imap-send: properly error out if imap.host is not set in config
[gitweb.git] / read-cache.c
index 8a7506d909421479eee30517ab3f3963c67cefd9..a92b25b59bf0e096942bca126542a1ea411b525b 100644 (file)
@@ -3,8 +3,11 @@
  *
  * Copyright (C) Linus Torvalds, 2005
  */
+#define NO_THE_INDEX_COMPATIBILITY_MACROS
 #include "cache.h"
 #include "cache-tree.h"
+#include "refs.h"
+#include "dir.h"
 
 /* Index extensions.
  *
 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
 #define CACHE_EXT_TREE 0x54524545      /* "TREE" */
 
-struct cache_entry **active_cache;
-static time_t index_file_timestamp;
-unsigned int active_nr, active_alloc, active_cache_changed;
+struct index_state the_index;
 
-struct cache_tree *active_cache_tree;
+static unsigned int hash_name(const char *name, int namelen)
+{
+       unsigned int hash = 0x123;
+
+       do {
+               unsigned char c = *name++;
+               hash = hash*101 + c;
+       } while (--namelen);
+       return hash;
+}
+
+static void hash_index_entry(struct index_state *istate, struct cache_entry *ce)
+{
+       void **pos;
+       unsigned int hash;
+
+       if (ce->ce_flags & CE_HASHED)
+               return;
+       ce->ce_flags |= CE_HASHED;
+       ce->next = NULL;
+       hash = hash_name(ce->name, ce_namelen(ce));
+       pos = insert_hash(hash, ce, &istate->name_hash);
+       if (pos) {
+               ce->next = *pos;
+               *pos = ce;
+       }
+}
+
+static void lazy_init_name_hash(struct index_state *istate)
+{
+       int nr;
+
+       if (istate->name_hash_initialized)
+               return;
+       for (nr = 0; nr < istate->cache_nr; nr++)
+               hash_index_entry(istate, istate->cache[nr]);
+       istate->name_hash_initialized = 1;
+}
+
+static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
+{
+       ce->ce_flags &= ~CE_UNHASHED;
+       istate->cache[nr] = ce;
+       if (istate->name_hash_initialized)
+               hash_index_entry(istate, ce);
+}
+
+static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
+{
+       struct cache_entry *old = istate->cache[nr];
+
+       remove_index_entry(old);
+       set_index_entry(istate, nr, ce);
+       istate->cache_changed = 1;
+}
+
+int index_name_exists(struct index_state *istate, const char *name, int namelen)
+{
+       unsigned int hash = hash_name(name, namelen);
+       struct cache_entry *ce;
 
-static int cache_errno;
+       lazy_init_name_hash(istate);
+       ce = lookup_hash(hash, &istate->name_hash);
 
-static void *cache_mmap;
-static size_t cache_mmap_size;
+       while (ce) {
+               if (!(ce->ce_flags & CE_UNHASHED)) {
+                       if (!cache_name_compare(name, namelen, ce->name, ce->ce_flags))
+                               return 1;
+               }
+               ce = ce->next;
+       }
+       return 0;
+}
 
 /*
  * This only updates the "non-critical" parts of the directory
@@ -36,20 +104,19 @@ static size_t cache_mmap_size;
  */
 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
 {
-       ce->ce_ctime.sec = htonl(st->st_ctime);
-       ce->ce_mtime.sec = htonl(st->st_mtime);
-#ifdef USE_NSEC
-       ce->ce_ctime.nsec = htonl(st->st_ctim.tv_nsec);
-       ce->ce_mtime.nsec = htonl(st->st_mtim.tv_nsec);
-#endif
-       ce->ce_dev = htonl(st->st_dev);
-       ce->ce_ino = htonl(st->st_ino);
-       ce->ce_uid = htonl(st->st_uid);
-       ce->ce_gid = htonl(st->st_gid);
-       ce->ce_size = htonl(st->st_size);
+       ce->ce_ctime = st->st_ctime;
+       ce->ce_mtime = st->st_mtime;
+       ce->ce_dev = st->st_dev;
+       ce->ce_ino = st->st_ino;
+       ce->ce_uid = st->st_uid;
+       ce->ce_gid = st->st_gid;
+       ce->ce_size = st->st_size;
 
        if (assume_unchanged)
-               ce->ce_flags |= htons(CE_VALID);
+               ce->ce_flags |= CE_VALID;
+
+       if (S_ISREG(st->st_mode))
+               ce_mark_uptodate(ce);
 }
 
 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
@@ -93,6 +160,23 @@ static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
        return match;
 }
 
+static int ce_compare_gitlink(struct cache_entry *ce)
+{
+       unsigned char sha1[20];
+
+       /*
+        * We don't actually require that the .git directory
+        * under GITLINK directory be a valid git directory. It
+        * might even be missing (in case nobody populated that
+        * sub-project).
+        *
+        * If so, we consider it always to match.
+        */
+       if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
+               return 0;
+       return hashcmp(sha1, ce->sha1);
+}
+
 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
 {
        switch (st->st_mode & S_IFMT) {
@@ -104,6 +188,9 @@ static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
                if (ce_compare_link(ce, xsize_t(st->st_size)))
                        return DATA_CHANGED;
                break;
+       case S_IFDIR:
+               if (S_ISGITLINK(ce->ce_mode))
+                       return 0;
        default:
                return TYPE_CHANGED;
        }
@@ -114,14 +201,17 @@ static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
 {
        unsigned int changed = 0;
 
-       switch (ntohl(ce->ce_mode) & S_IFMT) {
+       if (ce->ce_flags & CE_REMOVE)
+               return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
+
+       switch (ce->ce_mode & S_IFMT) {
        case S_IFREG:
                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
                /* We consider only the owner x bit to be relevant for
                 * "mode changes"
                 */
                if (trust_executable_bit &&
-                   (0100 & (ntohl(ce->ce_mode) ^ st->st_mode)))
+                   (0100 & (ce->ce_mode ^ st->st_mode)))
                        changed |= MODE_CHANGED;
                break;
        case S_IFLNK:
@@ -129,30 +219,24 @@ static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
                    (has_symlinks || !S_ISREG(st->st_mode)))
                        changed |= TYPE_CHANGED;
                break;
+       case S_IFGITLINK:
+               if (!S_ISDIR(st->st_mode))
+                       changed |= TYPE_CHANGED;
+               else if (ce_compare_gitlink(ce))
+                       changed |= DATA_CHANGED;
+               return changed;
        default:
-               die("internal error: ce_mode is %o", ntohl(ce->ce_mode));
+               die("internal error: ce_mode is %o", ce->ce_mode);
        }
-       if (ce->ce_mtime.sec != htonl(st->st_mtime))
+       if (ce->ce_mtime != (unsigned int) st->st_mtime)
                changed |= MTIME_CHANGED;
-       if (ce->ce_ctime.sec != htonl(st->st_ctime))
+       if (ce->ce_ctime != (unsigned int) st->st_ctime)
                changed |= CTIME_CHANGED;
 
-#ifdef USE_NSEC
-       /*
-        * nsec seems unreliable - not all filesystems support it, so
-        * as long as it is in the inode cache you get right nsec
-        * but after it gets flushed, you get zero nsec.
-        */
-       if (ce->ce_mtime.nsec != htonl(st->st_mtim.tv_nsec))
-               changed |= MTIME_CHANGED;
-       if (ce->ce_ctime.nsec != htonl(st->st_ctim.tv_nsec))
-               changed |= CTIME_CHANGED;
-#endif 
-
-       if (ce->ce_uid != htonl(st->st_uid) ||
-           ce->ce_gid != htonl(st->st_gid))
+       if (ce->ce_uid != (unsigned int) st->st_uid ||
+           ce->ce_gid != (unsigned int) st->st_gid)
                changed |= OWNER_CHANGED;
-       if (ce->ce_ino != htonl(st->st_ino))
+       if (ce->ce_ino != (unsigned int) st->st_ino)
                changed |= INODE_CHANGED;
 
 #ifdef USE_STDEV
@@ -161,27 +245,35 @@ static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
         * clients will have different views of what "device"
         * the filesystem is on
         */
-       if (ce->ce_dev != htonl(st->st_dev))
+       if (ce->ce_dev != (unsigned int) st->st_dev)
                changed |= INODE_CHANGED;
 #endif
 
-       if (ce->ce_size != htonl(st->st_size))
+       if (ce->ce_size != (unsigned int) st->st_size)
                changed |= DATA_CHANGED;
 
        return changed;
 }
 
-int ce_match_stat(struct cache_entry *ce, struct stat *st, int options)
+static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
+{
+       return (istate->timestamp &&
+               ((unsigned int)istate->timestamp) <= ce->ce_mtime);
+}
+
+int ie_match_stat(const struct index_state *istate,
+                 struct cache_entry *ce, struct stat *st,
+                 unsigned int options)
 {
        unsigned int changed;
-       int ignore_valid = options & 01;
-       int assume_racy_is_modified = options & 02;
+       int ignore_valid = options & CE_MATCH_IGNORE_VALID;
+       int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 
        /*
         * If it's marked as always valid in the index, it's
         * valid whatever the checked-out copy says.
         */
-       if (!ignore_valid && (ce->ce_flags & htons(CE_VALID)))
+       if (!ignore_valid && (ce->ce_flags & CE_VALID))
                return 0;
 
        changed = ce_match_stat_basic(ce, st);
@@ -202,9 +294,7 @@ int ce_match_stat(struct cache_entry *ce, struct stat *st, int options)
         * whose mtime are the same as the index file timestamp more
         * carefully than others.
         */
-       if (!changed &&
-           index_file_timestamp &&
-           index_file_timestamp <= ntohl(ce->ce_mtime.sec)) {
+       if (!changed && is_racy_timestamp(istate, ce)) {
                if (assume_racy_is_modified)
                        changed |= DATA_CHANGED;
                else
@@ -214,10 +304,12 @@ int ce_match_stat(struct cache_entry *ce, struct stat *st, int options)
        return changed;
 }
 
-int ce_modified(struct cache_entry *ce, struct stat *st, int really)
+int ie_modified(const struct index_state *istate,
+               struct cache_entry *ce, struct stat *st, unsigned int options)
 {
        int changed, changed_fs;
-       changed = ce_match_stat(ce, st, really);
+
+       changed = ie_match_stat(istate, ce, st, options);
        if (!changed)
                return 0;
        /*
@@ -231,7 +323,7 @@ int ce_modified(struct cache_entry *ce, struct stat *st, int really)
         * the length field is zero.  For other cases the ce_size
         * should match the SHA1 recorded in the index entry.
         */
-       if ((changed & DATA_CHANGED) && ce->ce_size != htonl(0))
+       if ((changed & DATA_CHANGED) && ce->ce_size != 0)
                return changed;
 
        changed_fs = ce_modified_check_fs(ce, st);
@@ -259,6 +351,41 @@ int base_name_compare(const char *name1, int len1, int mode1,
        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 }
 
+/*
+ * df_name_compare() is identical to base_name_compare(), except it
+ * compares conflicting directory/file entries as equal. Note that
+ * while a directory name compares as equal to a regular file, they
+ * then individually compare _differently_ to a filename that has
+ * a dot after the basename (because '\0' < '.' < '/').
+ *
+ * This is used by routines that want to traverse the git namespace
+ * but then handle conflicting entries together when possible.
+ */
+int df_name_compare(const char *name1, int len1, int mode1,
+                   const char *name2, int len2, int mode2)
+{
+       int len = len1 < len2 ? len1 : len2, cmp;
+       unsigned char c1, c2;
+
+       cmp = memcmp(name1, name2, len);
+       if (cmp)
+               return cmp;
+       /* Directories and files compare equal (same length, same name) */
+       if (len1 == len2)
+               return 0;
+       c1 = name1[len];
+       if (!c1 && S_ISDIR(mode1))
+               c1 = '/';
+       c2 = name2[len];
+       if (!c2 && S_ISDIR(mode2))
+               c2 = '/';
+       if (c1 == '/' && !c2)
+               return 0;
+       if (c2 == '/' && !c1)
+               return 0;
+       return c1 - c2;
+}
+
 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
 {
        int len1 = flags1 & CE_NAMEMASK;
@@ -285,16 +412,16 @@ int cache_name_compare(const char *name1, int flags1, const char *name2, int fla
        return 0;
 }
 
-int cache_name_pos(const char *name, int namelen)
+int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 {
        int first, last;
 
        first = 0;
-       last = active_nr;
+       last = istate->cache_nr;
        while (last > first) {
                int next = (last + first) >> 1;
-               struct cache_entry *ce = active_cache[next];
-               int cmp = cache_name_compare(name, namelen, ce->name, ntohs(ce->ce_flags));
+               struct cache_entry *ce = istate->cache[next];
+               int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
                if (!cmp)
                        return next;
                if (cmp < 0) {
@@ -307,43 +434,82 @@ int cache_name_pos(const char *name, int namelen)
 }
 
 /* Remove entry, return true if there are more entries to go.. */
-int remove_cache_entry_at(int pos)
+int remove_index_entry_at(struct index_state *istate, int pos)
 {
-       active_cache_changed = 1;
-       active_nr--;
-       if (pos >= active_nr)
+       struct cache_entry *ce = istate->cache[pos];
+
+       remove_index_entry(ce);
+       istate->cache_changed = 1;
+       istate->cache_nr--;
+       if (pos >= istate->cache_nr)
                return 0;
-       memmove(active_cache + pos, active_cache + pos + 1, (active_nr - pos) * sizeof(struct cache_entry *));
+       memmove(istate->cache + pos,
+               istate->cache + pos + 1,
+               (istate->cache_nr - pos) * sizeof(struct cache_entry *));
        return 1;
 }
 
-int remove_file_from_cache(const char *path)
+int remove_file_from_index(struct index_state *istate, const char *path)
 {
-       int pos = cache_name_pos(path, strlen(path));
+       int pos = index_name_pos(istate, path, strlen(path));
        if (pos < 0)
                pos = -pos-1;
-       while (pos < active_nr && !strcmp(active_cache[pos]->name, path))
-               remove_cache_entry_at(pos);
+       cache_tree_invalidate_path(istate->cache_tree, path);
+       while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
+               remove_index_entry_at(istate, pos);
        return 0;
 }
 
-int add_file_to_index(const char *path, int verbose)
+static int compare_name(struct cache_entry *ce, const char *path, int namelen)
+{
+       return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
+}
+
+static int index_name_pos_also_unmerged(struct index_state *istate,
+       const char *path, int namelen)
 {
-       int size, namelen;
+       int pos = index_name_pos(istate, path, namelen);
+       struct cache_entry *ce;
+
+       if (pos >= 0)
+               return pos;
+
+       /* maybe unmerged? */
+       pos = -1 - pos;
+       if (pos >= istate->cache_nr ||
+                       compare_name((ce = istate->cache[pos]), path, namelen))
+               return -1;
+
+       /* order of preference: stage 2, 1, 3 */
+       if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
+                       ce_stage((ce = istate->cache[pos + 1])) == 2 &&
+                       !compare_name(ce, path, namelen))
+               pos++;
+       return pos;
+}
+
+int add_file_to_index(struct index_state *istate, const char *path, int verbose)
+{
+       int size, namelen, pos;
        struct stat st;
        struct cache_entry *ce;
+       unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
 
        if (lstat(path, &st))
                die("%s: unable to stat (%s)", path, strerror(errno));
 
-       if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode))
-               die("%s: can only add regular files or symbolic links", path);
+       if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode) && !S_ISDIR(st.st_mode))
+               die("%s: can only add regular files, symbolic links or git-directories", path);
 
        namelen = strlen(path);
+       if (S_ISDIR(st.st_mode)) {
+               while (namelen && path[namelen-1] == '/')
+                       namelen--;
+       }
        size = cache_entry_size(namelen);
        ce = xcalloc(1, size);
        memcpy(ce->name, path, namelen);
-       ce->ce_flags = htons(namelen);
+       ce->ce_flags = namelen;
        fill_stat_cache_info(ce, &st);
 
        if (trust_executable_bit && has_symlinks)
@@ -353,22 +519,56 @@ int add_file_to_index(const char *path, int verbose)
                 * from it, otherwise assume unexecutable regular file.
                 */
                struct cache_entry *ent;
-               int pos = cache_name_pos(path, namelen);
+               int pos = index_name_pos_also_unmerged(istate, path, namelen);
 
-               ent = (0 <= pos) ? active_cache[pos] : NULL;
+               ent = (0 <= pos) ? istate->cache[pos] : NULL;
                ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
        }
 
+       pos = index_name_pos(istate, ce->name, namelen);
+       if (0 <= pos &&
+           !ce_stage(istate->cache[pos]) &&
+           !ie_match_stat(istate, istate->cache[pos], &st, ce_option)) {
+               /* Nothing changed, really */
+               free(ce);
+               ce_mark_uptodate(istate->cache[pos]);
+               return 0;
+       }
+
        if (index_path(ce->sha1, path, &st, 1))
                die("unable to index file %s", path);
-       if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
+       if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
                die("unable to add %s to index",path);
        if (verbose)
                printf("add '%s'\n", path);
-       cache_tree_invalidate_path(active_cache_tree, path);
        return 0;
 }
 
+struct cache_entry *make_cache_entry(unsigned int mode,
+               const unsigned char *sha1, const char *path, int stage,
+               int refresh)
+{
+       int size, len;
+       struct cache_entry *ce;
+
+       if (!verify_path(path))
+               return NULL;
+
+       len = strlen(path);
+       size = cache_entry_size(len);
+       ce = xcalloc(1, size);
+
+       hashcpy(ce->sha1, sha1);
+       memcpy(ce->name, path, len);
+       ce->ce_flags = create_ce_flags(len, stage);
+       ce->ce_mode = create_ce_mode(mode);
+
+       if (refresh)
+               return refresh_cache_entry(ce, 0);
+
+       return ce;
+}
+
 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 {
        int len = ce_namelen(a);
@@ -469,15 +669,16 @@ int verify_path(const char *path)
  * Do we have another file that has the beginning components being a
  * proper superset of the name we're trying to add?
  */
-static int has_file_name(const struct cache_entry *ce, int pos, int ok_to_replace)
+static int has_file_name(struct index_state *istate,
+                        const struct cache_entry *ce, int pos, int ok_to_replace)
 {
        int retval = 0;
        int len = ce_namelen(ce);
        int stage = ce_stage(ce);
        const char *name = ce->name;
 
-       while (pos < active_nr) {
-               struct cache_entry *p = active_cache[pos++];
+       while (pos < istate->cache_nr) {
+               struct cache_entry *p = istate->cache[pos++];
 
                if (len >= ce_namelen(p))
                        break;
@@ -487,10 +688,12 @@ static int has_file_name(const struct cache_entry *ce, int pos, int ok_to_replac
                        continue;
                if (p->name[len] != '/')
                        continue;
+               if (p->ce_flags & CE_REMOVE)
+                       continue;
                retval = -1;
                if (!ok_to_replace)
                        break;
-               remove_cache_entry_at(--pos);
+               remove_index_entry_at(istate, --pos);
        }
        return retval;
 }
@@ -499,7 +702,8 @@ static int has_file_name(const struct cache_entry *ce, int pos, int ok_to_replac
  * Do we have another file with a pathname that is a proper
  * subset of the name we're trying to add?
  */
-static int has_dir_name(const struct cache_entry *ce, int pos, int ok_to_replace)
+static int has_dir_name(struct index_state *istate,
+                       const struct cache_entry *ce, int pos, int ok_to_replace)
 {
        int retval = 0;
        int stage = ce_stage(ce);
@@ -517,29 +721,41 @@ static int has_dir_name(const struct cache_entry *ce, int pos, int ok_to_replace
                }
                len = slash - name;
 
-               pos = cache_name_pos(name, ntohs(create_ce_flags(len, stage)));
+               pos = index_name_pos(istate, name, create_ce_flags(len, stage));
                if (pos >= 0) {
-                       retval = -1;
-                       if (!ok_to_replace)
-                               break;
-                       remove_cache_entry_at(pos);
-                       continue;
+                       /*
+                        * Found one, but not so fast.  This could
+                        * be a marker that says "I was here, but
+                        * I am being removed".  Such an entry is
+                        * not a part of the resulting tree, and
+                        * it is Ok to have a directory at the same
+                        * path.
+                        */
+                       if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
+                               retval = -1;
+                               if (!ok_to_replace)
+                                       break;
+                               remove_index_entry_at(istate, pos);
+                               continue;
+                       }
                }
+               else
+                       pos = -pos-1;
 
                /*
                 * Trivial optimization: if we find an entry that
                 * already matches the sub-directory, then we know
                 * we're ok, and we can exit.
                 */
-               pos = -pos-1;
-               while (pos < active_nr) {
-                       struct cache_entry *p = active_cache[pos];
+               while (pos < istate->cache_nr) {
+                       struct cache_entry *p = istate->cache[pos];
                        if ((ce_namelen(p) <= len) ||
                            (p->name[len] != '/') ||
                            memcmp(p->name, name, len))
                                break; /* not our subdirectory */
-                       if (ce_stage(p) == stage)
-                               /* p is at the same stage as our entry, and
+                       if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
+                               /*
+                                * p is at the same stage as our entry, and
                                 * is a subdirectory of what we are looking
                                 * at, so we cannot have conflicts at our
                                 * level or anything shorter.
@@ -555,39 +771,50 @@ static int has_dir_name(const struct cache_entry *ce, int pos, int ok_to_replace
  * is being added, or we already have path and path/file is being
  * added.  Either one would result in a nonsense tree that has path
  * twice when git-write-tree tries to write it out.  Prevent it.
- * 
+ *
  * If ok-to-replace is specified, we remove the conflicting entries
  * from the cache so the caller should recompute the insert position.
  * When this happens, we return non-zero.
  */
-static int check_file_directory_conflict(const struct cache_entry *ce, int pos, int ok_to_replace)
+static int check_file_directory_conflict(struct index_state *istate,
+                                        const struct cache_entry *ce,
+                                        int pos, int ok_to_replace)
 {
+       int retval;
+
+       /*
+        * When ce is an "I am going away" entry, we allow it to be added
+        */
+       if (ce->ce_flags & CE_REMOVE)
+               return 0;
+
        /*
         * We check if the path is a sub-path of a subsequent pathname
         * first, since removing those will not change the position
-        * in the array
+        * in the array.
         */
-       int retval = has_file_name(ce, pos, ok_to_replace);
+       retval = has_file_name(istate, ce, pos, ok_to_replace);
+
        /*
         * Then check if the path might have a clashing sub-directory
         * before it.
         */
-       return retval + has_dir_name(ce, pos, ok_to_replace);
+       return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 }
 
-int add_cache_entry(struct cache_entry *ce, int option)
+static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 {
        int pos;
        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 
-       pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
+       cache_tree_invalidate_path(istate->cache_tree, ce->name);
+       pos = index_name_pos(istate, ce->name, ce->ce_flags);
 
        /* existing match? Just replace it. */
        if (pos >= 0) {
-               active_cache_changed = 1;
-               active_cache[pos] = ce;
+               replace_index_entry(istate, pos, ce);
                return 0;
        }
        pos = -pos-1;
@@ -596,10 +823,10 @@ int add_cache_entry(struct cache_entry *ce, int option)
         * Inserting a merged entry ("stage 0") into the index
         * will always replace all non-merged entries..
         */
-       if (pos < active_nr && ce_stage(ce) == 0) {
-               while (ce_same_name(active_cache[pos], ce)) {
+       if (pos < istate->cache_nr && ce_stage(ce) == 0) {
+               while (ce_same_name(istate->cache[pos], ce)) {
                        ok_to_add = 1;
-                       if (!remove_cache_entry_at(pos))
+                       if (!remove_index_entry_at(istate, pos))
                                break;
                }
        }
@@ -610,25 +837,45 @@ int add_cache_entry(struct cache_entry *ce, int option)
                return -1;
 
        if (!skip_df_check &&
-           check_file_directory_conflict(ce, pos, ok_to_replace)) {
+           check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
                if (!ok_to_replace)
-                       return error("'%s' appears as both a file and as a directory", ce->name);
-               pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
+                       return error("'%s' appears as both a file and as a directory",
+                                    ce->name);
+               pos = index_name_pos(istate, ce->name, ce->ce_flags);
                pos = -pos-1;
        }
+       return pos + 1;
+}
+
+int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
+{
+       int pos;
+
+       if (option & ADD_CACHE_JUST_APPEND)
+               pos = istate->cache_nr;
+       else {
+               int ret;
+               ret = add_index_entry_with_check(istate, ce, option);
+               if (ret <= 0)
+                       return ret;
+               pos = ret - 1;
+       }
 
        /* Make sure the array is big enough .. */
-       if (active_nr == active_alloc) {
-               active_alloc = alloc_nr(active_alloc);
-               active_cache = xrealloc(active_cache, active_alloc * sizeof(struct cache_entry *));
+       if (istate->cache_nr == istate->cache_alloc) {
+               istate->cache_alloc = alloc_nr(istate->cache_alloc);
+               istate->cache = xrealloc(istate->cache,
+                                       istate->cache_alloc * sizeof(struct cache_entry *));
        }
 
        /* Add it in.. */
-       active_nr++;
-       if (active_nr > pos)
-               memmove(active_cache + pos + 1, active_cache + pos, (active_nr - pos - 1) * sizeof(ce));
-       active_cache[pos] = ce;
-       active_cache_changed = 1;
+       istate->cache_nr++;
+       if (istate->cache_nr > pos + 1)
+               memmove(istate->cache + pos + 1,
+                       istate->cache + pos,
+                       (istate->cache_nr - pos - 1) * sizeof(ce));
+       set_index_entry(istate, pos, ce);
+       istate->cache_changed = 1;
        return 0;
 }
 
@@ -643,28 +890,50 @@ int add_cache_entry(struct cache_entry *ce, int option)
  * For example, you'd want to do this after doing a "git-read-tree",
  * to link up the stat cache details with the proper files.
  */
-struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
+static struct cache_entry *refresh_cache_ent(struct index_state *istate,
+                                            struct cache_entry *ce,
+                                            unsigned int options, int *err)
 {
        struct stat st;
        struct cache_entry *updated;
        int changed, size;
+       int ignore_valid = options & CE_MATCH_IGNORE_VALID;
+
+       if (ce_uptodate(ce))
+               return ce;
 
        if (lstat(ce->name, &st) < 0) {
-               cache_errno = errno;
+               if (err)
+                       *err = errno;
                return NULL;
        }
 
-       changed = ce_match_stat(ce, &st, really);
+       changed = ie_match_stat(istate, ce, &st, options);
        if (!changed) {
-               if (really && assume_unchanged &&
-                   !(ce->ce_flags & htons(CE_VALID)))
+               /*
+                * The path is unchanged.  If we were told to ignore
+                * valid bit, then we did the actual stat check and
+                * found that the entry is unmodified.  If the entry
+                * is not marked VALID, this is the place to mark it
+                * valid again, under "assume unchanged" mode.
+                */
+               if (ignore_valid && assume_unchanged &&
+                   !(ce->ce_flags & CE_VALID))
                        ; /* mark this one VALID again */
-               else
+               else {
+                       /*
+                        * We do not mark the index itself "modified"
+                        * because CE_UPTODATE flag is in-core only;
+                        * we are not going to write this change out.
+                        */
+                       ce_mark_uptodate(ce);
                        return ce;
+               }
        }
 
-       if (ce_modified(ce, &st, really)) {
-               cache_errno = EINVAL;
+       if (ie_modified(istate, ce, &st, options)) {
+               if (err)
+                       *err = EINVAL;
                return NULL;
        }
 
@@ -672,20 +941,20 @@ struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
        updated = xmalloc(size);
        memcpy(updated, ce, size);
        fill_stat_cache_info(updated, &st);
-
-       /* In this case, if really is not set, we should leave
-        * CE_VALID bit alone.  Otherwise, paths marked with
-        * --no-assume-unchanged (i.e. things to be edited) will
-        * reacquire CE_VALID bit automatically, which is not
-        * really what we want.
+       /*
+        * If ignore_valid is not set, we should leave CE_VALID bit
+        * alone.  Otherwise, paths marked with --no-assume-unchanged
+        * (i.e. things to be edited) will reacquire CE_VALID bit
+        * automatically, which is not really what we want.
         */
-       if (!really && assume_unchanged && !(ce->ce_flags & htons(CE_VALID)))
-               updated->ce_flags &= ~htons(CE_VALID);
+       if (!ignore_valid && assume_unchanged &&
+           !(ce->ce_flags & CE_VALID))
+               updated->ce_flags &= ~CE_VALID;
 
        return updated;
 }
 
-int refresh_cache(unsigned int flags)
+int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
 {
        int i;
        int has_errors = 0;
@@ -693,13 +962,16 @@ int refresh_cache(unsigned int flags)
        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
        int quiet = (flags & REFRESH_QUIET) != 0;
        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
+       unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
 
-       for (i = 0; i < active_nr; i++) {
+       for (i = 0; i < istate->cache_nr; i++) {
                struct cache_entry *ce, *new;
-               ce = active_cache[i];
+               int cache_errno = 0;
+
+               ce = istate->cache[i];
                if (ce_stage(ce)) {
-                       while ((i < active_nr) &&
-                              ! strcmp(active_cache[i]->name, ce->name))
+                       while ((i < istate->cache_nr) &&
+                              ! strcmp(istate->cache[i]->name, ce->name))
                                i++;
                        i--;
                        if (allow_unmerged)
@@ -709,7 +981,10 @@ int refresh_cache(unsigned int flags)
                        continue;
                }
 
-               new = refresh_cache_entry(ce, really);
+               if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
+                       continue;
+
+               new = refresh_cache_ent(istate, ce, options, &cache_errno);
                if (new == ce)
                        continue;
                if (!new) {
@@ -719,8 +994,8 @@ int refresh_cache(unsigned int flags)
                                /* If we are doing --really-refresh that
                                 * means the index is not valid anymore.
                                 */
-                               ce->ce_flags &= ~htons(CE_VALID);
-                               active_cache_changed = 1;
+                               ce->ce_flags &= ~CE_VALID;
+                               istate->cache_changed = 1;
                        }
                        if (quiet)
                                continue;
@@ -728,15 +1003,17 @@ int refresh_cache(unsigned int flags)
                        has_errors = 1;
                        continue;
                }
-               active_cache_changed = 1;
-               /* You can NOT just free active_cache[i] here, since it
-                * might not be necessarily malloc()ed but can also come
-                * from mmap(). */
-               active_cache[i] = new;
+
+               replace_index_entry(istate, i, new);
        }
        return has_errors;
 }
 
+struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
+{
+       return refresh_cache_ent(&the_index, ce, really, NULL);
+}
+
 static int verify_hdr(struct cache_header *hdr, unsigned long size)
 {
        SHA_CTX c;
@@ -754,11 +1031,12 @@ static int verify_hdr(struct cache_header *hdr, unsigned long size)
        return 0;
 }
 
-static int read_index_extension(const char *ext, void *data, unsigned long sz)
+static int read_index_extension(struct index_state *istate,
+                               const char *ext, void *data, unsigned long sz)
 {
        switch (CACHE_EXT(ext)) {
        case CACHE_EXT_TREE:
-               active_cache_tree = cache_tree_read(data, sz);
+               istate->cache_tree = cache_tree_read(data, sz);
                break;
        default:
                if (*ext < 'A' || 'Z' < *ext)
@@ -770,25 +1048,67 @@ static int read_index_extension(const char *ext, void *data, unsigned long sz)
        return 0;
 }
 
-int read_cache(void)
+int read_index(struct index_state *istate)
+{
+       return read_index_from(istate, get_index_file());
+}
+
+static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
+{
+       size_t len;
+
+       ce->ce_ctime = ntohl(ondisk->ctime.sec);
+       ce->ce_mtime = ntohl(ondisk->mtime.sec);
+       ce->ce_dev   = ntohl(ondisk->dev);
+       ce->ce_ino   = ntohl(ondisk->ino);
+       ce->ce_mode  = ntohl(ondisk->mode);
+       ce->ce_uid   = ntohl(ondisk->uid);
+       ce->ce_gid   = ntohl(ondisk->gid);
+       ce->ce_size  = ntohl(ondisk->size);
+       /* On-disk flags are just 16 bits */
+       ce->ce_flags = ntohs(ondisk->flags);
+       hashcpy(ce->sha1, ondisk->sha1);
+
+       len = ce->ce_flags & CE_NAMEMASK;
+       if (len == CE_NAMEMASK)
+               len = strlen(ondisk->name);
+       /*
+        * NEEDSWORK: If the original index is crafted, this copy could
+        * go unchecked.
+        */
+       memcpy(ce->name, ondisk->name, len + 1);
+}
+
+static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
 {
-       return read_cache_from(get_index_file());
+       long per_entry;
+
+       per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
+
+       /*
+        * Alignment can cause differences. This should be "alignof", but
+        * since that's a gcc'ism, just use the size of a pointer.
+        */
+       per_entry += sizeof(void *);
+       return ondisk_size + entries*per_entry;
 }
 
 /* remember to discard_cache() before reading a different cache! */
-int read_cache_from(const char *path)
+int read_index_from(struct index_state *istate, const char *path)
 {
        int fd, i;
        struct stat st;
-       unsigned long offset;
+       unsigned long src_offset, dst_offset;
        struct cache_header *hdr;
+       void *mmap;
+       size_t mmap_size;
 
        errno = EBUSY;
-       if (cache_mmap)
-               return active_nr;
+       if (istate->alloc)
+               return istate->cache_nr;
 
        errno = ENOENT;
-       index_file_timestamp = 0;
+       istate->timestamp = 0;
        fd = open(path, O_RDONLY);
        if (fd < 0) {
                if (errno == ENOENT)
@@ -796,33 +1116,51 @@ int read_cache_from(const char *path)
                die("index file open failed (%s)", strerror(errno));
        }
 
-       if (!fstat(fd, &st)) {
-               cache_mmap_size = xsize_t(st.st_size);
-               errno = EINVAL;
-               if (cache_mmap_size >= sizeof(struct cache_header) + 20)
-                       cache_mmap = xmmap(NULL, cache_mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
-               else
-                       die("index file smaller than expected");
-       } else
+       if (fstat(fd, &st))
                die("cannot stat the open index (%s)", strerror(errno));
+
+       errno = EINVAL;
+       mmap_size = xsize_t(st.st_size);
+       if (mmap_size < sizeof(struct cache_header) + 20)
+               die("index file smaller than expected");
+
+       mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
        close(fd);
+       if (mmap == MAP_FAILED)
+               die("unable to map index file");
 
-       hdr = cache_mmap;
-       if (verify_hdr(hdr, cache_mmap_size) < 0)
+       hdr = mmap;
+       if (verify_hdr(hdr, mmap_size) < 0)
                goto unmap;
 
-       active_nr = ntohl(hdr->hdr_entries);
-       active_alloc = alloc_nr(active_nr);
-       active_cache = xcalloc(active_alloc, sizeof(struct cache_entry *));
+       istate->cache_nr = ntohl(hdr->hdr_entries);
+       istate->cache_alloc = alloc_nr(istate->cache_nr);
+       istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
 
-       offset = sizeof(*hdr);
-       for (i = 0; i < active_nr; i++) {
-               struct cache_entry *ce = (struct cache_entry *) ((char *) cache_mmap + offset);
-               offset = offset + ce_size(ce);
-               active_cache[i] = ce;
+       /*
+        * The disk format is actually larger than the in-memory format,
+        * due to space for nsec etc, so even though the in-memory one
+        * has room for a few  more flags, we can allocate using the same
+        * index size
+        */
+       istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
+
+       src_offset = sizeof(*hdr);
+       dst_offset = 0;
+       for (i = 0; i < istate->cache_nr; i++) {
+               struct ondisk_cache_entry *disk_ce;
+               struct cache_entry *ce;
+
+               disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
+               ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
+               convert_from_disk(disk_ce, ce);
+               set_index_entry(istate, i, ce);
+
+               src_offset += ondisk_ce_size(ce);
+               dst_offset += ce_size(ce);
        }
-       index_file_timestamp = st.st_mtime;
-       while (offset <= cache_mmap_size - 20 - 8) {
+       istate->timestamp = st.st_mtime;
+       while (src_offset <= mmap_size - 20 - 8) {
                /* After an array of active_nr index entries,
                 * there can be arbitrary number of extended
                 * sections, each of which is prefixed with
@@ -830,38 +1168,47 @@ int read_cache_from(const char *path)
                 * in 4-byte network byte order.
                 */
                unsigned long extsize;
-               memcpy(&extsize, (char *) cache_mmap + offset + 4, 4);
+               memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
                extsize = ntohl(extsize);
-               if (read_index_extension(((const char *) cache_mmap) + offset,
-                                        (char *) cache_mmap + offset + 8,
+               if (read_index_extension(istate,
+                                        (const char *) mmap + src_offset,
+                                        (char *) mmap + src_offset + 8,
                                         extsize) < 0)
                        goto unmap;
-               offset += 8;
-               offset += extsize;
+               src_offset += 8;
+               src_offset += extsize;
        }
-       return active_nr;
+       munmap(mmap, mmap_size);
+       return istate->cache_nr;
 
 unmap:
-       munmap(cache_mmap, cache_mmap_size);
+       munmap(mmap, mmap_size);
        errno = EINVAL;
        die("index file corrupt");
 }
 
-int discard_cache(void)
+int discard_index(struct index_state *istate)
 {
-       int ret;
-
-       active_nr = active_cache_changed = 0;
-       index_file_timestamp = 0;
-       cache_tree_free(&active_cache_tree);
-       if (cache_mmap == NULL)
-               return 0;
-       ret = munmap(cache_mmap, cache_mmap_size);
-       cache_mmap = NULL;
-       cache_mmap_size = 0;
+       istate->cache_nr = 0;
+       istate->cache_changed = 0;
+       istate->timestamp = 0;
+       free_hash(&istate->name_hash);
+       cache_tree_free(&(istate->cache_tree));
+       free(istate->alloc);
+       istate->alloc = NULL;
 
        /* no need to throw away allocated active_cache */
-       return ret;
+       return 0;
+}
+
+int unmerged_index(const struct index_state *istate)
+{
+       int i;
+       for (i = 0; i < istate->cache_nr; i++) {
+               if (ce_stage(istate->cache[i]))
+                       return 1;
+       }
+       return 0;
 }
 
 #define WRITE_BUFFER_SIZE 8192
@@ -898,8 +1245,8 @@ static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
                write_buffer_len = buffered;
                len -= partial;
                data = (char *) data + partial;
-       }
-       return 0;
+       }
+       return 0;
 }
 
 static int write_index_ext_header(SHA_CTX *context, int fd,
@@ -965,7 +1312,7 @@ static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
                 * size to zero here, then the object name recorded
                 * in index is the 6-byte file but the cached stat information
                 * becomes zero --- which would then match what we would
-                * obtain from the filesystem next time we stat("frotz"). 
+                * obtain from the filesystem next time we stat("frotz").
                 *
                 * However, the second update-index, before calling
                 * this function, notices that the cached size is 6
@@ -973,18 +1320,42 @@ static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
                 * file, and never calls us, so the cached size information
                 * for "frotz" stays 6 which does not match the filesystem.
                 */
-               ce->ce_size = htonl(0);
+               ce->ce_size = 0;
        }
 }
 
-int write_cache(int newfd, struct cache_entry **cache, int entries)
+static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
+{
+       int size = ondisk_ce_size(ce);
+       struct ondisk_cache_entry *ondisk = xcalloc(1, size);
+
+       ondisk->ctime.sec = htonl(ce->ce_ctime);
+       ondisk->ctime.nsec = 0;
+       ondisk->mtime.sec = htonl(ce->ce_mtime);
+       ondisk->mtime.nsec = 0;
+       ondisk->dev  = htonl(ce->ce_dev);
+       ondisk->ino  = htonl(ce->ce_ino);
+       ondisk->mode = htonl(ce->ce_mode);
+       ondisk->uid  = htonl(ce->ce_uid);
+       ondisk->gid  = htonl(ce->ce_gid);
+       ondisk->size = htonl(ce->ce_size);
+       hashcpy(ondisk->sha1, ce->sha1);
+       ondisk->flags = htons(ce->ce_flags);
+       memcpy(ondisk->name, ce->name, ce_namelen(ce));
+
+       return ce_write(c, fd, ondisk, size);
+}
+
+int write_index(const struct index_state *istate, int newfd)
 {
        SHA_CTX c;
        struct cache_header hdr;
-       int i, removed;
+       int i, err, removed;
+       struct cache_entry **cache = istate->cache;
+       int entries = istate->cache_nr;
 
        for (i = removed = 0; i < entries; i++)
-               if (!cache[i]->ce_mode)
+               if (cache[i]->ce_flags & CE_REMOVE)
                        removed++;
 
        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
@@ -997,27 +1368,25 @@ int write_cache(int newfd, struct cache_entry **cache, int entries)
 
        for (i = 0; i < entries; i++) {
                struct cache_entry *ce = cache[i];
-               if (!ce->ce_mode)
+               if (ce->ce_flags & CE_REMOVE)
                        continue;
-               if (index_file_timestamp &&
-                   index_file_timestamp <= ntohl(ce->ce_mtime.sec))
+               if (is_racy_timestamp(istate, ce))
                        ce_smudge_racily_clean_entry(ce);
-               if (ce_write(&c, newfd, ce, ce_size(ce)) < 0)
+               if (ce_write_entry(&c, newfd, ce) < 0)
                        return -1;
        }
 
        /* Write extension data here */
-       if (active_cache_tree) {
-               unsigned long sz;
-               void *data = cache_tree_write(active_cache_tree, &sz);
-               if (data &&
-                   !write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sz) &&
-                   !ce_write(&c, newfd, data, sz))
-                       free(data);
-               else {
-                       free(data);
+       if (istate->cache_tree) {
+               struct strbuf sb;
+
+               strbuf_init(&sb, 0);
+               cache_tree_write(&sb, istate->cache_tree);
+               err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
+                       || ce_write(&c, newfd, sb.buf, sb.len) < 0;
+               strbuf_release(&sb);
+               if (err)
                        return -1;
-               }
        }
        return ce_flush(&c, newfd);
 }