Improve performance of max_exp10

Author: omus

src/FixedPointDecimals.jl

@@ -393,7 +393,19 @@ end
 The highest value of `x` which does not result in an overflow when evaluating `T(10)^x`.
 """
 function max_exp10{T <: Integer}(::Type{T})
-    length(digits(typemax(T))) - 1
+    W = widen(T)
+    type_max = W(typemax(T))
+
+    powt = one(W)
+    ten = W(10)
+    exponent = 0
+
+    while type_max > powt
+        powt *= ten
+        exponent += 1
+    end
+
+    exponent - 1
 end
 
 """

test/runtests.jl

@@ -74,6 +74,21 @@ if VERSION < v"0.6.0-dev.1849"
     Base.:/(x::UInt128, y::BigInt) = /(promote(x, y)...)
 end
 
+@testset "max_exp10" begin
+    @test FixedPointDecimals.max_exp10(Int8) == 2
+    @test FixedPointDecimals.max_exp10(Int64) == 18
+    @test FixedPointDecimals.max_exp10(Int128) == 38
+    @test FixedPointDecimals.max_exp10(UInt8) == 2
+    @test FixedPointDecimals.max_exp10(UInt64) == 19
+    @test FixedPointDecimals.max_exp10(UInt128) == 38
+    @test_throws MethodError FixedPointDecimals.max_exp10(BigInt)
+
+    for T in CONTAINER_TYPES
+        x = FixedPointDecimals.max_exp10(T)
+        @test T(10)^x == widen(T(10))^x
+    end
+end
+
 # ensure that the coefficient multiplied by the highest and lowest representable values of
 # the container type do not result in overflow.
 @testset "coefficient" begin