fully automated package publishing
Trust us, this will change your workflow for the better.
– egghead.io
Out of the box this is just about commit-messages, but you can do so much more.
- Detect breaking changes using the test suite of your last release: cracks
- Detect breaking changes using your dependents’ test suites: Help out! Implement the dont-break plugin
- Detect breaking changes diffing your JSDoc interface: Help out! Implement the india plugin
- Abort releases with insufficient test coverage: Help out! Implement the istanbul plugin
- Abort releases with vulnerable dependencies in the tree: Help out! Implement the nsp plugin
- Everything you can imagine: Build Plugins!
| Commands | Comment | |
|---|---|---|
| manual/before | |
You manually decide what the next version is. You have to remember what major, minor and patch means. You have to remember to push both commits and tags. You have to wait for the CI to pass. |
| semantic-release/after | |
You describe the changes you’ve made. A new version is automatically published with the correct version number. |
This removes the immediate connection between human emotions and version numbers, so strictly following the SemVer spec is not a problem anymore – and that’s ultimately semantic-release’s goal.
| “How to Write a JavaScript Library - Automating Releases with semantic-release” – egghead.io | |
|---|---|
|
|
| A free egghead.io tutorial series on how to write an open source library featuring semantic-release. | |
| “We fail to follow SemVer – and why it needn’t matter” | “semantic-release Q&A with Kent C. Dodds” |
|
|
| This talk gives you a complete introduction to the underlying concepts of this module. 38:30 | A “Hangouts on Air” conversation with hands on questions and answers about how to use semantic-release. 53:52 |
Instead of writing meaningless commit messages, we can take our time to think about the changes in the codebase and write them down. Following formalized conventions it is then possible to generate a helpful changelog and to derive the next semantic version number from them.
When semantic-release is setup it will do that after every successful continuous integration build of your master branch (or any other branch you specify) and publish the new version for you. This way no human is directly involved in the release process and your releases are guaranteed to be unromantic and unsentimental.
If you fear the loss of control over timing and marketing implications of software releases you should know that semantic-release supports release channels using npm’s dist-tags. This way you can keep control over what your users end up using by default, you can decide when to promote an automatically released version to the stable channel, and you can choose which versions to write blogposts and tweets about. You can use the same mechanism to support older versions of your software, for example with important security fixes.
This is what happens in series:
1. git push |
2. semantic-release pre |
3. npm publish |
4. semantic-release post |
|---|---|---|---|
| New code is pushed and triggers a CI build. | Based on all commits that happened since the last release, the new version number gets written to the package.json. |
The new version gets published to npm. |
A changelog gets generated and a release (including a git tag) on GitHub gets created. |
Note: The current release/tag implementation is tied to GitHub, but could be opened up to Bitbucket, GitLab, et al. Feel free to send PRs for these services.
This module ships with the AngularJS Commit Message Conventions and changelog generator, but you can define your own style.
Each commit message consists of a header, a body and a footer. The header has a special format that includes a type, a scope and a subject (full explanation):
<type>(<scope>): <subject>
<BLANK LINE>
<body>
<BLANK LINE>
<footer>
You can simplify using this convention for yourself and contributors by using commitizen and validate-commit-msg.
fix(pencil): stop graphite breaking when too much pressure applied
feat(pencil): add 'graphiteWidth' option
perf(pencil): remove graphiteWidth option
BREAKING CHANGE: The graphiteWidth option has been removed. The default graphite width of 10mm is always used for performance reason.
npm install -g semantic-release-cli
cd your-module
semantic-release-cli setup
This is what happens under the hood.
You can pass options either via command line (in kebab-case) or in the release field of your package.json (in camelCase). The following two examples are the same, but CLI arguments take precedence.
| CLI | package.json |
|---|---|
|
|
These options are currently available:
branch: The branch on which releases should happen. Default:'master'debug: If true doesn’t actually publish to npm or write things to file. Default:!process.env.CIgithubToken: The token used to authenticate with GitHub. Default:process.env.GH_TOKENgithubUrl: Optional. Pass your GitHub Enterprise endpoint.githubApiPathPrefix: Optional. The path prefix for your GitHub Enterprise API.
A few notes on npm config:
-
The
npmtoken can only be defined in the environment asNPM_TOKEN, because that’s wherenpmitself is going to read it from. -
In order to publish to a different
npmregistry you can specify that inside thepackage.json’spublishConfigfield. -
If you want to use another dist-tag for your publishes than
'latest'you can specify that inside thepackage.json’spublishConfigfield. -
semantic-releasegenerally tries to orientate itself towardsnpm– it inherits the loglevel for example.
There are numerous steps where you can customize semantic-release’s behaviour using plugins. A plugin is a regular option, but passed inside the release block of package.json:
{
"release": {
"analyzeCommits": "npm-module-name",
"generateNotes": "./path/to/a/local/module",
"verifyConditions": {
"path": "./path/to/a/module",
"additional": "config"
}
}semantic-release pre --analyze-commits="npm-module-name"
A plugin itself is an async function that always receives three arguments.
module.exports = function (pluginConfig, config, callback) {}pluginConfig: If the user of your plugin specifies additional plugin config in thepackage.json(see theverifyConditionsexample above) then it’s this object.config: A config object containing a lot of information to act upon.env: All environment variablesnpm: Select npm configuration bits likeregistry,tagandauthoptions:semantic-releaseoptions likedebug, orbranchpkg: Parsedpackage.json- For certain plugins the
configobject contains even more information. See below.
callback: If an error occurs pass it as first argument. Otherwise pass your result as second argument.
This plugin is responsible for determining the type of the next release. It additionally receives a commits array inside config. One commit is an object with a message and hash property. Call the callback with 'major', 'premajor', 'minor', 'preminor', 'patch', 'prepatch', 'prerelease', or null if nothing changed.
While it may be tempting to use 'prepatch', 'preminor' & 'prerelease' as part of a release process, this is strongly discouraged. A better approach is to use dist-tags to create release channels (such as 'latest', 'next', 'stable') and to return only 'major', 'premajor' and 'minor' from the commit analyzer.
Have a look at the default implementation.
This plugin is responsible for generating release notes. Call the callback with the notes as a string. Have a look at the default implementation.
This plugins is responsible for verifying that a release should happen in the first place. For example, the default implementation verifies that the publish is happening on Travis, that it’s the right branch, and that all other build jobs succeeded. There are more use cases for this, e.g. verifying that test coverage is above a certain threshold or that there are no vulnerabilities in your dependencies. Be creative.
Passing an array of plugins will run them in series.
This plugin is responsible for verifying a release that was determined before and is about to be published. There is no default implementation. It additionally receives nextRelease, lastRelease and commits inside config. While commits is the same as with analyzeCommits, nextRelease contains a type (e.g. 'major') and the new version (e.g. '1.0.0') and lastRelease contains the old version, the gitHead at the time of the release and the npm dist-tag (e.g. 'latest'). Using this information you could detect breaking changes or hold back certain types of releases. Again: Be creative.
Passing an array of plugins will run them in series.
This plugin is responsible for determining a package’s last release version. The default implementation uses the last published version on a npm registry.
I think you might frequently ask questions like these
The npm docs even state:
The most important things in your package.json are the name and version fields. Those are actually required, and your package won’t install without them. – npm docs
While this entirely true the version number doesn’t have to be checked into source control. semantic-release takes care of the version field right before npm publish uses it – and this is the only point where it really is required.
If you run npm run semantic-release locally a dry run gets performed, which logs the version that would currently get published.
Of course you can, but this doesn’t necessarily mean you should. Running your tests on an independent machine before releasing software is a crucial part of this workflow. Also it is a pain to set this up locally, with tokens lying around and everything. That said, you can run the scripts with --debug=false explicitly. You have to export GH_TOKEN=<your_token> and NPM_TOKEN=<your_other_token>.
You can trigger a release by pushing to your GitHub repository. You deliberately cannot trigger a specific version release, because this is the whole point of semantic-release. Start your packages with 1.0.0 and semver on.
It is indeed a great idea because it forces you to follow best practices. If you don’t feel comfortable making every passing feature or fix on your master branch addressable via npm you might not treat your master right. Have a look at branch workflows. If you still think you should have control over the exact point in time of your release, e.g. because you are following a release schedule, you can release only on the production/deploy/release branch and push your code there in certain intervals, or better yet use dist-tags.
semantic-release has a full unit- and integration-test-suite that tests actual npm publishes against the npm-registry-couchapp on all major node.js versions from ^0.10 on. A new version won’t get published if it doesn’t pass on all these engines.
Use this in one of your projects? Include one of these badges in your README.md to let people know that your package is published using semantic-release.
[](https://github.com/semantic-release/semantic-release)
[](https://github.com/semantic-release/semantic-release)
[](https://github.com/semantic-release/semantic-release)MIT License 2015 © Stephan Bönnemann and contributors
