powf for NaN returns 1

[CodeSandwich]

The powf for floating point numbers seems to have optimization: when base is 1, always return 1. This optimization does not check if exponent is a number, so

1f32.powf(std::f32::NAN) == 1.

This behavior is same for debug and release modes, f32 and f64. playground

[nikic]

Relevant code in LLVM: https://github.com/llvm-mirror/llvm/blob/b79c73bd103edafbaff703c7230fcf1fe4e3b920/lib/Transforms/Utils/SimplifyLibCalls.cpp#L1421

[nagisa]

The contract for the powf function allows requires this behaviour (cited from linux’s man-pages for powf(3)):

  Except as specified below, if x or y is a NaN, the result is a NaN.
  If x is +1, the result is 1.0 (even if y is a NaN).
  If y is 0, the result is 1.0 (even if x is a NaN).

[nagisa]

Interestingly, that does not correspond to any IEEE-754 pow* operation, but it does properly follow the POSIX and [C99] standards, so closing as not-a-bug.

[jonas-schievink]

This could use additional documentation though, the docs for powf are very bare-bones right now

[nagisa]

Copying the documentation for most functions from the POSIX standard would make sense for me, but that probably wants a separate issue specifically about that.

[qnighy]

Interestingly, that does not correspond to any IEEE-754 pow* operation, but...

It seems to comply with IEEE754 pow behavior described in IEEE754-2008 p.44 (§9.2.1 Special values):

https://play.rust-lang.org/?version=stable&mode=debug&edition=2018&gist=7a8b40ddf93beb95fdc19932762f4ee3

[hkBst]

@qnighy your sample seems to be missing f64::NAN.