Using an alias for a PEP 593 Annotation with multiple arguments.

injector.py

"""
Using an alias for a PEP 593 Annotation with multiple arguments.

I'm rewriting my injector.
I currently put the injector information in dataclass field metadata.
I'd like to allow other callables such as functions, so I'm looking at PEP 593 Annotations, similar to what [the inject package](https://github.com/alecthomas/injector/blob/master/injector/__init__.py#L1116) does.

My injection though needs zero or more arguments, sort of like an RxPY pipeline.

# Update 1: Changed from Markdown science fiction to using code in a running test
"""

def test_handle_field_injected_customer_experiment(this_container, this_injector, this_injector2):
    this_container.register_singleton(Customer(), Customer)

    InjectT = TypeVar('InjectT')

    def simple_factory(customer: Customer):
        return customer

    result = this_injector(target=simple_factory)
    assert 'Base Customer' == result.name

    # ==============================================================
    # Let's start using Annotated so we can bring in the other
    # features from the current injector.
    def annotated_factory(customer: Annotated[Customer, _inject_marker]):
        return customer

    result = this_injector(target=annotated_factory)
    assert 'Base Customer' == result.name

    # ==============================================================
    # The current injector lets the return type be different than
    # the lookup type. We want the result to be FrenchCustomer,
    # but ask the injector for the registered Customer.

    def frenchcustomer_factory(customer: Annotated[FrenchCustomer, Customer, _inject_marker]):
        # The injector looked up "Customer" which
        return customer

    result = this_injector2(target=frenchcustomer_factory)
    assert 'French Customer' == result.name

    # ==============================================================
    # The current injector also lets you pluck off just an
    # attribute, which is handy for building a props-based
    # component system.

    def attr_factory(customer_name: Annotated[str, Customer, Attr('name'), _inject_marker]):
        # The injector looked up "Customer" which
        return f'Name: {customer_name}'

    result = this_injector2(target=attr_factory)
    assert 'Name: French Customer' == result

    # ==============================================================
    # That's getting a little noisy. Would be nice to use a TypeAlias
    # to at least cut down that last part. But the alias fails at
    # runtime with:
    #   TypeError: Too many parameters for typing.Annotated[~InjectT,
    #      <object object at 0x10a4ce8c0>]; actual 3, expected 1
    #
    # mypy fails with:
    # tests/test_593_injector.py:177: error: Bad number of arguments for type alias, expected: 1, given: 3
    # tests/test_593_injector.py:177: error: Invalid type comment or annotation
    # tests/test_593_injector.py:177: note: Suggestion: use Attr[...] instead of Attr(...)

    Injected = Annotated[InjectT, _inject_marker]

    def injected_factory(customer_name: Injected[str, Customer, Attr('name')]):
        # The injector looked up "Customer" which
        return f'Name: {customer_name}'

    result = this_injector2(target=injected_factory)
    assert 'Name: French Customer' == result

Are you using mypy? You could post this in the mypy tracker -- or in the typing tracker, if it's independent from which type checker is used. (Have you tried others? You don't show any error messages, you just say "if fails" -- the errors usually give some hints as to what's wrong.)

I suspect that the type alias creates a straightforward generic type, just like

T = TypeVar("T")
A = Dict[str, T]
x: A[int]  # Dict[str, int]
y: A  # Dict[str, Any]

I don't think variadic aliases are supported, not even with PEP 613's TypeAlias.

@gvanrossum Yes, I'm basing this on mypy at first. I was really hoping to not post a tracker question, as it might just be noise at this point. The error is about the generic getting too many arguments. I will link to something in a repo that shows passing/failing tests.

Perhaps I shouldn't worry about using Annotated without a marker saying its for my system. I'm only doing the alias to avoid that.