Stricter subgraph schema validation

Do you want to request a feature or report a bug?

Bug

What is the current behavior?

We currently allow a lot of names in schemas that either cause problems or that we want to deliberately forbid in order to give us a reserved namespace. Examples are:

• Entity fields called set and unset. AssemblyScript apparently allows things like

  entity.set = "foo"
  

  which will replace the set(key: string, value: Value) method in the Entity base class with a string, which breaks all later attempts to assign fields in the same handler.

• Entity types and fields with a leading underscore (we have already introduced the special _Schema_ type and we want to move towards reserving names starting with underscores for Graph features.

If the current behavior is a bug, please provide the steps to reproduce and if possible a minimal demo of the problem.

Create a subgraph with a schema like this:

type Set @entity {
  id: ID!
}

type SetCreation @entity {
  id: ID!
  set: Set!
  name: String!
}

then write a handler that does the following:

let creation = new SetCreation("foo")
creation.set = "bar"
creation.name = "baz"

this will fail to compile.

What is the expected behavior?

We should catch problematic names at deploy (or even at codegen/build time) in graph-cli and graph-node.

The issue names graph-cli and graph-node as the validation entry points; start by tracing where subgraph schemas are checked during deploy or code generation. Reproduce the provided schema and handler, then ensure reserved names such as set, unset, and leading-underscore entity or field names are rejected before deployment or build.