MArray compatibility

[YingboMa]

There is

    f₀ = zero.(u0./t)
    u0 isa MArray && (f₀ = MArray(f₀))

because broadcasting an MArray results an SArray.

julia> ode.u0
2-element MArray{Tuple{2},Float64,1,2}:
 1.0
 2.0

julia> ode.u0 ./ 1
2-element SArray{Tuple{2},Float64,1,2}:
 1.0
 2.0

[ChrisRackauckas]

IMO this is just a bug JuliaArrays/StaticArrays.jl#327 and we shouldn't work around this.

[ChrisRackauckas]

If we can, we should just give an upstream PR

[ChrisRackauckas]

looks like SArray tests fail here?

[YingboMa]

Ah, right. Maybe we should explicitly cast the type to be MArray(m) in the meantime? I really want to get OrdinaryDiffEq.jl to work with MArrays.

[ChrisRackauckas]

Similar with element types doesn't exist for MArrays?

[YingboMa]

It does, but we cannot do oneunit(m::MArray), because Julia doesn't have that method anymore...

[ChrisRackauckas]

oneunit(first(x))? That won't do well with GPUArrays...

[ChrisRackauckas]

oneunit(eltype(m))

[YingboMa]

eltype(m) can be an MArray, too. https://github.com/JuliaDiffEq/OrdinaryDiffEq.jl/blob/master/test/static_array_tests.jl#L4

[ChrisRackauckas]

recursive_bottom_eltype(m).

[YingboMa]

That doesn't seem to work either...

julia> u0 = fill(zero(MVector{2,Float64}), 2); t = 0.1;

julia> similar(u0)
2-element Array{MArray{Tuple{2},Float64,1,2},1}:
 #undef
 #undef

julia> similar(u0, eltype(eltype(u0)))
2-element Array{Float64,1}:
 6.9502444917253e-310
 6.9502444933561e-310

julia> similar(u0, typeof(oneunit(eltype(eltype(u0)))/oneunit(t)))
2-element Array{Float64,1}:
 6.9501798101687e-310
 6.9501798097537e-310

[ChrisRackauckas]

This fix is really sketch and fails tests.

[YingboMa]

In favor of JuliaArrays/StaticArrays.jl#536