[test] Trigger device-builder downstream CI

DO NOT MERGE -- chained test PR to verify the new device-builder CI
job from #16214 actually runs and exercises esphome/device-builder
against the PR's Python code.

Touches a Python file under esphome/ to trip the should_run_device_builder
gate in determine-jobs.py.

.github/workflows/auto-label-pr.yml

@@ -27,9 +27,9 @@ jobs:
 
       - name: Generate a token
         id: generate-token
-        uses: actions/create-github-app-token@1b10c78c7865c340bc4f6099eb2f838309f1e8c3 # v2
+        uses: actions/create-github-app-token@1b10c78c7865c340bc4f6099eb2f838309f1e8c3 # v3.1.1
         with:
-          app-id: ${{ secrets.ESPHOME_GITHUB_APP_ID }}
+          client-id: ${{ vars.ESPHOME_GITHUB_APP_CLIENT_ID }}
           private-key: ${{ secrets.ESPHOME_GITHUB_APP_PRIVATE_KEY }}
 
       - name: Auto Label PR

.github/workflows/ci.yml

@@ -136,6 +136,53 @@ jobs:
           if-no-files-found: ignore
           retention-days: 14
 
+  device-builder:
+    name: Test downstream esphome/device-builder
+    runs-on: ubuntu-24.04
+    needs:
+      - common
+      - determine-jobs
+    if: needs.determine-jobs.outputs.device-builder == 'true'
+    steps:
+      - name: Check out esphome (this PR)
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+        with:
+          path: esphome
+      - name: Check out esphome/device-builder
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+        with:
+          repository: esphome/device-builder
+          ref: main
+          path: device-builder
+      - name: Set up Python
+        uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6.2.0
+        with:
+          python-version: "3.13"
+      - name: Set up uv
+        # Mirrors the install shape device-builder's own CI uses
+        # (esphome/device-builder#192): uv replaces pip for the
+        # install step (order-of-magnitude faster on cold boots,
+        # with its own wheel cache). actions/setup-python still
+        # provides the interpreter.
+        uses: astral-sh/setup-uv@08807647e7069bb48b6ef5acd8ec9567f424441b # v8.1.0
+        with:
+          enable-cache: true
+      - name: Install device-builder + esphome from PR
+        # Install device-builder with its esphome + test extras
+        # first so its pinned versions of pytest/etc. land, then
+        # overlay the PR's esphome so the downstream tests run
+        # against this PR's Python code. ``--system`` installs into
+        # the runner's Python instead of a venv.
+        run: |
+          uv pip install --system -e './device-builder[esphome,test]'
+          uv pip install --system -e ./esphome
+      - name: Run device-builder pytest
+        # ``-n auto`` runs under pytest-xdist (matches device-builder's
+        # own CI). No ``--cov`` here -- this is purely a downstream
+        # smoke check against this PR's esphome code.
+        working-directory: device-builder
+        run: pytest -q -n auto --maxfail=5 --durations=10 --no-cov --ignore=tests/benchmarks
+
   pytest:
     name: Run pytest
     strategy:
@@ -199,12 +246,12 @@ jobs:
       - common
     outputs:
       integration-tests: ${{ steps.determine.outputs.integration-tests }}
-      integration-tests-run-all: ${{ steps.determine.outputs.integration-tests-run-all }}
-      integration-test-files: ${{ steps.determine.outputs.integration-test-files }}
+      integration-test-buckets: ${{ steps.determine.outputs.integration-test-buckets }}
       clang-tidy: ${{ steps.determine.outputs.clang-tidy }}
       clang-tidy-mode: ${{ steps.determine.outputs.clang-tidy-mode }}
       python-linters: ${{ steps.determine.outputs.python-linters }}
       import-time: ${{ steps.determine.outputs.import-time }}
+      device-builder: ${{ steps.determine.outputs.device-builder }}
       changed-components: ${{ steps.determine.outputs.changed-components }}
       changed-components-with-tests: ${{ steps.determine.outputs.changed-components-with-tests }}
       directly-changed-components-with-tests: ${{ steps.determine.outputs.directly-changed-components-with-tests }}
@@ -243,12 +290,12 @@ jobs:
 
           # Extract individual fields
           echo "integration-tests=$(echo "$output" | jq -r '.integration_tests')" >> $GITHUB_OUTPUT
-          echo "integration-tests-run-all=$(echo "$output" | jq -r '.integration_tests_run_all')" >> $GITHUB_OUTPUT
-          echo "integration-test-files=$(echo "$output" | jq -c '.integration_test_files')" >> $GITHUB_OUTPUT
+          echo "integration-test-buckets=$(echo "$output" | jq -c '.integration_test_buckets')" >> $GITHUB_OUTPUT
           echo "clang-tidy=$(echo "$output" | jq -r '.clang_tidy')" >> $GITHUB_OUTPUT
           echo "clang-tidy-mode=$(echo "$output" | jq -r '.clang_tidy_mode')" >> $GITHUB_OUTPUT
           echo "python-linters=$(echo "$output" | jq -r '.python_linters')" >> $GITHUB_OUTPUT
           echo "import-time=$(echo "$output" | jq -r '.import_time')" >> $GITHUB_OUTPUT
+          echo "device-builder=$(echo "$output" | jq -r '.device_builder')" >> $GITHUB_OUTPUT
           echo "changed-components=$(echo "$output" | jq -c '.changed_components')" >> $GITHUB_OUTPUT
           echo "changed-components-with-tests=$(echo "$output" | jq -c '.changed_components_with_tests')" >> $GITHUB_OUTPUT
           echo "directly-changed-components-with-tests=$(echo "$output" | jq -c '.directly_changed_components_with_tests')" >> $GITHUB_OUTPUT
@@ -267,12 +314,16 @@ jobs:
           key: components-graph-${{ hashFiles('esphome/components/**/*.py') }}
 
   integration-tests:
-    name: Run integration tests
+    name: Run integration tests (${{ matrix.bucket.name }})
     runs-on: ubuntu-latest
     needs:
       - common
       - determine-jobs
     if: needs.determine-jobs.outputs.integration-tests == 'true'
+    strategy:
+      fail-fast: false
+      matrix:
+        bucket: ${{ fromJson(needs.determine-jobs.outputs.integration-test-buckets) }}
     steps:
       - name: Check out code from GitHub
         uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
@@ -299,19 +350,14 @@ jobs:
         run: echo "::add-matcher::.github/workflows/matchers/pytest.json"
       - name: Run integration tests
         env:
-          INTEGRATION_TEST_FILES: ${{ needs.determine-jobs.outputs.integration-test-files }}
-          INTEGRATION_TESTS_RUN_ALL: ${{ needs.determine-jobs.outputs.integration-tests-run-all }}
+          # JSON array of test paths; parsed into a bash array below to avoid
+          # shell word-splitting / glob hazards.
+          BUCKET_TESTS: ${{ toJson(matrix.bucket.tests) }}
         run: |
           . venv/bin/activate
-          if [[ "$INTEGRATION_TESTS_RUN_ALL" == "true" ]]; then
-            echo "Running all integration tests"
-            pytest -vv --no-cov --tb=native -n auto tests/integration/
-          else
-            # Parse JSON array into bash array to avoid shell expansion issues
-            mapfile -t test_files < <(echo "$INTEGRATION_TEST_FILES" | jq -r '.[]')
-            echo "Running ${#test_files[@]} specific integration tests"
-            pytest -vv --no-cov --tb=native -n auto "${test_files[@]}"
-          fi
+          mapfile -t test_files < <(echo "$BUCKET_TESTS" | jq -r '.[]')
+          echo "Bucket ${{ matrix.bucket.name }}: running ${#test_files[@]} integration tests"
+          pytest -vv --no-cov --tb=native -n auto "${test_files[@]}"
 
   cpp-unit-tests:
     name: Run C++ unit tests
@@ -1066,6 +1112,7 @@ jobs:
       - clang-tidy-nosplit
       - clang-tidy-split
       - determine-jobs
+      - device-builder
       - test-build-components-split
       - pre-commit-ci-lite
       - memory-impact-target-branch

.github/workflows/codeql.yml

@@ -58,7 +58,7 @@ jobs:
 
       # Initializes the CodeQL tools for scanning.
       - name: Initialize CodeQL
-        uses: github/codeql-action/init@95e58e9a2cdfd71adc6e0353d5c52f41a045d225 # v4.35.2
+        uses: github/codeql-action/init@e46ed2cbd01164d986452f91f178727624ae40d7 # v4.35.3
         with:
           languages: ${{ matrix.language }}
           build-mode: ${{ matrix.build-mode }}
@@ -86,6 +86,6 @@ jobs:
           exit 1
 
       - name: Perform CodeQL Analysis
-        uses: github/codeql-action/analyze@95e58e9a2cdfd71adc6e0353d5c52f41a045d225 # v4.35.2
+        uses: github/codeql-action/analyze@e46ed2cbd01164d986452f91f178727624ae40d7 # v4.35.3
         with:
           category: "/language:${{matrix.language}}"

.github/workflows/lock.yml

@@ -8,4 +8,4 @@ on:
 
 jobs:
   lock:
-    uses: esphome/workflows/.github/workflows/lock.yml@3c4e8446aa1029f1c346a482034b3ee1489077ca  # 2026.4.0
+    uses: esphome/workflows/.github/workflows/lock.yml@025a1e6255610c498ed590403b7e510b69e474df  # 2026.4.1

.github/workflows/release.yml

@@ -223,7 +223,7 @@ jobs:
         id: generate-token
         uses: actions/create-github-app-token@1b10c78c7865c340bc4f6099eb2f838309f1e8c3 # v3.1.1
         with:
-          app-id: ${{ secrets.ESPHOME_GITHUB_APP_ID }}
+          client-id: ${{ vars.ESPHOME_GITHUB_APP_CLIENT_ID }}
           private-key: ${{ secrets.ESPHOME_GITHUB_APP_PRIVATE_KEY }}
           owner: esphome
           repositories: home-assistant-addon
@@ -258,7 +258,7 @@ jobs:
         id: generate-token
         uses: actions/create-github-app-token@1b10c78c7865c340bc4f6099eb2f838309f1e8c3 # v3.1.1
         with:
-          app-id: ${{ secrets.ESPHOME_GITHUB_APP_ID }}
+          client-id: ${{ vars.ESPHOME_GITHUB_APP_CLIENT_ID }}
           private-key: ${{ secrets.ESPHOME_GITHUB_APP_PRIVATE_KEY }}
           owner: esphome
           repositories: esphome-schema
@@ -289,7 +289,7 @@ jobs:
         id: generate-token
         uses: actions/create-github-app-token@1b10c78c7865c340bc4f6099eb2f838309f1e8c3 # v3.1.1
         with:
-          app-id: ${{ secrets.ESPHOME_GITHUB_APP_ID }}
+          client-id: ${{ vars.ESPHOME_GITHUB_APP_CLIENT_ID }}
           private-key: ${{ secrets.ESPHOME_GITHUB_APP_PRIVATE_KEY }}
           owner: esphome
           repositories: version-notifier

esphome/__main__.py

@@ -1125,15 +1125,16 @@ def upload_program(
 
     remote_port = int(ota_conf[CONF_PORT])
     password = ota_conf.get(CONF_PASSWORD)
-    if getattr(args, "file", None) is not None:
-        binary = Path(args.file)
-    else:
-        binary = CORE.firmware_bin
 
     # Resolve MQTT magic strings to actual IP addresses
     network_devices = _resolve_network_devices(devices, config, args)
 
-    return espota2.run_ota(network_devices, remote_port, password, binary)
+    binary = CORE.firmware_bin
+    ota_type = espota2.OTA_TYPE_UPDATE_APP
+    if getattr(args, "file", None) is not None:
+        binary = Path(args.file)
+
+    return espota2.run_ota(network_devices, remote_port, password, binary, ota_type)
 
 
 def show_logs(config: ConfigType, args: ArgsProtocol, devices: list[str]) -> int | None:

esphome/components/audio/__init__.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 
 import esphome.codegen as cg
 from esphome.components.esp32 import (
@@ -7,7 +7,12 @@ from esphome.components.esp32 import (
     include_builtin_idf_component,
 )
 import esphome.config_validation as cv
-from esphome.const import CONF_BITS_PER_SAMPLE, CONF_NUM_CHANNELS, CONF_SAMPLE_RATE
+from esphome.const import (
+    CONF_BITS_PER_SAMPLE,
+    CONF_NUM_CHANNELS,
+    CONF_SAMPLE_RATE,
+    CONF_SIZE,
+)
 from esphome.core import CORE
 import esphome.final_validate as fv
 
@@ -25,13 +30,46 @@ AUDIO_FILE_TYPE_ENUM = {
     "OPUS": AudioFileType.OPUS,
 }
 
+MEMORY_PSRAM = "psram"
+MEMORY_INTERNAL = "internal"
+MEMORY_LOCATIONS = [MEMORY_PSRAM, MEMORY_INTERNAL]
+
+
+@dataclass
+class FlacOptions:
+    buffer_memory: str | None = None
+
+
+@dataclass
+class Mp3Options:
+    buffer_memory: str | None = None
+
+
+@dataclass
+class OpusPseudostackOptions:
+    threadsafe: bool | None = None
+    buffer_memory: str | None = None
+    size: int | None = None
+
+
+@dataclass
+class OpusOptions:
+    floating_point: bool | None = None
+    state_memory: str | None = None
+    pseudostack: OpusPseudostackOptions = field(default_factory=OpusPseudostackOptions)
+
 
 @dataclass
 class AudioData:
     flac_support: bool = False
     mp3_support: bool = False
     opus_support: bool = False
+    # WAV defaults to True for backward compatibility; will become opt-in in a future release
+    wav_support: bool = True
     micro_decoder_support: bool = False
+    flac: FlacOptions = field(default_factory=FlacOptions)
+    mp3: Mp3Options = field(default_factory=Mp3Options)
+    opus: OpusOptions = field(default_factory=OpusOptions)
 
 
 def _get_data() -> AudioData:
@@ -55,6 +93,11 @@ def request_opus_support() -> None:
     _get_data().opus_support = True
 
 
+def request_wav_support() -> None:
+    """Request WAV codec support for audio decoding."""
+    _get_data().wav_support = True
+
+
 def request_micro_decoder_support() -> None:
     """Request micro-decoder library support for audio decoding."""
     _get_data().micro_decoder_support = True
@@ -67,9 +110,78 @@ CONF_MAX_CHANNELS = "max_channels"
 CONF_MIN_SAMPLE_RATE = "min_sample_rate"
 CONF_MAX_SAMPLE_RATE = "max_sample_rate"
 
+CONF_CODECS = "codecs"
+CONF_WAV = "wav"
+CONF_FLAC = "flac"
+CONF_MP3 = "mp3"
+CONF_OPUS = "opus"
+CONF_BUFFER_MEMORY = "buffer_memory"
+CONF_FLOATING_POINT = "floating_point"
+CONF_STATE_MEMORY = "state_memory"
+CONF_PSEUDOSTACK = "pseudostack"
+CONF_THREADSAFE = "threadsafe"
+
+
+_MEMORY_LOCATION_VALIDATOR = cv.one_of(*MEMORY_LOCATIONS, lower=True)
+
+
+def _maybe_empty_codec(schema):
+    """Wrap a codec dict schema so that a bare key (None value) is treated as an empty dict."""
+
+    def validator(value):
+        if value is None:
+            value = {}
+        return schema(value)
+
+    return validator
+
+
+CODEC_FLAC_SCHEMA = cv.Schema(
+    {
+        cv.Optional(CONF_BUFFER_MEMORY): _MEMORY_LOCATION_VALIDATOR,
+    }
+)
+
+CODEC_MP3_SCHEMA = cv.Schema(
+    {
+        cv.Optional(CONF_BUFFER_MEMORY): _MEMORY_LOCATION_VALIDATOR,
+    }
+)
+
+OPUS_PSEUDOSTACK_SCHEMA = cv.Schema(
+    {
+        cv.Optional(CONF_THREADSAFE): cv.boolean,
+        cv.Optional(CONF_BUFFER_MEMORY): _MEMORY_LOCATION_VALIDATOR,
+        cv.Optional(CONF_SIZE): cv.int_range(60000, 240000),
+    }
+)
+
+CODEC_OPUS_SCHEMA = cv.Schema(
+    {
+        cv.Optional(CONF_FLOATING_POINT): cv.boolean,
+        cv.Optional(CONF_STATE_MEMORY): _MEMORY_LOCATION_VALIDATOR,
+        cv.Optional(CONF_PSEUDOSTACK): _maybe_empty_codec(OPUS_PSEUDOSTACK_SCHEMA),
+    }
+)
+
+CODEC_WAV_SCHEMA = cv.Schema({})
+
+CODECS_SCHEMA = cv.Schema(
+    {
+        cv.Optional(CONF_FLAC): _maybe_empty_codec(CODEC_FLAC_SCHEMA),
+        cv.Optional(CONF_MP3): _maybe_empty_codec(CODEC_MP3_SCHEMA),
+        cv.Optional(CONF_OPUS): _maybe_empty_codec(CODEC_OPUS_SCHEMA),
+        cv.Optional(CONF_WAV): _maybe_empty_codec(CODEC_WAV_SCHEMA),
+    }
+)
 
 CONFIG_SCHEMA = cv.All(
-    cv.Schema({}),
+    cv.Schema(
+        {
+            cv.Optional(CONF_CODECS): _maybe_empty_codec(CODECS_SCHEMA),
+        }
+    ),
+    cv.only_on_esp32,
 )
 
 AUDIO_COMPONENT_SCHEMA = cv.Schema(
@@ -208,6 +320,15 @@ def final_validate_audio_schema(
     )
 
 
+def _emit_memory_pair(value: str | None, psram_key: str, internal_key: str) -> None:
+    if value == MEMORY_PSRAM:
+        add_idf_sdkconfig_option(psram_key, True)
+        add_idf_sdkconfig_option(internal_key, False)
+    elif value == MEMORY_INTERNAL:
+        add_idf_sdkconfig_option(psram_key, False)
+        add_idf_sdkconfig_option(internal_key, True)
+
+
 async def to_code(config):
     # Re-enable ESP-IDF's HTTP client (excluded by default to save compile time)
     include_builtin_idf_component("esp_http_client")
@@ -219,6 +340,36 @@ async def to_code(config):
 
     data = _get_data()
 
+    # Merge user-supplied codec configuration (additive: presence enables the codec)
+    if codecs_config := config.get(CONF_CODECS):
+        if (flac_config := codecs_config.get(CONF_FLAC)) is not None:
+            data.flac_support = True
+            if (buffer_memory := flac_config.get(CONF_BUFFER_MEMORY)) is not None:
+                data.flac.buffer_memory = buffer_memory
+        if (mp3_config := codecs_config.get(CONF_MP3)) is not None:
+            data.mp3_support = True
+            if (buffer_memory := mp3_config.get(CONF_BUFFER_MEMORY)) is not None:
+                data.mp3.buffer_memory = buffer_memory
+        if (opus_config := codecs_config.get(CONF_OPUS)) is not None:
+            data.opus_support = True
+            floating_point = opus_config.get(CONF_FLOATING_POINT)
+            if floating_point is not None:
+                data.opus.floating_point = floating_point
+            if (state_memory := opus_config.get(CONF_STATE_MEMORY)) is not None:
+                data.opus.state_memory = state_memory
+            if (pseudostack_config := opus_config.get(CONF_PSEUDOSTACK)) is not None:
+                threadsafe = pseudostack_config.get(CONF_THREADSAFE)
+                if threadsafe is not None:
+                    data.opus.pseudostack.threadsafe = threadsafe
+                if (
+                    buffer_memory := pseudostack_config.get(CONF_BUFFER_MEMORY)
+                ) is not None:
+                    data.opus.pseudostack.buffer_memory = buffer_memory
+                if (size := pseudostack_config.get(CONF_SIZE)) is not None:
+                    data.opus.pseudostack.size = size
+        if CONF_WAV in codecs_config:
+            data.wav_support = True
+
     if data.micro_decoder_support:
         add_idf_component(name="esphome/micro-decoder", ref="0.2.0")
 
@@ -229,13 +380,50 @@ async def to_code(config):
             add_idf_sdkconfig_option("CONFIG_MICRO_DECODER_CODEC_MP3", False)
         if not data.opus_support:
             add_idf_sdkconfig_option("CONFIG_MICRO_DECODER_CODEC_OPUS", False)
+        if not data.wav_support:
+            add_idf_sdkconfig_option("CONFIG_MICRO_DECODER_CODEC_WAV", False)
 
-    # Legacy audio_decoder.cpp support defines and components
+    # Configure each codec library.
+    # Adds a define and IDF component for legacy `audio_decoder.cpp`.
     if data.flac_support:
         cg.add_define("USE_AUDIO_FLAC_SUPPORT")
         add_idf_component(name="esphome/micro-flac", ref="0.1.1")
+        _emit_memory_pair(
+            data.flac.buffer_memory,
+            "CONFIG_MICRO_FLAC_PREFER_PSRAM",
+            "CONFIG_MICRO_FLAC_PREFER_INTERNAL",
+        )
     if data.mp3_support:
         cg.add_define("USE_AUDIO_MP3_SUPPORT")
+        _emit_memory_pair(
+            data.mp3.buffer_memory,
+            "CONFIG_MP3_DECODER_PREFER_PSRAM",
+            "CONFIG_MP3_DECODER_PREFER_INTERNAL",
+        )
     if data.opus_support:
         cg.add_define("USE_AUDIO_OPUS_SUPPORT")
-        add_idf_component(name="esphome/micro-opus", ref="0.3.6")
+        add_idf_component(name="esphome/micro-opus", ref="0.4.0")
+        if data.opus.floating_point is not None:
+            add_idf_sdkconfig_option(
+                "CONFIG_OPUS_FLOATING_POINT", data.opus.floating_point
+            )
+        _emit_memory_pair(
+            data.opus.state_memory,
+            "CONFIG_OPUS_STATE_PREFER_PSRAM",
+            "CONFIG_OPUS_STATE_PREFER_INTERNAL",
+        )
+        if data.opus.pseudostack.threadsafe is True:
+            add_idf_sdkconfig_option("CONFIG_OPUS_THREADSAFE_PSEUDOSTACK", True)
+            add_idf_sdkconfig_option("CONFIG_OPUS_NONTHREADSAFE_PSEUDOSTACK", False)
+        elif data.opus.pseudostack.threadsafe is False:
+            add_idf_sdkconfig_option("CONFIG_OPUS_THREADSAFE_PSEUDOSTACK", False)
+            add_idf_sdkconfig_option("CONFIG_OPUS_NONTHREADSAFE_PSEUDOSTACK", True)
+        _emit_memory_pair(
+            data.opus.pseudostack.buffer_memory,
+            "CONFIG_OPUS_PSEUDOSTACK_PREFER_PSRAM",
+            "CONFIG_OPUS_PSEUDOSTACK_PREFER_INTERNAL",
+        )
+        if data.opus.pseudostack.size is not None:
+            add_idf_sdkconfig_option(
+                "CONFIG_OPUS_PSEUDOSTACK_SIZE", data.opus.pseudostack.size
+            )

esphome/components/beken_spi_led_strip/light.py

@@ -62,6 +62,7 @@ CONF_IS_WRGB = "is_wrgb"
 SUPPORTED_PINS = {
     libretiny.const.FAMILY_BK7231N: [16],
     libretiny.const.FAMILY_BK7231T: [16],
+    libretiny.const.FAMILY_BK7238: [16],
     libretiny.const.FAMILY_BK7251: [16],
 }
 

esphome/components/binary_sensor/__init__.py

@@ -143,15 +143,15 @@ BinarySensorCondition = binary_sensor_ns.class_("BinarySensorCondition", Conditi
 
 # Filters
 Filter = binary_sensor_ns.class_("Filter")
-TimeoutFilter = binary_sensor_ns.class_("TimeoutFilter", Filter, cg.Component)
-DelayedOnOffFilter = binary_sensor_ns.class_("DelayedOnOffFilter", Filter, cg.Component)
-DelayedOnFilter = binary_sensor_ns.class_("DelayedOnFilter", Filter, cg.Component)
-DelayedOffFilter = binary_sensor_ns.class_("DelayedOffFilter", Filter, cg.Component)
+TimeoutFilter = binary_sensor_ns.class_("TimeoutFilter", Filter)
+DelayedOnOffFilter = binary_sensor_ns.class_("DelayedOnOffFilter", Filter)
+DelayedOnFilter = binary_sensor_ns.class_("DelayedOnFilter", Filter)
+DelayedOffFilter = binary_sensor_ns.class_("DelayedOffFilter", Filter)
 InvertFilter = binary_sensor_ns.class_("InvertFilter", Filter)
 AutorepeatFilter = binary_sensor_ns.class_("AutorepeatFilter", Filter, cg.Component)
 LambdaFilter = binary_sensor_ns.class_("LambdaFilter", Filter)
 StatelessLambdaFilter = binary_sensor_ns.class_("StatelessLambdaFilter", Filter)
-SettleFilter = binary_sensor_ns.class_("SettleFilter", Filter, cg.Component)
+SettleFilter = binary_sensor_ns.class_("SettleFilter", Filter)
 
 _LOGGER = getLogger(__name__)
 
@@ -175,7 +175,6 @@ async def invert_filter_to_code(config, filter_id):
 )
 async def timeout_filter_to_code(config, filter_id):
     var = cg.new_Pvariable(filter_id)
-    await cg.register_component(var, {})
     template_ = await cg.templatable(config, [], cg.uint32)
     cg.add(var.set_timeout_value(template_))
     return var
@@ -203,7 +202,6 @@ async def timeout_filter_to_code(config, filter_id):
 )
 async def delayed_on_off_filter_to_code(config, filter_id):
     var = cg.new_Pvariable(filter_id)
-    await cg.register_component(var, {})
     if isinstance(config, dict):
         template_ = await cg.templatable(config[CONF_TIME_ON], [], cg.uint32)
         cg.add(var.set_on_delay(template_))
@@ -221,7 +219,6 @@ async def delayed_on_off_filter_to_code(config, filter_id):
 )
 async def delayed_on_filter_to_code(config, filter_id):
     var = cg.new_Pvariable(filter_id)
-    await cg.register_component(var, {})
     template_ = await cg.templatable(config, [], cg.uint32)
     cg.add(var.set_delay(template_))
     return var
@@ -234,7 +231,6 @@ async def delayed_on_filter_to_code(config, filter_id):
 )
 async def delayed_off_filter_to_code(config, filter_id):
     var = cg.new_Pvariable(filter_id)
-    await cg.register_component(var, {})
     template_ = await cg.templatable(config, [], cg.uint32)
     cg.add(var.set_delay(template_))
     return var
@@ -306,7 +302,6 @@ async def lambda_filter_to_code(config, filter_id):
 )
 async def settle_filter_to_code(config, filter_id):
     var = cg.new_Pvariable(filter_id)
-    await cg.register_component(var, {})
     template_ = await cg.templatable(config, [], cg.uint32)
     cg.add(var.set_delay(template_))
     return var

esphome/components/binary_sensor/filter.cpp

@@ -4,16 +4,14 @@
 #include "filter.h"
 
 #include "binary_sensor.h"
+#include "esphome/core/application.h"
 
 namespace esphome::binary_sensor {
 
 static const char *const TAG = "sensor.filter";
 
-// Timeout IDs for filter classes.
-// Each filter is its own Component instance, so the scheduler scopes
-// IDs by component pointer — no risk of collisions between instances.
-constexpr uint32_t FILTER_TIMEOUT_ID = 0;
-// AutorepeatFilter needs two distinct IDs (both timeouts on the same component)
+// AutorepeatFilter still inherits Component (it schedules two distinct timer
+// purposes), so it keeps the (Component *, id) scheduler API.
 constexpr uint32_t AUTOREPEAT_TIMING_ID = 0;
 constexpr uint32_t AUTOREPEAT_ON_OFF_ID = 1;
 
@@ -34,46 +32,40 @@ void Filter::input(bool value) {
 }
 
 void TimeoutFilter::input(bool value) {
-  this->set_timeout(FILTER_TIMEOUT_ID, this->timeout_delay_.value(), [this]() { this->parent_->invalidate_state(); });
+  App.scheduler.set_timeout(this, this->timeout_delay_.value(), [this]() { this->parent_->invalidate_state(); });
   // we do not de-dup here otherwise changes from invalid to valid state will not be output
   this->output(value);
 }
 
 optional<bool> DelayedOnOffFilter::new_value(bool value) {
   if (value) {
-    this->set_timeout(FILTER_TIMEOUT_ID, this->on_delay_.value(), [this]() { this->output(true); });
+    App.scheduler.set_timeout(this, this->on_delay_.value(), [this]() { this->output(true); });
   } else {
-    this->set_timeout(FILTER_TIMEOUT_ID, this->off_delay_.value(), [this]() { this->output(false); });
+    App.scheduler.set_timeout(this, this->off_delay_.value(), [this]() { this->output(false); });
   }
   return {};
 }
 
-float DelayedOnOffFilter::get_setup_priority() const { return setup_priority::HARDWARE; }
-
 optional<bool> DelayedOnFilter::new_value(bool value) {
   if (value) {
-    this->set_timeout(FILTER_TIMEOUT_ID, this->delay_.value(), [this]() { this->output(true); });
+    App.scheduler.set_timeout(this, this->delay_.value(), [this]() { this->output(true); });
     return {};
   } else {
-    this->cancel_timeout(FILTER_TIMEOUT_ID);
+    App.scheduler.cancel_timeout(this);
     return false;
   }
 }
 
-float DelayedOnFilter::get_setup_priority() const { return setup_priority::HARDWARE; }
-
 optional<bool> DelayedOffFilter::new_value(bool value) {
   if (!value) {
-    this->set_timeout(FILTER_TIMEOUT_ID, this->delay_.value(), [this]() { this->output(false); });
+    App.scheduler.set_timeout(this, this->delay_.value(), [this]() { this->output(false); });
     return {};
   } else {
-    this->cancel_timeout(FILTER_TIMEOUT_ID);
+    App.scheduler.cancel_timeout(this);
     return true;
   }
 }
 
-float DelayedOffFilter::get_setup_priority() const { return setup_priority::HARDWARE; }
-
 optional<bool> InvertFilter::new_value(bool value) { return !value; }
 
 // AutorepeatFilterBase
@@ -118,20 +110,18 @@ optional<bool> LambdaFilter::new_value(bool value) { return this->f_(value); }
 
 optional<bool> SettleFilter::new_value(bool value) {
   if (!this->steady_) {
-    this->set_timeout(FILTER_TIMEOUT_ID, this->delay_.value(), [this, value]() {
+    App.scheduler.set_timeout(this, this->delay_.value(), [this, value]() {
       this->steady_ = true;
       this->output(value);
     });
     return {};
   } else {
     this->steady_ = false;
-    this->set_timeout(FILTER_TIMEOUT_ID, this->delay_.value(), [this]() { this->steady_ = true; });
+    App.scheduler.set_timeout(this, this->delay_.value(), [this]() { this->steady_ = true; });
     return value;
   }
 }
 
-float SettleFilter::get_setup_priority() const { return setup_priority::HARDWARE; }
-
 }  // namespace esphome::binary_sensor
 
 #endif  // USE_BINARY_SENSOR_FILTER

esphome/components/binary_sensor/filter.h

@@ -29,7 +29,7 @@ class Filter {
   Deduplicator<bool> dedup_;
 };
 
-class TimeoutFilter : public Filter, public Component {
+class TimeoutFilter : public Filter {
  public:
   optional<bool> new_value(bool value) override { return value; }
   void input(bool value) override;
@@ -39,12 +39,10 @@ class TimeoutFilter : public Filter, public Component {
   TemplatableFn<uint32_t> timeout_delay_{};
 };
 
-class DelayedOnOffFilter final : public Filter, public Component {
+class DelayedOnOffFilter final : public Filter {
  public:
   optional<bool> new_value(bool value) override;
 
-  float get_setup_priority() const override;
-
   template<typename T> void set_on_delay(T delay) { this->on_delay_ = delay; }
   template<typename T> void set_off_delay(T delay) { this->off_delay_ = delay; }
 
@@ -53,24 +51,20 @@ class DelayedOnOffFilter final : public Filter, public Component {
   TemplatableFn<uint32_t> off_delay_{};
 };
 
-class DelayedOnFilter : public Filter, public Component {
+class DelayedOnFilter : public Filter {
  public:
   optional<bool> new_value(bool value) override;
 
-  float get_setup_priority() const override;
-
   template<typename T> void set_delay(T delay) { this->delay_ = delay; }
 
  protected:
   TemplatableFn<uint32_t> delay_{};
 };
 
-class DelayedOffFilter : public Filter, public Component {
+class DelayedOffFilter : public Filter {
  public:
   optional<bool> new_value(bool value) override;
 
-  float get_setup_priority() const override;
-
   template<typename T> void set_delay(T delay) { this->delay_ = delay; }
 
  protected:
@@ -146,12 +140,10 @@ class StatelessLambdaFilter : public Filter {
   optional<bool> (*f_)(bool);
 };
 
-class SettleFilter : public Filter, public Component {
+class SettleFilter : public Filter {
  public:
   optional<bool> new_value(bool value) override;
 
-  float get_setup_priority() const override;
-
   template<typename T> void set_delay(T delay) { this->delay_ = delay; }
 
  protected:

esphome/components/climate/__init__.py

@@ -48,13 +48,13 @@ from esphome.const import (
     CONF_VISUAL,
     CONF_WEB_SERVER,
 )
-from esphome.core import CORE, CoroPriority, coroutine_with_priority
+from esphome.core import CORE, CoroPriority, Lambda, coroutine_with_priority
 from esphome.core.entity_helpers import (
     entity_duplicate_validator,
     queue_entity_register,
     setup_entity,
 )
-from esphome.cpp_generator import MockObjClass
+from esphome.cpp_generator import LambdaExpression, MockObjClass
 
 IS_PLATFORM_COMPONENT = True
 
@@ -487,38 +487,57 @@ CLIMATE_CONTROL_ACTION_SCHEMA = cv.Schema(
 )
 async def climate_control_to_code(config, action_id, template_arg, args):
     paren = await cg.get_variable(config[CONF_ID])
-    var = cg.new_Pvariable(action_id, template_arg, paren)
-    if (mode := config.get(CONF_MODE)) is not None:
-        template_ = await cg.templatable(mode, args, ClimateMode)
-        cg.add(var.set_mode(template_))
-    if (target_temp := config.get(CONF_TARGET_TEMPERATURE)) is not None:
-        template_ = await cg.templatable(target_temp, args, cg.float_)
-        cg.add(var.set_target_temperature(template_))
-    if (target_temp_low := config.get(CONF_TARGET_TEMPERATURE_LOW)) is not None:
-        template_ = await cg.templatable(target_temp_low, args, cg.float_)
-        cg.add(var.set_target_temperature_low(template_))
-    if (target_temp_high := config.get(CONF_TARGET_TEMPERATURE_HIGH)) is not None:
-        template_ = await cg.templatable(target_temp_high, args, cg.float_)
-        cg.add(var.set_target_temperature_high(template_))
-    if (target_humidity := config.get(CONF_TARGET_HUMIDITY)) is not None:
-        template_ = await cg.templatable(target_humidity, args, cg.float_)
-        cg.add(var.set_target_humidity(template_))
-    if (fan_mode := config.get(CONF_FAN_MODE)) is not None:
-        template_ = await cg.templatable(fan_mode, args, ClimateFanMode)
-        cg.add(var.set_fan_mode(template_))
-    if (custom_fan_mode := config.get(CONF_CUSTOM_FAN_MODE)) is not None:
-        template_ = await cg.templatable(custom_fan_mode, args, cg.std_string)
-        cg.add(var.set_custom_fan_mode(template_))
-    if (preset := config.get(CONF_PRESET)) is not None:
-        template_ = await cg.templatable(preset, args, ClimatePreset)
-        cg.add(var.set_preset(template_))
-    if (custom_preset := config.get(CONF_CUSTOM_PRESET)) is not None:
-        template_ = await cg.templatable(custom_preset, args, cg.std_string)
-        cg.add(var.set_custom_preset(template_))
-    if (swing_mode := config.get(CONF_SWING_MODE)) is not None:
-        template_ = await cg.templatable(swing_mode, args, ClimateSwingMode)
-        cg.add(var.set_swing_mode(template_))
-    return var
+
+    # All configured fields are folded into a single stateless lambda whose
+    # constants live in flash; the action stores only a function pointer.
+    # For custom_fan_mode/custom_preset the static-string path emits the
+    # (const char *, size_t) overload of set_fan_mode/set_preset to avoid
+    # constructing a std::string and calling runtime strlen.
+    FIELDS = (
+        (CONF_MODE, "set_mode", ClimateMode),
+        (CONF_TARGET_TEMPERATURE, "set_target_temperature", cg.float_),
+        (CONF_TARGET_TEMPERATURE_LOW, "set_target_temperature_low", cg.float_),
+        (CONF_TARGET_TEMPERATURE_HIGH, "set_target_temperature_high", cg.float_),
+        (CONF_TARGET_HUMIDITY, "set_target_humidity", cg.float_),
+        (CONF_FAN_MODE, "set_fan_mode", ClimateFanMode),
+        (CONF_CUSTOM_FAN_MODE, "set_fan_mode", cg.std_string),
+        (CONF_PRESET, "set_preset", ClimatePreset),
+        (CONF_CUSTOM_PRESET, "set_preset", cg.std_string),
+        (CONF_SWING_MODE, "set_swing_mode", ClimateSwingMode),
+    )
+
+    fwd_args = ", ".join(name for _, name in args)
+    body_lines: list[str] = []
+
+    for conf_key, setter, type_ in FIELDS:
+        if (value := config.get(conf_key)) is None:
+            continue
+        if isinstance(value, Lambda):
+            inner = await cg.process_lambda(value, args, return_type=type_)
+            body_lines.append(f"call.{setter}(({inner})({fwd_args}));")
+        elif type_ is cg.std_string:
+            # Static custom strings: emit a flash literal and pass the
+            # UTF-8 byte length to skip the runtime strlen inside
+            # set_fan_mode/set_preset.
+            literal = cg.safe_exp(value)
+            body_lines.append(
+                f"call.{setter}({literal}, {len(value.encode('utf-8'))});"
+            )
+        else:
+            body_lines.append(f"call.{setter}({cg.safe_exp(value)});")
+
+    # Match ControlAction::ApplyFn signature: const Ts &... for trigger args.
+    apply_args = [
+        (ClimateCall.operator("ref"), "call"),
+        *((t.operator("const").operator("ref"), n) for t, n in args),
+    ]
+    apply_lambda = LambdaExpression(
+        ["\n".join(body_lines)],
+        apply_args,
+        capture="",
+        return_type=cg.void,
+    )
+    return cg.new_Pvariable(action_id, template_arg, paren, apply_lambda)
 
 
 @coroutine_with_priority(CoroPriority.CORE)

esphome/components/climate/automation.h

@@ -5,42 +5,25 @@
 
 namespace esphome::climate {
 
+// All configured fields are baked into a single stateless lambda whose
+// constants live in flash. The action only stores one function pointer
+// plus one parent pointer, regardless of how many fields the user set.
+// Trigger args are forwarded to the apply function so user lambdas
+// (e.g. `target_temperature: !lambda "return x;"`) keep working.
 template<typename... Ts> class ControlAction : public Action<Ts...> {
  public:
-  explicit ControlAction(Climate *climate) : climate_(climate) {}
-
-  TEMPLATABLE_VALUE(ClimateMode, mode)
-  TEMPLATABLE_VALUE(float, target_temperature)
-  TEMPLATABLE_VALUE(float, target_temperature_low)
-  TEMPLATABLE_VALUE(float, target_temperature_high)
-  TEMPLATABLE_VALUE(float, target_humidity)
-  TEMPLATABLE_VALUE(bool, away)
-  TEMPLATABLE_VALUE(ClimateFanMode, fan_mode)
-  TEMPLATABLE_VALUE(std::string, custom_fan_mode)
-  TEMPLATABLE_VALUE(ClimatePreset, preset)
-  TEMPLATABLE_VALUE(std::string, custom_preset)
-  TEMPLATABLE_VALUE(ClimateSwingMode, swing_mode)
+  using ApplyFn = void (*)(ClimateCall &, const Ts &...);
+  ControlAction(Climate *climate, ApplyFn apply) : climate_(climate), apply_(apply) {}
 
   void play(const Ts &...x) override {
     auto call = this->climate_->make_call();
-    call.set_mode(this->mode_.optional_value(x...));
-    call.set_target_temperature(this->target_temperature_.optional_value(x...));
-    call.set_target_temperature_low(this->target_temperature_low_.optional_value(x...));
-    call.set_target_temperature_high(this->target_temperature_high_.optional_value(x...));
-    call.set_target_humidity(this->target_humidity_.optional_value(x...));
-    if (away_.has_value()) {
-      call.set_preset(away_.value(x...) ? CLIMATE_PRESET_AWAY : CLIMATE_PRESET_HOME);
-    }
-    call.set_fan_mode(this->fan_mode_.optional_value(x...));
-    call.set_fan_mode(this->custom_fan_mode_.optional_value(x...));
-    call.set_preset(this->preset_.optional_value(x...));
-    call.set_preset(this->custom_preset_.optional_value(x...));
-    call.set_swing_mode(this->swing_mode_.optional_value(x...));
+    this->apply_(call, x...);
     call.perform();
   }
 
  protected:
   Climate *climate_;
+  ApplyFn apply_;
 };
 
 class ControlTrigger : public Trigger<ClimateCall &> {

esphome/components/esp32/__init__.py

@@ -1753,7 +1753,17 @@ async def to_code(config):
 
         # Wrap FILE*-based printf functions to eliminate newlib's _vfprintf_r
         # (~11 KB). See printf_stubs.cpp for implementation.
-        if conf[CONF_ADVANCED][CONF_ENABLE_FULL_PRINTF]:
+        #
+        # The wrap is only beneficial against newlib. Picolibc's tinystdio
+        # implements vsnprintf by building a string-output FILE and calling
+        # vfprintf, so vfprintf is unconditionally linked in by any caller
+        # of snprintf/vsnprintf — effectively every build — and the wrap
+        # saves nothing while costing ~170 B of shim. IDF 5.x defaults to
+        # newlib on every variant; IDF 6.0+ switches to picolibc on every
+        # variant.
+        if conf[CONF_ADVANCED][CONF_ENABLE_FULL_PRINTF] or idf_version() >= cv.Version(
+            6, 0, 0
+        ):
             cg.add_define("USE_FULL_PRINTF")
         else:
             for symbol in ("vprintf", "printf", "fprintf", "vfprintf"):

esphome/components/esp32/hal.h

@@ -15,6 +15,8 @@
 #define PROGMEM
 #endif
 
+namespace esphome::esp32 {}
+
 namespace esphome {
 
 // Forward decl from helpers.h (esphome/core/helpers.h) — kept here so this
@@ -42,6 +44,9 @@ __attribute__((always_inline)) inline void delayMicroseconds(uint32_t us) { dela
 __attribute__((always_inline)) inline void arch_feed_wdt() { esp_task_wdt_reset(); }
 __attribute__((always_inline)) inline uint32_t arch_get_cpu_cycle_count() { return esp_cpu_get_cycle_count(); }
 
+void arch_init();
+uint32_t arch_get_cpu_freq_hz();
+
 }  // namespace esphome
 
 #endif  // USE_ESP32

esphome/components/esp32/printf_stubs.cpp

@@ -1,32 +1,38 @@
 /*
- * Linker wrap stubs for FILE*-based printf functions.
+ * Linker wrap stubs for FILE*-based printf functions (newlib only).
  *
  * ESP-IDF SDK components (gpio driver, ringbuf, log_write) reference
- * fprintf(), printf(), vprintf(), and vfprintf() which pull in the full
- * printf implementation (~11 KB on newlib's _vfprintf_r, ~2.8 KB on
- * picolibc's vfprintf). This is a separate implementation from the one
- * used by snprintf/vsnprintf that handles FILE* stream I/O with buffering
- * and locking.
+ * fprintf(), printf(), vprintf(), and vfprintf(), which on newlib pull
+ * in _vfprintf_r (~11 KB) — a separate implementation from the one used
+ * by snprintf/vsnprintf that handles FILE* stream I/O with buffering.
  *
  * ESPHome replaces the ESP-IDF log handler via esp_log_set_vprintf_(),
  * so the SDK's vprintf() path is dead code at runtime. The fprintf()
  * and printf() calls in SDK components are only in debug/assert paths
  * (gpio_dump_io_configuration, ringbuf diagnostics) that are either
  * GC'd or never called. Crash backtraces and panic output are
- * unaffected — they use esp_rom_printf() which is a ROM function
- * and does not go through libc.
+ * unaffected; they use esp_rom_printf() which is a ROM function and
+ * does not go through libc.
  *
- * These stubs redirect through vsnprintf() (which uses _svfprintf_r
- * already in the binary) and fwrite(), allowing the linker to
- * dead-code eliminate _vfprintf_r.
+ * This wrap is newlib-only. On picolibc, vsnprintf is implemented as
+ * vfprintf into a string-output FILE, so vfprintf is unconditionally
+ * linked in by any caller of snprintf/vsnprintf and the wrap can never
+ * elide it — it just adds shim cost. Codegen forces USE_FULL_PRINTF
+ * on picolibc builds (IDF 6.0+ on all variants) so this file compiles
+ * to nothing there; the #error below catches a desynchronised gate.
  *
- * Saves ~11 KB of flash.
+ * Saves ~11 KB of flash on newlib.
  *
- * To disable these wraps, set enable_full_printf: true in the esp32
- * advanced config section.
+ * To disable this wrap on newlib, set enable_full_printf: true in the
+ * esp32 advanced config section.
  */
 
 #if defined(USE_ESP_IDF) && !defined(USE_FULL_PRINTF)
+
+#ifdef __PICOLIBC__
+#error "printf wrap is net-negative on picolibc; codegen should set USE_FULL_PRINTF"
+#endif
+
 #include <cstdarg>
 #include <cstdio>
 
@@ -34,6 +40,9 @@
 
 namespace esphome::esp32 {}
 
+// NOLINTBEGIN(bugprone-reserved-identifier,cert-dcl37-c,cert-dcl51-cpp,readability-identifier-naming)
+extern "C" {
+
 static constexpr size_t PRINTF_BUFFER_SIZE = 512;
 
 // These stubs are essentially dead code at runtime — ESPHome replaces the
@@ -55,14 +64,16 @@ static int write_printf_buffer(FILE *stream, char *buf, int len) {
   return len;
 }
 
-// NOLINTBEGIN(bugprone-reserved-identifier,cert-dcl37-c,cert-dcl51-cpp,readability-identifier-naming)
-extern "C" {
-
 int __wrap_vprintf(const char *fmt, va_list ap) {
   char buf[PRINTF_BUFFER_SIZE];
   return write_printf_buffer(stdout, buf, vsnprintf(buf, sizeof(buf), fmt, ap));
 }
 
+int __wrap_vfprintf(FILE *stream, const char *fmt, va_list ap) {
+  char buf[PRINTF_BUFFER_SIZE];
+  return write_printf_buffer(stream, buf, vsnprintf(buf, sizeof(buf), fmt, ap));
+}
+
 int __wrap_printf(const char *fmt, ...) {
   va_list ap;
   va_start(ap, fmt);
@@ -71,11 +82,6 @@ int __wrap_printf(const char *fmt, ...) {
   return len;
 }
 
-int __wrap_vfprintf(FILE *stream, const char *fmt, va_list ap) {
-  char buf[PRINTF_BUFFER_SIZE];
-  return write_printf_buffer(stream, buf, vsnprintf(buf, sizeof(buf), fmt, ap));
-}
-
 int __wrap_fprintf(FILE *stream, const char *fmt, ...) {
   va_list ap;
   va_start(ap, fmt);

esphome/components/esp32_hosted/__init__.py

@@ -246,9 +246,10 @@ async def to_code(config):
     idf_ver = esp32.idf_version()
     os.environ["ESP_IDF_VERSION"] = f"{idf_ver.major}.{idf_ver.minor}"
     if idf_ver >= cv.Version(5, 5, 0):
-        esp32.add_idf_component(name="espressif/esp_wifi_remote", ref="1.4.0")
-        esp32.add_idf_component(name="espressif/eppp_link", ref="1.1.4")
-        esp32.add_idf_component(name="espressif/esp_hosted", ref="2.12.1")
+        esp32.add_idf_component(name="espressif/esp_wifi_remote", ref="1.5.1")
+        esp32.add_idf_component(name="espressif/wifi_remote_over_eppp", ref="0.3.2")
+        esp32.add_idf_component(name="espressif/eppp_link", ref="1.1.5")
+        esp32.add_idf_component(name="espressif/esp_hosted", ref="2.12.6")
     else:
         esp32.add_idf_component(name="espressif/esp_wifi_remote", ref="0.13.0")
         esp32.add_idf_component(name="espressif/eppp_link", ref="0.2.0")

esphome/components/esp8266/core.cpp

@@ -3,98 +3,12 @@
 #include "core.h"
 #include "esphome/core/defines.h"
 #include "esphome/core/hal.h"
-#include "esphome/core/time_64.h"
-#include "esphome/core/helpers.h"
-#include "preferences.h"
 #include <Arduino.h>
-#include <core_esp8266_features.h>
-
-extern "C" {
-#include <user_interface.h>
-}
 
 namespace esphome {
 
-// yield(), micros(), millis_64() inlined in hal.h.
-// Fast accumulator replacement for Arduino's millis() (~3.3 μs via 4× 64-bit
-// multiplies on the LX106). Tracks a running ms counter from 32-bit
-// system_get_time() deltas using pure 32-bit ops. Installed as __wrap_millis
-// (via -Wl,--wrap=millis) so Arduino libs and IRAM_ATTR ISR handlers (e.g.
-// Wiegand, ZyAura) also get the fast version. xt_rsil(15) guards the static
-// state against ISR re-entry; the critical section is bounded (≤10 while-loop
-// iterations, ~100 ns on the common path, or a constant-time /1000 ~2.5 μs on
-// the rare path — well under WiFi's ~10 μs ISR latency budget). NMIs (level
-// >15) are not masked, but the ESP8266 SDK's NMI handlers don't call millis().
-//
-// system_get_time() wraps every ~71.6 min; unsigned (now_us - last_us) handles
-// one wrap. The main loop calls millis() at 60+ Hz, so delta stays tiny — a
-// >71 min block would trip the watchdog long before it could matter here.
-static constexpr uint32_t MILLIS_RARE_PATH_THRESHOLD_US = 10000;
-static constexpr uint32_t US_PER_MS = 1000;
-
-uint32_t IRAM_ATTR HOT millis() {
-  // Struct packs the three statics so the compiler loads one base address
-  // instead of three separate literal pool entries (saves ~8 bytes IRAM).
-  static struct {
-    uint32_t cache;
-    uint32_t remainder;
-    uint32_t last_us;
-  } state = {0, 0, 0};
-  uint32_t ps = xt_rsil(15);
-  uint32_t now_us = system_get_time();
-  uint32_t delta = now_us - state.last_us;
-  state.last_us = now_us;
-  state.remainder += delta;
-  if (state.remainder >= MILLIS_RARE_PATH_THRESHOLD_US) {
-    // Rare path: large gap (WiFi scan, boot, long block). Constant-time
-    // conversion keeps the critical section bounded.
-    uint32_t ms = state.remainder / US_PER_MS;
-    state.cache += ms;
-    // Reuse ms instead of `remainder %= US_PER_MS` — `%` would compile to a
-    // second __umodsi3 call on the LX106 (no hardware divide).
-    state.remainder -= ms * US_PER_MS;
-  } else {
-    // Common path: small gap. At most ~10 iterations since remainder was
-    // < threshold (10 ms) on entry and delta adds at most one more threshold
-    // before exiting this branch.
-    while (state.remainder >= US_PER_MS) {
-      state.cache++;
-      state.remainder -= US_PER_MS;
-    }
-  }
-  uint32_t result = state.cache;
-  xt_wsr_ps(ps);
-  return result;
-}
-// Poll-based delay that avoids ::delay() — Arduino's __delay has an intra-object
-// call to the original millis() that --wrap can't intercept, so calling ::delay()
-// would keep the slow Arduino millis body alive in IRAM. optimistic_yield still
-// enters esp_schedule()/esp_suspend_within_cont() via yield(), so SDK tasks and
-// WiFi run correctly. Theoretically less power-efficient than Arduino's
-// os_timer-based delay() for long waits, but nearly all ESPHome delays are short
-// (sensor/I²C/SPI settling in the 1–100 ms range) where the difference is
-// negligible.
-void HOT delay(uint32_t ms) {
-  if (ms == 0) {
-    optimistic_yield(1000);
-    return;
-  }
-  uint32_t start = millis();
-  while (millis() - start < ms) {
-    optimistic_yield(1000);
-  }
-}
-// delayMicroseconds(), arch_feed_wdt(), and progmem_read_*() are inlined in hal/hal_esp8266.h.
-void arch_restart() {
-  system_restart();
-  // restart() doesn't always end execution
-  while (true) {  // NOLINT(clang-diagnostic-unreachable-code)
-    yield();
-  }
-}
-void arch_init() {}
-uint32_t IRAM_ATTR HOT arch_get_cpu_cycle_count() { return esp_get_cycle_count(); }
-uint32_t arch_get_cpu_freq_hz() { return F_CPU; }
+// HAL functions live in hal.cpp. This file keeps only the ESP8266-specific
+// firmware bootstrap (Tasmota OTA magic bytes, optional GPIO pre-init).
 
 void force_link_symbols() {
   // Tasmota uses magic bytes in the binary to check if an OTA firmware is compatible
@@ -131,12 +45,4 @@ extern "C" void resetPins() {  // NOLINT
 
 }  // namespace esphome
 
-// Linker wrap: redirect all ::millis() calls (Arduino libs, ISRs) to our accumulator.
-// Requires -Wl,--wrap=millis in build flags (added by __init__.py).
-// NOLINTNEXTLINE(bugprone-reserved-identifier,cert-dcl37-c,cert-dcl51-cpp,readability-identifier-naming)
-extern "C" uint32_t IRAM_ATTR __wrap_millis() { return esphome::millis(); }
-// Note: Arduino's init() registers a 60-second overflow timer for micros64().
-// We leave it running — wrapping init() as a no-op would break micros64()'s
-// overflow tracking, and the timer's cost is negligible (~3 μs per 60 s).
-
 #endif  // USE_ESP8266

esphome/components/esp8266/hal.cpp

@@ -0,0 +1,111 @@
+#ifdef USE_ESP8266
+
+#include "esphome/core/hal.h"
+#include "esphome/core/helpers.h"
+
+#include <Arduino.h>
+#include <core_esp8266_features.h>
+
+extern "C" {
+#include <user_interface.h>
+}
+
+// Empty esp8266 namespace block to satisfy ci-custom's lint_namespace check.
+// HAL functions live in namespace esphome (root) — they are not part of the
+// esp8266 component's API.
+namespace esphome::esp8266 {}  // namespace esphome::esp8266
+
+namespace esphome {
+
+// yield(), micros(), millis_64(), delayMicroseconds(), arch_feed_wdt(),
+// progmem_read_*() are inlined in components/esp8266/hal.h.
+//
+// Fast accumulator replacement for Arduino's millis() (~3.3 μs via 4× 64-bit
+// multiplies on the LX106). Tracks a running ms counter from 32-bit
+// system_get_time() deltas using pure 32-bit ops. Installed as __wrap_millis
+// (via -Wl,--wrap=millis) so Arduino libs and IRAM_ATTR ISR handlers (e.g.
+// Wiegand, ZyAura) also get the fast version. xt_rsil(15) guards the static
+// state against ISR re-entry; the critical section is bounded (≤10 while-loop
+// iterations, ~100 ns on the common path, or a constant-time /1000 ~2.5 μs on
+// the rare path — well under WiFi's ~10 μs ISR latency budget). NMIs (level
+// >15) are not masked, but the ESP8266 SDK's NMI handlers don't call millis().
+//
+// system_get_time() wraps every ~71.6 min; unsigned (now_us - last_us) handles
+// one wrap. The main loop calls millis() at 60+ Hz, so delta stays tiny — a
+// >71 min block would trip the watchdog long before it could matter here.
+static constexpr uint32_t MILLIS_RARE_PATH_THRESHOLD_US = 10000;
+static constexpr uint32_t US_PER_MS = 1000;
+
+uint32_t IRAM_ATTR HOT millis() {
+  // Struct packs the three statics so the compiler loads one base address
+  // instead of three separate literal pool entries (saves ~8 bytes IRAM).
+  static struct {
+    uint32_t cache;
+    uint32_t remainder;
+    uint32_t last_us;
+  } state = {0, 0, 0};
+  uint32_t ps = xt_rsil(15);
+  uint32_t now_us = system_get_time();
+  uint32_t delta = now_us - state.last_us;
+  state.last_us = now_us;
+  state.remainder += delta;
+  if (state.remainder >= MILLIS_RARE_PATH_THRESHOLD_US) {
+    // Rare path: large gap (WiFi scan, boot, long block). Constant-time
+    // conversion keeps the critical section bounded.
+    uint32_t ms = state.remainder / US_PER_MS;
+    state.cache += ms;
+    // Reuse ms instead of `remainder %= US_PER_MS` — `%` would compile to a
+    // second __umodsi3 call on the LX106 (no hardware divide).
+    state.remainder -= ms * US_PER_MS;
+  } else {
+    // Common path: small gap. At most ~10 iterations since remainder was
+    // < threshold (10 ms) on entry and delta adds at most one more threshold
+    // before exiting this branch.
+    while (state.remainder >= US_PER_MS) {
+      state.cache++;
+      state.remainder -= US_PER_MS;
+    }
+  }
+  uint32_t result = state.cache;
+  xt_wsr_ps(ps);
+  return result;
+}
+
+// Poll-based delay that avoids ::delay() — Arduino's __delay has an intra-object
+// call to the original millis() that --wrap can't intercept, so calling ::delay()
+// would keep the slow Arduino millis body alive in IRAM. optimistic_yield still
+// enters esp_schedule()/esp_suspend_within_cont() via yield(), so SDK tasks and
+// WiFi run correctly. Theoretically less power-efficient than Arduino's
+// os_timer-based delay() for long waits, but nearly all ESPHome delays are short
+// (sensor/I²C/SPI settling in the 1–100 ms range) where the difference is
+// negligible.
+void HOT delay(uint32_t ms) {
+  if (ms == 0) {
+    optimistic_yield(1000);
+    return;
+  }
+  uint32_t start = millis();
+  while (millis() - start < ms) {
+    optimistic_yield(1000);
+  }
+}
+
+void arch_restart() {
+  system_restart();
+  // restart() doesn't always end execution
+  while (true) {  // NOLINT(clang-diagnostic-unreachable-code)
+    yield();
+  }
+}
+
+}  // namespace esphome
+
+// Linker wrap: redirect all ::millis() calls (Arduino libs, ISRs) to our accumulator.
+// Requires -Wl,--wrap=millis in build flags (added by __init__.py).
+// NOLINTNEXTLINE(bugprone-reserved-identifier,cert-dcl37-c,cert-dcl51-cpp,readability-identifier-naming)
+extern "C" uint32_t IRAM_ATTR __wrap_millis() { return esphome::millis(); }
+// Note: Arduino's init() registers a 60-second overflow timer for micros64().
+// We leave it running — wrapping init() as a no-op would break micros64()'s
+// overflow tracking, and the timer's cost is negligible (~3 μs per 60 s).
+
+#endif  // USE_ESP8266

esphome/components/esp8266/hal.h

@@ -3,6 +3,7 @@
 #ifdef USE_ESP8266
 
 #include <c_types.h>
+#include <core_esp8266_features.h>
 #include <cstdint>
 #include <pgmspace.h>
 
@@ -24,6 +25,8 @@ extern "C" unsigned long millis(void);
 // NOLINTNEXTLINE(readability-redundant-declaration)
 extern "C" void system_soft_wdt_feed(void);
 
+namespace esphome::esp8266 {}
+
 namespace esphome {
 
 // Forward decl from helpers.h so this header stays cheap.
@@ -59,8 +62,11 @@ __attribute__((always_inline)) inline uint16_t progmem_read_uint16(const uint16_
 // NOLINTNEXTLINE(readability-identifier-naming)
 __attribute__((always_inline)) inline void delayMicroseconds(uint32_t us) { delay_microseconds_safe(us); }
 __attribute__((always_inline)) inline void arch_feed_wdt() { system_soft_wdt_feed(); }
-
-uint32_t arch_get_cpu_cycle_count();
+__attribute__((always_inline)) inline void arch_init() {}
+// esp_get_cycle_count() declared in <core_esp8266_features.h>; F_CPU is a
+// compiler-driven macro from the ESP8266 Arduino board defs (-DF_CPU=...).
+__attribute__((always_inline)) inline uint32_t arch_get_cpu_cycle_count() { return esp_get_cycle_count(); }
+__attribute__((always_inline)) inline uint32_t arch_get_cpu_freq_hz() { return F_CPU; }
 
 }  // namespace esphome
 

esphome/components/esphome/ota/ota_esphome.cpp

@@ -114,8 +114,10 @@ void ESPHomeOTAComponent::loop() {
   this->handle_handshake_();
 }
 
-static const uint8_t FEATURE_SUPPORTS_COMPRESSION = 0x01;
-static const uint8_t FEATURE_SUPPORTS_SHA256_AUTH = 0x02;
+static constexpr uint8_t CLIENT_FEATURE_SUPPORTS_COMPRESSION = 0x01;
+static constexpr uint8_t CLIENT_FEATURE_SUPPORTS_SHA256_AUTH = 0x02;
+static constexpr uint8_t CLIENT_FEATURE_SUPPORTS_EXTENDED_PROTOCOL = 0x04;
+static constexpr uint8_t SERVER_FEATURE_SUPPORTS_COMPRESSION = 0x01;
 
 void ESPHomeOTAComponent::handle_handshake_() {
   /// Handle the OTA handshake and authentication.
@@ -201,16 +203,30 @@ void ESPHomeOTAComponent::handle_handshake_() {
       this->ota_features_ = this->handshake_buf_[0];
       ESP_LOGV(TAG, "Features: 0x%02X", this->ota_features_);
       this->transition_ota_state_(OTAState::FEATURE_ACK);
-      this->handshake_buf_[0] =
-          ((this->ota_features_ & FEATURE_SUPPORTS_COMPRESSION) != 0 && this->backend_->supports_compression())
-              ? ota::OTA_RESPONSE_SUPPORTS_COMPRESSION
-              : ota::OTA_RESPONSE_HEADER_OK;
+
+      const bool supports_compression =
+          (this->ota_features_ & CLIENT_FEATURE_SUPPORTS_COMPRESSION) != 0 && this->backend_->supports_compression();
+
+      // Compose the feature-ack response. When the client negotiates the extended protocol we emit
+      // a 2-byte response (marker + server feature flags); otherwise we emit the single-byte
+      // legacy response.
+      this->extended_proto_ = (this->ota_features_ & CLIENT_FEATURE_SUPPORTS_EXTENDED_PROTOCOL) != 0;
+      if (this->extended_proto_) {
+        static_assert(HANDSHAKE_BUF_SIZE >= 2, "handshake_buf_ must hold the 2-byte extended-protocol feature ack");
+        this->handshake_buf_[0] = ota::OTA_RESPONSE_FEATURE_FLAGS;
+        this->handshake_buf_[1] = (supports_compression ? SERVER_FEATURE_SUPPORTS_COMPRESSION : 0);
+      } else {
+        this->handshake_buf_[0] =
+            supports_compression ? ota::OTA_RESPONSE_SUPPORTS_COMPRESSION : ota::OTA_RESPONSE_HEADER_OK;
+      }
       [[fallthrough]];
     }
 
     case OTAState::FEATURE_ACK: {
-      // Acknowledge header - 1 byte
-      if (!this->try_write_(1, LOG_STR("ack feature"))) {
+      static constexpr size_t STANDARD_PROTO_ACK_SIZE = 1;
+      static constexpr size_t EXTENDED_PROTO_ACK_SIZE = 2;
+      const size_t ack_size = this->extended_proto_ ? EXTENDED_PROTO_ACK_SIZE : STANDARD_PROTO_ACK_SIZE;
+      if (!this->try_write_(ack_size, LOG_STR("ack feature"))) {
         return;
       }
 #ifdef USE_OTA_PASSWORD
@@ -296,6 +312,7 @@ void ESPHomeOTAComponent::handle_data_() {
   uint8_t buf[OTA_BUFFER_SIZE];
   char *sbuf = reinterpret_cast<char *>(buf);
   size_t ota_size;
+  ota::OTAType ota_type = ota::OTA_TYPE_UPDATE_APP;
 #if USE_OTA_VERSION == 2
   size_t size_acknowledged = 0;
 #endif
@@ -311,6 +328,16 @@ void ESPHomeOTAComponent::handle_data_() {
   // Acknowledge auth OK - 1 byte
   this->write_byte_(ota::OTA_RESPONSE_AUTH_OK);
 
+  if (this->extended_proto_) {
+    // Read ota type, 1 byte
+    if (!this->readall_(buf, 1)) {
+      this->log_read_error_(LOG_STR("OTA type"));
+      goto error;  // NOLINT(cppcoreguidelines-avoid-goto)
+    }
+    ota_type = static_cast<ota::OTAType>(buf[0]);
+  }
+  ESP_LOGV(TAG, "OTA type is 0x%02x", ota_type);
+
   // Read size, 4 bytes MSB first
   if (!this->readall_(buf, 4)) {
     this->log_read_error_(LOG_STR("size"));
@@ -320,6 +347,11 @@ void ESPHomeOTAComponent::handle_data_() {
              (static_cast<size_t>(buf[2]) << 8) | buf[3];
   ESP_LOGV(TAG, "Size is %u bytes", ota_size);
 
+  if (ota_type != ota::OTA_TYPE_UPDATE_APP) {
+    error_code = ota::OTA_RESPONSE_ERROR_UNSUPPORTED_OTA_TYPE;
+    goto error;  // NOLINT(cppcoreguidelines-avoid-goto)
+  }
+
   // Now that we've passed authentication and are actually
   // starting the update, set the warning status and notify
   // listeners. This ensures that port scanners do not
@@ -616,7 +648,7 @@ void ESPHomeOTAComponent::yield_and_feed_watchdog_() {
 void ESPHomeOTAComponent::log_auth_warning_(const LogString *msg) { ESP_LOGW(TAG, "Auth: %s", LOG_STR_ARG(msg)); }
 
 bool ESPHomeOTAComponent::select_auth_type_() {
-  bool client_supports_sha256 = (this->ota_features_ & FEATURE_SUPPORTS_SHA256_AUTH) != 0;
+  bool client_supports_sha256 = (this->ota_features_ & CLIENT_FEATURE_SUPPORTS_SHA256_AUTH) != 0;
 
   // Require SHA256
   if (!client_supports_sha256) {

esphome/components/esphome/ota/ota_esphome.h

@@ -97,8 +97,9 @@ class ESPHomeOTAComponent final : public ota::OTAComponent {
   ota::OTABackendPtr backend_;
 
   uint32_t client_connect_time_{0};
+  static constexpr size_t HANDSHAKE_BUF_SIZE = 5;
   uint16_t port_;
-  uint8_t handshake_buf_[5];
+  uint8_t handshake_buf_[HANDSHAKE_BUF_SIZE];
   OTAState ota_state_{OTAState::IDLE};
   uint8_t handshake_buf_pos_{0};
   uint8_t ota_features_{0};
@@ -106,6 +107,7 @@ class ESPHomeOTAComponent final : public ota::OTAComponent {
   uint8_t auth_buf_pos_{0};
   uint8_t auth_type_{0};  // Store auth type to know which hasher to use
 #endif                    // USE_OTA_PASSWORD
+  bool extended_proto_{false};
 };
 
 }  // namespace esphome

esphome/components/host/core.cpp

@@ -1,74 +1,16 @@
 #ifdef USE_HOST
 
 #include "esphome/core/application.h"
-#include "esphome/core/hal.h"
-#include "esphome/core/helpers.h"
 #include "preferences.h"
 
 #include <csignal>
-#include <sched.h>
-#include <time.h>
-#include <cstdlib>
 
 namespace {
 volatile sig_atomic_t s_signal_received = 0;  // NOLINT(cppcoreguidelines-avoid-non-const-global-variables)
 void signal_handler(int signal) { s_signal_received = signal; }
 }  // namespace
 
-namespace esphome {
-
-void HOT yield() { ::sched_yield(); }
-uint32_t IRAM_ATTR HOT millis() {
-  struct timespec spec;
-  clock_gettime(CLOCK_MONOTONIC, &spec);
-  return static_cast<uint32_t>(spec.tv_sec * 1000ULL + spec.tv_nsec / 1000000);
-}
-uint64_t millis_64() {
-  struct timespec spec;
-  clock_gettime(CLOCK_MONOTONIC, &spec);
-  return static_cast<uint64_t>(spec.tv_sec) * 1000ULL + static_cast<uint64_t>(spec.tv_nsec) / 1000000ULL;
-}
-void HOT delay(uint32_t ms) {
-  struct timespec ts;
-  ts.tv_sec = ms / 1000;
-  ts.tv_nsec = (ms % 1000) * 1000000;
-  int res;
-  do {
-    res = nanosleep(&ts, &ts);
-  } while (res != 0 && errno == EINTR);
-}
-uint32_t IRAM_ATTR HOT micros() {
-  struct timespec spec;
-  clock_gettime(CLOCK_MONOTONIC, &spec);
-  return static_cast<uint32_t>(spec.tv_sec * 1000000ULL + spec.tv_nsec / 1000);
-}
-void IRAM_ATTR HOT delayMicroseconds(uint32_t us) {
-  struct timespec ts;
-  ts.tv_sec = us / 1000000U;
-  ts.tv_nsec = (us % 1000000U) * 1000U;
-  int res;
-  do {
-    res = nanosleep(&ts, &ts);
-  } while (res != 0 && errno == EINTR);
-}
-void arch_restart() { exit(0); }
-void arch_init() {
-  // pass
-}
-void HOT arch_feed_wdt() {
-  // pass
-}
-
-uint32_t arch_get_cpu_cycle_count() {
-  struct timespec spec;
-  clock_gettime(CLOCK_MONOTONIC, &spec);
-  time_t seconds = spec.tv_sec;
-  uint32_t us = spec.tv_nsec;
-  return ((uint32_t) seconds) * 1000000000U + us;
-}
-uint32_t arch_get_cpu_freq_hz() { return 1000000000U; }
-
-}  // namespace esphome
+// HAL functions live in hal.cpp.
 
 void setup();
 void loop();

esphome/components/host/hal.cpp

@@ -0,0 +1,65 @@
+#ifdef USE_HOST
+
+#include "esphome/core/hal.h"
+#include "esphome/core/helpers.h"
+
+#include <time.h>
+#include <cerrno>
+#include <cstdlib>
+
+// Empty host namespace block to satisfy ci-custom's lint_namespace check.
+// HAL functions live in namespace esphome (root) — they are not part of the
+// host component's API.
+namespace esphome::host {}  // namespace esphome::host
+
+namespace esphome {
+
+// yield(), arch_init(), arch_feed_wdt(), arch_get_cpu_freq_hz() inlined in
+// components/host/hal.h.
+
+uint32_t IRAM_ATTR HOT millis() {
+  struct timespec spec;
+  clock_gettime(CLOCK_MONOTONIC, &spec);
+  return static_cast<uint32_t>(spec.tv_sec * 1000ULL + spec.tv_nsec / 1000000);
+}
+uint64_t millis_64() {
+  struct timespec spec;
+  clock_gettime(CLOCK_MONOTONIC, &spec);
+  return static_cast<uint64_t>(spec.tv_sec) * 1000ULL + static_cast<uint64_t>(spec.tv_nsec) / 1000000ULL;
+}
+void HOT delay(uint32_t ms) {
+  struct timespec ts;
+  ts.tv_sec = ms / 1000;
+  ts.tv_nsec = (ms % 1000) * 1000000;
+  int res;
+  do {
+    res = nanosleep(&ts, &ts);
+  } while (res != 0 && errno == EINTR);
+}
+uint32_t IRAM_ATTR HOT micros() {
+  struct timespec spec;
+  clock_gettime(CLOCK_MONOTONIC, &spec);
+  return static_cast<uint32_t>(spec.tv_sec * 1000000ULL + spec.tv_nsec / 1000);
+}
+void IRAM_ATTR HOT delayMicroseconds(uint32_t us) {
+  struct timespec ts;
+  ts.tv_sec = us / 1000000U;
+  ts.tv_nsec = (us % 1000000U) * 1000U;
+  int res;
+  do {
+    res = nanosleep(&ts, &ts);
+  } while (res != 0 && errno == EINTR);
+}
+void arch_restart() { exit(0); }
+
+uint32_t arch_get_cpu_cycle_count() {
+  struct timespec spec;
+  clock_gettime(CLOCK_MONOTONIC, &spec);
+  time_t seconds = spec.tv_sec;
+  uint32_t ns = static_cast<uint32_t>(spec.tv_nsec);
+  return static_cast<uint32_t>(seconds) * 1000000000U + ns;
+}
+
+}  // namespace esphome
+
+#endif  // USE_HOST

esphome/components/host/hal.h

@@ -3,26 +3,32 @@
 #ifdef USE_HOST
 
 #include <cstdint>
+#include <sched.h>
 
 #define IRAM_ATTR
 #define PROGMEM
 
+namespace esphome::host {}
+
 namespace esphome {
 
 /// Returns true when executing inside an interrupt handler.
 /// Host has no ISR concept.
 __attribute__((always_inline)) inline bool in_isr_context() { return false; }
 
-void yield();
+__attribute__((always_inline)) inline void yield() { ::sched_yield(); }
+
 void delay(uint32_t ms);
 uint32_t micros();
 uint32_t millis();
 uint64_t millis_64();
-
 void delayMicroseconds(uint32_t us);  // NOLINT(readability-identifier-naming)
-void arch_feed_wdt();
 uint32_t arch_get_cpu_cycle_count();
 
+__attribute__((always_inline)) inline void arch_init() {}
+__attribute__((always_inline)) inline void arch_feed_wdt() {}
+__attribute__((always_inline)) inline uint32_t arch_get_cpu_freq_hz() { return 1000000000U; }
+
 }  // namespace esphome
 
 #endif  // USE_HOST

esphome/components/libretiny/__init__.py

@@ -37,6 +37,7 @@ from .const import (
     CONF_UART_PORT,
     FAMILIES,
     FAMILY_BK7231N,
+    FAMILY_BK7238,
     FAMILY_COMPONENT,
     FAMILY_FRIENDLY,
     FAMILY_RTL8710B,
@@ -56,19 +57,22 @@ CODEOWNERS = ["@kuba2k2"]
 AUTO_LOAD = ["preferences"]
 IS_TARGET_PLATFORM = True
 
-# BK7231N SDK options to disable unused features.
+# BLE 5.x BK SDK options to disable unused features.
 # Disabling BLE saves ~21KB RAM and ~200KB Flash because BLE init code is
 # called unconditionally by the SDK. ESPHome doesn't use BLE on LibreTiny.
 #
-# This only works on BK7231N (BLE 5.x). Other BK72XX chips using BLE 4.2
-# (BK7231T, BK7231Q, BK7251; BK7252 boards use the BK7251 family) have a bug
-# where the BLE library still links and references undefined symbols when
-# CFG_SUPPORT_BLE=0.
+# This only works on BLE 5.x BK chips (BK7231N, BK7238). Other BK72XX chips
+# using BLE 4.2 (BK7231T, BK7231Q, BK7251; BK7252 boards use the BK7251 family)
+# have a bug where the BLE library still links and references undefined symbols
+# when CFG_SUPPORT_BLE=0.
+#
+# On BK7238 the SDK also hangs at WiFi STA enable when BLE init runs, so
+# disabling it is required for reliable boot, not just an optimization.
 #
 # Other options like CFG_TX_EVM_TEST, CFG_RX_SENSITIVITY_TEST, CFG_SUPPORT_BKREG,
 # CFG_SUPPORT_OTA_HTTP, and CFG_USE_SPI_SLAVE were evaluated but provide no  # NOLINT
 # measurable benefit - the linker already strips unreferenced code via -gc-sections.
-_BK7231N_SYS_CONFIG_OPTIONS = [
+_BLE5_BK_SYS_CONFIG_OPTIONS = [
     "CFG_SUPPORT_BLE=0",
 ]
 
@@ -549,9 +553,9 @@ async def component_to_code(config):
         cg.add_platformio_option("custom_fw_version", __version__)
 
     # Apply chip-specific SDK options to save RAM/Flash
-    if config[CONF_FAMILY] == FAMILY_BK7231N:
+    if config[CONF_FAMILY] in (FAMILY_BK7231N, FAMILY_BK7238):
         cg.add_platformio_option(
-            "custom_options.sys_config#h", _BK7231N_SYS_CONFIG_OPTIONS
+            "custom_options.sys_config#h", _BLE5_BK_SYS_CONFIG_OPTIONS
         )
 
     # Tune lwIP for ESPHome's actual needs.

esphome/components/libretiny/core.cpp

@@ -1,55 +1,6 @@
 #ifdef USE_LIBRETINY
 
-#include "core.h"
-#include "esphome/core/defines.h"
-#include "esphome/core/hal.h"
-#include "esphome/core/helpers.h"
-#include "preferences.h"
-
-#include <FreeRTOS.h>
-#include <task.h>
-
-void setup();
-void loop();
-
-namespace esphome {
-
-// yield(), delay(), micros(), millis(), millis_64() inlined in hal.h.
-void IRAM_ATTR HOT delayMicroseconds(uint32_t us) { ::delayMicroseconds(us); }
-
-void arch_init() {
-  libretiny::setup_preferences();
-  lt_wdt_enable(10000L);
-#ifdef USE_BK72XX
-  // BK72xx SDK creates the main Arduino task at priority 3, which is lower than
-  // all WiFi (4-5), LwIP (4), and TCP/IP (7) tasks. This causes ~100ms loop
-  // stalls whenever WiFi background processing runs, because the main task
-  // cannot resume until every higher-priority task finishes.
-  //
-  // By contrast, RTL87xx creates the main task at osPriorityRealtime (highest).
-  //
-  // Raise to priority 6: above WiFi/LwIP tasks (4-5) so they don't preempt the
-  // main loop, but below the TCP/IP thread (7) so packet processing keeps priority.
-  // This is safe because ESPHome yields voluntarily via wakeable_delay() and
-  // the Arduino mainTask yield() after each loop() iteration.
-  static constexpr UBaseType_t MAIN_TASK_PRIORITY = 6;
-  static_assert(MAIN_TASK_PRIORITY < configMAX_PRIORITIES, "MAIN_TASK_PRIORITY must be less than configMAX_PRIORITIES");
-  vTaskPrioritySet(nullptr, MAIN_TASK_PRIORITY);
-#endif
-#if LT_GPIO_RECOVER
-  lt_gpio_recover();
-#endif
-}
-
-void arch_restart() {
-  lt_reboot();
-  while (1) {
-  }
-}
-void HOT arch_feed_wdt() { lt_wdt_feed(); }
-uint32_t arch_get_cpu_cycle_count() { return lt_cpu_get_cycle_count(); }
-uint32_t arch_get_cpu_freq_hz() { return lt_cpu_get_freq(); }
-
-}  // namespace esphome
+// HAL functions live in hal.cpp. core.cpp is intentionally empty for
+// libretiny — there is no extra component bootstrap to keep here.
 
 #endif  // USE_LIBRETINY

esphome/components/libretiny/hal.cpp

@@ -0,0 +1,53 @@
+#ifdef USE_LIBRETINY
+
+#include "core.h"
+#include "esphome/core/hal.h"
+#include "preferences.h"
+
+#include <FreeRTOS.h>
+#include <task.h>
+
+// Empty libretiny namespace block to satisfy ci-custom's lint_namespace check.
+// HAL functions live in namespace esphome (root) — they are not part of the
+// libretiny component's API.
+namespace esphome::libretiny {}  // namespace esphome::libretiny
+
+namespace esphome {
+
+// yield(), delay(), micros(), millis(), millis_64(), delayMicroseconds(),
+// arch_feed_wdt(), arch_get_cpu_cycle_count(), arch_get_cpu_freq_hz()
+// inlined in components/libretiny/hal.h.
+
+void arch_init() {
+  libretiny::setup_preferences();
+  lt_wdt_enable(10000L);
+#ifdef USE_BK72XX
+  // BK72xx SDK creates the main Arduino task at priority 3, which is lower than
+  // all WiFi (4-5), LwIP (4), and TCP/IP (7) tasks. This causes ~100ms loop
+  // stalls whenever WiFi background processing runs, because the main task
+  // cannot resume until every higher-priority task finishes.
+  //
+  // By contrast, RTL87xx creates the main task at osPriorityRealtime (highest).
+  //
+  // Raise to priority 6: above WiFi/LwIP tasks (4-5) so they don't preempt the
+  // main loop, but below the TCP/IP thread (7) so packet processing keeps priority.
+  // This is safe because ESPHome yields voluntarily via wakeable_delay() and
+  // the Arduino mainTask yield() after each loop() iteration.
+  static constexpr UBaseType_t MAIN_TASK_PRIORITY = 6;
+  static_assert(MAIN_TASK_PRIORITY < configMAX_PRIORITIES, "MAIN_TASK_PRIORITY must be less than configMAX_PRIORITIES");
+  vTaskPrioritySet(nullptr, MAIN_TASK_PRIORITY);
+#endif
+#if LT_GPIO_RECOVER
+  lt_gpio_recover();
+#endif
+}
+
+void arch_restart() {
+  lt_reboot();
+  while (1) {
+  }
+}
+
+}  // namespace esphome
+
+#endif  // USE_LIBRETINY

esphome/components/libretiny/hal.h

@@ -51,8 +51,18 @@ extern "C" void yield(void);
 extern "C" void delay(unsigned long ms);
 extern "C" unsigned long micros(void);
 extern "C" unsigned long millis(void);
+extern "C" void delayMicroseconds(unsigned int us);
 // NOLINTEND(google-runtime-int,readability-identifier-naming,readability-redundant-declaration)
 
+// Forward decls from libretiny's <lt_api.h> family for the inline arch_*
+// wrappers below. Pulling the full header would drag in the rest of the
+// LibreTiny C API.
+extern "C" void lt_wdt_feed(void);
+extern "C" uint32_t lt_cpu_get_cycle_count(void);
+extern "C" uint32_t lt_cpu_get_freq(void);
+
+namespace esphome::libretiny {}
+
 namespace esphome {
 
 /// Returns true when executing inside an interrupt handler.
@@ -88,9 +98,13 @@ __attribute__((always_inline)) inline uint32_t millis() { return static_cast<uin
 #endif
 __attribute__((always_inline)) inline uint64_t millis_64() { return Millis64Impl::compute(millis()); }
 
-void delayMicroseconds(uint32_t us);  // NOLINT(readability-identifier-naming)
-void arch_feed_wdt();
-uint32_t arch_get_cpu_cycle_count();
+// NOLINTNEXTLINE(readability-identifier-naming)
+__attribute__((always_inline)) inline void delayMicroseconds(uint32_t us) { ::delayMicroseconds(us); }
+__attribute__((hot, always_inline)) inline void arch_feed_wdt() { lt_wdt_feed(); }
+__attribute__((always_inline)) inline uint32_t arch_get_cpu_cycle_count() { return lt_cpu_get_cycle_count(); }
+__attribute__((always_inline)) inline uint32_t arch_get_cpu_freq_hz() { return lt_cpu_get_freq(); }
+
+void arch_init();
 
 }  // namespace esphome
 

esphome/components/lvgl/lvgl_esphome.cpp

@@ -454,10 +454,12 @@ void LVTouchListener::update(const touchscreen::TouchPoints_t &tpoints) {
 
 #ifdef USE_LVGL_METER
 
-int16_t lv_get_needle_angle_for_value(lv_obj_t *obj, int value) {
+int16_t lv_get_needle_angle_for_value(lv_obj_t *obj, int32_t value) {
   auto *scale = lv_obj_get_parent(obj);
   auto min_value = lv_scale_get_range_min_value(scale);
-  return ((value - min_value) * lv_scale_get_angle_range(scale) / (lv_scale_get_range_max_value(scale) - min_value) +
+  auto max_value = lv_scale_get_range_max_value(scale);
+  value = clamp(value, min_value, max_value);
+  return ((value - min_value) * lv_scale_get_angle_range(scale) / (max_value - min_value) +
           lv_scale_get_rotation((scale))) %
          360;
 }

esphome/components/lvgl/lvgl_esphome.h

@@ -112,7 +112,7 @@ inline void lv_animimg_set_src(lv_obj_t *img, std::vector<image::Image *> images
 #endif  // USE_LVGL_ANIMIMG
 
 #ifdef USE_LVGL_METER
-int16_t lv_get_needle_angle_for_value(lv_obj_t *obj, int value);
+int16_t lv_get_needle_angle_for_value(lv_obj_t *obj, int32_t value);
 #endif
 
 #ifdef USE_LVGL_GRADIENT

esphome/components/mdns/mdns_component.cpp

@@ -39,7 +39,39 @@ MDNS_STATIC_CONST_CHAR(SERVICE_TCP, "_tcp");
 // Wrap build-time defines into flash storage
 MDNS_STATIC_CONST_CHAR(VALUE_VERSION, ESPHOME_VERSION);
 
-void MDNSComponent::compile_records_(StaticVector<MDNSService, MDNS_SERVICE_COUNT> &services, char *mac_address_buf) {
+void MDNSComponent::setup_buffers_and_register_(PlatformRegisterFn platform_register) {
+#ifdef USE_MDNS_STORE_SERVICES
+  auto &services = this->services_;
+#else
+  StaticVector<MDNSService, MDNS_SERVICE_COUNT> services_storage;
+  auto &services = services_storage;
+#endif
+
+#ifdef USE_API
+#ifdef USE_MDNS_STORE_SERVICES
+  get_mac_address_into_buffer(this->mac_address_);
+  char *mac_ptr = this->mac_address_;
+  format_hex_to(this->config_hash_str_, App.get_config_hash());
+  char *cfg_ptr = this->config_hash_str_;
+#else
+  char mac_address[MAC_ADDRESS_BUFFER_SIZE];
+  char config_hash_str[CONFIG_HASH_STR_SIZE];
+  get_mac_address_into_buffer(mac_address);
+  format_hex_to(config_hash_str, App.get_config_hash());
+  char *mac_ptr = mac_address;
+  char *cfg_ptr = config_hash_str;
+#endif
+#else
+  char *mac_ptr = nullptr;
+  char *cfg_ptr = nullptr;
+#endif
+
+  this->compile_records_(services, mac_ptr, cfg_ptr);
+  platform_register(this, services);
+}
+
+void MDNSComponent::compile_records_(StaticVector<MDNSService, MDNS_SERVICE_COUNT> &services, char *mac_address_buf,
+                                     char *config_hash_buf) {
   // IMPORTANT: The #ifdef blocks below must match COMPONENTS_WITH_MDNS_SERVICES
   // in mdns/__init__.py. If you add a new service here, update both locations.
 
@@ -47,6 +79,7 @@ void MDNSComponent::compile_records_(StaticVector<MDNSService, MDNS_SERVICE_COUN
   MDNS_STATIC_CONST_CHAR(SERVICE_ESPHOMELIB, "_esphomelib");
   MDNS_STATIC_CONST_CHAR(TXT_FRIENDLY_NAME, "friendly_name");
   MDNS_STATIC_CONST_CHAR(TXT_VERSION, "version");
+  MDNS_STATIC_CONST_CHAR(TXT_CONFIG_HASH, "config_hash");
   MDNS_STATIC_CONST_CHAR(TXT_MAC, "mac");
   MDNS_STATIC_CONST_CHAR(TXT_PLATFORM, "platform");
   MDNS_STATIC_CONST_CHAR(TXT_BOARD, "board");
@@ -63,7 +96,7 @@ void MDNSComponent::compile_records_(StaticVector<MDNSService, MDNS_SERVICE_COUN
     bool friendly_name_empty = friendly_name.empty();
 
     // Calculate exact capacity for txt_records
-    size_t txt_count = 3;  // version, mac, board (always present)
+    size_t txt_count = 4;  // version, config_hash, mac, board (always present)
     if (!friendly_name_empty) {
       txt_count++;  // friendly_name
     }
@@ -91,6 +124,9 @@ void MDNSComponent::compile_records_(StaticVector<MDNSService, MDNS_SERVICE_COUN
     }
     txt_records.push_back({MDNS_STR(TXT_VERSION), MDNS_STR(VALUE_VERSION)});
 
+    // Config hash: passed from caller (either member buffer or stack buffer depending on USE_MDNS_STORE_SERVICES)
+    txt_records.push_back({MDNS_STR(TXT_CONFIG_HASH), MDNS_STR(config_hash_buf)});
+
     // MAC address: passed from caller (either member buffer or stack buffer depending on USE_MDNS_STORE_SERVICES)
     txt_records.push_back({MDNS_STR(TXT_MAC), MDNS_STR(mac_address_buf)});
 

esphome/components/mdns/mdns_component.h

@@ -70,6 +70,9 @@ class MDNSComponent final : public Component
   void setup() override;
   void dump_config() override;
 
+  /// Size of buffer required for config hash hex string (8 hex chars + null terminator)
+  static constexpr size_t CONFIG_HASH_STR_SIZE = format_hex_size(sizeof(uint32_t));
+
 #ifdef USE_MDNS_EVENT_DRIVEN_POLLING
   // LEAmDNS has meaningful work only during the probe+announce phase (3×250ms probes +
   // 8×1000ms announces, ~9s). Afterwards every internal timer is resetToNeverExpires()
@@ -124,30 +127,7 @@ class MDNSComponent final : public Component
   /// Helper to set up services and MAC buffers, then call platform-specific registration
   using PlatformRegisterFn = void (*)(MDNSComponent *, StaticVector<MDNSService, MDNS_SERVICE_COUNT> &);
 
-  void setup_buffers_and_register_(PlatformRegisterFn platform_register) {
-#ifdef USE_MDNS_STORE_SERVICES
-    auto &services = this->services_;
-#else
-    StaticVector<MDNSService, MDNS_SERVICE_COUNT> services_storage;
-    auto &services = services_storage;
-#endif
-
-#ifdef USE_API
-#ifdef USE_MDNS_STORE_SERVICES
-    get_mac_address_into_buffer(this->mac_address_);
-    char *mac_ptr = this->mac_address_;
-#else
-    char mac_address[MAC_ADDRESS_BUFFER_SIZE];
-    get_mac_address_into_buffer(mac_address);
-    char *mac_ptr = mac_address;
-#endif
-#else
-    char *mac_ptr = nullptr;
-#endif
-
-    this->compile_records_(services, mac_ptr);
-    platform_register(this, services);
-  }
+  void setup_buffers_and_register_(PlatformRegisterFn platform_register);
 
 #ifdef USE_MDNS_DYNAMIC_TXT
   /// Storage for runtime-generated TXT values from user lambdas
@@ -159,6 +139,8 @@ class MDNSComponent final : public Component
 #if defined(USE_API) && defined(USE_MDNS_STORE_SERVICES)
   /// Fixed buffer for MAC address (only needed when services are stored)
   char mac_address_[MAC_ADDRESS_BUFFER_SIZE];
+  /// Fixed buffer for config hash hex string (only needed when services are stored)
+  char config_hash_str_[CONFIG_HASH_STR_SIZE];
 #endif
 #ifdef USE_MDNS_STORE_SERVICES
   StaticVector<MDNSService, MDNS_SERVICE_COUNT> services_{};
@@ -167,7 +149,8 @@ class MDNSComponent final : public Component
   // RP2040 defers MDNS.begin() until the first IP-up event; this tracks that.
   bool initialized_{false};
 #endif
-  void compile_records_(StaticVector<MDNSService, MDNS_SERVICE_COUNT> &services, char *mac_address_buf);
+  void compile_records_(StaticVector<MDNSService, MDNS_SERVICE_COUNT> &services, char *mac_address_buf,
+                        char *config_hash_buf);
 };
 
 }  // namespace esphome::mdns

esphome/components/mdns/mdns_host.cpp

@@ -3,6 +3,8 @@
 
 #include "esphome/components/network/ip_address.h"
 #include "esphome/components/network/util.h"
+#include "esphome/core/application.h"
+#include "esphome/core/helpers.h"
 #include "esphome/core/log.h"
 #include "mdns_component.h"
 
@@ -13,10 +15,13 @@ void MDNSComponent::setup() {
 #ifdef USE_API
   get_mac_address_into_buffer(this->mac_address_);
   char *mac_ptr = this->mac_address_;
+  format_hex_to(this->config_hash_str_, App.get_config_hash());
+  char *cfg_ptr = this->config_hash_str_;
 #else
   char *mac_ptr = nullptr;
+  char *cfg_ptr = nullptr;
 #endif
-  this->compile_records_(this->services_, mac_ptr);
+  this->compile_records_(this->services_, mac_ptr, cfg_ptr);
 #endif
   // Host platform doesn't have actual mDNS implementation
 }

esphome/components/nextion/nextion_upload_arduino.cpp

@@ -88,6 +88,33 @@ int Nextion::upload_by_chunks_(HTTPClient &http_client, uint32_t &range_start) {
       this->write_array(buffer, buffer_size);
       App.feed_wdt();
       this->recv_ret_string_(recv_string, NEXTION_UPLOAD_ACK_TIMEOUT_MS, true);
+
+      // Some Nextion firmware variants (notably bootloader/recovery mode on panels
+      // with no installed TFT) emit the 5-byte 0x08+position fast-mode ack with a
+      // multi-second gap between the leading 0x08 byte and the 4 trailing position
+      // bytes. recv_ret_string_ returns after the first byte; manually drain the
+      // trailing bytes from the UART before continuing.
+      if (!recv_string.empty() && recv_string[0] == 0x08 && recv_string.size() < 5) {
+        const uint32_t deadline = millis() + NEXTION_UPLOAD_ACK_TIMEOUT_MS;
+        while (recv_string.size() < 5 && millis() < deadline) {
+          if (this->available()) {
+            uint8_t b = 0;
+            if (this->read_byte(&b)) {
+              recv_string.push_back(static_cast<char>(b));
+            }
+          } else {
+            delay(5);  // NOLINT
+            App.feed_wdt();
+          }
+        }
+        if (recv_string.size() < 5) {
+          ESP_LOGE(TAG, "Truncated 0x08 response: got %zu bytes within %" PRIu32 "ms", recv_string.size(),
+                   NEXTION_UPLOAD_ACK_TIMEOUT_MS);
+          allocator.deallocate(buffer, 4096);
+          buffer = nullptr;
+          return -1;
+        }
+      }
       this->content_length_ -= read_len;
       const float upload_percentage = 100.0f * (this->tft_size_ - this->content_length_) / this->tft_size_;
       ESP_LOGD(TAG, "Upload: %0.2f%% (%" PRIu32 " left, heap: %" PRIu32 ")", upload_percentage, this->content_length_,

esphome/components/nextion/nextion_upload_esp32.cpp

@@ -104,6 +104,33 @@ int Nextion::upload_by_chunks_(esp_http_client_handle_t http_client, uint32_t &r
       this->write_array(buffer, buffer_size);
       App.feed_wdt();
       this->recv_ret_string_(recv_string, NEXTION_UPLOAD_ACK_TIMEOUT_MS, true);
+
+      // Some Nextion firmware variants (notably bootloader/recovery mode on panels
+      // with no installed TFT) emit the 5-byte 0x08+position fast-mode ack with a
+      // multi-second gap between the leading 0x08 byte and the 4 trailing position
+      // bytes. recv_ret_string_ returns after the first byte; manually drain the
+      // trailing bytes from the UART before continuing.
+      if (!recv_string.empty() && recv_string[0] == 0x08 && recv_string.size() < 5) {
+        const uint32_t deadline = millis() + NEXTION_UPLOAD_ACK_TIMEOUT_MS;
+        while (recv_string.size() < 5 && millis() < deadline) {
+          if (this->available()) {
+            uint8_t b = 0;
+            if (this->read_byte(&b)) {
+              recv_string.push_back(static_cast<char>(b));
+            }
+          } else {
+            vTaskDelay(pdMS_TO_TICKS(5));  // NOLINT
+            App.feed_wdt();
+          }
+        }
+        if (recv_string.size() < 5) {
+          ESP_LOGE(TAG, "Truncated 0x08 response: got %zu bytes within %" PRIu32 "ms", recv_string.size(),
+                   NEXTION_UPLOAD_ACK_TIMEOUT_MS);
+          allocator.deallocate(buffer, 4096);
+          buffer = nullptr;
+          return -1;
+        }
+      }
       this->content_length_ -= read_len;
       const float upload_percentage = 100.0f * (this->tft_size_ - this->content_length_) / this->tft_size_;
 #ifdef USE_PSRAM

esphome/components/ota/ota_backend.h

@@ -4,6 +4,8 @@
 #include "esphome/core/defines.h"
 #include "esphome/core/helpers.h"
 
+#include <cstdint>
+
 #ifdef USE_OTA_STATE_LISTENER
 #include <vector>
 #endif
@@ -23,6 +25,7 @@ enum OTAResponseTypes {
   OTA_RESPONSE_UPDATE_END_OK = 0x45,
   OTA_RESPONSE_SUPPORTS_COMPRESSION = 0x46,
   OTA_RESPONSE_CHUNK_OK = 0x47,
+  OTA_RESPONSE_FEATURE_FLAGS = 0x48,
 
   OTA_RESPONSE_ERROR_MAGIC = 0x80,
   OTA_RESPONSE_ERROR_UPDATE_PREPARE = 0x81,
@@ -38,6 +41,7 @@ enum OTAResponseTypes {
   OTA_RESPONSE_ERROR_MD5_MISMATCH = 0x8B,
   OTA_RESPONSE_ERROR_RP2040_NOT_ENOUGH_SPACE = 0x8C,
   OTA_RESPONSE_ERROR_SIGNATURE_INVALID = 0x8D,
+  OTA_RESPONSE_ERROR_UNSUPPORTED_OTA_TYPE = 0x8E,
   OTA_RESPONSE_ERROR_UNKNOWN = 0xFF,
 };
 
@@ -49,6 +53,10 @@ enum OTAState {
   OTA_ERROR,
 };
 
+enum OTAType : uint8_t {
+  OTA_TYPE_UPDATE_APP = 0x00,
+};
+
 /** Listener interface for OTA state changes.
  *
  * Components can implement this interface to receive OTA state updates

esphome/components/rp2040/core.cpp

@@ -1,41 +1,6 @@
 #ifdef USE_RP2040
 
-#include "core.h"
-#include "esphome/core/defines.h"
-#ifdef USE_RP2040_CRASH_HANDLER
-#include "crash_handler.h"
-#endif
-#include "esphome/core/hal.h"
-#include "esphome/core/helpers.h"
-
-#include "hardware/timer.h"
-#include "hardware/watchdog.h"
-
-namespace esphome {
-
-// yield(), delay(), micros(), millis(), millis_64() inlined in hal.h.
-void HOT delayMicroseconds(uint32_t us) { delay_microseconds_safe(us); }
-void arch_restart() {
-  watchdog_reboot(0, 0, 10);
-  while (1) {
-    continue;
-  }
-}
-
-void arch_init() {
-#ifdef USE_RP2040_CRASH_HANDLER
-  rp2040::crash_handler_read_and_clear();
-#endif
-#if USE_RP2040_WATCHDOG_TIMEOUT > 0
-  watchdog_enable(USE_RP2040_WATCHDOG_TIMEOUT, false);
-#endif
-}
-
-void HOT arch_feed_wdt() { watchdog_update(); }
-
-uint32_t HOT arch_get_cpu_cycle_count() { return ulMainGetRunTimeCounterValue(); }
-uint32_t arch_get_cpu_freq_hz() { return RP2040::f_cpu(); }
-
-}  // namespace esphome
+// HAL functions live in hal.cpp. core.cpp is intentionally empty for
+// rp2040 — there is no extra component bootstrap to keep here.
 
 #endif  // USE_RP2040

esphome/components/rp2040/hal.cpp

@@ -0,0 +1,41 @@
+#ifdef USE_RP2040
+
+#include "core.h"
+#include "esphome/core/defines.h"
+#include "esphome/core/hal.h"
+#ifdef USE_RP2040_CRASH_HANDLER
+#include "crash_handler.h"
+#endif
+
+#include "hardware/watchdog.h"
+
+// Empty rp2040 namespace block to satisfy ci-custom's lint_namespace check.
+// HAL functions live in namespace esphome (root) — they are not part of the
+// rp2040 component's API.
+namespace esphome::rp2040 {}  // namespace esphome::rp2040
+
+namespace esphome {
+
+// yield(), delay(), micros(), millis(), millis_64(), delayMicroseconds(),
+// arch_feed_wdt(), arch_get_cpu_cycle_count() inlined in components/rp2040/hal.h.
+void arch_restart() {
+  watchdog_reboot(0, 0, 10);
+  while (1) {
+    continue;
+  }
+}
+
+void arch_init() {
+#ifdef USE_RP2040_CRASH_HANDLER
+  rp2040::crash_handler_read_and_clear();
+#endif
+#if USE_RP2040_WATCHDOG_TIMEOUT > 0
+  watchdog_enable(USE_RP2040_WATCHDOG_TIMEOUT, false);
+#endif
+}
+
+uint32_t arch_get_cpu_freq_hz() { return RP2040::f_cpu(); }
+
+}  // namespace esphome
+
+#endif  // USE_RP2040

esphome/components/rp2040/hal.h

@@ -20,8 +20,19 @@ extern "C" unsigned long millis(void);
 // Forward decl from <pico/time.h>.
 extern "C" uint64_t time_us_64(void);
 
+// Forward decls from pico-sdk / FreeRTOS port for the inline arch_*
+// wrappers below.
+extern "C" void watchdog_update(void);
+extern "C" unsigned long ulMainGetRunTimeCounterValue(void);
+
+namespace esphome::rp2040 {}
+
 namespace esphome {
 
+// Forward decl from helpers.h.
+// NOLINTNEXTLINE(readability-redundant-declaration)
+void delay_microseconds_safe(uint32_t us);
+
 /// Returns true when executing inside an interrupt handler.
 __attribute__((always_inline)) inline bool in_isr_context() {
   uint32_t ipsr;
@@ -35,9 +46,15 @@ __attribute__((always_inline)) inline uint32_t micros() { return static_cast<uin
 __attribute__((always_inline)) inline uint32_t millis() { return micros_to_millis(::time_us_64()); }
 __attribute__((always_inline)) inline uint64_t millis_64() { return micros_to_millis<uint64_t>(::time_us_64()); }
 
-void delayMicroseconds(uint32_t us);  // NOLINT(readability-identifier-naming)
-void arch_feed_wdt();
-uint32_t arch_get_cpu_cycle_count();
+// NOLINTNEXTLINE(readability-identifier-naming)
+__attribute__((always_inline)) inline void delayMicroseconds(uint32_t us) { delay_microseconds_safe(us); }
+__attribute__((always_inline)) inline void arch_feed_wdt() { watchdog_update(); }
+__attribute__((always_inline)) inline uint32_t arch_get_cpu_cycle_count() {
+  return static_cast<uint32_t>(ulMainGetRunTimeCounterValue());
+}
+
+void arch_init();
+uint32_t arch_get_cpu_freq_hz();
 
 }  // namespace esphome
 

esphome/components/sendspin/__init__.py

@@ -24,6 +24,7 @@ CONF_SENDSPIN_ID = "sendspin_id"
 
 CONF_INITIAL_STATIC_DELAY = "initial_static_delay"
 CONF_FIXED_DELAY = "fixed_delay"
+CONF_DECODE_MEMORY = "decode_memory"
 
 # sendspin-cpp library lives in the global `sendspin` namespace.
 sendspin_library_ns = cg.global_ns.namespace("sendspin")
@@ -39,6 +40,18 @@ CODEC_FORMAT_UNSUPPORTED = SendspinCodecFormat.enum("UNSUPPORTED")
 AudioSupportedFormatObject = sendspin_library_ns.struct("AudioSupportedFormatObject")
 PlayerRoleConfig = sendspin_library_ns.struct("PlayerRoleConfig")
 
+# MemoryLocation enum (from sendspin/types.h) controls SPIRAM-vs-internal-RAM placement
+# preference for the player role's transfer buffers.
+SendspinMemoryLocation = sendspin_library_ns.enum("MemoryLocation", is_class=True)
+
+MEMORY_PSRAM = "psram"
+MEMORY_INTERNAL = "internal"
+MEMORY_LOCATIONS = [MEMORY_PSRAM, MEMORY_INTERNAL]
+MEMORY_LOCATION_ENUM = {
+    MEMORY_PSRAM: SendspinMemoryLocation.PREFER_EXTERNAL,
+    MEMORY_INTERNAL: SendspinMemoryLocation.PREFER_INTERNAL,
+}
+
 # Trailing underscore avoids clashing with sendspin-cpp's global `sendspin` namespace.
 # Analysis tools strip the trailing underscore (same pattern as `template_`).
 sendspin_ns = cg.esphome_ns.namespace("sendspin_")
@@ -193,7 +206,7 @@ async def to_code(config: ConfigType) -> None:
         )
 
     # sendspin-cpp library
-    esp32.add_idf_component(name="sendspin/sendspin-cpp", ref="0.3.1")
+    esp32.add_idf_component(name="sendspin/sendspin-cpp", ref="0.4.0")
 
     cg.add_define("USE_SENDSPIN", True)  # for MDNS
 
@@ -249,14 +262,23 @@ async def to_code(config: ConfigType) -> None:
                 "CONFIG_SPIRAM_ALLOW_STACK_EXTERNAL_MEMORY", True
             )
 
-        player_config_struct = cg.StructInitializer(
-            PlayerRoleConfig,
+        # Library defaults: priority 18 (one above httpd_priority 17 so the decoder is not
+        # starved by the HTTP server during the initial encoded-audio burst at stream start),
+        # interpolation/decode buffer locations PREFER_EXTERNAL.
+        player_struct_fields = [
             ("audio_formats", audio_format_structs),
             ("audio_buffer_capacity", player_cfg[CONF_BUFFER_SIZE]),
             ("fixed_delay_us", player_cfg[CONF_FIXED_DELAY]),
             ("initial_static_delay_ms", player_cfg[CONF_INITIAL_STATIC_DELAY]),
             ("psram_stack", psram_stack),
-            ("priority", 2),
+        ]
+        if (decode_memory := player_cfg.get(CONF_DECODE_MEMORY)) is not None:
+            player_struct_fields.append(
+                ("decode_buffer_location", MEMORY_LOCATION_ENUM[decode_memory])
+            )
+        player_config_struct = cg.StructInitializer(
+            PlayerRoleConfig,
+            *player_struct_fields,
         )
         cg.add(var.set_player_config(player_config_struct))
     else:

esphome/components/sendspin/media_source/__init__.py

@@ -13,9 +13,11 @@ from esphome.cpp_generator import MockObj, TemplateArgsType
 from esphome.types import ConfigType
 
 from .. import (
+    CONF_DECODE_MEMORY,
     CONF_FIXED_DELAY,
     CONF_INITIAL_STATIC_DELAY,
     CONF_SENDSPIN_ID,
+    MEMORY_LOCATIONS,
     SendspinHub,
     _validate_task_stack_in_psram,
     register_player_config,
@@ -57,6 +59,7 @@ def _register(config: ConfigType) -> ConfigType:
             CONF_INITIAL_STATIC_DELAY: config[CONF_INITIAL_STATIC_DELAY],
             CONF_FIXED_DELAY: config[CONF_FIXED_DELAY],
             CONF_TASK_STACK_IN_PSRAM: config.get(CONF_TASK_STACK_IN_PSRAM, False),
+            CONF_DECODE_MEMORY: config.get(CONF_DECODE_MEMORY),
         }
     )
     return config
@@ -82,6 +85,7 @@ CONFIG_SCHEMA = cv.All(
             cv.Optional(CONF_SAMPLE_RATE, default=48000): cv.int_range(
                 min=16000, max=96000
             ),
+            cv.Optional(CONF_DECODE_MEMORY): cv.one_of(*MEMORY_LOCATIONS, lower=True),
         }
     ),
     cv.only_on_esp32,

esphome/components/sensor/__init__.py

@@ -266,7 +266,7 @@ StreamingMovingAverageFilter = sensor_ns.class_("StreamingMovingAverageFilter",
 ExponentialMovingAverageFilter = sensor_ns.class_(
     "ExponentialMovingAverageFilter", Filter
 )
-ThrottleAverageFilter = sensor_ns.class_("ThrottleAverageFilter", Filter, cg.Component)
+ThrottleAverageFilter = sensor_ns.class_("ThrottleAverageFilter", Filter)
 LambdaFilter = sensor_ns.class_("LambdaFilter", Filter)
 StatelessLambdaFilter = sensor_ns.class_("StatelessLambdaFilter", Filter)
 OffsetFilter = sensor_ns.class_("OffsetFilter", Filter)
@@ -283,8 +283,8 @@ ThrottleWithPriorityNanFilter = sensor_ns.class_(
 TimeoutFilterBase = sensor_ns.class_("TimeoutFilterBase", Filter, cg.Component)
 TimeoutFilterLast = sensor_ns.class_("TimeoutFilterLast", TimeoutFilterBase)
 TimeoutFilterConfigured = sensor_ns.class_("TimeoutFilterConfigured", TimeoutFilterBase)
-DebounceFilter = sensor_ns.class_("DebounceFilter", Filter, cg.Component)
-HeartbeatFilter = sensor_ns.class_("HeartbeatFilter", Filter, cg.Component)
+DebounceFilter = sensor_ns.class_("DebounceFilter", Filter)
+HeartbeatFilter = sensor_ns.class_("HeartbeatFilter", Filter)
 DeltaFilter = sensor_ns.class_("DeltaFilter", Filter)
 OrFilter = sensor_ns.class_("OrFilter", Filter)
 CalibrateLinearFilter = sensor_ns.class_("CalibrateLinearFilter", Filter)
@@ -564,12 +564,15 @@ async def exponential_moving_average_filter_to_code(config, filter_id):
 
 
 @FILTER_REGISTRY.register(
-    "throttle_average", ThrottleAverageFilter, cv.positive_time_period_milliseconds
+    "throttle_average",
+    ThrottleAverageFilter,
+    cv.All(
+        cv.positive_time_period_milliseconds,
+        cv.Range(max=cv.TimePeriod(hours=24)),
+    ),
 )
 async def throttle_average_filter_to_code(config, filter_id):
-    var = cg.new_Pvariable(filter_id, config)
-    await cg.register_component(var, {})
-    return var
+    return cg.new_Pvariable(filter_id, config)
 
 
 @FILTER_REGISTRY.register("lambda", LambdaFilter, cv.returning_lambda)
@@ -698,13 +701,10 @@ HEARTBEAT_SCHEMA = cv.Schema(
 async def heartbeat_filter_to_code(config, filter_id):
     if isinstance(config, dict):
         var = cg.new_Pvariable(filter_id, config[CONF_PERIOD])
-        await cg.register_component(var, {})
         cg.add(var.set_optimistic(config[CONF_OPTIMISTIC]))
         return var
 
-    var = cg.new_Pvariable(filter_id, config)
-    await cg.register_component(var, {})
-    return var
+    return cg.new_Pvariable(filter_id, config)
 
 
 TIMEOUT_SCHEMA = cv.maybe_simple_value(
@@ -738,9 +738,7 @@ async def timeout_filter_to_code(config, filter_id):
     "debounce", DebounceFilter, cv.positive_time_period_milliseconds
 )
 async def debounce_filter_to_code(config, filter_id):
-    var = cg.new_Pvariable(filter_id, config)
-    await cg.register_component(var, {})
-    return var
+    return cg.new_Pvariable(filter_id, config)
 
 
 CONF_DATAPOINTS = "datapoints"

esphome/components/sensor/filter.cpp

@@ -13,11 +13,6 @@ namespace esphome::sensor {
 
 static const char *const TAG = "sensor.filter";
 
-// Filter scheduler IDs.
-// Each filter is its own Component instance, so the scheduler scopes
-// IDs by component pointer — no risk of collisions between instances.
-constexpr uint32_t FILTER_ID = 0;
-
 // Filter
 void Filter::input(float value) {
   ESP_LOGVV(TAG, "Filter(%p)::input(%f)", this, value);
@@ -185,8 +180,9 @@ optional<float> ThrottleAverageFilter::new_value(float value) {
   }
   return {};
 }
-void ThrottleAverageFilter::setup() {
-  this->set_interval(FILTER_ID, this->time_period_, [this]() {
+void ThrottleAverageFilter::initialize(Sensor *parent, Filter *next) {
+  Filter::initialize(parent, next);
+  App.scheduler.set_interval(this, this->time_period_, [this]() {
     ESP_LOGVV(TAG, "ThrottleAverageFilter(%p)::interval(sum=%f, n=%i)", this, this->sum_, this->n_);
     if (this->n_ == 0) {
       if (this->have_nan_)
@@ -199,7 +195,6 @@ void ThrottleAverageFilter::setup() {
     this->have_nan_ = false;
   });
 }
-float ThrottleAverageFilter::get_setup_priority() const { return setup_priority::HARDWARE; }
 
 // LambdaFilter
 LambdaFilter::LambdaFilter(lambda_filter_t lambda_filter) : lambda_filter_(std::move(lambda_filter)) {}
@@ -362,13 +357,12 @@ optional<float> TimeoutFilterConfigured::new_value(float value) {
 
 // DebounceFilter
 optional<float> DebounceFilter::new_value(float value) {
-  this->set_timeout(FILTER_ID, this->time_period_, [this, value]() { this->output(value); });
+  App.scheduler.set_timeout(this, this->time_period_, [this, value]() { this->output(value); });
 
   return {};
 }
 
 DebounceFilter::DebounceFilter(uint32_t time_period) : time_period_(time_period) {}
-float DebounceFilter::get_setup_priority() const { return setup_priority::HARDWARE; }
 
 // HeartbeatFilter
 HeartbeatFilter::HeartbeatFilter(uint32_t time_period) : time_period_(time_period), last_input_(NAN) {}
@@ -384,8 +378,9 @@ optional<float> HeartbeatFilter::new_value(float value) {
   return {};
 }
 
-void HeartbeatFilter::setup() {
-  this->set_interval(FILTER_ID, this->time_period_, [this]() {
+void HeartbeatFilter::initialize(Sensor *parent, Filter *next) {
+  Filter::initialize(parent, next);
+  App.scheduler.set_interval(this, this->time_period_, [this]() {
     ESP_LOGVV(TAG, "HeartbeatFilter(%p)::interval(has_value=%s, last_input=%f)", this, YESNO(this->has_value_),
               this->last_input_);
     if (!this->has_value_)
@@ -395,8 +390,6 @@ void HeartbeatFilter::setup() {
   });
 }
 
-float HeartbeatFilter::get_setup_priority() const { return setup_priority::HARDWARE; }
-
 optional<float> calibrate_linear_compute(const std::array<float, 3> *functions, size_t count, float value) {
   for (size_t i = 0; i < count; i++) {
     if (!std::isfinite(functions[i][2]) || value < functions[i][2])

esphome/components/sensor/filter.h

@@ -254,21 +254,22 @@ class ExponentialMovingAverageFilter : public Filter {
  *
  * It takes the average of all the values received in a period of time.
  */
-class ThrottleAverageFilter : public Filter, public Component {
+class ThrottleAverageFilter : public Filter {
  public:
   explicit ThrottleAverageFilter(uint32_t time_period);
 
-  void setup() override;
+  void initialize(Sensor *parent, Filter *next) override;
 
   optional<float> new_value(float value) override;
 
-  float get_setup_priority() const override;
-
  protected:
   float sum_{0.0f};
-  unsigned int n_{0};
   uint32_t time_period_;
-  bool have_nan_{false};
+  // Sample count packed with NaN-seen flag in a single 32-bit word.
+  // n_ is bounded by YAML cap on time_period_ (24 h) × max plausible source
+  // rate (1 kHz) = 86.4M ≪ 2^31, so 31 bits has 25x headroom.
+  uint32_t n_ : 31 {0};
+  uint32_t have_nan_ : 1 {0};
 };
 
 using lambda_filter_t = std::function<optional<float>(float)>;
@@ -454,25 +455,22 @@ class TimeoutFilterConfigured : public TimeoutFilterBase {
   // Total: 8 (base) + 4 = 12 bytes + vtable ptr + Component overhead
 };
 
-class DebounceFilter : public Filter, public Component {
+class DebounceFilter : public Filter {
  public:
   explicit DebounceFilter(uint32_t time_period);
 
   optional<float> new_value(float value) override;
 
-  float get_setup_priority() const override;
-
  protected:
   uint32_t time_period_;
 };
 
-class HeartbeatFilter : public Filter, public Component {
+class HeartbeatFilter : public Filter {
  public:
   explicit HeartbeatFilter(uint32_t time_period);
 
-  void setup() override;
+  void initialize(Sensor *parent, Filter *next) override;
   optional<float> new_value(float value) override;
-  float get_setup_priority() const override;
 
   void set_optimistic(bool optimistic) { this->optimistic_ = optimistic; }
 

esphome/components/st7789v/__init__.py

@@ -1,3 +1,8 @@
 import esphome.codegen as cg
 
 st7789v_ns = cg.esphome_ns.namespace("st7789v")
+
+DEPRECATED_COMPONENT = """
+The 'st7789v' component is deprecated and no new functionality will be added to it.
+PRs should target the newer and more performant 'mipi_spi' component.
+"""

esphome/components/st7789v/display.py

@@ -1,3 +1,5 @@
+import logging
+
 from esphome import pins
 import esphome.codegen as cg
 from esphome.components import display, power_supply, spi
@@ -26,6 +28,8 @@ CODEOWNERS = ["@kbx81"]
 
 DEPENDENCIES = ["spi"]
 
+LOGGER = logging.getLogger(__name__)
+
 ST7789V = st7789v_ns.class_(
     "ST7789V", cg.PollingComponent, spi.SPIDevice, display.DisplayBuffer
 )
@@ -175,6 +179,9 @@ FINAL_VALIDATE_SCHEMA = spi.final_validate_device_schema(
 
 
 async def to_code(config):
+    LOGGER.warning(
+        "The 'st7789v' component is deprecated, it is recommended to use 'mipi_spi' instead."
+    )
     var = cg.new_Pvariable(config[CONF_ID])
     await display.register_display(var, config)
     await spi.register_spi_device(var, config, write_only=True)

esphome/components/usb_host/__init__.py

@@ -10,6 +10,7 @@ from esphome.components.esp32 import (
 )
 import esphome.config_validation as cv
 from esphome.const import CONF_DEVICES, CONF_ID
+from esphome.core import CORE
 from esphome.cpp_types import Component
 from esphome.types import ConfigType
 
@@ -19,14 +20,15 @@ DEPENDENCIES = ["esp32"]
 usb_host_ns = cg.esphome_ns.namespace("usb_host")
 USBHost = usb_host_ns.class_("USBHost", Component)
 USBClient = usb_host_ns.class_("USBClient", Component)
-
+DOMAIN = "usb_host"
 CONF_VID = "vid"
 CONF_PID = "pid"
 CONF_ENABLE_HUBS = "enable_hubs"
 CONF_MAX_TRANSFER_REQUESTS = "max_transfer_requests"
+CONF_MAX_PACKET_SIZE = "max_packet_size"
 
 
-def usb_device_schema(cls=USBClient, vid: int = None, pid: [int] = None) -> cv.Schema:
+def usb_device_schema(cls=USBClient, vid: int = None, pid: int = None) -> cv.Schema:
     schema = cv.COMPONENT_SCHEMA.extend(
         {
             cv.GenerateID(): cv.declare_id(cls),
@@ -43,6 +45,17 @@ def usb_device_schema(cls=USBClient, vid: int = None, pid: [int] = None) -> cv.S
     return schema
 
 
+def _set_max_packet_size(config: dict) -> dict:
+    CORE.data.setdefault(DOMAIN, {})[CONF_MAX_PACKET_SIZE] = config[
+        CONF_MAX_PACKET_SIZE
+    ]
+    return config
+
+
+def get_max_packet_size() -> int:
+    return CORE.data.get(DOMAIN, {}).get(CONF_MAX_PACKET_SIZE, 64)
+
+
 CONFIG_SCHEMA = cv.All(
     cv.COMPONENT_SCHEMA.extend(
         {
@@ -51,10 +64,14 @@ CONFIG_SCHEMA = cv.All(
             cv.Optional(CONF_MAX_TRANSFER_REQUESTS, default=16): cv.int_range(
                 min=1, max=32
             ),
+            cv.Optional(CONF_MAX_PACKET_SIZE, default=64): cv.one_of(
+                64, 128, 256, 512, 1024, int=True
+            ),
             cv.Optional(CONF_DEVICES): cv.ensure_list(usb_device_schema()),
         }
     ),
     only_on_variant(supported=[VARIANT_ESP32P4, VARIANT_ESP32S2, VARIANT_ESP32S3]),
+    _set_max_packet_size,
 )
 
 
@@ -72,8 +89,8 @@ async def to_code(config: ConfigType) -> None:
     if config.get(CONF_ENABLE_HUBS):
         add_idf_sdkconfig_option("CONFIG_USB_HOST_HUBS_SUPPORTED", True)
 
-    max_requests = config[CONF_MAX_TRANSFER_REQUESTS]
-    cg.add_define("USB_HOST_MAX_REQUESTS", max_requests)
+    cg.add_define("USB_HOST_MAX_REQUESTS", config[CONF_MAX_TRANSFER_REQUESTS])
+    cg.add_define("USB_HOST_MAX_PACKET_SIZE", config[CONF_MAX_PACKET_SIZE])
 
     var = cg.new_Pvariable(config[CONF_ID])
     await cg.register_component(var, config)

esphome/components/usb_host/usb_host.h

@@ -66,6 +66,8 @@ static_assert(MAX_REQUESTS >= 1 && MAX_REQUESTS <= 32, "MAX_REQUESTS must be bet
 using trq_bitmask_t = std::conditional<(MAX_REQUESTS <= 16), uint16_t, uint32_t>::type;
 static constexpr trq_bitmask_t ALL_REQUESTS_IN_USE = MAX_REQUESTS == 32 ? ~0 : (1 << MAX_REQUESTS) - 1;
 
+static constexpr size_t USB_MAX_PACKET_SIZE =
+    USB_HOST_MAX_PACKET_SIZE;                        // Max USB packet size (64 for FS, 512 for P4 HS)
 static constexpr size_t USB_EVENT_QUEUE_SIZE = 32;   // Size of event queue between USB task and main loop
 static constexpr size_t USB_TASK_STACK_SIZE = 4096;  // Stack size for USB task (same as ESP-IDF USB examples)
 static constexpr UBaseType_t USB_TASK_PRIORITY = 5;  // Higher priority than main loop (tskIDLE_PRIORITY + 5)

esphome/components/usb_host/usb_host_client.cpp

@@ -217,7 +217,7 @@ void USBClient::setup() {
   // Pre-allocate USB transfer buffers for all slots at startup
   // This avoids any dynamic allocation during runtime
   for (auto &request : this->requests_) {
-    usb_host_transfer_alloc(64, 0, &request.transfer);
+    usb_host_transfer_alloc(USB_MAX_PACKET_SIZE, 0, &request.transfer);
     request.client = this;  // Set once, never changes
   }
 

esphome/components/usb_uart/__init__.py

@@ -1,7 +1,11 @@
 import esphome.codegen as cg
 from esphome.components.const import CONF_DATA_BITS, CONF_PARITY, CONF_STOP_BITS
 from esphome.components.uart import CONF_DEBUG_PREFIX, CONF_FLUSH_TIMEOUT, UARTComponent
-from esphome.components.usb_host import register_usb_client, usb_device_schema
+from esphome.components.usb_host import (
+    get_max_packet_size,
+    register_usb_client,
+    usb_device_schema,
+)
 import esphome.config_validation as cv
 from esphome.const import (
     CONF_BAUD_RATE,
@@ -118,14 +122,14 @@ CONFIG_SCHEMA = cv.ensure_list(
 async def to_code(config):
     # The output chunk pool/queue are compile-time-sized templates shared by all
     # USBUartChannel instances, so use the largest buffer_size across every channel
-    # of every device.  Each chunk is 64 bytes (USB FS MPS); add one extra slot
-    # because LockFreeQueue<T,N> is a ring buffer that wastes one entry.
+    # of every device. Add one extra slot because LockFreeQueue<T,N> is a ring
+    # buffer that wastes one entry.
     max_buffer_size = max(
         channel[CONF_BUFFER_SIZE]
         for device in config
         for channel in device[CONF_CHANNELS]
     )
-    output_chunk_count = max_buffer_size // 64 + 1
+    output_chunk_count = max(max_buffer_size // get_max_packet_size(), 2) + 1
     cg.add_define("USB_UART_OUTPUT_CHUNK_COUNT", output_chunk_count)
 
     for device in config:

esphome/components/usb_uart/usb_uart.cpp

@@ -157,7 +157,7 @@ void USBUartChannel::write_array(const uint8_t *data, size_t len) {
       ESP_LOGE(TAG, "Output pool full - lost %zu bytes", len);
       break;
     }
-    size_t chunk_len = std::min(len, UsbOutputChunk::MAX_CHUNK_SIZE);
+    uint16_t chunk_len = std::min(len, UsbOutputChunk::MAX_CHUNK_SIZE);
     memcpy(chunk->data, data, chunk_len);
     chunk->length = static_cast<uint8_t>(chunk_len);
     // Push always succeeds: pool is sized to queue capacity (SIZE-1), so if
@@ -222,7 +222,7 @@ void USBUartComponent::loop() {
 
 #ifdef USE_UART_DEBUGGER
     if (channel->debug_) {
-      char buf[4 + format_hex_pretty_size(UsbDataChunk::MAX_CHUNK_SIZE)];  // "<<< " + hex
+      char buf[4 + format_hex_pretty_size(usb_host::USB_MAX_PACKET_SIZE)];  // "<<< " + hex
       memcpy(buf, "<<< ", 4);
       format_hex_pretty_to(buf + 4, sizeof(buf) - 4, chunk->data, chunk->length, ',');
       ESP_LOGD(TAG, "%s%s", channel->debug_prefix_.c_str(), buf);
@@ -377,7 +377,7 @@ void USBUartComponent::start_output(USBUartChannel *channel) {
     this->start_output(channel);
   };
 
-  const uint8_t len = chunk->length;
+  const auto len = chunk->length;
   if (!this->transfer_out(ep->bEndpointAddress, callback, chunk->data, len)) {
     // Transfer submission failed — return chunk and release flag so callers can retry.
     channel->output_pool_.release(chunk);
@@ -394,10 +394,10 @@ void USBUartComponent::start_output(USBUartChannel *channel) {
 static void fix_mps(const usb_ep_desc_t *ep) {
   if (ep != nullptr) {
     auto *ep_mutable = const_cast<usb_ep_desc_t *>(ep);
-    if (ep->wMaxPacketSize > 64) {
-      ESP_LOGW(TAG, "Corrected MPS of EP 0x%02X from %u to 64", static_cast<uint8_t>(ep->bEndpointAddress & 0xFF),
-               ep->wMaxPacketSize);
-      ep_mutable->wMaxPacketSize = 64;
+    if (ep->wMaxPacketSize > usb_host::USB_MAX_PACKET_SIZE) {
+      ESP_LOGW(TAG, "Corrected MPS of EP 0x%02X from %u to %u", static_cast<uint8_t>(ep->bEndpointAddress & 0xFF),
+               ep->wMaxPacketSize, usb_host::USB_MAX_PACKET_SIZE);
+      ep_mutable->wMaxPacketSize = usb_host::USB_MAX_PACKET_SIZE;
     }
   }
 }

esphome/components/usb_uart/usb_uart.h

@@ -106,20 +106,19 @@ class RingBuffer {
 
 // Structure for queuing received USB data chunks
 struct UsbDataChunk {
-  static constexpr size_t MAX_CHUNK_SIZE = 64;  // USB packet size
-  uint8_t data[MAX_CHUNK_SIZE];
-  uint8_t length;  // Max 64 bytes, so uint8_t is sufficient
+  uint8_t data[usb_host::USB_MAX_PACKET_SIZE];
+  uint16_t length;
   USBUartChannel *channel;
 
   // Required for EventPool - no cleanup needed for POD types
   void release() {}
 };
 
-// Structure for queuing outgoing USB data chunks (one per USB FS packet)
+// Structure for queuing outgoing USB data chunks (one per USB packet)
 struct UsbOutputChunk {
-  static constexpr size_t MAX_CHUNK_SIZE = 64;  // USB FS MPS
+  static constexpr size_t MAX_CHUNK_SIZE = usb_host::USB_MAX_PACKET_SIZE;
   uint8_t data[MAX_CHUNK_SIZE];
-  uint8_t length;
+  uint16_t length;
 
   // Required for EventPool - no cleanup needed for POD types
   void release() {}

esphome/components/zephyr/core.cpp

@@ -1,8 +1,6 @@
 #ifdef USE_ZEPHYR
 
 #include <zephyr/kernel.h>
-#include <zephyr/drivers/watchdog.h>
-#include <zephyr/sys/reboot.h>
 #include <zephyr/random/random.h>
 #include "esphome/core/hal.h"
 #include "esphome/core/helpers.h"
@@ -10,55 +8,7 @@
 
 namespace esphome {
 
-#ifdef CONFIG_WATCHDOG
-static int wdt_channel_id = -1;  // NOLINT(cppcoreguidelines-avoid-non-const-global-variables)
-static const device *const WDT = DEVICE_DT_GET(DT_ALIAS(watchdog0));
-#endif
-
-void yield() { ::k_yield(); }
-uint32_t millis() { return static_cast<uint32_t>(millis_64()); }
-uint64_t millis_64() { return static_cast<uint64_t>(k_uptime_get()); }
-uint32_t micros() { return k_ticks_to_us_floor32(k_uptime_ticks()); }
-void delayMicroseconds(uint32_t us) { ::k_usleep(us); }
-void delay(uint32_t ms) { ::k_msleep(ms); }
-
-void arch_init() {
-#ifdef CONFIG_WATCHDOG
-  if (device_is_ready(WDT)) {
-    static wdt_timeout_cfg wdt_config{};
-    wdt_config.flags = WDT_FLAG_RESET_SOC;
-#ifdef USE_ZIGBEE
-    // zboss thread use a lot of cpu cycles during start
-    wdt_config.window.max = 10000;
-#else
-    wdt_config.window.max = 2000;
-#endif
-    wdt_channel_id = wdt_install_timeout(WDT, &wdt_config);
-    if (wdt_channel_id >= 0) {
-      uint8_t options = 0;
-#ifdef USE_DEBUG
-      options |= WDT_OPT_PAUSE_HALTED_BY_DBG;
-#endif
-#ifdef USE_DEEP_SLEEP
-      options |= WDT_OPT_PAUSE_IN_SLEEP;
-#endif
-      wdt_setup(WDT, options);
-    }
-  }
-#endif
-}
-
-void arch_feed_wdt() {
-#ifdef CONFIG_WATCHDOG
-  if (wdt_channel_id >= 0) {
-    wdt_feed(WDT, wdt_channel_id);
-  }
-#endif
-}
-
-void arch_restart() { sys_reboot(SYS_REBOOT_COLD); }
-uint32_t arch_get_cpu_cycle_count() { return k_cycle_get_32(); }
-uint32_t arch_get_cpu_freq_hz() { return sys_clock_hw_cycles_per_sec(); }
+// HAL functions live in hal.cpp.
 
 Mutex::Mutex() {
   auto *mutex = new k_mutex();

esphome/components/zephyr/hal.cpp

@@ -0,0 +1,63 @@
+#ifdef USE_ZEPHYR
+
+#include "esphome/core/defines.h"
+#include "esphome/core/hal.h"
+
+#include <zephyr/drivers/watchdog.h>
+#include <zephyr/sys/reboot.h>
+
+// Empty zephyr namespace block to satisfy ci-custom's lint_namespace check.
+// HAL functions live in namespace esphome (root) — they are not part of the
+// zephyr component's API.
+namespace esphome::zephyr {}  // namespace esphome::zephyr
+
+namespace esphome {
+
+#ifdef CONFIG_WATCHDOG
+static int wdt_channel_id = -1;  // NOLINT(cppcoreguidelines-avoid-non-const-global-variables)
+static const device *const WDT = DEVICE_DT_GET(DT_ALIAS(watchdog0));
+#endif
+
+// yield(), delay(), micros(), millis(), millis_64(), delayMicroseconds(),
+// arch_get_cpu_cycle_count(), arch_get_cpu_freq_hz() inlined in
+// components/zephyr/hal.h.
+
+void arch_init() {
+#ifdef CONFIG_WATCHDOG
+  if (device_is_ready(WDT)) {
+    static wdt_timeout_cfg wdt_config{};
+    wdt_config.flags = WDT_FLAG_RESET_SOC;
+#ifdef USE_ZIGBEE
+    // zboss thread uses a lot of CPU cycles during startup
+    wdt_config.window.max = 10000;
+#else
+    wdt_config.window.max = 2000;
+#endif
+    wdt_channel_id = wdt_install_timeout(WDT, &wdt_config);
+    if (wdt_channel_id >= 0) {
+      uint8_t options = 0;
+#ifdef USE_DEBUG
+      options |= WDT_OPT_PAUSE_HALTED_BY_DBG;
+#endif
+#ifdef USE_DEEP_SLEEP
+      options |= WDT_OPT_PAUSE_IN_SLEEP;
+#endif
+      wdt_setup(WDT, options);
+    }
+  }
+#endif
+}
+
+void arch_feed_wdt() {
+#ifdef CONFIG_WATCHDOG
+  if (wdt_channel_id >= 0) {
+    wdt_feed(WDT, wdt_channel_id);
+  }
+#endif
+}
+
+void arch_restart() { sys_reboot(SYS_REBOOT_COLD); }
+
+}  // namespace esphome
+
+#endif  // USE_ZEPHYR

esphome/components/zephyr/hal.h

@@ -0,0 +1,36 @@
+#pragma once
+
+#ifdef USE_ZEPHYR
+
+#include <cstdint>
+
+#include <zephyr/kernel.h>
+
+#define IRAM_ATTR
+#define PROGMEM
+
+namespace esphome::zephyr {}
+
+namespace esphome {
+
+/// Returns true when executing inside an interrupt handler.
+/// Zephyr/nRF52: not currently consulted — wake path is platform-specific.
+__attribute__((always_inline)) inline bool in_isr_context() { return false; }
+
+__attribute__((always_inline)) inline void yield() { ::k_yield(); }
+__attribute__((always_inline)) inline void delay(uint32_t ms) { ::k_msleep(ms); }
+__attribute__((always_inline)) inline uint32_t micros() { return k_ticks_to_us_floor32(k_uptime_ticks()); }
+__attribute__((always_inline)) inline uint64_t millis_64() { return static_cast<uint64_t>(k_uptime_get()); }
+__attribute__((always_inline)) inline uint32_t millis() { return static_cast<uint32_t>(millis_64()); }
+
+// NOLINTNEXTLINE(readability-identifier-naming)
+__attribute__((always_inline)) inline void delayMicroseconds(uint32_t us) { ::k_usleep(us); }
+__attribute__((always_inline)) inline uint32_t arch_get_cpu_cycle_count() { return k_cycle_get_32(); }
+__attribute__((always_inline)) inline uint32_t arch_get_cpu_freq_hz() { return sys_clock_hw_cycles_per_sec(); }
+
+void arch_feed_wdt();
+void arch_init();
+
+}  // namespace esphome
+
+#endif  // USE_ZEPHYR

esphome/components/zigbee/zigbee_esp32.py

@@ -9,6 +9,7 @@ from esphome.components.esp32 import (
     add_idf_component,
     add_idf_sdkconfig_option,
     add_partition,
+    idf_version,
     require_vfs_select,
 )
 import esphome.config_validation as cv
@@ -186,6 +187,10 @@ async def _zigbee_add_sdkconfigs(config: ConfigType) -> None:
     # The pre-built Zigbee library uses esp_log_default_level which requires
     # dynamic log level control to be enabled
     add_idf_sdkconfig_option("CONFIG_LOG_DYNAMIC_LEVEL_CONTROL", True)
+    # The pre-built Zigbee library is compiled against newlib which requires newlib
+    # reentrancy to be enabled with picolibc compatibility.
+    if idf_version() >= cv.Version(6, 0, 0):
+        add_idf_sdkconfig_option("CONFIG_LIBC_PICOLIBC_NEWLIB_COMPATIBILITY", True)
 
 
 async def attributes_to_code(

esphome/const.py

@@ -1,4 +1,4 @@
-"""Constants used by esphome."""
+"""Constants used by ESPHome."""
 
 from enum import Enum
 

esphome/core/application.h

@@ -637,10 +637,12 @@ inline void ESPHOME_ALWAYS_INLINE Application::loop() {
   // flag preserves it. wake_request_take() exchange-clears the flag; wakes
   // that arrive during Phase B re-set it and run Phase B again on the next
   // iteration.
-  const bool high_frequency = HighFrequencyLoopRequester::is_high_frequency();
-  const uint32_t elapsed = now - this->last_loop_;
-  const bool woke = esphome::wake_request_take();
-  const bool do_component_phase = high_frequency || woke || (elapsed >= this->loop_interval_);
+  //
+  // wake_request_take() must always be called first since it does an
+  // atomic exchange to clear the flag, and we want to run the component phase
+  // if either the flag was set or the scheduler requested a high-frequency loop.
+  const bool do_component_phase = esphome::wake_request_take() || HighFrequencyLoopRequester::is_high_frequency() ||
+                                  (now - this->last_loop_ >= this->loop_interval_);
 
   if (do_component_phase) {
     ComponentPhaseGuard phase_guard{*this};

esphome/core/component.h

@@ -65,7 +65,6 @@ inline constexpr uint32_t SCHEDULER_DONT_RUN = 4294967295UL;
 /// with component-level NUMERIC_ID values, even if the uint32_t values overlap.
 enum class InternalSchedulerID : uint32_t {
   POLLING_UPDATE = 0,  // PollingComponent interval
-  DELAY_ACTION = 1,    // DelayAction timeout
 };
 
 // Forward declaration

esphome/core/hal.h

@@ -8,22 +8,22 @@
 
 // Per-platform HAL bits (IRAM_ATTR / PROGMEM macros, in_isr_context(),
 // inline yield/delay/micros/millis/millis_64 wrappers, ESP8266 progmem
-// helpers) live under esphome/core/hal/ and are dispatched here based on
-// the active USE_* platform define. Each header guards its body with the
-// matching #ifdef USE_<platform> and re-enters namespace esphome {} so it
-// is safe to be re-included.
+// helpers) live next to each platform component as components/<platform>/hal.h
+// and are dispatched here based on the active USE_* platform define. Each
+// header guards its body with the matching #ifdef USE_<platform> and re-enters
+// namespace esphome {} so it is safe to be re-included.
 #if defined(USE_ESP32)
-#include "esphome/core/hal/hal_esp32.h"
+#include "esphome/components/esp32/hal.h"
 #elif defined(USE_ESP8266)
-#include "esphome/core/hal/hal_esp8266.h"
+#include "esphome/components/esp8266/hal.h"
 #elif defined(USE_LIBRETINY)
-#include "esphome/core/hal/hal_libretiny.h"
+#include "esphome/components/libretiny/hal.h"
 #elif defined(USE_RP2040)
-#include "esphome/core/hal/hal_rp2040.h"
+#include "esphome/components/rp2040/hal.h"
 #elif defined(USE_HOST)
-#include "esphome/core/hal/hal_host.h"
+#include "esphome/components/host/hal.h"
 #elif defined(USE_ZEPHYR)
-#include "esphome/core/hal/hal_zephyr.h"
+#include "esphome/components/zephyr/hal.h"
 #else
 #error "hal.h: not implemented for this platform"
 #endif
@@ -31,15 +31,14 @@
 namespace esphome {
 
 // Cross-platform declarations. delayMicroseconds(), arch_feed_wdt(),
-// arch_get_cpu_cycle_count() vary per platform (some inline, some
-// out-of-line) so they live in hal/hal_<platform>.h.
+// arch_get_cpu_cycle_count(), arch_init(), arch_get_cpu_freq_hz() vary
+// per platform (some inline, some out-of-line) so they live in
+// components/<platform>/hal.h.
 void __attribute__((noreturn)) arch_restart();
-void arch_init();
-uint32_t arch_get_cpu_freq_hz();
 
 #ifndef USE_ESP8266
 // All non-ESP8266 platforms: PROGMEM is a no-op, so these are direct dereferences.
-// ESP8266's out-of-line declarations live in hal/hal_esp8266.h.
+// ESP8266's out-of-line declarations live in components/esp8266/hal.h.
 inline uint8_t progmem_read_byte(const uint8_t *addr) { return *addr; }
 inline const char *progmem_read_ptr(const char *const *addr) { return *addr; }
 inline uint16_t progmem_read_uint16(const uint16_t *addr) { return *addr; }

esphome/core/hal/hal_zephyr.h

@@ -1,28 +0,0 @@
-#pragma once
-
-#ifdef USE_ZEPHYR
-
-#include <cstdint>
-
-#define IRAM_ATTR
-#define PROGMEM
-
-namespace esphome {
-
-/// Returns true when executing inside an interrupt handler.
-/// Zephyr/nRF52: not currently consulted — wake path is platform-specific.
-__attribute__((always_inline)) inline bool in_isr_context() { return false; }
-
-void yield();
-void delay(uint32_t ms);
-uint32_t micros();
-uint32_t millis();
-uint64_t millis_64();
-
-void delayMicroseconds(uint32_t us);  // NOLINT(readability-identifier-naming)
-void arch_feed_wdt();
-uint32_t arch_get_cpu_cycle_count();
-
-}  // namespace esphome
-
-#endif  // USE_ZEPHYR

esphome/core/helpers.h

@@ -2045,7 +2045,8 @@ void delay_microseconds_safe(uint32_t us);
  * Returns `nullptr` in case no memory is available.
  *
  * By setting flags, it can be configured to:
- * - perform external allocation falling back to main memory if SPI RAM is full or unavailable
+ * - perform external allocation falling back to internal memory if SPI RAM is full or unavailable (default)
+ * - perform internal allocation falling back to external memory (with PREFER_INTERNAL)
  * - perform external allocation only
  * - perform internal allocation only
  */
@@ -2054,16 +2055,26 @@ template<class T> class RAMAllocator {
   using value_type = T;
 
   enum Flags {
-    NONE = 0,                 // Perform external allocation and fall back to internal memory
-    ALLOC_EXTERNAL = 1 << 0,  // Perform external allocation only.
-    ALLOC_INTERNAL = 1 << 1,  // Perform internal allocation only.
-    ALLOW_FAILURE = 1 << 2,   // Does nothing. Kept for compatibility.
+    NONE = 0,                  // Perform external allocation and fall back to internal memory
+    ALLOC_EXTERNAL = 1 << 0,   // Perform external allocation only.
+    ALLOC_INTERNAL = 1 << 1,   // Perform internal allocation only.
+    ALLOW_FAILURE = 1 << 2,    // Does nothing. Kept for compatibility.
+    PREFER_INTERNAL = 1 << 3,  // Perform internal allocation and fall back to external memory
   };
 
   constexpr RAMAllocator() = default;
-  constexpr RAMAllocator(uint8_t flags)
-      : flags_((flags & (ALLOC_INTERNAL | ALLOC_EXTERNAL)) != 0 ? (flags & (ALLOC_INTERNAL | ALLOC_EXTERNAL))
-                                                                : (ALLOC_INTERNAL | ALLOC_EXTERNAL)) {}
+  constexpr RAMAllocator(uint8_t flags) {
+    if (flags & PREFER_INTERNAL) {
+      this->flags_ = ALLOC_INTERNAL | ALLOC_EXTERNAL | PREFER_INTERNAL;
+      return;
+    }
+    const uint8_t alloc_bits = flags & (ALLOC_INTERNAL | ALLOC_EXTERNAL);
+    if (alloc_bits != 0) {
+      this->flags_ = alloc_bits;
+      return;
+    }
+    this->flags_ = ALLOC_INTERNAL | ALLOC_EXTERNAL;
+  }
   template<class U> constexpr RAMAllocator(const RAMAllocator<U> &other) : flags_{other.flags_} {}
 
   T *allocate(size_t n) { return this->allocate(n, sizeof(T)); }
@@ -2072,12 +2083,8 @@ template<class T> class RAMAllocator {
     size_t size = n * manual_size;
     T *ptr = nullptr;
 #ifdef USE_ESP32
-    if (this->flags_ & Flags::ALLOC_EXTERNAL) {
-      ptr = static_cast<T *>(heap_caps_malloc(size, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT));
-    }
-    if (ptr == nullptr && this->flags_ & Flags::ALLOC_INTERNAL) {
-      ptr = static_cast<T *>(heap_caps_malloc(size, MALLOC_CAP_INTERNAL | MALLOC_CAP_8BIT));
-    }
+    const auto caps = this->get_caps_();
+    ptr = static_cast<T *>(heap_caps_malloc_prefer(size, 2, caps[0], caps[1]));
 #else
     // Ignore ALLOC_EXTERNAL/ALLOC_INTERNAL flags if external allocation is not supported
     ptr = static_cast<T *>(malloc(size));  // NOLINT(cppcoreguidelines-owning-memory,cppcoreguidelines-no-malloc)
@@ -2091,12 +2098,8 @@ template<class T> class RAMAllocator {
     size_t size = n * manual_size;
     T *ptr = nullptr;
 #ifdef USE_ESP32
-    if (this->flags_ & Flags::ALLOC_EXTERNAL) {
-      ptr = static_cast<T *>(heap_caps_realloc(p, size, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT));
-    }
-    if (ptr == nullptr && this->flags_ & Flags::ALLOC_INTERNAL) {
-      ptr = static_cast<T *>(heap_caps_realloc(p, size, MALLOC_CAP_INTERNAL | MALLOC_CAP_8BIT));
-    }
+    const auto caps = this->get_caps_();
+    ptr = static_cast<T *>(heap_caps_realloc_prefer(p, size, 2, caps[0], caps[1]));
 #else
     // Ignore ALLOC_EXTERNAL/ALLOC_INTERNAL flags if external allocation is not supported
     ptr = static_cast<T *>(realloc(p, size));  // NOLINT(cppcoreguidelines-owning-memory,cppcoreguidelines-no-malloc)
@@ -2147,6 +2150,24 @@ template<class T> class RAMAllocator {
   }
 
  private:
+#ifdef USE_ESP32
+  /// Returns {primary_caps, fallback_caps} for heap_caps_*_prefer based on the configured flags.
+  /// PREFER_INTERNAL implies both regions are enabled (enforced by the constructor), so when it is set
+  /// the primary is internal and the fallback is external. Otherwise the primary is whichever region
+  /// is enabled (external preferred when both are enabled), and the fallback is the other region (or
+  /// the same region when only one is enabled, making the second attempt a no-op).
+  std::array<uint32_t, 2> get_caps_() const {
+    constexpr uint32_t external_caps = MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT;
+    constexpr uint32_t internal_caps = MALLOC_CAP_INTERNAL | MALLOC_CAP_8BIT;
+    if (this->flags_ & PREFER_INTERNAL) {
+      return {internal_caps, external_caps};
+    }
+    const uint32_t primary = (this->flags_ & ALLOC_EXTERNAL) ? external_caps : internal_caps;
+    const uint32_t fallback = (this->flags_ & ALLOC_INTERNAL) ? internal_caps : external_caps;
+    return {primary, fallback};
+  }
+#endif
+
   uint8_t flags_{ALLOC_INTERNAL | ALLOC_EXTERNAL};
 };
 

esphome/core/ring_buffer.cpp

@@ -1,11 +1,9 @@
 #include "ring_buffer.h"
 
-#include "esphome/core/helpers.h"
-#include "esphome/core/log.h"
-
 #ifdef USE_ESP32
 
-#include "helpers.h"
+#include "esphome/core/helpers.h"
+#include "esphome/core/log.h"
 
 namespace esphome {
 
@@ -19,12 +17,15 @@ RingBuffer::~RingBuffer() {
   }
 }
 
-std::unique_ptr<RingBuffer> RingBuffer::create(size_t len) {
+std::unique_ptr<RingBuffer> RingBuffer::create(size_t len, MemoryPreference preference) {
   std::unique_ptr<RingBuffer> rb = make_unique<RingBuffer>();
 
   rb->size_ = len;
 
-  RAMAllocator<uint8_t> allocator;
+  const uint8_t type = (preference == MemoryPreference::INTERNAL_FIRST) ? RAMAllocator<uint8_t>::PREFER_INTERNAL
+                                                                        : RAMAllocator<uint8_t>::NONE;
+
+  RAMAllocator<uint8_t> allocator(type);
   rb->storage_ = allocator.allocate(rb->size_);
   if (rb->storage_ == nullptr) {
     return nullptr;

esphome/core/ring_buffer.h

@@ -80,7 +80,12 @@ class RingBuffer {
    */
   BaseType_t reset();
 
-  static std::unique_ptr<RingBuffer> create(size_t len);
+  enum class MemoryPreference {
+    EXTERNAL_FIRST,  // External RAM preferred, fall back to internal (default)
+    INTERNAL_FIRST,  // Internal RAM preferred, fall back to external
+  };
+
+  static std::unique_ptr<RingBuffer> create(size_t len, MemoryPreference preference = MemoryPreference::EXTERNAL_FIRST);
 
  protected:
   /// @brief Discards data from the ring buffer.

esphome/espota2.py

@@ -15,6 +15,8 @@ from typing import Any
 from esphome.core import EsphomeError
 from esphome.helpers import ProgressBar, resolve_ip_address
 
+OTA_TYPE_UPDATE_APP = 0x00
+
 RESPONSE_OK = 0x00
 RESPONSE_REQUEST_AUTH = 0x01
 RESPONSE_REQUEST_SHA256_AUTH = 0x02
@@ -27,6 +29,7 @@ RESPONSE_RECEIVE_OK = 0x44
 RESPONSE_UPDATE_END_OK = 0x45
 RESPONSE_SUPPORTS_COMPRESSION = 0x46
 RESPONSE_CHUNK_OK = 0x47
+RESPONSE_FEATURE_FLAGS = 0x48
 
 RESPONSE_ERROR_MAGIC = 0x80
 RESPONSE_ERROR_UPDATE_PREPARE = 0x81
@@ -42,6 +45,7 @@ RESPONSE_ERROR_NO_UPDATE_PARTITION = 0x8A
 RESPONSE_ERROR_MD5_MISMATCH = 0x8B
 RESPONSE_ERROR_RP2040_NOT_ENOUGH_SPACE = 0x8C
 RESPONSE_ERROR_SIGNATURE_INVALID = 0x8D
+RESPONSE_ERROR_UNSUPPORTED_OTA_TYPE = 0x8E
 RESPONSE_ERROR_UNKNOWN = 0xFF
 
 OTA_VERSION_1_0 = 1
@@ -49,9 +53,16 @@ OTA_VERSION_2_0 = 2
 
 MAGIC_BYTES = [0x6C, 0x26, 0xF7, 0x5C, 0x45]
 
-FEATURE_SUPPORTS_COMPRESSION = 0x01
-FEATURE_SUPPORTS_SHA256_AUTH = 0x02
+CLIENT_FEATURE_SUPPORTS_COMPRESSION = 0x01
+CLIENT_FEATURE_SUPPORTS_SHA256_AUTH = 0x02
+CLIENT_FEATURE_SUPPORTS_EXTENDED_PROTOCOL = 0x04
+SERVER_FEATURE_SUPPORTS_COMPRESSION = 0x01
+SERVER_FEATURE_SUPPORTS_PARTITION_ACCESS = 0x02
 
+# OTA types this client knows how to send. Future PRs that add bootloader/partition
+# updates extend this set. Anything outside the set is rejected up front so callers
+# of perform_ota/run_ota get a clear error instead of a post-auth 0x8E from the device.
+_SUPPORTED_OTA_TYPES: frozenset[int] = frozenset({OTA_TYPE_UPDATE_APP})
 
 UPLOAD_BLOCK_SIZE = 8192
 UPLOAD_BUFFER_SIZE = UPLOAD_BLOCK_SIZE * 8
@@ -64,6 +75,62 @@ _AUTH_METHODS: dict[int, tuple[Callable[..., Any], int, str]] = {
     RESPONSE_REQUEST_AUTH: (hashlib.md5, 32, "MD5"),
 }
 
+# Error response code -> human-readable message (without the "Error: " prefix; check_error()
+# prepends it uniformly). Looked up by check_error() to translate a single byte from the device
+# into an OTAError. Add new error codes here rather than extending the if-chain in check_error().
+_ERROR_MESSAGES: dict[int, str] = {
+    RESPONSE_ERROR_MAGIC: "Invalid magic byte",
+    RESPONSE_ERROR_UPDATE_PREPARE: (
+        "Couldn't prepare flash memory for update. Is the binary too big? "
+        "Please try restarting the ESP."
+    ),
+    RESPONSE_ERROR_AUTH_INVALID: "Authentication invalid. Is the password correct?",
+    RESPONSE_ERROR_WRITING_FLASH: (
+        "Writing OTA data to flash memory failed. See USB logs for more information."
+    ),
+    RESPONSE_ERROR_UPDATE_END: (
+        "Finishing update failed. See the MQTT/USB logs for more information."
+    ),
+    RESPONSE_ERROR_INVALID_BOOTSTRAPPING: (
+        "Please press the reset button on the ESP. A manual reset is "
+        "required on the first OTA-Update after flashing via USB."
+    ),
+    RESPONSE_ERROR_WRONG_CURRENT_FLASH_CONFIG: (
+        "ESP has been flashed with wrong flash size. Please choose the "
+        "correct 'board' option (esp01_1m always works) and then flash over USB."
+    ),
+    RESPONSE_ERROR_WRONG_NEW_FLASH_CONFIG: (
+        "ESP does not have the requested flash size (wrong board). Please "
+        "choose the correct 'board' option (esp01_1m always works) and try "
+        "uploading again."
+    ),
+    RESPONSE_ERROR_ESP8266_NOT_ENOUGH_SPACE: (
+        "ESP does not have enough space to store OTA file. Please try "
+        "flashing a minimal firmware (remove everything except ota)"
+    ),
+    RESPONSE_ERROR_ESP32_NOT_ENOUGH_SPACE: (
+        "The OTA partition on the ESP is too small. ESPHome needs to resize "
+        "this partition, please flash over USB."
+    ),
+    RESPONSE_ERROR_NO_UPDATE_PARTITION: (
+        "The OTA partition on the ESP couldn't be found. ESPHome needs to "
+        "create this partition, please flash over USB."
+    ),
+    RESPONSE_ERROR_MD5_MISMATCH: (
+        "Application MD5 code mismatch. Please try again "
+        "or flash over USB with a good quality cable."
+    ),
+    RESPONSE_ERROR_SIGNATURE_INVALID: (
+        "Firmware signature verification failed. The firmware was not signed "
+        "with the correct key. Ensure the signing key matches the one used to build "
+        "the firmware currently running on the device."
+    ),
+    RESPONSE_ERROR_UNSUPPORTED_OTA_TYPE: (
+        "The requested OTA type is not supported by the device."
+    ),
+    RESPONSE_ERROR_UNKNOWN: "Unknown error from ESP",
+}
+
 
 class OTAError(EsphomeError):
     pass
@@ -130,8 +197,10 @@ def check_error(data: list[int] | bytes, expect: int | list[int] | None) -> None
     :param expect: Expected response code(s), None to skip validation.
     :raises OTAError: If an error code is detected or response doesn't match expected.
     """
-    if expect is None:
-        return
+    # Detect device errors and connection-closed cases regardless of `expect`. If we
+    # only ran these checks when expect was set, error bytes returned during
+    # accept-any-response reads (e.g. feature negotiation, auth nonces) would be
+    # silently passed through and surface later as cryptic decode/timeout failures.
     if not data:
         raise OTAError(
             "Error: Device closed connection without responding. "
@@ -139,69 +208,11 @@ def check_error(data: list[int] | bytes, expect: int | list[int] | None) -> None
             "a network issue, or the connection was interrupted."
         )
     dat = data[0]
-    if dat == RESPONSE_ERROR_MAGIC:
-        raise OTAError("Error: Invalid magic byte")
-    if dat == RESPONSE_ERROR_UPDATE_PREPARE:
-        raise OTAError(
-            "Error: Couldn't prepare flash memory for update. Is the binary too big? "
-            "Please try restarting the ESP."
-        )
-    if dat == RESPONSE_ERROR_AUTH_INVALID:
-        raise OTAError("Error: Authentication invalid. Is the password correct?")
-    if dat == RESPONSE_ERROR_WRITING_FLASH:
-        raise OTAError(
-            "Error: Writing OTA data to flash memory failed. See USB logs for more "
-            "information."
-        )
-    if dat == RESPONSE_ERROR_UPDATE_END:
-        raise OTAError(
-            "Error: Finishing update failed. See the MQTT/USB logs for more "
-            "information."
-        )
-    if dat == RESPONSE_ERROR_INVALID_BOOTSTRAPPING:
-        raise OTAError(
-            "Error: Please press the reset button on the ESP. A manual reset is "
-            "required on the first OTA-Update after flashing via USB."
-        )
-    if dat == RESPONSE_ERROR_WRONG_CURRENT_FLASH_CONFIG:
-        raise OTAError(
-            "Error: ESP has been flashed with wrong flash size. Please choose the "
-            "correct 'board' option (esp01_1m always works) and then flash over USB."
-        )
-    if dat == RESPONSE_ERROR_WRONG_NEW_FLASH_CONFIG:
-        raise OTAError(
-            "Error: ESP does not have the requested flash size (wrong board). Please "
-            "choose the correct 'board' option (esp01_1m always works) and try "
-            "uploading again."
-        )
-    if dat == RESPONSE_ERROR_ESP8266_NOT_ENOUGH_SPACE:
-        raise OTAError(
-            "Error: ESP does not have enough space to store OTA file. Please try "
-            "flashing a minimal firmware (remove everything except ota)"
-        )
-    if dat == RESPONSE_ERROR_ESP32_NOT_ENOUGH_SPACE:
-        raise OTAError(
-            "Error: The OTA partition on the ESP is too small. ESPHome needs to resize "
-            "this partition, please flash over USB."
-        )
-    if dat == RESPONSE_ERROR_NO_UPDATE_PARTITION:
-        raise OTAError(
-            "Error: The OTA partition on the ESP couldn't be found. ESPHome needs to create "
-            "this partition, please flash over USB."
-        )
-    if dat == RESPONSE_ERROR_MD5_MISMATCH:
-        raise OTAError(
-            "Error: Application MD5 code mismatch. Please try again "
-            "or flash over USB with a good quality cable."
-        )
-    if dat == RESPONSE_ERROR_SIGNATURE_INVALID:
-        raise OTAError(
-            "Error: Firmware signature verification failed. The firmware was not signed "
-            "with the correct key. Ensure the signing key matches the one used to build "
-            "the firmware currently running on the device."
-        )
-    if dat == RESPONSE_ERROR_UNKNOWN:
-        raise OTAError("Unknown error from ESP")
+    error_msg = _ERROR_MESSAGES.get(dat)
+    if error_msg is not None:
+        raise OTAError(f"Error: {error_msg}")
+    if expect is None:
+        return
     if not isinstance(expect, (list, tuple)):
         expect = [expect]
     if dat not in expect:
@@ -232,8 +243,25 @@ def send_check(
 
 
 def perform_ota(
-    sock: socket.socket, password: str | None, file_handle: io.IOBase, filename: Path
+    sock: socket.socket,
+    password: str | None,
+    file_handle: io.IOBase,
+    filename: Path,
+    ota_type: int = OTA_TYPE_UPDATE_APP,
 ) -> None:
+    # Validate ota_type up front. It travels as a single byte on the wire, and
+    # passing an out-of-range value would only surface as a ValueError from
+    # bytes([ota_type]) deep inside send_check, bypassing OTAError handling.
+    if not isinstance(ota_type, int) or not 0 <= ota_type <= 0xFF:
+        raise OTAError(
+            f"Invalid ota_type {ota_type!r}; expected an integer in range 0-255"
+        )
+    if ota_type not in _SUPPORTED_OTA_TYPES:
+        supported = ", ".join(f"0x{t:02X}" for t in sorted(_SUPPORTED_OTA_TYPES))
+        raise OTAError(
+            f"Unsupported OTA type 0x{ota_type:02X}; this ESPHome supports: {supported}"
+        )
+
     file_contents = file_handle.read()
     file_size = len(file_contents)
     _LOGGER.info("Uploading %s (%s bytes)", filename, file_size)
@@ -251,7 +279,11 @@ def perform_ota(
         )
 
     # Features - send both compression and SHA256 auth support
-    features_to_send = FEATURE_SUPPORTS_COMPRESSION | FEATURE_SUPPORTS_SHA256_AUTH
+    features_to_send = (
+        CLIENT_FEATURE_SUPPORTS_COMPRESSION
+        | CLIENT_FEATURE_SUPPORTS_SHA256_AUTH
+        | CLIENT_FEATURE_SUPPORTS_EXTENDED_PROTOCOL
+    )
     send_check(sock, features_to_send, "features")
     features = receive_exactly(
         sock,
@@ -260,7 +292,36 @@ def perform_ota(
         None,  # Accept any response
     )[0]
 
-    if features == RESPONSE_SUPPORTS_COMPRESSION:
+    extended_proto = False
+    if features == RESPONSE_FEATURE_FLAGS:
+        extended_proto = True
+        features = receive_exactly(
+            sock,
+            1,
+            "feature flags",
+            None,  # Accept any response
+        )[0]
+    elif features == RESPONSE_SUPPORTS_COMPRESSION:
+        features = SERVER_FEATURE_SUPPORTS_COMPRESSION
+    else:
+        features = 0
+
+    if ota_type != OTA_TYPE_UPDATE_APP:
+        # Any non-app OTA type requires the extended protocol and the
+        # partition-access server feature. Reject up front so the user gets
+        # a clear capability error instead of a post-auth 0x8E from the device.
+        if not extended_proto:
+            raise OTAError(
+                f"Device does not support extended OTA protocol; "
+                f"OTA type 0x{ota_type:02X} requires it"
+            )
+        if not (features & SERVER_FEATURE_SUPPORTS_PARTITION_ACCESS):
+            raise OTAError(
+                f"Device does not support partition access; "
+                f"OTA type 0x{ota_type:02X} cannot be used"
+            )
+
+    if features & SERVER_FEATURE_SUPPORTS_COMPRESSION:
         upload_contents = gzip.compress(file_contents, compresslevel=9)
         _LOGGER.info("Compressed to %s bytes", len(upload_contents))
     else:
@@ -315,6 +376,9 @@ def perform_ota(
     # Timeout must match device-side OTA_SOCKET_TIMEOUT_DATA to prevent premature failures
     sock.settimeout(90.0)
 
+    if extended_proto:
+        send_check(sock, ota_type, "ota type")
+
     upload_size = len(upload_contents)
     upload_size_encoded = [
         (upload_size >> 24) & 0xFF,
@@ -375,7 +439,11 @@ def perform_ota(
 
 
 def run_ota_impl_(
-    remote_host: str | list[str], remote_port: int, password: str | None, filename: Path
+    remote_host: str | list[str],
+    remote_port: int,
+    password: str | None,
+    filename: Path,
+    ota_type: int = OTA_TYPE_UPDATE_APP,
 ) -> tuple[int, str | None]:
     from esphome.core import CORE
 
@@ -413,7 +481,7 @@ def run_ota_impl_(
         _LOGGER.info("Connected to %s", sa[0])
         with open(filename, "rb") as file_handle:
             try:
-                perform_ota(sock, password, file_handle, filename)
+                perform_ota(sock, password, file_handle, filename, ota_type)
             except OTAError as err:
                 _LOGGER.error(str(err))
                 return 1, None
@@ -428,10 +496,14 @@ def run_ota_impl_(
 
 
 def run_ota(
-    remote_host: str | list[str], remote_port: int, password: str | None, filename: Path
+    remote_host: str | list[str],
+    remote_port: int,
+    password: str | None,
+    filename: Path,
+    ota_type: int = OTA_TYPE_UPDATE_APP,
 ) -> tuple[int, str | None]:
     try:
-        return run_ota_impl_(remote_host, remote_port, password, filename)
+        return run_ota_impl_(remote_host, remote_port, password, filename, ota_type)
     except OTAError as err:
         _LOGGER.error(err)
         return 1, None

esphome/idf_component.yml

@@ -10,7 +10,7 @@ dependencies:
   esphome/micro-flac:
     version: 0.1.1
   esphome/micro-opus:
-    version: 0.3.6
+    version: 0.4.0
   espressif/esp-dsp:
     version: "1.7.1"
   espressif/esp-tflite-micro:
@@ -20,15 +20,19 @@ dependencies:
   espressif/mdns:
     version: 1.11.0
   espressif/esp_wifi_remote:
-    version: 1.4.0
+    version: 1.5.1
+    rules:
+      - if: "target in [esp32h2, esp32p4]"
+  espressif/wifi_remote_over_eppp:
+    version: 0.3.2
     rules:
       - if: "target in [esp32h2, esp32p4]"
   espressif/eppp_link:
-    version: 1.1.4
+    version: 1.1.5
     rules:
       - if: "target in [esp32h2, esp32p4]"
   espressif/esp_hosted:
-    version: 2.12.1
+    version: 2.12.6
     rules:
       - if: "target in [esp32h2, esp32p4]"
   zorxx/multipart-parser:
@@ -92,6 +96,6 @@ dependencies:
   esp32async/asynctcp:
     version: 3.4.91
   sendspin/sendspin-cpp:
-    version: 0.3.1
+    version: 0.4.0
   lvgl/lvgl:
     version: 9.5.0

esphome/platformio_api.py

@@ -14,6 +14,37 @@ from esphome.util import run_external_process
 _LOGGER = logging.getLogger(__name__)
 
 
+def _strip_win_long_path_prefix(path: str) -> str:
+    r"""Strip the Windows extended-length path prefix from ``path``.
+
+    Handles both forms documented at
+    https://learn.microsoft.com/windows/win32/fileio/naming-a-file:
+
+    * ``\\?\C:\path\to\file`` -> ``C:\path\to\file``
+    * ``\\?\UNC\server\share\path`` -> ``\\server\share\path``
+
+    The NSIS-installed ``esphome.exe`` launcher on Windows starts Python with
+    ``sys.executable`` already prefixed with ``\\?\``. That prefix propagates
+    into PlatformIO's ``$PYTHONEXE`` (PlatformIO reads ``PYTHONEXEPATH`` from
+    the environment, falling back to ``os.path.normpath(sys.executable)``)
+    and ends up baked into SCons-emitted command lines for build steps such
+    as the esp8266 ``elf2bin`` invocation. ``cmd.exe`` does not understand
+    the ``\\?\`` prefix, so the build fails with
+    "The system cannot find the path specified." Stripping the prefix early
+    keeps the path shell-quotable.
+
+    No-op on non-Windows platforms.
+    """
+    if sys.platform != "win32":
+        return path
+    if path.startswith("\\\\?\\UNC\\"):
+        # \\?\UNC\server\share\... -> \\server\share\...
+        return "\\\\" + path[len("\\\\?\\UNC\\") :]
+    if path.startswith("\\\\?\\"):
+        return path[len("\\\\?\\") :]
+    return path
+
+
 def run_platformio_cli(*args, **kwargs) -> str | int:
     os.environ["PLATFORMIO_FORCE_COLOR"] = "true"
     os.environ["PLATFORMIO_BUILD_DIR"] = str(CORE.relative_pioenvs_path().absolute())
@@ -24,7 +55,18 @@ def run_platformio_cli(*args, **kwargs) -> str | int:
     os.environ.setdefault("PYTHONWARNINGS", "ignore::SyntaxWarning")
     # Increase uv retry count to handle transient network errors (default is 3)
     os.environ.setdefault("UV_HTTP_RETRIES", "10")
-    cmd = [sys.executable, "-m", "esphome.platformio_runner"] + list(args)
+    # Strip the Windows extended-length path prefix from sys.executable so it
+    # doesn't propagate into PlatformIO's $PYTHONEXE and break SCons-emitted
+    # command lines run through cmd.exe.
+    python_exe = _strip_win_long_path_prefix(sys.executable)
+    if python_exe != sys.executable:
+        # Only override PYTHONEXEPATH when we actually stripped a prefix.
+        # PlatformIO's get_pythonexe_path() reads this and falls back to
+        # sys.executable otherwise; setting it unconditionally would clobber
+        # a user-provided value (or the unmodified path on platforms that
+        # don't need the strip).
+        os.environ["PYTHONEXEPATH"] = python_exe
+    cmd = [python_exe, "-m", "esphome.platformio_runner"] + list(args)
 
     return run_external_process(*cmd, **kwargs)
 

esphome/platformio_runner.py

@@ -101,7 +101,7 @@ def patch_file_downloader() -> None:
     FileDownloader.__init__ = patched_init
 
 
-_IGNORE_LIB_WARNINGS = f"(?:{'|'.join(['Hash', 'Update'])})"
+_IGNORE_LIB_WARNINGS = "(?:Hash|Update)"
 # Regex patterns matched against each line of PlatformIO output. Lines that
 # match are dropped by RedirectText before they reach the parent process.
 # Patterns are anchored at the start of the line (RedirectText uses

pyproject.toml

@@ -113,6 +113,7 @@ exclude = ['generated']
 select = [
   "E",    # pycodestyle
   "F",    # pyflakes/autoflake
+  "FLY",  # flynt: convert string formatting to f-strings
   "FURB", # refurb
   "I",    # isort
   "PERF", # performance

requirements.txt

@@ -12,7 +12,7 @@ platformio==6.1.19
 esptool==5.2.0
 click==8.3.3
 esphome-dashboard==20260425.0
-aioesphomeapi==44.22.0
+aioesphomeapi==44.23.0
 zeroconf==0.148.0
 puremagic==1.30
 ruamel.yaml==0.19.1 # dashboard_import

script/determine-jobs.py

@@ -6,11 +6,11 @@ what files have changed. It outputs JSON with the following structure:
 
 {
   "integration_tests": true/false,
-  "integration_tests_run_all": true/false,
-  "integration_test_files": ["tests/integration/test_foo.py", ...],
+  "integration_test_buckets": [{"name": "1/3", "tests": ["tests/integration/test_foo.py", ...]}, ...],
   "clang_tidy": true/false,
   "clang_format": true/false,
   "python_linters": true/false,
+  "device_builder": true/false,
   "changed_components": ["component1", "component2", ...],
   "component_test_count": 5,
   "memory_impact": {
@@ -26,6 +26,7 @@ The CI workflow uses this information to:
 - Skip or run clang-tidy (and whether to do a full scan)
 - Skip or run clang-format
 - Skip or run Python linters (ruff, flake8, pylint, pyupgrade)
+- Skip or run downstream esphome/device-builder tests against the PR's Python code
 - Determine which components to test individually
 - Decide how to split component tests (if there are many)
 - Run memory impact analysis whenever there are changed components (merged config), and also for core-only changes
@@ -81,6 +82,62 @@ CLANG_TIDY_SPLIT_THRESHOLD = 65
 # Isolated components count as 10x, groupable components count as 1x
 COMPONENT_TEST_BATCH_SIZE = 40
 
+# Integration test bucketing: when more than the threshold tests are scheduled,
+# fan out across this many parallel jobs. Below the threshold, a single job runs.
+INTEGRATION_TESTS_SPLIT_THRESHOLD = 10
+INTEGRATION_TESTS_SPLIT_BUCKETS = 3
+
+
+def _split_list(items: list[str], n: int) -> list[list[str]]:
+    """Split a list into n roughly-equal contiguous parts (matches script/clang-tidy)."""
+    k, m = divmod(len(items), n)
+    return [items[i * k + min(i, m) : (i + 1) * k + min(i + 1, m)] for i in range(n)]
+
+
+def _all_integration_test_files() -> list[str]:
+    """Return all integration test file paths, sorted, relative to repo root."""
+    return sorted(
+        str(p.relative_to(root_path))
+        for p in (Path(root_path) / "tests" / "integration").glob("test_*.py")
+    )
+
+
+def _compute_integration_test_buckets(
+    integration_run_all: bool,
+    integration_test_files: list[str],
+) -> tuple[bool, list[dict[str, Any]]]:
+    """Compute (run_integration, buckets) from the determine_integration_tests result.
+
+    Pure function for unit testing — no I/O beyond `_all_integration_test_files`
+    when `integration_run_all` is set.
+
+    `buckets` is a list of `{name, tests}` dicts where `tests` is a JSON-friendly
+    list of file paths so the workflow can build a bash array via jq, avoiding
+    shell word-splitting / glob hazards.
+    """
+    if integration_run_all:
+        files = _all_integration_test_files()
+    else:
+        files = sorted(integration_test_files)
+
+    # Empty list (e.g. run_all expansion with no files on disk) would otherwise
+    # cause the workflow to invoke pytest with no path argument and collect
+    # tests outside tests/integration/. Suppress the run instead.
+    if not files:
+        return False, []
+
+    if len(files) > INTEGRATION_TESTS_SPLIT_THRESHOLD:
+        parts = [
+            part for part in _split_list(files, INTEGRATION_TESTS_SPLIT_BUCKETS) if part
+        ]
+        buckets = [
+            {"name": f"{i + 1}/{len(parts)}", "tests": part}
+            for i, part in enumerate(parts)
+        ]
+    else:
+        buckets = [{"name": "1/1", "tests": files}]
+    return True, buckets
+
 
 class Platform(StrEnum):
     """Platform identifiers for memory impact analysis."""
@@ -385,6 +442,56 @@ def should_run_import_time(branch: str | None = None) -> bool:
     return False
 
 
+# Files outside esphome/**/*.py whose changes can affect the downstream
+# device-builder build. requirements.txt / pyproject.toml change the runtime
+# dependency graph that device-builder picks up when it installs esphome.
+DEVICE_BUILDER_TRIGGER_FILES = frozenset(
+    {
+        "requirements.txt",
+        "pyproject.toml",
+    }
+)
+
+
+def should_run_device_builder(branch: str | None = None) -> bool:
+    """Determine if downstream esphome/device-builder tests should run.
+
+    device-builder imports esphome as a library, so whenever the importable
+    Python surface, the runtime dependencies, or any non-C++ file packaged
+    with esphome (pyproject.toml has ``include-package-data = true``, so
+    things like esphome/idf_component.yml ship and can affect installs)
+    changes we re-run its test suite against the PR's code to catch
+    breakage we'd otherwise only see after a release.
+
+    Skipped on beta/release branches: those branches typically lag behind
+    device-builder@main, so a new device-builder API dependency would
+    falsely fail the run without reflecting any problem in the PR itself.
+
+    Args:
+        branch: Branch to compare against. If None, uses default.
+
+    Returns:
+        True if the device-builder downstream tests should run, False otherwise.
+    """
+    target_branch = get_target_branch()
+    if target_branch and (
+        target_branch.startswith("release") or target_branch.startswith("beta")
+    ):
+        return False
+
+    for file in changed_files(branch):
+        if file in DEVICE_BUILDER_TRIGGER_FILES:
+            return True
+        # Anything under esphome/ that isn't C++ source can change the
+        # importable / packaged surface device-builder consumes
+        # (Python sources, packaged YAML/JSON like idf_component.yml,
+        # etc.). C++ files only affect compiled firmware, not the
+        # Python install device-builder pulls in.
+        if file.startswith("esphome/") and not file.endswith(CPP_FILE_EXTENSIONS):
+            return True
+    return False
+
+
 def determine_cpp_unit_tests(
     branch: str | None = None,
 ) -> tuple[bool, list[str]]:
@@ -812,11 +919,14 @@ def main() -> None:
     integration_run_all, integration_test_files = determine_integration_tests(
         args.branch
     )
-    run_integration = integration_run_all or bool(integration_test_files)
+    run_integration, integration_test_buckets = _compute_integration_test_buckets(
+        integration_run_all, integration_test_files
+    )
     run_clang_tidy = should_run_clang_tidy(args.branch)
     run_clang_format = should_run_clang_format(args.branch)
     run_python_linters = should_run_python_linters(args.branch)
     run_import_time = should_run_import_time(args.branch)
+    run_device_builder = should_run_device_builder(args.branch)
     changed_cpp_file_count = count_changed_cpp_files(args.branch)
 
     # Get changed components
@@ -944,13 +1054,13 @@ def main() -> None:
 
     output: dict[str, Any] = {
         "integration_tests": run_integration,
-        "integration_tests_run_all": integration_run_all,
-        "integration_test_files": integration_test_files,
+        "integration_test_buckets": integration_test_buckets,
         "clang_tidy": run_clang_tidy,
         "clang_tidy_mode": clang_tidy_mode,
         "clang_format": run_clang_format,
         "python_linters": run_python_linters,
         "import_time": run_import_time,
+        "device_builder": run_device_builder,
         "changed_components": changed_components,
         "changed_components_with_tests": changed_components_with_tests,
         "directly_changed_components_with_tests": list(directly_changed_with_tests),

tests/benchmarks/components/api/__init__.py

@@ -11,11 +11,19 @@ def override_manifest(manifest: ComponentManifestOverride) -> None:
 
     async def to_code(config):
         await original_to_code(config)
-        # Enable BLE proto message types for benchmarks.  The real
-        # bluetooth_proxy component is ESP32-only; a lightweight stub
-        # header in tests/benchmarks/stubs/ satisfies the include.
+        # Enable proxy proto message types for benchmarks.  The real
+        # components have hardware dependencies (BLE/UART/RMT); lightweight
+        # stub headers in tests/benchmarks/stubs/ satisfy the includes.
         cg.add_define("USE_BLUETOOTH_PROXY")
         cg.add_define("BLUETOOTH_PROXY_MAX_CONNECTIONS", 3)
         cg.add_define("BLUETOOTH_PROXY_ADVERTISEMENT_BATCH_SIZE", 16)
+        cg.add_define("USE_ZWAVE_PROXY")
+        cg.add_define("USE_INFRARED")
+        cg.add_define("USE_IR_RF")
+        cg.add_define("USE_RADIO_FREQUENCY")
+        cg.add_define("USE_SERIAL_PROXY")
+        cg.add_define("SERIAL_PROXY_COUNT", 0)
+        cg.add_define("ESPHOME_ENTITY_INFRARED_COUNT", 0)
+        cg.add_define("ESPHOME_ENTITY_RADIO_FREQUENCY_COUNT", 0)
 
     manifest.to_code = to_code

tests/benchmarks/components/api/bench_proto_proxy.cpp

@@ -0,0 +1,280 @@
+// Encode/decode microbenchmarks for proxy message families that carry
+// high-volume traffic (Z-Wave, IR/RF, serial). Mirrors the existing
+// BluetoothLERawAdvertisementsResponse benchmarks in bench_proto_encode.cpp.
+
+#include <benchmark/benchmark.h>
+
+#include <cstring>
+
+#include "esphome/components/api/api_pb2.h"
+#include "esphome/components/api/api_buffer.h"
+
+namespace esphome::api::benchmarks {
+
+static constexpr int kInnerIterations = 2000;
+
+// Encodes `src` into `out`. Caller owns `out` and must keep it alive across
+// the decode loop (decoded messages may store pointers back into its bytes).
+template<typename T> static void encode_into(APIBuffer &out, const T &src) {
+  out.resize(src.calculate_size());
+  ProtoWriteBuffer writer(&out, 0);
+  src.encode(writer);
+}
+
+// --- ZWaveProxyFrame (Z-Wave frame, ~16 bytes payload) ---
+
+#ifdef USE_ZWAVE_PROXY
+
+static const uint8_t kZWaveFrameData[] = {0x01, 0x09, 0x00, 0x13, 0x01, 0x02, 0x00, 0x00,
+                                          0x25, 0x00, 0x05, 0xC4, 0x00, 0x00, 0x00, 0x00};
+
+static void Encode_ZWaveProxyFrame(benchmark::State &state) {
+  ZWaveProxyFrame msg;
+  msg.data = kZWaveFrameData;
+  msg.data_len = sizeof(kZWaveFrameData);
+  APIBuffer buffer;
+  buffer.resize(msg.calculate_size());
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_ZWaveProxyFrame);
+
+static void Decode_ZWaveProxyFrame(benchmark::State &state) {
+  ZWaveProxyFrame source;
+  source.data = kZWaveFrameData;
+  source.data_len = sizeof(kZWaveFrameData);
+  APIBuffer encoded;
+  encode_into(encoded, source);
+  const uint8_t *data = encoded.data();
+  size_t size = encoded.size();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ZWaveProxyFrame msg;
+      msg.decode(data, size);
+      benchmark::DoNotOptimize(msg);
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Decode_ZWaveProxyFrame);
+
+static const uint8_t kZWaveRequestData[] = {0xDE, 0xAD, 0xBE, 0xEF};
+
+static void Decode_ZWaveProxyRequest(benchmark::State &state) {
+  ZWaveProxyRequest source;
+  source.type = enums::ZWAVE_PROXY_REQUEST_TYPE_HOME_ID_CHANGE;
+  source.data = kZWaveRequestData;
+  source.data_len = sizeof(kZWaveRequestData);
+  APIBuffer encoded;
+  encode_into(encoded, source);
+  const uint8_t *data = encoded.data();
+  size_t size = encoded.size();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ZWaveProxyRequest msg;
+      msg.decode(data, size);
+      benchmark::DoNotOptimize(msg);
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Decode_ZWaveProxyRequest);
+
+#endif  // USE_ZWAVE_PROXY
+
+// --- SerialProxyDataReceived encode + SerialProxyWriteRequest decode ---
+//
+// SerialProxyWriteRequest is decode-only (SOURCE_CLIENT) but has the same
+// wire layout as SerialProxyDataReceived, so we encode via the latter and
+// decode as the former.
+
+#ifdef USE_SERIAL_PROXY
+
+static constexpr size_t kSerialPayloadSize = 64;
+static const uint8_t kSerialPayload[kSerialPayloadSize] = {
+    0x55, 0xAA, 0x12, 0x34, 0x56, 0x78, 0x9A, 0xBC, 0xDE, 0xF0, 0x01, 0x23, 0x45, 0x67, 0x89, 0xAB,
+    0xCD, 0xEF, 0x11, 0x22, 0x33, 0x44, 0x55, 0x66, 0x77, 0x88, 0x99, 0xAA, 0xBB, 0xCC, 0xDD, 0xEE,
+    0xFF, 0x00, 0x10, 0x20, 0x30, 0x40, 0x50, 0x60, 0x70, 0x80, 0x90, 0xA0, 0xB0, 0xC0, 0xD0, 0xE0,
+    0xF0, 0x0F, 0x1F, 0x2F, 0x3F, 0x4F, 0x5F, 0x6F, 0x7F, 0x8F, 0x9F, 0xAF, 0xBF, 0xCF, 0xDF, 0xEF};
+
+static void Encode_SerialProxyDataReceived(benchmark::State &state) {
+  SerialProxyDataReceived msg;
+  msg.instance = 0;
+  msg.set_data(kSerialPayload, kSerialPayloadSize);
+  APIBuffer buffer;
+  buffer.resize(msg.calculate_size());
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_SerialProxyDataReceived);
+
+static void Decode_SerialProxyWriteRequest(benchmark::State &state) {
+  SerialProxyDataReceived source;
+  source.instance = 0;
+  source.set_data(kSerialPayload, kSerialPayloadSize);
+  APIBuffer encoded;
+  encode_into(encoded, source);
+  const uint8_t *data = encoded.data();
+  size_t size = encoded.size();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      SerialProxyWriteRequest msg;
+      msg.decode(data, size);
+      benchmark::DoNotOptimize(msg);
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Decode_SerialProxyWriteRequest);
+
+#endif  // USE_SERIAL_PROXY
+
+// --- InfraredRFReceiveEvent encode (100 sint32 timings) +
+//     InfraredRFTransmitRawTimingsRequest decode (hand-built wire bytes) ---
+
+#if defined(USE_IR_RF) || defined(USE_RADIO_FREQUENCY)
+
+// Mark/space pairs simulating a typical RC-5 / NEC capture (100 timings).
+static std::vector<int32_t> make_ir_timings_100() {
+  std::vector<int32_t> v;
+  v.reserve(100);
+  for (int i = 0; i < 100; i++) {
+    v.push_back((i % 2 == 0) ? 560 : -560);
+  }
+  return v;
+}
+
+static const std::vector<int32_t> &get_ir_timings_100() {
+  static const std::vector<int32_t> timings = make_ir_timings_100();
+  return timings;
+}
+
+static void Encode_InfraredRFReceiveEvent(benchmark::State &state) {
+  InfraredRFReceiveEvent msg;
+  msg.key = 0xDEADBEEF;
+  msg.timings = &get_ir_timings_100();
+  APIBuffer buffer;
+  buffer.resize(msg.calculate_size());
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_InfraredRFReceiveEvent);
+
+static void CalculateSize_InfraredRFReceiveEvent(benchmark::State &state) {
+  InfraredRFReceiveEvent msg;
+  msg.key = 0xDEADBEEF;
+  msg.timings = &get_ir_timings_100();
+
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result += msg.calculate_size();
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalculateSize_InfraredRFReceiveEvent);
+
+// Hand-built wire bytes for InfraredRFTransmitRawTimingsRequest (decode-only,
+// no sister message with identical wire layout).
+//   field 2 (key, fixed32):           tag=0x15, 4 LE bytes
+//   field 3 (carrier_frequency):      tag=0x18, varint
+//   field 4 (repeat_count):           tag=0x20, varint
+//   field 5 (timings, packed sint32): tag=0x2A, length varint, packed payload
+//   field 6 (modulation):             tag=0x30, varint
+static APIBuffer build_infrared_rf_transmit_wire() {
+  uint8_t bytes[256];
+  size_t len = 0;
+
+  auto put_byte = [&](uint8_t b) { bytes[len++] = b; };
+  auto put_varint = [&](uint32_t v) {
+    while (v >= 0x80) {
+      bytes[len++] = static_cast<uint8_t>((v & 0x7F) | 0x80);
+      v >>= 7;
+    }
+    bytes[len++] = static_cast<uint8_t>(v);
+  };
+  auto encode_zigzag = [](int32_t v) -> uint32_t {
+    return (static_cast<uint32_t>(v) << 1) ^ static_cast<uint32_t>(v >> 31);
+  };
+
+  put_byte(0x15);
+  put_byte(0xEF);
+  put_byte(0xBE);
+  put_byte(0xAD);
+  put_byte(0xDE);
+  put_byte(0x18);
+  put_varint(38000);
+  put_byte(0x20);
+  put_varint(2);
+
+  uint8_t packed[200];
+  size_t packed_len = 0;
+  for (int i = 0; i < 100; i++) {
+    int32_t value = (i % 2 == 0) ? 560 : -560;
+    uint32_t zz = encode_zigzag(value);
+    while (zz >= 0x80) {
+      packed[packed_len++] = static_cast<uint8_t>((zz & 0x7F) | 0x80);
+      zz >>= 7;
+    }
+    packed[packed_len++] = static_cast<uint8_t>(zz);
+  }
+  put_byte(0x2A);
+  put_varint(static_cast<uint32_t>(packed_len));
+  std::memcpy(bytes + len, packed, packed_len);
+  len += packed_len;
+  // field 6: modulation = 1 (non-zero so it's actually emitted and exercises
+  // decode_varint for this field, matching the documented layout above).
+  put_byte(0x30);
+  put_varint(1);
+
+  APIBuffer buf;
+  buf.resize(len);
+  std::memcpy(buf.data(), bytes, len);
+  return buf;
+}
+
+static void Decode_InfraredRFTransmitRawTimingsRequest(benchmark::State &state) {
+  auto encoded = build_infrared_rf_transmit_wire();
+  const uint8_t *data = encoded.data();
+  size_t size = encoded.size();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      InfraredRFTransmitRawTimingsRequest msg;
+      msg.decode(data, size);
+      benchmark::DoNotOptimize(msg);
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Decode_InfraredRFTransmitRawTimingsRequest);
+
+#endif  // USE_IR_RF || USE_RADIO_FREQUENCY
+
+}  // namespace esphome::api::benchmarks

tests/benchmarks/stubs/esphome/components/infrared/infrared.h

@@ -0,0 +1,45 @@
+// Stub for benchmark builds — provides the minimal interface that
+// api_connection.cpp and Application need when USE_INFRARED is defined,
+// without pulling in the real remote_base/RMT dependencies.
+#pragma once
+
+#include <cstdint>
+#include "esphome/core/component.h"
+#include "esphome/core/entity_base.h"
+
+namespace esphome::infrared {
+
+class Infrared;
+
+class InfraredCall {
+ public:
+  explicit InfraredCall(Infrared *parent) : parent_(parent) {}
+  InfraredCall &set_carrier_frequency(uint32_t /*frequency*/) { return *this; }
+  InfraredCall &set_raw_timings_packed(const uint8_t * /*data*/, uint16_t /*length*/, uint16_t /*count*/) {
+    return *this;
+  }
+  InfraredCall &set_repeat_count(uint32_t /*count*/) { return *this; }
+  void perform() {}
+
+ protected:
+  Infrared *parent_;
+};
+
+class InfraredTraits {
+ public:
+  uint32_t get_receiver_frequency_hz() const { return 0; }
+};
+
+class Infrared : public Component, public EntityBase {
+ public:
+  Infrared() = default;
+  InfraredTraits &get_traits() { return this->traits_; }
+  const InfraredTraits &get_traits() const { return this->traits_; }
+  InfraredCall make_call() { return InfraredCall(this); }
+  uint32_t get_capability_flags() const { return 0; }
+
+ protected:
+  InfraredTraits traits_;
+};
+
+}  // namespace esphome::infrared

tests/benchmarks/stubs/esphome/components/radio_frequency/radio_frequency.h

@@ -0,0 +1,51 @@
+// Stub for benchmark builds — provides the minimal interface that
+// api_connection.cpp and Application need when USE_RADIO_FREQUENCY is defined.
+#pragma once
+
+#include <cstdint>
+#include "esphome/core/component.h"
+#include "esphome/core/entity_base.h"
+
+namespace esphome::radio_frequency {
+
+enum RadioFrequencyModulation : uint32_t {
+  RADIO_FREQUENCY_MODULATION_OOK = 0,
+};
+
+class RadioFrequency;
+
+class RadioFrequencyCall {
+ public:
+  explicit RadioFrequencyCall(RadioFrequency *parent) : parent_(parent) {}
+  RadioFrequencyCall &set_frequency(uint32_t /*frequency*/) { return *this; }
+  RadioFrequencyCall &set_modulation(RadioFrequencyModulation /*mod*/) { return *this; }
+  RadioFrequencyCall &set_repeat_count(uint32_t /*count*/) { return *this; }
+  RadioFrequencyCall &set_raw_timings_packed(const uint8_t * /*data*/, uint16_t /*length*/, uint16_t /*count*/) {
+    return *this;
+  }
+  void perform() {}
+
+ protected:
+  RadioFrequency *parent_;
+};
+
+class RadioFrequencyTraits {
+ public:
+  uint32_t get_frequency_min_hz() const { return 0; }
+  uint32_t get_frequency_max_hz() const { return 0; }
+  uint32_t get_supported_modulations() const { return 0; }
+};
+
+class RadioFrequency : public Component, public EntityBase {
+ public:
+  RadioFrequency() = default;
+  RadioFrequencyTraits &get_traits() { return this->traits_; }
+  const RadioFrequencyTraits &get_traits() const { return this->traits_; }
+  RadioFrequencyCall make_call() { return RadioFrequencyCall(this); }
+  uint32_t get_capability_flags() const { return 0; }
+
+ protected:
+  RadioFrequencyTraits traits_;
+};
+
+}  // namespace esphome::radio_frequency

tests/benchmarks/stubs/esphome/components/serial_proxy/serial_proxy.h

@@ -0,0 +1,46 @@
+// Stub for benchmark builds — provides the minimal interface that
+// api_connection.cpp and Application need when USE_SERIAL_PROXY is defined,
+// without pulling in the real UART implementation.
+#pragma once
+
+#include <cstdint>
+#include <cstddef>
+#include "esphome/components/api/api_pb2.h"
+
+namespace esphome {
+
+namespace api {
+class APIConnection;
+}  // namespace api
+
+namespace uart {
+enum class UARTFlushResult : uint8_t {
+  UART_FLUSH_RESULT_SUCCESS,
+  UART_FLUSH_RESULT_ASSUMED_SUCCESS,
+  UART_FLUSH_RESULT_TIMEOUT,
+  UART_FLUSH_RESULT_FAILED,
+};
+}  // namespace uart
+
+namespace serial_proxy {
+
+class SerialProxy {
+ public:
+  void set_instance_index(uint32_t index) { this->instance_index_ = index; }
+  uint32_t get_instance_index() const { return this->instance_index_; }
+  const char *get_name() const { return ""; }
+  api::enums::SerialProxyPortType get_port_type() const { return {}; }
+  api::APIConnection *get_api_connection() { return nullptr; }
+  void serial_proxy_request(api::APIConnection *conn, api::enums::SerialProxyRequestType type) {}
+  void configure(uint32_t baudrate, bool flow_control, uint8_t parity, uint32_t stop_bits, uint32_t data_size) {}
+  void write_from_client(const uint8_t *data, size_t len) {}
+  void set_modem_pins(uint32_t line_states) {}
+  uint32_t get_modem_pins() const { return 0; }
+  uart::UARTFlushResult flush_port() { return uart::UARTFlushResult::UART_FLUSH_RESULT_SUCCESS; }
+
+ protected:
+  uint32_t instance_index_{0};
+};
+
+}  // namespace serial_proxy
+}  // namespace esphome

tests/benchmarks/stubs/esphome/components/zwave_proxy/zwave_proxy.h

@@ -0,0 +1,29 @@
+// Stub for benchmark builds — provides the minimal interface that
+// api_connection.cpp needs when USE_ZWAVE_PROXY is defined,
+// without pulling in the real UART-based ZWaveProxy implementation.
+#pragma once
+
+#include "esphome/components/api/api_pb2.h"
+
+namespace esphome {
+namespace api {
+class APIConnection;
+}  // namespace api
+
+namespace zwave_proxy {
+
+class ZWaveProxy {
+ public:
+  api::APIConnection *get_api_connection() { return nullptr; }
+  void zwave_proxy_request(api::APIConnection *conn, api::enums::ZWaveProxyRequestType type) {}
+  void send_frame(const uint8_t *data, size_t length) {}
+  void api_connection_authenticated(api::APIConnection *conn) {}
+  uint32_t get_feature_flags() const { return 0; }
+  uint32_t get_home_id() { return 0; }
+};
+
+// NOLINTNEXTLINE(cppcoreguidelines-avoid-non-const-global-variables)
+extern ZWaveProxy *global_zwave_proxy;
+
+}  // namespace zwave_proxy
+}  // namespace esphome

tests/components/audio/common.yaml

@@ -0,0 +1,14 @@
+audio:
+  codecs:
+    flac:
+      buffer_memory: internal
+    mp3:
+      buffer_memory: psram
+    opus:
+      floating_point: false
+      state_memory: psram
+      pseudostack:
+        threadsafe: false
+        buffer_memory: internal
+        size: 80000
+    wav:

tests/components/audio/test.esp32-idf.yaml

@@ -0,0 +1 @@
+<<: !include common.yaml

tests/components/sendspin/common-media_source.yaml

@@ -7,3 +7,4 @@ media_source:
     initial_static_delay: 5ms
     static_delay_adjustable: true
     fixed_delay: 480us
+    decode_memory: internal

tests/integration/fixtures/climate_control_action.yaml

@@ -0,0 +1,92 @@
+esphome:
+  name: climate-control-action-test
+host:
+api:
+logger:
+  level: DEBUG
+
+globals:
+  - id: test_target_temp
+    type: float
+    initial_value: "21.5"
+
+sensor:
+  - platform: template
+    id: temp_sensor
+    name: "Temp"
+    lambda: 'return 20.0;'
+    update_interval: 60s
+
+climate:
+  - platform: thermostat
+    id: test_climate
+    name: "Test Climate"
+    sensor: temp_sensor
+    min_idle_time: 30s
+    min_heating_off_time: 300s
+    min_heating_run_time: 300s
+    min_cooling_off_time: 300s
+    min_cooling_run_time: 300s
+    heat_action:
+      - logger.log: heating
+    idle_action:
+      - logger.log: idle
+    cool_action:
+      - logger.log: cooling
+    heat_cool_mode:
+      - logger.log: heat_cool
+    preset:
+      - name: Default
+        default_target_temperature_low: 18 °C
+        default_target_temperature_high: 22 °C
+    visual:
+      min_temperature: 10 °C
+      max_temperature: 30 °C
+
+button:
+  # mode only
+  - platform: template
+    id: btn_mode
+    name: "Set Mode Heat"
+    on_press:
+      - climate.control:
+          id: test_climate
+          mode: HEAT
+
+  # mode + target_temperature_low + target_temperature_high
+  - platform: template
+    id: btn_mode_temps
+    name: "Set Mode Temps"
+    on_press:
+      - climate.control:
+          id: test_climate
+          mode: HEAT_COOL
+          target_temperature_low: 19.0 °C
+          target_temperature_high: 23.0 °C
+
+  # target_temperature_low only
+  - platform: template
+    id: btn_low_only
+    name: "Set Low Only"
+    on_press:
+      - climate.control:
+          id: test_climate
+          target_temperature_low: 17.5 °C
+
+  # Lambda path: target_temperature_high computed at runtime
+  - platform: template
+    id: btn_lambda_high
+    name: "Lambda High"
+    on_press:
+      - climate.control:
+          id: test_climate
+          target_temperature_high: !lambda "return id(test_target_temp);"
+
+  # mode only — turn off via mode
+  - platform: template
+    id: btn_off
+    name: "Set Off"
+    on_press:
+      - climate.control:
+          id: test_climate
+          mode: "OFF"

tests/integration/test_climate_control_action.py

@@ -0,0 +1,84 @@
+"""Integration test for climate ControlAction.
+
+Tests that climate.control automation actions work correctly with the
+single stateless apply lambda/function pointer implementation. Exercises
+multiple field combinations and the lambda path.
+"""
+
+from __future__ import annotations
+
+import asyncio
+
+from aioesphomeapi import (
+    ButtonInfo,
+    ClimateInfo,
+    ClimateMode,
+    ClimateState,
+    EntityState,
+)
+import pytest
+
+from .state_utils import InitialStateHelper, require_entity
+from .types import APIClientConnectedFactory, RunCompiledFunction
+
+
+@pytest.mark.asyncio
+async def test_climate_control_action(
+    yaml_config: str,
+    run_compiled: RunCompiledFunction,
+    api_client_connected: APIClientConnectedFactory,
+) -> None:
+    """Test climate ControlAction with constants and lambdas."""
+    loop = asyncio.get_running_loop()
+    async with run_compiled(yaml_config), api_client_connected() as client:
+        climate_state_future: asyncio.Future[ClimateState] | None = None
+
+        def on_state(state: EntityState) -> None:
+            if (
+                isinstance(state, ClimateState)
+                and climate_state_future is not None
+                and not climate_state_future.done()
+            ):
+                climate_state_future.set_result(state)
+
+        async def wait_for_climate_state(timeout: float = 5.0) -> ClimateState:
+            nonlocal climate_state_future
+            climate_state_future = loop.create_future()
+            try:
+                return await asyncio.wait_for(climate_state_future, timeout)
+            finally:
+                climate_state_future = None
+
+        entities, _ = await client.list_entities_services()
+        initial_state_helper = InitialStateHelper(entities)
+        client.subscribe_states(initial_state_helper.on_state_wrapper(on_state))
+        await initial_state_helper.wait_for_initial_states()
+
+        require_entity(entities, "test_climate", ClimateInfo)
+
+        async def press_and_wait(name: str) -> ClimateState:
+            btn = require_entity(entities, name.lower().replace(" ", "_"), ButtonInfo)
+            client.button_command(btn.key)
+            return await wait_for_climate_state()
+
+        # mode only — set HEAT
+        state = await press_and_wait("Set Mode Heat")
+        assert state.mode == ClimateMode.HEAT
+
+        # mode + target_temperature_low + target_temperature_high
+        state = await press_and_wait("Set Mode Temps")
+        assert state.mode == ClimateMode.HEAT_COOL
+        assert state.target_temperature_low == pytest.approx(19.0, abs=0.5)
+        assert state.target_temperature_high == pytest.approx(23.0, abs=0.5)
+
+        # target_temperature_low only
+        state = await press_and_wait("Set Low Only")
+        assert state.target_temperature_low == pytest.approx(17.5, abs=0.5)
+
+        # lambda path: target_temperature_high computed at runtime
+        state = await press_and_wait("Lambda High")
+        assert state.target_temperature_high == pytest.approx(21.5, abs=0.5)
+
+        # mode only — turn off via mode
+        state = await press_and_wait("Set Off")
+        assert state.mode == ClimateMode.OFF

tests/script/test_determine_jobs.py

@@ -63,6 +63,13 @@ def mock_should_run_import_time() -> Generator[Mock, None, None]:
         yield mock
 
 
+@pytest.fixture
+def mock_should_run_device_builder() -> Generator[Mock, None, None]:
+    """Mock should_run_device_builder from determine_jobs."""
+    with patch.object(determine_jobs, "should_run_device_builder") as mock:
+        yield mock
+
+
 @pytest.fixture
 def mock_determine_cpp_unit_tests() -> Generator[Mock, None, None]:
     """Mock determine_cpp_unit_tests from helpers."""
@@ -99,6 +106,7 @@ def test_main_all_tests_should_run(
     mock_should_run_clang_format: Mock,
     mock_should_run_python_linters: Mock,
     mock_should_run_import_time: Mock,
+    mock_should_run_device_builder: Mock,
     mock_changed_files: Mock,
     mock_determine_cpp_unit_tests: Mock,
     capsys: pytest.CaptureFixture[str],
@@ -113,6 +121,7 @@ def test_main_all_tests_should_run(
     mock_should_run_clang_format.return_value = True
     mock_should_run_python_linters.return_value = True
     mock_should_run_import_time.return_value = True
+    mock_should_run_device_builder.return_value = True
     mock_determine_cpp_unit_tests.return_value = (False, ["wifi", "api", "sensor"])
 
     # Mock changed_files to return non-component files (to avoid memory impact)
@@ -122,10 +131,19 @@ def test_main_all_tests_should_run(
         "esphome/helpers.py",
     ]
 
+    # Stable, deterministic stand-in for the tests/integration/ glob so the
+    # bucket assertions don't drift with the real test count.
+    fake_test_files = [f"tests/integration/test_{i:03d}.py" for i in range(15)]
+
     # Run main function with mocked argv
     with (
         patch("sys.argv", ["determine-jobs.py"]),
         patch.object(determine_jobs, "_is_clang_tidy_full_scan", return_value=False),
+        patch.object(
+            determine_jobs,
+            "_all_integration_test_files",
+            return_value=fake_test_files,
+        ),
         patch.object(
             determine_jobs,
             "get_changed_components",
@@ -161,13 +179,30 @@ def test_main_all_tests_should_run(
     output = json.loads(captured.out)
 
     assert output["integration_tests"] is True
-    assert output["integration_tests_run_all"] is True
-    assert output["integration_test_files"] == []
+    # run_all=True expands to the full glob and pre-buckets into 3 parts.
+    # Each bucket's `tests` is a JSON list of file paths.
+    assert isinstance(output["integration_test_buckets"], list)
+    assert len(output["integration_test_buckets"]) == 3
+    assert [b["name"] for b in output["integration_test_buckets"]] == [
+        "1/3",
+        "2/3",
+        "3/3",
+    ]
+    for bucket in output["integration_test_buckets"]:
+        assert isinstance(bucket["tests"], list)
+        for path in bucket["tests"]:
+            assert isinstance(path, str)
+    bucket_files = [f for b in output["integration_test_buckets"] for f in b["tests"]]
+    assert bucket_files == fake_test_files
+    # Bucket sizes are balanced (max-min difference at most 1).
+    sizes = [len(b["tests"]) for b in output["integration_test_buckets"]]
+    assert max(sizes) - min(sizes) <= 1
     assert output["clang_tidy"] is True
     assert output["clang_tidy_mode"] in ["nosplit", "split"]
     assert output["clang_format"] is True
     assert output["python_linters"] is True
     assert output["import_time"] is True
+    assert output["device_builder"] is True
     assert output["changed_components"] == ["wifi", "api", "sensor"]
     # changed_components_with_tests will only include components that actually have test files
     assert "changed_components_with_tests" in output
@@ -200,6 +235,7 @@ def test_main_no_tests_should_run(
     mock_should_run_clang_format: Mock,
     mock_should_run_python_linters: Mock,
     mock_should_run_import_time: Mock,
+    mock_should_run_device_builder: Mock,
     mock_changed_files: Mock,
     mock_determine_cpp_unit_tests: Mock,
     capsys: pytest.CaptureFixture[str],
@@ -214,6 +250,7 @@ def test_main_no_tests_should_run(
     mock_should_run_clang_format.return_value = False
     mock_should_run_python_linters.return_value = False
     mock_should_run_import_time.return_value = False
+    mock_should_run_device_builder.return_value = False
     mock_determine_cpp_unit_tests.return_value = (False, [])
 
     # Mock changed_files to return no component files
@@ -247,13 +284,13 @@ def test_main_no_tests_should_run(
     output = json.loads(captured.out)
 
     assert output["integration_tests"] is False
-    assert output["integration_tests_run_all"] is False
-    assert output["integration_test_files"] == []
+    assert output["integration_test_buckets"] == []
     assert output["clang_tidy"] is False
     assert output["clang_tidy_mode"] == "disabled"
     assert output["clang_format"] is False
     assert output["python_linters"] is False
     assert output["import_time"] is False
+    assert output["device_builder"] is False
     assert output["changed_components"] == []
     assert output["changed_components_with_tests"] == []
     assert output["component_test_count"] == 0
@@ -275,6 +312,7 @@ def test_main_with_branch_argument(
     mock_should_run_clang_format: Mock,
     mock_should_run_python_linters: Mock,
     mock_should_run_import_time: Mock,
+    mock_should_run_device_builder: Mock,
     mock_changed_files: Mock,
     mock_determine_cpp_unit_tests: Mock,
     capsys: pytest.CaptureFixture[str],
@@ -289,6 +327,7 @@ def test_main_with_branch_argument(
     mock_should_run_clang_format.return_value = False
     mock_should_run_python_linters.return_value = True
     mock_should_run_import_time.return_value = True
+    mock_should_run_device_builder.return_value = True
     mock_determine_cpp_unit_tests.return_value = (False, ["mqtt"])
 
     # Mock changed_files to return non-component files (to avoid memory impact)
@@ -326,19 +365,20 @@ def test_main_with_branch_argument(
     mock_should_run_clang_format.assert_called_once_with("main")
     mock_should_run_python_linters.assert_called_once_with("main")
     mock_should_run_import_time.assert_called_once_with("main")
+    mock_should_run_device_builder.assert_called_once_with("main")
 
     # Check output
     captured = capsys.readouterr()
     output = json.loads(captured.out)
 
     assert output["integration_tests"] is False
-    assert output["integration_tests_run_all"] is False
-    assert output["integration_test_files"] == []
+    assert output["integration_test_buckets"] == []
     assert output["clang_tidy"] is True
     assert output["clang_tidy_mode"] in ["nosplit", "split"]
     assert output["clang_format"] is False
     assert output["python_linters"] is True
     assert output["import_time"] is True
+    assert output["device_builder"] is True
     assert output["changed_components"] == ["mqtt"]
     # changed_components_with_tests will only include components that actually have test files
     assert "changed_components_with_tests" in output
@@ -357,6 +397,59 @@ def test_main_with_branch_argument(
     assert output["cpp_unit_tests_components"] == ["mqtt"]
 
 
+def test_compute_integration_test_buckets_empty() -> None:
+    """No integration tests scheduled => (False, [])."""
+    run, buckets = determine_jobs._compute_integration_test_buckets(False, [])
+    assert run is False
+    assert buckets == []
+
+
+def test_compute_integration_test_buckets_below_threshold() -> None:
+    """A small explicit list (<= threshold) => single 1/1 bucket with that list."""
+    files = [f"tests/integration/test_{name}.py" for name in ("c", "a", "b")]
+    run, buckets = determine_jobs._compute_integration_test_buckets(False, files)
+    assert run is True
+    assert buckets == [{"name": "1/1", "tests": sorted(files)}]
+
+
+def test_compute_integration_test_buckets_at_threshold_stays_single() -> None:
+    """Exactly INTEGRATION_TESTS_SPLIT_THRESHOLD files => still one bucket
+    (the split kicks in only when count is strictly greater than threshold)."""
+    files = [
+        f"tests/integration/test_{i:02d}.py"
+        for i in range(determine_jobs.INTEGRATION_TESTS_SPLIT_THRESHOLD)
+    ]
+    run, buckets = determine_jobs._compute_integration_test_buckets(False, files)
+    assert run is True
+    assert len(buckets) == 1
+    assert buckets[0]["name"] == "1/1"
+    assert buckets[0]["tests"] == sorted(files)
+
+
+def test_compute_integration_test_buckets_just_over_threshold_splits() -> None:
+    """One file over the threshold triggers the 3-bucket fan-out, balanced."""
+    n = determine_jobs.INTEGRATION_TESTS_SPLIT_THRESHOLD + 1
+    files = [f"tests/integration/test_{i:02d}.py" for i in range(n)]
+    run, buckets = determine_jobs._compute_integration_test_buckets(False, files)
+    assert run is True
+    assert [b["name"] for b in buckets] == ["1/3", "2/3", "3/3"]
+    union = [path for b in buckets for path in b["tests"]]
+    assert union == sorted(files)
+    sizes = [len(b["tests"]) for b in buckets]
+    assert max(sizes) - min(sizes) <= 1
+
+
+def test_compute_integration_test_buckets_run_all_with_empty_glob_disables_run() -> (
+    None
+):
+    """run_all=True but glob returns no files => run suppressed (otherwise
+    pytest would collect tests outside tests/integration/)."""
+    with patch.object(determine_jobs, "_all_integration_test_files", return_value=[]):
+        run, buckets = determine_jobs._compute_integration_test_buckets(True, [])
+    assert run is False
+    assert buckets == []
+
+
 def test_determine_integration_tests(
     monkeypatch: pytest.MonkeyPatch,
 ) -> None:
@@ -658,6 +751,82 @@ def test_should_run_import_time_with_branch() -> None:
         mock_changed.assert_called_once_with("release")
 
 
+@pytest.mark.parametrize(
+    ("changed_files", "expected_result"),
+    [
+        # esphome Python files trigger downstream device-builder tests
+        (["esphome/__main__.py"], True),
+        (["esphome/components/wifi/__init__.py"], True),
+        (["esphome/core/config.py"], True),
+        (["esphome/types.pyi"], True),
+        # Runtime dependency changes trigger
+        (["requirements.txt"], True),
+        (["pyproject.toml"], True),
+        # Non-C++ files packaged with esphome trigger -- device-builder
+        # picks them up because esphome's pyproject sets
+        # include-package-data = true.
+        (["esphome/idf_component.yml"], True),
+        (["esphome/dashboard/templates/index.html"], True),
+        (["esphome/components/api/api_pb2_service.json"], True),
+        # Mixed: any triggering file is enough
+        (["docs/README.md", "esphome/config.py"], True),
+        # Dev/test-only dependency changes don't trigger device-builder
+        # (they don't affect the importable surface device-builder uses)
+        (["requirements_dev.txt"], False),
+        (["requirements_test.txt"], False),
+        # Files outside esphome/ don't trigger
+        (["script/some_other_script.py"], False),
+        (["tests/script/test_determine_jobs.py"], False),
+        # C++ files under esphome/ don't trigger -- they only affect
+        # compiled firmware, not the Python install device-builder pulls in.
+        (["esphome/core/component.cpp"], False),
+        (["esphome/core/component.h"], False),
+        (["esphome/components/wifi/wifi_component.cpp"], False),
+        # Files outside esphome/ entirely
+        (["tests/components/wifi/test.esp32-idf.yaml"], False),
+        (["README.md"], False),
+        ([], False),
+    ],
+)
+def test_should_run_device_builder(
+    changed_files: list[str], expected_result: bool
+) -> None:
+    """Test should_run_device_builder function (non-beta/release target)."""
+    with (
+        patch.object(determine_jobs, "changed_files", return_value=changed_files),
+        # Mock target branch to "dev" so the beta/release skip is bypassed
+        # for these per-file behavior checks.
+        patch.object(determine_jobs, "get_target_branch", return_value="dev"),
+    ):
+        result = determine_jobs.should_run_device_builder()
+        assert result == expected_result
+
+
+def test_should_run_device_builder_with_branch() -> None:
+    """Test should_run_device_builder with branch argument."""
+    with (
+        patch.object(determine_jobs, "changed_files") as mock_changed,
+        patch.object(determine_jobs, "get_target_branch", return_value="dev"),
+    ):
+        mock_changed.return_value = []
+        determine_jobs.should_run_device_builder("release")
+        mock_changed.assert_called_once_with("release")
+
+
+@pytest.mark.parametrize("target_branch", ["beta", "release", "release-2026.5"])
+def test_should_run_device_builder_skips_beta_release(target_branch: str) -> None:
+    """Beta/release target branches skip device-builder (lag behind device-builder@main)."""
+    with (
+        patch.object(determine_jobs, "get_target_branch", return_value=target_branch),
+        patch.object(determine_jobs, "changed_files") as mock_changed,
+    ):
+        # Even with a triggering file present, the target-branch guard wins.
+        mock_changed.return_value = ["esphome/__main__.py"]
+        assert determine_jobs.should_run_device_builder() is False
+        # changed_files shouldn't even be consulted -- the guard short-circuits.
+        mock_changed.assert_not_called()
+
+
 @pytest.mark.parametrize(
     ("changed_files", "expected_result"),
     [

tests/unit_tests/test_espota2.py

@@ -185,6 +185,14 @@ def test_receive_exactly_socket_error(mock_socket: Mock) -> None:
             "Error: The OTA partition on the ESP couldn't be found",
         ),
         (espota2.RESPONSE_ERROR_MD5_MISMATCH, "Error: Application MD5 code mismatch"),
+        (
+            espota2.RESPONSE_ERROR_SIGNATURE_INVALID,
+            "Error: Firmware signature verification failed",
+        ),
+        (
+            espota2.RESPONSE_ERROR_UNSUPPORTED_OTA_TYPE,
+            "Error: The requested OTA type is not supported by the device",
+        ),
         (espota2.RESPONSE_ERROR_UNKNOWN, "Unknown error from ESP"),
     ],
 )
@@ -270,12 +278,13 @@ def test_perform_ota_successful_md5_auth(
     # Verify magic bytes were sent
     assert mock_socket.sendall.call_args_list[0] == call(bytes(espota2.MAGIC_BYTES))
 
-    # Verify features were sent (compression + SHA256 support)
+    # Verify features were sent (compression + SHA256 support + extended protocol)
     assert mock_socket.sendall.call_args_list[1] == call(
         bytes(
             [
-                espota2.FEATURE_SUPPORTS_COMPRESSION
-                | espota2.FEATURE_SUPPORTS_SHA256_AUTH
+                espota2.CLIENT_FEATURE_SUPPORTS_COMPRESSION
+                | espota2.CLIENT_FEATURE_SUPPORTS_SHA256_AUTH
+                | espota2.CLIENT_FEATURE_SUPPORTS_EXTENDED_PROTOCOL
             ]
         )
     )
@@ -640,12 +649,13 @@ def test_perform_ota_successful_sha256_auth(
     # Verify magic bytes were sent
     assert mock_socket.sendall.call_args_list[0] == call(bytes(espota2.MAGIC_BYTES))
 
-    # Verify features were sent (compression + SHA256 support)
+    # Verify features were sent (compression + SHA256 support + extended protocol)
     assert mock_socket.sendall.call_args_list[1] == call(
         bytes(
             [
-                espota2.FEATURE_SUPPORTS_COMPRESSION
-                | espota2.FEATURE_SUPPORTS_SHA256_AUTH
+                espota2.CLIENT_FEATURE_SUPPORTS_COMPRESSION
+                | espota2.CLIENT_FEATURE_SUPPORTS_SHA256_AUTH
+                | espota2.CLIENT_FEATURE_SUPPORTS_EXTENDED_PROTOCOL
             ]
         )
     )
@@ -699,8 +709,9 @@ def test_perform_ota_sha256_fallback_to_md5(
     assert mock_socket.sendall.call_args_list[1] == call(
         bytes(
             [
-                espota2.FEATURE_SUPPORTS_COMPRESSION
-                | espota2.FEATURE_SUPPORTS_SHA256_AUTH
+                espota2.CLIENT_FEATURE_SUPPORTS_COMPRESSION
+                | espota2.CLIENT_FEATURE_SUPPORTS_SHA256_AUTH
+                | espota2.CLIENT_FEATURE_SUPPORTS_EXTENDED_PROTOCOL
             ]
         )
     )
@@ -765,3 +776,220 @@ def test_perform_ota_version_differences(
 
     # For v2.0, verify more recv calls due to chunk acknowledgments
     assert mock_socket.recv.call_count == 9  # v2.0 has 9 recv calls (includes chunk OK)
+
+
+@pytest.mark.usefixtures("mock_time")
+def test_perform_ota_extended_protocol_app(
+    mock_socket: Mock, mock_file: io.BytesIO
+) -> None:
+    """Test OTA extended protocol app update."""
+    recv_responses = [
+        bytes([espota2.RESPONSE_OK]),  # First byte of version response
+        bytes([espota2.OTA_VERSION_2_0]),  # Version number
+        bytes([espota2.RESPONSE_FEATURE_FLAGS]),  # Device supports extended protocol
+        bytes(
+            [
+                espota2.SERVER_FEATURE_SUPPORTS_COMPRESSION
+                | espota2.SERVER_FEATURE_SUPPORTS_PARTITION_ACCESS
+            ]
+        ),  # Device feature flags
+        bytes([espota2.RESPONSE_AUTH_OK]),  # No auth required
+        bytes([espota2.RESPONSE_UPDATE_PREPARE_OK]),  # Binary size OK
+        bytes([espota2.RESPONSE_BIN_MD5_OK]),  # MD5 checksum OK
+        bytes([espota2.RESPONSE_CHUNK_OK]),  # Chunk OK
+        bytes([espota2.RESPONSE_RECEIVE_OK]),  # Receive OK
+        bytes([espota2.RESPONSE_UPDATE_END_OK]),  # Update end OK
+    ]
+
+    mock_socket.recv.side_effect = recv_responses
+
+    espota2.perform_ota(
+        mock_socket,
+        "testpass",
+        mock_file,
+        "test.bin",
+        espota2.OTA_TYPE_UPDATE_APP,
+    )
+
+    # Verify magic bytes were sent
+    assert mock_socket.sendall.call_args_list[0] == call(bytes(espota2.MAGIC_BYTES))
+
+    # Verify features were sent (compression + SHA256 support + extended protocol)
+    assert mock_socket.sendall.call_args_list[1] == call(
+        bytes(
+            [
+                espota2.CLIENT_FEATURE_SUPPORTS_COMPRESSION
+                | espota2.CLIENT_FEATURE_SUPPORTS_SHA256_AUTH
+                | espota2.CLIENT_FEATURE_SUPPORTS_EXTENDED_PROTOCOL
+            ]
+        )
+    )
+
+    # Verify ota type was sent
+    assert mock_socket.sendall.call_args_list[2] == call(
+        bytes([espota2.OTA_TYPE_UPDATE_APP])
+    )
+
+
+@pytest.mark.usefixtures("mock_time")
+def test_perform_ota_device_rejects_with_unsupported_ota_type(
+    mock_socket: Mock, mock_file: io.BytesIO
+) -> None:
+    """End-to-end: device returns 0x8E after the size byte; perform_ota must
+    surface the human-readable 'unsupported OTA type' error from the lookup
+    table in check_error()."""
+    recv_responses = [
+        bytes([espota2.RESPONSE_OK]),  # First byte of version response
+        bytes([espota2.OTA_VERSION_2_0]),  # Version number
+        bytes([espota2.RESPONSE_FEATURE_FLAGS]),  # Extended protocol marker
+        bytes(
+            [
+                espota2.SERVER_FEATURE_SUPPORTS_COMPRESSION
+                | espota2.SERVER_FEATURE_SUPPORTS_PARTITION_ACCESS
+            ]
+        ),  # Feature flags
+        bytes([espota2.RESPONSE_AUTH_OK]),  # No auth required
+        bytes([espota2.RESPONSE_ERROR_UNSUPPORTED_OTA_TYPE]),  # Reject at size step
+    ]
+
+    mock_socket.recv.side_effect = recv_responses
+
+    with pytest.raises(
+        espota2.OTAError,
+        match="The requested OTA type is not supported by the device",
+    ):
+        espota2.perform_ota(
+            mock_socket,
+            "testpass",
+            mock_file,
+            "test.bin",
+            espota2.OTA_TYPE_UPDATE_APP,
+        )
+
+    # Verify the client did send the OTA type byte before the size step
+    assert mock_socket.sendall.call_args_list[2] == call(
+        bytes([espota2.OTA_TYPE_UPDATE_APP])
+    )
+
+
+@pytest.mark.usefixtures("mock_time")
+def test_perform_ota_unsupported_type_rejected_early(
+    mock_socket: Mock, mock_file: io.BytesIO
+) -> None:
+    """ota_type values not in _SUPPORTED_OTA_TYPES are rejected before any I/O."""
+    with pytest.raises(espota2.OTAError, match="Unsupported OTA type 0xFF"):
+        espota2.perform_ota(
+            mock_socket,
+            "testpass",
+            mock_file,
+            "test.bin",
+            0xFF,
+        )
+    # No bytes should have been transmitted to the device.
+    mock_socket.sendall.assert_not_called()
+
+
+@pytest.mark.parametrize("bad_type", [-1, 256, 0x10000, "app", None, 1.5])
+def test_perform_ota_rejects_out_of_range_type(
+    mock_socket: Mock, mock_file: io.BytesIO, bad_type: object
+) -> None:
+    """Out-of-range or non-int ota_type must raise OTAError, not ValueError."""
+    with pytest.raises(espota2.OTAError, match="Invalid ota_type"):
+        espota2.perform_ota(
+            mock_socket,
+            "testpass",
+            mock_file,
+            "test.bin",
+            bad_type,  # type: ignore[arg-type]
+        )
+    mock_socket.sendall.assert_not_called()
+
+
+@pytest.mark.usefixtures("mock_time")
+def test_perform_ota_non_app_type_requires_extended_protocol(
+    mock_socket: Mock, mock_file: io.BytesIO, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """Non-app OTA type must fail when device only supports the legacy protocol."""
+    monkeypatch.setattr(
+        espota2,
+        "_SUPPORTED_OTA_TYPES",
+        frozenset({espota2.OTA_TYPE_UPDATE_APP, 0xFF}),
+    )
+    recv_responses = [
+        bytes([espota2.RESPONSE_OK]),  # First byte of version response
+        bytes([espota2.OTA_VERSION_2_0]),  # Version number
+        bytes([espota2.RESPONSE_HEADER_OK]),  # Legacy single-byte feature ack
+    ]
+
+    mock_socket.recv.side_effect = recv_responses
+
+    with pytest.raises(
+        espota2.OTAError, match="Device does not support extended OTA protocol"
+    ):
+        espota2.perform_ota(
+            mock_socket,
+            "testpass",
+            mock_file,
+            "test.bin",
+            0xFF,
+        )
+
+
+@pytest.mark.usefixtures("mock_time")
+def test_perform_ota_non_app_type_requires_partition_access(
+    mock_socket: Mock, mock_file: io.BytesIO, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """Non-app OTA type must fail when device advertises extended protocol but
+    not the partition-access feature."""
+    monkeypatch.setattr(
+        espota2,
+        "_SUPPORTED_OTA_TYPES",
+        frozenset({espota2.OTA_TYPE_UPDATE_APP, 0xFF}),
+    )
+    recv_responses = [
+        bytes([espota2.RESPONSE_OK]),  # First byte of version response
+        bytes([espota2.OTA_VERSION_2_0]),  # Version number
+        bytes([espota2.RESPONSE_FEATURE_FLAGS]),  # Extended protocol marker
+        bytes(
+            [espota2.SERVER_FEATURE_SUPPORTS_COMPRESSION]
+        ),  # Compression only, no partition access
+    ]
+
+    mock_socket.recv.side_effect = recv_responses
+
+    with pytest.raises(
+        espota2.OTAError, match="Device does not support partition access"
+    ):
+        espota2.perform_ota(
+            mock_socket,
+            "testpass",
+            mock_file,
+            "test.bin",
+            0xFF,
+        )
+
+
+def test_check_error_detects_errors_when_expect_is_none() -> None:
+    """check_error must surface device error bytes even when expect is None.
+
+    Regression test: previously, receive_exactly(..., expect=None) calls (used
+    during feature negotiation and nonce reads) silently passed error bytes
+    through, turning clean device errors into confusing later failures.
+    """
+    with pytest.raises(espota2.OTAError, match="Error: Authentication invalid"):
+        espota2.check_error([espota2.RESPONSE_ERROR_AUTH_INVALID], None)
+
+
+def test_check_error_detects_empty_when_expect_is_none() -> None:
+    """Empty data with expect=None must still raise (connection closed)."""
+    with pytest.raises(
+        espota2.OTAError, match="Device closed connection without responding"
+    ):
+        espota2.check_error([], None)
+
+
+def test_check_error_passes_non_error_when_expect_is_none() -> None:
+    """Non-error bytes with expect=None must pass through silently."""
+    espota2.check_error([espota2.RESPONSE_OK], None)
+    espota2.check_error([espota2.RESPONSE_HEADER_OK], None)
+    espota2.check_error([espota2.RESPONSE_FEATURE_FLAGS], None)

tests/unit_tests/test_main.py

@@ -83,6 +83,7 @@ from esphome.const import (
     PLATFORM_RP2040,
 )
 from esphome.core import CORE, EsphomeError
+from esphome.espota2 import OTA_TYPE_UPDATE_APP
 from esphome.util import BootselResult
 from esphome.zeroconf import _await_discovery, discover_mdns_devices
 
@@ -1593,7 +1594,7 @@ def test_upload_program_ota_success(
         tmp_path / ".esphome" / "build" / "test" / ".pioenvs" / "test" / "firmware.bin"
     )
     mock_run_ota.assert_called_once_with(
-        ["192.168.1.100"], 3232, "secret", expected_firmware
+        ["192.168.1.100"], 3232, "secret", expected_firmware, OTA_TYPE_UPDATE_APP
     )
 
 
@@ -1624,7 +1625,7 @@ def test_upload_program_ota_with_file_arg(
     assert exit_code == 0
     assert host == "192.168.1.100"
     mock_run_ota.assert_called_once_with(
-        ["192.168.1.100"], 3232, None, Path("custom.bin")
+        ["192.168.1.100"], 3232, None, Path("custom.bin"), OTA_TYPE_UPDATE_APP
     )
 
 
@@ -1682,7 +1683,7 @@ def test_upload_program_ota_with_mqtt_resolution(
         tmp_path / ".esphome" / "build" / "test" / ".pioenvs" / "test" / "firmware.bin"
     )
     mock_run_ota.assert_called_once_with(
-        ["192.168.1.100"], 3232, None, expected_firmware
+        ["192.168.1.100"], 3232, None, expected_firmware, OTA_TYPE_UPDATE_APP
     )
 
 
@@ -1730,7 +1731,7 @@ def test_upload_program_ota_with_mqtt_empty_broker(
         tmp_path / ".esphome" / "build" / "test" / ".pioenvs" / "test" / "firmware.bin"
     )
     mock_run_ota.assert_called_once_with(
-        ["192.168.1.50"], 3232, None, expected_firmware
+        ["192.168.1.50"], 3232, None, expected_firmware, OTA_TYPE_UPDATE_APP
     )
     # Verify warning was logged
     assert "MQTT IP discovery failed" in caplog.text
@@ -3207,7 +3208,11 @@ def test_upload_program_ota_static_ip_with_mqttip(
         tmp_path / ".esphome" / "build" / "test" / ".pioenvs" / "test" / "firmware.bin"
     )
     mock_run_ota.assert_called_once_with(
-        ["192.168.1.100", "192.168.2.50"], 3232, None, expected_firmware
+        ["192.168.1.100", "192.168.2.50"],
+        3232,
+        None,
+        expected_firmware,
+        OTA_TYPE_UPDATE_APP,
     )
 
 
@@ -3250,7 +3255,11 @@ def test_upload_program_ota_multiple_mqttip_resolves_once(
         tmp_path / ".esphome" / "build" / "test" / ".pioenvs" / "test" / "firmware.bin"
     )
     mock_run_ota.assert_called_once_with(
-        ["192.168.2.50", "192.168.2.51", "192.168.1.100"], 3232, None, expected_firmware
+        ["192.168.2.50", "192.168.2.51", "192.168.1.100"],
+        3232,
+        None,
+        expected_firmware,
+        OTA_TYPE_UPDATE_APP,
     )
 
 
@@ -3415,7 +3424,7 @@ def test_upload_program_ota_mqtt_timeout_fallback(
         tmp_path / ".esphome" / "build" / "test" / ".pioenvs" / "test" / "firmware.bin"
     )
     mock_run_ota.assert_called_once_with(
-        ["192.168.1.100"], 3232, None, expected_firmware
+        ["192.168.1.100"], 3232, None, expected_firmware, OTA_TYPE_UPDATE_APP
     )
 
 

tests/unit_tests/test_platformio_api.py

@@ -311,6 +311,105 @@ def test_run_platformio_cli_sets_environment_variables(
         assert "arg" in args
 
 
+@pytest.mark.parametrize(
+    ("platform", "input_path", "expected"),
+    [
+        # win32: drive-letter extended-length prefix is stripped
+        (
+            "win32",
+            "\\\\?\\C:\\Users\\jesse\\AppData\\Local\\ESPHome Builder\\python\\python.exe",
+            "C:\\Users\\jesse\\AppData\\Local\\ESPHome Builder\\python\\python.exe",
+        ),
+        # win32: UNC extended-length prefix is translated to a regular UNC path
+        (
+            "win32",
+            "\\\\?\\UNC\\server\\share\\python.exe",
+            "\\\\server\\share\\python.exe",
+        ),
+        # win32: paths without the prefix are returned unchanged
+        (
+            "win32",
+            "C:\\Users\\jesse\\AppData\\Local\\ESPHome Builder\\python\\python.exe",
+            "C:\\Users\\jesse\\AppData\\Local\\ESPHome Builder\\python\\python.exe",
+        ),
+        # non-win32: prefix is left alone (no-op)
+        ("linux", "\\\\?\\C:\\python.exe", "\\\\?\\C:\\python.exe"),
+        ("darwin", "/usr/bin/python3", "/usr/bin/python3"),
+    ],
+)
+def test_strip_win_long_path_prefix(
+    platform: str, input_path: str, expected: str
+) -> None:
+    r"""``\\?\`` and ``\\?\UNC\`` prefixes are stripped only on win32."""
+    with patch("esphome.platformio_api.sys.platform", platform):
+        assert platformio_api._strip_win_long_path_prefix(input_path) == expected
+
+
+def test_run_platformio_cli_strips_win_long_path_prefix(
+    setup_core: Path, mock_run_external_process: Mock
+) -> None:
+    r"""Windows ``\\?\`` prefix on sys.executable does not leak into the subprocess.
+
+    The NSIS-installed esphome.exe launcher starts Python with
+    ``sys.executable`` already prefixed by the extended-length path marker.
+    That prefix would otherwise propagate into PlatformIO's ``PYTHONEXE`` and
+    break SCons-emitted command lines run through ``cmd.exe``.
+    """
+    CORE.build_path = str(setup_core / "build" / "test")
+    prefixed_exe = (
+        "\\\\?\\C:\\Users\\jesse\\AppData\\Local\\ESPHome Builder\\python\\python.exe"
+    )
+    stripped_exe = (
+        "C:\\Users\\jesse\\AppData\\Local\\ESPHome Builder\\python\\python.exe"
+    )
+
+    with (
+        patch.dict(os.environ, {}, clear=False),
+        patch("esphome.platformio_api.sys.platform", "win32"),
+        patch("esphome.platformio_api.sys.executable", prefixed_exe),
+    ):
+        # Pop any pre-existing PYTHONEXEPATH so the assertion below reflects
+        # what run_platformio_cli set, not whatever the test runner's
+        # environment happened to contain.
+        os.environ.pop("PYTHONEXEPATH", None)
+        mock_run_external_process.return_value = 0
+        platformio_api.run_platformio_cli("test", "arg")
+
+        # The subprocess is invoked with the stripped executable path.
+        mock_run_external_process.assert_called_once()
+        args = mock_run_external_process.call_args[0]
+        assert args[0] == stripped_exe
+        # PYTHONEXEPATH is exported with the stripped path so PlatformIO's
+        # get_pythonexe_path() picks it up in the subprocess.
+        assert os.environ["PYTHONEXEPATH"] == stripped_exe
+
+
+def test_run_platformio_cli_does_not_set_pythonexepath_without_strip(
+    setup_core: Path, mock_run_external_process: Mock
+) -> None:
+    r"""PYTHONEXEPATH is not touched when sys.executable has no ``\\?\`` prefix.
+
+    Setting it unconditionally would clobber a user-provided value (or
+    interfere with non-Windows tooling that has no prefix to strip).
+    """
+    CORE.build_path = str(setup_core / "build" / "test")
+    plain_exe = "/usr/bin/python3"
+
+    with (
+        patch.dict(os.environ, {}, clear=False),
+        patch("esphome.platformio_api.sys.platform", "linux"),
+        patch("esphome.platformio_api.sys.executable", plain_exe),
+    ):
+        os.environ.pop("PYTHONEXEPATH", None)
+        mock_run_external_process.return_value = 0
+        platformio_api.run_platformio_cli("test", "arg")
+
+        mock_run_external_process.assert_called_once()
+        args = mock_run_external_process.call_args[0]
+        assert args[0] == plain_exe
+        assert "PYTHONEXEPATH" not in os.environ
+
+
 def test_run_platformio_cli_run_builds_command(
     setup_core: Path, mock_run_platformio_cli: Mock
 ) -> None: