Convert NaN to correct type in sqrt

[DanielVandH]

No description provided.

[DanielVandH]

Ah, I'm guessing the integer issues are a reason for not converting it initially? Unless perhaps

sqrt(x::T) where {T<:Number}

and

sqrt(x::Integer)

ought to be separated? The point being for floats to at least be able to avoid e.g.

julia> @code_warntype NaNMath.sqrt(-0.3)
MethodInstance for NaNMath.sqrt(::Float64)
  from sqrt(x::Real) in NaNMath at C:\Users\licer\.julia\packages\NaNMath\MNJRI\src\NaNMath.jl:21
Arguments
  #self#::Core.Const(NaNMath.sqrt)
  x::Float64
Body::Float64
1 ─ %1 = (x < 0.0)::Bool
└──      goto #3 if not %1
2 ─      return NaNMath.NaN
t(::Float32)
  from sqrt(x::Real) in NaNMath at C:\Users\licer\.julia\packages\NaNMath\MNJRI\src\NaNMath.jl:21
Arguments
  #self#::Core.Const(NaNMath.sqrt)
  x::Float32
Body::Union{Float32, Float64}
1 ─ %1 = (x < 0.0)::Bool
└──      goto #3 if not %1
2 ─      return NaNMath.NaN
3 ─ %4 = Base.getproperty(NaNMath.Base, :sqrt)::Core.Const(sqrt)
│   %5 = (%4)(x)::Float32
└──      return %5

This doesn't happen for the other functions like log since the methods for Floats and others are separated into

f(x::Float64) = ...
f(x::Float32) = ...
f(x::Real) = f(float(x))

which could also be done here, though you still get Unions as above.

[mlubin]

For which types T is T(NaN) supposed to work? I'd guess T <: AbstractFloat is ok, but we'd need to check to be sure.

[DanielVandH]

I believe only AbstractFloats, based on this definition of isnan from Base

isnan(x::AbstractFloat) = (x != x)::Bool
isnan(x::Number) = false

checking only AbstractFloats.

[mlubin]

sqrt(x::T) where {T<:AbstractFloat} = x < 0.0 ? T(NaN) : Base.sqrt(x)
sqrt(x::Real) = sqrt(float(x))

seems reasonable then

[codecov]

Codecov Report

Base: 97.02% // Head: 97.05% // Increases project coverage by +0.02% 🎉

Coverage data is based on head (2727977) compared to base (65a5928).
Patch coverage: 100.00% of modified lines in pull request are covered.

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #64      +/-   ##
==========================================
+ Coverage   97.02%   97.05%   +0.02%     
==========================================
  Files           1        1              
  Lines         101      102       +1     
==========================================
+ Hits           98       99       +1     
  Misses          3        3              

Impacted Files	Coverage Δ
src/NaNMath.jl	97.05% <100.00%> (+0.02%)	⬆️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.