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 struct repository *sort_ambiguous_repo;
 407 static int sort_ambiguous(const void *a, const void *b)
 408 {
 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         /* mutex will be needed if this code is to be made thread safe */
 438         sort_ambiguous_repo = r;
 439         QSORT(a->oid, a->nr, sort_ambiguous);
 440         sort_ambiguous_repo = NULL;
 441 }
 442
 443 static enum get_oid_result get_short_oid(struct repository *r,
 444                                          const char *name, int len,
 445                                          struct object_id *oid,
 446                                          unsigned flags)
 447 {
 448         int status;
 449         struct disambiguate_state ds;
 450         int quietly = !!(flags & GET_OID_QUIETLY);
 451
 452         if (init_object_disambiguation(r, name, len, &ds) < 0)
 453                 return -1;
 454
 455         if (HAS_MULTI_BITS(flags & GET_OID_DISAMBIGUATORS))
 456                 BUG("multiple get_short_oid disambiguator flags");
 457
 458         if (flags & GET_OID_COMMIT)
 459                 ds.fn = disambiguate_commit_only;
 460         else if (flags & GET_OID_COMMITTISH)
 461                 ds.fn = disambiguate_committish_only;
 462         else if (flags & GET_OID_TREE)
 463                 ds.fn = disambiguate_tree_only;
 464         else if (flags & GET_OID_TREEISH)
 465                 ds.fn = disambiguate_treeish_only;
 466         else if (flags & GET_OID_BLOB)
 467                 ds.fn = disambiguate_blob_only;
 468         else
 469                 ds.fn = default_disambiguate_hint;
 470
 471         find_short_object_filename(&ds);
 472         find_short_packed_object(&ds);
 473         status = finish_object_disambiguation(&ds, oid);
 474
 475         /*
 476          * If we didn't find it, do the usual reprepare() slow-path,
 477          * since the object may have recently been added to the repository
 478          * or migrated from loose to packed.
 479          */
 480         if (status == MISSING_OBJECT) {
 481                 reprepare_packed_git(r);
 482                 find_short_object_filename(&ds);
 483                 find_short_packed_object(&ds);
 484                 status = finish_object_disambiguation(&ds, oid);
 485         }
 486
 487         if (!quietly && (status == SHORT_NAME_AMBIGUOUS)) {
 488                 struct oid_array collect = OID_ARRAY_INIT;
 489
 490                 error(_("short SHA1 %s is ambiguous"), ds.hex_pfx);
 491
 492                 /*
 493                  * We may still have ambiguity if we simply saw a series of
 494                  * candidates that did not satisfy our hint function. In
 495                  * that case, we still want to show them, so disable the hint
 496                  * function entirely.
 497                  */
 498                 if (!ds.ambiguous)
 499                         ds.fn = NULL;
 500
 501                 advise(_("The candidates are:"));
 502                 repo_for_each_abbrev(r, ds.hex_pfx, collect_ambiguous, &collect);
 503                 sort_ambiguous_oid_array(r, &collect);
 504
 505                 if (oid_array_for_each(&collect, show_ambiguous_object, &ds))
 506                         BUG("show_ambiguous_object shouldn't return non-zero");
 507                 oid_array_clear(&collect);
 508         }
 509
 510         return status;
 511 }
 512
 513 int repo_for_each_abbrev(struct repository *r, const char *prefix,
 514                          each_abbrev_fn fn, void *cb_data)
 515 {
 516         struct oid_array collect = OID_ARRAY_INIT;
 517         struct disambiguate_state ds;
 518         int ret;
 519
 520         if (init_object_disambiguation(r, prefix, strlen(prefix), &ds) < 0)
 521                 return -1;
 522
 523         ds.always_call_fn = 1;
 524         ds.fn = repo_collect_ambiguous;
 525         ds.cb_data = &collect;
 526         find_short_object_filename(&ds);
 527         find_short_packed_object(&ds);
 528
 529         ret = oid_array_for_each_unique(&collect, fn, cb_data);
 530         oid_array_clear(&collect);
 531         return ret;
 532 }
 533
 534 /*
 535  * Return the slot of the most-significant bit set in "val". There are various
 536  * ways to do this quickly with fls() or __builtin_clzl(), but speed is
 537  * probably not a big deal here.
 538  */
 539 static unsigned msb(unsigned long val)
 540 {
 541         unsigned r = 0;
 542         while (val >>= 1)
 543                 r++;
 544         return r;
 545 }
 546
 547 struct min_abbrev_data {
 548         unsigned int init_len;
 549         unsigned int cur_len;
 550         char *hex;
 551         struct repository *repo;
 552         const struct object_id *oid;
 553 };
 554
 555 static inline char get_hex_char_from_oid(const struct object_id *oid,
 556                                          unsigned int pos)
 557 {
 558         static const char hex[] = "0123456789abcdef";
 559
 560         if ((pos & 1) == 0)
 561                 return hex[oid->hash[pos >> 1] >> 4];
 562         else
 563                 return hex[oid->hash[pos >> 1] & 0xf];
 564 }
 565
 566 static int extend_abbrev_len(const struct object_id *oid, void *cb_data)
 567 {
 568         struct min_abbrev_data *mad = cb_data;
 569
 570         unsigned int i = mad->init_len;
 571         while (mad->hex[i] && mad->hex[i] == get_hex_char_from_oid(oid, i))
 572                 i++;
 573
 574         if (i < GIT_MAX_RAWSZ && i >= mad->cur_len)
 575                 mad->cur_len = i + 1;
 576
 577         return 0;
 578 }
 579
 580 static int repo_extend_abbrev_len(struct repository *r,
 581                                   const struct object_id *oid,
 582                                   void *cb_data)
 583 {
 584         return extend_abbrev_len(oid, cb_data);
 585 }
 586
 587 static void find_abbrev_len_for_midx(struct multi_pack_index *m,
 588                                      struct min_abbrev_data *mad)
 589 {
 590         int match = 0;
 591         uint32_t num, first = 0;
 592         struct object_id oid;
 593         const struct object_id *mad_oid;
 594
 595         if (!m->num_objects)
 596                 return;
 597
 598         num = m->num_objects;
 599         mad_oid = mad->oid;
 600         match = bsearch_midx(mad_oid, m, &first);
 601
 602         /*
 603          * first is now the position in the packfile where we would insert
 604          * mad->hash if it does not exist (or the position of mad->hash if
 605          * it does exist). Hence, we consider a maximum of two objects
 606          * nearby for the abbreviation length.
 607          */
 608         mad->init_len = 0;
 609         if (!match) {
 610                 if (nth_midxed_object_oid(&oid, m, first))
 611                         extend_abbrev_len(&oid, mad);
 612         } else if (first < num - 1) {
 613                 if (nth_midxed_object_oid(&oid, m, first + 1))
 614                         extend_abbrev_len(&oid, mad);
 615         }
 616         if (first > 0) {
 617                 if (nth_midxed_object_oid(&oid, m, first - 1))
 618                         extend_abbrev_len(&oid, mad);
 619         }
 620         mad->init_len = mad->cur_len;
 621 }
 622
 623 static void find_abbrev_len_for_pack(struct packed_git *p,
 624                                      struct min_abbrev_data *mad)
 625 {
 626         int match = 0;
 627         uint32_t num, first = 0;
 628         struct object_id oid;
 629         const struct object_id *mad_oid;
 630
 631         if (p->multi_pack_index)
 632                 return;
 633
 634         if (open_pack_index(p) || !p->num_objects)
 635                 return;
 636
 637         num = p->num_objects;
 638         mad_oid = mad->oid;
 639         match = bsearch_pack(mad_oid, p, &first);
 640
 641         /*
 642          * first is now the position in the packfile where we would insert
 643          * mad->hash if it does not exist (or the position of mad->hash if
 644          * it does exist). Hence, we consider a maximum of two objects
 645          * nearby for the abbreviation length.
 646          */
 647         mad->init_len = 0;
 648         if (!match) {
 649                 if (nth_packed_object_oid(&oid, p, first))
 650                         extend_abbrev_len(&oid, mad);
 651         } else if (first < num - 1) {
 652                 if (nth_packed_object_oid(&oid, p, first + 1))
 653                         extend_abbrev_len(&oid, mad);
 654         }
 655         if (first > 0) {
 656                 if (nth_packed_object_oid(&oid, p, first - 1))
 657                         extend_abbrev_len(&oid, mad);
 658         }
 659         mad->init_len = mad->cur_len;
 660 }
 661
 662 static void find_abbrev_len_packed(struct min_abbrev_data *mad)
 663 {
 664         struct multi_pack_index *m;
 665         struct packed_git *p;
 666
 667         for (m = get_multi_pack_index(mad->repo); m; m = m->next)
 668                 find_abbrev_len_for_midx(m, mad);
 669         for (p = get_packed_git(mad->repo); p; p = p->next)
 670                 find_abbrev_len_for_pack(p, mad);
 671 }
 672
 673 int repo_find_unique_abbrev_r(struct repository *r, char *hex,
 674                               const struct object_id *oid, int len)
 675 {
 676         struct disambiguate_state ds;
 677         struct min_abbrev_data mad;
 678         struct object_id oid_ret;
 679         const unsigned hexsz = r->hash_algo->hexsz;
 680
 681         if (len < 0) {
 682                 unsigned long count = repo_approximate_object_count(r);
 683                 /*
 684                  * Add one because the MSB only tells us the highest bit set,
 685                  * not including the value of all the _other_ bits (so "15"
 686                  * is only one off of 2^4, but the MSB is the 3rd bit.
 687                  */
 688                 len = msb(count) + 1;
 689                 /*
 690                  * We now know we have on the order of 2^len objects, which
 691                  * expects a collision at 2^(len/2). But we also care about hex
 692                  * chars, not bits, and there are 4 bits per hex. So all
 693                  * together we need to divide by 2 and round up.
 694                  */
 695                 len = DIV_ROUND_UP(len, 2);
 696                 /*
 697                  * For very small repos, we stick with our regular fallback.
 698                  */
 699                 if (len < FALLBACK_DEFAULT_ABBREV)
 700                         len = FALLBACK_DEFAULT_ABBREV;
 701         }
 702
 703         oid_to_hex_r(hex, oid);
 704         if (len == hexsz || !len)
 705                 return hexsz;
 706
 707         mad.repo = r;
 708         mad.init_len = len;
 709         mad.cur_len = len;
 710         mad.hex = hex;
 711         mad.oid = oid;
 712
 713         find_abbrev_len_packed(&mad);
 714
 715         if (init_object_disambiguation(r, hex, mad.cur_len, &ds) < 0)
 716                 return -1;
 717
 718         ds.fn = repo_extend_abbrev_len;
 719         ds.always_call_fn = 1;
 720         ds.cb_data = (void *)&mad;
 721
 722         find_short_object_filename(&ds);
 723         (void)finish_object_disambiguation(&ds, &oid_ret);
 724
 725         hex[mad.cur_len] = 0;
 726         return mad.cur_len;
 727 }
 728
 729 const char *repo_find_unique_abbrev(struct repository *r,
 730                                     const struct object_id *oid,
 731                                     int len)
 732 {
 733         static int bufno;
 734         static char hexbuffer[4][GIT_MAX_HEXSZ + 1];
 735         char *hex = hexbuffer[bufno];
 736         bufno = (bufno + 1) % ARRAY_SIZE(hexbuffer);
 737         repo_find_unique_abbrev_r(r, hex, oid, len);
 738         return hex;
 739 }
 740
 741 static int ambiguous_path(const char *path, int len)
 742 {
 743         int slash = 1;
 744         int cnt;
 745
 746         for (cnt = 0; cnt < len; cnt++) {
 747                 switch (*path++) {
 748                 case '\0':
 749                         break;
 750                 case '/':
 751                         if (slash)
 752                                 break;
 753                         slash = 1;
 754                         continue;
 755                 case '.':
 756                         continue;
 757                 default:
 758                         slash = 0;
 759                         continue;
 760                 }
 761                 break;
 762         }
 763         return slash;
 764 }
 765
 766 static inline int at_mark(const char *string, int len,
 767                           const char **suffix, int nr)
 768 {
 769         int i;
 770
 771         for (i = 0; i < nr; i++) {
 772                 int suffix_len = strlen(suffix[i]);
 773                 if (suffix_len <= len
 774                     && !strncasecmp(string, suffix[i], suffix_len))
 775                         return suffix_len;
 776         }
 777         return 0;
 778 }
 779
 780 static inline int upstream_mark(const char *string, int len)
 781 {
 782         const char *suffix[] = { "@{upstream}", "@{u}" };
 783         return at_mark(string, len, suffix, ARRAY_SIZE(suffix));
 784 }
 785
 786 static inline int push_mark(const char *string, int len)
 787 {
 788         const char *suffix[] = { "@{push}" };
 789         return at_mark(string, len, suffix, ARRAY_SIZE(suffix));
 790 }
 791
 792 static enum get_oid_result get_oid_1(struct repository *r, const char *name, int len, struct object_id *oid, unsigned lookup_flags);
 793 static int interpret_nth_prior_checkout(struct repository *r, const char *name, int namelen, struct strbuf *buf);
 794
 795 static int get_oid_basic(struct repository *r, const char *str, int len,
 796                          struct object_id *oid, unsigned int flags)
 797 {
 798         static const char *warn_msg = "refname '%.*s' is ambiguous.";
 799         static const char *object_name_msg = N_(
 800         "Git normally never creates a ref that ends with 40 hex characters\n"
 801         "because it will be ignored when you just specify 40-hex. These refs\n"
 802         "may be created by mistake. For example,\n"
 803         "\n"
 804         "  git switch -c $br $(git rev-parse ...)\n"
 805         "\n"
 806         "where \"$br\" is somehow empty and a 40-hex ref is created. Please\n"
 807         "examine these refs and maybe delete them. Turn this message off by\n"
 808         "running \"git config advice.objectNameWarning false\"");
 809         struct object_id tmp_oid;
 810         char *real_ref = NULL;
 811         int refs_found = 0;
 812         int at, reflog_len, nth_prior = 0;
 813
 814         if (len == r->hash_algo->hexsz && !get_oid_hex(str, oid)) {
 815                 if (warn_ambiguous_refs && warn_on_object_refname_ambiguity) {
 816                         refs_found = repo_dwim_ref(r, str, len, &tmp_oid, &real_ref);
 817                         if (refs_found > 0) {
 818                                 warning(warn_msg, len, str);
 819                                 if (advice_object_name_warning)
 820                                         fprintf(stderr, "%s\n", _(object_name_msg));
 821                         }
 822                         free(real_ref);
 823                 }
 824                 return 0;
 825         }
 826
 827         /* basic@{time or number or -number} format to query ref-log */
 828         reflog_len = at = 0;
 829         if (len && str[len-1] == '}') {
 830                 for (at = len-4; at >= 0; at--) {
 831                         if (str[at] == '@' && str[at+1] == '{') {
 832                                 if (str[at+2] == '-') {
 833                                         if (at != 0)
 834                                                 /* @{-N} not at start */
 835                                                 return -1;
 836                                         nth_prior = 1;
 837                                         continue;
 838                                 }
 839                                 if (!upstream_mark(str + at, len - at) &&
 840                                     !push_mark(str + at, len - at)) {
 841                                         reflog_len = (len-1) - (at+2);
 842                                         len = at;
 843                                 }
 844                                 break;
 845                         }
 846                 }
 847         }
 848
 849         /* Accept only unambiguous ref paths. */
 850         if (len && ambiguous_path(str, len))
 851                 return -1;
 852
 853         if (nth_prior) {
 854                 struct strbuf buf = STRBUF_INIT;
 855                 int detached;
 856
 857                 if (interpret_nth_prior_checkout(r, str, len, &buf) > 0) {
 858                         detached = (buf.len == r->hash_algo->hexsz && !get_oid_hex(buf.buf, oid));
 859                         strbuf_release(&buf);
 860                         if (detached)
 861                                 return 0;
 862                 }
 863         }
 864
 865         if (!len && reflog_len)
 866                 /* allow "@{...}" to mean the current branch reflog */
 867                 refs_found = repo_dwim_ref(r, "HEAD", 4, oid, &real_ref);
 868         else if (reflog_len)
 869                 refs_found = repo_dwim_log(r, str, len, oid, &real_ref);
 870         else
 871                 refs_found = repo_dwim_ref(r, str, len, oid, &real_ref);
 872
 873         if (!refs_found)
 874                 return -1;
 875
 876         if (warn_ambiguous_refs && !(flags & GET_OID_QUIETLY) &&
 877             (refs_found > 1 ||
 878              !get_short_oid(r, str, len, &tmp_oid, GET_OID_QUIETLY)))
 879                 warning(warn_msg, len, str);
 880
 881         if (reflog_len) {
 882                 int nth, i;
 883                 timestamp_t at_time;
 884                 timestamp_t co_time;
 885                 int co_tz, co_cnt;
 886
 887                 /* Is it asking for N-th entry, or approxidate? */
 888                 for (i = nth = 0; 0 <= nth && i < reflog_len; i++) {
 889                         char ch = str[at+2+i];
 890                         if ('0' <= ch && ch <= '9')
 891                                 nth = nth * 10 + ch - '0';
 892                         else
 893                                 nth = -1;
 894                 }
 895                 if (100000000 <= nth) {
 896                         at_time = nth;
 897                         nth = -1;
 898                 } else if (0 <= nth)
 899                         at_time = 0;
 900                 else {
 901                         int errors = 0;
 902                         char *tmp = xstrndup(str + at + 2, reflog_len);
 903                         at_time = approxidate_careful(tmp, &errors);
 904                         free(tmp);
 905                         if (errors) {
 906                                 free(real_ref);
 907                                 return -1;
 908                         }
 909                 }
 910                 if (read_ref_at(get_main_ref_store(r),
 911                                 real_ref, flags, at_time, nth, oid, NULL,
 912                                 &co_time, &co_tz, &co_cnt)) {
 913                         if (!len) {
 914                                 if (starts_with(real_ref, "refs/heads/")) {
 915                                         str = real_ref + 11;
 916                                         len = strlen(real_ref + 11);
 917                                 } else {
 918                                         /* detached HEAD */
 919                                         str = "HEAD";
 920                                         len = 4;
 921                                 }
 922                         }
 923                         if (at_time) {
 924                                 if (!(flags & GET_OID_QUIETLY)) {
 925                                         warning("Log for '%.*s' only goes "
 926                                                 "back to %s.", len, str,
 927                                                 show_date(co_time, co_tz, DATE_MODE(RFC2822)));
 928                                 }
 929                         } else {
 930                                 if (flags & GET_OID_QUIETLY) {
 931                                         exit(128);
 932                                 }
 933                                 die("Log for '%.*s' only has %d entries.",
 934                                     len, str, co_cnt);
 935                         }
 936                 }
 937         }
 938
 939         free(real_ref);
 940         return 0;
 941 }
 942
 943 static enum get_oid_result get_parent(struct repository *r,
 944                                       const char *name, int len,
 945                                       struct object_id *result, int idx)
 946 {
 947         struct object_id oid;
 948         enum get_oid_result ret = get_oid_1(r, name, len, &oid,
 949                                             GET_OID_COMMITTISH);
 950         struct commit *commit;
 951         struct commit_list *p;
 952
 953         if (ret)
 954                 return ret;
 955         commit = lookup_commit_reference(r, &oid);
 956         if (parse_commit(commit))
 957                 return MISSING_OBJECT;
 958         if (!idx) {
 959                 oidcpy(result, &commit->object.oid);
 960                 return FOUND;
 961         }
 962         p = commit->parents;
 963         while (p) {
 964                 if (!--idx) {
 965                         oidcpy(result, &p->item->object.oid);
 966                         return FOUND;
 967                 }
 968                 p = p->next;
 969         }
 970         return MISSING_OBJECT;
 971 }
 972
 973 static enum get_oid_result get_nth_ancestor(struct repository *r,
 974                                             const char *name, int len,
 975                                             struct object_id *result,
 976                                             int generation)
 977 {
 978         struct object_id oid;
 979         struct commit *commit;
 980         int ret;
 981
 982         ret = get_oid_1(r, name, len, &oid, GET_OID_COMMITTISH);
 983         if (ret)
 984                 return ret;
 985         commit = lookup_commit_reference(r, &oid);
 986         if (!commit)
 987                 return MISSING_OBJECT;
 988
 989         while (generation--) {
 990                 if (parse_commit(commit) || !commit->parents)
 991                         return MISSING_OBJECT;
 992                 commit = commit->parents->item;
 993         }
 994         oidcpy(result, &commit->object.oid);
 995         return FOUND;
 996 }
 997
 998 struct object *repo_peel_to_type(struct repository *r, const char *name, int namelen,
 999                                  struct object *o, enum object_type expected_type)
1000 {
1001         if (name && !namelen)
1002                 namelen = strlen(name);
1003         while (1) {
1004                 if (!o || (!o->parsed && !parse_object(r, &o->oid)))
1005                         return NULL;
1006                 if (expected_type == OBJ_ANY || o->type == expected_type)
1007                         return o;
1008                 if (o->type == OBJ_TAG)
1009                         o = ((struct tag*) o)->tagged;
1010                 else if (o->type == OBJ_COMMIT)
1011                         o = &(repo_get_commit_tree(r, ((struct commit *)o))->object);
1012                 else {
1013                         if (name)
1014                                 error("%.*s: expected %s type, but the object "
1015                                       "dereferences to %s type",
1016                                       namelen, name, type_name(expected_type),
1017                                       type_name(o->type));
1018                         return NULL;
1019                 }
1020         }
1021 }
1022
1023 static int peel_onion(struct repository *r, const char *name, int len,
1024                       struct object_id *oid, unsigned lookup_flags)
1025 {
1026         struct object_id outer;
1027         const char *sp;
1028         unsigned int expected_type = 0;
1029         struct object *o;
1030
1031         /*
1032          * "ref^{type}" dereferences ref repeatedly until you cannot
1033          * dereference anymore, or you get an object of given type,
1034          * whichever comes first.  "ref^{}" means just dereference
1035          * tags until you get a non-tag.  "ref^0" is a shorthand for
1036          * "ref^{commit}".  "commit^{tree}" could be used to find the
1037          * top-level tree of the given commit.
1038          */
1039         if (len < 4 || name[len-1] != '}')
1040                 return -1;
1041
1042         for (sp = name + len - 1; name <= sp; sp--) {
1043                 int ch = *sp;
1044                 if (ch == '{' && name < sp && sp[-1] == '^')
1045                         break;
1046         }
1047         if (sp <= name)
1048                 return -1;
1049
1050         sp++; /* beginning of type name, or closing brace for empty */
1051         if (starts_with(sp, "commit}"))
1052                 expected_type = OBJ_COMMIT;
1053         else if (starts_with(sp, "tag}"))
1054                 expected_type = OBJ_TAG;
1055         else if (starts_with(sp, "tree}"))
1056                 expected_type = OBJ_TREE;
1057         else if (starts_with(sp, "blob}"))
1058                 expected_type = OBJ_BLOB;
1059         else if (starts_with(sp, "object}"))
1060                 expected_type = OBJ_ANY;
1061         else if (sp[0] == '}')
1062                 expected_type = OBJ_NONE;
1063         else if (sp[0] == '/')
1064                 expected_type = OBJ_COMMIT;
1065         else
1066                 return -1;
1067
1068         lookup_flags &= ~GET_OID_DISAMBIGUATORS;
1069         if (expected_type == OBJ_COMMIT)
1070                 lookup_flags |= GET_OID_COMMITTISH;
1071         else if (expected_type == OBJ_TREE)
1072                 lookup_flags |= GET_OID_TREEISH;
1073
1074         if (get_oid_1(r, name, sp - name - 2, &outer, lookup_flags))
1075                 return -1;
1076
1077         o = parse_object(r, &outer);
1078         if (!o)
1079                 return -1;
1080         if (!expected_type) {
1081                 o = deref_tag(r, o, name, sp - name - 2);
1082                 if (!o || (!o->parsed && !parse_object(r, &o->oid)))
1083                         return -1;
1084                 oidcpy(oid, &o->oid);
1085                 return 0;
1086         }
1087
1088         /*
1089          * At this point, the syntax look correct, so
1090          * if we do not get the needed object, we should
1091          * barf.
1092          */
1093         o = repo_peel_to_type(r, name, len, o, expected_type);
1094         if (!o)
1095                 return -1;
1096
1097         oidcpy(oid, &o->oid);
1098         if (sp[0] == '/') {
1099                 /* "$commit^{/foo}" */
1100                 char *prefix;
1101                 int ret;
1102                 struct commit_list *list = NULL;
1103
1104                 /*
1105                  * $commit^{/}. Some regex implementation may reject.
1106                  * We don't need regex anyway. '' pattern always matches.
1107                  */
1108                 if (sp[1] == '}')
1109                         return 0;
1110
1111                 prefix = xstrndup(sp + 1, name + len - 1 - (sp + 1));
1112                 commit_list_insert((struct commit *)o, &list);
1113                 ret = get_oid_oneline(r, prefix, oid, list);
1114                 free(prefix);
1115                 return ret;
1116         }
1117         return 0;
1118 }
1119
1120 static int get_describe_name(struct repository *r,
1121                              const char *name, int len,
1122                              struct object_id *oid)
1123 {
1124         const char *cp;
1125         unsigned flags = GET_OID_QUIETLY | GET_OID_COMMIT;
1126
1127         for (cp = name + len - 1; name + 2 <= cp; cp--) {
1128                 char ch = *cp;
1129                 if (!isxdigit(ch)) {
1130                         /* We must be looking at g in "SOMETHING-g"
1131                          * for it to be describe output.
1132                          */
1133                         if (ch == 'g' && cp[-1] == '-') {
1134                                 cp++;
1135                                 len -= cp - name;
1136                                 return get_short_oid(r,
1137                                                      cp, len, oid, flags);
1138                         }
1139                 }
1140         }
1141         return -1;
1142 }
1143
1144 static enum get_oid_result get_oid_1(struct repository *r,
1145                                      const char *name, int len,
1146                                      struct object_id *oid,
1147                                      unsigned lookup_flags)
1148 {
1149         int ret, has_suffix;
1150         const char *cp;
1151
1152         /*
1153          * "name~3" is "name^^^", "name~" is "name~1", and "name^" is "name^1".
1154          */
1155         has_suffix = 0;
1156         for (cp = name + len - 1; name <= cp; cp--) {
1157                 int ch = *cp;
1158                 if ('0' <= ch && ch <= '9')
1159                         continue;
1160                 if (ch == '~' || ch == '^')
1161                         has_suffix = ch;
1162                 break;
1163         }
1164
1165         if (has_suffix) {
1166                 int num = 0;
1167                 int len1 = cp - name;
1168                 cp++;
1169                 while (cp < name + len)
1170                         num = num * 10 + *cp++ - '0';
1171                 if (!num && len1 == len - 1)
1172                         num = 1;
1173                 if (has_suffix == '^')
1174                         return get_parent(r, name, len1, oid, num);
1175                 /* else if (has_suffix == '~') -- goes without saying */
1176                 return get_nth_ancestor(r, name, len1, oid, num);
1177         }
1178
1179         ret = peel_onion(r, name, len, oid, lookup_flags);
1180         if (!ret)
1181                 return FOUND;
1182
1183         ret = get_oid_basic(r, name, len, oid, lookup_flags);
1184         if (!ret)
1185                 return FOUND;
1186
1187         /* It could be describe output that is "SOMETHING-gXXXX" */
1188         ret = get_describe_name(r, name, len, oid);
1189         if (!ret)
1190                 return FOUND;
1191
1192         return get_short_oid(r, name, len, oid, lookup_flags);
1193 }
1194
1195 /*
1196  * This interprets names like ':/Initial revision of "git"' by searching
1197  * through history and returning the first commit whose message starts
1198  * the given regular expression.
1199  *
1200  * For negative-matching, prefix the pattern-part with '!-', like: ':/!-WIP'.
1201  *
1202  * For a literal '!' character at the beginning of a pattern, you have to repeat
1203  * that, like: ':/!!foo'
1204  *
1205  * For future extension, all other sequences beginning with ':/!' are reserved.
1206  */
1207
1208 /* Remember to update object flag allocation in object.h */
1209 #define ONELINE_SEEN (1u<<20)
1210
1211 struct handle_one_ref_cb {
1212         struct repository *repo;
1213         struct commit_list **list;
1214 };
1215
1216 static int handle_one_ref(const char *path, const struct object_id *oid,
1217                           int flag, void *cb_data)
1218 {
1219         struct handle_one_ref_cb *cb = cb_data;
1220         struct commit_list **list = cb->list;
1221         struct object *object = parse_object(cb->repo, oid);
1222         if (!object)
1223                 return 0;
1224         if (object->type == OBJ_TAG) {
1225                 object = deref_tag(cb->repo, object, path,
1226                                    strlen(path));
1227                 if (!object)
1228                         return 0;
1229         }
1230         if (object->type != OBJ_COMMIT)
1231                 return 0;
1232         commit_list_insert((struct commit *)object, list);
1233         return 0;
1234 }
1235
1236 static int get_oid_oneline(struct repository *r,
1237                            const char *prefix, struct object_id *oid,
1238                            struct commit_list *list)
1239 {
1240         struct commit_list *backup = NULL, *l;
1241         int found = 0;
1242         int negative = 0;
1243         regex_t regex;
1244
1245         if (prefix[0] == '!') {
1246                 prefix++;
1247
1248                 if (prefix[0] == '-') {
1249                         prefix++;
1250                         negative = 1;
1251                 } else if (prefix[0] != '!') {
1252                         return -1;
1253                 }
1254         }
1255
1256         if (regcomp(&regex, prefix, REG_EXTENDED))
1257                 return -1;
1258
1259         for (l = list; l; l = l->next) {
1260                 l->item->object.flags |= ONELINE_SEEN;
1261                 commit_list_insert(l->item, &backup);
1262         }
1263         while (list) {
1264                 const char *p, *buf;
1265                 struct commit *commit;
1266                 int matches;
1267
1268                 commit = pop_most_recent_commit(&list, ONELINE_SEEN);
1269                 if (!parse_object(r, &commit->object.oid))
1270                         continue;
1271                 buf = get_commit_buffer(commit, NULL);
1272                 p = strstr(buf, "\n\n");
1273                 matches = negative ^ (p && !regexec(&regex, p + 2, 0, NULL, 0));
1274                 unuse_commit_buffer(commit, buf);
1275
1276                 if (matches) {
1277                         oidcpy(oid, &commit->object.oid);
1278                         found = 1;
1279                         break;
1280                 }
1281         }
1282         regfree(&regex);
1283         free_commit_list(list);
1284         for (l = backup; l; l = l->next)
1285                 clear_commit_marks(l->item, ONELINE_SEEN);
1286         free_commit_list(backup);
1287         return found ? 0 : -1;
1288 }
1289
1290 struct grab_nth_branch_switch_cbdata {
1291         int remaining;
1292         struct strbuf buf;
1293 };
1294
1295 static int grab_nth_branch_switch(struct object_id *ooid, struct object_id *noid,
1296                                   const char *email, timestamp_t timestamp, int tz,
1297                                   const char *message, void *cb_data)
1298 {
1299         struct grab_nth_branch_switch_cbdata *cb = cb_data;
1300         const char *match = NULL, *target = NULL;
1301         size_t len;
1302
1303         if (skip_prefix(message, "checkout: moving from ", &match))
1304                 target = strstr(match, " to ");
1305
1306         if (!match || !target)
1307                 return 0;
1308         if (--(cb->remaining) == 0) {
1309                 len = target - match;
1310                 strbuf_reset(&cb->buf);
1311                 strbuf_add(&cb->buf, match, len);
1312                 return 1; /* we are done */
1313         }
1314         return 0;
1315 }
1316
1317 /*
1318  * Parse @{-N} syntax, return the number of characters parsed
1319  * if successful; otherwise signal an error with negative value.
1320  */
1321 static int interpret_nth_prior_checkout(struct repository *r,
1322                                         const char *name, int namelen,
1323                                         struct strbuf *buf)
1324 {
1325         long nth;
1326         int retval;
1327         struct grab_nth_branch_switch_cbdata cb;
1328         const char *brace;
1329         char *num_end;
1330
1331         if (namelen < 4)
1332                 return -1;
1333         if (name[0] != '@' || name[1] != '{' || name[2] != '-')
1334                 return -1;
1335         brace = memchr(name, '}', namelen);
1336         if (!brace)
1337                 return -1;
1338         nth = strtol(name + 3, &num_end, 10);
1339         if (num_end != brace)
1340                 return -1;
1341         if (nth <= 0)
1342                 return -1;
1343         cb.remaining = nth;
1344         strbuf_init(&cb.buf, 20);
1345
1346         retval = refs_for_each_reflog_ent_reverse(get_main_ref_store(r),
1347                         "HEAD", grab_nth_branch_switch, &cb);
1348         if (0 < retval) {
1349                 strbuf_reset(buf);
1350                 strbuf_addbuf(buf, &cb.buf);
1351                 retval = brace - name + 1;
1352         } else
1353                 retval = 0;
1354
1355         strbuf_release(&cb.buf);
1356         return retval;
1357 }
1358
1359 int repo_get_oid_mb(struct repository *r,
1360                     const char *name,
1361                     struct object_id *oid)
1362 {
1363         struct commit *one, *two;
1364         struct commit_list *mbs;
1365         struct object_id oid_tmp;
1366         const char *dots;
1367         int st;
1368
1369         dots = strstr(name, "...");
1370         if (!dots)
1371                 return repo_get_oid(r, name, oid);
1372         if (dots == name)
1373                 st = repo_get_oid(r, "HEAD", &oid_tmp);
1374         else {
1375                 struct strbuf sb;
1376                 strbuf_init(&sb, dots - name);
1377                 strbuf_add(&sb, name, dots - name);
1378                 st = repo_get_oid_committish(r, sb.buf, &oid_tmp);
1379                 strbuf_release(&sb);
1380         }
1381         if (st)
1382                 return st;
1383         one = lookup_commit_reference_gently(r, &oid_tmp, 0);
1384         if (!one)
1385                 return -1;
1386
1387         if (repo_get_oid_committish(r, dots[3] ? (dots + 3) : "HEAD", &oid_tmp))
1388                 return -1;
1389         two = lookup_commit_reference_gently(r, &oid_tmp, 0);
1390         if (!two)
1391                 return -1;
1392         mbs = repo_get_merge_bases(r, one, two);
1393         if (!mbs || mbs->next)
1394                 st = -1;
1395         else {
1396                 st = 0;
1397                 oidcpy(oid, &mbs->item->object.oid);
1398         }
1399         free_commit_list(mbs);
1400         return st;
1401 }
1402
1403 /* parse @something syntax, when 'something' is not {.*} */
1404 static int interpret_empty_at(const char *name, int namelen, int len, struct strbuf *buf)
1405 {
1406         const char *next;
1407
1408         if (len || name[1] == '{')
1409                 return -1;
1410
1411         /* make sure it's a single @, or @@{.*}, not @foo */
1412         next = memchr(name + len + 1, '@', namelen - len - 1);
1413         if (next && next[1] != '{')
1414                 return -1;
1415         if (!next)
1416                 next = name + namelen;
1417         if (next != name + 1)
1418                 return -1;
1419
1420         strbuf_reset(buf);
1421         strbuf_add(buf, "HEAD", 4);
1422         return 1;
1423 }
1424
1425 static int reinterpret(struct repository *r,
1426                        const char *name, int namelen, int len,
1427                        struct strbuf *buf, unsigned allowed)
1428 {
1429         /* we have extra data, which might need further processing */
1430         struct strbuf tmp = STRBUF_INIT;
1431         int used = buf->len;
1432         int ret;
1433
1434         strbuf_add(buf, name + len, namelen - len);
1435         ret = repo_interpret_branch_name(r, buf->buf, buf->len, &tmp, allowed);
1436         /* that data was not interpreted, remove our cruft */
1437         if (ret < 0) {
1438                 strbuf_setlen(buf, used);
1439                 return len;
1440         }
1441         strbuf_reset(buf);
1442         strbuf_addbuf(buf, &tmp);
1443         strbuf_release(&tmp);
1444         /* tweak for size of {-N} versus expanded ref name */
1445         return ret - used + len;
1446 }
1447
1448 static void set_shortened_ref(struct repository *r, struct strbuf *buf, const char *ref)
1449 {
1450         char *s = refs_shorten_unambiguous_ref(get_main_ref_store(r), ref, 0);
1451         strbuf_reset(buf);
1452         strbuf_addstr(buf, s);
1453         free(s);
1454 }
1455
1456 static int branch_interpret_allowed(const char *refname, unsigned allowed)
1457 {
1458         if (!allowed)
1459                 return 1;
1460
1461         if ((allowed & INTERPRET_BRANCH_LOCAL) &&
1462             starts_with(refname, "refs/heads/"))
1463                 return 1;
1464         if ((allowed & INTERPRET_BRANCH_REMOTE) &&
1465             starts_with(refname, "refs/remotes/"))
1466                 return 1;
1467
1468         return 0;
1469 }
1470
1471 static int interpret_branch_mark(struct repository *r,
1472                                  const char *name, int namelen,
1473                                  int at, struct strbuf *buf,
1474                                  int (*get_mark)(const char *, int),
1475                                  const char *(*get_data)(struct branch *,
1476                                                          struct strbuf *),
1477                                  unsigned allowed)
1478 {
1479         int len;
1480         struct branch *branch;
1481         struct strbuf err = STRBUF_INIT;
1482         const char *value;
1483
1484         len = get_mark(name + at, namelen - at);
1485         if (!len)
1486                 return -1;
1487
1488         if (memchr(name, ':', at))
1489                 return -1;
1490
1491         if (at) {
1492                 char *name_str = xmemdupz(name, at);
1493                 branch = branch_get(name_str);
1494                 free(name_str);
1495         } else
1496                 branch = branch_get(NULL);
1497
1498         value = get_data(branch, &err);
1499         if (!value)
1500                 die("%s", err.buf);
1501
1502         if (!branch_interpret_allowed(value, allowed))
1503                 return -1;
1504
1505         set_shortened_ref(r, buf, value);
1506         return len + at;
1507 }
1508
1509 int repo_interpret_branch_name(struct repository *r,
1510                                const char *name, int namelen,
1511                                struct strbuf *buf,
1512                                unsigned allowed)
1513 {
1514         char *at;
1515         const char *start;
1516         int len;
1517
1518         if (!namelen)
1519                 namelen = strlen(name);
1520
1521         if (!allowed || (allowed & INTERPRET_BRANCH_LOCAL)) {
1522                 len = interpret_nth_prior_checkout(r, name, namelen, buf);
1523                 if (!len) {
1524                         return len; /* syntax Ok, not enough switches */
1525                 } else if (len > 0) {
1526                         if (len == namelen)
1527                                 return len; /* consumed all */
1528                         else
1529                                 return reinterpret(r, name, namelen, len, buf, allowed);
1530                 }
1531         }
1532
1533         for (start = name;
1534              (at = memchr(start, '@', namelen - (start - name)));
1535              start = at + 1) {
1536
1537                 if (!allowed || (allowed & INTERPRET_BRANCH_HEAD)) {
1538                         len = interpret_empty_at(name, namelen, at - name, buf);
1539                         if (len > 0)
1540                                 return reinterpret(r, name, namelen, len, buf,
1541                                                    allowed);
1542                 }
1543
1544                 len = interpret_branch_mark(r, name, namelen, at - name, buf,
1545                                             upstream_mark, branch_get_upstream,
1546                                             allowed);
1547                 if (len > 0)
1548                         return len;
1549
1550                 len = interpret_branch_mark(r, name, namelen, at - name, buf,
1551                                             push_mark, branch_get_push,
1552                                             allowed);
1553                 if (len > 0)
1554                         return len;
1555         }
1556
1557         return -1;
1558 }
1559
1560 void strbuf_branchname(struct strbuf *sb, const char *name, unsigned allowed)
1561 {
1562         int len = strlen(name);
1563         int used = interpret_branch_name(name, len, sb, allowed);
1564
1565         if (used < 0)
1566                 used = 0;
1567         strbuf_add(sb, name + used, len - used);
1568 }
1569
1570 int strbuf_check_branch_ref(struct strbuf *sb, const char *name)
1571 {
1572         if (startup_info->have_repository)
1573                 strbuf_branchname(sb, name, INTERPRET_BRANCH_LOCAL);
1574         else
1575                 strbuf_addstr(sb, name);
1576
1577         /*
1578          * This splice must be done even if we end up rejecting the
1579          * name; builtin/branch.c::copy_or_rename_branch() still wants
1580          * to see what the name expanded to so that "branch -m" can be
1581          * used as a tool to correct earlier mistakes.
1582          */
1583         strbuf_splice(sb, 0, 0, "refs/heads/", 11);
1584
1585         if (*name == '-' ||
1586             !strcmp(sb->buf, "refs/heads/HEAD"))
1587                 return -1;
1588
1589         return check_refname_format(sb->buf, 0);
1590 }
1591
1592 /*
1593  * This is like "get_oid_basic()", except it allows "object ID expressions",
1594  * notably "xyz^" for "parent of xyz"
1595  */
1596 int repo_get_oid(struct repository *r, const char *name, struct object_id *oid)
1597 {
1598         struct object_context unused;
1599         return get_oid_with_context(r, name, 0, oid, &unused);
1600 }
1601
1602 /*
1603  * This returns a non-zero value if the string (built using printf
1604  * format and the given arguments) is not a valid object.
1605  */
1606 int get_oidf(struct object_id *oid, const char *fmt, ...)
1607 {
1608         va_list ap;
1609         int ret;
1610         struct strbuf sb = STRBUF_INIT;
1611
1612         va_start(ap, fmt);
1613         strbuf_vaddf(&sb, fmt, ap);
1614         va_end(ap);
1615
1616         ret = get_oid(sb.buf, oid);
1617         strbuf_release(&sb);
1618
1619         return ret;
1620 }
1621
1622 /*
1623  * Many callers know that the user meant to name a commit-ish by
1624  * syntactical positions where the object name appears.  Calling this
1625  * function allows the machinery to disambiguate shorter-than-unique
1626  * abbreviated object names between commit-ish and others.
1627  *
1628  * Note that this does NOT error out when the named object is not a
1629  * commit-ish. It is merely to give a hint to the disambiguation
1630  * machinery.
1631  */
1632 int repo_get_oid_committish(struct repository *r,
1633                             const char *name,
1634                             struct object_id *oid)
1635 {
1636         struct object_context unused;
1637         return get_oid_with_context(r, name, GET_OID_COMMITTISH,
1638                                     oid, &unused);
1639 }
1640
1641 int repo_get_oid_treeish(struct repository *r,
1642                          const char *name,
1643                          struct object_id *oid)
1644 {
1645         struct object_context unused;
1646         return get_oid_with_context(r, name, GET_OID_TREEISH,
1647                                     oid, &unused);
1648 }
1649
1650 int repo_get_oid_commit(struct repository *r,
1651                         const char *name,
1652                         struct object_id *oid)
1653 {
1654         struct object_context unused;
1655         return get_oid_with_context(r, name, GET_OID_COMMIT,
1656                                     oid, &unused);
1657 }
1658
1659 int repo_get_oid_tree(struct repository *r,
1660                       const char *name,
1661                       struct object_id *oid)
1662 {
1663         struct object_context unused;
1664         return get_oid_with_context(r, name, GET_OID_TREE,
1665                                     oid, &unused);
1666 }
1667
1668 int repo_get_oid_blob(struct repository *r,
1669                       const char *name,
1670                       struct object_id *oid)
1671 {
1672         struct object_context unused;
1673         return get_oid_with_context(r, name, GET_OID_BLOB,
1674                                     oid, &unused);
1675 }
1676
1677 /* Must be called only when object_name:filename doesn't exist. */
1678 static void diagnose_invalid_oid_path(struct repository *r,
1679                                       const char *prefix,
1680                                       const char *filename,
1681                                       const struct object_id *tree_oid,
1682                                       const char *object_name,
1683                                       int object_name_len)
1684 {
1685         struct object_id oid;
1686         unsigned short mode;
1687
1688         if (!prefix)
1689                 prefix = "";
1690
1691         if (file_exists(filename))
1692                 die("Path '%s' exists on disk, but not in '%.*s'.",
1693                     filename, object_name_len, object_name);
1694         if (is_missing_file_error(errno)) {
1695                 char *fullname = xstrfmt("%s%s", prefix, filename);
1696
1697                 if (!get_tree_entry(r, tree_oid, fullname, &oid, &mode)) {
1698                         die("Path '%s' exists, but not '%s'.\n"
1699                             "Did you mean '%.*s:%s' aka '%.*s:./%s'?",
1700                             fullname,
1701                             filename,
1702                             object_name_len, object_name,
1703                             fullname,
1704                             object_name_len, object_name,
1705                             filename);
1706                 }
1707                 die("Path '%s' does not exist in '%.*s'",
1708                     filename, object_name_len, object_name);
1709         }
1710 }
1711
1712 /* Must be called only when :stage:filename doesn't exist. */
1713 static void diagnose_invalid_index_path(struct repository *r,
1714                                         int stage,
1715                                         const char *prefix,
1716                                         const char *filename)
1717 {
1718         struct index_state *istate = r->index;
1719         const struct cache_entry *ce;
1720         int pos;
1721         unsigned namelen = strlen(filename);
1722         struct strbuf fullname = STRBUF_INIT;
1723
1724         if (!prefix)
1725                 prefix = "";
1726
1727         /* Wrong stage number? */
1728         pos = index_name_pos(istate, filename, namelen);
1729         if (pos < 0)
1730                 pos = -pos - 1;
1731         if (pos < istate->cache_nr) {
1732                 ce = istate->cache[pos];
1733                 if (ce_namelen(ce) == namelen &&
1734                     !memcmp(ce->name, filename, namelen))
1735                         die("Path '%s' is in the index, but not at stage %d.\n"
1736                             "Did you mean ':%d:%s'?",
1737                             filename, stage,
1738                             ce_stage(ce), filename);
1739         }
1740
1741         /* Confusion between relative and absolute filenames? */
1742         strbuf_addstr(&fullname, prefix);
1743         strbuf_addstr(&fullname, filename);
1744         pos = index_name_pos(istate, fullname.buf, fullname.len);
1745         if (pos < 0)
1746                 pos = -pos - 1;
1747         if (pos < istate->cache_nr) {
1748                 ce = istate->cache[pos];
1749                 if (ce_namelen(ce) == fullname.len &&
1750                     !memcmp(ce->name, fullname.buf, fullname.len))
1751                         die("Path '%s' is in the index, but not '%s'.\n"
1752                             "Did you mean ':%d:%s' aka ':%d:./%s'?",
1753                             fullname.buf, filename,
1754                             ce_stage(ce), fullname.buf,
1755                             ce_stage(ce), filename);
1756         }
1757
1758         if (repo_file_exists(r, filename))
1759                 die("Path '%s' exists on disk, but not in the index.", filename);
1760         if (is_missing_file_error(errno))
1761                 die("Path '%s' does not exist (neither on disk nor in the index).",
1762                     filename);
1763
1764         strbuf_release(&fullname);
1765 }
1766
1767
1768 static char *resolve_relative_path(struct repository *r, const char *rel)
1769 {
1770         if (!starts_with(rel, "./") && !starts_with(rel, "../"))
1771                 return NULL;
1772
1773         if (r != the_repository || !is_inside_work_tree())
1774                 die("relative path syntax can't be used outside working tree.");
1775
1776         /* die() inside prefix_path() if resolved path is outside worktree */
1777         return prefix_path(startup_info->prefix,
1778                            startup_info->prefix ? strlen(startup_info->prefix) : 0,
1779                            rel);
1780 }
1781
1782 static enum get_oid_result get_oid_with_context_1(struct repository *repo,
1783                                   const char *name,
1784                                   unsigned flags,
1785                                   const char *prefix,
1786                                   struct object_id *oid,
1787                                   struct object_context *oc)
1788 {
1789         int ret, bracket_depth;
1790         int namelen = strlen(name);
1791         const char *cp;
1792         int only_to_die = flags & GET_OID_ONLY_TO_DIE;
1793
1794         if (only_to_die)
1795                 flags |= GET_OID_QUIETLY;
1796
1797         memset(oc, 0, sizeof(*oc));
1798         oc->mode = S_IFINVALID;
1799         strbuf_init(&oc->symlink_path, 0);
1800         ret = get_oid_1(repo, name, namelen, oid, flags);
1801         if (!ret)
1802                 return ret;
1803         /*
1804          * sha1:path --> object name of path in ent sha1
1805          * :path -> object name of absolute path in index
1806          * :./path -> object name of path relative to cwd in index
1807          * :[0-3]:path -> object name of path in index at stage
1808          * :/foo -> recent commit matching foo
1809          */
1810         if (name[0] == ':') {
1811                 int stage = 0;
1812                 const struct cache_entry *ce;
1813                 char *new_path = NULL;
1814                 int pos;
1815                 if (!only_to_die && namelen > 2 && name[1] == '/') {
1816                         struct handle_one_ref_cb cb;
1817                         struct commit_list *list = NULL;
1818
1819                         cb.repo = repo;
1820                         cb.list = &list;
1821                         refs_for_each_ref(repo->refs, handle_one_ref, &cb);
1822                         refs_head_ref(repo->refs, handle_one_ref, &cb);
1823                         commit_list_sort_by_date(&list);
1824                         return get_oid_oneline(repo, name + 2, oid, list);
1825                 }
1826                 if (namelen < 3 ||
1827                     name[2] != ':' ||
1828                     name[1] < '0' || '3' < name[1])
1829                         cp = name + 1;
1830                 else {
1831                         stage = name[1] - '0';
1832                         cp = name + 3;
1833                 }
1834                 new_path = resolve_relative_path(repo, cp);
1835                 if (!new_path) {
1836                         namelen = namelen - (cp - name);
1837                 } else {
1838                         cp = new_path;
1839                         namelen = strlen(cp);
1840                 }
1841
1842                 if (flags & GET_OID_RECORD_PATH)
1843                         oc->path = xstrdup(cp);
1844
1845                 if (!repo->index || !repo->index->cache)
1846                         repo_read_index(repo);
1847                 pos = index_name_pos(repo->index, cp, namelen);
1848                 if (pos < 0)
1849                         pos = -pos - 1;
1850                 while (pos < repo->index->cache_nr) {
1851                         ce = repo->index->cache[pos];
1852                         if (ce_namelen(ce) != namelen ||
1853                             memcmp(ce->name, cp, namelen))
1854                                 break;
1855                         if (ce_stage(ce) == stage) {
1856                                 oidcpy(oid, &ce->oid);
1857                                 oc->mode = ce->ce_mode;
1858                                 free(new_path);
1859                                 return 0;
1860                         }
1861                         pos++;
1862                 }
1863                 if (only_to_die && name[1] && name[1] != '/')
1864                         diagnose_invalid_index_path(repo, stage, prefix, cp);
1865                 free(new_path);
1866                 return -1;
1867         }
1868         for (cp = name, bracket_depth = 0; *cp; cp++) {
1869                 if (*cp == '{')
1870                         bracket_depth++;
1871                 else if (bracket_depth && *cp == '}')
1872                         bracket_depth--;
1873                 else if (!bracket_depth && *cp == ':')
1874                         break;
1875         }
1876         if (*cp == ':') {
1877                 struct object_id tree_oid;
1878                 int len = cp - name;
1879                 unsigned sub_flags = flags;
1880
1881                 sub_flags &= ~GET_OID_DISAMBIGUATORS;
1882                 sub_flags |= GET_OID_TREEISH;
1883
1884                 if (!get_oid_1(repo, name, len, &tree_oid, sub_flags)) {
1885                         const char *filename = cp+1;
1886                         char *new_filename = NULL;
1887
1888                         new_filename = resolve_relative_path(repo, filename);
1889                         if (new_filename)
1890                                 filename = new_filename;
1891                         if (flags & GET_OID_FOLLOW_SYMLINKS) {
1892                                 ret = get_tree_entry_follow_symlinks(repo, &tree_oid,
1893                                         filename, oid, &oc->symlink_path,
1894                                         &oc->mode);
1895                         } else {
1896                                 ret = get_tree_entry(repo, &tree_oid, filename, oid,
1897                                                      &oc->mode);
1898                                 if (ret && only_to_die) {
1899                                         diagnose_invalid_oid_path(repo, prefix,
1900                                                                    filename,
1901                                                                    &tree_oid,
1902                                                                    name, len);
1903                                 }
1904                         }
1905                         if (flags & GET_OID_RECORD_PATH)
1906                                 oc->path = xstrdup(filename);
1907
1908                         free(new_filename);
1909                         return ret;
1910                 } else {
1911                         if (only_to_die)
1912                                 die("Invalid object name '%.*s'.", len, name);
1913                 }
1914         }
1915         return ret;
1916 }
1917
1918 /*
1919  * Call this function when you know "name" given by the end user must
1920  * name an object but it doesn't; the function _may_ die with a better
1921  * diagnostic message than "no such object 'name'", e.g. "Path 'doc' does not
1922  * exist in 'HEAD'" when given "HEAD:doc", or it may return in which case
1923  * you have a chance to diagnose the error further.
1924  */
1925 void maybe_die_on_misspelt_object_name(struct repository *r,
1926                                        const char *name,
1927                                        const char *prefix)
1928 {
1929         struct object_context oc;
1930         struct object_id oid;
1931         get_oid_with_context_1(r, name, GET_OID_ONLY_TO_DIE,
1932                                prefix, &oid, &oc);
1933 }
1934
1935 enum get_oid_result get_oid_with_context(struct repository *repo,
1936                                          const char *str,
1937                                          unsigned flags,
1938                                          struct object_id *oid,
1939                                          struct object_context *oc)
1940 {
1941         if (flags & GET_OID_FOLLOW_SYMLINKS && flags & GET_OID_ONLY_TO_DIE)
1942                 BUG("incompatible flags for get_sha1_with_context");
1943         return get_oid_with_context_1(repo, str, flags, NULL, oid, oc);
1944 }