I am currently digging in some CI scripts using github actions. They seem to be installing elixir/erlang using asdf and cache them afterwards. It seems they work pretty well currently, however this makes me question whether it is smart to compile OTP from source in a CI environment instead of using a pre-built image?

There are few clear benefits to the asdf approach:

1. You can use .tools-versions to keep the versions for both dev and CI envs;
2. You don’t have to rely on pre-built images availability, this can be both for older versions of OTP or bleeding edge ones.

My main concerns when I saw usage of asdf is: how controllable the env where OTP is compiled is? I can see how this can easily backfire when you want to use a older OTP version that doesn’t support openssl 3+.

I’ve took a look at some OSS projects like phoenix and it seems they are using erlef/setup-beam@v1 usually:

- name: Set up Elixir
  uses: erlef/setup-beam@v1
  with:
    elixir-version: ${{ matrix.elixir }}
    otp-version: ${{ matrix.otp }}

Any thoughts on this? My aim is to find the version of config that is the easiest to use and configure.