How to compile multiple schemas into a single AST?

[x0k]

I am currently using the following approach, but it has several typing issues:

for (const schema of schemas) {
  registerSchema(
    Object.assign(
      { $schema: "http://json-schema.org/draft-07/schema" },
      schema as SchemaObject,
    ),
  );
}
// AST declares `plugins` as an array, but the property is used as a `Set`
// https://github.com/hyperjump-io/json-schema/blob/50dd7df599011bcd1df3283b2073507e9fc02bf9/lib/keywords/validation.js#L39
const ast = { metaData: {}, plugins: new Set() } as unknown as AST;
for (const schema of schemas) {
  const s = await getSchema(schema.$id!);
  // `Validation.compile` does not use the third argument
  // https://github.com/hyperjump-io/json-schema/blob/50dd7df599011bcd1df3283b2073507e9fc02bf9/lib/keywords/validation.js#L10
  await Validation.compile(s, ast, s);
}

[jdesrosiers]

// AST declares plugins as an array, but the property is used as a Set

Sorry about that. We've discovered the error in #115 and it's going to be fixed soon.

// Validation.compile does not use the third argument

It doesn't happen to be used in Validation, but it's used in other keyword handlers.

I am currently using the following approach

That approach should work. However, keep in mind that any referenced schemas will automatically be included in the AST. You should just need to compile the main entry schema. If you your goal is to have on AST that includes multiple unrelated schemas, this approach should work. But, generally you want a separate AST for each entry schema. I think the only reason you might want to combine them is if you have a lot of helper schemas that multiple entry schemas use. In that case, the helper schemas would be be duplicated in each entry schema AST. Then you could optimize by creating a single shared AST.

[x0k]

I mean the following set of schemas:

• Each schema can be used for validation
• The schemas reference each other
• Multiple root schemas (v0 and ag schemas in example)

[
  {
    "type": "string",
    "$id": "https://example.com/v2"
  },
  {
    "type": "object",
    "properties": {
      "a": {
        "$ref": "https://example.com/v2#"
      }
    },
    "required": [
      "a"
    ],
    "$id": "https://example.com/v1"
  },
  {
    "allOf": [
      {
        "type": "object",
        "properties": {
          "a": {
            "$ref": "https://example.com/v2#"
          }
        }
      },
      {
        "anyOf": [
          {
            "required": [
              "a"
            ]
          }
        ]
      }
    ],
    "$id": "https://example.com/v1ag"
  },
  {
    "type": "string",
    "$id": "https://example.com/v4"
  },
  {
    "type": "object",
    "properties": {
      "b": {
        "$ref": "https://example.com/v4#"
      }
    },
    "required": [
      "b"
    ],
    "$id": "https://example.com/v3"
  },
  {
    "allOf": [
      {
        "type": "object",
        "properties": {
          "b": {
            "$ref": "https://example.com/v4#"
          }
        }
      },
      {
        "anyOf": [
          {
            "required": [
              "b"
            ]
          }
        ]
      }
    ],
    "$id": "https://example.com/v3ag"
  },
  {
    "oneOf": [
      {
        "$ref": "https://example.com/v1#"
      },
      {
        "$ref": "https://example.com/v3#"
      }
    ],
    "$id": "https://example.com/v0"
  }
]

Then you could optimize by creating a single shared AST.

So, this is not within the scope of this library?

[jdesrosiers]

So, this is not within the scope of this library?

There's nothing wrong with making one big AST for everything. The experimental API is designed to be a set of low-level utilities you can use to build any kind of JSON Schema related tooling. I didn't design the AST specifically for more than one root schema, but I always knew it could be used that way. There's nothing wrong with using it that way if that's what fits your application.