Skip to main content

Introduce Strict and Legacy All Variable Usages Are Allowed

At a glance​

Timeline​

For the following schema:

type Query {
  sum(numbers:[Int!]!): Int
}

The following query is currently valid:

query Q ($number: Int = 3) {
  sum(numbers: [1, $number, 3])
}

This query will accept the variables {"number": null} and result in a runtime field error when it turns out that null cannot be used in this non-nullable list value position. This was discussed in depth in the December 2022 Secondary EU WG meeting, resulting in this action item https://github.com/graphql/graphql-wg/issues/1337. Timestamped link to the relevant part of the discussion: https://youtu.be/nkPn-F_UBJo?list=PLP1igyLx8foH30_sDnEZnxV_8pYW3SDtb&t=2702

This PR implements the agreed solution: it introduces a stricter version of the All Variable Usages Are Allowed algorithm which forbids a nullable variable from being used in a non-nullable position; and to support existing documents a legacy version of the old algorithm is maintained (basically a copy/paste).

Under the new strict algorithm, the previous query becomes invalid and you'd need to make the variable type non-nullable:

query Q ($number: Int! = 3) {
  sum(numbers: [1, $number, 3])
}

This still allows the variable $number to be omitted, but it does not allow it to be explicitly null.

IMPORTANT NOTE: this new algorithm does not allow an argument/input object field's default value to be used if a variable is used as the value for that argument/input object field. A workaround is to copy the argument/input object field's default value to the variable definition, but this will mean that changes to the argument/input object field's default value will not be reflected by existing queries. I think this is acceptable, since there's no way to leverage the default value of an argument when passing a literal to it either - I'm in favour of literal/variable equivalence where possible, and I think that the benefits of turning this runtime error into a validation error outweigh the costs.

cc @leebyron @mjmahone @IvanGoncharov as they were participants in the discussion above and contributed to this decision.