[PATCH 3/4] split core-git.txt and update
[gitweb.git] / fsck-cache.c
index 27c9ccdeb121bff4a071e9b621ddede099878428..a00702b79f7169ef3da7d60e9bbbe9ccf237e673 100644 (file)
@@ -1,15 +1,18 @@
-#include "cache.h"
-
 #include <sys/types.h>
 #include <dirent.h>
 
+#include "cache.h"
 #include "commit.h"
 #include "tree.h"
 #include "blob.h"
+#include "tag.h"
 
 #define REACHABLE 0x0001
 
+static int show_root = 0;
+static int show_tags = 0;
 static int show_unreachable = 0;
+static int keep_cache_objects = 0; 
 static unsigned char head_sha1[20];
 
 static void check_connectivity(void)
@@ -19,16 +22,31 @@ static void check_connectivity(void)
        /* Look up all the requirements, warn about missing objects.. */
        for (i = 0; i < nr_objs; i++) {
                struct object *obj = objs[i];
+               struct object_list *refs;
+
+               if (!obj->parsed) {
+                       printf("missing %s %s\n", obj->type, sha1_to_hex(obj->sha1));
+                       continue;
+               }
+
+               for (refs = obj->refs; refs; refs = refs->next) {
+                       if (refs->item->parsed)
+                               continue;
+                       printf("broken link from %7s %s\n",
+                              obj->type, sha1_to_hex(obj->sha1));
+                       printf("              to %7s %s\n",
+                              refs->item->type, sha1_to_hex(refs->item->sha1));
+               }
+
+               /* Don't bother with tag reachability. */
+               if (obj->type == tag_type)
+                       continue;
 
                if (show_unreachable && !(obj->flags & REACHABLE)) {
                        printf("unreachable %s %s\n", obj->type, sha1_to_hex(obj->sha1));
                        continue;
                }
 
-               if (!obj->parsed) {
-                       printf("missing %s %s\n", obj->type, 
-                              sha1_to_hex(obj->sha1));
-               }
                if (!obj->used) {
                        printf("dangling %s %s\n", obj->type, 
                               sha1_to_hex(obj->sha1));
@@ -36,118 +54,210 @@ static void check_connectivity(void)
        }
 }
 
-static int fsck_tree(unsigned char *sha1, void *data, unsigned long size)
+/*
+ * The entries in a tree are ordered in the _path_ order,
+ * which means that a directory entry is ordered by adding
+ * a slash to the end of it.
+ *
+ * So a directory called "a" is ordered _after_ a file
+ * called "a.c", because "a/" sorts after "a.c".
+ */
+#define TREE_UNORDERED (-1)
+#define TREE_HAS_DUPS  (-2)
+
+static int verify_ordered(struct tree_entry_list *a, struct tree_entry_list *b)
 {
-       struct tree *item = lookup_tree(sha1);
-       if (parse_tree(item))
-               return -1;
-       if (item->has_full_path) {
+       int len1 = strlen(a->name);
+       int len2 = strlen(b->name);
+       int len = len1 < len2 ? len1 : len2;
+       unsigned char c1, c2;
+       int cmp;
+
+       cmp = memcmp(a->name, b->name, len);
+       if (cmp < 0)
+               return 0;
+       if (cmp > 0)
+               return TREE_UNORDERED;
+
+       /*
+        * Ok, the first <len> characters are the same.
+        * Now we need to order the next one, but turn
+        * a '\0' into a '/' for a directory entry.
+        */
+       c1 = a->name[len];
+       c2 = b->name[len];
+       if (!c1 && !c2)
+               /*
+                * git-write-tree used to write out a nonsense tree that has
+                * entries with the same name, one blob and one tree.  Make
+                * sure we do not have duplicate entries.
+                */
+               return TREE_HAS_DUPS;
+       if (!c1 && a->directory)
+               c1 = '/';
+       if (!c2 && b->directory)
+               c2 = '/';
+       return c1 < c2 ? 0 : TREE_UNORDERED;
+}
+
+static int fsck_tree(struct tree *item)
+{
+       int has_full_path = 0;
+       struct tree_entry_list *entry, *last;
+
+       last = NULL;
+       for (entry = item->entries; entry; entry = entry->next) {
+               if (strchr(entry->name, '/'))
+                       has_full_path = 1;
+
+               switch (entry->mode) {
+               /*
+                * Standard modes.. 
+                */
+               case S_IFREG | 0755:
+               case S_IFREG | 0644:
+               case S_IFLNK:
+               case S_IFDIR:
+                       break;
+               /*
+                * This is nonstandard, but we had a few of these
+                * early on when we honored the full set of mode
+                * bits..
+                */
+               case S_IFREG | 0664:
+                       break;
+               default:
+                       printf("tree %s has entry %o %s\n",
+                               sha1_to_hex(item->object.sha1),
+                               entry->mode, entry->name);
+               }
+
+               if (last) {
+                       switch (verify_ordered(last, entry)) {
+                       case TREE_UNORDERED:
+                               fprintf(stderr, "tree %s not ordered\n",
+                                       sha1_to_hex(item->object.sha1));
+                               return -1;
+                       case TREE_HAS_DUPS:
+                               fprintf(stderr, "tree %s has duplicate entries for '%s'\n",
+                                       sha1_to_hex(item->object.sha1),
+                                       entry->name);
+                               return -1;
+                       default:
+                               break;
+                       }
+               }
+
+               last = entry;
+       }
+
+       if (has_full_path) {
                fprintf(stderr, "warning: fsck-cache: tree %s "
-                       "has full pathnames in it\n", sha1_to_hex(sha1));
+                       "has full pathnames in it\n", 
+                       sha1_to_hex(item->object.sha1));
        }
+
        return 0;
 }
 
-static int fsck_commit(unsigned char *sha1, void *data, unsigned long size)
+static int fsck_commit(struct commit *commit)
 {
-       struct commit *commit = lookup_commit(sha1);
-       if (parse_commit(commit))
-               return -1;
        if (!commit->tree)
                return -1;
-       if (!commit->parents)
-               printf("root %s\n", sha1_to_hex(sha1));
+       if (!commit->parents && show_root)
+               printf("root %s\n", sha1_to_hex(commit->object.sha1));
        if (!commit->date)
-               printf("bad commit date in %s\n", sha1_to_hex(sha1));
+               printf("bad commit date in %s\n", 
+                      sha1_to_hex(commit->object.sha1));
        return 0;
 }
 
-static int fsck_blob(unsigned char *sha1, void *data, unsigned long size)
+static int fsck_tag(struct tag *tag)
 {
-       struct blob *blob = lookup_blob(sha1);
-       blob->object.parsed = 1;
+       struct object *tagged = tag->tagged;
+
+       if (!tagged) {
+               printf("bad object in tag %s\n", sha1_to_hex(tag->object.sha1));
+               return -1;
+       }
+       if (!show_tags)
+               return 0;
+
+       printf("tagged %s %s", tagged->type, sha1_to_hex(tagged->sha1));
+       printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
        return 0;
 }
 
-static int fsck_tag(unsigned char *sha1, void *data, unsigned long size)
+static int fsck_sha1(unsigned char *sha1)
 {
-       int typelen, taglen;
-       unsigned char object[20];
-       char object_hex[60];
-       const char *type_line, *tag_line, *sig_line;
-
-       if (size < 64)
-               return -1;
-       if (memcmp("object ", data, 7) || get_sha1_hex(data + 7, object))
+       struct object *obj = parse_object(sha1);
+       if (!obj)
                return -1;
+       if (obj->type == blob_type)
+               return 0;
+       if (obj->type == tree_type)
+               return fsck_tree((struct tree *) obj);
+       if (obj->type == commit_type)
+               return fsck_commit((struct commit *) obj);
+       if (obj->type == tag_type)
+               return fsck_tag((struct tag *) obj);
+       return -1;
+}
 
-       type_line = data + 48;
-       if (memcmp("\ntype ", type_line-1, 6))
-               return -1;
+/*
+ * This is the sorting chunk size: make it reasonably
+ * big so that we can sort well..
+ */
+#define MAX_SHA1_ENTRIES (1024)
 
-       tag_line = strchr(type_line, '\n');
-       if (!tag_line || memcmp("tag ", ++tag_line, 4))
-               return -1;
+struct sha1_entry {
+       unsigned long ino;
+       unsigned char sha1[20];
+};
 
-       sig_line = strchr(tag_line, '\n');
-       if (!sig_line)
-               return -1;
-       sig_line++;
+static struct {
+       unsigned long nr;
+       struct sha1_entry *entry[MAX_SHA1_ENTRIES];
+} sha1_list;
 
-       typelen = tag_line - type_line - strlen("type \n");
-       if (typelen >= 20)
-               return -1;
-       taglen = sig_line - tag_line - strlen("tag \n");
-
-       strcpy(object_hex, sha1_to_hex(object));
-       printf("tagged %.*s %s (%.*s) in %s\n",
-               typelen, type_line + 5,
-               object_hex,
-               taglen, tag_line + 4,
-               sha1_to_hex(sha1));
-       return 0;
+static int ino_compare(const void *_a, const void *_b)
+{
+       const struct sha1_entry *a = _a, *b = _b;
+       unsigned long ino1 = a->ino, ino2 = b->ino;
+       return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
 }
 
-static int fsck_entry(unsigned char *sha1, char *tag, void *data, 
-                     unsigned long size)
+static void fsck_sha1_list(void)
 {
-       if (!strcmp(tag, "blob")) {
-               if (fsck_blob(sha1, data, size) < 0)
-                       return -1;
-       } else if (!strcmp(tag, "tree")) {
-               if (fsck_tree(sha1, data, size) < 0)
-                       return -1;
-       } else if (!strcmp(tag, "commit")) {
-               if (fsck_commit(sha1, data, size) < 0)
-                       return -1;
-       } else if (!strcmp(tag, "tag")) {
-               if (fsck_tag(sha1, data, size) < 0)
-                       return -1;
-       } else
-               return -1;
-       return 0;
+       int i, nr = sha1_list.nr;
+
+       qsort(sha1_list.entry, nr, sizeof(struct sha1_entry *), ino_compare);
+       for (i = 0; i < nr; i++) {
+               struct sha1_entry *entry = sha1_list.entry[i];
+               unsigned char *sha1 = entry->sha1;
+
+               sha1_list.entry[i] = NULL;
+               if (fsck_sha1(sha1) < 0)
+                       fprintf(stderr, "bad sha1 entry '%s'\n", sha1_to_hex(sha1));
+               free(entry);
+       }
+       sha1_list.nr = 0;
 }
 
-static int fsck_name(char *hex)
+static void add_sha1_list(unsigned char *sha1, unsigned long ino)
 {
-       unsigned char sha1[20];
-       if (!get_sha1_hex(hex, sha1)) {
-               unsigned long mapsize;
-               void *map = map_sha1_file(sha1, &mapsize);
-               if (map) {
-                       char type[100];
-                       unsigned long size;
-                       void *buffer = unpack_sha1_file(map, mapsize, type, &size);
-                       if (!buffer)
-                               return -1;
-                       if (check_sha1_signature(sha1, buffer, size, type) < 0)
-                               printf("sha1 mismatch %s\n", sha1_to_hex(sha1));
-                       munmap(map, mapsize);
-                       if (!fsck_entry(sha1, type, buffer, size))
-                               return 0;
-               }
+       struct sha1_entry *entry = xmalloc(sizeof(*entry));
+       int nr;
+
+       entry->ino = ino;
+       memcpy(entry->sha1, sha1, 20);
+       nr = sha1_list.nr;
+       if (nr == MAX_SHA1_ENTRIES) {
+               fsck_sha1_list();
+               nr = 0;
        }
-       return -1;
+       sha1_list.entry[nr] = entry;
+       sha1_list.nr = ++nr;
 }
 
 static int fsck_dir(int i, char *path)
@@ -161,6 +271,7 @@ static int fsck_dir(int i, char *path)
 
        while ((de = readdir(dir)) != NULL) {
                char name[100];
+               unsigned char sha1[20];
                int len = strlen(de->d_name);
 
                switch (len) {
@@ -174,8 +285,10 @@ static int fsck_dir(int i, char *path)
                case 38:
                        sprintf(name, "%02x", i);
                        memcpy(name+2, de->d_name, len+1);
-                       if (!fsck_name(name))
-                               continue;
+                       if (get_sha1_hex(name, sha1) < 0)
+                               break;
+                       add_sha1_list(sha1, de->d_ino);
+                       continue;
                }
                fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
        }
@@ -188,35 +301,80 @@ int main(int argc, char **argv)
        int i, heads;
        char *sha1_dir;
 
-       sha1_dir = getenv(DB_ENVIRONMENT) ? : DEFAULT_DB_ENVIRONMENT;
+       for (i = 1; i < argc; i++) {
+               const char *arg = argv[i];
+
+               if (!strcmp(arg, "--unreachable")) {
+                       show_unreachable = 1;
+                       continue;
+               }
+               if (!strcmp(arg, "--tags")) {
+                       show_tags = 1;
+                       continue;
+               }
+               if (!strcmp(arg, "--root")) {
+                       show_root = 1;
+                       continue;
+               }
+               if (!strcmp(arg, "--cache")) {
+                       keep_cache_objects = 1;
+                       continue;
+               }
+               if (*arg == '-')
+                       usage("fsck-cache [--tags] [[--unreachable] [--cache] <head-sha1>*]");
+       }
+
+       sha1_dir = get_object_directory();
        for (i = 0; i < 256; i++) {
                static char dir[4096];
                sprintf(dir, "%s/%02x", sha1_dir, i);
                fsck_dir(i, dir);
        }
+       fsck_sha1_list();
 
        heads = 0;
        for (i = 1; i < argc; i++) {
-               if (!strcmp(argv[i], "--unreachable")) {
-                       show_unreachable = 1;
+               const char *arg = argv[i]; 
+
+               if (*arg == '-')
                        continue;
-               }
-               if (!get_sha1_hex(argv[i], head_sha1)) {
-                       struct object *obj = &lookup_commit(head_sha1)->object;
+
+               if (!get_sha1(arg, head_sha1)) {
+                       struct object *obj = lookup_object(head_sha1);
+
+                       /* Error is printed by lookup_object(). */
+                       if (!obj)
+                               continue;
+
                        obj->used = 1;
                        mark_reachable(obj, REACHABLE);
                        heads++;
                        continue;
                }
-               error("fsck-cache [[--unreachable] <head-sha1>*]");
+               error("expected sha1, got %s", arg);
+       }
+
+       if (keep_cache_objects) {
+               int i;
+               read_cache();
+               for (i = 0; i < active_nr; i++) {
+                       struct blob *blob = lookup_blob(active_cache[i]->sha1);
+                       struct object *obj;
+                       if (!blob)
+                               continue;
+                       obj = &blob->object;
+                       obj->used = 1;
+                       mark_reachable(obj, REACHABLE);
+               }
        }
 
-       if (!heads) {
+       if (!heads && !keep_cache_objects) {
                if (show_unreachable) {
-                       fprintf(stderr, "unable to do reachability without a head\n");
+                       fprintf(stderr, "unable to do reachability without a head nor --cache\n");
                        show_unreachable = 0; 
                }
-               fprintf(stderr, "expect dangling commits - potential heads - due to lack of head information\n");
+               if (!heads)
+                       fprintf(stderr, "expect dangling commits - potential heads - due to lack of head information\n");
        }
 
        check_connectivity();