• Only services require AWS. Clients don't.
• All services require DDB (I think so)
• Almost all services require secrets manager (feature flags don't)

Initially I thought about extracting this feature to ddb / database / dynamodb and separating out aws-sdk-secretsmanager.

But you merged them here and perhaps that makes sense. Curious about your reasoning here

Don't you need to use this mod in clients? Everything compiles though, I got confused 🤨

I don't really have a strong reason, this was just a simplest solution. Before these changes all services included all of aws dependencies so I didn't thought there was a reason to split them up.

But we definitely could have something like:

[features]
aws = ["dep:aws-config", "dep:aws-types"]
dynamodb = ["aws", "aws-sdk-dynamodb"]
secretsmanager = ["aws", "aws-sdk-secrets-manager"]

but I don't feel like we gain much from it.

To use trait methods you only need to import the trait itself (assuming that's what you meant), the actual implementation can be anywhere. If we defined a trait in db_conversions than we would have to import it.

Simplified example:

trait Trait {
    fn trait_fn(){}
}
struct A;

mod inner {
    use super::{A, Trait};
    impl Trait for A {}
}

mod client {
    use super::{A, Trait};
    fn _run_client() {
        A::trait_fn();
    }
}