🎉 Explore GraphQLConf 2026 • May 19-21 • Fremont, CA • View the schedule
Upgrade Guidesv16 to v17

What Changed in GraphQL.js v17

GraphQL.js v17 has a release candidate available for final testing and feedback. Use this guide to prepare v16 projects for testing the release-candidate line.

GraphQL.js v17 keeps the core programming model: build a schema, parse a document, validate it, and execute it. Most changes make boundaries more explicit. Stable single-result execution is separate from experimental incremental delivery, input coercion is split from diagnostic validation, and development checks are opt-in. Host integration features such as harnesses, abort signals, execution hooks, and Node.js tracing channels are explicit GraphQL.js runtime APIs.

Contents

Using this Guide

This guide is for projects that depend on GraphQL.js, including applications, servers, libraries, and tools. It compares the latest v16 release line with the current v17 release candidate, and focuses on differences you will see when testing that RC from an up-to-date v16 project. If a change is already present in both lines, including v17 work that also shipped in v16, it is not listed here as v16-to-v17 migration work.

Changes described below use these labels:

  • Breaking change: v16 code may need to change before it runs on v17.
  • Breaking change for earlier v17 alpha users: code that adopted an experimental v17 alpha feature may need to change; it is not stable v16-to-v17 migration work.
  • Behavioral tightening: v17 validates or reports a case more precisely.
  • Deprecation: the v16 API still works in v17, but should be migrated before v18.
  • New stable API: a new public API that can be adopted independently.
  • Experimental or opt-in: available in v17, but proposal-backed or outside the default execution path.

Platform and Package Shape

Node.js and TypeScript

Breaking change. GraphQL.js v17 requires Node.js 22, 24, 25, or 26 and later:

{
  "engines": {
    "node": "^22.0.0 || ^24.0.0 || ^25.0.0 || >=26.0.0"
  }
}

Upgrade Node.js before upgrading GraphQL.js. This separates runtime and package-manager errors from GraphQL.js migration errors.

Breaking change. The published type definitions target TypeScript 4.4 and newer.

Conditional exports

Breaking change. v17 uses package exports and modern package conditions to select the right build. Use public entry points such as graphql, graphql/execution, graphql/language, graphql/type, graphql/utilities, and graphql/validation.

GraphQL.js also exposes files below those package modules through exports, so deep imports can resolve. Treat those deep imports as file-level access rather than semver-stable API boundaries: GraphQL.js only promises semver compatibility for the root package and package-module entry points. Prefer the public entry points above for application and library code.

If you import individual files, v17 has fewer true moves than older release lines. The moved or renamed cases are summarized in Deep Import Moves. Removed APIs and compatibility paths are covered in the sections where they affect public imports.

The v17 package also contains a matching __dev__/ build tree. When your runtime or bundler honors package conditions, import the normal graphql/... specifier and select the development condition so package resolution chooses that build. If you are loading files in an environment that does not use package conditions, such as direct CDN file URLs, import from the matching __dev__/ file explicitly when you need the development build.

Breaking change. The deprecated graphql/subscription compatibility subpath is gone. Import subscription APIs from graphql or graphql/execution.

- import { subscribe } from 'graphql/subscription';
+ import { subscribe } from 'graphql/execution';

Development mode

Breaking change. Development mode is disabled by default and no longer depends on NODE_ENV. Enable it with the development package condition or by calling enableDevMode() during application startup.

import { enableDevMode, isDevModeEnabled } from 'graphql';
 
if (process.env.NODE_ENV === 'development') {
  enableDevMode();
}
 
console.log(isDevModeEnabled());

Development mode currently helps diagnose accidental use of multiple GraphQL.js module instances. See Development Mode for runtime and bundler setup.

Request Pipeline and Harnesses

graphql() and graphqlSync()

New stable API. The high-level graphql() and graphqlSync() APIs still use the object-argument form from v16. In v17, that object can also carry parser options, validation options, execution options, hideSuggestions, abortSignal, hooks, and a custom harness.

const result = await graphql({
  schema,
  source,
  variableValues,
  operationName: 'Viewer',
  hideSuggestions: true,
  abortSignal,
});

This does not make graphql() a framework. It remains the convenience API for a single-result “parse, validate, execute” request. The new arguments let simple hosts use common v17 options without rebuilding the whole pipeline.

GraphQL Harness

New stable API. The GraphQLHarness TypeScript interface and defaultHarness runtime object let hosts replace the parse, validate, execute, and subscribe phases used by graphql() and graphqlSync().

The harness is modeled after Envelop-style plugin pipelines. It brings the broader phase types used by that ecosystem closer to the reference implementation, so frameworks can accept a custom harness and plugin systems can interoperate around a shared request-pipeline shape.

For example, GraphQLParseFn can return either a parsed DocumentNode or a promise for one, even though the built-in parse() function is synchronous. The harness execute function follows execute() and does not include the experimental incremental delivery return type.

Use GraphQL Harness for examples, and call experimentalExecuteIncrementally() directly when an operation may use @defer or @stream.

Execution and Incremental Delivery

Single-result execution

Breaking change for earlier v17 alpha users. execute() is now the stable single-result executor. It does not support incremental delivery. If a schema or operation opts into @defer or @stream, execute it with experimentalExecuteIncrementally() instead.

- import { execute } from 'graphql';
+ import { experimentalExecuteIncrementally } from 'graphql';

This keeps the execute() contract simple: callers receive one ExecutionResult, or a promise for one.

Incremental delivery

The incremental-delivery notes in this section compare the v17 release candidate with earlier v17 alpha releases of this experimental feature. The current payload model follows the GraphQL WG defer/stream RFC payload format.

Experimental or opt-in. experimentalExecuteIncrementally() returns either a normal ExecutionResult or an object with initialResult and an async iterator of subsequentResults.

const result = await experimentalExecuteIncrementally({ schema, document });
 
if ('initialResult' in result) {
  send(result.initialResult);
 
  for await (const subsequentResult of result.subsequentResults) {
    send(subsequentResult);
  }
} else {
  send(result);
}

Experimental or opt-in. legacyExecuteIncrementally() remains available for hosts that still need the older incremental delivery payload shape from earlier v17 alpha releases. The legacy shape identifies deferred and streamed payloads with fields such as path and optional label, and can duplicate field data across payloads. The current experimental shape registers pending work by id and reports completion with completed entries.

Breaking change for earlier v17 alpha users. The legacy executor preserves that older incremental payload format, but the even older return type shape with singleResult as a discriminator is gone. Remove branches that check for singleResult.

Schema setup, directive validation, result shapes, and transport guidance are covered in Defer and Stream.

Behavioral tightening. Defer/stream validation now tracks named fragment spreads through the selected operation rather than treating every fragment use in a document the same way. Subscription operations reject active @defer and @stream usage found through named fragments. Root-field restrictions apply only to mutation and subscription root selections, including through fragments whose type condition is an interface implemented by the root type. Query root selections may use incremental delivery.

When a fragment is shared between query and subscription root fields, disable incremental behavior on the subscription path with a variable-backed if argument on @defer or @stream, or with applicable @skip and @include directives. Root-selection restrictions are separate: do not place @defer or @stream on mutation or subscription root selections themselves.

Resolver return values

New stable API. List fields can resolve to async iterables. This is useful for values that naturally arrive over time and is especially relevant when a host opts into @stream.

Custom execution helpers

New stable API. v17 exposes the helper boundary used by execute() itself. validateExecutionArgs() validates and normalizes ExecutionArgs, including schema checks, operation selection, variable coercion, fragment information, default resolvers, and execution options. It returns either ValidatedExecutionArgs or a list of GraphQLError values.

Then use the root selection set helper that matches the executor you are building:

  • executeRootSelectionSet() for stable single-result execution.
  • experimentalExecuteRootSelectionSet() for current incremental delivery.
  • legacyExecuteRootSelectionSet() for the legacy incremental payload shape.

These helpers do not replace operation validation with validate(). They are for hosts that already parsed and validated a document and need a custom execution function.

The native execute() and subscribe() functions preserve synchronous results when execution completes synchronously; they do not wrap every result in a promise. Custom wrappers that want the same behavior need to handle both sync and async paths. See Advanced Execution Pipelines for complete examples.

Subscriptions and Source Event Streams

Subscription return type

Breaking change. subscribe() returns a PromiseOrValue in v17 instead of always returning a promise. Existing await subscribe(args) code continues to work, but TypeScript code that assumed a promise return type must be updated to handle the synchronous path.

Breaking change for earlier v17 alpha users. subscribe() does not support incremental delivery. If a fragment is shared between query and subscription root fields, use a variable-backed if argument on @defer or @stream, or applicable @skip and @include directives, to disable incremental behavior in subscription operations.

Lower-level subscription helpers

Breaking change. createSourceEventStream() now accepts only ValidatedSubscriptionArgs. If you call it directly, call validateSubscriptionArgs() first and handle validation errors before passing the result onward. Use subscribe() when you want GraphQL.js to run the full subscription pipeline.

validateSubscriptionArgs() builds on validateExecutionArgs() and additionally asserts that the selected operation is a subscription.

New stable API. mapSourceToResponseEvent() maps a subscription source stream to execution results. Its third argument is a RootSelectionSetExecutor, which customizes how each source event is executed after the source stream has already been created. If you omit it, GraphQL.js uses the default subscription-event executor, executeSubscriptionEvent(). Call executeSubscriptionEvent() directly only when you already have ValidatedSubscriptionArgs and want to execute one subscription source event.

See Advanced Execution Pipelines for complete helper examples.

Abort Signals

Experimental or opt-in. GraphQL.js v17 adds AbortSignal support as its JavaScript runtime API for cancellation.

The GraphQL specification discusses cancellation in narrower execution cases. During non-null error propagation, sibling response positions that have not executed or yielded a value may be cancelled to avoid unnecessary work; in the specification’s normative sections, lowercase may has RFC 2119 MAY force. The subscription algorithms also describe cancelling response and source streams. The specification does not define a user or host cancelling an already issued query or mutation request.

GraphQL.js also accepts an external abortSignal on graphql(), execute(), subscribe(), and experimentalExecuteIncrementally(). Resolvers read the resolver-scoped signal with info.getAbortSignal() and pass it to downstream APIs that support cancellation.

When an external abort stops execution after a partial result exists, GraphQL.js rejects with AbortedGraphQLExecutionError. The error exposes the abort cause and the partial execution result so hosts can decide whether to log, discard, or surface partial data according to their transport policy.

At this time, GraphQL.js does not expose fine-grained per-field cancellation. Resolvers in an operation share one signal. For internally cancelled portions of an operation, GraphQL.js aborts that shared resolver signal when the result that will actually be returned has finished, notifying still-pending resolver work together. This may change in future versions.

See Abort Signals for resolver examples, HTTP request lifecycle wiring, and aborted execution handling.

Execution Hooks

Experimental or opt-in. Execution hooks are separate from abort signals. They let a host observe execution lifecycle boundaries; they do not force JavaScript work to stop.

The first hook is asyncWorkFinished. It fires after GraphQL.js has stopped producing payloads and all tracked async execution work has settled. This is useful for cleanup, logging, telemetry, tests that need a strong lifecycle boundary, or hosts that choose to delay returning a result until tracked async work is finished.

Resolvers can use info.getAsyncHelpers() to make additional work visible to that tracking. See Execution Hooks for examples.

Node.js Tracing Channels

New stable API. GraphQL.js v17 publishes lifecycle events on Node.js node:diagnostics_channel tracing channels. These channels are mainly for application performance monitoring (APM) tools, which can subscribe to channels such as graphql:parse, graphql:validate, graphql:execute, graphql:execute:variableCoercion, graphql:execute:rootSelectionSet, graphql:subscribe, and graphql:resolve without changing application execution code.

The channels are resolved at module load and no-op on runtimes that do not provide node:diagnostics_channel. GraphQL.js also exports TypeScript context types for strongly typed subscribers.

See the Tracing Channels guide for a full walkthrough, and the Diagnostics API reference for the channel names and payload shapes.

Input Coercion, Defaults, and Custom Scalars

Default values

Behavioral tightening. Argument, input-field, and directive-argument defaults are now validated as part of schema validation performed by validateSchema(). Invalid defaults that v16 could leave latent now make the schema invalid with a targeted validation error.

Deprecation. defaultValue is the legacy programmatic default format. It represents an already-coerced JavaScript value and remains available as a migration bridge.

New stable API. Prefer the new default model for programmatic schemas. Use default: { value } when you have the raw JavaScript input value before coercion. Use default: { literal } when you have the GraphQL literal. When GraphQL.js builds a schema from SDL, it now uses the literal form internally.

const field = {
  type: GraphQLString,
  args: {
    format: {
      type: GraphQLString,
      default: { value: 'short' },
    },
  },
};

This fixes subtle cases where introspection could report default values incorrectly from an already-coerced internal value.

See Passing Arguments and Mutations and Input Types.

undefined and omitted input values

Behavioral tightening. v17 continues the cleanup of JavaScript-specific undefined handling by treating undefined as absence at user-visible input boundaries. Known input-object fields with value undefined were already generally treated as omitted in the latest v16 line; v17 closes inconsistent cases around variables and unknown fields.

The v16-to-v17 differences to audit are:

  • variableValues: { x: undefined } is treated as if x was not provided. Operation and fragment variable defaults can apply, and nullable variables with no default are omitted from the coerced variable map instead of being coerced to null.
  • In input object literals such as { a: $x }, an omitted or explicitly undefined variable omits the field, so an input-field default can apply. An explicit null variable still overrides the default for nullable fields.
  • Unknown fields with value undefined in JavaScript input objects are ignored by coerceInputValue(), validateInputValue(), and valueToLiteral(). Unknown fields with any other value are still invalid.

Coercion and validation helpers

Behavioral change. coerceInputValue() and coerceInputLiteral() now return undefined when coercion fails. They are optimized for callers that want either a coerced value or failure.

New stable API. Use validateInputValue() and validateInputLiteral() when you need diagnostic errors. Use valueToLiteral() when converting an external JavaScript input value into a GraphQL literal.

New stable API. replaceVariables() replaces variables inside complex scalar literals. GraphQL.js calls it automatically during execution. If you use literal coercion helpers directly outside execution, call replaceVariables() yourself before coercing literals that may contain variables.

Built-in scalar bigint values

New stable behavior. The built-in scalar coercers accept JavaScript bigint values in the cases where the scalar can represent them.

  • GraphQLInt accepts bigint input and output values within the GraphQL 32-bit integer range.
  • GraphQLFloat accepts bigint input and output values that can be converted to a JavaScript number without losing precision.
  • GraphQLID accepts bigint input and output values and serializes them as strings.
  • GraphQLString and GraphQLBoolean accept bigint values during result coercion only.

GraphQL variable JSON and GraphQL literals still do not have a separate BigInt literal format. This is a JavaScript runtime coercion change for hosts that pass bigint values directly.

Custom scalar method names

Deprecation. v17 introduces scalar method names that match the GraphQL coercion model:

v16 namev17 namePurpose
serializecoerceOutputValueConvert resolver values into response values.
parseValuecoerceInputValueConvert variable values into internal values.
parseLiteralcoerceInputLiteralConvert GraphQL literals into internal values.
astFromValue()valueToLiteral()Convert external input values into GraphQL literals.

The v16 scalar method names still work in v17 and are deprecated for removal in v18. See Advanced Custom Scalars.

Variable Values and Resolver Info

Breaking change. getVariableValues() now returns { variableValues } on success. That value contains source information and coerced runtime values.

- const { coerced } = getVariableValues(schema, variableDefinitions, inputs);
+ const { variableValues } = getVariableValues(schema, variableDefinitions, inputs);
+ const coerced = variableValues.coerced;

This matters because v17 supports local variables on named fragments and more precise default handling. Passing only the coerced object loses information about where values came from.

Breaking change. info.variableValues follows the same model. Use info.variableValues.coerced for runtime values inside resolvers.

Behavioral tightening. v16 already used null-prototype object maps internally while preparing execution and coercion data. v17 keeps that shape through the user-visible boundary for performance, so resolver argument objects, coerced input object values, and the coerced maps inside VariableValues are null-prototype objects when user code receives them. Use Object.keys(), Object.entries(), Object.hasOwn(), or other APIs that do not rely on inherited Object.prototype methods. Avoid calling methods such as args.hasOwnProperty(...) directly.

New stable API. GraphQLResolveInfo adds getAbortSignal() and getAsyncHelpers() for the abort-signal and execution-hook APIs.

Schema, Type System, Directives, and Introspection

Schema validation

Behavioral tightening. GraphQLSchema.toConfig().assumeValid now preserves the original assumeValid setting instead of changing after schema validation has run.

Behavioral tightening. Schema validation reports duplicate use of the same object type for more than one operation root.

Behavioral tightening. Directive argument defaults are validated as part of schema validation.

Behavioral tightening. The built-in @deprecated(reason:) argument is now non-null. In v16 the argument type was nullable String, so a directive use could explicitly pass reason: null. In v17, omit reason to use the default deprecation reason; explicit null is no longer valid.

Behavioral tightening. The includeDeprecated arguments on __Type.fields, __Type.enumValues, __Type.inputFields, and __Field.args are now Boolean! = false. The includeDeprecated argument on __Schema.directives already had that shape in the latest v16 line. Clients should omit the argument or pass a Boolean value; queries that pass includeDeprecated: null to those fields are invalid in v17.

Programmatic schema APIs

New stable API. v17 exposes TypeScript schema element types and runtime assertions for programmatic schema work, including GraphQLField, GraphQLArgument, GraphQLInputField, GraphQLEnumValue, assertField(), assertArgument(), assertInputField(), and assertEnumValue().

New stable API. GraphQLSchema.getField(parentType, fieldName) resolves ordinary fields and GraphQL meta fields such as __typename, __schema, and __type.

New stable API. Custom extensions maps on schema elements support symbol keys as well as string keys.

New stable API. printDirective() prints a directive definition without printing an entire schema.

Breaking change. If you imported GraphQLInterfaceTypeNormalizedConfig from a public entry point, replace that import with ReturnType<GraphQLInterfaceType['toConfig']>.

Language, AST, Parser, Printer, and Visitor APIs

AST constants and visitors

Breaking change. The v16 alias types KindEnum, TokenKindEnum, and DirectiveLocationEnum are gone. Use Kind, TokenKind, and DirectiveLocation, which are const objects with matching union types.

- import type { KindEnum } from 'graphql';
+ import type { Kind } from 'graphql';

Breaking change. getVisitFn() is gone. Use getEnterLeaveForKind() instead.

const { enter, leave } = getEnterLeaveForKind(visitor, Kind.FIELD);

Behavioral change. The parser may omit optional AST collection properties as undefined instead of storing empty arrays. Code that reads properties such as arguments, directives, variableDefinitions, interfaces, fields, types, or operationTypes should treat them as optional.

New stable API. isSubscriptionOperationDefinitionNode() narrows subscription operation nodes.

Fragment arguments

Experimental or opt-in. v17 adds proposal-backed fragment arguments behind the experimentalFragmentArguments parser option. Older GraphQL.js versions had an experimental allowLegacyFragmentVariables option, but that was parser-only and did not work at runtime.

When enabled, named fragments can declare local variables and named fragment spreads can provide values. See Fragment Arguments.

Validation

Breaking change. The public validate() signature no longer accepts a custom TypeInfo instance as a fifth argument. JavaScript callers can still pass an extra argument, but v17 ignores it. If you need a custom traversal with custom type tracking, compose your visitor with visitWithTypeInfo().

- const errors = validate(schema, document, rules, options, customTypeInfo);
+ const errors = validate(schema, document, rules, options);

New stable API. hideSuggestions removes “Did you mean …” suggestion text from diagnostics. This is useful for public APIs that do not want to leak schema shape through error messages.

New stable API. KnownOperationTypesRule validates that an operation’s root type exists in the schema. For example, a mutation operation is invalid when the schema has no mutation root type.

The defer/stream validation rules are documented with the experimental incremental delivery feature rather than repeated here. See Defer and Stream.

Utilities and Error Handling

Schema Change utilities

Deprecation. findBreakingChanges() and findDangerousChanges() still exist in v17 and are deprecated for removal in v18. Use findSchemaChanges() for new schema registry and CI tooling.

New stable API. findSchemaChanges() reports breaking, dangerous, and safe changes from one call. Code that switches to it should handle all three categories; safe changes use the SafeChangeType and SafeChange shapes.

See Schema Evolution for schema comparison examples.

Removed helpers

Breaking change. The following deprecated helpers were removed:

  • assertValidName() and isValidNameError(); use assertName().
  • assertValidExecutionArguments(); use assertValidSchema() for schema validation and validateExecutionArgs() for execution argument validation.
  • getOperationRootType(); use schema.getRootType(operation).
  • getFieldDefFn from TypeInfo.
  • printError() and formatError(); use error.toString() or error.toJSON().

GraphQLError

Breaking change. The positional GraphQLError constructor was removed. Pass a message and an options object.

- new GraphQLError(message, nodes, source, positions, path, originalError);
+ new GraphQLError(message, {
+   nodes,
+   source,
+   positions,
+   path,
+   cause,
+   extensions,
+ });

Deprecation. GraphQLErrorOptions.originalError and GraphQLError.originalError still exist as a migration bridge, but v17 prefers the standard Error.cause field. Passing cause sets error.cause; when the cause is an Error, GraphQL.js also exposes it through originalError for compatibility. Unlike originalError, a cause stack is not copied onto the GraphQLError, because JavaScript tooling can already report cause chains.

Practical Migration Order

Use this order to keep the upgrade reviewable. The checklists below expand each category into concrete items.

  1. Update platform support and package-entry imports.
  2. Replace APIs that were removed in v17 and update changed AST/type surfaces.
  3. Update schema validation, defaults, input coercion, and variable-value handling.
  4. Update execution and subscription hosts, especially incremental-delivery paths.
  5. Migrate deprecated compatibility APIs that still work in v17 but should be gone before v18.
  6. Adopt optional runtime integrations and experimental features only where they match your server design.

Run schema validation, operation validation, execution tests, and TypeScript checks after each group. Mechanical changes are easiest to review when they are separated from behavior changes.

Detailed Migration Checklists

Platform and imports

  • Run v17 on Node.js 22, 24, 25, or 26 and later.
  • Keep TypeScript at 4.4 or newer.
  • Prefer package-module entry points such as graphql, graphql/execution, graphql/language, graphql/type, graphql/utilities, and graphql/validation for semver-stable imports. Audit file-level imports against the appendix below, since they are exposed but are not semver-stable API boundaries.
  • Replace graphql/subscription imports with graphql or graphql/execution.
  • Enable development mode explicitly in development environments, either with the development package condition or enableDevMode().

Removed APIs and required replacements

  • Replace positional GraphQLError constructor calls with the options object.
  • Replace KindEnum, TokenKindEnum, and DirectiveLocationEnum with Kind, TokenKind, and DirectiveLocation.
  • Replace getVisitFn() with getEnterLeaveForKind().
  • Remove the custom TypeInfo fifth argument from validate() call sites; use visitWithTypeInfo() for custom traversals.
  • Replace assertValidName() and isValidNameError() with assertName().
  • Replace assertValidExecutionArguments() with assertValidSchema() or validateExecutionArgs(), depending on whether you are validating a schema or execution arguments.
  • Replace getOperationRootType() with schema.getRootType(operation).
  • Remove getFieldDefFn customizations from TypeInfo.
  • Replace printError() and formatError() with error.toString() and error.toJSON().
  • Replace GraphQLInterfaceTypeNormalizedConfig with ReturnType<GraphQLInterfaceType['toConfig']>.
  • Treat empty AST collections as optional, because v17 may represent them as undefined rather than empty arrays.

Schema, defaults, and input values

  • Run validateSchema() and fix invalid argument, input-field, and directive-argument defaults.
  • Prefer default: { value } for raw JavaScript input values and default: { literal } for GraphQL literals.
  • Audit tests and integrations that pass undefined in variableValues or JavaScript input objects; v17 treats those values as omitted in more places.
  • Update direct coerceInputValue() and coerceInputLiteral() callers that expect diagnostic errors; use validateInputValue() or validateInputLiteral() when you need errors.
  • Use valueToLiteral() for converting external JavaScript input values to GraphQL literals.
  • Call replaceVariables() before direct scalar literal coercion when a literal may contain variables outside GraphQL.js execution.
  • Update getVariableValues() callers to read result.variableValues.coerced instead of result.coerced.
  • Update resolvers that read info.variableValues to use info.variableValues.coerced.
  • Treat user-visible resolver args, coerced input object values, and VariableValues.coerced as null-prototype maps.
  • Update clients or tests that passed includeDeprecated: null in introspection queries.
  • Update built-in scalar tests for JavaScript bigint values if your host passes bigint values directly.

Execution and subscriptions

  • Use execute() only for stable single-result execution.
  • Use experimentalExecuteIncrementally() for operations that may use active @defer or @stream.
  • If you tested earlier v17 alpha releases, remove code that checks the old incremental singleResult discriminator.
  • Keep legacyExecuteIncrementally() only for hosts that still need the older incremental payload shape from earlier v17 alpha releases.
  • Use validateExecutionArgs() before lower-level execution helpers such as executeRootSelectionSet(), experimentalExecuteRootSelectionSet(), and legacyExecuteRootSelectionSet().
  • Handle subscribe() returning either a value or a promise.
  • Call validateSubscriptionArgs() before createSourceEventStream().
  • Use executeSubscriptionEvent() only when executing one validated subscription event directly.
  • Replace subscription perEventExecutor usage with mapSourceToResponseEvent() and its root-selection-set executor argument.
  • Disable incremental behavior in fragments shared between query and subscription root fields with a variable-backed if argument on @defer or @stream, or applicable @skip and @include directives.

Deprecated compatibility APIs

These v16 APIs still work in v17, but are deprecated for removal in v18:

  • Programmatic defaultValue on arguments and input fields, including their config objects; use default: { value } or default: { literal }.
  • GraphQLScalarType legacy coercion hooks serialize, parseValue, and parseLiteral; use coerceOutputValue, coerceInputValue, and coerceInputLiteral.
  • GraphQLEnumType legacy coercion methods serialize(), parseValue(), and parseLiteral(); use coerceOutputValue(), coerceInputValue(), and coerceInputLiteral().
  • GraphQLScalarSerializer, GraphQLScalarValueParser, and GraphQLScalarLiteralParser; use GraphQLScalarOutputValueCoercer, GraphQLScalarInputValueCoercer, and GraphQLScalarInputLiteralCoercer.
  • astFromValue(); use valueToLiteral().
  • valueFromAST(); use coerceInputLiteral().
  • findBreakingChanges() and findDangerousChanges(); use findSchemaChanges().
  • GraphQLErrorOptions.originalError and GraphQLError.originalError; use cause.

Optional v17 runtime and experimental features

  • Use hideSuggestions when public diagnostics should omit schema suggestions.
  • Use abortSignal and info.getAbortSignal() when a host can propagate cancellation to downstream work.
  • Handle AbortedGraphQLExecutionError if your host needs access to partial execution results after an external abort.
  • Use info.getAsyncHelpers() and asyncWorkFinished hooks when a host needs a cleanup or telemetry boundary after tracked async work settles.
  • Use Node.js tracing channels primarily for application performance monitoring (APM) integrations.
  • Consider a custom GraphQLHarness object when a host needs to customize graphql() parse, validate, execute, or subscribe phases.
  • Enable experimentalFragmentArguments only for hosts that intentionally support arguments on named fragment spreads.

Deep Import Moves

These notes apply only to file-level imports. Prefer package-module imports where possible because file-level paths are not semver-stable API boundaries.

  • graphql/execution/subscribe was folded into graphql/execution/execute. The public function names did not change; prefer importing subscribe and createSourceEventStream from the graphql/execution package module.
  • The internal helper graphql/execution/mapAsyncIterator was renamed to graphql/execution/mapAsyncIterable. It remains internal, and the v17 helper no longer preserves a custom async-generator return value.
  • graphql/utilities/findBreakingChanges moved to graphql/utilities/findSchemaChanges. The deprecated findBreakingChanges() and findDangerousChanges() exports still exist there as migration bridges; prefer findSchemaChanges().
  • graphql/utilities/assertValidName has no same-name file replacement. Use assertName from graphql or graphql/type, or from the file graphql/type/assertName if you must keep a file-level import. There is no direct replacement for isValidNameError().