Functions

fn add(x: i32, y: i32) -> i32 {
    return x +% y;
}

// No return type means the unit type ().
fn shout(msg: str) {
    #println(msg);
}

// Functions are public by default; a leading `_` makes one file-private.
fn answer() -> i32 { return 42; }

Every function body must end with return EXPR;. There is no implicit tail return at the function level (the rule is E0333). Block expressions can still be tail expressions inside a return or a let:

fn classify(n: i32) -> i32 {
    return if n < 0 { -1 } else if n == 0 { 0 } else { 1 };
}

Generics use square brackets, not angle brackets (see Ownership for how arguments are passed):

fn identity[T](x: T) -> T { return x; }
fn max[T: Ord](a: T, b: T) -> T { ... }

There is no function overloading. A name has one signature, period. That is what lets a reader, or a model, resolve a call to exactly one definition.

Functions whose safety rests on a caller-side invariant

There is no unsafe keyword and no unsafe { } block. A function whose contract the compiler cannot verify is just an ordinary fn: it carries no extra marker, and its call site stays bare.

fn as_str(this) -> str { ... }   // caller must uphold the invariant

let view: str = t.as_str();

What makes such an operation visible is the operation itself, not a wrapper: a raw deref or index is *p / p[i], making a pointer is x as *T, a pointer-to-int is the loud #addr(p) intrinsic, and a foreign call cannot appear without a preceding extern fn declaration. The declaration is the marker; the call stays bare. So a method like Text::as_str or a raw FFI return reads as plain C+ — the invariant lives in the function's documentation, not in syntax.