To create a plugin for semantic-release
, you need to decide which parts of the release lifecycle are important to that plugin. For example, it is best to always have a verifyConditions
step because you may be receiving inputs from a user and want to make sure they exist. A plugin can abide by any of the following lifecycles:
verifyConditions
analyzeCommits
verifyRelease
generateNotes
addChannel
prepare
publish
success
fail
semantic-release
will require the plugin via node
and look through the required object for methods named like the lifecycles stated above. For example, if your plugin only had a verifyConditions
and success
step, the main
file for your object would need to export
an object with verifyConditions
and success
functions.
In addition to the lifecycle methods, each lifecycle is passed two objects:
pluginConfig
- an object containing the options that a user may pass in via theirrelease.config.js
file (or similar)context
- provided bysemantic-release
for access to things likeenv
variables set on the running process.
For each lifecycle you create, you will want to ensure it can accept pluginConfig
and context
as parameters.
It is recommended that you generate a new project with yarn init
. This will provide you with a basic node project to get started with. From there, create an index.js
file, and make sure it is specified as the main
in the package.json
. We will use this file to orchestrate the lifecycle methods later on.
Next, create a src
or lib
folder in the root of the project. This is where we will store our logic and code for how our lifecycle methods work. Finally, create a test
folder so you can write tests related to your logic.
We recommend you setup a linting system to ensure good javascript practices are enforced. ESLint is usually the system of choice, and the configuration can be whatever you or your team fancies.
In your index.js
file, you can start by writing the following code
const verify = require("./src/verify");
let verified;
/**
* Called by semantic-release during the verification step
* @param {*} pluginConfig The semantic-release plugin config
* @param {*} context The context provided by semantic-release
*/
async function verifyConditions(pluginConfig, context) {
await verify(pluginConfig, context);
verified = true;
}
module.exports = { verifyConditions };
Then, in your src
folder, create a file called verify.js
and add the following
const AggregateError = require("aggregate-error");
/**
* A method to verify that the user has given us a slack webhook url to post to
*/
module.exports = async (pluginConfig, context) => {
const { logger } = context;
const errors = [];
// Throw any errors we accumulated during the validation
if (errors.length > 0) {
throw new AggregateError(errors);
}
};
As of right now, this code won't do anything. However, if you were to run this plugin via semantic-release
, it would run when the verify
step occurred.
Following this structure, you can create different steps and checks to run through out the release process.
Let's say we want to verify that an option
is passed. An option
is a configuration object that is specific to your plugin. For example, the user may set an option
in their release config like:
{
prepare: {
path: "@semantic-release/my-special-plugin";
message: "My cool release message";
}
}
This message
option will be passed to the pluginConfig
object mentioned earlier. We can use the validation method we created to verify this option exists so we can perform logic based on that knowledge. In our verify
file, we can add the following:
const { message } = pluginConfig;
if (message.length) {
//...
}
stdout
stderr
logger
Initially the context object contains the following keys (verifyConditions
lifecycle):
cwd
- Current working directory
env
- Environment variables
envCi
- Information about CI environment
- Contains (at least) the following keys:
isCi
- Boolean, true if the environment is a CI environment
commit
- Commit hash
branch
- Current branch
options
- Options passed to
semantic-release
via CLI, configuration files etc.
- Options passed to
branch
- Information on the current branch
- Object keys:
channel
tags
type
name
range
accept
main
branches
- Information on branches
- List of branch objects (see above)
Compared to the verifyConditions, analyzeCommits
lifecycle context has keys
commits
(List)- List of commits taken into account when determining the new version.
- Keys:
commit
(Object)- Keys:
long
(String, Commit hash)short
(String, Commit hash)
- Keys:
tree
(Object)- Keys:
long
(String, Commit hash)short
(String, Commit hash)
- Keys:
author
(Object)- Keys:
name
(String)email
(String)date
(String, ISO 8601 timestamp)
- Keys:
committer
(Object)- Keys:
name
(String)email
(String)date
(String, ISO 8601 timestamp)
- Keys:
subject
(String, Commit message subject)body
(String, Commit message body)hash
(String, Commit hash)committerDate
(String, ISO 8601 timestamp)message
(String)gitTags
(String, List of git tags)
releases
(List)lastRelease
(Object)- Keys
version
(String)gitTag
(String)channels
(List)gitHead
(String, Commit hash)name
(String)
- Keys
Additional keys:
nextRelease
(Object)type
(String)channel
(String)gitHead
(String, Git hash)version
(String, version withoutv
)gitTag
(String, version withv
)name
(String)
No new content in the context.
This is run only if there are releases that have been merged from a higher branch but not added on the channel of the current branch.
Context content is similar to lifecycle verifyRelease
.
Only change is that generateNotes
has populated nextRelease.notes
.
No new content in the context.
Lifecycles success
and fail
are mutually exclusive, only one of them will be run.
Additional keys:
releases
- Populated by
publish
lifecycle
- Populated by
Lifecycles success
and fail
are mutually exclusive, only one of them will be run.
Additional keys:
errors
Similar to options
, environment variables exist to allow users to pass tokens and set special URLs. These are set on the context
object instead of the pluginConfig
object. Let's say we wanted to check for GITHUB_TOKEN
in the environment because we want to post to GitHub on the user's behalf. To do this, we can add the following to our verify
command:
const { env } = context;
if (env.GITHUB_TOKEN) {
//...
}
Use context.logger
to provide debug logging in the plugin. Available logging functions: log
, warn
, success
, error
.
const { logger } = context;
logger.log('Some message from plugin.').
The above usage yields the following where PLUGIN_PACKAGE_NAME
is automatically inferred.
[3:24:04 PM] [semantic-release] [PLUGIN_PACKAGE_NAME] › ℹ Some message from plugin.
For the lifecycles, the list at the top of the readme contains the order. If there are multiple plugins for the same lifecycle, then the order of the plugins determines the order in which they are executed.
In order to be able to detect and handle errors properly, the errors thrown from the must be of type SemanticReleaseError or extend it as described in the package readme. This way the errors are handled properly and plugins using the fail
lifecycle receive the errors correctly. For any other types of errors the internal error handling does nothing, lets them through up until the final catch and does not call any fail
plugins.
Knowledge that might be useful for plugin developers.
While it may be trivial that multiple analyzeCommits (or any lifecycle plugins) can be defined, it is not that self-evident that the plugins executed AFTER the first one (for example, the default one: commit-analyzer
) can change the result. This way it is possible to create more advanced rules or situations, e.g. if none of the commits would result in new release, then a default can be defined.
The commit must be a known release type, for example the commit-analyzer has the following default types:
- major
- premajor
- minor
- preminor
- patch
- prepatch
- prerelease
If the analyzeCommits-lifecycle plugin does not return anything, then the earlier result is used, but if it returns a supported string value, then that overrides the previous result.