I'm putting this one in separately as it's currently breaking all of ch_utils tests.
There were two main issues in here:
The newly refactored method was trying to access the .data attribute, but that's not available when copying an h5py.Dataset. Switching to [:] should be equivalent, albeit a little slower for the MemDataset's.
h5py returns attributes as bytestrings as so the lookup for time axes was failing as we pass them in as unicode strings. This explicitly converts the strings from the dataset.
I'm putting this one in separately as it's currently breaking all of ch_utils tests.
There were two main issues in here:
.dataattribute, but that's not available when copying anh5py.Dataset. Switching to[:]should be equivalent, albeit a little slower for theMemDataset's.h5pyreturns attributes as bytestrings as so the lookup for time axes was failing as we pass them in as unicode strings. This explicitly converts the strings from the dataset.