2019-09 (aka Draft 08) support

[rb-john]

Is 2019-09 support planned?

support for format (such as duration) should be minimal, and the use of $defs over definitions seems not to be strictly enforced.

Thanks

[horejsek]

I missed this update. :-) Support is not planned yet, but yeah, it will be added one day. If you want to take care of it, go ahead.

[handrews]

Draft 2020-12 is now out :-) It is also being used in OpenAPI 3.1 (as is! No restrictions or modifications like the old nullable anymore), so I'd very much recommend targeting 2020-12 and only doing 2019-09 later if at all. In addition to fixing some awkward bits of 2019-09, the OpenAPI 3.1 integration means the user base for 2020-12 will be much larger.

[coderextreme]

My boss said, oh, let's update to 2020-12, and I hesitantly said we kind of have it working for draft07. So now we're kind of stuck with jsonschema instead of fastjsonschema. We normally validate 4000 example JSON files. We really need the fastest possible correct schema validator in python.

I don't really have an axe to grid, but Ajv is much faster than jsonschema. I would love to have fastjsonschema at least competitive with Ajv as far as version support.

[vEpiphyte]

Draft 2019-09 / 2020-12 support would be fantastic. Right now, writing oneOf schemas is rather difficult and I've found it practically requires some post-processing if you are allowing additionalProperties to be true.

A schema like this won't allow a object to only have the sku and only one of the evaluated options, since the oneOf properties are not included when considering the additionalProperties check ( this is a generic jsonschema problem, not a fastjsonschema problem).

{
    "$schema": "http://json-schema.org/draft-07/schema#",
    "type": "object",
    "required": ["sku"],
    "additionalProperties": false,
    "properties": {
        "sku": {"type": "string"}
    },
    "oneOf": [
        {
            "required": ["count"],
            "properties": {
                "count": {"type": "integer"}
            },
            "not": {
                "required": ["bundles"]
            }
        },
        {
            "required": ["bundles"],
            "properties": {
                "bundles": {"type": "integer"}
            },
            "not": {
                "required": ["count"]
            }
        }
    ]
}

In the updates we can utilize the unevaluatedProperties option to clean this up considerably and still only allow the evaluated properties through:

{
    "$schema": "https://json-schema.org/draft/2020-12/schema",
    "unevaluatedProperties": false,
    "type": "object",
    "required": ["sku"],
    "properties": {
        "sku": {"type": "string"},
    },
    "oneOf": [
        {
            "required": ["count"],
            "properties": {
                "count": {"type": "integer"}
            },
        },
        {
            "required": ["bundles"],
            "properties": {
                "bundles": {"type": "integer"}
            },
        }
    ]
}

( credit for this example goes to https://medium.com/@smikulcik/oneof-explained-c0264c429735 )

[vEpiphyte]

@horejsek I've come back across the desire to add 2019-09 / 2020-12 support to the library. I tasked claude code with adding that and have an example PR with that work here vEpiphyte#1 . I would be happy to repoint it to your source if the contribution is welcome!

That does allow for the simpler data-dependency processing noted in my above example.

import fastjsonschema

schema = {
    "$schema": "https://json-schema.org/draft/2020-12/schema",
    "unevaluatedProperties": False,
    "type": "object",
    "required": ["sku"],
    "properties": {
        "sku": {"type": "string"},
    },
    "oneOf": [
        {
            "required": ["count"],
            "properties": {
                "count": {"type": "integer"}
            },
        },
        {
            "required": ["bundles"],
            "properties": {
                "bundles": {"type": "integer"}
            },
        }
    ]
}

def main():
    func = fastjsonschema.compile(schema)

    tests = (
        ({'sku': '1234',}, True),
        ({'sku': '1234', 'count': 1}, False),
        ({'sku': '1234', 'bundles': 2}, False),
        ({'sku': '1234', 'bundles': 2, 'undefined': True}, False),
        ({'sku': '1234', 'count': 3, 'bundles': 2}, True),
    )
    for (data, willerr) in tests:
        try:
            result = func(data)
        except fastjsonschema.exceptions.JsonSchemaException as e:
            if willerr:
                print(f'{data=} errored as expected')
            else:
                raise
        else:
            print(f'{result=}')

if __name__ == '__main__':
    main()

can run and produce the following output:

> python foo.py 
data={'sku': '1234'} errored as expected
result={'sku': '1234', 'count': 1}
result={'sku': '1234', 'bundles': 2}
result={'sku': '1234', 'bundles': 2, 'undefined': True}
data={'sku': '1234', 'count': 3, 'bundles': 2} errored as expected

[horejsek]

Thank you, @vEpiphyte . Would you mind splitting the MR into smaller pieces for easier review and validation?

[vEpiphyte]

@horejsek Would you want them as sequential changes ( adding 2019-09 support, then 2020-12 support ) ? From my memory the refs and annotation tracking were linked.

[horejsek]

I do not mind as much as per JSON schema version as feature-wise. Would that make sense?