Merge branch 'ps/ref-backend-migration-optim' into seen

Optimize migration procedure between two ref backends.

* ps/ref-backend-migration-optim:
  reftable/block: optimize allocations by using scratch buffer
  reftable/block: rename `block_writer::buf` variable
  reftable/writer: optimize allocations by using a scratch buffer
  refs: don't normalize log messages with `REF_SKIP_CREATE_REFLOG`
  refs: skip collision checks in initial transactions
  refs: use "initial" transaction semantics to migrate refs
  refs/files: support symbolic and root refs in initial transaction
  refs: introduce "initial" transaction flag
  refs/files: move logic to commit initial transaction
  refs: allow passing flags when setting up a transaction

Author: gitster

branch.c

@@ -627,7 +627,7 @@ void create_branch(struct repository *r,
 	else
 		msg = xstrfmt("branch: Created from %s", start_name);
 	transaction = ref_store_transaction_begin(get_main_ref_store(the_repository),
-						  &err);
+						  0, &err);
 	if (!transaction ||
 		ref_transaction_update(transaction, ref.buf,
 					&oid, forcing ? NULL : null_oid(),

builtin/clone.c

@@ -574,7 +574,7 @@ static void write_remote_refs(const struct ref *local_refs)
 	struct strbuf err = STRBUF_INIT;
 
 	t = ref_store_transaction_begin(get_main_ref_store(the_repository),
-					&err);
+					REF_TRANSACTION_FLAG_INITIAL, &err);
 	if (!t)
 		die("%s", err.buf);
 
@@ -586,7 +586,7 @@ static void write_remote_refs(const struct ref *local_refs)
 			die("%s", err.buf);
 	}
 
-	if (initial_ref_transaction_commit(t, &err))
+	if (ref_transaction_commit(t, &err))
 		die("%s", err.buf);
 
 	strbuf_release(&err);

builtin/fast-import.c

@@ -1635,7 +1635,7 @@ static int update_branch(struct branch *b)
 		}
 	}
 	transaction = ref_store_transaction_begin(get_main_ref_store(the_repository),
-						  &err);
+						  0, &err);
 	if (!transaction ||
 	    ref_transaction_update(transaction, b->name, &b->oid, &old_oid,
 				   NULL, NULL, 0, msg, &err) ||
@@ -1670,7 +1670,7 @@ static void dump_tags(void)
 	struct ref_transaction *transaction;
 
 	transaction = ref_store_transaction_begin(get_main_ref_store(the_repository),
-						  &err);
+						  0, &err);
 	if (!transaction) {
 		failure |= error("%s", err.buf);
 		goto cleanup;

builtin/fetch.c

@@ -665,7 +665,7 @@ static int s_update_ref(const char *action,
 	 */
 	if (!transaction) {
 		transaction = our_transaction = ref_store_transaction_begin(get_main_ref_store(the_repository),
-									    &err);
+									    0, &err);
 		if (!transaction) {
 			ret = STORE_REF_ERROR_OTHER;
 			goto out;
@@ -1756,7 +1756,7 @@ static int do_fetch(struct transport *transport,
 
 	if (atomic_fetch) {
 		transaction = ref_store_transaction_begin(get_main_ref_store(the_repository),
-							  &err);
+							  0, &err);
 		if (!transaction) {
 			retcode = -1;
 			goto cleanup;

builtin/receive-pack.c

@@ -1849,7 +1849,7 @@ static void execute_commands_non_atomic(struct command *commands,
 			continue;
 
 		transaction = ref_store_transaction_begin(get_main_ref_store(the_repository),
-							  &err);
+							  0, &err);
 		if (!transaction) {
 			rp_error("%s", err.buf);
 			strbuf_reset(&err);
@@ -1878,7 +1878,7 @@ static void execute_commands_atomic(struct command *commands,
 	const char *reported_error = "atomic push failure";
 
 	transaction = ref_store_transaction_begin(get_main_ref_store(the_repository),
-						  &err);
+						  0, &err);
 	if (!transaction) {
 		rp_error("%s", err.buf);
 		strbuf_reset(&err);

builtin/replace.c

@@ -201,7 +201,7 @@ static int replace_object_oid(const char *object_ref,
 	}
 
 	transaction = ref_store_transaction_begin(get_main_ref_store(the_repository),
-						  &err);
+						  0, &err);
 	if (!transaction ||
 	    ref_transaction_update(transaction, ref.buf, repl, &prev,
 				   NULL, NULL, 0, NULL, &err) ||

builtin/tag.c

@@ -681,7 +681,7 @@ int cmd_tag(int argc,
 	}
 
 	transaction = ref_store_transaction_begin(get_main_ref_store(the_repository),
-						  &err);
+						  0, &err);
 	if (!transaction ||
 	    ref_transaction_update(transaction, ref.buf, &object, &prev,
 				   NULL, NULL,

builtin/update-ref.c

@@ -612,7 +612,7 @@ static void update_refs_stdin(void)
 	int i, j;
 
 	transaction = ref_store_transaction_begin(get_main_ref_store(the_repository),
-						  &err);
+						  0, &err);
 	if (!transaction)
 		die("%s", err.buf);
 
@@ -680,7 +680,7 @@ static void update_refs_stdin(void)
 			 */
 			state = cmd->state;
 			transaction = ref_store_transaction_begin(get_main_ref_store(the_repository),
-								  &err);
+								  0, &err);
 			if (!transaction)
 				die("%s", err.buf);
 

refs.c

@@ -919,7 +919,7 @@ int refs_delete_ref(struct ref_store *refs, const char *msg,
 	struct ref_transaction *transaction;
 	struct strbuf err = STRBUF_INIT;
 
-	transaction = ref_store_transaction_begin(refs, &err);
+	transaction = ref_store_transaction_begin(refs, 0, &err);
 	if (!transaction ||
 	    ref_transaction_delete(transaction, refname, old_oid,
 				   NULL, flags, msg, &err) ||
@@ -1117,13 +1117,15 @@ int read_ref_at(struct ref_store *refs, const char *refname,
 }
 
 struct ref_transaction *ref_store_transaction_begin(struct ref_store *refs,
+						    unsigned int flags,
 						    struct strbuf *err)
 {
 	struct ref_transaction *tr;
 	assert(err);
 
 	CALLOC_ARRAY(tr, 1);
 	tr->ref_store = refs;
+	tr->flags = flags;
 	return tr;
 }
 
@@ -1187,8 +1189,9 @@ struct ref_update *ref_transaction_add_update(
 		oidcpy(&update->new_oid, new_oid);
 	if ((flags & REF_HAVE_OLD) && old_oid)
 		oidcpy(&update->old_oid, old_oid);
+	if (!(flags & REF_SKIP_CREATE_REFLOG))
+		update->msg = normalize_reflog_message(msg);
 
-	update->msg = normalize_reflog_message(msg);
 	return update;
 }
 
@@ -1310,7 +1313,7 @@ int refs_update_ref(struct ref_store *refs, const char *msg,
 	struct strbuf err = STRBUF_INIT;
 	int ret = 0;
 
-	t = ref_store_transaction_begin(refs, &err);
+	t = ref_store_transaction_begin(refs, 0, &err);
 	if (!t ||
 	    ref_transaction_update(t, refname, new_oid, old_oid, NULL, NULL,
 				   flags, msg, &err) ||
@@ -2128,7 +2131,7 @@ int refs_update_symref_extended(struct ref_store *refs, const char *ref,
 	struct strbuf err = STRBUF_INIT;
 	int ret = 0, prepret = 0;
 
-	transaction = ref_store_transaction_begin(refs, &err);
+	transaction = ref_store_transaction_begin(refs, 0, &err);
 	if (!transaction) {
 	error_return:
 		ret = error("%s", err.buf);
@@ -2342,7 +2345,7 @@ int ref_transaction_commit(struct ref_transaction *transaction,
 	}
 
 	ret = refs->be->transaction_finish(refs, transaction, err);
-	if (!ret)
+	if (!ret && !(transaction->flags & REF_TRANSACTION_FLAG_INITIAL))
 		run_transaction_hook(transaction, "committed");
 	return ret;
 }
@@ -2351,6 +2354,7 @@ int refs_verify_refname_available(struct ref_store *refs,
 				  const char *refname,
 				  const struct string_list *extras,
 				  const struct string_list *skip,
+				  int initial_transaction,
 				  struct strbuf *err)
 {
 	const char *slash;
@@ -2359,8 +2363,6 @@ int refs_verify_refname_available(struct ref_store *refs,
 	struct strbuf referent = STRBUF_INIT;
 	struct object_id oid;
 	unsigned int type;
-	struct ref_iterator *iter;
-	int ok;
 	int ret = -1;
 
 	/*
@@ -2390,7 +2392,8 @@ int refs_verify_refname_available(struct ref_store *refs,
 		if (skip && string_list_has_string(skip, dirname.buf))
 			continue;
 
-		if (!refs_read_raw_ref(refs, dirname.buf, &oid, &referent,
+		if (!initial_transaction &&
+		    !refs_read_raw_ref(refs, dirname.buf, &oid, &referent,
 				       &type, &ignore_errno)) {
 			strbuf_addf(err, _("'%s' exists; cannot create '%s'"),
 				    dirname.buf, refname);
@@ -2415,21 +2418,26 @@ int refs_verify_refname_available(struct ref_store *refs,
 	strbuf_addstr(&dirname, refname + dirname.len);
 	strbuf_addch(&dirname, '/');
 
-	iter = refs_ref_iterator_begin(refs, dirname.buf, NULL, 0,
-				       DO_FOR_EACH_INCLUDE_BROKEN);
-	while ((ok = ref_iterator_advance(iter)) == ITER_OK) {
-		if (skip &&
-		    string_list_has_string(skip, iter->refname))
-			continue;
+	if (!initial_transaction) {
+		struct ref_iterator *iter;
+		int ok;
 
-		strbuf_addf(err, _("'%s' exists; cannot create '%s'"),
-			    iter->refname, refname);
-		ref_iterator_abort(iter);
-		goto cleanup;
-	}
+		iter = refs_ref_iterator_begin(refs, dirname.buf, NULL, 0,
+					       DO_FOR_EACH_INCLUDE_BROKEN);
+		while ((ok = ref_iterator_advance(iter)) == ITER_OK) {
+			if (skip &&
+			    string_list_has_string(skip, iter->refname))
+				continue;
 
-	if (ok != ITER_DONE)
-		BUG("error while iterating over references");
+			strbuf_addf(err, _("'%s' exists; cannot create '%s'"),
+				    iter->refname, refname);
+			ref_iterator_abort(iter);
+			goto cleanup;
+		}
+
+		if (ok != ITER_DONE)
+			BUG("error while iterating over references");
+	}
 
 	extra_refname = find_descendant_ref(dirname.buf, extras, skip);
 	if (extra_refname)
@@ -2513,14 +2521,6 @@ int refs_reflog_expire(struct ref_store *refs,
 				       cleanup_fn, policy_cb_data);
 }
 
-int initial_ref_transaction_commit(struct ref_transaction *transaction,
-				   struct strbuf *err)
-{
-	struct ref_store *refs = transaction->ref_store;
-
-	return refs->be->initial_transaction_commit(refs, transaction, err);
-}
-
 void ref_transaction_for_each_queued_update(struct ref_transaction *transaction,
 					    ref_transaction_for_each_queued_update_fn cb,
 					    void *cb_data)
@@ -2556,7 +2556,7 @@ int refs_delete_refs(struct ref_store *refs, const char *logmsg,
 	 * individual updates can't fail, so we can pack all of the
 	 * updates into a single transaction.
 	 */
-	transaction = ref_store_transaction_begin(refs, &err);
+	transaction = ref_store_transaction_begin(refs, 0, &err);
 	if (!transaction) {
 		ret = error("%s", err.buf);
 		goto out;
@@ -2862,7 +2862,8 @@ int repo_migrate_ref_storage_format(struct repository *repo,
 	if (ret < 0)
 		goto done;
 
-	transaction = ref_store_transaction_begin(new_refs, errbuf);
+	transaction = ref_store_transaction_begin(new_refs, REF_TRANSACTION_FLAG_INITIAL,
+						  errbuf);
 	if (!transaction)
 		goto done;
 
@@ -2887,13 +2888,6 @@ int repo_migrate_ref_storage_format(struct repository *repo,
 	if (ret < 0)
 		goto done;
 
-	/*
-	 * TODO: we might want to migrate to `initial_ref_transaction_commit()`
-	 * here, which is more efficient for the files backend because it would
-	 * write new refs into the packed-refs file directly. At this point,
-	 * the files backend doesn't handle pseudo-refs and symrefs correctly
-	 * though, so this requires some more work.
-	 */
 	ret = ref_transaction_commit(transaction, errbuf);
 	if (ret < 0)
 		goto done;

refs.h

@@ -101,13 +101,16 @@ int refs_read_symbolic_ref(struct ref_store *ref_store, const char *refname,
  * both "foo" and with "foo/bar/baz" but not with "foo/bar" or
  * "foo/barbados".
  *
+ * If `initial_transaction` is truish, then all collision checks with
+ * preexisting refs are skipped.
+ *
  * extras and skip must be sorted.
  */
-
 int refs_verify_refname_available(struct ref_store *refs,
 				  const char *refname,
 				  const struct string_list *extras,
 				  const struct string_list *skip,
+				  int initial_transaction,
 				  struct strbuf *err);
 
 int refs_ref_exists(struct ref_store *refs, const char *refname);
@@ -214,11 +217,9 @@ char *repo_default_branch_name(struct repository *r, int quiet);
  *
  *   Or
  *
- *   - Call `initial_ref_transaction_commit()` if the ref database is
- *     known to be empty and have no other writers (e.g. during
- *     clone). This is likely to be much faster than
- *     `ref_transaction_commit()`. `ref_transaction_prepare()` should
- *     *not* be called before `initial_ref_transaction_commit()`.
+ *   - Call `ref_transaction_begin()` with REF_TRANSACTION_FLAG_INITIAL if the
+ *     ref database is known to be empty and have no other writers (e.g. during
+ *     clone). This is likely to be much faster than without the flag.
  *
  * - Then finally, call `ref_transaction_free()` to free the
  *   `ref_transaction` data structure.
@@ -234,7 +235,7 @@ char *repo_default_branch_name(struct repository *r, int quiet);
  *         struct strbuf err = STRBUF_INIT;
  *         int ret = 0;
  *
- *         transaction = ref_store_transaction_begin(refs, &err);
+ *         transaction = ref_store_transaction_begin(refs, 0, &err);
  *         if (!transaction ||
  *             ref_transaction_update(...) ||
  *             ref_transaction_create(...) ||
@@ -584,11 +585,27 @@ enum action_on_err {
 	UPDATE_REFS_QUIET_ON_ERR
 };
 
+enum ref_transaction_flag {
+	/*
+	 * The ref transaction is part of the initial creation of the ref store
+	 * and can thus assume that the ref store is completely empty. This
+	 * allows the backend to perform the transaction more efficiently by
+	 * skipping certain checks.
+	 *
+	 * It is a bug to set this flag when there might be other processes
+	 * accessing the repository or if there are existing references that
+	 * might conflict with the ones being created. All old_oid values must
+	 * either be absent or null_oid.
+	 */
+	REF_TRANSACTION_FLAG_INITIAL = (1 << 0),
+};
+
 /*
  * Begin a reference transaction.  The reference transaction must
  * be freed by calling ref_transaction_free().
  */
 struct ref_transaction *ref_store_transaction_begin(struct ref_store *refs,
+						    unsigned int flags,
 						    struct strbuf *err);
 
 /*
@@ -804,20 +821,6 @@ int ref_transaction_commit(struct ref_transaction *transaction,
 int ref_transaction_abort(struct ref_transaction *transaction,
 			  struct strbuf *err);
 
-/*
- * Like ref_transaction_commit(), but optimized for creating
- * references when originally initializing a repository (e.g., by "git
- * clone"). It writes the new references directly to packed-refs
- * without locking the individual references.
- *
- * It is a bug to call this function when there might be other
- * processes accessing the repository or if there are existing
- * references that might conflict with the ones being created. All
- * old_oid values must either be absent or null_oid.
- */
-int initial_ref_transaction_commit(struct ref_transaction *transaction,
-				   struct strbuf *err);
-
 /*
  * Execute the given callback function for each of the reference updates which
  * have been queued in the given transaction. `old_oid` and `new_oid` may be