Printing error message for huge ChainerX array blocks REPL

When the type of function’s argument is incompatible, chainerx tries to show TypeError with the details of the invoked array X in the Invoked with: section. If the X is huge, ChainerX still shows all data contained in X. It takes much time and blocks REPL.

>>> import chainerx as chx
>>> X = chx.arange(10000000)  # huge data
>>> x = X.take([0], axis=1)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: take(): incompatible function arguments. The following argument types are supported:
    1. (self: chainerx.ndarray, indices: chainerx.ndarray, axis: Optional[int]=None) -> chainerx.ndarray

Invoked with: array([[0., 0.],
       [0., 0.]], shape=(2, 2), dtype=float64, device='cuda:0'), [0]; kwargs: axis=1

[beam2d]

Thanks for the catch! The message is auto-generated by pybind11, so there are two options.

1. Switch to a custom type error message in which the whole array repr is not printed.
2. Fix __repr__ to omit middle items when the array is too large.

I’m not sure if we can accomplish the option 1 (I think we should anyway have a way to customize it, though…). The option 2 is anyway needed for other places. We can follow the format used in NumPy, whose __repr__ also omits middle items of large arrays.

[okapies]

It is still a problem in invoking such as .reshape(-1).