Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

__slots__ requires explicit type definition and must be Any (and nothing else) when used with Protocols #11013

Closed
posita opened this issue Aug 24, 2021 · 3 comments · Fixed by #15490
Labels
bug mypy got something wrong topic-protocols

Comments

@posita
Copy link
Contributor

posita commented Aug 24, 2021

# test_case1.py
from __future__ import annotations
from typing import Protocol, runtime_checkable

@runtime_checkable
class Foo(Protocol):
    __slots__ = ()
    def foo(self):
        ...

class Bar:
    __slots__ = ("_t",)
    def __init__(self):
        self._t = True
    def foo(self):
        pass

print(f"isinstance(Bar(), Foo) == {isinstance(Bar(), Foo)}")  # <- prints … == True
foo: Foo = Bar()  # <- errors here
% python test_case1.py
isinstance(Bar(), Foo) == True
% mypy --version
mypy 0.910
% mypy --config-file=/dev/null test_case1.py
/dev/null: No [mypy] section in config file
test_case1.py:19: error: Incompatible types in assignment (expression has type "Bar", variable has type "Foo")
test_case1.py:19: note: Following member(s) of "Bar" have conflicts:
test_case1.py:19: note:     __slots__: expected "Tuple[]", got "Tuple[str]"

The "fix" is to provide Tuple[str, ...] as the type for __slots__:

# test_case2.py
from __future__ import annotations
from typing import Protocol, Tuple, runtime_checkable

@runtime_checkable
class Foo(Protocol):
    __slots__: Tuple[str, ...] = ()
    def foo(self):
        ...

class Bar:
    __slots__ : Tuple[str, ...] = ("_t",)
    def __init__(self):
        self._t = True
    def foo(self):
        pass

print(f"isinstance(Bar(), Foo) == {isinstance(Bar(), Foo)}")  # still True
foo: Foo = Bar()  # <- works now?
% mypy --config-file=/dev/null test_case2.py
/dev/null: No [mypy] section in config file
Success: no issues found in 1 source file

Shouldn't mypy already know the type of __slots__?

@posita posita added the bug mypy got something wrong label Aug 24, 2021
@posita posita changed the title __slots__ requires type definition when used with Protocols __slots__ requires explicit type definition when used with Protocols Aug 24, 2021
@posita
Copy link
Contributor Author

posita commented Aug 24, 2021

Here's another oddity not fixed by typing __slots___:

# test_case3.py
from fractions import Fraction
from typing import Protocol, Tuple, TypeVar, runtime_checkable
_T_co = TypeVar("_T_co", covariant=True)

@runtime_checkable
class RationalInitializerT(Protocol[_T_co]):
    __slots__: Tuple[str, ...] = ()
    def __call__(self, numerator: int, denominator: int) -> _T_co:
        ...

f: RationalInitializerT = Fraction  # <- errors
% mypy --config-file=/dev/null test_case3.py
/dev/null: No [mypy] section in config file
test_case3.py:12: error: Incompatible types in assignment (expression has type "Type[Fraction]", variable has type "RationalInitializerT[Any]")
Found 1 error in 1 file (checked 1 source file)

Now, remove __slots__:

# test_case4.py
from fractions import Fraction
from typing import Protocol, Tuple, TypeVar, runtime_checkable
_T_co = TypeVar("_T_co", covariant=True)

@runtime_checkable
class RationalInitializerT(Protocol[_T_co]):
    # __slots__: Tuple[str, ...] = ()  # <- commented out
    def __call__(self, numerator: int, denominator: int) -> _T_co:
        ...

f: RationalInitializerT = Fraction  # <- works now?
% mypy --config-file=/dev/null test_case4.py
/dev/null: No [mypy] section in config file
Success: no issues found in 1 source file

Another example:

# test_case5.py
from typing import Callable, Iterable, Protocol, Tuple, Union, runtime_checkable
from mypy_extensions import VarArg

@runtime_checkable
class ProcessorT(Protocol):
    __slots__: Tuple[str, ...] = ()
    def __call__(
        self,
        *inputs: int,
    ) -> Union[int, Iterable[int]]:
        ...

def narrower_return_proc(*inputs: int) -> Iterable[int]:
    return inputs

proc1: Callable[[VarArg(int)], Union[int, Iterable[int]]] = narrower_return_proc  # <- works
proc2: ProcessorT = narrower_return_proc  # <- errors
% mypy --config-file=/dev/null test_case5.py
/dev/null: No [mypy] section in config file
test_case5.py:18: error: Incompatible types in assignment (expression has type "Callable[[VarArg(int)], Iterable[int]]", variable has type "ProcessorT")
Found 1 error in 1 file (checked 1 source file)

Removing/commenting out __slots__ in the above example results in no errors.

@posita
Copy link
Contributor Author

posita commented Aug 24, 2021

BTW, adding __slots__ to a Protocol doesn't seem that strange, since one might want to subclass it, and it seems to enjoy standard(ish) practice, given this [source]:

@runtime_checkable
class SupportsInt(Protocol):
    """An ABC with one abstract method __int__."""
    __slots__ = ()

    @abstractmethod
    def __int__(self) -> int:
        pass

But then I found this [source]:

def runtime_checkable(cls: _TC) -> _TC: ...
@runtime_checkable
class SupportsInt(Protocol, metaclass=ABCMeta):
    @abstractmethod
    def __int__(self) -> int: ...

I guess that's another way to "work around" trivial versions of the above problem, but it's not accessible (much less intuitive). I also have no idea how to make it work with forward references without reproducing entire swaths of code in .pyi files. For example:

# foo.py
from __future__ import annotations
from typing import Iterable, Protocol, Tuple, Union, runtime_checkable

class Foo:
    ...  # lots of of members and methods here

@runtime_checkable
class FooProcessorT(Protocol):
    __slots__: Tuple[str, ...] = ()
    def __call__(
        self,
        *inputs: Foo,
    ) -> Union[Foo, Iterable[Foo]]:
        ...
# foo.pyi
from __future__ import annotations
from typing import Protocol, Iterable, Union, runtime_checkable
from abc import ABCMeta

# What do I do? Redefine all of Foo's interfaces here?

@runtime_checkable
class FooProcessorT(Protocol, metaclass=ABCMeta):
    def __call__(
        self,
        *roll_outcomes: "Foo",  # <- undefined name 'Foo'
    ) -> Union["Foo", Iterable["Foo"]]:  # <- undefined name 'Foo'
        ...

What blows my mind is that this discovery makes the Protocols in typing.py look like a lie (at least with respect to Mypy)? I'm curious why "types" defined in typing.py would ever need ("clarifying"?) definitions in typing.pyi. Doesn't that kind of defeat the point?

@posita
Copy link
Contributor Author

posita commented Mar 22, 2022

I found another case for which I cannot find a work-around, specifically where __slots__ for the value's type doesn't precisely match __slots__ for the protocol's type:

# test_case.py
from abc import abstractmethod
from typing import Any, Iterable, Protocol, Tuple, Union, runtime_checkable


class Foo:
    __slots__ = ("bar", "baz")

    def foo(self) -> None:
        pass


@runtime_checkable
class SupportsFoo(Protocol):
    __slots__: Union[
        str,
        Tuple[str],
        Tuple[str, str],
        Tuple[str, str, str],
        Tuple[str, ...],
        Iterable[str],
        Any,  # <-- adding this doesn't even help
    ] = ()

    @abstractmethod
    def foo(self) -> None:
        pass


@runtime_checkable
class SupportsFooExactMatchWorks(Protocol):
    __slots__: Tuple[str, str] = ()  # type: ignore

    @abstractmethod
    def foo(self) -> None:
        pass


@runtime_checkable
class SupportsFooAnyWorks(Protocol):
    __slots__: Any = ()

    @abstractmethod
    def foo(self) -> None:
        pass


v1: SupportsFoo = Foo()  # <-- error
v2: SupportsFooExactMatchWorks = Foo()  # works: precise type match on __slots__
v3: SupportsFooAnyWorks = Foo()  # works: __slots__: Any, and nothing else
% mypy --config-file=/dev/null test_case.py
/dev/null: No [mypy] section in config file
test_case.py:48: error: Incompatible types in assignment (expression has type "Foo", variable has type "SupportsFoo")
test_case.py:48: note: Following member(s) of "Foo" have conflicts:
test_case.py:48: note:     __slots__: expected "Union[str, Tuple[str], Tuple[str, str], Tuple[str, str, str], Tuple[str, ...], Iterable[str], Any]", got "Tuple[str, str]"
Found 1 error in 1 file (checked 1 source file)

What this effectively means is that if Protocols define __slots__ they must be explicitly typed as Any and nothing else. That seems wrong.

Collecting related issues:

#11821 (marked as fixed)
#11885 (still in draft, blocked on #11891)

@posita posita changed the title __slots__ requires explicit type definition when used with Protocols __slots__ requires explicit type definition and must be Any (and nothing else) when used with Protocols Mar 22, 2022
posita added a commit to posita/beartype that referenced this issue Apr 5, 2022
See python/mypy#11013 for details, but we may impose downstream heartache on those
deriving from our protocols if we don't explicitly set the type of `__slots__` to `Any`.

This corrects for my previously botched metaclass TypeVar in accordance with [this
recommendation](python/mypy#9282 (comment)).
posita added a commit to posita/beartype that referenced this issue Apr 5, 2022
See python/mypy#11013 for details, but we may impose downstream heartache on those
deriving from our protocols if we don't explicitly set the type of `__slots__` to `Any`.

This *also* corrects for my previously botched metaclass TypeVar in accordance with
[this recommendation](python/mypy#9282 (comment)).
Could I have included that fix as a separate PR? Yes. Should I have? Yes. Did I? No.
`#GottaGoFast` or something the kids say. (Translation: Because I'm lazy.)
leycec pushed a commit to beartype/beartype that referenced this issue Apr 7, 2022
See python/mypy#11013 for details, but we may impose downstream heartache on those
deriving from our protocols if we don't explicitly set the type of `__slots__` to `Any`.

This *also* corrects for my previously botched metaclass TypeVar in accordance with
[this recommendation](python/mypy#9282 (comment)).
Could I have included that fix as a separate PR? Yes. Should I have? Yes. Did I? No.
`#GottaGoFast` or something the kids say. (Translation: Because I'm lazy.)
posita added a commit to beartype/numerary that referenced this issue May 3, 2022
leycec added a commit to beartype/beartype that referenced this issue Sep 18, 2022
This minor release unleashes a major firestorm of support for **class
decoration,** **colourful exceptions,** **pyright + PyLance + VSCode,**
[PEP 484][PEP 484], [PEP 544][PEP 544], [PEP 561][PEP 561],
[PEP 563][PEP 563], [PEP 585][PEP 585], [PEP 604][PEP 604],
[PEP 612][PEP 612], and [PEP 647][PEP 647].

This minor release resolves a mammoth **29 issues** and merges **12 pull
requests.** Noteworthy changes include:

## Compatibility Improved

* **Class decoration.** The `@beartype` decorator now decorates both
  higher-level classes *and* lower-level callables (i.e., functions,
  methods), resolving feature request #152 kindly submitted by @posita
  the positively sublime. All possible edge cases are supported,
  including:
  * Classes defining methods decorated by builtin decorators: i.e.,
    * Class methods via `@classmethod`.
    * Static methods via `@staticmethod`.
    * Property getters, setters, and deleters via `@property`.
  * Arbitrarily deeply nested (i.e., inner) classes.
  * Arbitrarily deeply nested (i.e., inner) classes whose type hints are
    postponed under [PEP 563][PEP 563].
  Since this was surprisingly trivial, @leycec
  probably should have done this a few years ago. He didn't. This is why
  he laments into his oatmeal in late 2022.
* **[PEP 484][PEP 484]- and [PEP 585][PEP 585]-compliant nested
  generics.** @beartype now supports arbitrarily complex [PEP 484][PEP
  484]- and [PEP 585][PEP 585]-compliant inheritance trees subclassing
  non-trivial combinations of the `typing.Generic` superclass and other
  `typing` pseudo-superclasses, resolving issue #140 kindly submitted by
  @langfield (William Blake – yes, *that* William Blake). Notably,
  this release extricated our transitive visitation of the tree of all
  pseudo-superclasses of any PEP 484- and 585-compliant generic type
  hint (*...don't ask*) from its prior hidden sacred cave deep within
  the private `beartype._decor._code._pep._pephint` submodule into a new
  reusable `iter_hint_pep484585_generic_bases_unerased_tree()`
  generator, which is now believed to be the most fully-compliant
  algorithm for traversing generic inheritance trees at runtime. This
  cleanly resolved all lingering issues surrounding generics,
  dramatically reduced the likelihood of more issues surrounding
  generics, and streamlined the resolution of any more issues
  surrounding generics should they arise... *which they won't.*
  Generics: we have resoundingly beaten you. Stay down, please.
* **[PEP 544][PEP 544] compatibility.** @beartype now supports
  arbitrarily complex [PEP 544][PEP 544]-compliant inheritance trees
  subclassing non-trivial combinations of the `typing.Protocol` +
  `abc.ABC` superclasses, resolving #117 kindly submitted by
  too-entertaining pun master @twoertwein (Torsten Wörtwein).
  Notably, `@beartype` now:
  * Correctly detects non-protocols as non-protocols. Previously,
    @beartype erroneously detected a subset of non-protocols as
    PEP 544-compliant protocols. It's best not to ask why.
  * Ignores both the unsubscripted `beartype.typing.Protocol` superclass
    *and* parametrizations of that superclass by one or more type
    variables (e.g., `beartype.typing.Protocol[typing.TypeVar('T')]`) as
    semantically meaningless in accordance with similar treatment of the
    `typing.Protocol` superclass.
  * Permits caller-defined abstract protocols subclassing our caching
    `beartype.typing.Protocol` superclass to themselves be subclassed by
    one or more concrete subclasses. Previously, attempting to do so
    would raise non-human-readable exceptions from the `typing` module;
    now, doing so behaves as expected.
  * Relaxed our prior bad assumption that the second-to-last superclass
    of all generics – and thus protocols – is the `typing.Generic`
    superclass. That assumption *only* holds for standard generics and
    protocols; non-standard protocols subclassing non-`typing`
    superclasses (e.g., the `abc.ABC` superclass) *after* the list
    `typing` superclass in their method resolution order (MRO)
    flagrantly violate this assumption. Well, that's fine. We're fine
    with that. What's not fine about that? **Fine. This is fine.**
  * Avoids a circular import dependency. Previously, our caching
    `beartype.typing.Protocol` superclass leveraged the general-purpose
    `@beartype._util.cache.utilcachecall.callable_cached decorator` to
    memoize its subscription; however, since that decorator transitively
    imports from the `beartype.typing` subpackage, doing so induced a
    circular import dependency. To circumvent this, a new
    `@beartype.typing._typingcache.callable_cached_minimal` decorator
    implementing only the minimal subset of the full
    `@beartype._util.cache.utilcachecall.callable_cached` decorator has
    been defined; the `beartype.typing` subpackage now safely defers to
    this minimal variant for all its caching needs.
* **[PEP 563][PEP 563] compatibility.** @beartype now resolves [PEP
  563][PEP 563]-postponed **self-referential type hints** (i.e., type
  hints circularly referring to the class currently being decorated).
  **Caveat:** this support requires that external callers decorate the
  *class* being referred to (rather than the *method* doing the
  referring) by the `@beartype` decorator. For this and similar reasons,
  users are advised to begin refactoring their object-oriented codebases
  to decorate their *classes* rather than *methods* with `@beartype`.
* **[PEP 612][PEP 612] partial shallow compatibility.** @beartype now
  shallowly detects [PEP 612][PEP 612]-compliant `typing.ParamSpec`
  objects by internally associating such objects with our
  `beartype._data.hint.pep.sign.datapepsigns.HintSignParamSpec`
  singleton, enabling @beartype to portably introspect
  `Callable[typing.ParamSpec(...), ...]` type hints.
* **Static type-checking.** @beartype is now substantially more
  compliant with static type-checkers, including:
  * **Microsoft [pyright](https://github.com/microsoft/pyright) +
    [PyLance](https://marketplace.visualstudio.com/items?itemName=ms-python.vscode-pylance)
    + [VSCode](https://visualstudio.com).** @beartype now officially
    supports pyright, Microsoft's in-house static type-checker oddly
    implemented in pure-TypeScript, <sup>*gulp*</sup> resolving issues
    #126 and #127 kindly submitted by fellow Zelda aficionado @rbroderi.
    Specifically, this release resolves several hundred false warnings
    and errors issued by pyright against the @beartype codebase. It is,
    indeed, dangerous to go alone – but we did it anyway.
  * **mypy `beartype.typing.Protocol` compatibility.** The
    @beartype-specific `beartype.typing.Protocol` superclass implementing
    [PEP 544][PEP 544]-compliant fast caching protocols is now fully
    compatible with mypy, Python's official static type-checker.
    Specifically, `beartype.typing.Protocol` now circumvents:
    * python/mypy#11013 by explicitly annotating the type of its
      `__slots__` as `Any`.
    * python/mypy#9282 by explicitly setting the `typing.TypeVar()`
      `bounds` parameter to this superclass.
* **[PEP 647][PEP 647] compatibility.** @beartype now supports
  arbitrarily complex **[type
  narrowing](https://mypy.readthedocs.io/en/latest/type_narrowing.html)**
  in [PEP 647][PEP 647]-compliant static type-checkers (e.g., mypy,
  pyright), resolving issues #164 and #165 kindly submitted in parallel
  by foxy machine learning gurus @justinchuby (Justin Chuby) and @rsokl
  (Ryan Soklaski). Thanks to their earnest dedication, @beartype is now
  believed to be the most fully complete type narrower. Specifically,
  the return of both the `beartype.door.is_bearable()` function and
  corresponding `beartype.door.TypeHint.is_bearable()` method are now
  annotated by the [PEP 647][PEP 647]-compliant `typing.TypeGuard[...]`
  type hint under both Python ≥ 3.10 *and* Python < 3.10 when the
  optional third-party `typing_extensions` dependency is installed.
  Doing so substantially reduces false positives from static type
  checkers on downstream codebases deferring to these callables.
  Thanks so much for improving @beartype so much, @justinchuby and
  @rsokl!
* **`@{classmethod,staticmethod,property}` chaining.** The `@beartype`
  decorator now implicitly supports callables decorated by both
  `@beartype` *and* one of the builtin method decorators `@classmethod`,
  `@staticmethod`, or `@property` regardless of decoration order,
  resolving issue #80 kindly requested by @qiujiangkun (AKA, Type
  Genius-kun). Previously, `@beartype` explicitly raised an exception
  when ordered *after* one of those builtin method decorators. This
  releseae relaxes this constraint, enabling callers to list `@beartype`
  either before or after one of those builtin method decorators.
* **`beartype.vale.Is[...]` integration.** Functional validators (i.e.,
  `beartype.vale.Is[...]`) now integrate more cleanly with the remainder
  of the Python ecosystem, including:
  * **IPython.** Functional validators localized to a sufficiently
    intelligent REPL (e.g., IPython) that caches locally defined
    callables to the standard `linecache` module now raise
    human-readable errors on type-checking, resolving issue #123 kindly
    submitted by typing brain-child @braniii. Relatedly, @beartype now
    permissively accepts both physical on-disk files and dynamic
    in-memory fake files cached with `linecache` as the files defining
    an arbitrary callable.
  * **NumPy,** which publishes various **bool-like tester functions**
    (i.e., functions returning a non-`bool` object whose class defines
    at least one of the `__bool__()` or `__len__()` dunder methods and
    is thus implicitly convertible into a `bool`). Functional validators
    now support subscription by these functions, resolving issue #153
    kindly submitted by molecular luminary @braniii (Daniel Nagel).
    Specifically, @beartype now unconditionally wraps *all* tester
    callables subscripting (indexing) `beartype.vale.Is` with a new
    private `_is_valid_bool()` closure that (in order):
    1. Detects when those tester callables return bool-like objects.
    2. Coerces those objects into corresponding `bool` values.
    3. Returns those `bool` values instead.
* **Moar fake builtin types.**@beartype now detects all known **fake
  builtin types** (i.e., C-based types falsely advertising themselves as
  being builtin and thus *not* require explicit importation), succinctly
  resolving issue #158 kindly submitted by the decorous typing gentleman
  @langfield. Specifically, @beartype now recognizes instances of all of
  the following as fake builtin types:
  * `beartype.cave.AsyncCoroutineCType`.
  * `beartype.cave.AsyncGeneratorCType`.
  * `beartype.cave.CallableCodeObjectType`.
  * `beartype.cave.CallableFrameType`.
  * `beartype.cave.ClassDictType`.
  * `beartype.cave.ClassType`.
  * `beartype.cave.ClosureVarCellType`.
  * `beartype.cave.EllipsisType`.
  * `beartype.cave.ExceptionTracebackType`.
  * `beartype.cave.FunctionType`.
  * `beartype.cave.FunctionOrMethodCType`.
  * `beartype.cave.GeneratorCType`.
  * `beartype.cave.MethodBoundInstanceDunderCType`.
  * `beartype.cave.MethodBoundInstanceOrClassType`.
  * `beartype.cave.MethodDecoratorBuiltinTypes`.
  * `beartype.cave.MethodUnboundClassCType`.
  * `beartype.cave.MethodUnboundInstanceDunderCType`.
  * `beartype.cave.MethodUnboundInstanceNondunderCType`.
  * `beartype.cave.MethodUnboundPropertyNontrivialCExtensionType`.
  * `beartype.cave.MethodUnboundPropertyTrivialCExtensionType`.

## Compatibility Broken

* **Python 3.6.x support dropped.** This release unilaterally drops
  support for the Python 3.6.x series, which somnambulantly collided
  with its End-of-Life (EOL) a year ago and now constitutes a compelling
  security risk. Doing so substantially streamlines the codebase, whose
  support for Python 3.6.x required an unmaintainable writhing nest of
  wicked corner cases. We all now breathe a sigh of contentment in the
  temporary stillness of morning.
* **`beartype.cave` deprecation removals.** This release removes all
  deprecated third-party attributes from the `beartype.cave` submodule.
  The continued existence of these attributes substantially increased
  the cost of importing *anything* from our mostly undocumented
  `beartype.cave` submodule, rendering that submodule even less useful
  than it already is. Specifically, this release removes these
  previously deprecated attributes:
  * `beartype.cave.NumpyArrayType`.
  * `beartype.cave.NumpyScalarType`.
  * `beartype.cave.SequenceOrNumpyArrayTypes`.
  * `beartype.cave.SequenceMutableOrNumpyArrayTypes`.
  * `beartype.cave.SetuptoolsVersionTypes`.
  * `beartype.cave.VersionComparableTypes`.
  * `beartype.cave.VersionTypes`.

## Exceptions Improved

* **Colour** – the sensation formerly known as "color." @beartype now
  emits colourized type-checking violations (i.e.,
  `beartype.roar.BeartypeCallHintViolation` exceptions) raised by both
  `@beartype`-decorated callables *and* statement-level type-checkers
  (e.g., `beartype.door.die_if_unbearable()`,
  `beartype.door.TypeHint.die_if_unbearable()`), resolving issue #161
  kindly submitted by foxy machine learning expert @justinchuby (Justin
  Chu). When standard output is attached to an interactive terminal
  (TTY), ANSII-flavoured colours now syntactically highlight various
  substrings of those violations for improved visibility, readability,
  and debuggability. Since *all* actively maintained versions of Windows
  (i.e., Windows ≥ 10) now widely support ANSII escape sequences across
  both Microsoft-managed terminals (e.g., Windows Terminal) and
  Microsoft-managed Integrated Development Environments (IDEs) (e.g.,
  VSCode), this supports extends to Windows as well. The bad old days of
  non-standard behaviour are behind us all. Thanks *so* much to
  @justinchuby for his immense contribution to the righteous cause of
  eye-pleasing user experience (UX)!
* **Types disambiguated.** @beartype now explicitly disambiguates the
  types of parameters and returns that violate type-checking in
  exception messages raised by the `@beartype` decorator, resolving
  issue #124 kindly submitted by typing brain-child @braniii. Thus was
  justice restored to the QAverse.
* **Stack frame squelched.** @beartype now intentionally squelches
  (i.e., hides) the ignorable stack frame encapsulating the call to our
  private `beartype._decor._error.errormain.get_beartype_violation()`
  getter from the parent type-checking wrapper function generated by the
  :mod:`beartype.beartype` decorator, resolving issue #140 kindly
  submitted by @langfield (William Blake – yes, *that* William Blake).
  That stack frame only needlessly complicated visual inspection of
  type-checking violations in tracebacks – especially from testing
  frameworks like :mod:`pytest` that recapitulate the full definition of
  the `get_beartype_violation()` getter (including verbose docstring) in
  those tracebacks. Specifically, this release:
  * Renamed the poorly named `raise_pep_call_exception()` function to
    `get_beartype_violation()` for clarity.
  * Refactored `get_beartype_violation()` to return rather than raise
    `BeartypeCallHintViolation` exceptions (while still raising all
    other types of unexpected exceptions for robustness).
  * Refactored type-checking wrapper functions to directly raise the
    exception returned by calling `get_beartype_violation()`.
* **``None`` type.** The type of the ``None`` singleton is no longer
  erroneously labelled as a PEP 544-compliant protocol in type-checking
  violations. Let's pretend that never happened.
* **`beartype.abby.die_if_unbearable()` violations.** The
  `beartype.abby.die_if_unbearable()` validator function no longer
  raises non-human-readable exception messages prefixed by the
  unexpected substring `"@beartyped
  beartype.abby._abbytest._get_type_checker._die_if_unbearable()
  return"`. "Surely that never happened, @beartype!"

## Features Added

* **`beartype.door.** @beartype now provides a new public framework for
  introspecting, sorting, and type-checking type hints at runtime in
  constant time. N-n-now... hear me out here. @leycec came up with a
  ludicrous acronym and we're going to have to learn to live with it:
  the **D**ecidedly **O**bject-**O**rientedly **R**ecursive (DOOR) API.
  Or, `beartype.door` for short. Open the door to a whole new
  type-hinting world, everyone. `beartype.door` enables type hint
  arithmetic via an object-oriented type hint class hierarchy
  encapsulating the crude non-object-oriented type hint declarative API
  standardized by the :mod:`typing` module, resolving issues #133 and
  #138 kindly submitted by Harvard microscopist and general genius
  @tlambert03. The new `beartype.door` subpackage defines a public:
  * `TypeHint({type_hint})` superclass, enabling rich comparisons
    between pairs of arbitrary type hints. Altogether, this class
    implements a partial ordering over the countably infinite set of all
    type hints. Pedagogical excitement ensues. Instances of this class
    efficiently satisfy both the `collections.abc.Sequence` and
    `collections.abc.FrozenSet` abstract base classes (ABC) and thus
    behave just like tuples and frozen sets over child type hints.
    Public attributes defined by this class include:
    * A pair of `die_if_unbearable()` and `is_bearable()` runtime
      type-checking methods, analogous in behaviour to the existing
      `beartype.abby.die_if_unbearable()` and
      `beartype.abby.is_bearable()` runtime type-checking functions.
    * `TypeHint.is_bearable()`, currently implemented in terms of the
      procedural `beartype.abby.is_bearable()` tester.
    * An `is_ignorable` property evaluating to `True` only if the
      current type hint is semantically ignorable (e.g., `object`,
      `typing.Any`). There exist a countably infinite number of
      semantically ignorable type hints. The more you know, the less you
      want to read this changeset.
    * The equality comparison operator (e.g., `==`), enabling type hints
      to be compared according to semantic equivalence.
    * Rich comparison operators (e.g., `<=`, `>`), enabling type hints
      to be compared and sorted according to semantic narrowing.
    * A sane `__bool__()` dunder method, enabling type hint wrappers to
      be trivially evaluated as booleans according to the child type
      hints subscripting the wrapped type hints.
    * A sane `__len__()` dunder method, enabling type hint wrappers to
      be trivially sized according to the child type hints subscripting
      the wrapped type hints.
    * A sane `__contains__()` dunder method, enabling type hint wrappers
      to be tested for child type hint membership – just like builtin
      sets, frozen sets, and dictionaries.
    * A sane `__getindex__()` dunder method, enabling type hint wrappers
      to be subscripted by both positive and negative indices as well as
      slices of such indices – just like builtin tuples.
  * `beartype.door.AnnotatedTypeHint` subclass.
  * `beartype.door.CallableTypeHint` subclass.
  * `beartype.door.LiteralTypeHint` subclass.
  * `beartype.door.NewTypeTypeHint` subclass.
  * `beartype.door.TupleTypeHint` subclass.
  * `beartype.door.TypeVarTypeHint` subclass.
  * `beartype.door.UnionTypeHint` subclass.
  * `is_subtype({type_hint_a}, {type_hint_b})` function, enabling
    @beartype users to decide whether any type hint is a **subtype**
    (i.e., narrower type hint) of any other type hint.
  * `beartype.roar.BeartypeDoorNonpepException` type, raised when the
    `beartype.door.TypeHint` constructor is passed an object that is
    *not* a PEP-compliant type hint currently supported by the DOOR API.
  Thanks so much to @tlambert03 for his phenomenal work here. He ran
  GitHub's PR gauntlet so that you did not have to. Praise be to him.
  Some people are the living embodiment of quality. @tlambert03 is one
  such people.
* **`beartype.peps`.** @beartype now publicizes runtime support for
  `typing`-centric Python Enhancement Proposals (PEPs) that currently
  lack official runtime support via a new public subpackage:
  `beartype.peps`. Notably, @beartype now provides:
  . Specifically, this commit:
  * A new public `beartype.peps.resolve_pep563()` function resolving
    [PEP 563][PEP 563]-postponed type hints on behalf of third-party
    Python packages. This function is intended to be "the final word" on
    runtime resolution of [PEP 563][PEP 563]. May no other third-party
    package suffer as we have suffered. This commit is for you,
    everyone. And "by everyone," we of course mostly mean @wesselb of
    [Plum](github.com/wesselb/plum) fame. See also beartype/plum#53.
* **`beartype.vale.Is*[...] {&,|}` short-circuiting.** `&`- and
  `|`-chained beartype validators now explicitly short-circuit when
  raising human-readable exceptions from type-checking violations
  against those validators, resolving issue #125 kindly submitted by
  typing brain-child @braniii.

## Features Optimized

* **`beartype.abby.is_bearable()` when returning `False`.** Previously,
  the public `beartype.abby.is_bearable()` runtime type-checker behaved
  reasonably optimally when the passed object satisfied the passed type
  hint but *extremely* suboptimally when that object violated that hint;
  this was due to our current naive implementation of that tester using
  the standard Easier to Ask for Permission than Forgiveness (EAFP)
  approach. This release fundamentally refactored
  `beartype.abby.is_bearable()` in terms of our new private
  `beartype._check.checkmake.make_func_tester()` type-checking tester
  function factory function. Ad-hoc profiling shows a speedup on
  the order of eight orders of magnitude – the single most intense
  optimization @beartype has ever brought to bear (*heh*). Our core code
  generation API now transparently generates both:
  * **Runtime type-checking testers** (i.e., functions merely returning
    ``False`` on type-checking violations).
  * **Runtime type-checking validators** (i.e., functions raising
    exceptions on type-checking violations).
* **[PEP 604][PEP 604]-compliant new unions** (e.g., `int | str |
  None`). Since these unions are **non-self-caching type hints** (i.e.,
  hints that do *not* implicitly cache themselves to reduce space and
  time consumption), @beartype now efficiently coerces these unions into
  singletons in the same manner as [PEP 585][PEP 585]-compliant type
  hints – which are similarly non-self-caching.

## Features Deprecated

* **`beartype.abby` → `beartype.door`.** This release officially
  deprecates the poorly named `beartype.abby` subpackage in favour of
  the sorta less poorly named `beartype.door` subpackage, whose name
  actually means something – even if that something is a punny acronym
  no one will ever find funny. Specifically:
  * `beartype.abby.die_if_unbearable()` has been moved to
    `beartype.door.die_if_unbearable()`.
  * `beartype.abby.is_bearable()` has been moved to
    `beartype.door.is_bearable()`.
  To preserve backward compatibility, the `beartype.abby` subpackage
  continues to dynamically exist (and thus be importable from) – albeit
  as a deprecated alias of the `beartype.door` subpackage.

## Deprecations Resolved

* **Setuptools licensing.** This release resolves a mostly negligible
  `setuptools` deprecation warning concerning the deprecated
  `license_file` setting in the top-level `setup.cfg` file. *Next!*

## Tests Improved

* **[PEP 544][PEP 544] compatibility.** All [PEP 544][PEP 544]-specific
  test type hints have been generalized to apply to both the non-caching
  `typing.Protocol` superclass *and* our caching
  `beartype.typing.Protocol` superclass.
* **[PEP 561][PEP 561] compatibility via pyright.** Our test suite now
  enforces static type-checking with `pyright`. Notably:
  * A new `test_pep561_pyright` functional test statically type-checks
    the @beartype codebase against the external `pyright` command in the
    current `${PATH}` (if available) specific to the version of the
    active Python interpreter currently being tested. For personal
    sanity, this test is currently ignored on remote continuous
    integration (CI) workflows. Let this shrieking demon finally die!
  * The private `beartype_test.util.cmd.pytcmdrun` submodule underlying
    our cross-platform portable forking of testing subprocesses now
    transparently supports vanilla Windows shells (e.g., `CMD.exe`,
    PowerShell).
* **Tarball compatibility.** `beartype` may now be fully tested from
  non-`git` repositories, including source tarballs containing the
  `beartype_test` package. Previously, three functional tests making
  inappropriate assumptions about the existence of a top-level `.git/`
  directory failed when exercised from a source tarball.
* **Sphinx documentation.** Our test suite now exercises that our
  documentation successfully builds with Sphinx via a new
  `test_sphinx_build()` functional test. This was surprisingly
  non-trivial – thanks to the `pytest`-specific `sphinx.testing`
  subpackage being mostly undocumented, behaving non-orthogonally, and
  suffering a host of unresolved issues that required we monkey-patch
  the core `pathlib.Path` class. Insanity, thy name is Sphinx.
* **GitHub Actions dependencies bumped.** This release bumps our GitHub
  Actions-based continuous integration (CI) workflows to both the
  recently released `checkout@v3` and `setup-python@v3` actions,
  inspired by a pair of sadly closed PRs by @RotekHandelsGmbH CTO
  @bitranox (Robert Nowotny). Thanks so much for the great idea,
  @bitranox!
* **`beartype.door` conformance.** A new smoke test guarantees
  conformance between our DOOR API and abstract base classes (ABCs)
  published by the standard `typing` module.
* **python/mypy#13627 circumvention.** This release pins our GitHub
  Actions-based CI workflow to Python 3.10.6 rather than 3.10.7,
  resolving a mypy-specific complaint inducing spurious test failures.

## Documentation Improved

* **[`beartype.abby`
  documented](https://github.com/beartype/beartype#beartype-at-any-time-api).**
  The new "Beartype At Any Time API" subsection of our front-facing
  `README.rst` file now documents our public `beartype.abby` API,
  resolving issue #139 kindly submitted by @gelatinouscube42 (i.e., the
  user whose username is the answer to the question: "What is the
  meaning of collagen sustainably harvested from animal body parts?").
* **[GitHub Sponsors activated](https://github.com/sponsors/leycec).**
  @beartype is now proudly financially supported by **GitHub Sponsors.**
  Specifically, this release:
  * Defines a new GitHub-specific funding configuration (i.e.,
    `.github/FUNDING.yml`).
  * Injects a hopefully non-intrusive advertising template
    <sup>*gulp*</sup> at the head of our `README.rst` documentation.
* **Sphinx configuration sanitized.** As the first tentative step
  towards chain refactoring our documentation from its current
  monolithic home in our top-level `README.rst` file to its eventual
  modular home at [ReadTheDocs (RTD)](https://beartype.readthedocs.io),
  en-route to resolving issue #8 (!) kindly submitted a literal lifetime
  ago by visionary computer vision export and long-standing phenomenal
  Finn @felix-hilden (Felix Hildén):
  * Our core Sphinx configuration has been resurrected from its early
    grave – which now actually builds nothing without raising errors. Is
    this an accomplishment? In 2022, mere survival is an accomplishment!
    So... *yes.* Significant improvements include:
    * Activation and configuration of the effectively mandatory
      `autosectionlabels` builtin Sphinx extension.
  * Our `doc/source/404.rst` file has been temporarily moved aside,
    resolving a non-fatal warning pertaining to that file. Look, we're
    not here to actually solve deep issues; we're here to just get
    documentation building, which it's not. Sphinx, you have much to
    answer for.
  * Our top-level `sphinx` entry point now:
    * Temporarily disables Sphinx's nit-picky mode (i.e., the `-n`
      option previously passed to `sphinx-build`) due to Sphinx's
      `autodoc` extension locally failing to generate working
      references.
    * Unconditionally disables Sphinx caching by forcing *all* target
      documentation files to be rebuilt regardless of whether their
      underlying source files have since been modified or not, obviating
      spurious build issues.

  [PEP 484]: https://www.python.org/dev/peps/pep-0484/
  [PEP 544]: https://www.python.org/dev/peps/pep-0544/
  [PEP 561]: https://www.python.org/dev/peps/pep-0561/
  [PEP 563]: https://www.python.org/dev/peps/pep-0563/
  [PEP 585]: https://www.python.org/dev/peps/pep-0585/
  [PEP 604]: https://www.python.org/dev/peps/pep-0604/
  [PEP 612]: https://www.python.org/dev/peps/pep-0612/
  [PEP 647]: https://www.python.org/dev/peps/pep-0647/

(*Impossible journey on an implacable placard-studded gurney!*)
HexDecimal added a commit to HexDecimal/mypy that referenced this issue Jun 22, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug mypy got something wrong topic-protocols
Projects
None yet
Development

Successfully merging a pull request may close this issue.

2 participants