`#![no_implicit_prelude]` should only apply to the current module, not to all children

[Kimundi]

Currently, the #![no_implicit_prelude] attribute can be used to disable the libstd prelude imports for a module. However, it seems to disable them for all submodules as well, which makes no sense seeing how the prelude is just a bunch of use statements, and those don't propagate into child modules.

Example:

fn a() {
    let _: Vec<usize> = vec![];
}

mod foo {
    #![no_implicit_prelude]

    fn b() {
        let _: Vec<usize> = vec![];
    }

    mod bar {
        fn c() {
            let _: Vec<usize> = vec![];
        }   
    }
}

fn main() {}

... which results in an error for the bar module as well:

<anon>:9:16: 9:25 error: use of undeclared type name `Vec`
<anon>:9         let _: Vec<usize> = vec![];
                        ^~~~~~~~~
<anon>:14:20: 14:29 error: use of undeclared type name `Vec`
<anon>:14             let _: Vec<usize> = vec![];
                             ^~~~~~~~~

As a fix, #![no_implicit_prelude] should be changed to only apply to the module it is applied on.

[huonw]

I originally implemented it with these semantics because it seemed reasonable/common* for someone to not want the prelude in their entire crate (e.g. if they're using a custom prelude), and having to type #[no_implicit_prelude] in each and every module seems annoying.

However, I have no evidence for this claim.

* Common among uses of #[no_implicit_prelude] anyway.

[Kimundi]

@huonw Well, as a contra point, I just had to manually import prelude items in like 5 submodules because of that , so I personally think it should be changed :)

I'd understand it the attribute where a crate-root only attribute but the current semantic is neither here nor there imo.

[Kimundi]

About avoiding writing that attribute all the time: Now that macros can expand to multible items, something like prelude!{foo::bar} that expands to #![no_implicit_prelude] use foo::bar::*; should be possible, right?

[steveklabnik]

Triage: this still works this way today.

[Nemo157]

I agree that disabling for the whole crate seems like it would be the common case (as a contra contra point it's how I want to use it 😛).

If it does get changed then it would be nice to be able to apply it as a crate attribute as well that disables the prelude throughout the crate (although, I guess that would mean not being able to apply it to just the root module, maybe it would need to be a different name for the crate only attribute?).

[Mark-Simulacrum]

From RFC 501:

Make name and behavior of the #![no_std] and #![no_implicit_prelude] attributes consistent by renaming the latter to #![no_prelude] and having it only apply to the current module.

Once #20561 is implemented, this will no longer be an issue, since the #![no_prelude] attribute will apply only to the current module. As such, this isn't tracking anything more than that is, so I'm going to close this out.