Hi Steve V., I would definitely suggest looking at base/sort.jl and base/order.jl to see how ordering is done in the current sorting framework. I'm hoping you'll submit your work to DataStructures.jl once you're comfortable, and it would be great if it worked with the existing sort infrastructure.
Cheers, Kevin On Thursday, July 24, 2014, Steven G. Johnson <[email protected]> wrote: > First, you should realize that == is defined for all types, but it may not > do what you want for some types because it defaults to object equality. > So checking whether == is defined is not helpful, even if it happens at > compile time. > > (Currently, method_exists is evaluated at runtime, but it's possible that > this will be evaluated at compile-time for some future Julia version; see > https://github.com/JuliaLang/julia/issues/7060) > > Normally, sorting uses isless(x,y) rather than x<y, since the isless > function is guaranteed to be a total order. For example, < is not a total > order on floating-point values because of NaN. > > for equality and ≤ consistent with isless, you could define generic > functions > > isless_eq(x,y) = !(isless(x,y) || isless(y,x)) > isless_le(x,y) = !isless(y,x) > > and then allow users to override isless_eq and isless_le if they have a > more efficient implementation. Then when you call isless_eq or isless_le, > the method-dispatch system will call the most efficient/specific version of > that function automatically for you, with the decision being made at > compile-time. >
