moon v1.28 - Task presets, OS tasks, meta tokens, and more

In this release, we've focused on developer experience and quality-of-life improvements.

New task presets

For the longest time, moon has supported the local setting for tasks. This setting was added early on to reduce the amount of options configuration necessary for tasks that "should only run locally". However, after years of supporting this setting, we've discovered a few short-comings with it:

• The name is confusing. It's not apparent what "local" really means for newcomers.
• This setting also changes non-local related options, like persistent and cache, which isn't apparent to the user.
• Setting the value to false doesn't inverse the options values. For example, enabling runInCI.

At this point we thought to ourselves "how can we improve this experience?". Instead of reworking local, we've opted to introduce a new task setting called preset, which configures a collection of task options, and deprecate the local setting. Presets are a much better pattern, as it allows us to support multiple presets, and uniquely name our presets. For example, old local tasks can be rewritten as such:

moon.yml

# Before
tasks:
  start:
    command: 'vite dev'
    local: true

# After
tasks:
  start:
    command: 'vite dev'
    preset: 'server'

To start, we're introducing 2 presets: server and watcher. Check out the documentation for more information on what these presets configure! If you have ideas for other presets, we would love to hear them.

New OS specific tasks

When moon was originally designed, it was meant to be a very simple but easily configurable task runner. But since then, moon has grown drastically, and so has the requirements of its users. One such request was being able to define tasks that only run for certain operating systems.

This wasn't possible before; but is now thanks to the new os task option. This option can be configured with "linux", "macos", or "windows" to target those specific systems, and can be a single value, or list of values.

moon.yml

tasks:
  build-linux:
    command: 'cargo build --target x86_64-unknown-linux-gnu'
    options:
      os: 'linux'
  build-macos:
    command: 'cargo build --target x86_64-apple-darwin'
    options:
      os: 'macos'
  build-windows:
    command: 'cargo build --target x86_64-pc-windows-msvc'
    options:
      os: 'windows'

When a task is ran on a non-compatible system, it will be converted to a no-operation task, so that dependency chains still resolve correctly.

In the future, we will look into supporting system architecture as well.

New @meta token function

In our last release, we introduced the project.metadata setting as a means for consumers to define their own metadata. This is great but it provided no other benefit, as it required consumers to open up config files to discover this metadata, or consumers to write custom abstractions/tooling. We want to improve the integration of this information within moon, so to start, we're introducing a new token function @meta(key).

The @meta(key) token function can be used to access project metdata from both the project and project.metadata settings. The former values will be used as-is, while the latter values will be JSON stringified (since consumers can define anything).

With this new token, our task composition has grown even deeper. For example, you can define a task at the top-level that a project inherits, and then define metadata within that project for the task! With that said, this is more of a stopgap solution until we support true variables within configuration.

.moon/tasks.yml

tasks:
  build:
    script: 'compile --label @meta(name) --compress @meta(compression)'

<project>/moon.yml

project:
  name: 'example'
  metadata:
    compression: 9

And lastly, while we're on the topic of tokens, we're also introducing $projectName, $projectChannel, and $projectOwner token variables, as an easier alternative solution.

New disallowRunInCiMismatch experiment

Have you ever ran moon ci in CI and encountered a scenario where tasks that should have run didn't run? Or the dependency of a task never runs causing the dependent to fail? If so, this may apply to you!

This scenario is most likely caused by an invalid relationship where the dependent is runInCI enabled, but the dependency is runInCI disabled. For example:

moon.yml

tasks:
  build:
    # ...
    options:
      runInCI: false

  analyze:
    # ...
    deps: ['build']
    options:
      runInCI: true

Because build is disabled, its outputs may not be created. Additionally, if build is affected by touched files, the analyze task will not run automatically unless it was also affected. We try our best to accomdate these scenarios and "detect" when things should run or not, but it's fallible, as there's many ways that tasks can relate to each other.

To try and remedy this going forward, we're introducing the new disallowRunInCiMismatch experiment that will hard error if an invalid relationship is configured. We want to try and capture how often users are configuring this, whether intentionally or accidentally. If the error becomes a blocker, the experiment can be disabled as so:

.moon/workspace.yml

experiments:
  disallowRunInCiMismatch: false

Other changes

View the official release for a full list of changes.

• We've sunset the CI insights feature in moonbase. We will no longer be tracking CI run history. This is a retroactive change that applies to all moon versions.
• Added codeowners.requiredApprovals setting to .moon/workspace.yml, that will apply requirements to all paths.
• Updated the terminal checkpoint (4 squares) colors to better reflect their state.
• Updated MOON_TOOLCHAIN_FORCE_GLOBALS to support filtering by tool IDs, instead of applying to everything.