Reduce allocation of error

[Moelf]

https://github.com/JuliaCollections/LRUCache.jl/blob/c55095c4a48ace9b49040570dd02c64942d11947/src/cyclicorderedset.jl#L111 probably needs the same treatment as: https://github.com/JuliaLang/julia/pull/41826

see discussion there.

[Jutho]

I am happy to change this, but just a question: is the allocation always there with this paradigm, or only if you combine it with interpolation?

[Moelf]

expanding: https://github.com/JuliaLang/julia/pull/41826#issuecomment-894843480

julia> using LRUCache

julia> function func_lru(ctr, a)
           _getter() = a + 1
           get!(_getter, cache_lru, (ctr, a))
       end

julia> const cache_lru = LRU{Tuple{Vector{Int},Int},Int}(maxsize = 10)
LRU{Tuple{Vector{Int64}, Int64}, Int64}(; maxsize = 10)

julia> function main()
           ctr = Int[0]
           for i = 1:10
               println("Run $i")
               println("Using Dict: $(@allocated func_dict(ctr, 1))")
               println("Using IdDict: $(@allocated func_iddict(ctr, 1))")
               println("Using LRU: $(@allocated func_lru(ctr, 1))")
           end
       end

Run 4
Using Dict: 0
Using IdDict: 32
Using LRU: 80
...