Merge branch 'dev' into api-stringref-user-services

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

.github/actions/build-image/action.yaml

@@ -47,7 +47,7 @@ runs:
 
     - name: Build and push to ghcr by digest
       id: build-ghcr
-      uses: docker/build-push-action@601a80b39c9405e50806ae38af30926f9d957c47 # v6.19.1
+      uses: docker/build-push-action@10e90e3645eae34f1e60eeb005ba3a3d33f178e8 # v6.19.2
       env:
         DOCKER_BUILD_SUMMARY: false
         DOCKER_BUILD_RECORD_UPLOAD: false
@@ -73,7 +73,7 @@ runs:
 
     - name: Build and push to dockerhub by digest
       id: build-dockerhub
-      uses: docker/build-push-action@601a80b39c9405e50806ae38af30926f9d957c47 # v6.19.1
+      uses: docker/build-push-action@10e90e3645eae34f1e60eeb005ba3a3d33f178e8 # v6.19.2
       env:
         DOCKER_BUILD_SUMMARY: false
         DOCKER_BUILD_RECORD_UPLOAD: false

docker/Dockerfile

@@ -9,7 +9,8 @@ FROM ghcr.io/esphome/docker-base:${BUILD_OS}-ha-addon-${BUILD_BASE_VERSION} AS b
 ARG BUILD_TYPE
 FROM base-source-${BUILD_TYPE} AS base
 
-RUN git config --system --add safe.directory "*"
+RUN git config --system --add safe.directory "*" \
+    && git config --system advice.detachedHead false
 
 # Install build tools for Python packages that require compilation
 # (e.g., ruamel.yaml.clibz used by ESP-IDF's idf-component-manager)

esphome/automation.py

@@ -57,8 +57,14 @@ def maybe_conf(conf, *validators):
     return validate
 
 
-def register_action(name: str, action_type: MockObjClass, schema: cv.Schema):
+def register_action(
-    return ACTION_REGISTRY.register(name, action_type, schema)
+    name: str,
+    action_type: MockObjClass,
+    schema: cv.Schema,
+    *,
+    deferred: bool = False,
+):
+    return ACTION_REGISTRY.register(name, action_type, schema, deferred=deferred)
 
 
 def register_condition(name: str, condition_type: MockObjClass, schema: cv.Schema):
@@ -335,7 +341,10 @@ async def component_is_idle_condition_to_code(
 
 
 @register_action(
-    "delay", DelayAction, cv.templatable(cv.positive_time_period_milliseconds)
+    "delay",
+    DelayAction,
+    cv.templatable(cv.positive_time_period_milliseconds),
+    deferred=True,
 )
 async def delay_action_to_code(
     config: ConfigType,
@@ -445,7 +454,7 @@ _validate_wait_until = cv.maybe_simple_value(
 )
 
 
-@register_action("wait_until", WaitUntilAction, _validate_wait_until)
+@register_action("wait_until", WaitUntilAction, _validate_wait_until, deferred=True)
 async def wait_until_action_to_code(
     config: ConfigType,
     action_id: ID,
@@ -578,6 +587,26 @@ async def build_condition_list(
     return conditions
 
 
+def has_deferred_actions(actions: ConfigType) -> bool:
+    """Check if a validated action list contains any deferred actions.
+
+    Deferred actions (delay, wait_until, script.wait) store trigger args
+    for later execution, making non-owning types like StringRef unsafe.
+    """
+    if isinstance(actions, list):
+        return any(has_deferred_actions(item) for item in actions)
+    if isinstance(actions, dict):
+        for key in actions:
+            if key in ACTION_REGISTRY and ACTION_REGISTRY[key].deferred:
+                return True
+        return any(
+            has_deferred_actions(v)
+            for v in actions.values()
+            if isinstance(v, (list, dict))
+        )
+    return False
+
+
 async def build_automation(
     trigger: MockObj, args: TemplateArgsType, config: ConfigType
 ) -> MockObj:

esphome/components/api/__init__.py

@@ -76,7 +76,7 @@ SERVICE_ARG_NATIVE_TYPES: dict[str, MockObj] = {
     "bool": cg.bool_,
     "int": cg.int32,
     "float": cg.float_,
-    "string": cg.std_string,
+    "string": cg.StringRef,
     "bool[]": cg.FixedVector.template(cg.bool_).operator("const").operator("ref"),
     "int[]": cg.FixedVector.template(cg.int32).operator("const").operator("ref"),
     "float[]": cg.FixedVector.template(cg.float_).operator("const").operator("ref"),
@@ -380,9 +380,16 @@ async def to_code(config: ConfigType) -> None:
                 if is_optional:
                     func_args.append((cg.bool_, "return_response"))
 
+            # Check if action chain has deferred actions that would make
+            # non-owning StringRef dangle (rx_buf_ reused after delay)
+            has_deferred = automation.has_deferred_actions(conf.get(CONF_THEN, []))
+
             service_arg_names: list[str] = []
             for name, var_ in conf[CONF_VARIABLES].items():
                 native = SERVICE_ARG_NATIVE_TYPES[var_]
+                # Fall back to std::string for string args if deferred actions exist
+                if has_deferred and native is cg.StringRef:
+                    native = cg.std_string
                 service_template_args.append(native)
                 func_args.append((native, name))
                 service_arg_names.append(name)

esphome/components/api/api.proto

@@ -824,7 +824,7 @@ message HomeAssistantStateResponse {
   option (ifdef) = "USE_API_HOMEASSISTANT_STATES";
 
   string entity_id = 1;
-  string state = 2;
+  string state = 2 [(null_terminate) = true];
   string attribute = 3;
 }
 
@@ -882,7 +882,7 @@ message ExecuteServiceArgument {
   bool bool_ = 1;
   int32 legacy_int = 2;
   float float_ = 3;
-  string string_ = 4;
+  string string_ = 4 [(null_terminate) = true];
   // ESPHome 1.14 (api v1.3) make int a signed value
   sint32 int_ = 5;
   repeated bool bool_array = 6 [packed=false, (fixed_vector) = true];

esphome/components/api/api_connection.cpp

@@ -1683,31 +1683,18 @@ void APIConnection::on_home_assistant_state_response(const HomeAssistantStateRes
   }
 
   for (auto &it : this->parent_->get_state_subs()) {
-    // Compare entity_id: check length matches and content matches
+    if (msg.entity_id != it.entity_id) {
-    size_t entity_id_len = strlen(it.entity_id);
-    if (entity_id_len != msg.entity_id.size() ||
-        memcmp(it.entity_id, msg.entity_id.c_str(), msg.entity_id.size()) != 0) {
       continue;
     }
 
     // Compare attribute: either both have matching attribute, or both have none
-    size_t sub_attr_len = it.attribute != nullptr ? strlen(it.attribute) : 0;
+    // it.attribute can be nullptr (meaning no attribute filter)
-    if (sub_attr_len != msg.attribute.size() ||
+    if (it.attribute != nullptr ? msg.attribute != it.attribute : !msg.attribute.empty()) {
-        (sub_attr_len > 0 && memcmp(it.attribute, msg.attribute.c_str(), sub_attr_len) != 0)) {
       continue;
     }
 
-    // Create null-terminated state for callback (parse_number needs null-termination)
+    // msg.state is already null-terminated in-place after protobuf decode
-    // HA state max length is 255 characters, but attributes can be much longer
+    it.callback(msg.state);
-    // Use stack buffer for common case (states), heap fallback for large attributes
-    size_t state_len = msg.state.size();
-    SmallBufferWithHeapFallback<MAX_STATE_LEN + 1> state_buf_alloc(state_len + 1);
-    char *state_buf = reinterpret_cast<char *>(state_buf_alloc.get());
-    if (state_len > 0) {
-      memcpy(state_buf, msg.state.c_str(), state_len);
-    }
-    state_buf[state_len] = '\0';
-    it.callback(StringRef(state_buf, state_len));
   }
 }
 #endif
@@ -1864,6 +1851,8 @@ void APIConnection::on_fatal_error() {
   this->flags_.remove = true;
 }
 
+void __attribute__((flatten)) APIConnection::DeferredBatch::push_item(const BatchItem &item) { items.push_back(item); }
+
 void APIConnection::DeferredBatch::add_item(EntityBase *entity, uint8_t message_type, uint8_t estimated_size,
                                             uint8_t aux_data_index) {
   // Check if we already have a message of this type for this entity
@@ -1880,7 +1869,7 @@ void APIConnection::DeferredBatch::add_item(EntityBase *entity, uint8_t message_
     }
   }
   // No existing item found (or event), add new one
-  items.push_back({entity, message_type, estimated_size, aux_data_index});
+  this->push_item({entity, message_type, estimated_size, aux_data_index});
 }
 
 void APIConnection::DeferredBatch::add_item_front(EntityBase *entity, uint8_t message_type, uint8_t estimated_size) {
@@ -1888,7 +1877,7 @@ void APIConnection::DeferredBatch::add_item_front(EntityBase *entity, uint8_t me
   // This avoids expensive vector::insert which shifts all elements
   // Note: We only ever have one high-priority message at a time (ping OR disconnect)
   // If we're disconnecting, pings are blocked, so this simple swap is sufficient
-  items.push_back({entity, message_type, estimated_size, AUX_DATA_UNUSED});
+  this->push_item({entity, message_type, estimated_size, AUX_DATA_UNUSED});
   if (items.size() > 1) {
     // Swap the new high-priority item to the front
     std::swap(items.front(), items.back());

esphome/components/api/api_connection.h

@@ -541,6 +541,8 @@ class APIConnection final : public APIServerConnectionBase {
                   uint8_t aux_data_index = AUX_DATA_UNUSED);
     // Add item to the front of the batch (for high priority messages like ping)
     void add_item_front(EntityBase *entity, uint8_t message_type, uint8_t estimated_size);
+    // Single push_back site to avoid duplicate _M_realloc_insert instantiation
+    void push_item(const BatchItem &item);
 
     // Clear all items
     void clear() {

esphome/components/api/api_frame_helper_noise.cpp

@@ -138,10 +138,12 @@ APIError APINoiseFrameHelper::handle_noise_error_(int err, const LogString *func
 
 /// Run through handshake messages (if in that phase)
 APIError APINoiseFrameHelper::loop() {
-  // During handshake phase, process as many actions as possible until we can't progress
+  // Cache ready() outside the loop. On ESP8266 LWIP raw TCP, ready() returns false once
-  // socket_->ready() stays true until next main loop, but state_action() will return
+  // the rx buffer is consumed. Re-checking each iteration would block handshake writes
-  // WOULD_BLOCK when no more data is available to read
+  // that must follow reads, deadlocking the handshake. state_action() will return
-  while (state_ != State::DATA && this->socket_->ready()) {
+  // WOULD_BLOCK when no more data is available to read.
+  bool socket_ready = this->socket_->ready();
+  while (state_ != State::DATA && socket_ready) {
     APIError err = state_action_();
     if (err == APIError::WOULD_BLOCK) {
       break;
@@ -199,9 +201,10 @@ APIError APINoiseFrameHelper::try_read_frame_() {
     return (state_ == State::DATA) ? APIError::BAD_DATA_PACKET : APIError::BAD_HANDSHAKE_PACKET_LEN;
   }
 
-  // Reserve space for body
+  // Reserve space for body (+1 for null terminator so protobuf StringRef fields
-  if (this->rx_buf_.size() != msg_size) {
+  // can be safely null-terminated in-place after decode)
-    this->rx_buf_.resize(msg_size);
+  if (this->rx_buf_.size() != msg_size + 1) {
+    this->rx_buf_.resize(msg_size + 1);
   }
 
   if (rx_buf_len_ < msg_size) {

esphome/components/api/api_frame_helper_plaintext.cpp

@@ -163,9 +163,10 @@ APIError APIPlaintextFrameHelper::try_read_frame_() {
   }
   // header reading done
 
-  // Reserve space for body
+  // Reserve space for body (+1 for null terminator so protobuf StringRef fields
-  if (this->rx_buf_.size() != this->rx_header_parsed_len_) {
+  // can be safely null-terminated in-place after decode)
-    this->rx_buf_.resize(this->rx_header_parsed_len_);
+  if (this->rx_buf_.size() != this->rx_header_parsed_len_ + 1) {
+    this->rx_buf_.resize(this->rx_header_parsed_len_ + 1);
   }
 
   if (rx_buf_len_ < rx_header_parsed_len_) {

esphome/components/api/api_options.proto

@@ -90,4 +90,13 @@ extend google.protobuf.FieldOptions {
     //   - uint16_t <field>_length_{0};
     //   - uint16_t <field>_count_{0};
     optional bool packed_buffer = 50015 [default=false];
+
+    // null_terminate: Write a null byte after string data in the decode buffer.
+    // When set on a string field in a SOURCE_CLIENT (decodable) message, the
+    // generated decode() override writes '\0' at data[length] after decoding.
+    // This makes the StringRef safe for c_str() usage without copying.
+    // Safe because: (1) frame helpers reserve +1 byte in rx_buf_, and
+    // (2) the overwritten byte was already consumed during decode.
+    // Only mark fields that actually need null-terminated access.
+    optional bool null_terminate = 50016 [default=false];
 }

esphome/components/api/api_pb2.cpp

@@ -953,6 +953,12 @@ bool HomeAssistantStateResponse::decode_length(uint32_t field_id, ProtoLengthDel
   }
   return true;
 }
+void HomeAssistantStateResponse::decode(const uint8_t *buffer, size_t length) {
+  ProtoDecodableMessage::decode(buffer, length);
+  if (!this->state.empty()) {
+    const_cast<char *>(this->state.c_str())[this->state.size()] = '\0';
+  }
+}
 #endif
 bool GetTimeResponse::decode_length(uint32_t field_id, ProtoLengthDelimited value) {
   switch (field_id) {
@@ -1057,6 +1063,9 @@ void ExecuteServiceArgument::decode(const uint8_t *buffer, size_t length) {
   uint32_t count_string_array = ProtoDecodableMessage::count_repeated_field(buffer, length, 9);
   this->string_array.init(count_string_array);
   ProtoDecodableMessage::decode(buffer, length);
+  if (!this->string_.empty()) {
+    const_cast<char *>(this->string_.c_str())[this->string_.size()] = '\0';
+  }
 }
 bool ExecuteServiceRequest::decode_varint(uint32_t field_id, ProtoVarInt value) {
   switch (field_id) {

esphome/components/api/api_pb2.h

@@ -1095,6 +1095,7 @@ class HomeAssistantStateResponse final : public ProtoDecodableMessage {
   StringRef entity_id{};
   StringRef state{};
   StringRef attribute{};
+  void decode(const uint8_t *buffer, size_t length) override;
 #ifdef HAS_PROTO_MESSAGE_DUMP
   const char *dump_to(DumpBuffer &out) const override;
 #endif

esphome/components/api/api_server.cpp

@@ -148,12 +148,16 @@ void APIServer::loop() {
   while (client_index < this->clients_.size()) {
     auto &client = this->clients_[client_index];
 
+    // Common case: process active client
+    if (!client->flags_.remove) {
+      client->loop();
+    }
+    // Handle disconnection promptly - close socket to free LWIP PCB
+    // resources and prevent retransmit crashes on ESP8266.
     if (client->flags_.remove) {
       // Rare case: handle disconnection (don't increment - swapped element needs processing)
       this->remove_client_(client_index);
     } else {
-      // Common case: process active client
-      client->loop();
       client_index++;
     }
   }
@@ -195,7 +199,7 @@ void APIServer::remove_client_(size_t client_index) {
 #endif
 }
 
-void APIServer::accept_new_connections_() {
+void __attribute__((flatten)) APIServer::accept_new_connections_() {
   while (true) {
     struct sockaddr_storage source_addr;
     socklen_t addr_len = sizeof(source_addr);

esphome/components/api/user_services.cpp

@@ -1,5 +1,6 @@
 #include "user_services.h"
 #include "esphome/core/log.h"
+#include "esphome/core/string_ref.h"
 
 namespace esphome::api {
 
@@ -11,6 +12,8 @@ template<> int32_t get_execute_arg_value<int32_t>(const ExecuteServiceArgument &
 }
 template<> float get_execute_arg_value<float>(const ExecuteServiceArgument &arg) { return arg.float_; }
 template<> std::string get_execute_arg_value<std::string>(const ExecuteServiceArgument &arg) { return arg.string_; }
+// Zero-copy StringRef version for YAML-generated services (string_ is null-terminated after decode)
+template<> StringRef get_execute_arg_value<StringRef>(const ExecuteServiceArgument &arg) { return arg.string_; }
 
 // Legacy std::vector versions for external components using custom_api_device.h - optimized with reserve
 template<> std::vector<bool> get_execute_arg_value<std::vector<bool>>(const ExecuteServiceArgument &arg) {
@@ -61,6 +64,8 @@ template<> enums::ServiceArgType to_service_arg_type<bool>() { return enums::SER
 template<> enums::ServiceArgType to_service_arg_type<int32_t>() { return enums::SERVICE_ARG_TYPE_INT; }
 template<> enums::ServiceArgType to_service_arg_type<float>() { return enums::SERVICE_ARG_TYPE_FLOAT; }
 template<> enums::ServiceArgType to_service_arg_type<std::string>() { return enums::SERVICE_ARG_TYPE_STRING; }
+// Zero-copy StringRef version for YAML-generated services
+template<> enums::ServiceArgType to_service_arg_type<StringRef>() { return enums::SERVICE_ARG_TYPE_STRING; }
 
 // Legacy std::vector versions for external components using custom_api_device.h
 template<> enums::ServiceArgType to_service_arg_type<std::vector<bool>>() { return enums::SERVICE_ARG_TYPE_BOOL_ARRAY; }

esphome/components/script/__init__.py

@@ -219,6 +219,7 @@ async def script_stop_action_to_code(config, action_id, template_arg, args):
     "script.wait",
     ScriptWaitAction,
     maybe_simple_id({cv.Required(CONF_ID): cv.use_id(Script)}),
+    deferred=True,
 )
 async def script_wait_action_to_code(config, action_id, template_arg, args):
     full_id, paren = await cg.get_variable_with_full_id(config[CONF_ID])

esphome/components/usb_host/usb_host.h

@@ -148,6 +148,7 @@ class USBClient : public Component {
   EventPool<UsbEvent, USB_EVENT_QUEUE_SIZE> event_pool;
 
  protected:
+  void handle_open_state_();
   TransferRequest *get_trq_();  // Lock-free allocation using atomic bitmask (multi-consumer safe)
   virtual void disconnect();
   virtual void on_connected() {}

esphome/components/usb_host/usb_host_client.cpp

@@ -9,6 +9,7 @@
 #include <cinttypes>
 #include <cstring>
 #include <atomic>
+#include <span>
 namespace esphome {
 namespace usb_host {
 
@@ -142,18 +143,23 @@ static void usb_client_print_config_descriptor(const usb_config_desc_t *cfg_desc
   } while (next_desc != NULL);
 }
 #endif
-static std::string get_descriptor_string(const usb_str_desc_t *desc) {
+// USB string descriptors: bLength (uint8_t, max 255) includes the 2-byte header (bLength and bDescriptorType).
-  char buffer[256];
+// Character count = (bLength - 2) / 2, max 126 chars + null terminator.
-  if (desc == nullptr)
+static constexpr size_t DESC_STRING_BUF_SIZE = 128;
+
+static const char *get_descriptor_string(const usb_str_desc_t *desc, std::span<char, DESC_STRING_BUF_SIZE> buffer) {
+  if (desc == nullptr || desc->bLength < 2)
     return "(unspecified)";
-  char *p = buffer;
+  int char_count = (desc->bLength - 2) / 2;
-  for (int i = 0; i != desc->bLength / 2; i++) {
+  char *p = buffer.data();
+  char *end = p + buffer.size() - 1;
+  for (int i = 0; i != char_count && p < end; i++) {
     auto c = desc->wData[i];
     if (c < 0x100)
       *p++ = static_cast<char>(c);
   }
   *p = '\0';
-  return {buffer};
+  return buffer.data();
 }
 
 // CALLBACK CONTEXT: USB task (called from usb_host_client_handle_events in USB task)
@@ -259,60 +265,63 @@ void USBClient::loop() {
     ESP_LOGW(TAG, "Dropped %u USB events due to queue overflow", dropped);
   }
 
-  switch (this->state_) {
+  if (this->state_ == USB_CLIENT_OPEN) {
-    case USB_CLIENT_OPEN: {
+    this->handle_open_state_();
-      int err;
+  }
-      ESP_LOGD(TAG, "Open device %d", this->device_addr_);
+}
-      err = usb_host_device_open(this->handle_, this->device_addr_, &this->device_handle_);
+
-      if (err != ESP_OK) {
+void USBClient::handle_open_state_() {
-        ESP_LOGW(TAG, "Device open failed: %s", esp_err_to_name(err));
+  int err;
-        this->state_ = USB_CLIENT_INIT;
+  ESP_LOGD(TAG, "Open device %d", this->device_addr_);
-        break;
+  err = usb_host_device_open(this->handle_, this->device_addr_, &this->device_handle_);
-      }
+  if (err != ESP_OK) {
-      ESP_LOGD(TAG, "Get descriptor device %d", this->device_addr_);
+    ESP_LOGW(TAG, "Device open failed: %s", esp_err_to_name(err));
-      const usb_device_desc_t *desc;
+    this->state_ = USB_CLIENT_INIT;
-      err = usb_host_get_device_descriptor(this->device_handle_, &desc);
+    return;
-      if (err != ESP_OK) {
+  }
-        ESP_LOGW(TAG, "Device get_desc failed: %s", esp_err_to_name(err));
+  ESP_LOGD(TAG, "Get descriptor device %d", this->device_addr_);
-        this->disconnect();
+  const usb_device_desc_t *desc;
-      } else {
+  err = usb_host_get_device_descriptor(this->device_handle_, &desc);
-        ESP_LOGD(TAG, "Device descriptor: vid %X pid %X", desc->idVendor, desc->idProduct);
+  if (err != ESP_OK) {
-        if (desc->idVendor == this->vid_ && desc->idProduct == this->pid_ || this->vid_ == 0 && this->pid_ == 0) {
+    ESP_LOGW(TAG, "Device get_desc failed: %s", esp_err_to_name(err));
-          usb_device_info_t dev_info;
+    this->disconnect();
-          err = usb_host_device_info(this->device_handle_, &dev_info);
+    return;
-          if (err != ESP_OK) {
+  }
-            ESP_LOGW(TAG, "Device info failed: %s", esp_err_to_name(err));
+  ESP_LOGD(TAG, "Device descriptor: vid %X pid %X", desc->idVendor, desc->idProduct);
-            this->disconnect();
+  if (desc->idVendor != this->vid_ || desc->idProduct != this->pid_) {
-            break;
+    if (this->vid_ != 0 || this->pid_ != 0) {
-          }
+      ESP_LOGD(TAG, "Not our device, closing");
-          this->state_ = USB_CLIENT_CONNECTED;
+      this->disconnect();
-          ESP_LOGD(TAG, "Device connected: Manuf: %s; Prod: %s; Serial: %s",
+      return;
-                   get_descriptor_string(dev_info.str_desc_manufacturer).c_str(),
+    }
-                   get_descriptor_string(dev_info.str_desc_product).c_str(),
+  }
-                   get_descriptor_string(dev_info.str_desc_serial_num).c_str());
+  usb_device_info_t dev_info;
+  err = usb_host_device_info(this->device_handle_, &dev_info);
+  if (err != ESP_OK) {
+    ESP_LOGW(TAG, "Device info failed: %s", esp_err_to_name(err));
+    this->disconnect();
+    return;
+  }
+  this->state_ = USB_CLIENT_CONNECTED;
+  char buf_manuf[DESC_STRING_BUF_SIZE];
+  char buf_product[DESC_STRING_BUF_SIZE];
+  char buf_serial[DESC_STRING_BUF_SIZE];
+  ESP_LOGD(TAG, "Device connected: Manuf: %s; Prod: %s; Serial: %s",
+           get_descriptor_string(dev_info.str_desc_manufacturer, buf_manuf),
+           get_descriptor_string(dev_info.str_desc_product, buf_product),
+           get_descriptor_string(dev_info.str_desc_serial_num, buf_serial));
 
 #if ESPHOME_LOG_LEVEL >= ESPHOME_LOG_LEVEL_VERBOSE
-          const usb_device_desc_t *device_desc;
+  const usb_device_desc_t *device_desc;
-          err = usb_host_get_device_descriptor(this->device_handle_, &device_desc);
+  err = usb_host_get_device_descriptor(this->device_handle_, &device_desc);
-          if (err == ESP_OK)
+  if (err == ESP_OK)
-            usb_client_print_device_descriptor(device_desc);
+    usb_client_print_device_descriptor(device_desc);
-          const usb_config_desc_t *config_desc;
+  const usb_config_desc_t *config_desc;
-          err = usb_host_get_active_config_descriptor(this->device_handle_, &config_desc);
+  err = usb_host_get_active_config_descriptor(this->device_handle_, &config_desc);
-          if (err == ESP_OK)
+  if (err == ESP_OK)
-            usb_client_print_config_descriptor(config_desc, nullptr);
+    usb_client_print_config_descriptor(config_desc, nullptr);
 #endif
-          this->on_connected();
+  this->on_connected();
-        } else {
-          ESP_LOGD(TAG, "Not our device, closing");
-          this->disconnect();
-        }
-      }
-      break;
-    }
-
-    default:
-      break;
-  }
 }
 
 void USBClient::on_opened(uint8_t addr) {

esphome/components/web_server/web_server.cpp

@@ -198,7 +198,8 @@ EntityMatchResult UrlMatch::match_entity(EntityBase *entity) const {
 
 #if !defined(USE_ESP32) && defined(USE_ARDUINO)
 // helper for allowing only unique entries in the queue
-void DeferredUpdateEventSource::deq_push_back_with_dedup_(void *source, message_generator_t *message_generator) {
+void __attribute__((flatten))
+DeferredUpdateEventSource::deq_push_back_with_dedup_(void *source, message_generator_t *message_generator) {
   DeferredEvent item(source, message_generator);
 
   // Use range-based for loop instead of std::find_if to reduce template instantiation overhead and binary size

esphome/components/wifi/wifi_component.cpp

@@ -3,7 +3,6 @@
 #include <cassert>
 #include <cinttypes>
 #include <cmath>
-#include <type_traits>
 
 #ifdef USE_ESP32
 #if (ESP_IDF_VERSION_MAJOR >= 5 && ESP_IDF_VERSION_MINOR >= 1)
@@ -488,6 +487,19 @@ bool WiFiComponent::matches_configured_network_(const char *ssid, const uint8_t
   return false;
 }
 
+void __attribute__((flatten)) WiFiComponent::set_sta_priority(bssid_t bssid, int8_t priority) {
+  for (auto &it : this->sta_priorities_) {
+    if (it.bssid == bssid) {
+      it.priority = priority;
+      return;
+    }
+  }
+  this->sta_priorities_.push_back(WiFiSTAPriority{
+      .bssid = bssid,
+      .priority = priority,
+  });
+}
+
 void WiFiComponent::log_discarded_scan_result_(const char *ssid, const uint8_t *bssid, int8_t rssi, uint8_t channel) {
 #if ESPHOME_LOG_LEVEL >= ESPHOME_LOG_LEVEL_VERBOSE
   // Skip logging during roaming scans to avoid log buffer overflow
@@ -1320,61 +1332,20 @@ void WiFiComponent::start_scanning() {
 // Using insertion sort instead of std::stable_sort saves flash memory
 // by avoiding template instantiations (std::rotate, std::stable_sort, lambdas)
 // IMPORTANT: This sort is stable (preserves relative order of equal elements)
-//
-// Uses raw memcpy instead of copy assignment to avoid CompactString's
-// destructor/constructor overhead (heap delete[]/new[] for long SSIDs).
-// Copy assignment calls ~CompactString() then placement-new for every shift,
-// which means delete[]/new[] per shift for heap-allocated SSIDs. With 70+
-// networks (e.g., captive portal showing full scan results), this caused
-// event loop blocking from hundreds of heap operations in a tight loop.
-//
-// This is safe because we're permuting elements within the same array —
-// each slot is overwritten exactly once, so no ownership duplication occurs.
-// All members of WiFiScanResult are either trivially copyable (bssid, channel,
-// rssi, priority, flags) or CompactString, which stores either inline data or
-// a heap pointer — never a self-referential pointer (unlike std::string's SSO
-// on some implementations). This was not possible before PR#13472 replaced
-// std::string with CompactString, since std::string's internal layout is
-// implementation-defined and may use self-referential pointers.
-//
-// TODO: If C++ standardizes std::trivially_relocatable, add the assertion for
-// WiFiScanResult/CompactString here to formally express the memcpy safety guarantee.
 template<typename VectorType> static void insertion_sort_scan_results(VectorType &results) {
-  // memcpy-based sort requires no self-referential pointers or virtual dispatch.
-  // These static_asserts guard the assumptions. If any fire, the memcpy sort
-  // must be reviewed for safety before updating the expected values.
-  //
-  // No vtable pointers (memcpy would corrupt vptr)
-  static_assert(!std::is_polymorphic<WiFiScanResult>::value, "WiFiScanResult must not have vtable");
-  static_assert(!std::is_polymorphic<CompactString>::value, "CompactString must not have vtable");
-  // Standard layout ensures predictable memory layout with no virtual bases
-  // and no mixed-access-specifier reordering
-  static_assert(std::is_standard_layout<WiFiScanResult>::value, "WiFiScanResult must be standard layout");
-  static_assert(std::is_standard_layout<CompactString>::value, "CompactString must be standard layout");
-  // Size checks catch added/removed fields that may need safety review
-  static_assert(sizeof(WiFiScanResult) == 32, "WiFiScanResult size changed - verify memcpy sort is still safe");
-  static_assert(sizeof(CompactString) == 20, "CompactString size changed - verify memcpy sort is still safe");
-  // Alignment must match for reinterpret_cast of key_buf to be valid
-  static_assert(alignof(WiFiScanResult) <= alignof(std::max_align_t), "WiFiScanResult alignment exceeds max_align_t");
   const size_t size = results.size();
-  constexpr size_t elem_size = sizeof(WiFiScanResult);
-  // Suppress warnings for intentional memcpy on non-trivially-copyable type.
-  // Safety is guaranteed by the static_asserts above and the permutation invariant.
-  // NOLINTNEXTLINE(bugprone-undefined-memory-manipulation)
-  auto *memcpy_fn = &memcpy;
   for (size_t i = 1; i < size; i++) {
-    alignas(WiFiScanResult) uint8_t key_buf[elem_size];
+    // Make a copy to avoid issues with move semantics during comparison
-    memcpy_fn(key_buf, &results[i], elem_size);
+    WiFiScanResult key = results[i];
-    const auto &key = *reinterpret_cast<const WiFiScanResult *>(key_buf);
     int32_t j = i - 1;
 
     // Move elements that are worse than key to the right
     // For stability, we only move if key is strictly better than results[j]
     while (j >= 0 && wifi_scan_result_is_better(key, results[j])) {
-      memcpy_fn(&results[j + 1], &results[j], elem_size);
+      results[j + 1] = results[j];
       j--;
     }
-    memcpy_fn(&results[j + 1], key_buf, elem_size);
+    results[j + 1] = key;
   }
 }
 

esphome/components/wifi/wifi_component.h

@@ -10,7 +10,6 @@
 
 #include <span>
 #include <string>
-#include <type_traits>
 #include <vector>
 
 #ifdef USE_LIBRETINY
@@ -220,14 +219,6 @@ class CompactString {
 };
 
 static_assert(sizeof(CompactString) == 20, "CompactString must be exactly 20 bytes");
-// CompactString is not trivially copyable (non-trivial destructor/copy for heap case).
-// However, its layout has no self-referential pointers: storage_[] contains either inline
-// data or an external heap pointer — never a pointer to itself. This is unlike libstdc++
-// std::string SSO where _M_p points to _M_local_buf within the same object.
-// This property allows memcpy-based permutation sorting where each element ends up in
-// exactly one slot (no ownership duplication). These asserts document that layout property.
-static_assert(std::is_standard_layout<CompactString>::value, "CompactString must be standard layout");
-static_assert(!std::is_polymorphic<CompactString>::value, "CompactString must not have vtable");
 
 class WiFiAP {
   friend class WiFiComponent;
@@ -497,18 +488,7 @@ class WiFiComponent : public Component {
     }
     return 0;
   }
-  void set_sta_priority(const bssid_t bssid, int8_t priority) {
+  void set_sta_priority(bssid_t bssid, int8_t priority);
-    for (auto &it : this->sta_priorities_) {
-      if (it.bssid == bssid) {
-        it.priority = priority;
-        return;
-      }
-    }
-    this->sta_priorities_.push_back(WiFiSTAPriority{
-        .bssid = bssid,
-        .priority = priority,
-    });
-  }
 
   network::IPAddresses wifi_sta_ip_addresses();
   // Remove before 2026.9.0

esphome/core/string_ref.h

@@ -81,6 +81,19 @@ class StringRef {
 
   operator std::string() const { return str(); }
 
+  /// Compare with a null-terminated C string (compatible with std::string::compare)
+  int compare(const char *s) const {
+    size_t s_len = std::strlen(s);
+    int result = std::memcmp(base_, s, std::min(len_, s_len));
+    if (result != 0)
+      return result;
+    if (len_ < s_len)
+      return -1;
+    if (len_ > s_len)
+      return 1;
+    return 0;
+  }
+
   /// Find first occurrence of substring, returns std::string::npos if not found.
   /// Note: Requires the underlying string to be null-terminated.
   size_type find(const char *s, size_type pos = 0) const {

esphome/util.py

@@ -24,11 +24,14 @@ class RegistryEntry:
         fun: Callable[..., Any],
         type_id: "MockObjClass",
         schema: "Schema",
+        *,
+        deferred: bool = False,
     ):
         self.name = name
         self.fun = fun
         self.type_id = type_id
         self.raw_schema = schema
+        self.deferred = deferred
 
     @property
     def coroutine_fun(self):
@@ -49,9 +52,16 @@ class Registry(dict[str, RegistryEntry]):
         self.base_schema = base_schema or {}
         self.type_id_key = type_id_key
 
-    def register(self, name: str, type_id: "MockObjClass", schema: "Schema"):
+    def register(
+        self,
+        name: str,
+        type_id: "MockObjClass",
+        schema: "Schema",
+        *,
+        deferred: bool = False,
+    ):
         def decorator(fun: Callable[..., Any]):
-            self[name] = RegistryEntry(name, fun, type_id, schema)
+            self[name] = RegistryEntry(name, fun, type_id, schema, deferred=deferred)
             return fun
 
         return decorator

script/api_protobuf/api_protobuf.py

@@ -2020,6 +2020,8 @@ def build_message_type(
 
     # Collect fixed_vector fields for custom decode generation
     fixed_vector_fields = []
+    # Collect fields with (null_terminate) = true option
+    null_terminate_fields = []
 
     for field in desc.field:
         # Skip deprecated fields completely
@@ -2062,6 +2064,10 @@ def build_message_type(
 
         ti = create_field_type_info(field, needs_decode, needs_encode)
 
+        # Collect fields with (null_terminate) = true for post-decode null-termination
+        if needs_decode and get_field_opt(field, pb.null_terminate, False):
+            null_terminate_fields.append(ti.field_name)
+
         # Skip field declarations for fields that are in the base class
         # but include their encode/decode logic
         if field.name not in common_field_names:
@@ -2168,8 +2174,8 @@ def build_message_type(
         prot = "bool decode_64bit(uint32_t field_id, Proto64Bit value) override;"
         protected_content.insert(0, prot)
 
-    # Generate custom decode() override for messages with FixedVector fields
+    # Generate custom decode() override for messages with FixedVector or null_terminate fields
-    if fixed_vector_fields:
+    if fixed_vector_fields or null_terminate_fields:
         # Generate the decode() implementation in cpp
         o = f"void {desc.name}::decode(const uint8_t *buffer, size_t length) {{\n"
         # Count and init each FixedVector field
@@ -2178,6 +2184,13 @@ def build_message_type(
             o += f"  this->{field_name}.init(count_{field_name});\n"
         # Call parent decode to populate the fields
         o += "  ProtoDecodableMessage::decode(buffer, length);\n"
+        # Null-terminate fields marked with (null_terminate) = true in-place.
+        # Safe: decode is complete, byte after string was already parsed (next field tag)
+        # or is the +1 reserved byte at end of rx_buf_.
+        for field_name in null_terminate_fields:
+            o += f"  if (!this->{field_name}.empty()) {{\n"
+            o += f"    const_cast<char *>(this->{field_name}.c_str())[this->{field_name}.size()] = '\\0';\n"
+            o += "  }\n"
         o += "}\n"
         cpp += o
         # Generate the decode() declaration in header (public method)

tests/integration/test_alarm_control_panel_state_transitions.py

@@ -270,6 +270,14 @@ async def test_alarm_control_panel_state_transitions(
         # The chime_sensor has chime: true, so opening it while disarmed
         # should trigger on_chime callback
 
+        # Set up future for the on_ready from opening the chime sensor
+        # (alarm becomes "not ready" when chime sensor opens).
+        # We must wait for this BEFORE creating the close future, otherwise
+        # the open event's log can arrive late and resolve the close future,
+        # causing the test to proceed before the chime close is processed.
+        ready_after_chime_open: asyncio.Future[bool] = loop.create_future()
+        ready_futures.append(ready_after_chime_open)
+
         # We're currently DISARMED - open the chime sensor
         client.switch_command(chime_switch_info.key, True)
 
@@ -279,11 +287,18 @@ async def test_alarm_control_panel_state_transitions(
         except TimeoutError:
             pytest.fail(f"on_chime callback not fired. Log lines: {log_lines[-20:]}")
 
-        # Close the chime sensor and wait for alarm to become ready again
+        # Wait for the on_ready from the chime sensor opening
-        # We need to wait for this transition before testing door sensor,
+        try:
-        # otherwise there's a race where the door sensor state change could
+            await asyncio.wait_for(ready_after_chime_open, timeout=2.0)
-        # arrive before the chime sensor state change, leaving the alarm in
+        except TimeoutError:
-        # a continuous "not ready" state with no on_ready callback fired.
+            pytest.fail(
+                f"on_ready callback not fired when chime sensor opened. "
+                f"Log lines: {log_lines[-20:]}"
+            )
+
+        # Now create the future for the close event and close the sensor.
+        # Since we waited for the open event above, the close event's
+        # on_ready log cannot be confused with the open event's.
         ready_after_chime_close: asyncio.Future[bool] = loop.create_future()
         ready_futures.append(ready_after_chime_close)