y1

y1 is our task runner that pretends to be yarn@1.22.22, but only does task running.

We use it because it's supported by turborepo and we don't want to mix task running and dependency management. Our workspace package.json says:

  "packageManager": "yarn@1.22.22",

Task runner only

LLMs and developers alike will typically try to manage dependencies using the declared packageManager. We need to be a lot more flexible with dependency management than with task running.

Performance

We run hundreds of thousands of tasks per day. Here's yarn's cost for a trivial task on a regular macbook (using nodejs):

% time yarn run test
yarn run v1.22.22
$ true
✨  Done in 0.02s.
yarn run test  0.11s user 0.03s system 98% cpu 0.141 total

Output ordering

y1's own messages (the yarn-emulated header, status, and error lines) are printed stdout-first then stderr. Task passthrough output preserves the task's original stream ordering.

Reliable signals

We need a way to catch the result of every single task through logs. It's great that yarn v1 prints the time taken for a task. It's also great that it prints the exit code on error.

Additional args

We emulate yarn v1's r, but also support single double dashes.

Configuration

y1 does not read yarn config. It's meant to behave the same on any system. It's basically unaffected by env.

Supported subcommands

• yarn run

  • Runs as: yarn run --non-interactive

• yarn run [task]

  • Omits: stdout line "info Visit https://yarnpkg.com/en/docs/cli/run for documentation about this command."
  • Accepts -- -- as separator for custom args (the original behavior)
  • Deviation: Also accepts -- with the same effect as -- -- to work like other task runners.

• yarn [task]

  • like yarn [task] except for any subcommand listed in yarn@1.22.22's help.

• yarn help | yarn --help | yarn -h

  • see Help section

• yarn help run

  • see yarn help run

Notably rejected

Every invocation that isn't documented as supported is an error.

The error message is: y1 yarn port rejected: $@

The exit code is: 99

• yarn without args in the actual yarn v1 does dependency management. The error message here is y1 yarn port requires a subcommand. Exit code is 99 here too.

Help section

Similar to the original help but:

• Omits: every line that's an unsupported flag or subcommand.
• Omits: "Visit https://yarnpkg.com/en/docs/cli/ to learn more about Yarn."
• Adds: a line identifying the y1 tool with version.

yarn help run

Samp principle as Help section.

npx gate mode

y1 doubles as an npx gate when invoked as npx or y-npx (detected via argv[0]). Instead of running npx freely, it blocks any invocation whose arguments are not explicitly whitelisted.

Y_NPX_ALLOWED_CMDS

Comma-separated list of allowed arg strings. Each entry is compared exactly (after whitespace normalization) against the full args passed to npx.

Y_NPX_ALLOWED_CMDS="tsc --noEmit,eslint ."

Only npx tsc --noEmit and npx eslint . are allowed. Everything else is blocked with:

y-npx blocked npx because `[args]` not found in Y_NPX_ALLOWED_CMDS (use installed tools)

Exit code on block: 1.

Y_NPX_ALLOWED_CMDS_SEPARATOR

Override the separator (default: ,). Useful if an allowed command itself contains commas.

Allowed invocations

When the args match an entry, y1 finds the system npx (skipping itself in PATH) and exec's into it, preserving all original arguments.

Development tooling

A CLI written in latest Rust.

Multi-arch builds (linux + mac and arm64 + amd64) using github action with pegged action versions.

Releases on tags using github actions, with uncompressed single binary downloads and a checksum file.

Running original yarn during development

For example: npm run refresh >/dev/null && (cd examples/minimal; ../../node_modules/.bin/yarn test)