Make Field constructors keyword-only

[akx]

Description

This PR makes the Field constructor keyword-only by adding a * (and turning *args, **kwargs into just **kwargs in subclasses). I don't believe anyone is (or should be) constructing fields as serializers.CharField(False, True, False, serializers.empty, ...)

The root issue this fixes is the nonobvious problem in e.g.

aliases = serializers.ListField(serializers.CharField(), default=list)

– you wouldn't expect that yields a non-validated read-only list field, would you? 😄 What happens is the positional serializers.CharField() argument gets passed up into the serializers.Field() call as the first argument, which just so happens to be read_only, and the field instance is a truthy value...

/cc @magdapoppins for having initially bumped into this issue.

[auvipy]

could you please fix the conflicts?

[akx]

@auvipy Rebased.

[akx]

@auvipy Can we get this merged? :)

[auvipy]

I am not maintainer :D

[akx]

Fair. @tomchristie or someone else, can you take a look?

[tomchristie]

Yup seems pretty reasonable.
We'll want to call it out clearly in the release notes, tho.

[tomchristie]

Based on past experience I'd expect that someone somewhere may well have some code updating to do after this change, but let's see. And yes it clearly makes sense.

[tomchristie]

Based on past experience I'd expect that someone somewhere may well have some code updating to do after this change, but let's see.

And there we go: fdb4931#r54532785

We might reconsider this given that we've already seen a user hit a pain point with the change, even without having released a new version. Unsure.

[auvipy]

documenting this would be the right choice, IMHO

[akx]

Cc @afparsons for having bumped into that issue in fdb4931#r54532785:

I believe the change in this PR is net positive and will likely uncover some bugs in downstream code even if it might need some work to adapt. In that sense I agree with @auvipy that it'd be better to just document this loudly for the next release.

[asottile-sentry]

I'm extremely late to the party but this now makes MultipleChoiceField and ChoiceField have incompatible signatures -- ChoiceField has choices as a positional argument and now it's required to be named only in MultipleChoiceField

unclear if this was intended -- but it certainly strikes me as odd