ICE "thread 'rustc' has overflowed its stack" on Windows (1.5 nightly x86_64 msvc toolchain)

[randomPoison]

I've got the following trait and impl:

pub trait Dot<Other=Self> {
    type Output;

    fn dot(self, rhs: Other) -> Self::Output;
}

impl<'a, T> Dot<&'a T> for T where T: Dot<T> + Copy {
    type Output = T::Output;

    fn dot(self, rhs: &T) -> Self::Output {
        self.dot(*rhs)
    }
}

When I change the impl to:

impl<'a, T, U> Dot<&'a U> for T where T: Dot<U>, U: Copy {
    type Output = T::Output;

    fn dot(self, rhs: &U) -> Self::Output {
        self.dot(*rhs)
    }
}

I get the error thread 'rustc' has overflowed its stack.

This is happening on the current nightly build (10/10/2015) with the x86_64 Windows MSVC toolchain. I tried to reproduce the error with just the above code but that didn't cause the error, but compiling the entire crate it's in did. The full crate (including commented out code that causes error) can be found here, the code that causes the error is in src/lib.rs.

Misc. Notes

• Using the 1.5 nightly build (10/10) of the x86_64 MSVC toolchain.
• Running Windows 10.

[apasel422]

I can't reproduce on

rustc 1.5.0-nightly (1210fb9bc 2015-10-23)
binary: rustc
commit-hash: 1210fb9bc61b351cb0441cc48dac9e18987fb88b
commit-date: 2015-10-23
host: x86_64-unknown-linux-gnu
release: 1.5.0-nightly

[steveklabnik]

If anyone needs help adding a test for this, please let me know!

[apasel422]

I guess I missed the part about this occurring only when compiling the entire crate. I'm getting an error when I try to build the whole repo, so I'm not sure if this was actually fixed.

[randomPoison]

@steveklabnik I would love to put together a better test case for this! Just let me know what I need to do :)

[steveklabnik]

@excaliburHisSheath so, the basic idea is to put a file with the program into https://github.com/rust-lang/rust/tree/master/src/test/run-pass . Everything in that directory should compile successfully, so by just copying it in, we can test it.

However, @apasel422 seems to suggest that we haven't actually figured out a small reproduction, it requires the whole crate. So, for now, I'm going to remove those tags, as that's what we need first before we can do that.

[randomPoison]

Cool, I'll work on paring down the case until I have a minimal example.

[randomPoison]

As of the latest nightly (Rust 1.6 MSVC 10/31/2015) this no longer causes an ICE, instead giving the following compile error:

src\vector.rs:91:15: 91:24 error: overflow evaluating the requirement `<&vector::Vector3 as Dot<_>>::Output` [E0275]
src\vector.rs:91         (self.dot(self) - 1.0).is_zero()
                               ^~~~~~~~~
src\vector.rs:91:15: 91:24 help: run `rustc --explain E0275` to see a detailed explanation
src\vector.rs:91:15: 91:24 note: consider adding a `#![recursion_limit="128"]` attribute to your crate

If it would still be useful to have a reproducible case as a regression test I can still put one together with an older version of the compiler.