Skip to content

Latest commit

 

History

History
115 lines (76 loc) · 7.62 KB

CONTRIBUTING.md

File metadata and controls

115 lines (76 loc) · 7.62 KB
Raw

Contributing to Prettier

To get up and running, install the dependencies and run the tests:

yarn
yarn test

Tests

The tests use Jest snapshots. You can make changes and run jest -u (or yarn test -u) to update the snapshots. Then run git diff to take a look at what changed. Always update the snapshots when opening a PR.

Each test directory in tests/format has a jsfmt.spec.js file that controls how exactly the rest of the files in the directory are used for tests. This file must contain one or more calls to the run_spec global function. For example, in directories with JavaScript formatting tests, jsfmt.spec.js generally looks like this:

run_spec(__dirname, ["babel", "flow", "typescript"]);

This verifies that for each file in the directory, the output matches the snapshot and is the same for each listed parser.

You can also pass options as the third argument:

run_spec(__dirname, ["babel"], { trailingComma: "es5" });

Signature:

function run_spec(
  fixtures:
    | string
    | {
        dirname: string;
        snippets?: Array<
          | string
          | { code: string; name?: string; filename?: string; output?: string }
        >;
      },
  parsers: string[],
  options?: PrettierOptions & {
    errors: true | { [parserName: string]: true | string[] };
  }
): void;

Parameters:

  • fixtures: Must be set to __dirname or to an object of the shape { dirname: __dirname, ... }. The object may have the snippets property to specify an array of extra input entries in addition to the files in the current directory. For each input entry (a file or a snippet), run_spec configures and runs a number of tests. The main check is that for a given input the output should match the snapshot (for snippets, the expected output can also be specified directly). Additional checks are controlled by options and environment variables.
  • parsers: A list of parser names. The tests verify that the parsers in this list produce the same output. If the list includes typescript, then babel-ts is included implicitly. If the list includes babel, and the current directory is inside tests/format/js, then acorn, espree, and meriyah are included implicitly.
  • options: In addition to Prettier's formatting options, can contain the errors property to specify that it's expected that the formatting shouldn't be successful and an error should be thrown for all (errors: true) or some combinations of input entries and parsers.

The implementation of run_spec can be found in tests/config/format-test.js.

tests/format/flow-repo/ contains the Flow test suite and is not supposed to be edited by hand. To update it, clone the Flow repo next to the Prettier repo and run: node scripts/sync-flow-tests.js ../flow/tests/.

Debugging

To debug Prettier locally, you can either debug it in Node (recommended) or the browser.

  • The easiest way to debug it in Node is to create a local test file with some example code you want formatted and either run it in an editor like VS Code or run it directly via ./bin/prettier.js <your_test_file>.
  • The easiest way to debug it in the browser is to build Prettier's website locally (see website/README.md).

Other

The project uses ESLint for linting and Prettier for formatting. If your editor isn't set up to work with them, you can lint and format all files from the command line using yarn fix.

After opening a PR, describe your changes in a file in the changelog_unreleased directory following the template changelog_unreleased/TEMPLATE.md and commit this file to your PR. You can use ./scripts/generate-changelog.mjs to create changelog file. Please see comments of the script file for usage.

Take a look at commands.md and, if you know Haskell, check out Wadler's paper to understand how Prettier works.

If you want to know more about Prettier's GitHub labels, see the Issue Labels page on the Wiki.

Advanced topics

Performance

If you're contributing a performance improvement, the following Prettier CLI options can help:

  • --debug-repeat N uses a naïve loop to repeat the formatting N times and measures the average run duration. It can be useful to highlight hot functions in the profiler. The measurements are printed at the debug log level, use --loglevel debug to see them.
  • --debug-benchmark uses benchmark module to produce statistically significant duration measurements. The measurements are printed at the debug log level, use --loglevel debug to see them.

For convenience, the following commands for profiling are available via package.json scripts.

Unfortunately, yarn simply appends passed arguments to commands, cannot reference them by name, so we have to use inline environment variables to pass them.

  • PERF_FILE=<filename> PERF_REPEAT=[number-of-repetitions:1000] yarn perf:repeat starts the naïve loop. See the CLI output for when the measurements finish, and stop profiling at that moment.
  • PERF_FILE=<filename> PERF_REPEAT=[number-of-repetitions:1000] yarn perf:repeat-inspect starts the naïve loop with node --inspect-brk flag that pauses execution and waits for Chromium/Chrome/Node Inspector to attach. Open chrome://inspect, select the process to inspect, and activate the CPU Profiler, this will unpause execution. See the CLI output for when the measurements finish, and stop the CPU Profiler at that moment to avoid collecting more data than needed.
  • PERF_FILE=<filename> yarn perf:benchmark starts the benchmark-powered measurements. See the CLI output for when the measurements finish.

In the above commands:

  • yarn && yarn build ensures the compiler-optimized version of Prettier is built prior to launching it. Prettier's own environment checks are defaulted to production and removed during the build. The build output is cached, so a rebuild will happen only if the source code changes.
  • NODE_ENV=production ensures Prettier and its dependencies run in production mode.
  • node --inspect-brk pauses the script execution until Inspector is connected to the Node process.
  • --loglevel debug ensures the --debug-repeat or --debug-benchmark measurements are printed to stderr.
  • > /dev/null ensures the formatted output is discarded.

In addition to the options above, you can use node --prof and node --prof-process, as well as node --trace-opt --trace-deopt, to get more advanced performance insights.

Regression testing

We have a cool tool for regression testing that runs on GitHub Actions. Have a look: https://github.com/prettier/prettier-regression-testing

Deeper testing

You can run FULL_TEST=1 jest for a more robust test run, which includes the following additional checks:

  • compare AST - re-parses the output and makes sure the new AST is equivalent to the original one.
  • second format - formats the output again and checks that the second output is the same as the first.
  • EOL '\r\n' and EOL '\r' - check that replacing line endings with \r\n or \r in the input doesn't affect the output.
  • BOM - checks that adding BOM (U+FEFF) to the input affects the output in only one way: the BOM is preserved.

Usually there is no need to run these extra checks locally, since they're run on the CI anyway.