`store.list` overloads not working in pylance when `arro3-core` not installed

[ion-elgreco]

The typehint for store.list doesn't seem to work well with pyright:

for batch in list.stream(), always makes batch an unkown type

[kylebarron]

I've noticed this. Can you check whether mypy handles it correctly? I believe it does IIRC.

It might be an edge case that pyright doesn't support? Or perhaps we should create an issue in pyright, assuming that mypy indeed handles it correctly.

[ion-elgreco]

@kylebarron I think it does yeah. I've used overloads as well in deltalake and there pyright shows them correctly, haven't had time to dig into it deeper.

[kylebarron]

Actually I can't reproduce this: https://github.com/kylebarron/obstore-typing-repro

In particular, it was saying unknown originally for me until I installed arro3-core. arro3-core isn't a required dependency because the Arrow support is optional, but the type hint will fail if it's not installed.

[ion-elgreco]

I see, yeah arro3 wasn't installed in my venv. So makes sense

[kylebarron]

Ok cool, so I think this issue is resolved then? The type hints work as long as arro3 is installed? I think this is reasonable because arro3 has to be installed for it to work at runtime anyways

[ion-elgreco]

Well not entirely, the type hint also didn't work with arrow= False

[kylebarron]

Oh ok, I can reproduce that when arro3-core is not installed, none of the overloads work in the pylance case

I think that's a valid bug to file upstream in pylance. I don't think there's anything we can do about it in obstore? Any ideas? These are static overloads so we can't switch on whether something is installed.

[kylebarron]

Looks like we might be able to work around this by not defining the constraints in the type hint here:

obstore/obstore/python/obstore/_list.pyi

Lines 48 to 54 in fbaeb7f

	ListChunkType = TypeVar( # noqa: PYI001, PLC0105
	"ListChunkType",
	Sequence[ObjectMeta],
	RecordBatch,
	Table,
	covariant=True,
	)

See #578

It looks like the type check against obspec is succeeding but CI is failing for other reasons (#579)