simplify ser-as-any mechanism

[davidhewitt]

Change Summary

Built atop #1359

The concern I had with that PR is that the "fix" seemed to be a strange special case, which to me implied that there was probably a simplification which would decouple the problem.

Indeed, in this PR I instead propose that when serialize_as_any=True then the schema is immediately dropped in favour of an AnySerializer, and the extra records serialize_as_any so that if any serializers are re-entered (via inference) then they immediately discard their schema too.

Related issue number

Fix pydantic/pydantic#9670

Checklist

• Unit tests for the changes exist
• Documentation reflects the changes where applicable
• Pydantic tests pass with this pydantic-core (except for expected changes)
• My PR is ready to review, please add a comment including the phrase "please review" to assign reviewers

[davidhewitt]

Hmm, looks like this needs another pass. Will investigate later.

[codspeed-hq]

CodSpeed Performance Report

Merging #1478 will not alter performance

_{Comparing dh/simpler-ser-as-any (c58b54f) with main (dc4846e)}

Summary

✅ 155 untouched benchmarks

[davidhewitt]

This now works, and it also fixes the bug, but I'm not sure it's really any simpler. To make things hang together I had to rework all serializer calls inside fields.rs to dispatch on serialize_as_any. At least it means that the logic is more concentrated than it was before, which helps avoid the original recursive-ref issue.

@sydney-runkle happy to take your call on whether this PR is a good idea.

(The pydantic-integration test probably needs running, so this might want to wait until we've dealt with the invalid-schema changes.)

[sydney-runkle]

(The pydantic-integration test probably needs running, so this might want to wait until we've dealt with the invalid-schema changes.)

I think we could probably just do a patch release with that change + this one to make CI happy :).

[sydney-runkle]

This LGTM, definitely feels like a simplification here and is much easier to understand if we just use the boolean pattern 👍

[davidhewitt]

Maybe merging #1480 will give us all we need to measure here.

[davidhewitt]

Ungh, will investigate the test failures when I get a chance.