[Stitching] Issues with enums

[orta]

From Slack

So, take this query

Causes issues like:

  "errors": [
    {
      "message": "Expected a value of type \"HomePageHeroUnitMode\" but received: LEFT_DARK"
    },
    {
      "message": "Expected a value of type \"HomePageHeroUnitMode\" but received: LEFT_LIGHT"
    },
    {
      "message": "Expected a value of type \"HomePageHeroUnitMode\" but received: LEFT_DARK"
    },
    {
      "message": "Expected a value of type \"HomePageHeroUnitMode\" but received: LEFT_DARK"
    },
    {
      "message": "Expected a value of type \"HomePageHeroUnitMode\" but received: LEFT_LIGHT"
    }
  ],

A quick look inside the enum shows they're defined as:

const HomePageHeroUnitType = new GraphQLObjectType({
  name: "HomePageHeroUnit",
  fields: {
    ...GravityIDFields,
    cached,
    mode: {
      type: new GraphQLEnumType({
        name: "HomePageHeroUnitMode",
        values: {
          LEFT_DARK: {
            value: "left white",
          },
          LEFT_LIGHT: {
            value: "left black",
          },
          CENTERED_DARK: {
            value: "center white",
          },
          CENTERED_LIGHT: {
            value: "center black",
          },
          RIGHT_DARK: {
            value: "right white",
          },
          RIGHT_LIGHT: {
            value: "right black",
          },
        },
      }),
      resolve: ({ type, menu_color_class }) => {
        return type.toLowerCase() + " " + menu_color_class.toLowerCase()
      },
    },

it looks like in turning on stitching, the value being passed into the completeLeafValue is the actual enum name LEFT_DARK not the value left white (sidenote: they're backwards?)

[orta]

Possibly - ardatan/graphql-tools#363

[orta]

Doesn't look like this fixes it.

Going to look at the merged resolvers, to verify that the resolvers in metaphysics' code are correct. This tells us whether it's on the way in, or on the way back.

If it's on the way back then look at potential verification passes inside the schema checks

[orta]

Looks like it's doing two sets of "resolves" or verification steps, the results of the first one are put into the second one. So the enum "complete" function will mutate the results (from "left white" to "LEFT_DARK") and on the second check you have the name instead of the value

I'd imagine the two sets of completions comes from first one being in "our" API, and the latter one coming from the merged schema "above".

Left = stitched

[orta]

I chatted privately with @freiksenet about this back in november - his recommendation was to try treating Metaphysics as an external API also (so that stitching doesn't use the real resolvers initially)

So more like this

metaphysics/src/lib/mergeSchemas.js

Lines 10 to 34 in 362efca

	export function createConvectionLink() {
	const httpLink = createHttpLink({
	fetch,
	uri: `${process.env.CONVECTION_API_BASE}/graphql`,
	})
	const middlewareLink = new ApolloLink((operation, forward) => forward(operation))
	const authMiddleware = setContext((_request, context) => {
	const locals = context.graphqlContext && context.graphqlContext.res.locals
	const tokenLoader = locals && locals.dataLoaders.convectionTokenLoader
	const headers = { ...(locals && requestIDHeaders(locals.requestIDs)) }
	// If a token loader exists for Convection (i.e. this is an authenticated request), use that token to make
	// authenticated requests to Convection.
	if (tokenLoader) {
	return tokenLoader().then(({ token }) => ({
	headers: Object.assign(headers, { Authorization: `Bearer ${token}` }),
	}))
	}
	// Otherwise use no authentication, which is also meant for fetching the service’s (public) schema.
	return { headers }
	})
	return middlewareLink.concat(authMiddleware).concat(httpLink)
	}

Instead of

metaphysics/src/lib/mergeSchemas.js

Line 7 in 362efca

import localSchema from "../schema"

metaphysics/src/lib/mergeSchemas.js

Lines 50 to 51 in 362efca

	return _mergeSchemas({
	schemas: [localSchema, convectionSchema, linkTypeDefs],

[orta]

unblocked by ardatan/graphql-tools#613

[orta]

Fixed 👍