enum struct variants can mask regular struct constructors

[erickt]

@thestinger found this odd bug. This code:

enum Foo { C { a: int, b: int } }
struct C { a: int, b: int }
fn main() { }

properly errors out with:

test.rs:2:0: 2:27 error: duplicate definition of type C
test.rs:2 struct C { a: int, b: int }
          ^~~~~~~~~~~~~~~~~~~~~~~~~~~
test.rs:1:11: 1:31 note: first definition of type C here:
test.rs:1 enum Foo { C { a: int, b: int } }
                     ^~~~~~~~~~~~~~~~~~~~

However, if you swap the enum and struct definition, it compiles fine with rust HEAD:

struct C { a: int, b: int }
enum Foo { C { a: int, b: int } }
fn main() { }

[pnkfelix]

still reproduces. Looks like there are a couple bugs here:

• The error messages talk about duplicate type definitions, but the enum is not declaring a type C; it is declaring a constructor named C, which belongs to a different namespace.
• On a related note: I'm not sure yet whether our intention is to allow the shadowing of the constructor names (which means we should not be signaling an error in either case), or if the bug really is as described in the description.

[sanxiyn]

See #7044.

[luqmana]

So, the reason why swapping the order works is because in resolve struct C { ... } by itself checks for any duplicate names, but struct C { ... } as an enum variant doesn't.

Also, the reason why the error messages refer to duplicate type definitions is because they're added to the Type namespace as opposed to the Value namespace. (Though, new type and unit structs are added to both the type and value namespaces.)

[luqmana]

Oh and rustc doesn't actually consider struct C { ... } as having a constructor. (i.e. the ctor_id field on the struct_def is None).

[pnkfelix]

I came out of the meeting from 2013-06-18 thinking that we might want to allow the shadowing here, but the meeting notes don't actually reflect much/any justification for that conclusion that I can see. I guess its good to be consistent, in any case.

[luqmana]

Oh, I had thought the consensus was it should be an error given pcwalton's comment.