Teach "git-read-tree -u" to check out submodules as a directory
[gitweb.git] / read-cache.c
index 3c32aae7e8ae63bccad83a603f2b86c66965cd26..d2f332a6222cc3543bde8fa661c54c5c904cf561 100644 (file)
@@ -5,6 +5,7 @@
  */
 #include "cache.h"
 #include "cache-tree.h"
+#include "refs.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 = NULL;
+struct cache_entry **active_cache;
 static time_t index_file_timestamp;
-unsigned int active_nr = 0, active_alloc = 0, active_cache_changed = 0;
+unsigned int active_nr, active_alloc, active_cache_changed;
 
-struct cache_tree *active_cache_tree = NULL;
+struct cache_tree *active_cache_tree;
+
+static void *cache_mmap;
+static size_t cache_mmap_size;
 
 /*
  * This only updates the "non-critical" parts of the directory
@@ -54,20 +58,20 @@ static int ce_compare_data(struct cache_entry *ce, struct stat *st)
 
        if (fd >= 0) {
                unsigned char sha1[20];
-               if (!index_fd(sha1, fd, st, 0, NULL))
-                       match = memcmp(sha1, ce->sha1, 20);
-               close(fd);
+               if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
+                       match = hashcmp(sha1, ce->sha1);
+               /* index_fd() closed the file descriptor already */
        }
        return match;
 }
 
-static int ce_compare_link(struct cache_entry *ce, unsigned long expected_size)
+static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
 {
        int match = -1;
        char *target;
        void *buffer;
        unsigned long size;
-       char type[10];
+       enum object_type type;
        int len;
 
        target = xmalloc(expected_size);
@@ -76,7 +80,7 @@ static int ce_compare_link(struct cache_entry *ce, unsigned long expected_size)
                free(target);
                return -1;
        }
-       buffer = read_sha1_file(ce->sha1, type, &size);
+       buffer = read_sha1_file(ce->sha1, &type, &size);
        if (!buffer) {
                free(target);
                return -1;
@@ -88,6 +92,23 @@ static int ce_compare_link(struct cache_entry *ce, unsigned long 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 DIRLNK 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) {
@@ -96,9 +117,12 @@ static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
                        return DATA_CHANGED;
                break;
        case S_IFLNK:
-               if (ce_compare_link(ce, st->st_size))
+               if (ce_compare_link(ce, xsize_t(st->st_size)))
                        return DATA_CHANGED;
                break;
+       case S_IFDIR:
+               if (S_ISDIRLNK(ntohl(ce->ce_mode)))
+                       return 0;
        default:
                return TYPE_CHANGED;
        }
@@ -120,8 +144,16 @@ static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
                        changed |= MODE_CHANGED;
                break;
        case S_IFLNK:
-               changed |= !S_ISLNK(st->st_mode) ? TYPE_CHANGED : 0;
+               if (!S_ISLNK(st->st_mode) &&
+                   (has_symlinks || !S_ISREG(st->st_mode)))
+                       changed |= TYPE_CHANGED;
                break;
+       case S_IFDIRLNK:
+               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));
        }
@@ -164,9 +196,11 @@ static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
        return changed;
 }
 
-int ce_match_stat(struct cache_entry *ce, struct stat *st, int ignore_valid)
+int ce_match_stat(struct cache_entry *ce, struct stat *st, int options)
 {
        unsigned int changed;
+       int ignore_valid = options & 01;
+       int assume_racy_is_modified = options & 02;
 
        /*
         * If it's marked as always valid in the index, it's
@@ -195,8 +229,12 @@ int ce_match_stat(struct cache_entry *ce, struct stat *st, int ignore_valid)
         */
        if (!changed &&
            index_file_timestamp &&
-           index_file_timestamp <= ntohl(ce->ce_mtime.sec))
-               changed |= ce_modified_check_fs(ce, st);
+           index_file_timestamp <= ntohl(ce->ce_mtime.sec)) {
+               if (assume_racy_is_modified)
+                       changed |= DATA_CHANGED;
+               else
+                       changed |= ce_modified_check_fs(ce, st);
+       }
 
        return changed;
 }
@@ -314,6 +352,52 @@ int remove_file_from_cache(const char *path)
        return 0;
 }
 
+int add_file_to_cache(const char *path, int verbose)
+{
+       int size, namelen;
+       struct stat st;
+       struct cache_entry *ce;
+
+       if (lstat(path, &st))
+               die("%s: unable to stat (%s)", path, strerror(errno));
+
+       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);
+       fill_stat_cache_info(ce, &st);
+
+       if (trust_executable_bit && has_symlinks)
+               ce->ce_mode = create_ce_mode(st.st_mode);
+       else {
+               /* If there is an existing entry, pick the mode bits and type
+                * from it, otherwise assume unexecutable regular file.
+                */
+               struct cache_entry *ent;
+               int pos = cache_name_pos(path, namelen);
+
+               ent = (0 <= pos) ? active_cache[pos] : NULL;
+               ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
+       }
+
+       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))
+               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;
+}
+
 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 {
        int len = ce_namelen(a);
@@ -432,6 +516,8 @@ static int has_file_name(const struct cache_entry *ce, int pos, int ok_to_replac
                        continue;
                if (p->name[len] != '/')
                        continue;
+               if (!ce_stage(p) && !p->ce_mode)
+                       continue;
                retval = -1;
                if (!ok_to_replace)
                        break;
@@ -464,26 +550,37 @@ static int has_dir_name(const struct cache_entry *ce, int pos, int ok_to_replace
 
                pos = cache_name_pos(name, ntohs(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 (stage || active_cache[pos]->ce_mode) {
+                               retval = -1;
+                               if (!ok_to_replace)
+                                       break;
+                               remove_cache_entry_at(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];
                        if ((ce_namelen(p) <= len) ||
                            (p->name[len] != '/') ||
                            memcmp(p->name, name, len))
                                break; /* not our subdirectory */
-                       if (ce_stage(p) == stage)
+                       if (ce_stage(p) == stage && (stage || p->ce_mode))
                                /* 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
@@ -507,12 +604,21 @@ static int has_dir_name(const struct cache_entry *ce, int pos, int ok_to_replace
  */
 static int check_file_directory_conflict(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_stage(ce) && !ce->ce_mode)
+               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(ce, pos, ok_to_replace);
+
        /*
         * Then check if the path might have a clashing sub-directory
         * before it.
@@ -557,7 +663,7 @@ int add_cache_entry(struct cache_entry *ce, int option)
        if (!skip_df_check &&
            check_file_directory_conflict(ce, pos, ok_to_replace)) {
                if (!ok_to_replace)
-                       return -1;
+                       return error("'%s' appears as both a file and as a directory", ce->name);
                pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
                pos = -pos-1;
        }
@@ -577,22 +683,6 @@ int add_cache_entry(struct cache_entry *ce, int option)
        return 0;
 }
 
-/* Three functions to allow overloaded pointer return; see linux/err.h */
-static inline void *ERR_PTR(long error)
-{
-       return (void *) error;
-}
-
-static inline long PTR_ERR(const void *ptr)
-{
-       return (long) ptr;
-}
-
-static inline long IS_ERR(const void *ptr)
-{
-       return (unsigned long)ptr > (unsigned long)-1000L;
-}
-
 /*
  * "refresh" does not calculate a new sha1 file or bring the
  * cache up-to-date for mode/content changes. But what it
@@ -604,14 +694,17 @@ static inline long IS_ERR(const void *ptr)
  * 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.
  */
-static struct cache_entry *refresh_entry(struct cache_entry *ce, int really)
+static struct cache_entry *refresh_cache_ent(struct cache_entry *ce, int really, int *err)
 {
        struct stat st;
        struct cache_entry *updated;
        int changed, size;
 
-       if (lstat(ce->name, &st) < 0)
-               return ERR_PTR(-errno);
+       if (lstat(ce->name, &st) < 0) {
+               if (err)
+                       *err = errno;
+               return NULL;
+       }
 
        changed = ce_match_stat(ce, &st, really);
        if (!changed) {
@@ -619,11 +712,14 @@ static struct cache_entry *refresh_entry(struct cache_entry *ce, int really)
                    !(ce->ce_flags & htons(CE_VALID)))
                        ; /* mark this one VALID again */
                else
-                       return NULL;
+                       return ce;
        }
 
-       if (ce_modified(ce, &st, really))
-               return ERR_PTR(-EINVAL);
+       if (ce_modified(ce, &st, really)) {
+               if (err)
+                       *err = EINVAL;
+               return NULL;
+       }
 
        size = ce_size(ce);
        updated = xmalloc(size);
@@ -653,6 +749,8 @@ int refresh_cache(unsigned int flags)
 
        for (i = 0; i < active_nr; i++) {
                struct cache_entry *ce, *new;
+               int cache_errno = 0;
+
                ce = active_cache[i];
                if (ce_stage(ce)) {
                        while ((i < active_nr) &&
@@ -666,13 +764,13 @@ int refresh_cache(unsigned int flags)
                        continue;
                }
 
-               new = refresh_entry(ce, really);
-               if (!new)
+               new = refresh_cache_ent(ce, really, &cache_errno);
+               if (new == ce)
                        continue;
-               if (IS_ERR(new)) {
-                       if (not_new && PTR_ERR(new) == -ENOENT)
+               if (!new) {
+                       if (not_new && cache_errno == ENOENT)
                                continue;
-                       if (really && PTR_ERR(new) == -EINVAL) {
+                       if (really && cache_errno == EINVAL) {
                                /* If we are doing --really-refresh that
                                 * means the index is not valid anymore.
                                 */
@@ -694,6 +792,11 @@ int refresh_cache(unsigned int flags)
        return has_errors;
 }
 
+struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
+{
+       return refresh_cache_ent(ce, really, NULL);
+}
+
 static int verify_hdr(struct cache_header *hdr, unsigned long size)
 {
        SHA_CTX c;
@@ -706,7 +809,7 @@ static int verify_hdr(struct cache_header *hdr, unsigned long size)
        SHA1_Init(&c);
        SHA1_Update(&c, hdr, size - 20);
        SHA1_Final(sha1, &c);
-       if (memcmp(sha1, (char *) hdr + size - 20, 20))
+       if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
                return error("bad index file sha1 signature");
        return 0;
 }
@@ -728,40 +831,44 @@ static int read_index_extension(const char *ext, void *data, unsigned long sz)
 }
 
 int read_cache(void)
+{
+       return read_cache_from(get_index_file());
+}
+
+/* remember to discard_cache() before reading a different cache! */
+int read_cache_from(const char *path)
 {
        int fd, i;
        struct stat st;
-       unsigned long size, offset;
-       void *map;
+       unsigned long offset;
        struct cache_header *hdr;
 
        errno = EBUSY;
-       if (active_cache)
+       if (cache_mmap)
                return active_nr;
 
        errno = ENOENT;
        index_file_timestamp = 0;
-       fd = open(get_index_file(), O_RDONLY);
+       fd = open(path, O_RDONLY);
        if (fd < 0) {
                if (errno == ENOENT)
                        return 0;
                die("index file open failed (%s)", strerror(errno));
        }
 
-       size = 0; // avoid gcc warning
-       map = MAP_FAILED;
        if (!fstat(fd, &st)) {
-               size = st.st_size;
+               cache_mmap_size = xsize_t(st.st_size);
                errno = EINVAL;
-               if (size >= sizeof(struct cache_header) + 20)
-                       map = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
-       }
+               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
+               die("cannot stat the open index (%s)", strerror(errno));
        close(fd);
-       if (map == MAP_FAILED)
-               die("index file mmap failed (%s)", strerror(errno));
 
-       hdr = map;
-       if (verify_hdr(hdr, size) < 0)
+       hdr = cache_mmap;
+       if (verify_hdr(hdr, cache_mmap_size) < 0)
                goto unmap;
 
        active_nr = ntohl(hdr->hdr_entries);
@@ -770,12 +877,12 @@ int read_cache(void)
 
        offset = sizeof(*hdr);
        for (i = 0; i < active_nr; i++) {
-               struct cache_entry *ce = (struct cache_entry *) ((char *) map + offset);
+               struct cache_entry *ce = (struct cache_entry *) ((char *) cache_mmap + offset);
                offset = offset + ce_size(ce);
                active_cache[i] = ce;
        }
        index_file_timestamp = st.st_mtime;
-       while (offset <= size - 20 - 8) {
+       while (offset <= cache_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
@@ -783,10 +890,10 @@ int read_cache(void)
                 * in 4-byte network byte order.
                 */
                unsigned long extsize;
-               memcpy(&extsize, (char *) map + offset + 4, 4);
+               memcpy(&extsize, (char *) cache_mmap + offset + 4, 4);
                extsize = ntohl(extsize);
-               if (read_index_extension(((const char *) map) + offset,
-                                        (char *) map + offset + 8,
+               if (read_index_extension(((const char *) cache_mmap) + offset,
+                                        (char *) cache_mmap + offset + 8,
                                         extsize) < 0)
                        goto unmap;
                offset += 8;
@@ -795,15 +902,44 @@ int read_cache(void)
        return active_nr;
 
 unmap:
-       munmap(map, size);
+       munmap(cache_mmap, cache_mmap_size);
        errno = EINVAL;
        die("index file corrupt");
 }
 
+int discard_cache(void)
+{
+       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;
+
+       /* no need to throw away allocated active_cache */
+       return ret;
+}
+
 #define WRITE_BUFFER_SIZE 8192
 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
 static unsigned long write_buffer_len;
 
+static int ce_write_flush(SHA_CTX *context, int fd)
+{
+       unsigned int buffered = write_buffer_len;
+       if (buffered) {
+               SHA1_Update(context, write_buffer, buffered);
+               if (write_in_full(fd, write_buffer, buffered) != buffered)
+                       return -1;
+               write_buffer_len = 0;
+       }
+       return 0;
+}
+
 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
 {
        while (len) {
@@ -814,8 +950,8 @@ static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
                memcpy(write_buffer + buffered, data, partial);
                buffered += partial;
                if (buffered == WRITE_BUFFER_SIZE) {
-                       SHA1_Update(context, write_buffer, WRITE_BUFFER_SIZE);
-                       if (write(fd, write_buffer, WRITE_BUFFER_SIZE) != WRITE_BUFFER_SIZE)
+                       write_buffer_len = buffered;
+                       if (ce_write_flush(context, fd))
                                return -1;
                        buffered = 0;
                }
@@ -831,10 +967,8 @@ static int write_index_ext_header(SHA_CTX *context, int fd,
 {
        ext = htonl(ext);
        sz = htonl(sz);
-       if ((ce_write(context, fd, &ext, 4) < 0) ||
-           (ce_write(context, fd, &sz, 4) < 0))
-               return -1;
-       return 0;
+       return ((ce_write(context, fd, &ext, 4) < 0) ||
+               (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
 }
 
 static int ce_flush(SHA_CTX *context, int fd)
@@ -848,7 +982,7 @@ static int ce_flush(SHA_CTX *context, int fd)
 
        /* Flush first if not enough space for SHA1 signature */
        if (left + 20 > WRITE_BUFFER_SIZE) {
-               if (write(fd, write_buffer, left) != left)
+               if (write_in_full(fd, write_buffer, left) != left)
                        return -1;
                left = 0;
        }
@@ -856,9 +990,7 @@ static int ce_flush(SHA_CTX *context, int fd)
        /* Append the SHA1 signature at the end */
        SHA1_Final(write_buffer + left, context);
        left += 20;
-       if (write(fd, write_buffer, left) != left)
-               return -1;
-       return 0;
+       return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
 }
 
 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
@@ -887,7 +1019,7 @@ static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
                 * $ echo filfre >nitfol
                 * $ git-update-index --add nitfol
                 *
-                * but it does not.  Whe the second update-index runs,
+                * but it does not.  When the second update-index runs,
                 * it notices that the entry "frotz" has the same timestamp
                 * as index, and if we were to smudge it by resetting its
                 * size to zero here, then the object name recorded
@@ -941,7 +1073,7 @@ int write_cache(int newfd, struct cache_entry **cache, int entries)
                if (data &&
                    !write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sz) &&
                    !ce_write(&c, newfd, data, sz))
-                       ;
+                       free(data);
                else {
                        free(data);
                        return -1;