Fix _Fresh benchmark consistency and address review nits

- Remove inner loop from CalcAndEncode_DeviceInfoResponse_Fresh to match
  SensorStateResponse_Fresh — both now measure single alloc+encode per
  iteration as intended for heap allocation cost benchmarking
- Add comments documenting why _Fresh variants skip inner loops
- Add first-wins comment to load_component_yaml_configs
- Clean up .gitignore template comment

script/test_helpers.py

@@ -139,6 +139,11 @@ def load_component_yaml_configs(components: list[str], tests_dir: Path) -> dict:
     Returns:
         Merged dict of component configs to add to the base config
     """
+    # Note: components are processed in sorted order. For conflicting keys
+    # (e.g. two benchmark.yaml files both declaring sensor:), the first
+    # component alphabetically wins via setdefault(). This is fine for now
+    # with a single benchmark component (api) but would need a real merge
+    # strategy if multiple components declare overlapping configs.
     merged: dict = {}
     for component in components:
         yaml_path = tests_dir / component / BENCHMARK_YAML_FILENAME

tests/benchmarks/components/.gitignore

@@ -1,5 +1,2 @@
-# Gitignore settings for ESPHome
-# This is an example and may include too much for your use-case.
-# You can modify this file to suit your needs.
 /.esphome/
 /secrets.yaml

tests/benchmarks/components/api/bench_proto_encode.cpp

@@ -70,7 +70,8 @@ static void CalcAndEncode_SensorStateResponse(benchmark::State &state) {
 }
 BENCHMARK(CalcAndEncode_SensorStateResponse);
 
-// Cold path: fresh buffer each iteration (measures heap allocation)
+// Cold path: fresh buffer each iteration (measures heap allocation).
+// No inner loop — the point is to measure one alloc+encode per iteration.
 static void CalcAndEncode_SensorStateResponse_Fresh(benchmark::State &state) {
   SensorStateResponse msg;
   msg.key = 0x12345678;
@@ -268,21 +269,19 @@ static void CalcAndEncode_DeviceInfoResponse(benchmark::State &state) {
 }
 BENCHMARK(CalcAndEncode_DeviceInfoResponse);
 
-// Cold path: fresh buffer each iteration (measures heap allocation)
+// Cold path: fresh buffer each iteration (measures heap allocation).
+// No inner loop — the point is to measure one alloc+encode per iteration.
 static void CalcAndEncode_DeviceInfoResponse_Fresh(benchmark::State &state) {
   auto msg = make_device_info_response();
 
   for (auto _ : state) {
-    for (int i = 0; i < kInnerIterations; i++) {
-      APIBuffer buffer;
-      uint32_t size = msg.calculate_size();
-      buffer.resize(size);
-      ProtoWriteBuffer writer(&buffer, 0);
-      msg.encode(writer);
-      benchmark::DoNotOptimize(buffer.data());
-    }
+    APIBuffer buffer;
+    uint32_t size = msg.calculate_size();
+    buffer.resize(size);
+    ProtoWriteBuffer writer(&buffer, 0);
+    msg.encode(writer);
+    benchmark::DoNotOptimize(buffer.data());
   }
-  state.SetItemsProcessed(state.iterations() * kInnerIterations);
 }
 BENCHMARK(CalcAndEncode_DeviceInfoResponse_Fresh);