asn1: Support type aliases

[loewexy]

While working with the new asn1 module i stubled upon one limitation when using type aliases. For example this excerpt from the documentation:

@asn1.sequence
class Validity:
    not_before: asn1.UTCTime | asn1.GeneralizedTime
    not_after: asn1.UTCTime | asn1.GeneralizedTime

I would be nice to instead write:

type Time = asn1.UTCTime | asn1.GeneralizedTime

@asn1.sequence
class Validity:
    not_before: Time
    not_after: Time

However this is currently not supported by cryptography. I think it would be nice to also support type aliases, so common combinations can be declared and reused.

[alex]

This seems reasonable -- assuming Python provides a way to map a type alias to the underlying resolved types at runtime. Would you be interested in submitting a PR for this?

[loewexy]

Oh i just understood that the asn1 implementation is fine with the older syntax:

Time = typing.Union[asn1.UTCTime, asn1.GeneralizedTime]

@asn1.sequence
class Validity:
    not_before: Time
    not_after: Time

Therefore the problem is only interpreting the new PEP 695 type aliases. They do not create a typing.Union but instead a typing.TypeAliasType which we would have to resolve for its value. If i am not mistaken this has to be done in _normalize_field_type once before parsing anotations once afterwards, because an Annotated type can be within a type alias, but also a type alias could be in the annotation.

I will prepare a PR for the implementation. But currently I am not sure where a test case for that would fit the rest of the codebase.

[alex]

Testing is an interesting question, since we need a file that's only imported on newer pythons.... Will require some creativity

[loewexy]

I have opened a pull request with the implementation I built. However, tests are still missing and i think the feature should get some.