Encoding sum types with json-schema

1_encoding_sum_types_in_json-schema.md

The goal is to express an ADT like the following in json schema:

data Sum 
  = Foo { hi :: Int }
  | Bar { x :: Bool }

or in rust syntax:

enum Sum {
  Foo { hi : i64 },
  Bar { x : bool }
}

This can be done by encoding each of these alternatives in a “tagged” way in json:

{
  "tag": "Foo",
  "hi": 234
}

or

{ 
  "tag": "Bar",
  "x": true
}

Of course this means the field tag cannot be used for any of the field names, but it’s a good encoding for most things.

Now, how to express this in json-schema? oneOf to the rescue.

2_sum-schema.json

{
	"$schema": "https://json-schema.org/draft/2019-09/schema",
    "oneOf": [
      { "type": "object",
        "required": [ "tag", "hi" ],
        "properties": {
          "tag": {
            "const": "foo"
          },
          "hi": {
            "type": "integer"
        }
        }
       
      },
      { "type": "object",
        "required": [ "tag", "x" ],
        "properties": {
          "tag": {
            "const": "bar"
          },
          "x": {
            "type": "boolean"
          }
        }
       
      } 
  
    ]
  
}

3.md

This will use the const type to differentiate the tag field for the record options. It’s slightly verbose, but should work with any standard-compliant parsers. Don’t forget to add "tag" to required in every branch, otherwise you get weird matches/a weird machine.

If your parser doesn’t yet support the draft 6 of json-schema, you can use "enum": [ "tag" ] instead of "const": "tag".