Description

We recently found an oversight of selection sets: we don't check for duplicates. In the unlikely (but, as evidenced, not impossible) case where we don't prevent a conflict from arising, said conflict should be detected when we build the schema. And in turns out that in some degenerate cases, we would construct a schema without detecting the conflict (specifically: when the conflicting root fields would yield types with the exact same fields, that therefore can't be differentiated).

(See discussion with @codingkarthik on Slack for more info, the test case, and the context.)

Affected components

• Server

Solution and Design

This is a very simple fix, that only fixes the problem for root fields. There are two other ways we could prevent this from happening:

• make all combinators monadic and throw on invalid schema (bad, no)
• make all combinators require a hashmap of field parsers, to delegate the responsibility of checking for duplicates to the callers (better, but perhaps overkill)