Single valued flag_values causes an Exception

[pp-mo]

Version of compliance checker running:
5.0.2.dev90+gcad1a7c
https://github.com/ioos/compliance-checker/tree/cad1a7c9900459fd33a403e58d79868a54d907b7

Describe the checker this affects:
CF

Attach a minimal CDL or NetCDF file which is able to reproduce the issue

netcdf singleflag {
variables:
  int x ;
    x:flag_values = -1 ;
    x:flag_meanings = "any" ;

}

To Reproduce:

$ ncgen -k3 singleflag.cdl.txt -o singleflag.nc
$ compliance-checker -vvv -t=cf singleflag.nc
Running Compliance Checker on the datasets from: ['singleflag.nc']
   . . .
* §2.6.2 global attribute history should exist and be a non-empty string
* §2.6.1 Conventions field is not present
WARNING: The following exceptions occurred during the cf checker (possibly indicate compliance checker issues):
cf.check_flags: 'numpy.int32' object is not iterable
  File "/net/home/h05/itpp/git/ioos_compliance-checker/compliance_checker/cf/cf_1_6.py", line 1072, in check_flags
    valid_values = self._check_flag_values(ds, name)
  File "/net/home/h05/itpp/git/ioos_compliance-checker/compliance_checker/cf/cf_1_6.py", line 1125, in _check_flag_values
    flag_set = set(flag_values)

$ 

Describe the issue below:
The error occurs trying to iterate over 'flag_values', but the code requires also the presence of a 'flag_meanings' to hit the error.

The CF validity of the example is, I think, questionable : Is a single flag value/meaning acceptable or not?
(update : on this, see #914 (comment) and #914 (comment) )
Having said that, I did have a real file from another person which contained this.
Arguably, the checker code needs to be more robust.

[pp-mo]

Possibly related :

• CF flag_values check #792
• Wrong error with flag_masks containing only one element #914
  • fixed in Handle scalar flag attributes #915 : we probably want something simliar (though I think this one ins simpler ?)

[pp-mo]

I can probably propose a fix, but someone else may be better placed.
Would anyone else agree that this is worth fixing ?

[benjwadams]

Yes, this is worth fixing. NetCDF's data model treats attributes as an array of a certain type. NetCDF4-Python will take a single element attribute and cast it to scalar.

[pp-mo]

#924 fixes