from __future__ import annotations

import pprint
import re
import sys
import sysconfig
from typing import Any, Callable, Final, Mapping, Pattern

from mypy import defaults
from mypy.errorcodes import ErrorCode, error_codes
from mypy.util import get_class_descriptors, replace_object_state


class BuildType:
    STANDARD: Final = 0
    MODULE: Final = 1
    PROGRAM_TEXT: Final = 2


PER_MODULE_OPTIONS: Final = {
    # Please keep this list sorted
    "allow_redefinition",
    "allow_untyped_globals",
    "always_false",
    "always_true",
    "check_untyped_defs",
    "debug_cache",
    "disable_error_code",
    "disabled_error_codes",
    "disallow_any_decorated",
    "disallow_any_explicit",
    "disallow_any_expr",
    "disallow_any_generics",
    "disallow_any_unimported",
    "disallow_incomplete_defs",
    "disallow_subclassing_any",
    "disallow_untyped_calls",
    "disallow_untyped_decorators",
    "disallow_untyped_defs",
    "enable_error_code",
    "enabled_error_codes",
    "extra_checks",
    "follow_imports_for_stubs",
    "follow_imports",
    "follow_untyped_imports",
    "ignore_errors",
    "ignore_missing_imports",
    "implicit_optional",
    "implicit_reexport",
    "local_partial_types",
    "mypyc",
    "strict_concatenate",
    "strict_equality",
    "strict_optional",
    "warn_no_return",
    "warn_return_any",
    "warn_unreachable",
    "warn_unused_ignores",
}

OPTIONS_AFFECTING_CACHE: Final = (
    PER_MODULE_OPTIONS
    | {
        "platform",
        "bazel",
        "old_type_inference",
        "plugins",
        "disable_bytearray_promotion",
        "disable_memoryview_promotion",
    }
) - {"debug_cache"}

# Features that are currently (or were recently) incomplete/experimental
TYPE_VAR_TUPLE: Final = "TypeVarTuple"
UNPACK: Final = "Unpack"
PRECISE_TUPLE_TYPES: Final = "PreciseTupleTypes"
NEW_GENERIC_SYNTAX: Final = "NewGenericSyntax"
INLINE_TYPEDDICT: Final = "InlineTypedDict"
INCOMPLETE_FEATURES: Final = frozenset((PRECISE_TUPLE_TYPES, INLINE_TYPEDDICT))
COMPLETE_FEATURES: Final = frozenset((TYPE_VAR_TUPLE, UNPACK, NEW_GENERIC_SYNTAX))


class Options:
    """Options collected from flags."""

    def __init__(self) -> None:
        # Cache for clone_for_module()
        self._per_module_cache: dict[str, Options] | None = None

        # -- build options --
        self.build_type = BuildType.STANDARD
        self.python_version: tuple[int, int] = sys.version_info[:2]
        # The executable used to search for PEP 561 packages. If this is None,
        # then mypy does not search for PEP 561 packages.
        self.python_executable: str | None = sys.executable

        # When cross compiling to emscripten, we need to rely on MACHDEP because
        # sys.platform is the host build platform, not emscripten.
        MACHDEP = sysconfig.get_config_var("MACHDEP")
        if MACHDEP == "emscripten":
            self.platform = MACHDEP
        else:
            self.platform = sys.platform

        self.custom_typing_module: str | None = None
        self.custom_typeshed_dir: str | None = None
        # The abspath() version of the above, we compute it once as an optimization.
        self.abs_custom_typeshed_dir: str | None = None
        self.mypy_path: list[str] = []
        self.report_dirs: dict[str, str] = {}
        # Show errors in PEP 561 packages/site-packages modules
        self.no_silence_site_packages = False
        self.no_site_packages = False
        self.ignore_missing_imports = False
        # Is ignore_missing_imports set in a per-module section
        self.ignore_missing_imports_per_module = False
        # Typecheck modules without stubs or py.typed marker
        self.follow_untyped_imports = False
        self.follow_imports = "normal"  # normal|silent|skip|error
        # Whether to respect the follow_imports setting even for stub files.
        # Intended to be used for disabling specific stubs.
        self.follow_imports_for_stubs = False
        # PEP 420 namespace packages
        # This allows definitions of packages without __init__.py and allows packages to span
        # multiple directories. This flag affects both import discovery and the association of
        # input files/modules/packages to the relevant file and fully qualified module name.
        self.namespace_packages = True
        # Use current directory and MYPYPATH to determine fully qualified module names of files
        # passed by automatically considering their subdirectories as packages. This is only
        # relevant if namespace packages are enabled, since otherwise examining __init__.py's is
        # sufficient to determine module names for files. As a possible alternative, add a single
        # top-level __init__.py to your packages.
        self.explicit_package_bases = False
        # File names, directory names or subpaths to avoid checking
        self.exclude: list[str] = []

        # disallow_any options
        self.disallow_any_generics = False
        self.disallow_any_unimported = False
        self.disallow_any_expr = False
        self.disallow_any_decorated = False
        self.disallow_any_explicit = False

        # Disallow calling untyped functions from typed ones
        self.disallow_untyped_calls = False

        # Always allow untyped calls for function coming from modules/packages
        # in this list (each item effectively acts as a prefix match)
        self.untyped_calls_exclude: list[str] = []

        # Disallow defining untyped (or incompletely typed) functions
        self.disallow_untyped_defs = False

        # Disallow defining incompletely typed functions
        self.disallow_incomplete_defs = False

        # Type check unannotated functions
        self.check_untyped_defs = False

        # Disallow decorating typed functions with untyped decorators
        self.disallow_untyped_decorators = False

        # Disallow subclassing values of type 'Any'
        self.disallow_subclassing_any = False

        # Also check typeshed for missing annotations
        self.warn_incomplete_stub = False

        # Warn about casting an expression to its inferred type
        self.warn_redundant_casts = False

        # Warn about falling off the end of a function returning non-None
        self.warn_no_return = True

        # Warn about returning objects of type Any when the function is
        # declared with a precise type
        self.warn_return_any = False

        # Report importing or using deprecated features as errors instead of notes.
        self.report_deprecated_as_note = False

        # Warn about unused '# type: ignore' comments
        self.warn_unused_ignores = False

        # Warn about unused '[mypy-<pattern>]'  or '[[tool.mypy.overrides]]' config sections
        self.warn_unused_configs = False

        # Files in which to ignore all non-fatal errors
        self.ignore_errors = False

        # Apply strict None checking
        self.strict_optional = True

        # Show "note: In function "foo":" messages.
        self.show_error_context = False

        # Use nicer output (when possible).
        self.color_output = True
        self.error_summary = True

        # Assume arguments with default values of None are Optional
        self.implicit_optional = False

        # Don't re-export names unless they are imported with `from ... as ...`
        self.implicit_reexport = True

        # Suppress toplevel errors caused by missing annotations
        self.allow_untyped_globals = False

        # Allow variable to be redefined with an arbitrary type in the same block
        # and the same nesting level as the initialization
        self.allow_redefinition = False

        # Prohibit equality, identity, and container checks for non-overlapping types.
        # This makes 1 == '1', 1 in ['1'], and 1 is '1' errors.
        self.strict_equality = False

        # Deprecated, use extra_checks instead.
        self.strict_concatenate = False

        # Enable additional checks that are technically correct but impractical.
        self.extra_checks = False

        # Report an error for any branches inferred to be unreachable as a result of
        # type analysis.
        self.warn_unreachable = False

        # Variable names considered True
        self.always_true: list[str] = []

        # Variable names considered False
        self.always_false: list[str] = []

        # Error codes to disable
        self.disable_error_code: list[str] = []
        self.disabled_error_codes: set[ErrorCode] = set()

        # Error codes to enable
        self.enable_error_code: list[str] = []
        self.enabled_error_codes: set[ErrorCode] = set()

        # Use script name instead of __main__
        self.scripts_are_modules = False

        # Config file name
        self.config_file: str | None = None

        # A filename containing a JSON mapping from filenames to
        # mtime/size/hash arrays, used to avoid having to recalculate
        # source hashes as often.
        self.quickstart_file: str | None = None

        # A comma-separated list of files/directories for mypy to type check;
        # supports globbing
        self.files: list[str] | None = None

        # A list of packages for mypy to type check
        self.packages: list[str] | None = None

        # A list of modules for mypy to type check
        self.modules: list[str] | None = None

        # Write junit.xml to given file
        self.junit_xml: str | None = None

        self.junit_format: str = "global"  # global|per_file

        # Caching and incremental checking options
        self.incremental = True
        self.cache_dir = defaults.CACHE_DIR
        self.sqlite_cache = False
        self.debug_cache = False
        self.skip_version_check = False
        self.skip_cache_mtime_checks = False
        self.fine_grained_incremental = False
        # Include fine-grained dependencies in written cache files
        self.cache_fine_grained = False
        # Read cache files in fine-grained incremental mode (cache must include dependencies)
        self.use_fine_grained_cache = False

        # Run tree.serialize() even if cache generation is disabled
        self.debug_serialize = False

        # Tune certain behaviors when being used as a front-end to mypyc. Set per-module
        # in modules being compiled. Not in the config file or command line.
        self.mypyc = False

        # An internal flag to modify some type-checking logic while
        # running inspections (e.g. don't expand function definitions).
        # Not in the config file or command line.
        self.inspections = False

        # Disable the memory optimization of freeing ASTs when
        # possible. This isn't exposed as a command line option
        # because it is intended for software integrating with
        # mypy. (Like mypyc.)
        self.preserve_asts = False

        # If True, function and class docstrings will be extracted and retained.
        # This isn't exposed as a command line option
        # because it is intended for software integrating with
        # mypy. (Like stubgen.)
        self.include_docstrings = False

        # Paths of user plugins
        self.plugins: list[str] = []

        # Per-module options (raw)
        self.per_module_options: dict[str, dict[str, object]] = {}
        self._glob_options: list[tuple[str, Pattern[str]]] = []
        self.unused_configs: set[str] = set()

        # -- development options --
        self.verbosity = 0  # More verbose messages (for troubleshooting)
        self.pdb = False
        self.show_traceback = False
        self.raise_exceptions = False
        self.dump_type_stats = False
        self.dump_inference_stats = False
        self.dump_build_stats = False
        self.enable_incomplete_feature: list[str] = []
        self.timing_stats: str | None = None
        self.line_checking_stats: str | None = None

        # -- test options --
        # Stop after the semantic analysis phase
        self.semantic_analysis_only = False

        # Use stub builtins fixtures to speed up tests
        self.use_builtins_fixtures = False

        # This should only be set when running certain mypy tests.
        # Use this sparingly to avoid tests diverging from non-test behavior.
        self.test_env = False

        # -- experimental options --
        self.shadow_file: list[list[str]] | None = None
        self.show_column_numbers: bool = False
        self.show_error_end: bool = False
        self.hide_error_codes = False
        self.show_error_code_links = False
        # Use soft word wrap and show trimmed source snippets with error location markers.
        self.pretty = False
        self.dump_graph = False
        self.dump_deps = False
        self.logical_deps = False
        # If True, partial types can't span a module top level and a function
        self.local_partial_types = False
        # Some behaviors are changed when using Bazel (https://bazel.build).
        self.bazel = False
        # If True, export inferred types for all expressions as BuildResult.types
        self.export_types = False
        # List of package roots -- directories under these are packages even
        # if they don't have __init__.py.
        self.package_root: list[str] = []
        self.cache_map: dict[str, tuple[str, str]] = {}
        # Don't properly free objects on exit, just kill the current process.
        self.fast_exit = True
        # fast path for finding modules from source set
        self.fast_module_lookup = False
        # Allow empty function bodies even if it is not safe, used for testing only.
        self.allow_empty_bodies = False
        # Used to transform source code before parsing if not None
        # TODO: Make the type precise (AnyStr -> AnyStr)
        self.transform_source: Callable[[Any], Any] | None = None
        # Print full path to each file in the report.
        self.show_absolute_path: bool = False
        # Install missing stub packages if True
        self.install_types = False
        # Install missing stub packages in non-interactive mode (don't prompt for
        # confirmation, and don't show any errors)
        self.non_interactive = False
        # When we encounter errors that may cause many additional errors,
        # skip most errors after this many messages have been reported.
        # -1 means unlimited.
        self.many_errors_threshold = defaults.MANY_ERRORS_THRESHOLD
        # Disable new experimental type inference algorithm.
        self.old_type_inference = False
        # Deprecated reverse version of the above, do not use.
        self.new_type_inference = False
        # Export line-level, limited, fine-grained dependency information in cache data
        # (undocumented feature).
        self.export_ref_info = False

        self.disable_bytearray_promotion = False
        self.disable_memoryview_promotion = False
        self.force_uppercase_builtins = False
        self.force_union_syntax = False

        # Sets custom output format
        self.output: str | None = None

    def use_lowercase_names(self) -> bool:
        if self.python_version >= (3, 9):
            return not self.force_uppercase_builtins
        return False

    def use_or_syntax(self) -> bool:
        if self.python_version >= (3, 10):
            return not self.force_union_syntax
        return False

    def use_star_unpack(self) -> bool:
        return self.python_version >= (3, 11)

    def snapshot(self) -> dict[str, object]:
        """Produce a comparable snapshot of this Option"""
        # Under mypyc, we don't have a __dict__, so we need to do worse things.
        d = dict(getattr(self, "__dict__", ()))
        for k in get_class_descriptors(Options):
            if hasattr(self, k):
                d[k] = getattr(self, k)
        # Remove private attributes from snapshot
        d = {k: v for k, v in d.items() if not k.startswith("_")}
        return d

    def __repr__(self) -> str:
        return f"Options({pprint.pformat(self.snapshot())})"

    def process_error_codes(self, *, error_callback: Callable[[str], Any]) -> None:
        # Process `--enable-error-code` and `--disable-error-code` flags
        disabled_codes = set(self.disable_error_code)
        enabled_codes = set(self.enable_error_code)

        valid_error_codes = set(error_codes.keys())

        invalid_codes = (enabled_codes | disabled_codes) - valid_error_codes
        if invalid_codes:
            error_callback(f"Invalid error code(s): {', '.join(sorted(invalid_codes))}")

        self.disabled_error_codes |= {error_codes[code] for code in disabled_codes}
        self.enabled_error_codes |= {error_codes[code] for code in enabled_codes}

        # Enabling an error code always overrides disabling
        self.disabled_error_codes -= self.enabled_error_codes

    def process_incomplete_features(
        self, *, error_callback: Callable[[str], Any], warning_callback: Callable[[str], Any]
    ) -> None:
        # Validate incomplete features.
        for feature in self.enable_incomplete_feature:
            if feature not in INCOMPLETE_FEATURES | COMPLETE_FEATURES:
                error_callback(f"Unknown incomplete feature: {feature}")
            if feature in COMPLETE_FEATURES:
                warning_callback(f"Warning: {feature} is already enabled by default")

    def apply_changes(self, changes: dict[str, object]) -> Options:
        # Note: effects of this method *must* be idempotent.
        new_options = Options()
        # Under mypyc, we don't have a __dict__, so we need to do worse things.
        replace_object_state(new_options, self, copy_dict=True)
        for key, value in changes.items():
            setattr(new_options, key, value)
        if changes.get("ignore_missing_imports"):
            # This is the only option for which a per-module and a global
            # option sometimes beheave differently.
            new_options.ignore_missing_imports_per_module = True

        # These two act as overrides, so apply them when cloning.
        # Similar to global codes enabling overrides disabling, so we start from latter.
        new_options.disabled_error_codes = self.disabled_error_codes.copy()
        new_options.enabled_error_codes = self.enabled_error_codes.copy()
        for code_str in new_options.disable_error_code:
            code = error_codes[code_str]
            new_options.disabled_error_codes.add(code)
            new_options.enabled_error_codes.discard(code)
        for code_str in new_options.enable_error_code:
            code = error_codes[code_str]
            new_options.enabled_error_codes.add(code)
            new_options.disabled_error_codes.discard(code)

        return new_options

    def compare_stable(self, other_snapshot: dict[str, object]) -> bool:
        """Compare options in a way that is stable for snapshot() -> apply_changes() roundtrip.

        This is needed because apply_changes() has non-trivial effects for some flags, so
        Options().apply_changes(options.snapshot()) may result in a (slightly) different object.
        """
        return (
            Options().apply_changes(self.snapshot()).snapshot()
            == Options().apply_changes(other_snapshot).snapshot()
        )

    def build_per_module_cache(self) -> None:
        self._per_module_cache = {}

        # Config precedence is as follows:
        #  1. Concrete section names: foo.bar.baz
        #  2. "Unstructured" glob patterns: foo.*.baz, in the order
        #     they appear in the file (last wins)
        #  3. "Well-structured" wildcard patterns: foo.bar.*, in specificity order.

        # Since structured configs inherit from structured configs above them in the hierarchy,
        # we need to process per-module configs in a careful order.
        # We have to process foo.* before foo.bar.* before foo.bar,
        # and we need to apply *.bar to foo.bar but not to foo.bar.*.
        # To do this, process all well-structured glob configs before non-glob configs and
        # exploit the fact that foo.* sorts earlier ASCIIbetically (unicodebetically?)
        # than foo.bar.*.
        # (A section being "processed last" results in its config "winning".)
        # Unstructured glob configs are stored and are all checked for each module.
        unstructured_glob_keys = [k for k in self.per_module_options.keys() if "*" in k[:-1]]
        structured_keys = [k for k in self.per_module_options.keys() if "*" not in k[:-1]]
        wildcards = sorted(k for k in structured_keys if k.endswith(".*"))
        concrete = [k for k in structured_keys if not k.endswith(".*")]

        for glob in unstructured_glob_keys:
            self._glob_options.append((glob, self.compile_glob(glob)))

        # We (for ease of implementation) treat unstructured glob
        # sections as used if any real modules use them or if any
        # concrete config sections use them. This means we need to
        # track which get used while constructing.
        self.unused_configs = set(unstructured_glob_keys)

        for key in wildcards + concrete:
            # Find what the options for this key would be, just based
            # on inheriting from parent configs.
            options = self.clone_for_module(key)
            # And then update it with its per-module options.
            self._per_module_cache[key] = options.apply_changes(self.per_module_options[key])

        # Add the more structured sections into unused configs, since
        # they only count as used if actually used by a real module.
        self.unused_configs.update(structured_keys)

    def clone_for_module(self, module: str) -> Options:
        """Create an Options object that incorporates per-module options.

        NOTE: Once this method is called all Options objects should be
        considered read-only, else the caching might be incorrect.
        """
        if self._per_module_cache is None:
            self.build_per_module_cache()
        assert self._per_module_cache is not None

        # If the module just directly has a config entry, use it.
        if module in self._per_module_cache:
            self.unused_configs.discard(module)
            return self._per_module_cache[module]

        # If not, search for glob paths at all the parents. So if we are looking for
        # options for foo.bar.baz, we search foo.bar.baz.*, foo.bar.*, foo.*,
        # in that order, looking for an entry.
        # This is technically quadratic in the length of the path, but module paths
        # don't actually get all that long.
        options = self
        path = module.split(".")
        for i in range(len(path), 0, -1):
            key = ".".join(path[:i] + ["*"])
            if key in self._per_module_cache:
                self.unused_configs.discard(key)
                options = self._per_module_cache[key]
                break

        # OK and *now* we need to look for unstructured glob matches.
        # We only do this for concrete modules, not structured wildcards.
        if not module.endswith(".*"):
            for key, pattern in self._glob_options:
                if pattern.match(module):
                    self.unused_configs.discard(key)
                    options = options.apply_changes(self.per_module_options[key])

        # We could update the cache to directly point to modules once
        # they have been looked up, but in testing this made things
        # slower and not faster, so we don't bother.

        return options

    def compile_glob(self, s: str) -> Pattern[str]:
        # Compile one of the glob patterns to a regex so that '.*' can
        # match *zero or more* module sections. This means we compile
        # '.*' into '(\..*)?'.
        parts = s.split(".")
        expr = re.escape(parts[0]) if parts[0] != "*" else ".*"
        for part in parts[1:]:
            expr += re.escape("." + part) if part != "*" else r"(\..*)?"
        return re.compile(expr + "\\Z")

    def select_options_affecting_cache(self) -> Mapping[str, object]:
        result: dict[str, object] = {}
        for opt in OPTIONS_AFFECTING_CACHE:
            val = getattr(self, opt)
            if opt in ("disabled_error_codes", "enabled_error_codes"):
                val = sorted([code.code for code in val])
            result[opt] = val
        return result
