Merge branch 'wc/find-commit-with-pattern-on-detached-head'
authorJunio C Hamano <gitster@pobox.com>
Tue, 24 Jul 2018 21:50:49 +0000 (14:50 -0700)
committerJunio C Hamano <gitster@pobox.com>
Tue, 24 Jul 2018 21:50:49 +0000 (14:50 -0700)
"git rev-parse ':/substring'" did not consider the history leading
only to HEAD when looking for a commit with the given substring,
when the HEAD is detached. This has been fixed.

* wc/find-commit-with-pattern-on-detached-head:
sha1-name.c: for ":/", find detached HEAD commits

1  2 
Documentation/revisions.txt
sha1-name.c
t/t4208-log-magic-pathspec.sh
Simple merge
diff --cc sha1-name.c
index 60d9ef3c7e7108c859647656972c171cce4e7d7f,0000000000000000000000000000000000000000..641ca12f91bad8f5aa5739ac92a3a9498e4d42e9
mode 100644,000000..100644
--- /dev/null
@@@ -1,1767 -1,0 +1,1768 @@@
 +#include "cache.h"
 +#include "config.h"
 +#include "tag.h"
 +#include "commit.h"
 +#include "tree.h"
 +#include "blob.h"
 +#include "tree-walk.h"
 +#include "refs.h"
 +#include "remote.h"
 +#include "dir.h"
 +#include "sha1-array.h"
 +#include "packfile.h"
 +#include "object-store.h"
 +#include "repository.h"
 +
 +static int get_oid_oneline(const char *, struct object_id *, struct commit_list *);
 +
 +typedef int (*disambiguate_hint_fn)(const struct object_id *, void *);
 +
 +struct disambiguate_state {
 +      int len; /* length of prefix in hex chars */
 +      char hex_pfx[GIT_MAX_HEXSZ + 1];
 +      struct object_id bin_pfx;
 +
 +      disambiguate_hint_fn fn;
 +      void *cb_data;
 +      struct object_id candidate;
 +      unsigned candidate_exists:1;
 +      unsigned candidate_checked:1;
 +      unsigned candidate_ok:1;
 +      unsigned disambiguate_fn_used:1;
 +      unsigned ambiguous:1;
 +      unsigned always_call_fn:1;
 +};
 +
 +static void update_candidates(struct disambiguate_state *ds, const struct object_id *current)
 +{
 +      if (ds->always_call_fn) {
 +              ds->ambiguous = ds->fn(current, ds->cb_data) ? 1 : 0;
 +              return;
 +      }
 +      if (!ds->candidate_exists) {
 +              /* this is the first candidate */
 +              oidcpy(&ds->candidate, current);
 +              ds->candidate_exists = 1;
 +              return;
 +      } else if (!oidcmp(&ds->candidate, current)) {
 +              /* the same as what we already have seen */
 +              return;
 +      }
 +
 +      if (!ds->fn) {
 +              /* cannot disambiguate between ds->candidate and current */
 +              ds->ambiguous = 1;
 +              return;
 +      }
 +
 +      if (!ds->candidate_checked) {
 +              ds->candidate_ok = ds->fn(&ds->candidate, ds->cb_data);
 +              ds->disambiguate_fn_used = 1;
 +              ds->candidate_checked = 1;
 +      }
 +
 +      if (!ds->candidate_ok) {
 +              /* discard the candidate; we know it does not satisfy fn */
 +              oidcpy(&ds->candidate, current);
 +              ds->candidate_checked = 0;
 +              return;
 +      }
 +
 +      /* if we reach this point, we know ds->candidate satisfies fn */
 +      if (ds->fn(current, ds->cb_data)) {
 +              /*
 +               * if both current and candidate satisfy fn, we cannot
 +               * disambiguate.
 +               */
 +              ds->candidate_ok = 0;
 +              ds->ambiguous = 1;
 +      }
 +
 +      /* otherwise, current can be discarded and candidate is still good */
 +}
 +
 +static int append_loose_object(const struct object_id *oid, const char *path,
 +                             void *data)
 +{
 +      oid_array_append(data, oid);
 +      return 0;
 +}
 +
 +static int match_sha(unsigned, const unsigned char *, const unsigned char *);
 +
 +static void find_short_object_filename(struct disambiguate_state *ds)
 +{
 +      int subdir_nr = ds->bin_pfx.hash[0];
 +      struct alternate_object_database *alt;
 +      static struct alternate_object_database *fakeent;
 +
 +      if (!fakeent) {
 +              /*
 +               * Create a "fake" alternate object database that
 +               * points to our own object database, to make it
 +               * easier to get a temporary working space in
 +               * alt->name/alt->base while iterating over the
 +               * object databases including our own.
 +               */
 +              fakeent = alloc_alt_odb(get_object_directory());
 +      }
 +      fakeent->next = the_repository->objects->alt_odb_list;
 +
 +      for (alt = fakeent; alt && !ds->ambiguous; alt = alt->next) {
 +              int pos;
 +
 +              if (!alt->loose_objects_subdir_seen[subdir_nr]) {
 +                      struct strbuf *buf = alt_scratch_buf(alt);
 +                      for_each_file_in_obj_subdir(subdir_nr, buf,
 +                                                  append_loose_object,
 +                                                  NULL, NULL,
 +                                                  &alt->loose_objects_cache);
 +                      alt->loose_objects_subdir_seen[subdir_nr] = 1;
 +              }
 +
 +              pos = oid_array_lookup(&alt->loose_objects_cache, &ds->bin_pfx);
 +              if (pos < 0)
 +                      pos = -1 - pos;
 +              while (!ds->ambiguous && pos < alt->loose_objects_cache.nr) {
 +                      const struct object_id *oid;
 +                      oid = alt->loose_objects_cache.oid + pos;
 +                      if (!match_sha(ds->len, ds->bin_pfx.hash, oid->hash))
 +                              break;
 +                      update_candidates(ds, oid);
 +                      pos++;
 +              }
 +      }
 +}
 +
 +static int match_sha(unsigned len, const unsigned char *a, const unsigned char *b)
 +{
 +      do {
 +              if (*a != *b)
 +                      return 0;
 +              a++;
 +              b++;
 +              len -= 2;
 +      } while (len > 1);
 +      if (len)
 +              if ((*a ^ *b) & 0xf0)
 +                      return 0;
 +      return 1;
 +}
 +
 +static void unique_in_pack(struct packed_git *p,
 +                         struct disambiguate_state *ds)
 +{
 +      uint32_t num, i, first = 0;
 +      const struct object_id *current = NULL;
 +
 +      if (open_pack_index(p) || !p->num_objects)
 +              return;
 +
 +      num = p->num_objects;
 +      bsearch_pack(&ds->bin_pfx, p, &first);
 +
 +      /*
 +       * At this point, "first" is the location of the lowest object
 +       * with an object name that could match "bin_pfx".  See if we have
 +       * 0, 1 or more objects that actually match(es).
 +       */
 +      for (i = first; i < num && !ds->ambiguous; i++) {
 +              struct object_id oid;
 +              current = nth_packed_object_oid(&oid, p, i);
 +              if (!match_sha(ds->len, ds->bin_pfx.hash, current->hash))
 +                      break;
 +              update_candidates(ds, current);
 +      }
 +}
 +
 +static void find_short_packed_object(struct disambiguate_state *ds)
 +{
 +      struct packed_git *p;
 +
 +      for (p = get_packed_git(the_repository); p && !ds->ambiguous;
 +           p = p->next)
 +              unique_in_pack(p, ds);
 +}
 +
 +#define SHORT_NAME_NOT_FOUND (-1)
 +#define SHORT_NAME_AMBIGUOUS (-2)
 +
 +static int finish_object_disambiguation(struct disambiguate_state *ds,
 +                                      struct object_id *oid)
 +{
 +      if (ds->ambiguous)
 +              return SHORT_NAME_AMBIGUOUS;
 +
 +      if (!ds->candidate_exists)
 +              return SHORT_NAME_NOT_FOUND;
 +
 +      if (!ds->candidate_checked)
 +              /*
 +               * If this is the only candidate, there is no point
 +               * calling the disambiguation hint callback.
 +               *
 +               * On the other hand, if the current candidate
 +               * replaced an earlier candidate that did _not_ pass
 +               * the disambiguation hint callback, then we do have
 +               * more than one objects that match the short name
 +               * given, so we should make sure this one matches;
 +               * otherwise, if we discovered this one and the one
 +               * that we previously discarded in the reverse order,
 +               * we would end up showing different results in the
 +               * same repository!
 +               */
 +              ds->candidate_ok = (!ds->disambiguate_fn_used ||
 +                                  ds->fn(&ds->candidate, ds->cb_data));
 +
 +      if (!ds->candidate_ok)
 +              return SHORT_NAME_AMBIGUOUS;
 +
 +      oidcpy(oid, &ds->candidate);
 +      return 0;
 +}
 +
 +static int disambiguate_commit_only(const struct object_id *oid, void *cb_data_unused)
 +{
 +      int kind = oid_object_info(the_repository, oid, NULL);
 +      return kind == OBJ_COMMIT;
 +}
 +
 +static int disambiguate_committish_only(const struct object_id *oid, void *cb_data_unused)
 +{
 +      struct object *obj;
 +      int kind;
 +
 +      kind = oid_object_info(the_repository, oid, NULL);
 +      if (kind == OBJ_COMMIT)
 +              return 1;
 +      if (kind != OBJ_TAG)
 +              return 0;
 +
 +      /* We need to do this the hard way... */
 +      obj = deref_tag(parse_object(oid), NULL, 0);
 +      if (obj && obj->type == OBJ_COMMIT)
 +              return 1;
 +      return 0;
 +}
 +
 +static int disambiguate_tree_only(const struct object_id *oid, void *cb_data_unused)
 +{
 +      int kind = oid_object_info(the_repository, oid, NULL);
 +      return kind == OBJ_TREE;
 +}
 +
 +static int disambiguate_treeish_only(const struct object_id *oid, void *cb_data_unused)
 +{
 +      struct object *obj;
 +      int kind;
 +
 +      kind = oid_object_info(the_repository, oid, NULL);
 +      if (kind == OBJ_TREE || kind == OBJ_COMMIT)
 +              return 1;
 +      if (kind != OBJ_TAG)
 +              return 0;
 +
 +      /* We need to do this the hard way... */
 +      obj = deref_tag(parse_object(oid), NULL, 0);
 +      if (obj && (obj->type == OBJ_TREE || obj->type == OBJ_COMMIT))
 +              return 1;
 +      return 0;
 +}
 +
 +static int disambiguate_blob_only(const struct object_id *oid, void *cb_data_unused)
 +{
 +      int kind = oid_object_info(the_repository, oid, NULL);
 +      return kind == OBJ_BLOB;
 +}
 +
 +static disambiguate_hint_fn default_disambiguate_hint;
 +
 +int set_disambiguate_hint_config(const char *var, const char *value)
 +{
 +      static const struct {
 +              const char *name;
 +              disambiguate_hint_fn fn;
 +      } hints[] = {
 +              { "none", NULL },
 +              { "commit", disambiguate_commit_only },
 +              { "committish", disambiguate_committish_only },
 +              { "tree", disambiguate_tree_only },
 +              { "treeish", disambiguate_treeish_only },
 +              { "blob", disambiguate_blob_only }
 +      };
 +      int i;
 +
 +      if (!value)
 +              return config_error_nonbool(var);
 +
 +      for (i = 0; i < ARRAY_SIZE(hints); i++) {
 +              if (!strcasecmp(value, hints[i].name)) {
 +                      default_disambiguate_hint = hints[i].fn;
 +                      return 0;
 +              }
 +      }
 +
 +      return error("unknown hint type for '%s': %s", var, value);
 +}
 +
 +static int init_object_disambiguation(const char *name, int len,
 +                                    struct disambiguate_state *ds)
 +{
 +      int i;
 +
 +      if (len < MINIMUM_ABBREV || len > GIT_SHA1_HEXSZ)
 +              return -1;
 +
 +      memset(ds, 0, sizeof(*ds));
 +
 +      for (i = 0; i < len ;i++) {
 +              unsigned char c = name[i];
 +              unsigned char val;
 +              if (c >= '0' && c <= '9')
 +                      val = c - '0';
 +              else if (c >= 'a' && c <= 'f')
 +                      val = c - 'a' + 10;
 +              else if (c >= 'A' && c <='F') {
 +                      val = c - 'A' + 10;
 +                      c -= 'A' - 'a';
 +              }
 +              else
 +                      return -1;
 +              ds->hex_pfx[i] = c;
 +              if (!(i & 1))
 +                      val <<= 4;
 +              ds->bin_pfx.hash[i >> 1] |= val;
 +      }
 +
 +      ds->len = len;
 +      ds->hex_pfx[len] = '\0';
 +      prepare_alt_odb(the_repository);
 +      return 0;
 +}
 +
 +static int show_ambiguous_object(const struct object_id *oid, void *data)
 +{
 +      const struct disambiguate_state *ds = data;
 +      struct strbuf desc = STRBUF_INIT;
 +      int type;
 +
 +      if (ds->fn && !ds->fn(oid, ds->cb_data))
 +              return 0;
 +
 +      type = oid_object_info(the_repository, oid, NULL);
 +      if (type == OBJ_COMMIT) {
 +              struct commit *commit = lookup_commit(oid);
 +              if (commit) {
 +                      struct pretty_print_context pp = {0};
 +                      pp.date_mode.type = DATE_SHORT;
 +                      format_commit_message(commit, " %ad - %s", &desc, &pp);
 +              }
 +      } else if (type == OBJ_TAG) {
 +              struct tag *tag = lookup_tag(oid);
 +              if (!parse_tag(tag) && tag->tag)
 +                      strbuf_addf(&desc, " %s", tag->tag);
 +      }
 +
 +      advise("  %s %s%s",
 +             find_unique_abbrev(oid, DEFAULT_ABBREV),
 +             type_name(type) ? type_name(type) : "unknown type",
 +             desc.buf);
 +
 +      strbuf_release(&desc);
 +      return 0;
 +}
 +
 +static int collect_ambiguous(const struct object_id *oid, void *data)
 +{
 +      oid_array_append(data, oid);
 +      return 0;
 +}
 +
 +static int sort_ambiguous(const void *a, const void *b)
 +{
 +      int a_type = oid_object_info(the_repository, a, NULL);
 +      int b_type = oid_object_info(the_repository, b, NULL);
 +      int a_type_sort;
 +      int b_type_sort;
 +
 +      /*
 +       * Sorts by hash within the same object type, just as
 +       * oid_array_for_each_unique() would do.
 +       */
 +      if (a_type == b_type)
 +              return oidcmp(a, b);
 +
 +      /*
 +       * Between object types show tags, then commits, and finally
 +       * trees and blobs.
 +       *
 +       * The object_type enum is commit, tree, blob, tag, but we
 +       * want tag, commit, tree blob. Cleverly (perhaps too
 +       * cleverly) do that with modulus, since the enum assigns 1 to
 +       * commit, so tag becomes 0.
 +       */
 +      a_type_sort = a_type % 4;
 +      b_type_sort = b_type % 4;
 +      return a_type_sort > b_type_sort ? 1 : -1;
 +}
 +
 +static int get_short_oid(const char *name, int len, struct object_id *oid,
 +                        unsigned flags)
 +{
 +      int status;
 +      struct disambiguate_state ds;
 +      int quietly = !!(flags & GET_OID_QUIETLY);
 +
 +      if (init_object_disambiguation(name, len, &ds) < 0)
 +              return -1;
 +
 +      if (HAS_MULTI_BITS(flags & GET_OID_DISAMBIGUATORS))
 +              BUG("multiple get_short_oid disambiguator flags");
 +
 +      if (flags & GET_OID_COMMIT)
 +              ds.fn = disambiguate_commit_only;
 +      else if (flags & GET_OID_COMMITTISH)
 +              ds.fn = disambiguate_committish_only;
 +      else if (flags & GET_OID_TREE)
 +              ds.fn = disambiguate_tree_only;
 +      else if (flags & GET_OID_TREEISH)
 +              ds.fn = disambiguate_treeish_only;
 +      else if (flags & GET_OID_BLOB)
 +              ds.fn = disambiguate_blob_only;
 +      else
 +              ds.fn = default_disambiguate_hint;
 +
 +      find_short_object_filename(&ds);
 +      find_short_packed_object(&ds);
 +      status = finish_object_disambiguation(&ds, oid);
 +
 +      if (!quietly && (status == SHORT_NAME_AMBIGUOUS)) {
 +              struct oid_array collect = OID_ARRAY_INIT;
 +
 +              error(_("short SHA1 %s is ambiguous"), ds.hex_pfx);
 +
 +              /*
 +               * We may still have ambiguity if we simply saw a series of
 +               * candidates that did not satisfy our hint function. In
 +               * that case, we still want to show them, so disable the hint
 +               * function entirely.
 +               */
 +              if (!ds.ambiguous)
 +                      ds.fn = NULL;
 +
 +              advise(_("The candidates are:"));
 +              for_each_abbrev(ds.hex_pfx, collect_ambiguous, &collect);
 +              QSORT(collect.oid, collect.nr, sort_ambiguous);
 +
 +              if (oid_array_for_each(&collect, show_ambiguous_object, &ds))
 +                      BUG("show_ambiguous_object shouldn't return non-zero");
 +              oid_array_clear(&collect);
 +      }
 +
 +      return status;
 +}
 +
 +int for_each_abbrev(const char *prefix, each_abbrev_fn fn, void *cb_data)
 +{
 +      struct oid_array collect = OID_ARRAY_INIT;
 +      struct disambiguate_state ds;
 +      int ret;
 +
 +      if (init_object_disambiguation(prefix, strlen(prefix), &ds) < 0)
 +              return -1;
 +
 +      ds.always_call_fn = 1;
 +      ds.fn = collect_ambiguous;
 +      ds.cb_data = &collect;
 +      find_short_object_filename(&ds);
 +      find_short_packed_object(&ds);
 +
 +      ret = oid_array_for_each_unique(&collect, fn, cb_data);
 +      oid_array_clear(&collect);
 +      return ret;
 +}
 +
 +/*
 + * Return the slot of the most-significant bit set in "val". There are various
 + * ways to do this quickly with fls() or __builtin_clzl(), but speed is
 + * probably not a big deal here.
 + */
 +static unsigned msb(unsigned long val)
 +{
 +      unsigned r = 0;
 +      while (val >>= 1)
 +              r++;
 +      return r;
 +}
 +
 +struct min_abbrev_data {
 +      unsigned int init_len;
 +      unsigned int cur_len;
 +      char *hex;
 +      const struct object_id *oid;
 +};
 +
 +static inline char get_hex_char_from_oid(const struct object_id *oid,
 +                                       unsigned int pos)
 +{
 +      static const char hex[] = "0123456789abcdef";
 +
 +      if ((pos & 1) == 0)
 +              return hex[oid->hash[pos >> 1] >> 4];
 +      else
 +              return hex[oid->hash[pos >> 1] & 0xf];
 +}
 +
 +static int extend_abbrev_len(const struct object_id *oid, void *cb_data)
 +{
 +      struct min_abbrev_data *mad = cb_data;
 +
 +      unsigned int i = mad->init_len;
 +      while (mad->hex[i] && mad->hex[i] == get_hex_char_from_oid(oid, i))
 +              i++;
 +
 +      if (i < GIT_MAX_RAWSZ && i >= mad->cur_len)
 +              mad->cur_len = i + 1;
 +
 +      return 0;
 +}
 +
 +static void find_abbrev_len_for_pack(struct packed_git *p,
 +                                   struct min_abbrev_data *mad)
 +{
 +      int match = 0;
 +      uint32_t num, first = 0;
 +      struct object_id oid;
 +      const struct object_id *mad_oid;
 +
 +      if (open_pack_index(p) || !p->num_objects)
 +              return;
 +
 +      num = p->num_objects;
 +      mad_oid = mad->oid;
 +      match = bsearch_pack(mad_oid, p, &first);
 +
 +      /*
 +       * first is now the position in the packfile where we would insert
 +       * mad->hash if it does not exist (or the position of mad->hash if
 +       * it does exist). Hence, we consider a maximum of two objects
 +       * nearby for the abbreviation length.
 +       */
 +      mad->init_len = 0;
 +      if (!match) {
 +              if (nth_packed_object_oid(&oid, p, first))
 +                      extend_abbrev_len(&oid, mad);
 +      } else if (first < num - 1) {
 +              if (nth_packed_object_oid(&oid, p, first + 1))
 +                      extend_abbrev_len(&oid, mad);
 +      }
 +      if (first > 0) {
 +              if (nth_packed_object_oid(&oid, p, first - 1))
 +                      extend_abbrev_len(&oid, mad);
 +      }
 +      mad->init_len = mad->cur_len;
 +}
 +
 +static void find_abbrev_len_packed(struct min_abbrev_data *mad)
 +{
 +      struct packed_git *p;
 +
 +      for (p = get_packed_git(the_repository); p; p = p->next)
 +              find_abbrev_len_for_pack(p, mad);
 +}
 +
 +int find_unique_abbrev_r(char *hex, const struct object_id *oid, int len)
 +{
 +      struct disambiguate_state ds;
 +      struct min_abbrev_data mad;
 +      struct object_id oid_ret;
 +      if (len < 0) {
 +              unsigned long count = approximate_object_count();
 +              /*
 +               * Add one because the MSB only tells us the highest bit set,
 +               * not including the value of all the _other_ bits (so "15"
 +               * is only one off of 2^4, but the MSB is the 3rd bit.
 +               */
 +              len = msb(count) + 1;
 +              /*
 +               * We now know we have on the order of 2^len objects, which
 +               * expects a collision at 2^(len/2). But we also care about hex
 +               * chars, not bits, and there are 4 bits per hex. So all
 +               * together we need to divide by 2 and round up.
 +               */
 +              len = DIV_ROUND_UP(len, 2);
 +              /*
 +               * For very small repos, we stick with our regular fallback.
 +               */
 +              if (len < FALLBACK_DEFAULT_ABBREV)
 +                      len = FALLBACK_DEFAULT_ABBREV;
 +      }
 +
 +      oid_to_hex_r(hex, oid);
 +      if (len == GIT_SHA1_HEXSZ || !len)
 +              return GIT_SHA1_HEXSZ;
 +
 +      mad.init_len = len;
 +      mad.cur_len = len;
 +      mad.hex = hex;
 +      mad.oid = oid;
 +
 +      find_abbrev_len_packed(&mad);
 +
 +      if (init_object_disambiguation(hex, mad.cur_len, &ds) < 0)
 +              return -1;
 +
 +      ds.fn = extend_abbrev_len;
 +      ds.always_call_fn = 1;
 +      ds.cb_data = (void *)&mad;
 +
 +      find_short_object_filename(&ds);
 +      (void)finish_object_disambiguation(&ds, &oid_ret);
 +
 +      hex[mad.cur_len] = 0;
 +      return mad.cur_len;
 +}
 +
 +const char *find_unique_abbrev(const struct object_id *oid, int len)
 +{
 +      static int bufno;
 +      static char hexbuffer[4][GIT_MAX_HEXSZ + 1];
 +      char *hex = hexbuffer[bufno];
 +      bufno = (bufno + 1) % ARRAY_SIZE(hexbuffer);
 +      find_unique_abbrev_r(hex, oid, len);
 +      return hex;
 +}
 +
 +static int ambiguous_path(const char *path, int len)
 +{
 +      int slash = 1;
 +      int cnt;
 +
 +      for (cnt = 0; cnt < len; cnt++) {
 +              switch (*path++) {
 +              case '\0':
 +                      break;
 +              case '/':
 +                      if (slash)
 +                              break;
 +                      slash = 1;
 +                      continue;
 +              case '.':
 +                      continue;
 +              default:
 +                      slash = 0;
 +                      continue;
 +              }
 +              break;
 +      }
 +      return slash;
 +}
 +
 +static inline int at_mark(const char *string, int len,
 +                        const char **suffix, int nr)
 +{
 +      int i;
 +
 +      for (i = 0; i < nr; i++) {
 +              int suffix_len = strlen(suffix[i]);
 +              if (suffix_len <= len
 +                  && !strncasecmp(string, suffix[i], suffix_len))
 +                      return suffix_len;
 +      }
 +      return 0;
 +}
 +
 +static inline int upstream_mark(const char *string, int len)
 +{
 +      const char *suffix[] = { "@{upstream}", "@{u}" };
 +      return at_mark(string, len, suffix, ARRAY_SIZE(suffix));
 +}
 +
 +static inline int push_mark(const char *string, int len)
 +{
 +      const char *suffix[] = { "@{push}" };
 +      return at_mark(string, len, suffix, ARRAY_SIZE(suffix));
 +}
 +
 +static int get_oid_1(const char *name, int len, struct object_id *oid, unsigned lookup_flags);
 +static int interpret_nth_prior_checkout(const char *name, int namelen, struct strbuf *buf);
 +
 +static int get_oid_basic(const char *str, int len, struct object_id *oid,
 +                        unsigned int flags)
 +{
 +      static const char *warn_msg = "refname '%.*s' is ambiguous.";
 +      static const char *object_name_msg = N_(
 +      "Git normally never creates a ref that ends with 40 hex characters\n"
 +      "because it will be ignored when you just specify 40-hex. These refs\n"
 +      "may be created by mistake. For example,\n"
 +      "\n"
 +      "  git checkout -b $br $(git rev-parse ...)\n"
 +      "\n"
 +      "where \"$br\" is somehow empty and a 40-hex ref is created. Please\n"
 +      "examine these refs and maybe delete them. Turn this message off by\n"
 +      "running \"git config advice.objectNameWarning false\"");
 +      struct object_id tmp_oid;
 +      char *real_ref = NULL;
 +      int refs_found = 0;
 +      int at, reflog_len, nth_prior = 0;
 +
 +      if (len == GIT_SHA1_HEXSZ && !get_oid_hex(str, oid)) {
 +              if (warn_ambiguous_refs && warn_on_object_refname_ambiguity) {
 +                      refs_found = dwim_ref(str, len, &tmp_oid, &real_ref);
 +                      if (refs_found > 0) {
 +                              warning(warn_msg, len, str);
 +                              if (advice_object_name_warning)
 +                                      fprintf(stderr, "%s\n", _(object_name_msg));
 +                      }
 +                      free(real_ref);
 +              }
 +              return 0;
 +      }
 +
 +      /* basic@{time or number or -number} format to query ref-log */
 +      reflog_len = at = 0;
 +      if (len && str[len-1] == '}') {
 +              for (at = len-4; at >= 0; at--) {
 +                      if (str[at] == '@' && str[at+1] == '{') {
 +                              if (str[at+2] == '-') {
 +                                      if (at != 0)
 +                                              /* @{-N} not at start */
 +                                              return -1;
 +                                      nth_prior = 1;
 +                                      continue;
 +                              }
 +                              if (!upstream_mark(str + at, len - at) &&
 +                                  !push_mark(str + at, len - at)) {
 +                                      reflog_len = (len-1) - (at+2);
 +                                      len = at;
 +                              }
 +                              break;
 +                      }
 +              }
 +      }
 +
 +      /* Accept only unambiguous ref paths. */
 +      if (len && ambiguous_path(str, len))
 +              return -1;
 +
 +      if (nth_prior) {
 +              struct strbuf buf = STRBUF_INIT;
 +              int detached;
 +
 +              if (interpret_nth_prior_checkout(str, len, &buf) > 0) {
 +                      detached = (buf.len == GIT_SHA1_HEXSZ && !get_oid_hex(buf.buf, oid));
 +                      strbuf_release(&buf);
 +                      if (detached)
 +                              return 0;
 +              }
 +      }
 +
 +      if (!len && reflog_len)
 +              /* allow "@{...}" to mean the current branch reflog */
 +              refs_found = dwim_ref("HEAD", 4, oid, &real_ref);
 +      else if (reflog_len)
 +              refs_found = dwim_log(str, len, oid, &real_ref);
 +      else
 +              refs_found = dwim_ref(str, len, oid, &real_ref);
 +
 +      if (!refs_found)
 +              return -1;
 +
 +      if (warn_ambiguous_refs && !(flags & GET_OID_QUIETLY) &&
 +          (refs_found > 1 ||
 +           !get_short_oid(str, len, &tmp_oid, GET_OID_QUIETLY)))
 +              warning(warn_msg, len, str);
 +
 +      if (reflog_len) {
 +              int nth, i;
 +              timestamp_t at_time;
 +              timestamp_t co_time;
 +              int co_tz, co_cnt;
 +
 +              /* Is it asking for N-th entry, or approxidate? */
 +              for (i = nth = 0; 0 <= nth && i < reflog_len; i++) {
 +                      char ch = str[at+2+i];
 +                      if ('0' <= ch && ch <= '9')
 +                              nth = nth * 10 + ch - '0';
 +                      else
 +                              nth = -1;
 +              }
 +              if (100000000 <= nth) {
 +                      at_time = nth;
 +                      nth = -1;
 +              } else if (0 <= nth)
 +                      at_time = 0;
 +              else {
 +                      int errors = 0;
 +                      char *tmp = xstrndup(str + at + 2, reflog_len);
 +                      at_time = approxidate_careful(tmp, &errors);
 +                      free(tmp);
 +                      if (errors) {
 +                              free(real_ref);
 +                              return -1;
 +                      }
 +              }
 +              if (read_ref_at(real_ref, flags, at_time, nth, oid, NULL,
 +                              &co_time, &co_tz, &co_cnt)) {
 +                      if (!len) {
 +                              if (starts_with(real_ref, "refs/heads/")) {
 +                                      str = real_ref + 11;
 +                                      len = strlen(real_ref + 11);
 +                              } else {
 +                                      /* detached HEAD */
 +                                      str = "HEAD";
 +                                      len = 4;
 +                              }
 +                      }
 +                      if (at_time) {
 +                              if (!(flags & GET_OID_QUIETLY)) {
 +                                      warning("Log for '%.*s' only goes "
 +                                              "back to %s.", len, str,
 +                                              show_date(co_time, co_tz, DATE_MODE(RFC2822)));
 +                              }
 +                      } else {
 +                              if (flags & GET_OID_QUIETLY) {
 +                                      exit(128);
 +                              }
 +                              die("Log for '%.*s' only has %d entries.",
 +                                  len, str, co_cnt);
 +                      }
 +              }
 +      }
 +
 +      free(real_ref);
 +      return 0;
 +}
 +
 +static int get_parent(const char *name, int len,
 +                    struct object_id *result, int idx)
 +{
 +      struct object_id oid;
 +      int ret = get_oid_1(name, len, &oid, GET_OID_COMMITTISH);
 +      struct commit *commit;
 +      struct commit_list *p;
 +
 +      if (ret)
 +              return ret;
 +      commit = lookup_commit_reference(&oid);
 +      if (parse_commit(commit))
 +              return -1;
 +      if (!idx) {
 +              oidcpy(result, &commit->object.oid);
 +              return 0;
 +      }
 +      p = commit->parents;
 +      while (p) {
 +              if (!--idx) {
 +                      oidcpy(result, &p->item->object.oid);
 +                      return 0;
 +              }
 +              p = p->next;
 +      }
 +      return -1;
 +}
 +
 +static int get_nth_ancestor(const char *name, int len,
 +                          struct object_id *result, int generation)
 +{
 +      struct object_id oid;
 +      struct commit *commit;
 +      int ret;
 +
 +      ret = get_oid_1(name, len, &oid, GET_OID_COMMITTISH);
 +      if (ret)
 +              return ret;
 +      commit = lookup_commit_reference(&oid);
 +      if (!commit)
 +              return -1;
 +
 +      while (generation--) {
 +              if (parse_commit(commit) || !commit->parents)
 +                      return -1;
 +              commit = commit->parents->item;
 +      }
 +      oidcpy(result, &commit->object.oid);
 +      return 0;
 +}
 +
 +struct object *peel_to_type(const char *name, int namelen,
 +                          struct object *o, enum object_type expected_type)
 +{
 +      if (name && !namelen)
 +              namelen = strlen(name);
 +      while (1) {
 +              if (!o || (!o->parsed && !parse_object(&o->oid)))
 +                      return NULL;
 +              if (expected_type == OBJ_ANY || o->type == expected_type)
 +                      return o;
 +              if (o->type == OBJ_TAG)
 +                      o = ((struct tag*) o)->tagged;
 +              else if (o->type == OBJ_COMMIT)
 +                      o = &(get_commit_tree(((struct commit *)o))->object);
 +              else {
 +                      if (name)
 +                              error("%.*s: expected %s type, but the object "
 +                                    "dereferences to %s type",
 +                                    namelen, name, type_name(expected_type),
 +                                    type_name(o->type));
 +                      return NULL;
 +              }
 +      }
 +}
 +
 +static int peel_onion(const char *name, int len, struct object_id *oid,
 +                    unsigned lookup_flags)
 +{
 +      struct object_id outer;
 +      const char *sp;
 +      unsigned int expected_type = 0;
 +      struct object *o;
 +
 +      /*
 +       * "ref^{type}" dereferences ref repeatedly until you cannot
 +       * dereference anymore, or you get an object of given type,
 +       * whichever comes first.  "ref^{}" means just dereference
 +       * tags until you get a non-tag.  "ref^0" is a shorthand for
 +       * "ref^{commit}".  "commit^{tree}" could be used to find the
 +       * top-level tree of the given commit.
 +       */
 +      if (len < 4 || name[len-1] != '}')
 +              return -1;
 +
 +      for (sp = name + len - 1; name <= sp; sp--) {
 +              int ch = *sp;
 +              if (ch == '{' && name < sp && sp[-1] == '^')
 +                      break;
 +      }
 +      if (sp <= name)
 +              return -1;
 +
 +      sp++; /* beginning of type name, or closing brace for empty */
 +      if (starts_with(sp, "commit}"))
 +              expected_type = OBJ_COMMIT;
 +      else if (starts_with(sp, "tag}"))
 +              expected_type = OBJ_TAG;
 +      else if (starts_with(sp, "tree}"))
 +              expected_type = OBJ_TREE;
 +      else if (starts_with(sp, "blob}"))
 +              expected_type = OBJ_BLOB;
 +      else if (starts_with(sp, "object}"))
 +              expected_type = OBJ_ANY;
 +      else if (sp[0] == '}')
 +              expected_type = OBJ_NONE;
 +      else if (sp[0] == '/')
 +              expected_type = OBJ_COMMIT;
 +      else
 +              return -1;
 +
 +      lookup_flags &= ~GET_OID_DISAMBIGUATORS;
 +      if (expected_type == OBJ_COMMIT)
 +              lookup_flags |= GET_OID_COMMITTISH;
 +      else if (expected_type == OBJ_TREE)
 +              lookup_flags |= GET_OID_TREEISH;
 +
 +      if (get_oid_1(name, sp - name - 2, &outer, lookup_flags))
 +              return -1;
 +
 +      o = parse_object(&outer);
 +      if (!o)
 +              return -1;
 +      if (!expected_type) {
 +              o = deref_tag(o, name, sp - name - 2);
 +              if (!o || (!o->parsed && !parse_object(&o->oid)))
 +                      return -1;
 +              oidcpy(oid, &o->oid);
 +              return 0;
 +      }
 +
 +      /*
 +       * At this point, the syntax look correct, so
 +       * if we do not get the needed object, we should
 +       * barf.
 +       */
 +      o = peel_to_type(name, len, o, expected_type);
 +      if (!o)
 +              return -1;
 +
 +      oidcpy(oid, &o->oid);
 +      if (sp[0] == '/') {
 +              /* "$commit^{/foo}" */
 +              char *prefix;
 +              int ret;
 +              struct commit_list *list = NULL;
 +
 +              /*
 +               * $commit^{/}. Some regex implementation may reject.
 +               * We don't need regex anyway. '' pattern always matches.
 +               */
 +              if (sp[1] == '}')
 +                      return 0;
 +
 +              prefix = xstrndup(sp + 1, name + len - 1 - (sp + 1));
 +              commit_list_insert((struct commit *)o, &list);
 +              ret = get_oid_oneline(prefix, oid, list);
 +              free(prefix);
 +              return ret;
 +      }
 +      return 0;
 +}
 +
 +static int get_describe_name(const char *name, int len, struct object_id *oid)
 +{
 +      const char *cp;
 +      unsigned flags = GET_OID_QUIETLY | GET_OID_COMMIT;
 +
 +      for (cp = name + len - 1; name + 2 <= cp; cp--) {
 +              char ch = *cp;
 +              if (!isxdigit(ch)) {
 +                      /* We must be looking at g in "SOMETHING-g"
 +                       * for it to be describe output.
 +                       */
 +                      if (ch == 'g' && cp[-1] == '-') {
 +                              cp++;
 +                              len -= cp - name;
 +                              return get_short_oid(cp, len, oid, flags);
 +                      }
 +              }
 +      }
 +      return -1;
 +}
 +
 +static int get_oid_1(const char *name, int len, struct object_id *oid, unsigned lookup_flags)
 +{
 +      int ret, has_suffix;
 +      const char *cp;
 +
 +      /*
 +       * "name~3" is "name^^^", "name~" is "name~1", and "name^" is "name^1".
 +       */
 +      has_suffix = 0;
 +      for (cp = name + len - 1; name <= cp; cp--) {
 +              int ch = *cp;
 +              if ('0' <= ch && ch <= '9')
 +                      continue;
 +              if (ch == '~' || ch == '^')
 +                      has_suffix = ch;
 +              break;
 +      }
 +
 +      if (has_suffix) {
 +              int num = 0;
 +              int len1 = cp - name;
 +              cp++;
 +              while (cp < name + len)
 +                      num = num * 10 + *cp++ - '0';
 +              if (!num && len1 == len - 1)
 +                      num = 1;
 +              if (has_suffix == '^')
 +                      return get_parent(name, len1, oid, num);
 +              /* else if (has_suffix == '~') -- goes without saying */
 +              return get_nth_ancestor(name, len1, oid, num);
 +      }
 +
 +      ret = peel_onion(name, len, oid, lookup_flags);
 +      if (!ret)
 +              return 0;
 +
 +      ret = get_oid_basic(name, len, oid, lookup_flags);
 +      if (!ret)
 +              return 0;
 +
 +      /* It could be describe output that is "SOMETHING-gXXXX" */
 +      ret = get_describe_name(name, len, oid);
 +      if (!ret)
 +              return 0;
 +
 +      return get_short_oid(name, len, oid, lookup_flags);
 +}
 +
 +/*
 + * This interprets names like ':/Initial revision of "git"' by searching
 + * through history and returning the first commit whose message starts
 + * the given regular expression.
 + *
 + * For negative-matching, prefix the pattern-part with '!-', like: ':/!-WIP'.
 + *
 + * For a literal '!' character at the beginning of a pattern, you have to repeat
 + * that, like: ':/!!foo'
 + *
 + * For future extension, all other sequences beginning with ':/!' are reserved.
 + */
 +
 +/* Remember to update object flag allocation in object.h */
 +#define ONELINE_SEEN (1u<<20)
 +
 +static int handle_one_ref(const char *path, const struct object_id *oid,
 +                        int flag, void *cb_data)
 +{
 +      struct commit_list **list = cb_data;
 +      struct object *object = parse_object(oid);
 +      if (!object)
 +              return 0;
 +      if (object->type == OBJ_TAG) {
 +              object = deref_tag(object, path, strlen(path));
 +              if (!object)
 +                      return 0;
 +      }
 +      if (object->type != OBJ_COMMIT)
 +              return 0;
 +      commit_list_insert((struct commit *)object, list);
 +      return 0;
 +}
 +
 +static int get_oid_oneline(const char *prefix, struct object_id *oid,
 +                          struct commit_list *list)
 +{
 +      struct commit_list *backup = NULL, *l;
 +      int found = 0;
 +      int negative = 0;
 +      regex_t regex;
 +
 +      if (prefix[0] == '!') {
 +              prefix++;
 +
 +              if (prefix[0] == '-') {
 +                      prefix++;
 +                      negative = 1;
 +              } else if (prefix[0] != '!') {
 +                      return -1;
 +              }
 +      }
 +
 +      if (regcomp(&regex, prefix, REG_EXTENDED))
 +              return -1;
 +
 +      for (l = list; l; l = l->next) {
 +              l->item->object.flags |= ONELINE_SEEN;
 +              commit_list_insert(l->item, &backup);
 +      }
 +      while (list) {
 +              const char *p, *buf;
 +              struct commit *commit;
 +              int matches;
 +
 +              commit = pop_most_recent_commit(&list, ONELINE_SEEN);
 +              if (!parse_object(&commit->object.oid))
 +                      continue;
 +              buf = get_commit_buffer(commit, NULL);
 +              p = strstr(buf, "\n\n");
 +              matches = negative ^ (p && !regexec(&regex, p + 2, 0, NULL, 0));
 +              unuse_commit_buffer(commit, buf);
 +
 +              if (matches) {
 +                      oidcpy(oid, &commit->object.oid);
 +                      found = 1;
 +                      break;
 +              }
 +      }
 +      regfree(&regex);
 +      free_commit_list(list);
 +      for (l = backup; l; l = l->next)
 +              clear_commit_marks(l->item, ONELINE_SEEN);
 +      free_commit_list(backup);
 +      return found ? 0 : -1;
 +}
 +
 +struct grab_nth_branch_switch_cbdata {
 +      int remaining;
 +      struct strbuf buf;
 +};
 +
 +static int grab_nth_branch_switch(struct object_id *ooid, struct object_id *noid,
 +                                const char *email, timestamp_t timestamp, int tz,
 +                                const char *message, void *cb_data)
 +{
 +      struct grab_nth_branch_switch_cbdata *cb = cb_data;
 +      const char *match = NULL, *target = NULL;
 +      size_t len;
 +
 +      if (skip_prefix(message, "checkout: moving from ", &match))
 +              target = strstr(match, " to ");
 +
 +      if (!match || !target)
 +              return 0;
 +      if (--(cb->remaining) == 0) {
 +              len = target - match;
 +              strbuf_reset(&cb->buf);
 +              strbuf_add(&cb->buf, match, len);
 +              return 1; /* we are done */
 +      }
 +      return 0;
 +}
 +
 +/*
 + * Parse @{-N} syntax, return the number of characters parsed
 + * if successful; otherwise signal an error with negative value.
 + */
 +static int interpret_nth_prior_checkout(const char *name, int namelen,
 +                                      struct strbuf *buf)
 +{
 +      long nth;
 +      int retval;
 +      struct grab_nth_branch_switch_cbdata cb;
 +      const char *brace;
 +      char *num_end;
 +
 +      if (namelen < 4)
 +              return -1;
 +      if (name[0] != '@' || name[1] != '{' || name[2] != '-')
 +              return -1;
 +      brace = memchr(name, '}', namelen);
 +      if (!brace)
 +              return -1;
 +      nth = strtol(name + 3, &num_end, 10);
 +      if (num_end != brace)
 +              return -1;
 +      if (nth <= 0)
 +              return -1;
 +      cb.remaining = nth;
 +      strbuf_init(&cb.buf, 20);
 +
 +      retval = 0;
 +      if (0 < for_each_reflog_ent_reverse("HEAD", grab_nth_branch_switch, &cb)) {
 +              strbuf_reset(buf);
 +              strbuf_addbuf(buf, &cb.buf);
 +              retval = brace - name + 1;
 +      }
 +
 +      strbuf_release(&cb.buf);
 +      return retval;
 +}
 +
 +int get_oid_mb(const char *name, struct object_id *oid)
 +{
 +      struct commit *one, *two;
 +      struct commit_list *mbs;
 +      struct object_id oid_tmp;
 +      const char *dots;
 +      int st;
 +
 +      dots = strstr(name, "...");
 +      if (!dots)
 +              return get_oid(name, oid);
 +      if (dots == name)
 +              st = get_oid("HEAD", &oid_tmp);
 +      else {
 +              struct strbuf sb;
 +              strbuf_init(&sb, dots - name);
 +              strbuf_add(&sb, name, dots - name);
 +              st = get_oid_committish(sb.buf, &oid_tmp);
 +              strbuf_release(&sb);
 +      }
 +      if (st)
 +              return st;
 +      one = lookup_commit_reference_gently(&oid_tmp, 0);
 +      if (!one)
 +              return -1;
 +
 +      if (get_oid_committish(dots[3] ? (dots + 3) : "HEAD", &oid_tmp))
 +              return -1;
 +      two = lookup_commit_reference_gently(&oid_tmp, 0);
 +      if (!two)
 +              return -1;
 +      mbs = get_merge_bases(one, two);
 +      if (!mbs || mbs->next)
 +              st = -1;
 +      else {
 +              st = 0;
 +              oidcpy(oid, &mbs->item->object.oid);
 +      }
 +      free_commit_list(mbs);
 +      return st;
 +}
 +
 +/* parse @something syntax, when 'something' is not {.*} */
 +static int interpret_empty_at(const char *name, int namelen, int len, struct strbuf *buf)
 +{
 +      const char *next;
 +
 +      if (len || name[1] == '{')
 +              return -1;
 +
 +      /* make sure it's a single @, or @@{.*}, not @foo */
 +      next = memchr(name + len + 1, '@', namelen - len - 1);
 +      if (next && next[1] != '{')
 +              return -1;
 +      if (!next)
 +              next = name + namelen;
 +      if (next != name + 1)
 +              return -1;
 +
 +      strbuf_reset(buf);
 +      strbuf_add(buf, "HEAD", 4);
 +      return 1;
 +}
 +
 +static int reinterpret(const char *name, int namelen, int len,
 +                     struct strbuf *buf, unsigned allowed)
 +{
 +      /* we have extra data, which might need further processing */
 +      struct strbuf tmp = STRBUF_INIT;
 +      int used = buf->len;
 +      int ret;
 +
 +      strbuf_add(buf, name + len, namelen - len);
 +      ret = interpret_branch_name(buf->buf, buf->len, &tmp, allowed);
 +      /* that data was not interpreted, remove our cruft */
 +      if (ret < 0) {
 +              strbuf_setlen(buf, used);
 +              return len;
 +      }
 +      strbuf_reset(buf);
 +      strbuf_addbuf(buf, &tmp);
 +      strbuf_release(&tmp);
 +      /* tweak for size of {-N} versus expanded ref name */
 +      return ret - used + len;
 +}
 +
 +static void set_shortened_ref(struct strbuf *buf, const char *ref)
 +{
 +      char *s = shorten_unambiguous_ref(ref, 0);
 +      strbuf_reset(buf);
 +      strbuf_addstr(buf, s);
 +      free(s);
 +}
 +
 +static int branch_interpret_allowed(const char *refname, unsigned allowed)
 +{
 +      if (!allowed)
 +              return 1;
 +
 +      if ((allowed & INTERPRET_BRANCH_LOCAL) &&
 +          starts_with(refname, "refs/heads/"))
 +              return 1;
 +      if ((allowed & INTERPRET_BRANCH_REMOTE) &&
 +          starts_with(refname, "refs/remotes/"))
 +              return 1;
 +
 +      return 0;
 +}
 +
 +static int interpret_branch_mark(const char *name, int namelen,
 +                               int at, struct strbuf *buf,
 +                               int (*get_mark)(const char *, int),
 +                               const char *(*get_data)(struct branch *,
 +                                                       struct strbuf *),
 +                               unsigned allowed)
 +{
 +      int len;
 +      struct branch *branch;
 +      struct strbuf err = STRBUF_INIT;
 +      const char *value;
 +
 +      len = get_mark(name + at, namelen - at);
 +      if (!len)
 +              return -1;
 +
 +      if (memchr(name, ':', at))
 +              return -1;
 +
 +      if (at) {
 +              char *name_str = xmemdupz(name, at);
 +              branch = branch_get(name_str);
 +              free(name_str);
 +      } else
 +              branch = branch_get(NULL);
 +
 +      value = get_data(branch, &err);
 +      if (!value)
 +              die("%s", err.buf);
 +
 +      if (!branch_interpret_allowed(value, allowed))
 +              return -1;
 +
 +      set_shortened_ref(buf, value);
 +      return len + at;
 +}
 +
 +int interpret_branch_name(const char *name, int namelen, struct strbuf *buf,
 +                        unsigned allowed)
 +{
 +      char *at;
 +      const char *start;
 +      int len;
 +
 +      if (!namelen)
 +              namelen = strlen(name);
 +
 +      if (!allowed || (allowed & INTERPRET_BRANCH_LOCAL)) {
 +              len = interpret_nth_prior_checkout(name, namelen, buf);
 +              if (!len) {
 +                      return len; /* syntax Ok, not enough switches */
 +              } else if (len > 0) {
 +                      if (len == namelen)
 +                              return len; /* consumed all */
 +                      else
 +                              return reinterpret(name, namelen, len, buf, allowed);
 +              }
 +      }
 +
 +      for (start = name;
 +           (at = memchr(start, '@', namelen - (start - name)));
 +           start = at + 1) {
 +
 +              if (!allowed || (allowed & INTERPRET_BRANCH_HEAD)) {
 +                      len = interpret_empty_at(name, namelen, at - name, buf);
 +                      if (len > 0)
 +                              return reinterpret(name, namelen, len, buf,
 +                                                 allowed);
 +              }
 +
 +              len = interpret_branch_mark(name, namelen, at - name, buf,
 +                                          upstream_mark, branch_get_upstream,
 +                                          allowed);
 +              if (len > 0)
 +                      return len;
 +
 +              len = interpret_branch_mark(name, namelen, at - name, buf,
 +                                          push_mark, branch_get_push,
 +                                          allowed);
 +              if (len > 0)
 +                      return len;
 +      }
 +
 +      return -1;
 +}
 +
 +void strbuf_branchname(struct strbuf *sb, const char *name, unsigned allowed)
 +{
 +      int len = strlen(name);
 +      int used = interpret_branch_name(name, len, sb, allowed);
 +
 +      if (used < 0)
 +              used = 0;
 +      strbuf_add(sb, name + used, len - used);
 +}
 +
 +int strbuf_check_branch_ref(struct strbuf *sb, const char *name)
 +{
 +      if (startup_info->have_repository)
 +              strbuf_branchname(sb, name, INTERPRET_BRANCH_LOCAL);
 +      else
 +              strbuf_addstr(sb, name);
 +
 +      /*
 +       * This splice must be done even if we end up rejecting the
 +       * name; builtin/branch.c::copy_or_rename_branch() still wants
 +       * to see what the name expanded to so that "branch -m" can be
 +       * used as a tool to correct earlier mistakes.
 +       */
 +      strbuf_splice(sb, 0, 0, "refs/heads/", 11);
 +
 +      if (*name == '-' ||
 +          !strcmp(sb->buf, "refs/heads/HEAD"))
 +              return -1;
 +
 +      return check_refname_format(sb->buf, 0);
 +}
 +
 +/*
 + * This is like "get_oid_basic()", except it allows "object ID expressions",
 + * notably "xyz^" for "parent of xyz"
 + */
 +int get_oid(const char *name, struct object_id *oid)
 +{
 +      struct object_context unused;
 +      return get_oid_with_context(name, 0, oid, &unused);
 +}
 +
 +
 +/*
 + * Many callers know that the user meant to name a commit-ish by
 + * syntactical positions where the object name appears.  Calling this
 + * function allows the machinery to disambiguate shorter-than-unique
 + * abbreviated object names between commit-ish and others.
 + *
 + * Note that this does NOT error out when the named object is not a
 + * commit-ish. It is merely to give a hint to the disambiguation
 + * machinery.
 + */
 +int get_oid_committish(const char *name, struct object_id *oid)
 +{
 +      struct object_context unused;
 +      return get_oid_with_context(name, GET_OID_COMMITTISH,
 +                                  oid, &unused);
 +}
 +
 +int get_oid_treeish(const char *name, struct object_id *oid)
 +{
 +      struct object_context unused;
 +      return get_oid_with_context(name, GET_OID_TREEISH,
 +                                  oid, &unused);
 +}
 +
 +int get_oid_commit(const char *name, struct object_id *oid)
 +{
 +      struct object_context unused;
 +      return get_oid_with_context(name, GET_OID_COMMIT,
 +                                  oid, &unused);
 +}
 +
 +int get_oid_tree(const char *name, struct object_id *oid)
 +{
 +      struct object_context unused;
 +      return get_oid_with_context(name, GET_OID_TREE,
 +                                  oid, &unused);
 +}
 +
 +int get_oid_blob(const char *name, struct object_id *oid)
 +{
 +      struct object_context unused;
 +      return get_oid_with_context(name, GET_OID_BLOB,
 +                                  oid, &unused);
 +}
 +
 +/* Must be called only when object_name:filename doesn't exist. */
 +static void diagnose_invalid_oid_path(const char *prefix,
 +                                    const char *filename,
 +                                    const struct object_id *tree_oid,
 +                                    const char *object_name,
 +                                    int object_name_len)
 +{
 +      struct object_id oid;
 +      unsigned mode;
 +
 +      if (!prefix)
 +              prefix = "";
 +
 +      if (file_exists(filename))
 +              die("Path '%s' exists on disk, but not in '%.*s'.",
 +                  filename, object_name_len, object_name);
 +      if (is_missing_file_error(errno)) {
 +              char *fullname = xstrfmt("%s%s", prefix, filename);
 +
 +              if (!get_tree_entry(tree_oid, fullname, &oid, &mode)) {
 +                      die("Path '%s' exists, but not '%s'.\n"
 +                          "Did you mean '%.*s:%s' aka '%.*s:./%s'?",
 +                          fullname,
 +                          filename,
 +                          object_name_len, object_name,
 +                          fullname,
 +                          object_name_len, object_name,
 +                          filename);
 +              }
 +              die("Path '%s' does not exist in '%.*s'",
 +                  filename, object_name_len, object_name);
 +      }
 +}
 +
 +/* Must be called only when :stage:filename doesn't exist. */
 +static void diagnose_invalid_index_path(int stage,
 +                                      const char *prefix,
 +                                      const char *filename)
 +{
 +      const struct cache_entry *ce;
 +      int pos;
 +      unsigned namelen = strlen(filename);
 +      struct strbuf fullname = STRBUF_INIT;
 +
 +      if (!prefix)
 +              prefix = "";
 +
 +      /* Wrong stage number? */
 +      pos = cache_name_pos(filename, namelen);
 +      if (pos < 0)
 +              pos = -pos - 1;
 +      if (pos < active_nr) {
 +              ce = active_cache[pos];
 +              if (ce_namelen(ce) == namelen &&
 +                  !memcmp(ce->name, filename, namelen))
 +                      die("Path '%s' is in the index, but not at stage %d.\n"
 +                          "Did you mean ':%d:%s'?",
 +                          filename, stage,
 +                          ce_stage(ce), filename);
 +      }
 +
 +      /* Confusion between relative and absolute filenames? */
 +      strbuf_addstr(&fullname, prefix);
 +      strbuf_addstr(&fullname, filename);
 +      pos = cache_name_pos(fullname.buf, fullname.len);
 +      if (pos < 0)
 +              pos = -pos - 1;
 +      if (pos < active_nr) {
 +              ce = active_cache[pos];
 +              if (ce_namelen(ce) == fullname.len &&
 +                  !memcmp(ce->name, fullname.buf, fullname.len))
 +                      die("Path '%s' is in the index, but not '%s'.\n"
 +                          "Did you mean ':%d:%s' aka ':%d:./%s'?",
 +                          fullname.buf, filename,
 +                          ce_stage(ce), fullname.buf,
 +                          ce_stage(ce), filename);
 +      }
 +
 +      if (file_exists(filename))
 +              die("Path '%s' exists on disk, but not in the index.", filename);
 +      if (is_missing_file_error(errno))
 +              die("Path '%s' does not exist (neither on disk nor in the index).",
 +                  filename);
 +
 +      strbuf_release(&fullname);
 +}
 +
 +
 +static char *resolve_relative_path(const char *rel)
 +{
 +      if (!starts_with(rel, "./") && !starts_with(rel, "../"))
 +              return NULL;
 +
 +      if (!is_inside_work_tree())
 +              die("relative path syntax can't be used outside working tree.");
 +
 +      /* die() inside prefix_path() if resolved path is outside worktree */
 +      return prefix_path(startup_info->prefix,
 +                         startup_info->prefix ? strlen(startup_info->prefix) : 0,
 +                         rel);
 +}
 +
 +static int get_oid_with_context_1(const char *name,
 +                                unsigned flags,
 +                                const char *prefix,
 +                                struct object_id *oid,
 +                                struct object_context *oc)
 +{
 +      int ret, bracket_depth;
 +      int namelen = strlen(name);
 +      const char *cp;
 +      int only_to_die = flags & GET_OID_ONLY_TO_DIE;
 +
 +      if (only_to_die)
 +              flags |= GET_OID_QUIETLY;
 +
 +      memset(oc, 0, sizeof(*oc));
 +      oc->mode = S_IFINVALID;
 +      strbuf_init(&oc->symlink_path, 0);
 +      ret = get_oid_1(name, namelen, oid, flags);
 +      if (!ret)
 +              return ret;
 +      /*
 +       * sha1:path --> object name of path in ent sha1
 +       * :path -> object name of absolute path in index
 +       * :./path -> object name of path relative to cwd in index
 +       * :[0-3]:path -> object name of path in index at stage
 +       * :/foo -> recent commit matching foo
 +       */
 +      if (name[0] == ':') {
 +              int stage = 0;
 +              const struct cache_entry *ce;
 +              char *new_path = NULL;
 +              int pos;
 +              if (!only_to_die && namelen > 2 && name[1] == '/') {
 +                      struct commit_list *list = NULL;
 +
 +                      for_each_ref(handle_one_ref, &list);
++                      head_ref(handle_one_ref, &list);
 +                      commit_list_sort_by_date(&list);
 +                      return get_oid_oneline(name + 2, oid, list);
 +              }
 +              if (namelen < 3 ||
 +                  name[2] != ':' ||
 +                  name[1] < '0' || '3' < name[1])
 +                      cp = name + 1;
 +              else {
 +                      stage = name[1] - '0';
 +                      cp = name + 3;
 +              }
 +              new_path = resolve_relative_path(cp);
 +              if (!new_path) {
 +                      namelen = namelen - (cp - name);
 +              } else {
 +                      cp = new_path;
 +                      namelen = strlen(cp);
 +              }
 +
 +              if (flags & GET_OID_RECORD_PATH)
 +                      oc->path = xstrdup(cp);
 +
 +              if (!active_cache)
 +                      read_cache();
 +              pos = cache_name_pos(cp, namelen);
 +              if (pos < 0)
 +                      pos = -pos - 1;
 +              while (pos < active_nr) {
 +                      ce = active_cache[pos];
 +                      if (ce_namelen(ce) != namelen ||
 +                          memcmp(ce->name, cp, namelen))
 +                              break;
 +                      if (ce_stage(ce) == stage) {
 +                              oidcpy(oid, &ce->oid);
 +                              oc->mode = ce->ce_mode;
 +                              free(new_path);
 +                              return 0;
 +                      }
 +                      pos++;
 +              }
 +              if (only_to_die && name[1] && name[1] != '/')
 +                      diagnose_invalid_index_path(stage, prefix, cp);
 +              free(new_path);
 +              return -1;
 +      }
 +      for (cp = name, bracket_depth = 0; *cp; cp++) {
 +              if (*cp == '{')
 +                      bracket_depth++;
 +              else if (bracket_depth && *cp == '}')
 +                      bracket_depth--;
 +              else if (!bracket_depth && *cp == ':')
 +                      break;
 +      }
 +      if (*cp == ':') {
 +              struct object_id tree_oid;
 +              int len = cp - name;
 +              unsigned sub_flags = flags;
 +
 +              sub_flags &= ~GET_OID_DISAMBIGUATORS;
 +              sub_flags |= GET_OID_TREEISH;
 +
 +              if (!get_oid_1(name, len, &tree_oid, sub_flags)) {
 +                      const char *filename = cp+1;
 +                      char *new_filename = NULL;
 +
 +                      new_filename = resolve_relative_path(filename);
 +                      if (new_filename)
 +                              filename = new_filename;
 +                      if (flags & GET_OID_FOLLOW_SYMLINKS) {
 +                              ret = get_tree_entry_follow_symlinks(&tree_oid,
 +                                      filename, oid, &oc->symlink_path,
 +                                      &oc->mode);
 +                      } else {
 +                              ret = get_tree_entry(&tree_oid, filename, oid,
 +                                                   &oc->mode);
 +                              if (ret && only_to_die) {
 +                                      diagnose_invalid_oid_path(prefix,
 +                                                                 filename,
 +                                                                 &tree_oid,
 +                                                                 name, len);
 +                              }
 +                      }
 +                      if (flags & GET_OID_RECORD_PATH)
 +                              oc->path = xstrdup(filename);
 +
 +                      free(new_filename);
 +                      return ret;
 +              } else {
 +                      if (only_to_die)
 +                              die("Invalid object name '%.*s'.", len, name);
 +              }
 +      }
 +      return ret;
 +}
 +
 +/*
 + * Call this function when you know "name" given by the end user must
 + * name an object but it doesn't; the function _may_ die with a better
 + * diagnostic message than "no such object 'name'", e.g. "Path 'doc' does not
 + * exist in 'HEAD'" when given "HEAD:doc", or it may return in which case
 + * you have a chance to diagnose the error further.
 + */
 +void maybe_die_on_misspelt_object_name(const char *name, const char *prefix)
 +{
 +      struct object_context oc;
 +      struct object_id oid;
 +      get_oid_with_context_1(name, GET_OID_ONLY_TO_DIE, prefix, &oid, &oc);
 +}
 +
 +int get_oid_with_context(const char *str, unsigned flags, struct object_id *oid, struct object_context *oc)
 +{
 +      if (flags & GET_OID_FOLLOW_SYMLINKS && flags & GET_OID_ONLY_TO_DIE)
 +              BUG("incompatible flags for get_sha1_with_context");
 +      return get_oid_with_context_1(str, flags, NULL, oid, oc);
 +}
Simple merge