[Aschenbath]

Description

Use each GraphQL operation's name when one input document contains multiple operations. Apply the source name override only when there is one source with one operation.

Related Issue

Fixes #3532.

Motivation and Context

A document can contain several named queries that share a fragment. The parser already builds a response type for each operation, but the input layer registers every response under the source name when there is only one source file. The second registration fails.

Previous Behaviour / Output

For example, a document with ViewerQuery, UserQuery, and a shared UserFields fragment fails with Internal error: Trying to add top-level with existing name. This also affects a query and a mutation in the same document.

New Behaviour / Output

The document produces separate ViewerQuery and UserQuery response types and converters. A single named or anonymous operation still uses the requested source name. Multiple source files retain their existing operation naming behavior.

How Has This Been Tested?

• Before the fix, four new regression cases failed with the reported duplicate-name error; three compatibility cases passed. All seven cases pass after the fix.
• The regression reads a shared-fragment GraphQL fixture and JSON response fixtures, generates JavaScript, round-trips both responses, and rejects each response through the other operation's converter. These tests use the input API because the existing GraphQL language drivers assume one fixed top-level response name.
• Windows, Node 24.22.0: npm run build, all 248 unit tests, and npm run lint passed.
• Linux, Node 24.22.0: build, all seven new regression cases, and all 18 existing JavaScript/TypeScript GraphQL fixture cases passed (QUICKTEST=true FIXTURE=graphql-javascript,graphql-typescript CPUs=2 npm run test:fixtures).
• CLI checks generated JavaScript, Swift, and Kotlin for the same document. The generated JavaScript round-tripped both responses and rejected crossed responses. Swift and Kotlin output was inspected; native compilation of those outputs was not run locally.

Screenshots (if appropriate):

Not applicable.

AI disclosure: implementation, tests, and this PR description were produced with OpenAI GPT-6 through Codex. The change was reviewed against the source and verified with the checks listed above.