Add source generator for HandleInputCache

[smoogipoo]

Prereqs:

• Refactor source generation for abstraction #5760

Description	Before	After
Startup
Options

[bdach]

I think we're gonna need to disable the suspicious type check inspections here because it doesn't look like R# can figure them out in the context of sourcegen...? Unless they're gonna go away after the sourcegen package is bumped I guess. (I wouldn't be surprised if they didn't, though.)

[smoogipoo]

They'll go away with a package bump :)

[bdach]

I've re-read this - all code review suggestions look applied, source looks sane. That said, I'd probably want to spend a bit of time setting up local unit test runs for both framework and game with this sourcegen enabled, to see if there are any regressions, before I'd be 100% comfortable with leaving an approval.

If you've already done this, or don't think that's necessary and that we can play it by ear whenever, then I can approve this and we can move forward. Otherwise, I'll set a few test runs up, but probably no sooner than this weekend.

[smoogipoo]

Reviewing this over again myself for the 5th time now... I believe there's still an issue if a non-partial class is the leaf node.

./sigh

I think this is going to have to end up something like dependency injection after all.

[bdach]

I'll hold off on the testing for now then until the code is a final state.

[smoogipoo]

Added back KnownType as a method to handle non-partial leaf types.

This PR is now once again ready to go... I hope.

Just to be clear, this structure should be reviewed as a basis for all other such "simple" SGens. The IsLongRunning SGen is another likely candidate to use this model.

[bdach]

So today I made a local sourcegen package, hooked it up to framework and a game checkout, ran full test suites on both (all passing), and confirmed the performance gains. I'm not sure it's time effective to be doing any further manual testing of this, so as far as I'm concerned this is probably as good to go as can get, and I'm gonna leave an approval.

Not sure how we want to approach releasing this, however. What I'd probably do is:

• get next game release out without this change
• force merge this to bypass code quality
• immediately after, get a new sourcegen package out and bump it in a PR / on master even to make the code quality inspection go away
• after that, get a framework package out with the sourcegen active, and bump framework on game-side so that we can dogfood this change during the interval between next release and release-after-next and see if anything funny falls out

But not sure my opinion holds any water here. @ppy/team-client thoughts?

[smoogipoo]

What we've done in the past is pushed an SGen nuget package from the PR, and updated the package ref in the PR.

[peppy]

I'll go ahead with making the nuget package.

@bdach fwiw you didn't approve

[bdach]

whoops fixed