Support @view or slice etc in C++ arguments

[tallakt]

Hi

I am working on a project where I am passing large data sets between Julia and C++.

It would be nice if I did not have to collect(...) my sliced/viewed data into arrays before sending them to C++. Maybe something like:

function test(jlcxx::GenericMatrix<float, 2> mat) {
  int rows = mat.size(0);
  int cols = mat.size(1);
  for (...) {
    float cpp_tmp = mat.at(i, j);
  }
}

[barche]

Hi, could you provide a more elaborate use case? It is not clear to me what kind of type is to be used on the C++ side? The best I can think of for now is to add methods for the std::vector wrappers that accept a Julia view and insert element by element into an std::vector that can then be passed to C++ . Of course then there still is a copy of the data. Wrapping the Julia SubArray type directly seems fragile to me, since its implementation is not public Julia API.

[tallakt]

I am using Julia to test point cloud c++ libraries.

In general I am taking many point data and putting them in a C++ structure or GPU tendor.

I could only get CxxWrap to work with float vectors or matrices. So to normalize the data, i would have to copy the data in Julie, eg if the data going to C++ was a transposed matrix or a slice of a bigger data structure (having filtered
unused points)

In C++ I would almost always have to copy the data into an application specific data structure meaning two extra memory copies one in Julia one in C++.

Solutions could be to provide an iterator with begin() and end() functions or something based on std::ranges. The Julia side could maybe be based on lazy iterators, though Im not sure if they are performant. A slow iterator implementation could make the existing double copy a better option