Glob importing causes unrelated compiler error

[dylanmckay]

When using globs to import everything from a module where the module itself imports things from the global namespace causes an error:

error: unresolved import
...
error: unresolved import (maybe you meant `std::*`?)

That is to say, if some_mod.rs glob imports other_mod and other_mod imports something from :: (say, std), the module cannot be found. If the glob import is replaced by manually importing every type, compilation succeeds.

The error occurs on the line in the imported module that tries to import the module.

Test Case (link to Playpen):

#![feature(globs)]

// Import everything from `test_module`
// This causes an error.
use self::test_module::*;

// Import what we want from `test_module`
// This succeeds.
// TODO: uncomment this line
//use self::test_module::Dummy;


mod test_module
{
    // Fails on this line, but only if `test_module` is glob imported.
    // Compilation is successful if globs are not used.
    use std;

    pub struct Dummy;
}

fn main() { }

Several months ago when I started learning Rust, this was actually the error that made modules so confusing to me in the beginning - I couldn't understand why sometimes (apparently when I used globs) I couldn't import std but instead I had to import every function and trait I wanted to use into my module's namespace.

[nrc]

Interestingly, the use std; doesn't even have to be in the submodule

#![feature(globs)]
use std;
use self::test_module::*;
mod test_module {}

fn main() { }

Gives the same error, but commenting out use self::test_module::*; gives a conflict error for std instead (which I think is correct).

[jpernst]

I've been hitting this bug recently as well. Globs seem to interact strangely with non-public use statements. After some more experimentation I've reduced my situation to this:

pub mod a {
    pub use self::aa::*;

    pub mod aa {
        use b::YY; // comment out this...

        pub struct XX;
    }
}

pub mod b {
    pub use self::bb::*;

    pub mod bb {
        use a::XX; // ...or this

        pub struct YY;
    }
}

If either of the marked 'use' lines are commented out, it will compile, but will fail with unresolved import errors if both are present. If the globs are replaced by named imports, it compiles fine.
rustc version:

rustc 1.0.0-nightly (ea6f65c5f 2015-01-06 19:47:08 +0000)
binary: rustc
commit-hash: ea6f65c5f1a3f84e010d2cef02a0160804e9567a
commit-date: 2015-01-06 19:47:08 +0000
host: x86_64-unknown-linux-gnu
release: 1.0.0-nightly

[Marckvdv]

I might have found another variant of this error:

pub mod a {
    use ::b::B;
    use std::*;
    pub struct A {
        i: isize
    }
}
pub mod b {
    use ::a::A;
    use std::*;
    pub struct B {
        i: isize
    }
}

[alexcrichton]

This has since been fixed, yay!

[gkoz]

As it doesn't seem to be fixed in beta (at least the one on play.rlo), the earliest release to get the fix would be 1.4?

[alexcrichton]

Yeah I think this landed relatively recently on nightly, so it should make its way into 1.4

[eddyb]

Is this the cyclic globs issue? Does this mean we can finally clean up the import mess in rustc_trans?