Add 'unstructured_grid' support to write function

[guidocioni]

Currently cfgrib can handle the gridType unstructured_grid when reading files (see #28). However, when writing a GRIB file using this data an error is raised NotImplementedError: Unsupported 'gridType': u'unstructured_grid'. Is it to difficult to implement this feature?

[alexamici]

@guidocioni can you please check out branch https://github.com/ecmwf/cfgrib/tree/write-unstructured_grid and test if it writes a GRIB file of you liking.

The change is trivial, if it works for you it'll go in the next release.

[guidocioni]

ValueErrorTraceback (most recent call last)
<ipython-input-8-d316a23fef91> in <module>()
      1 import cfgrib
----> 2 cfgrib.to_grib(ds_subset, 'out1.grib')

/work/mh0731/m300382/conda_envs/my_base/lib/python2.7/site-packages/cfgrib/xarray_to_grib.py in to_grib(*args, **kwargs)
    261 
    262 def to_grib(*args, **kwargs):
--> 263     return canonical_dataset_to_grib(*args, **kwargs)

/work/mh0731/m300382/conda_envs/my_base/lib/python2.7/site-packages/cfgrib/xarray_to_grib.py in canonical_dataset_to_grib(dataset, path, mode, no_warn, grib_keys, **kwargs)
    257     with open(path, mode=mode) as file:
    258         for data_var in dataset.data_vars.values():
--> 259             canonical_dataarray_to_grib(data_var, file, grib_keys=real_grib_keys, **kwargs)
    260 
    261 

/work/mh0731/m300382/conda_envs/my_base/lib/python2.7/site-packages/cfgrib/xarray_to_grib.py in canonical_dataarray_to_grib(data_var, file, grib_keys, default_grib_keys, **kwargs)
    209         raise ValueError("required grib_key 'gridType' not passed nor auto-detected")
    210 
--> 211     template_message = make_template_message(merged_grib_keys, **kwargs)
    212 
    213     coords_names, data_var = expand_dims(data_var)

/work/mh0731/m300382/conda_envs/my_base/lib/python2.7/site-packages/cfgrib/xarray_to_grib.py in make_template_message(merged_grib_keys, template_path, sample_name)
    179         if sample_name is None:
    180             sample_name = detect_sample_name(merged_grib_keys)
--> 181         template_message = cfgrib.CfMessage.from_sample_name(sample_name)
    182 
    183     for key in MESSAGE_DEFINITION_KEYS:

/work/mh0731/m300382/conda_envs/my_base/lib/python2.7/site-packages/cfgrib/messages.pyc in from_sample_name(cls, sample_name, **kwargs)
     62     @classmethod
     63     def from_sample_name(cls, sample_name, **kwargs):
---> 64         codes_id = eccodes.codes_new_from_samples(sample_name.encode('ASCII'))
     65         return cls(codes_id=codes_id, **kwargs)
     66 

/work/mh0731/m300382/conda_envs/my_base/lib/python2.7/site-packages/cfgrib/eccodes.pyc in codes_new_from_samples(samplename, product_kind)
    590     handle = lib.codes_grib_handle_new_from_samples(ffi.NULL, samplename)
    591     if handle == ffi.NULL:
--> 592         raise ValueError("sample not found: %r" % samplename)
    593     return handle
    594 

ValueError: sample not found: 'unstructured_grid_sfc_grib2'

[alexamici]

Ok, it was not so trivial after all.

Can I download a file with unstructured_grid grid anywhere? So I can test locally.

[guidocioni]

Sure, I'd choose something easy from the opendata of DWD, like a file here https://opendata.dwd.de/weather/nwp/icon/grib/00/qv_s/

[alexamici]

@guidocioni cfgrib now falls back to use a basic template when it can't detect a better one. The downside is that you need to specify all necessary GRIB keys. This should be enough to save to_grib a dataset opened with cfgrib.

Please report back if current master fixes the issue for you.

[guidocioni]

Sorry for the delay... I will try to test the behaviour in the following days.