Points on a `ProductManifold` are not `AbstractManifoldPoint`s

[NKlug]

I hope this issue is in the correct repository.
Essentially, points on a ProductManifold points are not AbstractManifoldPoints and tangent vectors are not AbstractTangentVectors.

MWE:

using Manifolds
using RecursiveArrayTools

M = Euclidean(3) × Euclidean(2)

println(M isa ProductManifold) # true
println(M isa AbstractManifold) # true

p = rand(M)
println(p isa AbstractManifoldPoint) # false

X = rand(M, vector_at=p)
println(X isa AbstractTangentVector) # false

This is a bit inconvenient at times, since functions typed f(::AbstractTangentVector) do not work automatically. Is there any way to resolve this?

[mateuszbaran]

Yes, this is the right repository. The point here is that we don't require points and tangent vectors to actually subtype AbstractManifoldPoint and AbstractTangentVector. I don't have a single good use case for these types but Ronny seems to like them. The only two solutions here are 1) don't constrain point/tangent vector types anywhere (like basically all of JuliaManifolds) or 2) wrap everything in a thin struct wrapper that does the subtyping, which is inconvenient because you have to unwrap and rewrap it constantly.

[kellertuer]

Yes, I think it is nice if our own point/vector types are subtypes of our abstract type.

It is not 100% necessary, Mateusz is right, but since we have that type we could also actually use it.

[kellertuer]

...and one solution is of course to add the super type to those two types. I can take a look, that should be relatively easy. We should be consistent within our packages to use that super type.

But in general I think it is not necessary to add types to function definitions f(p) since we also are just fine with arrays in general as well.
Unless of course there is a reason to just allow for a single (or a set of) point/vector types to accept.

edit: Oh I see, there is actual not even anything to fix here. The points you have are ArrayPartitions.

To be more clear: The super types are meant to be super types of points/tangents for the cases where we need points (like the PoincareBallPoint) – e.g. to distinguish point types / representations. So that is more like “all points where we need our own types are subtypes of this”. So it is more like just for an own internal hierarchy.

If you come along with your own types like ArrayPartitions – that is simply not possible and we also enforce that nowhere.

So there is nothing to fix here.