sha1_name.c

   1 #include "cache.h"
   2 #include "tag.h"
   3 #include "commit.h"
   4 #include "tree.h"
   5 #include "blob.h"
   6 #include "tree-walk.h"
   7 #include "refs.h"
   8 #include "remote.h"
   9
  10 static int get_sha1_oneline(const char *, unsigned char *, struct commit_list *);
  11
  12 typedef int (*disambiguate_hint_fn)(const unsigned char *, void *);
  13
  14 struct disambiguate_state {
  15         disambiguate_hint_fn fn;
  16         void *cb_data;
  17         unsigned char candidate[20];
  18         unsigned candidate_exists:1;
  19         unsigned candidate_checked:1;
  20         unsigned candidate_ok:1;
  21         unsigned disambiguate_fn_used:1;
  22         unsigned ambiguous:1;
  23 };
  24
  25 static void update_candidates(struct disambiguate_state *ds, const unsigned char *current)
  26 {
  27         if (!ds->candidate_exists) {
  28                 /* this is the first candidate */
  29                 hashcpy(ds->candidate, current);
  30                 ds->candidate_exists = 1;
  31                 return;
  32         } else if (!hashcmp(ds->candidate, current)) {
  33                 /* the same as what we already have seen */
  34                 return;
  35         }
  36
  37         if (!ds->fn) {
  38                 /* cannot disambiguate between ds->candidate and current */
  39                 ds->ambiguous = 1;
  40                 return;
  41         }
  42
  43         if (!ds->candidate_checked) {
  44                 ds->candidate_ok = ds->fn(ds->candidate, ds->cb_data);
  45                 ds->disambiguate_fn_used = 1;
  46                 ds->candidate_checked = 1;
  47         }
  48
  49         if (!ds->candidate_ok) {
  50                 /* discard the candidate; we know it does not satisify fn */
  51                 hashcpy(ds->candidate, current);
  52                 ds->candidate_checked = 0;
  53                 return;
  54         }
  55
  56         /* if we reach this point, we know ds->candidate satisfies fn */
  57         if (ds->fn(current, ds->cb_data)) {
  58                 /*
  59                  * if both current and candidate satisfy fn, we cannot
  60                  * disambiguate.
  61                  */
  62                 ds->candidate_ok = 0;
  63                 ds->ambiguous = 1;
  64         }
  65
  66         /* otherwise, current can be discarded and candidate is still good */
  67 }
  68
  69 static void find_short_object_filename(int len, const char *hex_pfx, struct disambiguate_state *ds)
  70 {
  71         struct alternate_object_database *alt;
  72         char hex[40];
  73         static struct alternate_object_database *fakeent;
  74
  75         if (!fakeent) {
  76                 /*
  77                  * Create a "fake" alternate object database that
  78                  * points to our own object database, to make it
  79                  * easier to get a temporary working space in
  80                  * alt->name/alt->base while iterating over the
  81                  * object databases including our own.
  82                  */
  83                 const char *objdir = get_object_directory();
  84                 int objdir_len = strlen(objdir);
  85                 int entlen = objdir_len + 43;
  86                 fakeent = xmalloc(sizeof(*fakeent) + entlen);
  87                 memcpy(fakeent->base, objdir, objdir_len);
  88                 fakeent->name = fakeent->base + objdir_len + 1;
  89                 fakeent->name[-1] = '/';
  90         }
  91         fakeent->next = alt_odb_list;
  92
  93         sprintf(hex, "%.2s", hex_pfx);
  94         for (alt = fakeent; alt && !ds->ambiguous; alt = alt->next) {
  95                 struct dirent *de;
  96                 DIR *dir;
  97                 sprintf(alt->name, "%.2s/", hex_pfx);
  98                 dir = opendir(alt->base);
  99                 if (!dir)
 100                         continue;
 101
 102                 while (!ds->ambiguous && (de = readdir(dir)) != NULL) {
 103                         unsigned char sha1[20];
 104
 105                         if (strlen(de->d_name) != 38)
 106                                 continue;
 107                         if (memcmp(de->d_name, hex_pfx + 2, len - 2))
 108                                 continue;
 109                         memcpy(hex + 2, de->d_name, 38);
 110                         if (!get_sha1_hex(hex, sha1))
 111                                 update_candidates(ds, sha1);
 112                 }
 113                 closedir(dir);
 114         }
 115 }
 116
 117 static int match_sha(unsigned len, const unsigned char *a, const unsigned char *b)
 118 {
 119         do {
 120                 if (*a != *b)
 121                         return 0;
 122                 a++;
 123                 b++;
 124                 len -= 2;
 125         } while (len > 1);
 126         if (len)
 127                 if ((*a ^ *b) & 0xf0)
 128                         return 0;
 129         return 1;
 130 }
 131
 132 static void unique_in_pack(int len,
 133                           const unsigned char *bin_pfx,
 134                            struct packed_git *p,
 135                            struct disambiguate_state *ds)
 136 {
 137         uint32_t num, last, i, first = 0;
 138         const unsigned char *current = NULL;
 139
 140         open_pack_index(p);
 141         num = p->num_objects;
 142         last = num;
 143         while (first < last) {
 144                 uint32_t mid = (first + last) / 2;
 145                 const unsigned char *current;
 146                 int cmp;
 147
 148                 current = nth_packed_object_sha1(p, mid);
 149                 cmp = hashcmp(bin_pfx, current);
 150                 if (!cmp) {
 151                         first = mid;
 152                         break;
 153                 }
 154                 if (cmp > 0) {
 155                         first = mid+1;
 156                         continue;
 157                 }
 158                 last = mid;
 159         }
 160
 161         /*
 162          * At this point, "first" is the location of the lowest object
 163          * with an object name that could match "bin_pfx".  See if we have
 164          * 0, 1 or more objects that actually match(es).
 165          */
 166         for (i = first; i < num && !ds->ambiguous; i++) {
 167                 current = nth_packed_object_sha1(p, i);
 168                 if (!match_sha(len, bin_pfx, current))
 169                         break;
 170                 update_candidates(ds, current);
 171         }
 172 }
 173
 174 static void find_short_packed_object(int len, const unsigned char *bin_pfx,
 175                                      struct disambiguate_state *ds)
 176 {
 177         struct packed_git *p;
 178
 179         prepare_packed_git();
 180         for (p = packed_git; p && !ds->ambiguous; p = p->next)
 181                 unique_in_pack(len, bin_pfx, p, ds);
 182 }
 183
 184 #define SHORT_NAME_NOT_FOUND (-1)
 185 #define SHORT_NAME_AMBIGUOUS (-2)
 186
 187 static int finish_object_disambiguation(struct disambiguate_state *ds,
 188                                         unsigned char *sha1)
 189 {
 190         if (ds->ambiguous)
 191                 return SHORT_NAME_AMBIGUOUS;
 192
 193         if (!ds->candidate_exists)
 194                 return SHORT_NAME_NOT_FOUND;
 195
 196         if (!ds->candidate_checked)
 197                 /*
 198                  * If this is the only candidate, there is no point
 199                  * calling the disambiguation hint callback.
 200                  *
 201                  * On the other hand, if the current candidate
 202                  * replaced an earlier candidate that did _not_ pass
 203                  * the disambiguation hint callback, then we do have
 204                  * more than one objects that match the short name
 205                  * given, so we should make sure this one matches;
 206                  * otherwise, if we discovered this one and the one
 207                  * that we previously discarded in the reverse order,
 208                  * we would end up showing different results in the
 209                  * same repository!
 210                  */
 211                 ds->candidate_ok = (!ds->disambiguate_fn_used ||
 212                                     ds->fn(ds->candidate, ds->cb_data));
 213
 214         if (!ds->candidate_ok)
 215                 return SHORT_NAME_AMBIGUOUS;
 216
 217         hashcpy(sha1, ds->candidate);
 218         return 0;
 219 }
 220
 221 static int disambiguate_commit_only(const unsigned char *sha1, void *cb_data_unused)
 222 {
 223         int kind = sha1_object_info(sha1, NULL);
 224         return kind == OBJ_COMMIT;
 225 }
 226
 227 static int disambiguate_committish_only(const unsigned char *sha1, void *cb_data_unused)
 228 {
 229         struct object *obj;
 230         int kind;
 231
 232         kind = sha1_object_info(sha1, NULL);
 233         if (kind == OBJ_COMMIT)
 234                 return 1;
 235         if (kind != OBJ_TAG)
 236                 return 0;
 237
 238         /* We need to do this the hard way... */
 239         obj = deref_tag(lookup_object(sha1), NULL, 0);
 240         if (obj && obj->type == OBJ_COMMIT)
 241                 return 1;
 242         return 0;
 243 }
 244
 245 static int get_short_sha1(const char *name, int len, unsigned char *sha1,
 246                           unsigned flags)
 247 {
 248         int i, status;
 249         char hex_pfx[40];
 250         unsigned char bin_pfx[20];
 251         struct disambiguate_state ds;
 252         int quietly = !!(flags & GET_SHA1_QUIETLY);
 253
 254         if (len < MINIMUM_ABBREV || len > 40)
 255                 return -1;
 256         hashclr(bin_pfx);
 257         memset(hex_pfx, 'x', 40);
 258         for (i = 0; i < len ;i++) {
 259                 unsigned char c = name[i];
 260                 unsigned char val;
 261                 if (c >= '0' && c <= '9')
 262                         val = c - '0';
 263                 else if (c >= 'a' && c <= 'f')
 264                         val = c - 'a' + 10;
 265                 else if (c >= 'A' && c <='F') {
 266                         val = c - 'A' + 10;
 267                         c -= 'A' - 'a';
 268                 }
 269                 else
 270                         return -1;
 271                 hex_pfx[i] = c;
 272                 if (!(i & 1))
 273                         val <<= 4;
 274                 bin_pfx[i >> 1] |= val;
 275         }
 276
 277         prepare_alt_odb();
 278
 279         memset(&ds, 0, sizeof(ds));
 280         if (flags & GET_SHA1_COMMIT)
 281                 ds.fn = disambiguate_commit_only;
 282         else if (flags & GET_SHA1_COMMITTISH)
 283                 ds.fn = disambiguate_committish_only;
 284
 285         find_short_object_filename(len, hex_pfx, &ds);
 286         find_short_packed_object(len, bin_pfx, &ds);
 287         status = finish_object_disambiguation(&ds, sha1);
 288
 289         if (!quietly && (status == SHORT_NAME_AMBIGUOUS))
 290                 return error("short SHA1 %.*s is ambiguous.", len, hex_pfx);
 291         return status;
 292 }
 293
 294 const char *find_unique_abbrev(const unsigned char *sha1, int len)
 295 {
 296         int status, exists;
 297         static char hex[41];
 298
 299         exists = has_sha1_file(sha1);
 300         memcpy(hex, sha1_to_hex(sha1), 40);
 301         if (len == 40 || !len)
 302                 return hex;
 303         while (len < 40) {
 304                 unsigned char sha1_ret[20];
 305                 status = get_short_sha1(hex, len, sha1_ret, GET_SHA1_QUIETLY);
 306                 if (exists
 307                     ? !status
 308                     : status == SHORT_NAME_NOT_FOUND) {
 309                         hex[len] = 0;
 310                         return hex;
 311                 }
 312                 len++;
 313         }
 314         return hex;
 315 }
 316
 317 static int ambiguous_path(const char *path, int len)
 318 {
 319         int slash = 1;
 320         int cnt;
 321
 322         for (cnt = 0; cnt < len; cnt++) {
 323                 switch (*path++) {
 324                 case '\0':
 325                         break;
 326                 case '/':
 327                         if (slash)
 328                                 break;
 329                         slash = 1;
 330                         continue;
 331                 case '.':
 332                         continue;
 333                 default:
 334                         slash = 0;
 335                         continue;
 336                 }
 337                 break;
 338         }
 339         return slash;
 340 }
 341
 342 static inline int upstream_mark(const char *string, int len)
 343 {
 344         const char *suffix[] = { "@{upstream}", "@{u}" };
 345         int i;
 346
 347         for (i = 0; i < ARRAY_SIZE(suffix); i++) {
 348                 int suffix_len = strlen(suffix[i]);
 349                 if (suffix_len <= len
 350                     && !memcmp(string, suffix[i], suffix_len))
 351                         return suffix_len;
 352         }
 353         return 0;
 354 }
 355
 356 static int get_sha1_1(const char *name, int len, unsigned char *sha1, unsigned lookup_flags);
 357
 358 static int get_sha1_basic(const char *str, int len, unsigned char *sha1)
 359 {
 360         static const char *warn_msg = "refname '%.*s' is ambiguous.";
 361         char *real_ref = NULL;
 362         int refs_found = 0;
 363         int at, reflog_len;
 364
 365         if (len == 40 && !get_sha1_hex(str, sha1))
 366                 return 0;
 367
 368         /* basic@{time or number or -number} format to query ref-log */
 369         reflog_len = at = 0;
 370         if (len && str[len-1] == '}') {
 371                 for (at = len-2; at >= 0; at--) {
 372                         if (str[at] == '@' && str[at+1] == '{') {
 373                                 if (!upstream_mark(str + at, len - at)) {
 374                                         reflog_len = (len-1) - (at+2);
 375                                         len = at;
 376                                 }
 377                                 break;
 378                         }
 379                 }
 380         }
 381
 382         /* Accept only unambiguous ref paths. */
 383         if (len && ambiguous_path(str, len))
 384                 return -1;
 385
 386         if (!len && reflog_len) {
 387                 struct strbuf buf = STRBUF_INIT;
 388                 int ret;
 389                 /* try the @{-N} syntax for n-th checkout */
 390                 ret = interpret_branch_name(str+at, &buf);
 391                 if (ret > 0) {
 392                         /* substitute this branch name and restart */
 393                         return get_sha1_1(buf.buf, buf.len, sha1, 0);
 394                 } else if (ret == 0) {
 395                         return -1;
 396                 }
 397                 /* allow "@{...}" to mean the current branch reflog */
 398                 refs_found = dwim_ref("HEAD", 4, sha1, &real_ref);
 399         } else if (reflog_len)
 400                 refs_found = dwim_log(str, len, sha1, &real_ref);
 401         else
 402                 refs_found = dwim_ref(str, len, sha1, &real_ref);
 403
 404         if (!refs_found)
 405                 return -1;
 406
 407         if (warn_ambiguous_refs && refs_found > 1)
 408                 warning(warn_msg, len, str);
 409
 410         if (reflog_len) {
 411                 int nth, i;
 412                 unsigned long at_time;
 413                 unsigned long co_time;
 414                 int co_tz, co_cnt;
 415
 416                 /* a @{-N} placed anywhere except the start is an error */
 417                 if (str[at+2] == '-')
 418                         return -1;
 419
 420                 /* Is it asking for N-th entry, or approxidate? */
 421                 for (i = nth = 0; 0 <= nth && i < reflog_len; i++) {
 422                         char ch = str[at+2+i];
 423                         if ('0' <= ch && ch <= '9')
 424                                 nth = nth * 10 + ch - '0';
 425                         else
 426                                 nth = -1;
 427                 }
 428                 if (100000000 <= nth) {
 429                         at_time = nth;
 430                         nth = -1;
 431                 } else if (0 <= nth)
 432                         at_time = 0;
 433                 else {
 434                         int errors = 0;
 435                         char *tmp = xstrndup(str + at + 2, reflog_len);
 436                         at_time = approxidate_careful(tmp, &errors);
 437                         free(tmp);
 438                         if (errors)
 439                                 return -1;
 440                 }
 441                 if (read_ref_at(real_ref, at_time, nth, sha1, NULL,
 442                                 &co_time, &co_tz, &co_cnt)) {
 443                         if (at_time)
 444                                 warning("Log for '%.*s' only goes "
 445                                         "back to %s.", len, str,
 446                                         show_date(co_time, co_tz, DATE_RFC2822));
 447                         else {
 448                                 free(real_ref);
 449                                 die("Log for '%.*s' only has %d entries.",
 450                                     len, str, co_cnt);
 451                         }
 452                 }
 453         }
 454
 455         free(real_ref);
 456         return 0;
 457 }
 458
 459 static int get_parent(const char *name, int len,
 460                       unsigned char *result, int idx)
 461 {
 462         unsigned char sha1[20];
 463         int ret = get_sha1_1(name, len, sha1, GET_SHA1_COMMITTISH);
 464         struct commit *commit;
 465         struct commit_list *p;
 466
 467         if (ret)
 468                 return ret;
 469         commit = lookup_commit_reference(sha1);
 470         if (!commit)
 471                 return -1;
 472         if (parse_commit(commit))
 473                 return -1;
 474         if (!idx) {
 475                 hashcpy(result, commit->object.sha1);
 476                 return 0;
 477         }
 478         p = commit->parents;
 479         while (p) {
 480                 if (!--idx) {
 481                         hashcpy(result, p->item->object.sha1);
 482                         return 0;
 483                 }
 484                 p = p->next;
 485         }
 486         return -1;
 487 }
 488
 489 static int get_nth_ancestor(const char *name, int len,
 490                             unsigned char *result, int generation)
 491 {
 492         unsigned char sha1[20];
 493         struct commit *commit;
 494         int ret;
 495
 496         ret = get_sha1_1(name, len, sha1, GET_SHA1_COMMITTISH);
 497         if (ret)
 498                 return ret;
 499         commit = lookup_commit_reference(sha1);
 500         if (!commit)
 501                 return -1;
 502
 503         while (generation--) {
 504                 if (parse_commit(commit) || !commit->parents)
 505                         return -1;
 506                 commit = commit->parents->item;
 507         }
 508         hashcpy(result, commit->object.sha1);
 509         return 0;
 510 }
 511
 512 struct object *peel_to_type(const char *name, int namelen,
 513                             struct object *o, enum object_type expected_type)
 514 {
 515         if (name && !namelen)
 516                 namelen = strlen(name);
 517         while (1) {
 518                 if (!o || (!o->parsed && !parse_object(o->sha1)))
 519                         return NULL;
 520                 if (o->type == expected_type)
 521                         return o;
 522                 if (o->type == OBJ_TAG)
 523                         o = ((struct tag*) o)->tagged;
 524                 else if (o->type == OBJ_COMMIT)
 525                         o = &(((struct commit *) o)->tree->object);
 526                 else {
 527                         if (name)
 528                                 error("%.*s: expected %s type, but the object "
 529                                       "dereferences to %s type",
 530                                       namelen, name, typename(expected_type),
 531                                       typename(o->type));
 532                         return NULL;
 533                 }
 534         }
 535 }
 536
 537 static int peel_onion(const char *name, int len, unsigned char *sha1)
 538 {
 539         unsigned char outer[20];
 540         const char *sp;
 541         unsigned int expected_type = 0;
 542         unsigned lookup_flags = 0;
 543         struct object *o;
 544
 545         /*
 546          * "ref^{type}" dereferences ref repeatedly until you cannot
 547          * dereference anymore, or you get an object of given type,
 548          * whichever comes first.  "ref^{}" means just dereference
 549          * tags until you get a non-tag.  "ref^0" is a shorthand for
 550          * "ref^{commit}".  "commit^{tree}" could be used to find the
 551          * top-level tree of the given commit.
 552          */
 553         if (len < 4 || name[len-1] != '}')
 554                 return -1;
 555
 556         for (sp = name + len - 1; name <= sp; sp--) {
 557                 int ch = *sp;
 558                 if (ch == '{' && name < sp && sp[-1] == '^')
 559                         break;
 560         }
 561         if (sp <= name)
 562                 return -1;
 563
 564         sp++; /* beginning of type name, or closing brace for empty */
 565         if (!strncmp(commit_type, sp, 6) && sp[6] == '}')
 566                 expected_type = OBJ_COMMIT;
 567         else if (!strncmp(tree_type, sp, 4) && sp[4] == '}')
 568                 expected_type = OBJ_TREE;
 569         else if (!strncmp(blob_type, sp, 4) && sp[4] == '}')
 570                 expected_type = OBJ_BLOB;
 571         else if (sp[0] == '}')
 572                 expected_type = OBJ_NONE;
 573         else if (sp[0] == '/')
 574                 expected_type = OBJ_COMMIT;
 575         else
 576                 return -1;
 577
 578         if (expected_type == OBJ_COMMIT)
 579                 lookup_flags = GET_SHA1_COMMITTISH;
 580
 581         if (get_sha1_1(name, sp - name - 2, outer, lookup_flags))
 582                 return -1;
 583
 584         o = parse_object(outer);
 585         if (!o)
 586                 return -1;
 587         if (!expected_type) {
 588                 o = deref_tag(o, name, sp - name - 2);
 589                 if (!o || (!o->parsed && !parse_object(o->sha1)))
 590                         return -1;
 591                 hashcpy(sha1, o->sha1);
 592                 return 0;
 593         }
 594
 595         /*
 596          * At this point, the syntax look correct, so
 597          * if we do not get the needed object, we should
 598          * barf.
 599          */
 600         o = peel_to_type(name, len, o, expected_type);
 601         if (!o)
 602                 return -1;
 603
 604         hashcpy(sha1, o->sha1);
 605         if (sp[0] == '/') {
 606                 /* "$commit^{/foo}" */
 607                 char *prefix;
 608                 int ret;
 609                 struct commit_list *list = NULL;
 610
 611                 /*
 612                  * $commit^{/}. Some regex implementation may reject.
 613                  * We don't need regex anyway. '' pattern always matches.
 614                  */
 615                 if (sp[1] == '}')
 616                         return 0;
 617
 618                 prefix = xstrndup(sp + 1, name + len - 1 - (sp + 1));
 619                 commit_list_insert((struct commit *)o, &list);
 620                 ret = get_sha1_oneline(prefix, sha1, list);
 621                 free(prefix);
 622                 return ret;
 623         }
 624         return 0;
 625 }
 626
 627 static int get_describe_name(const char *name, int len, unsigned char *sha1)
 628 {
 629         const char *cp;
 630         unsigned flags = GET_SHA1_QUIETLY | GET_SHA1_COMMIT;
 631
 632         for (cp = name + len - 1; name + 2 <= cp; cp--) {
 633                 char ch = *cp;
 634                 if (hexval(ch) & ~0377) {
 635                         /* We must be looking at g in "SOMETHING-g"
 636                          * for it to be describe output.
 637                          */
 638                         if (ch == 'g' && cp[-1] == '-') {
 639                                 cp++;
 640                                 len -= cp - name;
 641                                 return get_short_sha1(cp, len, sha1, flags);
 642                         }
 643                 }
 644         }
 645         return -1;
 646 }
 647
 648 static int get_sha1_1(const char *name, int len, unsigned char *sha1, unsigned lookup_flags)
 649 {
 650         int ret, has_suffix;
 651         const char *cp;
 652
 653         /*
 654          * "name~3" is "name^^^", "name~" is "name~1", and "name^" is "name^1".
 655          */
 656         has_suffix = 0;
 657         for (cp = name + len - 1; name <= cp; cp--) {
 658                 int ch = *cp;
 659                 if ('0' <= ch && ch <= '9')
 660                         continue;
 661                 if (ch == '~' || ch == '^')
 662                         has_suffix = ch;
 663                 break;
 664         }
 665
 666         if (has_suffix) {
 667                 int num = 0;
 668                 int len1 = cp - name;
 669                 cp++;
 670                 while (cp < name + len)
 671                         num = num * 10 + *cp++ - '0';
 672                 if (!num && len1 == len - 1)
 673                         num = 1;
 674                 if (has_suffix == '^')
 675                         return get_parent(name, len1, sha1, num);
 676                 /* else if (has_suffix == '~') -- goes without saying */
 677                 return get_nth_ancestor(name, len1, sha1, num);
 678         }
 679
 680         ret = peel_onion(name, len, sha1);
 681         if (!ret)
 682                 return 0;
 683
 684         ret = get_sha1_basic(name, len, sha1);
 685         if (!ret)
 686                 return 0;
 687
 688         /* It could be describe output that is "SOMETHING-gXXXX" */
 689         ret = get_describe_name(name, len, sha1);
 690         if (!ret)
 691                 return 0;
 692
 693         return get_short_sha1(name, len, sha1, lookup_flags);
 694 }
 695
 696 /*
 697  * This interprets names like ':/Initial revision of "git"' by searching
 698  * through history and returning the first commit whose message starts
 699  * the given regular expression.
 700  *
 701  * For future extension, ':/!' is reserved. If you want to match a message
 702  * beginning with a '!', you have to repeat the exclamation mark.
 703  */
 704 #define ONELINE_SEEN (1u<<20)
 705
 706 static int handle_one_ref(const char *path,
 707                 const unsigned char *sha1, int flag, void *cb_data)
 708 {
 709         struct commit_list **list = cb_data;
 710         struct object *object = parse_object(sha1);
 711         if (!object)
 712                 return 0;
 713         if (object->type == OBJ_TAG) {
 714                 object = deref_tag(object, path, strlen(path));
 715                 if (!object)
 716                         return 0;
 717         }
 718         if (object->type != OBJ_COMMIT)
 719                 return 0;
 720         commit_list_insert_by_date((struct commit *)object, list);
 721         return 0;
 722 }
 723
 724 static int get_sha1_oneline(const char *prefix, unsigned char *sha1,
 725                             struct commit_list *list)
 726 {
 727         struct commit_list *backup = NULL, *l;
 728         int found = 0;
 729         regex_t regex;
 730
 731         if (prefix[0] == '!') {
 732                 if (prefix[1] != '!')
 733                         die ("Invalid search pattern: %s", prefix);
 734                 prefix++;
 735         }
 736
 737         if (regcomp(&regex, prefix, REG_EXTENDED))
 738                 die("Invalid search pattern: %s", prefix);
 739
 740         for (l = list; l; l = l->next) {
 741                 l->item->object.flags |= ONELINE_SEEN;
 742                 commit_list_insert(l->item, &backup);
 743         }
 744         while (list) {
 745                 char *p, *to_free = NULL;
 746                 struct commit *commit;
 747                 enum object_type type;
 748                 unsigned long size;
 749                 int matches;
 750
 751                 commit = pop_most_recent_commit(&list, ONELINE_SEEN);
 752                 if (!parse_object(commit->object.sha1))
 753                         continue;
 754                 if (commit->buffer)
 755                         p = commit->buffer;
 756                 else {
 757                         p = read_sha1_file(commit->object.sha1, &type, &size);
 758                         if (!p)
 759                                 continue;
 760                         to_free = p;
 761                 }
 762
 763                 p = strstr(p, "\n\n");
 764                 matches = p && !regexec(&regex, p + 2, 0, NULL, 0);
 765                 free(to_free);
 766
 767                 if (matches) {
 768                         hashcpy(sha1, commit->object.sha1);
 769                         found = 1;
 770                         break;
 771                 }
 772         }
 773         regfree(&regex);
 774         free_commit_list(list);
 775         for (l = backup; l; l = l->next)
 776                 clear_commit_marks(l->item, ONELINE_SEEN);
 777         free_commit_list(backup);
 778         return found ? 0 : -1;
 779 }
 780
 781 struct grab_nth_branch_switch_cbdata {
 782         long cnt, alloc;
 783         struct strbuf *buf;
 784 };
 785
 786 static int grab_nth_branch_switch(unsigned char *osha1, unsigned char *nsha1,
 787                                   const char *email, unsigned long timestamp, int tz,
 788                                   const char *message, void *cb_data)
 789 {
 790         struct grab_nth_branch_switch_cbdata *cb = cb_data;
 791         const char *match = NULL, *target = NULL;
 792         size_t len;
 793         int nth;
 794
 795         if (!prefixcmp(message, "checkout: moving from ")) {
 796                 match = message + strlen("checkout: moving from ");
 797                 target = strstr(match, " to ");
 798         }
 799
 800         if (!match || !target)
 801                 return 0;
 802
 803         len = target - match;
 804         nth = cb->cnt++ % cb->alloc;
 805         strbuf_reset(&cb->buf[nth]);
 806         strbuf_add(&cb->buf[nth], match, len);
 807         return 0;
 808 }
 809
 810 /*
 811  * Parse @{-N} syntax, return the number of characters parsed
 812  * if successful; otherwise signal an error with negative value.
 813  */
 814 static int interpret_nth_prior_checkout(const char *name, struct strbuf *buf)
 815 {
 816         long nth;
 817         int i, retval;
 818         struct grab_nth_branch_switch_cbdata cb;
 819         const char *brace;
 820         char *num_end;
 821
 822         if (name[0] != '@' || name[1] != '{' || name[2] != '-')
 823                 return -1;
 824         brace = strchr(name, '}');
 825         if (!brace)
 826                 return -1;
 827         nth = strtol(name+3, &num_end, 10);
 828         if (num_end != brace)
 829                 return -1;
 830         if (nth <= 0)
 831                 return -1;
 832         cb.alloc = nth;
 833         cb.buf = xmalloc(nth * sizeof(struct strbuf));
 834         for (i = 0; i < nth; i++)
 835                 strbuf_init(&cb.buf[i], 20);
 836         cb.cnt = 0;
 837         retval = 0;
 838         for_each_recent_reflog_ent("HEAD", grab_nth_branch_switch, 40960, &cb);
 839         if (cb.cnt < nth) {
 840                 cb.cnt = 0;
 841                 for_each_reflog_ent("HEAD", grab_nth_branch_switch, &cb);
 842         }
 843         if (cb.cnt < nth)
 844                 goto release_return;
 845         i = cb.cnt % nth;
 846         strbuf_reset(buf);
 847         strbuf_add(buf, cb.buf[i].buf, cb.buf[i].len);
 848         retval = brace-name+1;
 849
 850 release_return:
 851         for (i = 0; i < nth; i++)
 852                 strbuf_release(&cb.buf[i]);
 853         free(cb.buf);
 854
 855         return retval;
 856 }
 857
 858 int get_sha1_mb(const char *name, unsigned char *sha1)
 859 {
 860         struct commit *one, *two;
 861         struct commit_list *mbs;
 862         unsigned char sha1_tmp[20];
 863         const char *dots;
 864         int st;
 865
 866         dots = strstr(name, "...");
 867         if (!dots)
 868                 return get_sha1(name, sha1);
 869         if (dots == name)
 870                 st = get_sha1("HEAD", sha1_tmp);
 871         else {
 872                 struct strbuf sb;
 873                 strbuf_init(&sb, dots - name);
 874                 strbuf_add(&sb, name, dots - name);
 875                 st = get_sha1(sb.buf, sha1_tmp);
 876                 strbuf_release(&sb);
 877         }
 878         if (st)
 879                 return st;
 880         one = lookup_commit_reference_gently(sha1_tmp, 0);
 881         if (!one)
 882                 return -1;
 883
 884         if (get_sha1(dots[3] ? (dots + 3) : "HEAD", sha1_tmp))
 885                 return -1;
 886         two = lookup_commit_reference_gently(sha1_tmp, 0);
 887         if (!two)
 888                 return -1;
 889         mbs = get_merge_bases(one, two, 1);
 890         if (!mbs || mbs->next)
 891                 st = -1;
 892         else {
 893                 st = 0;
 894                 hashcpy(sha1, mbs->item->object.sha1);
 895         }
 896         free_commit_list(mbs);
 897         return st;
 898 }
 899
 900 /*
 901  * This reads short-hand syntax that not only evaluates to a commit
 902  * object name, but also can act as if the end user spelled the name
 903  * of the branch from the command line.
 904  *
 905  * - "@{-N}" finds the name of the Nth previous branch we were on, and
 906  *   places the name of the branch in the given buf and returns the
 907  *   number of characters parsed if successful.
 908  *
 909  * - "<branch>@{upstream}" finds the name of the other ref that
 910  *   <branch> is configured to merge with (missing <branch> defaults
 911  *   to the current branch), and places the name of the branch in the
 912  *   given buf and returns the number of characters parsed if
 913  *   successful.
 914  *
 915  * If the input is not of the accepted format, it returns a negative
 916  * number to signal an error.
 917  *
 918  * If the input was ok but there are not N branch switches in the
 919  * reflog, it returns 0.
 920  */
 921 int interpret_branch_name(const char *name, struct strbuf *buf)
 922 {
 923         char *cp;
 924         struct branch *upstream;
 925         int namelen = strlen(name);
 926         int len = interpret_nth_prior_checkout(name, buf);
 927         int tmp_len;
 928
 929         if (!len)
 930                 return len; /* syntax Ok, not enough switches */
 931         if (0 < len && len == namelen)
 932                 return len; /* consumed all */
 933         else if (0 < len) {
 934                 /* we have extra data, which might need further processing */
 935                 struct strbuf tmp = STRBUF_INIT;
 936                 int used = buf->len;
 937                 int ret;
 938
 939                 strbuf_add(buf, name + len, namelen - len);
 940                 ret = interpret_branch_name(buf->buf, &tmp);
 941                 /* that data was not interpreted, remove our cruft */
 942                 if (ret < 0) {
 943                         strbuf_setlen(buf, used);
 944                         return len;
 945                 }
 946                 strbuf_reset(buf);
 947                 strbuf_addbuf(buf, &tmp);
 948                 strbuf_release(&tmp);
 949                 /* tweak for size of {-N} versus expanded ref name */
 950                 return ret - used + len;
 951         }
 952
 953         cp = strchr(name, '@');
 954         if (!cp)
 955                 return -1;
 956         tmp_len = upstream_mark(cp, namelen - (cp - name));
 957         if (!tmp_len)
 958                 return -1;
 959         len = cp + tmp_len - name;
 960         cp = xstrndup(name, cp - name);
 961         upstream = branch_get(*cp ? cp : NULL);
 962         if (!upstream
 963             || !upstream->merge
 964             || !upstream->merge[0]->dst)
 965                 return error("No upstream branch found for '%s'", cp);
 966         free(cp);
 967         cp = shorten_unambiguous_ref(upstream->merge[0]->dst, 0);
 968         strbuf_reset(buf);
 969         strbuf_addstr(buf, cp);
 970         free(cp);
 971         return len;
 972 }
 973
 974 int strbuf_branchname(struct strbuf *sb, const char *name)
 975 {
 976         int len = strlen(name);
 977         if (interpret_branch_name(name, sb) == len)
 978                 return 0;
 979         strbuf_add(sb, name, len);
 980         return len;
 981 }
 982
 983 int strbuf_check_branch_ref(struct strbuf *sb, const char *name)
 984 {
 985         strbuf_branchname(sb, name);
 986         if (name[0] == '-')
 987                 return -1;
 988         strbuf_splice(sb, 0, 0, "refs/heads/", 11);
 989         return check_refname_format(sb->buf, 0);
 990 }
 991
 992 /*
 993  * This is like "get_sha1_basic()", except it allows "sha1 expressions",
 994  * notably "xyz^" for "parent of xyz"
 995  */
 996 int get_sha1(const char *name, unsigned char *sha1)
 997 {
 998         struct object_context unused;
 999         return get_sha1_with_context(name, 0, sha1, &unused);
1000 }
1001
1002 /* Must be called only when object_name:filename doesn't exist. */
1003 static void diagnose_invalid_sha1_path(const char *prefix,
1004                                        const char *filename,
1005                                        const unsigned char *tree_sha1,
1006                                        const char *object_name)
1007 {
1008         struct stat st;
1009         unsigned char sha1[20];
1010         unsigned mode;
1011
1012         if (!prefix)
1013                 prefix = "";
1014
1015         if (!lstat(filename, &st))
1016                 die("Path '%s' exists on disk, but not in '%s'.",
1017                     filename, object_name);
1018         if (errno == ENOENT || errno == ENOTDIR) {
1019                 char *fullname = xmalloc(strlen(filename)
1020                                              + strlen(prefix) + 1);
1021                 strcpy(fullname, prefix);
1022                 strcat(fullname, filename);
1023
1024                 if (!get_tree_entry(tree_sha1, fullname,
1025                                     sha1, &mode)) {
1026                         die("Path '%s' exists, but not '%s'.\n"
1027                             "Did you mean '%s:%s' aka '%s:./%s'?",
1028                             fullname,
1029                             filename,
1030                             object_name,
1031                             fullname,
1032                             object_name,
1033                             filename);
1034                 }
1035                 die("Path '%s' does not exist in '%s'",
1036                     filename, object_name);
1037         }
1038 }
1039
1040 /* Must be called only when :stage:filename doesn't exist. */
1041 static void diagnose_invalid_index_path(int stage,
1042                                         const char *prefix,
1043                                         const char *filename)
1044 {
1045         struct stat st;
1046         struct cache_entry *ce;
1047         int pos;
1048         unsigned namelen = strlen(filename);
1049         unsigned fullnamelen;
1050         char *fullname;
1051
1052         if (!prefix)
1053                 prefix = "";
1054
1055         /* Wrong stage number? */
1056         pos = cache_name_pos(filename, namelen);
1057         if (pos < 0)
1058                 pos = -pos - 1;
1059         if (pos < active_nr) {
1060                 ce = active_cache[pos];
1061                 if (ce_namelen(ce) == namelen &&
1062                     !memcmp(ce->name, filename, namelen))
1063                         die("Path '%s' is in the index, but not at stage %d.\n"
1064                             "Did you mean ':%d:%s'?",
1065                             filename, stage,
1066                             ce_stage(ce), filename);
1067         }
1068
1069         /* Confusion between relative and absolute filenames? */
1070         fullnamelen = namelen + strlen(prefix);
1071         fullname = xmalloc(fullnamelen + 1);
1072         strcpy(fullname, prefix);
1073         strcat(fullname, filename);
1074         pos = cache_name_pos(fullname, fullnamelen);
1075         if (pos < 0)
1076                 pos = -pos - 1;
1077         if (pos < active_nr) {
1078                 ce = active_cache[pos];
1079                 if (ce_namelen(ce) == fullnamelen &&
1080                     !memcmp(ce->name, fullname, fullnamelen))
1081                         die("Path '%s' is in the index, but not '%s'.\n"
1082                             "Did you mean ':%d:%s' aka ':%d:./%s'?",
1083                             fullname, filename,
1084                             ce_stage(ce), fullname,
1085                             ce_stage(ce), filename);
1086         }
1087
1088         if (!lstat(filename, &st))
1089                 die("Path '%s' exists on disk, but not in the index.", filename);
1090         if (errno == ENOENT || errno == ENOTDIR)
1091                 die("Path '%s' does not exist (neither on disk nor in the index).",
1092                     filename);
1093
1094         free(fullname);
1095 }
1096
1097
1098 static char *resolve_relative_path(const char *rel)
1099 {
1100         if (prefixcmp(rel, "./") && prefixcmp(rel, "../"))
1101                 return NULL;
1102
1103         if (!startup_info)
1104                 die("BUG: startup_info struct is not initialized.");
1105
1106         if (!is_inside_work_tree())
1107                 die("relative path syntax can't be used outside working tree.");
1108
1109         /* die() inside prefix_path() if resolved path is outside worktree */
1110         return prefix_path(startup_info->prefix,
1111                            startup_info->prefix ? strlen(startup_info->prefix) : 0,
1112                            rel);
1113 }
1114
1115 static int get_sha1_with_context_1(const char *name,
1116                                    unsigned flags,
1117                                    const char *prefix,
1118                                    unsigned char *sha1,
1119                                    struct object_context *oc)
1120 {
1121         int ret, bracket_depth;
1122         int namelen = strlen(name);
1123         const char *cp;
1124         int only_to_die = flags & GET_SHA1_ONLY_TO_DIE;
1125
1126         memset(oc, 0, sizeof(*oc));
1127         oc->mode = S_IFINVALID;
1128         ret = get_sha1_1(name, namelen, sha1, flags);
1129         if (!ret)
1130                 return ret;
1131         /*
1132          * sha1:path --> object name of path in ent sha1
1133          * :path -> object name of absolute path in index
1134          * :./path -> object name of path relative to cwd in index
1135          * :[0-3]:path -> object name of path in index at stage
1136          * :/foo -> recent commit matching foo
1137          */
1138         if (name[0] == ':') {
1139                 int stage = 0;
1140                 struct cache_entry *ce;
1141                 char *new_path = NULL;
1142                 int pos;
1143                 if (!only_to_die && namelen > 2 && name[1] == '/') {
1144                         struct commit_list *list = NULL;
1145                         for_each_ref(handle_one_ref, &list);
1146                         return get_sha1_oneline(name + 2, sha1, list);
1147                 }
1148                 if (namelen < 3 ||
1149                     name[2] != ':' ||
1150                     name[1] < '0' || '3' < name[1])
1151                         cp = name + 1;
1152                 else {
1153                         stage = name[1] - '0';
1154                         cp = name + 3;
1155                 }
1156                 new_path = resolve_relative_path(cp);
1157                 if (!new_path) {
1158                         namelen = namelen - (cp - name);
1159                 } else {
1160                         cp = new_path;
1161                         namelen = strlen(cp);
1162                 }
1163
1164                 strncpy(oc->path, cp,
1165                         sizeof(oc->path));
1166                 oc->path[sizeof(oc->path)-1] = '\0';
1167
1168                 if (!active_cache)
1169                         read_cache();
1170                 pos = cache_name_pos(cp, namelen);
1171                 if (pos < 0)
1172                         pos = -pos - 1;
1173                 while (pos < active_nr) {
1174                         ce = active_cache[pos];
1175                         if (ce_namelen(ce) != namelen ||
1176                             memcmp(ce->name, cp, namelen))
1177                                 break;
1178                         if (ce_stage(ce) == stage) {
1179                                 hashcpy(sha1, ce->sha1);
1180                                 oc->mode = ce->ce_mode;
1181                                 free(new_path);
1182                                 return 0;
1183                         }
1184                         pos++;
1185                 }
1186                 if (only_to_die && name[1] && name[1] != '/')
1187                         diagnose_invalid_index_path(stage, prefix, cp);
1188                 free(new_path);
1189                 return -1;
1190         }
1191         for (cp = name, bracket_depth = 0; *cp; cp++) {
1192                 if (*cp == '{')
1193                         bracket_depth++;
1194                 else if (bracket_depth && *cp == '}')
1195                         bracket_depth--;
1196                 else if (!bracket_depth && *cp == ':')
1197                         break;
1198         }
1199         if (*cp == ':') {
1200                 unsigned char tree_sha1[20];
1201                 char *object_name = NULL;
1202                 if (only_to_die) {
1203                         object_name = xmalloc(cp-name+1);
1204                         strncpy(object_name, name, cp-name);
1205                         object_name[cp-name] = '\0';
1206                 }
1207                 if (!get_sha1_1(name, cp-name, tree_sha1, 0)) {
1208                         const char *filename = cp+1;
1209                         char *new_filename = NULL;
1210
1211                         new_filename = resolve_relative_path(filename);
1212                         if (new_filename)
1213                                 filename = new_filename;
1214                         ret = get_tree_entry(tree_sha1, filename, sha1, &oc->mode);
1215                         if (only_to_die) {
1216                                 diagnose_invalid_sha1_path(prefix, filename,
1217                                                            tree_sha1, object_name);
1218                                 free(object_name);
1219                         }
1220                         hashcpy(oc->tree, tree_sha1);
1221                         strncpy(oc->path, filename,
1222                                 sizeof(oc->path));
1223                         oc->path[sizeof(oc->path)-1] = '\0';
1224
1225                         free(new_filename);
1226                         return ret;
1227                 } else {
1228                         if (only_to_die)
1229                                 die("Invalid object name '%s'.", object_name);
1230                 }
1231         }
1232         return ret;
1233 }
1234
1235 /*
1236  * Call this function when you know "name" given by the end user must
1237  * name an object but it doesn't; the function _may_ die with a better
1238  * diagnostic message than "no such object 'name'", e.g. "Path 'doc' does not
1239  * exist in 'HEAD'" when given "HEAD:doc", or it may return in which case
1240  * you have a chance to diagnose the error further.
1241  */
1242 void maybe_die_on_misspelt_object_name(const char *name, const char *prefix)
1243 {
1244         struct object_context oc;
1245         unsigned char sha1[20];
1246         get_sha1_with_context_1(name, GET_SHA1_ONLY_TO_DIE, prefix, sha1, &oc);
1247 }
1248
1249 int get_sha1_with_context(const char *str, unsigned flags, unsigned char *sha1, struct object_context *orc)
1250 {
1251         return get_sha1_with_context_1(str, flags, NULL, sha1, orc);
1252 }