- How to Contribute
- For Project Maintainers
First of all, thank you for your interest in Puppeteer! We'd love to accept your patches and contributions!
Contributions to this project must be accompanied by a Contributor License Agreement. You (or your employer) retain the copyright to your contribution, this simply gives us permission to use and redistribute your contributions as part of the project. Head over to https://cla.developers.google.com/ to see your current agreements on file or to sign a new one.
You generally only need to submit a CLA once, so if you've already submitted one (even if it was for a different project), you probably don't need to do it again.
- Clone this repository
git clone https://github.com/puppeteer/puppeteer
cd puppeteer
- Install dependencies
npm install
- Run Puppeteer tests locally. For more information about tests, read Running & Writing Tests.
npm run unit
All submissions, including submissions by project members, require review. We use GitHub pull requests for this purpose. Consult GitHub Help for more information on using pull requests.
- Coding style is fully defined in
.eslintrc
and we automatically format our code with Prettier. - It's recommended to set-up Prettier into your editor, or you can run
npm run eslint-fix
to automatically format any files. - If you're working in a JS file, code should be annotated with closure annotations.
- If you're working in a TS file, you should explicitly type all variables and return types. You'll get ESLint warnings if you don't so if you're not sure use them as guidelines, and feel free to ask us for help!
To run ESLint, use:
npm run eslint
You can check your code (both JS & TS) type-checks by running:
npm run tsc
- Try to avoid the use of
any
when possible. Considerunknown
as a better alternative. You are able to useany
if needbe, but it will generate an ESLint warning.
The code in Puppeteer is split primarily into two folders:
src
contains all source codevendor
contains all dependencies that we've vendored into the codebase. See thevendor/README.md
for details.
We structure these using TypeScript's project references, which lets us treat each folder like a standalone TypeScript project.
Currently Puppeteer ships two bundles; a CommonJS version for Node and an ESM bundle for the browser. Therefore we maintain two tsconfig
files for each project; tsconfig.esm.json
and tsconfig.cjs.json
. At build time we compile twice, once outputting to CJS and another time to output to ESM.
We compile into the lib
directory which is what we publish on the npm repository and it's structured like so:
lib
- cjs
- puppeteer <== the output of compiling `src/tsconfig.cjs.json`
- vendor <== the output of compiling `vendor/tsconfig.cjs.json`
- esm
- puppeteer <== the output of compiling `src/tsconfig.esm.json`
- vendor <== the output of compiling `vendor/tsconfig.esm.json`
The main entry point for the Node module Puppeteer is cjs-entry.js
. This imports lib/cjs/puppeteer/index.js
and exposes it to Node users.
We also maintain test/tsconfig.test.json
. This is only used to compile the unit test *.spec.ts
files. When the tests are run, we first compile Puppeteer as normal before running the unit tests against the compiled output. Doing this lets the test run against the compiled code we ship to users so it gives us more confidence in our compiled output being correct.
The root tsconfig.json
exists for the API Extractor; it has to find a tsconfig.json
in the project's root directory. It is not used for anything else.
When authoring new API methods, consider the following:
- Expose as little information as needed. When in doubt, don’t expose new information.
- Methods are used in favor of getters/setters.
- The only exception is namespaces, e.g.
page.keyboard
andpage.coverage
- The only exception is namespaces, e.g.
- All string literals must be small case. This includes event names and option values.
- Avoid adding "sugar" API (API that is trivially implementable in user-space) unless they're very demanded.
Commit messages should follow the Conventional Commits format. This is enforced via npm run lint
.
In particular, breaking changes should clearly be noted as “BREAKING CHANGE:” in the commit message footer. Example:
fix(page): fix page.pizza method
This patch fixes page.pizza so that it works with iframes.
Issues: #123, #234
BREAKING CHANGE: page.pizza now delivers pizza at home by default.
To deliver to a different location, use the "deliver" option:
`page.pizza({deliver: 'work'})`.
All public API should have a descriptive entry in docs/api.md
. There's a documentation linter which makes sure documentation is aligned with the codebase.
To run the documentation linter, use:
npm run doc
To format the documentation markdown and its code snippets, use:
npm run markdownlint-fix
For all dependencies (both installation and development):
- Do not add a dependency if the desired functionality is easily implementable.
- If adding a dependency, it should be well-maintained and trustworthy.
A barrier for introducing new installation dependencies is especially high:
- Do not add installation dependency unless it's critical to project success.
There are additional considerations for dependencies that are environment agonistic. See the vendor/README.md
for details.
- Every feature should be accompanied by a test.
- Every public api event/method should be accompanied by a test.
- Tests should not depend on external services.
- Tests should work on all three platforms: Mac, Linux and Win. This is especially important for screenshot tests.
Puppeteer tests are located in the test
directory and are written using Mocha. See test/README.md
for more details.
Despite being named 'unit', these are integration tests, making sure public API methods and events work as expected.
- To run all tests:
npm run unit
- To run a specific test, substitute the
it
withit.only
:
...
it.only('should work', async function({server, page}) {
const response = await page.goto(server.EMPTY_PAGE);
expect(response.ok).toBe(true);
});
- To disable a specific test, substitute the
it
withxit
(mnemonic rule: 'cross it'):
...
// Using "xit" to skip specific test
xit('should work', async function({server, page}) {
const response = await page.goto(server.EMPTY_PAGE);
expect(response.ok).toBe(true);
});
- To run tests in non-headless mode:
HEADLESS=false npm run unit
- To run Firefox tests, firstly ensure you have Firefox installed locally (you only need to do this once, not on every test run) and then you can run the tests:
PUPPETEER_PRODUCT=firefox node install.js
PUPPETEER_PRODUCT=firefox npm run unit
- To run tests with custom browser executable:
BINARY=<path-to-executable> npm run unit
Every public API method or event should be called at least once in tests. To ensure this, there's a coverage
command which tracks calls to public API and reports back if some methods/events were not called.
Run coverage:
npm run coverage
See Debugging Tips in the readme.
The following steps are needed to update the Chromium version.
- Find a suitable Chromium revision
Not all revisions have builds for all platforms, so we need to find one that does.
To do so, run
utils/check_availability.js -rd
to find the latest suitabledev
Chromium revision (seeutils/check_availability.js -help
for more options). - Update
src/revisions.ts
with the found revision number. - Update
versions.js
with the new Chromium-to-Puppeteer version mapping. - Run
npm run ensure-correct-devtools-protocol-revision
. If it fails, updatepackage.json
with the expecteddevtools-protocol
version. - Run
npm run tsc
andnpm install
. - Run
npm run unit
and ensure that all tests pass. If a test fails, bisect the upstream cause of the failure, and either update the test expectations accordingly (if it was an intended change) or work around the changes in Puppeteer (if it’s not desirable to change Puppeteer’s observable behavior). - Commit and push your changes and open a pull request.
The commit message must contain the version in
Chromium <version> (<revision>)
format to ensure that pptr.dev can parse it correctly, e.g.'feat(chromium): roll to Chromium 90.0.4427.0 (r856583)'
.
Sometimes, performing a Chromium roll causes tests to fail. To figure out the cause, you need to bisect Chromium revisions to figure out the earliest possible revision that changed the behavior. The script in utils/bisect.js
can be helpful here. Given a Node.js script that calls process.exit(1)
for bad revisions, run this from the Puppeteer repository’s root directory:
node utils/bisect.js --good 686378 --bad 706915 script.js
Releasing to npm consists of the following phases:
- Source Code: mark a release.
- Run
npm run release
. (This automatically bumps the version number inpackage.json
, populates the changelog, updates the docs, and creates a Git commit for the next step.) - Send a PR for the commit created in the previous step.
- Make sure the PR passes all checks.
- WHY: there are linters in place that help to avoid unnecessary errors, e.g. like this
- Merge the PR.
- Once merged, publish the release notes from
CHANGELOG.md
using GitHub’s “draft new release tag” option.- NOTE: tag names are prefixed with
'v'
, e.g. for version1.4.0
the tag isv1.4.0
.
- NOTE: tag names are prefixed with
- As soon as the Git tag is created by completing the previous step, our CI automatically
npm publish
es the new releases for both thepuppeteer
andpuppeteer-core
packages.
- Run
- Source Code: mark post-release.
- Bump
package.json
version to the-post
version, runnpm run doc
to update the “released APIs” section at the top ofdocs/api.md
accordingly, and send a PR titled'chore: bump version to vXXX.YYY.ZZZ-post'
(example)- NOTE: no other commits should be landed in-between release commit and bump commit.
- Bump