After a few primitive attempts at developing some basic JSON
Schema validation systems for Committee, I
finally mustered a bit of initiative and developed a fully spec-compliant
schema and hyper-schema validator which I’ve officially released as
During a validation,
json_schema will build JSON Pointers to any data that
failed, and to the exact schema rule that failed it. This is designed to help
with the validation debugging process:
schema.json#/properties: failed schema #/properties/properties: Expected data to be of type "object"; value was: . schema.json#: failed schema #: Data did not match all subschemas of "allOf" condition.
I tried to pay special attention to the more difficult features that weren’t
well covered by my previous implementations like self-referencing schemas,
tuple validation in
dependencies, along with a multitude of
json_schema should also provide a nice way to introspect
and traverse schemas programmatically. Committee’s already been reworked to use
it, and I hope to soon start moving over other Interagent
A final thought: there is no better way to understand a spec than to write an implementation for it. I thought that I had an okay understanding the V4 JSON Schema spec before starting the project, but then as you might expect, I ran into every mental grey area, gap in understanding, obscure feature, and special case. I now realize just how tenuous my understanding actually was.
Check it out on GitHub.
Did I make a mistake? Please consider sending a pull request.