#!/usr/bin/env python3 """Generator of dynamically typed draft stubs for arbitrary modules. The logic of this script can be split in three steps: * parsing options and finding sources: - use runtime imports be default (to find also C modules) - or use mypy's mechanisms, if importing is prohibited * (optionally) semantically analysing the sources using mypy (as a single set) * emitting the stubs text: - for Python modules: from ASTs using ASTStubGenerator - for C modules using runtime introspection and (optionally) Sphinx docs During first and third steps some problematic files can be skipped, but any blocking error during second step will cause the whole program to stop. Basic usage: $ stubgen foo.py bar.py some_directory => Generate out/foo.pyi, out/bar.pyi, and stubs for some_directory (recursively). $ stubgen -m urllib.parse => Generate out/urllib/parse.pyi. $ stubgen -p urllib => Generate stubs for whole urllib package (recursively). For C modules, you can get more precise function signatures by parsing .rst (Sphinx) documentation for extra information. For this, use the --doc-dir option: $ stubgen --doc-dir /Python-3.4.2/Doc/library -m curses Note: The generated stubs should be verified manually. TODO: - maybe use .rst docs also for Python modules - maybe export more imported names if there is no __all__ (this affects ssl.SSLError, for example) - a quick and dirty heuristic would be to turn this on if a module has something like 'from x import y as _y' - we don't seem to always detect properties ('closed' in 'io', for example) """ from __future__ import annotations import argparse import keyword import os import os.path import sys import traceback from collections.abc import Iterable, Iterator from typing import Final import mypy.build import mypy.mixedtraverser import mypy.parse import mypy.traverser import mypy.util import mypy.version from mypy.build import build from mypy.errors import CompileError, Errors from mypy.find_sources import InvalidSourceList, create_source_list from mypy.modulefinder import ( BuildSource, FindModuleCache, ModuleNotFoundReason, SearchPaths, default_lib_path, ) from mypy.moduleinspect import ModuleInspect, is_pyc_only from mypy.nodes import ( ARG_NAMED, ARG_POS, ARG_STAR, ARG_STAR2, IS_ABSTRACT, NOT_ABSTRACT, AssignmentStmt, Block, BytesExpr, CallExpr, CastExpr, ClassDef, ComparisonExpr, ComplexExpr, ConditionalExpr, Decorator, DictExpr, DictionaryComprehension, EllipsisExpr, Expression, ExpressionStmt, FloatExpr, FuncBase, FuncDef, GeneratorExpr, IfStmt, Import, ImportAll, ImportFrom, IndexExpr, IntExpr, LambdaExpr, ListComprehension, ListExpr, MemberExpr, MypyFile, NameExpr, OpExpr, OverloadedFuncDef, SetComprehension, SetExpr, SliceExpr, StarExpr, Statement, StrExpr, TempNode, TupleExpr, TypeAliasStmt, TypeInfo, UnaryExpr, Var, ) from mypy.options import Options as MypyOptions from mypy.plugins.dataclasses import DATACLASS_FIELD_SPECIFIERS from mypy.semanal_shared import find_dataclass_transform_spec from mypy.sharedparse import MAGIC_METHODS_POS_ARGS_ONLY from mypy.stubdoc import ArgSig, FunctionSig from mypy.stubgenc import InspectionStubGenerator, generate_stub_for_c_module from mypy.stubutil import ( TYPING_BUILTIN_REPLACEMENTS, BaseStubGenerator, CantImport, ClassInfo, FunctionContext, common_dir_prefix, fail_missing, find_module_path_and_all_py3, generate_guarded, infer_method_arg_types, infer_method_ret_type, remove_misplaced_type_comments, report_missing, walk_packages, ) from mypy.traverser import ( all_yield_expressions, has_return_statement, has_yield_expression, has_yield_from_expression, ) from mypy.types import ( DATACLASS_TRANSFORM_NAMES, OVERLOAD_NAMES, TPDICT_NAMES, TYPE_VAR_LIKE_NAMES, TYPED_NAMEDTUPLE_NAMES, AnyType, CallableType, Instance, TupleType, Type, UnboundType, get_proper_type, ) from mypy.visitor import NodeVisitor # Common ways of naming package containing vendored modules. VENDOR_PACKAGES: Final = ["packages", "vendor", "vendored", "_vendor", "_vendored_packages"] # Avoid some file names that are unnecessary or likely to cause trouble (\n for end of path). BLACKLIST: Final = [ "/six.py\n", # Likely vendored six; too dynamic for us to handle "/vendored/", # Vendored packages "/vendor/", # Vendored packages "/_vendor/", "/_vendored_packages/", ] # These methods are expected to always return a non-trivial value. METHODS_WITH_RETURN_VALUE: Final = { "__ne__", "__eq__", "__lt__", "__le__", "__gt__", "__ge__", "__hash__", "__iter__", } class Options: """Represents stubgen options. This class is mutable to simplify testing. """ def __init__( self, pyversion: tuple[int, int], no_import: bool, inspect: bool, doc_dir: str, search_path: list[str], interpreter: str, parse_only: bool, ignore_errors: bool, include_private: bool, output_dir: str, modules: list[str], packages: list[str], files: list[str], verbose: bool, quiet: bool, export_less: bool, include_docstrings: bool, ) -> None: # See parse_options for descriptions of the flags. self.pyversion = pyversion self.no_import = no_import self.inspect = inspect self.doc_dir = doc_dir self.search_path = search_path self.interpreter = interpreter self.decointerpreter = interpreter self.parse_only = parse_only self.ignore_errors = ignore_errors self.include_private = include_private self.output_dir = output_dir self.modules = modules self.packages = packages self.files = files self.verbose = verbose self.quiet = quiet self.export_less = export_less self.include_docstrings = include_docstrings class StubSource: """A single source for stub: can be a Python or C module. A simple extension of BuildSource that also carries the AST and the value of __all__ detected at runtime. """ def __init__( self, module: str, path: str | None = None, runtime_all: list[str] | None = None ) -> None: self.source = BuildSource(path, module, None) self.runtime_all = runtime_all self.ast: MypyFile | None = None def __repr__(self) -> str: return f"StubSource({self.source})" @property def module(self) -> str: return self.source.module @property def path(self) -> str | None: return self.source.path # What was generated previously in the stub file. We keep track of these to generate # nicely formatted output (add empty line between non-empty classes, for example). EMPTY: Final = "EMPTY" FUNC: Final = "FUNC" CLASS: Final = "CLASS" EMPTY_CLASS: Final = "EMPTY_CLASS" VAR: Final = "VAR" NOT_IN_ALL: Final = "NOT_IN_ALL" # Indicates that we failed to generate a reasonable output # for a given node. These should be manually replaced by a user. ERROR_MARKER: Final = "" class AliasPrinter(NodeVisitor[str]): """Visitor used to collect type aliases _and_ type variable definitions. Visit r.h.s of the definition to get the string representation of type alias. """ def __init__(self, stubgen: ASTStubGenerator) -> None: self.stubgen = stubgen super().__init__() def visit_call_expr(self, node: CallExpr) -> str: # Call expressions are not usually types, but we also treat `X = TypeVar(...)` as a # type alias that has to be preserved (even if TypeVar is not the same as an alias) callee = node.callee.accept(self) args = [] for name, arg, kind in zip(node.arg_names, node.args, node.arg_kinds): if kind == ARG_POS: args.append(arg.accept(self)) elif kind == ARG_STAR: args.append("*" + arg.accept(self)) elif kind == ARG_STAR2: args.append("**" + arg.accept(self)) elif kind == ARG_NAMED: args.append(f"{name}={arg.accept(self)}") else: raise ValueError(f"Unknown argument kind {kind} in call") return f"{callee}({', '.join(args)})" def _visit_ref_expr(self, node: NameExpr | MemberExpr) -> str: fullname = self.stubgen.get_fullname(node) if fullname in TYPING_BUILTIN_REPLACEMENTS: return self.stubgen.add_name(TYPING_BUILTIN_REPLACEMENTS[fullname], require=False) qualname = get_qualified_name(node) self.stubgen.import_tracker.require_name(qualname) return qualname def visit_name_expr(self, node: NameExpr) -> str: return self._visit_ref_expr(node) def visit_member_expr(self, o: MemberExpr) -> str: return self._visit_ref_expr(o) def _visit_literal_node( self, node: StrExpr | BytesExpr | IntExpr | FloatExpr | ComplexExpr ) -> str: return repr(node.value) def visit_str_expr(self, node: StrExpr) -> str: return self._visit_literal_node(node) def visit_bytes_expr(self, node: BytesExpr) -> str: return f"b{self._visit_literal_node(node)}" def visit_int_expr(self, node: IntExpr) -> str: return self._visit_literal_node(node) def visit_float_expr(self, node: FloatExpr) -> str: return self._visit_literal_node(node) def visit_complex_expr(self, node: ComplexExpr) -> str: return self._visit_literal_node(node) def visit_index_expr(self, node: IndexExpr) -> str: base_fullname = self.stubgen.get_fullname(node.base) if base_fullname == "typing.Union": if isinstance(node.index, TupleExpr): return " | ".join([item.accept(self) for item in node.index.items]) return node.index.accept(self) if base_fullname == "typing.Optional": if isinstance(node.index, TupleExpr): return self.stubgen.add_name("_typeshed.Incomplete") return f"{node.index.accept(self)} | None" base = node.base.accept(self) index = node.index.accept(self) if len(index) > 2 and index.startswith("(") and index.endswith(")"): index = index[1:-1].rstrip(",") return f"{base}[{index}]" def visit_tuple_expr(self, node: TupleExpr) -> str: suffix = "," if len(node.items) == 1 else "" return f"({', '.join(n.accept(self) for n in node.items)}{suffix})" def visit_list_expr(self, node: ListExpr) -> str: return f"[{', '.join(n.accept(self) for n in node.items)}]" def visit_set_expr(self, node: SetExpr) -> str: return f"{{{', '.join(n.accept(self) for n in node.items)}}}" def visit_dict_expr(self, o: DictExpr) -> str: dict_items = [] for key, value in o.items: # This is currently only used for TypedDict where all keys are strings. assert isinstance(key, StrExpr) dict_items.append(f"{key.accept(self)}: {value.accept(self)}") return f"{{{', '.join(dict_items)}}}" def visit_ellipsis(self, node: EllipsisExpr) -> str: return "..." def visit_op_expr(self, o: OpExpr) -> str: return f"{o.left.accept(self)} {o.op} {o.right.accept(self)}" def visit_unary_expr(self, o: UnaryExpr, /) -> str: return f"{o.op}{o.expr.accept(self)}" def visit_slice_expr(self, o: SliceExpr, /) -> str: blocks = [ o.begin_index.accept(self) if o.begin_index is not None else "", o.end_index.accept(self) if o.end_index is not None else "", ] if o.stride is not None: blocks.append(o.stride.accept(self)) return ":".join(blocks) def visit_star_expr(self, o: StarExpr) -> str: return f"*{o.expr.accept(self)}" def visit_lambda_expr(self, o: LambdaExpr) -> str: # TODO: Required for among other things dataclass.field default_factory return self.stubgen.add_name("_typeshed.Incomplete") def _visit_unsupported_expr(self, o: object) -> str: # Something we do not understand. return self.stubgen.add_name("_typeshed.Incomplete") def visit_comparison_expr(self, o: ComparisonExpr) -> str: return self._visit_unsupported_expr(o) def visit_cast_expr(self, o: CastExpr) -> str: return self._visit_unsupported_expr(o) def visit_conditional_expr(self, o: ConditionalExpr) -> str: return self._visit_unsupported_expr(o) def visit_list_comprehension(self, o: ListComprehension) -> str: return self._visit_unsupported_expr(o) def visit_set_comprehension(self, o: SetComprehension) -> str: return self._visit_unsupported_expr(o) def visit_dictionary_comprehension(self, o: DictionaryComprehension) -> str: return self._visit_unsupported_expr(o) def visit_generator_expr(self, o: GeneratorExpr) -> str: return self._visit_unsupported_expr(o) def find_defined_names(file: MypyFile) -> set[str]: finder = DefinitionFinder() file.accept(finder) return finder.names def get_assigned_names(lvalues: Iterable[Expression]) -> Iterator[str]: for lvalue in lvalues: if isinstance(lvalue, NameExpr): yield lvalue.name elif isinstance(lvalue, TupleExpr): yield from get_assigned_names(lvalue.items) class DefinitionFinder(mypy.traverser.TraverserVisitor): """Find names of things defined at the top level of a module.""" def __init__(self) -> None: # Short names of things defined at the top level. self.names: set[str] = set() def visit_class_def(self, o: ClassDef) -> None: # Don't recurse into classes, as we only keep track of top-level definitions. self.names.add(o.name) def visit_func_def(self, o: FuncDef) -> None: # Don't recurse, as we only keep track of top-level definitions. self.names.add(o.name) def visit_assignment_stmt(self, o: AssignmentStmt) -> None: for name in get_assigned_names(o.lvalues): self.names.add(name) def visit_type_alias_stmt(self, o: TypeAliasStmt) -> None: self.names.add(o.name.name) def find_referenced_names(file: MypyFile) -> set[str]: finder = ReferenceFinder() file.accept(finder) return finder.refs def is_none_expr(expr: Expression) -> bool: return isinstance(expr, NameExpr) and expr.name == "None" class ReferenceFinder(mypy.mixedtraverser.MixedTraverserVisitor): """Find all name references (both local and global).""" # TODO: Filter out local variable and class attribute references def __init__(self) -> None: # Short names of things defined at the top level. self.refs: set[str] = set() def visit_block(self, block: Block) -> None: if not block.is_unreachable: super().visit_block(block) def visit_name_expr(self, e: NameExpr) -> None: self.refs.add(e.name) def visit_instance(self, t: Instance) -> None: self.add_ref(t.type.name) super().visit_instance(t) def visit_unbound_type(self, t: UnboundType) -> None: if t.name: self.add_ref(t.name) def visit_tuple_type(self, t: TupleType) -> None: # Ignore fallback for item in t.items: item.accept(self) def visit_callable_type(self, t: CallableType) -> None: # Ignore fallback for arg in t.arg_types: arg.accept(self) t.ret_type.accept(self) def add_ref(self, fullname: str) -> None: self.refs.add(fullname) while "." in fullname: fullname = fullname.rsplit(".", 1)[0] self.refs.add(fullname) class ASTStubGenerator(BaseStubGenerator, mypy.traverser.TraverserVisitor): """Generate stub text from a mypy AST.""" def __init__( self, _all_: list[str] | None = None, include_private: bool = False, analyzed: bool = False, export_less: bool = False, include_docstrings: bool = False, ) -> None: super().__init__(_all_, include_private, export_less, include_docstrings) self._decorators: list[str] = [] # Stack of defined variables (per scope). self._vars: list[list[str]] = [[]] # What was generated previously in the stub file. self._state = EMPTY self._class_stack: list[ClassDef] = [] # Was the tree semantically analysed before? self.analyzed = analyzed # Short names of methods defined in the body of the current class self.method_names: set[str] = set() self.processing_enum = False self.processing_dataclass = False self.dataclass_field_specifier: tuple[str, ...] = () @property def _current_class(self) -> ClassDef | None: return self._class_stack[-1] if self._class_stack else None def visit_mypy_file(self, o: MypyFile) -> None: self.module_name = o.fullname # Current module being processed self.path = o.path self.set_defined_names(find_defined_names(o)) self.referenced_names = find_referenced_names(o) super().visit_mypy_file(o) self.check_undefined_names() def visit_overloaded_func_def(self, o: OverloadedFuncDef) -> None: """@property with setters and getters, @overload chain and some others.""" overload_chain = False for item in o.items: if not isinstance(item, Decorator): continue if self.is_private_name(item.func.name, item.func.fullname): continue self.process_decorator(item) if not overload_chain: self.visit_func_def(item.func) if item.func.is_overload: overload_chain = True elif item.func.is_overload: self.visit_func_def(item.func) else: # skip the overload implementation and clear the decorator we just processed self.clear_decorators() def get_default_function_sig(self, func_def: FuncDef, ctx: FunctionContext) -> FunctionSig: args = self._get_func_args(func_def, ctx) retname = self._get_func_return(func_def, ctx) type_args = self.format_type_args(func_def) return FunctionSig(func_def.name, args, retname, type_args) def _get_func_args(self, o: FuncDef, ctx: FunctionContext) -> list[ArgSig]: args: list[ArgSig] = [] # Ignore pos-only status of magic methods whose args names are elided by mypy at parse actually_pos_only_args = o.name not in MAGIC_METHODS_POS_ARGS_ONLY pos_only_marker_position = 0 # Where to insert "/", if any for i, arg_ in enumerate(o.arguments): var = arg_.variable kind = arg_.kind name = var.name annotated_type = ( o.unanalyzed_type.arg_types[i] if isinstance(o.unanalyzed_type, CallableType) else None ) # I think the name check is incorrect: there are libraries which # name their 0th argument other than self/cls is_self_arg = i == 0 and name == "self" is_cls_arg = i == 0 and name == "cls" typename: str | None = None if annotated_type and not is_self_arg and not is_cls_arg: # Luckily, an argument explicitly annotated with "Any" has # type "UnboundType" and will not match. if not isinstance(get_proper_type(annotated_type), AnyType): typename = self.print_annotation(annotated_type) if actually_pos_only_args and arg_.pos_only: pos_only_marker_position += 1 if kind.is_named() and not any(arg.name.startswith("*") for arg in args): args.append(ArgSig("*")) default = "..." if arg_.initializer: if not typename: typename = self.get_str_type_of_node(arg_.initializer, can_be_incomplete=False) potential_default, valid = self.get_str_default_of_node(arg_.initializer) if valid and len(potential_default) <= 200: default = potential_default elif kind == ARG_STAR: name = f"*{name}" elif kind == ARG_STAR2: name = f"**{name}" args.append( ArgSig(name, typename, default=bool(arg_.initializer), default_value=default) ) if pos_only_marker_position: args.insert(pos_only_marker_position, ArgSig("/")) if ctx.class_info is not None and all( arg.type is None and arg.default is False for arg in args ): new_args = infer_method_arg_types( ctx.name, ctx.class_info.self_var, [arg.name for arg in args] ) if ctx.name == "__exit__": self.import_tracker.add_import("types") self.import_tracker.require_name("types") if new_args is not None: args = new_args return args def _get_func_return(self, o: FuncDef, ctx: FunctionContext) -> str | None: if o.name != "__init__" and isinstance(o.unanalyzed_type, CallableType): if isinstance(get_proper_type(o.unanalyzed_type.ret_type), AnyType): # Luckily, a return type explicitly annotated with "Any" has # type "UnboundType" and will enter the else branch. return None # implicit Any else: return self.print_annotation(o.unanalyzed_type.ret_type) if o.abstract_status == IS_ABSTRACT or o.name in METHODS_WITH_RETURN_VALUE: # Always assume abstract methods return Any unless explicitly annotated. Also # some dunder methods should not have a None return type. return None # implicit Any retname = infer_method_ret_type(o.name) if retname is not None: return retname if has_yield_expression(o) or has_yield_from_expression(o): generator_name = self.add_name("collections.abc.Generator") yield_name = "None" send_name: str | None = None return_name: str | None = None if has_yield_from_expression(o): yield_name = send_name = self.add_name("_typeshed.Incomplete") else: for expr, in_assignment in all_yield_expressions(o): if expr.expr is not None and not is_none_expr(expr.expr): yield_name = self.add_name("_typeshed.Incomplete") if in_assignment: send_name = self.add_name("_typeshed.Incomplete") if has_return_statement(o): return_name = self.add_name("_typeshed.Incomplete") if return_name is not None: if send_name is None: send_name = "None" return f"{generator_name}[{yield_name}, {send_name}, {return_name}]" elif send_name is not None: return f"{generator_name}[{yield_name}, {send_name}]" else: return f"{generator_name}[{yield_name}]" if not has_return_statement(o) and o.abstract_status == NOT_ABSTRACT: return "None" return None def _get_func_docstring(self, node: FuncDef) -> str | None: if not node.body.body: return None expr = node.body.body[0] if isinstance(expr, ExpressionStmt) and isinstance(expr.expr, StrExpr): return expr.expr.value return None def visit_func_def(self, o: FuncDef) -> None: is_dataclass_generated = ( self.analyzed and self.processing_dataclass and o.info.names[o.name].plugin_generated ) if is_dataclass_generated: # Skip methods generated by the @dataclass decorator return if ( self.is_private_name(o.name, o.fullname) or self.is_not_in_all(o.name) or (self.is_recorded_name(o.name) and not o.is_overload) ): self.clear_decorators() return if self.is_top_level() and self._state not in (EMPTY, FUNC): self.add("\n") if not self.is_top_level(): self_inits = find_self_initializers(o) for init, value, annotation in self_inits: if init in self.method_names: # Can't have both an attribute and a method/property with the same name. continue init_code = self.get_init(init, value, annotation) if init_code: self.add(init_code) if self._class_stack: if len(o.arguments): self_var = o.arguments[0].variable.name else: self_var = "self" class_info: ClassInfo | None = None for class_def in self._class_stack: class_info = ClassInfo(class_def.name, self_var, parent=class_info) else: class_info = None ctx = FunctionContext( module_name=self.module_name, name=o.name, docstring=self._get_func_docstring(o), is_abstract=o.abstract_status != NOT_ABSTRACT, class_info=class_info, ) self.record_name(o.name) default_sig = self.get_default_function_sig(o, ctx) sigs = self.get_signatures(default_sig, self.sig_generators, ctx) for output in self.format_func_def( sigs, is_coroutine=o.is_coroutine, decorators=self._decorators, docstring=ctx.docstring ): self.add(output + "\n") self.clear_decorators() self._state = FUNC def visit_decorator(self, o: Decorator) -> None: if self.is_private_name(o.func.name, o.func.fullname): return self.process_decorator(o) self.visit_func_def(o.func) def process_decorator(self, o: Decorator) -> None: """Process a series of decorators. Only preserve certain special decorators such as @abstractmethod. """ o.func.is_overload = False for decorator in o.original_decorators: d = decorator if isinstance(d, CallExpr): d = d.callee if not isinstance(d, (NameExpr, MemberExpr)): continue qualname = get_qualified_name(d) fullname = self.get_fullname(d) if fullname in ( "builtins.property", "builtins.staticmethod", "builtins.classmethod", "functools.cached_property", ): self.add_decorator(qualname, require_name=True) elif fullname in ( "asyncio.coroutine", "asyncio.coroutines.coroutine", "types.coroutine", ): o.func.is_awaitable_coroutine = True self.add_decorator(qualname, require_name=True) elif fullname == "abc.abstractmethod": self.add_decorator(qualname, require_name=True) o.func.abstract_status = IS_ABSTRACT elif fullname in ( "abc.abstractproperty", "abc.abstractstaticmethod", "abc.abstractclassmethod", ): abc_module = qualname.rpartition(".")[0] if not abc_module: self.import_tracker.add_import("abc") builtin_decorator_replacement = fullname[len("abc.abstract") :] self.add_decorator(builtin_decorator_replacement, require_name=False) self.add_decorator(f"{abc_module or 'abc'}.abstractmethod", require_name=True) o.func.abstract_status = IS_ABSTRACT elif fullname in OVERLOAD_NAMES: self.add_decorator(qualname, require_name=True) o.func.is_overload = True elif qualname.endswith((".setter", ".deleter")): self.add_decorator(qualname, require_name=False) elif fullname in DATACLASS_TRANSFORM_NAMES: p = AliasPrinter(self) self._decorators.append(f"@{decorator.accept(p)}") elif isinstance(decorator, (NameExpr, MemberExpr)): p = AliasPrinter(self) self._decorators.append(f"@{decorator.accept(p)}") def get_fullname(self, expr: Expression) -> str: """Return the expression's full name.""" if ( self.analyzed and isinstance(expr, (NameExpr, MemberExpr)) and expr.fullname and not (isinstance(expr.node, Var) and expr.node.is_suppressed_import) ): return expr.fullname name = get_qualified_name(expr) return self.resolve_name(name) def visit_class_def(self, o: ClassDef) -> None: self._class_stack.append(o) self.method_names = find_method_names(o.defs.body) sep: int | None = None if self.is_top_level() and self._state != EMPTY: sep = len(self._output) self.add("\n") decorators = self.get_class_decorators(o) for d in decorators: self.add(f"{self._indent}@{d}\n") self.record_name(o.name) base_types = self.get_base_types(o) if base_types: for base in base_types: self.import_tracker.require_name(base) if self.analyzed and o.info.is_enum: self.processing_enum = True if isinstance(o.metaclass, (NameExpr, MemberExpr)): meta = o.metaclass.accept(AliasPrinter(self)) base_types.append("metaclass=" + meta) elif self.analyzed and o.info.is_abstract and not o.info.is_protocol: base_types.append("metaclass=abc.ABCMeta") self.import_tracker.add_import("abc") self.import_tracker.require_name("abc") bases = f"({', '.join(base_types)})" if base_types else "" type_args = self.format_type_args(o) self.add(f"{self._indent}class {o.name}{type_args}{bases}:\n") self.indent() if self._include_docstrings and o.docstring: docstring = mypy.util.quote_docstring(o.docstring) self.add(f"{self._indent}{docstring}\n") n = len(self._output) self._vars.append([]) if self.analyzed and (spec := find_dataclass_transform_spec(o)): self.processing_dataclass = True self.dataclass_field_specifier = spec.field_specifiers super().visit_class_def(o) self.dedent() self._vars.pop() self._vars[-1].append(o.name) if len(self._output) == n: if self._state == EMPTY_CLASS and sep is not None: self._output[sep] = "" if not (self._include_docstrings and o.docstring): self._output[-1] = self._output[-1][:-1] + " ...\n" self._state = EMPTY_CLASS else: self._state = CLASS self.method_names = set() self.processing_dataclass = False self.dataclass_field_specifier = () self._class_stack.pop(-1) self.processing_enum = False def get_base_types(self, cdef: ClassDef) -> list[str]: """Get list of base classes for a class.""" base_types: list[str] = [] p = AliasPrinter(self) for base in cdef.base_type_exprs + cdef.removed_base_type_exprs: if isinstance(base, (NameExpr, MemberExpr)): if self.get_fullname(base) != "builtins.object": base_types.append(get_qualified_name(base)) elif isinstance(base, IndexExpr): base_types.append(base.accept(p)) elif isinstance(base, CallExpr): # namedtuple(typename, fields), NamedTuple(typename, fields) calls can # be used as a base class. The first argument is a string literal that # is usually the same as the class name. # # Note: # A call-based named tuple as a base class cannot be safely converted to # a class-based NamedTuple definition because class attributes defined # in the body of the class inheriting from the named tuple call are not # namedtuple fields at runtime. if self.is_namedtuple(base): nt_fields = self._get_namedtuple_fields(base) assert isinstance(base.args[0], StrExpr) typename = base.args[0].value if nt_fields is None: # Invalid namedtuple() call, cannot determine fields base_types.append(self.add_name("_typeshed.Incomplete")) continue fields_str = ", ".join(f"({f!r}, {t})" for f, t in nt_fields) namedtuple_name = self.add_name("typing.NamedTuple") base_types.append(f"{namedtuple_name}({typename!r}, [{fields_str}])") elif self.is_typed_namedtuple(base): base_types.append(base.accept(p)) else: # At this point, we don't know what the base class is, so we # just use Incomplete as the base class. base_types.append(self.add_name("_typeshed.Incomplete")) for name, value in cdef.keywords.items(): if name == "metaclass": continue # handled separately processed_value = value.accept(p) or "..." # at least, don't crash base_types.append(f"{name}={processed_value}") return base_types def get_class_decorators(self, cdef: ClassDef) -> list[str]: decorators: list[str] = [] p = AliasPrinter(self) for d in cdef.decorators: if self.is_dataclass(d): decorators.append(d.accept(p)) self.import_tracker.require_name(get_qualified_name(d)) self.processing_dataclass = True if self.is_dataclass_transform(d): decorators.append(d.accept(p)) self.import_tracker.require_name(get_qualified_name(d)) return decorators def is_dataclass(self, expr: Expression) -> bool: if isinstance(expr, CallExpr): expr = expr.callee return self.get_fullname(expr) == "dataclasses.dataclass" def is_dataclass_transform(self, expr: Expression) -> bool: if isinstance(expr, CallExpr): expr = expr.callee if self.get_fullname(expr) in DATACLASS_TRANSFORM_NAMES: return True if (spec := find_dataclass_transform_spec(expr)) is not None: self.processing_dataclass = True self.dataclass_field_specifier = spec.field_specifiers return True return False def visit_block(self, o: Block) -> None: # Unreachable statements may be partially uninitialized and that may # cause trouble. if not o.is_unreachable: super().visit_block(o) def visit_assignment_stmt(self, o: AssignmentStmt) -> None: foundl = [] for lvalue in o.lvalues: if isinstance(lvalue, NameExpr) and isinstance(o.rvalue, CallExpr): if self.is_namedtuple(o.rvalue) or self.is_typed_namedtuple(o.rvalue): self.process_namedtuple(lvalue, o.rvalue) foundl.append(False) # state is updated in process_namedtuple continue if self.is_typeddict(o.rvalue): self.process_typeddict(lvalue, o.rvalue) foundl.append(False) # state is updated in process_typeddict continue if ( isinstance(lvalue, NameExpr) and self.is_alias_expression(o.rvalue) and not self.is_private_name(lvalue.name) ): is_explicit_type_alias = ( o.unanalyzed_type and getattr(o.type, "name", None) == "TypeAlias" ) if is_explicit_type_alias: self.process_typealias(lvalue, o.rvalue, is_explicit_type_alias=True) continue if not o.unanalyzed_type: self.process_typealias(lvalue, o.rvalue) continue if isinstance(lvalue, (TupleExpr, ListExpr)): items = lvalue.items if isinstance(o.unanalyzed_type, TupleType): # type: ignore[misc] annotations: Iterable[Type | None] = o.unanalyzed_type.items else: annotations = [None] * len(items) else: items = [lvalue] annotations = [o.unanalyzed_type] sep = False found = False for item, annotation in zip(items, annotations): if isinstance(item, NameExpr): init = self.get_init(item.name, o.rvalue, annotation) if init: found = True if not sep and self.is_top_level() and self._state not in (EMPTY, VAR): init = "\n" + init sep = True self.add(init) self.record_name(item.name) foundl.append(found) if all(foundl): self._state = VAR def is_namedtuple(self, expr: CallExpr) -> bool: return self.get_fullname(expr.callee) == "collections.namedtuple" def is_typed_namedtuple(self, expr: CallExpr) -> bool: return self.get_fullname(expr.callee) in TYPED_NAMEDTUPLE_NAMES def _get_namedtuple_fields(self, call: CallExpr) -> list[tuple[str, str]] | None: if self.is_namedtuple(call): fields_arg = call.args[1] if isinstance(fields_arg, StrExpr): field_names = fields_arg.value.replace(",", " ").split() elif isinstance(fields_arg, (ListExpr, TupleExpr)): field_names = [] for field in fields_arg.items: if not isinstance(field, StrExpr): return None field_names.append(field.value) else: return None # Invalid namedtuple fields type if field_names: incomplete = self.add_name("_typeshed.Incomplete") return [(field_name, incomplete) for field_name in field_names] else: return [] elif self.is_typed_namedtuple(call): fields_arg = call.args[1] if not isinstance(fields_arg, (ListExpr, TupleExpr)): return None fields: list[tuple[str, str]] = [] p = AliasPrinter(self) for field in fields_arg.items: if not (isinstance(field, TupleExpr) and len(field.items) == 2): return None field_name, field_type = field.items if not isinstance(field_name, StrExpr): return None fields.append((field_name.value, field_type.accept(p))) return fields else: return None # Not a named tuple call def process_namedtuple(self, lvalue: NameExpr, rvalue: CallExpr) -> None: if self._state == CLASS: self.add("\n") if not isinstance(rvalue.args[0], StrExpr): self.annotate_as_incomplete(lvalue) return fields = self._get_namedtuple_fields(rvalue) if fields is None: self.annotate_as_incomplete(lvalue) return bases = self.add_name("typing.NamedTuple") # TODO: Add support for generic NamedTuples. Requires `Generic` as base class. class_def = f"{self._indent}class {lvalue.name}({bases}):" if len(fields) == 0: self.add(f"{class_def} ...\n") self._state = EMPTY_CLASS else: if self._state not in (EMPTY, CLASS): self.add("\n") self.add(f"{class_def}\n") for f_name, f_type in fields: self.add(f"{self._indent} {f_name}: {f_type}\n") self._state = CLASS def is_typeddict(self, expr: CallExpr) -> bool: return self.get_fullname(expr.callee) in TPDICT_NAMES def process_typeddict(self, lvalue: NameExpr, rvalue: CallExpr) -> None: if self._state == CLASS: self.add("\n") if not isinstance(rvalue.args[0], StrExpr): self.annotate_as_incomplete(lvalue) return items: list[tuple[str, Expression]] = [] total: Expression | None = None if len(rvalue.args) > 1 and rvalue.arg_kinds[1] == ARG_POS: if not isinstance(rvalue.args[1], DictExpr): self.annotate_as_incomplete(lvalue) return for attr_name, attr_type in rvalue.args[1].items: if not isinstance(attr_name, StrExpr): self.annotate_as_incomplete(lvalue) return items.append((attr_name.value, attr_type)) if len(rvalue.args) > 2: if rvalue.arg_kinds[2] != ARG_NAMED or rvalue.arg_names[2] != "total": self.annotate_as_incomplete(lvalue) return total = rvalue.args[2] else: for arg_name, arg in zip(rvalue.arg_names[1:], rvalue.args[1:]): if not isinstance(arg_name, str): self.annotate_as_incomplete(lvalue) return if arg_name == "total": total = arg else: items.append((arg_name, arg)) p = AliasPrinter(self) if any(not key.isidentifier() or keyword.iskeyword(key) for key, _ in items): # Keep the call syntax if there are non-identifier or reserved keyword keys. self.add(f"{self._indent}{lvalue.name} = {rvalue.accept(p)}\n") self._state = VAR else: bases = self.add_name("typing_extensions.TypedDict") # TODO: Add support for generic TypedDicts. Requires `Generic` as base class. if total is not None: bases += f", total={total.accept(p)}" class_def = f"{self._indent}class {lvalue.name}({bases}):" if len(items) == 0: self.add(f"{class_def} ...\n") self._state = EMPTY_CLASS else: if self._state not in (EMPTY, CLASS): self.add("\n") self.add(f"{class_def}\n") for key, key_type in items: self.add(f"{self._indent} {key}: {key_type.accept(p)}\n") self._state = CLASS def annotate_as_incomplete(self, lvalue: NameExpr) -> None: incomplete = self.add_name("_typeshed.Incomplete") self.add(f"{self._indent}{lvalue.name}: {incomplete}\n") self._state = VAR def is_alias_expression(self, expr: Expression, top_level: bool = True) -> bool: """Return True for things that look like target for an alias. Used to know if assignments look like type aliases, function alias, or module alias. """ # Assignment of TypeVar(...) and other typevar-likes are passed through if isinstance(expr, CallExpr) and self.get_fullname(expr.callee) in TYPE_VAR_LIKE_NAMES: return True elif isinstance(expr, EllipsisExpr): return not top_level elif isinstance(expr, NameExpr): if expr.name in ("True", "False"): return False elif expr.name == "None": return not top_level else: return not self.is_private_name(expr.name) elif isinstance(expr, MemberExpr) and self.analyzed: # Also add function and module aliases. return ( top_level and isinstance(expr.node, (FuncDef, Decorator, MypyFile)) or isinstance(expr.node, TypeInfo) ) and not self.is_private_member(expr.node.fullname) elif isinstance(expr, IndexExpr) and ( (isinstance(expr.base, NameExpr) and not self.is_private_name(expr.base.name)) or ( # Also some known aliases that could be member expression isinstance(expr.base, MemberExpr) and not self.is_private_member(get_qualified_name(expr.base)) and self.get_fullname(expr.base).startswith( ("builtins.", "typing.", "typing_extensions.", "collections.abc.") ) ) ): if isinstance(expr.index, TupleExpr): indices = expr.index.items else: indices = [expr.index] if expr.base.name == "Callable" and len(indices) == 2: args, ret = indices if isinstance(args, EllipsisExpr): indices = [ret] elif isinstance(args, ListExpr): indices = args.items + [ret] else: return False return all(self.is_alias_expression(i, top_level=False) for i in indices) elif isinstance(expr, OpExpr) and expr.op == "|": return self.is_alias_expression( expr.left, top_level=False ) and self.is_alias_expression(expr.right, top_level=False) else: return False def process_typealias( self, lvalue: NameExpr, rvalue: Expression, is_explicit_type_alias: bool = False ) -> None: p = AliasPrinter(self) if is_explicit_type_alias: self.import_tracker.require_name("TypeAlias") self.add(f"{self._indent}{lvalue.name}: TypeAlias = {rvalue.accept(p)}\n") else: self.add(f"{self._indent}{lvalue.name} = {rvalue.accept(p)}\n") self.record_name(lvalue.name) self._vars[-1].append(lvalue.name) def visit_type_alias_stmt(self, o: TypeAliasStmt) -> None: """Type aliases defined with the `type` keyword (PEP 695).""" p = AliasPrinter(self) name = o.name.name rvalue = o.value.expr() type_args = self.format_type_args(o) self.add(f"{self._indent}type {name}{type_args} = {rvalue.accept(p)}\n") self.record_name(name) self._vars[-1].append(name) def visit_if_stmt(self, o: IfStmt) -> None: # Ignore if __name__ == '__main__'. expr = o.expr[0] if ( isinstance(expr, ComparisonExpr) and isinstance(expr.operands[0], NameExpr) and isinstance(expr.operands[1], StrExpr) and expr.operands[0].name == "__name__" and "__main__" in expr.operands[1].value ): return super().visit_if_stmt(o) def visit_import_all(self, o: ImportAll) -> None: self.add_import_line(f"from {'.' * o.relative}{o.id} import *\n") def visit_import_from(self, o: ImportFrom) -> None: exported_names: set[str] = set() import_names = [] module, relative = translate_module_name(o.id, o.relative) if self.module_name: full_module, ok = mypy.util.correct_relative_import( self.module_name, relative, module, self.path.endswith(".__init__.py") ) if not ok: full_module = module else: full_module = module if module == "__future__": return # Not preserved for name, as_name in o.names: if name == "six": # Vendored six -- translate into plain 'import six'. self.visit_import(Import([("six", None)])) continue if self.should_reexport(name, full_module, as_name is not None): self.import_tracker.reexport(name) as_name = name import_names.append((name, as_name)) self.import_tracker.add_import_from("." * relative + module, import_names) self._vars[-1].extend(alias or name for name, alias in import_names) for name, alias in import_names: self.record_name(alias or name) if self._all_: # Include "import from"s that import names defined in __all__. names = [ name for name, alias in o.names if name in self._all_ and alias is None and name not in self.IGNORED_DUNDERS ] exported_names.update(names) def visit_import(self, o: Import) -> None: for id, as_id in o.ids: self.import_tracker.add_import(id, as_id) if as_id is None: target_name = id.split(".")[0] else: target_name = as_id self._vars[-1].append(target_name) self.record_name(target_name) def get_init( self, lvalue: str, rvalue: Expression, annotation: Type | None = None ) -> str | None: """Return initializer for a variable. Return None if we've generated one already or if the variable is internal. """ if lvalue in self._vars[-1]: # We've generated an initializer already for this variable. return None # TODO: Only do this at module top level. if self.is_private_name(lvalue) or self.is_not_in_all(lvalue): return None self._vars[-1].append(lvalue) if annotation is not None: typename = self.print_annotation(annotation) if ( isinstance(annotation, UnboundType) and not annotation.args and annotation.name == "Final" and self.import_tracker.module_for.get("Final") in self.TYPING_MODULE_NAMES ): # Final without type argument is invalid in stubs. final_arg = self.get_str_type_of_node(rvalue) typename += f"[{final_arg}]" elif self.processing_enum: initializer, _ = self.get_str_default_of_node(rvalue) return f"{self._indent}{lvalue} = {initializer}\n" elif self.processing_dataclass: # attribute without annotation is not a dataclass field, don't add annotation. return f"{self._indent}{lvalue} = ...\n" else: typename = self.get_str_type_of_node(rvalue) initializer = self.get_assign_initializer(rvalue) return f"{self._indent}{lvalue}: {typename}{initializer}\n" def get_assign_initializer(self, rvalue: Expression) -> str: """Does this rvalue need some special initializer value?""" if not self._current_class: return "" # Current rules # 1. Return `...` if we are dealing with `NamedTuple` or `dataclass` field and # it has an existing default value if ( self._current_class.info and self._current_class.info.is_named_tuple and not isinstance(rvalue, TempNode) ): return " = ..." if self.processing_dataclass: if isinstance(rvalue, CallExpr): fullname = self.get_fullname(rvalue.callee) if fullname in (self.dataclass_field_specifier or DATACLASS_FIELD_SPECIFIERS): p = AliasPrinter(self) return f" = {rvalue.accept(p)}" if not (isinstance(rvalue, TempNode) and rvalue.no_rhs): return " = ..." # TODO: support other possible cases, where initializer is important # By default, no initializer is required: return "" def add_decorator(self, name: str, require_name: bool = False) -> None: if require_name: self.import_tracker.require_name(name) self._decorators.append(f"@{name}") def clear_decorators(self) -> None: self._decorators.clear() def is_private_member(self, fullname: str) -> bool: parts = fullname.split(".") return any(self.is_private_name(part) for part in parts) def get_str_type_of_node(self, rvalue: Expression, *, can_be_incomplete: bool = True) -> str: rvalue = self.maybe_unwrap_unary_expr(rvalue) if isinstance(rvalue, IntExpr): return "int" if isinstance(rvalue, StrExpr): return "str" if isinstance(rvalue, BytesExpr): return "bytes" if isinstance(rvalue, FloatExpr): return "float" if isinstance(rvalue, ComplexExpr): # 1j return "complex" if isinstance(rvalue, OpExpr) and rvalue.op in ("-", "+"): # -1j + 1 if isinstance(self.maybe_unwrap_unary_expr(rvalue.left), ComplexExpr) or isinstance( self.maybe_unwrap_unary_expr(rvalue.right), ComplexExpr ): return "complex" if isinstance(rvalue, NameExpr) and rvalue.name in ("True", "False"): return "bool" if can_be_incomplete: return self.add_name("_typeshed.Incomplete") else: return "" def maybe_unwrap_unary_expr(self, expr: Expression) -> Expression: """Unwrap (possibly nested) unary expressions. But, some unary expressions can change the type of expression. While we want to preserve it. For example, `~True` is `int`. So, we only allow a subset of unary expressions to be unwrapped. """ if not isinstance(expr, UnaryExpr): return expr # First, try to unwrap `[+-]+ (int|float|complex)` expr: math_ops = ("+", "-") if expr.op in math_ops: while isinstance(expr, UnaryExpr): if expr.op not in math_ops or not isinstance( expr.expr, (IntExpr, FloatExpr, ComplexExpr, UnaryExpr) ): break expr = expr.expr return expr # Next, try `not bool` expr: if expr.op == "not": while isinstance(expr, UnaryExpr): if expr.op != "not" or not isinstance(expr.expr, (NameExpr, UnaryExpr)): break if isinstance(expr.expr, NameExpr) and expr.expr.name not in ("True", "False"): break expr = expr.expr return expr # This is some other unary expr, we cannot do anything with it (yet?). return expr def get_str_default_of_node(self, rvalue: Expression) -> tuple[str, bool]: """Get a string representation of the default value of a node. Returns a 2-tuple of the default and whether or not it is valid. """ if isinstance(rvalue, NameExpr): if rvalue.name in ("None", "True", "False"): return rvalue.name, True elif isinstance(rvalue, (IntExpr, FloatExpr)): return f"{rvalue.value}", True elif isinstance(rvalue, UnaryExpr): if isinstance(rvalue.expr, (IntExpr, FloatExpr)): return f"{rvalue.op}{rvalue.expr.value}", True elif isinstance(rvalue, StrExpr): return repr(rvalue.value), True elif isinstance(rvalue, BytesExpr): return "b" + repr(rvalue.value).replace("\\\\", "\\"), True elif isinstance(rvalue, TupleExpr): items_defaults = [] for e in rvalue.items: e_default, valid = self.get_str_default_of_node(e) if not valid: break items_defaults.append(e_default) else: closing = ",)" if len(items_defaults) == 1 else ")" default = "(" + ", ".join(items_defaults) + closing return default, True elif isinstance(rvalue, ListExpr): items_defaults = [] for e in rvalue.items: e_default, valid = self.get_str_default_of_node(e) if not valid: break items_defaults.append(e_default) else: default = "[" + ", ".join(items_defaults) + "]" return default, True elif isinstance(rvalue, SetExpr): items_defaults = [] for e in rvalue.items: e_default, valid = self.get_str_default_of_node(e) if not valid: break items_defaults.append(e_default) else: if items_defaults: default = "{" + ", ".join(items_defaults) + "}" return default, True elif isinstance(rvalue, DictExpr): items_defaults = [] for k, v in rvalue.items: if k is None: break k_default, k_valid = self.get_str_default_of_node(k) v_default, v_valid = self.get_str_default_of_node(v) if not (k_valid and v_valid): break items_defaults.append(f"{k_default}: {v_default}") else: default = "{" + ", ".join(items_defaults) + "}" return default, True return "...", False def should_reexport(self, name: str, full_module: str, name_is_alias: bool) -> bool: is_private = self.is_private_name(name, full_module + "." + name) if ( not name_is_alias and name not in self.referenced_names and (not self._all_ or name in self.IGNORED_DUNDERS) and not is_private and full_module not in ("abc", "asyncio") + self.TYPING_MODULE_NAMES ): # An imported name that is never referenced in the module is assumed to be # exported, unless there is an explicit __all__. Note that we need to special # case 'abc' since some references are deleted during semantic analysis. return True return super().should_reexport(name, full_module, name_is_alias) def find_method_names(defs: list[Statement]) -> set[str]: # TODO: Traverse into nested definitions result = set() for defn in defs: if isinstance(defn, FuncDef): result.add(defn.name) elif isinstance(defn, Decorator): result.add(defn.func.name) elif isinstance(defn, OverloadedFuncDef): for item in defn.items: result.update(find_method_names([item])) return result class SelfTraverser(mypy.traverser.TraverserVisitor): def __init__(self) -> None: self.results: list[tuple[str, Expression, Type | None]] = [] def visit_assignment_stmt(self, o: AssignmentStmt) -> None: lvalue = o.lvalues[0] if ( isinstance(lvalue, MemberExpr) and isinstance(lvalue.expr, NameExpr) and lvalue.expr.name == "self" ): self.results.append((lvalue.name, o.rvalue, o.unanalyzed_type)) def find_self_initializers(fdef: FuncBase) -> list[tuple[str, Expression, Type | None]]: """Find attribute initializers in a method. Return a list of pairs (attribute name, r.h.s. expression). """ traverser = SelfTraverser() fdef.accept(traverser) return traverser.results def get_qualified_name(o: Expression) -> str: if isinstance(o, NameExpr): return o.name elif isinstance(o, MemberExpr): return f"{get_qualified_name(o.expr)}.{o.name}" else: return ERROR_MARKER def remove_blacklisted_modules(modules: list[StubSource]) -> list[StubSource]: return [ module for module in modules if module.path is None or not is_blacklisted_path(module.path) ] def split_pyc_from_py(modules: list[StubSource]) -> tuple[list[StubSource], list[StubSource]]: py_modules = [] pyc_modules = [] for mod in modules: if is_pyc_only(mod.path): pyc_modules.append(mod) else: py_modules.append(mod) return pyc_modules, py_modules def is_blacklisted_path(path: str) -> bool: return any(substr in (normalize_path_separators(path) + "\n") for substr in BLACKLIST) def normalize_path_separators(path: str) -> str: return path.replace("\\", "/") if sys.platform == "win32" else path def collect_build_targets( options: Options, mypy_opts: MypyOptions ) -> tuple[list[StubSource], list[StubSource], list[StubSource]]: """Collect files for which we need to generate stubs. Return list of py modules, pyc modules, and C modules. """ if options.packages or options.modules: if options.no_import: py_modules = find_module_paths_using_search( options.modules, options.packages, options.search_path, options.pyversion ) c_modules: list[StubSource] = [] else: # Using imports is the default, since we can also find C modules. py_modules, c_modules = find_module_paths_using_imports( options.modules, options.packages, options.verbose, options.quiet ) else: # Use mypy native source collection for files and directories. try: source_list = create_source_list(options.files, mypy_opts) except InvalidSourceList as e: raise SystemExit(str(e)) from e py_modules = [StubSource(m.module, m.path) for m in source_list] c_modules = [] py_modules = remove_blacklisted_modules(py_modules) pyc_mod, py_mod = split_pyc_from_py(py_modules) return py_mod, pyc_mod, c_modules def find_module_paths_using_imports( modules: list[str], packages: list[str], verbose: bool, quiet: bool ) -> tuple[list[StubSource], list[StubSource]]: """Find path and runtime value of __all__ (if possible) for modules and packages. This function uses runtime Python imports to get the information. """ with ModuleInspect() as inspect: py_modules: list[StubSource] = [] c_modules: list[StubSource] = [] found = list(walk_packages(inspect, packages, verbose)) modules = modules + found modules = [ mod for mod in modules if not is_non_library_module(mod) ] # We don't want to run any tests or scripts for mod in modules: try: result = find_module_path_and_all_py3(inspect, mod, verbose) except CantImport as e: tb = traceback.format_exc() if verbose: sys.stderr.write(tb) if not quiet: report_missing(mod, e.message, tb) continue if not result: c_modules.append(StubSource(mod)) else: path, runtime_all = result py_modules.append(StubSource(mod, path, runtime_all)) return py_modules, c_modules def is_non_library_module(module: str) -> bool: """Does module look like a test module or a script?""" if module.endswith( ( ".tests", ".test", ".testing", "_tests", "_test_suite", "test_util", "test_utils", "test_base", ".__main__", ".conftest", # Used by pytest ".setup", # Typically an install script ) ): return True if module.split(".")[-1].startswith("test_"): return True if ( ".tests." in module or ".test." in module or ".testing." in module or ".SelfTest." in module ): return True return False def translate_module_name(module: str, relative: int) -> tuple[str, int]: for pkg in VENDOR_PACKAGES: for alt in "six.moves", "six": substr = f"{pkg}.{alt}" if module.endswith("." + substr) or (module == substr and relative): return alt, 0 if "." + substr + "." in module: return alt + "." + module.partition("." + substr + ".")[2], 0 return module, relative def find_module_paths_using_search( modules: list[str], packages: list[str], search_path: list[str], pyversion: tuple[int, int] ) -> list[StubSource]: """Find sources for modules and packages requested. This function just looks for source files at the file system level. This is used if user passes --no-import, and will not find C modules. Exit if some of the modules or packages can't be found. """ result: list[StubSource] = [] typeshed_path = default_lib_path(mypy.build.default_data_dir(), pyversion, None) search_paths = SearchPaths((".",) + tuple(search_path), (), (), tuple(typeshed_path)) cache = FindModuleCache(search_paths, fscache=None, options=None) for module in modules: m_result = cache.find_module(module) if isinstance(m_result, ModuleNotFoundReason): fail_missing(module, m_result) module_path = None else: module_path = m_result result.append(StubSource(module, module_path)) for package in packages: p_result = cache.find_modules_recursive(package) if p_result: fail_missing(package, ModuleNotFoundReason.NOT_FOUND) sources = [StubSource(m.module, m.path) for m in p_result] result.extend(sources) result = [m for m in result if not is_non_library_module(m.module)] return result def mypy_options(stubgen_options: Options) -> MypyOptions: """Generate mypy options using the flag passed by user.""" options = MypyOptions() options.follow_imports = "skip" options.incremental = False options.ignore_errors = True options.semantic_analysis_only = True options.python_version = stubgen_options.pyversion options.show_traceback = True options.transform_source = remove_misplaced_type_comments options.preserve_asts = True options.include_docstrings = stubgen_options.include_docstrings # Override cache_dir if provided in the environment environ_cache_dir = os.getenv("MYPY_CACHE_DIR", "") if environ_cache_dir.strip(): options.cache_dir = environ_cache_dir options.cache_dir = os.path.expanduser(options.cache_dir) return options def parse_source_file(mod: StubSource, mypy_options: MypyOptions) -> None: """Parse a source file. On success, store AST in the corresponding attribute of the stub source. If there are syntax errors, print them and exit. """ assert mod.path is not None, "Not found module was not skipped" with open(mod.path, "rb") as f: data = f.read() source = mypy.util.decode_python_encoding(data) errors = Errors(mypy_options) mod.ast = mypy.parse.parse( source, fnam=mod.path, module=mod.module, errors=errors, options=mypy_options ) mod.ast._fullname = mod.module if errors.is_blockers(): # Syntax error! for m in errors.new_messages(): sys.stderr.write(f"{m}\n") sys.exit(1) def generate_asts_for_modules( py_modules: list[StubSource], parse_only: bool, mypy_options: MypyOptions, verbose: bool ) -> None: """Use mypy to parse (and optionally analyze) source files.""" if not py_modules: return # Nothing to do here, but there may be C modules if verbose: print(f"Processing {len(py_modules)} files...") if parse_only: for mod in py_modules: parse_source_file(mod, mypy_options) return # Perform full semantic analysis of the source set. try: res = build([module.source for module in py_modules], mypy_options) except CompileError as e: raise SystemExit(f"Critical error during semantic analysis: {e}") from e for mod in py_modules: mod.ast = res.graph[mod.module].tree # Use statically inferred __all__ if there is no runtime one. if mod.runtime_all is None: mod.runtime_all = res.manager.semantic_analyzer.export_map[mod.module] def generate_stub_for_py_module( mod: StubSource, target: str, *, parse_only: bool = False, inspect: bool = False, include_private: bool = False, export_less: bool = False, include_docstrings: bool = False, doc_dir: str = "", all_modules: list[str], ) -> None: """Use analysed (or just parsed) AST to generate type stub for single file. If directory for target doesn't exist it will created. Existing stub will be overwritten. """ if inspect: ngen = InspectionStubGenerator( module_name=mod.module, known_modules=all_modules, _all_=mod.runtime_all, doc_dir=doc_dir, include_private=include_private, export_less=export_less, include_docstrings=include_docstrings, ) ngen.generate_module() output = ngen.output() else: gen = ASTStubGenerator( mod.runtime_all, include_private=include_private, analyzed=not parse_only, export_less=export_less, include_docstrings=include_docstrings, ) assert mod.ast is not None, "This function must be used only with analyzed modules" mod.ast.accept(gen) output = gen.output() # Write output to file. subdir = os.path.dirname(target) if subdir and not os.path.isdir(subdir): os.makedirs(subdir) with open(target, "w", encoding="utf-8") as file: file.write(output) def generate_stubs(options: Options) -> None: """Main entry point for the program.""" mypy_opts = mypy_options(options) py_modules, pyc_modules, c_modules = collect_build_targets(options, mypy_opts) all_modules = py_modules + pyc_modules + c_modules all_module_names = sorted(m.module for m in all_modules) # Use parsed sources to generate stubs for Python modules. generate_asts_for_modules(py_modules, options.parse_only, mypy_opts, options.verbose) files = [] for mod in py_modules + pyc_modules: assert mod.path is not None, "Not found module was not skipped" target = mod.module.replace(".", "/") if os.path.basename(mod.path) in ["__init__.py", "__init__.pyc"]: target += "/__init__.pyi" else: target += ".pyi" target = os.path.join(options.output_dir, target) files.append(target) with generate_guarded(mod.module, target, options.ignore_errors, options.verbose): generate_stub_for_py_module( mod, target, parse_only=options.parse_only, inspect=options.inspect or mod in pyc_modules, include_private=options.include_private, export_less=options.export_less, include_docstrings=options.include_docstrings, doc_dir=options.doc_dir, all_modules=all_module_names, ) # Separately analyse C modules using different logic. for mod in c_modules: if any(py_mod.module.startswith(mod.module + ".") for py_mod in all_modules): target = mod.module.replace(".", "/") + "/__init__.pyi" else: target = mod.module.replace(".", "/") + ".pyi" target = os.path.join(options.output_dir, target) files.append(target) with generate_guarded(mod.module, target, options.ignore_errors, options.verbose): generate_stub_for_c_module( mod.module, target, known_modules=all_module_names, doc_dir=options.doc_dir, include_private=options.include_private, export_less=options.export_less, include_docstrings=options.include_docstrings, ) num_modules = len(all_modules) if not options.quiet and num_modules > 0: print("Processed %d modules" % num_modules) if len(files) == 1: print(f"Generated {files[0]}") else: print(f"Generated files under {common_dir_prefix(files)}" + os.sep) HEADER = """%(prog)s [-h] [more options, see -h] [-m MODULE] [-p PACKAGE] [files ...]""" DESCRIPTION = """ Generate draft stubs for modules. Stubs are generated in directory ./out, to avoid overriding files with manual changes. This directory is assumed to exist. """ def parse_options(args: list[str]) -> Options: parser = argparse.ArgumentParser( prog="stubgen", usage=HEADER, description=DESCRIPTION, fromfile_prefix_chars="@" ) parser.add_argument( "--ignore-errors", action="store_true", help="ignore errors when trying to generate stubs for modules", ) parser.add_argument( "--no-import", action="store_true", help="don't import the modules, just parse and analyze them " "(doesn't work with C extension modules and might not " "respect __all__)", ) parser.add_argument( "--no-analysis", "--parse-only", dest="parse_only", action="store_true", help="don't perform semantic analysis of sources, just parse them " "(only applies to Python modules, might affect quality of stubs. " "Not compatible with --inspect-mode)", ) parser.add_argument( "--inspect-mode", dest="inspect", action="store_true", help="import and inspect modules instead of parsing source code." "This is the default behavior for c modules and pyc-only packages, but " "it is also useful for pure python modules with dynamically generated members.", ) parser.add_argument( "--include-private", action="store_true", help="generate stubs for objects and members considered private " "(single leading underscore and no trailing underscores)", ) parser.add_argument( "--export-less", action="store_true", help="don't implicitly export all names imported from other modules in the same package", ) parser.add_argument( "--include-docstrings", action="store_true", help="include existing docstrings with the stubs", ) parser.add_argument("-v", "--verbose", action="store_true", help="show more verbose messages") parser.add_argument("-q", "--quiet", action="store_true", help="show fewer messages") parser.add_argument( "--doc-dir", metavar="PATH", default="", help="use .rst documentation in PATH (this may result in " "better stubs in some cases; consider setting this to " "DIR/Python-X.Y.Z/Doc/library)", ) parser.add_argument( "--search-path", metavar="PATH", default="", help="specify module search directories, separated by ':' " "(currently only used if --no-import is given)", ) parser.add_argument( "-o", "--output", metavar="PATH", dest="output_dir", default="out", help="change the output directory [default: %(default)s]", ) parser.add_argument( "-m", "--module", action="append", metavar="MODULE", dest="modules", default=[], help="generate stub for module; can repeat for more modules", ) parser.add_argument( "-p", "--package", action="append", metavar="PACKAGE", dest="packages", default=[], help="generate stubs for package recursively; can be repeated", ) parser.add_argument( metavar="files", nargs="*", dest="files", help="generate stubs for given files or directories", ) parser.add_argument( "--version", action="version", version="%(prog)s " + mypy.version.__version__ ) ns = parser.parse_args(args) pyversion = sys.version_info[:2] ns.interpreter = sys.executable if ns.modules + ns.packages and ns.files: parser.error("May only specify one of: modules/packages or files.") if ns.quiet and ns.verbose: parser.error("Cannot specify both quiet and verbose messages") if ns.inspect and ns.parse_only: parser.error("Cannot specify both --parse-only/--no-analysis and --inspect-mode") # Create the output folder if it doesn't already exist. os.makedirs(ns.output_dir, exist_ok=True) return Options( pyversion=pyversion, no_import=ns.no_import, inspect=ns.inspect, doc_dir=ns.doc_dir, search_path=ns.search_path.split(":"), interpreter=ns.interpreter, ignore_errors=ns.ignore_errors, parse_only=ns.parse_only, include_private=ns.include_private, output_dir=ns.output_dir, modules=ns.modules, packages=ns.packages, files=ns.files, verbose=ns.verbose, quiet=ns.quiet, export_less=ns.export_less, include_docstrings=ns.include_docstrings, ) def main(args: list[str] | None = None) -> None: mypy.util.check_python_version("stubgen") # Make sure that the current directory is in sys.path so that # stubgen can be run on packages in the current directory. if not ("" in sys.path or "." in sys.path): sys.path.insert(0, "") options = parse_options(sys.argv[1:] if args is None else args) generate_stubs(options) if __name__ == "__main__": main()