Tool that detects fill value issues in a netCDF file

[corviday]

One might give files a once-over with ncdump or ncview before adding them to the database, and make sure that variables have a fill value assigned. However, a fill value attribute visible in ncdump or proper display in ncview are not sufficient to guarantee proper file indexing. An additional tool would be helpful.

Situations that were seriously annoying to resolve, and would be nice to pre-detect:

• fill value attribute for a packed variable was not packed and has the wrong data type
• fill value was not declared when the variable was created, but later a fill value attribute was added with ncattd or something. Most tools will handle this retroactive adjustment just fine, but not index_netcdf.

Another option might be to have index_netcdf warn you if the detected minimum or maximum is the same as the fill value, or to modify var_range in nchelpers somehow.

[rod-glover]

Questions:

1. Is the root of the problem really in the files themselves then? As in, this is not really a legitimate state of affairs? It looks so. In that case, having a detection tool might help. Better would be a detect-and-fix tool (with some human intervention likely needed).
2. Does nchelpers.var_range give wrong answers when the fill values are wonky as you describe? If so, it might make sense to fix it. Depends I suppose on whether we are willing and able to correct problematic files or not.

[corviday]

When I wrote this issue, I was treating nchelpers.var_range() as the definition of what a correct fill value is - "wrong fill value" means "nchelpers doesn't understand it as intended" even if other netCDF tools do.

This is somewhat complicated by the discovery that var_range() is currently buggy due to an upgrade to the netCDF4 python library.