Cirrus Config via Starlark

[fkorotkov]

YAML is pretty limiting and doesn't allow easily to have custom logic and makes it harder to be concise in some cases.

It will be great to have an option to generate tasks via .cirrus.star written in Starlark.

def main(ctx):
  return {
    "name": "Test Task",
    "instance": container("alpine:latest")
    "steps": [
      clone(),
      script(name = "env",  content = "printenv")
    ]
  }

ctx will include ctx.env for getting environment variables and ctx.functions to invoke helper functions while generating a set of tasks.

[fkorotkov]

It will be also possible to expose helper functions like fileExists(relativePath) and fileContent(relativePath) for even more advanced customization like generating a lint task only if lint config is presented. This will make it easy to start sharing common configs like described in cirruslabs/cirrus-ci-docs#106

[RDIL]

To be honest, I'm a fan of this approach, except for the proposed language. I feel like new users shouldn't need to learn a whole new language just to do advanced configuration. I feel like Python would work much better for this.

[fkorotkov]

Starlark is a deterministic subset of Python basically. 👌

[RDIL]

Doesn't look too much like python :(

[RDIL]

Actually it does, but that is awful following of EP 8

[fkorotkov]

Plus, let's give the ability to call GitHub APIs via ctx.github and let's evaluate the script for every GitHub Webhook Event. That way it will be possible to also write something like GitHub Actions but because of the reproducibility and ability to run Starlark in-process securely this "Cirrus Actions" will be instant and will cost nothing until the output of the evaluation is a Cirrus Build.