This is a set of zero-dependency rules to be used with the graphql-schema-linter package.

These rules are based on schema conventions we have established at Pinterest.

For instance, to package with your project, run:

npm install -D @pinterest/graphql-lint-rules

Or, if you want to use these rules locally across multiple projects, run:

yarn global add @pinterest/graphql-lint-rules

// The exact format here depends on what configuration format (JSON or JavaScript) you use
"customRulePaths": ["/path/to/npm_modules/@pinterest/graphql-lint-rules/rules/*.js"]

If you prefer, pass --custom-rule-paths '/path/to/npm_modules/@pinterest/graphql-lint-rules/rules/*.js' to graphql-schema-linter instead.

For example:

"rules": [
  "composite-fields-are-nullable"
]

When a field has a composite GraphQL type (object, interface, union), it must not be declared as a non-null field.

Composite fields are more often tied to a resolver that needs to fetch some additional data than are scalar fields. Any failure that results from that data fetch will cause a null value to bubble up the query tree unless the field is marked as nullable.

While it is possible to change a nullable field to non-null in schema updates, the reverse is a breaking change. The "safe" default behavior is to have these fields be nullable.

Caleb Meredith has written a good summary of issues around nullability.

The pageInfo field is specified by the Relay cursor connections specification as a non-null field, and therefore violates this rule. Since it's more important to follow the spec than this rule, fields that use the PageInfo type are skipped.

As many schemas will have a nonzero number of exceptions to this rule, fields can be marked with an @allowNonNull decorator (which you must add to your schema) to skip this rule as well. Standard graphql-schema-linter overrides will work as well.

Also, a list of exceptions can be added to the rulesOptions, so that entire scenarios can be excluded at once.

"rulesOptions": [
  "composite-fields-are-nullable": {
    "exceptions": [
      "ErrorInfo",
    ]
  },
]

This rule will validate that object type field and interface type field names are camel-cased or start with an underscore.

GraphQL field names should be valid GraphQL identifiers. For field names that have numeric names, one can use the approach to add an underscore to the start of the field name to make it a valid GraphQL identifier. This rule checks that field names follow the camelCase format, but also allowing for them to start with an underscore.

This rule will validate that fields do not return JSON or any other variation of it.

When a type implements the Node interface, it must contain a non-nullable string named entityId.

For cross-compatibility with other APIs, our model types need to expose the primary identifier that can be used to identify the object across services. We do not, however, want to use Node's id field for this identifier, as clients expect id to be globally unique across model and type (which our primary identifier is not). By enforcing the presence of entityId, we ensure there is a field to expose both, without changing the standard Node interface and still allowing for exceptional cases.

This rule will validate that types that end in ConnectionContainer return types that end in Connection.

This rule will validate the schema adheres to section 2 (Connection Types) of the Relay Cursor Connections Specification.

More specifically:

• Only object type names may end in Connection. These object types are considered connection types.
• Connection types must have a edges field that returns a list type.
• Connection types must have a pageInfo field that returns a non-null PageInfo object.

This is basically the same validation done in graphql-schema-linter, but with some extra validation:

• edges field must return a list type of an object type that ends in Edge.

This rule will validate the schema adheres to section 3 (Edge Types) of the Relay Cursor Connections Specification.

More specifically:

• Edge types must have a cursor field that returns a type that serializes as a String.
• Edge types must have a node field return either a Scalar, Enum, Object, Interface, Union, or a Non-Null wrapper around one of those types.

Also, it adds some validation regarding Pinterest-specific rules:

• Only edge type names may end in Edge. These object types are considered edge types.

This rule validates that when an id field is used, it is of an ID scalar type that is assumed to exist in the schema as a string scalar.

Relay imposes this requirement because it makes assumptions about any id field that it sees.

See CONTRIBUTING.md.