How to refactor type-specific but almost same redundant operations without full class templating

Question

I'm working on a C++ class that needs to perform type-specific operations on different data structures. For specific reasons, making class template parameterized is not a solution for me.

but I'm struggling to find a clean solution that doesn't involve significant code duplication. Here's a simplified version of my current code. Note that in do_something I will write and delete the cache, also get reference of some value of the cache and do computation using it.

Do you have any suggestion to refactor the class?

#include <vector>
// Note that I dont want to use a template class
struct DataBase {
  std::vector<double> cached; // std::vector is just an example. 
  std::vector<double> cache2d;
  std::vector<float> cachef;
  std::vector<float> cache2f;
  // many more data structures with different types (float or double)
  template <typename T>
  void do_something() {
    if (std::is_same<T, double>::value) {
      // long lines of code using cached or cache2d
      // in my actual code, I have a lot of these like cache3d, cache4d, etc.
      // and I use all of them in the function.
    } else if (std::is_same<T, float>::value) {
      // all the same long lines of code as above
      // except using float version of caches rather than double
    }
  }
};
int main() {
  DataBase db;
  db.do_something<double>();
  db.do_something<float>();
  return 0;
}

Answer 1

I would suggest you decouple the real data source with the operations. I would do something like:

struct DataBase {
  std::vector<double> cached;
  std::vector<double> cache2d;
  std::vector<float> cachef;
  std::vector<float> cache2f;
};

template <typename T>
struct DbAccessor {
  std::vector<T>& cache;
  std::vector<T>& cache2;

  DbAccessor(std::vector<T>& c, std::vector<T>& c2) : cache(c), cache2(c2) {}
  
  void do_something() {
    // db operations to
  }
};


int main() {
  DataBase db;

  // here is the new complexity, but hopefully less complex than your do_something logic

  DbAccessor<double>(db.cached, db.cache2d).do_something();
  DbAccessor<float>(db.cachef, db.cache2f).do_something();

  
  return 0;
}

I think for a simple application above is a reasonable trade off. However, based on your usage pattern, I may argue that cached and cachef can be abstracted to begin with. Check out the following:

template <typename T>
struct TypedDatabase {
  std::vector<T> c;
  std::vector<T> c2;
};

struct Database {
   TypedDatabase<float> cachef;
   TypedDatabase<double> cached;

   template <typename T>
   TypedDatabase<T>& get_cache();

   template <>
   TypedDatabase<float>& get_cache() { return cachef; } 

   template <>
   TypedDatabase<double>& get_cache() { return cached; } 
};

// optionally write this as DbAccessor method
template <typename T>
void do_something(TypedDatabase<T>& cache) {
   // operate on cache.c and cache.c2
}