Adding aliases xt::xtensor_pointer and xt::xarray_pointer

[tdegeus]

Fixes #2663

The name is up for discussion!

[JohanMabille]

The names sound a bit like "pointers to tensor or array", but I don't have a better alternative for now.

[tdegeus]

@JohanMabille Is seems to me that xt::xtensor_pointer<double, 2> is possible as I propose, but that getting xt::xarray_pointer from xt::adapt is ugly.

In particular, xarray_adaptor takes a shape type template, and it seems that there are no sufficient implicit conversion for xt::svector. With the current implementation I get

no viable conversion from 'xarray_adaptor<[2 * ...], std::vector<unsigned long>>' to 'xarray_adaptor<[2 * ...], (default) xt::svector<unsigned long, 4, std::allocator<unsigned long>, true>>'

So the only way I see is to have

template <class T, layout_type L, class SC>
using xarray_pointer = xarray_adaptor<xbuffer_adaptor<xtl::closure_type_t<T*>, xt::no_ownership, detail::default_allocator_for_ptr_t<T>>, L, SC>;

which could be called as

std::vector<size_t> shape = { 2, 2 };
std::vector<double> data = { 1, 2, 3, 4};
xt::xarray_pointer<double, XTENSOR_DEFAULT_LAYOUT, decltype(shape)> a = xt::adapt(data.data(), 4, xt::no_ownership(), shape);

which I find very ugly. Do you know a way around this? Can a proper conversion be presented to the compiler?

[JohanMabille]

You can use xt::svector instead of std::vector for the shape to avoid passing the shape parameter type. But you should keep it in the alias definition and default it (to the same value as in xarray_adaptor).

Can a proper conversion be presented to the compiler?

The only way to do it would be to provide an implicitconversion from xarray_adaptor<T, L, SC1> to xarray_adaptor<T, L, SC2> but I am definitely not for this. That would involve a copy hidden to the user.

[tdegeus]

Thanks. I went for the way of least resistance

[tdegeus]

Thanks! Quite happy that we now have this. I have worked around it a few times already, but new enough to implement it. In particular because I did not realise it could be done so simply