Expose FixtureDef and SubRequest

[asvetlov]

Export names required by pytest_fixture_setup and pytest_fixture_post_finalizer hook specs.

[RonnyPfannschmidt]

@bluetech im wondering, would there be a simple way to type the Sub-request with a FixtureParameter type as well

the idea bein that one coud use a type for the fixture param instead of any

[asvetlov]

@RonnyPfannschmidt FixtureParameter doesn't exist on the current main branch.
I assume you mean FixtureValue.

Now SubRequest is initialized with fixturedef=FixtureDef[object].
Sure, we can derive SubRequest from Generic[FixtureValue] and change contructor's argument to fixturedef=FixtureDef[FixtureValue].

If you mean something like this -- I can prepare a separate pull request.

The PR for exposing new public names is orthogonal to type system improvement and can be approved/merged independently.

[bluetech]

Hi @asvetlov,

Regarding SubRequest, please see #8763 (comment). The situation hasn't changed since then unfortunately, it needs work.

Regarding FixtureDef, it needs a bit of an audit before exposing it. From looking at it now:

• The docstring "A container for a factory definition." is a bit obscure, we should improve it (but that's preexisting, not blocking).

• Need to decide if we want the constructor to be public or private. If it is public and exposed in pytest, that means pytest is committing to keeping it stable/backward compatible. Also, in this case the ctor parameters need to be documented.

  The conservative approach would be to make it private for now. Our mechanism for doing it is described here. But before doing so we should check if there are many plugins which construct FixtureDef. (I have a large pytest plugins corpus which I use to check for such things, but I'm not on that machine ATM).

• Should document the non-underscored attributes, if they should be public, or make them private, if not. Need to audit them...

• Audit the methods:

  • scope -> Good
  • addfinalizer -> Good, should add a docstring
  • finish and execute -> Not sure about these. They also need SubRequest to be public. Need to see if/how the fixtures are meant to be using these.
  • cache_key -> I think it's probably better making it private. Might need a deprecation.

[bluetech]

Let's make the changelog refer to issue #7469 instead of the PR.

[bluetech]

@RonnyPfannschmidt FixtureParameter doesn't exist on the current main branch. I assume you mean FixtureValue.

There are two things:

First, FixtureValue, also present on FixtureRequest, which is the type of the return value of the fixture.

@RonnyPfannschmidt is talking about request.param, which is the value of the parameter in a parametrized fixture. It is present only on SubRequest and is currently not typed. (Additional complication is that it's not always set - only for parametrized fixtures).

[RonnyPfannschmidt]

Hence the idea of the adding a generic for the case of having a parameter and a normal type when not

[bluetech]

FixtureDef is exposed now with making the ctor private and minimal exposing of internals (fb55615).

[bluetech]

I'll close this since just exposing SubRequest as-is is not going to be good.