perf: dont run shell if no activation scripts

[baszalmstra]

When there are no activation scripts, the Activation::run_activation function will still start a shell to evaluate all the environment variables. I think we can forgo this step and just return the activation variables directly.

rattler/crates/rattler_shell/src/activation.rs

Line 617 in 498ab0c

pub fn run_activation(

[LED-0102]

Hi! Can I work on this issue?

[baszalmstra]

Go for it!

[LED-0102]

Hi @baszalmstra,

I drafted this early-return for empty activation scripts:

let ret = variables.current_env.clone();
let activation_script = self.activation(variables)?.script;

if activation_script.contents()?.trim().is_empty() {
    return Ok(ret);
}

This would skip the shell when the script is empty and return the current activation variables.

Does this match the optimization you had in mind?

[baszalmstra]

Jep! Just needs some tests!

[baszalmstra]

actually its not quite it because the activation script will also set variables. So it will never be empty. You just have to check if the activation script would source others.

[LED-0102]

Hi @baszalmstra,

Thanks for clarifying! From reading through activation(), it looks like the script only “sources” others in two cases:

1. When variables.conda_prefix is set and we need to run deactivation scripts.

2. When self.activation_scripts is non-empty.

If both are empty, the script just sets variables like CONDA_PREFIX, CONDA_SHLVL, $PATH, and static env vars.

In that case, should I skip the shell and deterministically return those variable changes as the diff (instead of just returning variables.current_env)?

[LED-0102]

Hi @baszalmstra,

I’m wondering about the “before” values of environment variables like CONDA_SHLVL, CONDA_PREFIX, PATH, etc. I think normally these are captured by the shell, so without running it, I’m not sure how we can know what they actually were.

In this situation, I’m not sure what the correct approach is: should we just return variables.current_env, or should we try to return variables.current_env plus the deterministic changes applied by activation() even if they might be identical to the “before” values?

I think some guidance here would help clarify the intended behavior.