ensure that the assert! macro doesn't ICE on non-booleans

[oli-obk]

cc #31592

Note that this is not a fix, it exposes the span-bug even more. During stage1 command-before-exec.rs fails to run, because

{...}x86_64-unknown-linux-gnu/test/run-pass/command-before-exec.stage1-x86_64-unknown-linux-gnu: symbol lookup error: {...}x86_64-unknown-linux-gnu/test/run-pass/command-before-exec.stage1-x86_64-unknown-linux-gnu: undefined symbol: _ZN6option15Option$LT$T$GT$6unwrap14_MSG_FILE_LINE20hc561c4c240f9e4ddnxOE

it works fine in stage2

---

This cannot be done for assert_eq, because there's no way I couldn't find a way to ensure two expressions have the same types without moving them into an array. decltype ftw

[rust-highfive]

r? @sfackler

(rust_highfive has picked a reviewer for you, use r? to override)

[oli-obk]

@alexcrichton You created the before_exec function. Any thoughts?

[ollie27]

This will break this code:

use std::ops::Not;

struct Foo;

impl Not for Foo {
    type Output = bool;

    fn not(self) -> Self::Output {
        false
    }
}

fn main() {
    assert!(Foo);
}

I don't know if that breakage is acceptable but in this case it can be fixed by putting the ! back in front of the $cond.

[oli-obk]

ugh :( I so want to break this code. But that's probably not acceptable breakage.

[ollie27]

Yeah it's a similar situation to #30143.