Type System & SDL

Because the published field names are derived from server-side identifiers. A refactor that renames a property renames the public field with it, and the rename looks like ordinary internal cleanup in review — no API artefact changed for anyone to notice.

Give the derived schema a committed artefact so surface changes show up in review, agree the shape in SDL before it is built, and name an owner for the shared graph's conventions rather than assuming the implementing team reviews the contract.

It attaches a URL to a custom scalar definition, pointing at a human-readable specification of how that scalar's values are serialized. The server never fetches or enforces it — humans and tooling read it through introspection.

Put a string literal — usually a triple-quoted block string — immediately before the definition. That description is part of the schema and readable through introspection. A # comment is ignored by the parser and never reaches a client.

It marks an input object type whose caller must supply exactly one of its fields, with a non-null value. Every field of such an input must be nullable and carry no default. It is a working-draft addition, not part of the last ratified specification edition.

Each field is a __Field record whose args are __InputValue records: name, description, type as a full wrapper chain, and defaultValue. The default is a String holding the value printed as GraphQL source, not a typed JSON value.

Introspection describes the type system, so it lists a schema's directive definitions but never their applications. The only applications visible are two the meta-schema was given dedicated fields for: deprecated, as isDeprecated and deprecationReason, and specifiedBy, as specifiedByURL.

A chain of Non-Null fields that returns to the same input object can never be satisfied by a finite value. The specification requires at least one link in such a cycle to be nullable or a list type.