C++ 模板，用于通过常量引用和原始指针传递向量

Question

(Trivial template specialization isn't a silver bullet in this case!)

I am working on some image processing code and am interfacing with some Cython. To interact with Cython I need to pass a raw pointer to a function to avoid copy overhead. I realized that it would be nice to work with boolean images to save memory in some cases, so ideally I could change the declaration from:

template <typename T>
uint32_t* connected_components(
    T* in_labels, ...
)

To this:

template <typename T>
uint32_t* connected_components(
    const std::vector<T> &in_labels, ...
)

However, to support the bit packed case, I can't just cast the vector's container to a pointer in a specialization, I need to preserve the vector class semantics.

Is there a way to write this template such that it will declare in_labels as both a pointer and a vector? I can just copy the code, but I'd rather not as it is a rather complex implementation of a fast algorithm.

Edit: I suppose a giant macro would work... but it's ugly.

Answer 1

I think I figured out a simpler solution. I'll just make the vector specialization contain the algorithmic code and cast the raw pointer to a vector in its own specialization. I'll then let Cython interact with the pointer specialization while other uses can default to the vector specialization.

template <typename T>
uint32_t* connected_components3d_26(
    const std::vector<T> &in_labels, 
    const int64_t sx, const int64_t sy, const int64_t sz,
    int64_t max_labels, uint32_t *out_labels = NULL
  ) {

    // algorithm goes here
}


template <typename T>
uint32_t* connected_components3d_26(
    T* in_labels, 
    const int64_t sx, const int64_t sy, const int64_t sz,
    int64_t max_labels, uint32_t *out_labels = NULL
  ) {

  std::vector<T> vec_in_labels;
  vec_in_labels.assign(in_labels, in_labels + sx * sy * sz);

  return connected_components3d_26(
    vec_in_labels, 
    sx, sy, sz,
    max_labels, out_labels
  );
}