enso/docs/distribution/release-policy.md

layout	title	category	tags	order
developer-doc	Release Policy	distribution	distribution release release-policy policy	3

Release Policy

As an open-source project and programming language, it is incredibly important that we have a well-defined release policy. This document defines said policy.

Once a release has been made it is immutable. The release should only ever be edited to mark it as broken. Nothing else should ever be changed.

No two release workflows can be running at once, to avoid race conditions since releases must update files in S3. Make sure that tags which trigger release builds are pushed sequentially, only pushing the next one after the previous build has finished.

• Versioning
  • Launcher Versioning
• Release Branches
• Release Workflow
  • Tag Naming
  • Manifest Files
  • Breaking Changes to Launcher Upgrade
  • GitHub Releases
  • Release Notes
• Version Support
• Working on the Current Release
• Backporting Fixes

Versioning

Releases of Enso are versioned using semantic versioning. Where a.b.c-tag is the version string, a is the major version, b, is the minor version, c is the patch version, and tag is additional metadata, the following hold:

• Breaking changes to language behaviour or the public API will result in a major version increase.
• Addition of functionality in a backwards-compatible manner will result in a minor version increase.
• Backwards-compatible bug fixes will result in a patch version increase.
• The tag will indicate pre-release or beta versions, and will increase when any pre-release change is made. These are not intended to be stable.

Launcher Versioning

The launcher is released alongside Enso releases, so the launcher version is tied to the Enso version that it is released with.

Release Branches

A release branch in the Enso repository is a branch prefixed with release/. Release branches obey the following rules:

• One release branch exists per major version, and is named release/n.x, where n is the major version, and the rest is literal.
• A release branch must contain tags corresponding to released versions of Enso. Once a release has been made, no further changes may be made to that release.
• A tagged release must contain a RELEASES.md file that describes the changes contained in that release.

It should be noted that general development still takes place on the main branch of the repository.

Release Workflow

Cutting a release for Enso proceeds as follows:

1. If no release branch exists for the current major version, one should be created.
2. Create a branch called wip/<initials/release-bump. On this branch, ensure that the release notes are up to date in RELEASES.md (follow the existing format), and that the new version number has been set in build.sbt. This version should not contain SNAPSHOT.
3. Open a PR for this branch into main.
4. Once the changes have been reviewed, merge the PR into main (getting commit hash xxxxxxx). The message should be Prepare for the $version release.
5. Immediately push a commit to main that updates the version in build.sbt to the new snapshot version. If unclear, bump the patch version by one and append -SNAPSHOT (e.g. 0.2.10 becomes 0.2.11-SNAPSHOT). The message should be Bump the snapshot version.
6. Find the commit hash of the last "Bump the snapshot version" commit. Let's say this is yyyyyyy.
7. Run a rebase --onto the release branch from the yyyyyyy commit to the xxxxxxx commit. For example:

git rebase --onto origin/release/0.x yyyyyyy~1 xxxxxxx

8. This will put you into a "detached HEAD" state at commit zzzzzzz, so you need to make a new branch: git branch release-update zzzzzzz
9. This new branch is a fast-forward merge away from the release branch. Check out the release branch and then fast-forward merge release-update into it. For example:

git checkout release/0.x
git merge --ff-only release-update

10. As long as the fast-forward proceeds cleanly, you can push the updated release branch to the origin.
11. Create a tag for the commit at the HEAD of the release branch. It should be named as above. For example:

git tag --sign enso-0.2.11

12. Push the tag to the remote. This will start the release build.
13. CI will create a draft release for this tag, as well as build and upload the appropriate artefacts. Do not create a release for this tag manually.
14. The release notes for the version being released should be copied from the new section in RELEASES.md into the GitHub release description with the line breaks removed.
15. The title of the release should be Enso Engine <version> (e.g. Enso Engine 0.2.11).
16. The release must be verified by at least one member of the engine team and the QA team.
17. Once verification has been performed, the release can be published. It should not be a pre-releases as we reserve these for nightly builds.

Tag Naming

Tags for releases are named as follows enso-version, where version is the semver string (see versioning) representing the version being released.

Manifest Files

Manifest files are used to describe metadata about various releases for use by the Enso tooling.

Engine Manifest

Each GitHub release contains an asset named manifest.yaml which is a YAML file containing metadata regarding the release. The manifest is also included in the root of an Enso version package. It has at least the following fields:

• minimum-launcher-version - specifies the minimum version of the launcher that should be used with this release of Enso,
• minimum-project-manager-version - specifies the minimum version of the project manager that should be used with this release of Enso; currently it is the same as the launcher version but this may change in the future,
• graal-vm-version - specifies the exact version of GraalVM that should be used with this release of Enso,
• graal-java-version - as GraalVM versions may have different variants for different Java versions, this specifies which variant to use.

The minimum launcher and project manager versions are kept as separate fields, because at some point the same runtime version management logic may be associated with different versions of these components.

It can also contain the following additional fields:

• jvm-options - specifies a list of options that should be passed to the JVM running the engine. These options can be used to fine-tune version specific optimization settings etc. Each option must have a key called value which specifies what option should be passed. That value can include a variable $enginePackagePath which is substituted with the absolute path to the root of the engine package that is being launched. Optionally, the option may define os which will restrict this option only to the provided operating system. Possible os values are linux, macos and windows.
• broken - can be set to true to mark this release as broken. This field is never set in a release. Instead, when the launcher is installing a release marked as broken using the broken file, it adds this property to the manifest to preserve that information.

For example:

minimum-launcher-version: 0.0.1
minimum-project-manager-version: 0.0.1
jvm-options:
  - value: "-Dpolyglot.engine.IterativePartialEscape=true"
  - value: "-Dtruffle.class.path.append=$enginePackagePath\\component\\runtime.jar"
    os: "windows"
  - value: "-Dtruffle.class.path.append=$enginePackagePath/component/runtime.jar"
    os: "linux"
  - value: "-Dtruffle.class.path.append=$enginePackagePath/component/runtime.jar"
    os: "macos"
graal-vm-version: 20.2.0
graal-java-version: 11

The minimum-launcher-version should be updated whenever a new version of Enso introduces changes that require a more recent launcher version. This value is stored in distribution/manifest.template.yaml and other values are added to this template at build time.

Launcher Manifest

Additionally, each release should contain an asset named launcher-manifest.yaml which contains launcher-specific release metadata.

It contains the following fields:

• minimum-version-for-upgrade - specifies the minimum version of the launcher that is allowed to upgrade to this launcher version. If a launcher is older than the version specified here it must perform the upgrade in steps, first upgrading to an older version newer than minimum-version-for-upgrade and only then, using that version, to the target version. This logic ensures that if a newer launcher version required custom upgrade logic not present in older versions, the upgrade can still be performed by first upgrading to a newer version that does not require the new logic but knows about it and continuing the upgrade with that knowledge.
• files-to-copy - a list of files that should be copied into the distribution's data root. This may include the README and similar files, so that after the upgrade these additional files are also up-to-date. These files are treated as non-essential, i.e. an error when copying them will not cancel the upgrade (but it should be reported).
• directories-to-copy - a list of directories that should be copied into the distribution's data root. Acts similarly to files-to-copy.

A template manifest file, located in distribution/launcher-manifest.yaml, is automatically copied to the release. If any new files or directories are added or a breaking change to the upgrade mechanism is being made, this manifest template must be updated accordingly.

Breaking Changes to Launcher Upgrade

If at any point the launcher's upgrade mechanism needs an update, i.e. additional logic must be added that was not present before, special action is required.

First, the additional logic has to be implemented and a new launcher version should be released which includes this additional logic, but does not require it yet. Then, another version can be released that can depend on this new logic and its minimum-version-for-upgrade has to be bumped to that previous version which already includes new logic but does not depend on it.

This way, old launcher versions can first upgrade to a version that contains the new logic (as it does not depend on it yet, the upgrade is possible) and using that new version, upgrade to the target version that depends on that logic.

GitHub Releases

A release is considered official once it has been made into a release on GitHub. Once official, a release may not be changed in any way, except to mark it as broken.

Release Assets Structure

Each release contains a build of the Enso engine and native launcher binaries for each supported platform. Moreover, for convenience, it should include bundles containing native launcher binaries and the latest engine build for each platform. So each release should contain the following assets:

• enso-bundle-<version>-linux-amd64.tar.gz
• enso-bundle-<version>-macos-amd64.tar.gz
• enso-bundle-<version>-windows-amd64.zip
• enso-engine-<version>-linux-amd64.tar.gz
• enso-engine-<version>-macos-amd64.tar.gz
• enso-engine-<version>-windows-amd64.zip
• enso-launcher-<version>-linux-amd64.tar.gz
• enso-launcher-<version>-macos-amd64.tar.gz
• enso-launcher-<version>-windows-amd64.zip
• manifest.yaml

Marking a Release as Broken

We intend to never delete a release from GitHub, as users may have projects that depend on specific versions of Enso. Instead, we provide a mechanism for marking releases as broken that works as follows:

• An empty file named broken is uploaded to the release.
• The release description is edited to visibly mark the release as broken.

A broken release is one that must not be downloaded by the launcher unless a project specifies an exact version match, and it must not be used in new projects by the launcher unless explicitly specified by the user as an exact version match.

When the release is marked as broken at GitHub, a GitHub Actions Workflow is triggered that also updates the release in the fallback mechanism. Given its current implementation is prone to race conditions when updating releases, the broken file should be added to releases one by one, making sure that only one update workflow is running at the same time and that no release workflows are running in parallel with it.

In an unusual situation in which you want to upload a release that is marked as broken from the start, you should first publish it in a non-broken state and only mark it as broken after publishing. That is because the GitHub Workflow that will persist the broken mark to S3 is not triggered for release drafts.

When marking the release as broken, you should make sure that the workflow persisting the broken mark to Se has succeeded and re-run it if necessary.

Release Notes

Release notes should contain a summary of the changes made between the last release and the current release. They should follow the template given below, and are contained in the RELEASES.md file in the repository root.

# Enso x.y.z (YYYY-MM-DD)

## Language

- A list of language-level changes.

## Interpreter/Runtime

- A list of changes to the Enso interpreter.

## Type System

- A list of type-system changes.

## Tooling

- A list of changes to the Enso language tooling.

## Libraries

- A list of changes to the Enso core libraries.

## Stabilised Features

- A list of stabilised APIs and/or features.

## Misc

- A list of miscellaneous changes.

## Internal Only

- A list of changes that do not have user-facing impact, but represent
  significant improvements to the internals of Enso and related tools.

If there are no changes for a section, the section may be removed.

The releases file is an ongoing record of changes, and may diverge between main and the various release branches.

Version Support

We aim to support a given major version for some period of time after the release of the next major version. For a detailed breakdown of the major versions that are currently supported, please see the security document.

Working on the Current Release

When working on the current release, development should take place against the main branch. When it is time to cut a release, the new commits on the main branch are cherry-picked onto the current release branch. From there, the release proceeds as described in release workflow above.

Backporting Fixes

Supporting a major version for some time after the release of the next major version will sometimes require backporting a fix to the previous major version from the current version or from main.

Backporting should only be used for applying fixes, not the addition of new features.

The process for performing such a backport is as follows:

1. Create a new branch called backport/version/fix-name, where version matches the version string of the corresponding release branch. This branch should branch off the corresponding release branch.
2. Back-port the fix to the newly created backport branch. This can be done by:
   • Cherry-picking the commit and performing fixups (preferred).
   • Re-implementing the fix manually (if cherry-picking will not work due to progression of the codebase).
3. Submit your backport/version/fix-name branch for review as a pull-request into the release/version branch.
4. Once the PR has passed CI and been approved by the appropriate reviewers, it can be merged into the release branch.