make 'git clone' ask the remote only for objects it cares about
authorNicolas Pitre <nico@fluxnic.net>
Sat, 26 Sep 2009 03:54:42 +0000 (25 23:54 -0400)
committerShawn O. Pearce <spearce@spearce.org>
Sat, 26 Sep 2009 19:51:04 +0000 (26 12:51 -0700)
Current behavior of 'git clone' when not using --mirror is to fetch
everything from the peer, and then filter out unwanted refs just before
writing them out to the cloned repository.  This may become highly
inefficient if the peer has an unusual ref namespace, or if it simply
has "remotes" refs of its own, and those locally unwanted refs are
connecting to a large set of objects which becomes unreferenced as soon
as they are fetched.

Let's filter out those unwanted refs from the peer _before_ asking it
what refs we want to fetch instead, which is the most logical thing to
do anyway.

Signed-off-by: Nicolas Pitre <nico@fluxnic.net>
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
builtin-clone.c

index bab2d84..4992c25 100644 (file)
@@ -329,24 +329,28 @@ static void remove_junk_on_signal(int signo)
        raise(signo);
 }
 
-static struct ref *write_remote_refs(const struct ref *refs,
-               struct refspec *refspec, const char *reflog)
+static struct ref *wanted_peer_refs(const struct ref *refs,
+               struct refspec *refspec)
 {
        struct ref *local_refs = NULL;
        struct ref **tail = &local_refs;
-       struct ref *r;
 
        get_fetch_map(refs, refspec, &tail, 0);
        if (!option_mirror)
                get_fetch_map(refs, tag_refspec, &tail, 0);
 
+       return local_refs;
+}
+
+static void write_remote_refs(const struct ref *local_refs)
+{
+       const struct ref *r;
+
        for (r = local_refs; r; r = r->next)
                add_extra_ref(r->peer_ref->name, r->old_sha1, 0);
 
        pack_refs(PACK_REFS_ALL);
        clear_extra_refs();
-
-       return local_refs;
 }
 
 int cmd_clone(int argc, const char **argv, const char *prefix)
@@ -495,9 +499,10 @@ int cmd_clone(int argc, const char **argv, const char *prefix)
 
        strbuf_reset(&value);
 
-       if (path && !is_bundle)
+       if (path && !is_bundle) {
                refs = clone_local(path, git_dir);
-       else {
+               mapped_refs = wanted_peer_refs(refs, refspec);
+       } else {
                struct remote *remote = remote_get(argv[0]);
                transport = transport_get(remote, remote->url[0]);
 
@@ -520,14 +525,16 @@ int cmd_clone(int argc, const char **argv, const char *prefix)
                                             option_upload_pack);
 
                refs = transport_get_remote_refs(transport);
-               if (refs)
-                       transport_fetch_refs(transport, refs);
+               if (refs) {
+                       mapped_refs = wanted_peer_refs(refs, refspec);
+                       transport_fetch_refs(transport, mapped_refs);
+               }
        }
 
        if (refs) {
                clear_extra_refs();
 
-               mapped_refs = write_remote_refs(refs, refspec, reflog_msg.buf);
+               write_remote_refs(mapped_refs);
 
                remote_head = find_ref_by_name(refs, "HEAD");
                remote_head_points_at =