Merge branch 'esp8266-native-toolchain-plumbing' into esp8266-native-build-infra

# Conflicts:
#	tests/unit_tests/test_main.py

esphome/__main__.py

@@ -857,7 +857,20 @@ def compile_program(args: ArgsProtocol, config: ConfigType) -> int:
         toolchain.create_factory_bin()
         toolchain.create_ota_bin()
         toolchain.create_elf_copy()
-        toolchain.get_idedata()
+        from esphome.build_helpers.idedata import IDEDATA_BEST_EFFORT_ERRORS
+
+        try:
+            if toolchain.get_idedata() is None:
+                _LOGGER.warning("No idedata was generated for this build")
+        except IDEDATA_BEST_EFFORT_ERRORS as err:
+            # The firmware already built; an idedata failure must not fail
+            # a successful build.
+            _LOGGER.warning(
+                "Could not generate idedata: %s (IDE, clang-tidy, and "
+                "memory-analysis data will be unavailable for this build)",
+                err,
+            )
+            _LOGGER.debug("Idedata failure detail", exc_info=True)
     else:
         from esphome.platformio import toolchain
 
@@ -2721,10 +2734,14 @@ def run_esphome(argv):
     # Skipped when -s overrides are passed, since the cache was written
     # against the previous substitution set.
     config: ConfigType | None = None
-    cache_eligible = (
+    cache_write_eligible = (
         args.command in ("upload", "logs") and not command_line_substitutions
     )
-    if cache_eligible:
+    # An explicit --toolchain must re-run the per-platform validators, so
+    # gate only the cache read; the refresh below saves the result unless
+    # the sidecar records a different toolchain.
+    cache_read_eligible = cache_write_eligible and args.toolchain is None
+    if cache_read_eligible:
         from esphome.compiled_config import load_compiled_config
 
         config = load_compiled_config(conf_path)
@@ -2748,17 +2765,14 @@ def run_esphome(argv):
             return 2
     CORE.config = config
 
-    # Fallback for platforms whose validators didn't set the toolchain
-    # (only the esp32 component reads esp32.framework.toolchain). All
-    # other platforms only support PlatformIO today. Must run before the
-    # cache refresh below so its sidecar records the same toolchain a
-    # compile would.
+    # The cache fast path skips validation, and legacy sidecars lack the
+    # toolchain field. Must run before the cache refresh below.
     if CORE.toolchain is None:
         CORE.toolchain = Toolchain.PLATFORMIO
 
     # Refresh the cache so the next upload/logs hits the fast path
     # instead of re-running read_config.
-    if cache_eligible and cache_missed:
+    if cache_write_eligible and cache_missed:
         from esphome.compiled_config import save_compiled_config_and_sidecar
 
         save_compiled_config_and_sidecar(config)

esphome/build_helpers/__init__.py

@@ -0,0 +1 @@
+"""Build helpers shared by the native (non-PlatformIO) toolchains."""

esphome/build_helpers/ccache.py

@@ -0,0 +1,93 @@
+"""Shared ccache policy for build backends: env-knob parsing, binary
+resolution, and default ``CCACHE_*`` values."""
+
+from __future__ import annotations
+
+import logging
+import os
+from pathlib import Path
+
+from esphome.framework_helpers import strip_win_long_path_prefix, tool_version_runs
+from esphome.helpers import FALSY_BOOL_STRINGS, TRUTHY_BOOL_STRINGS
+
+_LOGGER = logging.getLogger(__name__)
+
+# cv.boolean's spelling tables plus the 1/0 env convention
+TRUTHY_ENV_STRINGS = TRUTHY_BOOL_STRINGS | {"1"}
+FALSY_ENV_STRINGS = FALSY_BOOL_STRINGS | {"0"}
+
+
+def _ccache_runs(ccache: str) -> bool:
+    """Return True when the ``ccache`` found on PATH actually runs."""
+    return tool_version_runs(
+        ccache,
+        "Ignoring ccache at %s because it failed to run; compiling without ccache",
+    )
+
+
+def parse_enable_env(name: str) -> bool | None:
+    """Strictly parse an on/off environment knob; None when unset or invalid.
+
+    ``bool(str)`` truthiness would flip ``no``/``off`` to enabled, so only
+    1/true/yes/on and 0/false/no/off count; anything else warns and reads
+    as unset so the caller's default policy applies.
+    """
+    raw = os.environ.get(name)
+    if raw is None:
+        return None
+    lowered = raw.strip().lower()
+    if lowered in TRUTHY_ENV_STRINGS:
+        return True
+    if lowered in FALSY_ENV_STRINGS:
+        return False
+    _LOGGER.warning("Ignoring unrecognized %s=%r; use 1 or 0", name, raw)
+    return None
+
+
+def resolve_ccache_path() -> str | None:
+    """The ccache binary to wrap compiles with, or None when disabled.
+
+    An explicit ``ESPHOME_CCACHE_ENABLE=1`` skips the runnability probe; the
+    Windows extended-length prefix is stripped before probing (#18399).
+    """
+    import shutil
+
+    explicit = parse_enable_env("ESPHOME_CCACHE_ENABLE")
+    if explicit is False:
+        return None
+    ccache = shutil.which("ccache")
+    if ccache is None:
+        if explicit:
+            _LOGGER.warning(
+                "ESPHOME_CCACHE_ENABLE is set but no ccache binary is on PATH; "
+                "compiling without ccache"
+            )
+        return None
+    ccache = strip_win_long_path_prefix(ccache)
+    if not explicit and not _ccache_runs(ccache):
+        return None
+    return ccache
+
+
+def ccache_defaults_env(cache_dir: Path) -> dict[str, str]:
+    """Default ``CCACHE_*`` values for a build subprocess (not os.environ).
+
+    Values the user already set in the environment are respected. Depend
+    mode is on: both native backends emit depfiles (-MMD / CMake), which
+    keeps cache-miss overhead low.
+    """
+    from esphome.core import CORE
+
+    # An unset build_path means the env was built before preload; fail loudly
+    # rather than silently drop CCACHE_BASEDIR.
+    if CORE.build_path is None:
+        raise ValueError(
+            "CORE.build_path must be set before constructing the build environment"
+        )
+    defaults = {
+        "CCACHE_DIR": str(cache_dir),
+        "CCACHE_NOHASHDIR": "true",
+        "CCACHE_DEPEND": "1",
+        "CCACHE_BASEDIR": str(Path(CORE.build_path).resolve()),
+    }
+    return {k: v for k, v in defaults.items() if k not in os.environ}

esphome/build_helpers/idedata.py

@@ -1,10 +1,10 @@
-"""Derive idedata from an ESP-IDF native-toolchain ``compile_commands.json``.
+"""Derive idedata from a native (non-PlatformIO) build's ``compile_commands.json``.
 
-PlatformIO exposes a curated ``pio run -t idedata`` JSON; the native ESP-IDF
-toolchain has no such command, but its CMake build emits
-``build/compile_commands.json`` (CMAKE_EXPORT_COMPILE_COMMANDS). This module
-turns that file into the same fields consumers (IDE integration, clang-tidy)
-expect:
+PlatformIO exposes a curated ``pio run -t idedata`` JSON; the native
+toolchains have no such command, but each build produces a
+``compile_commands.json`` (CMAKE_EXPORT_COMPILE_COMMANDS for ESP-IDF, ninja's
+compdb tool otherwise). This module turns that file into the same fields
+consumers (IDE integration, clang-tidy) expect:
 
     {cc_path, cxx_path, cxx_flags, defines, includes: {build, toolchain}}
 """
@@ -18,6 +18,20 @@ from pathlib import Path
 import shlex
 import subprocess
 
+from esphome.core import EsphomeError
+from esphome.helpers import write_file
+
+# Everything idedata generation may raise after a successful link. Broad on
+# purpose, and shared by every consumer: idedata is a bonus artifact, so
+# these must be caught and warned about, never allowed to fail the build.
+IDEDATA_BEST_EFFORT_ERRORS = (
+    EsphomeError,
+    LookupError,
+    OSError,
+    RuntimeError,
+    ValueError,
+)
+
 _LOGGER = logging.getLogger(__name__)
 
 # C++ translation-unit suffixes used to identify ESPHome source files.
@@ -120,7 +134,18 @@ def _pick_entry(entries: list[dict]) -> dict:
     raise ValueError("no C++ translation unit found in compile_commands.json")
 
 
-def _parse_entry(entry: dict) -> tuple[str, list[str], list[str], list[str]]:
+# Compiler launchers that may prefix a compile command; a closed launcher
+# denylist beats enumerating compiler names, an open set.
+_LAUNCHER_STEMS = frozenset({"ccache", "sccache", "distcc", "icecc", "buildcache"})
+
+
+def _is_launcher(token: str) -> bool:
+    return Path(token).stem.lower() in _LAUNCHER_STEMS
+
+
+def parse_entry(
+    entry: dict, launcher: str | None = None
+) -> tuple[str, list[str], list[str], list[str]]:
     """Parse one compile_commands entry -> (cxx_path, defines, includes, cxx_flags)."""
     directory = Path(entry["directory"])
     tokens = _expand_response_files(_split_command(entry["command"]), directory)
@@ -136,6 +161,20 @@ def _parse_entry(entry: dict) -> tuple[str, list[str], list[str], list[str]]:
             raw = os.path.normpath(directory / raw)
         return raw.replace("\\", "/")
 
+    # A launcher-wrapped command ("ccache g++ ...") names the compiler second
+    if launcher is not None and tokens[:1] == [launcher]:
+        tokens = tokens[1:]
+    if not tokens:
+        # _split_command("") is [] by design, and a command that is only
+        # the launcher strips to nothing; fail like _pick_entry does
+        # instead of an IndexError traceback
+        raise ValueError(f"empty compile command for {entry.get('file')}")
+    if _is_launcher(tokens[0]) and len(tokens) > 1 and not tokens[1].startswith("-"):
+        # A stale compile DB built with a launcher the current run no longer
+        # configures: the real compiler is the next token. Warn: the DB is
+        # stale and worth regenerating.
+        _LOGGER.warning("Stripping unconfigured launcher %s", tokens[0])
+        tokens = tokens[1:]
     # token0 is the compiler path; the rest of the command already uses forward
     # slashes on Windows, so normalize it too for a consistent idedata file.
     cxx_path = tokens[0].replace("\\", "/")
@@ -168,7 +207,7 @@ def _parse_entry(entry: dict) -> tuple[str, list[str], list[str], list[str]]:
     return cxx_path, defines, includes, cxx_flags
 
 
-def _get_toolchain_includes(cxx_path: str) -> list[str]:
+def get_toolchain_includes(cxx_path: str) -> list[str]:
     """Query the compiler for its builtin ``#include <...>`` search dirs."""
     result = subprocess.run(
         [cxx_path, "-E", "-x", "c++", "-", "-v"],
@@ -219,26 +258,114 @@ def _cc_path_from_cxx(cxx_path: str) -> str:
     return f"{stem}{suffix}"
 
 
-def idedata_from_build(compile_commands: Path) -> dict:
+def load_or_build_idedata(
+    compile_commands: Path,
+    elf_path: Path,
+    cache: Path,
+    launcher: str | None = None,
+) -> dict | None:
+    """Return idedata for a compile_commands.json build, cached on mtime.
+
+    Shared by the native ESP-IDF and ESP8266 Arduino toolchains. Returns None
+    when the compile DB doesn't exist yet (nothing was built). ``launcher``
+    is the compiler-launcher path (ccache) the build was generated with, if
+    any; commands in the compile DB are prefixed with it.
+    """
+    if not compile_commands.is_file():
+        _LOGGER.debug("No %s yet; skipping idedata generation", compile_commands)
+        return None
+
+    if cache.is_file() and cache.stat().st_mtime >= compile_commands.stat().st_mtime:
+        try:
+            cached = json.loads(cache.read_text(encoding="utf-8"))
+        except (ValueError, OSError) as err:
+            # A recurring cause (interrupted write, disk full) would otherwise
+            # look like unexplained slow builds
+            _LOGGER.warning("Discarding unreadable idedata cache %s: %s", cache, err)
+        else:
+            # Rebuild pre-cc_path caches on the field, not the timestamp;
+            # the type check keeps "in" from substring-matching a string
+            if isinstance(cached, dict) and "cc_path" in cached:
+                return cached
+
+    data = idedata_from_build(compile_commands, launcher)
+    data["prog_path"] = str(elf_path)
+    cache.parent.mkdir(parents=True, exist_ok=True)
+    # Atomic so a crash mid-write cannot leave a truncated cache
+    write_file(cache, json.dumps(data, indent=2) + "\n")
+    return data
+
+
+def reject_launcher_compiler(cxx_path: str) -> None:
+    """Reject a compile DB that names a launcher (ccache) as the compiler.
+
+    Reject before the toolchain probe, which would fail opaquely on a
+    launcher; the unusable compile DB must never be cached or consumed.
+    """
+    if _is_launcher(cxx_path):
+        raise EsphomeError(
+            f"compile_commands.json names the launcher {cxx_path} as the "
+            "compiler; the compile database is unusable"
+        )
+
+
+def idedata_from_build(compile_commands: Path, launcher: str | None = None) -> dict:
     """Parse compile_commands.json into the idedata fields consumers expect.
 
-    A single ESP-IDF compile entry only carries its own component's REQUIRES
-    include set, but consumers (clang-tidy) analyze ESPHome headers that
-    transitively pull in other components. So take cxx_path / cxx_flags /
-    defines from a representative ESPHome TU, but union the include dirs across
-    all ESPHome TUs to get a project-wide superset (as PlatformIO's idedata
-    provides).
+    A single compile entry only carries the include set its own translation
+    unit was built with (per-component under ESP-IDF), but consumers
+    (clang-tidy) analyze ESPHome headers that transitively pull in other
+    components. So take cxx_path / cxx_flags / defines from a representative
+    ESPHome TU, but union the include dirs across all ESPHome TUs to get a
+    project-wide superset (as PlatformIO's idedata provides).
     """
     entries = json.loads(Path(compile_commands).read_text(encoding="utf-8"))
-    cxx_path, defines, _, cxx_flags = _parse_entry(_pick_entry(entries))
 
-    build_includes: dict[str, None] = {}
+    representative = _pick_entry(entries)
+    cxx_path, defines, rep_includes, cxx_flags = parse_entry(representative, launcher)
+    reject_launcher_compiler(cxx_path)
+
+    # Seed with the representative's includes so it is not parsed twice
+    has_esphome_tu = _is_esphome_src(representative["file"])
+    build_includes: dict[str, None] = dict.fromkeys(
+        rep_includes if has_esphome_tu else ()
+    )
+
+    def _shape(entry: dict) -> str:
+        # The command minus its TU-specific paths: entries sharing a shape
+        # carry identical include sets (one ninja rule), so tokenize once
+        # per shape instead of once per TU. Response-file commands never
+        # dedupe: per-object .rsp names strip to one shape while the files
+        # may hold different include sets.
+        command = entry["command"]
+        if "@" in command:
+            return f"unique:{entry['file']}"
+        return command.replace(entry.get("file", ""), "").replace(
+            entry.get("output", ""), ""
+        )
+
+    seen_shapes = {_shape(representative)}
     for entry in entries:
-        if not _is_esphome_src(entry["file"]):
+        if entry is representative or not _is_esphome_src(entry["file"]):
             continue
-        for inc in _parse_entry(entry)[2]:
+        has_esphome_tu = True
+        if (shape := _shape(entry)) in seen_shapes:
+            _LOGGER.debug("Include union: %s shares a command shape", entry["file"])
+            continue
+        seen_shapes.add(shape)
+        for inc in parse_entry(entry, launcher)[2]:
             build_includes.setdefault(inc, None)
 
+    if not has_esphome_tu:
+        # _pick_entry fell back to an arbitrary C++ entry: idedata built
+        # from it breaks clang-tidy/IDE consumers, and a one-time warning
+        # would be cached into permanence. The best-effort call sites
+        # downgrade this to a build warning.
+        raise EsphomeError(
+            f"No ESPHome translation unit found in {compile_commands}; "
+            "refusing to cache unusable idedata"
+        )
+
     return {
         "cc_path": _cc_path_from_cxx(cxx_path),
         "cxx_path": cxx_path,
@@ -246,6 +373,6 @@ def idedata_from_build(compile_commands: Path) -> dict:
         "defines": defines,
         "includes": {
             "build": list(build_includes),
-            "toolchain": _get_toolchain_includes(cxx_path),
+            "toolchain": get_toolchain_includes(cxx_path),
         },
     }

esphome/build_helpers/ninja.py

@@ -0,0 +1,92 @@
+"""Platform-neutral helpers for ninja-driven native builds."""
+
+from __future__ import annotations
+
+import logging
+import os
+from pathlib import Path
+import re
+import shutil
+
+from esphome.core import EsphomeError
+from esphome.framework_helpers import strip_win_long_path_prefix, tool_version_runs
+
+_LOGGER = logging.getLogger(__name__)
+
+
+def _ninja_runs(binary: str) -> bool:
+    """Whether the ninja found on PATH actually runs (see tool_version_runs)."""
+    return tool_version_runs(
+        binary,
+        "Ignoring ninja at %s because it failed to run; "
+        "falling back to the bundled wheel",
+    )
+
+
+def find_ninja() -> Path:
+    """Locate the ninja binary: a runnable PATH hit first, else the ninja
+    PyPI wheel."""
+    if binary := shutil.which("ninja"):
+        binary = strip_win_long_path_prefix(binary)
+        if _ninja_runs(binary):
+            return Path(binary)
+    import_error: ImportError | None = None
+    try:
+        import ninja
+    except ImportError as err:
+        import_error = err
+        wheel_binary = None
+    else:
+        wheel_binary = Path(ninja.BIN_DIR) / (
+            "ninja.exe" if os.name == "nt" else "ninja"
+        )
+    if wheel_binary is None or not wheel_binary.is_file():
+        raise EsphomeError(
+            "ninja not found on PATH or in the ninja package; reinstall the "
+            "esphome Python environment"
+        ) from import_error
+    return wheel_binary
+
+
+def escape(value: Path | str) -> str:
+    """Escape a path or token for a ninja file."""
+    return str(value).replace("$", "$$").replace(":", "$:").replace(" ", "$ ")
+
+
+def quote_arg(tok: str) -> str:
+    """Quote with the CreateProcess argv rule (as ``subprocess.list2cmdline``):
+    backslash runs double only before a quote. Windows-only; ``$`` must
+    already be doubled for ninja.
+    """
+    quoted = re.sub(r'(\\*)"', lambda m: m.group(1) * 2 + '\\"', tok)
+    quoted = re.sub(r"(\\+)\Z", lambda m: m.group(1) * 2, quoted)
+    return f'"{quoted}"'
+
+
+# Force-quote any token containing a character outside the shlex.quote-style
+# safe set: ninja hands POSIX commands to /bin/sh -c, so bare (, ;, <, *, `
+# and friends would be re-parsed as shell syntax.
+_NEEDS_QUOTE = re.compile(r"[^\w@%+=:,./-]")
+
+
+def shell_token(tok: str, force: bool = False) -> str:
+    """Re-quote a lexed token for the platform shell; ``force`` always quotes.
+
+    Single quotes on POSIX (/bin/sh), the argv rule on Windows
+    (CreateProcess). ``$`` is doubled first because ninja expands it before
+    the command reaches the shell.
+    """
+    tok = tok.replace("$", "$$")  # ninja would expand a bare $ to nothing
+    if not (force or not tok or _NEEDS_QUOTE.search(tok)):
+        return tok
+    # An empty token must become '' / "" or it vanishes from the argv
+    if os.name == "nt":
+        return quote_arg(tok)
+    # shlex.quote's rule; inlined because the $-doubled token must not be
+    # re-examined for safe characters
+    return "'" + tok.replace("'", "'\"'\"'") + "'"
+
+
+def quote_path(value: Path | str) -> str:
+    """Force-quote a path for the ninja command line (shell/CreateProcess)."""
+    return shell_token(str(value), force=True)

esphome/build_helpers/size_summary.py

@@ -0,0 +1,24 @@
+"""The PlatformIO-format size bar shared by the native toolchains."""
+
+from __future__ import annotations
+
+
+def format_bar(used: int, total: int) -> str:
+    """Match PlatformIO's ``_format_availale_bytes`` (sic, pioupload.py) exactly."""
+    pct_raw = used / total if total else 0
+    blocks = 10
+    filled = min(int(round(blocks * pct_raw)), blocks)
+    progress = "=" * filled
+    return (
+        f"[{progress:<{blocks}}] {pct_raw: 6.1%} "
+        f"(used {used:d} bytes from {total:d} bytes)"
+    )
+
+
+def print_size_line(label: str, used: int, total: int) -> None:
+    """One PlatformIO-format summary line (``RAM``/``Flash``).
+
+    The label padding is part of the format: ``script/ci_memory_impact_extract.py``
+    matches these lines verbatim.
+    """
+    print(f"{label + ':':<7}{format_bar(used, total)}")

esphome/build_helpers/tools_cache.py

@@ -0,0 +1,36 @@
+"""Machine-global tools cache location shared by the native backends."""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+
+def tools_cache_path(env_var: str, subdir: str) -> Path:
+    """A backend's machine-global tools directory, with an env override.
+
+    A blank/whitespace override is treated as unset: ``Path("")`` resolves
+    to the CWD, which ``clean-all`` would then delete.
+    """
+    import platformdirs
+
+    from esphome.helpers import get_str_env
+
+    if prefix := get_str_env(env_var, "").strip():
+        # resolve(): symlinked prefixes otherwise trip idf.py's
+        # venv-mismatch warning on every build
+        return Path(prefix).expanduser().resolve()
+    # appauthor=False keeps the Windows path short (no vendor segment);
+    # deep IDF trees run into MAX_PATH otherwise
+    return (
+        Path(platformdirs.user_cache_dir("esphome", appauthor=False)) / subdir
+    ).resolve()
+
+
+# (env override, cache subdir) per native backend. writer.clean_all wipes
+# every entry via tools_cache_path, so listing a cache here is the single
+# step that registers it for removal; the backends' own path getters use
+# the same named pairs so the two cannot drift.
+IDF_TOOLS_CACHE = ("ESPHOME_ESP_IDF_PREFIX", "idf")
+SDK_NRF_TOOLS_CACHE = ("ESPHOME_SDK_NRF_PREFIX", "sdk-nrf")
+ARDUINO8266_TOOLS_CACHE = ("ESPHOME_ARDUINO8266_PREFIX", "arduino8266")
+TOOLS_CACHE_SPECS = (IDF_TOOLS_CACHE, SDK_NRF_TOOLS_CACHE, ARDUINO8266_TOOLS_CACHE)

esphome/compiled_config.py

@@ -100,6 +100,21 @@ def _refresh_sidecar() -> bool:
             )
             return False
         if old is not None and old.can_apply_to_core():
+            if (
+                old.toolchain is not None
+                and CORE.toolchain is not None
+                and old.toolchain != CORE.toolchain.value
+            ):
+                # Platforms normalize toolchain-sensitive keys differently;
+                # never cache a config validated under a different toolchain
+                # than the compile's
+                _LOGGER.debug(
+                    "Not caching: config validated with toolchain %r but the "
+                    "last compile used %r",
+                    CORE.toolchain.value,
+                    old.toolchain,
+                )
+                return False
             # Compile-written; nothing to refresh.
             return True
         if CORE.build_path is not None and CORE.build_path.exists():

esphome/components/esp32/__init__.py

@@ -1072,19 +1072,11 @@ def _check_esp_idf_versions(config: ConfigType) -> ConfigType:
     return config
 
 
-def _validate_toolchain(value) -> Toolchain:
-    return Toolchain(
-        cv.one_of(Toolchain.PLATFORMIO, Toolchain.ESP_IDF, lower=True)(value)
-    )
-
-
-def _resolve_toolchain(value: ConfigType) -> ConfigType:
-    # Resolve toolchain: CLI (already on CORE.toolchain) > YAML > default.
-    # Runs before _detect_variant so downstream validators can rely on
-    # CORE.toolchain instead of re-resolving it from the config dict.
-    if CORE.toolchain is None:
-        CORE.toolchain = value.get(CONF_TOOLCHAIN, Toolchain.ESP_IDF)
-    return value
+_TOOLCHAINS = (Toolchain.PLATFORMIO, Toolchain.ESP_IDF)
+_validate_toolchain = cv.toolchain_enum(_TOOLCHAINS)
+# Runs before _detect_variant so downstream validators can rely on
+# CORE.toolchain instead of re-resolving it from the config dict.
+_resolve_toolchain = cv.resolve_toolchain("ESP32", _TOOLCHAINS, Toolchain.ESP_IDF)
 
 
 def _check_versions(config: ConfigType) -> ConfigType:

esphome/components/esp8266/__init__.py

@@ -246,6 +246,9 @@ CONFIG_SCHEMA = cv.All(
             cv.Optional(CONF_ENABLE_SCANF_FLOAT): cv.boolean,
         }
     ),
+    # Until the native toolchain lands, PlatformIO is the only backend;
+    # reject a --toolchain this platform cannot serve yet.
+    cv.require_platformio_toolchain("ESP8266"),
     set_core_data,
 )
 
@@ -397,8 +400,8 @@ async def to_code(config: ConfigType) -> None:
     )
 
     if config[CONF_BOARD] in BOARDS:
-        flash_size = BOARDS[config[CONF_BOARD]][KEY_FLASH_SIZE]
-        ld_scripts = ESP8266_LD_SCRIPTS[flash_size]
+        board_data = BOARDS[config[CONF_BOARD]]
+        ld_scripts = ESP8266_LD_SCRIPTS[board_data[KEY_FLASH_SIZE]]
 
         if ver <= cv.Version(2, 3, 0):
             # No ld script support
@@ -407,7 +410,9 @@ async def to_code(config: ConfigType) -> None:
             # Old ld script path
             ld_script = ld_scripts[0]
         else:
-            ld_script = ld_scripts[1]
+            # A per-board override preserves a layout the board shipped
+            # with (see d1_wroom_02 in boards.py)
+            ld_script = board_data.get("ldscript", ld_scripts[1])
 
         if ld_script is not None:
             cg.add_platformio_option("board_build.ldscript", ld_script)

esphome/components/esp8266/boards.py

@@ -199,6 +199,15 @@ BOARDS = {
         "name": "WeMos D1 mini Pro",
         "flash_size": FLASH_SIZE_16_MB,
     },
+    "d1_wroom_02": {
+        "name": "WeMos D1 ESP-WROOM-02",
+        "flash_size": FLASH_SIZE_2_MB,
+        # This board joined BOARDS after shipping with the manifest default
+        # (64 KB filesystem region); the flash-size default (2m.ld) would
+        # move _FS_end and with it the preferences sector, wiping existing
+        # devices' flash-backed state on update.
+        "ldscript": "eagle.flash.2m64.ld",
+    },
     "d1": {
         "name": "WEMOS D1 R1",
         "flash_size": FLASH_SIZE_4_MB,
@@ -360,3 +369,112 @@ BOARDS = {
         "flash_size": FLASH_SIZE_4_MB,
     },
 }
+
+
+# Per-board variant dir + identity defines from platform-espressif8266 4.x
+# build.extra_flags; the shared -DESP8266/-DARDUINO_ARCH_ESP8266 are added
+# by the generator.
+#
+# Regenerate ESP8266_BOARD_BUILD with (v4.2.1 is the platform version the
+# native toolchain mirrors; regenerate against the tag when bumping it):
+#
+# git clone -b v4.2.1 https://github.com/platformio/platform-espressif8266
+# python3 - <<'EOF'
+# import json, glob, os
+# for f in sorted(glob.glob("platform-espressif8266/boards/*.json")):
+#     b = json.load(open(f))["build"]
+#     extra = b["extra_flags"]
+#     extra = extra.split() if isinstance(extra, str) else extra
+#     defines = [
+#         e[2:] for e in extra if e not in ("-DESP8266", "-DARDUINO_ARCH_ESP8266")
+#     ]
+#     entries = ", ".join(f'"{d}"' for d in defines) + ("," if len(defines) == 1 else "")
+#     board = os.path.splitext(os.path.basename(f))[0]
+#     print(f'    "{board}": {{"variant": "{b["variant"]}", "defines": ({entries})}},')
+# EOF
+ESP8266_BOARD_BUILD = {
+    "agruminolemon": {
+        "variant": "agruminolemonv4",
+        "defines": ("ARDUINO_ESP8266_AGRUMINO_LEMON_V4",),
+    },
+    "d1": {"variant": "d1", "defines": ("ARDUINO_ESP8266_WEMOS_D1R1",)},
+    "d1_mini": {"variant": "d1_mini", "defines": ("ARDUINO_ESP8266_WEMOS_D1MINI",)},
+    "d1_mini_lite": {
+        "variant": "d1_mini",
+        "defines": ("ARDUINO_ESP8266_WEMOS_D1MINILITE",),
+    },
+    "d1_mini_pro": {
+        "variant": "d1_mini",
+        "defines": ("ARDUINO_ESP8266_WEMOS_D1MINIPRO",),
+    },
+    "d1_wroom_02": {
+        "variant": "d1_mini",
+        "defines": ("ARDUINO_ESP8266_WEMOS_D1WROOM02",),
+    },
+    "eduinowifi": {
+        "variant": "eduinowifi",
+        "defines": ("ARDUINO_ESP8266_SCHIRMILABS_EDUINO_WIFI",),
+    },
+    "esp01": {"variant": "generic", "defines": ("ARDUINO_ESP8266_ESP01",)},
+    "esp01_1m": {"variant": "generic", "defines": ("ARDUINO_ESP8266_ESP01",)},
+    "esp07": {"variant": "generic", "defines": ("ARDUINO_ESP8266_ESP07",)},
+    "esp07s": {"variant": "nodemcu", "defines": ("ARDUINO_ESP8266_ESP07",)},
+    "esp12e": {"variant": "nodemcu", "defines": ("ARDUINO_ESP8266_ESP12",)},
+    "esp210": {"variant": "generic", "defines": ("ARDUINO_ESP8266_ESP210",)},
+    "esp8285": {"variant": "esp8285", "defines": ("ARDUINO_ESP8266_ESP01",)},
+    "esp_wroom_02": {
+        "variant": "nodemcu",
+        "defines": ("ARDUINO_ESP8266_ESP_WROOM_02",),
+    },
+    "espduino": {"variant": "ESPDuino", "defines": ("ARDUINO_ESP8266_ESP13",)},
+    "espectro": {"variant": "espectro", "defines": ("ARDUINO_ESP8266_ESPECTRO_CORE",)},
+    "espino": {"variant": "espino", "defines": ("ARDUINO_ESP8266_ESP12",)},
+    "espinotee": {"variant": "espinotee", "defines": ("ARDUINO_ESP8266_ESP13",)},
+    "espmxdevkit": {
+        "variant": "esp8285",
+        "defines": ("ARDUINO_ESP8266_ESP01", "LED_BUILTIN=16"),
+    },
+    "espresso_lite_v1": {
+        "variant": "espresso_lite_v1",
+        "defines": ("ARDUINO_ESP8266_ESPRESSO_LITE_V1",),
+    },
+    "espresso_lite_v2": {
+        "variant": "espresso_lite_v2",
+        "defines": ("ARDUINO_ESP8266_ESPRESSO_LITE_V2",),
+    },
+    "gen4iod": {"variant": "generic", "defines": ("ARDUINO_GEN4_IOD",)},
+    "heltec_wifi_kit_8": {
+        "variant": "wifi_kit_8",
+        "defines": ("ARDUINO_wifi_kit_8",),
+    },
+    "huzzah": {"variant": "adafruit", "defines": ("ARDUINO_ESP8266_ADAFRUIT_HUZZAH",)},
+    "inventone": {"variant": "inventone", "defines": ("ARDUINO_ESP8266_INVENT_ONE",)},
+    "modwifi": {"variant": "generic", "defines": ("ARDUINO_MOD_WIFI_ESP8266",)},
+    "nodemcu": {"variant": "nodemcu", "defines": ("ARDUINO_ESP8266_NODEMCU",)},
+    "nodemcuv2": {"variant": "nodemcu", "defines": ("ARDUINO_ESP8266_NODEMCU_ESP12E",)},
+    "oak": {"variant": "oak", "defines": ("ARDUINO_ESP8266_OAK",)},
+    "phoenix_v1": {
+        "variant": "phoenix_v1",
+        "defines": ("ARDUINO_ESP8266_PHOENIX_V1",),
+    },
+    "phoenix_v2": {
+        "variant": "phoenix_v2",
+        "defines": ("ARDUINO_ESP8266_PHOENIX_V2",),
+    },
+    "sonoff_basic": {"variant": "itead", "defines": ("ARDUINO_ESP8266_SONOFF_BASIC",)},
+    "sonoff_s20": {"variant": "itead", "defines": ("ARDUINO_ESP8266_SONOFF_S20",)},
+    "sonoff_sv": {"variant": "itead", "defines": ("ARDUINO_ESP8266_SONOFF_SV",)},
+    "sonoff_th": {"variant": "itead", "defines": ("ARDUINO_ESP8266_SONOFF_TH",)},
+    "sparkfunBlynk": {"variant": "thing", "defines": ("ARDUINO_ESP8266_THING",)},
+    "thing": {"variant": "thing", "defines": ("ARDUINO_ESP8266_THING",)},
+    "thingdev": {"variant": "thing", "defines": ("ARDUINO_ESP8266_THING_DEV",)},
+    "wifi_slot": {"variant": "wifi_slot", "defines": ("ARDUINO_AMPERKA_WIFI_SLOT",)},
+    "wifiduino": {"variant": "wifiduino", "defines": ("ARDUINO_WIFIDUINO_ESP8266",)},
+    "wifinfo": {"variant": "wifinfo", "defines": ("ARDUINO_WIFINFO",)},
+    "wio_link": {"variant": "wiolink", "defines": ("ARDUINO_ESP8266_WIO_LINK",)},
+    "wio_node": {"variant": "nodemcu", "defines": ("ARDUINO_ESP8266_ESP_WROOM_02",)},
+    "xinabox_cw01": {
+        "variant": "xinabox",
+        "defines": ("ARDUINO_ESP8266_XINABOX_CW01",),
+    },
+}

esphome/components/esp8266/build_surgery.py

@@ -0,0 +1,118 @@
+"""Linker-script surgery shared with the native (PlatformIO-free) toolchain.
+
+These mirror the PlatformIO extra scripts in this directory
+(``relocate_ratetable.py.script`` and ``testing_mode.py.script``), which run
+inside SCons and must stay self-contained. The native build generator applies
+the same patches to the linker scripts it generates, so the logic lives here
+as plain functions. Keep both in sync when changing either.
+``segment_length`` is native-toolchain-only and has no script twin.
+"""
+
+from __future__ import annotations
+
+from collections.abc import Collection
+import hashlib
+import re
+
+# Move the NONOS SDK wifi rate tables from flash to DRAM; see
+# relocate_ratetable.py.script for the full background (NONOS SDK issue 320).
+RATETABLE_RULE = "*libnet80211.a:ieee80211_phy.o(.irom.text .irom.text.*)"
+_RATETABLE_COMMENT = (
+    "/* ESPHome: wifi rate tables must live in DRAM, see NONOS SDK issue 320 */"
+)
+# Match the whole line: "_data_start" is also a substring of the
+# "_dport0_data_start" line in the earlier .dport0.data section
+_RATETABLE_ANCHOR = re.compile(r"^\s*_data_start = ABSOLUTE\(\.\);", re.MULTILINE)
+
+# Memory sizes for testing mode (allow larger builds for CI component grouping)
+TESTING_IRAM_SIZE = "0x200000"  # 2MB
+TESTING_DRAM_SIZE = "0x200000"  # 2MB
+TESTING_FLASH_SIZE = "0x2000000"  # 32MB
+
+
+def relocate_ratetable(content: str) -> str:
+    """Insert the rate-table DRAM rule into a generated common linker script."""
+    if RATETABLE_RULE in content:
+        return content
+    match = _RATETABLE_ANCHOR.search(content)
+    if match is None:
+        raise RuntimeError(
+            "'_data_start' anchor not found in the generated linker script; "
+            "cannot apply wifi rate table DRAM relocation "
+            "(has the Arduino core linker script changed?)"
+        )
+    insert_pos = match.end()
+    return (
+        content[:insert_pos]
+        + f"\n    {_RATETABLE_COMMENT}"
+        + f"\n    {RATETABLE_RULE}"
+        + content[insert_pos:]
+    )
+
+
+_TESTING_SEGMENT_SIZES = {
+    "iram1_0_seg": TESTING_IRAM_SIZE,
+    "dram0_0_seg": TESTING_DRAM_SIZE,
+    "irom0_0_seg": TESTING_FLASH_SIZE,
+}
+
+
+def _segment_line_re(segment_name: str) -> re.Pattern[str]:
+    """The MEMORY line for one segment: ``<seg> : org = 0x..., len = 0x...``.
+
+    Anchored to the start of the line so a name never matches inside a
+    longer one (``ram0_0_seg`` must not read ``dram0_0_seg``). The size
+    group stops at the hex digits, leaving any ``ul`` suffix (from the
+    preprocessed ``MMU_IRAM_SIZE``) in place.
+    """
+    return re.compile(
+        rf"(^[ \t]*{re.escape(segment_name)}"
+        r"\s*:\s*org\s*=\s*0x[0-9a-fA-F]+\s*,\s*len\s*=\s*)"
+        r"(0x[0-9a-fA-F]+)",
+        re.MULTILINE,
+    )
+
+
+def apply_testing_memory_patches(content: str, segments: Collection[str]) -> str:
+    """Enlarge the named memory segments so grouped CI test builds can link.
+
+    Each caller passes the segments its linker script defines: the
+    generated common ld carries ``iram1_0_seg``; the flash ld carries
+    ``dram0_0_seg`` and ``irom0_0_seg``. A segment that fails to match
+    raises, since a silently kept real memory limit would fail grouped
+    builds far from the cause.
+    """
+    for segment in _TESTING_SEGMENT_SIZES:
+        if segment not in segments and _segment_line_re(segment).search(content):
+            raise RuntimeError(
+                f"Testing-mode segment {segment} is present in the linker "
+                "script but was not selected for patching"
+            )
+    for segment in segments:
+        if segment not in _TESTING_SEGMENT_SIZES:
+            raise RuntimeError(f"Unknown testing-mode segment {segment!r}")
+        content, count = _segment_line_re(segment).subn(
+            rf"\g<1>{_TESTING_SEGMENT_SIZES[segment]}", content
+        )
+        if count == 0:
+            raise RuntimeError(
+                f"Testing-mode memory patch failed: segment {segment} "
+                "not found (has the Arduino core linker script changed?)"
+            )
+    return content
+
+
+def segment_length(content: str, segment_name: str) -> int | None:
+    """Read a memory segment's length from linker script content."""
+    match = _segment_line_re(segment_name).search(content)
+    return int(match.group(2), 16) if match else None
+
+
+def surgery_fingerprint() -> str:
+    """Hash of this module's source; linker-script caches include it so an
+    edit here invalidates them."""
+    import inspect
+    import sys
+
+    source = inspect.getsource(sys.modules[__name__])
+    return hashlib.sha256(source.encode()).hexdigest()

esphome/components/esp8266/const.py

@@ -15,6 +15,9 @@ CONF_ENABLE_SERIAL1 = "enable_serial1"
 KEY_WAVEFORM_REQUIRED = "waveform_required"
 KEY_SERIAL_REQUIRED = "serial_required"
 KEY_SERIAL1_REQUIRED = "serial1_required"
+# Set for the native (non-PlatformIO) toolchain's build generator
+KEY_FLASH_MODE = "flash_mode"
+KEY_SCANF_FLOAT = "scanf_float"
 
 # esp8266 namespace is already defined by arduino, manually prefix esphome
 esp8266_ns = cg.global_ns.namespace("esphome").namespace("esp8266")

esphome/components/host/__init__.py

@@ -37,6 +37,7 @@ CONFIG_SCHEMA = cv.All(
             cv.Optional(CONF_MAC_ADDRESS, default="98:35:69:ab:f6:79"): cv.mac_address,
         }
     ),
+    cv.require_platformio_toolchain("host"),
     set_core_data,
 )
 

esphome/components/libretiny/__init__.py

@@ -300,7 +300,7 @@ FRAMEWORK_SCHEMA = cv.All(
     _check_debug_order,
 )
 
-CONFIG_SCHEMA = cv.All(_notify_old_style)
+CONFIG_SCHEMA = cv.All(_notify_old_style, cv.require_platformio_toolchain("LibreTiny"))
 
 BASE_SCHEMA = cv.Schema(
     {
@@ -314,6 +314,7 @@ BASE_SCHEMA = cv.Schema(
 )
 
 BASE_SCHEMA.add_extra(_detect_variant)
+BASE_SCHEMA.add_extra(cv.require_platformio_toolchain("LibreTiny"))
 BASE_SCHEMA.add_extra(_update_core_data)
 
 

esphome/components/nrf52/__init__.py

@@ -125,10 +125,8 @@ def set_core_data(config: ConfigType) -> ConfigType:
     return config
 
 
-def _resolve_toolchain(config: ConfigType) -> ConfigType:
-    if CORE.toolchain is None:
-        CORE.toolchain = config.get(CONF_TOOLCHAIN, Toolchain.SDK_NRF)
-    return config
+_TOOLCHAINS = (Toolchain.PLATFORMIO, Toolchain.SDK_NRF)
+_resolve_toolchain = cv.resolve_toolchain("nRF52", _TOOLCHAINS, Toolchain.SDK_NRF)
 
 
 def set_framework(config: ConfigType) -> ConfigType:
@@ -170,10 +168,7 @@ BOOTLOADERS = [
 ]
 
 
-def _validate_toolchain(value) -> Toolchain:
-    return Toolchain(
-        cv.one_of(Toolchain.PLATFORMIO, Toolchain.SDK_NRF, lower=True)(value)
-    )
+_validate_toolchain = cv.toolchain_enum(_TOOLCHAINS)
 
 
 def _detect_bootloader(config: ConfigType) -> ConfigType:

esphome/components/nrf52/framework.py

@@ -7,8 +7,7 @@ import shutil
 import sys
 import tempfile
 
-import platformdirs
-
+from esphome.build_helpers.tools_cache import SDK_NRF_TOOLS_CACHE, tools_cache_path
 import esphome.config_validation as cv
 from esphome.const import KEY_CORE, KEY_FRAMEWORK_VERSION
 from esphome.core import CORE, EsphomeError
@@ -21,7 +20,6 @@ from esphome.framework_helpers import (
     run_command_ok,
     str_to_lst_of_str,
 )
-from esphome.helpers import get_str_env
 
 _LOGGER = logging.getLogger(__name__)
 
@@ -51,15 +49,9 @@ SDK_NG_MINIMAL_MIRRORS = str_to_lst_of_str(
 
 
 def get_sdk_nrf_tools_path() -> Path:
-    # A blank ESPHOME_SDK_NRF_PREFIX must be treated as unset: Path("")
-    # resolves to the CWD, which clean-all would then delete.
-    if prefix := get_str_env("ESPHOME_SDK_NRF_PREFIX", "").strip():
-        path = Path(prefix).expanduser()
-    else:
-        # Machine-global (OS user cache dir) so all projects share one install;
-        # see espidf.framework.get_idf_tools_path for the location rationale.
-        path = Path(platformdirs.user_cache_dir("esphome", appauthor=False)) / "sdk-nrf"
-    return path.resolve()
+    # Machine-global (OS user cache dir) so all projects share one install;
+    # see espidf.framework.get_idf_tools_path for the location rationale.
+    return tools_cache_path(*SDK_NRF_TOOLS_CACHE)
 
 
 def _needs_venv_rebuild(

esphome/components/rp2/__init__.py

@@ -312,6 +312,7 @@ CONFIG_SCHEMA = cv.All(
     ),
     cv.has_at_least_one_key(CONF_BOARD, CONF_VARIANT),
     _detect_variant,
+    cv.require_platformio_toolchain("RP2"),
     set_core_data,
 )
 

esphome/config_validation.py

@@ -53,6 +53,7 @@ from esphome.const import (
     CONF_SETUP_PRIORITY,
     CONF_STATE_TOPIC,
     CONF_SUBSCRIBE_QOS,
+    CONF_TOOLCHAIN,
     CONF_TOPIC,
     CONF_TYPE,
     CONF_TYPE_ID,
@@ -75,6 +76,7 @@ from esphome.const import (
     TYPE_GIT,
     TYPE_LOCAL,
     Framework,
+    Toolchain,
     __version__ as ESPHOME_VERSION,
 )
 from esphome.core import (
@@ -91,7 +93,13 @@ from esphome.core import (
 )
 from esphome.enum import StrEnum
 from esphome.expression import SUBSTITUTION_VARIABLE_PROG as VARIABLE_PROG
-from esphome.helpers import add_class_to_obj, docs_url, list_starts_with
+from esphome.helpers import (
+    FALSY_BOOL_STRINGS,
+    TRUTHY_BOOL_STRINGS,
+    add_class_to_obj,
+    docs_url,
+    list_starts_with,
+)
 from esphome.schema_extractors import (
     SCHEMA_EXTRACT,
     schema_extractor,
@@ -106,6 +114,9 @@ from esphome.util import parse_esphome_version  # noqa: F401
 from esphome.voluptuous_schema import _Schema
 from esphome.yaml_util import SensitiveStr, make_data_base
 
+if typing.TYPE_CHECKING:
+    from esphome.types import ConfigType
+
 _LOGGER = logging.getLogger(__name__)
 
 # pylint: disable=invalid-name
@@ -576,9 +587,9 @@ def boolean(value):
         return value
     if isinstance(value, str):
         value = value.lower()
-        if value in ("true", "yes", "on", "enable"):
+        if value in TRUTHY_BOOL_STRINGS:
             return True
-        if value in ("false", "no", "off", "disable"):
+        if value in FALSY_BOOL_STRINGS:
             return False
     raise Invalid(
         f"Expected boolean value, but cannot convert {value} to a boolean. Please use 'true' or 'false'"
@@ -2532,6 +2543,63 @@ def platformio_version_constraint(value):
     return constraints
 
 
+def _check_supported_toolchain(
+    platform_name: str, supported: tuple[Toolchain, ...]
+) -> None:
+    """Raise when the resolved ``CORE.toolchain`` is not in ``supported``
+    (one message shape for every platform)."""
+    toolchain = CORE.toolchain
+    if toolchain is None:
+        # A caller ran the check before resolving; an ordering bug, not a
+        # user error
+        raise Invalid(f"Toolchain was not resolved before {platform_name} validation")
+    if toolchain not in supported:
+        names = ", ".join(f"'{tc.value}'" for tc in supported)
+        raise Invalid(
+            f"Unsupported toolchain "
+            f"'{toolchain.value}' for "
+            f"{platform_name}. Supported: {names}."
+        )
+
+
+def toolchain_enum(supported: tuple[Toolchain, ...]) -> Callable[[str], Toolchain]:
+    """Schema validator for a platform's ``toolchain`` config key."""
+
+    def validator(value: str) -> Toolchain:
+        return Toolchain(one_of(*supported, lower=True)(value))
+
+    return validator
+
+
+def resolve_toolchain(
+    platform_name: str, supported: tuple[Toolchain, ...], default: Toolchain
+) -> Callable[[ConfigType], ConfigType]:
+    """Resolve ``CORE.toolchain`` (CLI > YAML > default) and reject one the
+    platform cannot serve.
+
+    Add to the platform's validation chain before anything that reads
+    ``CORE.toolchain``.
+    """
+
+    def validator(config: ConfigType) -> ConfigType:
+        if CORE.toolchain is None:
+            CORE.toolchain = config.get(CONF_TOOLCHAIN, default)
+        _check_supported_toolchain(platform_name, supported)
+        return config
+
+    return validator
+
+
+def require_platformio_toolchain(
+    platform_name: str,
+) -> Callable[[ConfigType], ConfigType]:
+    """Reject a CLI-selected toolchain other than PlatformIO, for platforms
+    with only the PlatformIO backend."""
+    return resolve_toolchain(
+        platform_name, (Toolchain.PLATFORMIO,), Toolchain.PLATFORMIO
+    )
+
+
 def require_framework_version(
     *,
     max_version=False,

esphome/const.py

@@ -21,6 +21,14 @@ class Toolchain(StrEnum):
     PLATFORMIO = "platformio"
     ESP_IDF = "esp-idf"
     SDK_NRF = "sdk-nrf"
+    # ESP8266: the Arduino core built directly (no PlatformIO)
+    ARDUINO = "arduino"
+
+
+# Toolchains that drive their build natively and never read platformio.ini.
+# SDK_NRF is absent on purpose: the zephyr backend keeps consuming
+# platformio_options.
+NATIVE_TOOLCHAINS = frozenset({Toolchain.ESP_IDF, Toolchain.ARDUINO})
 
 
 class Platform(StrEnum):

esphome/core/__init__.py

@@ -21,6 +21,7 @@ from esphome.const import (
     KEY_CORE,
     KEY_TARGET_FRAMEWORK,
     KEY_TARGET_PLATFORM,
+    NATIVE_TOOLCHAINS,
     PLATFORM_BK72XX,
     PLATFORM_ESP32,
     PLATFORM_ESP8266,
@@ -982,6 +983,19 @@ class EsphomeCore:
     def using_toolchain_sdk_nrf(self):
         return self.toolchain == Toolchain.SDK_NRF
 
+    @property
+    def using_toolchain_arduino(self):
+        """The native ESP8266 Arduino build toolchain (unlike
+        ``using_arduino``, which is the target framework)."""
+        return self.toolchain == Toolchain.ARDUINO
+
+    @property
+    def using_native_toolchain(self):
+        """Whether the selected toolchain builds natively, without reading
+        ``platformio.ini`` (see ``NATIVE_TOOLCHAINS`` in ``esphome.const``;
+        keep its membership in sync with ``write_cpp_file``'s dispatch)."""
+        return self.toolchain in NATIVE_TOOLCHAINS
+
     @property
     def using_zephyr(self):
         return self.target_framework == "zephyr"
@@ -1095,6 +1109,8 @@ class EsphomeCore:
         return build_flag
 
     def add_build_unflag(self, build_unflag: str) -> None:
+        # No warning for using_toolchain_arduino: the native ESP8266 build
+        # honors build_unflags (token-level, matching PlatformIO).
         if self.using_toolchain_esp_idf:
             # The native ESP-IDF build generator does not consume build_unflags
             _LOGGER.warning(

esphome/core/config.py

@@ -555,12 +555,24 @@ def _add_library_str(lib: str) -> None:
         cg.add_library(lib, None)
 
 
+# platformio_options keys the native ESP8266 Arduino generator (a later PR
+# in this chain) will honor; its ignored-option warning will consume the same
+# list so the two cannot drift
+NATIVE_ARDUINO_PIO_OPTIONS = frozenset({"board_build.f_cpu", "board_build.ldscript"})
+# The full set that survives into CORE.platformio_options under the native
+# arduino toolchain: lib_ignore is the only specially-translated key below
+# that is stored rather than translated away. Consumed by the esp8266 native
+# backend (later in this chain) for its ignored-option warning; defined here
+# so it stays adjacent to the routing.
+NATIVE_ARDUINO_CONSUMED_PIO_OPTIONS = NATIVE_ARDUINO_PIO_OPTIONS | {"lib_ignore"}
+
+
 @coroutine_with_priority(CoroPriority.FINAL)
 async def _add_platformio_options(pio_options: dict[str, str | list[str]]) -> None:
-    if CORE.using_toolchain_esp_idf:
-        # The native ESP-IDF build doesn't read platformio.ini; honor the
-        # options with a native equivalent and warn about the rest, which
-        # would otherwise be silently ignored.
+    if CORE.using_native_toolchain:
+        # The native builds don't read platformio.ini; honor the options
+        # with a native equivalent and warn about the rest, which would
+        # otherwise be silently ignored.
         for key, val in pio_options.items():
             vals = [val] if isinstance(val, str) else val
             if key == CONF_BUILD_FLAGS:
@@ -573,23 +585,41 @@ async def _add_platformio_options(pio_options: dict[str, str | list[str]]) -> No
                 )
                 for flag in vals:
                     cg.add_build_flag(flag)
+            elif key == "build_unflags":
+                # Native equivalent: add_build_unflag (honored token-level by
+                # the arduino generator; the IDF generator warns there)
+                for flag in vals:
+                    CORE.add_build_unflag(flag)
             elif key == "lib_deps":
-                # Routed through the regular library mechanism so the libraries
-                # are converted to IDF components like any other PIO library
+                # Routed through the regular library mechanism so the
+                # libraries reach the native backend's converter (IDF
+                # components, or the ESP8266 native library resolution)
                 for lib in vals:
                     _add_library_str(lib)
             elif key == "lib_ignore":
-                # Read by the PIO-library-to-IDF-component conversion
-                # (generate_idf_components); filters both top-level libraries
-                # and dependencies discovered during conversion
+                # Read by the shared library conversion (lib_ignore_set in
+                # platformio/library.py); filters top-level libraries and
+                # discovered dependencies
                 cg.add_platformio_option(key, vals)
+            elif (
+                key in NATIVE_ARDUINO_PIO_OPTIONS
+                and CORE.using_toolchain_arduino
+                and vals
+            ):
+                # The esp8266 native generator reads these as scalars; the
+                # schema also permits the list form, where the last value
+                # wins like a later platformio.ini line (an empty list falls
+                # through to the ignored-option warning). Other native
+                # toolchains have no equivalent and fall through too.
+                cg.add_platformio_option(key, vals[-1])
             elif key != "upload_speed":
                 # upload_speed needs no handling: it is read from the raw
                 # config at upload time (upload_using_esptool)
                 _LOGGER.warning(
                     "esphome->platformio_options->%s is ignored when building with "
-                    "the native ESP-IDF toolchain",
+                    "the native '%s' toolchain",
                     key,
+                    CORE.toolchain.value,
                 )
         return
     # Add includes at the very end, so that they override everything

esphome/espidf/clang_tidy.py

@@ -23,6 +23,12 @@ from dataclasses import dataclass
 import os
 from pathlib import Path
 
+from esphome.build_helpers.idedata import (
+    get_toolchain_includes,
+    parse_entry,
+    reject_launcher_compiler,
+)
+
 TIDY_PROJECT_NAME = "esphome_tidy"
 
 # A do-nothing C++ app: just enough for IDF to configure a valid project. It's
@@ -415,13 +421,12 @@ def _idedata_from_tidy_project(compile_commands: Path) -> dict:
     """
     import json
 
-    from esphome.espidf.idedata import _get_toolchain_includes, _parse_entry
-
     entries = json.loads(Path(compile_commands).read_text(encoding="utf-8"))
     entry = next((e for e in entries if e["file"].endswith("tidy.cpp")), None)
     if entry is None:
         raise RuntimeError(f"tidy.cpp not found in {compile_commands}")
-    cxx_path, defines, includes, cxx_flags = _parse_entry(entry)
+    cxx_path, defines, includes, cxx_flags = parse_entry(entry)
+    reject_launcher_compiler(cxx_path)
 
     return {
         "cxx_path": cxx_path,
@@ -429,7 +434,7 @@ def _idedata_from_tidy_project(compile_commands: Path) -> dict:
         "defines": defines,
         "includes": {
             "build": includes,
-            "toolchain": _get_toolchain_includes(cxx_path),
+            "toolchain": get_toolchain_includes(cxx_path),
         },
     }
 

esphome/espidf/component.py

@@ -27,6 +27,7 @@ from esphome.platformio.library import (
     collect_filtered_files,
     convert_libraries,
     ensure_list,
+    lex_build_flags,
     split_list_by_condition,
 )
 
@@ -40,37 +41,6 @@ def _idf_framework() -> str:
     return "arduino" if CORE.using_arduino else "espidf"
 
 
-def _apply_extra_script(component: IDFComponent) -> None:
-    """Run a PIO ``extraScript`` and fold its captured env vars into
-    ``component.data["build"]["flags"]`` so the existing -L/-l/-D
-    extraction in ``generate_cmakelists_txt`` picks them up."""
-    extra_script = component.data.get("build", {}).get("extraScript")
-    if not extra_script:
-        return
-    # Resolve and confine to the library's source dir so a malicious
-    # library.json can't escape (e.g. ``"extraScript": "../../etc/passwd"``).
-    source_path = component.source_dir
-    library_root = source_path.resolve()
-    script_path = (source_path / extra_script).resolve()
-    if not script_path.is_relative_to(library_root) or not script_path.is_file():
-        return
-    from esphome.components.esp32 import get_esp32_variant
-    from esphome.espidf.extra_script import captured_as_build_flags, run_extra_script
-
-    idf_target = variant_to_idf_target(get_esp32_variant())
-    result = run_extra_script(
-        script_path, library_dir=source_path, idf_target=idf_target
-    )
-    extra_flags = captured_as_build_flags(result, library_dir=source_path)
-    if not extra_flags:
-        return
-    flags = component.data.setdefault("build", {}).setdefault("flags", [])
-    if isinstance(flags, str):
-        flags = [flags]
-    flags.extend(extra_flags)
-    component.data["build"]["flags"] = flags
-
-
 def generate_cmakelists_txt(component: IDFComponent) -> str:
     """
     Generate a CMakeLists.txt file for an ESP-IDF component.
@@ -85,10 +55,6 @@ def generate_cmakelists_txt(component: IDFComponent) -> str:
     Returns:
         str: The complete CMakeLists.txt content as a string
     """
-    # Late import: this module loads with the esp32 platform on every
-    # validate/compile, but shlex is only needed when generating component
-    # CMakeLists.
-    import shlex
 
     def escape_entry(p: PathType) -> str:
         # In CMakeLists.txt, backslashes need to be escaped
@@ -122,26 +88,12 @@ def generate_cmakelists_txt(component: IDFComponent) -> str:
     build_src_filter = ensure_list(
         component.data.get("build", {}).get("srcFilter", DEFAULT_BUILD_SRC_FILTER)
     )
-    build_flags = ensure_list(
-        component.data.get("build", {}).get("flags", DEFAULT_BUILD_FLAGS)
+    # PlatformIO shell-lexes each build.flags entry; bare -I/-L/-l/-D tokens
+    # re-glue to their argument so the prefix classifiers below route them.
+    build_flags = lex_build_flags(
+        component.data.get("build", {}).get("flags", DEFAULT_BUILD_FLAGS),
+        f"library {component.name}",
     )
-    # PlatformIO shell-lexes each build.flags entry, so one entry can carry a
-    # flag and its argument (e.g. "-include cp_custom_alloc.h"). Split the
-    # same way; emitting such an entry as a single quoted compile option
-    # hands the compiler one argv with an embedded space.
-    build_flags = [token for entry in build_flags for token in shlex.split(entry)]
-    # Re-glue bare -I/-L/-l tokens to their argument ("-I foo" -> "-Ifoo") so
-    # the prefix classifiers below still route them to INCLUDE_DIRS and the
-    # link handling.
-    tokens, build_flags = build_flags, []
-    i = 0
-    while i < len(tokens):
-        if tokens[i] in ("-I", "-L", "-l") and i + 1 < len(tokens):
-            build_flags.append(tokens[i] + tokens[i + 1])
-            i += 2
-        else:
-            build_flags.append(tokens[i])
-            i += 1
 
     # List all sources files
     build_src_files = collect_filtered_files(
@@ -299,7 +251,14 @@ def generate_idf_component_yml(component: IDFComponent) -> str:
 
 def _emit_idf_component(component: IDFComponent) -> None:
     """Write the ESP-IDF build files for a resolved library into its cache dir."""
-    _apply_extra_script(component)
+    from esphome.components.esp32 import get_esp32_variant
+    from esphome.platformio.extra_script import apply_extra_script
+
+    apply_extra_script(
+        component,
+        board_mcu=lambda: variant_to_idf_target(get_esp32_variant()),
+        pio_platform="espressif32",
+    )
     write_file_if_changed(
         component.path / "CMakeLists.txt",
         generate_cmakelists_txt(component),

esphome/espidf/extra_script.py

@@ -1,161 +0,0 @@
-"""Run a PlatformIO ``extraScript`` against a captured SCons-env stand-in.
-
-PlatformIO libraries occasionally configure per-target link/build state
-via a Python ``extraScript`` declared in ``library.json``'s ``build``
-section instead of static fields. The script runs under SCons during
-PIO's build and mutates the active ``Environment`` (``env.Append``,
-``env.Replace``, …) — chiefly to set ``LIBPATH``/``LIBS`` per chip MCU.
-
-ESPHome's PIO→IDF converter doesn't run SCons, so these scripts were
-previously ignored and any library
-relying on them failed to link under ``toolchain: esp-idf``. This
-module provides a small shim that ``exec``s an extra-script with a
-fake ``env`` object, captures the common ``env.Append(...)`` calls,
-and returns the captured vars so the caller can fold them back into
-the library's generated CMakeLists.
-
-Caveats
--------
-* Only the ``env.Append`` API is captured. ``env.Replace``,
-  ``env.Prepend``, ``env.AddPreAction``, SCons file generators, and any
-  arbitrary I/O are silently no-ops. Scripts that depend on those will
-  produce incomplete output.
-* Running arbitrary Python from third-party libraries is a non-trivial
-  trust decision. The shim does no sandboxing — anything in the
-  script's process can run. Use only with libraries whose source you
-  trust.
-"""
-
-from __future__ import annotations
-
-from dataclasses import dataclass, field
-import logging
-import os
-from pathlib import Path
-
-_LOGGER = logging.getLogger(__name__)
-
-# Keys we know how to translate back into ESPHome's build-flag pipeline.
-# Other env.Append kwargs are recorded but ignored downstream.
-_CAPTURED_KEYS = frozenset({"LIBPATH", "LIBS", "CPPDEFINES", "LINKFLAGS", "CPPFLAGS"})
-
-
-@dataclass
-class ExtraScriptResult:
-    """Build-var deltas captured from a PIO extra-script ``env.Append`` call."""
-
-    libpath: list[str] = field(default_factory=list)
-    libs: list[str] = field(default_factory=list)
-    cppdefines: list[str | tuple[str, str]] = field(default_factory=list)
-    linkflags: list[str] = field(default_factory=list)
-    cppflags: list[str] = field(default_factory=list)
-
-
-class _FakeSConsEnv:
-    """Minimal stand-in for SCons ``Environment`` exposed to extra-scripts.
-
-    Implements just enough surface area to let scripts query ``BOARD_MCU``
-    / ``PIOENV`` and call ``env.Append(LIBPATH=…, LIBS=…, …)``. Every
-    other env method swallows silently so unrelated calls don't raise
-    ``AttributeError`` and abort the script.
-    """
-
-    def __init__(self, *, board_mcu: str, pio_env: str) -> None:
-        self._vars: dict[str, str] = {
-            "BOARD_MCU": board_mcu,
-            "PIOPLATFORM": "espressif32",
-            "PIOENV": pio_env,
-        }
-        self.result = ExtraScriptResult()
-
-    # ----- SCons env API the common scripts use -----
-
-    def get(self, key: str, default: str | None = None) -> str | None:
-        return self._vars.get(key, default)
-
-    def Append(self, **kwargs) -> None:  # noqa: N802 (SCons API name)
-        for key, value in kwargs.items():
-            if key not in _CAPTURED_KEYS:
-                continue
-            items = list(value) if isinstance(value, (list, tuple)) else [value]
-            bucket = getattr(self.result, key.lower())
-            bucket.extend(items)
-
-    # ----- Everything else is a no-op so unsupported scripts don't crash -----
-
-    def __getattr__(self, name: str):
-        def _noop(*args, **kwargs):
-            return None
-
-        return _noop
-
-
-def run_extra_script(
-    script_path: Path, *, library_dir: Path, idf_target: str
-) -> ExtraScriptResult:
-    """Execute ``script_path`` with a fake SCons env and return captured vars.
-
-    ``idf_target`` is the active ESP-IDF target name (e.g. ``esp32``,
-    ``esp32s3``); it's exposed to the script as PlatformIO's
-    ``BOARD_MCU`` so chip-conditional logic resolves the same way it
-    would under PIO. The script runs with ``library_dir`` as the
-    process CWD so relative-path lookups (``join``, ``realpath``,
-    ``open``) resolve against the library tree.
-
-    On any exception inside the script we log at debug level and return
-    an empty result — extra-scripts are best-effort, and an unsupported
-    script shouldn't block the build.
-    """
-    env = _FakeSConsEnv(board_mcu=idf_target, pio_env=f"esphome_{idf_target}")
-    code = compile(script_path.read_text(encoding="utf-8"), str(script_path), "exec")
-    old_cwd = Path.cwd()
-    try:
-        os.chdir(library_dir)
-        exec(  # noqa: S102 pylint: disable=exec-used
-            code,
-            {
-                "Import": lambda *_args: None,  # SCons-side import; harmless here
-                "env": env,
-                "__file__": str(script_path),
-                "__name__": "__pio_extra_script__",
-            },
-        )
-    except Exception as e:  # noqa: BLE001  # pylint: disable=broad-exception-caught
-        _LOGGER.warning("PIO extra-script %s raised %s; skipping", script_path, e)
-        return ExtraScriptResult()
-    finally:
-        os.chdir(old_cwd)
-    return env.result
-
-
-def captured_as_build_flags(
-    result: ExtraScriptResult, *, library_dir: Path
-) -> list[str]:
-    """Translate captured env vars into the ``-L`` / ``-l`` / ``-D`` /
-    raw-flag form ``_generate_cmakelists_txt`` already knows how to consume.
-
-    ``LIBPATH`` entries are made relative to ``library_dir`` so the
-    generated CMakeLists is portable; absolute paths outside the library
-    tree are kept as-is (CMake handles absolute paths in
-    ``target_link_directories`` fine).
-    """
-    flags: list[str] = []
-    library_root = library_dir.resolve()
-    for path in result.libpath:
-        # Anchor relative paths to library_dir (not the current CWD, which
-        # has been restored by the time we get here). Joining an absolute
-        # path against library_dir returns the absolute path unchanged.
-        resolved = (library_dir / path).resolve()
-        try:
-            flags.append(f"-L{resolved.relative_to(library_root)}")
-        except ValueError:
-            flags.append(f"-L{resolved}")
-    flags.extend(f"-l{lib}" for lib in result.libs)
-    for define in result.cppdefines:
-        if isinstance(define, tuple) and len(define) == 2:
-            flags.append(f"-D{define[0]}={define[1]}")
-        else:
-            flags.append(f"-D{define}")
-    flags.extend(result.linkflags)
-    flags.extend(result.cppflags)
-    return flags

esphome/espidf/framework.py

@@ -11,10 +11,15 @@ import re
 import shutil
 from typing import Any, NoReturn
 
-import platformdirs
-
-from esphome.core import CORE, Version
+from esphome.build_helpers.ccache import (
+    ccache_defaults_env,
+    parse_enable_env,
+    resolve_ccache_path,
+)
+from esphome.build_helpers.tools_cache import IDF_TOOLS_CACHE, tools_cache_path
+from esphome.core import Version
 from esphome.framework_helpers import (
+    BatchDownloadProgress,
     PathType,
     archive_extract_all,
     create_venv,
@@ -23,11 +28,12 @@ from esphome.framework_helpers import (
     get_python_env_executable_path,
     get_system_python_path,
     rmdir,
+    run_batch_downloads,
     run_command,
     run_command_ok,
     str_to_lst_of_str,
 )
-from esphome.helpers import get_bool_env, get_str_env, write_file_if_changed
+from esphome.helpers import write_file_if_changed
 
 _LOGGER = logging.getLogger(__name__)
 
@@ -88,22 +94,10 @@ def get_idf_tools_path() -> Path:
     Returns:
         Path object pointing to the ESP-IDF tools directory
     """
-    # Treat an empty/whitespace ESPHOME_ESP_IDF_PREFIX as unset: Path("")
-    # resolves to the CWD, which would install into (and let clean-all delete)
-    # the working directory by accident.
-    if prefix := get_str_env("ESPHOME_ESP_IDF_PREFIX", "").strip():
-        path = Path(prefix).expanduser()
-    else:
-        # Machine-global so all projects share the multi-GB install instead of
-        # a per-config-directory copy. The user cache dir (not ~/.esphome)
-        # avoids colliding with data_dir when configs live in the home dir.
-        # appauthor=False drops the redundant <author>\ segment on Windows
-        # (which otherwise repeats "esphome\esphome\") to keep the path short.
-        path = Path(platformdirs.user_cache_dir("esphome", appauthor=False)) / "idf"
-    # Resolve so an unnormalized config path (e.g. compiling ``../config/x.yaml``)
-    # doesn't leave ``..`` segments in the IDF_TOOLS_PATH handed to idf.py, which
-    # otherwise warns that the venv interpreter path doesn't match the install.
-    return path.resolve()
+    # Machine-global so all projects share the multi-GB install instead of
+    # a per-config-directory copy; see build_helpers.tools_cache.tools_cache_path
+    # for the env-override and normalization rules.
+    return tools_cache_path(*IDF_TOOLS_CACHE)
 
 
 # Windows' default MAX_PATH is 260 characters. ESP-IDF toolchains nest deeply
@@ -702,10 +696,10 @@ def _prefetch_idf_tool_archives(
     which makes large archives effectively impossible to fetch on unstable
     connections (#17703). This asks the framework's idf_tools (via
     ``get_tool_downloads.py``) which archives the coming install needs, then
-    downloads each into ``<IDF_TOOLS_PATH>/dist`` with
-    ``download_with_resume``. The installer then finds the verified archives
-    already in place ("file ... is already downloaded") and never touches the
-    network.
+    downloads them into ``<IDF_TOOLS_PATH>/dist`` with
+    ``download_with_resume``, a few at a time under one combined progress
+    bar. The installer then finds the verified archives already in place
+    ("file ... is already downloaded") and never touches the network.
 
     Strictly best-effort: any failure here just logs and returns, leaving
     ``idf_tools.py install`` to download whatever is missing exactly as
@@ -727,26 +721,58 @@ def _prefetch_idf_tool_archives(
             )
             return
         dist_path = get_idf_tools_path() / "dist"
-        entries = [
-            entry
-            for entry in json.loads(stdout)
-            if not (dist_path / entry["dest"]).is_file()
-        ]
-        for index, entry in enumerate(entries, start=1):
-            _LOGGER.info(
-                "Downloading %s (%d/%d) ...", entry["name"], index, len(entries)
-            )
-            try:
-                download_with_resume(
-                    entry["url"],
-                    dist_path / entry["dest"],
-                    sha256=entry["sha256"],
-                    size=entry["size"],
+        entries = []
+        seen_dests: set[str] = set()
+        for entry in json.loads(stdout):
+            if (dist_path / entry["dest"]).is_file():
+                continue
+            if entry["dest"] in seen_dests:
+                # Two workers on one .part file would interleave
+                # seek/truncate writes; mirror the library prefetch's dedupe
+                continue
+            seen_dests.add(entry["dest"])
+            # tools.json always carries sha256 and size; an entry missing
+            # either must not be downloaded unverified here, so leave it to
+            # the installer (which fails loudly on a bad archive).
+            if entry.get("sha256") and entry.get("size"):
+                entries.append(entry)
+            else:
+                _LOGGER.warning(
+                    "Tool %s has no sha256/size in the download list; "
+                    "leaving it to the installer",
+                    entry["name"],
                 )
-            except Exception as e:  # noqa: BLE001  # pylint: disable=broad-exception-caught
-                # Keep prefetching the remaining archives; the installer
-                # will retry this one itself (without resume).
-                _LOGGER.warning("Could not prefetch %s: %s", entry["name"], e)
+        if not entries:
+            return
+        _LOGGER.info(
+            "Downloading %d ESP-IDF tool archive(s): %s",
+            len(entries),
+            ", ".join(entry["name"] for entry in entries),
+        )
+
+        # Every entry carries a size (checked above), so the combined bar can
+        # be trusted. Unlike the library prefetch there is no sequential
+        # fallback: per-file bars from several threads would interleave, and
+        # skipping the prefetch would lose the resume workaround for #17703.
+        def _download(entry: dict):
+            return lambda tracker: download_with_resume(
+                entry["url"],
+                dist_path / entry["dest"],
+                sha256=entry["sha256"],
+                size=entry["size"],
+                progress=tracker,
+            )
+
+        # A failed archive is retried by the installer itself (without
+        # resume); keep prefetching the rest.
+        failures = run_batch_downloads(
+            BatchDownloadProgress(
+                "Downloading ESP-IDF tools", sum(entry["size"] for entry in entries)
+            ),
+            [(entry["name"], _download(entry)) for entry in entries],
+        )
+        for name, e in failures:
+            _LOGGER.warning("Could not prefetch %s: %s", name, e)
     except Exception as e:  # noqa: BLE001  # pylint: disable=broad-exception-caught
         # The installer downloads anything missing itself; never let the
         # prefetch become a new way for the install to fail.
@@ -1145,8 +1171,10 @@ def check_esp_idf_install(
 def _ccache_env() -> dict[str, str]:
     """Return ccache settings for ESP-IDF compiles.
 
-    Enabled by default whenever the ``ccache`` binary is on PATH; set
-    ``IDF_CCACHE_ENABLE=0`` in the environment to opt out. The cache lives under
+    Enabled by default whenever a runnable ``ccache`` binary is on PATH.
+    ``IDF_CCACHE_ENABLE=0`` opts out and ``=1`` forces it on; when that knob
+    is unset the shared ``ESPHOME_CCACHE_ENABLE`` applies (same 0/1 forms,
+    unrecognized values warn and count as unset). The cache lives under
     the IDF tools path (the machine-global cache dir, or
     ``ESPHOME_ESP_IDF_PREFIX``), so it is shared across all projects and removed
     by ``esphome clean-all`` along with the framework.
@@ -1161,33 +1189,23 @@ def _ccache_env() -> dict[str, str]:
     Only values the user has not already set in the environment are returned, so
     a custom ``CCACHE_DIR`` / ``CCACHE_MAXSIZE`` / etc. is respected.
     """
-    # Honor an explicit choice already in the environment (opt-out or opt-in).
-    if "IDF_CCACHE_ENABLE" in os.environ:
-        if not get_bool_env("IDF_CCACHE_ENABLE"):
-            return {}
-    elif shutil.which("ccache") is None:
+    # IDF_CCACHE_ENABLE (the backend-native knob) wins over the shared
+    # ESPHOME_CCACHE_ENABLE.
+    idf_knob = parse_enable_env("IDF_CCACHE_ENABLE")
+    if idf_knob is False:
+        # The raw value (e.g. "disable") is still inherited by idf.py via
+        # os.environ, where a non-false-constant string reads as truthy;
+        # export the canonical off spelling instead
+        return {"IDF_CCACHE_ENABLE": "0"}
+    if idf_knob is None and resolve_ccache_path() is None:
         # ESP-IDF silently skips ccache without the binary; don't enable it.
         return {}
 
-    # ccache is enabled past here. build_path is set during preload for every
-    # config-loading command, so it being unset means a caller built the IDF env
-    # too early -- fail loudly rather than silently drop CCACHE_BASEDIR (which
-    # would quietly cost cross-device cache hits).
-    if CORE.build_path is None:
-        raise ValueError(
-            "CORE.build_path must be set before constructing the ESP-IDF build "
-            "environment"
-        )
-
-    defaults = {
-        "IDF_CCACHE_ENABLE": "1",
-        "CCACHE_DIR": str(get_idf_tools_path() / "ccache"),
-        "CCACHE_NOHASHDIR": "true",
-        "CCACHE_DEPEND": "1",
-        "CCACHE_BASEDIR": str(Path(CORE.build_path).resolve()),
-    }
-    # Don't override CCACHE_* values the user already set in their environment.
-    return {k: v for k, v in defaults.items() if k not in os.environ}
+    env = ccache_defaults_env(get_idf_tools_path() / "ccache")
+    if idf_knob is None:
+        # An unparsable IDF_CCACHE_ENABLE must not leak to idf.py as truthy
+        env["IDF_CCACHE_ENABLE"] = "1"
+    return env
 
 
 def get_framework_env(

esphome/espidf/size_summary.py

@@ -28,6 +28,8 @@ import json
 import logging
 from pathlib import Path
 
+from esphome.build_helpers.size_summary import print_size_line
+
 _LOGGER = logging.getLogger(__name__)
 _SIZE_SUFFIXES = {"K": 1024, "M": 1024 * 1024}
 
@@ -67,31 +69,26 @@ def _find_app_partition_size(partitions_csv: Path) -> int:
     raise ValueError(f"No app+factory or app+ota_0 partition in {partitions_csv}")
 
 
-def _format_bar(used: int, total: int) -> str:
-    """Match PlatformIO's ``_format_availale_bytes`` (pioupload.py) exactly."""
-    pct_raw = used / total if total else 0
-    blocks = 10
-    filled = min(int(round(blocks * pct_raw)), blocks)
-    progress = "=" * filled
-    return (
-        f"[{progress:<{blocks}}] {pct_raw: 6.1%} "
-        f"(used {used:d} bytes from {total:d} bytes)"
-    )
-
-
 def print_summary(size_json: Path, partitions_csv: Path | None) -> None:
     """Print PlatformIO-shaped RAM and Flash one-liners.
 
     Failures are non-fatal: the build has already succeeded, we just couldn't
-    summarize. Logs the cause at debug level.
+    summarize. Logs the cause at warning level, so a missing RAM/Flash line
+    (which CI's memory-impact extraction greps for) is diagnosable.
     """
     if not size_json.is_file():
-        _LOGGER.debug("Skipping size summary: %s not found", size_json)
+        _LOGGER.warning("Skipping size summary: %s not found", size_json)
         return
     try:
         data = json.loads(size_json.read_text(encoding="utf-8"))
     except (OSError, json.JSONDecodeError) as e:
-        _LOGGER.debug("Skipping size summary: %s", e)
+        _LOGGER.warning("Skipping size summary: %s", e)
+        return
+
+    if not isinstance(data, dict):
+        # Valid JSON that is not an object (truncated tool output) must
+        # not raise past a build that already linked
+        _LOGGER.warning("Skipping size summary: unexpected shape in %s", size_json)
         return
 
     memory_types = data.get("memory_types", {})
@@ -99,14 +96,32 @@ def print_summary(size_json: Path, partitions_csv: Path | None) -> None:
     ram_used = ram_region.get("used")
     ram_total = ram_region.get("size")
     if ram_total and ram_used is not None:
-        print(f"RAM:   {_format_bar(ram_used, ram_total)}")
+        print_size_line("RAM", ram_used, ram_total)
+    else:
+        _LOGGER.warning(
+            "Skipping RAM summary: no usable DRAM/DIRAM region in %s", size_json
+        )
 
     image_size = data.get("image_size")
-    if image_size is None or partitions_csv is None:
+    if image_size is None:
+        _LOGGER.warning("Skipping Flash summary: no image_size in %s", size_json)
+        return
+    if partitions_csv is None:
+        _LOGGER.warning("Skipping Flash summary: no partition table given")
         return
     try:
         app_size = _find_app_partition_size(partitions_csv)
-    except ValueError as e:
-        _LOGGER.debug("Skipping Flash summary: %s", e)
+    except (ValueError, OSError) as e:
+        _LOGGER.warning("Skipping Flash summary: %s", e)
         return
-    print(f"Flash: {_format_bar(image_size, app_size)}")
+    if app_size <= 0:
+        # A "from 0 bytes" denominator is meaningless to a reader. The skip
+        # costs CI's memory-impact extraction its Flash match, which is the
+        # loud outcome a broken partition table deserves.
+        _LOGGER.warning(
+            "Skipping Flash summary: app partition size is %s in %s",
+            app_size,
+            partitions_csv,
+        )
+        return
+    print_size_line("Flash", image_size, app_size)

esphome/espidf/toolchain.py

@@ -526,32 +526,15 @@ def get_idedata() -> dict | None:
     idedata fields IDE integrations and clang-tidy expect, cached alongside the
     PlatformIO idedata path. Returns None if the compile DB doesn't exist yet.
     """
-    from esphome.espidf.idedata import idedata_from_build
+    from esphome.build_helpers.idedata import load_or_build_idedata
 
-    compile_commands = CORE.relative_build_path("build", "compile_commands.json")
-    if not compile_commands.is_file():
-        _LOGGER.debug("No %s yet; skipping idedata generation", compile_commands)
-        return None
-
-    cache = CORE.relative_internal_path("idedata", f"{CORE.name}.json")
-    if cache.is_file() and cache.stat().st_mtime >= compile_commands.stat().st_mtime:
-        try:
-            cached = json.loads(cache.read_text(encoding="utf-8"))
-        except ValueError:
-            pass
-        else:
-            # Caches written before cc_path was emitted stay newer than
-            # compile_commands.json forever, so rebuild them on the field rather
-            # than on the timestamp. Check the type too: a corrupted cache can
-            # still be valid JSON, and "in" would match a substring of a string.
-            if isinstance(cached, dict) and "cc_path" in cached:
-                return cached
-
-    data = idedata_from_build(compile_commands)
-    data["prog_path"] = str(get_elf_path())
-    cache.parent.mkdir(parents=True, exist_ok=True)
-    cache.write_text(json.dumps(data, indent=2) + "\n", encoding="utf-8")
-    return data
+    # No launcher: CMake excludes CMAKE_<LANG>_COMPILER_LAUNCHER (ccache)
+    # from the exported compile database, unlike ninja's compdb dump.
+    return load_or_build_idedata(
+        CORE.relative_build_path("build", "compile_commands.json"),
+        get_elf_path(),
+        CORE.relative_internal_path("idedata", f"{CORE.name}.json"),
+    )
 
 
 def create_factory_bin() -> bool:

esphome/framework_helpers.py

@@ -1,6 +1,7 @@
 """Generic toolchain installation helpers shared across framework implementations."""
 
-from collections.abc import Iterable
+from collections.abc import Callable, Iterable
+from concurrent.futures import ThreadPoolExecutor
 from contextlib import ExitStack
 import hashlib
 import io
@@ -10,6 +11,7 @@ import os
 from pathlib import Path
 import subprocess
 import sys
+import threading
 import time
 from typing import IO, TYPE_CHECKING
 
@@ -23,6 +25,7 @@ PathType = str | os.PathLike
 
 _LOGGER = logging.getLogger(__name__)
 
+
 # Attempts per mirror URL before falling through to the next mirror; only
 # mid-stream drops retry (resuming when the server gave a validator),
 # connect errors move on to the next mirror immediately.
@@ -196,6 +199,30 @@ def run_command(
         return False, None, None
 
 
+def tool_version_runs(binary: str, warning: str) -> bool:
+    """Probe ``binary --version``; on failure warn with ``warning`` % binary.
+
+    ``shutil.which`` proves existence, not runnability (Windows .bat/.cmd
+    shims, stale package-manager shims).
+    """
+    try:
+        subprocess.run(
+            [binary, "--version"],
+            check=True,
+            stdout=subprocess.DEVNULL,
+            stderr=subprocess.DEVNULL,
+            timeout=15,
+            # Repo-wide convention (posix_spawn fast path)
+            close_fds=False,
+        )
+    except (OSError, subprocess.SubprocessError) as err:
+        # The cause (permission denied, missing DLL, timeout) is the one
+        # detail the user needs to fix it
+        _LOGGER.warning("%s (%s)", warning % binary, err)
+        return False
+    return True
+
+
 def run_command_ok(*args, **kwargs) -> bool:
     """
     Execute a command and return only the success status.
@@ -697,7 +724,11 @@ def _response_validator(resp: "requests.Response") -> str | None:
 
 
 def _stream_response_to_file(
-    resp: "requests.Response", f: IO[bytes], offset: int, size: int | None = None
+    resp: "requests.Response",
+    f: IO[bytes],
+    offset: int,
+    size: int | None = None,
+    progress: Callable[[int], None] | None = None,
 ) -> None:
     """Stream an open ``_open_ranged`` response body into ``f`` at ``offset``.
 
@@ -705,21 +736,112 @@ def _stream_response_to_file(
     (effective offset 0) discards the stale bytes. ``offset`` also seeds the
     progress bar so a resumed download shows overall progress. ``size`` is
     the known full file size; when None it is derived from the response's
-    content-length, and without either there is no progress bar.
+    content-length, and without either there is no progress bar. With
+    ``progress`` set, no bar is drawn here; the callback gets the absolute
+    byte count, seeded with ``offset`` and then after each chunk.
     """
     f.seek(offset)
     f.truncate(offset)
     total_size = size or offset + _content_length(resp)
     downloaded = offset
-    progress = ProgressBar("Downloading") if total_size > 0 else None
+    own_bar: ProgressBar | None = None
+    if progress is None:
+        own_bar = ProgressBar("Downloading") if total_size > 0 else None
+        progress = (
+            (lambda done: own_bar.update(done / total_size))
+            if own_bar
+            else (lambda _: None)
+        )
+    progress(downloaded)
     for chunk in resp.iter_content(chunk_size=256 * 1024):
         if chunk:
             f.write(chunk)
             downloaded += len(chunk)
-        if progress is not None:
-            progress.update(downloaded / total_size)
-    if progress is not None:
-        progress.update(1)
+        progress(downloaded)
+    if own_bar is not None:
+        own_bar.update(1)
+
+
+# Concurrent downloads per batch; enough to hide latency without
+# hammering the host or the mirrors.
+BATCH_DOWNLOAD_WORKERS = 4
+
+
+def run_batch_downloads(
+    progress: "BatchDownloadProgress",
+    jobs: list[tuple[str, Callable[[Callable[[int], None]], None]]],
+    max_workers: int = BATCH_DOWNLOAD_WORKERS,
+) -> list[tuple[str, Exception]]:
+    """Run download jobs concurrently, reporting into one combined bar.
+
+    ``jobs`` holds ``(name, fetch)`` pairs where ``fetch(tracker)`` performs
+    one download reporting absolute byte counts to ``tracker``. Failures are
+    collected (list.append is atomic under the GIL) and returned after the
+    bar is done, so the caller's warnings never land on the bar's row; a
+    failed job credits its tracker 0 so the bar can still complete. Ctrl-C
+    drops queued jobs instead of downloading them all before the process
+    can exit; in-flight ones still finish. ``jobs`` must be non-empty.
+    """
+    failures: list[tuple[str, Exception]] = []
+
+    def _run(name: str, fetch: Callable[[Callable[[int], None]], None]) -> None:
+        tracker = progress.tracker()
+        try:
+            fetch(tracker)
+        except Exception as err:  # noqa: BLE001  # pylint: disable=broad-exception-caught
+            failures.append((name, err))
+            tracker(0)
+
+    ex = ThreadPoolExecutor(max_workers=min(max_workers, len(jobs)))
+    try:
+        for future in [ex.submit(_run, name, fetch) for name, fetch in jobs]:
+            future.result()
+    finally:
+        ex.shutdown(wait=True, cancel_futures=True)
+        progress.done()
+    return failures
+
+
+class BatchDownloadProgress:
+    """One progress bar across several concurrent ``download_with_resume`` calls.
+
+    Each ``tracker()`` is a ``progress`` callback for one download; it reports
+    that file's absolute byte count and the bar shows the sum over ``total``.
+    The lock also serialises the bar's stderr writes, so worker threads never
+    interleave frames. With an unknown ``total`` (0) nothing is drawn. Call
+    ``done()`` once every download has finished (or failed) so a bar that
+    never reached 100% still ends its line before the next log message.
+    """
+
+    def __init__(self, header: str, total: int) -> None:
+        self._bar = ProgressBar(header) if total > 0 else None
+        self._total = total
+        self._sum = 0
+        self._lock = threading.Lock()
+
+    def tracker(self) -> Callable[[int], None]:
+        last = 0
+
+        def update(done: int) -> None:
+            nonlocal last
+            if self._bar is None:
+                return
+            with self._lock:
+                self._sum += done - last
+                last = done
+                self._bar.update(min(self._sum / self._total, 1))
+
+        return update
+
+    def done(self) -> None:
+        # Nothing to end unless a frame was drawn and it was not the final
+        # one (update(1) already emitted its own newline).
+        if (
+            self._bar is not None
+            and self._bar.last_progress is not None
+            and self._bar.last_progress != 100
+        ):
+            self._bar.done()
 
 
 def download_with_resume(
@@ -732,6 +854,7 @@ def download_with_resume(
     attempts: int = 5,
     timeout: int = 30,
     retry_connect_errors: bool = True,
+    progress: Callable[[int], None] | None = None,
 ) -> None:
     """Download ``url`` to ``dest``, resuming partial downloads.
 
@@ -754,6 +877,12 @@ def download_with_resume(
     of consuming attempts — for callers with their own fallback, like
     ``download_from_mirrors``.
 
+    ``progress``, when given, replaces the built-in progress bar: it is called
+    with the absolute number of bytes of ``dest`` obtained so far (including
+    a resumed prefix, and the final size once the file is verified), so a
+    caller running several downloads at once can draw one combined bar (see
+    ``BatchDownloadProgress``).
+
     Raises EsphomeError when all attempts are exhausted.
     """
     # Imported lazily: requests is a heavy import (~85ms) and is only needed
@@ -777,6 +906,8 @@ def download_with_resume(
     if dest.is_file() and (sha256 is not None or size is not None):
         try:
             _verify_file(dest, sha256, size)
+            if progress is not None:
+                progress(size if size is not None else dest.stat().st_size)
             return
         except EsphomeError:
             dest.unlink()
@@ -822,7 +953,7 @@ def download_with_resume(
                             # Recorded so a later run can prove an If-Range
                             # resume of this part file safe.
                             _write_download_meta(meta, url, validator, expected_total)
-                        _stream_response_to_file(resp, f, offset, size)
+                        _stream_response_to_file(resp, f, offset, size, progress)
             # else: a previous run already wrote every byte (or more) but
             # was killed before the rename below. Skip the network entirely
             # — a Range request past EOF would draw HTTP 416 — and let
@@ -831,6 +962,10 @@ def download_with_resume(
 
             expected_size = size if size is not None else expected_total
             _verify_file(part, sha256, expected_size or None)
+            if progress is not None:
+                # Also credits a part file an earlier run completed without
+                # streaming anything this time.
+                progress(expected_size or part.stat().st_size)
             if not expected_size and sha256 is None:
                 # No sha, no size, and the server sent no usable
                 # content-length: nothing can prove the download complete
@@ -933,6 +1068,7 @@ def _try_mirrors_once(
     f: IO[bytes] | None,
     timeout: int,
     failures: list[tuple[str, Exception]],
+    progress: Callable[[int], None] | None = None,
 ) -> str | None:
     """Single pass over the resolved mirror ``urls``, one try per URL.
 
@@ -961,6 +1097,7 @@ def _try_mirrors_once(
                     # next mirror immediately; only mid-stream drops
                     # retry-with-resume on the same URL.
                     retry_connect_errors=False,
+                    progress=progress,
                 )
                 return url
             except (requests.RequestException, OSError, EsphomeError) as e:
@@ -1002,7 +1139,7 @@ def _try_mirrors_once(
                         if offset == 0:
                             validator = _response_validator(resp)
                             expected_total = _content_length(resp)
-                        _stream_response_to_file(resp, f, offset)
+                        _stream_response_to_file(resp, f, offset, progress=progress)
 
                 if expected_total and f.tell() != expected_total:
                     raise EsphomeError(
@@ -1051,6 +1188,7 @@ def download_from_mirrors(
     substitutions: dict[str, str],
     target: io.RawIOBase | IO[bytes] | PathType,
     timeout: int = 30,
+    progress: Callable[[int], None] | None = None,
 ) -> str:
     """
     Download file from multiple mirrors with substitution support.
@@ -1060,6 +1198,8 @@ def download_from_mirrors(
         substitutions: Dictionary of substitutions to apply to URLs
         target: Target file path or file-like object
         timeout: Download timeout in seconds
+        progress: Passed through to the download (see ``download_with_resume``);
+            replaces the built-in per-file bar
 
     Returns:
         The source URL.
@@ -1124,7 +1264,9 @@ def download_from_mirrors(
     for sweep in range(1, _MIRROR_SWEEP_ATTEMPTS + 1):
         sweep_failures: list[tuple[str, Exception]] = []
         if (
-            url := _try_mirrors_once(urls, path_target, f, timeout, sweep_failures)
+            url := _try_mirrors_once(
+                urls, path_target, f, timeout, sweep_failures, progress
+            )
         ) is not None:
             return url
         failures.extend(sweep_failures)
@@ -1169,3 +1311,37 @@ def download_from_mirrors(
             f"No mirror URL template matched the provided substitutions:{details}"
         )
     raise ValueError("download_from_mirrors called with an empty mirrors list")
+
+
+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.
+
+    Also applied to the ccache path exported by the ccache helpers, which
+    ``shutil.which`` can return with the same prefix.
+
+    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

esphome/helpers.py

@@ -31,6 +31,12 @@ SockAddr = IPv4SockAddr | IPv6SockAddr
 
 _LOGGER = logging.getLogger(__name__)
 
+# cv.boolean's closed spelling tables, shared with the strict env-knob
+# parser (build_helpers.ccache.parse_enable_env). The legacy get_bool_env
+# below keeps its own laxer table for backward compatibility.
+TRUTHY_BOOL_STRINGS = frozenset({"true", "yes", "on", "enable"})
+FALSY_BOOL_STRINGS = frozenset({"false", "no", "off", "disable"})
+
 IS_MACOS = platform.system() == "Darwin"
 IS_WINDOWS = platform.system() == "Windows"
 IS_LINUX = platform.system() == "Linux"

esphome/platformio/extra_script.py

@@ -0,0 +1,263 @@
+"""Run a PlatformIO library ``extraScript`` against a fake SCons env.
+
+The shim execs the script with a stand-in ``env``, captures ``env.Append``
+calls (everything else is a logged no-op), and folds the result into the
+library's build flags. No sandboxing: the script runs with full process
+access, so it carries the same trust as the library's own source.
+"""
+
+from __future__ import annotations
+
+from collections.abc import Callable
+from dataclasses import dataclass, field
+import logging
+import os
+from pathlib import Path
+from typing import TYPE_CHECKING
+
+from esphome.core import EsphomeError
+
+if TYPE_CHECKING:
+    from esphome.platformio.library import ConvertedLibrary
+
+_LOGGER = logging.getLogger(__name__)
+
+
+def apply_extra_script(
+    component: ConvertedLibrary,
+    board_mcu: Callable[[], str],
+    pio_platform: str,
+) -> None:
+    """Run a library's ``extraScript`` and fold its captured env vars into
+    ``component.data["build"]["flags"]``.
+
+    ``board_mcu`` is a callable so its lookup runs only when a script will.
+    """
+    extra_script = component.data.get("build", {}).get("extraScript")
+    if not extra_script:
+        return
+    # Resolve and confine to the library's source dir so a malicious
+    # library.json can't escape (e.g. ``"extraScript": "../../etc/passwd"``).
+    source_path = component.source_dir
+    library_root = source_path.resolve()
+    script_path = (source_path / extra_script).resolve()
+    if not script_path.is_relative_to(library_root):
+        # More hostile than a missing script; must not be quieter than it
+        raise EsphomeError(
+            f"extraScript {extra_script} of library {component.name} escapes "
+            "the library directory"
+        )
+    if not script_path.is_file():
+        # A declared-but-absent script is a broken or half-downloaded
+        # package, not an unsupported script; PlatformIO fails on it too
+        raise EsphomeError(
+            f"extraScript {extra_script} of library {component.name} not found"
+        )
+    result = run_extra_script(
+        script_path,
+        library_dir=source_path,
+        board_mcu=board_mcu(),
+        pio_platform=pio_platform,
+    )
+    extra_flags = captured_as_build_flags(result, library_dir=source_path)
+    if not extra_flags:
+        return
+    flags = component.data.setdefault("build", {}).setdefault("flags", [])
+    if isinstance(flags, str):
+        flags = [flags]
+    elif not isinstance(flags, list):
+        # A null/dict value coerced through a list wrapper would inject a
+        # non-string into the compiler command line; fail naming the library
+        raise EsphomeError(
+            f"Library {component.name} has a malformed build.flags "
+            f"({type(flags).__name__}); expected a string or list"
+        )
+    component.data["build"]["flags"] = [*flags, *extra_flags]
+
+
+# Keys we know how to translate back into ESPHome's build-flag pipeline.
+# Other env.Append kwargs are recorded but ignored downstream.
+_CAPTURED_KEYS = frozenset({"LIBPATH", "LIBS", "CPPDEFINES", "LINKFLAGS", "CPPFLAGS"})
+
+
+@dataclass
+class ExtraScriptResult:
+    """Build-var deltas captured from a PIO extra-script ``env.Append`` call."""
+
+    libpath: list[str] = field(default_factory=list)
+    libs: list[str] = field(default_factory=list)
+    cppdefines: list[str | tuple[str, str]] = field(default_factory=list)
+    linkflags: list[str] = field(default_factory=list)
+    cppflags: list[str] = field(default_factory=list)
+
+
+class _FakeSConsEnv:
+    """Minimal SCons ``Environment`` stand-in: ``get`` and ``Append`` work;
+    every other method is a swallowed no-op so scripts don't abort."""
+
+    def __init__(self, *, board_mcu: str, pio_env: str, pio_platform: str) -> None:
+        self._vars: dict[str, str] = {
+            "BOARD_MCU": board_mcu,
+            "PIOPLATFORM": pio_platform,
+            "PIOENV": pio_env,
+        }
+        self.result = ExtraScriptResult()
+        self._warned_methods: set[str] = set()
+        self._warned_keys: set[str] = set()
+
+    # ----- SCons env API the common scripts use -----
+
+    def get(self, key: str, default: str | None = None) -> str | None:
+        return self._vars.get(key, default)
+
+    def __getitem__(self, key: str) -> str:
+        # Scripts also read env["BOARD_MCU"]; without this the broad
+        # handler would discard every flag the script captured
+        return self._vars[key]
+
+    def Append(self, **kwargs) -> None:  # noqa: N802 (SCons API name)
+        for key, value in kwargs.items():
+            if key not in _CAPTURED_KEYS:
+                # Warn once per key so a loop of Appends cannot spam
+                if key not in self._warned_keys:
+                    self._warned_keys.add(key)
+                    _LOGGER.warning(
+                        "PIO extra-script env.Append(%s=...) is not captured; ignoring",
+                        key,
+                    )
+                continue
+            items = list(value) if isinstance(value, (list, tuple)) else [value]
+            bucket = getattr(self.result, key.lower())
+            bucket.extend(items)
+
+    # ----- Everything else is a no-op so unsupported scripts don't crash -----
+
+    def __getattr__(self, name: str):
+        def _noop(*args, **kwargs):
+            # Once per method: a script whose whole effect is env.Replace()
+            # must be diagnosable from a normal build log
+            if name not in self._warned_methods:
+                self._warned_methods.add(name)
+                _LOGGER.warning(
+                    "PIO extra-script env.%s(...) is not supported; ignoring", name
+                )
+
+        return _noop
+
+
+def run_extra_script(
+    script_path: Path,
+    *,
+    library_dir: Path,
+    board_mcu: str,
+    pio_platform: str,
+) -> ExtraScriptResult:
+    """Execute ``script_path`` with a fake SCons env and return captured vars.
+
+    Runs with ``library_dir`` as CWD so relative lookups resolve against
+    the library tree. A crashed script warns and returns an empty result,
+    never a partial capture.
+    """
+    env = _FakeSConsEnv(
+        board_mcu=board_mcu,
+        pio_env=f"esphome_{board_mcu}",
+        pio_platform=pio_platform,
+    )
+    try:
+        source = script_path.read_text(encoding="utf-8")
+    except OSError as err:
+        # An unreadable declared script is a broken package, exactly like a
+        # missing one; must not be quieter than that case
+        raise EsphomeError(f"extraScript {script_path} is unreadable: {err}") from err
+    except UnicodeDecodeError as e:
+        # A content problem, best-effort like a SyntaxError below
+        _LOGGER.warning(
+            "PIO extra-script %s (in %s) is not UTF-8 (%r); ignoring its output",
+            script_path,
+            library_dir.name,
+            e,
+        )
+        return ExtraScriptResult()
+    old_cwd = Path.cwd()
+    try:
+        # Inside the try: a SyntaxError in a vendored script is just as
+        # best-effort as a runtime failure
+        code = compile(source, str(script_path), "exec")
+        os.chdir(library_dir)
+        exec(  # noqa: S102 pylint: disable=exec-used
+            code,
+            {
+                "Import": lambda *_args: None,  # SCons-side import; harmless here
+                "env": env,
+                "__file__": str(script_path),
+                "__name__": "__pio_extra_script__",
+            },
+        )
+    except SystemExit as e:
+        if not e.code:
+            # sys.exit() / sys.exit(0) is a normal PlatformIO script ending;
+            # the capture is complete
+            return env.result
+        _LOGGER.warning(
+            "PIO extra-script %s (in %s) exited with status %r; ignoring its output",
+            script_path,
+            library_dir.name,
+            e.code,
+        )
+        return ExtraScriptResult()
+    except Exception as e:  # noqa: BLE001  # pylint: disable=broad-exception-caught
+        # Discard any partial capture: half-applied flags could build wrong
+        # firmware that links cleanly.
+        _LOGGER.warning(
+            "PIO extra-script %s (in %s) raised %r; ignoring its output",
+            script_path,
+            library_dir.name,
+            e,
+        )
+        return ExtraScriptResult()
+    finally:
+        os.chdir(old_cwd)
+    return env.result
+
+
+def captured_as_build_flags(
+    result: ExtraScriptResult, *, library_dir: Path
+) -> list[str]:
+    """Translate captured env vars into -L/-l/-D/raw build flags.
+
+    ``LIBPATH`` entries are made relative to ``library_dir`` so the
+    generated build files stay portable.
+    """
+    flags: list[str] = []
+
+    def _strs(bucket: list, kind: str) -> list[str]:
+        # Third-party scripts legally append SCons nodes, ints, or dicts;
+        # stringifying those into flags would hand the compiler garbage
+        good = [entry for entry in bucket if isinstance(entry, str)]
+        for entry in bucket:
+            if not isinstance(entry, str):
+                _LOGGER.warning("Ignoring unsupported %s entry %r", kind, entry)
+        return good
+
+    library_root = library_dir.resolve()
+    for path in _strs(result.libpath, "LIBPATH"):
+        # Anchor relative paths to library_dir; the script's CWD has been
+        # restored by now
+        resolved = (library_dir / path).resolve()
+        try:
+            flags.append(f"-L{resolved.relative_to(library_root)}")
+        except ValueError:
+            flags.append(f"-L{resolved}")
+    flags.extend(f"-l{lib}" for lib in _strs(result.libs, "LIBS"))
+    for define in result.cppdefines:
+        # SCons also accepts dict/list CPPDEFINES; formatting those blind
+        # would hand the compiler garbage like -D{'FOO': '1'}
+        if isinstance(define, (tuple, list)) and len(define) == 2:
+            flags.append(f"-D{define[0]}={define[1]}")
+        elif isinstance(define, str):
+            flags.append(f"-D{define}")
+        else:
+            _LOGGER.warning("Ignoring unsupported CPPDEFINES entry %r", define)
+    flags.extend(_strs(result.linkflags, "LINKFLAGS"))
+    flags.extend(_strs(result.cppflags, "CPPFLAGS"))
+    return flags

esphome/platformio/library.py

@@ -13,7 +13,8 @@ regardless of which toolchain consumes the result.
 """
 
 from collections import deque
-from collections.abc import Callable
+from collections.abc import Callable, Iterable
+from concurrent.futures import ThreadPoolExecutor
 from dataclasses import dataclass, field
 import glob
 import hashlib
@@ -30,7 +31,14 @@ from urllib.request import url2pathname
 
 from esphome import git
 from esphome.core import CORE, EsphomeError, Library
-from esphome.framework_helpers import archive_extract_all, download_from_mirrors, rmdir
+from esphome.framework_helpers import (
+    BATCH_DOWNLOAD_WORKERS,
+    BatchDownloadProgress,
+    archive_extract_all,
+    download_from_mirrors,
+    rmdir,
+    run_batch_downloads,
+)
 
 _LOGGER = logging.getLogger(__name__)
 
@@ -47,20 +55,28 @@ DEFAULT_BUILD_SRC_FILTER = (
 DEFAULT_BUILD_SRC_DIRS = "src"
 DEFAULT_BUILD_INCLUDE_DIR = "include"
 DEFAULT_BUILD_FLAGS = []
-SRC_FILE_EXTENSIONS = [
-    ".c",
-    ".cpp",
-    ".cc",
-    ".cxx",
-    ".c++",
-    ".S",
-    ".spp",
-    ".SPP",
-    ".sx",
-    ".s",
-    ".asm",
-    ".ASM",
-]
+# Suffix -> compiler kind (PlatformIO's CSUFFIXES/CXXSUFFIXES/ASSUFFIXES).
+# "asm" merges SCons's AS and ASPP sets: all compile as assembler-with-cpp.
+# The kind values drive the ESP8266 native ninja rules (later in this
+# chain); existing backends consume only the keys. Note .C/.C++ join the
+# suffix set here, matching PlatformIO's CXXSUFFIXES.
+SOURCE_KIND_FOR_SUFFIX: dict[str, str] = {
+    ".c": "c",
+    ".cpp": "cxx",
+    ".cc": "cxx",
+    ".cxx": "cxx",
+    ".c++": "cxx",
+    ".C": "cxx",
+    ".C++": "cxx",
+    ".S": "asm",
+    ".spp": "asm",
+    ".SPP": "asm",
+    ".sx": "asm",
+    ".s": "asm",
+    ".asm": "asm",
+    ".ASM": "asm",
+}
+SRC_FILE_EXTENSIONS = list(SOURCE_KIND_FOR_SUFFIX)
 
 DOMAIN = "pio_components"
 
@@ -70,7 +86,12 @@ ESPHOME_DATA_EXTRA_CMAKE_KEY = "EXTRA_CMAKE"
 
 class Source:
     def download(
-        self, dir_suffix: str, force: bool = False, salt: str = "", namespace: str = ""
+        self,
+        dir_suffix: str,
+        force: bool = False,
+        salt: str = "",
+        namespace: str = "",
+        progress: Callable[[int], None] | None = None,
     ) -> Path:
         raise NotImplementedError
 
@@ -87,9 +108,7 @@ class URLSource(Source):
     def __init__(self, url: str):
         self.url = url
 
-    def download(
-        self, dir_suffix: str, force: bool = False, salt: str = "", namespace: str = ""
-    ) -> Path:
+    def _cache_dir(self, dir_suffix: str, salt: str, namespace: str) -> Path:
         # Namespace the cache per backend (e.g. pio_components/idf, .../zephyr) so
         # the build files each backend writes into the library dir can't collide.
         base_dir = Path(CORE.data_dir) / DOMAIN
@@ -99,7 +118,23 @@ class URLSource(Source):
         h.update(self.url.encode())
         if salt:
             h.update(salt.encode())
-        path = base_dir / h.hexdigest()[:8] / dir_suffix
+        return base_dir / h.hexdigest()[:8] / dir_suffix
+
+    def is_cached(self, dir_suffix: str, salt: str = "", namespace: str = "") -> bool:
+        """Whether a completed extraction already exists for this source."""
+        return (
+            self._cache_dir(dir_suffix, salt, namespace) / ".esphome_extracted"
+        ).is_file()
+
+    def download(
+        self,
+        dir_suffix: str,
+        force: bool = False,
+        salt: str = "",
+        namespace: str = "",
+        progress: Callable[[int], None] | None = None,
+    ) -> Path:
+        path = self._cache_dir(dir_suffix, salt, namespace)
         # Marker file written last to signal a complete extraction. Using a
         # marker (instead of just `path.is_dir()`) means an interrupted
         # extraction is correctly detected and re-run on the next invocation,
@@ -111,10 +146,12 @@ class URLSource(Source):
 
             # Download in temporary file
             with tempfile.NamedTemporaryFile() as tmp:
-                _LOGGER.info("Downloading %s ...", self.url)
+                if progress is None:
+                    # A batch caller draws one combined bar and logs the list
+                    _LOGGER.info("Downloading %s ...", self.url)
                 _LOGGER.debug("Location: %s", path)
 
-                download_from_mirrors([self.url], {}, tmp.file)
+                download_from_mirrors([self.url], {}, tmp.file, progress=progress)
 
                 _LOGGER.debug("Extracting archive to %s ...", path)
                 archive_extract_all(tmp.file, path)
@@ -131,7 +168,12 @@ class GitSource(Source):
         self.ref = ref
 
     def download(
-        self, dir_suffix: str, force: bool = False, salt: str = "", namespace: str = ""
+        self,
+        dir_suffix: str,
+        force: bool = False,
+        salt: str = "",
+        namespace: str = "",
+        progress: Callable[[int], None] | None = None,
     ) -> Path:
         domain = DOMAIN
         if namespace:
@@ -166,7 +208,12 @@ class LocalSource(Source):
         self.local_path = path
 
     def download(
-        self, dir_suffix: str, force: bool = False, salt: str = "", namespace: str = ""
+        self,
+        dir_suffix: str,
+        force: bool = False,
+        salt: str = "",
+        namespace: str = "",
+        progress: Callable[[int], None] | None = None,
     ) -> Path:
         src = Path(self.local_path)
         if not src.is_dir():
@@ -203,6 +250,14 @@ class InvalidLibrary(Exception):
     pass
 
 
+class IncompatiblePlatform(InvalidLibrary):
+    """The manifest's platform filter rejected the target platform.
+
+    A distinct type so callers can treat the routine cross-platform skip
+    differently from other manifest problems without matching message text.
+    """
+
+
 class ConvertedLibrary:
     """A resolved PlatformIO library plus its parsed manifest and on-disk path.
 
@@ -251,7 +306,13 @@ class ConvertedLibrary:
     def get_require_name(self):
         return self.get_sanitized_name().replace("/", "__")
 
-    def download(self, force: bool = False, salt: str = "", namespace: str = ""):
+    def download(
+        self,
+        force: bool = False,
+        salt: str = "",
+        namespace: str = "",
+        progress: Callable[[int], None] | None = None,
+    ):
         """Fetch the library into the shared cache and record its ``path``.
 
         The cache directory is named after the sanitized library name; backends
@@ -260,7 +321,11 @@ class ConvertedLibrary:
         ``get_require_name``). ``namespace`` keeps each backend's cache separate.
         """
         self.path = self.source.download(
-            self.get_sanitized_name(), force=force, salt=salt, namespace=namespace
+            self.get_sanitized_name(),
+            force=force,
+            salt=salt,
+            namespace=namespace,
+            progress=progress,
         )
         self.source_path = self.source.source_root(self.path)
 
@@ -432,7 +497,7 @@ def check_library_data(data: dict, platform: str | None, framework: str):
     valid_platforms = platform is None or "*" in platforms or platform in platforms
 
     if not valid_platforms:
-        raise InvalidLibrary(f"Unsupported library platforms: {platforms}")
+        raise IncompatiblePlatform(f"Unsupported library platforms: {platforms}")
 
     frameworks = data.get("frameworks", "*")
     if isinstance(frameworks, str):
@@ -455,7 +520,7 @@ def check_library_data(data: dict, platform: str | None, framework: str):
         )
 
 
-def _parse_library_json(library_json_path: PathType):
+def parse_library_json(library_json_path: PathType):
     """
     Load and parse a JSON file describing a library.
 
@@ -469,7 +534,7 @@ def _parse_library_json(library_json_path: PathType):
         return json.load(fp)
 
 
-def _parse_library_properties(library_properties_path: PathType):
+def parse_library_properties(library_properties_path: PathType):
     """
     Parse a key-value platformio .properties style file into a dictionary.
 
@@ -553,19 +618,147 @@ def _resolve_registry_version(
     return owner, name, best["name"], pkgfile["download_url"]
 
 
-def _normalize_dependencies(dependencies: Any) -> list[dict]:
+def split_flag_entry(entry: Any, owner: str) -> list[str]:
+    """``shlex.split`` with a clean error naming the offending flags entry."""
+    # Late import: shlex is only needed when actually lexing flags
+    import shlex
+
+    try:
+        return shlex.split(entry)
+    except (ValueError, AttributeError, TypeError) as err:
+        # AttributeError/TypeError: a dict or number from a third-party
+        # manifest; name the entry instead of an opaque shlex traceback
+        raise EsphomeError(f"Malformed build flag {entry!r} in {owner}: {err}") from err
+
+
+def lex_build_flags(entries: str | list[str], owner: str) -> list[str]:
+    """Shell-lex ``build.flags`` entries the way PlatformIO's ParseFlags
+    does; bare -I/-L/-l/-D tokens re-glue to their argument."""
+    # Join per entry, as SCons's ParseFlags lexes each string independently:
+    # a dangling -I ending one entry must warn, not absorb the next entry's
+    # first token.
+    return [
+        token
+        for entry in ensure_list(entries)
+        for token in join_flag_args(split_flag_entry(entry, owner), owner)
+    ]
+
+
+# Flags whose argument may follow as a separate token; ParseFlags glues them
+BARE_ARG_FLAGS = frozenset({"-I", "-L", "-l", "-D"})
+
+
+def raise_on_empty_arg_flags(tokens: list[str], owner: str) -> None:
+    """Reject bare ``-I``/``-D``/``-L``/``-l`` tokens left by an empty glued
+    argument (``-D ""``).
+
+    Consumed by the ESP8266 native build generator (later in this chain)
+    for user build_flags; library manifests deliberately stay warn-and-drop.
+
+    Lives next to ``join_flag_args`` because the bare token is its
+    postcondition: a trailing bare flag is warned and dropped there, so a
+    surviving one always means an empty argument. gcc would eat the next
+    flag as the argument (or add the CWD for ``-L``); always a typo.
+    """
+    if empty := sorted({tok for tok in tokens if tok in BARE_ARG_FLAGS}):
+        raise EsphomeError(
+            f"{owner} contain empty-argument flag(s): {', '.join(empty)}"
+        )
+
+
+def join_flag_args(tokens: Iterable[str], owner: str) -> list[str]:
+    """Join a bare ``-I``/``-L``/``-l``/``-D`` with its following token,
+    the way PlatformIO's ParseFlags lexes them."""
+    out: list[str] = []
+    it = iter(tokens)
+    for tok in it:
+        if tok in BARE_ARG_FLAGS:
+            arg = next(it, None)
+            if arg is None:
+                _LOGGER.warning("Ignoring trailing '%s' in %s build flags", tok, owner)
+                break
+            tok += arg
+        out.append(tok)
+    return out
+
+
+def warn_properties_depends(name: str, data: object) -> None:
+    """Warn when a manifest declares dependencies only as ``depends=``.
+
+    The dependency walk reads the JSON ``dependencies`` key; the raw
+    ``library.properties`` spelling would otherwise drop silently.
+    """
+    if isinstance(data, dict) and not data.get("dependencies") and data.get("depends"):
+        # INFO: common and unactionable for transitive libraries; a WARNING
+        # on every build would train users to ignore the stream
+        _LOGGER.info(
+            "Library %s declares dependencies via library.properties "
+            "depends=, which are not resolved automatically; add them with "
+            "add_library() if needed",
+            name,
+        )
+
+
+def dependency_is_usable(
+    dep: dict, platform: str | None, framework: str, requester: str
+) -> bool:
+    """Compatibility filter for a manifest dependency: platform mismatches
+    skip at debug, any other ``InvalidLibrary`` warns naming the requester."""
+    try:
+        check_library_data(dep, platform, framework)
+    except IncompatiblePlatform as e:
+        _LOGGER.debug("Skip dependency %s of %s: %s", dep.get("name"), requester, e)
+        return False
+    except InvalidLibrary as e:
+        _LOGGER.warning(
+            "Skipping dependency %s of %s: %s", dep.get("name"), requester, e
+        )
+        return False
+    return True
+
+
+def _valid_dependency_entry(entry: dict, manifest_name: str) -> bool:
+    """Whether a normalized entry carries a usable name and version.
+
+    The name must be a non-empty string (every consumer indexes or joins
+    it); a present version must be a string (a container would raise from
+    ``set.add()``, an int fails opaquely inside the registry resolution).
+    Invalid entries warn naming the manifest.
+    """
+    name = entry.get("name")
+    if (
+        isinstance(name, str)
+        and name
+        and ("version" not in entry or isinstance(entry["version"], str))
+    ):
+        return True
+    _LOGGER.warning(
+        "Ignoring unrecognized dependency entry %r of %s", entry, manifest_name
+    )
+    return False
+
+
+def normalize_dependencies(
+    dependencies: Any, manifest_name: str = "manifest"
+) -> list[dict]:
     """Normalize a library manifest's ``dependencies`` to a list of dicts.
 
-    PIO's library.json accepts both the list-of-dicts form and the shorthand
-    dict form (``{"owner/Name": "version_spec"}``); normalize the latter so
-    callers see a uniform list.
+    PIO's library.json accepts the list-of-dicts form, the shorthand dict
+    form (``{"owner/Name": "version_spec"}``), bare name strings inside the
+    list, and a plain (possibly comma-separated) string; normalize them all
+    so callers see a uniform list. ``manifest_name`` names the manifest in the
+    warning for entries that cannot be normalized.
     """
     if not dependencies:
         return []
+    if isinstance(dependencies, str):
+        # A plain string is one or more comma-separated names; iterating it
+        # as a list would shred it into one-character "libraries"
+        return [{"name": n.strip()} for n in dependencies.split(",") if n.strip()]
     if isinstance(dependencies, dict):
         normalized = []
         for raw_name, spec in dependencies.items():
-            if "/" in raw_name:
+            if isinstance(raw_name, str) and "/" in raw_name:
                 owner, pkgname = raw_name.split("/", 1)
             else:
                 owner, pkgname = None, raw_name
@@ -574,9 +767,31 @@ def _normalize_dependencies(dependencies: Any) -> list[dict]:
                 entry.update(spec)
             else:
                 entry["version"] = spec
-            normalized.append(entry)
+            if _valid_dependency_entry(entry, manifest_name):
+                normalized.append(entry)
         return normalized
-    return [d for d in dependencies if isinstance(d, dict)]
+    if not isinstance(dependencies, (list, tuple)):
+        _LOGGER.warning(
+            "Ignoring unrecognized dependencies %r of %s",
+            dependencies,
+            manifest_name,
+        )
+        return []
+    normalized = []
+    for entry in dependencies:
+        if isinstance(entry, dict):
+            if _valid_dependency_entry(entry, manifest_name):
+                normalized.append(entry)
+        elif isinstance(entry, str) and entry:
+            # PIO also accepts a bare list of names ("dependencies": ["Wire"])
+            normalized.append({"name": entry})
+        else:
+            _LOGGER.warning(
+                "Ignoring unrecognized dependency entry %r of %s",
+                entry,
+                manifest_name,
+            )
+    return normalized
 
 
 @dataclass
@@ -688,6 +903,112 @@ def _node_key(
     return name, "registry", (owner, pkgname)
 
 
+def lib_ignore_set() -> set[str]:
+    """The ``lib_ignore`` names from ``esphome->platformio_options``,
+    normalized to lowercase short names (the part after the ``/``)."""
+    return {
+        name.split("/")[-1].lower()
+        for name in CORE.platformio_options.get("lib_ignore", [])
+    }
+
+
+def is_lib_ignored(name: str | None, lib_ignore: set[str]) -> bool:
+    """Whether ``name`` matches the normalized ``lib_ignore`` set."""
+    return (
+        bool(lib_ignore)
+        and name is not None
+        and (name.split("/")[-1].lower() in lib_ignore)
+    )
+
+
+def _content_lengths(urls: list[str]) -> list[int | None]:
+    """Content-Length per URL via HEAD requests; None when unknown."""
+    import requests
+
+    def head(url: str) -> int | None:
+        try:
+            resp = requests.head(url, timeout=10, allow_redirects=True)
+            if not resp.ok:
+                _LOGGER.debug("HEAD %s returned %s", url, resp.status_code)
+                return None
+            return int(resp.headers.get("content-length", 0)) or None
+        except (requests.RequestException, ValueError) as err:
+            _LOGGER.debug("HEAD %s failed: %s", url, err)
+            return None
+
+    with ThreadPoolExecutor(max_workers=min(BATCH_DOWNLOAD_WORKERS, len(urls))) as ex:
+        return list(ex.map(head, urls))
+
+
+def _prefetch_wave(
+    wave: list[tuple[str, ConvertedLibrary]], salt: str, namespace: str
+) -> None:
+    """Best-effort parallel download of a wave's registry archives.
+
+    The walk's own ``download()`` call stays authoritative (it surfaces real
+    failures, with resume); bars are suppressed since parallel bars would
+    interleave. Duplicate URLs prefetch once so two threads never extract
+    into the same cache directory.
+    """
+    components: list[ConvertedLibrary] = []
+    seen: set[str] = set()
+    for _key, component in wave:
+        if not isinstance(component.source, URLSource):
+            continue
+        if component.source.url in seen:
+            continue
+        seen.add(component.source.url)
+        try:
+            cached = component.source.is_cached(
+                component.get_sanitized_name(), salt=salt, namespace=namespace
+            )
+        except Exception as err:  # noqa: BLE001  # pylint: disable=broad-exception-caught
+            # Best-effort: a failing probe prefetches (and re-downloads)
+            _LOGGER.debug("Cache probe for %s failed: %s", component.name, err)
+            cached = False
+        if cached:
+            # A completed extraction downloads nothing; a warm build must
+            # stay silent
+            continue
+        components.append(component)
+    if len(components) < 2:
+        return
+    # One combined bar over the batch, sized by HEAD requests. An unknown
+    # size would mean a silent multi-MB download; fall back to sequential
+    # downloads with their per-file bars instead.
+    sizes = _content_lengths([c.source.url for c in components])
+    if not all(sizes):
+        # Announced before the sequential per-file downloads take over, so
+        # the fallback is distinguishable from a hang
+        _LOGGER.info(
+            "No Content-Length for %s; downloading sequentially",
+            ", ".join(
+                c.source.url
+                for c, size in zip(components, sizes, strict=True)
+                if not size
+            ),
+        )
+        return
+    _LOGGER.info(
+        "Downloading %d libraries: %s",
+        len(components),
+        ", ".join(c.name for c in components),
+    )
+
+    def _fetch(component: ConvertedLibrary):
+        return lambda tracker: component.download(
+            salt=salt, namespace=namespace, progress=tracker
+        )
+
+    failures = run_batch_downloads(
+        BatchDownloadProgress("Downloading libraries", sum(sizes)),
+        [(component.name, _fetch(component)) for component in components],
+    )
+    for name, err in failures:
+        # The sequential call below retries and raises the real error
+        _LOGGER.warning("Prefetch of %s failed (retrying sequentially): %s", name, err)
+
+
 def convert_libraries(
     libraries: list[Library], backend: LibraryBackend
 ) -> list[ConvertedLibrary]:
@@ -713,10 +1034,7 @@ def convert_libraries(
     """
     nodes: dict[str, _LibNode] = {}
 
-    lib_ignore = {
-        name.split("/")[-1].lower()
-        for name in CORE.platformio_options.get("lib_ignore", [])
-    }
+    lib_ignore = lib_ignore_set()
 
     # The generated build files inside the shared cache bake in the dependency
     # wiring, which lib_ignore changes; salt the cache path so configs with
@@ -728,11 +1046,6 @@ def convert_libraries(
         else ""
     )
 
-    def is_ignored(name: str | None) -> bool:
-        if not lib_ignore or name is None:
-            return False
-        return name.split("/")[-1].lower() in lib_ignore
-
     def add_spec(name: str | None, version: str | None, repository: str | None) -> str:
         key, kind, locator = _node_key(name, version, repository)
         node = nodes.get(key) or _LibNode(key=key, is_git=kind == "git")
@@ -781,7 +1094,7 @@ def convert_libraries(
     top_level = [
         add_spec(library.name, library.version, library.repository)
         for library in libraries
-        if not is_ignored(library.name)
+        if not is_lib_ignored(library.name, lib_ignore)
     ]
 
     # Collect + resolve to a fixpoint: a node is (re)resolved whenever its
@@ -792,105 +1105,132 @@ def convert_libraries(
     top_level_keys = set(top_level)
     worklist = deque(dict.fromkeys(top_level))
     while worklist:
-        key = worklist.popleft()
-        node = nodes[key]
+        # Drain the frontier sequentially (spec resolution mutates shared
+        # node state), then prefetch the wave's registry archives in
+        # parallel; the per-component download() below stays authoritative.
+        wave: list[tuple[str, ConvertedLibrary]] = []
+        while worklist:
+            key = worklist.popleft()
+            node = nodes[key]
 
-        # A node is queued once per referring edge; skip the (uncached) registry
-        # lookup + download + dependency walk unless its requirement set grew
-        # since the last resolve. Requirements only ever grow, so this still
-        # converges the fixpoint and terminates dependency cycles.
-        requirements = frozenset(node.requirements)
-        if resolved_requirements.get(key) == requirements:
-            continue
-        resolved_requirements[key] = requirements
+            # Re-resolve only when the requirement set grew; requirements
+            # only ever grow, so the fixpoint converges and cycles terminate
+            requirements = frozenset(node.requirements)
+            if resolved_requirements.get(key) == requirements:
+                continue
+            resolved_requirements[key] = requirements
 
-        if node.is_git:
-            component = ConvertedLibrary(key, "*", GitSource(node.url, node.ref))
-        elif node.is_local:
-            component = ConvertedLibrary(key, "*", LocalSource(node.local_path))
-        else:
-            owner, name, version, url = _resolve_registry_version(
-                node.owner, node.pkgname, node.requirements
-            )
-            component = ConvertedLibrary(
-                _owner_pkgname_to_name(owner, name), version, URLSource(url)
-            )
-        component.download(salt=salt, namespace=backend.cache_key)
+            if node.is_git:
+                component = ConvertedLibrary(key, "*", GitSource(node.url, node.ref))
+            elif node.is_local:
+                component = ConvertedLibrary(key, "*", LocalSource(node.local_path))
+            else:
+                owner, name, version, url = _resolve_registry_version(
+                    node.owner, node.pkgname, node.requirements
+                )
+                component = ConvertedLibrary(
+                    _owner_pkgname_to_name(owner, name), version, URLSource(url)
+                )
+            wave.append((key, component))
+        _prefetch_wave(wave, salt, backend.cache_key)
+        for key, component in wave:
+            node = nodes[key]
+            if frozenset(node.requirements) != resolved_requirements[key]:
+                # An earlier wave entry grew this node's requirements after
+                # the drain resolved it; downloading the superseded version
+                # would be wasted work, and the next wave re-resolves it
+                worklist.append(key)
+                continue
+            component.download(salt=salt, namespace=backend.cache_key)
 
-        source_dir = component.source_dir
-        library_json_path = source_dir / "library.json"
-        library_properties_path = source_dir / "library.properties"
-        has_json = library_json_path.is_file()
-        has_properties = library_properties_path.is_file()
-        if not has_json and not has_properties and not node.is_local:
-            # The shared cache can hold a broken copy (e.g. a clone or an
-            # extraction interrupted by a killed process). Force one
-            # re-download so a bad cache entry self-heals instead of failing
-            # every build until the user runs a full clean. A local source is
-            # read in place, so there is nothing to re-download.
-            _LOGGER.warning(
-                "Library %s at %s is missing library.json and library.properties; "
-                "re-downloading",
-                key,
-                source_dir,
-            )
-            component.download(force=True, salt=salt, namespace=backend.cache_key)
+            source_dir = component.source_dir
+            library_json_path = source_dir / "library.json"
+            library_properties_path = source_dir / "library.properties"
             has_json = library_json_path.is_file()
             has_properties = library_properties_path.is_file()
-        if has_json:
-            component.data = _parse_library_json(library_json_path)
-        elif has_properties:
-            component.data = _parse_library_properties(library_properties_path)
-        else:
-            # For a local library a missing manifest is user input, so raise
-            # EsphomeError (clean CLI message) like the missing-directory case;
-            # for registry/git a missing manifest means a corrupt cache, which
-            # is not user error, so keep RuntimeError.
-            error_cls = EsphomeError if node.is_local else RuntimeError
-            raise error_cls(
-                f"Invalid PIO library {key}: missing library.json and "
-                f"library.properties in {source_dir}"
-            )
+            if not has_json and not has_properties and not node.is_local:
+                # An interrupted clone/extraction self-heals with one forced
+                # re-download; a local source has nothing to re-download
+                _LOGGER.warning(
+                    "Library %s at %s is missing library.json and library.properties; "
+                    "re-downloading",
+                    key,
+                    source_dir,
+                )
+                component.download(force=True, salt=salt, namespace=backend.cache_key)
+                has_json = library_json_path.is_file()
+                has_properties = library_properties_path.is_file()
+            if has_json:
+                component.data = parse_library_json(library_json_path)
+            elif has_properties:
+                component.data = parse_library_properties(library_properties_path)
+            else:
+                # Local sources are user input (EsphomeError); a registry/git
+                # miss means a corrupt cache (RuntimeError)
+                error_cls = EsphomeError if node.is_local else RuntimeError
+                raise error_cls(
+                    f"Invalid PIO library {key}: missing library.json and "
+                    f"library.properties in {source_dir}"
+                )
 
-        try:
-            check_library_data(component.data, backend.platform, backend.framework)
-        except InvalidLibrary as e:
-            # Skip an incompatible transitive dependency, but fail fast if a
-            # top-level library the build explicitly requested is incompatible.
-            if key in top_level_keys:
-                raise RuntimeError(
-                    f"Requested library {key} is not compatible with "
-                    f"{backend.framework}: {e}"
-                ) from e
-            _LOGGER.debug("Skip incompatible dependency %s: %s", key, str(e))
-            continue
-        components[key] = component
+            if not isinstance(component.data, dict) or not isinstance(
+                component.data.get("build", {}), dict
+            ):
+                # A bare json.load imposes no shape; every backend dereferences
+                # data/build, so validate once here and name the library
+                raise EsphomeError(f"Library {key} has a malformed manifest")
+            warn_properties_depends(component.name, component.data)
 
-        # Requirements changed (we got past the short-circuit above), so
-        # (re)walk this component's dependencies.
-        node.edges = set()
-        for dependency in _normalize_dependencies(component.data.get("dependencies")):
-            if "name" not in dependency or "version" not in dependency:
-                continue
             try:
-                check_library_data(dependency, backend.platform, backend.framework)
+                check_library_data(component.data, backend.platform, backend.framework)
             except InvalidLibrary as e:
-                _LOGGER.debug("Skip dependency %s: %s", dependency.get("name"), str(e))
+                # An explicitly requested library fails fast; the routine
+                # cross-platform skip stays at debug, other causes warn
+                if key in top_level_keys:
+                    raise RuntimeError(
+                        f"Requested library {key} is not compatible with "
+                        f"{backend.framework}: {e}"
+                    ) from e
+                if isinstance(e, IncompatiblePlatform):
+                    _LOGGER.debug("Skip incompatible dependency %s: %s", key, str(e))
+                else:
+                    _LOGGER.warning("Skipping dependency %s: %s", key, str(e))
                 continue
-            dep_name = _owner_pkgname_to_name(
-                dependency.get("owner"), dependency.get("name")
-            )
-            if is_ignored(dep_name):
-                _LOGGER.debug("Skip ignored dependency %s", dep_name)
-                continue
-            # The version field may actually be a URL (git/archive dependency).
-            dep_version = dependency["version"]
-            dep_url = _url_or_none(dep_version)
-            if dep_url is not None:
-                dep_version = None
-            dep_key = add_spec(dep_name, dep_version, dep_url)
-            node.edges.add(dep_key)
-            worklist.append(dep_key)
+            components[key] = component
+
+            # Requirements changed (we got past the short-circuit above), so
+            # (re)walk this component's dependencies.
+            node.edges = set()
+            for dependency in normalize_dependencies(
+                component.data.get("dependencies"), component.name
+            ):
+                if "version" not in dependency:
+                    # Cannot resolve from the registry; the arduino-backend
+                    # PR adds the reconciliation that reports real drops
+                    _LOGGER.debug(
+                        "Skip version-less dependency %r of %s",
+                        dependency.get("name"),
+                        component.name,
+                    )
+                    continue
+                if not dependency_is_usable(
+                    dependency, backend.platform, backend.framework, component.name
+                ):
+                    continue
+                dep_name = _owner_pkgname_to_name(
+                    dependency.get("owner"), dependency.get("name")
+                )
+                if is_lib_ignored(dep_name, lib_ignore):
+                    _LOGGER.debug("Skip ignored dependency %s", dep_name)
+                    continue
+                # The version field may actually be a URL (git/archive dependency).
+                dep_version = dependency["version"]
+                dep_url = _url_or_none(dep_version)
+                if dep_url is not None:
+                    dep_version = None
+                dep_key = add_spec(dep_name, dep_version, dep_url)
+                node.edges.add(dep_key)
+                worklist.append(dep_key)
 
     # A git or local source wins over the same component requested from the
     # registry. That's intentional, but warn so the dropped registry spec isn't

esphome/platformio/registry.py

@@ -0,0 +1,280 @@
+"""Install packages from the PlatformIO registry without importing the
+platformio package (identical bits, esphome's own download machinery)."""
+
+from __future__ import annotations
+
+from collections.abc import Collection
+import io
+import json
+import logging
+import os
+from pathlib import Path
+import platform
+
+from esphome.core import EsphomeError
+from esphome.framework_helpers import (
+    BatchDownloadProgress,
+    archive_extract_all,
+    download_from_mirrors,
+    download_with_resume,
+    rmdir,
+    run_batch_downloads,
+)
+
+_LOGGER = logging.getLogger(__name__)
+
+_REGISTRY_URL = (
+    "https://api.registry.platformio.org/v3/packages/platformio/tool/{package}"
+)
+
+
+def get_systype() -> str:
+    """The registry system tag for the current host.
+
+    Transliterates ``platformio.util.get_systype()`` (same
+    ``PLATFORMIO_SYSTEM_TYPE`` override). Deviation: windows-arm64 maps to
+    ``windows_amd64`` (no arm64 toolchains; x86 emulation).
+    """
+    if systype := os.environ.get("PLATFORMIO_SYSTEM_TYPE"):
+        return systype
+    system = platform.system().lower()
+    arch = platform.machine().lower()
+    if system == "windows":
+        if not arch:  # same fallback as upstream (platformio issue #4353)
+            arch = "x86_" + platform.architecture()[0]
+        if "x86" in arch:
+            arch = "amd64" if "64" in arch else "x86"
+        elif arch == "arm64":
+            arch = "amd64"
+    if arch == "aarch64" and platform.architecture()[0] == "32bit":
+        # 64-bit kernel with a 32-bit userland (e.g. 32-bit Raspberry Pi OS)
+        arch = "armv7l"
+    return f"{system}_{arch}" if arch else system
+
+
+def registry_download(package: str, version: str) -> tuple[str, str, int | None]:
+    """Resolve a package's download URL, sha256, and size via the registry.
+
+    The metadata fetch goes through ``download_from_mirrors`` so it shares
+    the retry, backoff, and error reporting of every other download here.
+    """
+    buf = io.BytesIO()
+    download_from_mirrors([_REGISTRY_URL], {"package": package}, buf)
+    try:
+        data = json.loads(buf.getvalue())
+    except ValueError as err:
+        raise EsphomeError(
+            f"The package registry returned invalid JSON for {package}: {err}"
+        ) from err
+    if not isinstance(data, dict):
+        raise EsphomeError(
+            f"Unexpected package registry response for {package}: {str(data)[:200]}"
+        )
+    systype = get_systype()
+    versions = data.get("versions")
+    if not isinstance(versions, list):
+        # A schema change or an error/captive-portal payload must not be
+        # reported as "version not found"
+        raise EsphomeError(
+            f"Unexpected package registry response for {package}: {str(data)[:200]}"
+        )
+    for ver in versions:
+        if not isinstance(ver, dict):
+            raise EsphomeError(
+                f"Unexpected package registry response for {package}: {str(data)[:200]}"
+            )
+        if ver.get("name") != version:
+            continue
+        files = ver.get("files")
+        if not isinstance(files, list):
+            raise EsphomeError(
+                f"Unexpected package registry response for {package}: {str(ver)[:200]}"
+            )
+        for file in files:
+            if not isinstance(file, dict):
+                raise EsphomeError(
+                    f"Unexpected package registry response for {package}: "
+                    f"{str(ver)[:200]}"
+                )
+            # Only a missing key means "any system"; an empty list must not
+            # match, and a bare string would make ``in`` a substring test.
+            systems = file.get("system")
+            if systems is None:
+                systems = ["*"]
+            elif isinstance(systems, str):
+                systems = [systems]
+            elif not isinstance(systems, list):
+                # An int would make ``in`` a TypeError and a dict a key test
+                raise EsphomeError(
+                    f"Unexpected package registry response for {package}: "
+                    f"{str(file)[:200]}"
+                )
+            if "*" in systems or systype in systems:
+                sha256 = (file.get("checksum") or {}).get("sha256")
+                if not sha256:
+                    # Never extract an unverified archive; the registry
+                    # publishes a checksum for every package file.
+                    raise EsphomeError(
+                        f"The package registry returned no sha256 for "
+                        f"{package} {version}; refusing the unverified download"
+                    )
+                url = file.get("download_url")
+                if not url:
+                    raise EsphomeError(
+                        f"The package registry returned no download URL for "
+                        f"{package} {version}"
+                    )
+                return (url, sha256, file.get("size"))
+        raise EsphomeError(
+            f"No {package} {version} build for this platform ({systype})"
+        )
+    raise EsphomeError(f"{package} {version} not found in the package registry")
+
+
+def _check_layout(name: str, dest: Path, expect: Collection[str]) -> None:
+    """Raise when an install tree is missing an expected directory (runs on
+    fresh extracts and on marker hits)."""
+    for rel in expect:
+        if not (dest / rel).is_dir():
+            raise EsphomeError(
+                f"{name} at {dest} is missing the expected {rel} "
+                "directory; run 'esphome clean-all' and retry"
+            )
+
+
+def prefetch_packages(
+    packages: list[tuple[str, str, Path, list[str]]], downloads_dir: Path
+) -> None:
+    """Download pending package archives in parallel under one combined bar.
+
+    ``packages`` holds ``(name, version, dest, mirrors)`` per package. Purely
+    an optimization: ``install_package`` verifies every archive and
+    re-downloads anything this pass left unfinished. Mirror overrides and
+    registry entries without a size stay on the sequential path so its
+    per-file bars remain trustworthy. Each fetch holds the same per-dest
+    lock as ``install_package``: the archive's ``.part`` file is shared, and
+    two concurrent writers would truncate each other's bytes.
+    """
+    from filelock import FileLock
+
+    pending: list[tuple[str, str, Path, str, str, int]] = []
+    seen: set[str] = set()
+    for name, version, dest, mirrors in packages:
+        if mirrors or (dest / ".esphome_extracted").is_file():
+            continue
+        archive_name = f"{name}-{version}"
+        if archive_name in seen:
+            # A duplicate entry would race itself between two workers
+            continue
+        seen.add(archive_name)
+        try:
+            url, sha256, size = registry_download(name, version)
+        except EsphomeError as err:
+            # The sequential install reports the real failure with context
+            _LOGGER.debug("Prefetch resolve for %s failed: %s", name, err)
+            continue
+        if not size:
+            continue
+        archive = downloads_dir / archive_name
+        if archive.is_file() and archive.stat().st_size == size:
+            continue
+        pending.append((name, version, dest, url, sha256, size))
+    if len(pending) < 2:
+        return
+    downloads_dir.mkdir(parents=True, exist_ok=True)
+    _LOGGER.info(
+        "Downloading %d package archive(s): %s",
+        len(pending),
+        ", ".join(name for name, *_ in pending),
+    )
+
+    def _fetch(entry: tuple[str, str, Path, str, str, int]):
+        name, version, dest, url, sha256, size = entry
+
+        def fetch(tracker):
+            dest.parent.mkdir(parents=True, exist_ok=True)
+            with FileLock(f"{dest}.lock", fallback_to_soft=False):
+                download_with_resume(
+                    url,
+                    downloads_dir / f"{name}-{version}",
+                    sha256=sha256,
+                    size=size,
+                    progress=tracker,
+                )
+
+        return fetch
+
+    failures = run_batch_downloads(
+        BatchDownloadProgress(
+            "Downloading packages", sum(size for *_, size in pending)
+        ),
+        [(entry[0], _fetch(entry)) for entry in pending],
+    )
+    for name, err in failures:
+        if isinstance(err, (EsphomeError, OSError)):
+            # Expected download failures: install_package retries this one
+            # itself, with a visible bar
+            _LOGGER.debug("Prefetch of %s failed: %s", name, err)
+        else:
+            # Anything else is a programming error that would otherwise
+            # become a permanent silent no-op
+            _LOGGER.warning("Prefetch of %s failed: %r", name, err)
+
+
+def install_package(
+    name: str,
+    version: str,
+    dest: Path,
+    mirrors: list[str],
+    downloads_dir: Path,
+    expect: Collection[str],
+) -> None:
+    """Download, verify, and extract one package if not already installed.
+
+    The registry path is integrity-checked against the sha256 the registry
+    publishes; a mirror override (URL templates with ``{VERSION}``/``{SYSTEM}``
+    substitution) is trusted as configured. ``downloads_dir`` holds the
+    archive between runs so an interrupted download resumes.
+    """
+    if not expect:
+        # Layout validation before marker.touch() is the only guard against
+        # caching a truncated mirror archive as a good install
+        raise ValueError("install_package requires a non-empty expect")
+    marker = dest / ".esphome_extracted"
+    if marker.is_file():
+        _check_layout(name, dest, expect)
+        return
+    from filelock import FileLock
+
+    # Serialize concurrent cold builds (same filelock pattern as git.py).
+    dest.parent.mkdir(parents=True, exist_ok=True)
+    # A soft-lock fallback would turn a hard-killed run into a permanent
+    # hang (see git.py).
+    with FileLock(f"{dest}.lock", fallback_to_soft=False):
+        if marker.is_file():
+            # Another process finished the install while we waited
+            return
+        rmdir(dest, msg=f"Clean up incomplete {name} install")
+        # Persistent location so an interrupted download resumes across runs.
+        downloads_dir.mkdir(parents=True, exist_ok=True)
+        archive = downloads_dir / f"{name}-{version}"
+        _LOGGER.info("Downloading %s %s ...", name, version)
+        if mirrors:
+            _LOGGER.warning(
+                "Downloading %s from a mirror override; checksum verification "
+                "is skipped for mirrors",
+                name,
+            )
+            download_from_mirrors(
+                mirrors, {"VERSION": version, "SYSTEM": get_systype()}, archive
+            )
+        else:
+            url, sha256, size = registry_download(name, version)
+            download_with_resume(url, archive, sha256=sha256, size=size)
+        _LOGGER.info("Extracting %s ...", name)
+        archive_extract_all(archive, dest, progress_header="Extracting")
+        # Validate the layout before recording success, so an unexpected
+        # package is never cached as a working install.
+        _check_layout(name, dest, expect)
+        marker.touch()
+        archive.unlink(missing_ok=True)

esphome/platformio/toolchain.py

@@ -4,19 +4,18 @@ import logging
 import os
 from pathlib import Path
 import re
-import shutil
-import subprocess
 import sys
 from typing import TYPE_CHECKING, Any
 
 import platformdirs
 
+from esphome.build_helpers.ccache import resolve_ccache_path
 from esphome.const import CONF_COMPILE_PROCESS_LIMIT, CONF_ESPHOME, KEY_CORE
 from esphome.core import CORE, EsphomeError
+from esphome.framework_helpers import strip_win_long_path_prefix
 from esphome.helpers import (
     add_git_ceiling_directory,
     copy_file_if_changed,
-    get_bool_env,
     rmtree,
     write_file,
 )
@@ -41,40 +40,6 @@ _PIO_PYTHON_STAMP_LOCK = ".esphome.pio.stamp.lock"
 _PIO_PYTHON_STAMP_SCHEMA = "0"
 
 
-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.
-
-    Also applied to the ccache path exported by ``_ccache_env()``, which
-    ``shutil.which`` can return with the same prefix.
-
-    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 get_platformio_config() -> "ProjectConfig | None":
     """Return PlatformIO's ``ProjectConfig``, or None when PlatformIO is absent."""
     try:
@@ -238,32 +203,6 @@ def _check_platformio_python_stamp(config: "ProjectConfig") -> None:
         _write_pio_stamp_python(stamp_file, current)
 
 
-def _ccache_runs(ccache: str) -> bool:
-    """Return True when the ``ccache`` found on PATH actually runs.
-
-    ``shutil.which`` proves existence, not runnability: on Windows it also
-    matches ``.bat``/``.cmd`` wrappers and stale package-manager shims whose
-    target is gone. Wrapping compiles around such a find fails every compile
-    step with an opaque OS error, so probe once and fall back to compiling
-    without ccache when the probe fails.
-    """
-    try:
-        subprocess.run(
-            [ccache, "--version"],
-            check=True,
-            stdout=subprocess.DEVNULL,
-            stderr=subprocess.DEVNULL,
-            timeout=15,
-        )
-    except (OSError, subprocess.SubprocessError):
-        _LOGGER.warning(
-            "Ignoring ccache at %s because it failed to run; compiling without ccache",
-            ccache,
-        )
-        return False
-    return True
-
-
 def _ccache_env() -> dict[str, str]:
     r"""Return ccache settings for PlatformIO builds.
 
@@ -282,7 +221,7 @@ def _ccache_env() -> dict[str, str]:
     runs fine through ``CreateProcess``, which is how ESP-IDF invokes it,
     but SCons runs every compile through ``cmd.exe``, which fails on it with
     "The system cannot find the path specified." (#18399), so the prefix is
-    stripped here with ``_strip_win_long_path_prefix()`` before the
+    stripped here with ``strip_win_long_path_prefix()`` before the
     runnability probe, which therefore validates the exact string the build
     will execute.
     ``ESPHOME_CCACHE_PATH`` is an internal channel, not a user setting: the
@@ -308,22 +247,8 @@ def _ccache_env() -> dict[str, str]:
     build dir. The other ``CCACHE_*`` values the user already set in the
     environment are respected.
     """
-    explicit = "ESPHOME_CCACHE_ENABLE" in os.environ
-    if explicit and not get_bool_env("ESPHOME_CCACHE_ENABLE"):
-        return {"ESPHOME_CCACHE_ENABLE": "0"}
-    ccache_path = shutil.which("ccache")
+    ccache_path = resolve_ccache_path()
     if ccache_path is None:
-        if explicit:
-            _LOGGER.warning(
-                "ESPHOME_CCACHE_ENABLE is set but no ccache binary is on PATH; "
-                "compiling without ccache"
-            )
-        return {"ESPHOME_CCACHE_ENABLE": "0"}
-    # Strip before probing so the probe validates (and the failure warning
-    # names) the exact string the build will execute through cmd.exe.
-    ccache_path = _strip_win_long_path_prefix(ccache_path)
-    # An explicit opt-in skips the runnability probe.
-    if not explicit and not _ccache_runs(ccache_path):
         return {"ESPHOME_CCACHE_ENABLE": "0"}
     env = {
         "ESPHOME_CCACHE_ENABLE": "1",
@@ -385,7 +310,7 @@ def run_platformio_cli(*args, **kwargs) -> str | int:
     # 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)
+    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

esphome/writer.py

@@ -288,11 +288,13 @@ def copy_src_tree():
             # Source file removed, delete target
             p.unlink()
             if target not in generated_files:
+                _LOGGER.debug("Source removed: %s", target)
                 sources_changed = True
         else:
             src_file = source_files_copy.pop(target)
             with src_file.path() as src_path:
                 if copy_file_if_changed(src_path, p) and target not in generated_files:
+                    _LOGGER.debug("Source changed: %s", target)
                     sources_changed = True
 
     # Now copy new files
@@ -303,21 +305,25 @@ def copy_src_tree():
                 copy_file_if_changed(src_path, dst_path)
                 and target not in generated_files
             ):
+                _LOGGER.debug("Source added: %s", target)
                 sources_changed = True
 
     # Finally copy defines
     if write_file_if_changed(
         CORE.relative_src_path("esphome", "core", "defines.h"), generate_defines_h()
     ):
+        _LOGGER.debug("Source changed: esphome/core/defines.h")
         sources_changed = True
     write_file_if_changed(CORE.relative_build_path("README.txt"), ESPHOME_README_TXT)
     if write_file_if_changed(
         CORE.relative_src_path("esphome.h"), ESPHOME_H_FORMAT.format(include_s)
     ):
+        _LOGGER.debug("Source changed: esphome.h")
         sources_changed = True
     if write_file_if_changed(
         CORE.relative_src_path("esphome", "core", "version.h"), generate_version_h()
     ):
+        _LOGGER.debug("Source changed: esphome/core/version.h")
         sources_changed = True
 
     # Generate new build_info files if needed
@@ -332,18 +338,13 @@ def copy_src_tree():
 
     # Defensively force a rebuild if the build_info files don't exist, or if
     # there was a config change which didn't actually cause a source change
-    if not build_info_data_h_path.exists() or not build_info_data_cpp_path.exists():
+    if _build_info_stale(
+        build_info_data_h_path,
+        build_info_data_cpp_path,
+        build_info_json_path,
+        config_hash,
+    ):
         sources_changed = True
-    else:
-        try:
-            existing = json.loads(build_info_json_path.read_text(encoding="utf-8"))
-            if (
-                existing.get("config_hash") != config_hash
-                or existing.get("esphome_version") != __version__
-            ):
-                sources_changed = True
-        except (json.JSONDecodeError, KeyError, OSError):
-            sources_changed = True
 
     # Write build_info header and JSON metadata
     if sources_changed:
@@ -397,6 +398,38 @@ def generate_version_h():
     )
 
 
+def _build_info_stale(
+    h_path: Path, cpp_path: Path, json_path: Path, config_hash: int
+) -> bool:
+    """Whether the build-info sources must regenerate (missing or stale)."""
+    if not h_path.exists() or not cpp_path.exists():
+        _LOGGER.debug("Build info files missing; regenerating")
+        return True
+    try:
+        existing = json.loads(json_path.read_text(encoding="utf-8"))
+    except (json.JSONDecodeError, OSError):
+        _LOGGER.debug("Build info JSON unreadable; regenerating")
+        return True
+    if not isinstance(existing, dict):
+        # Valid JSON that is not an object (truncated or hand-edited) is
+        # stale, not a traceback
+        _LOGGER.debug("Build info JSON malformed; regenerating")
+        return True
+    if (
+        existing.get("config_hash") != config_hash
+        or existing.get("esphome_version") != __version__
+    ):
+        _LOGGER.debug(
+            "Build info stale (config_hash %s -> %s, version %s -> %s)",
+            existing.get("config_hash"),
+            config_hash,
+            existing.get("esphome_version"),
+            __version__,
+        )
+        return True
+    return False
+
+
 def get_build_info() -> tuple[int, int, str, str]:
     """Calculate build_info values from current config.
 
@@ -657,14 +690,17 @@ def clean_all(configuration: list[str]):
     # the per-config loop above can't reach. Wipe the default cache root
     # (also catches leftovers from older install layouts), then the resolved
     # install paths for the ESPHOME_*_PREFIX overrides (docker/add-on/CI)
-    # that live outside it.
+    # that live outside it. Every backend's cache is listed in
+    # TOOLS_CACHE_SPECS, so registering one there is the only step.
     import platformdirs
 
-    from esphome.components.nrf52.framework import get_sdk_nrf_tools_path
-    from esphome.espidf.framework import get_idf_tools_path
+    from esphome.build_helpers.tools_cache import TOOLS_CACHE_SPECS, tools_cache_path
 
     cache_root = Path(platformdirs.user_cache_dir("esphome", appauthor=False)).resolve()
-    for install_path in (cache_root, get_idf_tools_path(), get_sdk_nrf_tools_path()):
+    install_paths = [cache_root] + [
+        tools_cache_path(*spec) for spec in TOOLS_CACHE_SPECS
+    ]
+    for install_path in install_paths:
         if install_path.is_dir():
             _LOGGER.info("Deleting %s", install_path)
             rmtree(install_path)

requirements.txt

@@ -28,6 +28,7 @@ smpclient==7.2.0
 requests==2.34.2
 py7zr==1.1.3
 platformdirs==4.11.3  # native esp-idf toolchain global cache dir
+ninja==1.13.0  # native esp8266 arduino toolchain build driver
 filelock==3.32.3  # inter-process locks (PlatformIO cache heal, git clone cache); >=3.32 for FileLock(fallback_to_soft=...), older versions silently drop the kwarg
 
 # esp-idf >= 5.0 requires this

script/determine-jobs.py

@@ -525,13 +525,21 @@ def _esp32_platformio_path_or_file_trigger(files: list[str]) -> bool:
     return False
 
 
-# ESP-IDF infra: changes under esphome/espidf/ or to the IDF build generator
-# affect every esp32 IDF build (now the default toolchain) but aren't
+# Native-build infra: changes under esphome/espidf/, the shared
+# esphome/build_helpers/ package, or the modules the native ESP-IDF build
+# imports affect every esp32 IDF build (now the default toolchain) but aren't
 # components, so the component matrix wouldn't otherwise force any esp32
 # compile. When they change we fold the `esp32` component into the matrix so
 # the default native-IDF build path is still compiled on an infra-only PR.
-ESP_IDF_INFRA_TRIGGER_PATH_PREFIXES = ("esphome/espidf/",)
-ESP_IDF_INFRA_TRIGGER_FILES = frozenset({"esphome/build_gen/espidf.py"})
+ESP_IDF_INFRA_TRIGGER_PATH_PREFIXES = ("esphome/espidf/", "esphome/build_helpers/")
+ESP_IDF_INFRA_TRIGGER_FILES = frozenset(
+    {
+        "esphome/build_gen/espidf.py",
+        "esphome/framework_helpers.py",
+        "esphome/platformio/library.py",
+        "esphome/platformio/extra_script.py",
+    }
+)
 
 
 def _esp_idf_infra_changed(files: list[str]) -> bool:

tests/component_tests/esp32/test_esp32.py

@@ -131,6 +131,20 @@ def test_esp32_rejects_unsupported_toolchains(
         CONFIG_SCHEMA({"variant": VARIANT_ESP32, "toolchain": config_toolchain})
 
 
+def test_esp32_rejects_unsupported_cli_toolchain(
+    set_core_config: SetCoreConfigCallable,
+) -> None:
+    """A --toolchain the platform cannot serve fails instead of silently
+    building with PlatformIO (the CLI path bypasses the YAML validator)."""
+    set_core_config(PlatformFramework.ESP32_IDF)
+
+    from esphome.components.esp32 import CONFIG_SCHEMA
+
+    CORE.toolchain = Toolchain.ARDUINO
+    with pytest.raises(cv.Invalid, match="Unsupported toolchain 'arduino'"):
+        CONFIG_SCHEMA({"variant": VARIANT_ESP32})
+
+
 @pytest.mark.parametrize(
     ("config", "error_match"),
     [

tests/dummy_main.cpp

@@ -29,6 +29,7 @@ void setup() {
 
   auto *ota = new esphome::ESPHomeOTAComponent();  // NOLINT
   ota->set_port(8266);
+  App.register_component_(ota);
 
   App.setup();
 }

tests/script/test_determine_jobs.py

@@ -1120,7 +1120,14 @@ def test_should_run_esp32_platformio_with_branch() -> None:
         (["esphome/espidf/runner.py"], True),
         (["esphome/espidf/framework.py"], True),
         (["esphome/build_gen/espidf.py"], True),
-        # PlatformIO build gen and esp32 component are NOT IDF-infra triggers
+        # Shared native-build modules the IDF build imports -> trigger
+        (["esphome/build_helpers/idedata.py"], True),
+        (["esphome/platformio/library.py"], True),
+        (["esphome/framework_helpers.py"], True),
+        (["esphome/platformio/extra_script.py"], True),
+        # PlatformIO build gen, its toolchain, and the esp32 component are
+        # NOT IDF-infra triggers
+        (["esphome/platformio/toolchain.py"], False),
         (["esphome/build_gen/platformio.py"], False),
         (["esphome/components/esp32/__init__.py"], False),
         (["README.md"], False),

tests/unit_tests/analyze_memory/test_build_artifacts.py

@@ -9,7 +9,7 @@ from esphome.analyze_memory.toolchain import (
     find_idedata_path,
     idedata_candidates,
 )
-from esphome.espidf.idedata import _cc_path_from_cxx
+from esphome.build_helpers.idedata import _cc_path_from_cxx
 from esphome.platformio.toolchain import IDEData
 
 

tests/unit_tests/build_helpers/test_ccache.py

@@ -0,0 +1,119 @@
+"""Tests for the shared ccache policy in esphome.build_helpers.ccache."""
+
+from __future__ import annotations
+
+import os
+from pathlib import Path
+from types import SimpleNamespace
+from unittest.mock import patch
+
+import pytest
+
+from esphome.build_helpers import ccache
+
+
+def test_resolve_opt_out() -> None:
+    with patch.dict(os.environ, {"ESPHOME_CCACHE_ENABLE": "0"}):
+        assert ccache.resolve_ccache_path() is None
+
+
+def test_resolve_no_binary(caplog: pytest.LogCaptureFixture) -> None:
+    with (
+        patch.dict(os.environ, {}, clear=True),
+        patch("shutil.which", return_value=None),
+    ):
+        assert ccache.resolve_ccache_path() is None
+    assert "no ccache binary" not in caplog.text
+
+
+def test_resolve_probe_failure() -> None:
+    with (
+        patch.dict(os.environ, {}, clear=True),
+        patch("shutil.which", return_value="/usr/bin/ccache"),
+        patch("esphome.framework_helpers.subprocess.run", side_effect=OSError("boom")),
+    ):
+        assert ccache.resolve_ccache_path() is None
+
+
+def test_resolve_explicit_skips_probe_and_warns_missing(
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    with (
+        patch.dict(os.environ, {"ESPHOME_CCACHE_ENABLE": "1"}, clear=True),
+        patch("shutil.which", return_value="/usr/bin/ccache"),
+        patch.object(ccache, "_ccache_runs", side_effect=AssertionError),
+    ):
+        assert ccache.resolve_ccache_path() == "/usr/bin/ccache"
+    with (
+        patch.dict(os.environ, {"ESPHOME_CCACHE_ENABLE": "1"}, clear=True),
+        patch("shutil.which", return_value=None),
+    ):
+        assert ccache.resolve_ccache_path() is None
+    assert "no ccache binary is on PATH" in caplog.text
+
+
+def test_probe_spawns_with_close_fds_false() -> None:
+    with patch("esphome.framework_helpers.subprocess.run") as mock_run:
+        assert ccache._ccache_runs("/usr/bin/ccache") is True
+    assert mock_run.call_args.kwargs["close_fds"] is False
+
+
+def test_defaults_env(tmp_path: Path) -> None:
+    with (
+        patch("esphome.core.CORE", SimpleNamespace(build_path=tmp_path / "b")),
+        patch.dict(os.environ, {"CCACHE_NOHASHDIR": "false"}, clear=True),
+    ):
+        env = ccache.ccache_defaults_env(tmp_path / "cache")
+    assert env["CCACHE_DIR"] == str(tmp_path / "cache")
+    assert env["CCACHE_DEPEND"] == "1"
+    assert "CCACHE_NOHASHDIR" not in env  # user value respected
+
+
+def test_defaults_env_requires_build_path() -> None:
+    with (
+        patch("esphome.core.CORE", SimpleNamespace(build_path=None)),
+        pytest.raises(ValueError, match="build_path"),
+    ):
+        ccache.ccache_defaults_env(Path("/x"))
+
+
+@pytest.mark.parametrize("value", ["no", "off", "false", "0"])
+def test_resolve_opt_out_synonyms(value: str) -> None:
+    """Every recognized falsy spelling disables ccache."""
+    with patch.dict(os.environ, {"ESPHOME_CCACHE_ENABLE": value}):
+        assert ccache.resolve_ccache_path() is None
+
+
+def test_resolve_unrecognized_value_warns_and_probes(
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    """An unparsable ESPHOME_CCACHE_ENABLE is treated as unset: it must not
+    silently enable ccache or skip the runnability probe."""
+    with (
+        patch.dict(os.environ, {"ESPHOME_CCACHE_ENABLE": "enabled"}),
+        patch("shutil.which", return_value="/usr/bin/ccache"),
+        patch.object(ccache, "_ccache_runs", return_value=False) as mock_probe,
+    ):
+        assert ccache.resolve_ccache_path() is None
+    mock_probe.assert_called_once()
+    assert "unrecognized ESPHOME_CCACHE_ENABLE" in caplog.text
+
+
+@pytest.mark.parametrize(
+    ("raw", "expected"),
+    [
+        ("1", True),
+        ("enable", True),
+        ("ON", True),
+        ("0", False),
+        ("disable", False),
+        ("Off", False),
+        ("maybe", None),
+    ],
+)
+def test_parse_enable_env_spelling_tables(
+    monkeypatch: pytest.MonkeyPatch, raw: str, expected: bool | None
+) -> None:
+    """cv.boolean's spelling tables plus the 1/0 env convention."""
+    monkeypatch.setenv("ESPHOME_CCACHE_ENABLE", raw)
+    assert ccache.parse_enable_env("ESPHOME_CCACHE_ENABLE") is expected

tests/unit_tests/build_helpers/test_idedata.py

@@ -0,0 +1,588 @@
+"""Tests for esphome.build_helpers.idedata (compile_commands.json -> idedata)."""
+
+# pylint: disable=protected-access
+
+import json
+import logging
+import os
+from pathlib import Path
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from esphome.build_helpers import idedata
+from esphome.core import EsphomeError
+
+# An absolute, forward-slash (shlex-safe) path prefix valid on the host OS, so
+# tests exercise the same is-absolute / normalize behavior as a real compile DB
+# (a drive-qualified path on Windows, a leading slash elsewhere).
+ABS = "C:/" if os.name == "nt" else "/"
+
+
+def _entry(directory: str, file: str, command: str) -> dict:
+    return {"directory": directory, "file": file, "command": command}
+
+
+def test_parse_entry_extracts_fields() -> None:
+    """cxx_path, defines, includes and remaining flags are split apart."""
+    entry = _entry(
+        f"{ABS}build",
+        f"{ABS}build/src/esphome/core/application.cpp",
+        f"/tools/xtensa-esp32-elf-g++ -DUSE_ESP32 -DESPHOME_LOG_LEVEL=5 "
+        f"-I{ABS}inc/a -isystem {ABS}sys/b -std=gnu++20 -c app.cpp -o app.cpp.o",
+    )
+
+    cxx_path, defines, includes, cxx_flags = idedata.parse_entry(entry)
+
+    assert cxx_path == "/tools/xtensa-esp32-elf-g++"
+    assert "USE_ESP32" in defines
+    assert "ESPHOME_LOG_LEVEL=5" in defines
+    assert f"{ABS}inc/a" in includes
+    assert f"{ABS}sys/b" in includes
+    assert "-std=gnu++20" in cxx_flags
+    # input/output files and their flags are not treated as flags
+    assert "-c" not in cxx_flags
+    assert "-o" not in cxx_flags
+    assert "app.cpp" not in cxx_flags
+    assert "app.cpp.o" not in cxx_flags
+
+
+def test_parse_entry_space_separated_args() -> None:
+    """``-D X`` / ``-I path`` (separate arg) and ``-isystem<path>`` (joined)."""
+    entry = _entry(
+        f"{ABS}build",
+        f"{ABS}build/src/esphome/x.cpp",
+        f"g++ -D FOO=1 -I {ABS}inc/sep -isystem{ABS}sys/joined -c x.cpp",
+    )
+
+    _, defines, includes, _ = idedata.parse_entry(entry)
+
+    assert "FOO=1" in defines
+    assert f"{ABS}inc/sep" in includes
+    assert f"{ABS}sys/joined" in includes
+
+
+def test_parse_entry_resolves_relative_includes() -> None:
+    """Relative includes are resolved against the entry's ``directory``."""
+    directory = f"{ABS}build/proj"
+    entry = _entry(
+        directory,
+        f"{directory}/src/esphome/x.cpp",
+        "g++ -Iconfig -I../shared -isystem rel/sys -c x.cpp",
+    )
+
+    _, _, includes, _ = idedata.parse_entry(entry)
+
+    def resolved(rel: str) -> str:
+        # parse_entry emits forward slashes for consistency (normpath would
+        # yield backslashes on Windows).
+        return os.path.normpath(Path(directory) / rel).replace("\\", "/")
+
+    assert resolved("config") in includes
+    assert resolved("../shared") in includes  # ../ normalized away
+    assert resolved("rel/sys") in includes
+    # nothing is left relative
+    assert all(Path(inc).is_absolute() for inc in includes)
+
+
+def test_parse_entry_skips_dependency_flags() -> None:
+    """Dependency-generation flags (and their args) are dropped."""
+    entry = _entry(
+        "/build",
+        "/build/src/esphome/x.cpp",
+        "g++ -MD -MT x.cpp.o -MF x.cpp.o.d -c x.cpp -o x.cpp.o",
+    )
+
+    _, _, _, cxx_flags = idedata.parse_entry(entry)
+
+    for tok in ("-MD", "-MT", "x.cpp.o", "-MF", "x.cpp.o.d", "-c", "-o", "x.cpp"):
+        assert tok not in cxx_flags
+
+
+def test_expand_response_files(tmp_path: Path) -> None:
+    """``@file`` arguments are inlined relative to the directory."""
+    rsp = tmp_path / "flags.rsp"
+    rsp.write_text("-DFROM_RSP -I/rsp/inc")
+
+    tokens = idedata._expand_response_files(
+        ["g++", f"@{rsp.name}", "-c", "x.cpp"], tmp_path
+    )
+
+    assert "-DFROM_RSP" in tokens
+    assert "-I/rsp/inc" in tokens
+    assert not any(t.startswith("@") for t in tokens)
+
+
+def test_expand_response_files_keeps_literal_when_missing(tmp_path: Path) -> None:
+    """An unreadable ``@file`` token is kept verbatim rather than dropped."""
+    tokens = idedata._expand_response_files(["g++", "@nope.rsp"], tmp_path)
+    assert "@nope.rsp" in tokens
+
+
+def test_pick_entry_prefers_esphome_tu() -> None:
+    """A ``/src/esphome/`` C++ TU is picked over other compile entries."""
+    entries = [
+        _entry("/b", "/b/managed_components/foo/foo.c", "gcc -c foo.c"),
+        _entry("/b", "/b/src/esphome/core/app.cpp", "g++ -c app.cpp"),
+    ]
+    assert idedata._pick_entry(entries)["file"].endswith("app.cpp")
+
+
+def test_pick_entry_falls_back_to_any_cxx_tu() -> None:
+    """With no ``/src/esphome/`` TU present, the first C++ entry is the fallback."""
+    entries = [
+        _entry("/b", "/b/managed_components/foo/foo.c", "gcc -c foo.c"),
+        _entry("/b", "/b/components/x/x.cpp", "g++ -c x.cpp"),
+    ]
+    assert idedata._pick_entry(entries)["file"].endswith("x.cpp")
+
+
+def test_is_esphome_src_handles_backslash_paths() -> None:
+    r"""The src marker must match Windows ``\src\esphome\`` paths too.
+
+    compile_commands ``file`` entries use the OS-native separator; if the
+    marker only matched forward slashes no source would match on Windows and
+    the build-include union would be silently empty.
+    """
+    assert idedata._is_esphome_src(r"C:\b\src\esphome\core\app.cpp")
+    assert idedata._is_esphome_src("/b/src/esphome/core/app.cpp")
+    # non-esphome and non-C++ still rejected regardless of separator
+    assert not idedata._is_esphome_src(r"C:\b\managed_components\x\x.cpp")
+    assert not idedata._is_esphome_src(r"C:\b\src\esphome\core\app.h")
+
+
+@pytest.mark.parametrize(
+    ("command", "launcher"),
+    [
+        ("", None),
+        # A command that is only the launcher strips to nothing
+        ("/usr/bin/ccache", "/usr/bin/ccache"),
+    ],
+)
+def test_parse_entry_empty_command_raises(command: str, launcher: str | None) -> None:
+    """A blank (or launcher-only) command fails with a named ValueError,
+    not an IndexError."""
+    entry = {"directory": "/b", "file": "/b/src/x.cpp", "command": command}
+    with pytest.raises(ValueError, match="empty compile command"):
+        idedata.parse_entry(entry, launcher)
+
+
+def test_idedata_from_build_empty_includes_raises(tmp_path: Path) -> None:
+    """A compile DB with no ESPHome TU is never usable idedata and must
+    not be cached (call sites downgrade the raise to a build warning)."""
+    compile_commands = tmp_path / "compile_commands.json"
+    compile_commands.write_text(
+        json.dumps(
+            [
+                _entry(
+                    f"{ABS}build",
+                    f"{ABS}build/other/lib.cpp",
+                    "/tools/g++ -c other/lib.cpp -o lib.o",
+                )
+            ]
+        )
+    )
+    with (
+        patch.object(idedata, "get_toolchain_includes", return_value=[]),
+        pytest.raises(EsphomeError, match="No ESPHome translation unit found"),
+    ):
+        idedata.idedata_from_build(compile_commands)
+
+
+def test_idedata_from_build_rsp_commands_never_dedupe(tmp_path: Path) -> None:
+    """Per-object response files strip to one shape while holding different
+    include sets; @-commands must tokenize per TU."""
+    entries = []
+    for name in ("a", "b"):
+        rsp = tmp_path / f"{name}.cpp.o.rsp"
+        rsp.write_text(f"-I{ABS}inc/{name}")
+        file = f"{ABS}build/src/esphome/core/{name}.cpp"
+        entries.append(
+            {
+                "directory": str(tmp_path),
+                "file": file,
+                "command": f"/tools/g++ @{rsp.name} -c {file} -o {name}.o",
+                "output": f"{name}.o",
+            }
+        )
+    compile_commands = tmp_path / "compile_commands.json"
+    compile_commands.write_text(json.dumps(entries))
+    with patch.object(idedata, "get_toolchain_includes", return_value=[]):
+        data = idedata.idedata_from_build(compile_commands)
+    joined = " ".join(data["includes"]["build"])
+    assert "inc/a" in joined and "inc/b" in joined
+
+
+def test_idedata_from_build_dedupes_identical_command_shapes(
+    tmp_path: Path,
+) -> None:
+    """Translation units sharing one ninja rule (same command modulo
+    file/output) carry
+    identical includes, so only one per shape is tokenized; a differing
+    shape still contributes its includes."""
+
+    def _tu(name: str, inc: str) -> dict:
+        # ninja's compdb embeds the file and output strings verbatim
+        file = f"{ABS}build/src/esphome/core/{name}.cpp"
+        return _entry(
+            f"{ABS}build", file, f"/tools/g++ -I{ABS}inc/{inc} -c {file} -o {name}.o"
+        ) | {"output": f"{name}.o"}
+
+    entries = [_tu(name, "shared") for name in ("application", "component", "helpers")]
+    entries.append(_tu("extra", "extra"))
+    compile_commands = tmp_path / "compile_commands.json"
+    compile_commands.write_text(json.dumps(entries))
+    with (
+        patch.object(idedata, "get_toolchain_includes", return_value=[]),
+        patch.object(idedata, "parse_entry", wraps=idedata.parse_entry) as spy,
+    ):
+        data = idedata.idedata_from_build(compile_commands)
+    includes = set(data["includes"]["build"])
+    assert f"{ABS}inc/shared".replace("\\", "/") in {
+        i.replace("\\", "/") for i in includes
+    }
+    assert any("inc/extra" in i for i in includes)
+    # Representative + one distinct shape; the two same-shape duplicates
+    # are never tokenized
+    assert spy.call_count == 2
+
+
+def test_idedata_from_build(tmp_path: Path) -> None:
+    """Full transform: representative entry + include union + toolchain dirs."""
+    compile_commands = tmp_path / "compile_commands.json"
+    entries = [
+        _entry(
+            f"{ABS}b",
+            f"{ABS}b/src/esphome/core/app.cpp",
+            f"g++ -DUSE_ESP32 -I{ABS}inc/core -std=gnu++20 -c app.cpp -o app.cpp.o",
+        ),
+        _entry(
+            f"{ABS}b",
+            f"{ABS}b/src/esphome/sensor/s.cpp",
+            f"g++ -DUSE_ESP32 -I{ABS}inc/sensor -c s.cpp -o s.cpp.o",
+        ),
+        # non-esphome TU: its includes must not leak into the union
+        _entry(
+            f"{ABS}b",
+            f"{ABS}b/managed_components/x/x.c",
+            f"gcc -I{ABS}inc/managed -c x.c",
+        ),
+    ]
+    compile_commands.write_text(json.dumps(entries))
+
+    fake_proc = MagicMock(
+        returncode=0,
+        stderr=(
+            "ignored\n"
+            "#include <...> search starts here:\n"
+            " /tc/inc/c++\n"
+            " /tc/inc\n"
+            "End of search list.\n"
+            "more ignored\n"
+        ),
+    )
+    with patch.object(idedata.subprocess, "run", return_value=fake_proc):
+        data = idedata.idedata_from_build(compile_commands)
+
+    assert data["cxx_path"] == "g++"
+    assert "USE_ESP32" in data["defines"]
+    assert "-std=gnu++20" in data["cxx_flags"]
+    # include dirs unioned across all esphome TUs
+    assert f"{ABS}inc/core" in data["includes"]["build"]
+    assert f"{ABS}inc/sensor" in data["includes"]["build"]
+    # the non-esphome TU is excluded from the union
+    assert f"{ABS}inc/managed" not in data["includes"]["build"]
+    # toolchain search dirs parsed from the compiler's -v output
+    assert data["includes"]["toolchain"] == ["/tc/inc/c++", "/tc/inc"]
+
+
+def test_get_toolchain_includes_raises_on_probe_failure() -> None:
+    """A failed compiler probe is a hard error, not a silent empty list."""
+    fake_proc = MagicMock(returncode=1, stderr="xtensa-esp32-elf-g++: not found")
+    with (
+        patch.object(idedata.subprocess, "run", return_value=fake_proc),
+        pytest.raises(RuntimeError, match="builtin include dirs"),
+    ):
+        idedata.get_toolchain_includes("/bad/compiler")
+
+
+def test_get_toolchain_includes_raises_when_no_dirs_found() -> None:
+    """Markers present but no dirs (anomalous output) also raises."""
+    fake_proc = MagicMock(
+        returncode=0,
+        stderr="#include <...> search starts here:\nEnd of search list.\n",
+    )
+    with (
+        patch.object(idedata.subprocess, "run", return_value=fake_proc),
+        pytest.raises(RuntimeError, match="builtin include dirs"),
+    ):
+        idedata.get_toolchain_includes("/some/compiler")
+
+
+# ESP-IDF's compile_commands.json on Windows mixes literal backslash path
+# separators in the compiler path with shell ``\"`` quote-escaping in defines,
+# which only the real Windows argv parser handles. These exercise that path.
+@pytest.mark.skipif(os.name != "nt", reason="Windows argv tokenization")
+def test_split_command_preserves_paths_and_unescapes_quotes() -> None:
+    r"""Backslash paths survive while ``\"`` define-quoting is unescaped."""
+    command = r"C:\esp\bin\riscv32-esp-elf-g++.exe -DVER=\"1.2.3\" -IC:/inc/a -c x.cpp"
+
+    tokens = idedata._split_command(command)
+
+    assert tokens[0] == r"C:\esp\bin\riscv32-esp-elf-g++.exe"
+    assert '-DVER="1.2.3"' in tokens
+    assert "-IC:/inc/a" in tokens
+
+
+@pytest.mark.skipif(os.name != "nt", reason="Windows argv tokenization")
+def test_split_command_empty_returns_empty() -> None:
+    """An empty or blank command tokenizes to ``[]`` (e.g. an empty response file).
+
+    Guards against ``CommandLineToArgvW("")`` returning the current process name
+    instead of an empty list.
+    """
+    assert idedata._split_command("") == []
+    assert idedata._split_command("   ") == []
+
+
+@pytest.mark.skipif(os.name != "nt", reason="Windows argv tokenization")
+def test_parse_entry_normalizes_windows_cxx_path() -> None:
+    """A backslash compiler path is emitted forward-slashed; define unescaped."""
+    entry = _entry(
+        r"C:\b",
+        r"C:\b\src\esphome\x.cpp",
+        r"C:\esp\bin\g++.exe -DVER=\"1.2.3\" -IC:/inc/a -c x.cpp",
+    )
+
+    cxx_path, defines, includes, _ = idedata.parse_entry(entry)
+
+    assert cxx_path == "C:/esp/bin/g++.exe"
+    assert "\\" not in cxx_path
+    assert 'VER="1.2.3"' in defines
+    assert "C:/inc/a" in includes
+
+
+def test_parse_entry_strips_launcher_prefix() -> None:
+    """A launcher-wrapped compile names the compiler second; the exact
+    configured launcher is stripped, not anything ccache-shaped."""
+    entry = _entry(
+        f"{ABS}build",
+        f"{ABS}build/src/esphome/core/application.cpp",
+        "/opt/homebrew/bin/ccache /tools/xtensa-lx106-elf-g++ -DUSE_ESP8266 "
+        "-c app.cpp -o app.cpp.o",
+    )
+    cxx_path, defines, _, _ = idedata.parse_entry(
+        entry, launcher="/opt/homebrew/bin/ccache"
+    )
+    assert cxx_path == "/tools/xtensa-lx106-elf-g++"
+    assert defines == ["USE_ESP8266"]
+
+
+def test_parse_entry_recovers_from_unconfigured_launcher(
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    """A stale compile DB built with a launcher this run no longer configures
+    still yields the real compiler (the next token), not the launcher."""
+    entry = _entry(
+        f"{ABS}build",
+        f"{ABS}build/src/esphome/core/application.cpp",
+        "/opt/homebrew/bin/ccache /tools/xtensa-lx106-elf-g++ -c a.cpp -o a.o",
+    )
+    caplog.set_level(logging.DEBUG)
+    cxx_path, _, _, _ = idedata.parse_entry(entry)
+    assert cxx_path == "/tools/xtensa-lx106-elf-g++"
+    assert "Stripping unconfigured launcher" in caplog.text
+
+
+def test_parse_entry_keeps_launcher_without_program() -> None:
+    """A launcher followed only by flags (no program to recover) stays as
+    token zero; the cache layer refuses to persist it."""
+    entry = _entry(
+        f"{ABS}build",
+        f"{ABS}build/src/esphome/core/application.cpp",
+        "/opt/homebrew/bin/ccache -c a.cpp -o a.o",
+    )
+    cxx_path, _, _, _ = idedata.parse_entry(entry)
+    assert cxx_path == "/opt/homebrew/bin/ccache"
+
+
+def _write_compile_commands(tmp_path: Path) -> Path:
+    compile_commands = tmp_path / "compile_commands.json"
+    compile_commands.write_text(
+        json.dumps(
+            [
+                _entry(
+                    f"{ABS}build",
+                    f"{ABS}build/src/esphome/core/application.cpp",
+                    "/tools/g++ -DUSE_ESP8266 -c app.cpp -o app.cpp.o",
+                )
+            ]
+        )
+    )
+    return compile_commands
+
+
+def test_load_or_build_idedata_missing_compile_db(tmp_path: Path) -> None:
+    assert (
+        idedata.load_or_build_idedata(
+            tmp_path / "compile_commands.json", tmp_path / "f.elf", tmp_path / "c.json"
+        )
+        is None
+    )
+
+
+def test_load_or_build_idedata_builds_and_caches(tmp_path: Path) -> None:
+    compile_commands = _write_compile_commands(tmp_path)
+    cache = tmp_path / "cache" / "test.json"
+    with patch.object(
+        idedata, "get_toolchain_includes", return_value=["/toolchain/include"]
+    ):
+        data = idedata.load_or_build_idedata(
+            compile_commands, tmp_path / "firmware.elf", cache
+        )
+    assert data["cc_path"] == "/tools/gcc"
+    assert data["prog_path"] == str(tmp_path / "firmware.elf")
+    assert json.loads(cache.read_text()) == data
+
+    # A fresh cache is served without re-parsing the compile DB
+    os.utime(cache, (compile_commands.stat().st_mtime + 10,) * 2)
+    with patch.object(idedata, "idedata_from_build") as mock_build:
+        assert (
+            idedata.load_or_build_idedata(
+                compile_commands, tmp_path / "firmware.elf", cache
+            )
+            == data
+        )
+    mock_build.assert_not_called()
+
+
+def test_load_or_build_idedata_rebuilds_bad_cache(tmp_path: Path) -> None:
+    compile_commands = _write_compile_commands(tmp_path)
+    cache = tmp_path / "cache.json"
+    for bad in ("not json", json.dumps({"no_cc_path": True})):
+        cache.write_text(bad)
+        os.utime(cache, (compile_commands.stat().st_mtime + 10,) * 2)
+        with patch.object(idedata, "get_toolchain_includes", return_value=[]):
+            data = idedata.load_or_build_idedata(
+                compile_commands, tmp_path / "f.elf", cache
+            )
+        assert "cc_path" in data
+
+
+def test_load_or_build_idedata_rebuilds_when_compile_db_newer(tmp_path: Path) -> None:
+    """A compile DB newer than the cache forces regeneration."""
+    compile_commands = _write_compile_commands(tmp_path)
+    cache = tmp_path / "cache.json"
+    cache.write_text(json.dumps({"cc_path": "stale"}))
+    os.utime(compile_commands, (cache.stat().st_mtime + 10,) * 2)
+    with patch.object(idedata, "get_toolchain_includes", return_value=[]):
+        data = idedata.load_or_build_idedata(
+            compile_commands, tmp_path / "f.elf", cache
+        )
+    assert data["cc_path"] != "stale"
+
+
+def test_load_or_build_idedata_rebuilds_non_dict_cache(tmp_path: Path) -> None:
+    """Valid JSON that is not an object is regenerated, never handed out.
+
+    A bare string would otherwise pass the cc_path check by substring.
+    """
+    compile_commands = _write_compile_commands(tmp_path)
+    cache = tmp_path / "cache.json"
+    for bad in ('"cc_path is a string"', "[]", "42"):
+        cache.write_text(bad)
+        os.utime(cache, (compile_commands.stat().st_mtime + 10,) * 2)
+        with patch.object(idedata, "get_toolchain_includes", return_value=[]):
+            data = idedata.load_or_build_idedata(
+                compile_commands, tmp_path / "f.elf", cache
+            )
+        assert isinstance(data, dict)
+        assert "cc_path" in data
+
+
+def test_is_launcher_matches_only_known_launchers() -> None:
+    """Compilers of any shape pass; only the closed launcher set matches."""
+    for token in ("/t/g++-13", "gcc-8.4.0", "clang++-17", "armcc", "icx", "cc"):
+        assert not idedata._is_launcher(token)
+    for token in ("/opt/homebrew/bin/ccache", "CCACHE.EXE", "distcc", "sccache"):
+        assert idedata._is_launcher(token)
+
+
+def test_load_or_build_idedata_corrupted_cache_is_logged(
+    tmp_path: Path, caplog: pytest.LogCaptureFixture
+) -> None:
+    """A truncated cache is diagnosable, not a silent slow-build cause."""
+    compile_commands = _write_compile_commands(tmp_path)
+    cache = tmp_path / "c.json"
+    cache.write_text('{"cc_path": trunc')
+    os.utime(cache, (compile_commands.stat().st_mtime + 5,) * 2)
+    with patch.object(idedata, "get_toolchain_includes", return_value=[]):
+        data = idedata.load_or_build_idedata(
+            compile_commands, tmp_path / "f.elf", cache
+        )
+    assert data["cxx_path"] == "/tools/g++"
+    assert "Discarding unreadable idedata cache" in caplog.text
+
+
+def test_load_or_build_idedata_discards_unreadable_cache_file(
+    tmp_path: Path, caplog: pytest.LogCaptureFixture
+) -> None:
+    """An OSError on the cache read (permissions, I/O) regenerates like a
+    parse failure instead of aborting the consumer."""
+    compile_commands = _write_compile_commands(tmp_path)
+    cache = tmp_path / "c.json"
+    cache.write_text("{}")
+    os.utime(cache, (compile_commands.stat().st_mtime + 5,) * 2)
+    real_read_text = Path.read_text
+
+    def fail_cache_read(self: Path, *args: object, **kwargs: object) -> str:
+        # chmod(0) cannot revoke read access on Windows, so fault the read
+        # itself for a platform-independent OSError
+        if self == cache:
+            raise OSError("permission denied")
+        return real_read_text(self, *args, **kwargs)
+
+    with (
+        patch.object(idedata, "get_toolchain_includes", return_value=[]),
+        patch.object(Path, "read_text", fail_cache_read),
+    ):
+        data = idedata.load_or_build_idedata(
+            compile_commands, tmp_path / "f.elf", cache
+        )
+    assert data["cxx_path"] == "/tools/g++"
+    assert "Discarding unreadable idedata cache" in caplog.text
+
+
+def test_load_or_build_idedata_never_caches_a_launcher(tmp_path: Path) -> None:
+    """A compile DB naming a launcher as the compiler is rejected, never cached."""
+    compile_commands = tmp_path / "compile_commands.json"
+    compile_commands.write_text(
+        json.dumps(
+            [
+                _entry(
+                    f"{ABS}build",
+                    f"{ABS}build/src/esphome/core/application.cpp",
+                    "/opt/homebrew/bin/ccache -c app.cpp -o app.cpp.o",
+                )
+            ]
+        )
+    )
+    cache = tmp_path / "c.json"
+    # No probe patch needed: the launcher is rejected before the probe runs
+    with pytest.raises(EsphomeError, match="compile database is unusable"):
+        idedata.load_or_build_idedata(compile_commands, tmp_path / "f.elf", cache)
+    assert not cache.exists()
+
+
+def test_load_or_build_idedata_cache_hit_skips_rebuild(tmp_path: Path) -> None:
+    """A valid cache newer than the compile DB is served without re-parsing."""
+    compile_commands = _write_compile_commands(tmp_path)
+    cache = tmp_path / "c.json"
+    cache.write_text(json.dumps({"cc_path": "/tools/gcc", "cached": True}))
+    os.utime(cache, (compile_commands.stat().st_mtime + 5,) * 2)
+    with patch.object(idedata, "idedata_from_build") as mock_build:
+        data = idedata.load_or_build_idedata(
+            compile_commands, tmp_path / "f.elf", cache
+        )
+    mock_build.assert_not_called()
+    assert data["cached"] is True

tests/unit_tests/build_helpers/test_ninja.py

@@ -0,0 +1,143 @@
+"""Tests for esphome.build_helpers.ninja."""
+
+from __future__ import annotations
+
+import os
+from pathlib import Path
+import subprocess
+import sys
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from esphome.build_helpers import ninja as ninja_helper
+from esphome.core import EsphomeError
+
+
+def test_find_ninja_prefers_path(tmp_path: Path) -> None:
+    with (
+        patch("shutil.which", return_value=str(tmp_path / "ninja")),
+        patch.object(ninja_helper, "_ninja_runs", return_value=True),
+    ):
+        assert ninja_helper.find_ninja() == tmp_path / "ninja"
+
+
+def test_find_ninja_falls_back_to_wheel(tmp_path: Path) -> None:
+    """Without a PATH entry, the ninja PyPI wheel's binary is used."""
+    binary_name = "ninja.exe" if os.name == "nt" else "ninja"
+    (tmp_path / binary_name).touch()
+    wheel = MagicMock(BIN_DIR=str(tmp_path))
+    with (
+        patch("shutil.which", return_value=None),
+        patch.dict(sys.modules, {"ninja": wheel}),
+    ):
+        assert ninja_helper.find_ninja() == tmp_path / binary_name
+
+
+def test_find_ninja_package_not_installed() -> None:
+    """A missing ninja package raises the actionable message, not ImportError."""
+    with (
+        patch("shutil.which", return_value=None),
+        patch.dict(sys.modules, {"ninja": None}),
+        pytest.raises(EsphomeError, match="ninja not found"),
+    ):
+        ninja_helper.find_ninja()
+
+
+def test_find_ninja_missing_everywhere(tmp_path: Path) -> None:
+    wheel = MagicMock(BIN_DIR=str(tmp_path))
+    with (
+        patch("shutil.which", return_value=None),
+        patch.dict(sys.modules, {"ninja": wheel}),
+        pytest.raises(EsphomeError, match="ninja not found"),
+    ):
+        ninja_helper.find_ninja()
+
+
+def test_escape_ninja_specials() -> None:
+    assert ninja_helper.escape("a b:c$d") == "a$ b$:c$$d"
+
+
+def _q(tok: str) -> str:
+    """The platform's shell_token quote wrapper (argv rule on Windows)."""
+    return f'"{tok}"' if os.name == "nt" else f"'{tok}'"
+
+
+def test_quote_arg_windows_argv_rule() -> None:
+    # Backslash runs double only before a quote (subprocess.list2cmdline rule)
+    assert ninja_helper.quote_arg('-DX=a\\"b c') == '"-DX=a\\\\\\"b c"'
+    assert ninja_helper.quote_arg("a b\\") == '"a b\\\\"'
+
+
+def test_shell_token_quotes_only_when_needed() -> None:
+    assert ninja_helper.shell_token("-Os") == "-Os"
+    assert ninja_helper.shell_token("-DP=C:\\x y") == _q("-DP=C:\\x y")
+    assert ninja_helper.shell_token("plain", force=True) == _q("plain")
+
+
+def test_shell_token_quotes_shell_metacharacters() -> None:
+    """Tokens like -DMASK=(1<<3) must not reach /bin/sh -c bare."""
+    assert ninja_helper.shell_token("-DMASK=(1<<3)") == _q("-DMASK=(1<<3)")
+    assert ninja_helper.shell_token("-DX=a;b") == _q("-DX=a;b")
+    assert ninja_helper.shell_token("-DX=$HOME") == _q("-DX=$$HOME")
+
+
+def test_shell_token_posix_roundtrips_through_sh() -> None:
+    """Backslash runs, $, backticks, and quotes must reach the compiler
+    exactly as lexed once ninja un-doubles $$ and /bin/sh strips quotes."""
+
+    if sys.platform == "win32":
+        pytest.skip("POSIX sh quoting")
+    for tok in ("-DP=a\\\\b", "-DX=$VAR", "-DY=`date`", "-DZ=it's", '-DC="q"'):
+        quoted = ninja_helper.shell_token(tok).replace("$$", "$")
+        out = subprocess.run(
+            ["/bin/sh", "-c", f'printf "%s" {quoted}'],
+            capture_output=True,
+            text=True,
+            check=True,
+        )
+        assert out.stdout == tok
+
+
+def test_quote_path_force_quotes() -> None:
+    assert ninja_helper.quote_path(Path("a b")) == _q("a b")
+    assert ninja_helper.quote_path("simple") == _q("simple")
+
+
+def test_shell_token_empty_token_is_quoted() -> None:
+    """An empty argv element must survive as an explicit pair of quotes."""
+    assert ninja_helper.shell_token("") == _q("")
+
+
+def test_find_ninja_probes_path_hit(tmp_path: Path) -> None:
+    """A broken PATH shim falls back to the wheel instead of failing every
+    build later."""
+    binary_name = "ninja.exe" if os.name == "nt" else "ninja"
+    (tmp_path / binary_name).touch()
+    wheel = MagicMock(BIN_DIR=str(tmp_path))
+    with (
+        patch("shutil.which", return_value="/broken/ninja"),
+        patch.object(ninja_helper, "_ninja_runs", return_value=False),
+        patch.dict(sys.modules, {"ninja": wheel}),
+    ):
+        assert ninja_helper.find_ninja() == tmp_path / binary_name
+
+
+def test_ninja_probe_failure_warns(caplog: pytest.LogCaptureFixture) -> None:
+    with patch("esphome.framework_helpers.subprocess.run", side_effect=OSError("boom")):
+        assert ninja_helper._ninja_runs("/broken/ninja") is False
+    assert "failed to run" in caplog.text
+
+
+def test_ninja_probe_success() -> None:
+    with patch("esphome.framework_helpers.subprocess.run") as mock_run:
+        assert ninja_helper._ninja_runs("/usr/bin/ninja") is True
+    assert mock_run.call_args.kwargs["close_fds"] is False
+
+
+def test_shell_token_windows_branch_uses_argv_rule() -> None:
+    """The nt branch quotes with the CreateProcess argv rule (the ubuntu
+    coverage run never takes it naturally)."""
+    with patch.object(os, "name", "nt"):
+        assert ninja_helper.shell_token("a b") == '"a b"'
+        assert ninja_helper.shell_token("", force=True) == '""'

tests/unit_tests/build_helpers/test_size_summary.py

@@ -0,0 +1,22 @@
+"""Tests for the shared PlatformIO-format size bar."""
+
+from __future__ import annotations
+
+import pytest
+
+from esphome.build_helpers.size_summary import format_bar, print_size_line
+
+
+def test_format_bar_zero_total() -> None:
+    """A zero total must not divide by zero."""
+    assert format_bar(0, 0) == "[          ]   0.0% (used 0 bytes from 0 bytes)"
+
+
+def test_print_size_line_label_padding(capsys: pytest.CaptureFixture[str]) -> None:
+    """The label column is exactly what ci_memory_impact_extract.py greps."""
+    print_size_line("RAM", 47932, 180736)
+    print_size_line("Flash", 888511, 1835008)
+    out = capsys.readouterr().out.splitlines()
+    assert out[0].startswith("RAM:   [")
+    assert out[1].startswith("Flash: [")
+    assert "26.5% (used 47932 bytes from 180736 bytes)" in out[0]

tests/unit_tests/components/esp8266/test_build_surgery.py

@@ -0,0 +1,145 @@
+"""Tests for the linker-script surgery shared with the native toolchain."""
+
+from __future__ import annotations
+
+import importlib.util
+from pathlib import Path
+import sys
+
+import pytest
+
+from esphome.components.esp8266 import build_surgery
+from esphome.components.esp8266.boards import BOARDS, ESP8266_BOARD_BUILD
+from esphome.components.esp8266.build_surgery import (
+    RATETABLE_RULE,
+    apply_testing_memory_patches,
+    relocate_ratetable,
+    segment_length,
+)
+
+_COMMON_LD_SNIPPET = """\
+  .dport0.data : ALIGN(4)
+  {
+    _dport0_data_start = ABSOLUTE(.);
+  } >dport0_0_seg :dport0_0_phdr
+  .data : ALIGN(4)
+  {
+    _data_start = ABSOLUTE(.);
+    *(.data)
+  } >dram0_0_seg :dram0_0_phdr
+"""
+
+# Shaped like the real SDK flash ld scripts: no iram1_0_seg (that lives in
+# the generated common ld only)
+_FLASH_LD_SNIPPET = """\
+MEMORY
+{
+  dport0_0_seg :                        org = 0x3FF00000, len = 0x10
+  dram0_0_seg :                         org = 0x3FFE8000, len = 0x14000
+  irom0_0_seg :                         org = 0x40201010, len = 0xfeff0
+}
+"""
+
+# Shaped like the preprocessed common ld: MMU_IRAM_SIZE expands with a ul
+# suffix the patcher must leave in place
+_COMMON_LD_MEMORY_SNIPPET = """\
+MEMORY
+{
+  iram1_0_seg :                         org = 0x40100000, len = 0x8000ul
+}
+"""
+
+
+def test_relocate_ratetable_inserts_after_data_start() -> None:
+    patched = relocate_ratetable(_COMMON_LD_SNIPPET)
+    assert RATETABLE_RULE in patched
+    # Inserted after the .data section's anchor, not the .dport0.data one
+    # (whose closing brace bounds the decoy block)
+    assert RATETABLE_RULE not in patched[: patched.index("} >dport0_0_seg")]
+    assert patched.index(RATETABLE_RULE) < patched.index("*(.data)")
+    # Idempotent on an already-patched script
+    assert relocate_ratetable(patched) == patched
+
+
+def test_relocate_ratetable_requires_anchor() -> None:
+    with pytest.raises(RuntimeError, match="_data_start"):
+        relocate_ratetable("SECTIONS { }")
+
+
+def test_testing_memory_patches_enlarge_segments() -> None:
+    patched = apply_testing_memory_patches(
+        _FLASH_LD_SNIPPET, ("dram0_0_seg", "irom0_0_seg")
+    )
+    assert segment_length(patched, "dram0_0_seg") == 0x200000
+    assert segment_length(patched, "irom0_0_seg") == 0x2000000
+    # Untouched segments keep their sizes
+    assert segment_length(patched, "dport0_0_seg") == 0x10
+
+
+def test_testing_memory_patches_keep_ul_suffix() -> None:
+    """The common ld's preprocessed sizes carry a ul suffix; the patch must
+    replace only the hex digits, as testing_mode.py.script does."""
+    patched = apply_testing_memory_patches(_COMMON_LD_MEMORY_SNIPPET, ("iram1_0_seg",))
+    assert "len = 0x200000ul" in patched
+    assert segment_length(patched, "iram1_0_seg") == 0x200000
+
+
+def test_segment_length_requires_whole_name() -> None:
+    """A name must match its own line, never inside a longer segment name."""
+    assert segment_length(_FLASH_LD_SNIPPET, "ram0_0_seg") is None
+
+
+def test_testing_memory_patches_unknown_segment_raises() -> None:
+    with pytest.raises(RuntimeError, match="Unknown testing-mode segment"):
+        apply_testing_memory_patches("MEMORY { }", ("bogus_seg",))
+
+
+def test_segment_length() -> None:
+    assert segment_length(_FLASH_LD_SNIPPET, "irom0_0_seg") == 0xFEFF0
+    assert segment_length(_FLASH_LD_SNIPPET, "missing_seg") is None
+
+
+def test_testing_memory_patches_missing_segment_raises() -> None:
+    """A named segment the patch could not find raises instead of silently
+    keeping the real memory limits."""
+    with pytest.raises(RuntimeError, match="dram0_0_seg"):
+        apply_testing_memory_patches("MEMORY { }", ("dram0_0_seg",))
+
+
+def test_board_build_covers_every_board() -> None:
+    """Every supported board has native build metadata (the table may carry
+    extras that BOARDS does not expose)."""
+    assert set(BOARDS) <= set(ESP8266_BOARD_BUILD)
+
+
+def test_surgery_fingerprint_is_stable_and_sensitive(tmp_path) -> None:
+    """The properties the linker-script cache depends on: the fingerprint is
+    stable across calls and changes when the module's source changes."""
+
+    first = build_surgery.surgery_fingerprint()
+    assert first == build_surgery.surgery_fingerprint()
+    assert len(first) == 64
+    int(first, 16)  # sha256 hex digest
+
+    # A modified copy of the module must fingerprint differently
+    copy = tmp_path / "build_surgery_variant.py"
+    copy.write_text(
+        Path(build_surgery.__file__).read_text(encoding="utf-8")
+        + "\nEXTRA_BEHAVIORAL_INPUT = 1\n",
+        encoding="utf-8",
+    )
+    spec = importlib.util.spec_from_file_location("build_surgery_variant", copy)
+    variant = importlib.util.module_from_spec(spec)
+    sys.modules[spec.name] = variant
+    try:
+        spec.loader.exec_module(variant)
+        assert variant.surgery_fingerprint() != first
+    finally:
+        del sys.modules[spec.name]
+
+
+def test_testing_memory_patches_present_but_unselected_raises() -> None:
+    """A known segment left off the caller's list must fail, not silently
+    keep its real memory limit."""
+    with pytest.raises(RuntimeError, match="not selected"):
+        apply_testing_memory_patches(_FLASH_LD_SNIPPET, ("dram0_0_seg",))

tests/unit_tests/core/test_config.py

@@ -1285,6 +1285,7 @@ async def test_add_platformio_options_native_idf(
     await config._add_platformio_options(
         {
             "build_flags": "-DSINGLE_FLAG",  # string and list forms both valid
+            "build_unflags": ["-Os"],
             "lib_deps": ["bblanchon/ArduinoJson@7.4.2"],
             "lib_ignore": "libsodium",
             "upload_speed": "115200",
@@ -1294,6 +1295,7 @@ async def test_add_platformio_options_native_idf(
 
     assert "-DSINGLE_FLAG" in CORE.build_flags
     assert "ArduinoJson" in CORE.platformio_libraries
+    assert "-Os" in CORE.build_unflags
     # lib_ignore is stored (listified) for generate_idf_components to read;
     # nothing else lands in platformio_options on the native toolchain.
     assert CORE.platformio_options == {"lib_ignore": ["libsodium"]}
@@ -1389,3 +1391,50 @@ def test_esphome_build_internals_are_yaml_only() -> None:
         assert markers[field].visibility is cv.Visibility.ADVANCED, field
     # A regular device-config field stays on the main form.
     assert markers[CONF_NAME_ADD_MAC_SUFFIX].visibility is None
+
+
+@pytest.mark.asyncio
+async def test_add_platformio_options_native_arduino(
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    """The native ESP8266 Arduino toolchain honors board_build.f_cpu (a
+    real-world overclock knob) and warns about the rest like native IDF."""
+    CORE.toolchain = Toolchain.ARDUINO
+    CORE.data[KEY_CORE] = {
+        KEY_TARGET_PLATFORM: "esp8266",
+        KEY_TARGET_FRAMEWORK: "arduino",
+    }
+
+    await config._add_platformio_options(
+        {
+            "board_build.f_cpu": "160000000L",
+            # The schema also permits the list form; the last value wins
+            # and reaches the generator as a scalar
+            "board_build.ldscript": ["eagle.flash.2m.ld", "eagle.flash.4m2m.ld"],
+            "board_build.filesystem": "littlefs",
+            "upload_speed": "115200",
+        }
+    )
+
+    assert CORE.platformio_options["board_build.f_cpu"] == "160000000L"
+    assert CORE.platformio_options["board_build.ldscript"] == "eagle.flash.4m2m.ld"
+    assert "board_build.f_cpu is ignored" not in caplog.text
+    assert "board_build.ldscript is ignored" not in caplog.text
+    assert (
+        "esphome->platformio_options->board_build.filesystem is ignored" in caplog.text
+    )
+    # An empty list for an honored key is not a scalar; it falls through
+    # to the ignored-option warning instead of an IndexError
+    await config._add_platformio_options({"board_build.ldscript": []})
+    assert "board_build.ldscript is ignored" in caplog.text
+    assert "'arduino' toolchain" in caplog.text
+    assert "upload_speed" not in caplog.text
+
+
+def test_esp8266_rejects_unsupported_cli_toolchain() -> None:
+    """Until the native backend lands, ESP8266 serves only PlatformIO."""
+    from esphome.components.esp8266 import CONFIG_SCHEMA
+
+    CORE.toolchain = Toolchain.ARDUINO
+    with pytest.raises(cv.Invalid, match="Unsupported toolchain 'arduino'"):
+        CONFIG_SCHEMA({"board": "nodemcuv2"})

tests/unit_tests/test_compiled_config.py

@@ -68,6 +68,7 @@ def _write_storage(
     esp_platform: str | None = "ESP32",
     core_platform: str | None = "esp32",
     build_path: str | None = "/build/lite_test",
+    toolchain: str | None = None,
 ) -> None:
     """Write a vanilla StorageJSON sidecar for the cache tests."""
     storage_path.parent.mkdir(parents=True, exist_ok=True)
@@ -88,6 +89,7 @@ def _write_storage(
         "no_mdns": False,
         "framework": "arduino",
         "core_platform": core_platform,
+        "toolchain": toolchain,
     }
     storage_path.write_text(json.dumps(data), encoding="utf-8")
 
@@ -629,6 +631,35 @@ def test_save_compiled_config_and_sidecar_builds_real_sidecar(tmp_path: Path) ->
     assert load_compiled_config(yaml_path) is not None
 
 
+@pytest.mark.parametrize(
+    ("sidecar_toolchain", "saved"),
+    [
+        ("esp-idf", False),
+        ("platformio", True),
+        (None, True),  # legacy sidecar without the field: guard is inert
+    ],
+)
+def test_save_compiled_config_and_sidecar_toolchain_mismatch(
+    tmp_path: Path, sidecar_toolchain: str | None, saved: bool
+) -> None:
+    """A config validated under a different toolchain than the compile's
+    must not overwrite the cache."""
+    yaml_path = _bare_yaml(tmp_path)
+    _prime_core(tmp_path)
+    CORE.config = {CONF_ESPHOME: {CONF_NAME: "lite_test"}}
+    CORE.toolchain = Toolchain.PLATFORMIO
+    _write_storage(
+        tmp_path / ".esphome" / "storage" / "lite_test.yaml.json",
+        toolchain=sidecar_toolchain,
+    )
+
+    save_compiled_config_and_sidecar(CORE.config)
+
+    cache = tmp_path / ".esphome" / "storage" / "lite_test.yaml.validated.json"
+    assert cache.exists() is saved
+    assert (load_compiled_config(yaml_path) is not None) is saved
+
+
 @pytest.mark.parametrize("command", ["upload", "logs"])
 def test_run_esphome_upload_and_logs_refresh_cache_on_fallback(
     tmp_path: Path, command: str

tests/unit_tests/test_config_validation.py

@@ -1,3 +1,4 @@
+import importlib
 import json
 import logging
 from pathlib import Path
@@ -48,6 +49,7 @@ from esphome.const import (
     TYPE_GIT,
     TYPE_LOCAL,
     Framework,
+    Toolchain,
 )
 from esphome.core import (
     CORE,
@@ -3165,3 +3167,46 @@ def test_file__remapped_path_is_directory_raises(setup_core: Path) -> None:
 
     with pytest.raises(Invalid, match="is not a file"):
         cv.file_("/original/config/headers")
+
+
+def test_require_platformio_toolchain() -> None:
+    """Platforms with only the PlatformIO backend reject other toolchains."""
+    validator = cv.require_platformio_toolchain("RP2")
+    CORE.toolchain = None
+    config: dict = {}
+    assert validator(config) is config
+    assert CORE.toolchain == Toolchain.PLATFORMIO
+
+    CORE.toolchain = Toolchain.ARDUINO
+    with pytest.raises(Invalid, match="Unsupported toolchain 'arduino' for RP2"):
+        validator(config)
+
+
+def test_check_supported_toolchain_unresolved_is_an_ordering_bug() -> None:
+    """Calling the check before resolution fails naming the ordering bug,
+    not a user-facing unsupported-toolchain error."""
+    CORE.toolchain = None
+    with pytest.raises(Invalid, match="not resolved before RP2 validation"):
+        cv._check_supported_toolchain("RP2", (Toolchain.PLATFORMIO,))
+
+
+@pytest.mark.parametrize(
+    ("platform", "minimal_config"),
+    [
+        ("host", {}),
+        ("rp2", {"board": "rpipicow"}),
+        ("bk72xx", {"board": "generic-bk7231n-qfn32-tuya"}),
+        ("rtl87xx", {"board": "generic-rtl8710bn-2mb-788k"}),
+        ("ln882x", {"board": "generic-ln882h"}),
+        # The legacy stub platform must reject too, not just the chip families
+        ("libretiny", {}),
+    ],
+)
+def test_every_platformio_only_platform_rejects_arduino_toolchain(
+    platform: str, minimal_config: dict
+) -> None:
+    """A platform that cannot serve a CLI toolchain rejects it at validation."""
+    module = importlib.import_module(f"esphome.components.{platform}")
+    CORE.toolchain = Toolchain.ARDUINO
+    with pytest.raises(Invalid, match="Unsupported toolchain 'arduino'"):
+        module.CONFIG_SCHEMA(dict(minimal_config))

tests/unit_tests/test_core.py

@@ -6,9 +6,9 @@ from unittest.mock import patch
 
 from hypothesis import given
 import pytest
-from strategies import mac_addr_strings
 
 from esphome import const, core
+from tests.unit_tests.strategies import mac_addr_strings
 
 
 class TestHexInt:
@@ -958,6 +958,24 @@ class TestEsphomeCore:
         target.toolchain = const.Toolchain.ESP_IDF
         assert target.using_toolchain_sdk_nrf is False
 
+    def test_using_toolchain_arduino(self, target):
+        """A toolchain choice, distinct from the arduino target framework."""
+        target.toolchain = const.Toolchain.ARDUINO
+        assert target.using_toolchain_arduino is True
+        target.toolchain = const.Toolchain.PLATFORMIO
+        assert target.using_toolchain_arduino is False
+
+    def test_using_native_toolchain(self, target):
+        """True exactly for the toolchains that never read platformio.ini."""
+        target.toolchain = const.Toolchain.ESP_IDF
+        assert target.using_native_toolchain is True
+        target.toolchain = const.Toolchain.ARDUINO
+        assert target.using_native_toolchain is True
+        target.toolchain = const.Toolchain.PLATFORMIO
+        assert target.using_native_toolchain is False
+        target.toolchain = const.Toolchain.SDK_NRF
+        assert target.using_native_toolchain is False
+
     def test_add_library__extracts_short_name_from_path(self, target):
         """Test add_library extracts short name from library paths like owner/lib."""
         target.data[const.KEY_CORE] = {

tests/unit_tests/test_espidf_clang_tidy.py

@@ -1,10 +1,13 @@
 """Tests for esphome.espidf.clang_tidy tidy-project generation."""
 
+import json
 import os
 from pathlib import Path
+from unittest.mock import patch
 
 import pytest
 
+from esphome.espidf import clang_tidy
 from esphome.espidf.clang_tidy import _Settings, _setup_core, _write_tidy_project
 
 REPO_ROOT = Path(__file__).resolve().parents[2]
@@ -64,3 +67,35 @@ def test_setup_core_sets_arduino_env(
     _setup_core(tmp_path / "proj", _settings(target_framework=target_framework))
 
     assert os.environ["ESPHOME_ARDUINO_COMPONENT"] == expected
+
+
+def test_idedata_from_tidy_project(tmp_path) -> None:
+    """The tidy TU's compile entry is assembled into consumer-shaped idedata."""
+    compile_commands = tmp_path / "compile_commands.json"
+    compile_commands.write_text(
+        json.dumps(
+            [
+                {
+                    "directory": str(tmp_path),
+                    "file": str(tmp_path / "main" / "tidy.cpp"),
+                    "command": "/tc/xtensa-esp32-elf-g++ -DUSE_ESP32 "
+                    f"-I{tmp_path}/inc -c main/tidy.cpp -o tidy.o",
+                }
+            ]
+        )
+    )
+    with patch(
+        "esphome.espidf.clang_tidy.get_toolchain_includes", return_value=["/tc/inc"]
+    ):
+        data = clang_tidy._idedata_from_tidy_project(compile_commands)
+    assert data["cxx_path"] == "/tc/xtensa-esp32-elf-g++"
+    assert data["defines"] == ["USE_ESP32"]
+    assert data["includes"]["toolchain"] == ["/tc/inc"]
+    assert any(inc.endswith("/inc") for inc in data["includes"]["build"])
+
+
+def test_idedata_from_tidy_project_missing_tu_raises(tmp_path) -> None:
+    compile_commands = tmp_path / "compile_commands.json"
+    compile_commands.write_text(json.dumps([]))
+    with pytest.raises(RuntimeError, match="tidy.cpp not found"):
+        clang_tidy._idedata_from_tidy_project(compile_commands)

tests/unit_tests/test_espidf_component.py

@@ -1,12 +1,12 @@
 import glob
 import hashlib
 import json
-import os
 from pathlib import Path
 from unittest.mock import MagicMock
 
 import pytest
 
+from esphome.components import esp32 as esp32_module
 from esphome.const import (
     KEY_CORE,
     KEY_TARGET_FRAMEWORK,
@@ -16,6 +16,7 @@ from esphome.const import (
 )
 from esphome.core import CORE, Library
 from esphome.espidf.component import (
+    _emit_idf_component,
     generate_cmakelists_txt,
     generate_idf_component_yml,
     generate_idf_components,
@@ -26,11 +27,11 @@ from esphome.platformio.library import (
     GitSource,
     URLSource,
     _node_key,
-    _normalize_dependencies,
-    _parse_library_json,
-    _parse_library_properties,
     _resolve_registry_version,
     collect_filtered_files,
+    normalize_dependencies,
+    parse_library_json,
+    parse_library_properties,
     split_list_by_condition,
 )
 
@@ -369,133 +370,11 @@ def test_generate_idf_component_yml_missing_path_raises(tmp_component):
         generate_idf_component_yml(tmp_component)
 
 
-def test_extra_script_captures_libpath_libs_and_defines(tmp_path):
-    from esphome.espidf.extra_script import captured_as_build_flags, run_extra_script
-
-    (tmp_path / "src" / "esp32").mkdir(parents=True)
-    script = tmp_path / "extra_script.py"
-    script.write_text(
-        "Import('env')\n"
-        "mcu = env.get('BOARD_MCU')\n"
-        "env.Append(\n"
-        "    LIBPATH=[join('src', mcu)],\n"
-        "    LIBS=['algobsec'],\n"
-        "    CPPDEFINES=['FOO', ('BAR', '1')],\n"
-        "    LINKFLAGS=['-Wl,--gc-sections'],\n"
-        ")\n"
-    )
-    # The script uses bare ``join`` (PIO's extra-scripts run inside SCons
-    # where this is in scope). Inject it via the script header so the
-    # shim's exec namespace can resolve it.
-    script.write_text("from os.path import join\n" + script.read_text())
-
-    result = run_extra_script(script, library_dir=tmp_path, idf_target="esp32")
-
-    assert result.libpath == [str(Path("src") / "esp32")]
-    assert result.libs == ["algobsec"]
-    assert ("BAR", "1") in result.cppdefines
-    assert "FOO" in result.cppdefines
-    assert result.linkflags == ["-Wl,--gc-sections"]
-
-    flags = captured_as_build_flags(result, library_dir=tmp_path)
-    sep = os.sep
-    assert f"-Lsrc{sep}esp32" in flags
-    assert "-lalgobsec" in flags
-    assert "-DFOO" in flags
-    assert "-DBAR=1" in flags
-    assert "-Wl,--gc-sections" in flags
-
-
-def test_extra_script_libpath_relative_resolves_against_library_dir(
-    tmp_path, monkeypatch
-):
-    """Relative LIBPATH entries must resolve against ``library_dir``, not the
-    caller's CWD (the shim restores CWD before ``captured_as_build_flags``
-    runs)."""
-    from esphome.espidf.extra_script import ExtraScriptResult, captured_as_build_flags
-
-    (tmp_path / "lib" / "esp32").mkdir(parents=True)
-    elsewhere = tmp_path.parent / "not_the_library_dir"
-    elsewhere.mkdir(exist_ok=True)
-    monkeypatch.chdir(elsewhere)
-
-    result = ExtraScriptResult(libpath=["lib/esp32"])
-    flags = captured_as_build_flags(result, library_dir=tmp_path)
-
-    sep = os.sep
-    assert flags == [f"-Llib{sep}esp32"]
-
-
-def test_extra_script_libpath_absolute_outside_library_dir(tmp_path):
-    from esphome.espidf.extra_script import ExtraScriptResult, captured_as_build_flags
-
-    outside = tmp_path.parent / "system_lib"
-    outside.mkdir(exist_ok=True)
-    result = ExtraScriptResult(libpath=[str(outside)])
-
-    flags = captured_as_build_flags(result, library_dir=tmp_path)
-    assert flags == [f"-L{outside.resolve()}"]
-
-
-def test_extra_script_failure_returns_empty_result(tmp_path, caplog):
-    from esphome.espidf.extra_script import run_extra_script
-
-    script = tmp_path / "broken.py"
-    script.write_text("raise RuntimeError('boom')\n")
-
-    with caplog.at_level("WARNING"):
-        result = run_extra_script(script, library_dir=tmp_path, idf_target="esp32")
-
-    assert result.libpath == []
-    assert result.libs == []
-    assert "broken.py" in caplog.text
-
-
-def test_apply_extra_script_path_traversal_is_rejected(tmp_path):
-    from esphome.espidf.component import _apply_extra_script
-
-    library_dir = tmp_path / "lib"
-    library_dir.mkdir()
-    outside = tmp_path / "evil.py"
-    outside.write_text("env.Append(LIBS=['pwned'])\n")
-
-    c = IDFComponent("owner/name", "1.0", source=URLSource("http://dummy"))
-    c.path = library_dir
-    c.data = {"build": {"extraScript": "../evil.py"}}
-
-    _apply_extra_script(c)
-
-    # Nothing was folded into flags: the traversal was rejected before
-    # the script could run.
-    assert "flags" not in c.data["build"]
-
-
-def test_apply_extra_script_merges_into_existing_flags(tmp_path, monkeypatch):
-    from esphome.components import esp32 as esp32_module
-
-    monkeypatch.setattr(esp32_module, "get_esp32_variant", lambda: "ESP32")
-
-    from esphome.espidf.component import _apply_extra_script
-
-    (tmp_path / "src").mkdir()
-    script = tmp_path / "extra.py"
-    script.write_text("env.Append(LIBS=['algobsec'])\n")
-
-    c = IDFComponent("owner/name", "1.0", source=URLSource("http://dummy"))
-    c.path = tmp_path
-    c.data = {"build": {"extraScript": "extra.py", "flags": ["-DEXISTING"]}}
-
-    _apply_extra_script(c)
-
-    assert "-DEXISTING" in c.data["build"]["flags"]
-    assert "-lalgobsec" in c.data["build"]["flags"]
-
-
 def test_parse_library_json(tmp_path):
     f = tmp_path / "library.json"
     f.write_text(json.dumps({"name": "test"}))
 
-    result = _parse_library_json(f)
+    result = parse_library_json(f)
     assert result["name"] == "test"
 
 
@@ -510,7 +389,7 @@ empty=
 """
     )
 
-    result = _parse_library_properties(f)
+    result = parse_library_properties(f)
 
     assert result["name"] == "Test"
     assert result["version"] == "1.0"
@@ -680,22 +559,22 @@ def test_node_key_registry_bare_name():
 
 
 def test_normalize_dependencies_none():
-    assert _normalize_dependencies(None) == []
+    assert normalize_dependencies(None) == []
 
 
 def test_normalize_dependencies_list_form():
     deps = [{"name": "foo", "version": "1.0"}]
-    assert _normalize_dependencies(deps) == [{"name": "foo", "version": "1.0"}]
+    assert normalize_dependencies(deps) == [{"name": "foo", "version": "1.0"}]
 
 
 def test_normalize_dependencies_dict_form():
-    out = _normalize_dependencies({"nanopb/Nanopb": "^0.4.91", "BareName": "1.2.3"})
+    out = normalize_dependencies({"nanopb/Nanopb": "^0.4.91", "BareName": "1.2.3"})
     assert {"name": "Nanopb", "owner": "nanopb", "version": "^0.4.91"} in out
     assert {"name": "BareName", "owner": None, "version": "1.2.3"} in out
 
 
 def test_normalize_dependencies_dict_form_nested_spec():
-    out = _normalize_dependencies(
+    out = normalize_dependencies(
         {"nanopb/Nanopb": {"version": "^0.4.91", "platforms": "espidf"}}
     )
     assert out == [
@@ -1187,6 +1066,36 @@ def test_idf_component_download_passes_salt() -> None:
     c.download(force=True, salt="abcd1234", namespace="idf")
 
     source.download.assert_called_once_with(
-        "owner/name", force=True, salt="abcd1234", namespace="idf"
+        "owner/name", force=True, salt="abcd1234", namespace="idf", progress=None
     )
     assert c.path == Path("/converted/owner/name")
+
+
+def test_emit_idf_component_wires_esp32_target(tmp_path, monkeypatch):
+    """Emitting a component resolves the esp32 variant into the shared
+    extraScript helper."""
+
+    monkeypatch.setattr(esp32_module, "get_esp32_variant", lambda: "ESP32")
+    (tmp_path / "src").mkdir()
+    script = tmp_path / "extra.py"
+    script.write_text("env.Append(LIBS=[env.get('BOARD_MCU')])\n")
+    c = IDFComponent("owner/name", "1.0", source=URLSource("http://dummy"))
+    c.path = tmp_path
+    c.data = {"build": {"extraScript": "extra.py"}}
+    _emit_idf_component(c)
+    assert c.data["build"]["flags"] == ["-lesp32"]
+
+
+def test_build_flags_dangling_flag_does_not_cross_entries(
+    tmp_path, caplog: pytest.LogCaptureFixture
+) -> None:
+    """Each entry is lexed independently, as ParseFlags does: a dangling -I ending one
+    entry warns instead of absorbing the next entry's first token."""
+    (tmp_path / "src").mkdir()
+    c = IDFComponent("owner/name", "1.0", source=URLSource("http://dummy"))
+    c.path = tmp_path
+    c.data = {"build": {"flags": ["-Wall -I", "-DFOO=1"]}}
+    content = generate_cmakelists_txt(c)
+    assert "FOO=1" in content
+    assert "-I-DFOO" not in content
+    assert "Ignoring trailing '-I'" in caplog.text

tests/unit_tests/test_espidf_framework.py

@@ -2,6 +2,7 @@
 
 # pylint: disable=protected-access
 
+from concurrent.futures import ThreadPoolExecutor
 from contextlib import contextmanager
 import importlib.util
 import io
@@ -14,7 +15,7 @@ import subprocess
 import sys
 import tarfile
 from types import SimpleNamespace
-from unittest.mock import patch
+from unittest.mock import MagicMock, patch
 
 import pytest
 
@@ -887,6 +888,78 @@ _PREFETCH_JSON = json.dumps(
 )
 
 
+def test_prefetch_leaves_unverifiable_entries_to_the_installer(
+    tmp_path: Path, caplog: pytest.LogCaptureFixture
+) -> None:
+    """An entry missing sha256 or size must not download unverified; the
+    installer handles it and fails loudly on a bad archive."""
+    entries = json.loads(_PREFETCH_JSON)
+    del entries[0]["sha256"]
+    del entries[1]["size"]
+    entries.append(
+        {
+            "name": "gcc@14.2.0",
+            "url": "https://example.com/gcc.tar.gz",
+            "size": 67,
+            "sha256": "ef" * 32,
+            "dest": "gcc.tar.gz",
+        }
+    )
+    with (
+        patch(
+            "esphome.espidf.framework.run_command",
+            return_value=(True, json.dumps(entries), ""),
+        ),
+        patch("esphome.espidf.framework.download_with_resume") as download,
+        patch("esphome.espidf.framework.get_system_python_path", return_value="python"),
+        patch("esphome.espidf.framework.BatchDownloadProgress") as progress_cls,
+    ):
+        _prefetch_idf_tool_archives(tmp_path, "esp32", ["required"], None)
+    assert [call[0][0] for call in download.call_args_list] == [
+        "https://example.com/gcc.tar.gz"
+    ]
+    assert download.call_args[1]["sha256"] == "ef" * 32
+    progress_cls.assert_called_once_with("Downloading ESP-IDF tools", 67)
+    assert "cmake@3.30.2 has no sha256/size" in caplog.text
+    assert "ninja@1.12.1 has no sha256/size" in caplog.text
+
+
+def test_prefetch_all_entries_unverifiable_is_a_noop(tmp_path: Path) -> None:
+    entries = json.loads(_PREFETCH_JSON)
+    for entry in entries:
+        del entry["sha256"]
+    with (
+        patch(
+            "esphome.espidf.framework.run_command",
+            return_value=(True, json.dumps(entries), ""),
+        ),
+        patch("esphome.espidf.framework.download_with_resume") as download,
+        patch("esphome.espidf.framework.get_system_python_path", return_value="python"),
+    ):
+        _prefetch_idf_tool_archives(tmp_path, "esp32", ["required"], None)
+    download.assert_not_called()
+
+
+def test_prefetch_dedupes_entries_by_dest(tmp_path: Path) -> None:
+    """Two entries resolving to one dest would interleave writes into the
+    same .part file; only the first downloads."""
+    entries = json.loads(_PREFETCH_JSON)
+    dup = dict(entries[0]) | {"name": "cmake-alias@3.30.2"}
+    entries.append(dup)
+    with (
+        patch(
+            "esphome.espidf.framework.run_command",
+            return_value=(True, json.dumps(entries), ""),
+        ),
+        patch("esphome.espidf.framework.download_with_resume") as download,
+        patch("esphome.espidf.framework.get_system_python_path", return_value="python"),
+        patch("esphome.espidf.framework.BatchDownloadProgress"),
+    ):
+        _prefetch_idf_tool_archives(tmp_path, "esp32", ["required"], None)
+    dests = [call[0][1].name for call in download.call_args_list]
+    assert dests.count("cmake-3.30.2.tar.gz") == 1
+
+
 def test_prefetch_downloads_each_archive_with_resume(tmp_path: Path) -> None:
     with (
         patch(
@@ -895,16 +968,73 @@ def test_prefetch_downloads_each_archive_with_resume(tmp_path: Path) -> None:
         ),
         patch("esphome.espidf.framework.download_with_resume") as download,
         patch("esphome.espidf.framework.get_system_python_path", return_value="python"),
+        patch("esphome.espidf.framework.BatchDownloadProgress") as progress_cls,
     ):
+        # Materialize the lazy mock before threads race its first creation
+        tracker = progress_cls.return_value.tracker.return_value
         _prefetch_idf_tool_archives(tmp_path, "esp32", ["required"], None)
 
     dist = get_idf_tools_path() / "dist"
-    assert download.call_count == 2
-    assert download.call_args_list[0][0] == (
-        "https://example.com/cmake.tar.gz",
-        dist / "cmake-3.30.2.tar.gz",
-    )
-    assert download.call_args_list[0][1] == {"sha256": "ab" * 32, "size": 123}
+    # Archives download concurrently, so the call order is not fixed.
+    calls = {call[0]: call[1] for call in download.call_args_list}
+    assert set(calls) == {
+        ("https://example.com/cmake.tar.gz", dist / "cmake-3.30.2.tar.gz"),
+        ("https://example.com/ninja.zip", dist / "ninja.zip"),
+    }
+    kwargs = calls[("https://example.com/cmake.tar.gz", dist / "cmake-3.30.2.tar.gz")]
+    assert kwargs["sha256"] == "ab" * 32
+    assert kwargs["size"] == 123
+    # every archive reports into the one combined progress bar
+    progress_cls.assert_called_once_with("Downloading ESP-IDF tools", 123 + 45)
+    assert all(kw["progress"] is tracker for kw in calls.values())
+
+
+def test_prefetch_downloads_archives_concurrently(tmp_path: Path) -> None:
+    """More than one archive fans out over a bounded thread pool."""
+    entries = [
+        {
+            "name": f"tool{i}@1",
+            "url": f"https://example.com/tool{i}.tar.gz",
+            "size": 10,
+            "sha256": "ab" * 32,
+            "dest": f"tool{i}.tar.gz",
+        }
+        for i in range(6)
+    ]
+    with (
+        patch(
+            "esphome.espidf.framework.run_command",
+            return_value=(True, json.dumps(entries), ""),
+        ),
+        patch("esphome.espidf.framework.download_with_resume") as download,
+        patch("esphome.espidf.framework.get_system_python_path", return_value="python"),
+        patch(
+            "esphome.framework_helpers.ThreadPoolExecutor", wraps=ThreadPoolExecutor
+        ) as pool,
+    ):
+        _prefetch_idf_tool_archives(tmp_path, "esp32", ["required"], None)
+
+    pool.assert_called_once_with(max_workers=4)
+    assert download.call_count == 6
+
+
+def test_prefetch_single_archive_uses_one_worker(tmp_path: Path) -> None:
+    entries = json.loads(_PREFETCH_JSON)[:1]
+    with (
+        patch(
+            "esphome.espidf.framework.run_command",
+            return_value=(True, json.dumps(entries), ""),
+        ),
+        patch("esphome.espidf.framework.download_with_resume") as download,
+        patch("esphome.espidf.framework.get_system_python_path", return_value="python"),
+        patch(
+            "esphome.framework_helpers.ThreadPoolExecutor", wraps=ThreadPoolExecutor
+        ) as pool,
+    ):
+        _prefetch_idf_tool_archives(tmp_path, "esp32", ["required"], None)
+
+    pool.assert_called_once_with(max_workers=1)
+    assert download.call_count == 1
 
 
 def test_prefetch_skips_already_downloaded_archives(tmp_path: Path) -> None:
@@ -964,6 +1094,11 @@ def test_prefetch_one_failed_archive_does_not_stop_the_rest(
 ) -> None:
     """A single archive failing its download must not abort the prefetch of
     the remaining archives."""
+
+    def _fail_cmake_download(url: str, *args, **kwargs) -> None:
+        if "cmake" in url:
+            raise OSError("network down")
+
     with (
         patch(
             "esphome.espidf.framework.run_command",
@@ -971,7 +1106,7 @@ def test_prefetch_one_failed_archive_does_not_stop_the_rest(
         ),
         patch(
             "esphome.espidf.framework.download_with_resume",
-            side_effect=[OSError("network down"), None],
+            side_effect=_fail_cmake_download,
         ) as download,
         patch("esphome.espidf.framework.get_system_python_path", return_value="python"),
     ):
@@ -981,6 +1116,29 @@ def test_prefetch_one_failed_archive_does_not_stop_the_rest(
     assert "Could not prefetch cmake@3.30.2" in caplog.text
 
 
+def test_prefetch_finishes_progress_bar_and_cancels_queue(tmp_path: Path) -> None:
+    """The batch bar is closed out after the pool, and the pool is shut down
+    with cancel_futures so Ctrl-C does not drain every queued archive."""
+    with (
+        patch(
+            "esphome.espidf.framework.run_command",
+            return_value=(True, _PREFETCH_JSON, ""),
+        ),
+        patch("esphome.espidf.framework.download_with_resume"),
+        patch("esphome.espidf.framework.get_system_python_path", return_value="python"),
+        patch("esphome.espidf.framework.BatchDownloadProgress") as progress_cls,
+        patch(
+            "esphome.framework_helpers.ThreadPoolExecutor", wraps=ThreadPoolExecutor
+        ) as pool_cls,
+    ):
+        pool = MagicMock(wraps=ThreadPoolExecutor(max_workers=2))
+        pool_cls.return_value = pool
+        _prefetch_idf_tool_archives(tmp_path, "esp32", ["required"], None)
+
+    pool.shutdown.assert_called_once_with(wait=True, cancel_futures=True)
+    progress_cls.return_value.done.assert_called_once_with()
+
+
 def test_prefetch_passes_targets_and_tools_to_script(tmp_path: Path) -> None:
     with (
         patch(
@@ -1395,13 +1553,14 @@ def test_get_framework_env_without_python_env_uses_os_path(tmp_path: Path) -> No
 
 def _ccache_patches(tmp_path: Path, which: str | None, build_path: Path | None):
     return (
-        patch("esphome.espidf.framework.shutil.which", return_value=which),
+        patch("esphome.espidf.framework.resolve_ccache_path", return_value=which),
         patch(
             "esphome.espidf.framework.get_idf_tools_path",
             return_value=tmp_path / "tools",
         ),
+        # ccache_defaults_env (build_helpers.ccache) reads CORE at call time
         patch(
-            "esphome.espidf.framework.CORE",
+            "esphome.core.CORE",
             SimpleNamespace(build_path=build_path),
         ),
     )
@@ -1430,7 +1589,9 @@ def test_ccache_env_opt_out_via_env(tmp_path: Path) -> None:
     # short-circuits before build_path is needed.
     p1, p2, p3 = _ccache_patches(tmp_path, "/usr/bin/ccache", None)
     with patch.dict("os.environ", {"IDF_CCACHE_ENABLE": "0"}, clear=True), p1, p2, p3:
-        assert _ccache_env() == {}
+        # The canonical off spelling is exported: the raw value is inherited
+        # by idf.py, where a spelling like "disable" would read as truthy
+        assert _ccache_env() == {"IDF_CCACHE_ENABLE": "0"}
 
 
 def test_ccache_env_opt_in_without_binary(tmp_path: Path) -> None:
@@ -1444,6 +1605,48 @@ def test_ccache_env_opt_in_without_binary(tmp_path: Path) -> None:
     assert env["CCACHE_DEPEND"] == "1"
 
 
+def test_ccache_env_honors_shared_esphome_opt_out(tmp_path: Path) -> None:
+    """ESPHOME_CCACHE_ENABLE=0 disables ccache here too; the shared policy
+    must not apply to every backend except this one."""
+    _p1, p2, p3 = _ccache_patches(tmp_path, "/usr/bin/ccache", tmp_path / "build")
+    env_vars = {"ESPHOME_CCACHE_ENABLE": "0", "PATH": "/usr/bin"}
+    with patch.dict("os.environ", env_vars, clear=True), p2, p3:
+        # The real resolver runs so the opt-out parse is exercised
+        assert _ccache_env() == {}
+
+
+@pytest.mark.parametrize("value", ["off", "no"])
+def test_ccache_env_idf_knob_parses_strictly(tmp_path: Path, value: str) -> None:
+    """IDF_CCACHE_ENABLE uses the same strict table as the shared knob, so
+    "off" disables instead of reading as truthy."""
+    p1, p2, p3 = _ccache_patches(tmp_path, "/usr/bin/ccache", tmp_path / "build")
+    with patch.dict("os.environ", {"IDF_CCACHE_ENABLE": value}, clear=True), p1, p2, p3:
+        assert _ccache_env() == {"IDF_CCACHE_ENABLE": "0"}
+
+
+def test_ccache_env_idf_knob_unrecognized_warns_and_defers(
+    tmp_path: Path, caplog: pytest.LogCaptureFixture
+) -> None:
+    """An unparsable IDF_CCACHE_ENABLE warns, defers to the shared resolver,
+    and is not forwarded to idf.py as truthy."""
+    p1, p2, p3 = _ccache_patches(tmp_path, "/usr/bin/ccache", tmp_path / "build")
+    env_vars = {"IDF_CCACHE_ENABLE": "enabled"}
+    with patch.dict("os.environ", env_vars, clear=True), p1, p2, p3:
+        env = _ccache_env()
+    assert "unrecognized IDF_CCACHE_ENABLE" in caplog.text
+    assert env["IDF_CCACHE_ENABLE"] == "1"
+
+
+def test_ccache_env_idf_knob_wins_over_shared_opt_out(tmp_path: Path) -> None:
+    """IDF_CCACHE_ENABLE=1 takes precedence over ESPHOME_CCACHE_ENABLE=0."""
+    p1, p2, p3 = _ccache_patches(tmp_path, None, tmp_path / "build")
+    env_vars = {"IDF_CCACHE_ENABLE": "1", "ESPHOME_CCACHE_ENABLE": "0"}
+    with patch.dict("os.environ", env_vars, clear=True), p1, p2, p3:
+        env = _ccache_env()
+    assert env["CCACHE_DIR"] == str(tmp_path / "tools" / "ccache")
+    assert "IDF_CCACHE_ENABLE" not in env
+
+
 def test_ccache_env_preserves_user_overrides(tmp_path: Path) -> None:
     # User-set CCACHE_* values must not be clobbered; unset ones still default.
     p1, p2, p3 = _ccache_patches(tmp_path, "/usr/bin/ccache", tmp_path / "build")

tests/unit_tests/test_espidf_idedata.py

@@ -1,264 +0,0 @@
-"""Tests for esphome.espidf.idedata (compile_commands.json -> idedata)."""
-
-# pylint: disable=protected-access
-
-import json
-import os
-from pathlib import Path
-from unittest.mock import MagicMock, patch
-
-import pytest
-
-from esphome.espidf import idedata
-
-# An absolute, forward-slash (shlex-safe) path prefix valid on the host OS, so
-# tests exercise the same is-absolute / normalize behavior as a real compile DB
-# (a drive-qualified path on Windows, a leading slash elsewhere).
-ABS = "C:/" if os.name == "nt" else "/"
-
-
-def _entry(directory: str, file: str, command: str) -> dict:
-    return {"directory": directory, "file": file, "command": command}
-
-
-def test_parse_entry_extracts_fields() -> None:
-    """cxx_path, defines, includes and remaining flags are split apart."""
-    entry = _entry(
-        f"{ABS}build",
-        f"{ABS}build/src/esphome/core/application.cpp",
-        f"/tools/xtensa-esp32-elf-g++ -DUSE_ESP32 -DESPHOME_LOG_LEVEL=5 "
-        f"-I{ABS}inc/a -isystem {ABS}sys/b -std=gnu++20 -c app.cpp -o app.cpp.o",
-    )
-
-    cxx_path, defines, includes, cxx_flags = idedata._parse_entry(entry)
-
-    assert cxx_path == "/tools/xtensa-esp32-elf-g++"
-    assert "USE_ESP32" in defines
-    assert "ESPHOME_LOG_LEVEL=5" in defines
-    assert f"{ABS}inc/a" in includes
-    assert f"{ABS}sys/b" in includes
-    assert "-std=gnu++20" in cxx_flags
-    # input/output files and their flags are not treated as flags
-    assert "-c" not in cxx_flags
-    assert "-o" not in cxx_flags
-    assert "app.cpp" not in cxx_flags
-    assert "app.cpp.o" not in cxx_flags
-
-
-def test_parse_entry_space_separated_args() -> None:
-    """``-D X`` / ``-I path`` (separate arg) and ``-isystem<path>`` (joined)."""
-    entry = _entry(
-        f"{ABS}build",
-        f"{ABS}build/src/esphome/x.cpp",
-        f"g++ -D FOO=1 -I {ABS}inc/sep -isystem{ABS}sys/joined -c x.cpp",
-    )
-
-    _, defines, includes, _ = idedata._parse_entry(entry)
-
-    assert "FOO=1" in defines
-    assert f"{ABS}inc/sep" in includes
-    assert f"{ABS}sys/joined" in includes
-
-
-def test_parse_entry_resolves_relative_includes() -> None:
-    """Relative includes are resolved against the entry's ``directory``."""
-    directory = f"{ABS}build/proj"
-    entry = _entry(
-        directory,
-        f"{directory}/src/esphome/x.cpp",
-        "g++ -Iconfig -I../shared -isystem rel/sys -c x.cpp",
-    )
-
-    _, _, includes, _ = idedata._parse_entry(entry)
-
-    def resolved(rel: str) -> str:
-        # _parse_entry emits forward slashes for consistency (normpath would
-        # yield backslashes on Windows).
-        return os.path.normpath(Path(directory) / rel).replace("\\", "/")
-
-    assert resolved("config") in includes
-    assert resolved("../shared") in includes  # ../ normalized away
-    assert resolved("rel/sys") in includes
-    # nothing is left relative
-    assert all(Path(inc).is_absolute() for inc in includes)
-
-
-def test_parse_entry_skips_dependency_flags() -> None:
-    """Dependency-generation flags (and their args) are dropped."""
-    entry = _entry(
-        "/build",
-        "/build/src/esphome/x.cpp",
-        "g++ -MD -MT x.cpp.o -MF x.cpp.o.d -c x.cpp -o x.cpp.o",
-    )
-
-    _, _, _, cxx_flags = idedata._parse_entry(entry)
-
-    for tok in ("-MD", "-MT", "x.cpp.o", "-MF", "x.cpp.o.d", "-c", "-o", "x.cpp"):
-        assert tok not in cxx_flags
-
-
-def test_expand_response_files(tmp_path: Path) -> None:
-    """``@file`` arguments are inlined relative to the directory."""
-    rsp = tmp_path / "flags.rsp"
-    rsp.write_text("-DFROM_RSP -I/rsp/inc")
-
-    tokens = idedata._expand_response_files(
-        ["g++", f"@{rsp.name}", "-c", "x.cpp"], tmp_path
-    )
-
-    assert "-DFROM_RSP" in tokens
-    assert "-I/rsp/inc" in tokens
-    assert not any(t.startswith("@") for t in tokens)
-
-
-def test_expand_response_files_keeps_literal_when_missing(tmp_path: Path) -> None:
-    """An unreadable ``@file`` token is kept verbatim rather than dropped."""
-    tokens = idedata._expand_response_files(["g++", "@nope.rsp"], tmp_path)
-    assert "@nope.rsp" in tokens
-
-
-def test_pick_entry_prefers_esphome_tu() -> None:
-    """A ``/src/esphome/`` C++ TU is picked over other compile entries."""
-    entries = [
-        _entry("/b", "/b/managed_components/foo/foo.c", "gcc -c foo.c"),
-        _entry("/b", "/b/src/esphome/core/app.cpp", "g++ -c app.cpp"),
-    ]
-    assert idedata._pick_entry(entries)["file"].endswith("app.cpp")
-
-
-def test_pick_entry_falls_back_to_any_cxx_tu() -> None:
-    """With no ``/src/esphome/`` TU present, the first C++ entry is the fallback."""
-    entries = [
-        _entry("/b", "/b/managed_components/foo/foo.c", "gcc -c foo.c"),
-        _entry("/b", "/b/components/x/x.cpp", "g++ -c x.cpp"),
-    ]
-    assert idedata._pick_entry(entries)["file"].endswith("x.cpp")
-
-
-def test_is_esphome_src_handles_backslash_paths() -> None:
-    r"""The src marker must match Windows ``\src\esphome\`` paths too.
-
-    compile_commands ``file`` entries use the OS-native separator; if the
-    marker only matched forward slashes no source would match on Windows and
-    the build-include union would be silently empty.
-    """
-    assert idedata._is_esphome_src(r"C:\b\src\esphome\core\app.cpp")
-    assert idedata._is_esphome_src("/b/src/esphome/core/app.cpp")
-    # non-esphome and non-C++ still rejected regardless of separator
-    assert not idedata._is_esphome_src(r"C:\b\managed_components\x\x.cpp")
-    assert not idedata._is_esphome_src(r"C:\b\src\esphome\core\app.h")
-
-
-def test_idedata_from_build(tmp_path: Path) -> None:
-    """Full transform: representative entry + include union + toolchain dirs."""
-    compile_commands = tmp_path / "compile_commands.json"
-    entries = [
-        _entry(
-            f"{ABS}b",
-            f"{ABS}b/src/esphome/core/app.cpp",
-            f"g++ -DUSE_ESP32 -I{ABS}inc/core -std=gnu++20 -c app.cpp -o app.cpp.o",
-        ),
-        _entry(
-            f"{ABS}b",
-            f"{ABS}b/src/esphome/sensor/s.cpp",
-            f"g++ -DUSE_ESP32 -I{ABS}inc/sensor -c s.cpp -o s.cpp.o",
-        ),
-        # non-esphome TU: its includes must not leak into the union
-        _entry(
-            f"{ABS}b",
-            f"{ABS}b/managed_components/x/x.c",
-            f"gcc -I{ABS}inc/managed -c x.c",
-        ),
-    ]
-    compile_commands.write_text(json.dumps(entries))
-
-    fake_proc = MagicMock(
-        returncode=0,
-        stderr=(
-            "ignored\n"
-            "#include <...> search starts here:\n"
-            " /tc/inc/c++\n"
-            " /tc/inc\n"
-            "End of search list.\n"
-            "more ignored\n"
-        ),
-    )
-    with patch.object(idedata.subprocess, "run", return_value=fake_proc):
-        data = idedata.idedata_from_build(compile_commands)
-
-    assert data["cxx_path"] == "g++"
-    assert "USE_ESP32" in data["defines"]
-    assert "-std=gnu++20" in data["cxx_flags"]
-    # include dirs unioned across all esphome TUs
-    assert f"{ABS}inc/core" in data["includes"]["build"]
-    assert f"{ABS}inc/sensor" in data["includes"]["build"]
-    # the non-esphome TU is excluded from the union
-    assert f"{ABS}inc/managed" not in data["includes"]["build"]
-    # toolchain search dirs parsed from the compiler's -v output
-    assert data["includes"]["toolchain"] == ["/tc/inc/c++", "/tc/inc"]
-
-
-def test_get_toolchain_includes_raises_on_probe_failure() -> None:
-    """A failed compiler probe is a hard error, not a silent empty list."""
-    fake_proc = MagicMock(returncode=1, stderr="xtensa-esp32-elf-g++: not found")
-    with (
-        patch.object(idedata.subprocess, "run", return_value=fake_proc),
-        pytest.raises(RuntimeError, match="builtin include dirs"),
-    ):
-        idedata._get_toolchain_includes("/bad/compiler")
-
-
-def test_get_toolchain_includes_raises_when_no_dirs_found() -> None:
-    """Markers present but no dirs (anomalous output) also raises."""
-    fake_proc = MagicMock(
-        returncode=0,
-        stderr="#include <...> search starts here:\nEnd of search list.\n",
-    )
-    with (
-        patch.object(idedata.subprocess, "run", return_value=fake_proc),
-        pytest.raises(RuntimeError, match="builtin include dirs"),
-    ):
-        idedata._get_toolchain_includes("/some/compiler")
-
-
-# ESP-IDF's compile_commands.json on Windows mixes literal backslash path
-# separators in the compiler path with shell ``\"`` quote-escaping in defines,
-# which only the real Windows argv parser handles. These exercise that path.
-@pytest.mark.skipif(os.name != "nt", reason="Windows argv tokenization")
-def test_split_command_preserves_paths_and_unescapes_quotes() -> None:
-    r"""Backslash paths survive while ``\"`` define-quoting is unescaped."""
-    command = r"C:\esp\bin\riscv32-esp-elf-g++.exe -DVER=\"1.2.3\" -IC:/inc/a -c x.cpp"
-
-    tokens = idedata._split_command(command)
-
-    assert tokens[0] == r"C:\esp\bin\riscv32-esp-elf-g++.exe"
-    assert '-DVER="1.2.3"' in tokens
-    assert "-IC:/inc/a" in tokens
-
-
-@pytest.mark.skipif(os.name != "nt", reason="Windows argv tokenization")
-def test_split_command_empty_returns_empty() -> None:
-    """An empty or blank command tokenizes to ``[]`` (e.g. an empty response file).
-
-    Guards against ``CommandLineToArgvW("")`` returning the current process name
-    instead of an empty list.
-    """
-    assert idedata._split_command("") == []
-    assert idedata._split_command("   ") == []
-
-
-@pytest.mark.skipif(os.name != "nt", reason="Windows argv tokenization")
-def test_parse_entry_normalizes_windows_cxx_path() -> None:
-    """A backslash compiler path is emitted forward-slashed; define unescaped."""
-    entry = _entry(
-        r"C:\b",
-        r"C:\b\src\esphome\x.cpp",
-        r"C:\esp\bin\g++.exe -DVER=\"1.2.3\" -IC:/inc/a -c x.cpp",
-    )
-
-    cxx_path, defines, includes, _ = idedata._parse_entry(entry)
-
-    assert cxx_path == "C:/esp/bin/g++.exe"
-    assert "\\" not in cxx_path
-    assert 'VER="1.2.3"' in defines
-    assert "C:/inc/a" in includes

tests/unit_tests/test_espidf_toolchain.py

@@ -140,7 +140,7 @@ def test_get_idedata_generates_and_caches(setup_core: Path) -> None:
     compile_commands.write_text("[]")
 
     with patch(
-        "esphome.espidf.idedata.idedata_from_build",
+        "esphome.build_helpers.idedata.idedata_from_build",
         return_value={"cxx_path": "g++"},
     ) as mock_transform:
         result = toolchain.get_idedata()
@@ -151,114 +151,6 @@ def test_get_idedata_generates_and_caches(setup_core: Path) -> None:
     assert json.loads(cache.read_text()) == {"cxx_path": "g++", "prog_path": prog_path}
 
 
-def test_get_idedata_uses_cache_when_valid(setup_core: Path) -> None:
-    """A cache at least as new as the compile DB is reused without regenerating."""
-    compile_commands, cache = _setup_build(setup_core)
-    compile_commands.parent.mkdir(parents=True, exist_ok=True)
-    compile_commands.write_text("[]")
-    cache.parent.mkdir(parents=True, exist_ok=True)
-    cache.write_text('{"cc_path": "cached-gcc", "cxx_path": "cached"}')
-    cc_mtime = compile_commands.stat().st_mtime
-    os.utime(cache, (cc_mtime + 1, cc_mtime + 1))
-
-    with patch("esphome.espidf.idedata.idedata_from_build") as mock_transform:
-        result = toolchain.get_idedata()
-
-    mock_transform.assert_not_called()
-    assert result == {"cc_path": "cached-gcc", "cxx_path": "cached"}
-
-
-def test_get_idedata_regenerates_cache_without_cc_path(setup_core: Path) -> None:
-    """A cache predating cc_path is rebuilt even though it is newer.
-
-    Such a cache stays newer than the compile DB forever, so consumers that
-    derive the binutils paths from cc_path would keep failing on it.
-    """
-    compile_commands, cache = _setup_build(setup_core)
-    compile_commands.parent.mkdir(parents=True, exist_ok=True)
-    compile_commands.write_text("[]")
-    cache.parent.mkdir(parents=True, exist_ok=True)
-    cache.write_text('{"cxx_path": "cached"}')
-    cc_mtime = compile_commands.stat().st_mtime
-    os.utime(cache, (cc_mtime + 1, cc_mtime + 1))
-
-    with patch(
-        "esphome.espidf.idedata.idedata_from_build",
-        return_value={"cc_path": "gcc", "cxx_path": "g++"},
-    ) as mock_transform:
-        result = toolchain.get_idedata()
-
-    mock_transform.assert_called_once()
-    assert result["cc_path"] == "gcc"
-
-
-def test_get_idedata_regenerates_when_compile_commands_newer(setup_core: Path) -> None:
-    """A compile DB newer than the cache forces regeneration."""
-    compile_commands, cache = _setup_build(setup_core)
-    cache.parent.mkdir(parents=True, exist_ok=True)
-    cache.write_text('{"cxx_path": "stale"}')
-    compile_commands.parent.mkdir(parents=True, exist_ok=True)
-    compile_commands.write_text("[]")
-    cache_mtime = cache.stat().st_mtime
-    os.utime(compile_commands, (cache_mtime + 1, cache_mtime + 1))
-
-    with patch(
-        "esphome.espidf.idedata.idedata_from_build",
-        return_value={"cxx_path": "fresh"},
-    ) as mock_transform:
-        result = toolchain.get_idedata()
-
-    mock_transform.assert_called_once()
-    assert result == {"cxx_path": "fresh", "prog_path": str(toolchain.get_elf_path())}
-
-
-@pytest.mark.parametrize("cached", ['"cc_path is a string"', "[]", "42"])
-def test_get_idedata_regenerates_on_non_dict_cache(
-    setup_core: Path, cached: str
-) -> None:
-    """A newer cache holding valid JSON that is not an object is regenerated.
-
-    A bare string would otherwise pass the cc_path check by substring and be
-    handed to consumers expecting a dict.
-    """
-    compile_commands, cache = _setup_build(setup_core)
-    compile_commands.parent.mkdir(parents=True, exist_ok=True)
-    compile_commands.write_text("[]")
-    cache.parent.mkdir(parents=True, exist_ok=True)
-    cache.write_text(cached)
-    cc_mtime = compile_commands.stat().st_mtime
-    os.utime(cache, (cc_mtime + 1, cc_mtime + 1))
-
-    with patch(
-        "esphome.espidf.idedata.idedata_from_build",
-        return_value={"cc_path": "gcc", "cxx_path": "g++"},
-    ) as mock_transform:
-        result = toolchain.get_idedata()
-
-    mock_transform.assert_called_once()
-    assert isinstance(result, dict)
-
-
-def test_get_idedata_regenerates_on_corrupted_cache(setup_core: Path) -> None:
-    """An unparseable (but newer) cache falls back to regeneration."""
-    compile_commands, cache = _setup_build(setup_core)
-    compile_commands.parent.mkdir(parents=True, exist_ok=True)
-    compile_commands.write_text("[]")
-    cache.parent.mkdir(parents=True, exist_ok=True)
-    cache.write_text("{not json")
-    cc_mtime = compile_commands.stat().st_mtime
-    os.utime(cache, (cc_mtime + 1, cc_mtime + 1))
-
-    with patch(
-        "esphome.espidf.idedata.idedata_from_build",
-        return_value={"cxx_path": "regen"},
-    ) as mock_transform:
-        result = toolchain.get_idedata()
-
-    mock_transform.assert_called_once()
-    assert result == {"cxx_path": "regen", "prog_path": str(toolchain.get_elf_path())}
-
-
 def test_get_idedata_prog_path_points_at_firmware_elf(setup_core: Path) -> None:
     """The idedata exposes prog_path (the ELF) so consumers like build-action
     can locate firmware.factory.bin / firmware.ota.bin as its siblings."""
@@ -267,7 +159,7 @@ def test_get_idedata_prog_path_points_at_firmware_elf(setup_core: Path) -> None:
     compile_commands.write_text("[]")
 
     with patch(
-        "esphome.espidf.idedata.idedata_from_build",
+        "esphome.build_helpers.idedata.idedata_from_build",
         return_value={"cxx_path": "g++"},
     ):
         result = toolchain.get_idedata()

tests/unit_tests/test_framework_helpers.py

@@ -21,6 +21,7 @@ import requests as req
 from esphome import framework_helpers
 from esphome.core import EsphomeError
 from esphome.framework_helpers import (
+    BatchDownloadProgress,
     _7z_extract_all,
     _detect_archive_root,
     _is_transient_download_error,
@@ -1112,6 +1113,108 @@ class TestDownloadWithResume:
         assert mock_get.call_args[1]["headers"] == {}
         assert dest.read_bytes() == b"data"
 
+    def test_progress_callback_reports_absolute_bytes(self, tmp_path: Path) -> None:
+        """With a callback no bar is drawn; the callback sees the running
+        byte count of this file, then its final verified size."""
+        dest = tmp_path / "tool.tar.gz"
+        resp = _mock_response(b"")
+        resp.headers = {"content-length": "7"}
+        resp.iter_content.return_value = [b"1234", b"567"]
+        seen: list[int] = []
+        with (
+            patch("requests.get", return_value=resp),
+            patch("esphome.framework_helpers.ProgressBar") as bar,
+        ):
+            download_with_resume(
+                "https://example.com/t", dest, size=7, progress=seen.append
+            )
+        assert seen == [0, 4, 7, 7]
+        bar.assert_not_called()
+
+    def test_progress_callback_seeds_with_resume_offset(self, tmp_path: Path) -> None:
+        dest = tmp_path / "tool.tar.gz"
+        (tmp_path / "tool.tar.gz.part").write_bytes(b"12345")
+        good = hashlib.sha256(b"12345678").hexdigest()
+        seen: list[int] = []
+        with patch("requests.get", return_value=_resumed_response(b"678")):
+            download_with_resume(
+                "https://example.com/t", dest, sha256=good, size=8, progress=seen.append
+            )
+        assert seen[0] == 5
+        assert seen[-1] == 8
+
+    def test_progress_callback_credits_already_complete_download(
+        self, tmp_path: Path
+    ) -> None:
+        """A verified dest from an earlier run still counts toward the batch."""
+        dest = tmp_path / "tool.tar.gz"
+        dest.write_bytes(b"12345678")
+        seen: list[int] = []
+        with patch("requests.get") as mock_get:
+            download_with_resume(
+                "https://example.com/t", dest, size=8, progress=seen.append
+            )
+        mock_get.assert_not_called()
+        assert seen == [8]
+
+
+class TestBatchDownloadProgress:
+    def test_sums_trackers_into_one_bar(self) -> None:
+        with patch("esphome.framework_helpers.ProgressBar") as bar_cls:
+            progress = BatchDownloadProgress("Downloading", 100)
+            a = progress.tracker()
+            b = progress.tracker()
+            a(10)
+            b(20)
+            a(30)
+            a(0)  # a restart from zero takes that file's bytes back out
+        bar_cls.assert_called_once_with("Downloading")
+        updates = [c[0][0] for c in bar_cls.return_value.update.call_args_list]
+        assert updates == [0.1, 0.3, 0.5, 0.2]
+
+    def test_clamps_at_one(self) -> None:
+        """Sizes are advisory; an over-delivering server never pushes past 100%."""
+        with patch("esphome.framework_helpers.ProgressBar") as bar_cls:
+            progress = BatchDownloadProgress("Downloading", 10)
+            progress.tracker()(25)
+        assert bar_cls.return_value.update.call_args[0][0] == 1
+
+    def test_unknown_total_draws_nothing(self) -> None:
+        with patch("esphome.framework_helpers.ProgressBar") as bar_cls:
+            progress = BatchDownloadProgress("Downloading", 0)
+            progress.tracker()(5)
+            progress.done()
+        bar_cls.assert_not_called()
+
+    def test_done_ends_an_unfinished_bar(self) -> None:
+        """A batch that stops short of 100% (a failed archive) still ends its
+        line so the next log message starts on a fresh row."""
+        stream = io.StringIO()
+        stream.isatty = lambda: True  # type: ignore[method-assign]
+        with patch("esphome.helpers.sys.stderr", stream):
+            progress = BatchDownloadProgress("Downloading", 10)
+            progress.tracker()(5)
+            progress.done()
+        assert stream.getvalue().endswith("50% \n")
+
+    def test_done_before_any_frame_writes_nothing(self) -> None:
+        """A batch aborted before any tracker fired must not emit a stray
+        newline for a bar that was never drawn."""
+        stream = io.StringIO()
+        stream.isatty = lambda: True  # type: ignore[method-assign]
+        with patch("esphome.helpers.sys.stderr", stream):
+            BatchDownloadProgress("Downloading", 10).done()
+        assert stream.getvalue() == ""
+
+    def test_done_after_full_bar_adds_nothing(self) -> None:
+        stream = io.StringIO()
+        stream.isatty = lambda: True  # type: ignore[method-assign]
+        with patch("esphome.helpers.sys.stderr", stream):
+            progress = BatchDownloadProgress("Downloading", 10)
+            progress.tracker()(10)
+            progress.done()
+        assert stream.getvalue().endswith("100% Done...\r\n")
+
 
 class TestDownloadFromMirrors:
     def test_success_returns_url_and_writes_content(self, tmp_path: Path) -> None:
@@ -2091,3 +2194,37 @@ class TestGetProjectCxxCompileFlags:
     def test_empty_flags(self) -> None:
         with patch("esphome.core.CORE", _make_core_cxx(set())):
             assert get_project_cxx_compile_flags() == []
+
+
+@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.framework_helpers.sys.platform", platform):
+        assert framework_helpers.strip_win_long_path_prefix(input_path) == expected

tests/unit_tests/test_main.py

@@ -17,6 +17,7 @@ from unittest.mock import AsyncMock, MagicMock, Mock, patch
 
 import pytest
 from pytest import CaptureFixture
+import serial
 from zeroconf import ServiceStateChange
 
 from esphome import __main__ as main, yaml_util
@@ -26,6 +27,7 @@ from esphome.__main__ import (
     _make_crystal_freq_callback,
     _redact_with_legacy_fallback,
     _resolve_network_devices,
+    _should_subscribe_states,
     _split_network_devices,
     _unresolved_default_error,
     _validate_bootloader_binary,
@@ -69,19 +71,21 @@ from esphome.__main__ import (
 )
 from esphome.address_cache import AddressCache
 from esphome.bundle import BUNDLE_EXTENSION, BundleFile, BundleResult
-from esphome.components import esp32, esp8266
+from esphome.components import esp32, esp8266, mqtt
 from esphome.components.esp32 import (
     KEY_ESP32,
     KEY_VARIANT,
     VARIANT_ESP32,
     get_esp32_variant,
 )
+from esphome.config import Config
 from esphome.const import (
     CONF_API,
     CONF_AUTH,
     CONF_BAUD_RATE,
     CONF_BROKER,
     CONF_DISABLED,
+    CONF_DISCOVER_IP,
     CONF_ESPHOME,
     CONF_LEVEL,
     CONF_LOG,
@@ -103,6 +107,7 @@ from esphome.const import (
     CONF_WEB_SERVER,
     CONF_WIFI,
     KEY_CORE,
+    KEY_TARGET_FRAMEWORK,
     KEY_TARGET_PLATFORM,
     PLATFORM_BK72XX,
     PLATFORM_ESP32,
@@ -567,8 +572,6 @@ def test_command_config__no_defaults_dumps_user_snapshot(
 ) -> None:
     """``--no-defaults`` dumps ``config.user_config`` instead of the
     validated config, so schema defaults don't leak into the output."""
-    from esphome.config import Config
-
     setup_core(tmp_path=tmp_path, config={"esphome": {"name": "test"}})
     args = MockArgs()
     args.show_secrets = True
@@ -621,8 +624,6 @@ def test_command_config__no_defaults_skips_strip_default_ids(
 ) -> None:
     """When ``--no-defaults`` is set, ``strip_default_ids`` isn't run --
     the user snapshot is already free of schema-injected IDs."""
-    from esphome.config import Config
-
     setup_core(tmp_path=tmp_path, config={"esphome": {"name": "test"}})
     args = MockArgs()
     args.show_secrets = True
@@ -3440,9 +3441,6 @@ def test_get_port_type() -> None:
 
 def test_mqtt_reexports_discover_ip() -> None:
     """The old import path must keep working for external code."""
-    from esphome.components import mqtt
-    from esphome.const import CONF_DISCOVER_IP
-
     assert mqtt.CONF_DISCOVER_IP is CONF_DISCOVER_IP
 
 
@@ -5909,8 +5907,6 @@ class MockSerial:
             chunk = self.chunks[self.chunk_index]
             if chunk is MOCK_SERIAL_END:
                 # Sentinel means we're done - simulate port closed
-                import serial
-
                 raise serial.SerialException("Port closed")
             # Respect the requested size and keep any remaining bytes
             if size <= 0:
@@ -5924,8 +5920,6 @@ class MockSerial:
                 # Entire chunk consumed; advance to the next one
                 self.chunk_index += 1
             return data  # type: ignore[return-value]
-        import serial
-
         raise serial.SerialException("Port closed")
 
 
@@ -6784,8 +6778,6 @@ def test_parse_args_argcomplete_only_runs_when_completing() -> None:
 
 def test_should_subscribe_states_default() -> None:
     """Test that states are shown by default when nothing is set."""
-    from esphome.__main__ import _should_subscribe_states
-
     args = parse_args(["esphome", "logs", "device.yaml"])
     with patch.dict(os.environ, {}, clear=False):
         os.environ.pop("ESPHOME_LOG_STATES", None)
@@ -6794,8 +6786,6 @@ def test_should_subscribe_states_default() -> None:
 
 def test_should_subscribe_states_env_suppresses() -> None:
     """Test that ESPHOME_LOG_STATES=false suppresses states by default."""
-    from esphome.__main__ import _should_subscribe_states
-
     args = parse_args(["esphome", "logs", "device.yaml"])
     with patch.dict(os.environ, {"ESPHOME_LOG_STATES": "false"}):
         assert _should_subscribe_states(args) is False
@@ -6803,8 +6793,6 @@ def test_should_subscribe_states_env_suppresses() -> None:
 
 def test_should_subscribe_states_env_enables() -> None:
     """Test that ESPHOME_LOG_STATES=true enables states by default."""
-    from esphome.__main__ import _should_subscribe_states
-
     args = parse_args(["esphome", "logs", "device.yaml"])
     with patch.dict(os.environ, {"ESPHOME_LOG_STATES": "true"}):
         assert _should_subscribe_states(args) is True
@@ -6812,8 +6800,6 @@ def test_should_subscribe_states_env_enables() -> None:
 
 def test_should_subscribe_states_flag_overrides_env() -> None:
     """Test that --states overrides ESPHOME_LOG_STATES=false."""
-    from esphome.__main__ import _should_subscribe_states
-
     args = parse_args(["esphome", "logs", "--states", "device.yaml"])
     with patch.dict(os.environ, {"ESPHOME_LOG_STATES": "false"}):
         assert _should_subscribe_states(args) is True
@@ -6821,8 +6807,6 @@ def test_should_subscribe_states_flag_overrides_env() -> None:
 
 def test_should_subscribe_states_no_flag_overrides_env() -> None:
     """Test that --no-states overrides ESPHOME_LOG_STATES=true."""
-    from esphome.__main__ import _should_subscribe_states
-
     args = parse_args(["esphome", "logs", "--no-states", "device.yaml"])
     with patch.dict(os.environ, {"ESPHOME_LOG_STATES": "true"}):
         assert _should_subscribe_states(args) is False
@@ -7135,6 +7119,118 @@ def test_warn_source_tree_mismatch_falls_back_when_stat_fails(
     assert not caplog.text
 
 
+@pytest.mark.parametrize(
+    "error",
+    [
+        FileNotFoundError("no such compiler"),
+        RuntimeError("Could not query builtin include dirs"),
+        ValueError("no C++ translation unit found"),
+        KeyError("command"),
+        None,  # replaced with EsphomeError inside
+    ],
+)
+def test_compile_program_espidf_idedata_failure_does_not_fail_build(
+    error: Exception,
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    """A post-compile idedata error is a warning: the firmware already built."""
+    if error is None:
+        error = EsphomeError("compile database is unusable")
+    CORE.toolchain = Toolchain.ESP_IDF
+    CORE.data[KEY_CORE] = {
+        KEY_TARGET_PLATFORM: "esp32",
+        KEY_TARGET_FRAMEWORK: "esp-idf",
+    }
+    with (
+        patch("esphome.espidf.toolchain.run_compile", return_value=0),
+        patch("esphome.espidf.toolchain.create_factory_bin"),
+        patch("esphome.espidf.toolchain.create_ota_bin"),
+        patch("esphome.espidf.toolchain.create_elf_copy"),
+        patch("esphome.espidf.toolchain.get_idedata", side_effect=error),
+        patch("esphome.__main__._check_and_emit_build_info"),
+    ):
+        assert compile_program(MagicMock(), {}) == 0
+    assert "Could not generate idedata" in caplog.text
+
+
+def test_compile_program_espidf_idedata_success_is_silent(
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    """The healthy path: idedata generated, nothing to warn about."""
+    CORE.toolchain = Toolchain.ESP_IDF
+    CORE.data[KEY_CORE] = {
+        KEY_TARGET_PLATFORM: "esp32",
+        KEY_TARGET_FRAMEWORK: "esp-idf",
+    }
+    with (
+        patch("esphome.espidf.toolchain.run_compile", return_value=0),
+        patch("esphome.espidf.toolchain.create_factory_bin"),
+        patch("esphome.espidf.toolchain.create_ota_bin"),
+        patch("esphome.espidf.toolchain.create_elf_copy"),
+        patch("esphome.espidf.toolchain.get_idedata", return_value={"cc_path": "x"}),
+        patch("esphome.__main__._check_and_emit_build_info"),
+    ):
+        assert compile_program(MagicMock(), {}) == 0
+    assert "idedata" not in caplog.text
+
+
+def test_compile_program_espidf_idedata_none_warns(
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    """A silent None from the post-compile idedata refresh is made visible."""
+    CORE.toolchain = Toolchain.ESP_IDF
+    CORE.data[KEY_CORE] = {
+        KEY_TARGET_PLATFORM: "esp32",
+        KEY_TARGET_FRAMEWORK: "esp-idf",
+    }
+    with (
+        patch("esphome.espidf.toolchain.run_compile", return_value=0),
+        patch("esphome.espidf.toolchain.create_factory_bin"),
+        patch("esphome.espidf.toolchain.create_ota_bin"),
+        patch("esphome.espidf.toolchain.create_elf_copy"),
+        patch("esphome.espidf.toolchain.get_idedata", return_value=None),
+        patch("esphome.__main__._check_and_emit_build_info"),
+    ):
+        assert compile_program(MagicMock(), {}) == 0
+    assert "No idedata was generated" in caplog.text
+
+
+def test_cli_toolchain_skips_the_validated_config_cache(tmp_path: Path) -> None:
+    """An explicit --toolchain must run the per-platform validators, so the
+    upload/logs fast path becomes a cache miss."""
+    conf = tmp_path / "device.yaml"
+    conf.write_text("esphome:\n  name: t\n")
+    argv = ["esphome", "--toolchain", "arduino", "logs", str(conf)]
+    with (
+        patch("esphome.compiled_config.load_compiled_config") as mock_cache,
+        patch("esphome.config.read_config", return_value=None) as mock_read,
+    ):
+        assert run_esphome(argv) == 2
+    mock_cache.assert_not_called()
+    mock_read.assert_called_once()
+
+
+def test_cli_toolchain_still_refreshes_the_validated_config_cache(
+    tmp_path: Path,
+) -> None:
+    """An explicit --toolchain gates only the cache read; with a matching
+    sidecar the freshly validated config is still saved."""
+    conf = tmp_path / "device.yaml"
+    conf.write_text("esphome:\n  name: t\n")
+    argv = ["esphome", "--toolchain", "platformio", "logs", str(conf)]
+    with (
+        patch("esphome.compiled_config.load_compiled_config") as mock_load,
+        patch("esphome.config.read_config", return_value={CONF_ESPHOME: {}}),
+        patch("esphome.compiled_config.save_compiled_config_and_sidecar") as mock_save,
+        patch.dict(
+            "esphome.__main__.POST_CONFIG_ACTIONS", {"logs": Mock(return_value=0)}
+        ),
+    ):
+        assert run_esphome(argv) == 0
+    mock_load.assert_not_called()
+    mock_save.assert_called_once()
+
+
 @pytest.mark.asyncio
 async def test_wrap_to_code_comment_is_insertion_order_independent() -> None:
     """The config comment dumps with sorted keys: voluptuous fills schema

tests/unit_tests/test_nrf52_framework.py

@@ -7,8 +7,10 @@ import sys
 from types import SimpleNamespace
 from unittest.mock import patch
 
+import platformdirs
 import pytest
 
+from esphome.components.nrf52 import _resolve_toolchain
 from esphome.components.nrf52.framework import (
     _PLATFORMIO_PENV_REQUIREMENTS,
     _REQUIREMENTS,
@@ -22,8 +24,9 @@ from esphome.components.nrf52.framework import (
     get_sdk_nrf_tools_path,
     setup_platformio_python_env,
 )
+import esphome.config_validation as cv
 from esphome.config_validation import Version
-from esphome.const import KEY_CORE, KEY_FRAMEWORK_VERSION
+from esphome.const import KEY_CORE, KEY_FRAMEWORK_VERSION, Toolchain
 from esphome.core import CORE, EsphomeError
 from esphome.framework_helpers import get_python_env_executable_path
 
@@ -558,7 +561,6 @@ def testget_tools_path_blank_env_falls_back_to_default(
     Path("") would resolve to the working directory, which clean-all could
     then delete by accident.
     """
-    import platformdirs
 
     monkeypatch.setenv("ESPHOME_SDK_NRF_PREFIX", value)
     expected = (
@@ -570,7 +572,6 @@ def testget_tools_path_blank_env_falls_back_to_default(
 def testget_tools_path_default_is_global_cache(
     monkeypatch: pytest.MonkeyPatch,
 ) -> None:
-    import platformdirs
 
     monkeypatch.delenv("ESPHOME_SDK_NRF_PREFIX", raising=False)
     expected = (
@@ -619,3 +620,11 @@ def test_needs_venv_rebuild_on_dangling_interpreter_symlink(tmp_path: Path) -> N
     assert not python.exists()
 
     assert _needs_venv_rebuild(python, sentinel, "abc123")
+
+
+def test_resolve_toolchain_rejects_unsupported() -> None:
+    """A --toolchain nRF52 cannot serve fails instead of degrading silently."""
+
+    CORE.toolchain = Toolchain.ARDUINO
+    with pytest.raises(cv.Invalid, match="Unsupported toolchain 'arduino'"):
+        _resolve_toolchain({})

tests/unit_tests/test_platformio_extra_script.py

@@ -0,0 +1,401 @@
+"""Tests for the shared extraScript machinery (platformio.extra_script)."""
+
+from __future__ import annotations
+
+import logging
+import os
+from pathlib import Path
+from unittest.mock import patch
+
+import pytest
+
+from esphome.core import EsphomeError
+from esphome.platformio.extra_script import (
+    ExtraScriptResult,
+    _FakeSConsEnv,
+    apply_extra_script,
+    captured_as_build_flags,
+    run_extra_script,
+)
+from esphome.platformio.library import ConvertedLibrary as IDFComponent, URLSource
+
+
+def test_extra_script_captures_libpath_libs_and_defines(tmp_path):
+
+    (tmp_path / "src" / "esp32").mkdir(parents=True)
+    script = tmp_path / "extra_script.py"
+    script.write_text(
+        "Import('env')\n"
+        "mcu = env.get('BOARD_MCU')\n"
+        "env.Append(\n"
+        "    LIBPATH=[join('src', mcu)],\n"
+        "    LIBS=['algobsec'],\n"
+        "    CPPDEFINES=['FOO', ('BAR', '1')],\n"
+        "    LINKFLAGS=['-Wl,--gc-sections'],\n"
+        ")\n"
+    )
+    # The script uses bare ``join`` (PIO's extra-scripts run inside SCons
+    # where this is in scope). Inject it via the script header so the
+    # shim's exec namespace can resolve it.
+    script.write_text("from os.path import join\n" + script.read_text())
+
+    result = run_extra_script(
+        script, library_dir=tmp_path, board_mcu="esp32", pio_platform="espressif32"
+    )
+
+    assert result.libpath == [str(Path("src") / "esp32")]
+    assert result.libs == ["algobsec"]
+    assert ("BAR", "1") in result.cppdefines
+    assert "FOO" in result.cppdefines
+    assert result.linkflags == ["-Wl,--gc-sections"]
+
+    flags = captured_as_build_flags(result, library_dir=tmp_path)
+    sep = os.sep
+    assert f"-Lsrc{sep}esp32" in flags
+    assert "-lalgobsec" in flags
+    assert "-DFOO" in flags
+    assert "-DBAR=1" in flags
+    assert "-Wl,--gc-sections" in flags
+
+
+def test_extra_script_libpath_relative_resolves_against_library_dir(
+    tmp_path, monkeypatch
+):
+    """Relative LIBPATH entries must resolve against ``library_dir``, not the
+    caller's CWD (the shim restores CWD before ``captured_as_build_flags``
+    runs)."""
+
+    (tmp_path / "lib" / "esp32").mkdir(parents=True)
+    elsewhere = tmp_path.parent / "not_the_library_dir"
+    elsewhere.mkdir(exist_ok=True)
+    monkeypatch.chdir(elsewhere)
+
+    result = ExtraScriptResult(libpath=["lib/esp32"])
+    flags = captured_as_build_flags(result, library_dir=tmp_path)
+
+    sep = os.sep
+    assert flags == [f"-Llib{sep}esp32"]
+
+
+def test_extra_script_libpath_absolute_outside_library_dir(tmp_path):
+
+    outside = tmp_path.parent / "system_lib"
+    outside.mkdir(exist_ok=True)
+    result = ExtraScriptResult(libpath=[str(outside)])
+
+    flags = captured_as_build_flags(result, library_dir=tmp_path)
+    assert flags == [f"-L{outside.resolve()}"]
+
+
+def test_extra_script_failure_returns_empty_result(tmp_path, caplog):
+
+    script = tmp_path / "broken.py"
+    script.write_text("raise RuntimeError('boom')\n")
+
+    with caplog.at_level("WARNING"):
+        result = run_extra_script(
+            script, library_dir=tmp_path, board_mcu="esp32", pio_platform="espressif32"
+        )
+
+    assert result.libpath == []
+    assert result.libs == []
+    assert "broken.py" in caplog.text
+
+
+def test_apply_extra_script_path_traversal_is_rejected(tmp_path):
+
+    library_dir = tmp_path / "lib"
+    library_dir.mkdir()
+    outside = tmp_path / "evil.py"
+    outside.write_text("env.Append(LIBS=['pwned'])\n")
+
+    c = IDFComponent("owner/name", "1.0", source=URLSource("http://dummy"))
+    c.path = library_dir
+    c.data = {"build": {"extraScript": "../evil.py"}}
+
+    with pytest.raises(EsphomeError, match="escapes the library directory"):
+        apply_extra_script(c, board_mcu=lambda: "esp32", pio_platform="espressif32")
+    # Nothing was folded into flags: the traversal was rejected before
+    # the script could run.
+    assert "flags" not in c.data["build"]
+
+
+def test_apply_extra_script_merges_into_existing_flags(tmp_path):
+
+    (tmp_path / "src").mkdir()
+    script = tmp_path / "extra.py"
+    script.write_text("env.Append(LIBS=['algobsec'])\n")
+
+    c = IDFComponent("owner/name", "1.0", source=URLSource("http://dummy"))
+    c.path = tmp_path
+    c.data = {"build": {"extraScript": "extra.py", "flags": ["-DEXISTING"]}}
+
+    apply_extra_script(c, board_mcu=lambda: "esp32", pio_platform="espressif32")
+
+    assert "-DEXISTING" in c.data["build"]["flags"]
+    assert "-lalgobsec" in c.data["build"]["flags"]
+
+
+def test_apply_extra_script_malformed_flags_raises(tmp_path) -> None:
+    """A null/dict build.flags fails naming the library instead of injecting
+    a non-string into the compiler command line."""
+
+    (tmp_path / "src").mkdir()
+    script = tmp_path / "extra.py"
+    script.write_text("env.Append(LIBS=['algobsec'])\n")
+
+    c = IDFComponent("owner/name", "1.0", source=URLSource("http://dummy"))
+    c.path = tmp_path
+    c.data = {"build": {"extraScript": "extra.py", "flags": None}}
+
+    with pytest.raises(EsphomeError, match="malformed build.flags"):
+        apply_extra_script(c, board_mcu=lambda: "esp32", pio_platform="espressif32")
+
+
+def test_apply_extra_script_callable_target_and_str_flags(tmp_path) -> None:
+    """The shared helper resolves the board_mcu callable lazily and normalizes
+    a string ``build.flags`` value into a list before extending it."""
+
+    (tmp_path / "src").mkdir()
+    script = tmp_path / "extra.py"
+    script.write_text("env.Append(LIBS=[env.get('BOARD_MCU')])\n")
+
+    c = IDFComponent("owner/name", "1.0", source=URLSource("http://dummy"))
+    c.path = tmp_path
+    c.data = {"build": {"extraScript": "extra.py", "flags": "-DBASE=1"}}
+
+    apply_extra_script(c, board_mcu=lambda: "esp8266", pio_platform="espressif8266")
+
+    assert c.data["build"]["flags"] == ["-DBASE=1", "-lesp8266"]
+
+
+def test_captured_nonstring_buckets_warn_and_skip(tmp_path, caplog) -> None:
+    """Non-string LIBS/LINKFLAGS/CPPFLAGS/LIBPATH entries (legal SCons
+    nodes) are skipped by name instead of stringified into garbage flags."""
+    (tmp_path / "src").mkdir()
+    script = tmp_path / "extra.py"
+    script.write_text(
+        "env.Append(LIBS=['m', 42], LINKFLAGS=['-Wl,-x', {'no': 1}], "
+        "CPPFLAGS=['-Os', 3.5], LIBPATH=['libs', 7])\n"
+    )
+    (tmp_path / "libs").mkdir()
+
+    c = IDFComponent("owner/name", "1.0", source=URLSource("http://dummy"))
+    c.path = tmp_path
+    c.data = {"build": {"extraScript": "extra.py"}}
+
+    apply_extra_script(c, board_mcu=lambda: "esp8266", pio_platform="espressif8266")
+
+    flags = c.data["build"]["flags"]
+    assert "-lm" in flags and "-Wl,-x" in flags and "-Os" in flags
+    assert not any("42" in f or "no" in f or "3.5" in f for f in flags)
+    assert "Ignoring unsupported LIBS entry 42" in caplog.text
+    assert "Ignoring unsupported LIBPATH entry 7" in caplog.text
+
+
+def test_captured_dict_cppdefines_warn_and_skip(tmp_path, caplog) -> None:
+    """A dict CPPDEFINES entry (legal SCons) must warn and skip; formatting
+    it blind would hand the compiler -D{'FOO': '1'} garbage."""
+    (tmp_path / "src").mkdir()
+    script = tmp_path / "extra.py"
+    script.write_text(
+        "env.Append(CPPDEFINES=[{'FOO': '1'}, ('BAR', 2), ['BAZ', 3], 'PLAIN'])\n"
+    )
+
+    c = IDFComponent("owner/name", "1.0", source=URLSource("http://dummy"))
+    c.path = tmp_path
+    c.data = {"build": {"extraScript": "extra.py"}}
+
+    apply_extra_script(c, board_mcu=lambda: "esp8266", pio_platform="espressif8266")
+
+    assert c.data["build"]["flags"] == ["-DBAR=2", "-DBAZ=3", "-DPLAIN"]
+    assert "Ignoring unsupported CPPDEFINES entry" in caplog.text
+
+
+def test_apply_extra_script_subscript_env_read(tmp_path) -> None:
+    """Scripts also read env["BOARD_MCU"]; the subscript form must work or
+    the broad handler discards every flag the script captured."""
+    (tmp_path / "src").mkdir()
+    script = tmp_path / "extra.py"
+    script.write_text("env.Append(LIBS=[env['BOARD_MCU']])\n")
+
+    c = IDFComponent("owner/name", "1.0", source=URLSource("http://dummy"))
+    c.path = tmp_path
+    c.data = {"build": {"extraScript": "extra.py"}}
+
+    apply_extra_script(c, board_mcu=lambda: "esp8266", pio_platform="espressif8266")
+
+    assert c.data["build"]["flags"] == ["-lesp8266"]
+
+
+def test_apply_extra_script_no_script_and_no_flags(tmp_path) -> None:
+
+    # No extraScript declared: nothing happens, the target is never resolved
+    c = IDFComponent("owner/name", "1.0", source=URLSource("http://dummy"))
+    c.path = tmp_path
+    c.data = {"build": {}}
+    apply_extra_script(
+        c,
+        board_mcu=lambda: pytest.fail("target resolved without a script"),
+        pio_platform="espressif8266",
+    )
+
+    # A script that captures nothing leaves the flags untouched
+    script = tmp_path / "noop.py"
+    script.write_text("pass\n")
+    c.data = {"build": {"extraScript": "noop.py"}}
+    apply_extra_script(c, board_mcu=lambda: "esp8266", pio_platform="espressif8266")
+    assert "flags" not in c.data["build"]
+
+
+def test_apply_extra_script_ignores_uncaptured_env_calls(tmp_path, caplog) -> None:
+    """Un-captured env vars and unsupported env methods are skipped but
+    diagnosable from the build log."""
+
+    caplog.set_level(logging.DEBUG)
+    script = tmp_path / "extra.py"
+    script.write_text(
+        "env.Replace(CC='clang')\nenv.Append(UNCAPTURED=['x'], LIBS='single')\n"
+    )
+    c = IDFComponent("owner/name", "1.0", source=URLSource("http://dummy"))
+    c.path = tmp_path
+    c.data = {"build": {"extraScript": "extra.py"}}
+    apply_extra_script(c, board_mcu=lambda: "esp8266", pio_platform="espressif8266")
+    assert c.data["build"]["flags"] == ["-lsingle"]
+    assert "env.Append(UNCAPTURED=...) is not captured" in caplog.text
+    assert "env.Replace(...) is not supported" in caplog.text
+
+
+def test_apply_extra_script_swallows_script_errors(tmp_path, caplog) -> None:
+    """A raising extra-script is best-effort: logged and skipped."""
+
+    script = tmp_path / "extra.py"
+    script.write_text("raise RuntimeError('boom')\n")
+    c = IDFComponent("owner/name", "1.0", source=URLSource("http://dummy"))
+    c.path = tmp_path
+    c.data = {"build": {"extraScript": "extra.py"}}
+    apply_extra_script(c, board_mcu=lambda: "esp8266", pio_platform="espressif8266")
+    assert "flags" not in c.data["build"]
+    assert "ignoring its output" in caplog.text
+
+
+def test_apply_extra_script_pio_platform(tmp_path) -> None:
+    """The backend's platform token is exposed to the script as PIOPLATFORM."""
+
+    script = tmp_path / "extra.py"
+    script.write_text("env.Append(LIBS=[env.get('PIOPLATFORM')])\n")
+    c = IDFComponent("owner/name", "1.0", source=URLSource("http://dummy"))
+    c.path = tmp_path
+    c.data = {"build": {"extraScript": "extra.py"}}
+    apply_extra_script(c, board_mcu=lambda: "esp8266", pio_platform="espressif8266")
+    assert c.data["build"]["flags"] == ["-lespressif8266"]
+
+
+def test_apply_extra_script_missing_script_raises(tmp_path) -> None:
+    """A declared but absent extraScript is a broken package and fails by
+    name, as it would under PlatformIO."""
+
+    c = IDFComponent("owner/name", "1.0", source=URLSource("http://dummy"))
+    c.path = tmp_path
+    c.data = {"build": {"extraScript": "nope.py"}}
+    with pytest.raises(EsphomeError, match="nope.py of library owner/name not found"):
+        apply_extra_script(c, board_mcu=lambda: "esp8266", pio_platform="espressif8266")
+
+
+def test_run_extra_script_failure_discards_partial_capture(tmp_path, caplog) -> None:
+    """A crashed script yields an empty result: half-applied flags could
+    build wrong-output firmware that links cleanly."""
+
+    script = tmp_path / "extra.py"
+    script.write_text("env.Append(LIBS=['algobsec'])\nraise RuntimeError('boom')\n")
+    result = run_extra_script(
+        script, library_dir=tmp_path, board_mcu="esp32", pio_platform="espressif32"
+    )
+    assert result.libs == []
+    assert "ignoring its output" in caplog.text
+
+
+def test_run_extra_script_syntax_error_is_best_effort(tmp_path, caplog) -> None:
+    """A vendored script that does not even compile warns and skips instead
+    of aborting the build."""
+
+    script = tmp_path / "extra.py"
+    script.write_text("def broken(:\n")
+    result = run_extra_script(
+        script, library_dir=tmp_path, board_mcu="esp32", pio_platform="espressif32"
+    )
+    assert result.libs == []
+    assert "ignoring its output" in caplog.text
+
+
+def test_unsupported_env_method_warns_once(caplog) -> None:
+    """Repeated calls to the same unsupported method warn only once."""
+
+    env = _FakeSConsEnv(
+        board_mcu="esp8266", pio_env="esphome_esp8266", pio_platform="espressif8266"
+    )
+    env.Replace(CC="clang")
+    env.Replace(CC="gcc")
+    assert caplog.text.count("env.Replace(...) is not supported") == 1
+
+
+def test_run_extra_script_sys_exit_is_best_effort(tmp_path, caplog) -> None:
+    """A nonzero sys.exit() in a vendored script must not kill the esphome
+    run, and its output is discarded."""
+
+    script = tmp_path / "extra.py"
+    script.write_text("import sys\nenv.Append(LIBS=['x'])\nsys.exit(3)\n")
+    result = run_extra_script(
+        script, library_dir=tmp_path, board_mcu="esp32", pio_platform="espressif32"
+    )
+    assert result.libs == []
+    assert "exited with status 3" in caplog.text
+
+
+def test_run_extra_script_sys_exit_zero_is_success(tmp_path, caplog) -> None:
+    """sys.exit(0) is a normal PlatformIO script ending: the capture is kept."""
+
+    script = tmp_path / "extra.py"
+    script.write_text("import sys\nenv.Append(LIBS=['algobsec'])\nsys.exit(0)\n")
+    result = run_extra_script(
+        script, library_dir=tmp_path, board_mcu="esp32", pio_platform="espressif32"
+    )
+    assert result.libs == ["algobsec"]
+    assert "ignoring its output" not in caplog.text
+
+
+def test_run_extra_script_unreadable_raises(tmp_path) -> None:
+    """An unreadable declared script is a broken package, like a missing one."""
+
+    script = tmp_path / "extra.py"
+    script.write_text("")
+    with (
+        patch("pathlib.Path.read_text", side_effect=OSError("denied")),
+        pytest.raises(EsphomeError, match="is unreadable"),
+    ):
+        run_extra_script(
+            script, library_dir=tmp_path, board_mcu="esp32", pio_platform="espressif32"
+        )
+
+
+def test_run_extra_script_bad_encoding_is_best_effort(tmp_path, caplog) -> None:
+    """Undecodable content warns and skips, like a SyntaxError."""
+
+    script = tmp_path / "extra.py"
+    script.write_bytes(b"\xff\xfe\x00bad")
+    result = run_extra_script(
+        script, library_dir=tmp_path, board_mcu="esp32", pio_platform="espressif32"
+    )
+    assert result.libs == []
+    assert "is not UTF-8" in caplog.text
+
+
+def test_uncaptured_append_key_warns_once(caplog) -> None:
+    """A loop of Appends to the same uncaptured key warns once."""
+
+    env = _FakeSConsEnv(
+        board_mcu="esp8266", pio_env="esphome_esp8266", pio_platform="espressif8266"
+    )
+    env.Append(CPPPATH=["a"])
+    env.Append(CPPPATH=["b"])
+    assert caplog.text.count("env.Append(CPPPATH=...) is not captured") == 1

tests/unit_tests/test_platformio_library.py

@@ -4,15 +4,18 @@ Covers the shared download/parse/resolve/dependency-walk paths in
 ``esphome.platformio.library`` directly (the ESP-IDF and Zephyr backends are
 exercised in their own test modules)."""
 
+from contextlib import contextmanager
 import json
 import logging
 from pathlib import Path
+from types import SimpleNamespace
 
 import pytest
 
 from esphome.core import EsphomeError, Library
 import esphome.platformio.library as lib
 from esphome.platformio.library import (
+    SOURCE_KIND_FOR_SUFFIX,
     ConvertedLibrary,
     GitSource,
     InvalidLibrary,
@@ -23,6 +26,8 @@ from esphome.platformio.library import (
     _resolve_registry_version,
     check_library_data,
     convert_libraries,
+    join_flag_args,
+    split_flag_entry,
 )
 
 
@@ -150,11 +155,30 @@ def test_localsource_download_returns_empty_build_dir(setup_core: Path) -> None:
     assert plain != out
 
 
+@contextmanager
+def caplog_at_info():
+    records: list[logging.LogRecord] = []
+    handler = logging.Handler()
+    handler.emit = records.append
+    logger = logging.getLogger("esphome.platformio.library")
+    logger.addHandler(handler)
+    # The level must actually admit INFO or the no-INFO assertions are vacuous
+    old_level = logger.level
+    logger.setLevel(logging.INFO)
+    try:
+        yield records
+    finally:
+        logger.setLevel(old_level)
+        logger.removeHandler(handler)
+
+
 def test_urlsource_download_extracts_then_reuses_marker(setup_core, monkeypatch):
     monkeypatch.setattr(lib, "rmdir", lambda path, msg="": None)
     dl_calls: list[list[str]] = []
     monkeypatch.setattr(
-        lib, "download_from_mirrors", lambda urls, headers, f: dl_calls.append(urls)
+        lib,
+        "download_from_mirrors",
+        lambda urls, headers, f, progress=None: dl_calls.append(urls),
     )
 
     def fake_extract(fileobj, path):
@@ -173,6 +197,12 @@ def test_urlsource_download_extracts_then_reuses_marker(setup_core, monkeypatch)
     assert out2 == out
     assert len(dl_calls) == 1
 
+    # A batch caller passes a tracker and owns the messaging; no per-file INFO
+    with caplog_at_info() as records:
+        src.download("mylib-batch", progress=lambda done: None)
+    assert len(dl_calls) == 2
+    assert not [r for r in records if "Downloading" in r.message]
+
 
 def test_resolve_registry_version_raises_without_pkg_file(monkeypatch):
     registry = lib._make_registry_client()
@@ -213,7 +243,7 @@ def _patch_registry_resolve(monkeypatch: pytest.MonkeyPatch) -> None:
 def _patch_download_with_manifests(monkeypatch, tmp_path, manifests, *, properties=()):
     """Fake ConvertedLibrary.download to materialize canned manifests on disk."""
 
-    def fake_download(self, force=False, salt="", namespace=""):
+    def fake_download(self, force=False, salt="", namespace="", progress=None):
         self.path = tmp_path / self.get_require_name()
         self.path.mkdir(parents=True, exist_ok=True)
         if self.name in properties:
@@ -225,6 +255,37 @@ def _patch_download_with_manifests(monkeypatch, tmp_path, manifests, *, properti
     _patch_registry_resolve(monkeypatch)
 
 
+def test_wave_requirement_growth_defers_the_superseded_download(tmp_path, monkeypatch):
+    """A's manifest constrains B while B sits in the same wave: B's
+    drain-time resolution is superseded, so its download defers to the
+    next wave instead of fetching a version that is immediately replaced."""
+    download_names: list[str] = []
+    manifests = {
+        "esphome/A": {
+            "name": "A",
+            "build": {},
+            "dependencies": {"esphome/B": ">=1.0"},
+        },
+        "esphome/B": {"name": "B", "build": {}},
+    }
+
+    def fake_download(self, force=False, salt="", namespace="", progress=None):
+        download_names.append(self.name)
+        self.path = tmp_path / self.get_require_name()
+        self.path.mkdir(parents=True, exist_ok=True)
+        (self.path / "library.json").write_text(json.dumps(manifests[self.name]))
+
+    monkeypatch.setattr(ConvertedLibrary, "download", fake_download)
+    _patch_registry_resolve(monkeypatch)
+    top = convert_libraries(
+        [Library("esphome/A", "1.0.0", None), Library("esphome/B", None, None)],
+        _backend(),
+    )
+    assert sorted(c.name for c in top) == ["esphome/A", "esphome/B"]
+    # B downloads exactly once, after its requirement set stabilized
+    assert download_names.count("esphome/B") == 1
+
+
 def test_convert_libraries_parses_library_properties(tmp_path, monkeypatch):
     # A manifest provided as library.properties (Arduino style) instead of
     # library.json must still be parsed and converted.
@@ -292,7 +353,11 @@ def _patch_download_without_manifest(
     calls: list[bool] = []
 
     def fake_download(
-        self: ConvertedLibrary, force: bool = False, salt: str = "", namespace: str = ""
+        self: ConvertedLibrary,
+        force: bool = False,
+        salt: str = "",
+        namespace: str = "",
+        progress=None,
     ) -> None:
         calls.append(force)
         self.path = tmp_path / self.get_require_name()
@@ -531,3 +596,330 @@ def test_convert_libraries_skips_incompatible_dependency(tmp_path, monkeypatch):
     top = convert_libraries([Library("esphome/A", "1.0.0", None)], _backend())
 
     assert top[0].dependencies == []
+
+
+def test_split_flag_entry_unbalanced_quote_is_clean() -> None:
+    """A malformed flags entry raises EsphomeError, not a raw ValueError."""
+
+    assert split_flag_entry('-DX="a b"', "library x") == ["-DX=a b"]
+    with pytest.raises(EsphomeError, match=r"Malformed build flag.*library x"):
+        split_flag_entry('-DX="unclosed', "library x")
+
+
+def test_join_flag_args_reglues_spaced_define() -> None:
+    """A spaced -D re-glues to its argument, as ParseFlags does."""
+
+    assert join_flag_args(["-D", "FOO=1", "-Os"], "x") == ["-DFOO=1", "-Os"]
+
+
+def test_join_flag_args_trailing_bare_flag_warns(
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+
+    assert join_flag_args(["-Os", "-l"], "library x") == ["-Os"]
+    assert "Ignoring trailing '-l'" in caplog.text
+
+
+def test_lex_build_flags_dangling_flag_does_not_cross_entries(
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    """Each entry is lexed independently, as ParseFlags does: a dangling -I
+    ending one entry warns instead of absorbing the next entry's first token."""
+    from esphome.platformio.library import lex_build_flags
+
+    assert lex_build_flags(["-Wall -I", "-DFOO=1"], "lib x") == ["-Wall", "-DFOO=1"]
+    assert "Ignoring trailing '-I'" in caplog.text
+
+
+def test_prefetch_wave_downloads_registry_archives_in_parallel(
+    setup_core, monkeypatch: pytest.MonkeyPatch, caplog: pytest.LogCaptureFixture
+) -> None:
+    """Registry archives in one wave download concurrently, deduped by URL;
+    git/local sources and failures are left to the sequential call."""
+    calls: list[str] = []
+
+    def fake_download(self, force=False, salt="", namespace="", progress=None):
+        calls.append(self.source.url)
+        if progress is not None:
+            progress(0)
+        if "boom" in self.source.url:
+            raise RuntimeError("boom")
+
+    monkeypatch.setattr(ConvertedLibrary, "download", fake_download)
+    monkeypatch.setattr(lib, "_content_lengths", lambda urls: [1] * len(urls))
+    wave = [
+        ("a", ConvertedLibrary("a", "1.0", URLSource("https://x/a.tar.gz"))),
+        ("b", ConvertedLibrary("b", "1.0", URLSource("https://x/b.tar.gz"))),
+        # Duplicate URL must prefetch once (two threads must never extract
+        # into the same cache directory)
+        ("b2", ConvertedLibrary("b2", "1.0", URLSource("https://x/b.tar.gz"))),
+        ("c", ConvertedLibrary("c", "1.0", URLSource("https://x/boom.tar.gz"))),
+        ("g", ConvertedLibrary("g", "*", lib.GitSource("https://x/g.git", None))),
+    ]
+    lib._prefetch_wave(wave, "", "idf")
+    assert sorted(calls) == [
+        "https://x/a.tar.gz",
+        "https://x/b.tar.gz",
+        "https://x/boom.tar.gz",
+    ]
+    # The failure surfaces at default verbosity, after the bar
+    assert "Prefetch of c failed (retrying sequentially)" in caplog.text
+
+
+def test_prefetch_wave_unknown_size_falls_back_to_sequential(
+    setup_core, monkeypatch: pytest.MonkeyPatch, caplog: pytest.LogCaptureFixture
+) -> None:
+    """Any unknown HEAD size skips the parallel prefetch entirely so the
+    sequential downloads keep their per-file bars."""
+
+    def fail_download(self, force=False, salt="", namespace="", progress=None):
+        raise AssertionError("prefetched despite unknown size")
+
+    monkeypatch.setattr(ConvertedLibrary, "download", fail_download)
+    monkeypatch.setattr(lib, "_content_lengths", lambda urls: [1, None])
+    wave = [
+        ("a", ConvertedLibrary("a", "1.0", URLSource("https://x/a.tar.gz"))),
+        ("b", ConvertedLibrary("b", "1.0", URLSource("https://x/b.tar.gz"))),
+    ]
+    caplog.set_level("DEBUG")
+    lib._prefetch_wave(wave, "", "idf")
+    # The culprit URL is named so the fallback is traceable
+    assert "No Content-Length for https://x/b.tar.gz" in caplog.text
+
+
+def test_raise_on_empty_arg_flags() -> None:
+    """A surviving bare flag means an empty glued argument; reject by name."""
+    with pytest.raises(EsphomeError, match=r"build_flags contain empty-argument"):
+        lib.raise_on_empty_arg_flags(["-DFOO", "-D", "-l"], "build_flags")
+    lib.raise_on_empty_arg_flags(["-DFOO", "-Iinc"], "build_flags")
+
+
+def test_content_lengths_head_requests(monkeypatch: pytest.MonkeyPatch) -> None:
+    """Sizes come from HEAD Content-Length; a failing HEAD reads as 0 so
+    the combined bar is skipped rather than wrong."""
+    import requests
+
+    def fake_head(url, timeout, allow_redirects):
+        if "bad" in url:
+            raise requests.ConnectionError("down")
+        if "gone" in url:
+            return SimpleNamespace(ok=False, status_code=404, headers={})
+        if "garbage" in url:
+            # A proxy/CDN doubling the header ("123, 123") or emitting junk
+            # must degrade to unknown, not ValueError the build
+            return SimpleNamespace(ok=True, headers={"content-length": "123, 123"})
+        return SimpleNamespace(ok=True, headers={"content-length": "123"})
+
+    monkeypatch.setattr(
+        lib.requests if hasattr(lib, "requests") else requests, "head", fake_head
+    )
+    # None marks an unknown size (probe failure or non-2xx), distinct
+    # from a genuine zero
+    assert lib._content_lengths(
+        ["https://x/a", "https://x/bad", "https://x/gone", "https://x/garbage"]
+    ) == [
+        123,
+        None,
+        None,
+        None,
+    ]
+
+
+def test_prefetch_wave_cache_probe_failure_still_prefetches(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """The cache probe is best-effort; a failing probe prefetches anyway."""
+    calls: list[str] = []
+    monkeypatch.setattr(
+        ConvertedLibrary,
+        "download",
+        lambda self, **kw: calls.append(self.source.url),
+    )
+    monkeypatch.setattr(
+        URLSource,
+        "is_cached",
+        lambda self, *a, **kw: (_ for _ in ()).throw(RuntimeError("no core")),
+    )
+    monkeypatch.setattr(lib, "_content_lengths", lambda urls: [1] * len(urls))
+    wave = [
+        ("a", ConvertedLibrary("a", "1.0", URLSource("https://x/a.tar.gz"))),
+        ("b", ConvertedLibrary("b", "1.0", URLSource("https://x/b.tar.gz"))),
+    ]
+    lib._prefetch_wave(wave, "", "idf")
+    assert sorted(calls) == ["https://x/a.tar.gz", "https://x/b.tar.gz"]
+
+
+def test_prefetch_wave_warm_cache_is_silent(
+    setup_core, monkeypatch: pytest.MonkeyPatch, caplog: pytest.LogCaptureFixture
+) -> None:
+    """Already-extracted archives download nothing; a warm build must not
+    print a Downloading line or draw a bar."""
+    monkeypatch.setattr(
+        ConvertedLibrary,
+        "download",
+        lambda self, **kw: (_ for _ in ()).throw(AssertionError("downloaded")),
+    )
+    wave = []
+    for name in ("a", "b", "c"):
+        comp = ConvertedLibrary(name, "1.0", URLSource(f"https://x/{name}.tar.gz"))
+        marker_dir = comp.source._cache_dir(comp.get_sanitized_name(), "", "idf")
+        marker_dir.mkdir(parents=True)
+        (marker_dir / ".esphome_extracted").touch()
+        wave.append((name, comp))
+    lib._prefetch_wave(wave, "", "idf")
+    assert "Downloading" not in caplog.text
+
+
+def test_prefetch_wave_single_archive_skips_the_pool(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """One archive gains nothing from a pool; the sequential call keeps its
+    progress bar."""
+    monkeypatch.setattr(
+        ConvertedLibrary,
+        "download",
+        lambda self, **kw: (_ for _ in ()).throw(AssertionError("prefetched")),
+    )
+    lib._prefetch_wave(
+        [("a", ConvertedLibrary("a", "1.0", URLSource("https://x/a.tar.gz")))],
+        "",
+        "idf",
+    )
+
+
+def test_normalize_dependencies_forms(caplog) -> None:
+    """Every PIO-legal spelling normalizes; unrecognizable entries warn."""
+    from esphome.platformio.library import normalize_dependencies
+
+    assert normalize_dependencies(
+        ["Wire", {"name": "SPI"}, 5, "", {"version": "1.0"}], "libx"
+    ) == [
+        {"name": "Wire"},
+        {"name": "SPI"},
+    ]
+    # The int, the empty string, and the nameless dict all warn
+    assert caplog.text.count("unrecognized dependency entry") == 3
+    # A plain string is names, never iterated into characters
+    assert normalize_dependencies("Wire, SPI") == [
+        {"name": "Wire"},
+        {"name": "SPI"},
+    ]
+    assert normalize_dependencies("Wire") == [{"name": "Wire"}]
+    # A non-iterable value fails by manifest name, never a bare TypeError
+    assert normalize_dependencies(5, "libx") == []
+    assert "Ignoring unrecognized dependencies 5 of libx" in caplog.text
+    # The dict-shorthand form validates names like the list form: an empty
+    # key and a spec overriding name with a non-string both warn and drop
+    assert normalize_dependencies(
+        {"": "1.0", "Wire": {"name": 123, "version": "1.0"}, "SPI": "*"}, "libx"
+    ) == [{"name": "SPI", "owner": None, "version": "*"}]
+    assert caplog.text.count("unrecognized dependency entry") == 5
+    # A container or numeric version would raise from set.add() or fail
+    # opaquely in the registry; both spellings warn and drop
+    assert normalize_dependencies({"Foo": ["1.0", "2.0"]}, "libx") == []
+    assert normalize_dependencies([{"name": "Foo", "version": 1}], "libx") == []
+    assert caplog.text.count("unrecognized dependency entry") == 7
+
+
+@pytest.mark.parametrize(
+    "manifest", [["not", "a", "manifest"], {"name": "A", "build": "src"}]
+)
+def test_convert_libraries_malformed_manifest_raises(
+    tmp_path, monkeypatch, manifest
+) -> None:
+    """A manifest without the expected dict shape fails by library name
+    before any backend dereferences data/build."""
+    _patch_download_with_manifests(monkeypatch, tmp_path, {"esphome/A": manifest})
+    with pytest.raises(EsphomeError, match="has a malformed manifest"):
+        convert_libraries([Library("esphome/A", None, None)], _backend())
+
+
+def test_walk_warns_for_properties_only_depends(
+    tmp_path, monkeypatch, caplog: pytest.LogCaptureFixture
+) -> None:
+    """A manifest declaring dependencies only as library.properties depends=
+    warns in the shared walk, so every backend reports the drop."""
+    _patch_download_with_manifests(
+        monkeypatch,
+        tmp_path,
+        {"esphome/A": "name=A\nversion=1.0\ndepends=Wire, SPI\n"},
+        properties=("esphome/A",),
+    )
+    caplog.set_level("INFO")
+    convert_libraries([Library("esphome/A", "1.0.0", None)], _backend())
+    assert "declares dependencies via library.properties" in caplog.text
+
+
+def test_walk_warns_for_nonplatform_invalid_library(
+    tmp_path, monkeypatch, caplog: pytest.LogCaptureFixture
+) -> None:
+    """A dependency dropped for any cause other than the routine platform
+    filter is visible in every backend."""
+    _patch_download_with_manifests(
+        monkeypatch,
+        tmp_path,
+        {"esphome/A": {"name": "A", "dependencies": [{"name": "B", "version": "1.0"}]}},
+    )
+    calls = {"n": 0}
+    real = lib.check_library_data
+
+    def flaky(data, platform, framework):
+        calls["n"] += 1
+        if calls["n"] > 1:
+            raise InvalidLibrary("manifest is corrupt")
+        return real(data, platform, framework)
+
+    monkeypatch.setattr(lib, "check_library_data", flaky)
+    convert_libraries([Library("esphome/A", None, None)], _backend())
+    assert "Skipping dependency B of esphome/A: manifest is corrupt" in caplog.text
+
+
+def test_convert_libraries_warns_for_nonplatform_invalid_dependency_component(
+    tmp_path, monkeypatch, caplog: pytest.LogCaptureFixture
+) -> None:
+    """A dependency component dropped for any cause other than the platform
+    filter warns; only the routine cross-platform skip stays at debug."""
+    _patch_download_with_manifests(
+        monkeypatch,
+        tmp_path,
+        {
+            "esphome/A": {
+                "name": "A",
+                "dependencies": [{"name": "C", "owner": "esphome", "version": "1.0"}],
+            },
+            "esphome/C": {"name": "C"},
+        },
+    )
+    real = lib.check_library_data
+
+    def flaky(data, platform, framework):
+        # Fail only on C's resolved manifest, not on A's dependency entry
+        if data.get("name") == "C" and "version" not in data:
+            raise InvalidLibrary("manifest is corrupt")
+        return real(data, platform, framework)
+
+    monkeypatch.setattr(lib, "check_library_data", flaky)
+    convert_libraries([Library("esphome/A", "1.0.0", None)], _backend())
+    assert "manifest is corrupt" in caplog.text
+    assert "Skipping dependency" in caplog.text
+
+
+def test_split_flag_entry_non_string_is_clean() -> None:
+    """A dict or number from a third-party manifest fails naming the entry,
+    not with an opaque shlex traceback."""
+
+    with pytest.raises(EsphomeError, match="Malformed build flag"):
+        split_flag_entry({"esp32": ["-DX"]}, "lib x")
+    with pytest.raises(EsphomeError, match="Malformed build flag 5"):
+        split_flag_entry(5, "lib x")
+
+
+def test_source_kind_map_shape() -> None:
+    """The kind values the native compile rules key on, and the deliberate
+    AS/ASPP merge (.s and .S both map to asm)."""
+
+    assert set(SOURCE_KIND_FOR_SUFFIX.values()) == {"c", "cxx", "asm"}
+    assert SOURCE_KIND_FOR_SUFFIX[".s"] == "asm"
+    assert SOURCE_KIND_FOR_SUFFIX[".S"] == "asm"
+    assert SOURCE_KIND_FOR_SUFFIX[".c"] == "c"
+    assert SOURCE_KIND_FOR_SUFFIX[".cpp"] == "cxx"

tests/unit_tests/test_platformio_registry.py

@@ -0,0 +1,650 @@
+"""Tests for esphome.platformio.registry (PIO-registry package installs)."""
+
+from __future__ import annotations
+
+from contextlib import contextmanager
+import json
+import os
+from pathlib import Path
+from unittest.mock import patch
+
+import pytest
+
+from esphome.core import EsphomeError
+from esphome.platformio import registry
+
+
+@pytest.mark.parametrize(
+    ("system", "machine", "expected"),
+    [
+        ("Darwin", "arm64", "darwin_arm64"),
+        ("Darwin", "x86_64", "darwin_x86_64"),
+        ("Windows", "AMD64", "windows_amd64"),
+        # Deviation from upstream: auto-mapped to the emulated-x86 packages
+        ("Windows", "ARM64", "windows_amd64"),
+        ("Windows", "x86", "windows_x86"),
+        ("Linux", "x86_64", "linux_x86_64"),
+        ("Linux", "aarch64", "linux_aarch64"),
+        ("Linux", "i686", "linux_i686"),
+        ("Linux", "armv7l", "linux_armv7l"),
+        # Unknown hosts pass through like upstream; the registry lookup
+        # then fails naming the tag
+        ("FreeBSD", "amd64", "freebsd_amd64"),
+    ],
+)
+def test_get_systype(system: str, machine: str, expected: str) -> None:
+    with (
+        patch("platform.system", return_value=system),
+        patch("platform.machine", return_value=machine),
+        patch("platform.architecture", return_value=("64bit", "")),
+    ):
+        assert registry.get_systype() == expected
+
+
+def test_get_systype_env_override() -> None:
+    """PLATFORMIO_SYSTEM_TYPE wins, exactly as in upstream get_systype()."""
+    with patch.dict(os.environ, {"PLATFORMIO_SYSTEM_TYPE": "windows_amd64"}):
+        assert registry.get_systype() == "windows_amd64"
+
+
+def test_get_systype_aarch64_32bit_userland() -> None:
+    """A 32-bit userland on a 64-bit arm kernel gets armv7l binaries."""
+    with (
+        patch("platform.system", return_value="Linux"),
+        patch("platform.machine", return_value="aarch64"),
+        patch("platform.architecture", return_value=("32bit", "")),
+    ):
+        assert registry.get_systype() == "linux_armv7l"
+
+
+def test_get_systype_windows_empty_machine() -> None:
+    """An empty machine string falls back to the architecture bits."""
+    with (
+        patch("platform.system", return_value="Windows"),
+        patch("platform.machine", return_value=""),
+        patch("platform.architecture", return_value=("64bit", "")),
+    ):
+        assert registry.get_systype() == "windows_amd64"
+
+
+def _registry_response(files: list[dict]):
+    """Patch the shared downloader to serve a canned registry response."""
+    payload = {"versions": [{"name": "1.0.0", "files": files}]}
+
+    def fake_download(mirrors: list[str], substitutions: dict, target) -> str:
+        target.write(json.dumps(payload).encode())
+        return mirrors[0].format(**substitutions)
+
+    return patch.object(registry, "download_from_mirrors", side_effect=fake_download)
+
+
+def test_registry_download_uses_shared_downloader() -> None:
+    """The metadata fetch delegates its retries and error reporting to
+    download_from_mirrors; failures surface unchanged."""
+    with (
+        patch.object(
+            registry,
+            "download_from_mirrors",
+            side_effect=EsphomeError("Failed to download from all mirrors"),
+        ) as mock_download,
+        pytest.raises(EsphomeError, match="Failed to download from all mirrors"),
+    ):
+        registry.registry_download("pkg", "1.0.0")
+    (mirrors, substitutions, _), _ = mock_download.call_args
+    assert mirrors == [registry._REGISTRY_URL]
+    assert substitutions == {"package": "pkg"}
+
+
+def test_registry_download_invalid_json_is_clean() -> None:
+    def fake_download(mirrors: list[str], substitutions: dict, target) -> str:
+        target.write(b"<html>not json</html>")
+        return "http://x"
+
+    with (
+        patch.object(registry, "download_from_mirrors", side_effect=fake_download),
+        pytest.raises(EsphomeError, match="invalid JSON"),
+    ):
+        registry.registry_download("pkg", "1.0.0")
+
+
+def test_registry_download_matches_system() -> None:
+    with (
+        _registry_response(
+            [
+                {"system": ["windows_amd64"], "download_url": "http://x/win"},
+                {
+                    "system": ["linux_x86_64"],
+                    "download_url": "http://x/linux",
+                    "checksum": {"sha256": "abc123"},
+                    "size": 42,
+                },
+            ]
+        ),
+        patch.object(registry, "get_systype", return_value="linux_x86_64"),
+    ):
+        assert registry.registry_download("pkg", "1.0.0") == (
+            "http://x/linux",
+            "abc123",
+            42,
+        )
+
+
+def test_registry_download_bare_string_system() -> None:
+    """A bare-string system tag is an exact match, not a substring test."""
+    with (
+        _registry_response(
+            [
+                {"system": "linux_x86", "download_url": "http://x/x86"},
+                {
+                    "system": "linux_x86_64",
+                    "download_url": "http://x/x86_64",
+                    "checksum": {"sha256": "abc"},
+                },
+            ]
+        ),
+        patch.object(registry, "get_systype", return_value="linux_x86_64"),
+    ):
+        assert registry.registry_download("pkg", "1.0.0")[0] == "http://x/x86_64"
+
+
+def test_registry_download_wildcard_system() -> None:
+    with _registry_response(
+        [
+            {
+                "system": "*",
+                "download_url": "http://x/any",
+                "checksum": {"sha256": "abc"},
+                "size": 7,
+            }
+        ]
+    ):
+        assert registry.registry_download("pkg", "1.0.0") == (
+            "http://x/any",
+            "abc",
+            7,
+        )
+
+
+def test_registry_download_missing_checksum_raises() -> None:
+    """An unverifiable archive is refused, never silently extracted."""
+    with (
+        _registry_response([{"system": "*", "download_url": "http://x/any"}]),
+        pytest.raises(EsphomeError, match="no sha256"),
+    ):
+        registry.registry_download("pkg", "1.0.0")
+
+
+def test_registry_download_no_system_match() -> None:
+    with (
+        _registry_response(
+            [{"system": ["windows_amd64"], "download_url": "http://x/win"}]
+        ),
+        patch.object(registry, "get_systype", return_value="linux_x86_64"),
+        pytest.raises(EsphomeError, match="No pkg 1.0.0 build"),
+    ):
+        registry.registry_download("pkg", "1.0.0")
+
+
+def test_registry_download_version_not_found() -> None:
+    def fake_download(mirrors: list[str], substitutions: dict, target) -> str:
+        target.write(
+            json.dumps({"versions": [{"name": "2.0.0", "files": []}]}).encode()
+        )
+        return "http://x"
+
+    with (
+        patch.object(registry, "download_from_mirrors", side_effect=fake_download),
+        pytest.raises(EsphomeError, match="not found"),
+    ):
+        registry.registry_download("pkg", "1.0.0")
+
+
+def test_install_package_skips_when_marker_exists(tmp_path: Path) -> None:
+    dest = tmp_path / "pkg"
+    (dest / "payload").mkdir(parents=True)
+    (dest / ".esphome_extracted").touch()
+    with patch.object(registry, "download_from_mirrors") as mock_download:
+        registry.install_package(
+            "pkg", "1.0.0", dest, [], tmp_path / "dl", expect=("payload",)
+        )
+    mock_download.assert_not_called()
+
+
+def test_install_package_marker_hit_rechecks_layout(tmp_path: Path) -> None:
+    """A marked install that later lost files fails by name instead of
+    surfacing as an opaque toolchain error."""
+    dest = tmp_path / "pkg"
+    dest.mkdir()
+    (dest / ".esphome_extracted").touch()
+    with pytest.raises(EsphomeError, match="missing the expected payload"):
+        registry.install_package(
+            "pkg", "1.0.0", dest, [], tmp_path / "dl", expect=("payload",)
+        )
+
+
+def test_install_package_downloads_via_mirrors(tmp_path: Path) -> None:
+    dest = tmp_path / "pkg"
+    mirrors = ["http://mirror/{VERSION}/{SYSTEM}.tar.gz"]
+    with (
+        patch.object(registry, "download_from_mirrors") as mock_download,
+        patch.object(registry, "archive_extract_all") as mock_extract,
+        patch.object(registry, "get_systype", return_value="linux_x86_64"),
+    ):
+        # Extraction is expected to create the directory
+        mock_extract.side_effect = lambda *_a, **_kw: (dest / "payload").mkdir(
+            parents=True
+        )
+        registry.install_package(
+            "pkg", "1.0.0", dest, mirrors, tmp_path / "dl", expect=("payload",)
+        )
+    assert mock_download.call_args[0][0] is mirrors
+    assert mock_download.call_args[0][1] == {
+        "VERSION": "1.0.0",
+        "SYSTEM": "linux_x86_64",
+    }
+    assert (dest / ".esphome_extracted").is_file()
+
+
+def test_install_package_downloads_via_registry(tmp_path: Path) -> None:
+    """The registry path downloads with the registry's sha256 and size."""
+    dest = tmp_path / "pkg"
+    with (
+        patch.object(registry, "download_with_resume") as mock_download,
+        patch.object(registry, "archive_extract_all") as mock_extract,
+        patch.object(
+            registry,
+            "registry_download",
+            return_value=("http://x/pkg.tar.gz", "abc123", 42),
+        ),
+    ):
+        mock_extract.side_effect = lambda *_a, **_kw: (dest / "payload").mkdir(
+            parents=True
+        )
+        registry.install_package(
+            "pkg", "1.0.0", dest, [], tmp_path / "dl", expect=("payload",)
+        )
+    assert mock_download.call_args[0][0] == "http://x/pkg.tar.gz"
+    assert mock_download.call_args[1] == {"sha256": "abc123", "size": 42}
+
+
+def test_install_package_validates_expected_layout(tmp_path: Path) -> None:
+    """The success marker is only written when the extracted tree is usable."""
+    dest = tmp_path / "pkg"
+    with (
+        patch.object(registry, "download_from_mirrors"),
+        patch.object(registry, "archive_extract_all") as mock_extract,
+        patch.object(registry, "get_systype", return_value="linux_x86_64"),
+    ):
+        mock_extract.side_effect = lambda *_a, **_kw: (dest / "bin").mkdir(parents=True)
+        registry.install_package(
+            "pkg", "1.0.0", dest, ["http://m"], tmp_path / "dl", expect=("bin",)
+        )
+    assert (dest / ".esphome_extracted").is_file()
+
+
+def test_install_package_unexpected_layout_raises(tmp_path: Path) -> None:
+    dest = tmp_path / "pkg"
+    with (
+        patch.object(registry, "download_from_mirrors"),
+        patch.object(registry, "archive_extract_all") as mock_extract,
+        patch.object(registry, "get_systype", return_value="linux_x86_64"),
+        pytest.raises(EsphomeError, match="missing the expected bin"),
+    ):
+        mock_extract.side_effect = lambda *_a, **_kw: (dest / "payload").mkdir(
+            parents=True
+        )
+        registry.install_package(
+            "pkg", "1.0.0", dest, ["http://m"], tmp_path / "dl", expect=("bin",)
+        )
+    assert not (dest / ".esphome_extracted").exists()
+
+
+def test_install_package_marker_rechecked_under_lock(tmp_path: Path) -> None:
+    """A concurrent install finishing while we wait for the lock is detected."""
+    dest = tmp_path / "pkg"
+    marker = dest / ".esphome_extracted"
+
+    @contextmanager
+    def _fake_lock(*_a, **_kw):
+        dest.mkdir(parents=True, exist_ok=True)
+        marker.touch()
+        yield
+
+    with (
+        patch("filelock.FileLock", _fake_lock),
+        patch.object(registry, "download_from_mirrors") as mock_download,
+        patch.object(registry, "rmdir") as mock_rmdir,
+    ):
+        registry.install_package(
+            "pkg", "1.0.0", dest, ["http://m"], tmp_path / "dl", expect=("payload",)
+        )
+    mock_download.assert_not_called()
+    mock_rmdir.assert_not_called()
+
+
+def test_install_package_uses_hard_lock(tmp_path: Path) -> None:
+    """The install lock must never degrade to a soft (existence) lock."""
+    dest = tmp_path / "pkg"
+    with (
+        patch("filelock.FileLock") as mock_lock,
+        patch.object(registry, "download_from_mirrors"),
+        patch.object(registry, "archive_extract_all") as mock_extract,
+        patch.object(registry, "get_systype", return_value="linux_x86_64"),
+    ):
+        mock_extract.side_effect = lambda *_a, **_kw: (dest / "payload").mkdir(
+            parents=True, exist_ok=True
+        )
+        registry.install_package(
+            "pkg", "1.0.0", dest, ["http://m"], tmp_path / "dl", expect=("payload",)
+        )
+    assert mock_lock.call_args.kwargs["fallback_to_soft"] is False
+
+
+def test_registry_download_empty_system_list_does_not_match() -> None:
+    """An explicitly empty system list must not act as a wildcard."""
+    with (
+        _registry_response([{"system": [], "download_url": "http://x/any"}]),
+        patch.object(registry, "get_systype", return_value="linux_x86_64"),
+        pytest.raises(EsphomeError, match="No pkg 1.0.0 build"),
+    ):
+        registry.registry_download("pkg", "1.0.0")
+
+
+def test_registry_download_unexpected_payload_is_named() -> None:
+    """An error envelope without a versions list is not 'version not found'."""
+
+    def fake_download(mirrors: list[str], substitutions: dict, target) -> str:
+        target.write(json.dumps({"message": "rate limited"}).encode())
+        return "http://x"
+
+    with (
+        patch.object(registry, "download_from_mirrors", side_effect=fake_download),
+        pytest.raises(EsphomeError, match="Unexpected package registry response"),
+    ):
+        registry.registry_download("pkg", "1.0.0")
+
+
+def test_registry_download_missing_system_key_matches_any() -> None:
+    """A file with no system key at all serves every host."""
+    with _registry_response(
+        [{"download_url": "http://x/any", "checksum": {"sha256": "abc"}, "size": 1}]
+    ):
+        assert registry.registry_download("pkg", "1.0.0") == ("http://x/any", "abc", 1)
+
+
+def test_registry_download_missing_files_list_is_named() -> None:
+    """A version entry without a files list is an unexpected payload, not a
+    missing platform build."""
+    with (
+        _registry_response(None),
+        pytest.raises(EsphomeError, match="Unexpected package registry response"),
+    ):
+        registry.registry_download("pkg", "1.0.0")
+
+
+def test_registry_download_missing_download_url_is_named() -> None:
+    with (
+        _registry_response([{"system": "*", "checksum": {"sha256": "abc"}, "size": 1}]),
+        pytest.raises(EsphomeError, match="no download URL"),
+    ):
+        registry.registry_download("pkg", "1.0.0")
+
+
+def test_install_package_empty_expect_rejected(tmp_path: Path) -> None:
+    """Layout validation is the only guard before marker.touch(), so an
+    empty expect is a caller bug, not a lenient install."""
+    with pytest.raises(ValueError, match="non-empty expect"):
+        registry.install_package(
+            "pkg", "1.0.0", tmp_path / "pkg", [], tmp_path / "dl", expect=()
+        )
+
+
+def test_registry_download_non_dict_version_entry_is_named() -> None:
+    """A versions list of bare strings is an unexpected payload, not an
+    AttributeError traceback."""
+
+    def fake_download(mirrors: list[str], substitutions: dict, target) -> str:
+        target.write(json.dumps({"versions": ["1.0.0", "2.0.0"]}).encode())
+        return "http://x"
+
+    with (
+        patch.object(registry, "download_from_mirrors", side_effect=fake_download),
+        pytest.raises(EsphomeError, match="Unexpected package registry response"),
+    ):
+        registry.registry_download("pkg", "1.0.0")
+
+
+def test_registry_download_non_dict_file_entry_is_named() -> None:
+    def fake_download(mirrors: list[str], substitutions: dict, target) -> str:
+        target.write(
+            json.dumps(
+                {"versions": [{"name": "1.0.0", "files": ["a.tar.gz"]}]}
+            ).encode()
+        )
+        return "http://x"
+
+    with (
+        patch.object(registry, "download_from_mirrors", side_effect=fake_download),
+        pytest.raises(EsphomeError, match="Unexpected package registry response"),
+    ):
+        registry.registry_download("pkg", "1.0.0")
+
+
+def test_registry_download_non_dict_payload_is_named() -> None:
+    """A JSON array answer is an unexpected payload at the outermost level."""
+
+    def fake_download(mirrors: list[str], substitutions: dict, target) -> str:
+        target.write(json.dumps(["1.0.0"]).encode())
+        return "http://x"
+
+    with (
+        patch.object(registry, "download_from_mirrors", side_effect=fake_download),
+        pytest.raises(EsphomeError, match="Unexpected package registry response"),
+    ):
+        registry.registry_download("pkg", "1.0.0")
+
+
+def test_registry_download_non_list_system_is_named() -> None:
+    """A system field that is neither missing, str, nor list is an
+    unexpected payload, not a TypeError from the ``in`` test."""
+    with (
+        _registry_response([{"system": 5, "checksum": {"sha256": "abc"}, "size": 1}]),
+        pytest.raises(EsphomeError, match="Unexpected package registry response"),
+    ):
+        registry.registry_download("pkg", "1.0.0")
+
+
+def _resolve_for(sizes: dict[str, int | None]):
+    def resolve(name: str, version: str):
+        size = sizes[name]
+        if size == -1:
+            raise EsphomeError("registry down")
+        return (f"http://x/{name}.tar.gz", "abc123", size)
+
+    return resolve
+
+
+def test_prefetch_packages_downloads_pending_in_parallel(tmp_path: Path) -> None:
+    """Two uninstalled packages download together under one combined bar,
+    with the registry's sha256 and size and a batch progress tracker."""
+    with (
+        patch.object(registry, "download_with_resume") as mock_download,
+        patch.object(
+            registry, "registry_download", side_effect=_resolve_for({"a": 10, "b": 20})
+        ),
+    ):
+        registry.prefetch_packages(
+            [
+                ("a", "1.0", tmp_path / "a", []),
+                ("b", "2.0", tmp_path / "b", []),
+            ],
+            tmp_path / "dl",
+        )
+    assert mock_download.call_count == 2
+    # Locking makes worker completion order nondeterministic
+    calls = sorted(mock_download.call_args_list, key=lambda c: c[0][0])
+    for call, (name, version, size) in zip(
+        calls, [("a", "1.0", 10), ("b", "2.0", 20)], strict=True
+    ):
+        assert call[0][0] == f"http://x/{name}.tar.gz"
+        assert call[0][1] == tmp_path / "dl" / f"{name}-{version}"
+        assert call[1]["sha256"] == "abc123"
+        assert call[1]["size"] == size
+        assert callable(call[1]["progress"])
+
+
+def test_prefetch_packages_dedupes_duplicate_entries(tmp_path: Path) -> None:
+    """Duplicate (name, version) entries would race each other between two
+    workers; only one survives (and one is too few to parallelize)."""
+    with (
+        patch.object(registry, "download_with_resume") as mock_download,
+        patch.object(
+            registry, "registry_download", side_effect=_resolve_for({"a": 10})
+        ),
+    ):
+        registry.prefetch_packages(
+            [
+                ("a", "1.0", tmp_path / "a", []),
+                ("a", "1.0", tmp_path / "a", []),
+            ],
+            tmp_path / "dl",
+        )
+    mock_download.assert_not_called()
+
+
+def test_prefetch_packages_single_pending_skips(tmp_path: Path) -> None:
+    """One pending package has nothing to parallelize; the sequential
+    install keeps its own bar."""
+    marker_dest = tmp_path / "a"
+    marker_dest.mkdir()
+    (marker_dest / ".esphome_extracted").touch()
+    with (
+        patch.object(registry, "download_with_resume") as mock_download,
+        patch.object(
+            registry, "registry_download", side_effect=_resolve_for({"b": 20})
+        ),
+    ):
+        registry.prefetch_packages(
+            [
+                ("a", "1.0", marker_dest, []),
+                ("b", "2.0", tmp_path / "b", []),
+            ],
+            tmp_path / "dl",
+        )
+    mock_download.assert_not_called()
+
+
+def test_prefetch_packages_mirror_and_sizeless_stay_sequential(
+    tmp_path: Path,
+) -> None:
+    """Mirror overrides and size-less registry entries are left to the
+    sequential path so its per-file bars stay trustworthy."""
+    with (
+        patch.object(registry, "download_with_resume") as mock_download,
+        patch.object(
+            registry,
+            "registry_download",
+            side_effect=_resolve_for({"b": None, "c": 30}),
+        ),
+    ):
+        registry.prefetch_packages(
+            [
+                ("a", "1.0", tmp_path / "a", ["http://mirror/{VERSION}"]),
+                ("b", "2.0", tmp_path / "b", []),
+                ("c", "3.0", tmp_path / "c", []),
+            ],
+            tmp_path / "dl",
+        )
+    mock_download.assert_not_called()
+
+
+def test_prefetch_packages_resolve_failure_defers_to_install(
+    tmp_path: Path, caplog: pytest.LogCaptureFixture
+) -> None:
+    """A registry failure only skips the prefetch; install_package reports
+    the real error with context."""
+    caplog.set_level("DEBUG")
+    with (
+        patch.object(registry, "download_with_resume") as mock_download,
+        patch.object(
+            registry, "registry_download", side_effect=_resolve_for({"a": -1, "b": 20})
+        ),
+    ):
+        registry.prefetch_packages(
+            [
+                ("a", "1.0", tmp_path / "a", []),
+                ("b", "2.0", tmp_path / "b", []),
+            ],
+            tmp_path / "dl",
+        )
+    mock_download.assert_not_called()
+    assert "Prefetch resolve for a failed" in caplog.text
+
+
+def test_prefetch_packages_complete_archive_skipped(tmp_path: Path) -> None:
+    """An archive already fully downloaded is not re-fetched."""
+    dl = tmp_path / "dl"
+    dl.mkdir()
+    (dl / "a-1.0").write_bytes(b"x" * 10)
+    with (
+        patch.object(registry, "download_with_resume") as mock_download,
+        patch.object(
+            registry, "registry_download", side_effect=_resolve_for({"a": 10, "b": 20})
+        ),
+    ):
+        registry.prefetch_packages(
+            [
+                ("a", "1.0", tmp_path / "a", []),
+                ("b", "2.0", tmp_path / "b", []),
+            ],
+            dl,
+        )
+    mock_download.assert_not_called()
+
+
+def test_prefetch_packages_download_failure_is_debug(
+    tmp_path: Path, caplog: pytest.LogCaptureFixture
+) -> None:
+    """A failed prefetch download is logged and left for install_package."""
+    caplog.set_level("DEBUG")
+    with (
+        patch.object(
+            registry, "download_with_resume", side_effect=OSError("boom")
+        ) as mock_download,
+        patch.object(
+            registry, "registry_download", side_effect=_resolve_for({"a": 10, "b": 20})
+        ),
+    ):
+        registry.prefetch_packages(
+            [
+                ("a", "1.0", tmp_path / "a", []),
+                ("b", "2.0", tmp_path / "b", []),
+            ],
+            tmp_path / "dl",
+        )
+    assert mock_download.call_count == 2
+    assert "Prefetch of a failed" in caplog.text
+    assert "Prefetch of b failed" in caplog.text
+
+
+def test_prefetch_packages_unexpected_failure_warns(
+    tmp_path: Path, caplog: pytest.LogCaptureFixture
+) -> None:
+    """A programming error (not a download failure) surfaces at WARNING
+    instead of becoming a permanent silent no-op."""
+    with (
+        patch.object(
+            registry, "download_with_resume", side_effect=TypeError("bad call")
+        ),
+        patch.object(
+            registry, "registry_download", side_effect=_resolve_for({"a": 10, "b": 20})
+        ),
+    ):
+        registry.prefetch_packages(
+            [
+                ("a", "1.0", tmp_path / "a", []),
+                ("b", "2.0", tmp_path / "b", []),
+            ],
+            tmp_path / "dl",
+        )
+    assert "TypeError" in caplog.text

tests/unit_tests/test_platformio_toolchain.py

@@ -431,8 +431,8 @@ def test_ccache_env_enabled_by_default(setup_core: Path) -> None:
 
     with (
         patch.dict(os.environ, {}, clear=True),
-        patch.object(toolchain.shutil, "which", return_value="/usr/bin/ccache"),
-        patch.object(toolchain.subprocess, "run"),
+        patch("shutil.which", return_value="/usr/bin/ccache"),
+        patch("esphome.framework_helpers.subprocess.run"),
     ):
         env = toolchain._ccache_env()
 
@@ -469,7 +469,7 @@ def test_ccache_env_disabled_without_binary(
 
     with (
         patch.dict(os.environ, env_vars, clear=True),
-        patch.object(toolchain.shutil, "which", return_value=None),
+        patch("shutil.which", return_value=None),
         caplog.at_level("WARNING"),
     ):
         env = toolchain._ccache_env()
@@ -494,8 +494,8 @@ def test_ccache_env_disabled_when_probe_fails(
 
     with (
         patch.dict(os.environ, {}, clear=True),
-        patch.object(toolchain.shutil, "which", return_value="/usr/bin/ccache"),
-        patch.object(toolchain.subprocess, "run", side_effect=probe_error),
+        patch("shutil.which", return_value="/usr/bin/ccache"),
+        patch("esphome.framework_helpers.subprocess.run", side_effect=probe_error),
     ):
         env = toolchain._ccache_env()
 
@@ -508,8 +508,8 @@ def test_ccache_env_forced_on_skips_probe(setup_core: Path) -> None:
 
     with (
         patch.dict(os.environ, {"ESPHOME_CCACHE_ENABLE": "1"}, clear=True),
-        patch.object(toolchain.shutil, "which", return_value="/usr/bin/ccache"),
-        patch.object(toolchain.subprocess, "run") as mock_probe,
+        patch("shutil.which", return_value="/usr/bin/ccache"),
+        patch("esphome.framework_helpers.subprocess.run") as mock_probe,
     ):
         env = toolchain._ccache_env()
 
@@ -537,9 +537,9 @@ def test_ccache_env_strips_win_long_path_prefix(setup_core: Path) -> None:
         patch.dict(os.environ, {}, clear=True),
         # shutil.which is patched, so the win32 code path of the real
         # implementation (which crashes on a POSIX host) is never reached.
-        patch("esphome.platformio.toolchain.sys.platform", "win32"),
-        patch.object(toolchain.shutil, "which", return_value=prefixed),
-        patch.object(toolchain.subprocess, "run") as mock_probe,
+        patch("esphome.framework_helpers.sys.platform", "win32"),
+        patch("shutil.which", return_value=prefixed),
+        patch("esphome.framework_helpers.subprocess.run") as mock_probe,
     ):
         env = toolchain._ccache_env()
 
@@ -555,7 +555,7 @@ def test_ccache_env_opt_out(setup_core: Path) -> None:
 
     with (
         patch.dict(os.environ, {"ESPHOME_CCACHE_ENABLE": "0"}, clear=True),
-        patch.object(toolchain.shutil, "which", return_value="/usr/bin/ccache"),
+        patch("shutil.which", return_value="/usr/bin/ccache"),
     ):
         env = toolchain._ccache_env()
 
@@ -568,7 +568,7 @@ def test_ccache_env_normalizes_enable_value(setup_core: Path) -> None:
 
     with (
         patch.dict(os.environ, {"ESPHOME_CCACHE_ENABLE": "yes"}, clear=True),
-        patch.object(toolchain.shutil, "which", return_value="/usr/bin/ccache"),
+        patch("shutil.which", return_value="/usr/bin/ccache"),
     ):
         env = toolchain._ccache_env()
 
@@ -587,8 +587,8 @@ def test_ccache_env_respects_user_values_and_refreshes_basedir(
 
     with (
         patch.dict(os.environ, user_env, clear=True),
-        patch.object(toolchain.shutil, "which", return_value="/usr/bin/ccache"),
-        patch.object(toolchain.subprocess, "run"),
+        patch("shutil.which", return_value="/usr/bin/ccache"),
+        patch("esphome.framework_helpers.subprocess.run"),
     ):
         env = toolchain._ccache_env()
 
@@ -606,8 +606,8 @@ def test_run_platformio_cli_passes_ccache_env_to_subprocess_only(
 
     with (
         patch.dict(os.environ, {}, clear=False),
-        patch.object(toolchain.shutil, "which", return_value="/usr/bin/ccache"),
-        patch.object(toolchain.subprocess, "run"),
+        patch("shutil.which", return_value="/usr/bin/ccache"),
+        patch("esphome.framework_helpers.subprocess.run"),
     ):
         os.environ.pop("ESPHOME_CCACHE_ENABLE", None)
         mock_run_external_process.return_value = 0
@@ -628,8 +628,8 @@ def test_ccache_env_requires_build_path(setup_core: Path) -> None:
 
     with (
         patch.dict(os.environ, {}, clear=True),
-        patch.object(toolchain.shutil, "which", return_value="/usr/bin/ccache"),
-        patch.object(toolchain.subprocess, "run"),
+        patch("shutil.which", return_value="/usr/bin/ccache"),
+        patch("esphome.framework_helpers.subprocess.run"),
         pytest.raises(ValueError, match="CORE.build_path must be set"),
     ):
         toolchain._ccache_env()
@@ -642,8 +642,8 @@ def test_run_platformio_cli_merges_caller_env(
     CORE.build_path = str(setup_core / "build" / "test")
 
     with (
-        patch.object(toolchain.shutil, "which", return_value="/usr/bin/ccache"),
-        patch.object(toolchain.subprocess, "run"),
+        patch("shutil.which", return_value="/usr/bin/ccache"),
+        patch("esphome.framework_helpers.subprocess.run"),
     ):
         mock_run_external_process.return_value = 0
         toolchain.run_platformio_cli(
@@ -800,9 +800,7 @@ def test_ccache_env_real_probe_runs_stripped_path(setup_core: Path) -> None:
 
     with (
         patch.dict(os.environ, {}, clear=False),
-        patch.object(
-            toolchain.shutil, "which", return_value="\\\\?\\" + sys.executable
-        ),
+        patch("shutil.which", return_value="\\\\?\\" + sys.executable),
     ):
         os.environ.pop("ESPHOME_CCACHE_ENABLE", None)
         env = toolchain._ccache_env()
@@ -843,40 +841,6 @@ def test_ccache_wrapper_through_cmd_exe(
         assert marker.read_text() == "compiled"
 
 
-@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.toolchain.sys.platform", platform):
-        assert toolchain._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:
@@ -900,7 +864,7 @@ def test_run_platformio_cli_strips_win_long_path_prefix(
         # so the stdlib sees it too) would send shutil.which down the Windows
         # code path, which crashes on a POSIX host.
         patch.dict(os.environ, {"ESPHOME_CCACHE_ENABLE": "0"}, clear=False),
-        patch("esphome.platformio.toolchain.sys.platform", "win32"),
+        patch("esphome.framework_helpers.sys.platform", "win32"),
         patch("esphome.platformio.toolchain.sys.executable", prefixed_exe),
     ):
         # Pop any pre-existing PYTHONEXEPATH so the assertion below reflects
@@ -932,7 +896,7 @@ def test_run_platformio_cli_does_not_set_pythonexepath_without_strip(
 
     with (
         patch.dict(os.environ, {}, clear=False),
-        patch("esphome.platformio.toolchain.sys.platform", "linux"),
+        patch("esphome.framework_helpers.sys.platform", "linux"),
         patch("esphome.platformio.toolchain.sys.executable", plain_exe),
     ):
         os.environ.pop("PYTHONEXEPATH", None)

tests/unit_tests/test_size_summary.py

@@ -126,3 +126,92 @@ def test_print_summary_handles_no_memory_types(
     size_json = _write_size_json(tmp_path, {"image_size": 0})
     print_summary(size_json, partitions_csv=None)
     assert capsys.readouterr().out == ""
+
+
+def test_print_summary_flash_line(
+    tmp_path: Path, capsys: pytest.CaptureFixture[str]
+) -> None:
+    """image_size + a factory app partition produce the Flash line."""
+    size_json = tmp_path / "esp_idf_size.json"
+    size_json.write_text(
+        json.dumps(
+            {
+                "memory_types": {"DRAM": {"used": 100, "size": 200}},
+                "image_size": 500,
+            }
+        )
+    )
+    partitions = tmp_path / "partitions.csv"
+    partitions.write_text(
+        "# name, type, subtype, offset, size\napp0, app, factory, 0x10000, 0x100000\n"
+    )
+    print_summary(size_json, partitions)
+    out = capsys.readouterr().out
+    assert "RAM:   [=====     ]  50.0% (used 100 bytes from 200 bytes)" in out
+    assert "Flash: [          ]   0.0% (used 500 bytes from 1048576 bytes)" in out
+
+
+def test_print_summary_missing_ram_region_warns(
+    tmp_path: Path, caplog: pytest.LogCaptureFixture
+) -> None:
+    """A missing RAM line is diagnosable, not a silently absent CI metric."""
+    size_json = _write_size_json(tmp_path, {"memory_types": {}, "image_size": 100})
+    print_summary(size_json, partitions_csv=None)
+    assert "Skipping RAM summary" in caplog.text
+
+
+def test_print_summary_bad_partitions_warns(
+    tmp_path: Path, caplog: pytest.LogCaptureFixture
+) -> None:
+    """An unparseable partition table skips the Flash line with a warning."""
+    size_json = _write_size_json(tmp_path, _esp32_size_data())
+    partitions = tmp_path / "partitions.csv"
+    partitions.write_text("not,a,valid,partition,table\n")
+    print_summary(size_json, partitions_csv=partitions)
+    assert "Skipping Flash summary" in caplog.text
+
+
+def test_print_summary_corrupt_json_warns(
+    tmp_path: Path, caplog: pytest.LogCaptureFixture
+) -> None:
+    size_json = tmp_path / "size.json"
+    size_json.write_text("{not json")
+    print_summary(size_json, partitions_csv=None)
+    assert "Skipping size summary" in caplog.text
+
+
+def test_print_summary_missing_flash_inputs_warn(
+    tmp_path: Path, caplog: pytest.LogCaptureFixture
+) -> None:
+    """Both absent-input paths for the Flash line name their cause."""
+    size_json = _write_size_json(tmp_path, _esp32_size_data())
+    print_summary(size_json, partitions_csv=None)
+    assert "no partition table given" in caplog.text
+    caplog.clear()
+    data = _esp32_size_data()
+    data.pop("image_size", None)
+    size_json = _write_size_json(tmp_path, data)
+    print_summary(size_json, partitions_csv=tmp_path / "partitions.cssv")
+    assert "no image_size" in caplog.text
+
+
+def test_print_summary_non_dict_json_warns(tmp_path, caplog) -> None:
+    """Valid JSON that is not an object must warn, not raise past a build
+    that already linked."""
+    size_json = tmp_path / "size.json"
+    size_json.write_text("[]")
+    print_summary(size_json, tmp_path / "partitions.csv")
+    assert "unexpected shape" in caplog.text
+
+
+def test_print_summary_zero_app_partition_warns(tmp_path, caplog) -> None:
+    """A malformed partition row parsing to 0 must not render a 0% bar for
+    CI's memory-impact extraction to ingest."""
+    size_json = tmp_path / "size.json"
+    size_json.write_text(
+        '{"memory_types": {"DRAM": {"used": 1, "size": 2}}, "image_size": 100}'
+    )
+    partitions = tmp_path / "partitions.csv"
+    partitions.write_text("app0, app, ota_0, 0x10000, ,\n")
+    print_summary(size_json, partitions)
+    assert "app partition size is" in caplog.text

tests/unit_tests/test_writer.py

@@ -2462,3 +2462,31 @@ def test_copy_src_tree_ignores_removed_generated_file(
     # file was removed and regenerated, not that it triggered sources_changed.
     new_json = json.loads(build_info_json_path.read_text())
     assert new_json["config_hash"] == 0xDEADBEEF
+
+
+def test_build_info_stale_branches(tmp_path: Path) -> None:
+    """Missing files, an unreadable JSON, a hash or version mismatch each
+    regenerate; a matching record does not."""
+    import json as json_mod
+
+    from esphome.const import __version__
+    from esphome.writer import _build_info_stale
+
+    h = tmp_path / "build_info_data.h"
+    cpp = tmp_path / "build_info_data.cpp"
+    info = tmp_path / "build_info.json"
+    assert _build_info_stale(h, cpp, info, 1) is True  # files missing
+    h.write_text("")
+    cpp.write_text("")
+    assert _build_info_stale(h, cpp, info, 1) is True  # JSON unreadable
+    info.write_text("not json")
+    assert _build_info_stale(h, cpp, info, 1) is True
+    # Valid JSON that is not an object is stale, not an AttributeError
+    info.write_text("[]")
+    assert _build_info_stale(h, cpp, info, 1) is True
+    info.write_text(json_mod.dumps({"config_hash": 2, "esphome_version": __version__}))
+    assert _build_info_stale(h, cpp, info, 1) is True  # hash mismatch
+    info.write_text(json_mod.dumps({"config_hash": 1, "esphome_version": "0.0.0"}))
+    assert _build_info_stale(h, cpp, info, 1) is True  # version mismatch
+    info.write_text(json_mod.dumps({"config_hash": 1, "esphome_version": __version__}))
+    assert _build_info_stale(h, cpp, info, 1) is False