PEP647 typing.TypeGuard based is_ok() and is_err() helpers

PEP 647 (https://www.python.org/dev/peps/pep-0647/) makes it possible to write custom type guard functions.

this makes it possible to write shorthand is_ok(...) and is_err(...) functions that effectively work as shorthands for the longer isinstance(..., ...) forms, resulting in denser and more readable code, while still keeping all the benefits of strict type checking:

def is_ok(result: Result[T, E]) -> TypeGuard[Ok[T]]:
    return isinstance(result, Ok)


def is_err(result: Result[T, E]) -> TypeGuard[Err[E]]:
    return isinstance(result, Err)

not all supported python versions support TypeGuard, but the widely used typing_extensions package (https://github.com/python/typing/blob/master/typing_extensions/README.rst) has a functional backport that works with (at least) mypy. i personally do not see a problem adding a dependency on typing_extensions, since it’s maintained ‘upstream’, close to python itself, and it will already be pulled in by many other libraries in many real-world applications.

thoughts? opinions? (cc @francium @dbrgn)

[jack-michaud]

It seems like mypy generic typeguards are fixed, but there’s still no negation.

from result import Ok, Result
from typing import TypeVar
from typing_extensions import TypeGuard


E = TypeVar("E")
T = TypeVar("T")


def is_ok(result: Result[T, E]) -> TypeGuard[Ok[T]]:
    return isinstance(result, Ok)

def main(result: Result[str, int]):
    if is_ok(result):
        reveal_type(result)
    else:
        reveal_type(result)

test.py:20: note: Revealed type is "result.result.Ok[builtins.str]"
test.py:22: note: Revealed type is "Union[result.result.Ok[builtins.str], result.result.Err[builtins.int]]"

Doing something like this would work:

def main(result: Result[str, int]):
    if is_ok(result):
        reveal_type(result)
    elif is_err(result):
        reveal_type(result)

test.py:20: note: Revealed type is "result.result.Ok[builtins.str]"
test.py:22: note: Revealed type is "result.result.Err[builtins.int]"

…but ideally we wouldn’t need to do this.

Related discussion for typeguard negations: https://github.com/python/typing/discussions/1013

[dbrgn]

If only older Python versions need it, I don’t think adding a dependency is a problem. By upgrading your Python, you can get rid of the dependency.

We did the same thing with typing when we still supported Python versions that didn’t have this built-in yet.

From my experience with TS, type guards are pretty nice to have.