torch.Tensor.ndim is not typed

[KKawamura1]

🐛 Bug

import torch

a = torch.Tensor().ndim  # mypy says: `error: "Tensor" has no attribute "ndim"`

Environment

torch 1.6.0 and master (ecb9e79)

Why this occurs and how to solve

#38157 removed torch/__init__.pyi.in and introduced torch/_C/__init__.pyi.in, but some methods defined in torch/__init__.pyi.in (ndim, T, and grad_fn) was not salvaged and had gone away.
T and grad_fn was rescued in #40658. Only the ndim is still not typed.

The solution wolud be just adding ndim attribute to _TensorBase in torch/_C/__init__.pyi.in.

cc @ezyang @malfet @rgommers

[zhangguanheng66]

Can you share a code snippet? I can not reproduce the error when running:

import torch
a = torch.Tensor().ndim

[rgommers]

I can reproduce it, bug report seems correct. On current master:

$ cat tmp.py
import torch
reveal_type(torch.Tensor.ndim)
reveal_type(torch._C._TensorBase.ndim)

$ mypy tmp.py 
tmp.py:3: error: "Type[Tensor]" has no attribute "ndim"  [attr-defined]
tmp.py:3: note: Revealed type is 'Any'
tmp.py:4: error: "Type[_TensorBase]" has no attribute "ndim"  [attr-defined]
tmp.py:4: note: Revealed type is 'Any'