Description
Note: I was going to post this in
conda/schemasdirectly but I can't tag the steering team from there, hence the issue here.
We have a few CEPs coming up (see #51 #8 #50) where it has been suggested that we document the proposed standards with an accompanying JSON schema in the conda/schemas repo.
As I start writing some of these schemas down, I will be using pydantic to write reusable models that can auto export JSON documents. I wonder if there's an interest in doing this retroactively for existing models that have not been standardized yet, so if this catches up, I'd like to propose the following repository structure:
examples/: Real-life examples of the documented file types, that can be used in test suites if necessary.schemas/: Publishable JSON schema documents.models/:Python scripts that generate JSON schemas intoschemas/. Most likely usingpydantic.scripts/: Other helper utilities (e.g. refresh all JSON schemas from their models, publish to a HTML tree, etc).tests/: Automations to make sure the proposed models actually implement the existing "standards" (as inexamples/), but also reject non-compliant files (e.g. extra keys due to typos).
Eventually I can foresee how the models directory becomes a collection of well structured modules with no extra dependencies (other than Pydantic) that would be best served as an importable Python package. Downstream packages (maybe including conda!) could benefit from using these models to transparently add typing their code paths, or at least compare their custom implementations to accepted community standards in their test suites.
I'd love to hear thoughts from @conda-incubator/steering! Thanks!