There is a lot of complexity around HDF5 flavors used with Pandas. In particular "fixed" format hdf5 files are somewhat opaque and can not be cut easily. Dask.dataframe currently chokes on these.

We should do one of the following:

1. Raise informative error messages
2. Accept them as single-partition dask.dataframes
3. Accept them in glob strings as multi-partition dask.dataframes

cc pvanderlinden