Behavior of NAN as min/max for clamp_min/max

[jturner314]

#122 added clamp_min and clamp_max functions which preserve NANs in the input. I didn't see any discussion about what the best behavior is when the min/max argument is NAN, and I think it's worth considering before clamp_min/clamp_max are included in an official release. I think four different possible behaviors are reasonable when min/max is NAN:

1. Panic in debug mode
2. Panic in both debug and release mode
3. Return input
4. Return NAN

Option 1 matches num_traits::clamp. Option 2 matches (currently unstable) f32::clamp/f64::clamp. Option 3 is what the implementation currently does.

I lean towards option 1 for consistency with num_traits::clamp and because it avoids the question of whether option 3 or option 4 makes more sense.

Thoughts?

cc @termoshtt @cuviper

[cuviper]

This is important to consider -- thanks for bringing it up! I think consistency with num_traits::clamp would be best -- panic in debug mode, and return input in release mode. Maybe in a future release (semver bumped) we could consider making them all always panic like the std clamp methods.