[PATCH 0/3] reftable/stack: avoid reloading the stack when locked

[PATCH 0/3] reftable/stack: avoid reloading the stack when locked

[Karthik Nayak]

This patch series is based on the report by Jeff [1], where he noticed
that when creating a lot of refs within a single reference transaction,
the majority of the time was spent on fstat().

The issue stems from the fact that within the reftable library we do not
track Git reference transactions, as such any calls within the library
would potentially reload the stack to ensure that there are no
concurrent updates made to the stack. While this makes sense outside of
a reference transaction, within one, the stack is locked, so reloading
the stack is a no-op. The only time we want to reload the stack is
immediately after locking the list file, which is to catch any
concurrent updates made to the stack.

The first patch in this small series, cleans up the flow of reloading
the stack by providing a flag explicitly. The patch argues that since
all flows reload the stack, the flag can be safely removed. This
simplifies the flow of when to reload the stack.

The next two commits move the lock variable to the reftable_stack
structure and then use this information to decide if reloading of the
stack is necessary.

During benchmarking, I first tried to benchmark adding new references
against HEAD. This kicks in the DWIM ref resolution, and we iterate over
siz difference candidate ref names before settling on a match. Each such
lookup reloads the stack. This happens before the reference transaction
is created. I quickly realized that this would dominate the benchmarks,
so the benchmarks in the third patch are against a static commit OID.

The benchmarks show a consistent 1-2% improvement in clock time for
'git-update-ref(1)', but such low values could also be chalked to being
within an error rate. However, the syscall counts show that now the
calls to `newfstatat()` stay constant at around 55 calls regardless of
the number of refs to be created. Before this would grow linearly with
the number of refs.

[1]: https://lore.kernel.org/git/20260629203527.GA1895313@coredump.intra.peff.net/

Signed-off-by: Karthik Nayak <karthik.188@gmail.com>
---
Karthik Nayak (3):
      reftable/stack: remove `REFTABLE_STACK_NEW_ADDITION_RELOAD`
      reftable/stack: move list lock to `struct reftable_stack`
      reftable/stack: avoid reloading the stack when already locked

 refs/reftable-backend.c         | 18 ++++-------
 reftable/reftable-stack.h       | 17 ++--------
 reftable/stack.c                | 69 +++++++++++++++++------------------------
 reftable/stack.h                |  7 ++++-
 t/unit-tests/u-reftable-stack.c | 69 ++++++++++++++++++-----------------------
 5 files changed, 75 insertions(+), 105 deletions(-)


---
base-commit: 18e66859d87fb4b76599f73460b54f0848c76b16
change-id: 20260814-740-optimize-reloading-the-reftable-stack-f5f3adf0a0c0


Thanks
- Karthik

[PATCH 1/3] reftable/stack: remove `REFTABLE_STACK_NEW_ADDITION_RELOAD`

[Karthik Nayak]

In 80e7342ea8 (reftable/stack: allow locking of outdated stacks,
2024-09-24), the `REFTABLE_STACK_NEW_ADDITION_RELOAD` was introduced so
that callers of `reftable_stack_init_addition()` can also reload the
stack if there was a concurrent update made before the lock was
obtained.

Then 16684b6fae (refs/reftable: always reload stacks when creating
lock, 2025-08-12) updated all of the remaining call-sites to propagate
this flag to ensure that we always reload the stack whenever there was a
concurrent update.

As all calls to `reftable_stack_init_addition()` inevitably propagate
the flag, it is safe to remove the flag and its associated code and make
the reloading of the stack the default flow. This makes it easier to
follow the flow and simplifies the logic.

The only exceptions are:

  1. Unit tests, where we explicitly do not propagate the flag. These
     tests are now modified with the new status quo.

  2. `reftable_stack_clean_locked()`, which was propagating 0 to
     `reftable_stack_new_addition()` but was then manually reloading the
     stack after. Here the new flow will achieve the same, while also
     allowing us to remove the manual reload.

This also makes two checks for 'REFTABLE_OUTDATED_ERROR' redundant, so
remove them also.

Signed-off-by: Karthik Nayak <karthik.188@gmail.com>
---
 refs/reftable-backend.c         | 18 ++++-------
 reftable/reftable-stack.h       | 17 ++--------
 reftable/stack.c                | 37 ++++++----------------
 t/unit-tests/u-reftable-stack.c | 69 ++++++++++++++++++-----------------------
 4 files changed, 49 insertions(+), 92 deletions(-)

diff --git a/refs/reftable-backend.c b/refs/reftable-backend.c
index 028f0211af..5c87fd2d68 100644
--- a/refs/reftable-backend.c
+++ b/refs/reftable-backend.c
@@ -1003,8 +1003,7 @@ static int prepare_transaction_update(struct write_transaction_table_arg **out,
 		struct reftable_addition *addition;
 
 		ret = reftable_stack_new_addition(&addition, be->stack,
-						  &reftable_be_write_options(refs)->opts,
-						  REFTABLE_STACK_NEW_ADDITION_RELOAD);
+						  &reftable_be_write_options(refs)->opts);
 		if (ret) {
 			if (ret == REFTABLE_LOCK_ERROR)
 				strbuf_addstr(err, "cannot lock references");
@@ -2010,8 +2009,7 @@ static int reftable_be_rename_ref(struct ref_store *ref_store,
 	if (ret)
 		goto done;
 	ret = reftable_stack_add(arg.be->stack, &write_copy_table, &arg,
-				 &reftable_be_write_options(refs)->opts,
-				 REFTABLE_STACK_NEW_ADDITION_RELOAD);
+				 &reftable_be_write_options(refs)->opts);
 
 done:
 	assert(ret != REFTABLE_API_ERROR);
@@ -2041,8 +2039,7 @@ static int reftable_be_copy_ref(struct ref_store *ref_store,
 	if (ret)
 		goto done;
 	ret = reftable_stack_add(arg.be->stack, &write_copy_table, &arg,
-				 &reftable_be_write_options(refs)->opts,
-				 REFTABLE_STACK_NEW_ADDITION_RELOAD);
+				 &reftable_be_write_options(refs)->opts);
 
 done:
 	assert(ret != REFTABLE_API_ERROR);
@@ -2424,8 +2421,7 @@ static int reftable_be_create_reflog(struct ref_store *ref_store,
 	arg.stack = be->stack;
 
 	ret = reftable_stack_add(be->stack, &write_reflog_existence_table, &arg,
-				 &reftable_be_write_options(refs)->opts,
-				 REFTABLE_STACK_NEW_ADDITION_RELOAD);
+				 &reftable_be_write_options(refs)->opts);
 
 done:
 	return ret;
@@ -2499,8 +2495,7 @@ static int reftable_be_delete_reflog(struct ref_store *ref_store,
 	arg.stack = be->stack;
 
 	ret = reftable_stack_add(be->stack, &write_reflog_delete_table, &arg,
-				 &reftable_be_write_options(refs)->opts,
-				 REFTABLE_STACK_NEW_ADDITION_RELOAD);
+				 &reftable_be_write_options(refs)->opts);
 
 	assert(ret != REFTABLE_API_ERROR);
 	return ret;
@@ -2622,8 +2617,7 @@ static int reftable_be_reflog_expire(struct ref_store *ref_store,
 		goto done;
 
 	ret = reftable_stack_new_addition(&add, be->stack,
-					  &reftable_be_write_options(refs)->opts,
-					  REFTABLE_STACK_NEW_ADDITION_RELOAD);
+					  &reftable_be_write_options(refs)->opts);
 	if (ret < 0)
 		goto done;
 
diff --git a/reftable/reftable-stack.h b/reftable/reftable-stack.h
index 5d22d84e80..5d224f8079 100644
--- a/reftable/reftable-stack.h
+++ b/reftable/reftable-stack.h
@@ -58,22 +58,13 @@ uint64_t reftable_stack_next_update_index(struct reftable_stack *st);
 /* holds a transaction to add tables at the top of a stack. */
 struct reftable_addition;
 
-enum {
-	/*
-	 * Reload the stack when the stack is out-of-date after locking it.
-	 */
-	REFTABLE_STACK_NEW_ADDITION_RELOAD = (1 << 0),
-};
-
 /*
  * returns a new transaction to add reftables to the given stack. As a side
- * effect, the ref database is locked. Accepts REFTABLE_STACK_NEW_ADDITION_*
- * flags.
+ * effect, the ref database is locked.
  */
 int reftable_stack_new_addition(struct reftable_addition **dest,
 				struct reftable_stack *st,
-				const struct reftable_write_options *opts,
-				unsigned int flags);
+				const struct reftable_write_options *opts);
 
 /* Adds a reftable to transaction. */
 int reftable_addition_add(struct reftable_addition *add,
@@ -93,14 +84,12 @@ void reftable_addition_destroy(struct reftable_addition *add);
 /*
  * Add a new table to the stack. The write_table function must call
  * reftable_writer_set_limits, add refs and return an error value.
- * The flags are passed through to `reftable_stack_new_addition()`.
  */
 int reftable_stack_add(struct reftable_stack *st,
 		       int (*write_table)(struct reftable_writer *wr,
 					  void *write_arg),
 		       void *write_arg,
-		       const struct reftable_write_options *opts,
-		       unsigned flags);
+		       const struct reftable_write_options *opts);
 
 struct reftable_iterator;
 
diff --git a/reftable/stack.c b/reftable/stack.c
index 308f9578f0..540f5e77ac 100644
--- a/reftable/stack.c
+++ b/reftable/stack.c
@@ -659,8 +659,7 @@ static void reftable_addition_close(struct reftable_addition *add)
 
 static int reftable_stack_init_addition(struct reftable_addition *add,
 					struct reftable_stack *st,
-					const struct reftable_write_options *opts,
-					unsigned int flags)
+					const struct reftable_write_options *opts)
 {
 	struct reftable_buf lock_file_name = REFTABLE_BUF_INIT;
 	int err;
@@ -686,15 +685,11 @@ static int reftable_stack_init_addition(struct reftable_addition *add,
 	err = stack_uptodate(st);
 	if (err < 0)
 		goto done;
-	if (err > 0 && flags & REFTABLE_STACK_NEW_ADDITION_RELOAD) {
+	if (err > 0) {
 		err = reftable_stack_reload_maybe_reuse(add->stack, 1);
 		if (err)
 			goto done;
 	}
-	if (err > 0) {
-		err = REFTABLE_OUTDATED_ERROR;
-		goto done;
-	}
 
 	add->next_update_index = reftable_stack_next_update_index(st);
 done:
@@ -708,13 +703,12 @@ static int stack_try_add(struct reftable_stack *st,
 			 int (*write_table)(struct reftable_writer *wr,
 					    void *arg),
 			 void *arg,
-			 const struct reftable_write_options *opts,
-			 unsigned flags)
+			 const struct reftable_write_options *opts)
 {
 	struct reftable_addition add;
 	int err;
 
-	err = reftable_stack_init_addition(&add, st, opts, flags);
+	err = reftable_stack_init_addition(&add, st, opts);
 	if (err < 0)
 		goto done;
 
@@ -731,17 +725,10 @@ static int stack_try_add(struct reftable_stack *st,
 int reftable_stack_add(struct reftable_stack *st,
 		       int (*write)(struct reftable_writer *wr, void *arg),
 		       void *arg,
-		       const struct reftable_write_options *opts,
-		       unsigned flags)
+		       const struct reftable_write_options *opts)
 {
-	int err = stack_try_add(st, write, arg, opts, flags);
+	int err = stack_try_add(st, write, arg, opts);
 	if (err < 0) {
-		if (err == REFTABLE_OUTDATED_ERROR) {
-			/* Ignore error return, we want to propagate
-			   REFTABLE_OUTDATED_ERROR.
-			*/
-			reftable_stack_reload(st);
-		}
 		return err;
 	}
 
@@ -843,8 +830,7 @@ int reftable_addition_commit(struct reftable_addition *add)
 
 int reftable_stack_new_addition(struct reftable_addition **dest,
 				struct reftable_stack *st,
-				const struct reftable_write_options *opts,
-				unsigned int flags)
+				const struct reftable_write_options *opts)
 {
 	int err;
 
@@ -852,7 +838,7 @@ int reftable_stack_new_addition(struct reftable_addition **dest,
 	if (!*dest)
 		return REFTABLE_OUT_OF_MEMORY_ERROR;
 
-	err = reftable_stack_init_addition(*dest, st, opts, flags);
+	err = reftable_stack_init_addition(*dest, st, opts);
 	if (err) {
 		reftable_free(*dest);
 		*dest = NULL;
@@ -1840,12 +1826,7 @@ static int reftable_stack_clean_locked(struct reftable_stack *st)
 int reftable_stack_clean(struct reftable_stack *st)
 {
 	struct reftable_addition *add = NULL;
-	int err = reftable_stack_new_addition(&add, st, NULL, 0);
-	if (err < 0) {
-		goto done;
-	}
-
-	err = reftable_stack_reload(st);
+	int err = reftable_stack_new_addition(&add, st, NULL);
 	if (err < 0) {
 		goto done;
 	}
diff --git a/t/unit-tests/u-reftable-stack.c b/t/unit-tests/u-reftable-stack.c
index e6c1635940..c6254190e6 100644
--- a/t/unit-tests/u-reftable-stack.c
+++ b/t/unit-tests/u-reftable-stack.c
@@ -127,7 +127,7 @@ static void write_n_ref_tables(struct reftable_stack *st,
 		cl_reftable_set_hash(ref.value.val1, i, REFTABLE_HASH_SHA1);
 
 		cl_assert_equal_i(reftable_stack_add(st,
-						     &write_test_ref, &ref, &opts, 0), 0);
+						     &write_test_ref, &ref, &opts), 0);
 	}
 }
 
@@ -168,7 +168,7 @@ void test_reftable_stack__add_one(void)
 	err = reftable_new_stack(&st, dir, NULL);
 	cl_assert(!err);
 
-	err = reftable_stack_add(st, write_test_ref, &ref, &opts, 0);
+	err = reftable_stack_add(st, write_test_ref, &ref, &opts);
 	cl_assert(!err);
 
 	err = reftable_stack_read_ref(st, ref.refname, &dest);
@@ -231,12 +231,9 @@ void test_reftable_stack__uptodate(void)
 	cl_assert_equal_i(reftable_new_stack(&st1, dir, NULL), 0);
 	cl_assert_equal_i(reftable_new_stack(&st2, dir, NULL), 0);
 	cl_assert_equal_i(reftable_stack_add(st1, write_test_ref,
-					     &ref1, NULL, 0), 0);
+					     &ref1, NULL), 0);
 	cl_assert_equal_i(reftable_stack_add(st2, write_test_ref,
-					     &ref2, NULL, 0), REFTABLE_OUTDATED_ERROR);
-	cl_assert_equal_i(reftable_stack_reload(st2), 0);
-	cl_assert_equal_i(reftable_stack_add(st2, write_test_ref,
-					     &ref2, NULL, 0), 0);
+					     &ref2, NULL), 0);
 	reftable_stack_destroy(st1);
 	reftable_stack_destroy(st2);
 	clear_dir(dir);
@@ -260,7 +257,7 @@ void test_reftable_stack__transaction_api(void)
 
 	reftable_addition_destroy(add);
 
-	cl_assert_equal_i(reftable_stack_new_addition(&add, st, NULL, 0), 0);
+	cl_assert_equal_i(reftable_stack_new_addition(&add, st, NULL), 0);
 	cl_assert_equal_i(reftable_addition_add(add, write_test_ref,
 						&ref), 0);
 	cl_assert_equal_i(reftable_addition_commit(add), 0);
@@ -301,21 +298,17 @@ void test_reftable_stack__transaction_with_reload(void)
 
 	cl_assert_equal_i(reftable_new_stack(&st1, dir, NULL), 0);
 	cl_assert_equal_i(reftable_new_stack(&st2, dir, NULL), 0);
-	cl_assert_equal_i(reftable_stack_new_addition(&add, st1, NULL, 0), 0);
+	cl_assert_equal_i(reftable_stack_new_addition(&add, st1, NULL), 0);
 	cl_assert_equal_i(reftable_addition_add(add, write_test_ref,
 						&refs[0]), 0);
 	cl_assert_equal_i(reftable_addition_commit(add), 0);
 	reftable_addition_destroy(add);
 
 	/*
-	 * The second stack is now outdated, which we should notice. We do not
-	 * create the addition and lock the stack by default, but allow the
-	 * reload to happen when REFTABLE_STACK_NEW_ADDITION_RELOAD is set.
+	 * The second stack is now outdated, but it should automatically reload it
+	 * with the newer updates.
 	 */
-	cl_assert_equal_i(reftable_stack_new_addition(&add, st2, NULL, 0),
-						      REFTABLE_OUTDATED_ERROR);
-	cl_assert_equal_i(reftable_stack_new_addition(&add, st2, NULL,
-						      REFTABLE_STACK_NEW_ADDITION_RELOAD), 0);
+	cl_assert_equal_i(reftable_stack_new_addition(&add, st2, NULL), 0);
 	cl_assert_equal_i(reftable_addition_add(add, write_test_ref,
 						&refs[1]), 0);
 	cl_assert_equal_i(reftable_addition_commit(add), 0);
@@ -363,7 +356,7 @@ void test_reftable_stack__transaction_api_performs_auto_compaction(void)
 		 * better control over when exactly auto compaction runs.
 		 */
 		cl_assert_equal_i(reftable_stack_new_addition(&add,
-							      st, &write_opts, 0), 0);
+							      st, &write_opts), 0);
 		cl_assert_equal_i(reftable_addition_add(add,
 							write_test_ref, &ref), 0);
 		cl_assert_equal_i(reftable_addition_commit(add), 0);
@@ -400,7 +393,7 @@ void test_reftable_stack__auto_compaction_fails_gracefully(void)
 
 	cl_assert_equal_i(reftable_new_stack(&st, dir, NULL), 0);
 	cl_assert_equal_i(reftable_stack_add(st, write_test_ref,
-					     &ref, NULL, 0), 0);
+					     &ref, NULL), 0);
 	cl_assert_equal_i(st->merged->tables_len, 1);
 	cl_assert_equal_i(st->stats.attempts, 0);
 	cl_assert_equal_i(st->stats.failures, 0);
@@ -418,7 +411,7 @@ void test_reftable_stack__auto_compaction_fails_gracefully(void)
 	write_file_buf(table_path.buf, "", 0);
 
 	ref.update_index = 2;
-	err = reftable_stack_add(st, write_test_ref, &ref, NULL, 0);
+	err = reftable_stack_add(st, write_test_ref, &ref, NULL);
 	cl_assert(!err);
 	cl_assert_equal_i(st->merged->tables_len, 2);
 	cl_assert_equal_i(st->stats.attempts, 1);
@@ -453,9 +446,9 @@ void test_reftable_stack__update_index_check(void)
 
 	cl_assert_equal_i(reftable_new_stack(&st, dir, NULL), 0);
 	cl_assert_equal_i(reftable_stack_add(st, write_test_ref,
-					     &ref1, NULL, 0), 0);
+					     &ref1, NULL), 0);
 	cl_assert_equal_i(reftable_stack_add(st, write_test_ref,
-					     &ref2, NULL, 0), REFTABLE_API_ERROR);
+					     &ref2, NULL), REFTABLE_API_ERROR);
 	reftable_stack_destroy(st);
 	clear_dir(dir);
 }
@@ -469,7 +462,7 @@ void test_reftable_stack__lock_failure(void)
 	cl_assert_equal_i(reftable_new_stack(&st, dir, NULL), 0);
 	for (i = -1; i != REFTABLE_EMPTY_TABLE_ERROR; i--)
 		cl_assert_equal_i(reftable_stack_add(st, write_error,
-						     &i, NULL, 0), i);
+						     &i, NULL), i);
 
 	reftable_stack_destroy(st);
 	clear_dir(dir);
@@ -513,7 +506,7 @@ void test_reftable_stack__add(void)
 
 	for (i = 0; i < N; i++)
 		cl_assert_equal_i(reftable_stack_add(st, write_test_ref,
-						     &refs[i], &opts, 0), 0);
+						     &refs[i], &opts), 0);
 
 	for (i = 0; i < N; i++) {
 		struct write_log_arg arg = {
@@ -521,7 +514,7 @@ void test_reftable_stack__add(void)
 			.update_index = reftable_stack_next_update_index(st),
 		};
 		cl_assert_equal_i(reftable_stack_add(st, write_test_log,
-						     &arg, &opts, 0), 0);
+						     &arg, &opts), 0);
 	}
 
 	cl_assert_equal_i(reftable_stack_compact_all(st, &opts, NULL), 0);
@@ -604,7 +597,7 @@ void test_reftable_stack__iterator(void)
 
 	for (i = 0; i < N; i++)
 		cl_assert_equal_i(reftable_stack_add(st, write_test_ref,
-						     &refs[i], NULL, 0), 0);
+						     &refs[i], NULL), 0);
 
 	for (i = 0; i < N; i++) {
 		struct write_log_arg arg = {
@@ -613,7 +606,7 @@ void test_reftable_stack__iterator(void)
 		};
 
 		cl_assert_equal_i(reftable_stack_add(st, write_test_log,
-						     &arg, NULL, 0), 0);
+						     &arg, NULL), 0);
 	}
 
 	reftable_stack_init_ref_iterator(st, &it);
@@ -685,11 +678,11 @@ void test_reftable_stack__log_normalize(void)
 
 	input.value.update.message = (char *) "one\ntwo";
 	cl_assert_equal_i(reftable_stack_add(st, write_test_log,
-					     &arg, NULL, 0), REFTABLE_API_ERROR);
+					     &arg, NULL), REFTABLE_API_ERROR);
 
 	input.value.update.message = (char *) "one";
 	cl_assert_equal_i(reftable_stack_add(st, write_test_log,
-					     &arg, NULL, 0), 0);
+					     &arg, NULL), 0);
 	cl_assert_equal_i(reftable_stack_read_log(st, input.refname,
 						  &dest), 0);
 	cl_assert_equal_s(dest.value.update.message, "one\n");
@@ -697,7 +690,7 @@ void test_reftable_stack__log_normalize(void)
 	input.value.update.message = (char *) "two\n";
 	arg.update_index = 2;
 	cl_assert_equal_i(reftable_stack_add(st, write_test_log,
-					     &arg, NULL, 0), 0);
+					     &arg, NULL), 0);
 	cl_assert_equal_i(reftable_stack_read_log(st, input.refname,
 						  &dest), 0);
 	cl_assert_equal_s(dest.value.update.message, "two\n");
@@ -747,7 +740,7 @@ void test_reftable_stack__tombstone(void)
 	}
 	for (i = 0; i < N; i++)
 		cl_assert_equal_i(reftable_stack_add(st, write_test_ref,
-						     &refs[i], NULL, 0), 0);
+						     &refs[i], NULL), 0);
 
 	for (i = 0; i < N; i++) {
 		struct write_log_arg arg = {
@@ -755,7 +748,7 @@ void test_reftable_stack__tombstone(void)
 			.update_index = reftable_stack_next_update_index(st),
 		};
 		cl_assert_equal_i(reftable_stack_add(st, write_test_log,
-						     &arg, NULL, 0), 0);
+						     &arg, NULL), 0);
 	}
 
 	cl_assert_equal_i(reftable_stack_read_ref(st, "branch",
@@ -801,7 +794,7 @@ void test_reftable_stack__hash_id(void)
 
 	cl_assert_equal_i(reftable_new_stack(&st, dir, NULL), 0);
 	cl_assert_equal_i(reftable_stack_add(st, write_test_ref,
-					     &ref, NULL, 0), 0);
+					     &ref, NULL), 0);
 
 	/* can't read it with the wrong hash ID. */
 	cl_assert_equal_i(reftable_new_stack(&st32, dir,
@@ -869,7 +862,7 @@ void test_reftable_stack__reflog_expire(void)
 			.update_index = reftable_stack_next_update_index(st),
 		};
 		cl_assert_equal_i(reftable_stack_add(st, write_test_log,
-						     &arg, NULL, 0), 0);
+						     &arg, NULL), 0);
 	}
 
 	cl_assert_equal_i(reftable_stack_compact_all(st, NULL, NULL), 0);
@@ -908,7 +901,7 @@ void test_reftable_stack__empty_add(void)
 
 	cl_assert_equal_i(reftable_new_stack(&st, dir, NULL), 0);
 	cl_assert_equal_i(reftable_stack_add(st, write_nothing,
-					     NULL, NULL, 0), 0);
+					     NULL, NULL), 0);
 	cl_assert_equal_i(reftable_new_stack(&st2, dir, NULL), 0);
 	clear_dir(dir);
 	reftable_stack_destroy(st);
@@ -947,7 +940,7 @@ void test_reftable_stack__auto_compaction(void)
 		};
 		snprintf(name, sizeof(name), "branch%04"PRIuMAX, (uintmax_t)i);
 
-		err = reftable_stack_add(st, write_test_ref, &ref, &opts, 0);
+		err = reftable_stack_add(st, write_test_ref, &ref, &opts);
 		cl_assert(!err);
 
 		err = reftable_stack_auto_compact(st, &opts);
@@ -983,7 +976,7 @@ void test_reftable_stack__auto_compaction_factor(void)
 		};
 		xsnprintf(name, sizeof(name), "branch%04"PRIuMAX, (uintmax_t)i);
 
-		err = reftable_stack_add(st, &write_test_ref, &ref, &opts, 0);
+		err = reftable_stack_add(st, &write_test_ref, &ref, &opts);
 		cl_assert(!err);
 
 		cl_assert(i < 5 || st->merged->tables_len < 5 * fastlogN(i, 5));
@@ -1064,7 +1057,7 @@ void test_reftable_stack__add_performs_auto_compaction(void)
 		ref.refname = buf;
 
 		cl_assert_equal_i(reftable_stack_add(st, write_test_ref,
-						     &ref, &write_opts, 0), 0);
+						     &ref, &write_opts), 0);
 
 		/*
 		 * The stack length should grow continuously for all runs where
@@ -1303,7 +1296,7 @@ void test_reftable_stack__invalid_limit_updates(void)
 
 	reftable_addition_destroy(add);
 
-	cl_assert_equal_i(reftable_stack_new_addition(&add, st, &opts, 0), 0);
+	cl_assert_equal_i(reftable_stack_new_addition(&add, st, &opts), 0);
 
 	/*
 	 * write_limits_after_ref also updates the update indexes after adding

-- 
2.55.GIT

[PATCH 2/3] reftable/stack: move list lock to `struct reftable_stack`

[Karthik Nayak]

The struct `reftable_addition` is used to modify a given stack, as such,
it also includes a `struct reftable_flock` used to obtain the lock to
the list file. While the scope of the field lies within this struct, it
doesn't allow for optimizations to be made on `struct reftable_stack`
itself.

Move the field to `struct reftable_stack`, allowing us to make a simple
optimization around avoiding a stack reload when we have already
obtained a lock. While this is currently possible in the write path, the
write path also contains multiple branches to reads which only work
on top of `struct reftable_stack`, and we would miss the optimization in
such paths.

While here, remove an unused header file from 'reftable/stack.h'.

Signed-off-by: Karthik Nayak <karthik.188@gmail.com>
---
 reftable/stack.c | 15 ++++++++-------
 reftable/stack.h |  7 ++++++-
 2 files changed, 14 insertions(+), 8 deletions(-)

diff --git a/reftable/stack.c b/reftable/stack.c
index 540f5e77ac..e449af9c03 100644
--- a/reftable/stack.c
+++ b/reftable/stack.c
@@ -536,6 +536,8 @@ int reftable_new_stack(struct reftable_stack **dest, const char *dir,
 		goto out;
 	}
 
+	p->list_lock = REFTABLE_FLOCK_INIT;
+
 	err = reftable_stack_reload_maybe_reuse(p, 1);
 	if (err < 0)
 		goto out;
@@ -628,7 +630,6 @@ int reftable_stack_reload(struct reftable_stack *st)
 }
 
 struct reftable_addition {
-	struct reftable_flock tables_list_lock;
 	struct reftable_stack *stack;
 	struct reftable_write_options opts;
 
@@ -653,7 +654,7 @@ static void reftable_addition_close(struct reftable_addition *add)
 	add->new_tables_len = 0;
 	add->new_tables_cap = 0;
 
-	flock_release(&add->tables_list_lock);
+	flock_release(&add->stack->list_lock);
 	reftable_buf_release(&nm);
 }
 
@@ -669,13 +670,13 @@ static int reftable_stack_init_addition(struct reftable_addition *add,
 	if (opts)
 		add->opts = *opts;
 
-	err = flock_acquire(&add->tables_list_lock, st->list_file,
+	err = flock_acquire(&add->stack->list_lock, st->list_file,
 			    add->opts.lock_timeout_ms);
 	if (err < 0)
 		goto done;
 
 	if (add->opts.default_permissions) {
-		if (chmod(add->tables_list_lock.path,
+		if (chmod(add->stack->list_lock.path,
 			  add->opts.default_permissions) < 0) {
 			err = REFTABLE_IO_ERROR;
 			goto done;
@@ -774,7 +775,7 @@ int reftable_addition_commit(struct reftable_addition *add)
 			goto done;
 	}
 
-	err = reftable_write_data(add->tables_list_lock.fd,
+	err = reftable_write_data(add->stack->list_lock.fd,
 				  table_list.buf, table_list.len);
 	reftable_buf_release(&table_list);
 	if (err < 0) {
@@ -782,13 +783,13 @@ int reftable_addition_commit(struct reftable_addition *add)
 		goto done;
 	}
 
-	err = fsync(add->tables_list_lock.fd);
+	err = fsync(add->stack->list_lock.fd);
 	if (err < 0) {
 		err = REFTABLE_IO_ERROR;
 		goto done;
 	}
 
-	err = flock_commit(&add->tables_list_lock);
+	err = flock_commit(&add->stack->list_lock);
 	if (err < 0) {
 		err = REFTABLE_IO_ERROR;
 		goto done;
diff --git a/reftable/stack.h b/reftable/stack.h
index f7901e6c6f..52e07ad551 100644
--- a/reftable/stack.h
+++ b/reftable/stack.h
@@ -10,7 +10,6 @@
 #define STACK_H
 
 #include "system.h"
-#include "reftable-writer.h"
 #include "reftable-stack.h"
 
 struct reftable_stack {
@@ -18,6 +17,12 @@ struct reftable_stack {
 	char *list_file;
 	int list_fd;
 
+	/*
+	 * Set while an addition holds the stack locked. Used by
+	 * stack_uptodate() to skip reload checks while locked.
+	 */
+	struct reftable_flock list_lock;
+
 	char *reftable_dir;
 
 	struct reftable_stack_options opts;

-- 
2.55.GIT

[PATCH 3/3] reftable/stack: avoid reloading the stack when already locked

[Karthik Nayak]

When making modifications to the reftable stack, the stack obtains a
lock to the list file and removes the lock after the commit phase. Since
most operations reload the stack to ensure we have the latest state, any
branched operation during the locked phase could trigger a state reload.

To prevent data loss due to concurrent writes, state reload is necessary
right after obtaining the lock. But any reloads after that are just a
no-op. Now that the struct has access to the lock file status, simply
skip reloading if the lock is present.

Benchmarking with a fixed, non-symbolic target OID shows a modest but
consistent ~1-2% improvement in clock time for `update-ref` across ref
counts ranging from 2,000 to 100,000.

We can see better improvements in the number of syscall counts. On
master, the number of calls to `newfstatat()` grows linearly with the
number of refs created. With this patch, the number is now a constant:

  refcount   master   patch
  --------   ------   ------
  1,000      1,059       55
  5,000      5,059       55
  10,000     10,059      55
  20,000     20,059      55

Reported-by: Jeff King <peff@peff.net>
Signed-off-by: Karthik Nayak <karthik.188@gmail.com>
---
 reftable/stack.c | 17 ++++++++++++-----
 1 file changed, 12 insertions(+), 5 deletions(-)

diff --git a/reftable/stack.c b/reftable/stack.c
index e449af9c03..433a611ed1 100644
--- a/reftable/stack.c
+++ b/reftable/stack.c
@@ -553,14 +553,21 @@ int reftable_new_stack(struct reftable_stack **dest, const char *dir,
 
 /*
  * Check whether the given stack is up-to-date with what we have in memory.
+ * If skip_if_locked is set skip stack reloading if the stack is currently
+ * locked. Stack reloading must _not_ be skipped right after obtaining the
+ * lock, to check for concurrent updates which may have happened.
+ *
  * Returns 0 if so, 1 if the stack is out-of-date or a negative error code
  * otherwise.
  */
-static int stack_uptodate(struct reftable_stack *st)
+static int stack_uptodate(struct reftable_stack *st, int skip_if_locked)
 {
 	char **names = NULL;
 	int err;
 
+	if (skip_if_locked && st->list_lock.fd != -1)
+		return 0;
+
 	/*
 	 * When we have cached stat information available then we use it to
 	 * verify whether the file has been rewritten.
@@ -623,7 +630,7 @@ static int stack_uptodate(struct reftable_stack *st)
 
 int reftable_stack_reload(struct reftable_stack *st)
 {
-	int err = stack_uptodate(st);
+	int err = stack_uptodate(st, 1);
 	if (err > 0)
 		return reftable_stack_reload_maybe_reuse(st, 1);
 	return err;
@@ -683,7 +690,7 @@ static int reftable_stack_init_addition(struct reftable_addition *add,
 		}
 	}
 
-	err = stack_uptodate(st);
+	err = stack_uptodate(st, 0);
 	if (err < 0)
 		goto done;
 	if (err > 0) {
@@ -1189,7 +1196,7 @@ static int stack_compact_range(struct reftable_stack *st,
 	 * we could check that relevant tables still exist. But for now it's
 	 * good enough to just abort.
 	 */
-	err = stack_uptodate(st);
+	err = stack_uptodate(st, 0);
 	if (err < 0)
 		goto done;
 	if (err > 0) {
@@ -1308,7 +1315,7 @@ static int stack_compact_range(struct reftable_stack *st,
 	 * tables with our compacted version. If they don't, then we need to
 	 * abort.
 	 */
-	err = stack_uptodate(st);
+	err = stack_uptodate(st, 0);
 	if (err < 0)
 		goto done;
 	if (err > 0) {

-- 
2.55.GIT

Re: [PATCH 1/3] reftable/stack: remove `REFTABLE_STACK_NEW_ADDITION_RELOAD`

[Justin Tobler]

On 26/08/19 03:19PM, Karthik Nayak wrote:
> In 80e7342ea8 (reftable/stack: allow locking of outdated stacks,
> 2024-09-24), the `REFTABLE_STACK_NEW_ADDITION_RELOAD` was introduced so
> that callers of `reftable_stack_init_addition()` can also reload the
> stack if there was a concurrent update made before the lock was
> obtained.
> 
> Then 16684b6fae (refs/reftable: always reload stacks when creating
> lock, 2025-08-12) updated all of the remaining call-sites to propagate
> this flag to ensure that we always reload the stack whenever there was a
> concurrent update.

Ok, if all call sites already wire this flag, then we probably don't
need if anymore.

> As all calls to `reftable_stack_init_addition()` inevitably propagate
> the flag, it is safe to remove the flag and its associated code and make
> the reloading of the stack the default flow. This makes it easier to
> follow the flow and simplifies the logic.

Makes sense.

> The only exceptions are:
> 
>   1. Unit tests, where we explicitly do not propagate the flag. These
>      tests are now modified with the new status quo.

I assume this means we no longer need to test for the case where we
don't reload.

>   2. `reftable_stack_clean_locked()`, which was propagating 0 to

Did you mean `reftable_stack_clean()`?

>      `reftable_stack_new_addition()` but was then manually reloading the
>      stack after. Here the new flow will achieve the same, while also
>      allowing us to remove the manual reload.

Out of curiousity, was this call site just forgotten previously? Or was
there any reason a manual reload was useful?

> This also makes two checks for 'REFTABLE_OUTDATED_ERROR' redundant, so
> remove them also.
> 
> Signed-off-by: Karthik Nayak <karthik.188@gmail.com>
> ---
[snip]
> diff --git a/reftable/reftable-stack.h b/reftable/reftable-stack.h
> index 5d22d84e80..5d224f8079 100644
> --- a/reftable/reftable-stack.h
> +++ b/reftable/reftable-stack.h
> @@ -58,22 +58,13 @@ uint64_t reftable_stack_next_update_index(struct reftable_stack *st);
>  /* holds a transaction to add tables at the top of a stack. */
>  struct reftable_addition;
>  
> -enum {
> -	/*
> -	 * Reload the stack when the stack is out-of-date after locking it.
> -	 */
> -	REFTABLE_STACK_NEW_ADDITION_RELOAD = (1 << 0),
> -};

The flag is dropped now that it is the only behavior.

>  /*
>   * returns a new transaction to add reftables to the given stack. As a side
> - * effect, the ref database is locked. Accepts REFTABLE_STACK_NEW_ADDITION_*
> - * flags.
> + * effect, the ref database is locked.
>   */
>  int reftable_stack_new_addition(struct reftable_addition **dest,
>  				struct reftable_stack *st,
> -				const struct reftable_write_options *opts,
> -				unsigned int flags);
> +				const struct reftable_write_options *opts);

Signatures updated. Ok.

[snip]
> diff --git a/reftable/stack.c b/reftable/stack.c
> index 308f9578f0..540f5e77ac 100644
> --- a/reftable/stack.c
> +++ b/reftable/stack.c
> @@ -659,8 +659,7 @@ static void reftable_addition_close(struct reftable_addition *add)
>  
>  static int reftable_stack_init_addition(struct reftable_addition *add,
>  					struct reftable_stack *st,
> -					const struct reftable_write_options *opts,
> -					unsigned int flags)
> +					const struct reftable_write_options *opts)
>  {
>  	struct reftable_buf lock_file_name = REFTABLE_BUF_INIT;
>  	int err;
> @@ -686,15 +685,11 @@ static int reftable_stack_init_addition(struct reftable_addition *add,
>  	err = stack_uptodate(st);
>  	if (err < 0)
>  		goto done;
> -	if (err > 0 && flags & REFTABLE_STACK_NEW_ADDITION_RELOAD) {
> +	if (err > 0) {
>  		err = reftable_stack_reload_maybe_reuse(add->stack, 1);
>  		if (err)
>  			goto done;
>  	}
> -	if (err > 0) {
> -		err = REFTABLE_OUTDATED_ERROR;
> -		goto done;
> -	}

`reftable_stack_init_addition()` now reload unconditionally. Looks good.

>  	add->next_update_index = reftable_stack_next_update_index(st);
>  done:
> @@ -708,13 +703,12 @@ static int stack_try_add(struct reftable_stack *st,
>  			 int (*write_table)(struct reftable_writer *wr,
>  					    void *arg),
>  			 void *arg,
> -			 const struct reftable_write_options *opts,
> -			 unsigned flags)
> +			 const struct reftable_write_options *opts)
>  {
>  	struct reftable_addition add;
>  	int err;
>  
> -	err = reftable_stack_init_addition(&add, st, opts, flags);
> +	err = reftable_stack_init_addition(&add, st, opts);
>  	if (err < 0)
>  		goto done;
>  
> @@ -731,17 +725,10 @@ static int stack_try_add(struct reftable_stack *st,
>  int reftable_stack_add(struct reftable_stack *st,
>  		       int (*write)(struct reftable_writer *wr, void *arg),
>  		       void *arg,
> -		       const struct reftable_write_options *opts,
> -		       unsigned flags)
> +		       const struct reftable_write_options *opts)
>  {
> -	int err = stack_try_add(st, write, arg, opts, flags);
> +	int err = stack_try_add(st, write, arg, opts);
>  	if (err < 0) {
> -		if (err == REFTABLE_OUTDATED_ERROR) {
> -			/* Ignore error return, we want to propagate
> -			   REFTABLE_OUTDATED_ERROR.
> -			*/
> -			reftable_stack_reload(st);
> -		}

Since we always reload now, the REFTABLE_OUTDATED_ERROR is no longer a
possibility and doesn't need to be handled anymore.

[snip]
> diff --git a/t/unit-tests/u-reftable-stack.c b/t/unit-tests/u-reftable-stack.c
> index e6c1635940..c6254190e6 100644
> --- a/t/unit-tests/u-reftable-stack.c
> +++ b/t/unit-tests/u-reftable-stack.c
> @@ -127,7 +127,7 @@ static void write_n_ref_tables(struct reftable_stack *st,
>  		cl_reftable_set_hash(ref.value.val1, i, REFTABLE_HASH_SHA1);
>  
>  		cl_assert_equal_i(reftable_stack_add(st,
> -						     &write_test_ref, &ref, &opts, 0), 0);
> +						     &write_test_ref, &ref, &opts), 0);
>  	}
>  }
>  
> @@ -168,7 +168,7 @@ void test_reftable_stack__add_one(void)
>  	err = reftable_new_stack(&st, dir, NULL);
>  	cl_assert(!err);
>  
> -	err = reftable_stack_add(st, write_test_ref, &ref, &opts, 0);
> +	err = reftable_stack_add(st, write_test_ref, &ref, &opts);
>  	cl_assert(!err);
>  
>  	err = reftable_stack_read_ref(st, ref.refname, &dest);
> @@ -231,12 +231,9 @@ void test_reftable_stack__uptodate(void)
>  	cl_assert_equal_i(reftable_new_stack(&st1, dir, NULL), 0);
>  	cl_assert_equal_i(reftable_new_stack(&st2, dir, NULL), 0);
>  	cl_assert_equal_i(reftable_stack_add(st1, write_test_ref,
> -					     &ref1, NULL, 0), 0);
> +					     &ref1, NULL), 0);
>  	cl_assert_equal_i(reftable_stack_add(st2, write_test_ref,
> -					     &ref2, NULL, 0), REFTABLE_OUTDATED_ERROR);
> -	cl_assert_equal_i(reftable_stack_reload(st2), 0);
> -	cl_assert_equal_i(reftable_stack_add(st2, write_test_ref,
> -					     &ref2, NULL, 0), 0);
> +					     &ref2, NULL), 0);

We no longer need to check for REFTABLE_OUTDATED_ERROR since the stack
is always reloaded now. Makes sense.

>  	reftable_stack_destroy(st1);
>  	reftable_stack_destroy(st2);
>  	clear_dir(dir);
> @@ -260,7 +257,7 @@ void test_reftable_stack__transaction_api(void)
>  
>  	reftable_addition_destroy(add);
>  
> -	cl_assert_equal_i(reftable_stack_new_addition(&add, st, NULL, 0), 0);
> +	cl_assert_equal_i(reftable_stack_new_addition(&add, st, NULL), 0);
>  	cl_assert_equal_i(reftable_addition_add(add, write_test_ref,
>  						&ref), 0);
>  	cl_assert_equal_i(reftable_addition_commit(add), 0);
> @@ -301,21 +298,17 @@ void test_reftable_stack__transaction_with_reload(void)
>  
>  	cl_assert_equal_i(reftable_new_stack(&st1, dir, NULL), 0);
>  	cl_assert_equal_i(reftable_new_stack(&st2, dir, NULL), 0);
> -	cl_assert_equal_i(reftable_stack_new_addition(&add, st1, NULL, 0), 0);
> +	cl_assert_equal_i(reftable_stack_new_addition(&add, st1, NULL), 0);
>  	cl_assert_equal_i(reftable_addition_add(add, write_test_ref,
>  						&refs[0]), 0);
>  	cl_assert_equal_i(reftable_addition_commit(add), 0);
>  	reftable_addition_destroy(add);
>  
>  	/*
> -	 * The second stack is now outdated, which we should notice. We do not
> -	 * create the addition and lock the stack by default, but allow the
> -	 * reload to happen when REFTABLE_STACK_NEW_ADDITION_RELOAD is set.
> +	 * The second stack is now outdated, but it should automatically reload it
> +	 * with the newer updates.
>  	 */
> -	cl_assert_equal_i(reftable_stack_new_addition(&add, st2, NULL, 0),
> -						      REFTABLE_OUTDATED_ERROR);
> -	cl_assert_equal_i(reftable_stack_new_addition(&add, st2, NULL,
> -						      REFTABLE_STACK_NEW_ADDITION_RELOAD), 0);
> +	cl_assert_equal_i(reftable_stack_new_addition(&add, st2, NULL), 0);

Same here.

The rest of this patch is just updating call sites and looks good.

-Justin

Re: [PATCH 2/3] reftable/stack: move list lock to `struct reftable_stack`

[Justin Tobler]

On 26/08/19 03:19PM, Karthik Nayak wrote:
> The struct `reftable_addition` is used to modify a given stack, as such,
> it also includes a `struct reftable_flock` used to obtain the lock to
> the list file. While the scope of the field lies within this struct, it
> doesn't allow for optimizations to be made on `struct reftable_stack`
> itself.

Hmmm IIUC, there can only be a single lock for the reftable stack
correct? If that is the case, it sounds like `struct reftable_stack` may
conceptually be the better place for the field regardless.

> Move the field to `struct reftable_stack`, allowing us to make a simple
> optimization around avoiding a stack reload when we have already
> obtained a lock. While this is currently possible in the write path, the
> write path also contains multiple branches to reads which only work
> on top of `struct reftable_stack`, and we would miss the optimization in
> such paths.

Ok, so if we know the reftable stack is alreay locked, there is no need
to reload it since it can't change. Makes sense.

> While here, remove an unused header file from 'reftable/stack.h'.
> 
> Signed-off-by: Karthik Nayak <karthik.188@gmail.com>
> ---
[snip]
>  struct reftable_stack {
> @@ -18,6 +17,12 @@ struct reftable_stack {
>  	char *list_file;
>  	int list_fd;
>  
> +	/*
> +	 * Set while an addition holds the stack locked. Used by
> +	 * stack_uptodate() to skip reload checks while locked.
> +	 */
> +	struct reftable_flock list_lock;
> +

As mentioned in the log message, the lock is now tracked in `struct
reftable_stack` and the rest of this patch just wires it accordingly.
Looks good.

-Justin

Re: [PATCH 3/3] reftable/stack: avoid reloading the stack when already locked

[Justin Tobler]

On 26/08/19 03:19PM, Karthik Nayak wrote:
> When making modifications to the reftable stack, the stack obtains a
> lock to the list file and removes the lock after the commit phase. Since
> most operations reload the stack to ensure we have the latest state, any
> branched operation during the locked phase could trigger a state reload.
> 
> To prevent data loss due to concurrent writes, state reload is necessary
> right after obtaining the lock. But any reloads after that are just a
> no-op. Now that the struct has access to the lock file status, simply
> skip reloading if the lock is present.

Makes sense.

> Benchmarking with a fixed, non-symbolic target OID shows a modest but
> consistent ~1-2% improvement in clock time for `update-ref` across ref
> counts ranging from 2,000 to 100,000.
> 
> We can see better improvements in the number of syscall counts. On
> master, the number of calls to `newfstatat()` grows linearly with the
> number of refs created. With this patch, the number is now a constant:
> 
>   refcount   master   patch
>   --------   ------   ------
>   1,000      1,059       55
>   5,000      5,059       55
>   10,000     10,059      55
>   20,000     20,059      55
> 
> Reported-by: Jeff King <peff@peff.net>
> Signed-off-by: Karthik Nayak <karthik.188@gmail.com>
> ---
>  reftable/stack.c | 17 ++++++++++++-----
>  1 file changed, 12 insertions(+), 5 deletions(-)
> 
> diff --git a/reftable/stack.c b/reftable/stack.c
> index e449af9c03..433a611ed1 100644
> --- a/reftable/stack.c
> +++ b/reftable/stack.c
> @@ -553,14 +553,21 @@ int reftable_new_stack(struct reftable_stack **dest, const char *dir,
>  
>  /*
>   * Check whether the given stack is up-to-date with what we have in memory.
> + * If skip_if_locked is set skip stack reloading if the stack is currently
> + * locked. Stack reloading must _not_ be skipped right after obtaining the
> + * lock, to check for concurrent updates which may have happened.
> + *
>   * Returns 0 if so, 1 if the stack is out-of-date or a negative error code
>   * otherwise.
>   */
> -static int stack_uptodate(struct reftable_stack *st)
> +static int stack_uptodate(struct reftable_stack *st, int skip_if_locked)
>  {
>  	char **names = NULL;
>  	int err;
>  
> +	if (skip_if_locked && st->list_lock.fd != -1)
> +		return 0;
> +
>  	/*
>  	 * When we have cached stat information available then we use it to
>  	 * verify whether the file has been rewritten.
> @@ -623,7 +630,7 @@ static int stack_uptodate(struct reftable_stack *st)
>  
>  int reftable_stack_reload(struct reftable_stack *st)
>  {
> -	int err = stack_uptodate(st);
> +	int err = stack_uptodate(st, 1);

Ok, this appears to be the only call site where is actually want to skip
if there is a lock present. Could we instead just not invoke
`stack_uptodate()` in such cases? That way we don't have to change its
function signature and can leave all other existing call sites alone.

-Justin

Re: [PATCH 2/3] reftable/stack: move list lock to `struct reftable_stack`

[Junio C Hamano]

Karthik Nayak <karthik.188@gmail.com> writes:

> The struct `reftable_addition` is used to modify a given stack, as such,
> it also includes a `struct reftable_flock` used to obtain the lock to
> the list file. While the scope of the field lies within this struct, it
> doesn't allow for optimizations to be made on `struct reftable_stack`
> itself.
>
> Move the field to `struct reftable_stack`, allowing us to make a simple
> optimization around avoiding a stack reload when we have already
> obtained a lock. While this is currently possible in the write path, the
> write path also contains multiple branches to reads which only work
> on top of `struct reftable_stack`, and we would miss the optimization in
> such paths.

As long as nobody tries to open a nested or concurrent addition on
the same 'struct reftable_stack', this should be safe, but do we
give enough tools to help the API users avoid doing so?

I may be misreading the code completely, but when a caller already
holds a lock after calling reftable_stack_init_addition() on an
instance of reftable_stack, and then adds another reftable_addition
on the same reftable_stack, flock_acquire(add->stack->list_lock)
would fail because the lock is per stack now, unlike the original
code where the lock was per reftable_addition.  We jump to the
done: label and call reftable_addition_close(), which would release
the lock, which is now shared with other reftable_addition
instances that work on the same stack, which in turn would get the
holders of the lock into trouble, no?