Cannot access 'typeMap' before initialization

[gmac]

There appears to be finickiness with abstract type definitions in a merge. The example below errors with ReferenceError: Cannot access 'typeMap' before initialization when the union type is present. Interestingly though, I have a branch in QA right now that successfully implements a similar pattern. That points to one of a few scenarios:

1. My QA branch is on 6.0.18 still, so this could be a regression in 6.1.0 (<- this appears to be the case).
2. Maybe merging abstracts is order-dependent, and I happen to have a compatible order in my working branch but not here.

I know this works – I'm not sure why it doesn't work here.

const { makeExecutableSchema } = require('@graphql-tools/schema');
const { stitchSchemas } = require('@graphql-tools/stitch');
const { addMocksToSchema } = require('@graphql-tools/mock');

const imageSchema = makeExecutableSchema({
  typeDefs: `
    type Image {
      id: ID!
      url: String!
    }

    type Query {
      images(ids: [ID!]!): [Image]!
    }
  `
});

const contentSchema = makeExecutableSchema({
  typeDefs: `
    type Post {
      leadArt: LeadArt
    }

    type Image {
      id: ID!
    }

    type Video {
      id: ID!
      url: String!
    }

    union LeadArt = Image | Video

    type Query {
      posts(ids: [ID!]!): [Post]!
    }
  `
});

const gatewaySchema = stitchSchemas({
  subschemas: [
    {
      schema: addMocksToSchema({ schema: imageSchema }),
      merge: {
        Image: {
          selectionSet: '{ id }',
          fieldName: 'images',
          key: ({ id }) => id,
          argsFromKeys: (ids) => ({ ids }),
        },
      },
    },
    {
      schema: addMocksToSchema({ schema: contentSchema }),
      merge: {
        Post: {
          selectionSet: '{ id }',
          fieldName: 'posts',
          key: ({ id }) => id,
          argsFromKeys: (ids) => ({ ids }),
        },
      },
    },
  ],
  mergeTypes: true
});

Trace:

/Users/greg.macwilliam/Desktop/micromedia/node_modules/@graphql-tools/stitch/index.cjs.js:395
            typeMap[type.name] = type;
            ^

ReferenceError: Cannot access 'typeMap' before initialization
    at /Users/greg.macwilliam/Desktop/micromedia/node_modules/@graphql-tools/stitch/index.cjs.js:395:13
    at Array.forEach (<anonymous>)
    at /Users/greg.macwilliam/Desktop/micromedia/node_modules/@graphql-tools/stitch/index.cjs.js:394:22
    at Array.reduce (<anonymous>)
    at mergeUnionTypeCandidates (/Users/greg.macwilliam/Desktop/micromedia/node_modules/@graphql-tools/stitch/index.cjs.js:393:29)
    at mergeCandidates (/Users/greg.macwilliam/Desktop/micromedia/node_modules/@graphql-tools/stitch/index.cjs.js:277:16)
    at /Users/greg.macwilliam/Desktop/micromedia/node_modules/@graphql-tools/stitch/index.cjs.js:591:33
    at Array.forEach (<anonymous>)
    at buildTypeMap (/Users/greg.macwilliam/Desktop/micromedia/node_modules/@graphql-tools/stitch/index.cjs.js:583:33)
    at stitchSchemas (/Users/greg.macwilliam/Desktop/micromedia/node_modules/@graphql-tools/stitch/index.cjs.js:961:21)

[yaacovCR]

Hmmm. Below passes for me on master.

describe('new', () => {
  it('works', () => {
    const imageSchema = makeExecutableSchema({
      typeDefs: `
        type Image {
          id: ID!
          url: String!
        }

        type Query {
          images(ids: [ID!]!): [Image]!
        }
      `
    });

    const contentSchema = makeExecutableSchema({
      typeDefs: `
        type Post {
          leadArt: LeadArt
        }

        type Image {
          id: ID!
        }

        type Video {
          id: ID!
          url: String!
        }

        union LeadArt = Image | Video

        type Query {
          posts(ids: [ID!]!): [Post]!
        }
      `
    });

    const gatewaySchema = stitchSchemas({
      subschemas: [
        {
          schema: addMocksToSchema({ schema: imageSchema }),
          merge: {
            Image: {
              selectionSet: '{ id }',
              fieldName: 'images',
              key: ({ id }) => id,
              argsFromKeys: (ids) => ({ ids }),
            },
          },
        },
        {
          schema: addMocksToSchema({ schema: contentSchema }),
          merge: {
            Post: {
              selectionSet: '{ id }',
              fieldName: 'posts',
              key: ({ id }) => id,
              argsFromKeys: (ids) => ({ ids }),
            },
          },
        },
      ],
      mergeTypes: true
    });

    expect(gatewaySchema).toBeInstanceOf(GraphQLSchema);
  });
});

[yaacovCR]

What version of GraphQL are you using?

If you submit a PR with above test, ci will test with latest versions of 14 and 15

[gmac]

Curious... not rightly sure, I just did a “yarn install” on everything in a new project and assumed I got the latest of everything. Will have to do some more digging. Thanks for having a look.

[gmac]

Welp, I'm stumped. I've put the test into the GraphQL Tools repo and am still getting the same result: #1981.

See failure in https://github.com/ardatan/graphql-tools/pull/1981/checks?check_run_id=1056293407.

[yaacovCR]

Not sure why passed for me but looks easy though to correct

[gmac]

Is there a chance this could be the result of a race condition in schema definition order? I’m a bit thrown by this one having done it before.

…

On Tue, Sep 1, 2020 at 9:20 AM Yaacov Rydzinski ***@***.***> wrote: Not sure why passed for me but looks easy though to correct — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#1973 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAFRROGYP45NJ3FA267BYTLSDTYIBANCNFSM4QRHLWAQ> .

[yaacovCR]

I think just should have used acc instead of typemap in that reduce,

[gmac]

Uh oh, this definitely appears to be a version issue. I just setup a 6.2 upgrade branch and this error appeared... Last night when I tried downgrading to 6.0.18 to smoketest, I'll bet I bumped the packaged down but not the lock file, at which time my 0.18 test was still running a more recent version.

[gmac]

Should be good to go in #1981, with new test coverage. Thanks for mentioning the fix – that was it!