Error in object detection schema

[rohanakut]

In the dimensions field

"prefixItems": [
                        { "type": "number", "minimum": 0, "maximum": 1 },
                        { "type": "number", "minimum": 0, "maximum": 1 },
                        { "type": "number", "minimum": 0, "maximum": 1 },
                        { "type": "number", "minimum": 0, "maximum": 1 }
                    ],
                    "items": false,
                    "minItems": 4

needs to be changed to

"items": [
                        { "type": "number", "minimum": 0, "maximum": 1 },
                        { "type": "number", "minimum": 0, "maximum": 1 },
                        { "type": "number", "minimum": 0, "maximum": 1 },
                        { "type": "number", "minimum": 0, "maximum": 1 }
                    ],
                    "minItems": 4

@JRegimbal tagging you to double-check the modification

[JRegimbal]

Ah the prefixItems is a feature of the new draft for this (2020-12 IIRC), so it's actually valid but not backwards compatible.

[rohanakut]

should i create a new PR to make the changes?

[JRegimbal]

What's the actual issue that's happening?

[rohanakut]

when i try to validate it throws the following error:

False schema does not allow 0.026533996683250415

Failed validating None in schema['properties']['objects']['items']['properties']['dimensions']:
    False

[JRegimbal]

How are you trying to validate it?

[rohanakut]

I am validating using a draft7validator

This is my exact code:

try:
        validator = jsonschema.Draft7Validator(data_schema)
        validator.validate(things)
except jsonschema.exceptions.ValidationError as e:
          logging.error(e)
          return jsonify("Invalid Preprocessor JSON format"), 500

And the data schema is called in the following way:

with open('./schemas/preprocessors/object-detection.schema.json') as jsonfile:
            data_schema = json.load(jsonfile)

[JRegimbal]

Alright that explains it. This isn't a draft 7 schema and so it isn't understood by validator. A quick check suggests this is a much bigger issue as jsonschema doesn't support draft 2020-12 (although python-jsonschema/jsonschema#817 would add support). I'd hate to say it but it might be necessary to do rewriting of a lot of schema files if support isn't added soon.

[rohanakut]

The fix i suggested above works well as my schema gets validated with that fix. So for now should I start with the object detection schema and then move on to other schemas?

[JRegimbal]

A lot of other changes would be necessary, including to the containers that actually are validating it as a draft 2020-12 schema and would fail with this change.

[rohanakut]

As far as i know, only the object detection schema throws this error. this is because all the other models have been validated using draft7validator and none of them threw this error. So I don't think that any other schemas need to be changed

[JRegimbal]

The actual request schema itself uses the prefixItems key as well and all of them are draft 2020-12. Since I don't know how well validators handle multi-file schemas based on different drafts, it might be best to just drop everything to draft 7. This will fix your problem as well.

[rohanakut]

So as a solution should we go ahead with jschon or should we change the schemas?

[JRegimbal]

Changing the schemas now. I like jschon and I think it'll be an option later, but it is still new and I get the sense we might have to deal with bugs. I already have the schemas updated (downdated?) in the draft7 branch and am testing changes to my many containers that expected a 2020-12 schema.

[JRegimbal]

Fixed. Main now validates against draft7 and the schema branch contains all schemas in a draft 7 format.