Files
esphome/esphome/platformio/library.py
T

779 lines
29 KiB
Python

"""Toolchain-agnostic PlatformIO library converter.
Resolves a batch of PlatformIO/Arduino library specs (added via
``cg.add_library(...)``) into local, build-ready directories: it fetches each
library (registry/git/url), parses its ``library.json`` / ``library.properties``
manifest, resolves the whole dependency graph to a single version per name, and
caches the result under ``<data_dir>/pio_components``.
The toolchain-specific part — turning a resolved library into build files
(ESP-IDF ``idf_component_register`` CMakeLists, or a Zephyr module) — is supplied
by a :class:`LibraryBackend`. This module owns everything that is the same
regardless of which toolchain consumes the result.
"""
from collections import deque
from collections.abc import Callable
from dataclasses import dataclass, field
import glob
import hashlib
import itertools
import json
import logging
import os
from pathlib import Path
import re
import tempfile
from typing import Any
from urllib.parse import urlsplit, urlunsplit
from esphome import git
from esphome.core import CORE, Library
from esphome.framework_helpers import archive_extract_all, download_from_mirrors, rmdir
_LOGGER = logging.getLogger(__name__)
PathType = str | os.PathLike
#
# Constants from platformio
#
FILTER_REGEX = re.compile(r"([+-])<([^>]+)>")
DEFAULT_BUILD_SRC_FILTER = (
"+<*> -<.git/> -<.svn/> -<example/> -<examples/> -<test/> -<tests/>"
)
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",
]
DOMAIN = "pio_components"
ESPHOME_DATA_KEY = "ESPHOME"
ESPHOME_DATA_EXTRA_CMAKE_KEY = "EXTRA_CMAKE"
class Source:
def download(
self, dir_suffix: str, force: bool = False, salt: str = "", namespace: str = ""
) -> Path:
raise NotImplementedError
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:
# 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
if namespace:
base_dir = base_dir / namespace
h = hashlib.new("sha256")
h.update(self.url.encode())
if salt:
h.update(salt.encode())
path = base_dir / h.hexdigest()[:8] / dir_suffix
# 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,
# and lets us extract directly into ``path`` — avoiding a
# post-extraction rename that races with antivirus on Windows.
extracted_marker = path / ".esphome_extracted"
if not extracted_marker.is_file() or force:
rmdir(path, msg=f"Clean up library directory {path}")
# Download in temporary file
with tempfile.NamedTemporaryFile() as tmp:
_LOGGER.info("Downloading %s ...", self.url)
_LOGGER.debug("Location: %s", path)
download_from_mirrors([self.url], {}, tmp.file)
_LOGGER.debug("Extracting archive to %s ...", path)
archive_extract_all(tmp.file, path)
extracted_marker.touch()
return path
def __str__(self):
return self.url
class GitSource(Source):
def __init__(self, url: str, ref: str | None):
self.url = url
self.ref = ref
def download(
self, dir_suffix: str, force: bool = False, salt: str = "", namespace: str = ""
) -> Path:
domain = DOMAIN
if namespace:
domain = f"{domain}/{namespace}"
if salt:
domain = f"{domain}/{salt}"
path, _ = git.clone_or_update(
url=self.url,
ref=self.ref,
refresh=git.NEVER_REFRESH if not force else None,
domain=domain,
submodules=[],
subpath=Path(dir_suffix),
)
return path
def __str__(self):
return f"{self.url}#{self.ref}" if self.ref else self.url
class InvalidLibrary(Exception):
pass
class ConvertedLibrary:
"""A resolved PlatformIO library plus its parsed manifest and on-disk path.
Toolchain-neutral: ESP-IDF treats it as a component, Zephyr as a module. The
backend reads ``name``/``version``/``data``/``dependencies``/``path`` to emit
its build files.
"""
def __init__(self, name: str, version: str, source: Source | None):
self.name = name
self.version = version
self.source = source
self.data = {}
self.dependencies: list[ConvertedLibrary] = []
self._path: Path | None = None
def __str__(self):
return f"{self.name}@{self.version}={self.source}"
@property
def path(self) -> Path:
if self._path is None:
raise RuntimeError(f"path not set for library {self}")
return self._path
@path.setter
def path(self, value: Path) -> None:
self._path = value
def get_sanitized_name(self):
return re.sub(r"[^a-zA-Z0-9_.\-/]", "_", self.name)
def get_require_name(self):
return self.get_sanitized_name().replace("/", "__")
def download(self, force: bool = False, salt: str = "", namespace: str = ""):
"""Fetch the library into the shared cache and record its ``path``.
The cache directory is named after the sanitized library name; backends
rely on that name to identify the unit they build (e.g. ESP-IDF uses the
directory name as the component name, replacing ``/`` with ``__`` via
``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
)
@dataclass
class LibraryBackend:
"""Toolchain hooks for :func:`convert_libraries`.
``platform``/``framework`` drive the manifest compatibility check.
``emit`` writes the toolchain-specific build files into a resolved library's
``path`` (e.g. the ESP-IDF ``CMakeLists.txt`` + ``idf_component.yml``, or a
Zephyr ``module.yml`` + ``CMakeLists.txt``).
``cache_key`` namespaces the download cache (``pio_components/<cache_key>/``)
so the differing build files two backends emit into a library dir never
collide when the same config dir hosts both an ESP-IDF and a Zephyr build.
"""
platform: str | None
framework: str
emit: Callable[["ConvertedLibrary"], None]
cache_key: str
def ensure_list[T](obj: T | list[T]) -> list[T]:
"""
Convert an object to a list if it isn't already a list.
Args:
obj: Object that may or may not already be a list.
Returns:
list[T]: The original list if ``obj`` is a list, otherwise a single-item
list containing ``obj``.
"""
return [obj] if not isinstance(obj, list) else obj
def _owner_pkgname_to_name(owner: str | None, pkgname: str) -> str:
"""
Convert owner and package name to a standardized component name.
This function combines owner and package name with a forward slash when
both are provided, otherwise returns just the package name.
Args:
owner: The owner/username of the package (can be None)
pkgname: The name of the package
Returns:
str: The standardized component name in "owner/pkgname" format or just "pkgname"
"""
return f"{owner}/{pkgname}" if owner else pkgname
def collect_filtered_files(src_dir: PathType, src_filters: list[str]) -> list[str]:
"""
Recursively match files in a directory according to include/exclude patterns.
This function processes a list of filter strings that indicate which files
to include or exclude. Each filter is parsed into patterns with a sign:
'+' for inclusion and '-' for exclusion. Directory patterns ending with '/'
are normalized to include all their contents recursively.
Args:
src_dir (PathType): Root directory to search within.
src_filters (list[str]): List of filter strings, which may contain multiple
patterns. Each pattern can start with '+' or '-' to indicate inclusion
or exclusion.
Returns:
list[str]: List of matched file paths as strings. Only files (not directories)
are returned, even if a directory matches a pattern.
"""
matches = list(
itertools.chain.from_iterable(
FILTER_REGEX.findall(src_filter) for src_filter in src_filters
)
)
selected = set()
for sign, pattern in matches:
pattern = pattern.strip()
if pattern.endswith("/"):
pattern = pattern.rstrip("/") + "/**"
# glob.escape has no pathlib equivalent and the matcher works on raw
# path strings, so PTH118/PTH207 don't apply here.
full_pattern = os.path.join(glob.escape(str(src_dir)), pattern) # noqa: PTH118
matched = []
for item in glob.glob(full_pattern, recursive=True): # noqa: PTH207
if not Path(item).is_dir():
matched.append(item)
else:
# PlatformIO quirk: a directory matched with "*" should include all its
# nested files and subdirectories, not just the directory itself.
for root, _, files in os.walk(item):
matched.extend([str(Path(root) / f) for f in files])
# glob keeps the pattern's literal separators for non-wildcard path
# components, so on Windows the same file can surface with different
# separators depending on where the wildcards sit; normalize so the
# include/exclude set operations below compare equal paths.
matched = [os.path.normpath(m) for m in matched]
# FILTER_REGEX only ever captures "+" or "-", so the else is the "-" case.
if sign == "+":
selected.update(matched)
else:
selected.difference_update(matched)
return [r for r in selected if Path(r).is_file()]
def split_list_by_condition(
items: list[str], match_fn: Callable[[str], str | None]
) -> tuple[list[str], list[str]]:
"""
Splits a list into two lists based on a matching function.
Args:
items: List of items to split.
match_fn: Function that returns a value for items that should go into the "matched" list.
Returns:
A tuple (matched, non_matched)
"""
matched = []
non_matched = []
for item in items:
result = match_fn(item)
if result:
matched.append(result)
else:
non_matched.append(item)
return matched, non_matched
def check_library_data(data: dict, platform: str | None, framework: str):
"""
Check whether a library manifest is compatible with the target toolchain.
A platform mismatch (e.g. an AVR-only library on ESP32) raises
``InvalidLibrary`` so the caller skips the library. A framework mismatch only
logs a warning — PIO manifests often understate the frameworks they actually
compile under, and there's no opt-out at this layer, so we include the library
anyway.
Args:
data: PIO library manifest dict being processed.
platform: The PlatformIO platform token the build targets (e.g.
``espressif32``). ``None`` skips the platform check entirely — useful
for targets (e.g. Zephyr) where PIO manifests rarely declare the
platform yet portable libraries still build.
framework: The active framework name (e.g. ``espidf``, ``arduino``,
``zephyr``) the manifest is expected to declare.
Raises:
InvalidLibrary: If the library does not support the target platform.
"""
platforms = data.get("platforms", "*")
if isinstance(platforms, str):
platforms = [a.strip() for a in platforms.split(",")]
platforms = ensure_list(platforms)
# Check if library supports the target platform
valid_platforms = platform is None or "*" in platforms or platform in platforms
if not valid_platforms:
raise InvalidLibrary(f"Unsupported library platforms: {platforms}")
frameworks = data.get("frameworks", "*")
if isinstance(frameworks, str):
frameworks = [a.strip() for a in frameworks.split(",")]
frameworks = ensure_list(frameworks)
# Check if library declares the active framework. PIO library manifests
# often list only "arduino" even when the library actually compiles fine
# under the target framework, and there's no way to opt out of the check at
# this layer. Warn instead of failing so the user isn't forced to fork the
# library to fix the manifest.
valid_framework = "*" in frameworks or framework in frameworks
if not valid_framework:
_LOGGER.warning(
"Library %s declares frameworks %s that do not include '%s'; including anyway",
data.get("name", "<unknown>"),
frameworks,
framework,
)
def _parse_library_json(library_json_path: PathType):
"""
Load and parse a JSON file describing a library.
Args:
library_json_path (PathType): Path to the JSON file.
Returns:
dict: Parsed JSON content as a Python dictionary.
"""
with Path(library_json_path).open(encoding="utf8") as fp:
return json.load(fp)
def _parse_library_properties(library_properties_path: PathType):
"""
Parse a key-value platformio .properties style file into a dictionary.
Args:
library_properties_path (PathType): Path to the properties file.
Returns:
dict[str, str]: Mapping of parsed property keys to values.
"""
with Path(library_properties_path).open(encoding="utf8") as fp:
data = {}
for line in fp.read().splitlines():
line = line.strip()
if not line or "=" not in line:
continue
# skip comments
if line.startswith("#"):
continue
key, value = line.split("=", 1)
if not value.strip():
continue
data[key.strip()] = value.strip()
return data
def _make_registry_client() -> Any:
"""Create a minimal PlatformIO registry client with no system filtering.
``is_system_compatible`` is forced True so version selection is driven purely
by the requested version requirements -- target compatibility is handled
elsewhere, not by the PlatformIO registry.
"""
from platformio.package.manager._registry import PackageManagerRegistryMixin
class _Registry(PackageManagerRegistryMixin):
def __init__(self) -> None:
self._registry_client = None
self.pkg_type = "library"
@staticmethod
def is_system_compatible(value: Any, custom_system: Any = None) -> bool:
return True
return _Registry()
def _resolve_registry_version(
owner: str | None, pkgname: str, requirements: set[str]
) -> tuple[str, str, str, str]:
"""Resolve a registry package to the single highest version satisfying ALL
the given requirements; return ``(owner, name, version, download_url)``.
Intersecting every requirement (rather than resolving each consumer in
isolation) makes the result independent of processing order and guarantees
no stated constraint is violated -- e.g. ``esphome/libsodium`` requested as
both ``==1.10021.0`` and ``^1.10018.1`` resolves to ``1.10021.0``.
"""
from platformio.package.meta import PackageSpec
registry = _make_registry_client()
package = registry.fetch_registry_package(PackageSpec(owner=owner, name=pkgname))
owner = package["owner"]["username"]
name = package["name"]
# Chaining the per-requirement filter intersects all constraints.
versions = package.get("versions") or []
for requirement in sorted(requirements):
versions = registry.get_compatible_registry_versions(
versions, PackageSpec(owner=owner, name=name, requirements=requirement)
)
if not versions:
raise RuntimeError(
f"No version of {owner}/{name} satisfies all requirements "
f"{sorted(requirements)} requested across the library tree"
)
best = registry.pick_best_registry_version(versions)
pkgfile = registry.pick_compatible_pkg_file(best["files"])
if not pkgfile:
raise RuntimeError(f"No package file for {owner}/{name}@{best['name']}")
return owner, name, best["name"], pkgfile["download_url"]
def _normalize_dependencies(dependencies: Any) -> 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.
"""
if not dependencies:
return []
if isinstance(dependencies, dict):
normalized = []
for raw_name, spec in dependencies.items():
if "/" in raw_name:
owner, pkgname = raw_name.split("/", 1)
else:
owner, pkgname = None, raw_name
entry = {"name": pkgname, "owner": owner}
if isinstance(spec, dict):
entry.update(spec)
else:
entry["version"] = spec
normalized.append(entry)
return normalized
return [d for d in dependencies if isinstance(d, dict)]
@dataclass
class _LibNode:
"""A node in the library dependency graph being resolved as a batch."""
key: str
is_git: bool
owner: str | None = None
pkgname: str | None = None
requirements: set[str] = field(default_factory=set)
url: str | None = None
ref: str | None = None
edges: set[str] = field(default_factory=set)
def _url_or_none(value: Any) -> str | None:
"""Return ``value`` if it parses as a URL (scheme and host), else None."""
if not value or not isinstance(value, str):
return None
try:
parsed = urlsplit(value)
except ValueError:
return None
return value if parsed.scheme and parsed.netloc else None
def _node_key(
name: str | None, version: str | None, repository: str | None
) -> tuple[str, bool, tuple[str | None, str | None]]:
"""Return ``(key, is_git, locator)`` for a library or dependency spec.
The key is derived from the *input* spec (the registry name as written, or
the git URL path), not the resolved canonical name. So a package referenced
inconsistently -- bare ``name`` vs ``owner/name``, or git vs registry -- maps
to distinct keys and isn't deduplicated; ``convert_libraries`` warns about
that after resolution rather than merging the nodes.
PlatformIO's Library Manager also accepted a git URL in the *name*
position (``add_library("https://github.com/x/y", None)``), including the
``git+`` VCS prefix and the ``CustomName=URL`` form; recognize those here
so such specs resolve as git sources instead of failing a registry lookup.
"""
if not repository and name and "://" in name:
# Try the whole name first so a bare URL whose query contains ``=``
# stays intact; fall back to the ``CustomName=URL`` form, where the
# key derives from the URL path and the custom name is irrelevant.
repository = _url_or_none(name) or _url_or_none(name.split("=", 1)[-1])
if repository is None:
# Anything with ``://`` was meant to be a URL; failing it fast
# beats a confusing registry "package not found" error.
raise RuntimeError(f"Invalid PIO library URL: {name}")
if repository:
split_result = urlsplit(repository.removeprefix("git+"))
key = str(split_result.path).strip("/").removesuffix(".git")
ref = split_result.fragment.strip() or None
url = urlunsplit(split_result._replace(fragment=""))
return key, True, (url, ref)
if name and "/" in name:
owner, pkgname = name.split("/", 1)
else:
owner, pkgname = None, name
return name, False, (owner, pkgname)
def convert_libraries(
libraries: list[Library], backend: LibraryBackend
) -> list[ConvertedLibrary]:
"""Resolve and convert a batch of PlatformIO libraries for ``backend``.
Resolves the whole set together rather than each library independently: it
walks the dependency graph collecting every version *requirement* per
component name, then resolves each name once to a single version satisfying
all of them. So a transitive dependency shared under
different specs (e.g. ``esphome/libsodium``, pulled by both ``noise-c`` and
``esp_wireguard``) becomes one component instead of two clashing
``override_path`` entries -- order-independently, and without ever violating
a stated constraint.
The returned list holds the top-level components (those directly requested);
transitive dependencies are converted too and wired into each component's
generated manifest. ``backend.emit`` is called once per converted library to
write its toolchain-specific build files.
``lib_ignore`` from ``esphome->platformio_options`` excludes libraries by
short name (part after the ``/``), matched against both the top-level
libraries and every dependency discovered during the graph walk.
"""
nodes: dict[str, _LibNode] = {}
lib_ignore = {
name.split("/")[-1].lower()
for name in CORE.platformio_options.get("lib_ignore", [])
}
# The generated build files inside the shared cache bake in the dependency
# wiring, which lib_ignore changes; salt the cache path so configs with
# different lib_ignore values don't fight over (and constantly rewrite) the
# same converted component files.
salt = (
hashlib.sha256(",".join(sorted(lib_ignore)).encode()).hexdigest()[:8]
if lib_ignore
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, is_git, locator = _node_key(name, version, repository)
node = nodes.get(key) or _LibNode(key=key, is_git=is_git)
nodes[key] = node
if is_git:
node.is_git = True
node.url, node.ref = locator
else:
node.owner, node.pkgname = locator
if version:
node.requirements.add(version)
return key
top_level = [
add_spec(library.name, library.version, library.repository)
for library in libraries
if not is_ignored(library.name)
]
# Collect + resolve to a fixpoint: a node is (re)resolved whenever its
# requirement set has grown since the last time, so every requirement in the
# graph is accounted for before conversion.
components: dict[str, ConvertedLibrary] = {}
resolved_requirements: dict[str, frozenset[str]] = {}
top_level_keys = set(top_level)
worklist = deque(dict.fromkeys(top_level))
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
if node.is_git:
component = ConvertedLibrary(key, "*", GitSource(node.url, node.ref))
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)
library_json_path = component.path / "library.json"
library_properties_path = component.path / "library.properties"
has_json = library_json_path.is_file()
has_properties = library_properties_path.is_file()
if not has_json and not has_properties:
# 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.
_LOGGER.warning(
"Library %s at %s is missing library.json and library.properties; "
"re-downloading",
key,
component.path,
)
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:
raise RuntimeError(
f"Invalid PIO library {key}: missing library.json and "
f"library.properties in {component.path}"
)
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
# 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)
except InvalidLibrary as e:
_LOGGER.debug("Skip dependency %s: %s", dependency.get("name"), 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)
# A git source wins over any registry version requested for the same
# component. That's intentional, but warn so a dropped registry pin isn't a
# silent surprise.
for node in nodes.values():
if node.is_git and node.requirements:
_LOGGER.warning(
"Library %s is requested both from a git source (%s) and as "
"registry version(s) %s; using the git source.",
node.key,
node.url,
sorted(node.requirements),
)
# Two graph nodes that resolve to the same component name (e.g. a package
# referenced both bare and as ``owner/name``) are not deduplicated and can
# produce conflicting component definitions. Warn so it's not silent.
canonical_keys: dict[str, str] = {}
for node_key, component in components.items():
canonical = component.get_sanitized_name()
if canonical_keys.setdefault(canonical, node_key) != node_key:
_LOGGER.warning(
"Library %s is referenced under multiple names (%s and %s); these "
"are not deduplicated. Reference it consistently as %s.",
canonical,
canonical_keys[canonical],
node_key,
canonical,
)
# Wire each component's dependencies to the single resolved instances, then
# emit build files.
for key, component in components.items():
component.dependencies = [
components[dep_key]
for dep_key in sorted(nodes[key].edges)
if dep_key in components
]
for component in components.values():
backend.emit(component)
return [components[key] for key in top_level if key in components]