chore: Move to yarn 4 and workspaces by ibgreen-openai · Pull Request #10321 · visgl/deck.gl

ibgreen-openai · 2026-05-20T16:17:25Z

Summary

Migrates deck.gl to Yarn 4.15 workspaces with nodeLinker: node-modules, and installs all package-managed examples from the repo root.

Motivations

yarn 1 is ancient, time to move on, install and CI times will be meaningfully improved.
the fact that the examples do not have lock files is what causes problems in corporate environments that prevent recent high risk packages from being installed.
Having a common lock file through the workspace solves this elegantly
the development experience is much improved, a quick install and any example is ready to go. We can add website and test apps next.
We have moved to yarn 4 and workspaces in most of the other repos (luma, loaders, deck-community, ...) so this is aligned with that.

Changes

Set root packageManager to yarn@4.15.0.
Added .yarnrc.yml with:
- enableScripts: true
- nodeLinker: node-modules
- targeted Yarn peer-warning filters for known legacy/third-party peer noise.
Expanded root workspaces from modules/* to:
- modules/*
- examples/**
Converted internal deck.gl / @deck.gl/* example deps to workspace:*.
Added/fixed example workspace metadata and duplicate names.
Kept Lerna publish/build scope limited to modules/*.
Updated docs and example READMEs to use one root yarn install.
Regenerated yarn.lock with Yarn 4.15.
Updated stale internal @deck.gl/* peer ranges from ~9.3.0-alpha.1 to ~9.3.0-beta.2.

Verification

corepack yarn --version -> 4.15.0
corepack yarn install --mode=skip-build
corepack yarn install --immutable --mode=skip-build
corepack yarn workspaces list --json
- 81 total workspaces
- 64 examples
- 16 modules
- 0 duplicate names
corepack yarn test-fast
Representative example builds:
- deckgl-example-react-basic
- deckgl-example-pure-js-basic
- deckgl-examples-scatterplot
- deckgl-examples-gallery

Notes

Plain yarn install runs the existing root postinstall script (npx playwright install chromium). Lockfile verification used --mode=skip-build to avoid browser-install side effects during validation.

Note

Medium Risk
Touches all install/CI/release paths and dependency resolution across modules and many examples; incorrect workspace or resolution pins could break builds or publishing, though runtime library changes are minimal (one S2 helper return type).

Overview
Upgrades the monorepo from Yarn 1 to Yarn 4.15 with node-modules linking and a root packageManager pin, adding .yarnrc.yml (scripts enabled, peer-warning filters) and ignoring .yarn/ in git.

Workspaces now include examples/** alongside modules/*, so examples install from one root lockfile; bootstrap no longer runs yarn itself, and CI/release/website flows use corepack enable, yarn install --immutable, then yarn bootstrap (dropping setup-node’s yarn cache). The website keeps its own Yarn 4 config and uses immutable installs in test-website / deploy steps; gallery build scripts rely on the root install instead of per-folder yarn.

Root resolutions pin all @deck.gl/* / deck.gl to 9.3.0-beta.2; example package.json files get small metadata/order fixes. getIdFromToken in s2-utils returns a string id (via Long.toString()) instead of a Long object. Minor tooling tweaks: Docusaurus source-map rule excludes node_modules, jupyter-widget build script split, and a mapbox default-import eslint comment in a get-started example.

^{Reviewed by Cursor Bugbot for commit 258145b. Bugbot is set up for automated code reviews on this repo. Configure here.}

coveralls · 2026-05-20T16:50:42Z

coverage: 83.374% (+0.008%) from 83.366% — ib/yarn-4-workspace into master

chrisgervang

I'm not a fan of how fragile yarn tends to be - workspaces and compatibility issues with PnP can be frustrating to users. Also, it's currently easy to copy/paste an example into your own project to adopt deck, or codesandbox to reproduce a bug. I don't want to loss those adoption channels with the added complexity

What if we used npm 7? It has many of the same benefits as yarn, it's actively maintained (yarn v1 is quite old), and is very easy for users adopting deck to work with

chrisgervang · 2026-05-22T19:42:42Z

  // pad token with zeros to make the length 16
  const paddedToken = token.padEnd(16, '0');
-  return Long.fromString(paddedToken, 16);
+  return Long.fromString(paddedToken, 16).toString();


Let's split this out

I understand that having seemingly unrelated changes in a PR is annoying but this is related to typescript version being bumped as part of satisfying all the package.json.
To split this out I would need to make a separate PR to bump typescript and fix issue or downgrade typescript in other places.

I just want to make sure we're thinking through the implications and covering bug fixes with tests. I'm only focusing on testing the build system in this PR

chrisgervang · 2026-05-22T19:44:32Z


 ### Usage

-Copy the content of this folder to your project. 


The intent @Pessimistress shared with me years ago was that example folders are meant to be standalone so that they're easy to bring into you own projects.

Workspaces adds a layer of complexity for adopters, and I'm not convinced its the right direction to go

We definitely don't have to use workspace version specifiers in the examples I get that might be a separate discussion point.

That said, the version specifiers in the examples are often out of date which doesn't exactly help user, and during alpha/beta release cycles it gets worse.

ibgreen-openai · 2026-05-22T20:57:43Z

I'm not a fan of how fragile yarn tends to be - workspaces and compatibility issues with PnP can be frustrating to users.

FWIW, this branch is not using PnP. Both root and website are configured with nodeLinker: node-modules, so Yarn writes a normal node_modules install. I’m removing the .pnp.* ignore entry.

I personally consider workspace installs a real "luxury", have personally not had issues with it in our other repos. Just fast reliable installs across the repo and CI.

Also, it's currently easy to copy/paste an example into your own project to adopt deck, or codesandbox to reproduce a bug. I don't want to loss those adoption channels with the added complexity

Agreed, I am removing the "workspace" version specifiers, that was a step too far.

What if we used npm 7? It has many of the same benefits as yarn, it's actively maintained (yarn v1 is quite old), and is very easy for users adopting deck to work with

I have no experience with recent npm but understand that it has improved a lot. Most projects I am familiar with seem to be going pnpm these days though.

I am not aware of anything we would gain by switching that is meaningully better than yarn 4 + workspaces, but if someone thinks it is worth it and wants to put in the work, I am not against it.

We do use yarn 4 + workspace in our other repos so that does make the experience a little more consistent for a multi-repo developer like myself.

ibgreen-openai · 2026-05-22T21:18:59Z

@chrisgervang Appreciate the comments! Addressed the bigger issues, and cleaned up the PR. File churn now down from 158 to 36 files!

chrisgervang

Thanks for addressing my feedback. I think it's important to keep the adoption simple, so I appreciate not using the workspace version specifiers.

Approving to unblock but I would prefer to split out the bug fix found by a typescript upgrade so its documented well and covered, and split out module beta upgrades

chrisgervang · 2026-05-22T21:09:28Z

  "scripts": {
-    "watch": "npm run build -- --watch",
-    "build": "ocular-bundle ./src/index.js --output=./dist/index.js --globalName= --sourcemap --format=umd --externals=@jupyter-widgets/base",
+    "watch": "npm run build-bundle -- --watch",
+    "build": "npm run build-bundle",
+    "build-bundle": "ocular-bundle ./src/index.js --output=./dist/index.js --globalName= --sourcemap --format=umd --externals=@jupyter-widgets/base",


What does this change?

chrisgervang · 2026-05-22T21:28:02Z

  // pad token with zeros to make the length 16
  const paddedToken = token.padEnd(16, '0');
-  return Long.fromString(paddedToken, 16);
+  return Long.fromString(paddedToken, 16).toString();


I just want to make sure we're thinking through the implications and covering bug fixes with tests. I'm only focusing on testing the build system in this PR

chrisgervang · 2026-05-22T21:33:29Z

        uses: actions/setup-node@53b83947a5a98c8d113130e565377fae1a50d02f # v6.3.0
        with:
          node-version-file: '.nvmrc'
-          cache: 'yarn'


I'm looking into if this increases build times

chrisgervang · 2026-05-22T21:44:06Z

Could you take a pass on the PR description before merging for anything that is out of date?

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

chrisgervang · 2026-05-22T22:59:06Z

I've asked the LLM to review installation times and this is what it found:

Job	Install step	This PR	Other PRs	Regression
test-node	Install deps	~92s	~36s	~2.5x slower
test-website	Install deps	~91s	~36s	~2.5x slower
test-python	Install deps	~162s	~101s	~1.6x slower

Root cause: The cache: 'yarn' option was removed from actions/setup-node. Other PRs benefit from cached dependencies; without it, every run does a full network fetch.

Options to fix:

Re-add cache: 'yarn' to actions/setup-node — it should auto-detect Yarn 4's cache path via yarn config get cacheFolder
Add an explicit actions/cache step targeting .yarn/cache

The regression adds ~2-3 minutes of extra wall-clock time per CI run.

Resolved yarn.lock conflicts by regenerating lock files. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

cursor

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 258145b. Configure here.}

cursor · 2026-05-25T02:10:44Z

        uses: actions/setup-node@53b83947a5a98c8d113130e565377fae1a50d02f # v6.3.0
        with:
          node-version-file: '.nvmrc'
-          cache: 'yarn'


CI cache removed without replacement, causing slower builds

Medium Severity

The cache: 'yarn' option was removed from actions/setup-node across all CI workflows (test-node, test-website, test-python, release, website deploy) without adding any replacement caching mechanism such as actions/cache. PR discussion confirms this causes ~2.5x slower dependency install times on every CI run. While Yarn 4 may need different cache path configuration, the caching capability can be restored using actions/cache with the correct Yarn 4 global cache directory.

Additional Locations (2)

.github/workflows/release.yml#L47-L50

.github/workflows/website.yml#L35-L38

^{Reviewed by Cursor Bugbot for commit 258145b. Configure here.}

chore Move to yarn 4 and workspaces

7e98ba8

ibgreen-openai changed the title ~~chore Move to yarn 4 and workspaces~~ chore: Move to yarn 4 and workspaces May 20, 2026

ibgreen added 2 commits May 20, 2026 12:21

corepack

59276c4

fix

86232ca

website-yarn-4

ae5719c

ibgreen-openai marked this pull request as ready for review May 21, 2026 12:57

ibgreen-openai requested review from Pessimistress, chrisgervang and felixpalmer May 21, 2026 12:57

chrisgervang reviewed May 22, 2026

View reviewed changes

Address comments

2027e7f

chrisgervang mentioned this pull request May 22, 2026

chore: Upgrade module peer dependencies to ~9.3.0-beta.2 #10325

Closed

chrisgervang approved these changes May 22, 2026

View reviewed changes

chrisgervang and others added 4 commits May 22, 2026 15:15

Merge branch 'master' into ib/yarn-4-workspace

529e5a2

clean up after rebase

c0c1e63

fix

9555cb1

chore(examples): sort package.json dependencies alphabetically

61d9e8d

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Merge branch 'master' into ib/yarn-4-workspace

258145b

Resolved yarn.lock conflicts by regenerating lock files. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

cursor Bot reviewed May 25, 2026

View reviewed changes

Conversation

ibgreen-openai commented May 20, 2026 • edited by cursor Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Motivations

Changes

Verification

Notes

Uh oh!

coveralls commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chrisgervang left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ibgreen-openai commented May 22, 2026

Uh oh!

ibgreen-openai commented May 22, 2026

Uh oh!

chrisgervang left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

chrisgervang commented May 22, 2026

Uh oh!

chrisgervang commented May 22, 2026

Uh oh!

cursor Bot left a comment

Choose a reason for hiding this comment

Uh oh!

cursor Bot May 25, 2026

Choose a reason for hiding this comment

CI cache removed without replacement, causing slower builds

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

ibgreen-openai commented May 20, 2026 •

edited by cursor Bot

Loading

coveralls commented May 20, 2026 •

edited

Loading