Skip to content

README.md

Latest commit

 

History

History
206 lines (153 loc) · 6.39 KB

README.md

File metadata and controls

206 lines (153 loc) · 6.39 KB
Raw

Chromatic

index: examples

To use a given basic-example as a template for a new myproject:

$ npx degit pmndrs/examples/examples/basic-example myproject
$ code myproject

Example metadata

Every example has a pmndrs.json file containing the catalog metadata used by the website:

{
  "$schema": "../../schemas/pmndrs.schema.json",
  "title": "Basic Example",
  "description": "Shows how to form self-contained components with their own state and user interaction.",
  "tags": ["interaction", "pointer-events"],
  "authors": [],
  "source": "https://codesandbox.io/s/rrppl0y8l4",
  "libraries": ["@react-three/fiber", "@react-three/drei"],
  "assets": []
}

Use package names for libraries; each entry must also be a dependency of the example. publishedAt is optional and uses YYYY-MM-DD when the original publication date is known. Add externally sourced models, textures, fonts, audio, and other assets to assets with their creator, source, and license when available.

Run pnpm lint:metadata to validate every metadata file.

INSTALL

Prerequisites:

  • Install nvm - Node Version Manager, then:

    $ nvm install
$ nvm use
$ node -v # make sure your version satisfies package.json#engines.node

    nb: if you want this node version to be your default nvm's one: nvm alias default node

$ pnpm install

dev

$ pnpm dev

build

$ pnpm build

NB: pnpm build --force to ignore turbo cache

Then npx serve out.

Details

This will:

  1. execute ^build2 which will vite build each examples/* with:
  • a --base set to ${BASE_PATH}/${app_name}
  • a custom vite --config, whith a monkey() plugin that will:
    • deterministic script into src/index.jsx — seeds Math.random, and gives every WebGL context preserveDrawingBuffer so the archived canvas is not blank
    • monkeypatch the <Canvas> with CheesyCanvas for setting up the scene for playwright screenshots — in whichever src/**.[jt]sx imports it, which is App for most examples and Scene, Bananas, Canvas or index for nine of them
  1. build the Next.js apps/website
  2. copy final result into out folder

[!TIP] This is totally fine BASE_PATH/BASE_URL to be unset/empty. But for debug purposes(to be 1:1 with GitHub pages) you can:

$ BASE_PATH=/examples BASE_URL=http://localhost:4000 pnpm build
$ npx serve out -p 4000

deploy

.github/workflows/ci.yml publishes main to GitHub Pages, and uploads the same out folder to Vercel for each pull request, as a preview — Vercel only hosts it, the build always happens in the CI. Needs VERCEL_TOKEN, VERCEL_ORG_ID and VERCEL_PROJECT_ID as repo secrets; a pull request from a fork cannot read them, and gets no preview.

The preview is built with the same BASE_PATH as the Pages one, so it is served from <deployment-url>/examples and not from the domain root — same layout as production, and the same build2 cache entries, which is what keeps a preview an upload rather than a rebuild of all 167 examples.

test

$ pnpm test

Every example with a test script gets loaded, held at a fixed thirty frames and archived. The test asserts that the shot completed — an example that throws, or never mounts its <Canvas>, fails here. What the picture looks like is Chromatic's question, below; there are no baseline PNGs in this repo.

Details

You can also:

$ BASE_PATH=/examples pnpm test

The CI spreads this over eight shards (SHARDS in .github/workflows/ci.yml, bin/shard.mjs deciding who takes what), one example at a time per shard — a cold run is ~60s per example on a runner with no GPU, and two of them at once just starve each other.

Eight examples are excluded, each with its reason in bin/e2e-exceptions.mjs — three that were never built (bbuild2), three that throw, one behind a ▶️ button, one that needs more than the 180s budget. An example is in the run if and only if it has a test script, and test/e2e-exceptions.test.ts fails if that list and the scripts disagree.

Is it reproducible?

$ pnpm exec e2e-flaky @example/aquarium            # 3 shots, same canvas?
$ pnpm exec e2e-flaky --runs=10                    # every example with a test

Shoots the same example N times and compares the canvas byte for byte. This is what decides whether an example may be published to Chromatic, which has no pixel tolerance to hide behind. It has to be measured: useFrame tells you nothing either way.

The Flaky workflow runs it nightly, five shots per example, and goes red when an example starts drifting — before Chromatic starts flagging changes nobody made.

Chromatic

The same runs also archive each page — DOM, styles, assets, and the <canvas> as a still — for Chromatic, which re-renders them in its own browsers and asks a human to accept or reject what moved. Where pnpm test answers "did this change?", Chromatic answers "should it have?".

$ pnpm test          # writes examples/*/test-results/chromatic-archives/
$ pnpm chromatic     # collects them into one build and uploads it

Needs CHROMATIC_PROJECT_TOKEN (repo secret in the CI, your shell locally); a pull request from a fork cannot read it, and gets no Chromatic build.

[!IMPORTANT] Only the examples listed in bin/chromatic.mjs are published. A snapshot joins the list once pnpm exec e2e-flaky <example> says N shots of the same commit produce the same canvas — Chromatic has no pixel tolerance to hide behind, and a build that flags a change nobody made is a build nobody reads.

Docker

For generating reproductible snapshots, we use mcr.microsoft.com/playwright:v1.45.3-jammy Docker image.

$ docker run -it --rm  \
  -w /app -v "$(pwd)":/app -v /app/node_modules \
  mcr.microsoft.com/playwright:v1.45.3-jammy /bin/sh
#
# echo "Hey, I am acting like the CI"
#
# pnpm install
# pnpm test

Colophon