LLVM undef with overflowing floating-point -> integer casts

[arielb1]

An overflowing fp -> int cast, for example:

#[inline(never)]
fn show<T: std::fmt::Debug>(e: T) { println!("{:?}", e) }

fn main() {
    show({
        9.223372036854776e18f64 as i64
    });
}

playpen

Causes LLVM to not initialise the parameter passed to show:

trs: uwtable
define internal void @_ZN4main20he0a80c27bce3fc14RaaE() unnamed_addr #0 {
entry-block:
  tail call fastcc void @_ZN4show21h11375292364260386722E()
  ret void
}

[arielb1]

Shouldn't this be I-wrong?

[pnkfelix]

triage: P-high

[arielb1]

duplicate of #15536