v4 JSON schema requires non-empty string arrays

[StephenCleary]

The current spec allows empty required and empty arrays in dependencies.

However, the stringArray type defined in the Draft 04 schema does not:

"stringArray": {
    "type": "array",
    "items": { "type": "string" },
    "minItems": 1,
    "uniqueItems": true
}

Since these two places are the only locations in the schema that use stringArray, this can be fixed by removing the minItems requirement:

"stringArray": {
    "type": "array",
    "items": { "type": "string" },
    "uniqueItems": true
}

[StephenCleary]

To be clear, we are generating JSON schemas dynamically (based on user input) in our product, and using the official Draft 04 schema to validate these schemas. So this is a "real" issue for us.

We've updated our local copies of the Draft 04 schemas to remove minItems, but it would be better if the official copies were fixed so we don't fork.

[chaffin]

Also, given that stringArray MUST be unique, would also removing uniqueItems make any difference? If so, this would this create a validation dependency on a validation keyword, rather than the data the schema is to validate.

[handrews]

@StephenCleary the draft-04 meta-schema matches the "draft-04" spec (really draft-fge-json-schema-validation-00 in this case). Draft-04 does not allow for empty arrays for these keywords, so the meta-schema is correct as-is.

If you want to use empty arrays, you can move to draft-06 which supports empty arrays in its meta-schema.

[StephenCleary]

I was under the impression that draft-06 was not yet finalized/released yet. It's good to use?

[handrews]

Yes, it's been out for a few weeks and is good to use. I'm trying to rework the "software" section of the web site to highlight support for draft-06. So far I think only ajv (and therefore also JSON Schema Lint) support it, but I know several other validators are working on it now that it's out.

The validation changes aren't huge, so hopefully I'll be able to highlight more draft-06 compliant implementations soon (there is a test suite for draft-06 available to help implementors).

Hyper-schema changes are very substantial, but hyper-schema was never as broadly implemented (which is one reason we felt free to change it more drastically).

[handrews]

Thanks for filing #317, can this issue be closed with that one tracking the needed fix?

[StephenCleary]

Absolutely, thanks!

Part of my confusion doubtless stemmed from us starting to use JSON schemas about 6 weeks ago, when draft-04 was current.

[handrews]

Nice timing- just in time for the first significant, behavior-changing update to the spec in more than four years!

Here are some guides to the changes:
https://github.com/json-schema-org/json-schema-spec/wiki/FAQ:-draft-wright-json-schema%5B-validation%5D-01
https://github.com/json-schema-org/json-schema-spec/wiki/FAQ:-draft-wright-json-schema-hyperschema-01