strncasecmp(string, suffix[i], suffix_len))
 772                         return suffix_len;
 773         }
 774         return 0;
 775 }
 776
 777 static inline int upstream_mark(const char *string, int len)
 778 {
 779         const char *suffix[] = { "@{upstream}", "@{u}" };
 780         return at_mark(string, len, suffix, ARRAY_SIZE(suffix));
 781 }
 782
 783 static inline int push_mark(const char *string, int len)
 784 {
 785         const char *suffix[] = { "@{push}" };
 786         return at_mark(string, len, suffix, ARRAY_SIZE(suffix));
 787 }
 788
 789 static enum get_oid_result get_oid_1(struct repository *r, const char *name, int len, struct object_id *oid, unsigned lookup_flags);
 790 static int interpret_nth_prior_checkout(struct repository *r, const char *name, int namelen, struct strbuf *buf);
 791
 792 static int get_oid_basic(struct repository *r, const char *str, int len,
 793                          struct object_id *oid, unsigned int flags)
 794 {
 795         static const char *warn_msg = "refname '%.*s' is ambiguous.";
 796         static const char *object_name_msg = N_(
 797         "Git normally never creates a ref that ends with 40 hex characters\n"
 798         "because it will be ignored when you just specify 40-hex. These refs\n"
 799         "may be created by mistake. For example,\n"
 800         "\n"
 801         "  git switch -c $br $(git rev-parse ...)\n"
 802         "\n"
 803         "where \"$br\" is somehow empty and a 40-hex ref is created. Please\n"
 804         "examine these refs and maybe delete them. Turn this message off by\n"
 805         "running \"git config advice.objectNameWarning false\"");
 806         struct object_id tmp_oid;
 807         char *real_ref = NULL;
 808         int refs_found = 0;
 809         int at, reflog_len, nth_prior = 0;
 810
 811         if (len == r->hash_algo->hexsz && !get_oid_hex(str, oid)) {
 812                 if (warn_ambiguous_refs && warn_on_object_refname_ambiguity) {
 813                         refs_found = repo_dwim_ref(r, str, len, &tmp_oid, &real_ref);
 814                         if (refs_found > 0) {
 815                                 warning(warn_msg, len, str);
 816                                 if (advice_object_name_warning)
 817                                         fprintf(stderr, "%s\n", _(object_name_msg));
 818                         }
 819                         free(real_ref);
 820                 }
 821                 return 0;
 822         }
 823
 824         /* basic@{time or number or -number} format to query ref-log */
 825         reflog_len = at = 0;
 826         if (len && str[len-1] == '}') {
 827                 for (at = len-4; at >= 0; at--) {
 828                         if (str[at] == '@' && str[at+1] == '{') {
 829                                 if (str[at+2] == '-') {
 830                                         if (at != 0)
 831                                                 /* @{-N} not at start */
 832                                                 return -1;
 833                                         nth_prior = 1;
 834                                         continue;
 835                                 }
 836                                 if (!upstream_mark(str + at, len - at) &&
 837                                     !push_mark(str + at, len - at)) {
 838                                         reflog_len = (len-1) - (at+2);
 839                                         len = at;
 840                                 }
 841                                 break;
 842                         }
 843                 }
 844         }
 845
 846         /* Accept only unambiguous ref paths. */
 847         if (len && ambiguous_path(str, len))
 848                 return -1;
 849
 850         if (nth_prior) {
 851                 struct strbuf buf = STRBUF_INIT;
 852                 int detached;
 853
 854                 if (interpret_nth_prior_checkout(r, str, len, &buf) > 0) {
 855                         detached = (buf.len == r->hash_algo->hexsz && !get_oid_hex(buf.buf, oid));
 856                         strbuf_release(&buf);
 857                         if (detached)
 858                                 return 0;
 859                 }
 860         }
 861
 862         if (!len && reflog_len)
 863                 /* allow "@{...}" to mean the current branch reflog */
 864                 refs_found = repo_dwim_ref(r, "HEAD", 4, oid, &real_ref);
 865         else if (reflog_len)
 866                 refs_found = repo_dwim_log(r, str, len, oid, &real_ref);
 867         else
 868                 refs_found = repo_dwim_ref(r, str, len, oid, &real_ref);
 869
 870         if (!refs_found)
 871                 return -1;
 872
 873         if (warn_ambiguous_refs && !(flags & GET_OID_QUIETLY) &&
 874             (refs_found > 1 ||
 875              !get_short_oid(r, str, len, &tmp_oid, GET_OID_QUIETLY)))
 876                 warning(warn_msg, len, str);
 877
 878         if (reflog_len) {
 879                 int nth, i;
 880                 timestamp_t at_time;
 881                 timestamp_t co_time;
 882                 int co_tz, co_cnt;
 883
 884                 /* Is it asking for N-th entry, or approxidate? */
 885                 for (i = nth = 0; 0 <= nth && i < reflog_len; i++) {
 886                         char ch = str[at+2+i];
 887                         if ('0' <= ch && ch <= '9')
 888                                 nth = nth * 10 + ch - '0';
 889                         else
 890                                 nth = -1;
 891                 }
 892                 if (100000000 <= nth) {
 893                         at_time = nth;
 894                         nth = -1;
 895                 } else if (0 <= nth)
 896                         at_time = 0;
 897                 else {
 898                         int errors = 0;
 899                         char *tmp = xstrndup(str + at + 2, reflog_len);
 900                         at_time = approxidate_careful(tmp, &errors);
 901                         free(tmp);
 902                         if (errors) {
 903                                 free(real_ref);
 904                                 return -1;
 905                         }
 906                 }
 907                 if (read_ref_at(get_main_ref_store(r),
 908                                 real_ref, flags, at_time, nth, oid, NULL,
 909                                 &co_time, &co_tz, &co_cnt)) {
 910                         if (!len) {
 911                                 if (starts_with(real_ref, "refs/heads/")) {
 912                                         str = real_ref + 11;
 913                                         len = strlen(real_ref + 11);
 914                                 } else {
 915                                         /* detached HEAD */
 916                                         str = "HEAD";
 917                                         len = 4;
 918                                 }
 919                         }
 920                         if (at_time) {
 921                                 if (!(flags & GET_OID_QUIETLY)) {
 922                                         warning("Log for '%.*s' only goes "
 923                                                 "back to %s.", len, str,
 924                                                 show_date(co_time, co_tz, DATE_MODE(RFC2822)));
 925                                 }
 926                         } else {
 927                                 if (flags & GET_OID_QUIETLY) {
 928                                         exit(128);
 929                                 }
 930                                 die("Log for '%.*s' only has %d entries.",
 931                                     len, str, co_cnt);
 932                         }
 933                 }
 934         }
 935
 936         free(real_ref);
 937         return 0;
 938 }
 939
 940 static enum get_oid_result get_parent(struct repository *r,
 941                                       const char *name, int len,
 942                                       struct object_id *result, int idx)
 943 {
 944         struct object_id oid;
 945         enum get_oid_result ret = get_oid_1(r, name, len, &oid,
 946                                             GET_OID_COMMITTISH);
 947         struct commit *commit;
 948         struct commit_list *p;
 949
 950         if (ret)
 951                 return ret;
 952         commit = lookup_commit_reference(r, &oid);
 953         if (parse_commit(commit))
 954                 return MISSING_OBJECT;
 955         if (!idx) {
 956                 oidcpy(result, &commit->object.oid);
 957                 return FOUND;
 958         }
 959         p = commit->parents;
 960         while (p) {
 961                 if (!--idx) {
 962                         oidcpy(result, &p->item->object.oid);
 963                         return FOUND;
 964                 }
 965                 p = p->next;
 966         }
 967         return MISSING_OBJECT;
 968 }
 969
 970 static enum get_oid_result get_nth_ancestor(struct repository *r,
 971                                             const char *name, int len,
 972                                             struct object_id *result,
 973                                             int generation)
 974 {
 975         struct object_id oid;
 976         struct commit *commit;
 977         int ret;
 978
 979         ret = get_oid_1(r, name, len, &oid, GET_OID_COMMITTISH);
 980         if (ret)
 981                 return ret;
 982         commit = lookup_commit_reference(r, &oid);
 983         if (!commit)
 984                 return MISSING_OBJECT;
 985
 986         while (generation--) {
 987                 if (parse_commit(commit) || !commit->parents)
 988                         return MISSING_OBJECT;
 989                 commit = commit->parents->item;
 990         }
 991         oidcpy(result, &commit->object.oid);
 992         return FOUND;
 993 }
 994
 995 struct object *repo_peel_to_type(struct repository *r, const char *name, int namelen,
 996                                  struct object *o, enum object_type expected_type)
 997 {
 998         if (name && !namelen)
 999                 namelen = strlen(name);
1000         while (1) {
1001                 if (!o || (!o->parsed && !parse_object(r, &o->oid)))
1002                         return NULL;
1003                 if (expected_type == OBJ_ANY || o->type == expected_type)
1004                         return o;
1005                 if (o->type == OBJ_TAG)
1006                         o = ((struct tag*) o)->tagged;
1007                 else if (o->type == OBJ_COMMIT)
1008                         o = &(repo_get_commit_tree(r, ((struct commit *)o))->object);
1009                 else {
1010                         if (name)
1011                                 error("%.*s: expected %s type, but the object "
1012                                       "dereferences to %s type",
1013                                       namelen, name, type_name(expected_type),
1014                                       type_name(o->type));
1015                         return NULL;
1016                 }
1017         }
1018 }
1019
1020 static int peel_onion(struct repository *r, const char *name, int len,
1021                       struct object_id *oid, unsigned lookup_flags)
1022 {
1023         struct object_id outer;
1024         const char *sp;
1025         unsigned int expected_type = 0;
1026         struct object *o;
1027
1028         /*
1029          * "ref^{type}" dereferences ref repeatedly until you cannot
1030          * dereference anymore, or you get an object of given type,
1031          * whichever comes first.  "ref^{}" means just dereference
1032          * tags until you get a non-tag.  "ref^0" is a shorthand for
1033          * "ref^{commit}".  "commit^{tree}" could be used to find the
1034          * top-level tree of the given commit.
1035          */
1036         if (len < 4 || name[len-1] != '}')
1037                 return -1;
1038
1039         for (sp = name + len - 1; name <= sp; sp--) {
1040                 int ch = *sp;
1041                 if (ch == '{' && name < sp && sp[-1] == '^')
1042                         break;
1043         }
1044         if (sp <= name)
1045                 return -1;
1046
1047         sp++; /* beginning of type name, or closing brace for empty */
1048         if (starts_with(sp, "commit}"))
1049                 expected_type = OBJ_COMMIT;
1050         else if (starts_with(sp, "tag}"))
1051                 expected_type = OBJ_TAG;
1052         else if (starts_with(sp, "tree}"))
1053                 expected_type = OBJ_TREE;
1054         else if (starts_with(sp, "blob}"))
1055                 expected_type = OBJ_BLOB;
1056         else if (starts_with(sp, "object}"))
1057                 expected_type = OBJ_ANY;
1058         else if (sp[0] == '}')
1059                 expected_type = OBJ_NONE;
1060         else if (sp[0] == '/')
1061                 expected_type = OBJ_COMMIT;
1062         else
1063                 return -1;
1064
1065         lookup_flags &= ~GET_OID_DISAMBIGUATORS;
1066         if (expected_type == OBJ_COMMIT)
1067                 lookup_flags |= GET_OID_COMMITTISH;
1068         else if (expected_type == OBJ_TREE)
1069                 lookup_flags |= GET_OID_TREEISH;
1070
1071         if (get_oid_1(r, name, sp - name - 2, &outer, lookup_flags))
1072                 return -1;
1073
1074         o = parse_object(r, &outer);
1075         if (!o)
1076                 return -1;
1077         if (!expected_type) {
1078                 o = deref_tag(r, o, name, sp - name - 2);
1079                 if (!o || (!o->parsed && !parse_object(r, &o->oid)))
1080                         return -1;
1081                 oidcpy(oid, &o->oid);
1082                 return 0;
1083         }
1084
1085         /*
1086          * At this point, the syntax look correct, so
1087          * if we do not get the needed object, we should
1088          * barf.
1089          */
1090         o = repo_peel_to_type(r, name, len, o, expected_type);
1091         if (!o)
1092                 return -1;
1093
1094         oidcpy(oid, &o->oid);
1095         if (sp[0] == '/') {
1096                 /* "$commit^{/foo}" */
1097                 char *prefix;
1098                 int ret;
1099                 struct commit_list *list = NULL;
1100
1101                 /*
1102                  * $commit^{/}. Some regex implementation may reject.
1103                  * We don't need regex anyway. '' pattern always matches.
1104                  */
1105                 if (sp[1] == '}')
1106                         return 0;
1107
1108                 prefix = xstrndup(sp + 1, name + len - 1 - (sp + 1));
1109                 commit_list_insert((struct commit *)o, &list);
1110                 ret = get_oid_oneline(r, prefix, oid, list);
1111                 free(prefix);
1112                 return ret;
1113         }
1114         return 0;
1115 }
1116
1117 static int get_describe_name(struct repository *r,
1118                              const char *name, int len,
1119                              struct object_id *oid)
1120 {
1121         const char *cp;
1122         unsigned flags = GET_OID_QUIETLY | GET_OID_COMMIT;
1123
1124         for (cp = name + len - 1; name + 2 <= cp; cp--) {
1125                 char ch = *cp;
1126                 if (!isxdigit(ch)) {
1127                         /* We must be looking at g in "SOMETHING-g"
1128                          * for it to be describe output.
1129                          */
1130                         if (ch == 'g' && cp[-1] == '-') {
1131                                 cp++;
1132                                 len -= cp - name;
1133                                 return get_short_oid(r,
1134                                                      cp, len, oid, flags);
1135                         }
1136                 }
1137         }
1138         return -1;
1139 }
1140
1141 static enum get_oid_result get_oid_1(struct repository *r,
1142                                      const char *name, int len,
1143                                      struct object_id *oid,
1144                                      unsigned lookup_flags)
1145 {
1146         int ret, has_suffix;
1147         const char *cp;
1148
1149         /*
1150          * "name~3" is "name^^^", "name~" is "name~1", and "name^" is "name^1".
1151          */
1152         has_suffix = 0;
1153         for (cp = name + len - 1; name <= cp; cp--) {
1154                 int ch = *cp;
1155                 if ('0' <= ch && ch <= '9')
1156                         continue;
1157                 if (ch == '~' || ch == '^')
1158                         has_suffix = ch;
1159                 break;
1160         }
1161
1162         if (has_suffix) {
1163                 int num = 0;
1164                 int len1 = cp - name;
1165                 cp++;
1166                 while (cp < name + len)
1167                         num = num * 10 + *cp++ - '0';
1168                 if (!num && len1 == len - 1)
1169                         num = 1;
1170                 if (has_suffix == '^')
1171                         return get_parent(r, name, len1, oid, num);
1172                 /* else if (has_suffix == '~') -- goes without saying */
1173                 return get_nth_ancestor(r, name, len1, oid, num);
1174         }
1175
1176         ret = peel_onion(r, name, len, oid, lookup_flags);
1177         if (!ret)
1178                 return FOUND;
1179
1180         ret = get_oid_basic(r, name, len, oid, lookup_flags);
1181         if (!ret)
1182                 return FOUND;
1183
1184         /* It could be describe output that is "SOMETHING-gXXXX" */
1185         ret = get_describe_name(r, name, len, oid);
1186         if (!ret)
1187                 return FOUND;
1188
1189         return get_short_oid(r, name, len, oid, lookup_flags);
1190 }
1191
1192 /*
1193  * This interprets names like ':/Initial revision of "git"' by searching
1194  * through history and returning the first commit whose message starts
1195  * the given regular expression.
1196  *
1197  * For negative-matching, prefix the pattern-part with '!-', like: ':/!-WIP'.
1198  *
1199  * For a literal '!' character at the beginning of a pattern, you have to repeat
1200  * that, like: ':/!!foo'
1201  *
1202  * For future extension, all other sequences beginning with ':/!' are reserved.
1203  */
1204
1205 /* Remember to update object flag allocation in object.h */
1206 #define ONELINE_SEEN (1u<<20)
1207
1208 struct handle_one_ref_cb {
1209         struct repository *repo;
1210         struct commit_list **list;
1211 };
1212
1213 static int handle_one_ref(const char *path, const struct object_id *oid,
1214                           int flag, void *cb_data)
1215 {
1216         struct handle_one_ref_cb *cb = cb_data;
1217         struct commit_list **list = cb->list;
1218         struct object *object = parse_object(cb->repo, oid);
1219         if (!object)
1220                 return 0;
1221         if (object->type == OBJ_TAG) {
1222                 object = deref_tag(cb->repo, object, path,
1223                                    strlen(path));
1224                 if (!object)
1225                         return 0;
1226         }
1227         if (object->type != OBJ_COMMIT)
1228                 return 0;
1229         commit_list_insert((struct commit *)object, list);
1230         return 0;
1231 }
1232
1233 static int get_oid_oneline(struct repository *r,
1234                            const char *prefix, struct object_id *oid,
1235                            struct commit_list *list)
1236 {
1237         struct commit_list *backup = NULL, *l;
1238         int found = 0;
1239         int negative = 0;
1240         regex_t regex;
1241
1242         if (prefix[0] == '!') {
1243                 prefix++;
1244
1245                 if (prefix[0] == '-') {
1246                         prefix++;
1247                         negative = 1;
1248                 } else if (prefix[0] != '!') {
1249                         return -1;
1250                 }
1251         }
1252
1253         if (regcomp(&regex, prefix, REG_EXTENDED))
1254                 return -1;
1255
1256         for (l = list; l; l = l->next) {
1257                 l->item->object.flags |= ONELINE_SEEN;
1258                 commit_list_insert(l->item, &backup);
1259         }
1260         while (list) {
1261                 const char *p, *buf;
1262                 struct commit *commit;
1263                 int matches;
1264
1265                 commit = pop_most_recent_commit(&list, ONELINE_SEEN);
1266                 if (!parse_object(r, &commit->object.oid))
1267                         continue;
1268                 buf = get_commit_buffer(commit, NULL);
1269                 p = strstr(buf, "\n\n");
1270                 matches = negative ^ (p && !regexec(&regex, p + 2, 0, NULL, 0));
1271                 unuse_commit_buffer(commit, buf);
1272
1273                 if (matches) {
1274                         oidcpy(oid, &commit->object.oid);
1275                         found = 1;
1276                         break;
1277                 }
1278         }
1279         regfree(&regex);
1280         free_commit_list(list);
1281         for (l = backup; l; l = l->next)
1282                 clear_commit_marks(l->item, ONELINE_SEEN);
1283         free_commit_list(backup);
1284         return found ? 0 : -1;
1285 }
1286
1287 struct grab_nth_branch_switch_cbdata {
1288         int remaining;
1289         struct strbuf buf;
1290 };
1291
1292 static int grab_nth_branch_switch(struct object_id *ooid, struct object_id *noid,
1293                                   const char *email, timestamp_t timestamp, int tz,
1294                                   const char *message, void *cb_data)
1295 {
1296         struct grab_nth_branch_switch_cbdata *cb = cb_data;
1297         const char *match = NULL, *target = NULL;
1298         size_t len;
1299
1300         if (skip_prefix(message, "checkout: moving from ", &match))
1301                 target = strstr(match, " to ");
1302
1303         if (!match || !target)
1304                 return 0;
1305         if (--(cb->remaining) == 0) {
1306                 len = target - match;
1307                 strbuf_reset(&cb->buf);
1308                 strbuf_add(&cb->buf, match, len);
1309                 return 1; /* we are done */
1310         }
1311         return 0;
1312 }
1313
1314 /*
1315  * Parse @{-N} syntax, return the number of characters parsed
1316  * if successful; otherwise signal an error with negative value.
1317  */
1318 static int interpret_nth_prior_checkout(struct repository *r,
1319                                         const char *name, int namelen,
1320                                         struct strbuf *buf)
1321 {
1322         long nth;
1323         int retval;
1324         struct grab_nth_branch_switch_cbdata cb;
1325         const char *brace;
1326         char *num_end;
1327
1328         if (namelen < 4)
1329                 return -1;
1330         if (name[0] != '@' || name[1] != '{' || name[2] != '-')
1331                 return -1;
1332         brace = memchr(name, '}', namelen);
1333         if (!brace)
1334                 return -1;
1335         nth = strtol(name + 3, &num_end, 10);
1336         if (num_end != brace)
1337                 return -1;
1338         if (nth <= 0)
1339                 return -1;
1340         cb.remaining = nth;
1341         strbuf_init(&cb.buf, 20);
1342
1343         retval = refs_for_each_reflog_ent_reverse(get_main_ref_store(r),
1344                         "HEAD", grab_nth_branch_switch, &cb);
1345         if (0 < retval) {
1346                 strbuf_reset(buf);
1347                 strbuf_addbuf(buf, &cb.buf);
1348                 retval = brace - name + 1;
1349         } else
1350                 retval = 0;
1351
1352         strbuf_release(&cb.buf);
1353         return retval;
1354 }
1355
1356 int repo_get_oid_mb(struct repository *r,
1357                     const char *name,
1358                     struct object_id *oid)
1359 {
1360         struct commit *one, *two;
1361         struct commit_list *mbs;
1362         struct object_id oid_tmp;
1363         const char *dots;
1364         int st;
1365
1366         dots = strstr(name, "...");
1367         if (!dots)
1368                 return repo_get_oid(r, name, oid);
1369         if (dots == name)
1370                 st = repo_get_oid(r, "HEAD", &oid_tmp);
1371         else {
1372                 struct strbuf sb;
1373                 strbuf_init(&sb, dots - name);
1374                 strbuf_add(&sb, name, dots - name);
1375                 st = repo_get_oid_committish(r, sb.buf, &oid_tmp);
1376                 strbuf_release(&sb);
1377         }
1378         if (st)
1379                 return st;
1380         one = lookup_commit_reference_gently(r, &oid_tmp, 0);
1381         if (!one)
1382                 return -1;
1383
1384         if (repo_get_oid_committish(r, dots[3] ? (dots + 3) : "HEAD", &oid_tmp))
1385                 return -1;
1386         two = lookup_commit_reference_gently(r, &oid_tmp, 0);
1387         if (!two)
1388                 return -1;
1389         mbs = repo_get_merge_bases(r, one, two);
1390         if (!mbs || mbs->next)
1391                 st = -1;
1392         else {
1393                 st = 0;
1394                 oidcpy(oid, &mbs->item->object.oid);
1395         }
1396         free_commit_list(mbs);
1397         return st;
1398 }
1399
1400 /* parse @something syntax, when 'something' is not {.*} */
1401 static int interpret_empty_at(const char *name, int namelen, int len, struct strbuf *buf)
1402 {
1403         const char *next;
1404
1405         if (len || name[1] == '{')
1406                 return -1;
1407
1408         /* make sure it's a single @, or @@{.*}, not @foo */
1409         next = memchr(name + len + 1, '@', namelen - len - 1);
1410         if (next && next[1] != '{')
1411                 return -1;
1412         if (!next)
1413                 next = name + namelen;
1414         if (next != name + 1)
1415                 return -1;
1416
1417         strbuf_reset(buf);
1418         strbuf_add(buf, "HEAD", 4);
1419         return 1;
1420 }
1421
1422 static int reinterpret(struct repository *r,
1423                        const char *name, int namelen, int len,
1424                        struct strbuf *buf, unsigned allowed)
1425 {
1426         /* we have extra data, which might need further processing */
1427         struct strbuf tmp = STRBUF_INIT;
1428         int used = buf->len;
1429         int ret;
1430
1431         strbuf_add(buf, name + len, namelen - len);
1432         ret = repo_interpret_branch_name(r, buf->buf, buf->len, &tmp, allowed);
1433         /* that data was not interpreted, remove our cruft */
1434         if (ret < 0) {
1435                 strbuf_setlen(buf, used);
1436                 return len;
1437         }
1438         strbuf_reset(buf);
1439         strbuf_addbuf(buf, &tmp);
1440         strbuf_release(&tmp);
1441         /* tweak for size of {-N} versus expanded ref name */
1442         return ret - used + len;
1443 }
1444
1445 static void set_shortened_ref(struct repository *r, struct strbuf *buf, const char *ref)
1446 {
1447         char *s = refs_shorten_unambiguous_ref(get_main_ref_store(r), ref, 0);
1448         strbuf_reset(buf);
1449         strbuf_addstr(buf, s);
1450         free(s);
1451 }
1452
1453 static int branch_interpret_allowed(const char *refname, unsigned allowed)
1454 {
1455         if (!allowed)
1456                 return 1;
1457
1458         if ((allowed & INTERPRET_BRANCH_LOCAL) &&
1459             starts_with(refname, "refs/heads/"))
1460                 return 1;
1461         if ((allowed & INTERPRET_BRANCH_REMOTE) &&
1462             starts_with(refname, "refs/remotes/"))
1463                 return 1;
1464
1465         return 0;
1466 }
1467
1468 static int interpret_branch_mark(struct repository *r,
1469                                  const char *name, int namelen,
1470                                  int at, struct strbuf *buf,
1471                                  int (*get_mark)(const char *, int),
1472                                  const char *(*get_data)(struct branch *,
1473                                                          struct strbuf *),
1474                                  unsigned allowed)
1475 {
1476         int len;
1477         struct branch *branch;
1478         struct strbuf err = STRBUF_INIT;
1479         const char *value;
1480
1481         len = get_mark(name + at, namelen - at);
1482         if (!len)
1483                 return -1;
1484
1485         if (memchr(name, ':', at))
1486                 return -1;
1487
1488         if (at) {
1489                 char *name_str = xmemdupz(name, at);
1490                 branch = branch_get(name_str);
1491                 free(name_str);
1492         } else
1493                 branch = branch_get(NULL);
1494
1495         value = get_data(branch, &err);
1496         if (!value)
1497                 die("%s", err.buf);
1498
1499         if (!branch_interpret_allowed(value, allowed))
1500                 return -1;
1501
1502         set_shortened_ref(r, buf, value);
1503         return len + at;
1504 }
1505
1506 int repo_interpret_branch_name(struct repository *r,
1507                                const char *name, int namelen,
1508                                struct strbuf *buf,
1509                                unsigned allowed)
1510 {
1511         char *at;
1512         const char *start;
1513         int len;
1514
1515         if (!namelen)
1516                 namelen = strlen(name);
1517
1518         if (!allowed || (allowed & INTERPRET_BRANCH_LOCAL)) {
1519                 len = interpret_nth_prior_checkout(r, name, namelen, buf);
1520                 if (!len) {
1521                         return len; /* syntax Ok, not enough switches */
1522                 } else if (len > 0) {
1523                         if (len == namelen)
1524                                 return len; /* consumed all */
1525                         else
1526                                 return reinterpret(r, name, namelen, len, buf, allowed);
1527                 }
1528         }
1529
1530         for (start = name;
1531              (at = memchr(start, '@', namelen - (start - name)));
1532              start = at + 1) {
1533
1534                 if (!allowed || (allowed & INTERPRET_BRANCH_HEAD)) {
1535                         len = interpret_empty_at(name, namelen, at - name, buf);
1536                         if (len > 0)
1537                                 return reinterpret(r, name, namelen, len, buf,
1538                                                    allowed);
1539                 }
1540
1541                 len = interpret_branch_mark(r, name, namelen, at - name, buf,
1542                                             upstream_mark, branch_get_upstream,
1543                                             allowed);
1544                 if (len > 0)
1545                         return len;
1546
1547                 len = interpret_branch_mark(r, name, namelen, at - name, buf,
1548                                             push_mark, branch_get_push,
1549                                             allowed);
1550                 if (len > 0)
1551                         return len;
1552         }
1553
1554         return -1;
1555 }
1556
1557 void strbuf_branchname(struct strbuf *sb, const char *name, unsigned allowed)
1558 {
1559         int len = strlen(name);
1560         int used = interpret_branch_name(name, len, sb, allowed);
1561
1562         if (used < 0)
1563                 used = 0;
1564         strbuf_add(sb, name + used, len - used);
1565 }
1566
1567 int strbuf_check_branch_ref(struct strbuf *sb, const char *name)
1568 {
1569         if (startup_info->have_repository)
1570                 strbuf_branchname(sb, name, INTERPRET_BRANCH_LOCAL);
1571         else
1572                 strbuf_addstr(sb, name);
1573
1574         /*
1575          * This splice must be done even if we end up rejecting the
1576          * name; builtin/branch.c::copy_or_rename_branch() still wants
1577          * to see what the name expanded to so that "branch -m" can be
1578          * used as a tool to correct earlier mistakes.
1579          */
1580         strbuf_splice(sb, 0, 0, "refs/heads/", 11);
1581
1582         if (*name == '-' ||
1583             !strcmp(sb->buf, "refs/heads/HEAD"))
1584                 return -1;
1585
1586         return check_refname_format(sb->buf, 0);
1587 }
1588
1589 /*
1590  * This is like "get_oid_basic()", except it allows "object ID expressions",
1591  * notably "xyz^" for "parent of xyz"
1592  */
1593 int repo_get_oid(struct repository *r, const char *name, struct object_id *oid)
1594 {
1595         struct object_context unused;
1596         return get_oid_with_context(r, name, 0, oid, &unused);
1597 }
1598
1599 /*
1600  * This returns a non-zero value if the string (built using printf
1601  * format and the given arguments) is not a valid object.
1602  */
1603 int get_oidf(struct object_id *oid, const char *fmt, ...)
1604 {
1605         va_list ap;
1606         int ret;
1607         struct strbuf sb = STRBUF_INIT;
1608
1609         va_start(ap, fmt);
1610         strbuf_vaddf(&sb, fmt, ap);
1611         va_end(ap);
1612
1613         ret = get_oid(sb.buf, oid);
1614         strbuf_release(&sb);
1615
1616         return ret;
1617 }
1618
1619 /*
1620  * Many callers know that the user meant to name a commit-ish by
1621  * syntactical positions where the object name appears.  Calling this
1622  * function allows the machinery to disambiguate shorter-than-unique
1623  * abbreviated object names between commit-ish and others.
1624  *
1625  * Note that this does NOT error out when the named object is not a
1626  * commit-ish. It is merely to give a hint to the disambiguation
1627  * machinery.
1628  */
1629 int repo_get_oid_committish(struct repository *r,
1630                             const char *name,
1631                             struct object_id *oid)
1632 {
1633         struct object_context unused;
1634         return get_oid_with_context(r, name, GET_OID_COMMITTISH,
1635                                     oid, &unused);
1636 }
1637
1638 int repo_get_oid_treeish(struct repository *r,
1639                          const char *name,
1640                          struct object_id *oid)
1641 {
1642         struct object_context unused;
1643         return get_oid_with_context(r, name, GET_OID_TREEISH,
1644                                     oid, &unused);
1645 }
1646
1647 int repo_get_oid_commit(struct repository *r,
1648                         const char *name,
1649                         struct object_id *oid)
1650 {
1651         struct object_context unused;
1652         return get_oid_with_context(r, name, GET_OID_COMMIT,
1653                                     oid, &unused);
1654 }
1655
1656 int repo_get_oid_tree(struct repository *r,
1657                       const char *name,
1658                       struct object_id *oid)
1659 {
1660         struct object_context unused;
1661         return get_oid_with_context(r, name, GET_OID_TREE,
1662                                     oid, &unused);
1663 }
1664
1665 int repo_get_oid_blob(struct repository *r,
1666                       const char *name,
1667                       struct object_id *oid)
1668 {
1669         struct object_context unused;
1670         return get_oid_with_context(r, name, GET_OID_BLOB,
1671                                     oid, &unused);
1672 }
1673
1674 /* Must be called only when object_name:filename doesn't exist. */
1675 static void diagnose_invalid_oid_path(struct repository *r,
1676                                       const char *prefix,
1677                                       const char *filename,
1678                                       const struct object_id *tree_oid,
1679                                       const char *object_name,
1680                                       int object_name_len)
1681 {
1682         struct object_id oid;
1683         unsigned short mode;
1684
1685         if (!prefix)
1686                 prefix = "";
1687
1688         if (file_exists(filename))
1689                 die("Path '%s' exists on disk, but not in '%.*s'.",
1690                     filename, object_name_len, object_name);
1691         if (is_missing_file_error(errno)) {
1692                 char *fullname = xstrfmt("%s%s", prefix, filename);
1693
1694                 if (!get_tree_entry(r, tree_oid, fullname, &oid, &mode)) {
1695                         die("Path '%s' exists, but not '%s'.\n"
1696                             "Did you mean '%.*s:%s' aka '%.*s:./%s'?",
1697                             fullname,
1698                             filename,
1699                             object_name_len, object_name,
1700                             fullname,
1701                             object_name_len, object_name,
1702                             filename);
1703                 }
1704                 die("Path '%s' does not exist in '%.*s'",
1705                     filename, object_name_len, object_name);
1706         }
1707 }
1708
1709 /* Must be called only when :stage:filename doesn't exist. */
1710 static void diagnose_invalid_index_path(struct repository *r,
1711                                         int stage,
1712                                         const char *prefix,
1713                                         const char *filename)
1714 {
1715         struct index_state *istate = r->index;
1716         const struct cache_entry *ce;
1717         int pos;
1718         unsigned namelen = strlen(filename);
1719         struct strbuf fullname = STRBUF_INIT;
1720
1721         if (!prefix)
1722                 prefix = "";
1723
1724         /* Wrong stage number? */
1725         pos = index_name_pos(istate, filename, namelen);
1726         if (pos < 0)
1727                 pos = -pos - 1;
1728         if (pos < istate->cache_nr) {
1729                 ce = istate->cache[pos];
1730                 if (ce_namelen(ce) == namelen &&
1731                     !memcmp(ce->name, filename, namelen))
1732                         die("Path '%s' is in the index, but not at stage %d.\n"
1733                             "Did you mean ':%d:%s'?",
1734                             filename, stage,
1735                             ce_stage(ce), filename);
1736         }
1737
1738         /* Confusion between relative and absolute filenames? */
1739         strbuf_addstr(&fullname, prefix);
1740         strbuf_addstr(&fullname, filename);
1741         pos = index_name_pos(istate, fullname.buf, fullname.len);
1742         if (pos < 0)
1743                 pos = -pos - 1;
1744         if (pos < istate->cache_nr) {
1745                 ce = istate->cache[pos];
1746                 if (ce_namelen(ce) == fullname.len &&
1747                     !memcmp(ce->name, fullname.buf, fullname.len))
1748                         die("Path '%s' is in the index, but not '%s'.\n"
1749                             "Did you mean ':%d:%s' aka ':%d:./%s'?",
1750                             fullname.buf, filename,
1751                             ce_stage(ce), fullname.buf,
1752                             ce_stage(ce), filename);
1753         }
1754
1755         if (repo_file_exists(r, filename))
1756                 die("Path '%s' exists on disk, but not in the index.", filename);
1757         if (is_missing_file_error(errno))
1758                 die("Path '%s' does not exist (neither on disk nor in the index).",
1759                     filename);
1760
1761         strbuf_release(&fullname);
1762 }
1763
1764
1765 static char *resolve_relative_path(struct repository *r, const char *rel)
1766 {
1767         if (!starts_with(rel, "./") && !starts_with(rel, "../"))
1768                 return NULL;
1769
1770         if (r != the_repository || !is_inside_work_tree())
1771                 die("relative path syntax can't be used outside working tree.");
1772
1773         /* die() inside prefix_path() if resolved path is outside worktree */
1774         return prefix_path(startup_info->prefix,
1775                            startup_info->prefix ? strlen(startup_info->prefix) : 0,
1776                            rel);
1777 }
1778
1779 static enum get_oid_result get_oid_with_context_1(struct repository *repo,
1780                                   const char *name,
1781                                   unsigned flags,
1782                                   const char *prefix,
1783                                   struct object_id *oid,
1784                                   struct object_context *oc)
1785 {
1786         int ret, bracket_depth;
1787         int namelen = strlen(name);
1788         const char *cp;
1789         int only_to_die = flags & GET_OID_ONLY_TO_DIE;
1790
1791         if (only_to_die)
1792                 flags |= GET_OID_QUIETLY;
1793
1794         memset(oc, 0, sizeof(*oc));
1795         oc->mode = S_IFINVALID;
1796         strbuf_init(&oc->symlink_path, 0);
1797         ret = get_oid_1(repo, name, namelen, oid, flags);
1798         if (!ret)
1799                 return ret;
1800         /*
1801          * sha1:path --> object name of path in ent sha1
1802          * :path -> object name of absolute path in index
1803          * :./path -> object name of path relative to cwd in index
1804          * :[0-3]:path -> object name of path in index at stage
1805          * :/foo -> recent commit matching foo
1806          */
1807         if (name[0] == ':') {
1808                 int stage = 0;
1809                 const struct cache_entry *ce;
1810                 char *new_path = NULL;
1811                 int pos;
1812                 if (!only_to_die && namelen > 2 && name[1] == '/') {
1813                         struct handle_one_ref_cb cb;
1814                         struct commit_list *list = NULL;
1815
1816                         cb.repo = repo;
1817                         cb.list = &list;
1818                         refs_for_each_ref(repo->refs, handle_one_ref, &cb);
1819                         refs_head_ref(repo->refs, handle_one_ref, &cb);
1820                         commit_list_sort_by_date(&list);
1821                         return get_oid_oneline(repo, name + 2, oid, list);
1822                 }
1823                 if (namelen < 3 ||
1824                     name[2] != ':' ||
1825                     name[1] < '0' || '3' < name[1])
1826                         cp = name + 1;
1827                 else {
1828                         stage = name[1] - '0';
1829                         cp = name + 3;
1830                 }
1831                 new_path = resolve_relative_path(repo, cp);
1832                 if (!new_path) {
1833                         namelen = namelen - (cp - name);
1834                 } else {
1835                         cp = new_path;
1836                         namelen = strlen(cp);
1837                 }
1838
1839                 if (flags & GET_OID_RECORD_PATH)
1840                         oc->path = xstrdup(cp);
1841
1842                 if (!repo->index || !repo->index->cache)
1843                         repo_read_index(repo);
1844                 pos = index_name_pos(repo->index, cp, namelen);
1845                 if (pos < 0)
1846                         pos = -pos - 1;
1847                 while (pos < repo->index->cache_nr) {
1848                         ce = repo->index->cache[pos];
1849                         if (ce_namelen(ce) != namelen ||
1850                             memcmp(ce->name, cp, namelen))
1851                                 break;
1852                         if (ce_stage(ce) == stage) {
1853                                 oidcpy(oid, &ce->oid);
1854                                 oc->mode = ce->ce_mode;
1855                                 free(new_path);
1856                                 return 0;
1857                         }
1858                         pos++;
1859                 }
1860                 if (only_to_die && name[1] && name[1] != '/')
1861                         diagnose_invalid_index_path(repo, stage, prefix, cp);
1862                 free(new_path);
1863                 return -1;
1864         }
1865         for (cp = name, bracket_depth = 0; *cp; cp++) {
1866                 if (*cp == '{')
1867                         bracket_depth++;
1868                 else if (bracket_depth && *cp == '}')
1869                         bracket_depth--;
1870                 else if (!bracket_depth && *cp == ':')
1871                         break;
1872         }
1873         if (*cp == ':') {
1874                 struct object_id tree_oid;
1875                 int len = cp - name;
1876                 unsigned sub_flags = flags;
1877
1878                 sub_flags &= ~GET_OID_DISAMBIGUATORS;
1879                 sub_flags |= GET_OID_TREEISH;
1880
1881                 if (!get_oid_1(repo, name, len, &tree_oid, sub_flags)) {
1882                         const char *filename = cp+1;
1883                         char *new_filename = NULL;
1884
1885                         new_filename = resolve_relative_path(repo, filename);
1886                         if (new_filename)
1887                                 filename = new_filename;
1888                         if (flags & GET_OID_FOLLOW_SYMLINKS) {
1889                                 ret = get_tree_entry_follow_symlinks(repo, &tree_oid,
1890                                         filename, oid, &oc->symlink_path,
1891                                         &oc->mode);
1892                         } else {
1893                                 ret = get_tree_entry(repo, &tree_oid, filename, oid,
1894                                                      &oc->mode);
1895                                 if (ret && only_to_die) {
1896                                         diagnose_invalid_oid_path(repo, prefix,
1897                                                                    filename,
1898                                                                    &tree_oid,
1899                                                                    name, len);
1900                                 }
1901                         }
1902                         if (flags & GET_OID_RECORD_PATH)
1903                                 oc->path = xstrdup(filename);
1904
1905                         free(new_filename);
1906                         return ret;
1907                 } else {
1908                         if (only_to_die)
1909                                 die("Invalid object name '%.*s'.", len, name);
1910                 }
1911         }
1912         return ret;
1913 }
1914
1915 /*
1916  * Call this function when you know "name" given by the end user must
1917  * name an object but it doesn't; the function _may_ die with a better
1918  * diagnostic message than "no such object 'name'", e.g. "Path 'doc' does not
1919  * exist in 'HEAD'" when given "HEAD:doc", or it may return in which case
1920  * you have a chance to diagnose the error further.
1921  */
1922 void maybe_die_on_misspelt_object_name(struct repository *r,
1923                                        const char *name,
1924                                        const char *prefix)
1925 {
1926         struct object_context oc;
1927         struct object_id oid;
1928         get_oid_with_context_1(r, name, GET_OID_ONLY_TO_DIE,
1929                                prefix, &oid, &oc);
1930 }
1931
1932 enum get_oid_result get_oid_with_context(struct repository *repo,
1933                                          const char *str,
1934                                          unsigned flags,
1935                                          struct object_id *oid,
1936                                          struct object_context *oc)
1937 {
1938         if (flags & GET_OID_FOLLOW_SYMLINKS && flags & GET_OID_ONLY_TO_DIE)
1939                 BUG("incompatible flags for get_sha1_with_context");
1940         return get_oid_with_context_1(repo, str, flags, NULL, oid, oc);
1941 }