Fixes #482: cuda::span improvements

* `operator[]` method for `cuda::span`.
* Using the type definitions for various methods
* De-constexpr'ifying a few methods, seeing how we're not consistent in this respect anyway
* Dropped `cbegin()` and `cend()` and added an explanatory comment

src/cuda/api/types.hpp

@@ -148,25 +148,28 @@ struct span {
  using size_type = size_t;
  using difference_type = ::std::ptrdiff_t;
  using pointer = T*;
- using const_pointer = const T*;
+ using const_pointer = T const *;
  using reference = T&;
  using const_reference = const T&;
 
- T *data_;
+ pointer data_;
  size_t size_;
 
- T *data() const noexcept { return data_; }
- constexpr size_t size() const noexcept { return size_; }
+ pointer data() const noexcept { return data_; }
+ size_type  size() const noexcept { return size_; }
 
- T const *cbegin() const noexcept { return data(); }
- T const *cend() const noexcept { return data() + size_; }
- T *begin() const noexcept { return data(); }
- T *end() const noexcept { return data() + size_; }
+ // About cbegin() and cend() for spans, see:
+ // https://stackoverflow.com/q/62757700/1593077
+ // (for which reason, they're not implemented here)
+ pointer begin() const noexcept { return data(); }
+ pointer end() const noexcept { return data() + size_; }
+
+ reference operator[](size_t idx) const noexcept { return data_[idx]; }
 
  // Allows a non-const-element span to be used as its const-element equivalent. With pointers,
  // we get a T* to const T* casting for free, but the span has to take care of this for itself.
  template<
- typename U = T,
+ typename U = value_type,
  typename = typename ::std::enable_if<not ::std::is_const<U>::value>::type
  >
  operator span<const U>()
@@ -178,7 +181,6 @@ struct span {
 
 #endif
 
-
 /**
  * Indicates either the result (success or error index) of a CUDA Runtime or Driver API call,
  * or the overall status of the API (which is typically the last triggered error).