Initial pika stubs

[srittau]

https://github.com/pika/pika

These stubs are basically create_baseline_stubs + fixes. I have more annotations lined up in later PRs. This PR is best reviewed commit by commit.

[AlexWaygood]

I skimmed -- mostly looks good! Just a few things I spotted.

Do we need all the LOGGER annotations -- are these meant to be exposed to end users? If not, maybe it would be better to allowlist them?

[AlexWaygood]

I guess you could plausibly want to monkey-patch these for testing purposes, which would make the Final[] annotation a little annoying?

[srittau]

On the other hand, outside of testing, overriding these would potentially be fatal. Even for testing, changing these is potentially fragile.

[AlexWaygood]

I don't have a strong opinion — it was just a thought :) feel free to ignore

[srittau]

Do we need all the LOGGER annotations -- are these meant to be exposed to end users? If not, maybe it would be better to allowlist them?

Probably not, but the same is usually true for lots of implementation detail. I'm not sure why we should special case loggers.

[AlexWaygood]

I'm not sure why we should special case loggers.

Well, I don't see it as special-casing loggers exactly, since, in general, I think we should limit the implementation details we expose in stubs to the bare minimum :)

[AlexWaygood]

Looks good (other than my minor nit about the LOGGER variables)

[srittau]

This was discussed in the past. Personally, I agree with you. If I had my way, we'd just include public API (with well-documented exceptions). This would help with the maintenance nightmare that SQLAlchemy or just now fpdf2 is. But IIRC it was decided way back, that basically all "accessible" items should be typed. I can't find the exact discussion now, though.

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

[JelleZijlstra]

This was discussed in the past. Personally, I agree with you. If I had my way, we'd just include public API (with well-documented exceptions). This would help with the maintenance nightmare that SQLAlchemy or just now fpdf2 is. But IIRC it was decided way back, that basically all "accessible" items should be typed. I can't find the exact discussion now, though.

For what it's worth I don't recall such a discussion. I think we should include everything that's "clearly public" (e.g. if it's mentioned in documentation or __all__), and skip private-looking stuff unless people explicitly ask for it.

[srittau]

Maybe we should discuss it again then, since most of us seem to agree. But currently we do the opposite and stubtest kind of pushes us towards including everything that's not explicitly marked as private.