Impls should not allow methods not defined in the trait to be implemented

[mitsuhiko]

This code is wrong because ToStr does not define new. But what's odd is that the compiler error does not happen in the ToStr block but in the main:

struct Point {
 mut x: float,
 mut y: float,
}

impl Point : ToStr {
 static fn new(x: float, y: float) -> Point {
 Point { x: x, y: y }
 }

 pure fn to_str() -> ~str {
 fmt!("(%f, %f)", self.x, self.y)
 }
}

fn main() {
 let p = Point::new(0.0f, 0.0f);
 io::println(p.to_str());
}

Error message:

test.rs:17:12: 17:22 error: unresolved name
test.rs:17 let p = Point::new(0.0f, 0.0f);
 ^~~~~~~~~~
test.rs:17:12: 17:22 error: use of undeclared module `Point`
test.rs:17 let p = Point::new(0.0f, 0.0f);
 ^~~~~~~~~~
test.rs:17:12: 17:22 error: unresolved name: Point::new
test.rs:17 let p = Point::new(0.0f, 0.0f);
 ^~~~~~~~~~
error: aborting due to 3 previous errors

The expected error message would point to the impl block that needs to be split into two (one with ToStr and one without a trait).

[catamorphism]

Reproduced as of ae0ca9c

[catamorphism]

Repro'd with 373504 -- but not critical for 0.6, de-milestoning

[bstrie]

Appears to be fixed, as per this updated test case:

struct Point {
 x: float,
 y: float,
}

impl ToStr for Point {
 fn new(x: float, y: float) -> Point {
 Point { x: x, y: y }
 }

 fn to_str(&self) -> ~str {
 fmt!("(%f, %f)", self.x, self.y)
 }
}

fn main() {
}

...which fails with error: method new is not a member of trait ToStr. Closing.

[huonw]

We've regressed: there's a test case (compile-fail/issue-3973.rs), but it's been xfailed since it was added. With 6a3dd30 I see:

issue-3973.rs:30:12: 30:22 error: unresolved name
issue-3973.rs:30 let p = Point::new(0.0f, 0.0f);
 ^~~~~~~~~~
issue-3973.rs:30:12: 30:22 error: use of undeclared module `Point`
issue-3973.rs:30 let p = Point::new(0.0f, 0.0f);
 ^~~~~~~~~~
issue-3973.rs:30:12: 30:22 error: unresolved name `Point::new`.
issue-3973.rs:30 let p = Point::new(0.0f, 0.0f);
 ^~~~~~~~~~
error: aborting due to 3 previous errors

[alexcrichton]

Aaaand, we've un-regressed!

Removing all tags except for neesdtest.

[alexcrichton]

Whoops, had the wrong code example, regression still here.