Remove unnecessary restriction to StridedVecOrMat

[dlfivefifty]

The "Strided array interface" https://docs.julialang.org/en/v1/manual/interfaces/#man-interface-strided-arrays-1 means that this is useful beyond just StridedVecOrMat so there is no reason to be so restrictive.

[dlfivefifty]

Note ArrayLayouts.jl also has an approach for pointers to adjoints of complex arrays by defining ConjPtr so that, for example, pointer(view(A',1:3,2:4)') works. This could be moved to Base and thereby removing type piracy for ArrayLayouts.jl:

"""
    ConjPtr{T}
represents that the entry is the complex-conjugate of the pointed to entry.
"""
struct ConjPtr{T} 
    ptr::Ptr{T}
end

unsafe_convert(::Type{ConjPtr{T}}, Ac::Adjoint{<:Complex}) where T<:Complex = unsafe_convert(Ptr{T}, parent(Ac))    
function unsafe_convert(::Type{ConjPtr{T}}, V::SubArray{T,2}) where {T,N,P}
    kr, jr = parentindices(V)
    unsafe_convert(Ptr{T}, view(parent(V)', jr, kr))
end

[dlfivefifty]

Can this get merged so I can start testing all the packages I manage on Julia v1.5?

[KristofferC]

It seems a test was requested.

[dlfivefifty]

I'm a bit confused by the following:

julia> strides(randn(4)')
(4, 1)

julia> strides([1 2 3 4])
(1, 1)

Shouldn't strides(randn(4)') return (1,1)?

[mbauman]

Good question... but is there a case in which it matters? You're not allowed to multiply that first value by anything other than zero, are you?

Is there a way in which (1, 1) is any more correct or helpful than (4, 1)? I suppose it's easier to flag (1,1) as contiguous.

[dlfivefifty]

Only because stride(::AbstractVector, 2) throws an error at the moment. I changed the definition to work around this but if you really want (4,1) we can think of an alternative.

[mbauman]

This is my main motivation:

julia> strides(randn(4, 1)')
(4, 1)

julia> strides(randn(4)')
(4, 1)

[dlfivefifty]

OK that's a strong argument. So I generalised stride(::AbstractArray, k::Integer) to be defined for all k, as in Array.

I also added ConjPtr to deal with pointers of complex adjoints. This could be removed if you aren't happy with it (though would mean ArrayLayouts.jl would still have some type piracy).

@mbauman Could you review the changes again please?

[dlfivefifty]

I believe this is ready to merge now unless you want other changes made.

[KristofferC]

Whitespace needs fixing. If this is not merged soon, I'll just push for a 1.5-RC1 since the only thing this "fixes" is a package that was doing type piracy anyway and could just be worked around in the package.

[dlfivefifty]

The test failure looks unrelated

[dlfivefifty]

Does "backport-1.5" mean it's not going to be in v1.5.0? If so I'll start working around the issue in ArrayLayouts.jl for v1.5-beta

[KristofferC]

Does "backport-1.5" mean it's not going to be in v1.5.0?

No, if it gets merged before 1.5.0 is released, then it will be in.