Extending validators with custom callables not associated with specific properties

[tate-song-ai]

Hi,

It appears that to extend a validator class, one must pass in a mapping from schema property names to callables with custom validation logic. What if the callable I want to inject isn't associated with a singular schema property and must always run regardless of what schema properties exist?

[agammann]

The mapping passed to validators.extend() is from schema keywords to callables, rather than from individual instance property names. A custom keyword can check several instance properties. However, its callable is only invoked when that keyword is present in the schema being evaluated.

If the check must run once for every instance even when the schema is {}, put it in a validation wrapper:

from jsonschema import Draft202012Validator, ValidationError

def business_errors(instance):
    # Replace this example with your check across the instance.
    if instance == 'blocked':
        yield ValidationError('Custom check failed')

def iter_errors(instance, schema):
    yield from business_errors(instance)
    yield from Draft202012Validator(schema).iter_errors(instance)

errors = list(iter_errors('blocked', {}))

This runs your check at the root once per call and also collects normal schema errors. It does not automatically run the business check again for every nested subschema; add traversal explicitly if that is what you need.

I tested jsonschema 4.26.0: an extension keyword was not called for {}, but the wrapper above ran with both an empty schema and an ordinary typed schema. With 'blocked' and {'type': 'integer'}, it collected both the custom error and the built-in type error.

If you can require a keyword in each schema, validators.extend(Draft202012Validator, {'x-check': your_callable}) is the other option. The extension API describes that keyword-to-callable mapping.