Functions to distinguish signaling from non-signaling NaN's

[icefoxen]

It'd be nice to be able to distinguish signaling from non-signaling NaN's in some cases. In particular, writing a WebAssembly interpreter; WebAssembly disallows signaling NaN's, so one needs some way to detect and filter them out. Easy to do with bit masks, but it would be nice to be a little more abstract.

Thank you!

[icefoxen]

If you want 'em, here they are. :-P

/// Returns whether or not the float is a signaling NaN.
/// A signaling NaN has a format like:
/// `s111 1111 1nxx xxxx xxxx xxxx xxxx xxxx`
/// where the `x`'s represent a non-zero number (zero
/// would be infinity) and `n` is 0.
/// The sign bit `s` may be anything.
///
/// On some old-fashioned platforms (PA-RISC, some MIPS)
/// a signaling NaN is marked by `n=1`, but the 2008 revision of
/// IEEE754 defines it to be `n=0`.
pub fn f32_is_signaling_nan(f: f32) -> bool {
    let uf: u32 = f.to_bits();
    let signal_bit = 0b0000_0000_0100_0000_0000_0000_0000_0000;
    let signal_bit = 1 << 22;
    // signaling nan := is NAN and signal bit is clear
    let signal_bit_clear = (uf & signal_bit) == 0;
    f32::is_nan(f) && signal_bit_clear
}

/// Same as `f32_is_signaling_nan()` for `f64`'s.
/// The signaling-nan-bit is bit 51 instead of bit 22
pub fn f64_is_signaling_nan(f: f64) -> bool {
    let uf: u64 = f.to_bits();
    let signal_bit = 1 << 51;
    // signaling nan := is NAN and signal bit is clear
    let signal_bit_clear = (uf & signal_bit) == 0;
    f64::is_nan(f) && signal_bit_clear
}

[cuviper]

FloatCore or Float::integer_decode can also help safely get the mantissa, if you're on a pre-1.20 Rust without to_bits(). You still have to know where to find that is_quiet bit though.

I'm not sure if there's a good way we can add a method for this without a breaking change. Standalone methods aren't really in the spirit of num-traits. A whole new trait for this would be possible, but seems rather excessive.

[icefoxen]

I was imagining it just being a method on the Float trait, really. Not sure if that would count as a breaking change though, since implementors of Float would need to add a method. Honestly maybe this belongs in std instead...

[cuviper]

Not sure if that would count as a breaking change though, since implementors of Float would need to add a method.

Yes, that makes it breaking, unless we also add a default implementation for that method, but I don't know if that's possible here.

[icefoxen]

The obvious default implementation is "return false", but that's not really a correct default implementation, just a stub.

Due to that, and the fact that it is deathly platform-dependent, I think it might fit better in std. Might be something to consider for inclusion in num if/when you want to break the API anyway though.

[cuviper]

This issue was moved to rust-num/num-traits#51