Skip to content

Latest commit

 

History

History
137 lines (90 loc) · 6.23 KB

CONTRIBUTING.md

File metadata and controls

137 lines (90 loc) · 6.23 KB
Raw

This guide is best-effort and will be improved as necessary.

Dev environment

Tools I use:

  • yarn 3 for package management
  • volta (optional, node version management)

Features, bugfixes, and other code

We use package.json scripts for building, testing, and linting. Read the scripts to become familiar with our build process. The big ones are:

yarn
yarn build
yarn test
yarn fmt

yarn prepack / pnpm prepack/ npm prepare are maintained so that anyone can install ts-node from git, which is useful for testing experimental branches and unreleased features.

Source lives in src and is compiled to dist. Some shim files live outside of src so that they can be imported at certain paths. For example, to allow users to import ts-node/register, we have register/index.js which is a shim to compiled code in dist.

dist-raw is for larger chunks of code which are not compiled nor linted because they have been copy-pasted from node's source code.

Tests

Test cases are declared in src/test/*.spec.ts, and test fixtures live in ./tests. They can be run with yarn test.

To run a subset of tests:

# Use ava's --match flag to match the name of a test or suite
# https://github.com/avajs/ava/blob/main/docs/05-command-line.md
# Don't forget the * wildcards
yarn test --match '*esm loader*'

# Or pass a filename as it exists in the compiled output
# To run the tests in ./src/test/diagnostics.spec.ts
yarn test ./dist/test/diagnostics.spec.js

Tests are run with AVA, but using a custom wrapper API to enable some TS-friendly features and grouped test suites.

The tests yarn pack ts-node into a tarball and yarn install it into ./tests/node_modules. This makes ./tests a better testing environment because it more closely matches the end-user's environment. Complex require() / import / --require / --loader invocations behave the way they would in a users's project.

Historically, it has been difficult to test ts-node in-process because it mutates the node environment: installing require hooks, stack trace formatters, etc. nyc, ava, and ts-node all mutate the node environment, so it is tricky to setup and teardown individual tests in isolation, because ts-node's hooks need to be reset without disturbing nyc or ava hooks. For this reason, many tests are integration style, spawning ts-node's CLI in an external process, asking it to execute one of the fixture projects in ./tests.

Over time, I've gradually added setup and teardown logic so that more components can be tested in-process.

We have a debug configuration for VSCode.

  1. Open a *.spec.ts so it is the active/focused file.
  2. (optional) set breakpoints.
  3. Invoke debugger with F5.

Note that some tests might misbehave in the debugger. REPL tests in particular. I'm not sure why, but I think it is related to how console does not write to stdout when in a debug session.

Test Context

Ava has the concept of test "context", an object which can store reusable fields common across many tests.

By convention, any functions that setup reusable context start with ctx, making them easier to tab-complete and auto-import while writing tests.

See ctxTsNode for an example.

Context setup functions are re-executed for each test case. If you don't want this, wrap the context function in lodash once(). Each test will still get a unique context object, but the values placed onto each context will be identical.

Every ctx* function has a namespace with Ctx and T types. These make it easier/cleaner to write tests.

Test Macros

Ava has the concept of test "macros", reusable functions which can declare a type of test many times with different inputs.

Macro functions are created with test.macro().

By convention, if a macro function is meant to be imported and reused in multiple files, its name should start with macro.

Macros can also be declared to require a certain "context," thanks to the namespace types described in "Test Context" above.

See examples in helpers/*.ts.

Documentation

Documentation is written in markdown in website/docs and rendered into a website by Docusaurus. The README is also generated from these markdown files.

To edit documentation, modify the markdown files in ./website/docs and the sidebar declaration in ./website/sidebars.js

Docs for the latest stable release live in a docs branch. The "Edit this page" links on the website link to the docs branch so that the website can be improved in parallel with new feature work.

Docs changes for unreleased features are merged to main in the same PR which implements the feature, adds tests, etc. When we release a new version, we merge main with docs, unifying the two.

cd ./website
yarn
yarn start
# Will host live website locally

yarn build-readme # will rebuild the README.md

This site was used to generate the favicon from a high-res PNG export of the SVG. https://realfavicongenerator.net/

Release checklist

We publish using np: https://npm.im/np

  1. Merge docs into main using a pull request, ensuring a consistent squash-merge
  2. Rebuild the README (see instructions above, necessary because npmjs.com renders the readme)
  3. (optional) Update the api-extractor report; check for unexpected changes. See below
  4. Publish with np
  • np --branch main --no-tests
  • --no-tests because we must rely on CI to test ts-node. Even if you did run the tests locally, you would only be testing a single operating system, node version, and TypeScript version, so locally-run tests are insufficient.
  1. Add changelog to the Github Release; match formatting from previous releases
  2. Move docs branch to head of main
  • this rebuilds the website
  • git push --force origin main:docs
  • avoids merge messiness due to earlier squash-merge from docs to main
  1. If tsconfig schema has changed, send a pull request to schemastore. Example

APIExtractor

yarn api-extractor will update an API report generated by api-extractor which may be useful when generating release notes to detect (breaking) changes in our API surface.

I configured it for my own convenience; it is not a necessary part of our development process.