Compatibility with runtime type checkers about onnxscript HOT 4 OPEN

@gramalingam mypy doesn't seem to play nicely with these types

https://github.com/microsoft/onnx-script/blob/b85fb550c2cf7ab37d57e4e00c9a0085461ce985/onnxscript/onnx_types.py#L67-L82

is there a way we can reconcile this?

from onnxscript.

I think there is non-trivial work to be done for integrating the multiple requirements here. There are one or two things to be done first:
(a) Aaron has a first extension to enable intellisense in visual-code and other environments.
(b) We need to integrate the type annotation classes with the actual runtime class (tensor.Tensor).
As for integrating it with other runtime and static type-checkers, that could be more complex. I don't have an answer off the top of my head. It's worth looking into. If you have suggestions, please let me know, thanks.

from onnxscript.

With regards to the earlier question: for onnxscript code, I think we can do the checking ourselves (both statically and dynamically) and I think that will be better (than trying to use other systems), at least that is my first reaction. However, using the other checkers for our code, in general, may be useful. When we do that, we may run into some issues in the test-related code, where both code co-exist.

from onnxscript.

BTW, what is the specific issue with mypy? It is possible that Aaron's PR in-progress, which makes FLOAT etc. derived classes of TensorType, might fix it. (Though, generally, there are broader challenges in handling things like FLOAT["M", "N"] etc.)

from onnxscript.