重要提示: 此中文文档针对的是 Yarn 2 版本。
有关 1.x 版本的中文文档,请点击进入 yarn.bootcss.com。
Yarn
ArchitectureContributingError CodesLexiconLifecycle ScriptsPlugin TutorialPnP APIPnPifyRulebookTelemetry

Sherlock

Edit this page on GitHub

In order to simplify our work maintaining Yarn, we ask you to help us by providing reproductions for the bugs you encounter. We can't stress this enough: under normal circumstances, no reproduction will mean no help and your issue may be closed for this reason.

To help you help us, we use a tool called Sherlock which allows you to attach reproduction testcases to your issues. In a sense, it's a bit like if you were writing a Jest test directly in your issue. By using this system we can quickly check whether a problem can be reproduced, and whether our fix is the right one. It's a win-win for everyone!

Adding a reproduction to your issue

Adding a reproduction is as simple as adding a markdown codefence to your issue:

This is my reproduction case:

```js repro
const foo = await ...;
```

Important: Note that the codefence is not only tagged js, but also repro. This is critical, as we don't want to execute every single snippet!

Developing your reproduction

The easiest and most convenient way to develop your reproduction is by using the Playground (powered by CodeSandbox).

Note: We strongly encourage you to use the Playground. It saves both you and us important time. Also, in the future, we might autoclose issues without a reproduction after a period of time, unless a contributor manually labels them as reproducible, so be prepared!

Features

  • A Monaco Editor (the editor that powers VSCode) where you can write your reproduction.
  • A Terminal-like output where you can see the assertion or the error / playground tips if you haven't run anything yet.
  • A Menu for selecting reproduction templates. Note: Your last input is saved inside localStorage for future use.
  • The status of the playground (more information on hover).
  • An Export button that can be used to share the link of the playground and even copy/open an issue in our repo with the issue template filled with your reproduction (status has to be reproducible). You can also open the reproduction in CodeSandbox.
  • Shareable Playground URLs - You can export them by using the Export button and the code will be automatically imported inside the Playground when you access it through them.

Limitations

  • Reproductions are run inside a CodeSandbox container, so they're always running inside a Linux environment, with the Node version that CodeSandbox uses. This limitation also exists when running the reproduction with Sherlock inside an issue, so if you need to use a specific OS / Node version, you have to develop your reproduction offline.
  • Don't use very large packages unless you absolutely have to - the sandbox won't like it.
  • The sandbox may go to sleep after some time. If that happens, reload the page to reboot it.

Available functions

You can use all of Javascript inside the repro code, including async/await. In fact, you can even use top-level await! As far as the runtime goes you should have access to all of Node's APIs, plus some extras:

  • In order to simplify some tasks that would be very menial otherwise (like spawning the Yarn process, or running a Node script in the generated environment), we provide a set of builtin functions that abstract these semantics. You can find the full reference inside our repository, along with some documentation.

    Note: All core plugins are enabled by default, you don't need to run yarn import on them again.

  • The expect API is available globally and must be used to describe the assertions in your test (simply throwing will cause Sherlock to mark your reproduction case as broken).

Examples

In #478, Yarn used to print an incorrect help message.

const output = await yarn(`add`, `--help`);

expect(output).not.toContain(`yarn yarn`);

In #448, running pnpify --sdk used to fail when ESLint was missing.

await packageJsonAndInstall({
  devDependencies: {
    [`@yarnpkg/pnpify`]: `^2.0.0-rc.3`,
    [`typescript`]: `^3.6.3`
  }
});

await expect(yarn(`pnpify`, `--sdk`)).rejects.not.toThrow(`Command failed`);

In #361, running yarn add dep when dep was already a devDependency used to duplicate it in both fields:

const {promises: {readFile}} = require(`fs`);

await packageJsonAndInstall({
  devDependencies: {lodash: `*`},
});

await yarn(`add`, `lodash`);

const pkgJson = JSON.parse(await readFile(`package.json`, `utf8`));
expect(pkgJson.dependencies).not.toHaveProperty(`lodash`);

For more examples, consult the list of issues that have been tagged as having valid reproductions:

https://github.com/yarnpkg/berry/issues?q=is%3Aissue+label%3Areproducible+is%3Aclosed

Q&A

Can I develop my reproduction offline?

If you can't use the Playground, you can build a reproduction offline and only share it once it's ready:

  1. Clone our repository
  2. Go into it, and write your issue in a markdown file (as if you were on GitHub)
  3. Once ready, run yarn sherlock <my-file.md> to make Sherlock run your reproduction locally

Note that you must write the issue itself, not only the reproduction, so don't forget to add the codefence etc. Once you are done, copy-paste the result on GitHub and the bot will promptly review your code and approve it.

The Sherlock bot doesn't label my message

It likely means that it didn't see your issue as containing a reproduction case. Make sure you properly added the repro tag to your codefence:

This is my reproduction case:

```js repro
const foo = ...;
```

The Sherlock bot is labelling my issue with broken-repro

This means that your code is throwing an exception on our systems which doesn't seem to be part of the expected testsuite. Note that only exceptions thrown by the expect API will lead to your reproduction being correctly labelled.

If you wish to test that something is failing (for example to test that yarn install --immutable doesn't work when the lockfile isn't up-to-date, or something like this) you can use the following pattern:

await expect(async () => {
  // ...
}()).rejects.toThrow();

How can I run the bot again?

Editing the issue content will cause the bot to run again. If you're a maintainer, you can also remove the reproduction label from the issue and Sherlock will happilly execute it again.

That's really cool, can I do the same on my own project?

Sherlock is an open-source project that we develop on this repository. Try and let us know how it worked out for you!