test_numba_integration passes tuple into cuda instead of array

[code-review-doctor]

🐛 Describe the bug

There is an (as far as I can tell) unwanted comma on line 314:

pytorch/test/test_numba_integration.py

Lines 313 to 318 in 1b66915

	for dtype in dtypes:
	numpy_ary = numpy.arange(6).reshape(2, 3).astype(dtype),
	numba_ary = numba.cuda.to_device(numpy_ary)
	self.assertTrue(numba_ary.is_c_contiguous())
	torch_ary = torch.as_tensor(numba_ary, device="cuda")
	self.assertTrue(torch_ary.is_contiguous())

This makes the numpy_ary a tuple containing a numpy array rather than a numpy array ie.:

In [3]: numpy.arange(6).reshape(2, 3).astype(numpy.float64)  # no comma at end
Out[3]: 
array([[0., 1., 2.],
       [3., 4., 5.]])

In [4]: numpy.arange(6).reshape(2, 3).astype(numpy.float64),  # comma at end
Out[4]: 
(array([[0., 1., 2.],
        [3., 4., 5.]]),)

This is a problem for the test because the tuple is then passed into numba.cuda.to_device(numpy_ary) and assertions are done on the output. It is possible the test is affected by the fact a tuple is passed in instead of a numpy array. Indeed, if in "live" a numpy array is passed in but in tests a tuple containing a numpy array is passed in then this reduces the confidence the tests give us as it's not "like live".

Versions

NA

I can make a PR to fix this :)

cc @mruberry

[mruberry]

Hi @code-review-doctor, I agree that comma seems out of place and it's odd that the test passes. We would accept a PR removing it (if that really is the intention of the test).