`test_ind2ptr` fails after `torch.Tensor.tolist` was updated to error out when called on tensors with null storage

[anjali411]

Looks like

pytorch_sparse/test/test_storage.py

Lines 23 to 24 in eabdc58

	row = torch.ops.torch_sparse.ptr2ind(rowptr, 0)
	assert row.tolist() == []

started failing after pytorch/pytorch#75990 landed

        row = torch.ops.torch_sparse.ptr2ind(rowptr, 0)
>       assert row.tolist() == []
E       RuntimeError: tolist() shouldn't be called on a tensor with unallocated storage

[rusty1s]

Thanks for sharing. Do you have any pointers on how to fix this on our end?

[anjali411]

@rusty1s Hey so unfortunately torch.empty(0) in python is not the same as torch::empty(0) in C++. I think changing

pytorch_sparse/csrc/cpu/convert_cpu.cpp

Line 39 in 7a6c9ab

auto out = torch::empty(E, ptr.options());

to auto out = torch::empty({E}, ptr.options()); should fix it

Btw this test code was segfaulting before so it's not a bc breaking change

[rusty1s]

Thank you for digging into this. Can you send a PR?

[anjali411]

I don't think I will get to it until next week. It would be nice if this is fixed asap so that the PR doesn't get reverted

[rusty1s]

Fixed this via #222. Thank you!