Add JSON schema

[franzpoeschel]

TODO:

• Define particlePatches
• Fill in things like "description", "type", "title", ...
• Maybe add support for abbreviated JSON representations Not now
• Dataset layout

I have added a CLI tool openpmd-convert_json_toml, exposing internally existing functionality to convert between TOML and JSON. I think it's better to define and maintain the schema in terms of TOML (it can be commented and is more legible), and just compile it to JSON for verification.

x-ref #1436

[ax3l]

Great work, thank you!

Wondering a bit if validation of an XML backend would be easier.... not sure.

[ax3l]

Are we able to patch this in check.py?

[franzpoeschel]

Not very easily. The JSON schema is on the file system and the single .json files refer to each other by their file names. Changing this would require (1) traversing the entire JSON schema and overriding the meshes path, the particles path and the references and (2) somehow setting up python-jsonschema to cross-reference in-memory schemas which I don't even know if it supports that, both at runtime of check.py.

[franzpoeschel]

Wondering a bit if validation of an XML backend would be easier.... not sure.

The major issue of the JSON schema approach is that it is not a predictive parser (https://en.wikipedia.org/wiki/Recursive_descent_parser), meaning that it deals with {"anyOf": [<option1>, <option2>, ...]}-like statements very naively. If there is an error somewhere deep down the hierarchy, the JSON schema just sees that all options have failed and cannot give any more details where the failure was.

If we want to seriously consider XML as an alternative, my opinion is that it must at least solve this particular issue .. which I doubt it can properly, since this involves a bit of parser theory and grammar transformation.

This job is better done by handwritten parsers, which the openPMD-api and the openPMD-validator are.

However, if we add an XML backend at some point, an XML schema is definitely at least worth a consideration.