rational powers (fixes #18114) (#18118)

Fixes precision problem for rational powers of BigFloats (#18114) (cherry picked from commit db03824)
JuliaLang · Aug 20, 2016 · 51aadfe · 51aadfe
1 parent 6d0aa14
commit 51aadfe
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 3 deletions.
diff --git a/base/rational.jl b/base/rational.jl
@@ -375,8 +375,8 @@ function ^(x::Rational, n::Integer)
 end
 
 ^(x::Number, y::Rational) = x^(y.num/y.den)
-^{T<:AbstractFloat}(x::T, y::Rational) = x^(convert(T, y.num / y.den))
-^{T<:AbstractFloat}(x::Complex{T}, y::Rational) = x^(convert(T, y.num / y.den))
+^{T<:AbstractFloat}(x::T, y::Rational) = x^convert(T,y)
+^{T<:AbstractFloat}(x::Complex{T}, y::Rational) = x^convert(T,y)
 
 ^{T<:Rational}(z::Complex{T}, n::Bool) = n ? z : one(z) # to resolve ambiguity
 function ^{T<:Rational}(z::Complex{T}, n::Integer)

diff --git a/test/numbers.jl b/test/numbers.jl
@@ -2139,7 +2139,8 @@ rationalize(nextfloat(0.0)) == 0//1
 # rational-exponent promotion rules (issue #3155):
 @test 2.0f0^(1//3) == 2.0f0^(1.0f0/3)
 @test 2^(1//3) == 2^(1/3)
-
+# no loss of precision for rational powers (issue #18114)
+@test BigFloat(2)^(BigFloat(1)/BigFloat(3)) == BigFloat(2)^(1//3)
 
 # large shift amounts
 @test Int32(-1)>>31 == -1