dead_code warnings are wrong

[mahkoh]

In some cases, removing this dead code will cause an error.

[sfackler]

Do you have an example?

[mahkoh]

static A: u8 = 1;
fn f() { A; }
fn main() { }

test5.rs:1:1: 1:18 warning: static item is never used: `A`, #[warn(dead_code)] on by default
test5.rs:1 static A: u8 = 1;
           ^~~~~~~~~~~~~~~~~

[steveklabnik]

A is never used in that case, though.

[mahkoh]

test5.rs:1:10: 1:11 error: unresolved name `A`.
test5.rs:1 fn f() { A; }
                    ^

[sfackler]

The dead code warning is transitive by design:

fn foo() {}

fn bar() {
    foo()
}

fn main() {}

<anon>:1:1: 1:12 warning: function is never used: `foo`, #[warn(dead_code)] on by default
<anon>:1 fn foo() {}
         ^~~~~~~~~~~
<anon>:3:1: 5:2 warning: function is never used: `bar`, #[warn(dead_code)] on by default
<anon>:3 fn bar() {
<anon>:4     foo()
<anon>:5 }

There are two desired properties of the dead code lint that it would ideally exhibit:

1. Removing any subset of items classified as unused from a well-formed program always leaves the resulting program in a compiling state.
2. Removing any item classified as unused never introduces new "unused code" warnings.

Unfortunately, it is impossible to satisfy both so we're choosing 2) over 1). I'm not sure it's the right choice.