Enums with >32 bits on i686 are all canonicalized to 0

[alexcrichton]

The following test fails on i686, and passes on x86_64

enum A {
  A1 = 1 << 41,
}

fn main() {
  assert_eq!(A1 as u64, 1 << 41);
}

This is blocking #7115 until I have time to make a workaround.

[huonw]

@kemurphy is working in this area.

[kemurphy]

Oh boy, this'll be fun. My big patch is nearing completion but I'm going to be traveling all day tomorrow so it still won't be done for a few days I bet.

[catamorphism]

Reproduced on i686 with cf7b9eb . @kemurphy -- what do you need in order for your patch to be ready for submission?

[kemurphy]

My full-time job is keeping be busy enough to not have a whole lot of time to work on this, but the big patch I was working on ended up not working out anyway. I was trying to unify the representation of structs, tuple structs, and all the enum variants behind the scenes (so they all followed one path on their way to trans), but tuple structs and variants have a special case where they can get passed around as callable values and that conflicted with what I was trying to do. I'm planning on trying again with a different design that doesn't nuke the callable stuff but due to lack of time it'll be a while before it lands. Consequently, if someone else wants to pick up this bug they should go for it.

[catamorphism]

@kemurphy No worries. Thanks for trying and thanks for the update!

[alexcrichton]

Closed by #9613, tests appear to be sufficient.