Would "dachar" work better as a stand-alone library if fixes were language agnostic?

[agstephens]

• dachar version: *
• Python version: 3.7
• Operating System: Linux

Description

We have created fixes that are specific to our daops library, e.g.:

{
    "post_processors": [
        {
            "args": null,
            "func": "daops.data_utils.coord_utils.squeeze_dims",
            "kwargs": {
                "dims": ["lev"]
            }
        }
    ]
}

An alternative would be to define all fixes in a language-agnostic form, such as:

{ 
  "fixes": [
     {
         "operation": "squeeze singleton dimensions",
         "operands": {
               "dimensions": ["lev"]
      }
  ]
} 

The reason for taking this approach would be that the dachar library would exist as a stand-alone component that was entirely independent of any implementation of fixes.

Benefits of this approach would be:

1. Any external tools using the information generated by dachar could use the fix definitions.
2. dachar does not hard-code specific implementations.

Drawbacks of this approach would be:

1. An extra language/encoding of the fixes is required - this must be written by the project. (There might be existing standards/approaches we could use).
2. Our daops library will need to define mappings from the language-agnostic form to our own daops functions.

@cehbrecht and @huard: what are your thoughts on this issue?

[huard]

Agree with the benefits.
For the drawbacks:

1. Could you consider an extension to NcML that could become NcML-2.3 or NcML-3.0 ?
2. daops.mappings would essentially be xarray.open_ncml, and I think you'd get community support.
   read ncml files to create multifile datasets pydata/xarray#2697
   keep track of xarray ncml progress pangeo-data/esgf2xarray#2

[agstephens]

@huard, thanks for your response. We will have to look into it. NcML is certainly an option - rather than making up our own bespoke approach.

[huard]

Sorry to harp on this, I'll be happy to be proved wrong on this.

[agstephens]

@huard, I might be happy for you to be proved right :-)