The primary implementation of changesets. Helps you manage the versioning and changelog entries for your packages, with a focus on versioning within a mono-repository (though we support single-package repositories too).
This package is intended as a successor to @atlaskit/build-releases
with a more general focus.
If you are installing this in a monorepo run
yarn add @changesets/cli
yarn changeset init
otherwise run
yarn add --dev @changesets/cli
yarn changeset init
From here you are set up to use changesets. Add your first changeset by running
yarn changeset
and following the prompts that you are presented with.
Below you can find a basic workflow for maintainers to help them use changesets, which you can vary to meet your own needs.
The core concept that changesets
follows is that contributors to a repository should be able to declare an intent to release, and that multiple intents should be able to be combined sensibly. Sensibly here refers to if there is one intent to release button as a 'minor' and another to release button as a 'patch', only one release will be made, at the higher of the two versions.
A single changeset
is an intent to release stored as data, with the information we need to combine multiple changesets and coordinate releases. We also work along bolt's structure guidelines to make sure that packages within a mono-repository will all depend on the latest versions of each other. This approach comes from bolt.
Contributor runs:
yarn changeset
or
npx changeset
and answers the provided questions.
When the maintainer wants to release packages, they should run
yarn changeset bump
or
npx changeset bump
and then
yarn changeset release
or
npx changeset release
The commands are explained further below.
changeset init
This command sets up the .changeset
folder. It generates a readme and a config file. The config file includes the default options, as well as comments on what these options represent. You should run this command once, when you are setting up changesets
.
changeset [--commit]
or
changeset add [--commit]
This command will ask you a series of questions, first about what packages you want to release, then what version for each package, then it will ask for a summary of the entire changeset. At the final step it will show the changeset it will generate, and confirm that you want to add it.
Once confirmed, the changeset will be written into two files:
.changeset/{HASH}/changes.md
- this includes the summary message, and is safe to edit and expand on..changeset/{HASH}/changes.json
- this is the intent to update and should not be manually edited.
The information in the changes.json
will look like:
{
"releases": [
{ "name": "@atlaskit/analytics-listeners", "type": "major" },
{ "name": "@atlaskit/website", "type": "patch" }
],
"dependents": [
{
"name": "@atlaskit/global-navigation",
"type": "patch",
"dependencies": ["@atlaskit/analytics-listeners"]
}
]
}
You can pass the option --commit
, or provide this in the config. Commit is false by default. If it is true, the command will add the updated changeset files and then commit them.
changeset bump [--update-changelog] [--skipCI] [--commit]
Updates the versions for all packages (and depdendencies) described in changesets since last release.
Will also create/append to a CHANGELOG file for each package using the summaries from the changesets.
We recommend making sure changes made from this commmand are merged back into master before you run `release.
This command will read then delete changesets on disk, ensuring that they are only used once.
--update-changelog
(default true) - Sets whether you want changesets to write out changelog files.
--commit
(default false) - If true, running this command will automatically commit the changes it made.
[skip ci]
can be used to prevent this commit from triggering a CI build as the common use case would be to run this in master and then push back to master. We want to avoid the infinite loop there. If you are running version locally, you may need to make another commit after this to trigger your CI.
changeset release [--public]
Publishes to NPM repo, and creates tags. Because this command assumes that last commit is the release commit you should not commit any changes between calling version
and publish
. These commands are separate to enable you to check if release commit is acurate.
--public
- enables the --access-public
flag when publishing. This is required if trying to publish public scoped packages.
IF YOU HAVE 2FA TURNED ON FOR NPM you need to provide your auth token. Currently the way to do this is to run
NPM_CONFIG_OTP=123456 yarn changeset release
.
NOTE: You will still need to push your changes back to master after this
git push --follow-tags
status [--verbose] [--output={filePath}] [--since-master]
The status command provides information about the changesets that currently exist. If there are no changesets present, it exits with an error status code.
Use verbose if you want to know the new versions, and get a link to the relevant changeset summary.
You can use output
to write the json object of the status out, for consumption by other tools, such as CI.
You can use the `since-master flag to only display information about changesets since the master branch. While this can be used to add a CI check for changesets, we recommend not doing this. We instead recommend using the changeset bot to detect pull requests missing changesets, as not all pull requests need one.
In almost all circumstances, internal packages will be bumped as a patch. The one exception is when the dependency is a peerDependency
, in which case the change will become a major.