Submodules cannot access private members of parents

[brson]

mod foo {
    mod bar {
        // This doesn't resolve, nor does foo::S
        use super::S;

        fn make() -> S { fail }
    }

    priv struct S(());
}

fn main() { }

[bstrie]

Still reproduces:

⌁ rustc 4536.rs
4536.rs:4:12: 4:21 error: failed to resolve import: super::S
4536.rs:4         use super::S;
                      ^~~~~~~~~
4536.rs:5:12: 5:19 error: failed to resolve import: foo::S
4536.rs:5         use foo::S;
                      ^~~~~~~
error: failed to resolve imports
error: aborting due to 3 previous errors

[graydon]

looks like "just" a bug, de-milestoning

[graydon]

hm. this is part of a pattern. re-milestoning and collecting into a metabug.

[emberian]

Still reproduces:

[17:03:53]/tmp> rustc foo.rs
foo.rs:4:12: 4:20 error: unresolved import: found `S` in `foo` but it is private
foo.rs:4         use super::S;
                     ^~~~~~~~
foo.rs:4:12: 4:20 error: failed to resolve import `super::S`
foo.rs:4         use super::S;
                     ^~~~~~~~

Is this desired behavior? Should children be able to access its parent's privates? Should this extend to all ancestor modules?

If so, it should be easy to fix I think. For each resolve that "fails" due to privacy, check if the module is a child of wherever it resolved to, and if so, don't actually fail the resolve.

This happens in resolve_single_import, which is a huge function in resolve.rs. Should probably be handled in/around src/librustc/middle/resolve.rs:2434 (handily, I refactored this code to enable this enabling the "it's private" error).

It shouldn't affect performance much, I don't think, especially since it's so rare (the error path is hit very rarely afaict).

[pnkfelix]

linking to #6143 for unified tracking of resolve

[catamorphism]

Dup of #5238