fix(graphql): generate valid operations against large schemas

The selection-set builder emitted composite fields with no sub-selection
(when it hit the depth cap or cycle guard) and selected nested fields whose
required arguments it could not fill. Against a rich schema like GitLab's,
every generated tool over a non-trivial return type failed server-side
validation.

buildSelectionSet now skips nested fields with a required argument it cannot
supply, and drops composite fields whose sub-selection comes back empty
instead of printing them bare. Adds a regression test that drives the plugin
end to end against a GitLab-shaped graphql-js server.

Fixes #1146

Author: RhysSullivan

.changeset/graphql-valid-selections.md

@@ -0,0 +1,5 @@
+---
+"@executor-js/plugin-graphql": patch
+---
+
+Fix the GraphQL plugin generating invalid operations against large schemas. The auto-generated selection set no longer emits composite (object/connection) fields without a sub-selection, and no longer selects nested fields whose required arguments it cannot supply. Generated tools now validate against rich schemas (such as GitLab's) instead of failing on every call against a rich return type.

packages/plugins/graphql/src/sdk/plugin.test.ts

@@ -28,6 +28,7 @@ import { endpointForTelemetry } from "./invoke";
 import { introspect } from "./introspect";
 import type { IntrospectionResult } from "./introspect";
 import {
+  makeGitlab1146Schema,
   makeGreetingGraphqlSchema,
   serveGraphqlFailureTestServer,
   serveGraphqlTestServer,
@@ -885,3 +886,71 @@ describe("graphqlPlugin detect URL-token fallback", () => {
     }),
   );
 });
+
+// Issue #1146: against a large, real-world schema (GitLab) the auto-generated
+// operations were invalid GraphQL and every call against a rich object type
+// failed validation. These drive the real plugin (live introspection -> operation
+// generation -> invocation) against a GitLab-shaped graphql-yoga server, which
+// validates with graphql-js exactly like the @emulators/gitlab surface. A clean
+// `ok: true` therefore proves the generated operation is valid GraphQL.
+describe("graphqlPlugin generates valid operations against rich schemas (#1146)", () => {
+  const gitlabServer = serveGraphqlTestServer({ schema: makeGitlab1146Schema() });
+
+  const lastQuery = (requests: { readonly payload: { readonly query?: string } }[]): string =>
+    requests[requests.length - 1]?.payload.query ?? "";
+
+  it.effect("query.metadata: drops the nested field whose required argument it cannot fill", () =>
+    Effect.gen(function* () {
+      const server = yield* gitlabServer;
+      const executor = yield* makeExecutor();
+
+      yield* executor.graphql.addIntegration({ endpoint: server.endpoint, slug: "gitlab" });
+      yield* createOrgConnection(executor, {
+        integration: "gitlab",
+        name: "main",
+        template: "none",
+        value: "unused",
+      });
+
+      yield* server.clearRequests;
+      const result = yield* executor.execute(toolAddr("gitlab", "main", "query.metadata"), {});
+
+      // Valid GraphQL: the server accepts and resolves it (no validation errors).
+      expect(result).toMatchObject({ ok: true });
+      // `featureFlags(names: [String!]!)` has a required arg the generator cannot
+      // supply, so it is omitted rather than emitted without arguments.
+      const query = lastQuery(yield* server.requests);
+      expect(query).not.toContain("featureFlags");
+      // Fields it CAN select are still present (no over-pruning).
+      expect(query).toContain("kas {");
+    }),
+  );
+
+  it.effect("query.currentUser: never emits a composite field without a sub-selection", () =>
+    Effect.gen(function* () {
+      const server = yield* gitlabServer;
+      const executor = yield* makeExecutor();
+
+      yield* executor.graphql.addIntegration({ endpoint: server.endpoint, slug: "gitlab2" });
+      yield* createOrgConnection(executor, {
+        integration: "gitlab2",
+        name: "main",
+        template: "none",
+        value: "unused",
+      });
+
+      yield* server.clearRequests;
+      const result = yield* executor.execute(toolAddr("gitlab2", "main", "query.currentUser"), {});
+
+      expect(result).toMatchObject({ ok: true });
+      const query = lastQuery(yield* server.requests);
+      // The connection is traversed (proves recursion still works)...
+      expect(query).toContain("mergeRequests {");
+      expect(query).toContain("nodes {");
+      // ...but composite leaves past the depth cap (`author`, `assignees`) are
+      // dropped instead of being emitted bare, which the server would reject.
+      expect(query).not.toContain("author");
+      expect(query).not.toContain("assignees");
+    }),
+  );
+});

packages/plugins/graphql/src/sdk/plugin.ts

@@ -38,6 +38,7 @@ import {
   type IntrospectionResult,
   type IntrospectionType,
   type IntrospectionField,
+  type IntrospectionInputValue,
   type IntrospectionTypeRef,
 } from "./introspect";
 import { extract } from "./extract";
@@ -217,13 +218,40 @@ const unwrapTypeName = (ref: IntrospectionTypeRef): string => {
   return "Unknown";
 };
 
+// Bound the auto-generated selection so a huge schema (GitLab has 4000+ types)
+// does not produce an unbounded operation: cap recursion depth and the number
+// of fields emitted per level.
+const MAX_SELECTION_DEPTH = 2;
+const MAX_FIELDS_PER_LEVEL = 12;
+
+// Composite (output) types require a sub-selection; leaves (scalars/enums) must
+// not have one. Anything we cannot resolve in the type map is treated as a leaf
+// (custom scalars live in the map as SCALAR; truly-unknown types are rare).
+const isCompositeType = (
+  ref: IntrospectionTypeRef,
+  types: ReadonlyMap<string, IntrospectionType>,
+): boolean => {
+  const kind = types.get(unwrapTypeName(ref))?.kind;
+  return kind === "OBJECT" || kind === "INTERFACE" || kind === "UNION";
+};
+
+// A nested field whose argument is non-null without a default cannot be selected
+// by the generator: it has no value to pass and emitting the field without the
+// argument is invalid (e.g. GitLab's `metadata.featureFlags(names:)`). Required
+// root-field arguments are different: those are threaded as operation variables
+// (and surfaced on the tool's input schema) in `buildOperationStringForField`.
+const hasRequiredArgWithoutDefault = (field: IntrospectionField): boolean =>
+  field.args.some(
+    (arg: IntrospectionInputValue) => arg.type.kind === "NON_NULL" && arg.defaultValue == null,
+  );
+
 const buildSelectionSet = (
   ref: IntrospectionTypeRef,
   types: ReadonlyMap<string, IntrospectionType>,
   depth: number,
   seen: Set<string>,
 ): string => {
-  if (depth > 2) return "";
+  if (depth > MAX_SELECTION_DEPTH) return "";
 
   const leafName = unwrapTypeName(ref);
   if (seen.has(leafName)) return "";
@@ -236,13 +264,25 @@ const buildSelectionSet = (
 
   seen.add(leafName);
 
-  const subFields = objectType.fields
-    .filter((f: IntrospectionField) => !f.name.startsWith("__"))
-    .slice(0, 12)
-    .map((f: IntrospectionField) => {
+  const subFields: string[] = [];
+  for (const f of objectType.fields) {
+    if (subFields.length >= MAX_FIELDS_PER_LEVEL) break;
+    if (f.name.startsWith("__")) continue;
+    // Cannot synthesize a value for a required nested argument: skip the field
+    // rather than emit an invalid selection.
+    if (hasRequiredArgWithoutDefault(f)) continue;
+
+    if (isCompositeType(f.type, types)) {
+      // A composite field MUST have a sub-selection. When the recursion yields
+      // nothing (depth cap, cycle guard, or every child skipped) we drop the
+      // field instead of emitting it bare (which the server rejects).
       const sub = buildSelectionSet(f.type, types, depth + 1, seen);
-      return sub ? `${f.name} ${sub}` : f.name;
-    });
+      if (sub) subFields.push(`${f.name} ${sub}`);
+    } else {
+      // Scalar / enum leaf: select it directly.
+      subFields.push(f.name);
+    }
+  }
 
   seen.delete(leafName);
 

packages/plugins/graphql/src/testing/index.ts

@@ -251,6 +251,77 @@ export const makeGreetingGraphqlSchema = (
   });
 };
 
+// A small GitLab-shaped schema that reproduces the two failure modes from issue
+// #1146 when the plugin auto-generates selection sets against it:
+//   1. `metadata.featureFlags(names: [String!]!)` has a REQUIRED nested argument
+//      the generator cannot fill.
+//   2. `currentUser` nests connections deep enough that composite leaf fields
+//      (`author`, `assignees`) sit past the recursion depth cap, so a naive
+//      generator emits them with no sub-selection.
+// graphql-yoga validates with graphql-js, exactly like the @emulators/gitlab
+// surface, so a generated operation that validates clean here is valid GraphQL.
+export const makeGitlab1146Schema = (): GraphQLSchema =>
+  createSchema<GraphqlTestContext>({
+    typeDefs: /* GraphQL */ `
+      type Query {
+        metadata: Metadata
+        currentUser: User
+      }
+
+      type Metadata {
+        version: String
+        revision: String
+        enterprise: Boolean
+        featureFlags(names: [String!]!): [FeatureFlag!]!
+        kas: Kas
+      }
+
+      type FeatureFlag {
+        name: String
+        enabled: Boolean
+      }
+
+      type Kas {
+        enabled: Boolean
+        externalUrl: String
+      }
+
+      type User {
+        active: Boolean
+        admin: Boolean
+        mergeRequests: MergeRequestConnection
+      }
+
+      type MergeRequestConnection {
+        count: Int
+        nodes: [MergeRequest!]
+        pageInfo: PageInfo
+      }
+
+      type PageInfo {
+        hasNextPage: Boolean
+        endCursor: String
+      }
+
+      type MergeRequest {
+        id: ID
+        title: String
+        author: Author
+        assignees: AssigneeConnection
+      }
+
+      type Author {
+        name: String
+        username: String
+      }
+
+      type AssigneeConnection {
+        count: Int
+        nodes: [Author!]
+      }
+    `,
+  });
+
 export const TestLayers = {
   greeting: () => GraphqlTestServer.layer({ schema: makeGreetingGraphqlSchema() }),
   greetingWithOAuth: () =>