How to avoid code duplication implementing const and non-const iterators?

user_1386054 0
Viewed 17835

I'm implementing a custom container with an STL-like interface. I have to provide a regular iterator and a const iterator. Most of the code for the two versions of the iterators is identical . How can I avoid this duplication?

For example, my container class is Foo, and I'm implementating FooIterator and FooConstIterator. Both of the iterators have to provide methods like operator++() which are identical.

My question is similar to How do I remove code duplication between similar const and non-const member functions?, but the answer to that one is specific to const and non-const methods, especially accessors. I don't see how that might generalize to the iterator problem.

Should I have FooIterator derive from FooConstIterator and extend it with additional non-const methods? That either leads to virtual methods or method hiding, which seem inappropriate here.

Perhaps FooIterator should contain a FooConstIterator. Although that approach does reduce implementation duplication, it seems to re-introduce a lot of boilerplate method definitions.

Is there clever template technique for generating the two iterators from a single definition? Or perhaps there's a way to--shudder--use the preprocessor to stamp out these nearly identical classes.

I've tried looking at my local STL implementation to see how it handle this. There are so many helper classes that I'm having trouble grokking the design, but it looks like the functionality is simply duplicated.

In previous projects, my custom container was built on top of a standard STL container, so I didn't have to provide my own iterators. That's not an option in this case.

dry iterator constants c++
6 Answers

In addition to the suggestion that you might templatize the constness and non-constness, you could also reduce the amount of work by taking a look at Boost.Iterator tutorial - which also mentions the same solution.

user_1556930

Arthor O'Dwyer is answering this in detail in his blog post: https://quuxplusone.github.io/blog/2018/12/01/const-iterator-antipatterns/

In essence, 

template<bool IsConst>
class MyIterator {
    int *d_;
public:
    MyIterator(const MyIterator&) = default;  // REDUNDANT BUT GOOD STYLE

    template<bool IsConst_ = IsConst, class = std::enable_if_t<IsConst_>>
    MyIterator(const MyIterator<false>& rhs) : d_(rhs.d_) {}  // OK
};
using Iterator = MyIterator<false>;
using ConstIterator = MyIterator<true>;
};

Also, add static_assert(std::is_trivially_copy_constructible_v<ConstIterator>); to your code, to make sure your iterators stay trivially copy constructible:

Conclusion: If you are implementing your own container iterators — or any other pair of types with this “one-way implicit converting” behavior, such as the Networking TS’s const_buffers_type and mutable_buffers_type — then you should use one of the patterns above to implement converting constructors without accidentally disabling trivial copyability.

user_4615970

You can use CRTP and a common base to "inject" methods (but you still have to duplicate ctors in current C++), or just use the preprocessor (no shuddering required; handles ctors easily):

struct Container {

#define G(This) \
This operator++(int) { This copy (*this); ++*this; return copy; }
// example of postfix++ delegating to ++prefix

  struct iterator : std::iterator<...> {
    iterator& operator++();
    G(iterator)
  };
  struct const_iterator : std::iterator<...> {
    const_iterator& operator++();
    G(const_iterator)
  };

#undef G
// G is "nicely" scoped and treated as an implementation detail
};

Use std::iterator, the typedefs it gives you, and any other typedefs you might provide to make the macro straight-forward.

Admin1
Related