Segmentation fault occurs when I try to insert the topologyLayer to my own GNN module

[ljj-cyber]

I want to import your topology layer to my own neural network called GDN.
But when I try to use the operation of graph batch, the segmentation fault occurs.
Please let me know the input type of the TopologyLayer. So that I can adjust my code. Thanks a lot.
There is some detail of my code below:


This code works well. But if I use 'Batch' in pyg to encapsulate the graph data, it fails.

The segmentation fault occurs in the following position(#116).