Inconsistencies in processes (based on pydantic tests)

[m-mohr]

We created Pydantic Classes from the OpenEO API JSON process graph specification and tested them against the standard processes (from https://raw.githubusercontent.com/Open-EO/openeo-processes/1.x.0/)
We found that ['array_apply', 'array_filter'] use
schema": [ {"type": "number"}, . . . {"type": "null"}
instead of
"schema": { "type": ["number", "string", "null"]}

So,

1. Is there any interest in discussing this matter?
2. What would be the consequences of updating the standard processes accordingly?
3. Maybe the tests and models are of interest?

Running the tests against other lists of standard processes reveals other inconsistencies. Not sure how to proceed here in a way that benefits all. PS. the reason we used Pydantic is that FastAPI gives us automatic checking of the API parameters)

Originally posted by @danielFlemstrom in #204 (comment)

[m-mohr]

Hey @danielFlemstrom,

I'm a bit confused and I think I need a bit more context here.

• It seems the terminology is a bit off here. Did you create pydantic classes for the openEO process schema? Because I'm not exactly sure how you could validate processes against a process graph schema.
• Yes, processes use different ways to make schemas available (schema vs. array of schemas). It depends on the use case and the data types that will be made available. Especially, we don't try to conflate different basic data types (except for null) so that data types can be described clearly and unambiguous. Both options shown above are valid.
• Regarding your three points:
  1. Yes.
  2. I'm not sure exactly what you want to align and why.
  3. Not sure yet, I'm not using Python a lot right now. It seems they would need some fine-tuning though.
• We try to minimize inconsistencies, but it might be that not all things that look like inconsistencies are really inconsistencies. I'm happy to discuss though. I'd hope to spot things that we can improve. How can I access the results of these tests?

[danielFlemstrom]

Hi @m-mohr! Thank you for your quick response. Sorry for the confusion. I am relatively new to all this. As you correctly say, the input is not a python process directly but, one of the Pydantic classes "down the road" is. As a result, a process graph will not be accepted if it includes "non-conformant" processes. Maybe the most efficient way to access the results of the tests is that I walk you through it in a T-EAMS or Zoom session? Of course, I can zip the models and the test case and mail it to you as well.

[m-mohr]

Happy to join a Teams of Zoom session. That seems indeed like the most efficient way. I'm pretty flexible with meeting slots this week (e.g. today until 4), you could just send an invite. My e-mail address is in my GitHub profile. Thank you!

[m-mohr]

Clarified the issues on a video call. Thank you again for your contribution.