raw_view and quat_view are not lazy in the xtensor-sense

[nunupeke]

xtensor-quaternion/include/xquaternion.hpp

Lines 99 to 122 in 1073ae4

	// view an n-d xcontainer of quaternions as a "plain" (n+1)-d xcontainer
	// (the last dimension of the returned view will be 4)
	template <class E>
	auto raw_view(E&& in) {
	using value_type = typename std::decay_t<E>::value_type::value_type;
	auto shape = in.shape();
	std::vector<std::size_t> vshape(shape.begin(), shape.end());
	vshape.push_back(4);
	return xt::adapt(reinterpret_cast<value_type*>(in.data()),
	in.size()*4, xt::no_ownership(), vshape);
	}
	// view an n-d xcontainer as an (n-1)-d xcontainer of quaternions
	// (the last dimension of the input container must be 4)
	template <class E>
	auto quat_view(E&& in) {
	using value_type = typename std::decay_t<E>::value_type;
	auto shape = in.shape();
	std::vector<std::size_t> vshape(shape.begin(), shape.end());
	assert(vshape.back() == 4);
	vshape.pop_back();
	return xt::adapt(reinterpret_cast<xt::quaternion<value_type>*>(in.data()),
	in.size()/4, xt::no_ownership(), vshape);
	}

I need to evaluate before I can use q.data().

[nunupeke]

@wolfv, can you give me a hint on this one (again). See also here.

[wolfv]

Hi @nunupeke sorry for being late to respond here. Indeed, that's not lazy in the xtensor sense.

I am trying to figure out a way to do it but for the time being I also don't have an elegant solution. I think right now you'd probably need to implement your own view-like class for a quaternion-array.

Another possibility would be to implement a free function access operator like so:

template <class E1>
inline auto get_x(E1&& e1) noexcept
{
    auto fnct = [](auto& quat) -> decltype(quat.x()) {
        return quat.x();
    };
    return xt::make_lambda_xfunction(std::move(fnct), std::forward<E1>(e1));
}

This would be lazy, and you could even assign to it (if quat.x() returns a reference)...