`MVector` type yields `ImmutableDiffResult`

[RomeoV]

Hello, I'm using this package through LsqFit.jl. I have a 3d parameter vector, and would like to speed up the code by leveraging StaticArrays.jl, which basically come for free by passing a StaticVector instead of a Vector. In order to support in-place operations, we can further pass an MVector, i.e. a modifiable statically sized array. However, as seen below, Static Arrays always lead to ImmutableDiffResults.

DiffResults.jl/src/DiffResults.jl

Lines 31 to 52 in 80f085e

	"""
	DiffResult(value::Union{Number,AbstractArray}, derivs::Tuple{Vararg{Number}})
	DiffResult(value::Union{Number,AbstractArray}, derivs::Tuple{Vararg{AbstractArray}})
	Return `r::DiffResult`, with output value storage provided by `value` and output derivative
	storage provided by `derivs`.
	In reality, `DiffResult` is an abstract supertype of two concrete types, `MutableDiffResult`
	and `ImmutableDiffResult`. If all `value`/`derivs` are all `Number`s or `StaticArray`s,
	then `r` will be immutable (i.e. `r::ImmutableDiffResult`). Otherwise, `r` will be mutable
	(i.e. `r::MutableDiffResult`).
	Note that `derivs` can be provide in splatted form, i.e. `DiffResult(value, derivs...)`.
	"""
	DiffResult
	DiffResult(value::Number, derivs::Tuple{Vararg{Number}}) = ImmutableDiffResult(value, derivs)
	DiffResult(value::Number, derivs::Tuple{Vararg{StaticArray}}) = ImmutableDiffResult(value, derivs)
	DiffResult(value::StaticArray, derivs::Tuple{Vararg{StaticArray}}) = ImmutableDiffResult(value, derivs)
	DiffResult(value::Number, derivs::Tuple{Vararg{AbstractArray}}) = MutableDiffResult(value, derivs)
	DiffResult(value::AbstractArray, derivs::Tuple{Vararg{AbstractArray}}) = MutableDiffResult(value, derivs)
	DiffResult(value::Union{Number,AbstractArray}, derivs::Union{Number,AbstractArray}...) = DiffResult(value, derivs)

I've done the following overloads to make it work for my case:

DiffResult(value::MArray, derivs::Tuple{Vararg{MArray}}) = MutableDiffResult(value, derivs)
DiffResult(value::Union{Number, AbstractArray}, derivs::Tuple{Vararg{MVector}}) = MutableDiffResult(value, derivs)

I think this is a useful feature and currently can be considered a bug. If there's interest in merging this I can also open a PR.

[gdalle]

I agree that this behavior is not wanted, and I am actually questioning the utility of the mutable/immutable split (see #26)

If we do keep it, perhaps we should base the distinction on the output of ismutable?

[gdalle]

I have opened #28 to keep track

[gdalle]

Reopened cause the PR was temporarily reverted