18fac921dd7bb067b2c21d1ad51df3818bcdf261
   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
  10static int get_sha1_oneline(const char *, unsigned char *, struct commit_list *);
  11
  12typedef int (*disambiguate_hint_fn)(const unsigned char *, void *);
  13
  14struct 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
  25static 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
  69static 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
 117static 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
 132static 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
 174static 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
 187static 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
 221static 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
 227static 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
 245static int disambiguate_tree_only(const unsigned char *sha1, void *cb_data_unused)
 246{
 247        int kind = sha1_object_info(sha1, NULL);
 248        return kind == OBJ_TREE;
 249}
 250
 251static int disambiguate_treeish_only(const unsigned char *sha1, void *cb_data_unused)
 252{
 253        struct object *obj;
 254        int kind;
 255
 256        kind = sha1_object_info(sha1, NULL);
 257        if (kind == OBJ_TREE || kind == OBJ_COMMIT)
 258                return 1;
 259        if (kind != OBJ_TAG)
 260                return 0;
 261
 262        /* We need to do this the hard way... */
 263        obj = deref_tag(lookup_object(sha1), NULL, 0);
 264        if (obj && (obj->type == OBJ_TREE || obj->type == OBJ_COMMIT))
 265                return 1;
 266        return 0;
 267}
 268
 269static int disambiguate_blob_only(const unsigned char *sha1, void *cb_data_unused)
 270{
 271        int kind = sha1_object_info(sha1, NULL);
 272        return kind == OBJ_BLOB;
 273}
 274
 275static int get_short_sha1(const char *name, int len, unsigned char *sha1,
 276                          unsigned flags)
 277{
 278        int i, status;
 279        char hex_pfx[40];
 280        unsigned char bin_pfx[20];
 281        struct disambiguate_state ds;
 282        int quietly = !!(flags & GET_SHA1_QUIETLY);
 283
 284        if (len < MINIMUM_ABBREV || len > 40)
 285                return -1;
 286        hashclr(bin_pfx);
 287        memset(hex_pfx, 'x', 40);
 288        for (i = 0; i < len ;i++) {
 289                unsigned char c = name[i];
 290                unsigned char val;
 291                if (c >= '0' && c <= '9')
 292                        val = c - '0';
 293                else if (c >= 'a' && c <= 'f')
 294                        val = c - 'a' + 10;
 295                else if (c >= 'A' && c <='F') {
 296                        val = c - 'A' + 10;
 297                        c -= 'A' - 'a';
 298                }
 299                else
 300                        return -1;
 301                hex_pfx[i] = c;
 302                if (!(i & 1))
 303                        val <<= 4;
 304                bin_pfx[i >> 1] |= val;
 305        }
 306
 307        prepare_alt_odb();
 308
 309        memset(&ds, 0, sizeof(ds));
 310        if (flags & GET_SHA1_COMMIT)
 311                ds.fn = disambiguate_commit_only;
 312        else if (flags & GET_SHA1_COMMITTISH)
 313                ds.fn = disambiguate_committish_only;
 314        else if (flags & GET_SHA1_TREE)
 315                ds.fn = disambiguate_tree_only;
 316        else if (flags & GET_SHA1_TREEISH)
 317                ds.fn = disambiguate_treeish_only;
 318        else if (flags & GET_SHA1_BLOB)
 319                ds.fn = disambiguate_blob_only;
 320
 321        find_short_object_filename(len, hex_pfx, &ds);
 322        find_short_packed_object(len, bin_pfx, &ds);
 323        status = finish_object_disambiguation(&ds, sha1);
 324
 325        if (!quietly && (status == SHORT_NAME_AMBIGUOUS))
 326                return error("short SHA1 %.*s is ambiguous.", len, hex_pfx);
 327        return status;
 328}
 329
 330const char *find_unique_abbrev(const unsigned char *sha1, int len)
 331{
 332        int status, exists;
 333        static char hex[41];
 334
 335        exists = has_sha1_file(sha1);
 336        memcpy(hex, sha1_to_hex(sha1), 40);
 337        if (len == 40 || !len)
 338                return hex;
 339        while (len < 40) {
 340                unsigned char sha1_ret[20];
 341                status = get_short_sha1(hex, len, sha1_ret, GET_SHA1_QUIETLY);
 342                if (exists
 343                    ? !status
 344                    : status == SHORT_NAME_NOT_FOUND) {
 345                        hex[len] = 0;
 346                        return hex;
 347                }
 348                len++;
 349        }
 350        return hex;
 351}
 352
 353static int ambiguous_path(const char *path, int len)
 354{
 355        int slash = 1;
 356        int cnt;
 357
 358        for (cnt = 0; cnt < len; cnt++) {
 359                switch (*path++) {
 360                case '\0':
 361                        break;
 362                case '/':
 363                        if (slash)
 364                                break;
 365                        slash = 1;
 366                        continue;
 367                case '.':
 368                        continue;
 369                default:
 370                        slash = 0;
 371                        continue;
 372                }
 373                break;
 374        }
 375        return slash;
 376}
 377
 378static inline int upstream_mark(const char *string, int len)
 379{
 380        const char *suffix[] = { "@{upstream}", "@{u}" };
 381        int i;
 382
 383        for (i = 0; i < ARRAY_SIZE(suffix); i++) {
 384                int suffix_len = strlen(suffix[i]);
 385                if (suffix_len <= len
 386                    && !memcmp(string, suffix[i], suffix_len))
 387                        return suffix_len;
 388        }
 389        return 0;
 390}
 391
 392static int get_sha1_1(const char *name, int len, unsigned char *sha1, unsigned lookup_flags);
 393
 394static int get_sha1_basic(const char *str, int len, unsigned char *sha1)
 395{
 396        static const char *warn_msg = "refname '%.*s' is ambiguous.";
 397        char *real_ref = NULL;
 398        int refs_found = 0;
 399        int at, reflog_len;
 400
 401        if (len == 40 && !get_sha1_hex(str, sha1))
 402                return 0;
 403
 404        /* basic@{time or number or -number} format to query ref-log */
 405        reflog_len = at = 0;
 406        if (len && str[len-1] == '}') {
 407                for (at = len-2; at >= 0; at--) {
 408                        if (str[at] == '@' && str[at+1] == '{') {
 409                                if (!upstream_mark(str + at, len - at)) {
 410                                        reflog_len = (len-1) - (at+2);
 411                                        len = at;
 412                                }
 413                                break;
 414                        }
 415                }
 416        }
 417
 418        /* Accept only unambiguous ref paths. */
 419        if (len && ambiguous_path(str, len))
 420                return -1;
 421
 422        if (!len && reflog_len) {
 423                struct strbuf buf = STRBUF_INIT;
 424                int ret;
 425                /* try the @{-N} syntax for n-th checkout */
 426                ret = interpret_branch_name(str+at, &buf);
 427                if (ret > 0) {
 428                        /* substitute this branch name and restart */
 429                        return get_sha1_1(buf.buf, buf.len, sha1, 0);
 430                } else if (ret == 0) {
 431                        return -1;
 432                }
 433                /* allow "@{...}" to mean the current branch reflog */
 434                refs_found = dwim_ref("HEAD", 4, sha1, &real_ref);
 435        } else if (reflog_len)
 436                refs_found = dwim_log(str, len, sha1, &real_ref);
 437        else
 438                refs_found = dwim_ref(str, len, sha1, &real_ref);
 439
 440        if (!refs_found)
 441                return -1;
 442
 443        if (warn_ambiguous_refs && refs_found > 1)
 444                warning(warn_msg, len, str);
 445
 446        if (reflog_len) {
 447                int nth, i;
 448                unsigned long at_time;
 449                unsigned long co_time;
 450                int co_tz, co_cnt;
 451
 452                /* a @{-N} placed anywhere except the start is an error */
 453                if (str[at+2] == '-')
 454                        return -1;
 455
 456                /* Is it asking for N-th entry, or approxidate? */
 457                for (i = nth = 0; 0 <= nth && i < reflog_len; i++) {
 458                        char ch = str[at+2+i];
 459                        if ('0' <= ch && ch <= '9')
 460                                nth = nth * 10 + ch - '0';
 461                        else
 462                                nth = -1;
 463                }
 464                if (100000000 <= nth) {
 465                        at_time = nth;
 466                        nth = -1;
 467                } else if (0 <= nth)
 468                        at_time = 0;
 469                else {
 470                        int errors = 0;
 471                        char *tmp = xstrndup(str + at + 2, reflog_len);
 472                        at_time = approxidate_careful(tmp, &errors);
 473                        free(tmp);
 474                        if (errors)
 475                                return -1;
 476                }
 477                if (read_ref_at(real_ref, at_time, nth, sha1, NULL,
 478                                &co_time, &co_tz, &co_cnt)) {
 479                        if (at_time)
 480                                warning("Log for '%.*s' only goes "
 481                                        "back to %s.", len, str,
 482                                        show_date(co_time, co_tz, DATE_RFC2822));
 483                        else {
 484                                free(real_ref);
 485                                die("Log for '%.*s' only has %d entries.",
 486                                    len, str, co_cnt);
 487                        }
 488                }
 489        }
 490
 491        free(real_ref);
 492        return 0;
 493}
 494
 495static int get_parent(const char *name, int len,
 496                      unsigned char *result, int idx)
 497{
 498        unsigned char sha1[20];
 499        int ret = get_sha1_1(name, len, sha1, GET_SHA1_COMMITTISH);
 500        struct commit *commit;
 501        struct commit_list *p;
 502
 503        if (ret)
 504                return ret;
 505        commit = lookup_commit_reference(sha1);
 506        if (!commit)
 507                return -1;
 508        if (parse_commit(commit))
 509                return -1;
 510        if (!idx) {
 511                hashcpy(result, commit->object.sha1);
 512                return 0;
 513        }
 514        p = commit->parents;
 515        while (p) {
 516                if (!--idx) {
 517                        hashcpy(result, p->item->object.sha1);
 518                        return 0;
 519                }
 520                p = p->next;
 521        }
 522        return -1;
 523}
 524
 525static int get_nth_ancestor(const char *name, int len,
 526                            unsigned char *result, int generation)
 527{
 528        unsigned char sha1[20];
 529        struct commit *commit;
 530        int ret;
 531
 532        ret = get_sha1_1(name, len, sha1, GET_SHA1_COMMITTISH);
 533        if (ret)
 534                return ret;
 535        commit = lookup_commit_reference(sha1);
 536        if (!commit)
 537                return -1;
 538
 539        while (generation--) {
 540                if (parse_commit(commit) || !commit->parents)
 541                        return -1;
 542                commit = commit->parents->item;
 543        }
 544        hashcpy(result, commit->object.sha1);
 545        return 0;
 546}
 547
 548struct object *peel_to_type(const char *name, int namelen,
 549                            struct object *o, enum object_type expected_type)
 550{
 551        if (name && !namelen)
 552                namelen = strlen(name);
 553        while (1) {
 554                if (!o || (!o->parsed && !parse_object(o->sha1)))
 555                        return NULL;
 556                if (o->type == expected_type)
 557                        return o;
 558                if (o->type == OBJ_TAG)
 559                        o = ((struct tag*) o)->tagged;
 560                else if (o->type == OBJ_COMMIT)
 561                        o = &(((struct commit *) o)->tree->object);
 562                else {
 563                        if (name)
 564                                error("%.*s: expected %s type, but the object "
 565                                      "dereferences to %s type",
 566                                      namelen, name, typename(expected_type),
 567                                      typename(o->type));
 568                        return NULL;
 569                }
 570        }
 571}
 572
 573static int peel_onion(const char *name, int len, unsigned char *sha1)
 574{
 575        unsigned char outer[20];
 576        const char *sp;
 577        unsigned int expected_type = 0;
 578        unsigned lookup_flags = 0;
 579        struct object *o;
 580
 581        /*
 582         * "ref^{type}" dereferences ref repeatedly until you cannot
 583         * dereference anymore, or you get an object of given type,
 584         * whichever comes first.  "ref^{}" means just dereference
 585         * tags until you get a non-tag.  "ref^0" is a shorthand for
 586         * "ref^{commit}".  "commit^{tree}" could be used to find the
 587         * top-level tree of the given commit.
 588         */
 589        if (len < 4 || name[len-1] != '}')
 590                return -1;
 591
 592        for (sp = name + len - 1; name <= sp; sp--) {
 593                int ch = *sp;
 594                if (ch == '{' && name < sp && sp[-1] == '^')
 595                        break;
 596        }
 597        if (sp <= name)
 598                return -1;
 599
 600        sp++; /* beginning of type name, or closing brace for empty */
 601        if (!strncmp(commit_type, sp, 6) && sp[6] == '}')
 602                expected_type = OBJ_COMMIT;
 603        else if (!strncmp(tree_type, sp, 4) && sp[4] == '}')
 604                expected_type = OBJ_TREE;
 605        else if (!strncmp(blob_type, sp, 4) && sp[4] == '}')
 606                expected_type = OBJ_BLOB;
 607        else if (sp[0] == '}')
 608                expected_type = OBJ_NONE;
 609        else if (sp[0] == '/')
 610                expected_type = OBJ_COMMIT;
 611        else
 612                return -1;
 613
 614        if (expected_type == OBJ_COMMIT)
 615                lookup_flags = GET_SHA1_COMMITTISH;
 616
 617        if (get_sha1_1(name, sp - name - 2, outer, lookup_flags))
 618                return -1;
 619
 620        o = parse_object(outer);
 621        if (!o)
 622                return -1;
 623        if (!expected_type) {
 624                o = deref_tag(o, name, sp - name - 2);
 625                if (!o || (!o->parsed && !parse_object(o->sha1)))
 626                        return -1;
 627                hashcpy(sha1, o->sha1);
 628                return 0;
 629        }
 630
 631        /*
 632         * At this point, the syntax look correct, so
 633         * if we do not get the needed object, we should
 634         * barf.
 635         */
 636        o = peel_to_type(name, len, o, expected_type);
 637        if (!o)
 638                return -1;
 639
 640        hashcpy(sha1, o->sha1);
 641        if (sp[0] == '/') {
 642                /* "$commit^{/foo}" */
 643                char *prefix;
 644                int ret;
 645                struct commit_list *list = NULL;
 646
 647                /*
 648                 * $commit^{/}. Some regex implementation may reject.
 649                 * We don't need regex anyway. '' pattern always matches.
 650                 */
 651                if (sp[1] == '}')
 652                        return 0;
 653
 654                prefix = xstrndup(sp + 1, name + len - 1 - (sp + 1));
 655                commit_list_insert((struct commit *)o, &list);
 656                ret = get_sha1_oneline(prefix, sha1, list);
 657                free(prefix);
 658                return ret;
 659        }
 660        return 0;
 661}
 662
 663static int get_describe_name(const char *name, int len, unsigned char *sha1)
 664{
 665        const char *cp;
 666        unsigned flags = GET_SHA1_QUIETLY | GET_SHA1_COMMIT;
 667
 668        for (cp = name + len - 1; name + 2 <= cp; cp--) {
 669                char ch = *cp;
 670                if (hexval(ch) & ~0377) {
 671                        /* We must be looking at g in "SOMETHING-g"
 672                         * for it to be describe output.
 673                         */
 674                        if (ch == 'g' && cp[-1] == '-') {
 675                                cp++;
 676                                len -= cp - name;
 677                                return get_short_sha1(cp, len, sha1, flags);
 678                        }
 679                }
 680        }
 681        return -1;
 682}
 683
 684static int get_sha1_1(const char *name, int len, unsigned char *sha1, unsigned lookup_flags)
 685{
 686        int ret, has_suffix;
 687        const char *cp;
 688
 689        /*
 690         * "name~3" is "name^^^", "name~" is "name~1", and "name^" is "name^1".
 691         */
 692        has_suffix = 0;
 693        for (cp = name + len - 1; name <= cp; cp--) {
 694                int ch = *cp;
 695                if ('0' <= ch && ch <= '9')
 696                        continue;
 697                if (ch == '~' || ch == '^')
 698                        has_suffix = ch;
 699                break;
 700        }
 701
 702        if (has_suffix) {
 703                int num = 0;
 704                int len1 = cp - name;
 705                cp++;
 706                while (cp < name + len)
 707                        num = num * 10 + *cp++ - '0';
 708                if (!num && len1 == len - 1)
 709                        num = 1;
 710                if (has_suffix == '^')
 711                        return get_parent(name, len1, sha1, num);
 712                /* else if (has_suffix == '~') -- goes without saying */
 713                return get_nth_ancestor(name, len1, sha1, num);
 714        }
 715
 716        ret = peel_onion(name, len, sha1);
 717        if (!ret)
 718                return 0;
 719
 720        ret = get_sha1_basic(name, len, sha1);
 721        if (!ret)
 722                return 0;
 723
 724        /* It could be describe output that is "SOMETHING-gXXXX" */
 725        ret = get_describe_name(name, len, sha1);
 726        if (!ret)
 727                return 0;
 728
 729        return get_short_sha1(name, len, sha1, lookup_flags);
 730}
 731
 732/*
 733 * This interprets names like ':/Initial revision of "git"' by searching
 734 * through history and returning the first commit whose message starts
 735 * the given regular expression.
 736 *
 737 * For future extension, ':/!' is reserved. If you want to match a message
 738 * beginning with a '!', you have to repeat the exclamation mark.
 739 */
 740#define ONELINE_SEEN (1u<<20)
 741
 742static int handle_one_ref(const char *path,
 743                const unsigned char *sha1, int flag, void *cb_data)
 744{
 745        struct commit_list **list = cb_data;
 746        struct object *object = parse_object(sha1);
 747        if (!object)
 748                return 0;
 749        if (object->type == OBJ_TAG) {
 750                object = deref_tag(object, path, strlen(path));
 751                if (!object)
 752                        return 0;
 753        }
 754        if (object->type != OBJ_COMMIT)
 755                return 0;
 756        commit_list_insert_by_date((struct commit *)object, list);
 757        return 0;
 758}
 759
 760static int get_sha1_oneline(const char *prefix, unsigned char *sha1,
 761                            struct commit_list *list)
 762{
 763        struct commit_list *backup = NULL, *l;
 764        int found = 0;
 765        regex_t regex;
 766
 767        if (prefix[0] == '!') {
 768                if (prefix[1] != '!')
 769                        die ("Invalid search pattern: %s", prefix);
 770                prefix++;
 771        }
 772
 773        if (regcomp(&regex, prefix, REG_EXTENDED))
 774                die("Invalid search pattern: %s", prefix);
 775
 776        for (l = list; l; l = l->next) {
 777                l->item->object.flags |= ONELINE_SEEN;
 778                commit_list_insert(l->item, &backup);
 779        }
 780        while (list) {
 781                char *p, *to_free = NULL;
 782                struct commit *commit;
 783                enum object_type type;
 784                unsigned long size;
 785                int matches;
 786
 787                commit = pop_most_recent_commit(&list, ONELINE_SEEN);
 788                if (!parse_object(commit->object.sha1))
 789                        continue;
 790                if (commit->buffer)
 791                        p = commit->buffer;
 792                else {
 793                        p = read_sha1_file(commit->object.sha1, &type, &size);
 794                        if (!p)
 795                                continue;
 796                        to_free = p;
 797                }
 798
 799                p = strstr(p, "\n\n");
 800                matches = p && !regexec(&regex, p + 2, 0, NULL, 0);
 801                free(to_free);
 802
 803                if (matches) {
 804                        hashcpy(sha1, commit->object.sha1);
 805                        found = 1;
 806                        break;
 807                }
 808        }
 809        regfree(&regex);
 810        free_commit_list(list);
 811        for (l = backup; l; l = l->next)
 812                clear_commit_marks(l->item, ONELINE_SEEN);
 813        free_commit_list(backup);
 814        return found ? 0 : -1;
 815}
 816
 817struct grab_nth_branch_switch_cbdata {
 818        long cnt, alloc;
 819        struct strbuf *buf;
 820};
 821
 822static int grab_nth_branch_switch(unsigned char *osha1, unsigned char *nsha1,
 823                                  const char *email, unsigned long timestamp, int tz,
 824                                  const char *message, void *cb_data)
 825{
 826        struct grab_nth_branch_switch_cbdata *cb = cb_data;
 827        const char *match = NULL, *target = NULL;
 828        size_t len;
 829        int nth;
 830
 831        if (!prefixcmp(message, "checkout: moving from ")) {
 832                match = message + strlen("checkout: moving from ");
 833                target = strstr(match, " to ");
 834        }
 835
 836        if (!match || !target)
 837                return 0;
 838
 839        len = target - match;
 840        nth = cb->cnt++ % cb->alloc;
 841        strbuf_reset(&cb->buf[nth]);
 842        strbuf_add(&cb->buf[nth], match, len);
 843        return 0;
 844}
 845
 846/*
 847 * Parse @{-N} syntax, return the number of characters parsed
 848 * if successful; otherwise signal an error with negative value.
 849 */
 850static int interpret_nth_prior_checkout(const char *name, struct strbuf *buf)
 851{
 852        long nth;
 853        int i, retval;
 854        struct grab_nth_branch_switch_cbdata cb;
 855        const char *brace;
 856        char *num_end;
 857
 858        if (name[0] != '@' || name[1] != '{' || name[2] != '-')
 859                return -1;
 860        brace = strchr(name, '}');
 861        if (!brace)
 862                return -1;
 863        nth = strtol(name+3, &num_end, 10);
 864        if (num_end != brace)
 865                return -1;
 866        if (nth <= 0)
 867                return -1;
 868        cb.alloc = nth;
 869        cb.buf = xmalloc(nth * sizeof(struct strbuf));
 870        for (i = 0; i < nth; i++)
 871                strbuf_init(&cb.buf[i], 20);
 872        cb.cnt = 0;
 873        retval = 0;
 874        for_each_recent_reflog_ent("HEAD", grab_nth_branch_switch, 40960, &cb);
 875        if (cb.cnt < nth) {
 876                cb.cnt = 0;
 877                for_each_reflog_ent("HEAD", grab_nth_branch_switch, &cb);
 878        }
 879        if (cb.cnt < nth)
 880                goto release_return;
 881        i = cb.cnt % nth;
 882        strbuf_reset(buf);
 883        strbuf_add(buf, cb.buf[i].buf, cb.buf[i].len);
 884        retval = brace-name+1;
 885
 886release_return:
 887        for (i = 0; i < nth; i++)
 888                strbuf_release(&cb.buf[i]);
 889        free(cb.buf);
 890
 891        return retval;
 892}
 893
 894int get_sha1_mb(const char *name, unsigned char *sha1)
 895{
 896        struct commit *one, *two;
 897        struct commit_list *mbs;
 898        unsigned char sha1_tmp[20];
 899        const char *dots;
 900        int st;
 901
 902        dots = strstr(name, "...");
 903        if (!dots)
 904                return get_sha1(name, sha1);
 905        if (dots == name)
 906                st = get_sha1("HEAD", sha1_tmp);
 907        else {
 908                struct strbuf sb;
 909                strbuf_init(&sb, dots - name);
 910                strbuf_add(&sb, name, dots - name);
 911                st = get_sha1_committish(sb.buf, sha1_tmp);
 912                strbuf_release(&sb);
 913        }
 914        if (st)
 915                return st;
 916        one = lookup_commit_reference_gently(sha1_tmp, 0);
 917        if (!one)
 918                return -1;
 919
 920        if (get_sha1_committish(dots[3] ? (dots + 3) : "HEAD", sha1_tmp))
 921                return -1;
 922        two = lookup_commit_reference_gently(sha1_tmp, 0);
 923        if (!two)
 924                return -1;
 925        mbs = get_merge_bases(one, two, 1);
 926        if (!mbs || mbs->next)
 927                st = -1;
 928        else {
 929                st = 0;
 930                hashcpy(sha1, mbs->item->object.sha1);
 931        }
 932        free_commit_list(mbs);
 933        return st;
 934}
 935
 936/*
 937 * This reads short-hand syntax that not only evaluates to a commit
 938 * object name, but also can act as if the end user spelled the name
 939 * of the branch from the command line.
 940 *
 941 * - "@{-N}" finds the name of the Nth previous branch we were on, and
 942 *   places the name of the branch in the given buf and returns the
 943 *   number of characters parsed if successful.
 944 *
 945 * - "<branch>@{upstream}" finds the name of the other ref that
 946 *   <branch> is configured to merge with (missing <branch> defaults
 947 *   to the current branch), and places the name of the branch in the
 948 *   given buf and returns the number of characters parsed if
 949 *   successful.
 950 *
 951 * If the input is not of the accepted format, it returns a negative
 952 * number to signal an error.
 953 *
 954 * If the input was ok but there are not N branch switches in the
 955 * reflog, it returns 0.
 956 */
 957int interpret_branch_name(const char *name, struct strbuf *buf)
 958{
 959        char *cp;
 960        struct branch *upstream;
 961        int namelen = strlen(name);
 962        int len = interpret_nth_prior_checkout(name, buf);
 963        int tmp_len;
 964
 965        if (!len)
 966                return len; /* syntax Ok, not enough switches */
 967        if (0 < len && len == namelen)
 968                return len; /* consumed all */
 969        else if (0 < len) {
 970                /* we have extra data, which might need further processing */
 971                struct strbuf tmp = STRBUF_INIT;
 972                int used = buf->len;
 973                int ret;
 974
 975                strbuf_add(buf, name + len, namelen - len);
 976                ret = interpret_branch_name(buf->buf, &tmp);
 977                /* that data was not interpreted, remove our cruft */
 978                if (ret < 0) {
 979                        strbuf_setlen(buf, used);
 980                        return len;
 981                }
 982                strbuf_reset(buf);
 983                strbuf_addbuf(buf, &tmp);
 984                strbuf_release(&tmp);
 985                /* tweak for size of {-N} versus expanded ref name */
 986                return ret - used + len;
 987        }
 988
 989        cp = strchr(name, '@');
 990        if (!cp)
 991                return -1;
 992        tmp_len = upstream_mark(cp, namelen - (cp - name));
 993        if (!tmp_len)
 994                return -1;
 995        len = cp + tmp_len - name;
 996        cp = xstrndup(name, cp - name);
 997        upstream = branch_get(*cp ? cp : NULL);
 998        if (!upstream
 999            || !upstream->merge
1000            || !upstream->merge[0]->dst)
1001                return error("No upstream branch found for '%s'", cp);
1002        free(cp);
1003        cp = shorten_unambiguous_ref(upstream->merge[0]->dst, 0);
1004        strbuf_reset(buf);
1005        strbuf_addstr(buf, cp);
1006        free(cp);
1007        return len;
1008}
1009
1010int strbuf_branchname(struct strbuf *sb, const char *name)
1011{
1012        int len = strlen(name);
1013        if (interpret_branch_name(name, sb) == len)
1014                return 0;
1015        strbuf_add(sb, name, len);
1016        return len;
1017}
1018
1019int strbuf_check_branch_ref(struct strbuf *sb, const char *name)
1020{
1021        strbuf_branchname(sb, name);
1022        if (name[0] == '-')
1023                return -1;
1024        strbuf_splice(sb, 0, 0, "refs/heads/", 11);
1025        return check_refname_format(sb->buf, 0);
1026}
1027
1028/*
1029 * This is like "get_sha1_basic()", except it allows "sha1 expressions",
1030 * notably "xyz^" for "parent of xyz"
1031 */
1032int get_sha1(const char *name, unsigned char *sha1)
1033{
1034        struct object_context unused;
1035        return get_sha1_with_context(name, 0, sha1, &unused);
1036}
1037
1038/*
1039 * Many callers know that the user meant to name a committish by
1040 * syntactical positions where the object name appears.  Calling this
1041 * function allows the machinery to disambiguate shorter-than-unique
1042 * abbreviated object names between committish and others.
1043 *
1044 * Note that this does NOT error out when the named object is not a
1045 * committish. It is merely to give a hint to the disambiguation
1046 * machinery.
1047 */
1048int get_sha1_committish(const char *name, unsigned char *sha1)
1049{
1050        struct object_context unused;
1051        return get_sha1_with_context(name, GET_SHA1_COMMITTISH,
1052                                     sha1, &unused);
1053}
1054
1055int get_sha1_treeish(const char *name, unsigned char *sha1)
1056{
1057        struct object_context unused;
1058        return get_sha1_with_context(name, GET_SHA1_TREEISH,
1059                                     sha1, &unused);
1060}
1061
1062int get_sha1_commit(const char *name, unsigned char *sha1)
1063{
1064        struct object_context unused;
1065        return get_sha1_with_context(name, GET_SHA1_COMMIT,
1066                                     sha1, &unused);
1067}
1068
1069int get_sha1_tree(const char *name, unsigned char *sha1)
1070{
1071        struct object_context unused;
1072        return get_sha1_with_context(name, GET_SHA1_TREE,
1073                                     sha1, &unused);
1074}
1075
1076int get_sha1_blob(const char *name, unsigned char *sha1)
1077{
1078        struct object_context unused;
1079        return get_sha1_with_context(name, GET_SHA1_BLOB,
1080                                     sha1, &unused);
1081}
1082
1083/* Must be called only when object_name:filename doesn't exist. */
1084static void diagnose_invalid_sha1_path(const char *prefix,
1085                                       const char *filename,
1086                                       const unsigned char *tree_sha1,
1087                                       const char *object_name)
1088{
1089        struct stat st;
1090        unsigned char sha1[20];
1091        unsigned mode;
1092
1093        if (!prefix)
1094                prefix = "";
1095
1096        if (!lstat(filename, &st))
1097                die("Path '%s' exists on disk, but not in '%s'.",
1098                    filename, object_name);
1099        if (errno == ENOENT || errno == ENOTDIR) {
1100                char *fullname = xmalloc(strlen(filename)
1101                                             + strlen(prefix) + 1);
1102                strcpy(fullname, prefix);
1103                strcat(fullname, filename);
1104
1105                if (!get_tree_entry(tree_sha1, fullname,
1106                                    sha1, &mode)) {
1107                        die("Path '%s' exists, but not '%s'.\n"
1108                            "Did you mean '%s:%s' aka '%s:./%s'?",
1109                            fullname,
1110                            filename,
1111                            object_name,
1112                            fullname,
1113                            object_name,
1114                            filename);
1115                }
1116                die("Path '%s' does not exist in '%s'",
1117                    filename, object_name);
1118        }
1119}
1120
1121/* Must be called only when :stage:filename doesn't exist. */
1122static void diagnose_invalid_index_path(int stage,
1123                                        const char *prefix,
1124                                        const char *filename)
1125{
1126        struct stat st;
1127        struct cache_entry *ce;
1128        int pos;
1129        unsigned namelen = strlen(filename);
1130        unsigned fullnamelen;
1131        char *fullname;
1132
1133        if (!prefix)
1134                prefix = "";
1135
1136        /* Wrong stage number? */
1137        pos = cache_name_pos(filename, namelen);
1138        if (pos < 0)
1139                pos = -pos - 1;
1140        if (pos < active_nr) {
1141                ce = active_cache[pos];
1142                if (ce_namelen(ce) == namelen &&
1143                    !memcmp(ce->name, filename, namelen))
1144                        die("Path '%s' is in the index, but not at stage %d.\n"
1145                            "Did you mean ':%d:%s'?",
1146                            filename, stage,
1147                            ce_stage(ce), filename);
1148        }
1149
1150        /* Confusion between relative and absolute filenames? */
1151        fullnamelen = namelen + strlen(prefix);
1152        fullname = xmalloc(fullnamelen + 1);
1153        strcpy(fullname, prefix);
1154        strcat(fullname, filename);
1155        pos = cache_name_pos(fullname, fullnamelen);
1156        if (pos < 0)
1157                pos = -pos - 1;
1158        if (pos < active_nr) {
1159                ce = active_cache[pos];
1160                if (ce_namelen(ce) == fullnamelen &&
1161                    !memcmp(ce->name, fullname, fullnamelen))
1162                        die("Path '%s' is in the index, but not '%s'.\n"
1163                            "Did you mean ':%d:%s' aka ':%d:./%s'?",
1164                            fullname, filename,
1165                            ce_stage(ce), fullname,
1166                            ce_stage(ce), filename);
1167        }
1168
1169        if (!lstat(filename, &st))
1170                die("Path '%s' exists on disk, but not in the index.", filename);
1171        if (errno == ENOENT || errno == ENOTDIR)
1172                die("Path '%s' does not exist (neither on disk nor in the index).",
1173                    filename);
1174
1175        free(fullname);
1176}
1177
1178
1179static char *resolve_relative_path(const char *rel)
1180{
1181        if (prefixcmp(rel, "./") && prefixcmp(rel, "../"))
1182                return NULL;
1183
1184        if (!startup_info)
1185                die("BUG: startup_info struct is not initialized.");
1186
1187        if (!is_inside_work_tree())
1188                die("relative path syntax can't be used outside working tree.");
1189
1190        /* die() inside prefix_path() if resolved path is outside worktree */
1191        return prefix_path(startup_info->prefix,
1192                           startup_info->prefix ? strlen(startup_info->prefix) : 0,
1193                           rel);
1194}
1195
1196static int get_sha1_with_context_1(const char *name,
1197                                   unsigned flags,
1198                                   const char *prefix,
1199                                   unsigned char *sha1,
1200                                   struct object_context *oc)
1201{
1202        int ret, bracket_depth;
1203        int namelen = strlen(name);
1204        const char *cp;
1205        int only_to_die = flags & GET_SHA1_ONLY_TO_DIE;
1206
1207        memset(oc, 0, sizeof(*oc));
1208        oc->mode = S_IFINVALID;
1209        ret = get_sha1_1(name, namelen, sha1, flags);
1210        if (!ret)
1211                return ret;
1212        /*
1213         * sha1:path --> object name of path in ent sha1
1214         * :path -> object name of absolute path in index
1215         * :./path -> object name of path relative to cwd in index
1216         * :[0-3]:path -> object name of path in index at stage
1217         * :/foo -> recent commit matching foo
1218         */
1219        if (name[0] == ':') {
1220                int stage = 0;
1221                struct cache_entry *ce;
1222                char *new_path = NULL;
1223                int pos;
1224                if (!only_to_die && namelen > 2 && name[1] == '/') {
1225                        struct commit_list *list = NULL;
1226                        for_each_ref(handle_one_ref, &list);
1227                        return get_sha1_oneline(name + 2, sha1, list);
1228                }
1229                if (namelen < 3 ||
1230                    name[2] != ':' ||
1231                    name[1] < '0' || '3' < name[1])
1232                        cp = name + 1;
1233                else {
1234                        stage = name[1] - '0';
1235                        cp = name + 3;
1236                }
1237                new_path = resolve_relative_path(cp);
1238                if (!new_path) {
1239                        namelen = namelen - (cp - name);
1240                } else {
1241                        cp = new_path;
1242                        namelen = strlen(cp);
1243                }
1244
1245                strncpy(oc->path, cp,
1246                        sizeof(oc->path));
1247                oc->path[sizeof(oc->path)-1] = '\0';
1248
1249                if (!active_cache)
1250                        read_cache();
1251                pos = cache_name_pos(cp, namelen);
1252                if (pos < 0)
1253                        pos = -pos - 1;
1254                while (pos < active_nr) {
1255                        ce = active_cache[pos];
1256                        if (ce_namelen(ce) != namelen ||
1257                            memcmp(ce->name, cp, namelen))
1258                                break;
1259                        if (ce_stage(ce) == stage) {
1260                                hashcpy(sha1, ce->sha1);
1261                                oc->mode = ce->ce_mode;
1262                                free(new_path);
1263                                return 0;
1264                        }
1265                        pos++;
1266                }
1267                if (only_to_die && name[1] && name[1] != '/')
1268                        diagnose_invalid_index_path(stage, prefix, cp);
1269                free(new_path);
1270                return -1;
1271        }
1272        for (cp = name, bracket_depth = 0; *cp; cp++) {
1273                if (*cp == '{')
1274                        bracket_depth++;
1275                else if (bracket_depth && *cp == '}')
1276                        bracket_depth--;
1277                else if (!bracket_depth && *cp == ':')
1278                        break;
1279        }
1280        if (*cp == ':') {
1281                unsigned char tree_sha1[20];
1282                char *object_name = NULL;
1283                if (only_to_die) {
1284                        object_name = xmalloc(cp-name+1);
1285                        strncpy(object_name, name, cp-name);
1286                        object_name[cp-name] = '\0';
1287                }
1288                if (!get_sha1_1(name, cp-name, tree_sha1, GET_SHA1_TREEISH)) {
1289                        const char *filename = cp+1;
1290                        char *new_filename = NULL;
1291
1292                        new_filename = resolve_relative_path(filename);
1293                        if (new_filename)
1294                                filename = new_filename;
1295                        ret = get_tree_entry(tree_sha1, filename, sha1, &oc->mode);
1296                        if (only_to_die) {
1297                                diagnose_invalid_sha1_path(prefix, filename,
1298                                                           tree_sha1, object_name);
1299                                free(object_name);
1300                        }
1301                        hashcpy(oc->tree, tree_sha1);
1302                        strncpy(oc->path, filename,
1303                                sizeof(oc->path));
1304                        oc->path[sizeof(oc->path)-1] = '\0';
1305
1306                        free(new_filename);
1307                        return ret;
1308                } else {
1309                        if (only_to_die)
1310                                die("Invalid object name '%s'.", object_name);
1311                }
1312        }
1313        return ret;
1314}
1315
1316/*
1317 * Call this function when you know "name" given by the end user must
1318 * name an object but it doesn't; the function _may_ die with a better
1319 * diagnostic message than "no such object 'name'", e.g. "Path 'doc' does not
1320 * exist in 'HEAD'" when given "HEAD:doc", or it may return in which case
1321 * you have a chance to diagnose the error further.
1322 */
1323void maybe_die_on_misspelt_object_name(const char *name, const char *prefix)
1324{
1325        struct object_context oc;
1326        unsigned char sha1[20];
1327        get_sha1_with_context_1(name, GET_SHA1_ONLY_TO_DIE, prefix, sha1, &oc);
1328}
1329
1330int get_sha1_with_context(const char *str, unsigned flags, unsigned char *sha1, struct object_context *orc)
1331{
1332        return get_sha1_with_context_1(str, flags, NULL, sha1, orc);
1333}