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
 159         if (p->multi_pack_index)
 160                 return;
 161
 162         if (open_pack_index(p) || !p->num_objects)
 163                 return;
 164
 165         num = p->num_objects;
 166         bsearch_pack(&ds->bin_pfx, p, &first);
 167
 168         /*
 169          * At this point, "first" is the location of the lowest object
 170          * with an object name that could match "bin_pfx".  See if we have
 171          * 0, 1 or more objects that actually match(es).
 172          */
 173         for (i = first; i < num && !ds->ambiguous; i++) {
 174                 struct object_id oid;
 175                 nth_packed_object_id(&oid, p, i);
 176                 if (!match_sha(ds->len, ds->bin_pfx.hash, oid.hash))
 177                         break;
 178                 update_candidates(ds, &oid);
 179         }
 180 }
 181
 182 static void find_short_packed_object(struct disambiguate_state *ds)
 183 {
 184         struct multi_pack_index *m;
 185         struct packed_git *p;
 186
 187         for (m = get_multi_pack_index(ds->repo); m && !ds->ambiguous;
 188              m = m->next)
 189                 unique_in_midx(m, ds);
 190         for (p = get_packed_git(ds->repo); p && !ds->ambiguous;
 191              p = p->next)
 192                 unique_in_pack(p, ds);
 193 }
 194
 195 static int finish_object_disambiguation(struct disambiguate_state *ds,
 196                                         struct object_id *oid)
 197 {
 198         if (ds->ambiguous)
 199                 return SHORT_NAME_AMBIGUOUS;
 200
 201         if (!ds->candidate_exists)
 202                 return MISSING_OBJECT;
 203
 204         if (!ds->candidate_checked)
 205                 /*
 206                  * If this is the only candidate, there is no point
 207                  * calling the disambiguation hint callback.
 208                  *
 209                  * On the other hand, if the current candidate
 210                  * replaced an earlier candidate that did _not_ pass
 211                  * the disambiguation hint callback, then we do have
 212                  * more than one objects that match the short name
 213                  * given, so we should make sure this one matches;
 214                  * otherwise, if we discovered this one and the one
 215                  * that we previously discarded in the reverse order,
 216                  * we would end up showing different results in the
 217                  * same repository!
 218                  */
 219                 ds->candidate_ok = (!ds->disambiguate_fn_used ||
 220                                     ds->fn(ds->repo, &ds->candidate, ds->cb_data));
 221
 222         if (!ds->candidate_ok)
 223                 return SHORT_NAME_AMBIGUOUS;
 224
 225         oidcpy(oid, &ds->candidate);
 226         return 0;
 227 }
 228
 229 static int disambiguate_commit_only(struct repository *r,
 230                                     const struct object_id *oid,
 231                                     void *cb_data_unused)
 232 {
 233         int kind = oid_object_info(r, oid, NULL);
 234         return kind == OBJ_COMMIT;
 235 }
 236
 237 static int disambiguate_committish_only(struct repository *r,
 238                                         const struct object_id *oid,
 239                                         void *cb_data_unused)
 240 {
 241         struct object *obj;
 242         int kind;
 243
 244         kind = oid_object_info(r, oid, NULL);
 245         if (kind == OBJ_COMMIT)
 246                 return 1;
 247         if (kind != OBJ_TAG)
 248                 return 0;
 249
 250         /* We need to do this the hard way... */
 251         obj = deref_tag(r, parse_object(r, oid), NULL, 0);
 252         if (obj && obj->type == OBJ_COMMIT)
 253                 return 1;
 254         return 0;
 255 }
 256
 257 static int disambiguate_tree_only(struct repository *r,
 258                                   const struct object_id *oid,
 259                                   void *cb_data_unused)
 260 {
 261         int kind = oid_object_info(r, oid, NULL);
 262         return kind == OBJ_TREE;
 263 }
 264
 265 static int disambiguate_treeish_only(struct repository *r,
 266                                      const struct object_id *oid,
 267                                      void *cb_data_unused)
 268 {
 269         struct object *obj;
 270         int kind;
 271
 272         kind = oid_object_info(r, oid, NULL);
 273         if (kind == OBJ_TREE || kind == OBJ_COMMIT)
 274                 return 1;
 275         if (kind != OBJ_TAG)
 276                 return 0;
 277
 278         /* We need to do this the hard way... */
 279         obj = deref_tag(r, parse_object(r, oid), NULL, 0);
 280         if (obj && (obj->type == OBJ_TREE || obj->type == OBJ_COMMIT))
 281                 return 1;
 282         return 0;
 283 }
 284
 285 static int disambiguate_blob_only(struct repository *r,
 286                                   const struct object_id *oid,
 287                                   void *cb_data_unused)
 288 {
 289         int kind = oid_object_info(r, oid, NULL);
 290         return kind == OBJ_BLOB;
 291 }
 292
 293 static disambiguate_hint_fn default_disambiguate_hint;
 294
 295 int set_disambiguate_hint_config(const char *var, const char *value)
 296 {
 297         static const struct {
 298                 const char *name;
 299                 disambiguate_hint_fn fn;
 300         } hints[] = {
 301                 { "none", NULL },
 302                 { "commit", disambiguate_commit_only },
 303                 { "committish", disambiguate_committish_only },
 304                 { "tree", disambiguate_tree_only },
 305                 { "treeish", disambiguate_treeish_only },
 306                 { "blob", disambiguate_blob_only }
 307         };
 308         int i;
 309
 310         if (!value)
 311                 return config_error_nonbool(var);
 312
 313         for (i = 0; i < ARRAY_SIZE(hints); i++) {
 314                 if (!strcasecmp(value, hints[i].name)) {
 315                         default_disambiguate_hint = hints[i].fn;
 316                         return 0;
 317                 }
 318         }
 319
 320         return error("unknown hint type for '%s': %s", var, value);
 321 }
 322
 323 static int init_object_disambiguation(struct repository *r,
 324                                       const char *name, int len,
 325                                       struct disambiguate_state *ds)
 326 {
 327         int i;
 328
 329         if (len < MINIMUM_ABBREV || len > the_hash_algo->hexsz)
 330                 return -1;
 331
 332         memset(ds, 0, sizeof(*ds));
 333
 334         for (i = 0; i < len ;i++) {
 335                 unsigned char c = name[i];
 336                 unsigned char val;
 337                 if (c >= '0' && c <= '9')
 338                         val = c - '0';
 339                 else if (c >= 'a' && c <= 'f')
 340                         val = c - 'a' + 10;
 341                 else if (c >= 'A' && c <='F') {
 342                         val = c - 'A' + 10;
 343                         c -= 'A' - 'a';
 344                 }
 345                 else
 346                         return -1;
 347                 ds->hex_pfx[i] = c;
 348                 if (!(i & 1))
 349                         val <<= 4;
 350                 ds->bin_pfx.hash[i >> 1] |= val;
 351         }
 352
 353         ds->len = len;
 354         ds->hex_pfx[len] = '\0';
 355         ds->repo = r;
 356         prepare_alt_odb(r);
 357         return 0;
 358 }
 359
 360 static int show_ambiguous_object(const struct object_id *oid, void *data)
 361 {
 362         const struct disambiguate_state *ds = data;
 363         struct strbuf desc = STRBUF_INIT;
 364         int type;
 365
 366         if (ds->fn && !ds->fn(ds->repo, oid, ds->cb_data))
 367                 return 0;
 368
 369         type = oid_object_info(ds->repo, oid, NULL);
 370         if (type == OBJ_COMMIT) {
 371                 struct commit *commit = lookup_commit(ds->repo, oid);
 372                 if (commit) {
 373                         struct pretty_print_context pp = {0};
 374                         pp.date_mode.type = DATE_SHORT;
 375                         format_commit_message(commit, " %ad - %s", &desc, &pp);
 376                 }
 377         } else if (type == OBJ_TAG) {
 378                 struct tag *tag = lookup_tag(ds->repo, oid);
 379                 if (!parse_tag(tag) && tag->tag)
 380                         strbuf_addf(&desc, " %s", tag->tag);
 381         }
 382
 383         advise("  %s %s%s",
 384                repo_find_unique_abbrev(ds->repo, oid, DEFAULT_ABBREV),
 385                type_name(type) ? type_name(type) : "unknown type",
 386                desc.buf);
 387
 388         strbuf_release(&desc);
 389         return 0;
 390 }
 391
 392 static int collect_ambiguous(const struct object_id *oid, void *data)
 393 {
 394         oid_array_append(data, oid);
 395         return 0;
 396 }
 397
 398 static int repo_collect_ambiguous(struct repository *r,
 399                                   const struct object_id *oid,
 400                                   void *data)
 401 {
 402         return collect_ambiguous(oid, data);
 403 }
 404
 405 static int sort_ambiguous(const void *a, const void *b, void *ctx)
 406 {
 407         struct repository *sort_ambiguous_repo = ctx;
 408         int a_type = oid_object_info(sort_ambiguous_repo, a, NULL);
 409         int b_type = oid_object_info(sort_ambiguous_repo, b, NULL);
 410         int a_type_sort;
 411         int b_type_sort;
 412
 413         /*
 414          * Sorts by hash within the same object type, just as
 415          * oid_array_for_each_unique() would do.
 416          */
 417         if (a_type == b_type)
 418                 return oidcmp(a, b);
 419
 420         /*
 421          * Between object types show tags, then commits, and finally
 422          * trees and blobs.
 423          *
 424          * The object_type enum is commit, tree, blob, tag, but we
 425          * want tag, commit, tree blob. Cleverly (perhaps too
 426          * cleverly) do that with modulus, since the enum assigns 1 to
 427          * commit, so tag becomes 0.
 428          */
 429         a_type_sort = a_type % 4;
 430         b_type_sort = b_type % 4;
 431         return a_type_sort > b_type_sort ? 1 : -1;
 432 }
 433
 434 static void sort_ambiguous_oid_array(struct repository *r, struct oid_array *a)
 435 {
 436         QSORT_S(a->oid, a->nr, sort_ambiguous, r);
 437 }
 438
 439 static enum get_oid_result get_short_oid(struct repository *r,
 440                                          const char *name, int len,
 441                                          struct object_id *oid,
 442                                          unsigned flags)
 443 {
 444         int status;
 445         struct disambiguate_state ds;
 446         int quietly = !!(flags & GET_OID_QUIETLY);
 447
 448         if (init_object_disambiguation(r, name, len, &ds) < 0)
 449                 return -1;
 450
 451         if (HAS_MULTI_BITS(flags & GET_OID_DISAMBIGUATORS))
 452                 BUG("multiple get_short_oid disambiguator flags");
 453
 454         if (flags & GET_OID_COMMIT)
 455                 ds.fn = disambiguate_commit_only;
 456         else if (flags & GET_OID_COMMITTISH)
 457                 ds.fn = disambiguate_committish_only;
 458         else if (flags & GET_OID_TREE)
 459                 ds.fn = disambiguate_tree_only;
 460         else if (flags & GET_OID_TREEISH)
 461                 ds.fn = disambiguate_treeish_only;
 462         else if (flags & GET_OID_BLOB)
 463                 ds.fn = disambiguate_blob_only;
 464         else
 465                 ds.fn = default_disambiguate_hint;
 466
 467         find_short_object_filename(&ds);
 468         find_short_packed_object(&ds);
 469         status = finish_object_disambiguation(&ds, oid);
 470
 471         /*
 472          * If we didn't find it, do the usual reprepare() slow-path,
 473          * since the object may have recently been added to the repository
 474          * or migrated from loose to packed.
 475          */
 476         if (status == MISSING_OBJECT) {
 477                 reprepare_packed_git(r);
 478                 find_short_object_filename(&ds);
 479                 find_short_packed_object(&ds);
 480                 status = finish_object_disambiguation(&ds, oid);
 481         }
 482
 483         if (!quietly && (status == SHORT_NAME_AMBIGUOUS)) {
 484                 struct oid_array collect = OID_ARRAY_INIT;
 485
 486                 error(_("short SHA1 %s is ambiguous"), ds.hex_pfx);
 487
 488                 /*
 489                  * We may still have ambiguity if we simply saw a series of
 490                  * candidates that did not satisfy our hint function. In
 491                  * that case, we still want to show them, so disable the hint
 492                  * function entirely.
 493                  */
 494                 if (!ds.ambiguous)
 495                         ds.fn = NULL;
 496
 497                 advise(_("The candidates are:"));
 498                 repo_for_each_abbrev(r, ds.hex_pfx, collect_ambiguous, &collect);
 499                 sort_ambiguous_oid_array(r, &collect);
 500
 501                 if (oid_array_for_each(&collect, show_ambiguous_object, &ds))
 502                         BUG("show_ambiguous_object shouldn't return non-zero");
 503                 oid_array_clear(&collect);
 504         }
 505
 506         return status;
 507 }
 508
 509 int repo_for_each_abbrev(struct repository *r, const char *prefix,
 510                          each_abbrev_fn fn, void *cb_data)
 511 {
 512         struct oid_array collect = OID_ARRAY_INIT;
 513         struct disambiguate_state ds;
 514         int ret;
 515
 516         if (init_object_disambiguation(r, prefix, strlen(prefix), &ds) < 0)
 517                 return -1;
 518
 519         ds.always_call_fn = 1;
 520         ds.fn = repo_collect_ambiguous;
 521         ds.cb_data = &collect;
 522         find_short_object_filename(&ds);
 523         find_short_packed_object(&ds);
 524
 525         ret = oid_array_for_each_unique(&collect, fn, cb_data);
 526         oid_array_clear(&collect);
 527         return ret;
 528 }
 529
 530 /*
 531  * Return the slot of the most-significant bit set in "val". There are various
 532  * ways to do this quickly with fls() or __builtin_clzl(), but speed is
 533  * probably not a big deal here.
 534  */
 535 static unsigned msb(unsigned long val)
 536 {
 537         unsigned r = 0;
 538         while (val >>= 1)
 539                 r++;
 540         return r;
 541 }
 542
 543 struct min_abbrev_data {
 544         unsigned int init_len;
 545         unsigned int cur_len;
 546         char *hex;
 547         struct repository *repo;
 548         const struct object_id *oid;
 549 };
 550
 551 static inline char get_hex_char_from_oid(const struct object_id *oid,
 552                                          unsigned int pos)
 553 {
 554         static const char hex[] = "0123456789abcdef";
 555
 556         if ((pos & 1) == 0)
 557                 return hex[oid->hash[pos >> 1] >> 4];
 558         else
 559                 return hex[oid->hash[pos >> 1] & 0xf];
 560 }
 561
 562 static int extend_abbrev_len(const struct object_id *oid, void *cb_data)
 563 {
 564         struct min_abbrev_data *mad = cb_data;
 565
 566         unsigned int i = mad->init_len;
 567         while (mad->hex[i] && mad->hex[i] == get_hex_char_from_oid(oid, i))
 568                 i++;
 569
 570         if (i < GIT_MAX_RAWSZ && i >= mad->cur_len)
 571                 mad->cur_len = i + 1;
 572
 573         return 0;
 574 }
 575
 576 static int repo_extend_abbrev_len(struct repository *r,
 577                                   const struct object_id *oid,
 578                                   void *cb_data)
 579 {
 580         return extend_abbrev_len(oid, cb_data);
 581 }
 582
 583 static void find_abbrev_len_for_midx(struct multi_pack_index *m,
 584                                      struct min_abbrev_data *mad)
 585 {
 586         int match = 0;
 587         uint32_t num, first = 0;
 588         struct object_id oid;
 589         const struct object_id *mad_oid;
 590
 591         if (!m->num_objects)
 592                 return;
 593
 594         num = m->num_objects;
 595         mad_oid = mad->oid;
 596         match = bsearch_midx(mad_oid, m, &first);
 597
 598         /*
 599          * first is now the position in the packfile where we would insert
 600          * mad->hash if it does not exist (or the position of mad->hash if
 601          * it does exist). Hence, we consider a maximum of two objects
 602          * nearby for the abbreviation length.
 603          */
 604         mad->init_len = 0;
 605         if (!match) {
 606                 if (nth_midxed_object_oid(&oid, m, first))
 607                         extend_abbrev_len(&oid, mad);
 608         } else if (first < num - 1) {
 609                 if (nth_midxed_object_oid(&oid, m, first + 1))
 610                         extend_abbrev_len(&oid, mad);
 611         }
 612         if (first > 0) {
 613                 if (nth_midxed_object_oid(&oid, m, first - 1))
 614                         extend_abbrev_len(&oid, mad);
 615         }
 616         mad->init_len = mad->cur_len;
 617 }
 618
 619 static void find_abbrev_len_for_pack(struct packed_git *p,
 620                                      struct min_abbrev_data *mad)
 621 {
 622         int match = 0;
 623         uint32_t num, first = 0;
 624         struct object_id oid;
 625         const struct object_id *mad_oid;
 626
 627         if (p->multi_pack_index)
 628                 return;
 629
 630         if (open_pack_index(p) || !p->num_objects)
 631                 return;
 632
 633         num = p->num_objects;
 634         mad_oid = mad->oid;
 635         match = bsearch_pack(mad_oid, p, &first);
 636
 637         /*
 638          * first is now the position in the packfile where we would insert
 639          * mad->hash if it does not exist (or the position of mad->hash if
 640          * it does exist). Hence, we consider a maximum of two objects
 641          * nearby for the abbreviation length.
 642          */
 643         mad->init_len = 0;
 644         if (!match) {
 645                 if (!nth_packed_object_id(&oid, p, first))
 646                         extend_abbrev_len(&oid, mad);
 647         } else if (first < num - 1) {
 648                 if (!nth_packed_object_id(&oid, p, first + 1))
 649                         extend_abbrev_len(&oid, mad);
 650         }
 651         if (first > 0) {
 652                 if (!nth_packed_object_id(&oid, p, first - 1))
 653                         extend_abbrev_len(&oid, mad);
 654         }
 655         mad->init_len = mad->cur_len;
 656 }
 657
 658 static void find_abbrev_len_packed(struct min_abbrev_data *mad)
 659 {
 660         struct multi_pack_index *m;
 661         struct packed_git *p;
 662
 663         for (m = get_multi_pack_index(mad->repo); m; m = m->next)
 664                 find_abbrev_len_for_midx(m, mad);
 665         for (p = get_packed_git(mad->repo); p; p = p->next)
 666                 find_abbrev_len_for_pack(p, mad);
 667 }
 668
 669 int repo_find_unique_abbrev_r(struct repository *r, char *hex,
 670                               const struct object_id *oid, int len)
 671 {
 672         struct disambiguate_state ds;
 673         struct min_abbrev_data mad;
 674         struct object_id oid_ret;
 675         const unsigned hexsz = r->hash_algo->hexsz;
 676
 677         if (len < 0) {
 678                 unsigned long count = repo_approximate_object_count(r);
 679                 /*
 680                  * Add one because the MSB only tells us the highest bit set,
 681                  * not including the value of all the _other_ bits (so "15"
 682                  * is only one off of 2^4, but the MSB is the 3rd bit.
 683                  */
 684                 len = msb(count) + 1;
 685                 /*
 686                  * We now know we have on the order of 2^len objects, which
 687                  * expects a collision at 2^(len/2). But we also care about hex
 688                  * chars, not bits, and there are 4 bits per hex. So all
 689                  * together we need to divide by 2 and round up.
 690                  */
 691                 len = DIV_ROUND_UP(len, 2);
 692                 /*
 693                  * For very small repos, we stick with our regular fallback.
 694                  */
 695                 if (len < FALLBACK_DEFAULT_ABBREV)
 696                         len = FALLBACK_DEFAULT_ABBREV;
 697         }
 698
 699         oid_to_hex_r(hex, oid);
 700         if (len == hexsz || !len)
 701                 return hexsz;
 702
 703         mad.repo = r;
 704         mad.init_len = len;
 705         mad.cur_len = len;
 706         mad.hex = hex;
 707         mad.oid = oid;
 708
 709         find_abbrev_len_packed(&mad);
 710
 711         if (init_object_disambiguation(r, hex, mad.cur_len, &ds) < 0)
 712                 return -1;
 713
 714         ds.fn = repo_extend_abbrev_len;
 715         ds.always_call_fn = 1;
 716         ds.cb_data = (void *)&mad;
 717
 718         find_short_object_filename(&ds);
 719         (void)finish_object_disambiguation(&ds, &oid_ret);
 720
 721         hex[mad.cur_len] = 0;
 722         return mad.cur_len;
 723 }
 724
 725 const char *repo_find_unique_abbrev(struct repository *r,
 726                                     const struct object_id *oid,
 727                                     int len)
 728 {
 729         static int bufno;
 730         static char hexbuffer[4][GIT_MAX_HEXSZ + 1];
 731         char *hex = hexbuffer[bufno];
 732         bufno = (bufno + 1) % ARRAY_SIZE(hexbuffer);
 733         repo_find_unique_abbrev_r(r, hex, oid, len);
 734         return hex;
 735 }
 736
 737 static int ambiguous_path(const char *path, int len)
 738 {
 739         int slash = 1;
 740         int cnt;
 741
 742         for (cnt = 0; cnt < len; cnt++) {
 743                 switch (*path++) {
 744                 case '\0':
 745                         break;
 746                 case '/':
 747                         if (slash)
 748                                 break;
 749                         slash = 1;
 750                         continue;
 751                 case '.':
 752                         continue;
 753                 default:
 754                         slash = 0;
 755                         continue;
 756                 }
 757                 break;
 758         }
 759         return slash;
 760 }
 761
 762 static inline int at_mark(const char *string, int len,
 763                           const char **suffix, int nr)
 764 {
 765         int i;
 766
 767         for (i = 0; i < nr; i++) {
 768                 int suffix_len = strlen(suffix[i]);
 769                 if (suffix_len <= len
 770                     && !strncasecmp(string, suffix[i], suffix_len))
 771                         return suffix_len;
 772         }
 773         return 0;
 774 }
 775
 776 static inline int upstream_mark(const char *string, int len)
 777 {
 778         const char *suffix[] = { "@{upstream}", "@{u}" };
 779         return at_mark(string, len, suffix, ARRAY_SIZE(suffix));
 780 }
 781
 782 static inline int push_mark(const char *string, int len)
 783 {
 784         const char *suffix[] = { "@{push}" };
 785         return at_mark(string, len, suffix, ARRAY_SIZE(suffix));
 786 }
 787
 788 static enum get_oid_result get_oid_1(struct repository *r, const char *name, int len, struct object_id *oid, unsigned lookup_flags);
 789 static int interpret_nth_prior_checkout(struct repository *r, const char *name, int namelen, struct strbuf *buf);
 790
 791 static int get_oid_basic(struct repository *r, const char *str, int len,
 792                          struct object_id *oid, unsigned int flags)
 793 {
 794         static const char *warn_msg = "refname '%.*s' is ambiguous.";
 795         static const char *object_name_msg = N_(
 796         "Git normally never creates a ref that ends with 40 hex characters\n"
 797         "because it will be ignored when you just specify 40-hex. These refs\n"
 798         "may be created by mistake. For example,\n"
 799         "\n"
 800         "  git switch -c $br $(git rev-parse ...)\n"
 801         "\n"
 802         "where \"$br\" is somehow empty and a 40-hex ref is created. Please\n"
 803         "examine these refs and maybe delete them. Turn this message off by\n"
 804         "running \"git config advice.objectNameWarning false\"");
 805         struct object_id tmp_oid;
 806         char *real_ref = NULL;
 807         int refs_found = 0;
 808         int at, reflog_len, nth_prior = 0;
 809
 810         if (len == r->hash_algo->hexsz && !get_oid_hex(str, oid)) {
 811                 if (warn_ambiguous_refs && warn_on_object_refname_ambiguity) {
 812                         refs_found = repo_dwim_ref(r, str, len, &tmp_oid, &real_ref);
 813                         if (refs_found > 0) {
 814                                 warning(warn_msg, len, str);
 815                                 if (advice_object_name_warning)
 816                                         fprintf(stderr, "%s\n", _(object_name_msg));
 817                         }
 818                         free(real_ref);
 819                 }
 820                 return 0;
 821         }
 822
 823         /* basic@{time or number or -number} format to query ref-log */
 824         reflog_len = at = 0;
 825         if (len && str[len-1] == '}') {
 826                 for (at = len-4; at >= 0; at--) {
 827                         if (str[at] == '@' && str[at+1] == '{') {
 828                                 if (str[at+2] == '-') {
 829                                         if (at != 0)
 830                                                 /* @{-N} not at start */
 831                                                 return -1;
 832                                         nth_prior = 1;
 833                                         continue;
 834                                 }
 835                                 if (!upstream_mark(str + at, len - at) &&
 836                                     !push_mark(str + at, len - at)) {
 837                                         reflog_len = (len-1) - (at+2);
 838                                         len = at;
 839                                 }
 840                                 break;
 841                         }
 842                 }
 843         }
 844
 845         /* Accept only unambiguous ref paths. */
 846         if (len && ambiguous_path(str, len))
 847                 return -1;
 848
 849         if (nth_prior) {
 850                 struct strbuf buf = STRBUF_INIT;
 851                 int detached;
 852
 853                 if (interpret_nth_prior_checkout(r, str, len, &buf) > 0) {
 854                         detached = (buf.len == r->hash_algo->hexsz && !get_oid_hex(buf.buf, oid));
 855                         strbuf_release(&buf);
 856                         if (detached)
 857                                 return 0;
 858                 }
 859         }
 860
 861         if (!len && reflog_len)
 862                 /* allow "@{...}" to mean the current branch reflog */
 863                 refs_found = repo_dwim_ref(r, "HEAD", 4, oid, &real_ref);
 864         else if (reflog_len)
 865                 refs_found = repo_dwim_log(r, str, len, oid, &real_ref);
 866         else
 867                 refs_found = repo_dwim_ref(r, str, len, oid, &real_ref);
 868
 869         if (!refs_found)
 870                 return -1;
 871
 872         if (warn_ambiguous_refs && !(flags & GET_OID_QUIETLY) &&
 873             (refs_found > 1 ||
 874              !get_short_oid(r, str, len, &tmp_oid, GET_OID_QUIETLY)))
 875                 warning(warn_msg, len, str);
 876
 877         if (reflog_len) {
 878                 int nth, i;
 879                 timestamp_t at_time;
 880                 timestamp_t co_time;
 881                 int co_tz, co_cnt;
 882
 883                 /* Is it asking for N-th entry, or approxidate? */
 884                 for (i = nth = 0; 0 <= nth && i < reflog_len; i++) {
 885                         char ch = str[at+2+i];
 886                         if ('0' <= ch && ch <= '9')
 887                                 nth = nth * 10 + ch - '0';
 888                         else
 889                                 nth = -1;
 890                 }
 891                 if (100000000 <= nth) {
 892                         at_time = nth;
 893                         nth = -1;
 894                 } else if (0 <= nth)
 895                         at_time = 0;
 896                 else {
 897                         int errors = 0;
 898                         char *tmp = xstrndup(str + at + 2, reflog_len);
 899                         at_time = approxidate_careful(tmp, &errors);
 900                         free(tmp);
 901                         if (errors) {
 902                                 free(real_ref);
 903                                 return -1;
 904                         }
 905                 }
 906                 if (read_ref_at(get_main_ref_store(r),
 907                                 real_ref, flags, at_time, nth, oid, NULL,
 908                                 &co_time, &co_tz, &co_cnt)) {
 909                         if (!len) {
 910                                 if (!skip_prefix(real_ref, "refs/heads/", &str))
 911                                         str = "HEAD";
 912                                 len = strlen(str);
 913                         }
 914                         if (at_time) {
 915                                 if (!(flags & GET_OID_QUIETLY)) {
 916                                         warning(_("log for '%.*s' only goes back to %s"),
 917                                                 len, str,
 918                                                 show_date(co_time, co_tz, DATE_MODE(RFC2822)));
 919                                 }
 920                         } else {
 921                                 if (flags & GET_OID_QUIETLY) {
 922                                         exit(128);
 923                                 }
 924                                 die(_("log for '%.*s' only has %d entries"),
 925                                     len, str, co_cnt);
 926                         }
 927                 }
 928         }
 929
 930         free(real_ref);
 931         return 0;
 932 }
 933
 934 static enum get_oid_result get_parent(struct repository *r,
 935                                       const char *name, int len,
 936                                       struct object_id *result, int idx)
 937 {
 938         struct object_id oid;
 939         enum get_oid_result ret = get_oid_1(r, name, len, &oid,
 940                                             GET_OID_COMMITTISH);
 941         struct commit *commit;
 942         struct commit_list *p;
 943
 944         if (ret)
 945                 return ret;
 946         commit = lookup_commit_reference(r, &oid);
 947         if (parse_commit(commit))
 948                 return MISSING_OBJECT;
 949         if (!idx) {
 950                 oidcpy(result, &commit->object.oid);
 951                 return FOUND;
 952         }
 953         p = commit->parents;
 954         while (p) {
 955                 if (!--idx) {
 956                         oidcpy(result, &p->item->object.oid);
 957                         return FOUND;
 958                 }
 959                 p = p->next;
 960         }
 961         return MISSING_OBJECT;
 962 }
 963
 964 static enum get_oid_result get_nth_ancestor(struct repository *r,
 965                                             const char *name, int len,
 966                                             struct object_id *result,
 967                                             int generation)
 968 {
 969         struct object_id oid;
 970         struct commit *commit;
 971         int ret;
 972
 973         ret = get_oid_1(r, name, len, &oid, GET_OID_COMMITTISH);
 974         if (ret)
 975                 return ret;
 976         commit = lookup_commit_reference(r, &oid);
 977         if (!commit)
 978                 return MISSING_OBJECT;
 979
 980         while (generation--) {
 981                 if (parse_commit(commit) || !commit->parents)
 982                         return MISSING_OBJECT;
 983                 commit = commit->parents->item;
 984         }
 985         oidcpy(result, &commit->object.oid);
 986         return FOUND;
 987 }
 988
 989 struct object *repo_peel_to_type(struct repository *r, const char *name, int namelen,
 990                                  struct object *o, enum object_type expected_type)
 991 {
 992         if (name && !namelen)
 993                 namelen = strlen(name);
 994         while (1) {
 995                 if (!o || (!o->parsed && !parse_object(r, &o->oid)))
 996                         return NULL;
 997                 if (expected_type == OBJ_ANY || o->type == expected_type)
 998                         return o;
 999                 if (o->type == OBJ_TAG)
1000                         o = ((struct tag*) o)->tagged;
1001                 else if (o->type == OBJ_COMMIT)
1002                         o = &(repo_get_commit_tree(r, ((struct commit *)o))->object);
1003                 else {
1004                         if (name)
1005                                 error("%.*s: expected %s type, but the object "
1006                                       "dereferences to %s type",
1007                                       namelen, name, type_name(expected_type),
1008                                       type_name(o->type));
1009                         return NULL;
1010                 }
1011         }
1012 }
1013
1014 static int peel_onion(struct repository *r, const char *name, int len,
1015                       struct object_id *oid, unsigned lookup_flags)
1016 {
1017         struct object_id outer;
1018         const char *sp;
1019         unsigned int expected_type = 0;
1020         struct object *o;
1021
1022         /*
1023          * "ref^{type}" dereferences ref repeatedly until you cannot
1024          * dereference anymore, or you get an object of given type,
1025          * whichever comes first.  "ref^{}" means just dereference
1026          * tags until you get a non-tag.  "ref^0" is a shorthand for
1027          * "ref^{commit}".  "commit^{tree}" could be used to find the
1028          * top-level tree of the given commit.
1029          */
1030         if (len < 4 || name[len-1] != '}')
1031                 return -1;
1032
1033         for (sp = name + len - 1; name <= sp; sp--) {
1034                 int ch = *sp;
1035                 if (ch == '{' && name < sp && sp[-1] == '^')
1036                         break;
1037         }
1038         if (sp <= name)
1039                 return -1;
1040
1041         sp++; /* beginning of type name, or closing brace for empty */
1042         if (starts_with(sp, "commit}"))
1043                 expected_type = OBJ_COMMIT;
1044         else if (starts_with(sp, "tag}"))
1045                 expected_type = OBJ_TAG;
1046         else if (starts_with(sp, "tree}"))
1047                 expected_type = OBJ_TREE;
1048         else if (starts_with(sp, "blob}"))
1049                 expected_type = OBJ_BLOB;
1050         else if (starts_with(sp, "object}"))
1051                 expected_type = OBJ_ANY;
1052         else if (sp[0] == '}')
1053                 expected_type = OBJ_NONE;
1054         else if (sp[0] == '/')
1055                 expected_type = OBJ_COMMIT;
1056         else
1057                 return -1;
1058
1059         lookup_flags &= ~GET_OID_DISAMBIGUATORS;
1060         if (expected_type == OBJ_COMMIT)
1061                 lookup_flags |= GET_OID_COMMITTISH;
1062         else if (expected_type == OBJ_TREE)
1063                 lookup_flags |= GET_OID_TREEISH;
1064
1065         if (get_oid_1(r, name, sp - name - 2, &outer, lookup_flags))
1066                 return -1;
1067
1068         o = parse_object(r, &outer);
1069         if (!o)
1070                 return -1;
1071         if (!expected_type) {
1072                 o = deref_tag(r, o, name, sp - name - 2);
1073                 if (!o || (!o->parsed && !parse_object(r, &o->oid)))
1074                         return -1;
1075                 oidcpy(oid, &o->oid);
1076                 return 0;
1077         }
1078
1079         /*
1080          * At this point, the syntax look correct, so
1081          * if we do not get the needed object, we should
1082          * barf.
1083          */
1084         o = repo_peel_to_type(r, name, len, o, expected_type);
1085         if (!o)
1086                 return -1;
1087
1088         oidcpy(oid, &o->oid);
1089         if (sp[0] == '/') {
1090                 /* "$commit^{/foo}" */
1091                 char *prefix;
1092                 int ret;
1093                 struct commit_list *list = NULL;
1094
1095                 /*
1096                  * $commit^{/}. Some regex implementation may reject.
1097                  * We don't need regex anyway. '' pattern always matches.
1098                  */
1099                 if (sp[1] == '}')
1100                         return 0;
1101
1102                 prefix = xstrndup(sp + 1, name + len - 1 - (sp + 1));
1103                 commit_list_insert((struct commit *)o, &list);
1104                 ret = get_oid_oneline(r, prefix, oid, list);
1105                 free(prefix);
1106                 return ret;
1107         }
1108         return 0;
1109 }
1110
1111 static int get_describe_name(struct repository *r,
1112                              const char *name, int len,
1113                              struct object_id *oid)
1114 {
1115         const char *cp;
1116         unsigned flags = GET_OID_QUIETLY | GET_OID_COMMIT;
1117
1118         for (cp = name + len - 1; name + 2 <= cp; cp--) {
1119                 char ch = *cp;
1120                 if (!isxdigit(ch)) {
1121                         /* We must be looking at g in "SOMETHING-g"
1122                          * for it to be describe output.
1123                          */
1124                         if (ch == 'g' && cp[-1] == '-') {
1125                                 cp++;
1126                                 len -= cp - name;
1127                                 return get_short_oid(r,
1128                                                      cp, len, oid, flags);
1129                         }
1130                 }
1131         }
1132         return -1;
1133 }
1134
1135 static enum get_oid_result get_oid_1(struct repository *r,
1136                                      const char *name, int len,
1137                                      struct object_id *oid,
1138                                      unsigned lookup_flags)
1139 {
1140         int ret, has_suffix;
1141         const char *cp;
1142
1143         /*
1144          * "name~3" is "name^^^", "name~" is "name~1", and "name^" is "name^1".
1145          */
1146         has_suffix = 0;
1147         for (cp = name + len - 1; name <= cp; cp--) {
1148                 int ch = *cp;
1149                 if ('0' <= ch && ch <= '9')
1150                         continue;
1151                 if (ch == '~' || ch == '^')
1152                         has_suffix = ch;
1153                 break;
1154         }
1155
1156         if (has_suffix) {
1157                 unsigned int num = 0;
1158                 int len1 = cp - name;
1159                 cp++;
1160                 while (cp < name + len) {
1161                         unsigned int digit = *cp++ - '0';
1162                         if (unsigned_mult_overflows(num, 10))
1163                                 return MISSING_OBJECT;
1164                         num *= 10;
1165                         if (unsigned_add_overflows(num, digit))
1166                                 return MISSING_OBJECT;
1167                         num += digit;
1168                 }
1169                 if (!num && len1 == len - 1)
1170                         num = 1;
1171                 else if (num > INT_MAX)
1172                         return MISSING_OBJECT;
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 *sb;
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->sb);
1311                 strbuf_add(cb->sb, 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         cb.sb = buf;
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                 retval = brace - name + 1;
1350         } else
1351                 retval = 0;
1352
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                             "hint: 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                             "hint: 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                             "hint: 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 }