GraphQL Code Generator v5 Roadmap

[theguild-bot]

This page is synced automatically from The Guild's Notion
Notion page URL: https://www.notion.so/GraphQL-Code-Generator-v4-Roadmap-91923bfb2dee48eaa0d6a77666429968

At The Guild, we’ve decided to work as much as possible in public; that’s why we are opening the roadmaps for all of our projects.

The goals for this are:

1. So you will know what we are working on, what we see as a higher priority, and know what to expect from our projects in the future
2. So you can share your opinions and thoughts about what we do and influence our decisions
3. So you can join us and contribute to our efforts!

---

Before laying down the roadmap of GraphQL Code Generator v3, we would like to thank all of you for being so many who use codegen daily and for contributing to making it such a complete project! 🚀

While some people judge that GraphQL is difficult, GraphQL Code Generator v3 aims to change that perspective by providing a unified configuration along with a smaller and simpler generated code.

By providing a unified package and configuration for all client-side use cases, all existing and future plugin alternatives will be moved to community repos.

Let’s now cover these changes in detail.

A unified configuration and package for all GraphQL clients

Most of the existing client-side plugins (typescript-react-apollo, typescript-react-query, etc) rely on the generation of hooks or SDKs that wrap the underlying GraphQL Client in a type-safe way.

However, the generation of hooks or SDK code brings many downsides:

• an unnecessary increase of the final bundle size
• misalignment between the generated hooks signature and the underlying GraphQL Client
• inconsistencies of configuration options and preset compatibility across packages (ex: near-operation-file compatibility)

To make GraphQL code generation great and simple, the v3 version will introduce two major changes:

• a new unique preset for all GraphQL clients, which include better developer experience, smaller bundle size, stronger typings, and easier-to-follow best practices
• a TypeScript-first configuration file that will allow configuration autocompletion

Here is how you can already configure codegen for all GraphQL Clients:

import { CodegenConfig } from '@graphql-codegen/cli'

const config: CodegenConfig = {
  schema: 'http://localhost:4000/graphql',
  documents: ['src/**/*.tsx'],
  generates: {
    './src/gql/': {
      preset: 'client',
      plugins: []
    }
  }
}

export default config

The client preset comes with a simple opinionated configuration and a lightweight types-only generation.

To try the new client preset, please install the following dependencies:

yarn add graphql
yarn add -D typescript
yarn add -D @graphql-codegen/cli
yarn add -D @graphql-codegen/client-preset

First, start GraphQL Code Generator in watch mode:

yarn graphql-codegen --watch

Using GraphQL Code Generator will type your GraphQL Query and Mutations as you write them ⚡️

Now, each query or mutation written with the generated graphql() function will be automatically typed!

For example, with Apollo Client (React):

import React from 'react';
import { useQuery } from '@apollo/client';
import { graphql } from './gql/gql';

import Film from './Film';

// here, `allFilmsWithVariablesQueryDocument` is fully typed!
const allFilmsWithVariablesQueryDocument = graphql(/* GraphQL */ `
  query allFilmsWithVariablesQuery($first: Int!) {
    allFilms(first: $first) {
      edges {
        node {
          ...FilmItem
        }
      }
    }
  }
`);

function App() {
	// Most GraphQL Clients know how to deal with typed GraphQL documents,
	//   providing typed data and typed variables
  const { data } = useQuery(allFilmsWithVariablesQueryDocument, { variables: { first: 10 } });
  return (
    <div className="App">
      {data && <ul>{data.allFilms?.edges?.map((e, i) => e?.node && <Film film={e?.node} key={`film-${i}`} />)}</ul>}
    </div>
  );
}

export default App;

Thanks to work made to integrate TypeDocumentNode (the underlying plugin used by preset: client) with most of the popular GraphQL clients, you no longer need hooks or SDK, simple GraphQL documents works!

We believe that the preset: client approach is the way to get the best of TypeScript and GraphQL by:

• reducing the size of the generated bundle
• only the graphql() function needs to be imported (no type, hooks, document imports)
• removing layers between your application and your chosen GraphQL Client
• providing stronger typings that will stay aligned with your chosen GraphQL Client
• offering you the best component isolation design by leveraging Fragment Masking

Finally, this new preset: client has been properly tested on all popular GraphQL clients across most frameworks:

• React
  • @apollo/client (since 3.2.0, not when using React Components (<Query>))
  • @urql/core (since 1.15.0)
  • @urql/preact (since 1.4.0)
  • urql (since 1.11.0)
  • graphql-request (since 5.0.0)
  • react-query (with graphql-request@5.0.0)
  • swr (with graphql-request@5.0.0)
  • @urql/exchange-graphcache (since 3.1.11)
• Svelte
  • @urql/svelte (since 1.1.3)
• Vue
  • @vue/apollo-composable (since 4.0.0-alpha.13)
  • villus (since 1.0.0-beta.8)
  • @urql/vue (since 1.11.0)
• Others
  • graphql-js (since 15.2.0)
  • graphql-request (since 5.0.0)

You will find demos and code examples for each of them in the examples/front-end/ folder of the codegen repository.

You will also find a complete guide for React and Vue in codegen documentation.

We aim for GraphQL Code Generator 3.0’s client preset to become the official way to generate GraphQL Types for front-end use cases, replacing all existing hook and SDK-based plugins.

For this reason, we encourage you to already give a try at the codegen v3 client preset (@graphql-codegen/client-presec) and provide feedback on this issue.

The v3 stable release will be shipped once sufficient feedback is posted.

Finally, while the GraphQL Code Generator 3.0 milestone aims to provide a unified front-end experience through the preset: client, the 3.x versions aim to fully rewrite the core packages of codegen.

Some core parts of codegen are more than 6 years old and need to be rewritten (optimized, simplified, and more).

We plan to incorporate the pending issues related to the core packages in this gradual 3.x milestones.

Introduction of the “community plugins”

Historically, all plugins were pushed to the https://github.com/dotansimha/graphql-code-generator repository, making it hard for us to review all contributions in a reasonable timeframe and to enforce consistency across all the options introduced in the different packages.

We believe that the best way to keep codegen extensible and improve the contribution experience at scale is to introduce the concept of community plugins.

A community plugin offers a feature-set that diverges from the preset: client or a plugin created by the community.

Soon, all the existing plugins part of the list below and all the future plugins created by the community will live in their dedicated repository:

• @graphql-codegen/typescript-react-apollo
• @graphql-codegen/typescript-graphql-request
• @graphql-codegen/typescript-apollo-angular
• @graphql-codegen/typescript-apollo-client-helpers
• @graphql-codegen/typescript-react-query
• @graphql-codegen/typescript-urql
• @graphql-codegen/named-operations-object
• @graphql-codegen/urql-introspection
• @graphql-codegen/flow-resolvers
• @graphql-codegen/typescript-vue-apollo
• @graphql-codegen/typescript-rtk-query
• @graphql-codegen/flow-operations
• @graphql-codegen/typescript-msw
• @graphql-codegen/typescript-mongodb
• @graphql-codegen/typescript-type-graphql
• @graphql-codegen/jsdoc
• @graphql-codegen/typescript-vue-urql
• @graphql-codegen/kotlin
• @graphql-codegen/typescript-vue-apollo-smart-ops
• @graphql-codegen/java
• @graphql-codegen/c-sharp-operations
• @graphql-codegen/hasura-allow-list
• @graphql-codegen/typescript-stencil-apollo
• @graphql-codegen/relay-operation-optimizer
• @graphql-codegen/typescript-oclif
• @graphql-codegen/java-resolvers
• @graphql-codegen/java-apollo-android

All the above plugins will be eligible for repository ownership transfer based on relevant past contributions.

Of course, such a change will come with help from our side:

• We will create a new “Create a plugin” guide that will provide complete information and guidelines (ex: publishing, codegen APIs, adding your plugin to the codegen hub)
• Since each community plugin will live in its own repository, we will provide a proper Github repository template with building and publishing CI tools configured.

What about server-side plugins?

The 3.x milestones include some work on server-side plugins such as typescript-resolvers (ex: improving Federation support).

---

Milestones

Below are the details of the aforementioned plans for the 3.0 and 3.x milestones.

3.0

• client preset
  • Integrate with most of the GraphQL clients
    • graphql-request https://github.com/prisma-labs/graphql-request/pull/350/files
    • https://github.com/dotansimha/graphql-code-generator/issues/8061
    • Improve TypeScript support in libraries that already support TypedDocumentNode (variables should be required if there are required variables in the operation declaration and optional if there are none)
      • URQL
      • Apollo Client
    • React Query (works with graphql-request)
    • SWR (works with graphql-request)
    • Add support for AWS AppSync client
  • Fix pending issues on gql-tag-operation-preset
    • gql-tag-operations-preset babelPlugin doesn't work with vite #8206
  • canary release: GraphQL Code Generator v3: RFC umbrella PR #8248
    • TypeScript config support
    • documents preset preconfiguration
    • opinionated plugins configuration
    • deprecate gql-tag-operations-preset in favor of the client-preset
  • examples: How to use codegen in front-end applications, code examples using preset: front-end #8184
  • Better support for graphql-scalars #8413
  • Update documentation for front-end/GraphQL clients parts
    • Update graphql-request docs on their repo
      • chore(docs): update documentation on GraphQL Code Generator integration graffle-js/graffle#392
    • Update Apollo Client docs
      • doc(static-typing): rewrite with GraphQL Code Generator setup apollographql/apollo-client#10173
    • Update URQL docs
      • doc: add section for TypeScript types generation with GraphQL Code Generator urql-graphql/urql#2729
    • Update apollo-angular docs
      • → not enough time
    • Update SWR GraphQL documentation with a link to codegen doc
      • doc(data-fetching): add mention of GraphQL Code Generator for end-to-end type safety vercel/swr-site#359
  • official release 🚢
  • better TypeScript/VSCode integration
    • Potential solution: gql-tag-operations-preset and vscode typescript #8345 (comment)
  • create a Github template repository
  • new documentation for creating a plugin
    • + updated contributing guidelines
  • move all community plugins to a dedicated repository
    • move the related issues + communicate about new contribution guidelines
    • Reach out to potential people who want to support

3.x

preset: client improvements

• Only generate actually used types
• Support operation minification/compilation? (aka embed/re-invent relay-compiler) https://relay-compiler-repl.netlify.app/
  • Support (parameterized) Fragment Arguments? Parameterized fragments graphql/graphql-spec#204
• “Client Controlled Nullability” support
  → feat: override operation field types via directives #8071
• Create a TypedDocumentNode string alternative (TypedString) that does not require GraphQL AST on the Client (should be easily configurable within the preset)
• https://tortilla-hq.slack.com/archives/CCVDM3NCD/p1659508122835199
• Support @defer and @stream in operations plugin #7885
• Refactor preset: client to not use the old plugins (get rid of actual core packages)

Future of codegen CLI

• Better TypeScript config support
  • get typed options (config) → WIP 🚧
• Remove usage of DetailedError which is not well-handled by listr #8200
• Allow generating outputs even if there are no documents (so you can still generate the gql function for getting started)
• Better configuration
• Performance
  • Smarter watch #5642
  • Incremental builds (Incremental rebuilds #6142)
• Remove graphql dependency
• Better monorepo support?
  • multi-project, etc.
  • shareable configs? → (one codegen file versus many)

Back-end code generation issues

We will go over the following typescript-resolvers and graphql-modules pending plugins issues:

• [TypeScript Resolvers] feat: Generate Interface types without field resolvers #2194
• Federation Resolvers: Pick<ParentType, 'fieldName'> is incorrect for mapped parent types. #3207
• [typescript-resolvers] Enums ValueMaps don't get generated anymore #3619
• [typescript-resolvers] avoid input optional as typescript-operations #3815
• enumPrefix not working properly with typescript-resolvers #3873
• Resolve input types with preResolveTypes same as selection sets #3958
• [typescript-resolver] interface resolver should could use type mappers  #4121
• When federation is enabled for a type, its resolvers have a wrong ParentType #4722
• Multiple @key directives not supported when type is extended #4739
• Avoid optionals on fields influences variables #4788
• https://github.com/dotansimha/graphql-code-generator/issues/4901
• [typescript-resolvers] Allow passing parent types to Resolvers #5517
• Error "Cannot find name 'User'" in typescript-resolvers output #5841
• FieldWrapper not applied to fields in union resolver type #5776
• Possibility of passing an expression for the mappers config #5594
• Federation with custom mappers #5646
• Mapped types are dropped when combined with unions at non-trivial depth. #5998
• [typescript-resolvers] Missings args types in QueryResolvers #5968
• Types for input objects should not have a prototype #5964
• [typescript-resolvers] Map custom context type on a type-level #6483
• Subscription resolver args are of type any #6482
• TypeScript Resolvers: Easier way to get possible resolve types for interfaces #6443
• TypeScript Resolvers plugin doesn't allow null to be returned from a promise #6173
• [typescript-resolvers] Resolvers' default result type should allow async and resolvable object values #7358
• Allow typescript-resolvers plugin to export resolver function alias type #7373
• Directive resolver input arguments type ignores enumPrefix = false #7560
• [experimental-graphql-modules] no config in generation for enums #5242
• Support for Args and related module types in GraphQL Module Resolvers #5139
• graphql-modules-preset doesn't obey enumsAsTypes and doesn't generate per-module Resolvers types for enum internal values #6023
• Watch generates different files from no watch (graphql-modules) #6851
• Watch generates different files from no watch (graphql-modules) #6851
• graphql-modules-preset is not adding interface and union resolvers to Module.Resolver #7123
• graphql-modules-preset is not adding interface and union resolvers to Module.Resolver #7123
• [graphql-modules-preset] Import declarations in a namespace cannot reference a module #7671

[huv1k]

Will there be support for module augmentation still? 🤔 I really enjoy this feature, but there are some shortcomings like some frameworks don't provide for example useFragment etc.

[charlypoly]

Will there be support for module augmentation still? 🤔 I really enjoy this feature, but there are some shortcomings like some frameworks don't provide, for example, useFragment etc.

Hi @huv1k!
We are planning to remove the support for module augmentation since it is not working without the associated babel plugin properly set up.
Since not all front-end setups can provide custom babel configuration, we were planning to remove the augmentedModuleName options and also to simplify the installation flow.

However, this being an RFC, we will take your feedback into account! 👀

[franky47]

Based on your code example, does that mean that the preferred way to consume GraphQL operations is co-location within TS/JS files?

I just migrated a large codebase from Apollo CLI to graphql-codegen v2 and refactored it to move all GraphQL operation definitions to .graphql files (importing only typed documents to pass to Apollo Client), I wouldn't want to have to undo that all over again..

[charlypoly]

Based on your code example, does that mean that the preferred way to consume GraphQL operations is co-location within TS/JS files?

I just migrated a large codebase from Apollo CLI to graphql-codegen v2 and refactored it to move all GraphQL operation definitions to .graphql files (importing only typed documents to pass to Apollo Client), I wouldn't want to have to undo that all over again..

Hi @franky47,

We are exploring different migration path options; providing a codemod could be one; however, we have to look at the different setups to migrate from.
What plugins are you currently using?

[franky47]

Hi @charlypoly, here's my codegen.yml:

overwrite: true
schema: "../backend/schema.graphql"
documents: "src/**/*.graphql"
config: # The following configuration will adjust output to the defaults of Apollo-Codegen, and should be compatible with most use-cases.
  preResolveTypes: true # Simplifies the generated types
  namingConvention: keep # Keeps naming as-is
  avoidOptionals: # Avoids optionals on the level of the field
    field: true
  nonOptionalTypename: true # Forces `__typename` on all selection sets
  skipTypeNameForRoot: true # Don't generate __typename for root types
generates:
  src/modules/graphql.types.ts:
    plugins:
      - typescript
      - typescript-operations
      - typed-document-node
    hooks:
      afterOneFileWrite:
        - prettier --write

I have GraphQL code into .graphql files, which are parsed by the generator to give me the full schema types (I use them to type props in a React/Next.js app), and typed schemas to pass to Apollo Client, which provides type inference (variables & output), so that TypeScript has my back when matching the data & view layers.

I migrated from a setup where the Apollo CLI generated the TS types from gql tags co-located with their components in .ts(x) files, hence my remark that having to go back would be quite annoying.

I really like the typed document plugin, it removed the need to explicitly pass generic arguments to useQuery and useMutation, so most of my query/mutation hooks are now one-liners. The only thing that required a bit of refactoring to use those was that their naming convention does not seem to be configurable, so I had to rename all FOO_{QUERY|MUTATION} to FooDocument, not a big deal but the name can be confusing sometimes (especially when dealing with domain objects also named "Document").

Hope this helps you see a bit better about my particular use-case.

[charlypoly]

@franky47, thank you for sharing it!

We are also exploring options to still support .graphql files, we will keep you updated here.

[nandita1]

I think there are some problems with this:

1. The generation time has gone way up. It has doubled up infact.
2. And I see that there have been duplicate types being generated. The size of the file(s) have also gone way up.

[productdevbook]

vue 3 and vue graphql v4 problem

but it works when these lines are added

[ilijaNL]

If it is going full TypeDocumentNode way (which I like), is it not better to go with a builder pattern approach. The builder can be full type safe and be generated from the introspection query. Additionally you need to generate the builder only initially and on schema change and not when adding new documents. Some open source projects which apply this principle:

• https://github.com/typed-graphql-builder/typed-graphql-builder (my favorite)
• https://zeus.graphqleditor.com/page/plugins/typedDocumentNode.html
• https://github.com/timkendall/tql

In my opinion this provides best of the worlds. Small bundle size, no watchers necessary on frontend and much faster dev cycle.
Additionally the builder can be scoped per context (think about hasura roles or public/private introspection results)

[hrougier]

Will there be support for module augmentation still? 🤔 I really enjoy this feature, but there are some shortcomings like some frameworks don't provide, for example, useFragment etc.

Hi @huv1k! We are planning to remove the support for module augmentation since it is not working without the associated babel plugin properly set up. Since not all front-end setups can provide custom babel configuration, we were planning to remove the augmentedModuleName options and also to simplify the installation flow.

However, this being an RFC, we will take your feedback into account! 👀

Augmented module is a nice feature for people who don't like having their own source code polluted by/relying on generated code, why not keeping it for those willing to make advanced configuration? (babel plugin is just a one line configuration).

Inspired by what graphql-let does, I personally came up with some kind of hack that generates everything to a fake node_modules/@types of a monorepo:

schema: path/to/schema.json
documents:
  - src/**/*.ts
  - src/**/*.tsx
generates:
  # app operations are generated inside a fake @types module
  ../../node_modules/@types/myProject__graphql-client:
    preset: gql-tag-operations-preset
    presetConfig:
      augmentedModuleName: '@myProject/graphql-client'
      fragmentMasking:
        augmentedModuleName: '@myProject/graphql-client'
hooks:
  # here is the ugliest part of the "hack": generating a package.json
  afterAllFileWrite:
    - "printf '{ \"name\": \"@types/myProject__graphql-client\", \"types\": \"gql.d.ts\" }' > '../../node_modules/@types/myProject__graphql-client/package.json'"

This allows for everything within an app to be transparently imported from a @myProject/graphql-client (a preconfigured apollo-client module in the monorepo) without ever seeing/dealing with any ugly generated code 😅

Big advantage: I can remove codegen at any time without breaking the code, only losing types!

import { gql, useQuery } from '@myProject/graphql-client'

const MyQuery = gql(`
  query MyQuery {
    ...
  }
`)

const MyComponent = () => {
  const { loading, data, error } = useQuery(MyQuery)

  return <>...</>
}

It would be nice to keep this ability to hide the generated code somewhere.

[grischaerbe]

I really like the new approach with preset: client. It seems like a big step towards a simpler toolchain with less dependencies. I guess currently the configuration for a custom scalar types is not considered:

config:

overwrite: true
schema: 'http://localhost:8080/v1/graphql'
documents:
  - './src/**/*.graphql'
  - './src/**/*.svelte'
generates:
  src/lib/graphql/new/:
    preset: client
    config:
      strictScalars: true
      scalars:
        numeric: number
        jsonb: Array<any>
        bigint: number
        uuid: string
        timestamptz: string
        timestamp: string

generated scalars:

export type Scalars = {
  ID: string;
  String: string;
  Boolean: boolean;
  Int: number;
  Float: number;
  bigint: any;
  jsonb: any;
  numeric: any;
  timestamp: any;
  timestamptz: any;
  uuid: any;
};