Should I use public or private variables?

Question 1

private data members are generally considered good because they provide encapsulation.

Providing getters and setters for them breaks that encapsulation, but it's still better than public data members because there's only once access point to that data.

You'll notice this during debugging. If it's private, you know you can only modify the variable inside the class. If it's public, you'll have to search the whole code-base for where it might be modified.

As much as possible, ban getters/setters and make properties private. This follows the principle of information hiding - you shouldn't care about what properties a class has. It should be self-contained. Of course, in practice this isn't feasible, and if it is, a design that follows this will be more cluttered and harder to maintain than one that doesn't.

This is of course a rule of thumb - for example, I'd just use a struct (equivalent with a class with public access) for, say, a simple point class:

struct Point2D
{
   double x;
   double y;
};

Question 2

Since you say that you know the theory, and other answers have dug into the meaning of public/private, getters and setters, I'd like to focus myself on the why of using accessors instead of creating public attributes (member data in C++).

Imagine that you have a class Truck in a logistic project:

class Truck {
public:
    double capacity;

    // lots of more things...
};

Provided you are northamerican, you'll probably use gallons in order to represent the capacity of your trucks. Imagine that your project is finished, it works perfectly, though many direct uses of Truck::capacity are done. Actually, your project becomes a success, so some european firm asks you to adapt your project to them; unfortunately, the project should use the metric system now, so litres instead of gallons should be employed for capacity.

Now, this could be a mess. Of course, one possibility would be to prepare a codebase only for North America, and a codebase only for Europe. But this means that bug fixes should be applied in two different code sources, and that is decided to be unfeasible.

The solution is to create a configuration possibility in your project. The user should be able to set gallons or litres, instead of that being a fixed, hardwired choice of gallons.

With the approach seen above, this will mean a lot of work, you will have to track down all uses of Truck::capacity, and decide what to do with them. This will probably mean to modify files along the whole codebase. Let's suppose, as an alternative, that you decided a more theoretic approach.

class Truck {
public:
    double getCapacity() const
        { return capacity; }

    // lots of more things...
private:
    double capacity;
};

A possible, alternative change involves no modification to the interface of the class:

class Truck {
public:
    double getCapacity() const
        { if ( Configuration::Measure == Gallons ) {
            return capacity;
          } else {
             return ( capacity * 3.78 );
          }
        }


    // lots of more things...
private:
    double capacity;
};

(Please take int account that there are lots of ways for doing this, that one is only one possibility, and this is only an example)

You'll have to create the global utility class configuration (but you had to do it anyway), and add an include in truck.h for configuration.h, but these are all local changes, the remaining of your codebase stays unchanged, thus avoiding potential bugs.

Finally, you also state that you are working now in a big project, which I think it is the kind of field in which these reasons actually make more sense. Remember that the objective to keep in mind while working in large projects is to create maintainable code, i.e., code that you can correct and extend with new functionalities. You can forget about getters and setters in personal, small projects, though I'd try to make myself used to them.

Hope this helps.

Question 3

There is no hard rule as to what should be private/public or protected.

It depends on the role of your class and what it offers.

All the methods and members that constitute the internal workings of the class should be made private.
Everything that a class offers to the outside world should be public.
Members and methods that may have to be extended in a specialization of this class, could be declared as protected.

Question 4

From an OOP point of view getters/setters help with encapsulation and should therefore always be used. When you call a getter/setter the class can do whatever it wants behind the scenes and the internals of the class are not exposed to the outside.

On the other hand, from a C++ point of view, it can also be a disadvantage if the class does lots of unexpected things when you just want to get/set a value. People like to know if some access results in huge overhead or is simple and efficient. When you access a public variable you know exactly what you get, when you use a getter/setter you have no idea.

Especially if you only do a small project, spending your time writing getters/setters and adjusting them all accordingly when you decide to change your variable name/type/... produces lots of busywork for little gain. You'd better spend that time writing code that does something useful.

C++ code commonly doesn't use getters/setters when they don't provide real gain. If you design a 1,000,000-line project with lots of modules that have to be as independent as possible it might make sense, but for most normal-sized code you write day to day they are overkill.

Question 5

There are some data types whose sole purpose is to hold well-specified data. These can typically be written as structs with public data members. Aside from that, a class should define an abstraction. Public variables or trivial setters and getters suggest that the design hasn't been thought through sufficiently, resulting in an agglomeration of weak abstractions that don't abstract much of anything. Instead of thinking about data, think about behavior: this class should do X, Y, and Z. From there, decide what internal data is needed to support the desired behavior. That's not easy at first, but keep reminding yourself that it's behavior that matters, not data.

Question 6

Public variables are usually discouraged, and the better form is to make all variables private and access them with getters and setters:

private int var;

public int getVar() {
  return var;
}

public void setVar(int _var) {
  var = _var;
}

Modern IDEs like Eclipse and others help you doing this by providing features like "Implement Getters and Setters" and "Encapsulate Field" (which replaces all direct acccesses of variables with the corresponding getter and setter calls).

Question 7

Private member variables are preferred over public member variables, mainly for the reasons stated above (encapsulation, well-specified data, etc..). They also provide some data protection as well, since it guarantees that no outside entity can alter the member variable without going through the proper channel of a setter if need be.

Another benefit of getters and setters is that if you are using an IDE (like Eclipse or Netbeans), you can use the IDE's functionality to search for every place in the codebase where the function is called. They provide visibility as to where a piece of data in that particular class is being used or modified. Also, you can easily make the access to the member variables thread safe by having an internal mutex. The getter/setter functions would grab this mutex before accessing or modifying the variable.

I'm a proponent of abstraction to the point where it is still useful. Abstraction for the sake of abstraction usually results in a cluttered mess that is more complicated than its worth.