The previous behavior of blowing up with a confusing error message with empty resources were attempted to be included in a dataset was not good. However, as mentioned in https://issues.xnat.org/browse/XNAT-6669, there seemed to me to be 2 realistic paths forward: improve the error message (and say this isn't supported) or think pretty deeply about how we should handle this with respect to warning the user, dataset validation, etc. The change for
XNAT-6669 allows the dataset to include empty resources without warning. However, that means you can have datasets with files like:
This is unlikely to be what the user intended (and hard to interpret), but there's no sort of warning. Even the dataset validation step includes green checkmarks for these scans. If you try to feed a dataset with some empty resources like this into clara, I could see it likely failing there. If that's true, we've moved the bug from the dataset creation step to the ML training step except now the root cause of the issue is further removed so it's harder to track down.