Approach to implementing additional schema restrictions

[dvstans]

In my application, we are intending to allow end-users to author their own schemas, and those schemas will be used to validate user-authored metadata. The validated metadata will then be stored in a database (document store) where it can be queried with user-author metadata expressions. Due to how the DB / search engine works, I need to restrict some aspects of the schema/metadata:

1. Schema property field names must conform to a restricted character set (compared to what JSON allows), and
2. Fields with multiple types (i.e. ["string","number"] cannot be supported

Before I fork and dive into your code, do you feel this would be relatively straight-forward change? I would consider a pre-validation check, but I think this would be quite complex to implement correctly (i.e. I'm not deeply familiar with the JSON schema validation specification). Any advice would be appreciated!

[pboettch]

Hmm, from what I understand this could be done with restricted schema-validation-schema . This you use on user-schemas and if they are valid they will be used to validate user's data. https://github.com/pboettch/json-schema-validator/blob/master/schema

No need to change the code.

[dvstans]

Interesting. Are you saying I run the validate function on the user schema using the referenced (or modified) master schema? Or is it a different function to validate schemas? Also, I don't see anything the the JSON schema spec that restricts keys themselves (i.e. a regex pattern that defines what the property key, itself, can be - not just the value). I may have missed this as it's a large spec!

Thanks!

[pboettch]

Yes, a schema is nothing more that a JSON instance which can be validated with a schema (see the linked root-schema). I see that you already understood that. The same validate-function can be used, but the validator-instance has another schema. Btw. this is the reason the root-schema is committed to the repository: I use it from time to time to validate a schema I wrote using the app.

There is no limit to the key-value as long as it is a string. JSON-object is "key any string with": <json-value>. So if you want to restrict key name, nothing easier than that: Have a look at the difference between properties and patternProperties in the root-schema. You'll see that patternProperties has a propertyNames-field. This defines a (sub)-schema describing what names are allowed for property-names.

So, if you want to restrict key-names for your user's schemas. Just take the root-schema and add a propertyNames-field contains a list of allowed fields, a regex (via "type":"string", "pattern": "<regex>").

Hope this helps.