code-server/docs/CONTRIBUTING.md

Contributing

• Requirements
  • Linux-specific requirements
• Creating pull requests
  • Commits and commit history
• Development workflow
  • Version updates to Code
  • Patching Code
  • Build
    • Creating a Standalone Release
  • Troubleshooting
    • I see "Forbidden access" when I load code-server in the browser
    • "Can only have one anonymous define call per script"
  • Help
• Test
  • Unit tests
  • Script tests
  • Integration tests
  • End-to-end tests
• Structure
  • Modifications to Code
  • Currently Known Issues

• Detailed CI and build process docs

Requirements

The prerequisites for contributing to code-server are almost the same as those for VS Code. Here is what is needed:

• node v18.x
• git v2.x or greater
• git-lfs
• yarn
  • Used to install JS packages and run scripts
• nfpm
  • Used to build .deb and .rpm packages
• jq
  • Used to build code-server releases
• gnupg
  • All commits must be signed and verified; see GitHub's Managing commit signature verification or follow this tutorial
• quilt
  • Used to manage patches to Code
• rsync and unzip
  • Used for code-server releases
• bats
  • Used to run script unit tests

Linux-specific requirements

If you're developing code-server on Linux, make sure you have installed or install the following dependencies:

sudo apt-get install build-essential g++ libx11-dev libxkbfile-dev libsecret-1-dev libkrb5-dev python-is-python3

These are required by Code. See their Wiki for more information.

Creating pull requests

Please create a GitHub Issue that includes context for issues that you see. You can skip this if the proposed fix is minor.

In your pull requests (PR), link to the issue that the PR solves.

Please ensure that the base of your PR is the main branch.

Commits and commit history

We prefer a clean commit history. This means you should squash all fixups and fixup-type commits before asking for a review (e.g., clean up, squash, then force push). If you need help with this, feel free to leave a comment in your PR, and we'll guide you.

Development workflow

1. git clone https://github.com/coder/code-server.git - Clone code-server
2. git submodule update --init - Clone vscode submodule
3. quilt push -a - Apply patches to the vscode submodule.
4. yarn - Install dependencies
5. yarn watch - Launch code-server localhost:8080. code-server will be live reloaded when changes are made; the browser needs to be refreshed manually.

When pulling down changes that include modifications to the patches you will need to apply them with quilt. If you pull down changes that update the vscode submodule you will need to run git submodule update --init and re-apply the patches.

Version updates to Code

1. Update the lib/vscode submodule to the desired upstream version branch.
   1. cd lib/vscode && git checkout release/1.66 && cd ../..
   2. git add lib && git commit -m "chore: update Code"
2. Apply the patches (quilt push -a) or restore your stashed changes. At this stage you may need to resolve conflicts. For example use quilt push -f, manually apply the rejected portions, then quilt refresh.
3. From the code-server project root, run yarn install.
4. Test code-server locally to make sure everything works.
5. Check the Node.js version that's used by Electron (which is shipped with VS Code. If necessary, update your version of Node.js to match.
6. Commit the updated submodule and patches to code-server.
7. Open a PR.

Tip: if you're certain all patches are applied correctly and you simply need to refresh, you can use this trick:

while quilt push; do quilt refresh; done

Source

Patching Code

0. You can go through the patch stack with quilt push and quilt pop.
1. Create a new patch (quilt new {name}.diff) or use an existing patch.
2. Add the file(s) you are patching (quilt add [-P patch] {file}). A file must be added before you make changes to it.
3. Make your changes. Patches do not need to be independent of each other but each patch must result in a working code-server without any broken in-between states otherwise they are difficult to test and modify.
4. Add your changes to the patch (quilt refresh)
5. Add a comment in the patch about the reason for the patch and how to reproduce the behavior it fixes or adds. Every patch should have an e2e test as well.

Build

You can build as follows:

yarn build
yarn build:vscode
yarn release

NOTE: this does not keep node_modules. If you want them to be kept, use KEEP_MODULES=1 yarn release (if you're testing in Coder, you'll want to do this)

Run your build:

cd release
npm install --omit=dev # Skip if you used KEEP_MODULES=1
# Runs the built JavaScript with Node.
node .

Build the release packages (make sure that you run yarn release first):

yarn release:standalone
yarn test:integration
yarn package

On Linux, the currently running distro will become the minimum supported version. In our GitHub Actions CI, we use CentOS 7 for maximum compatibility. If you need your builds to support older distros, run the build commands inside a Docker container with all the build requirements installed.

Creating a Standalone Release

Part of the build process involves creating standalone releases. At the time of writing, we do this for the following platforms/architectures:

• Linux amd64 (.tar.gz, .deb, and .rpm)
• Linux arm64 (.tar.gz, .deb, and .rpm)
• Linux arm7l (.tar.gz)
• Linux armhf.deb
• Linux armhf.rpm
• macOS amd64 (Intel-based)

Currently, these are compiled in CI using the yarn release-standalone command in the release.yaml workflow. We then upload them to the draft release and distribute via GitHub Releases.

Troubleshooting

I see "Forbidden access" when I load code-server in the browser

This means your patches didn't apply correctly. We have a patch to remove the auth from vanilla Code because we use our own.

Try popping off the patches with quilt pop -a and reapplying with quilt push -a.

"Can only have one anonymous define call per script"

Code might be trying to use a dev or prod HTML in the wrong context. You can try re-running code-server and setting VSCODE_DEV=1.

Help

If you get stuck or need help, you can always start a new GitHub Discussion here. One of the maintainers will respond and help you out.

Test

There are four kinds of tests in code-server:

1. Unit tests
2. Script tests
3. Integration tests
4. End-to-end tests

Unit tests

Our unit tests are written in TypeScript and run using Jest, the testing framework].

These live under test/unit.

We use unit tests for functions and things that can be tested in isolation. The file structure is modeled closely after /src so it's easy for people to know where test files should live.

Script tests

Our script tests are written in bash and run using bats.

These tests live under test/scripts.

We use these to test anything related to our scripts (most of which live under ci).

Integration tests

These are a work in progress. We build code-server and run tests with yarn test:integration, which ensures that code-server builds work on their respective platforms.

Our integration tests look at components that rely on one another. For example, testing the CLI requires us to build and package code-server.

End-to-end tests

The end-to-end (e2e) tests are written in TypeScript and run using Playwright.

These live under test/e2e.

Before the e2e tests run, we run globalSetup, which eliminates the need to log in before each test by preserving the authentication state.

Take a look at codeServer.test.ts to see how you would use it (see test.use).

We also have a model where you can create helpers to use within tests. See models/CodeServer.ts for an example.

Generally speaking, e2e means testing code-server while running in the browser and interacting with it in a way that's similar to how a user would interact with it. When running these tests with yarn test:e2e, you must have code-server running locally. In CI, this is taken care of for you.

Structure

The code-server script serves as an HTTP API for login and starting a remote Code process.

The CLI code is in src/node and the HTTP routes are implemented in src/node/routes.

Most of the meaty parts are in the Code portion of the codebase under lib/vscode, which we describe next.

Modifications to Code

Our modifications to Code can be found in the patches directory. We pull in Code as a submodule pointing to an upstream release branch.

In v1 of code-server, we had Code as a submodule and used a single massive patch that split the codebase into a front-end and a server. The front-end consisted of the UI code, while the server ran the extensions and exposed an API to the front-end for file access and all UI needs.

Over time, Microsoft added support to Code to run it on the web. They had made the front-end open source, but not the server. As such, code-server v2 (and later) uses the Code front-end and implements the server. We did this by using a Git subtree to fork and modify Code.

Microsoft eventually made the server open source and we were able to reduce our changes significantly. Some time later we moved back to a submodule and patches (managed by quilt this time instead of the mega-patch).

As the web portion of Code continues to mature, we'll be able to shrink and possibly eliminate our patches. In the meantime, upgrading the Code version requires us to ensure that our changes are still applied correctly and work as intended. In the future, we'd like to run Code unit tests against our builds to ensure that features work as expected.

We have extension docs on the CI and build system.

If the functionality you're working on does NOT depend on code from Code, please move it out and into code-server.

Currently Known Issues

• Creating custom Code extensions and debugging them doesn't work
• Extension profiling and tips are currently disabled