If it's in a function, you can just make the function generic, but a bigger issue is that that push! pattern is bound to be quite slow. It's *much* better if you can allocate the entire array up front. I'm having a hard time coming up with examples based on what you wrote – it's a little too vague. Do you have a more specific example?
On Thu, May 1, 2014 at 12:19 PM, Ethan Anderes <ethanande...@gmail.com>wrote: > I have a question which is somewhat related to this question ( > https://groups.google.com/forum/m/?fromgroups#!topic/julia-users/bldp27KJCjY) > but mine is more of a style question rather than a language design question. > > I find myself often filling an array of Array{Float64,1}'s in julia. > I usually do this by initializing an empty array then using push! > > x = Array{Float64,1}[] > while someconditional > nextvec = getvec(...) > push!(x,nextvec) > end > > I'm wondering how to make the Float64 specification more flexible so that > the code works when used by someone on a machine using Float32, for > example, but still keep the performance of having the entries of x all a > homogeneous type. I would be tempted to initialize > x=Array{FloatingPoint,1}[] but that allows some entries of x to be > Array{Float64,1} and others to be Array{Float32,1} for example. > > > >
