feat/river schema v2 compatibility

[blast-hardcheese]

Why

River Schema v2 compatibility so we can start getting rid of the compatibility layer.

This PR is just the first, it does not introduce the half-close semantics yet.

What changed

• init is now required for all method types, input is now optional. This simplifies rpc and stream codegen.
• Fixed some bugs where init was being used in input position and vice versa.
• Swapping f"'{foo}'" encoding for f"{repr(foo)}", giving greater safety post-generation (to avoid situations where f"'{None}'" gets rendered as "None". f"{repr(None)}" would render as None which would fail typechecks.)
• Simplify optional return
• Swapping List and Dict to list and dict
• Swapping encode_... lambdas for def, to placate both line length as well as lint suggestions (and increase readability)

Test plan

Manually ran codegen against v2 generated schemas, everything typechecks.

[masad-frost]

Do you wanna set up a different base branch in case we need to send a patch against main?

[masad-frost]

huh, how did this even work lol

[blast-hardcheese]

It's equivalent in the happy path but repr will surface bugs in the not happy path

[masad-frost]

Ideally the type is if request exists then request_serializer should not be optional.

[blast-hardcheese]

Maybe it should be request: Optional[ tuple[ {request_type}, {request_serializer_type} ] ]

[masad-frost]

first_message should be no longer needed since we always send a STREAM_OPEN_BIT and we can remove the bit from send_close_stream

[blast-hardcheese]

Based on how the code worked previously, if we had an exception during sending the first message, then STREAM_OPEN_BIT was sent in the send_close_stream -- was that not intentional?

[masad-frost]

Suggested change

	control_flags = 0
	await self.send_message(
	stream_id=stream_id,
	service_name=service_name,
	procedure_name=procedure_name,
	control_flags=control_flags,
	await self.send_message(
	stream_id=stream_id,
	service_name=service_name,
	procedure_name=procedure_name,
	control_flags=0,

[masad-frost]

all this stuff is hard to review (and I'm sure iterate on), we should set up some simple tests. Snapshot testing is probably good enough https://pypi.org/project/pytest-snapshot/.

I did this for river js replit/river#261 using https://vitest.dev/guide/snapshot

[blast-hardcheese]

The best way I have found to iterate on this is to just run the codegen against different production schemas and verify that the set of changes is what I expect. I'm happy to wait until after I land the babel changes before merging this.

The codegen currently isn't compositional enough to make snapshot testing manageable. The more I work with it though I've been trying to come up with strategies to make it less imperative without increasing cognitive load too much, I'll land snapshot testing then.

[masad-frost]

What's the difference?

I wonder if it's like the stupid typescript thing where String (object) is different from string (literal/primitive) 😬

[blast-hardcheese]

Python 3.8 and below did not permit the builtin types to be subscripted, typing.List and typing.Dict were added as future-compatible shims during the migration. Python 3.8 is EOL as of the end of Spooky Season, so now we can 🚀.

[blast-hardcheese]

Do you wanna set up a different base branch in case we need to send a patch against main?

If we need to patch, we should do feature patches off previous release tags. Since we're reserving 1.x for GA, I can follow the 0.200.0 strategy in this repo for schema-v2 releases.

Doing that though would suggest that we do at least a halfway decent pass at the v2 server codegen as well though. Given what we discussed supporting v1/v2 using a naive shim doesn't seem too bad, probably can get that done this week.

Wdyt?