fast-import: Proper notes tree manipulation
[gitweb.git] / fast-import.c
index 7ef9865aa6da794ab52cfc50f21f9f41f861fc4f..f04cd1751ae268692b128e1c8b10370c5872d548 100644 (file)
@@ -22,8 +22,8 @@ Format of STDIN stream:
     ('author' sp name sp '<' email '>' sp when lf)?
     'committer' sp name sp '<' email '>' sp when lf
     commit_msg
-    ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
-    ('merge' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)*
+    ('from' sp committish lf)?
+    ('merge' sp committish lf)*
     file_change*
     lf?;
   commit_msg ::= data;
@@ -41,15 +41,18 @@ Format of STDIN stream:
   file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
   file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
     data;
+  note_obm ::= 'N' sp (hexsha1 | idnum) sp committish lf;
+  note_inm ::= 'N' sp 'inline' sp committish lf
+    data;
 
   new_tag ::= 'tag' sp tag_str lf
-    'from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf
+    'from' sp committish lf
     ('tagger' sp name sp '<' email '>' sp when lf)?
     tag_msg;
   tag_msg ::= data;
 
   reset_branch ::= 'reset' sp ref_str lf
-    ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
+    ('from' sp committish lf)?
     lf?;
 
   checkpoint ::= 'checkpoint' lf
@@ -88,6 +91,7 @@ Format of STDIN stream:
      # stream formatting is: \, " and LF.  Otherwise these values
      # are UTF8.
      #
+  committish  ::= (ref_str | hexsha1 | sha1exp_str | idnum);
   ref_str     ::= ref;
   sha1exp_str ::= sha1exp;
   tag_str     ::= tag;
@@ -241,6 +245,7 @@ struct branch
        const char *name;
        struct tree_entry branch_tree;
        uintmax_t last_commit;
+       uintmax_t num_notes;
        unsigned active : 1;
        unsigned pack_id : PACK_ID_BITS;
        unsigned char sha1[20];
@@ -689,6 +694,7 @@ static struct branch *new_branch(const char *name)
        b->table_next_branch = branch_table[hc];
        b->branch_tree.versions[0].mode = S_IFDIR;
        b->branch_tree.versions[1].mode = S_IFDIR;
+       b->num_notes = 0;
        b->active = 0;
        b->pack_id = MAX_PACK_ID;
        branch_table[hc] = b;
@@ -1744,10 +1750,12 @@ static int validate_raw_date(const char *src, char *result, int maxlen)
 {
        const char *orig_src = src;
        char *endp;
+       unsigned long num;
 
        errno = 0;
 
-       strtoul(src, &endp, 10);
+       num = strtoul(src, &endp, 10);
+       /* NEEDSWORK: perhaps check for reasonable values? */
        if (errno || endp == src || *endp != ' ')
                return -1;
 
@@ -1755,8 +1763,9 @@ static int validate_raw_date(const char *src, char *result, int maxlen)
        if (*src != '-' && *src != '+')
                return -1;
 
-       strtoul(src + 1, &endp, 10);
-       if (errno || endp == src || *endp || (endp - orig_src) >= maxlen)
+       num = strtoul(src + 1, &endp, 10);
+       if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
+           1400 < num)
                return -1;
 
        strcpy(result, orig_src);
@@ -1853,6 +1862,109 @@ static void load_branch(struct branch *b)
        }
 }
 
+static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
+{
+       unsigned char fanout = 0;
+       while ((num_notes >>= 8))
+               fanout++;
+       return fanout;
+}
+
+static void construct_path_with_fanout(const char *hex_sha1,
+               unsigned char fanout, char *path)
+{
+       unsigned int i = 0, j = 0;
+       if (fanout >= 20)
+               die("Too large fanout (%u)", fanout);
+       while (fanout) {
+               path[i++] = hex_sha1[j++];
+               path[i++] = hex_sha1[j++];
+               path[i++] = '/';
+               fanout--;
+       }
+       memcpy(path + i, hex_sha1 + j, 40 - j);
+       path[i + 40 - j] = '\0';
+}
+
+static uintmax_t do_change_note_fanout(
+               struct tree_entry *orig_root, struct tree_entry *root,
+               char *hex_sha1, unsigned int hex_sha1_len,
+               char *fullpath, unsigned int fullpath_len,
+               unsigned char fanout)
+{
+       struct tree_content *t = root->tree;
+       struct tree_entry *e, leaf;
+       unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
+       uintmax_t num_notes = 0;
+       unsigned char sha1[20];
+       char realpath[60];
+
+       for (i = 0; t && i < t->entry_count; i++) {
+               e = t->entries[i];
+               tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
+               tmp_fullpath_len = fullpath_len;
+
+               /*
+                * We're interested in EITHER existing note entries (entries
+                * with exactly 40 hex chars in path, not including directory
+                * separators), OR directory entries that may contain note
+                * entries (with < 40 hex chars in path).
+                * Also, each path component in a note entry must be a multiple
+                * of 2 chars.
+                */
+               if (!e->versions[1].mode ||
+                   tmp_hex_sha1_len > 40 ||
+                   e->name->str_len % 2)
+                       continue;
+
+               /* This _may_ be a note entry, or a subdir containing notes */
+               memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
+                      e->name->str_len);
+               if (tmp_fullpath_len)
+                       fullpath[tmp_fullpath_len++] = '/';
+               memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
+                      e->name->str_len);
+               tmp_fullpath_len += e->name->str_len;
+               fullpath[tmp_fullpath_len] = '\0';
+
+               if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
+                       /* This is a note entry */
+                       construct_path_with_fanout(hex_sha1, fanout, realpath);
+                       if (!strcmp(fullpath, realpath)) {
+                               /* Note entry is in correct location */
+                               num_notes++;
+                               continue;
+                       }
+
+                       /* Rename fullpath to realpath */
+                       if (!tree_content_remove(orig_root, fullpath, &leaf))
+                               die("Failed to remove path %s", fullpath);
+                       tree_content_set(orig_root, realpath,
+                               leaf.versions[1].sha1,
+                               leaf.versions[1].mode,
+                               leaf.tree);
+               } else if (S_ISDIR(e->versions[1].mode)) {
+                       /* This is a subdir that may contain note entries */
+                       if (!e->tree)
+                               load_tree(e);
+                       num_notes += do_change_note_fanout(orig_root, e,
+                               hex_sha1, tmp_hex_sha1_len,
+                               fullpath, tmp_fullpath_len, fanout);
+               }
+
+               /* The above may have reallocated the current tree_content */
+               t = root->tree;
+       }
+       return num_notes;
+}
+
+static uintmax_t change_note_fanout(struct tree_entry *root,
+               unsigned char fanout)
+{
+       char hex_sha1[40], path[60];
+       return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
+}
+
 static void file_change_m(struct branch *b)
 {
        const char *p = command_buf.buf + 2;
@@ -2003,12 +2115,98 @@ static void file_change_cr(struct branch *b, int rename)
                leaf.tree);
 }
 
+static void note_change_n(struct branch *b, unsigned char old_fanout)
+{
+       const char *p = command_buf.buf + 2;
+       static struct strbuf uq = STRBUF_INIT;
+       struct object_entry *oe = oe;
+       struct branch *s;
+       unsigned char sha1[20], commit_sha1[20];
+       char path[60];
+       uint16_t inline_data = 0;
+       unsigned char new_fanout;
+
+       /* <dataref> or 'inline' */
+       if (*p == ':') {
+               char *x;
+               oe = find_mark(strtoumax(p + 1, &x, 10));
+               hashcpy(sha1, oe->sha1);
+               p = x;
+       } else if (!prefixcmp(p, "inline")) {
+               inline_data = 1;
+               p += 6;
+       } else {
+               if (get_sha1_hex(p, sha1))
+                       die("Invalid SHA1: %s", command_buf.buf);
+               oe = find_object(sha1);
+               p += 40;
+       }
+       if (*p++ != ' ')
+               die("Missing space after SHA1: %s", command_buf.buf);
+
+       /* <committish> */
+       s = lookup_branch(p);
+       if (s) {
+               hashcpy(commit_sha1, s->sha1);
+       } else if (*p == ':') {
+               uintmax_t commit_mark = strtoumax(p + 1, NULL, 10);
+               struct object_entry *commit_oe = find_mark(commit_mark);
+               if (commit_oe->type != OBJ_COMMIT)
+                       die("Mark :%" PRIuMAX " not a commit", commit_mark);
+               hashcpy(commit_sha1, commit_oe->sha1);
+       } else if (!get_sha1(p, commit_sha1)) {
+               unsigned long size;
+               char *buf = read_object_with_reference(commit_sha1,
+                       commit_type, &size, commit_sha1);
+               if (!buf || size < 46)
+                       die("Not a valid commit: %s", p);
+               free(buf);
+       } else
+               die("Invalid ref name or SHA1 expression: %s", p);
+
+       if (inline_data) {
+               static struct strbuf buf = STRBUF_INIT;
+
+               if (p != uq.buf) {
+                       strbuf_addstr(&uq, p);
+                       p = uq.buf;
+               }
+               read_next_command();
+               parse_data(&buf);
+               store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
+       } else if (oe) {
+               if (oe->type != OBJ_BLOB)
+                       die("Not a blob (actually a %s): %s",
+                               typename(oe->type), command_buf.buf);
+       } else if (!is_null_sha1(sha1)) {
+               enum object_type type = sha1_object_info(sha1, NULL);
+               if (type < 0)
+                       die("Blob not found: %s", command_buf.buf);
+               if (type != OBJ_BLOB)
+                       die("Not a blob (actually a %s): %s",
+                           typename(type), command_buf.buf);
+       }
+
+       construct_path_with_fanout(sha1_to_hex(commit_sha1), old_fanout, path);
+       if (tree_content_remove(&b->branch_tree, path, NULL))
+               b->num_notes--;
+
+       if (is_null_sha1(sha1))
+               return; /* nothing to insert */
+
+       b->num_notes++;
+       new_fanout = convert_num_notes_to_fanout(b->num_notes);
+       construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
+       tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
+}
+
 static void file_change_deleteall(struct branch *b)
 {
        release_tree_content_recursive(b->branch_tree.tree);
        hashclr(b->branch_tree.versions[0].sha1);
        hashclr(b->branch_tree.versions[1].sha1);
        load_tree(&b->branch_tree);
+       b->num_notes = 0;
 }
 
 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
@@ -2132,6 +2330,7 @@ static void parse_new_commit(void)
        char *committer = NULL;
        struct hash_list *merge_list = NULL;
        unsigned int merge_count;
+       unsigned char prev_fanout, new_fanout;
 
        /* Obtain the branch name from the rest of our command */
        sp = strchr(command_buf.buf, ' ') + 1;
@@ -2162,6 +2361,8 @@ static void parse_new_commit(void)
                load_branch(b);
        }
 
+       prev_fanout = convert_num_notes_to_fanout(b->num_notes);
+
        /* file_change* */
        while (command_buf.len > 0) {
                if (!prefixcmp(command_buf.buf, "M "))
@@ -2172,6 +2373,8 @@ static void parse_new_commit(void)
                        file_change_cr(b, 1);
                else if (!prefixcmp(command_buf.buf, "C "))
                        file_change_cr(b, 0);
+               else if (!prefixcmp(command_buf.buf, "N "))
+                       note_change_n(b, prev_fanout);
                else if (!strcmp("deleteall", command_buf.buf))
                        file_change_deleteall(b);
                else {
@@ -2182,6 +2385,10 @@ static void parse_new_commit(void)
                        break;
        }
 
+       new_fanout = convert_num_notes_to_fanout(b->num_notes);
+       if (new_fanout != prev_fanout)
+               b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
+
        /* build the tree and the commit */
        store_tree(&b->branch_tree);
        hashcpy(b->branch_tree.versions[0].sha1,
@@ -2402,6 +2609,9 @@ int main(int argc, const char **argv)
 
        git_extract_argv0_path(argv[0]);
 
+       if (argc == 2 && !strcmp(argv[1], "-h"))
+               usage(fast_import_usage);
+
        setup_git_directory();
        git_config(git_pack_config, NULL);
        if (!pack_compression_seen && core_compression_seen)