Equivalent of PreallocationTools.jl for functions

General Usage
1

A Julia design paradigm I often see to avoid allocating temporary arrays in repeated calculations is to use a struct that contains the temporary array and a “functor” for performing the calculation:

struct MyStruct 
    data::Vector{Float64}
    f::FunctionWrapper{Float64, Tuple{Float64, Float64}}
end

function (m::MyStruct)(u)
    # mutate u using m.data and m.f
end

In the above example, I’ve also included a function as a field of MyStruct that is used in the calculation, too. I’ve used FunctionWrappers.jl so that this field has a concrete type.

To use this paradigm with ForwardDiff.jl, I know I can use PreallocationTools.jl to replace the temporary array with a cache that can store both, e.g., Float64 data as well as Dual data used by ForwardDiff.jl. Is there an equivalent approach for functions that would replace FunctionWrappers.jl, i.e., so I can have a function that takes and returns both Float64 and Dual data but can be made a field of a struct with a concrete type?

2

I’m not sure I fully understand what you want to do and why the following pattern doesn’t work?

struct MyWrapper{F}
    f::F
end
3

not sure if it fits your need but you may be interested in GitHub - MasonProtter/Bumper.jl: Bring Your Own Stack ?

4

Yes, that would work, but in my actual case I might have half a dozen different functions, so the list of parameters could get complicated.

5