typing: Improve FixtureDefinition and FixtureDef

[nicoddemus]

• Carry around parameters and return value in FixtureFunctionDefinition.
• Add FixtureParams to FixtureDef.

Follow up to #12473.

[nicoddemus]

cc @Glyphack

[nicoddemus]

Hmmm turns out pytest-bdd uses FixtureDef in a few places:

https://github.com/search?q=repo%3Apytest-dev%2Fpytest-bdd%20FixtureDef&type=code

The new Generic parameter breaks it, and there is no support for default values in Generic parameters yet...

[bluetech]

Is the FixtureDef change needed for the FixtureFunctionDefinition change?

[nicoddemus]

Is the FixtureDef change needed for the FixtureFunctionDefinition change?

I don't think so. Should I try to keep only the change to FixtureFunctionDefinition?

[RonnyPfannschmidt]

The paramspec types give me a slight headache as we basically resolve them internally and I wonder if there's a good way to make them work in passover

[RonnyPfannschmidt]

I'm starting to get the impression we need a type alias for those dicts as they land in too many places

[nicoddemus]

Done.

[bluetech]

I don't think so. Should I try to keep only the change to FixtureFunctionDefinition?

FixtureDef is internal and we mostly use object or Any as the type parameter, so if it breaks plugins we might as well keep it as is.

[nicoddemus]

FixtureDef is internal and we mostly use object or Any as the type parameter, so if it breaks plugins we might as well keep it as is.

Done.

[nicoddemus]

TBH I'm second guessing if this is useful at all. In the end we cannot do proper type checking because we need to keep all the FixtureDefinition instances in the same data structure, so it all ends up being Any anyway...

[bluetech]

I agree for FixtureDef, but FixtureFunctionDefinition is not stored in any datastructure, it's a decorator which wraps fixture functions. I think we should try to preserve the typing of fixture functions, even if we don't allow calling them. I can imagine some e.g. mypy plugin introspecting it. Also IDEs, etc.