[api] Add description and example metadata to user-defined actions (#18881)

Co-authored-by: J. Nick Koston <nick@home-assistant.io>

esphome/codegen.py

@@ -78,6 +78,7 @@ from esphome.cpp_types import (  # noqa: F401
     StringRef,
     arduino_json_ns,
     bool_,
+    char,
     const_char_ptr,
     double,
     esphome_ns,

esphome/components/api/__init__.py

@@ -5,6 +5,7 @@ from typing import Any
 from esphome import automation
 from esphome.automation import Condition
 import esphome.codegen as cg
+from esphome.components.const import CONF_DESCRIPTION
 from esphome.components.logger import request_log_listener
 
 # ENCRYPTION_SCHEMA and validate_encryption_key are re-exported for external
@@ -41,10 +42,12 @@ from esphome.const import (
     CONF_TAG,
     CONF_THEN,
     CONF_TRIGGER_ID,
+    CONF_TYPE,
     CONF_VARIABLES,
 )
 from esphome.core import CORE, ID, CoroPriority, EsphomeError, coroutine_with_priority
 from esphome.cpp_generator import MockObj, TemplateArgsType
+from esphome.helpers import fnv1_hash
 from esphome.types import ConfigFragmentType, ConfigType
 
 # Compat alias: downstream consumers (e.g. device-builder) referenced the
@@ -125,6 +128,7 @@ SERVICE_ARG_FALLBACK_TYPES: dict[str, MockObj] = {
 }
 CONF_BATCH_DELAY = "batch_delay"
 CONF_CUSTOM_SERVICES = "custom_services"
+CONF_EXAMPLE = "example"
 CONF_HOMEASSISTANT_SERVICES = "homeassistant_services"
 CONF_HOMEASSISTANT_STATES = "homeassistant_states"
 CONF_LISTEN_BACKLOG = "listen_backlog"
@@ -228,14 +232,30 @@ def _validate_supports_response(value: Any) -> str:
     return cv.enum(SUPPORTS_RESPONSE_OPTIONS, lower=True)(value)
 
 
+# ESP8266 copies every string of an action into a stack buffer sized by codegen; keep it small
+ESP8266_ACTION_STRINGS_MAX_TOTAL = 384
+
+VARIABLE_SCHEMA = cv.Schema(
+    {
+        cv.Required(CONF_TYPE): cv.one_of(*SERVICE_ARG_NATIVE_TYPES, lower=True),
+        cv.Optional(CONF_DESCRIPTION): cv.string_strict,
+        cv.Optional(CONF_EXAMPLE): cv.string_strict,
+    }
+)
+
+# Accepts the plain `name: type` shorthand or the full mapping form
+validate_variable = cv.maybe_simple_value(VARIABLE_SCHEMA, key=CONF_TYPE)
+
+
 ACTIONS_SCHEMA = automation.validate_automation(
     {
         cv.GenerateID(CONF_TRIGGER_ID): cv.declare_id(UserServiceTrigger),
         cv.Exclusive(CONF_SERVICE, group_of_exclusion=CONF_ACTION): cv.valid_name,
         cv.Exclusive(CONF_ACTION, group_of_exclusion=CONF_ACTION): cv.valid_name,
+        cv.Optional(CONF_DESCRIPTION): cv.string_strict,
         cv.Optional(CONF_VARIABLES, default={}): cv.Schema(
             {
-                cv.validate_id_name: cv.one_of(*SERVICE_ARG_NATIVE_TYPES, lower=True),
+                cv.validate_id_name: validate_variable,
             }
         ),
         # No default - auto-detected by _auto_detect_supports_response
@@ -352,6 +372,85 @@ CONFIG_SCHEMA = cv.All(
 )
 
 
+def _has_action_metadata(actions: list[ConfigType]) -> bool:
+    # Empty strings count as unset, matching _action_strings
+    return any(
+        conf.get(CONF_DESCRIPTION)
+        or any(
+            var_.get(CONF_DESCRIPTION) or var_.get(CONF_EXAMPLE)
+            for var_ in conf[CONF_VARIABLES].values()
+        )
+        for conf in actions
+    )
+
+
+def _action_strings(conf: ConfigType, has_metadata: bool) -> list[str | None]:
+    """Strings of one action in the table order UserServiceStatic (user_services.h) expects."""
+    # An empty description or example is treated as unset
+    strings: list[str | None] = [conf[CONF_ACTION]]
+    if has_metadata:
+        strings.append(conf.get(CONF_DESCRIPTION) or None)
+    for name, var_ in conf[CONF_VARIABLES].items():
+        strings.append(name)
+        if has_metadata:
+            strings += [
+                var_.get(CONF_DESCRIPTION) or None,
+                var_.get(CONF_EXAMPLE) or None,
+            ]
+    return strings
+
+
+def _action_strings_size(strings: list[str | None]) -> int:
+    """Bytes needed to copy every string out of flash, each with its terminator."""
+    return sum(
+        len(string.encode("utf-8")) + 1 for string in strings if string is not None
+    )
+
+
+def _validate_esp8266_action_strings(config: ConfigType) -> ConfigType:
+    if not CORE.is_esp8266:
+        return config
+    actions = config.get(CONF_ACTIONS, [])
+    has_metadata = _has_action_metadata(actions)
+    for conf in actions:
+        size = _action_strings_size(_action_strings(conf, has_metadata))
+        if size > ESP8266_ACTION_STRINGS_MAX_TOTAL:
+            raise cv.Invalid(
+                f"Action '{conf[CONF_ACTION]}' has {size} bytes of name, variable name, "
+                f"description and example text; ESP8266 allows at most "
+                f"{ESP8266_ACTION_STRINGS_MAX_TOTAL} bytes per action"
+            )
+    return config
+
+
+FINAL_VALIDATE_SCHEMA = _validate_esp8266_action_strings
+
+
+def _add_action_strings(
+    index: int, strings: list[str | None], interned: dict[str, MockObj]
+) -> MockObj:
+    """Emit the PROGMEM string table for one action.
+
+    Each string is its own PROGMEM array because on ESP8266 .rodata is RAM, and identical
+    strings are shared between actions through `interned`.
+    """
+    entries: list[MockObj] = []
+    for string in strings:
+        if string is None:
+            entries.append(cg.nullptr)
+            continue
+        if (var := interned.get(string)) is None:
+            var = interned[string] = cg.progmem_array(
+                ID(f"api_action_str{len(interned)}", is_declaration=True, type=cg.char),
+                string,
+            )
+        entries.append(var)
+    return cg.progmem_array(
+        ID(f"api_action{index}_strings", is_declaration=True, type=cg.const_char_ptr),
+        entries,
+    )
+
+
 @coroutine_with_priority(CoroPriority.WEB)
 async def to_code(config: ConfigType) -> None:
     var = cg.new_Pvariable(config[CONF_ID])
@@ -371,8 +470,10 @@ async def to_code(config: ConfigType) -> None:
     cg.add_define("MAX_API_CONNECTIONS", config[CONF_MAX_CONNECTIONS])
     cg.add_define("API_MAX_SEND_QUEUE", config[CONF_MAX_SEND_QUEUE])
 
+    actions = config.get(CONF_ACTIONS, [])
+    has_user_actions = bool(actions) or config[CONF_CUSTOM_SERVICES]
     # Set USE_API_USER_DEFINED_ACTIONS if any services are enabled
-    if config.get(CONF_ACTIONS) or config[CONF_CUSTOM_SERVICES]:
+    if has_user_actions:
         cg.add_define("USE_API_USER_DEFINED_ACTIONS")
 
     # Set USE_API_CUSTOM_SERVICES if external components need dynamic service registration
@@ -385,10 +486,17 @@ async def to_code(config: ConfigType) -> None:
     if config[CONF_HOMEASSISTANT_STATES]:
         cg.add_define("USE_API_HOMEASSISTANT_STATES")
 
-    if actions := config.get(CONF_ACTIONS, []):
+    scratch_size = 0
+    if actions:
+        # Metadata is compiled in for every action once any action declares it, because the
+        # string table layout is fixed by the define rather than per action
+        has_metadata = _has_action_metadata(actions)
+        if has_metadata:
+            cg.add_define("USE_API_USER_DEFINED_ACTION_METADATA")
+        interned_strings: dict[str, MockObj] = {}
         # Collect all triggers first, then register all at once with initializer_list
         triggers: list[cg.MockObj] = []
-        for conf in actions:
+        for index, conf in enumerate(actions):
             func_args: list[tuple[MockObj, str]] = []
             service_template_args: list[MockObj] = []  # User service argument types
 
@@ -421,22 +529,23 @@ async def to_code(config: ConfigType) -> None:
                 conf.get(CONF_THEN, [])
             )
 
-            service_arg_names: list[str] = []
             for name, var_ in conf[CONF_VARIABLES].items():
-                if has_non_synchronous and var_ in SERVICE_ARG_FALLBACK_TYPES:
-                    native = SERVICE_ARG_FALLBACK_TYPES[var_]
+                var_type = var_[CONF_TYPE]
+                if has_non_synchronous and var_type in SERVICE_ARG_FALLBACK_TYPES:
+                    native = SERVICE_ARG_FALLBACK_TYPES[var_type]
                 else:
-                    native = SERVICE_ARG_NATIVE_TYPES[var_]
+                    native = SERVICE_ARG_NATIVE_TYPES[var_type]
                 service_template_args.append(native)
                 func_args.append((native, name))
-                service_arg_names.append(name)
+            strings = _action_strings(conf, has_metadata)
+            table = _add_action_strings(index, strings, interned_strings)
+            if CORE.is_esp8266:
+                scratch_size = max(scratch_size, _action_strings_size(strings))
             # Template args: supports_response mode, then user service arg types
             templ = cg.TemplateArguments(supports_response, *service_template_args)
+            # Key is hashed here because the name is not readable at runtime on ESP8266
             trigger = cg.new_Pvariable(
-                conf[CONF_TRIGGER_ID],
-                templ,
-                conf[CONF_ACTION],
-                service_arg_names,
+                conf[CONF_TRIGGER_ID], templ, table, fnv1_hash(conf[CONF_ACTION])
             )
             triggers.append(trigger)
             auto = await automation.build_automation(trigger, func_args, conf)
@@ -458,6 +567,9 @@ async def to_code(config: ConfigType) -> None:
                 cg.add(auto.add_actions([unregister_action]))
         # Register all services at once - single allocation, no reallocations
         cg.add(var.initialize_user_services(triggers))
+    if CORE.is_esp8266 and has_user_actions:
+        # Stack buffer that list-entities copies PROGMEM strings into, sized for the largest action
+        cg.add_define("API_USER_ACTION_STRINGS_SCRATCH_SIZE", max(scratch_size, 1))
 
     if CONF_ON_CLIENT_CONNECTED in config:
         cg.add_define("USE_API_CLIENT_CONNECTED_TRIGGER")

esphome/components/api/api.proto

@@ -1034,6 +1034,8 @@ message ListEntitiesServicesArgument {
   option (ifdef) = "USE_API_USER_DEFINED_ACTIONS";
   string name = 1;
   ServiceArgType type = 2;
+  string description = 3 [(field_ifdef) = "USE_API_USER_DEFINED_ACTION_METADATA"];
+  string example = 4 [(field_ifdef) = "USE_API_USER_DEFINED_ACTION_METADATA"];
 }
 message ListEntitiesServicesResponse {
   option (id) = 41;
@@ -1044,6 +1046,7 @@ message ListEntitiesServicesResponse {
   fixed32 key = 2 [(force) = true];
   repeated ListEntitiesServicesArgument args = 3 [(fixed_vector) = true];
   SupportsResponseType supports_response = 4;
+  string description = 5 [(field_ifdef) = "USE_API_USER_DEFINED_ACTION_METADATA"];
 }
 message ExecuteServiceArgument {
   option (ifdef) = "USE_API_USER_DEFINED_ACTIONS";

esphome/components/api/api_pb2.cpp

@@ -1275,12 +1275,24 @@ uint8_t *ListEntitiesServicesArgument::encode(ProtoWriteBuffer &buffer PROTO_ENC
   uint8_t *__restrict__ pos = buffer.get_pos();
   ProtoEncode::encode_string(pos PROTO_ENCODE_DEBUG_ARG, 1, this->name);
   ProtoEncode::encode_uint32(pos PROTO_ENCODE_DEBUG_ARG, 2, static_cast<uint32_t>(this->type));
+#ifdef USE_API_USER_DEFINED_ACTION_METADATA
+  ProtoEncode::encode_string(pos PROTO_ENCODE_DEBUG_ARG, 3, this->description);
+#endif
+#ifdef USE_API_USER_DEFINED_ACTION_METADATA
+  ProtoEncode::encode_string(pos PROTO_ENCODE_DEBUG_ARG, 4, this->example);
+#endif
   return pos;
 }
 uint32_t ListEntitiesServicesArgument::calculate_size() const {
   uint32_t size = 0;
   size += ProtoSize::calc_length(1, this->name.size());
   size += this->type ? 2 : 0;
+#ifdef USE_API_USER_DEFINED_ACTION_METADATA
+  size += ProtoSize::calc_length(1, this->description.size());
+#endif
+#ifdef USE_API_USER_DEFINED_ACTION_METADATA
+  size += ProtoSize::calc_length(1, this->example.size());
+#endif
   return size;
 }
 uint8_t *ListEntitiesServicesResponse::encode(ProtoWriteBuffer &buffer PROTO_ENCODE_DEBUG_PARAM) const {
@@ -1291,6 +1303,9 @@ uint8_t *ListEntitiesServicesResponse::encode(ProtoWriteBuffer &buffer PROTO_ENC
     ProtoEncode::encode_sub_message(pos PROTO_ENCODE_DEBUG_ARG, buffer, 3, it);
   }
   ProtoEncode::encode_uint32(pos PROTO_ENCODE_DEBUG_ARG, 4, static_cast<uint32_t>(this->supports_response));
+#ifdef USE_API_USER_DEFINED_ACTION_METADATA
+  ProtoEncode::encode_string(pos PROTO_ENCODE_DEBUG_ARG, 5, this->description);
+#endif
   return pos;
 }
 uint32_t ListEntitiesServicesResponse::calculate_size() const {
@@ -1303,6 +1318,9 @@ uint32_t ListEntitiesServicesResponse::calculate_size() const {
     }
   }
   size += this->supports_response ? 2 : 0;
+#ifdef USE_API_USER_DEFINED_ACTION_METADATA
+  size += ProtoSize::calc_length(1, this->description.size());
+#endif
   return size;
 }
 bool ExecuteServiceArgument::decode_varint(uint32_t field_id, proto_varint_value_t value) {

esphome/components/api/api_pb2.h

@@ -1317,6 +1317,12 @@ class ListEntitiesServicesArgument final : public ProtoMessage {
  public:
   StringRef name{};
   enums::ServiceArgType type{};
+#ifdef USE_API_USER_DEFINED_ACTION_METADATA
+  StringRef description{};
+#endif
+#ifdef USE_API_USER_DEFINED_ACTION_METADATA
+  StringRef example{};
+#endif
   uint8_t *encode(ProtoWriteBuffer &buffer PROTO_ENCODE_DEBUG_PARAM) const;
   uint32_t calculate_size() const;
 #ifdef HAS_PROTO_MESSAGE_DUMP
@@ -1328,7 +1334,7 @@ class ListEntitiesServicesArgument final : public ProtoMessage {
 class ListEntitiesServicesResponse final : public ProtoMessage {
  public:
   static constexpr uint16_t MESSAGE_TYPE = 41;
-  static constexpr uint8_t ESTIMATED_SIZE = 50;
+  static constexpr uint8_t ESTIMATED_SIZE = 59;
 #ifdef HAS_PROTO_MESSAGE_DUMP
   const LogString *message_name() const override { return LOG_STR("list_entities_services_response"); }
 #endif
@@ -1336,6 +1342,9 @@ class ListEntitiesServicesResponse final : public ProtoMessage {
   uint32_t key{0};
   FixedVector<ListEntitiesServicesArgument> args{};
   enums::SupportsResponseType supports_response{};
+#ifdef USE_API_USER_DEFINED_ACTION_METADATA
+  StringRef description{};
+#endif
   uint8_t *encode(ProtoWriteBuffer &buffer PROTO_ENCODE_DEBUG_PARAM) const;
   uint32_t calculate_size() const;
 #ifdef HAS_PROTO_MESSAGE_DUMP

esphome/components/api/api_pb2_dump.cpp

@@ -1500,6 +1500,12 @@ const char *ListEntitiesServicesArgument::dump_to(DumpBuffer &out) const {
   MessageDumpHelper helper(out, ESPHOME_PSTR("ListEntitiesServicesArgument"));
   dump_field(out, ESPHOME_PSTR("name"), this->name);
   dump_field(out, ESPHOME_PSTR("type"), static_cast<enums::ServiceArgType>(this->type));
+#ifdef USE_API_USER_DEFINED_ACTION_METADATA
+  dump_field(out, ESPHOME_PSTR("description"), this->description);
+#endif
+#ifdef USE_API_USER_DEFINED_ACTION_METADATA
+  dump_field(out, ESPHOME_PSTR("example"), this->example);
+#endif
   return out.c_str();
 }
 const char *ListEntitiesServicesResponse::dump_to(DumpBuffer &out) const {
@@ -1512,6 +1518,9 @@ const char *ListEntitiesServicesResponse::dump_to(DumpBuffer &out) const {
     out.append("\n");
   }
   dump_field(out, ESPHOME_PSTR("supports_response"), static_cast<enums::SupportsResponseType>(this->supports_response));
+#ifdef USE_API_USER_DEFINED_ACTION_METADATA
+  dump_field(out, ESPHOME_PSTR("description"), this->description);
+#endif
   return out.c_str();
 }
 const char *ExecuteServiceArgument::dump_to(DumpBuffer &out) const {

esphome/components/api/list_entities.cpp

@@ -99,7 +99,8 @@ ListEntitiesIterator::ListEntitiesIterator(APIConnection *client) : client_(clie
 static constexpr uint8_t SERVICE_YIELD_INTERVAL = 3;
 
 bool ListEntitiesIterator::on_service(UserServiceDescriptor *service) {
-  auto resp = service->encode_list_service_response();
+  UserActionScratch scratch;
+  auto resp = service->encode_list_service_response(scratch);
   if (!this->client_->send_message(resp))
     return false;
   // at_ is this service's index

esphome/components/api/user_services.cpp

@@ -1,9 +1,52 @@
 #include "user_services.h"
+#include "esphome/core/hal.h"
 #include "esphome/core/log.h"
 #include "esphome/core/string_ref.h"
 
 namespace esphome::api {
 
+StringRef UserServiceStatic::str_(size_t idx, std::span<char> &scratch) const {
+  const char *s = progmem_read_ptr(&this->strings_[idx]);
+  if (s == nullptr)
+    return {};
+#ifdef USE_ESP8266
+  // Codegen sizes the scratch buffer for the largest service; the bound only guards other callers
+  if (scratch.empty())
+    return {};
+  size_t len = strnlen_P(s, scratch.size() - 1);
+  progmem_memcpy(scratch.data(), s, len);
+  scratch[len] = '\0';
+  StringRef ref(scratch.data(), len);
+  scratch = scratch.subspan(len + 1);
+  return ref;
+#else
+  return StringRef(s);
+#endif
+}
+
+ListEntitiesServicesResponse UserServiceStatic::encode_list_service_response_(
+    std::span<const enums::ServiceArgType> arg_types, std::span<char> scratch) const {
+  ListEntitiesServicesResponse msg;
+  msg.name = this->str_(0, scratch);
+  msg.key = this->key_;
+  msg.supports_response = this->supports_response_;
+#ifdef USE_API_USER_DEFINED_ACTION_METADATA
+  msg.description = this->str_(1, scratch);
+#endif
+  msg.args.init(arg_types.size());
+  for (size_t i = 0; i < arg_types.size(); i++) {
+    size_t base = USER_ACTION_HEADER_STRINGS + i * USER_ACTION_ARG_STRINGS;
+    auto &arg = msg.args.emplace_back();
+    arg.type = arg_types[i];
+    arg.name = this->str_(base, scratch);
+#ifdef USE_API_USER_DEFINED_ACTION_METADATA
+    arg.description = this->str_(base + 1, scratch);
+    arg.example = this->str_(base + 2, scratch);
+#endif
+  }
+  return msg;
+}
+
 template<> bool get_execute_arg_value<bool>(const ExecuteServiceArgument &arg) { return arg.bool_; }
 template<> int32_t get_execute_arg_value<int32_t>(const ExecuteServiceArgument &arg) {
   if (arg.legacy_int != 0)

esphome/components/api/user_services.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <span>
 #include <tuple>
 #include <utility>
 #include <vector>
@@ -19,7 +20,9 @@ class APIServer;
 
 class UserServiceDescriptor {
  public:
-  virtual ListEntitiesServicesResponse encode_list_service_response() = 0;
+  /// Build the list-entities message. On ESP8266 the strings live in PROGMEM and are copied into
+  /// `scratch`, so the returned message is only valid while `scratch` is; other platforms ignore it.
+  virtual ListEntitiesServicesResponse encode_list_service_response(std::span<char> scratch) = 0;
 
   virtual bool execute_service(const ExecuteServiceRequest &req) = 0;
 #ifdef USE_API_USER_DEFINED_ACTION_RESPONSES
@@ -34,29 +37,51 @@ template<typename T> T get_execute_arg_value(const ExecuteServiceArgument &arg);
 
 template<typename T> enums::ServiceArgType to_service_arg_type();
 
-// Base class for YAML-defined services (most common case)
-// Stores only pointers to string literals in flash - no heap allocation
-template<typename... Ts> class UserServiceBase : public UserServiceDescriptor {
- public:
-  UserServiceBase(const char *name, const std::array<const char *, sizeof...(Ts)> &arg_names,
-                  enums::SupportsResponseType supports_response = enums::SUPPORTS_RESPONSE_NONE)
-      : name_(name), arg_names_(arg_names), supports_response_(supports_response) {
-    this->key_ = fnv1_hash(name);
-  }
+// Scratch buffer list-entities hands to encode_list_service_response(); only ESP8266 copies into it
+#ifdef USE_ESP8266
+using UserActionScratch = std::array<char, API_USER_ACTION_STRINGS_SCRATCH_SIZE>;
+#else
+using UserActionScratch = std::array<char, 0>;
+#endif
 
-  ListEntitiesServicesResponse encode_list_service_response() override {
-    ListEntitiesServicesResponse msg;
-    msg.name = StringRef(this->name_);
-    msg.key = this->key_;
-    msg.supports_response = this->supports_response_;
+// Non-template base for YAML-defined services so the list-entities encoder is compiled once.
+// All strings live in one PROGMEM pointer table emitted by codegen (see _action_strings in
+// __init__.py), so each service costs a single pointer of RAM. Layout: the action name, then
+// each argument name; with USE_API_USER_DEFINED_ACTION_METADATA the action description follows
+// the name and every argument is (name, description, example). Unset metadata is nullptr.
+#ifdef USE_API_USER_DEFINED_ACTION_METADATA
+static constexpr size_t USER_ACTION_HEADER_STRINGS = 2;
+static constexpr size_t USER_ACTION_ARG_STRINGS = 3;
+#else
+static constexpr size_t USER_ACTION_HEADER_STRINGS = 1;
+static constexpr size_t USER_ACTION_ARG_STRINGS = 1;
+#endif
+class UserServiceStatic : public UserServiceDescriptor {
+ public:
+  UserServiceStatic(const char *const *strings, uint32_t key,
+                    enums::SupportsResponseType supports_response = enums::SUPPORTS_RESPONSE_NONE)
+      : strings_(strings), key_(key), supports_response_(supports_response) {}
+
+ protected:
+  ListEntitiesServicesResponse encode_list_service_response_(std::span<const enums::ServiceArgType> arg_types,
+                                                             std::span<char> scratch) const;
+  /// Reference table entry `idx`; nullptr gives an empty StringRef.
+  /// On ESP8266 the bytes are copied out of PROGMEM into `scratch` with a terminator, and the span
+  /// is advanced past the copy.
+  StringRef str_(size_t idx, std::span<char> &scratch) const;
+
+  const char *const *strings_;  // PROGMEM pointer table, read with progmem_read_ptr()
+  uint32_t key_;
+  enums::SupportsResponseType supports_response_;
+};
+
+template<typename... Ts> class UserServiceBase : public UserServiceStatic {
+ public:
+  using UserServiceStatic::UserServiceStatic;
+
+  ListEntitiesServicesResponse encode_list_service_response(std::span<char> scratch) override {
     std::array<enums::ServiceArgType, sizeof...(Ts)> arg_types = {to_service_arg_type<Ts>()...};
-    msg.args.init(sizeof...(Ts));
-    for (size_t i = 0; i < sizeof...(Ts); i++) {
-      auto &arg = msg.args.emplace_back();
-      arg.type = arg_types[i];
-      arg.name = StringRef(this->arg_names_[i]);
-    }
-    return msg;
+    return this->encode_list_service_response_(arg_types, scratch);
   }
 
   bool execute_service(const ExecuteServiceRequest &req) override {
@@ -89,12 +114,6 @@ template<typename... Ts> class UserServiceBase : public UserServiceDescriptor {
   void execute_(const ArgsContainer &args, uint32_t call_id, bool return_response, std::index_sequence<S...> /*type*/) {
     this->execute(call_id, return_response, (get_execute_arg_value<Ts>(args[S]))...);
   }
-
-  // Pointers to string literals in flash - no heap allocation
-  const char *name_;
-  std::array<const char *, sizeof...(Ts)> arg_names_;
-  uint32_t key_{0};
-  enums::SupportsResponseType supports_response_{enums::SUPPORTS_RESPONSE_NONE};
 };
 
 // Separate class for custom_api_device services (rare case)
@@ -106,7 +125,7 @@ template<typename... Ts> class UserServiceDynamic : public UserServiceDescriptor
     this->key_ = fnv1_hash(this->name_.c_str());
   }
 
-  ListEntitiesServicesResponse encode_list_service_response() override {
+  ListEntitiesServicesResponse encode_list_service_response(std::span<char> /*scratch*/) override {
     ListEntitiesServicesResponse msg;
     msg.name = StringRef(this->name_);
     msg.key = this->key_;
@@ -167,8 +186,8 @@ template<typename... Ts>
 class UserServiceTrigger<enums::SUPPORTS_RESPONSE_NONE, Ts...> final : public UserServiceBase<Ts...>,
                                                                        public Trigger<Ts...> {
  public:
-  UserServiceTrigger(const char *name, const std::array<const char *, sizeof...(Ts)> &arg_names)
-      : UserServiceBase<Ts...>(name, arg_names, enums::SUPPORTS_RESPONSE_NONE) {}
+  UserServiceTrigger(const char *const *strings, uint32_t key)
+      : UserServiceBase<Ts...>(strings, key, enums::SUPPORTS_RESPONSE_NONE) {}
 
  protected:
   void execute(uint32_t /*call_id*/, bool /*return_response*/, Ts... x) override { this->trigger(x...); }
@@ -179,8 +198,8 @@ template<typename... Ts>
 class UserServiceTrigger<enums::SUPPORTS_RESPONSE_OPTIONAL, Ts...> final : public UserServiceBase<Ts...>,
                                                                            public Trigger<uint32_t, bool, Ts...> {
  public:
-  UserServiceTrigger(const char *name, const std::array<const char *, sizeof...(Ts)> &arg_names)
-      : UserServiceBase<Ts...>(name, arg_names, enums::SUPPORTS_RESPONSE_OPTIONAL) {}
+  UserServiceTrigger(const char *const *strings, uint32_t key)
+      : UserServiceBase<Ts...>(strings, key, enums::SUPPORTS_RESPONSE_OPTIONAL) {}
 
  protected:
   void execute(uint32_t call_id, bool return_response, Ts... x) override {
@@ -193,8 +212,8 @@ template<typename... Ts>
 class UserServiceTrigger<enums::SUPPORTS_RESPONSE_ONLY, Ts...> final : public UserServiceBase<Ts...>,
                                                                        public Trigger<uint32_t, Ts...> {
  public:
-  UserServiceTrigger(const char *name, const std::array<const char *, sizeof...(Ts)> &arg_names)
-      : UserServiceBase<Ts...>(name, arg_names, enums::SUPPORTS_RESPONSE_ONLY) {}
+  UserServiceTrigger(const char *const *strings, uint32_t key)
+      : UserServiceBase<Ts...>(strings, key, enums::SUPPORTS_RESPONSE_ONLY) {}
 
  protected:
   void execute(uint32_t call_id, bool /*return_response*/, Ts... x) override { this->trigger(call_id, x...); }
@@ -205,8 +224,8 @@ template<typename... Ts>
 class UserServiceTrigger<enums::SUPPORTS_RESPONSE_STATUS, Ts...> final : public UserServiceBase<Ts...>,
                                                                          public Trigger<uint32_t, Ts...> {
  public:
-  UserServiceTrigger(const char *name, const std::array<const char *, sizeof...(Ts)> &arg_names)
-      : UserServiceBase<Ts...>(name, arg_names, enums::SUPPORTS_RESPONSE_STATUS) {}
+  UserServiceTrigger(const char *const *strings, uint32_t key)
+      : UserServiceBase<Ts...>(strings, key, enums::SUPPORTS_RESPONSE_STATUS) {}
 
  protected:
   void execute(uint32_t call_id, bool /*return_response*/, Ts... x) override { this->trigger(call_id, x...); }

esphome/components/const/__init__.py

@@ -16,6 +16,7 @@ CONF_CO2_EQUIVALENT = "co2_equivalent"
 CONF_COLOR_DEPTH = "color_depth"
 CONF_CRC_ENABLE = "crc_enable"
 CONF_DATA_BITS = "data_bits"
+CONF_DESCRIPTION = "description"
 CONF_DRAW_ROUNDING = "draw_rounding"
 CONF_ENABLE_OTA_DOWNGRADE_PROTECTION = "enable_ota_downgrade_protection"
 CONF_ENABLED = "enabled"

esphome/core/defines.h

@@ -218,9 +218,11 @@
 #define USE_API_PLAINTEXT
 #define USE_API_USER_DEFINED_ACTIONS
 #define USE_API_CUSTOM_SERVICES
+#define USE_API_USER_DEFINED_ACTION_METADATA
 #define USE_API_USER_DEFINED_ACTION_RESPONSES
 #define USE_API_USER_DEFINED_ACTION_RESPONSES_JSON
 #define API_MAX_SEND_QUEUE 8
+#define API_USER_ACTION_STRINGS_SCRATCH_SIZE 64
 #define MAX_API_CONNECTIONS 6
 // The Improv library is not in the Zephyr tidy environment
 #define USE_IMPROV_SERIAL

esphome/cpp_types.py

@@ -14,6 +14,7 @@ std_string_ref = std_ns.namespace("string &")
 std_vector = std_ns.class_("vector")
 std_span = std_ns.class_("span")
 int8 = global_ns.namespace("int8_t")
+char = global_ns.namespace("char")
 uint8 = global_ns.namespace("uint8_t")
 uint16 = global_ns.namespace("uint16_t")
 uint32 = global_ns.namespace("uint32_t")

tests/component_tests/api/test_action_metadata.py

@@ -0,0 +1,155 @@
+"""Tests for user-defined action field metadata (description / example)."""
+
+from collections.abc import Callable
+from pathlib import Path
+
+import pytest
+
+from esphome.components.api import (
+    _action_strings,
+    _action_strings_size,
+    _has_action_metadata,
+    _validate_esp8266_action_strings,
+    validate_variable,
+)
+from esphome.config_validation import Invalid
+from esphome.const import PlatformFramework
+from esphome.core import CORE
+from esphome.cpp_generator import safe_exp
+from esphome.helpers import fnv1_hash
+from tests.component_tests.helpers import get_define_value
+from tests.component_tests.types import SetCoreConfigCallable
+
+CONFIG = "tests/component_tests/api/test_action_metadata.yaml"
+CONFIG_ESP8266 = "tests/component_tests/api/test_action_metadata_esp8266.yaml"
+CONFIG_SHORTHAND = "tests/component_tests/api/test_action_metadata_shorthand.yaml"
+
+
+def test_metadata_is_emitted_as_progmem_table(
+    generate_main: Callable[[str | Path], str],
+) -> None:
+    """Every action string is a PROGMEM array referenced from one PROGMEM table."""
+    main_cpp = generate_main(CONFIG)
+
+    assert (
+        'static constexpr char api_action_str0[] PROGMEM = "play_buzzer";' in main_cpp
+    )
+    assert (
+        'static constexpr char api_action_str1[] PROGMEM = "Play an RTTTL melody on the buzzer";'
+        in main_cpp
+    )
+    assert (
+        'static constexpr char api_action_str4[] PROGMEM = "two_short:d=4,o=5,b=100:16e6,16e6";'
+        in main_cpp
+    )
+    assert (
+        "static constexpr const char * api_action0_strings[] PROGMEM = {"
+        "api_action_str0, api_action_str1, api_action_str2, api_action_str3, "
+        "api_action_str4, api_action_str5, nullptr, nullptr};" in main_cpp
+    )
+    # An action without metadata still carries the metadata slots (as nullptr)
+    assert (
+        "static constexpr const char * api_action1_strings[] PROGMEM = {"
+        "api_action_str6, nullptr, api_action_str7, nullptr, nullptr};" in main_cpp
+    )
+    assert f"(api_action0_strings, {safe_exp(fnv1_hash('play_buzzer'))});" in main_cpp
+    assert "USE_API_USER_DEFINED_ACTION_METADATA" in {d.name for d in CORE.defines}
+    assert get_define_value("API_USER_ACTION_STRINGS_SCRATCH_SIZE") is None
+
+
+def test_esp8266_sizes_scratch_buffer_for_largest_action(
+    generate_main: Callable[[str | Path], str],
+) -> None:
+    """ESP8266 gets a scratch buffer define equal to the byte total of the largest action."""
+    generate_main(CONFIG_ESP8266)
+
+    # play_buzzer: name, description, two variable names, one description, one example,
+    # each with a terminator
+    assert get_define_value("API_USER_ACTION_STRINGS_SCRATCH_SIZE") == "117"
+
+
+def test_shorthand_variables_emit_no_metadata(
+    generate_main: Callable[[str | Path], str],
+) -> None:
+    """The name: type shorthand emits a name-only table and no define."""
+    main_cpp = generate_main(CONFIG_SHORTHAND)
+
+    assert (
+        "static constexpr const char * api_action0_strings[] PROGMEM = "
+        "{api_action_str0, api_action_str1};" in main_cpp
+    )
+    assert "USE_API_USER_DEFINED_ACTION_METADATA" not in {d.name for d in CORE.defines}
+
+
+def test_variable_shorthand_normalizes_to_mapping() -> None:
+    """A bare type string validates to the mapping form."""
+    assert validate_variable("string") == {"type": "string"}
+
+
+@pytest.mark.parametrize(
+    "value",
+    [
+        {"description": "no type given"},
+        {"type": "string", "selector": "text"},
+        "stringy",
+        {"type": "stringy"},
+    ],
+)
+def test_variable_rejects_invalid(value: object) -> None:
+    """Missing or unknown type and unknown keys raise in both forms."""
+    with pytest.raises(Invalid):
+        validate_variable(value)
+
+
+def _oversized_action_config() -> dict:
+    return {
+        "actions": [
+            {
+                "action": "big",
+                "description": "x" * 300,
+                "variables": {"a": {"type": "string", "example": "y" * 300}},
+            }
+        ]
+    }
+
+
+def test_esp8266_rejects_actions_over_string_budget(
+    set_core_config: SetCoreConfigCallable,
+) -> None:
+    set_core_config(PlatformFramework.ESP8266_ARDUINO)
+    with pytest.raises(Invalid, match="ESP8266 allows at most 384 bytes"):
+        _validate_esp8266_action_strings(_oversized_action_config())
+
+
+def test_other_platforms_have_no_string_budget(
+    set_core_config: SetCoreConfigCallable,
+) -> None:
+    set_core_config(PlatformFramework.ESP32_IDF)
+    config = _oversized_action_config()
+    assert _validate_esp8266_action_strings(config) is config
+
+
+def test_empty_metadata_is_unset_and_not_counted() -> None:
+    """An empty description or example emits nullptr and takes no scratch space."""
+    conf = {
+        "action": "a",
+        "description": "",
+        "variables": {"b": {"type": "int", "description": "", "example": "ex"}},
+    }
+    strings = _action_strings(conf, has_metadata=True)
+    assert strings == ["a", None, "b", None, "ex"]
+    # Every emitted string counts its terminator: "a" + "b" + "ex"
+    assert _action_strings_size(strings) == 2 + 2 + 3
+
+
+def test_empty_metadata_does_not_enable_the_define() -> None:
+    actions = [
+        {
+            "action": "a",
+            "description": "",
+            "variables": {"b": {"type": "int", "example": ""}},
+        }
+    ]
+    assert not _has_action_metadata(actions)
+    actions[0]["variables"]["b"]["example"] = "1"
+    assert _has_action_metadata(actions)

tests/component_tests/api/test_action_metadata.yaml

@@ -0,0 +1,14 @@
+esphome:
+  name: test
+
+esp32:
+  board: esp32dev
+
+wifi:
+  ssid: MySSID
+  password: password1
+
+logger:
+
+packages:
+  api: !include test_action_metadata_common.yaml

tests/component_tests/api/test_action_metadata_common.yaml

@@ -0,0 +1,18 @@
+api:
+  actions:
+    - action: play_buzzer
+      description: Play an RTTTL melody on the buzzer
+      variables:
+        song_str:
+          type: string
+          description: RTTTL melody string
+          example: "two_short:d=4,o=5,b=100:16e6,16e6"
+        volume:
+          type: int
+      then:
+        - logger.log: Action Called
+    - action: plain_action
+      variables:
+        value: int
+      then:
+        - logger.log: Action Called

tests/component_tests/api/test_action_metadata_esp8266.yaml

@@ -0,0 +1,14 @@
+esphome:
+  name: test
+
+esp8266:
+  board: d1_mini
+
+wifi:
+  ssid: MySSID
+  password: password1
+
+logger:
+
+packages:
+  api: !include test_action_metadata_common.yaml

tests/component_tests/api/test_action_metadata_shorthand.yaml

@@ -0,0 +1,19 @@
+esphome:
+  name: test
+
+esp32:
+  board: esp32dev
+
+wifi:
+  ssid: MySSID
+  password: password1
+
+logger:
+
+api:
+  actions:
+    - action: plain_action
+      variables:
+        value: int
+      then:
+        - logger.log: Action Called

tests/component_tests/api/test_homeassistant_action.py

@@ -9,7 +9,7 @@ def test_synchronous_chain_keeps_zero_copy_args(generate_main):
 
     assert (
         "api::UserServiceTrigger<api::enums::SUPPORTS_RESPONSE_NONE, StringRef>"
-        '("zero_copy_args", {"message"})' in main_cpp
+        "(api_action0_strings," in main_cpp
     )
 
 
@@ -22,7 +22,7 @@ def test_response_callback_args_are_owning(generate_main):
 
     assert (
         "api::UserServiceTrigger<api::enums::SUPPORTS_RESPONSE_NONE, std::string>"
-        '("response_args", {"message"})' in main_cpp
+        "(api_action1_strings," in main_cpp
     )
     assert "api::HomeAssistantServiceCallAction<std::string>" in main_cpp
     assert "api::HomeAssistantServiceCallAction<StringRef>" not in main_cpp

tests/components/api/common-base.yaml

@@ -61,8 +61,12 @@ api:
   reboot_timeout: 0min
   actions:
     - action: hello_world
+      description: Log a greeting
       variables:
-        name: string
+        name:
+          type: string
+          description: Name to greet
+          example: World
       then:
         - logger.log:
             format: Hello World %s!

tests/components/api/common.yaml

@@ -1,4 +1,5 @@
-<<: !include common-base.yaml
+packages:
+  base: !include common-base.yaml
 
 api:
   encryption:

tests/integration/fixtures/api_action_metadata.yaml

@@ -0,0 +1,26 @@
+esphome:
+  name: api-action-metadata-test
+host:
+api:
+  batch_delay: 0ms
+  actions:
+    - action: play_buzzer
+      description: Play an RTTTL melody on the buzzer
+      variables:
+        song_str:
+          type: string
+          description: RTTTL melody string
+          example: "two_short:d=4,o=5,b=100:16e6,16e6"
+        volume:
+          type: int
+      then:
+        - logger.log:
+            format: "Buzzer: %s"
+            args: [song_str.c_str()]
+    - action: plain_action
+      variables:
+        value: int
+      then:
+        - logger.log: "Plain action called"
+
+logger:

tests/integration/test_api_action_metadata.py

@@ -0,0 +1,65 @@
+"""Integration test for user-defined action field metadata."""
+
+from __future__ import annotations
+
+import asyncio
+import re
+
+import pytest
+
+from esphome.helpers import fnv1_hash
+
+from .types import APIClientConnectedFactory, RunCompiledFunction
+
+
+@pytest.mark.asyncio
+async def test_api_action_metadata(
+    yaml_config: str,
+    run_compiled: RunCompiledFunction,
+    api_client_connected: APIClientConnectedFactory,
+) -> None:
+    """Action and argument metadata reach the client and the actions still run."""
+    loop = asyncio.get_running_loop()
+    buzzer_called = loop.create_future()
+    plain_called = loop.create_future()
+    buzzer_pattern = re.compile(r"Buzzer: two_short")
+    plain_pattern = re.compile(r"Plain action called")
+
+    def check_output(line: str) -> None:
+        if not buzzer_called.done() and buzzer_pattern.search(line):
+            buzzer_called.set_result(True)
+        elif not plain_called.done() and plain_pattern.search(line):
+            plain_called.set_result(True)
+
+    async with (
+        run_compiled(yaml_config, line_callback=check_output),
+        api_client_connected() as client,
+    ):
+        _, services = await client.list_entities_services()
+
+        by_name = {service.name: service for service in services}
+        assert set(by_name) == {"play_buzzer", "plain_action"}
+        # Keys are hashed at codegen time and must match what the client expects
+        for name, service in by_name.items():
+            assert service.key == fnv1_hash(name), name
+
+        buzzer = by_name["play_buzzer"]
+        assert buzzer.description == "Play an RTTTL melody on the buzzer"
+        args = {arg.name: arg for arg in buzzer.args}
+        assert args["song_str"].description == "RTTTL melody string"
+        assert args["song_str"].example == "two_short:d=4,o=5,b=100:16e6,16e6"
+        # An arg without metadata sends empty strings
+        assert args["volume"].description == ""
+        assert args["volume"].example == ""
+
+        # An action without metadata sends empty strings
+        plain = by_name["plain_action"]
+        assert plain.description == ""
+        assert plain.args[0].description == ""
+
+        await client.execute_service(
+            buzzer, {"song_str": "two_short:d=4,o=5,b=100:16e6,16e6", "volume": 3}
+        )
+        await client.execute_service(plain, {"value": 1})
+        await asyncio.wait_for(buzzer_called, timeout=5.0)
+        await asyncio.wait_for(plain_called, timeout=5.0)