radius_graph and knn_graph with zero size tensor leads to segmentation fault

[maabuu]

The following corner case leads to a segmentation fault.

import torch
from torch_cluster import radius_graph

x = torch.rand(0, 3)
radius_graph(x, r=5.0)

The same is true for knn_graph instead of radius_graph.

I am on linux and used conda to install python 3.10.8, pytorch 1.13.0 and pytorch_cluster 1.6.0.

[rusty1s]

Thanks for reporting. I fixed this via 3316bdc.