I addressed all comments. Let me know if you want to see any other change.

I messed up when addressing the comments so I had to squash a bunch of things and forced pushed. Sorry for the mess.

No new tests or changes to existing tests are needed?

I can add a bunch of type checking related tests but I have no idea where they are located.

See the ArrayTerator typing tests for a few examples; I'd suggest creating a new file specifically for nditer.

If you're willing to add tests, then the most import thing would be to check that the various legal statements
return the expected type (directly calling the methods is less interesting to check here). For example:

import numpy as np

nditer_obj: np.nditer

with nditer_obj as context:
    reveal_type(context)   # E: numpy.nditer

Thanks. For ``get/set/delitem`, are you fine with simply implicitly invoking the protocol but not using any reveal since everything is Any for the time being ?

Thanks. For get/set/delitem, are you fine with simply implicitly invoking the protocol but not using any reveal since everything is Any for the time being ?

I'd still use a reveal_type here. The actual revealed value is less interesting here (for now),
but it does still allows for some basic verification (and checking for errors).

I pushed some first tests for nditer however I am having a hard time running tests locally. Running numpy.typing.test("slow", verbose=3), test_reveal are deselected and I could not really figure why. I would appreciate any pointer you may have.

Failures look unrelated to this work, I believe. Let me know if I am wrong.

I pushed some first tests for nditer however I am having a hard time running tests locally. Running numpy.typing.test("slow", verbose=3), test_reveal are deselected and I could not really figure why. I would appreciate any pointer you may have.

I usually run the python ./runtests.py -t numpy/typing --mode=full command from the command line.
I think running numpy.typing.test("full") should also work.

In any case, the added tests look good to me.

Thanks @MatthieuDartiailh