Apollo Client 4.3

Orbit supports Apollo Client 4.3's custom scalars, cache and fragment-reference type overrides, and incremental result types. These features are opt-in.

Custom scalars

Follow Apollo's custom scalar guide for declarations, cache configuration, lists, nested inputs, and reusing GraphQLScalarType implementations.

Pass the configured cache through withApolloOptions. Orbit's observable and signal APIs expose parsed values, including with fetchPolicy: 'no-cache'. Variables can contain parsed values such as Date; Apollo serializes them for the network. Cache snapshots are serialized too.

Scalar field policies can also live in Orbit's state(...).typePolicies(...). A field's scalar policy cannot be combined with read or merge.

Map the scalar to its parsed type in your GraphQL Codegen configuration:

config: {
  scalars: { DateTime: { input: 'Date', output: 'Date' } }
}

Custom cache and fragment references

Use Apollo's cache type declaration to retain your cache's methods in apollo.cache, mutation updates, state initializers, and actions. Provide the declared implementation at runtime. Angular keeps Orbit's cache.watchQuery extension alongside those methods.

The same override applies to React state callbacks.

watchFragment and signal.fragment honor Apollo's FromOptionValue override for stricter identifiers. Orbit also accepts nullable and array from values; each non-null value must satisfy the override.

Incremental delivery

For operation types generated with deferred-field unions, enable Apollo's new type overrides:

import '@apollo/client';
import type { GraphQLCodegenIncremental } from '@apollo/client/incremental';

declare module '@apollo/client' {
  interface TypeOverrides extends GraphQLCodegenIncremental.TypeOverrides {}
}

Narrowing result().dataState to 'complete' then exposes all deferred fields. Complete cache results use the same assembled type. Follow Apollo's incremental delivery setup and set incrementalHandler through withApolloOptions; the type override alone does not enable delivery.

dataState describes the available fields, while loading and networkStatus describe the request. An @stream result can have dataState: 'complete' while networkStatus is NetworkStatus.streaming. Use loading() to track pending work. Query execution promises wait for the final chunk.