GitHub Action
release-train
release-train creates releases for every PR merge in your repository. No magic commit message is required. You only need to label your PRs with the appropriate labels and run release-train -- either from the command line or a GitHub Action.
Release-train is inspired by semantic-release and has a few advantages for somebody with my biases and proclivities:
- It doesn't require special commit messages, so you won't need to squash commits or ask contributors to rebase before merging. You are free to follow whatever commit conventions suit you, including xkcd commit conventions.
- No need for npm or other package managers.
- No plugin configuration. Release-train has no plugins. You can do anything a plugin would do from the pre-tag hook.
Labels and label aliases are not case-sensitive
Release-train uses pull request labels to determine the change level for each PR. The release updates the version based on the highest change level found in the PRs being released -- typically only one PR if you are doing continuous releases.
Label | Effect | Example |
---|---|---|
semver:breaking |
Increment major version | v0.1.2 -> v1.0.0 |
semver:minor |
Increment minor version | v0.1.2 -> v0.2.0 |
semver:patch |
Increment patch version | v0.1.2 -> v0.1.3 |
semver:none |
No version change | v0.1.2 -> v0.1.2 |
In addition there are prerelease and stable labels used to determine whether to
publish a prerelease or stable release. These are semver:prerelease
and
semver:stable
.
semver:prerelease
must be combined with version labels to determine what the
stable portion of the prerelease version will be.
semver:prerelease
can also specify the prerelease identifier. For example,
semver:prerelease:alpha
will create a prerelease with the identifier alpha
.
A prerelease cannot contain PRs with conflicting identifiers.
Previous release | Labels | Next release | Notes |
---|---|---|---|
v0.1.2 |
semver:breaking , semver:prerelease |
v1.0.0-0 |
|
v0.1.2 |
semver:breaking , semver:prerelease:alpha |
v1.0.0-alpha.0 |
|
v1.0.0-0 |
semver:breaking , semver:prerelease |
v1.0.0-1 |
Doesn't iterate major because minor and patch are 0 |
v1.1.0-0 |
semver:breaking , semver:prerelease |
v2.0.0-0 |
Iterates major because minor is not 0 |
v1.0.0-2 |
semver:breaking , semver:prerelease:alpha |
v1.0.0-alpha.0 |
New identifier resets the prerelease iterator |
When the most recent release is a prerelease, semver:stable
is used to
indicate that the next release should be stable. When the most recent release is
a prerelease and one PR in the next release is labeled semver:stable
, then all
PRs in the next release must be labeled semver:stable
.
If semver:stable
is combined with a version label, the version is incremented
after making the release stable. For instance, if the latest release
is v1.0.0-3
and you merge a PR labeled semver:stable
, the next release will
be v1.0.0
, but if the PR is labeled both semver:stable
and semver:minor
, the next release will be v1.1.0
.
The labels listed above are the canonical labels, but you can use aliases that are better suited to your project.
For example if you want something close to Conventional Commits your github action config might contain something like this:
labels: |
"breaking change"=semver:breaking
fix=semver:patch
feat=semver:minor
perf=semver:patch
chore=semver:none
docs=semver:none
style=semver:none
refactor=semver:patch
The pre-tag hook is a shell script that runs before the new release is tagged. It lets you do some customizations like creating release notes, building release artifacts, or validating the release.
See the action doc for more details.
When run from a release branch, release-train follows these steps to publish a
release. Some options such as --check-pr
will modify this behavior.
- Find the previous release tag by searching backward through git history
for the first tag formatted like
<prefix><semantic version>
. When no previous tag is found it usesv0.0.0
or the value from--initial-tag
. - Find the next release version by looking at the diff between HEAD and the previous release tag and inspecting the pull request where each commit was introduced. The previous version is incremented by the highest change level found.
- Run pre-tag-hook. This is where you can do things like validate the release, built release artifacts or generate a changelog.
- Create and push the new git tag if
--create-tag
is set. - Create a draft release if
--create-release
is set. It starts as a draft to avoid publishing a release that doesn't have all the necessary artifacts yet. - Upload release assets. Any files written to
$ASSETS_DIR
will be uploaded as release assets. - Publish the release.
- Emit output including release version, tag, change level, etc.
Recipes are only provided as GitHub Actions workflows. Adapting them to the command line is pretty straightforward.
Unlike Cooks Illustrated, we have no recipe testers. If you have trouble getting one to work, there may be an issue with the recipe. Please open an issue for help or a PR if you fix it.
This is probably the simplest possible usage of release-train. It will create a
release for every PR that is merged into main
with default release notes and
no release assets.
on:
push:
jobs:
release:
runs-on: ubuntu-latest
permissions:
contents: write # needed for creating tags and releases
steps:
- uses: actions/checkout@v3
with:
fetch-depth: 0
- uses: WillAbides/[email protected]
with:
create-release: true
release-refs: main
This is like Simple Release above, but it also checks that pull requests are appropriately labeled.
on:
push:
pull_request:
types: [ synchronize, opened, reopened, labeled, unlabeled ]
jobs:
release:
runs-on: ubuntu-latest
permissions:
contents: write
steps:
- uses: actions/checkout@v3
with:
fetch-depth: 0
- uses: WillAbides/[email protected]
with:
create-release: true
release-refs: main
Create custom release notes. This assumes you have a script called
script/generate-release-notes
that will echo the release notes to stdout.
on:
push:
jobs:
release:
runs-on: ubuntu-latest
permissions:
contents: write
steps:
- uses: actions/checkout@v3
with:
fetch-depth: 0
- uses: WillAbides/[email protected]
with:
create-release: true
release-refs: main
pre-tag-hook: |
set -e
script/generate-release-notes > "$RELEASE_NOTES_FILE"
This assumes you have a script called script/build-release-artifacts
that
creates tar.gz files and checksums.txt in the dist
directory.
on:
push:
jobs:
release:
runs-on: ubuntu-latest
permissions:
contents: write
steps:
- uses: actions/checkout@v3
with:
fetch-depth: 0
- uses: WillAbides/[email protected]
with:
create-release: true
release-refs: main
pre-tag-hook: |
set -e
script/build-release-artifacts
cp dist/checksums.txt dist/*.tar.gz "$ASSETS_DIR"
Sometimes you need to update a file before the release is tagged. For example, to update CHANGELOG.md with the new release notes or write the new version to a version file.
For simplicity, this recipe assumes there are no rules preventing an action from pushing a new commit to the release branch. In reality, you probably have your release branch protected. A later recipe will cover how to do this with a GitHub App that can override branch protection rules.
This recipe updates the file version.txt
with the new release tag.
on:
push:
jobs:
release:
runs-on: ubuntu-latest
permissions:
contents: write
steps:
- uses: actions/checkout@v3
with:
fetch-depth: 0
- name: git config
run: |
git config --local user.name '${{ github.actor }}'
git config --local user.email '${{ github.actor }}@users.noreply.github.com'
- uses: WillAbides/[email protected]
with:
create-release: true
release-refs: main
pre-tag-hook: |
set -e
echo "$RELEASE_TAG" > version.txt
git add version.txt
git commit -m "bump version.txt to $RELEASE_TAG"
git tag "$RELEASE_TAG"
This is the same as the previous recipe, but it authenticates as a GitHub App. It assumes the appropriate secrets are set and uses another action, tibdex/github-app-token, to create a token for the GitHub App.
on:
push:
jobs:
release:
runs-on: ubuntu-latest
permissions:
contents: write
steps:
- uses: tibdex/[email protected]
id: generate-token
with:
app_id: ${{ secrets.RELEASER_APP_ID }}
private_key: ${{ secrets.RELEASER_APP_KEY }}
- uses: actions/checkout@v3
with:
fetch-depth: 0
token: ${{ steps.generate-token.outputs.token }}
- name: git config
run: |
git config --local user.name '${{ github.actor }}'
git config --local user.email '${{ github.actor }}@users.noreply.github.com'
- uses: WillAbides/[email protected]
with:
create-release: true
release-refs: main
github-token: ${{ steps.generate-token.outputs.token }}
pre-tag-hook: |
set -e
echo "$RELEASE_TAG" > version.txt
git add version.txt
git commit -m "bump version.txt to $RELEASE_TAG"
git tag "$RELEASE_TAG"
This is similar to the previous recipe, but instead of pushing the change to the release branch, it creates a pull request with the change. When that PR is merged, the release will be created.
Once again you will need to use a GitHub App, and this recipe assumes you have the appropriate secrets set.
When the content of version.txt is not the same as the release tag, the hook exits with status code 10 to abort the release without causing release-train to error. After that, we use the third party action peter-evans/create-pull-request to create a pull request with the change, but only if the release was aborted. Then we set the newly created pull request to auto-merge.
on:
push:
pull_request:
types: [ synchronize, opened, reopened, labeled, unlabeled ]
jobs:
release:
runs-on: ubuntu-latest
permissions:
contents: write
steps:
- uses: tibdex/[email protected]
id: generate-token
with:
app_id: ${{ secrets.RELEASER_APP_ID }}
private_key: ${{ secrets.RELEASER_APP_KEY }}
- uses: actions/checkout@v3
with:
fetch-depth: 0
token: ${{ steps.generate-token.outputs.token }}
- name: git config
run: |
git config --local user.name '${{ github.actor }}'
git config --local user.email '${{ github.actor }}@users.noreply.github.com'
- uses: WillAbides/[email protected]
id: release-train
with:
create-release: true
release-refs: main
github-token: ${{ steps.generate-token.outputs.token }}
pre-tag-hook: |
set -e
echo "$RELEASE_TAG" > version.txt
# abort if the working tree is dirty
if ! git diff --quiet; then
exit 10
fi
- if: github.event_name == 'push' && steps.release-train.outputs.pre-release-hook-aborted == 'true'
name: create release pr
id: create-release-pr
uses: peter-evans/[email protected]
with:
token: "${{ steps.generate-token.outputs.token }}"
commit-message: "prepare to release ${{ steps.release-train.outputs.release-tag }}"
branch: "prepare-release"
delete-branch: true
title: "prepare to release ${{ steps.release-train.outputs.release-tag }}"
body: "prepare to release ${{ steps.release-train.outputs.release-tag }}"
labels: "semver:none"
- if: steps.create-release-pr.outputs.pull-request-number
name: auto-merge release pr
env:
GH_TOKEN: "${{ steps.generate-token.outputs.token }}"
run: gh pr merge --merge --auto ${{ steps.create-release-pr.outputs.pull-request-number }}
See action.md.
Install with bindown
bindown template-source add release-train https://github.com/WillAbides/release-train/releases/latest/download/bindown.yaml
bindown dependency add release-train --source release-train
go install github.com/willabides/release-train/v3@latest
Pick a release from the releases page and download the appropriate binary for your platform.
Usage: release-train
Release every PR merge. No magic commit message required.
Flags:
-h, --help Show context-sensitive help.
--version
--repo=STRING GitHub repository in the form of owner/repo.
--check-pr=INT Operates as if the given PR has already been merged. Useful
for making sure the PR is properly labeled. Skips tag and
release.
--label=<alias>=<label>;... PR label alias in the form of "<alias>=<label>" where <label>
is a canonical label.
-C, --checkout-dir="." The directory where the repository is checked out.
--ref="HEAD" git ref.
--create-tag Whether to create a tag for the release.
--create-release Whether to create a release. Implies create-tag.
--draft Leave the release as a draft.
--tag-prefix="v" The prefix to use for the tag.
--v0 Assert that current major version is 0 and treat breaking
changes as minor changes. Errors if the major version is not
0.
--initial-tag="v0.0.0" The tag to use if no previous version can be found. Set to ""
to cause an error instead.
--pre-tag-hook=<command> Command to run before tagging the release. You may abort the
release by exiting with a non-zero exit code. Exit code 0
will continue the release. Exit code 10 will skip the release
without error. Any other exit code will abort the release with
an error.
Environment variables available to the hook:
RELEASE_VERSION
The semantic version being released (e.g. 1.2.3).
RELEASE_TAG
The tag being created (e.g. v1.2.3).
PREVIOUS_VERSION
The previous semantic version (e.g. 1.2.2). Empty on
first release.
FIRST_RELEASE
Whether this is the first release. Either "true" or
"false".
GITHUB_TOKEN
The GitHub token that was provided to release-train.
RELEASE_NOTES_FILE
A file path where you can write custom release notes.
When nothing is written to this file, release-train
will use GitHub's default release notes.
RELEASE_TARGET
A file path where you can write an alternate git ref
to release instead of HEAD.
ASSETS_DIR
A directory where you can write release assets. All
files in this directory will be uploaded as release
assets.
In addition to the above environment variables, all variables
from release-train's environment are available to the hook.
When the hook creates a tag named $RELEASE_TAG, it will be
used as the release target instead of either HEAD or the value
written to $RELEASE_TARGET.
--pre-release-hook=<command> *deprecated* Will be removed in a future release. Alias for
pre-tag-hook.
--release-ref=<branch>,... Only allow tags and releases to be created from matching refs.
Refs can be patterns accepted by git-show-ref. If undefined,
any branch can be used.
--push-remote="origin" The remote to push tags to.
--tempdir=STRING The prefix to use with mktemp to create a temporary directory.
--github-api-url="https://api.github.com"
GitHub API URL.
--output-format="json" Output either json our GitHub action output.
--debug Enable debug logging.