548b383 Aug 31, 2018
1 contributor

Users who have contributed to this file

229 lines (169 sloc) 9.14 KB
title id
Config Files

Configuration File Types

Babel has two parallel config file formats, which can be used together, or independently.

  • Project-wide configuration
  • File-relative configuration
    • .babelrc (and .babelrc.js) files
    • package.json files with a "babel" key

Project-wide configuration

New in Babel 7.x, Babel has as concept of a "root" directory, which defaults to the current working directory. For project-wide configuration, Babel will automatically search for a "babel.config.js" in this root directory. Alternatively, users can use an explicit "configFile" value to override the default config file search behavior.

Because project-wide config files are separated from the physical location of the config file, they can be ideal for configuration that must apply broadly, even allowing plugins and presets to easily apply to files in node_modules or in symlinked packages, which were traditionally quite painful to configure in Babel 6.x.

The primary downside of this project-wide config is that, because it relies on the working directory, it can be more painful to use in monorepos if the working directory is not the monorepo root. For example, if you have


and the individual packages are responsible for running their builds (and their working directory for Babel is the individual packages), the babel.config.js file will not be automatically loaded, and users will be required to set the path to it manually.

Project-wide configs can also be disabled by setting "configFile" to false.

File-relative configuration

Babel loads .babelrc (and .babelrc.js / package.json#babel) files by searching up the directory structure starting from the "filename" being compiled. This can be powerful because it allows you to create independent configurations for subsections of a repository. File-relative configurations are also merged over top of project-wide config values, making them potentially useful for specific overrides, though that can also be accomplished through "overrides".

There are a few edge cases to consider when using a file-relative config:

  • Searching will stop once a directory containing a package.json is found, so a relative config only applies within a single package.
  • The "filename" being compiled must be inside of "babelrcRoots" packages, or else searching will be skipped entirely.

File-relative configs can also be disable by setting "babelrc" to false.

6.x vs 7.x .babelrc loading

Users coming from Babel 6.x will likely trip up on these two edge cases, which are new in Babel 7.x. These two restrictions were added to address common footguns in Babel 6.x:

  • .babelrc files applied to node_modules dependencies, often unexpectedly.
  • .babelrc files failed to apply to symlinked node_modules when people expected them to behave like normal dependencies.
  • .babelrc files in node_modules dependencies would be detected, even though the plugins and presets inside they were generally not installed, and may not even be valid in the version of Babel compiling the file.

These cases will primarily cause issues for users with a monorepo structure, because if you have


the config will now be entirely ignored, because it is across a package boundary.

One alternative would be to create a .babelrc in each sub-package that uses "extends" as

{ "extends": "../../.babelrc" }

Unfortunately, this approach can be a bit repetitive, and depending on how Babel is being used, could require setting "babelrcRoots".

Given that, it may be more desirable to rename the .babelrc to be a project-wide "babel.config.js". As mentioned in the project-wide section above, this may then require explicitly setting "configFile" since Babel will not find the config file if the working directory isn't correct.

Config Format

The format of individual config files themselves separates into JS files vs JSON5 files.


Any file that isn't a .js file will be parsed as JSON5 and should contain an object matching the options format that Babel accepts.


Any .js file will be require()ed and should export either a configuration object, or a function that will return a configuration object when called. The main benefit being that users can include JS logic to build up their config structures, potentially allowing config logic to be shared more easily. .js files can be used as project-wide configuration or via .babelrc.js files for file-relative configuration.

Function-returning configs are given a few special powers because they can access an API exposed by Babel itself. See Config Function API for more information.

Config Function API

JS config files may export a function that will be passed config function API:

module.exports = function(api) {
  return {};

The api object exposes everthing Babel itself exposes from its index module, along with config-file specific APIs:


Type: string

The version string for the Babel version that is loading the config file.


JS configs are great because they can compute a config on the fly, but the downside there is that it makes caching harder. Babel wants to avoid re-executing the config function every time a file is compiled, because then it would also need to re-execute any plugin and preset functions referenced in that config.

The avoid this, Babel expects users of config functions to tell it how to manage caching within a config file.

  • api.cache.forever() - Permacache the computed config and never call the function again.
  • api.cache.never() - Do not cache this config, and re-execute the function every time.
  • api.cache.using(() => process.env.NODE_ENV) - Cache based on the value of NODE_ENV. Any time the using callback returns a value other than the one that was expected, the overall config function will be called again and a new entry will be added to the cache.
  • api.cache.invalidate(() => process.env.NODE_ENV) - Cache based on the value of NODE_ENV. Any time the using callback returns a value other than the one that was expected, the overall config function will be called again and all entries in the cache will be replaced with the result.

Since the actual callback result is used to check if the cache entry is valid, it is recommended that:

  • Callbacks should be small and side-effect free.
  • Callbacks should return values with the smallest range possible. For example, the .using(() => process.env.NODE_ENV) usage above is not ideal because it would create an unknown number of cache entries depending on how many values of NODE_ENV are detected. It would be safer to do .using(() => process.env.NODE_ENV === "development") because then the cache entry can only ever be true or false.


Since NODE_ENV is a fairly common way to toggle behavior, Babel also includes an API function meant specifically for that. This API is used as a quick way to check the "envName" that Babel was loaded with, which takes NODE_ENV into account if no other overriding environment is set.

It has a few different forms:

  • api.env("production") returns true if envName === "production".
  • api.env(["development", "test"]) returns true if ["development", "test"].includes(envName).
  • api.env() returns the current envName string.
  • api.env(envName => envName.startsWith("test-")) returns true if the env starts with "test-".

This function internally makes use of api.cache mentioned below to ensure that Babel is aware that this build depends on a specific envName.


This API is used as a way to access the caller data that has been passed to Babel. Since many instances of Babel may be running in the same process with different caller values, this API is designed to automatically configure api.cache, the same way api.env() does.

The caller value is available as the first parameter of the callback function. It is best used with something like

function isBabelRegister(caller) {
  return !!(caller && === "@babel/register");

module.exports = function(api) {
  const isRegister = api.caller(isBabelRegister);

  return {
    // ...

to toggle configuration behavior based on a specific environment.


While api.version can be useful in general, it's sometimes nice to just declare your version. This API exposes a simple way to do that with:

module.exports = function(api) {

  return {
    // ...