Test format of Float types conversions

float_to_string.jl

# There is no default parse(Float16), this one is taken from
# https://github.com/JuliaLang/julia/issues/16411#issuecomment-256835385
import Base: parse

function parse(::Type{Float16}, str::String)
    fp = 0.0
    try
        fp = parse(Float64, str)
    catch
        throw(ArgumentError(string("invalid number format \"",str,"\" for Float16")))
    end
    return convert(Float16, fp)
end

function test_precision(T, n)
  vals = Array{Bool}(n)
  x = rand(T, n)
  for i in 1:n
    if T == Float16
      x_str = @sprintf("%.4e", x[i])
    elseif T == Float32
      x_str = @sprintf("%.8e", x[i])
    else
      x_str = @sprintf("%.16e", x[i])
    end
    x_num = parse(T, x_str)
    vals[i] = x_num == x[i]
  end

  return all(vals)
end

# All tests are green on my local machine (Julia Version 0.5.0 Commit 3c9d753 (2016-09-19 18:14 UTC))
@assert test_precision(Float16, 10000)
@assert test_precision(Float32, 10000)
@assert test_precision(Float64, 10000)