clone: support remote shallow repository

Cloning from a shallow repository does not follow the "8 steps for new
.git/shallow" because if it does we need to get through step 6 for all
refs. That means commit walking down to the bottom.

Instead the rule to create .git/shallow is simpler and, more
importantly, cheap: if a shallow commit is found in the pack, it's
probably used (i.e. reachable from some refs), so we add it. Others
are dropped.

One may notice this method seems flawed by the word "probably". A
shallow commit may not be reachable from any refs at all if it's
attached to an object island (a group of objects that are not
reachable by any refs).

If that object island is not complete, a new fetch request may send
more objects to connect it to some ref. At that time, because we
incorrectly installed the shallow commit in this island, the user will
not see anything after that commit (fsck is still ok). This is not
desired.

Given that object islands are rare (C Git never sends such islands for
security reasons) and do not really harm the repository integrity, a
tradeoff is made to surprise the user occasionally but work faster
everyday.

A new option --strict could be added later that follows exactly the 8
steps. "git prune" can also learn to remove dangling objects _and_ the
shallow commits that are attached to them from .git/shallow.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>

builtin/clone.c

@@ -889,6 +889,7 @@ int cmd_clone(int argc, const char **argv, const char *prefix)
 
 	remote = remote_get(option_origin);
 	transport = transport_get(remote, remote->url[0]);
+	transport->cloning = 1;
 
 	if (!transport->get_refs_list || (!is_local && !transport->fetch))
 		die(_("Don't know how to clone %s"), transport->url);

builtin/fetch-pack.c

@@ -153,7 +153,7 @@ int cmd_fetch_pack(int argc, const char **argv, const char *prefix)
 	get_remote_heads(fd[0], NULL, 0, &ref, 0, NULL, NULL);
 
 	ref = fetch_pack(&args, fd, conn, ref, dest,
-			 sought, nr_sought, pack_lockfile_ptr);
+			 sought, nr_sought, NULL, pack_lockfile_ptr);
 	if (pack_lockfile) {
 		printf("lock %s\n", pack_lockfile);
 		fflush(stdout);

fetch-pack.c

@@ -13,6 +13,7 @@
 #include "transport.h"
 #include "version.h"
 #include "prio-queue.h"
+#include "sha1-array.h"
 
 static int transfer_unpack_limit = -1;
 static int fetch_unpack_limit = -1;
@@ -774,6 +775,7 @@ static struct ref *do_fetch_pack(struct fetch_pack_args *args,
 				 int fd[2],
 				 const struct ref *orig_ref,
 				 struct ref **sought, int nr_sought,
+				 struct shallow_info *si,
 				 char **pack_lockfile)
 {
 	struct ref *ref = copy_ref_list(orig_ref);
@@ -852,6 +854,8 @@ static struct ref *do_fetch_pack(struct fetch_pack_args *args,
 	if (args->depth > 0)
 		setup_alternate_shallow(&shallow_lock, &alternate_shallow_file,
 					NULL);
+	else if (args->cloning && si->shallow && si->shallow->nr)
+		alternate_shallow_file = setup_temporary_shallow(si->shallow);
 	else
 		alternate_shallow_file = NULL;
 	if (get_pack(args, fd, pack_lockfile))
@@ -925,8 +929,11 @@ static int remove_duplicates_in_refs(struct ref **ref, int nr)
 	return dst;
 }
 
-static void update_shallow(struct fetch_pack_args *args)
+static void update_shallow(struct fetch_pack_args *args,
+			   struct shallow_info *si)
 {
+	int i;
+
 	if (args->depth > 0 && alternate_shallow_file) {
 		if (*alternate_shallow_file == '\0') { /* --unshallow */
 			unlink_or_warn(git_path("shallow"));
@@ -935,6 +942,42 @@ static void update_shallow(struct fetch_pack_args *args)
 			commit_lock_file(&shallow_lock);
 		return;
 	}
+
+	if (!si->shallow || !si->shallow->nr)
+		return;
+
+	if (alternate_shallow_file) {
+		/*
+		 * The temporary shallow file is only useful for
+		 * index-pack and unpack-objects because it may
+		 * contain more roots than we want. Delete it.
+		 */
+		if (*alternate_shallow_file)
+			unlink(alternate_shallow_file);
+		free((char *)alternate_shallow_file);
+	}
+
+	if (args->cloning) {
+		/*
+		 * remote is shallow, but this is a clone, there are
+		 * no objects in repo to worry about. Accept any
+		 * shallow points that exist in the pack (iow in repo
+		 * after get_pack() and reprepare_packed_git())
+		 */
+		struct sha1_array extra = SHA1_ARRAY_INIT;
+		unsigned char (*sha1)[20] = si->shallow->sha1;
+		for (i = 0; i < si->shallow->nr; i++)
+			if (has_sha1_file(sha1[i]))
+				sha1_array_append(&extra, sha1[i]);
+		if (extra.nr) {
+			setup_alternate_shallow(&shallow_lock,
+						&alternate_shallow_file,
+						&extra);
+			commit_lock_file(&shallow_lock);
+		}
+		sha1_array_clear(&extra);
+		return;
+	}
 }
 
 struct ref *fetch_pack(struct fetch_pack_args *args,
@@ -942,9 +985,11 @@ struct ref *fetch_pack(struct fetch_pack_args *args,
 		       const struct ref *ref,
 		       const char *dest,
 		       struct ref **sought, int nr_sought,
+		       struct sha1_array *shallow,
 		       char **pack_lockfile)
 {
 	struct ref *ref_cpy;
+	struct shallow_info si;
 
 	fetch_pack_setup();
 	if (nr_sought)
@@ -954,8 +999,11 @@ struct ref *fetch_pack(struct fetch_pack_args *args,
 		packet_flush(fd[1]);
 		die("no matching remote head");
 	}
-	ref_cpy = do_fetch_pack(args, fd, ref, sought, nr_sought, pack_lockfile);
+	prepare_shallow_info(&si, shallow);
-	update_shallow(args);
+	ref_cpy = do_fetch_pack(args, fd, ref, sought, nr_sought,
+				&si, pack_lockfile);
 	reprepare_packed_git();
+	update_shallow(args, &si);
+	clear_shallow_info(&si);
 	return ref_cpy;
 }

fetch-pack.h

@@ -4,6 +4,8 @@
 #include "string-list.h"
 #include "run-command.h"
 
+struct sha1_array;
+
 struct fetch_pack_args {
 	const char *uploadpack;
 	int unpacklimit;
@@ -20,6 +22,7 @@ struct fetch_pack_args {
 	unsigned stateless_rpc:1;
 	unsigned check_self_contained_and_connected:1;
 	unsigned self_contained_and_connected:1;
+	unsigned cloning:1;
 };
 
 /*
@@ -33,6 +36,7 @@ struct ref *fetch_pack(struct fetch_pack_args *args,
 		       const char *dest,
 		       struct ref **sought,
 		       int nr_sought,
+		       struct sha1_array *shallow,
 		       char **pack_lockfile);
 
 #endif

transport.c

@@ -456,6 +456,7 @@ struct git_transport_data {
 	int fd[2];
 	unsigned got_remote_heads : 1;
 	struct sha1_array extra_have;
+	struct sha1_array shallow;
 };
 
 static int set_git_option(struct git_transport_options *opts,
@@ -512,7 +513,9 @@ static struct ref *get_refs_via_connect(struct transport *transport, int for_pus
 
 	connect_setup(transport, for_push, 0);
 	get_remote_heads(data->fd[0], NULL, 0, &refs,
-			 for_push ? REF_NORMAL : 0, &data->extra_have, NULL);
+			 for_push ? REF_NORMAL : 0,
+			 &data->extra_have,
+			 transport->cloning ? &data->shallow : NULL);
 	data->got_remote_heads = 1;
 
 	return refs;
@@ -539,17 +542,19 @@ static int fetch_refs_via_pack(struct transport *transport,
 	args.depth = data->options.depth;
 	args.check_self_contained_and_connected =
 		data->options.check_self_contained_and_connected;
+	args.cloning = transport->cloning;
 
 	if (!data->got_remote_heads) {
 		connect_setup(transport, 0, 0);
 		get_remote_heads(data->fd[0], NULL, 0, &refs_tmp, 0,
-				 NULL, NULL);
+				 NULL,
+				 transport->cloning ? &data->shallow : NULL);
 		data->got_remote_heads = 1;
 	}
 
 	refs = fetch_pack(&args, data->fd, data->conn,
 			  refs_tmp ? refs_tmp : transport->remote_refs,
-			  dest, to_fetch, nr_heads,
+			  dest, to_fetch, nr_heads, &data->shallow,
 			  &transport->pack_lockfile);
 	close(data->fd[0]);
 	close(data->fd[1]);

transport.h

@@ -35,6 +35,12 @@ struct transport {
 	 */
 	unsigned cannot_reuse : 1;
 
+	/*
+	 * A hint from caller that it will be performing a clone, not
+	 * normal fetch. IOW the repository is guaranteed empty.
+	 */
+	unsigned cloning : 1;
+
 	/**
 	 * Returns 0 if successful, positive if the option is not
 	 * recognized or is inapplicable, and negative if the option