# This module contains the code allowing to find modules. # # Note: It might be possible to replace a good part of this module's logic # with utilities from `importlib` (however the util in question is private): # # ```pycon # >>> from importlib.util import _find_spec # >>> _find_spec("griffe.agents", _find_spec("griffe", None).submodule_search_locations) # ModuleSpec( # name='griffe.agents', # loader=<_frozen_importlib_external.SourceFileLoader object at 0x7fa5f34e8110>, # origin='/media/data/dev/griffe/src/griffe/agents/__init__.py', # submodule_search_locations=['/media/data/dev/griffe/src/griffe/agents'], # ) # ``` from __future__ import annotations import ast import os import re import sys from collections import defaultdict from contextlib import suppress from dataclasses import dataclass from itertools import chain from pathlib import Path from typing import TYPE_CHECKING, ClassVar from _griffe.exceptions import UnhandledEditableModuleError from _griffe.logger import logger if TYPE_CHECKING: from collections.abc import Iterator, Sequence from re import Pattern from _griffe.models import Module _editable_editables_patterns = [re.compile(pat) for pat in (r"^__editables_\w+\.py$", r"^_editable_impl_\w+\.py$")] _editable_setuptools_patterns = [re.compile(pat) for pat in (r"^__editable__\w+\.py$",)] _editable_scikit_build_core_patterns = [re.compile(pat) for pat in (r"^_\w+_editable.py$",)] _editable_meson_python_patterns = [re.compile(pat) for pat in (r"^_\w+_editable_loader.py$",)] NamePartsType = tuple[str, ...] """Type alias for the parts of a module name.""" NamePartsAndPathType = tuple[NamePartsType, Path] """Type alias for the parts of a module name and its path.""" def _match_pattern(string: str, patterns: Sequence[Pattern]) -> bool: return any(pattern.match(string) for pattern in patterns) @dataclass class Package: """This class is a simple placeholder used during the process of finding packages. Parameters: name: The package name. path: The package path(s). stubs: An optional path to the related stubs file (.pyi). """ name: str """Package name.""" path: Path """Package folder path.""" stubs: Path | None = None """Package stubs file.""" @dataclass class NamespacePackage: """This class is a simple placeholder used during the process of finding packages. Parameters: name: The package name. path: The package paths. """ name: str """Namespace package name.""" path: list[Path] """Namespace package folder paths.""" class ModuleFinder: """The Griffe finder, allowing to find modules on the file system. The module finder is generally not used directly. Each [`GriffeLoader`][griffe.GriffeLoader] instance creates its own module finder instance. The finder can be configured when instantiating the loader thanks to the [loader][griffe.GriffeLoader]'s `search_paths` parameter. """ accepted_py_module_extensions: ClassVar[list[str]] = [".py", ".pyc", ".pyo", ".pyd", ".pyi", ".so"] """List of extensions supported by the finder.""" extensions_set: ClassVar[set[str]] = set(accepted_py_module_extensions) """Set of extensions supported by the finder.""" def __init__(self, search_paths: Sequence[str | Path] | None = None) -> None: """Initialize the finder. Parameters: search_paths: Optional paths to search into. """ self._paths_contents: dict[Path, list[Path]] = {} self.search_paths: list[Path] = [] """The finder search paths.""" # Optimization: pre-compute Paths to relieve CPU when joining paths. for path in search_paths or sys.path: self.append_search_path(Path(path)) self._always_scan_for: dict[str, list[Path]] = defaultdict(list) self._extend_from_pth_files() def append_search_path(self, path: Path) -> None: """Append a search path. The path will be resolved (absolute, normalized). The path won't be appended if it is already in the search paths list. Parameters: path: The path to append. """ path = path.resolve() if path not in self.search_paths: self.search_paths.append(path) def insert_search_path(self, position: int, path: Path) -> None: """Insert a search path at the given position. The path will be resolved (absolute, normalized). The path won't be inserted if it is already in the search paths list. Parameters: position: The insert position in the list. path: The path to insert. """ path = path.resolve() if path not in self.search_paths: self.search_paths.insert(position, path) def find_spec( self, module: str | Path, *, try_relative_path: bool = True, find_stubs_package: bool = False, ) -> tuple[str, Package | NamespacePackage]: """Find the top-level parent module of a module. If a Path is passed, only try to find the module as a file path. If a string is passed, first try to find the module as a file path, then look into the search paths. Parameters: module: The module name or path. try_relative_path: Whether to try finding the module as a relative path, when the given module is not already a path. find_stubs_package: Whether to search for stubs-only package. If both the package and its stubs are found, they'll be merged together. If only the stubs are found, they'll be used as the package itself. Raises: FileNotFoundError: When a Path was passed and the module could not be found: - the directory has no `__init__.py` file in it - the path does not exist ModuleNotFoundError: When a string was passed and the module could not be found: - no `module/__init__.py` - no `module.py` - no `module.pth` - no `module` directory (namespace packages) - or unsupported .pth file Returns: The name of the module, and an instance representing its (namespace) package. """ module_path: Path | list[Path] if isinstance(module, Path): module_name, module_path = self._module_name_path(module) top_module_name = self._top_module_name(module_path) elif try_relative_path: try: module_name, module_path = self._module_name_path(Path(module)) except FileNotFoundError: module_name = module top_module_name = module.split(".", 1)[0] else: top_module_name = self._top_module_name(module_path) else: module_name = module top_module_name = module.split(".", 1)[0] # Only search for actual package, let exceptions bubble up. if not find_stubs_package: return module_name, self.find_package(top_module_name) # Search for both package and stubs-only package. try: package = self.find_package(top_module_name) except ModuleNotFoundError: package = None try: stubs = self.find_package(top_module_name + "-stubs") except ModuleNotFoundError: stubs = None # None found, raise error. if package is None and stubs is None: raise ModuleNotFoundError(top_module_name) # Both found, assemble them to be merged later. if package and stubs: if isinstance(package, Package) and isinstance(stubs, Package): package.stubs = stubs.path elif isinstance(package, NamespacePackage) and isinstance(stubs, NamespacePackage): package.path += stubs.path return module_name, package # Return either one. return module_name, package or stubs # type: ignore[return-value] def find_package(self, module_name: str) -> Package | NamespacePackage: """Find a package or namespace package. Parameters: module_name: The module name. Raises: ModuleNotFoundError: When the module cannot be found. Returns: A package or namespace package wrapper. """ filepaths = [ Path(module_name), # TODO: Handle .py[cod] and .so files? # This would be needed for package that are composed # solely of a file with such an extension. Path(f"{module_name}.py"), ] real_module_name = module_name if real_module_name.endswith("-stubs"): real_module_name = real_module_name[:-6] namespace_dirs = [] for path in self.search_paths: path_contents = self._contents(path) if path_contents: for choice in filepaths: abs_path = path / choice if abs_path in path_contents: if abs_path.suffix: stubs = abs_path.with_suffix(".pyi") return Package(real_module_name, abs_path, stubs if stubs.exists() else None) init_module = abs_path / "__init__.py" if init_module.exists() and not _is_pkg_style_namespace(init_module): stubs = init_module.with_suffix(".pyi") return Package(real_module_name, init_module, stubs if stubs.exists() else None) init_module = abs_path / "__init__.pyi" if init_module.exists(): # Stubs package return Package(real_module_name, init_module, None) namespace_dirs.append(abs_path) if namespace_dirs: return NamespacePackage(module_name, namespace_dirs) raise ModuleNotFoundError(module_name) def iter_submodules( self, path: Path | list[Path], seen: set | None = None, ) -> Iterator[NamePartsAndPathType]: """Iterate on a module's submodules, if any. Parameters: path: The module path. seen: If not none, this set is used to skip some files. The goal is to replicate the behavior of Python by only using the first packages (with `__init__` modules) of the same name found in different namespace packages. As soon as we find an `__init__` module, we add its parent path to the `seen` set, which will be reused when scanning the next namespace packages. Yields: name_parts (tuple[str, ...]): The parts of a submodule name. filepath (Path): A submodule filepath. """ if isinstance(path, list): # We never enter this condition again in recursive calls, # so we just have to set `seen` once regardless of its value. seen = set() for path_elem in path: yield from self.iter_submodules(path_elem, seen) return if path.stem == "__init__": path = path.parent # Optimization: just check if the file name ends with .py[icod]/.so # (to distinguish it from a directory), not if it's an actual file. elif path.suffix in self.extensions_set: return # `seen` is only set when we scan a list of paths (namespace package). # `skip` is used to prevent yielding modules # of a regular subpackage that we already yielded # from another part of the namespace. skip = set(seen or ()) for subpath in self._filter_py_modules(path): rel_subpath = subpath.relative_to(path) if rel_subpath.parent in skip: logger.debug("Skip %s, another module took precedence", subpath) continue py_file = rel_subpath.suffix == ".py" stem = rel_subpath.stem if not py_file: # .py[cod] and .so files look like `name.cpython-38-x86_64-linux-gnu.ext` stem = stem.split(".", 1)[0] if stem == "__init__": # Optimization: since it's a relative path, if it has only one part # and is named __init__, it means it's the starting path # (no need to compare it against starting path). if len(rel_subpath.parts) == 1: continue yield rel_subpath.parts[:-1], subpath if seen is not None: seen.add(rel_subpath.parent) elif py_file: yield rel_subpath.with_suffix("").parts, subpath else: yield rel_subpath.with_name(stem).parts, subpath def submodules(self, module: Module) -> list[NamePartsAndPathType]: """Return the list of a module's submodules. Parameters: module: The parent module. Returns: A list of tuples containing the parts of the submodule name and its path. """ return sorted( chain( self.iter_submodules(module.filepath), self.iter_submodules(self._always_scan_for[module.name]), ), key=_module_depth, ) def _module_name_path(self, path: Path) -> tuple[str, Path]: # Always return absolute paths to avoid working-directory-dependent issues. path = path.absolute() if path.is_dir(): for ext in self.accepted_py_module_extensions: module_path = path / f"__init__{ext}" if module_path.exists(): return path.name, module_path return path.name, path if path.exists(): if path.stem == "__init__": return path.parent.name, path return path.stem, path raise FileNotFoundError def _contents(self, path: Path) -> list[Path]: if path not in self._paths_contents: try: self._paths_contents[path] = list(path.iterdir()) except (FileNotFoundError, NotADirectoryError): self._paths_contents[path] = [] return self._paths_contents[path] def _append_search_path(self, path: Path) -> None: if path not in self.search_paths: self.search_paths.append(path) def _extend_from_pth_files(self) -> None: for path in self.search_paths: for item in self._contents(path): if item.suffix == ".pth": for directory in _handle_pth_file(item): if scan := directory.always_scan_for: self._always_scan_for[scan].append(directory.path.joinpath(scan)) self.append_search_path(directory.path) def _filter_py_modules(self, path: Path) -> Iterator[Path]: for root, dirs, files in os.walk(path, topdown=True): # Optimization: modify dirs in-place to exclude `__pycache__` directories. dirs[:] = [dir for dir in dirs if dir != "__pycache__"] for relfile in files: if os.path.splitext(relfile)[1] in self.extensions_set: # noqa: PTH122 yield Path(root, relfile) def _top_module_name(self, path: Path) -> str: # First find if a parent is in search paths. parent_path = path if path.is_dir() else path.parent # Always resolve parent path to compare for relativeness against resolved search paths. parent_path = parent_path.resolve() for search_path in self.search_paths: with suppress(ValueError, IndexError): rel_path = parent_path.relative_to(search_path.resolve()) return rel_path.parts[0] # If not, get the highest directory with an `__init__` module, # add its parent to search paths and return it. while parent_path.parent != parent_path and (parent_path.parent / "__init__.py").exists(): parent_path = parent_path.parent self.insert_search_path(0, parent_path.parent) return parent_path.name _re_pkgresources = re.compile(r"(?:__import__\([\"']pkg_resources[\"']\).declare_namespace\(__name__\))") _re_pkgutil = re.compile(r"(?:__path__ = __import__\([\"']pkgutil[\"']\).extend_path\(__path__, __name__\))") _re_import_line = re.compile(r"^import[ \t]+\w+$") # TODO: For more robustness, we should load and minify the AST # to search for particular call statements. def _is_pkg_style_namespace(init_module: Path) -> bool: code = init_module.read_text(encoding="utf8") return bool(_re_pkgresources.search(code) or _re_pkgutil.search(code)) def _module_depth(name_parts_and_path: NamePartsAndPathType) -> int: return len(name_parts_and_path[0]) @dataclass class _SP: path: Path always_scan_for: str = "" def _handle_pth_file(path: Path) -> list[_SP]: # Support for .pth files pointing to directories. # From https://docs.python.org/3/library/site.html: # A path configuration file is a file whose name has the form name.pth # and exists in one of the four directories mentioned above; # its contents are additional items (one per line) to be added to sys.path. # Non-existing items are never added to sys.path, # and no check is made that the item refers to a directory rather than a file. # No item is added to sys.path more than once. # Blank lines and lines beginning with # are skipped. # Lines starting with import (followed by space or tab) are executed. directories: list[_SP] = [] try: # It turns out PyTorch recommends its users to use `.pth` as the extension # when saving models on the disk. These model files are not encoded in UTF8. # If UTF8 decoding fails, we skip the .pth file. text = path.read_text(encoding="utf8") except UnicodeDecodeError: return directories for line in text.strip().replace(";", "\n").splitlines(keepends=False): line = line.strip() # noqa: PLW2901 if _re_import_line.match(line): editable_module = path.parent / f"{line[len('import'):].lstrip()}.py" with suppress(UnhandledEditableModuleError): return _handle_editable_module(editable_module) if line and not line.startswith("#") and os.path.exists(line): # noqa: PTH110 directories.append(_SP(Path(line))) return directories def _handle_editable_module(path: Path) -> list[_SP]: if _match_pattern(path.name, (*_editable_editables_patterns, *_editable_scikit_build_core_patterns)): # Support for how 'editables' write these files: # example line: `F.map_module('griffe', '/media/data/dev/griffe/src/griffe/__init__.py')`. # And how 'scikit-build-core' writes these files: # example line: `install({'griffe': '/media/data/dev/griffe/src/griffe/__init__.py'}, {'cmake_example': ...}, None, False, True)`. try: editable_lines = path.read_text(encoding="utf8").strip().splitlines(keepends=False) except FileNotFoundError as error: raise UnhandledEditableModuleError(path) from error new_path = Path(editable_lines[-1].split("'")[3]) if new_path.name.startswith("__init__"): return [_SP(new_path.parent.parent)] return [_SP(new_path)] if _match_pattern(path.name, _editable_setuptools_patterns): # Support for how 'setuptools' writes these files: # example line: `MAPPING = {'griffe': '/media/data/dev/griffe/src/griffe', 'briffe': '/media/data/dev/griffe/src/briffe'}`. # with annotation: `MAPPING: dict[str, str] = {...}`. parsed_module = ast.parse(path.read_text()) for node in parsed_module.body: if isinstance(node, ast.Assign): target = node.targets[0] elif isinstance(node, ast.AnnAssign): target = node.target else: continue if isinstance(target, ast.Name) and target.id == "MAPPING" and isinstance(node.value, ast.Dict): # type: ignore[attr-defined] return [_SP(Path(cst.value).parent) for cst in node.value.values if isinstance(cst, ast.Constant)] # type: ignore[attr-defined] if _match_pattern(path.name, _editable_meson_python_patterns): # Support for how 'meson-python' writes these files: # example line: `install({'package', 'module1'}, '/media/data/dev/griffe/build/cp311', ["path"], False)`. # Compiled modules then found in the cp311 folder, under src/package. parsed_module = ast.parse(path.read_text()) for node in parsed_module.body: if ( isinstance(node, ast.Expr) and isinstance(node.value, ast.Call) and isinstance(node.value.func, ast.Name) and node.value.func.id == "install" and isinstance(node.value.args[1], ast.Constant) ): build_path = Path(node.value.args[1].value, "src") # NOTE: What if there are multiple packages? pkg_name = next(build_path.iterdir()).name return [_SP(build_path, always_scan_for=pkg_name)] raise UnhandledEditableModuleError(path)