Working with index files and xr.open_mfdataset in a location without write permission

[matzech]

I work with multiple grib files in a location without write permission. In locations with writing permission, I usually work with open_mfdataset from xarray like

ds = xr.open_mfdataset("201506/*.grib", engine="cfgrib")

As I would like to work with index files, I tried to follow the approach proposed in #126 , but the short_hash is actually the same for the different files when applied like:

ds = xr.open_mfdataset("201506/*.grib", engine="cfgrib", backend_kwargs={"indexpath":'mygrib.{short_hash}.idx'}
Is there another solution?

[alexamici]

@matzech you may pass {"indexpath": ""} to disable writing the index, or you may try something along the line of (untested):

ds = xr.open_mfdataset("*.grib", engine="cfgrib", backend_kwargs={"indexpath": 'writable_folder{path}.{short_hash}.idx'})

[onnyyonn]

Is it possible to modify the {path} so that it contains only the basename? Also, what keys are available other than {path} and {short_hash}?

[FRidh]

As suggested in pydata/xarray#6512 I think we should use a temporary file/folder as default or disable it by default. It cannot be relied on that the folder the file is in is writeable nor that the index file is accurate. The index file is purely an optimization. In my opinion robustness is more important than performance.