Overflow when doing upreferred(1u"ϵ0")

[mdavezac]

julia> using Unitful

julia> upreferred(1u"ϵ0")
ERROR: OverflowError()
 in *(::Rational{Int64}, ::Rational{Int64}) at ./rational.jl:208
 in mapfoldl_impl(::Unitful.##7#9, ::Base.#*, ::Int64, ::Tuple{Tuple{Float64,Rational{Int64}},Tuple{Float64,Int64},Tuple{Float64,R
ational{Int64}}}, ::Int64) at ./reduce.jl:49
 in basefactor(::Unitful.FreeUnits{(Unitful.Unit{:Henry,Unitful.Dimensions{(Unitful.Dimension{:Current}(-2//1),Unitful.Dimension{:
Length}(2//1),Unitful.Dimension{:Mass}(1//1),Unitful.Dimension{:Time}(-2//1))}}(0,-1//1),Unitful.Unit{:Meter,Unitful.Dimensions{(U
nitful.Dimension{:Length}(1//1),)}}(0,1//1),Unitful.Unit{:SpeedOfLight,Unitful.Dimensions{(Unitful.Dimension{:Length}(1//1),Unitfu
l.Dimension{:Time}(-1//1))}}(0,-2//1)),Unitful.Dimensions{(Unitful.Dimension{:Current}(2//1),Unitful.Dimension{:Length}(-3//1),Uni
tful.Dimension{:Mass}(-1//1),Unitful.Dimension{:Time}(4//1))}}) at /Users/mdavezac/workspaces/crystal/julia/v0.5/Unitful/src/Unitf
ul.jl:453
 in convfact(...) at /Users/mdavezac/workspaces/crystal/julia/v0.5/Unitful/src/Conversion.jl:56
 in upreferred(::Unitful.Quantity{Float64,Unitful.Dimensions{(Unitful.Dimension{:Current}(2//1),Unitful.Dimension{:Length}(-3//1),
Unitful.Dimension{:Mass}(-1//1),Unitful.Dimension{:Time}(4//1))},Unitful.FreeUnits{(Unitful.Unit{:Henry,Unitful.Dimensions{(Unitfu
l.Dimension{:Current}(-2//1),Unitful.Dimension{:Length}(2//1),Unitful.Dimension{:Mass}(1//1),Unitful.Dimension{:Time}(-2//1))}}(0,
-1//1),Unitful.Unit{:Meter,Unitful.Dimensions{(Unitful.Dimension{:Length}(1//1),)}}(0,1//1),Unitful.Unit{:SpeedOfLight,Unitful.Dim
ensions{(Unitful.Dimension{:Length}(1//1),Unitful.Dimension{:Time}(-1//1))}}(0,-2//1)),Unitful.Dimensions{(Unitful.Dimension{:Curr
ent}(2//1),Unitful.Dimension{:Length}(-3//1),Unitful.Dimension{:Mass}(-1//1),Unitful.Dimension{:Time}(4//1))}}}) at /Users/mdaveza
c/workspaces/crystal/julia/v0.5/Unitful/src/User.jl:313

Trying to find where the overflow starts, I find:

julia> upreferred(1u"μ0*c^2")
-2.9634046534192514e9 A^-2 kg m^3 s^-4

Which is probably incorrect.

I'm finding a fair number of these overflow errors happening when trying to work in atomic units.

I did find that comment in the convfact about doing an overflow check :)

[mdavezac]

Trying to work out where the overflow comes from

julia> Unitful.basefactor(unit(Unitful.μ0 * 1Unitful.c^2))
(1.0,-2358202494865994080)

So, already in basefactor

[ajkeller34]

I've fixed this with a recent commit—I was being inconsistent in my strategy for overflow detection. I'll tag a new version soon.

As I've implemented it now, there's still no option to do exact conversions in Unitful when the numbers grow bigger than Int64. I consider that more of a feature enhancement than a bug fix though (see also #81). The basefactor methods could probably be rewritten to widen the type rather than giving up and using floating-point. This would be more in line with your PR.