Ryan May

I've seen errors with both C's `ncdump` and netCDF-java's Tools-UI when trying to get the `rhprs` array. Given that some requests work for you and some don't, it seems like...

@epifanio I get errors with those variables when I use netCDF-java (through ToolsUI) as well. Something is at fault in the server configuration used to aggregate the individual netcdf files...

If the data are malformed, that's bad--but it doesn't explain to me why writing to disk and reading it in that way then works.

Another option idea is to have `.first` and `.last` attributes on a dict-like object that holds the datasets.

@DennisHeimbigner @WardF Getting netCDF-C 4.9.0 for conda will require solving the build issues encountered on the update PR: https://github.com/conda-forge/libnetcdf-feedstock/pull/140 We'd love any help you all can give on that.

Siphon's not really playing a role here in terms of needing an explicit close. What's going on is that you're keeping references to every xarray dataset you ever open with...

I don't think you can. netCDF4-python is just wrapping the netCDF-C library. This requires either a filename or a buffer containing the contents of the file in memory.

cc @DennisHeimbigner

@DennisHeimbigner If you want any more datasets, you can go to the [AWS Public Dataset Registry](https://registry.opendata.aws) and search for "zarr". That gave me 6 datasets--might be good to at least...

netCDF4 cannot be *leading* the C library. Also, creating non-portable files isn't just a problem for an individual user, it can lead to support problems for the software projects. I'm...