[PATCH] Prevent git-rev-list without --merge-order producing duplicates in output
[gitweb.git] / update-cache.c
index 0bcd5d408191333e70f54fbf28676a7e3696ced5..d2f79fa4535d3ec718b60a13a9a8c67d74323d8e 100644 (file)
  */
 #include "cache.h"
 
-static int cache_name_compare(const char *name1, int len1, const char *name2, int len2)
-{
-       int len = len1 < len2 ? len1 : len2;
-       int cmp;
+/*
+ * Default to not allowing changes to the list of files. The
+ * tool doesn't actually care, but this makes it harder to add
+ * files to the revision control by mistake by doing something
+ * like "git-update-cache *" and suddenly having all the object
+ * files be revision controlled.
+ */
+static int allow_add = 0, allow_remove = 0, allow_replace = 0, not_new = 0;
+static int force_remove;
 
-       cmp = memcmp(name1, name2, len);
-       if (cmp)
-               return cmp;
-       if (len1 < len2)
-               return -1;
-       if (len1 > len2)
-               return 1;
-       return 0;
+/* Three functions to allow overloaded pointer return; see linux/err.h */
+static inline void *ERR_PTR(long error)
+{
+       return (void *) error;
 }
 
-static int cache_name_pos(const char *name, int namelen)
+static inline long PTR_ERR(const void *ptr)
 {
-       int first, last;
-
-       first = 0;
-       last = active_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, ce->namelen);
-               if (!cmp)
-                       return -next-1;
-               if (cmp < 0) {
-                       last = next;
-                       continue;
-               }
-               first = next+1;
-       }
-       return first;
+       return (long) ptr;
 }
 
-static int remove_file_from_cache(char *path)
+static inline long IS_ERR(const void *ptr)
 {
-       int pos = cache_name_pos(path, strlen(path));
-       if (pos < 0) {
-               pos = -pos-1;
-               active_nr--;
-               if (pos < active_nr)
-                       memmove(active_cache + pos, active_cache + pos + 1, (active_nr - pos - 1) * sizeof(struct cache_entry *));
-       }
+       return (unsigned long)ptr > (unsigned long)-1000L;
 }
 
-static int add_cache_entry(struct cache_entry *ce)
+static int add_file_to_cache(char *path)
 {
-       int pos;
-
-       pos = cache_name_pos(ce->name, ce->namelen);
+       int size, namelen, option, status;
+       struct cache_entry *ce;
+       struct stat st;
+       int fd;
+       char *target;
 
-       /* existing match? Just replace it */
-       if (pos < 0) {
-               active_cache[-pos-1] = ce;
-               return 0;
+       status = lstat(path, &st);
+       if (status < 0 || S_ISDIR(st.st_mode)) {
+               /* When we used to have "path" and now we want to add
+                * "path/file", we need a way to remove "path" before
+                * being able to add "path/file".  However,
+                * "git-update-cache --remove path" would not work.
+                * --force-remove can be used but this is more user
+                * friendly, especially since we can do the opposite
+                * case just fine without --force-remove.
+                */
+               if (status == 0 || (errno == ENOENT || errno == ENOTDIR)) {
+                       if (allow_remove)
+                               return remove_file_from_cache(path);
+               }
+               return error("open(\"%s\"): %s", path, strerror(errno));
        }
-
-       /* Make sure the array is big enough .. */
-       if (active_nr == active_alloc) {
-               active_alloc = alloc_nr(active_alloc);
-               active_cache = realloc(active_cache, active_alloc * sizeof(struct cache_entry *));
+       namelen = strlen(path);
+       size = cache_entry_size(namelen);
+       ce = xmalloc(size);
+       memset(ce, 0, size);
+       memcpy(ce->name, path, namelen);
+       fill_stat_cache_info(ce, &st);
+       ce->ce_mode = create_ce_mode(st.st_mode);
+       ce->ce_flags = htons(namelen);
+       switch (st.st_mode & S_IFMT) {
+       case S_IFREG:
+               fd = open(path, O_RDONLY);
+               if (fd < 0)
+                       return -1;
+               if (index_fd(ce->sha1, fd, &st) < 0)
+                       return -1;
+               break;
+       case S_IFLNK:
+               target = xmalloc(st.st_size+1);
+               if (readlink(path, target, st.st_size+1) != st.st_size) {
+                       free(target);
+                       return -1;
+               }
+               if (write_sha1_file(target, st.st_size, "blob", ce->sha1))
+                       return -1;
+               free(target);
+               break;
+       default:
+               return -1;
        }
+       option = allow_add ? ADD_CACHE_OK_TO_ADD : 0;
+       option |= allow_replace ? ADD_CACHE_OK_TO_REPLACE : 0;
+       return add_cache_entry(ce, option);
+}
 
-       /* 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;
+static int match_data(int fd, void *buffer, unsigned long size)
+{
+       while (size) {
+               char compare[1024];
+               int ret = read(fd, compare, sizeof(compare));
+
+               if (ret <= 0 || ret > size || memcmp(buffer, compare, ret))
+                       return -1;
+               size -= ret;
+               buffer += ret;
+       }
        return 0;
 }
 
-static int index_fd(const char *path, int namelen, struct cache_entry *ce, int fd, struct stat *st)
+static int compare_data(struct cache_entry *ce, unsigned long expected_size)
 {
-       z_stream stream;
-       int max_out_bytes = namelen + st->st_size + 200;
-       void *out = malloc(max_out_bytes);
-       void *metadata = malloc(namelen + 200);
-       void *in = mmap(NULL, st->st_size, PROT_READ, MAP_PRIVATE, fd, 0);
-       SHA_CTX c;
-
-       close(fd);
-       if (!out || (int)(long)in == -1)
-               return -1;
-
-       memset(&stream, 0, sizeof(stream));
-       deflateInit(&stream, Z_BEST_COMPRESSION);
+       int match = -1;
+       int fd = open(ce->name, O_RDONLY);
 
-       /*
-        * ASCII size + nul byte
-        */     
-       stream.next_in = metadata;
-       stream.avail_in = 1+sprintf(metadata, "blob %lu", (unsigned long) st->st_size);
-       stream.next_out = out;
-       stream.avail_out = max_out_bytes;
-       while (deflate(&stream, 0) == Z_OK)
-               /* nothing */;
+       if (fd >= 0) {
+               void *buffer;
+               unsigned long size;
+               char type[20];
 
-       /*
-        * File content
-        */
-       stream.next_in = in;
-       stream.avail_in = st->st_size;
-       while (deflate(&stream, Z_FINISH) == Z_OK)
-               /*nothing */;
-
-       deflateEnd(&stream);
-       
-       SHA1_Init(&c);
-       SHA1_Update(&c, out, stream.total_out);
-       SHA1_Final(ce->sha1, &c);
-
-       return write_sha1_buffer(ce->sha1, out, stream.total_out);
+               buffer = read_sha1_file(ce->sha1, type, &size);
+               if (buffer) {
+                       if (size == expected_size && !strcmp(type, "blob"))
+                               match = match_data(fd, buffer, size);
+                       free(buffer);
+               }
+               close(fd);
+       }
+       return match;
 }
 
-static int add_file_to_cache(char *path)
+static int compare_link(struct cache_entry *ce, unsigned long expected_size)
 {
-       int size, namelen;
-       struct cache_entry *ce;
-       struct stat st;
-       int fd;
+       int match = -1;
+       char *target;
+       void *buffer;
+       unsigned long size;
+       char type[10];
+       int len;
 
-       fd = open(path, O_RDONLY);
-       if (fd < 0) {
-               if (errno == ENOENT)
-                       return remove_file_from_cache(path);
+       target = xmalloc(expected_size);
+       len = readlink(ce->name, target, expected_size);
+       if (len != expected_size) {
+               free(target);
                return -1;
        }
-       if (fstat(fd, &st) < 0) {
-               close(fd);
+       buffer = read_sha1_file(ce->sha1, type, &size);
+       if (!buffer) {
+               free(target);
                return -1;
        }
-       namelen = strlen(path);
-       size = cache_entry_size(namelen);
-       ce = malloc(size);
-       memset(ce, 0, size);
-       memcpy(ce->name, path, namelen);
-       ce->ctime.sec = st.st_ctime;
-       ce->ctime.nsec = st.st_ctim.tv_nsec;
-       ce->mtime.sec = st.st_mtime;
-       ce->mtime.nsec = st.st_mtim.tv_nsec;
-       ce->st_dev = st.st_dev;
-       ce->st_ino = st.st_ino;
-       ce->st_mode = st.st_mode;
-       ce->st_uid = st.st_uid;
-       ce->st_gid = st.st_gid;
-       ce->st_size = st.st_size;
-       ce->namelen = namelen;
-
-       if (index_fd(path, namelen, ce, fd, &st) < 0)
-               return -1;
-
-       return add_cache_entry(ce);
+       if (size == expected_size)
+               match = memcmp(buffer, target, size);
+       free(buffer);
+       free(target);
+       return match;
 }
 
-static int write_cache(int newfd, struct cache_entry **cache, int entries)
+/*
+ * "refresh" does not calculate a new sha1 file or bring the
+ * cache up-to-date for mode/content changes. But what it
+ * _does_ do is to "re-match" the stat information of a file
+ * with the cache, so that you can refresh the cache for a
+ * file that hasn't been changed but where the stat entry is
+ * out of date.
+ *
+ * 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)
 {
-       SHA_CTX c;
-       struct cache_header hdr;
-       int i;
+       struct stat st;
+       struct cache_entry *updated;
+       int changed, size;
+
+       if (lstat(ce->name, &st) < 0)
+               return ERR_PTR(-errno);
 
-       hdr.signature = CACHE_SIGNATURE;
-       hdr.version = 1;
-       hdr.entries = entries;
+       changed = ce_match_stat(ce, &st);
+       if (!changed)
+               return ce;
 
-       SHA1_Init(&c);
-       SHA1_Update(&c, &hdr, offsetof(struct cache_header, sha1));
-       for (i = 0; i < entries; i++) {
-               struct cache_entry *ce = cache[i];
-               int size = ce_size(ce);
-               SHA1_Update(&c, ce, size);
+       /*
+        * If the mode or type has changed, there's no point in trying
+        * to refresh the entry - it's not going to match
+        */
+       if (changed & (MODE_CHANGED | TYPE_CHANGED))
+               return ERR_PTR(-EINVAL);
+
+       switch (st.st_mode & S_IFMT) {
+       case S_IFREG:
+               if (compare_data(ce, st.st_size))
+                       return ERR_PTR(-EINVAL);
+               break;
+       case S_IFLNK:
+               if (compare_link(ce, st.st_size))
+                       return ERR_PTR(-EINVAL);
+               break;
+       default:
+               return ERR_PTR(-EINVAL);
        }
-       SHA1_Final(hdr.sha1, &c);
 
-       if (write(newfd, &hdr, sizeof(hdr)) != sizeof(hdr))
-               return -1;
+       size = ce_size(ce);
+       updated = xmalloc(size);
+       memcpy(updated, ce, size);
+       fill_stat_cache_info(updated, &st);
+       return updated;
+}
 
-       for (i = 0; i < entries; i++) {
-               struct cache_entry *ce = cache[i];
-               int size = ce_size(ce);
-               if (write(newfd, ce, size) != size)
-                       return -1;
+static int refresh_cache(void)
+{
+       int i;
+       int has_errors = 0;
+
+       for (i = 0; i < active_nr; i++) {
+               struct cache_entry *ce, *new;
+               ce = active_cache[i];
+               if (ce_stage(ce)) {
+                       printf("%s: needs merge\n", ce->name);
+                       has_errors = 1;
+                       while ((i < active_nr) &&
+                              ! strcmp(active_cache[i]->name, ce->name))
+                               i++;
+                       i--;
+                       continue;
+               }
+
+               new = refresh_entry(ce);
+               if (IS_ERR(new)) {
+                       if (!(not_new && PTR_ERR(new) == -ENOENT)) {
+                               printf("%s: needs update\n", ce->name);
+                               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;
        }
-       return 0;
-}              
+       return has_errors;
+}
 
 /*
  * We fundamentally don't like some paths: we don't want
- * dot or dot-dot anywhere, and in fact, we don't even want
- * any other dot-files (.dircache or anything else). They
- * are hidden, for chist sake.
+ * dot or dot-dot anywhere, and for obvious reasons don't
+ * want to recurse into ".git" either.
  *
  * Also, we don't want double slashes or slashes at the
- * end that can make pathnames ambiguous. 
+ * end that can make pathnames ambiguous.
  */
+static int verify_dotfile(const char *rest)
+{
+       /*
+        * The first character was '.', but that
+        * has already been discarded, we now test
+        * the rest.
+        */
+       switch (*rest) {
+       /* "." is not allowed */
+       case '\0': case '/':
+               return 0;
+
+       /*
+        * ".git" followed by  NUL or slash is bad. This
+        * shares the path end test with the ".." case.
+        */
+       case 'g':
+               if (rest[1] != 'i')
+                       break;
+               if (rest[2] != 't')
+                       break;
+               rest += 2;
+       /* fallthrough */
+       case '.':
+               if (rest[1] == '\0' || rest[1] == '/')
+                       return 0;
+       }
+       return 1;
+}
+
 static int verify_path(char *path)
 {
        char c;
@@ -212,42 +286,122 @@ static int verify_path(char *path)
                if (c == '/') {
 inside:
                        c = *path++;
-                       if (c != '/' && c != '.' && c != '\0')
+                       switch (c) {
+                       default:
                                continue;
+                       case '/': case '\0':
+                               break;
+                       case '.':
+                               if (verify_dotfile(path))
+                                       continue;
+                       }
                        return 0;
                }
                c = *path++;
        }
 }
 
+static int add_cacheinfo(char *arg1, char *arg2, char *arg3)
+{
+       int size, len, option;
+       unsigned int mode;
+       unsigned char sha1[20];
+       struct cache_entry *ce;
+
+       if (sscanf(arg1, "%o", &mode) != 1)
+               return -1;
+       if (get_sha1_hex(arg2, sha1))
+               return -1;
+       if (!verify_path(arg3))
+               return -1;
+
+       len = strlen(arg3);
+       size = cache_entry_size(len);
+       ce = xmalloc(size);
+       memset(ce, 0, size);
+
+       memcpy(ce->sha1, sha1, 20);
+       memcpy(ce->name, arg3, len);
+       ce->ce_flags = htons(len);
+       ce->ce_mode = create_ce_mode(mode);
+       option = allow_add ? ADD_CACHE_OK_TO_ADD : 0;
+       option |= allow_replace ? ADD_CACHE_OK_TO_REPLACE : 0;
+       return add_cache_entry(ce, option);
+}
+
+static struct cache_file cache_file;
+
 int main(int argc, char **argv)
 {
-       int i, newfd, entries;
+       int i, newfd, entries, has_errors = 0;
+       int allow_options = 1;
+
+       newfd = hold_index_file_for_update(&cache_file, get_index_file());
+       if (newfd < 0)
+               die("unable to create new cachefile");
 
        entries = read_cache();
-       if (entries < 0) {
-               perror("cache corrupted");
-               return -1;
-       }
+       if (entries < 0)
+               die("cache corrupted");
 
-       newfd = open(".dircache/index.lock", O_RDWR | O_CREAT | O_EXCL, 0600);
-       if (newfd < 0) {
-               perror("unable to create new cachefile");
-               return -1;
-       }
        for (i = 1 ; i < argc; i++) {
                char *path = argv[i];
+
+               if (allow_options && *path == '-') {
+                       if (!strcmp(path, "--")) {
+                               allow_options = 0;
+                               continue;
+                       }
+                       if (!strcmp(path, "--add")) {
+                               allow_add = 1;
+                               continue;
+                       }
+                       if (!strcmp(path, "--replace")) {
+                               allow_replace = 1;
+                               continue;
+                       }
+                       if (!strcmp(path, "--remove")) {
+                               allow_remove = 1;
+                               continue;
+                       }
+                       if (!strcmp(path, "--refresh")) {
+                               has_errors |= refresh_cache();
+                               continue;
+                       }
+                       if (!strcmp(path, "--cacheinfo")) {
+                               if (i+3 >= argc)
+                                       die("git-update-cache: --cacheinfo <mode> <sha1> <path>");
+                               if (add_cacheinfo(argv[i+1], argv[i+2], argv[i+3]))
+                                       die("git-update-cache: --cacheinfo cannot add %s", argv[i+3]);
+                               i += 3;
+                               continue;
+                       }
+                       if (!strcmp(path, "--force-remove")) {
+                               force_remove = 1;
+                               continue;
+                       }
+
+                       if (!strcmp(path, "--ignore-missing")) {
+                               not_new = 1;
+                               continue;
+                       }
+                       die("unknown option %s", path);
+               }
                if (!verify_path(path)) {
                        fprintf(stderr, "Ignoring path %s\n", argv[i]);
                        continue;
                }
-               if (add_file_to_cache(path)) {
-                       fprintf(stderr, "Unable to add %s to database\n", path);
-                       goto out;
+               if (force_remove) {
+                       if (remove_file_from_cache(path))
+                               die("git-update-cache: --force-remove cannot remove %s", path);
+                       continue;
                }
+               if (add_file_to_cache(path))
+                       die("Unable to add %s to database", path);
        }
-       if (!write_cache(newfd, active_cache, active_nr) && !rename(".dircache/index.lock", ".dircache/index"))
-               return 0;
-out:
-       unlink(".dircache/index.lock");
+       if (write_cache(newfd, active_cache, active_nr) ||
+           commit_index_file(&cache_file))
+               die("Unable to write new cachefile");
+
+       return has_errors ? 1 : 0;
 }