[PATCH] Split external diff command interface to a separate file.
[gitweb.git] / update-cache.c
index 413e09d48fa0aad26ef4863a47e1c75c08844d81..02c213cfb65a72b3768a5d5d8eb96904d4e74ef6 100644 (file)
@@ -5,93 +5,56 @@
  */
 #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;
-
-       cmp = memcmp(name1, name2, len);
-       if (cmp)
-               return cmp;
-       if (len1 < len2)
-               return -1;
-       if (len1 > len2)
-               return 1;
-       return 0;
-}
+/*
+ * 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 "update-cache *" and suddenly having all the object
+ * files be revision controlled.
+ */
+static int allow_add = 0, allow_remove = 0, not_new = 0;
 
-static int cache_name_pos(const char *name, int namelen)
+/* Three functions to allow overloaded pointer return; see linux/err.h */
+static inline void *ERR_PTR(long error)
 {
-       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 (void *) error;
 }
 
-static int remove_file_from_cache(char *path)
+static inline long PTR_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 0;
+       return (long) ptr;
 }
 
-static int add_cache_entry(struct cache_entry *ce)
+static inline long IS_ERR(const void *ptr)
 {
-       int pos;
-
-       pos = cache_name_pos(ce->name, ce->namelen);
-
-       /* existing match? Just replace it */
-       if (pos < 0) {
-               active_cache[-pos-1] = ce;
-               return 0;
-       }
-
-       /* 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 *));
-       }
-
-       /* 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;
-       return 0;
+       return (unsigned long)ptr > (unsigned long)-1000L;
 }
 
-static int index_fd(const char *path, int namelen, struct cache_entry *ce, int fd, struct stat *st)
+static int index_fd(unsigned char *sha1, int fd, struct stat *st)
 {
        z_stream stream;
-       int max_out_bytes = namelen + st->st_size + 200;
+       unsigned long size = st->st_size;
+       int max_out_bytes = 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);
+       void *metadata = malloc(200);
+       int metadata_size;
+       void *in;
        SHA_CTX c;
 
+       in = "";
+       if (size)
+               in = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
        close(fd);
        if (!out || (int)(long)in == -1)
                return -1;
 
+       metadata_size = 1+sprintf(metadata, "blob %lu", size);
+
+       SHA1_Init(&c);
+       SHA1_Update(&c, metadata, metadata_size);
+       SHA1_Update(&c, in, size);
+       SHA1_Final(sha1, &c);
+
        memset(&stream, 0, sizeof(stream));
        deflateInit(&stream, Z_BEST_COMPRESSION);
 
@@ -99,7 +62,7 @@ static int index_fd(const char *path, int namelen, struct cache_entry *ce, int f
         * ASCII size + nul byte
         */     
        stream.next_in = metadata;
-       stream.avail_in = 1+sprintf(metadata, "blob %lu", (unsigned long) st->st_size);
+       stream.avail_in = metadata_size;
        stream.next_out = out;
        stream.avail_out = max_out_bytes;
        while (deflate(&stream, 0) == Z_OK)
@@ -109,17 +72,33 @@ static int index_fd(const char *path, int namelen, struct cache_entry *ce, int f
         * File content
         */
        stream.next_in = in;
-       stream.avail_in = st->st_size;
+       stream.avail_in = 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(sha1, out, stream.total_out);
+}
 
-       return write_sha1_buffer(ce->sha1, out, stream.total_out);
+/*
+ * This only updates the "non-critical" parts of the directory
+ * cache, ie the parts that aren't tracked by GIT, and only used
+ * to validate the cache.
+ */
+static 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 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);
 }
 
 static int add_file_to_cache(char *path)
@@ -131,8 +110,10 @@ static int add_file_to_cache(char *path)
 
        fd = open(path, O_RDONLY);
        if (fd < 0) {
-               if (errno == ENOENT)
-                       return remove_file_from_cache(path);
+               if (errno == ENOENT) {
+                       if (allow_remove)
+                               return remove_file_from_cache(path);
+               }
                return -1;
        }
        if (fstat(fd, &st) < 0) {
@@ -144,63 +125,126 @@ static int add_file_to_cache(char *path)
        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)
+       fill_stat_cache_info(ce, &st);
+       ce->ce_mode = create_ce_mode(st.st_mode);
+       ce->ce_flags = htons(namelen);
+
+       if (index_fd(ce->sha1, fd, &st) < 0)
                return -1;
 
-       return add_cache_entry(ce);
+       return add_cache_entry(ce, allow_add);
 }
 
-static int write_cache(int newfd, struct cache_entry **cache, int entries)
+static int match_data(int fd, void *buffer, unsigned long size)
 {
-       SHA_CTX c;
-       struct cache_header hdr;
-       int i;
+       while (size) {
+               char compare[1024];
+               int ret = read(fd, compare, sizeof(compare));
 
-       hdr.signature = CACHE_SIGNATURE;
-       hdr.version = 1;
-       hdr.entries = entries;
+               if (ret <= 0 || ret > size || memcmp(buffer, compare, ret))
+                       return -1;
+               size -= ret;
+               buffer += ret;
+       }
+       return 0;
+}
 
-       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);
+static int compare_data(struct cache_entry *ce, unsigned long expected_size)
+{
+       int match = -1;
+       int fd = open(ce->name, O_RDONLY);
+
+       if (fd >= 0) {
+               void *buffer;
+               unsigned long size;
+               char type[10];
+
+               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);
        }
-       SHA1_Final(hdr.sha1, &c);
+       return match;
+}
 
-       if (write(newfd, &hdr, sizeof(hdr)) != sizeof(hdr))
-               return -1;
+/*
+ * "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 "read-tree",
+ * to link up the stat cache details with the proper files.
+ */
+static struct cache_entry *refresh_entry(struct cache_entry *ce)
+{
+       struct stat st;
+       struct cache_entry *updated;
+       int changed, size;
 
-       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;
+       if (stat(ce->name, &st) < 0)
+               return ERR_PTR(-errno);
+
+       changed = cache_match_stat(ce, &st);
+       if (!changed)
+               return ce;
+
+       /*
+        * If the mode has changed, there's no point in trying
+        * to refresh the entry - it's not going to match
+        */
+       if (changed & MODE_CHANGED)
+               return ERR_PTR(-EINVAL);
+
+       if (compare_data(ce, st.st_size))
+               return ERR_PTR(-EINVAL);
+
+       size = ce_size(ce);
+       updated = malloc(size);
+       memcpy(updated, ce, size);
+       fill_stat_cache_info(updated, &st);
+       return updated;
+}
+
+static void refresh_cache(void)
+{
+       int i;
+
+       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);
+                       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);
+                       continue;
+               }
+               active_cache[i] = new;
        }
-       return 0;
-}              
+}
 
 /*
  * 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
+ * any other dot-files (.git or anything else). They
  * are hidden, for chist sake.
  *
  * 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_path(char *path)
 {
@@ -221,35 +265,102 @@ static int verify_path(char *path)
        }
 }
 
+static int add_cacheinfo(char *arg1, char *arg2, char *arg3)
+{
+       int size, len;
+       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 = malloc(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);
+       return add_cache_entry(ce, allow_add);
+}
+
+static const char *lockfile_name = NULL;
+
+static void remove_lock_file(void)
+{
+       if (lockfile_name)
+               unlink(lockfile_name);
+}
+
 int main(int argc, char **argv)
 {
        int i, newfd, entries;
+       int allow_options = 1;
+       static char lockfile[MAXPATHLEN+1];
+       const char *indexfile = get_index_file();
+
+       snprintf(lockfile, sizeof(lockfile), "%s.lock", indexfile);
+
+       newfd = open(lockfile, O_RDWR | O_CREAT | O_EXCL, 0600);
+       if (newfd < 0)
+               die("unable to create new cachefile");
+
+       atexit(remove_lock_file);
+       lockfile_name = lockfile;
 
        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, "--remove")) {
+                               allow_remove = 1;
+                               continue;
+                       }
+                       if (!strcmp(path, "--refresh")) {
+                               refresh_cache();
+                               continue;
+                       }
+                       if (!strcmp(path, "--cacheinfo")) {
+                               if (i+3 >= argc || add_cacheinfo(argv[i+1], argv[i+2], argv[i+3]))
+                                       die("update-cache: --cacheinfo <mode> <sha1> <path>");
+                               i += 3;
+                               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 (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) || rename(lockfile, indexfile))
+               die("Unable to write new cachefile");
+
+       lockfile_name = NULL;
        return 0;
 }