Merge branch 'jc/rw-prefix'
authorJunio C Hamano <junkio@cox.net>
Sun, 18 Jun 2006 00:56:52 +0000 (17:56 -0700)
committerJunio C Hamano <junkio@cox.net>
Sun, 18 Jun 2006 00:56:52 +0000 (17:56 -0700)
* jc/rw-prefix:
  read-tree: reorganize bind_merge code.
  write-tree: --prefix=<path>
  read-tree: --prefix=<path>/ option.

1  2 
Documentation/git-read-tree.txt
builtin-read-tree.c
cache-tree.c
write-tree.c

Simple merge
index bb50fbd,0000000..04506da
mode 100644,000000..100644
--- /dev/null
@@@ -1,987 -1,0 +1,1045 @@@
-               if (unpack_trees_rec(posns, len, "", fn, &indpos))
 +/*
 + * GIT - The information manager from hell
 + *
 + * Copyright (C) Linus Torvalds, 2005
 + */
 +#define DBRT_DEBUG 1
 +
 +#include "cache.h"
 +
 +#include "object.h"
 +#include "tree.h"
 +#include "tree-walk.h"
 +#include "cache-tree.h"
 +#include <sys/time.h>
 +#include <signal.h>
 +#include "builtin.h"
 +
 +static int reset = 0;
 +static int merge = 0;
 +static int update = 0;
 +static int index_only = 0;
 +static int nontrivial_merge = 0;
 +static int trivial_merges_only = 0;
 +static int aggressive = 0;
 +static int verbose_update = 0;
 +static volatile int progress_update = 0;
++static const char *prefix = NULL;
 +
 +static int head_idx = -1;
 +static int merge_size = 0;
 +
 +static struct object_list *trees = NULL;
 +
 +static struct cache_entry df_conflict_entry = {
 +};
 +
 +struct tree_entry_list {
 +      struct tree_entry_list *next;
 +      unsigned directory : 1;
 +      unsigned executable : 1;
 +      unsigned symlink : 1;
 +      unsigned int mode;
 +      const char *name;
 +      const unsigned char *sha1;
 +};
 +
 +static struct tree_entry_list df_conflict_list = {
 +      .name = NULL,
 +      .next = &df_conflict_list
 +};
 +
 +typedef int (*merge_fn_t)(struct cache_entry **src);
 +
 +static struct tree_entry_list *create_tree_entry_list(struct tree *tree)
 +{
 +      struct tree_desc desc;
 +      struct name_entry one;
 +      struct tree_entry_list *ret = NULL;
 +      struct tree_entry_list **list_p = &ret;
 +
 +      desc.buf = tree->buffer;
 +      desc.size = tree->size;
 +
 +      while (tree_entry(&desc, &one)) {
 +              struct tree_entry_list *entry;
 +
 +              entry = xmalloc(sizeof(struct tree_entry_list));
 +              entry->name = one.path;
 +              entry->sha1 = one.sha1;
 +              entry->mode = one.mode;
 +              entry->directory = S_ISDIR(one.mode) != 0;
 +              entry->executable = (one.mode & S_IXUSR) != 0;
 +              entry->symlink = S_ISLNK(one.mode) != 0;
 +              entry->next = NULL;
 +
 +              *list_p = entry;
 +              list_p = &entry->next;
 +      }
 +      return ret;
 +}
 +
 +static int entcmp(const char *name1, int dir1, const char *name2, int dir2)
 +{
 +      int len1 = strlen(name1);
 +      int len2 = strlen(name2);
 +      int len = len1 < len2 ? len1 : len2;
 +      int ret = memcmp(name1, name2, len);
 +      unsigned char c1, c2;
 +      if (ret)
 +              return ret;
 +      c1 = name1[len];
 +      c2 = name2[len];
 +      if (!c1 && dir1)
 +              c1 = '/';
 +      if (!c2 && dir2)
 +              c2 = '/';
 +      ret = (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 +      if (c1 && c2 && !ret)
 +              ret = len1 - len2;
 +      return ret;
 +}
 +
 +static int unpack_trees_rec(struct tree_entry_list **posns, int len,
 +                          const char *base, merge_fn_t fn, int *indpos)
 +{
 +      int baselen = strlen(base);
 +      int src_size = len + 1;
 +      do {
 +              int i;
 +              const char *first;
 +              int firstdir = 0;
 +              int pathlen;
 +              unsigned ce_size;
 +              struct tree_entry_list **subposns;
 +              struct cache_entry **src;
 +              int any_files = 0;
 +              int any_dirs = 0;
 +              char *cache_name;
 +              int ce_stage;
 +
 +              /* Find the first name in the input. */
 +
 +              first = NULL;
 +              cache_name = NULL;
 +
 +              /* Check the cache */
 +              if (merge && *indpos < active_nr) {
 +                      /* This is a bit tricky: */
 +                      /* If the index has a subdirectory (with
 +                       * contents) as the first name, it'll get a
 +                       * filename like "foo/bar". But that's after
 +                       * "foo", so the entry in trees will get
 +                       * handled first, at which point we'll go into
 +                       * "foo", and deal with "bar" from the index,
 +                       * because the base will be "foo/". The only
 +                       * way we can actually have "foo/bar" first of
 +                       * all the things is if the trees don't
 +                       * contain "foo" at all, in which case we'll
 +                       * handle "foo/bar" without going into the
 +                       * directory, but that's fine (and will return
 +                       * an error anyway, with the added unknown
 +                       * file case.
 +                       */
 +
 +                      cache_name = active_cache[*indpos]->name;
 +                      if (strlen(cache_name) > baselen &&
 +                          !memcmp(cache_name, base, baselen)) {
 +                              cache_name += baselen;
 +                              first = cache_name;
 +                      } else {
 +                              cache_name = NULL;
 +                      }
 +              }
 +
 +#if DBRT_DEBUG > 1
 +              if (first)
 +                      printf("index %s\n", first);
 +#endif
 +              for (i = 0; i < len; i++) {
 +                      if (!posns[i] || posns[i] == &df_conflict_list)
 +                              continue;
 +#if DBRT_DEBUG > 1
 +                      printf("%d %s\n", i + 1, posns[i]->name);
 +#endif
 +                      if (!first || entcmp(first, firstdir,
 +                                           posns[i]->name, 
 +                                           posns[i]->directory) > 0) {
 +                              first = posns[i]->name;
 +                              firstdir = posns[i]->directory;
 +                      }
 +              }
 +              /* No name means we're done */
 +              if (!first)
 +                      return 0;
 +
 +              pathlen = strlen(first);
 +              ce_size = cache_entry_size(baselen + pathlen);
 +
 +              src = xcalloc(src_size, sizeof(struct cache_entry *));
 +
 +              subposns = xcalloc(len, sizeof(struct tree_list_entry *));
 +
 +              if (cache_name && !strcmp(cache_name, first)) {
 +                      any_files = 1;
 +                      src[0] = active_cache[*indpos];
 +                      remove_cache_entry_at(*indpos);
 +              }
 +
 +              for (i = 0; i < len; i++) {
 +                      struct cache_entry *ce;
 +
 +                      if (!posns[i] ||
 +                          (posns[i] != &df_conflict_list &&
 +                           strcmp(first, posns[i]->name))) {
 +                              continue;
 +                      }
 +
 +                      if (posns[i] == &df_conflict_list) {
 +                              src[i + merge] = &df_conflict_entry;
 +                              continue;
 +                      }
 +
 +                      if (posns[i]->directory) {
 +                              struct tree *tree = lookup_tree(posns[i]->sha1);
 +                              any_dirs = 1;
 +                              parse_tree(tree);
 +                              subposns[i] = create_tree_entry_list(tree);
 +                              posns[i] = posns[i]->next;
 +                              src[i + merge] = &df_conflict_entry;
 +                              continue;
 +                      }
 +
 +                      if (!merge)
 +                              ce_stage = 0;
 +                      else if (i + 1 < head_idx)
 +                              ce_stage = 1;
 +                      else if (i + 1 > head_idx)
 +                              ce_stage = 3;
 +                      else
 +                              ce_stage = 2;
 +
 +                      ce = xcalloc(1, ce_size);
 +                      ce->ce_mode = create_ce_mode(posns[i]->mode);
 +                      ce->ce_flags = create_ce_flags(baselen + pathlen,
 +                                                     ce_stage);
 +                      memcpy(ce->name, base, baselen);
 +                      memcpy(ce->name + baselen, first, pathlen + 1);
 +
 +                      any_files = 1;
 +
 +                      memcpy(ce->sha1, posns[i]->sha1, 20);
 +                      src[i + merge] = ce;
 +                      subposns[i] = &df_conflict_list;
 +                      posns[i] = posns[i]->next;
 +              }
 +              if (any_files) {
 +                      if (merge) {
 +                              int ret;
 +
 +#if DBRT_DEBUG > 1
 +                              printf("%s:\n", first);
 +                              for (i = 0; i < src_size; i++) {
 +                                      printf(" %d ", i);
 +                                      if (src[i])
 +                                              printf("%s\n", sha1_to_hex(src[i]->sha1));
 +                                      else
 +                                              printf("\n");
 +                              }
 +#endif
 +                              ret = fn(src);
 +                              
 +#if DBRT_DEBUG > 1
 +                              printf("Added %d entries\n", ret);
 +#endif
 +                              *indpos += ret;
 +                      } else {
 +                              for (i = 0; i < src_size; i++) {
 +                                      if (src[i]) {
 +                                              add_cache_entry(src[i], ADD_CACHE_OK_TO_ADD|ADD_CACHE_SKIP_DFCHECK);
 +                                      }
 +                              }
 +                      }
 +              }
 +              if (any_dirs) {
 +                      char *newbase = xmalloc(baselen + 2 + pathlen);
 +                      memcpy(newbase, base, baselen);
 +                      memcpy(newbase + baselen, first, pathlen);
 +                      newbase[baselen + pathlen] = '/';
 +                      newbase[baselen + pathlen + 1] = '\0';
 +                      if (unpack_trees_rec(subposns, len, newbase, fn,
 +                                           indpos))
 +                              return -1;
 +                      free(newbase);
 +              }
 +              free(subposns);
 +              free(src);
 +      } while (1);
 +}
 +
 +static void reject_merge(struct cache_entry *ce)
 +{
 +      die("Entry '%s' would be overwritten by merge. Cannot merge.", 
 +          ce->name);
 +}
 +
 +/* Unlink the last component and attempt to remove leading
 + * directories, in case this unlink is the removal of the
 + * last entry in the directory -- empty directories are removed.
 + */
 +static void unlink_entry(char *name)
 +{
 +      char *cp, *prev;
 +
 +      if (unlink(name))
 +              return;
 +      prev = NULL;
 +      while (1) {
 +              int status;
 +              cp = strrchr(name, '/');
 +              if (prev)
 +                      *prev = '/';
 +              if (!cp)
 +                      break;
 +
 +              *cp = 0;
 +              status = rmdir(name);
 +              if (status) {
 +                      *cp = '/';
 +                      break;
 +              }
 +              prev = cp;
 +      }
 +}
 +
 +static void progress_interval(int signum)
 +{
 +      progress_update = 1;
 +}
 +
 +static void setup_progress_signal(void)
 +{
 +      struct sigaction sa;
 +      struct itimerval v;
 +
 +      memset(&sa, 0, sizeof(sa));
 +      sa.sa_handler = progress_interval;
 +      sigemptyset(&sa.sa_mask);
 +      sa.sa_flags = SA_RESTART;
 +      sigaction(SIGALRM, &sa, NULL);
 +
 +      v.it_interval.tv_sec = 1;
 +      v.it_interval.tv_usec = 0;
 +      v.it_value = v.it_interval;
 +      setitimer(ITIMER_REAL, &v, NULL);
 +}
 +
 +static void check_updates(struct cache_entry **src, int nr)
 +{
 +      static struct checkout state = {
 +              .base_dir = "",
 +              .force = 1,
 +              .quiet = 1,
 +              .refresh_cache = 1,
 +      };
 +      unsigned short mask = htons(CE_UPDATE);
 +      unsigned last_percent = 200, cnt = 0, total = 0;
 +
 +      if (update && verbose_update) {
 +              for (total = cnt = 0; cnt < nr; cnt++) {
 +                      struct cache_entry *ce = src[cnt];
 +                      if (!ce->ce_mode || ce->ce_flags & mask)
 +                              total++;
 +              }
 +
 +              /* Don't bother doing this for very small updates */
 +              if (total < 250)
 +                      total = 0;
 +
 +              if (total) {
 +                      fprintf(stderr, "Checking files out...\n");
 +                      setup_progress_signal();
 +                      progress_update = 1;
 +              }
 +              cnt = 0;
 +      }
 +
 +      while (nr--) {
 +              struct cache_entry *ce = *src++;
 +
 +              if (total) {
 +                      if (!ce->ce_mode || ce->ce_flags & mask) {
 +                              unsigned percent;
 +                              cnt++;
 +                              percent = (cnt * 100) / total;
 +                              if (percent != last_percent ||
 +                                  progress_update) {
 +                                      fprintf(stderr, "%4u%% (%u/%u) done\r",
 +                                              percent, cnt, total);
 +                                      last_percent = percent;
 +                                      progress_update = 0;
 +                              }
 +                      }
 +              }
 +              if (!ce->ce_mode) {
 +                      if (update)
 +                              unlink_entry(ce->name);
 +                      continue;
 +              }
 +              if (ce->ce_flags & mask) {
 +                      ce->ce_flags &= ~mask;
 +                      if (update)
 +                              checkout_entry(ce, &state, NULL);
 +              }
 +      }
 +      if (total) {
 +              signal(SIGALRM, SIG_IGN);
 +              fputc('\n', stderr);
 +      }
 +}
 +
 +static int unpack_trees(merge_fn_t fn)
 +{
 +      int indpos = 0;
 +      unsigned len = object_list_length(trees);
 +      struct tree_entry_list **posns;
 +      int i;
 +      struct object_list *posn = trees;
 +      merge_size = len;
 +
 +      if (len) {
 +              posns = xmalloc(len * sizeof(struct tree_entry_list *));
 +              for (i = 0; i < len; i++) {
 +                      posns[i] = create_tree_entry_list((struct tree *) posn->item);
 +                      posn = posn->next;
 +              }
- static const char read_tree_usage[] = "git-read-tree (<sha> | -m [--aggressive] [-u | -i] <sha1> [<sha2> [<sha3>]])";
++              if (unpack_trees_rec(posns, len, prefix ? prefix : "",
++                                   fn, &indpos))
 +                      return -1;
 +      }
 +
 +      if (trivial_merges_only && nontrivial_merge)
 +              die("Merge requires file-level merging");
 +
 +      check_updates(active_cache, active_nr);
 +      return 0;
 +}
 +
 +static int list_tree(unsigned char *sha1)
 +{
 +      struct tree *tree = parse_tree_indirect(sha1);
 +      if (!tree)
 +              return -1;
 +      object_list_append(&tree->object, &trees);
 +      return 0;
 +}
 +
 +static int same(struct cache_entry *a, struct cache_entry *b)
 +{
 +      if (!!a != !!b)
 +              return 0;
 +      if (!a && !b)
 +              return 1;
 +      return a->ce_mode == b->ce_mode && 
 +              !memcmp(a->sha1, b->sha1, 20);
 +}
 +
 +
 +/*
 + * When a CE gets turned into an unmerged entry, we
 + * want it to be up-to-date
 + */
 +static void verify_uptodate(struct cache_entry *ce)
 +{
 +      struct stat st;
 +
 +      if (index_only || reset)
 +              return;
 +
 +      if (!lstat(ce->name, &st)) {
 +              unsigned changed = ce_match_stat(ce, &st, 1);
 +              if (!changed)
 +                      return;
 +              errno = 0;
 +      }
 +      if (reset) {
 +              ce->ce_flags |= htons(CE_UPDATE);
 +              return;
 +      }
 +      if (errno == ENOENT)
 +              return;
 +      die("Entry '%s' not uptodate. Cannot merge.", ce->name);
 +}
 +
 +static void invalidate_ce_path(struct cache_entry *ce)
 +{
 +      if (ce)
 +              cache_tree_invalidate_path(active_cache_tree, ce->name);
 +}
 +
 +/*
 + * We do not want to remove or overwrite a working tree file that
 + * is not tracked.
 + */
 +static void verify_absent(const char *path, const char *action)
 +{
 +      struct stat st;
 +
 +      if (index_only || reset || !update)
 +              return;
 +      if (!lstat(path, &st))
 +              die("Untracked working tree file '%s' "
 +                  "would be %s by merge.", path, action);
 +}
 +
 +static int merged_entry(struct cache_entry *merge, struct cache_entry *old)
 +{
 +      merge->ce_flags |= htons(CE_UPDATE);
 +      if (old) {
 +              /*
 +               * See if we can re-use the old CE directly?
 +               * That way we get the uptodate stat info.
 +               *
 +               * This also removes the UPDATE flag on
 +               * a match.
 +               */
 +              if (same(old, merge)) {
 +                      *merge = *old;
 +              } else {
 +                      verify_uptodate(old);
 +                      invalidate_ce_path(old);
 +              }
 +      }
 +      else {
 +              verify_absent(merge->name, "overwritten");
 +              invalidate_ce_path(merge);
 +      }
 +
 +      merge->ce_flags &= ~htons(CE_STAGEMASK);
 +      add_cache_entry(merge, ADD_CACHE_OK_TO_ADD);
 +      return 1;
 +}
 +
 +static int deleted_entry(struct cache_entry *ce, struct cache_entry *old)
 +{
 +      if (old)
 +              verify_uptodate(old);
 +      else
 +              verify_absent(ce->name, "removed");
 +      ce->ce_mode = 0;
 +      add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
 +      invalidate_ce_path(ce);
 +      return 1;
 +}
 +
 +static int keep_entry(struct cache_entry *ce)
 +{
 +      add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
 +      return 1;
 +}
 +
 +#if DBRT_DEBUG
 +static void show_stage_entry(FILE *o,
 +                           const char *label, const struct cache_entry *ce)
 +{
 +      if (!ce)
 +              fprintf(o, "%s (missing)\n", label);
 +      else
 +              fprintf(o, "%s%06o %s %d\t%s\n",
 +                      label,
 +                      ntohl(ce->ce_mode),
 +                      sha1_to_hex(ce->sha1),
 +                      ce_stage(ce),
 +                      ce->name);
 +}
 +#endif
 +
 +static int threeway_merge(struct cache_entry **stages)
 +{
 +      struct cache_entry *index;
 +      struct cache_entry *head; 
 +      struct cache_entry *remote = stages[head_idx + 1];
 +      int count;
 +      int head_match = 0;
 +      int remote_match = 0;
 +      const char *path = NULL;
 +
 +      int df_conflict_head = 0;
 +      int df_conflict_remote = 0;
 +
 +      int any_anc_missing = 0;
 +      int no_anc_exists = 1;
 +      int i;
 +
 +      for (i = 1; i < head_idx; i++) {
 +              if (!stages[i])
 +                      any_anc_missing = 1;
 +              else {
 +                      if (!path)
 +                              path = stages[i]->name;
 +                      no_anc_exists = 0;
 +              }
 +      }
 +
 +      index = stages[0];
 +      head = stages[head_idx];
 +
 +      if (head == &df_conflict_entry) {
 +              df_conflict_head = 1;
 +              head = NULL;
 +      }
 +
 +      if (remote == &df_conflict_entry) {
 +              df_conflict_remote = 1;
 +              remote = NULL;
 +      }
 +
 +      if (!path && index)
 +              path = index->name;
 +      if (!path && head)
 +              path = head->name;
 +      if (!path && remote)
 +              path = remote->name;
 +
 +      /* First, if there's a #16 situation, note that to prevent #13
 +       * and #14.
 +       */
 +      if (!same(remote, head)) {
 +              for (i = 1; i < head_idx; i++) {
 +                      if (same(stages[i], head)) {
 +                              head_match = i;
 +                      }
 +                      if (same(stages[i], remote)) {
 +                              remote_match = i;
 +                      }
 +              }
 +      }
 +
 +      /* We start with cases where the index is allowed to match
 +       * something other than the head: #14(ALT) and #2ALT, where it
 +       * is permitted to match the result instead.
 +       */
 +      /* #14, #14ALT, #2ALT */
 +      if (remote && !df_conflict_head && head_match && !remote_match) {
 +              if (index && !same(index, remote) && !same(index, head))
 +                      reject_merge(index);
 +              return merged_entry(remote, index);
 +      }
 +      /*
 +       * If we have an entry in the index cache, then we want to
 +       * make sure that it matches head.
 +       */
 +      if (index && !same(index, head)) {
 +              reject_merge(index);
 +      }
 +
 +      if (head) {
 +              /* #5ALT, #15 */
 +              if (same(head, remote))
 +                      return merged_entry(head, index);
 +              /* #13, #3ALT */
 +              if (!df_conflict_remote && remote_match && !head_match)
 +                      return merged_entry(head, index);
 +      }
 +
 +      /* #1 */
 +      if (!head && !remote && any_anc_missing)
 +              return 0;
 +
 +      /* Under the new "aggressive" rule, we resolve mostly trivial
 +       * cases that we historically had git-merge-one-file resolve.
 +       */
 +      if (aggressive) {
 +              int head_deleted = !head && !df_conflict_head;
 +              int remote_deleted = !remote && !df_conflict_remote;
 +              /*
 +               * Deleted in both.
 +               * Deleted in one and unchanged in the other.
 +               */
 +              if ((head_deleted && remote_deleted) ||
 +                  (head_deleted && remote && remote_match) ||
 +                  (remote_deleted && head && head_match)) {
 +                      if (index)
 +                              return deleted_entry(index, index);
 +                      else if (path)
 +                              verify_absent(path, "removed");
 +                      return 0;
 +              }
 +              /*
 +               * Added in both, identically.
 +               */
 +              if (no_anc_exists && head && remote && same(head, remote))
 +                      return merged_entry(head, index);
 +
 +      }
 +
 +      /* Below are "no merge" cases, which require that the index be
 +       * up-to-date to avoid the files getting overwritten with
 +       * conflict resolution files. 
 +       */
 +      if (index) {
 +              verify_uptodate(index);
 +      }
 +      else if (path)
 +              verify_absent(path, "overwritten");
 +
 +      nontrivial_merge = 1;
 +
 +      /* #2, #3, #4, #6, #7, #9, #11. */
 +      count = 0;
 +      if (!head_match || !remote_match) {
 +              for (i = 1; i < head_idx; i++) {
 +                      if (stages[i]) {
 +                              keep_entry(stages[i]);
 +                              count++;
 +                              break;
 +                      }
 +              }
 +      }
 +#if DBRT_DEBUG
 +      else {
 +              fprintf(stderr, "read-tree: warning #16 detected\n");
 +              show_stage_entry(stderr, "head   ", stages[head_match]);
 +              show_stage_entry(stderr, "remote ", stages[remote_match]);
 +      }
 +#endif
 +      if (head) { count += keep_entry(head); }
 +      if (remote) { count += keep_entry(remote); }
 +      return count;
 +}
 +
 +/*
 + * Two-way merge.
 + *
 + * The rule is to "carry forward" what is in the index without losing
 + * information across a "fast forward", favoring a successful merge
 + * over a merge failure when it makes sense.  For details of the
 + * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
 + *
 + */
 +static int twoway_merge(struct cache_entry **src)
 +{
 +      struct cache_entry *current = src[0];
 +      struct cache_entry *oldtree = src[1], *newtree = src[2];
 +
 +      if (merge_size != 2)
 +              return error("Cannot do a twoway merge of %d trees",
 +                           merge_size);
 +
 +      if (current) {
 +              if ((!oldtree && !newtree) || /* 4 and 5 */
 +                  (!oldtree && newtree &&
 +                   same(current, newtree)) || /* 6 and 7 */
 +                  (oldtree && newtree &&
 +                   same(oldtree, newtree)) || /* 14 and 15 */
 +                  (oldtree && newtree &&
 +                   !same(oldtree, newtree) && /* 18 and 19*/
 +                   same(current, newtree))) {
 +                      return keep_entry(current);
 +              }
 +              else if (oldtree && !newtree && same(current, oldtree)) {
 +                      /* 10 or 11 */
 +                      return deleted_entry(oldtree, current);
 +              }
 +              else if (oldtree && newtree &&
 +                       same(current, oldtree) && !same(current, newtree)) {
 +                      /* 20 or 21 */
 +                      return merged_entry(newtree, current);
 +              }
 +              else {
 +                      /* all other failures */
 +                      if (oldtree)
 +                              reject_merge(oldtree);
 +                      if (current)
 +                              reject_merge(current);
 +                      if (newtree)
 +                              reject_merge(newtree);
 +                      return -1;
 +              }
 +      }
 +      else if (newtree)
 +              return merged_entry(newtree, current);
 +      else
 +              return deleted_entry(oldtree, current);
 +}
 +
 +/*
++ * Bind merge.
++ *
++ * Keep the index entries at stage0, collapse stage1 but make sure
++ * stage0 does not have anything there.
++ */
++static int bind_merge(struct cache_entry **src)
++{
++      struct cache_entry *old = src[0];
++      struct cache_entry *a = src[1];
++
++      if (merge_size != 1)
++              return error("Cannot do a bind merge of %d trees\n",
++                           merge_size);
++      if (a && old)
++              die("Entry '%s' overlaps.  Cannot bind.", a->name);
++      if (!a)
++              return keep_entry(old);
++      else
++              return merged_entry(a, NULL);
++}
++
++/*
 + * One-way merge.
 + *
 + * The rule is:
 + * - take the stat information from stage0, take the data from stage1
 + */
 +static int oneway_merge(struct cache_entry **src)
 +{
 +      struct cache_entry *old = src[0];
 +      struct cache_entry *a = src[1];
 +
 +      if (merge_size != 1)
 +              return error("Cannot do a oneway merge of %d trees",
 +                           merge_size);
 +
 +      if (!a)
 +              return deleted_entry(old, old);
 +      if (old && same(old, a)) {
 +              if (reset) {
 +                      struct stat st;
 +                      if (lstat(old->name, &st) ||
 +                          ce_match_stat(old, &st, 1))
 +                              old->ce_flags |= htons(CE_UPDATE);
 +              }
 +              return keep_entry(old);
 +      }
 +      return merged_entry(a, old);
 +}
 +
 +static int read_cache_unmerged(void)
 +{
 +      int i;
 +      struct cache_entry **dst;
 +      struct cache_entry *last = NULL;
 +
 +      read_cache();
 +      dst = active_cache;
 +      for (i = 0; i < active_nr; i++) {
 +              struct cache_entry *ce = active_cache[i];
 +              if (ce_stage(ce)) {
 +                      if (last && !strcmp(ce->name, last->name))
 +                              continue;
 +                      invalidate_ce_path(ce);
 +                      last = ce;
 +                      ce->ce_mode = 0;
 +                      ce->ce_flags &= ~htons(CE_STAGEMASK);
 +              }
 +              *dst++ = ce;
 +      }
 +      active_nr = dst - active_cache;
 +      return !!last;
 +}
 +
 +static void prime_cache_tree_rec(struct cache_tree *it, struct tree *tree)
 +{
 +      struct tree_desc desc;
 +      struct name_entry entry;
 +      int cnt;
 +
 +      memcpy(it->sha1, tree->object.sha1, 20);
 +      desc.buf = tree->buffer;
 +      desc.size = tree->size;
 +      cnt = 0;
 +      while (tree_entry(&desc, &entry)) {
 +              if (!S_ISDIR(entry.mode))
 +                      cnt++;
 +              else {
 +                      struct cache_tree_sub *sub;
 +                      struct tree *subtree = lookup_tree(entry.sha1);
 +                      if (!subtree->object.parsed)
 +                              parse_tree(subtree);
 +                      sub = cache_tree_sub(it, entry.path);
 +                      sub->cache_tree = cache_tree();
 +                      prime_cache_tree_rec(sub->cache_tree, subtree);
 +                      cnt += sub->cache_tree->entry_count;
 +              }
 +      }
 +      it->entry_count = cnt;
 +}
 +
 +static void prime_cache_tree(void)
 +{
 +      struct tree *tree = (struct tree *)trees->item;
 +      if (!tree)
 +              return;
 +      active_cache_tree = cache_tree();
 +      prime_cache_tree_rec(active_cache_tree, tree);
 +
 +}
 +
-                       if (stage || merge)
++static const char read_tree_usage[] = "git-read-tree (<sha> | [[-m [--aggressive] | --reset | --prefix=<prefix>] [-u | -i]] <sha1> [<sha2> [<sha3>]])";
 +
 +static struct lock_file lock_file;
 +
 +int cmd_read_tree(int argc, const char **argv, char **envp)
 +{
 +      int i, newfd, stage = 0;
 +      unsigned char sha1[20];
 +      merge_fn_t fn = NULL;
 +
 +      setup_git_directory();
 +      git_config(git_default_config);
 +
 +      newfd = hold_lock_file_for_update(&lock_file, get_index_file());
 +      if (newfd < 0)
 +              die("unable to create new index file");
 +
 +      git_config(git_default_config);
 +
 +      merge = 0;
 +      reset = 0;
 +      for (i = 1; i < argc; i++) {
 +              const char *arg = argv[i];
 +
 +              /* "-u" means "update", meaning that a merge will update
 +               * the working tree.
 +               */
 +              if (!strcmp(arg, "-u")) {
 +                      update = 1;
 +                      continue;
 +              }
 +
 +              if (!strcmp(arg, "-v")) {
 +                      verbose_update = 1;
 +                      continue;
 +              }
 +
 +              /* "-i" means "index only", meaning that a merge will
 +               * not even look at the working tree.
 +               */
 +              if (!strcmp(arg, "-i")) {
 +                      index_only = 1;
 +                      continue;
 +              }
 +
++              /* "--prefix=<subdirectory>/" means keep the current index
++               *  entries and put the entries from the tree under the
++               * given subdirectory.
++               */
++              if (!strncmp(arg, "--prefix=", 9)) {
++                      if (stage || merge || prefix)
++                              usage(read_tree_usage);
++                      prefix = arg + 9;
++                      merge = 1;
++                      stage = 1;
++                      if (read_cache_unmerged())
++                              die("you need to resolve your current index first");
++                      continue;
++              }
++
 +              /* This differs from "-m" in that we'll silently ignore
 +               * unmerged entries and overwrite working tree files that
 +               * correspond to them.
 +               */
 +              if (!strcmp(arg, "--reset")) {
-                       if (stage || merge)
++                      if (stage || merge || prefix)
 +                              usage(read_tree_usage);
 +                      reset = 1;
 +                      merge = 1;
 +                      stage = 1;
 +                      read_cache_unmerged();
 +                      continue;
 +              }
 +
 +              if (!strcmp(arg, "--trivial")) {
 +                      trivial_merges_only = 1;
 +                      continue;
 +              }
 +
 +              if (!strcmp(arg, "--aggressive")) {
 +                      aggressive = 1;
 +                      continue;
 +              }
 +
 +              /* "-m" stands for "merge", meaning we start in stage 1 */
 +              if (!strcmp(arg, "-m")) {
-                       fn = oneway_merge;
++                      if (stage || merge || prefix)
 +                              usage(read_tree_usage);
 +                      if (read_cache_unmerged())
 +                              die("you need to resolve your current index first");
 +                      stage = 1;
 +                      merge = 1;
 +                      continue;
 +              }
 +
 +              /* using -u and -i at the same time makes no sense */
 +              if (1 < index_only + update)
 +                      usage(read_tree_usage);
 +
 +              if (get_sha1(arg, sha1))
 +                      die("Not a valid object name %s", arg);
 +              if (list_tree(sha1) < 0)
 +                      die("failed to unpack tree object %s", arg);
 +              stage++;
 +      }
 +      if ((update||index_only) && !merge)
 +              usage(read_tree_usage);
 +
++      if (prefix) {
++              int pfxlen = strlen(prefix);
++              int pos;
++              if (prefix[pfxlen-1] != '/')
++                      die("prefix must end with /");
++              if (stage != 2)
++                      die("binding merge takes only one tree");
++              pos = cache_name_pos(prefix, pfxlen);
++              if (0 <= pos)
++                      die("corrupt index file");
++              pos = -pos-1;
++              if (pos < active_nr &&
++                  !strncmp(active_cache[pos]->name, prefix, pfxlen))
++                      die("subdirectory '%s' already exists.", prefix);
++              pos = cache_name_pos(prefix, pfxlen-1);
++              if (0 <= pos)
++                      die("file '%.*s' already exists.", pfxlen-1, prefix);
++      }
++
 +      if (merge) {
 +              if (stage < 2)
 +                      die("just how do you expect me to merge %d trees?", stage-1);
 +              switch (stage - 1) {
 +              case 1:
-       if (trees && trees->item && (!merge || (stage == 2))) {
++                      fn = prefix ? bind_merge : oneway_merge;
 +                      break;
 +              case 2:
 +                      fn = twoway_merge;
 +                      break;
 +              case 3:
 +              default:
 +                      fn = threeway_merge;
 +                      cache_tree_free(&active_cache_tree);
 +                      break;
 +              }
 +
 +              if (stage - 1 >= 3)
 +                      head_idx = stage - 2;
 +              else
 +                      head_idx = 1;
 +      }
 +
 +      unpack_trees(fn);
 +
 +      /*
 +       * When reading only one tree (either the most basic form,
 +       * "-m ent" or "--reset ent" form), we can obtain a fully
 +       * valid cache-tree because the index must match exactly
 +       * what came from the tree.
 +       */
++      if (trees && trees->item && !prefix && (!merge || (stage == 2))) {
 +              cache_tree_free(&active_cache_tree);
 +              prime_cache_tree();
 +      }
 +
 +      if (write_cache(newfd, active_cache, active_nr) ||
 +          commit_lock_file(&lock_file))
 +              die("unable to write new index file");
 +      return 0;
 +}
diff --cc cache-tree.c
Simple merge
diff --cc write-tree.c
@@@ -8,10 -8,12 +8,12 @@@
  #include "cache-tree.h"
  
  static int missing_ok = 0;
+ static char *prefix = NULL;
  
- static const char write_tree_usage[] = "git-write-tree [--missing-ok]";
+ static const char write_tree_usage[] =
+ "git-write-tree [--missing-ok] [--prefix=<prefix>/]";
  
 -static struct cache_file cache_file;
 +static struct lock_file lock_file;
  
  int main(int argc, char **argv)
  {
  
        setup_git_directory();
  
 -      newfd = hold_index_file_for_update(&cache_file, get_index_file());
 +      newfd = hold_lock_file_for_update(&lock_file, get_index_file());
        entries = read_cache();
-       if (argc == 2) {
-               if (!strcmp(argv[1], "--missing-ok"))
+       while (1 < argc) {
+               char *arg = argv[1];
+               if (!strcmp(arg, "--missing-ok"))
                        missing_ok = 1;
+               else if (!strncmp(arg, "--prefix=", 9))
+                       prefix = arg + 9;
                else
                        die(write_tree_usage);
+               argc--; argv++;
        }
-       
        if (argc > 2)
                die("too many options");