What are the semantics of a const member function?

Question

I understand that the function is not allowed to change the state of the object, but I thought I read somewhere that the compiler was allowed to assume that if the function was called with the same arguments, it would return the same value and thus could reuse a cached value if it was available. e.g.

class object
{
    int get_value(int n) const
    {
        ...
    }

...


object x;

int a = x.get_value(1);
    ...
int b = x.get_value(1);

then the compiler could optimize the second call away and either use the value in a register or simply do b = a;

Is this true?

Answer 1

const is about program semantics and not about implementation details. You should mark a member function const when it does not change the visible state of the object, and should be callable on an object that is itself const. Within a const member function on a class X, the type of this is X const *: pointer to constant X object. Thus all member variables are effectively const within that member function (except mutable ones). If you have a const object, you can only call const member functions on it.

You can use mutable to indicate that a member variable may change even within a const member function. This is typically used to identify variables used for caching results, or for variables that don't affect the actual observable state such as mutexes (you still need to lock the mutex in the const member functions) or use counters.

class X
{
    int data;
    mutable boost::mutex m;
public:
    void set_data(int i)
    {
        boost::lock_guard<boost::mutex> lk(m);
        data=i;
    }
    int get_data() const // we want to be able to get the data on a const object
    {
        boost::lock_guard<boost::mutex> lk(m); // this requires m to be non-const
        return data;
    }
};

If you hold the data by pointer rather than directly (including smart pointers such as std::auto_ptr or boost::shared_ptr) then the pointer becomes const in a const member function, but not the pointed-to data, so you can modify the pointed-to data.

As for caching: in general the compiler cannot do this because the state might change between calls (especially in my multi-threaded example with the mutex). However, if the definition is inline then the compiler can pull the code into the calling function and optimize what it can see there. This might result in the function effectively only being called once.

The next version of the C++ Standard (C++0x) will have a new keyword constexpr. Functions tagged constexpr return a constant value, so the results can be cached. There are limits on what you can do in such a function (in order that the compiler can verify this fact).

Answer 2

The keyword mutable on member variables allows for const functions to alter the state of the object at hand.

And no, it doesn't cache data (at least not all calls) since the following code is a valid const function that changes over time:

int something() const { return m_pSomeObject->NextValue(); }

Note that the pointer can be const, though the object pointed to is not const, therefore the call to NextValue on SomeObject may or may not alter it's own internal state. This causes the function something to return different values each time it's called.

However, I can't answer how the compiler works with const methods. I have heard that it can optimize certain things, though I'd have to look it up to be certain.