The reason I'm asking is that right above, in line 203, we set node->paths[nodeType] to NULL, so the subsequent free in line 209 should operate on a NULL pointer -- while many programmers like to check for that, in principle free(NULL) has a well defined semantics (it's harmless). Or am I missing something?
Anyway, this part of the code will have changed for the next NCCL release so your PR is not applicable (sorry!).
Thank you for your submission!
Your patch, while harmless, doesn't seem necessary to me -- is it based on an actual problem you encountered?
https://github.com/NVIDIA/nccl/blob/b9d748672478a1d2dd2e67b0c5e518b5e075dc98/src/graph/paths.cc#L202-L210
The reason I'm asking is that right above, in line 203, we set
node->paths[nodeType]toNULL, so the subsequentfreein line 209 should operate on aNULLpointer -- while many programmers like to check for that, in principlefree(NULL)has a well defined semantics (it's harmless). Or am I missing something?Anyway, this part of the code will have changed for the next NCCL release so your PR is not applicable (sorry!).