Enum vs Const

Introduction

I recently had to review some c++ code written by someone else. The code was well written. It checked for error conditions and did reasonable things if something went wrong. But there was something about it that irked me.

For obvious reasons I'm not showing the real classes. Instead I'll illustrate the irksome thing with contrived examples.

The Original Class

class CAClass
{
public:
    static const UINT aConstantOfInterestToThisClass1 = 0;
    static const UINT aConstantOfInterestToThisClass2 = 1;

    CAClass(UINT constValue, LPCTSTR someOtherParameter);

    void DoSomething();

private:
    UINT    m_const;
    CString m_parameter;
};

The class CAClass constructor takes a parameter that specifies something of interest to the class, constValue and a pointer to a string. DoSomething() does something relevant based on the parameters passed to the constructor. The class implementation was coded something like this.

CAClass::CAClass(UINT constValue, LPCTSTR someOtherParameter)
{
    m_const = constValue;
    m_parameter = someOtherParameter;
}
    
void CAClass::DoSomething()
{
    switch (m_const)
    {
    case aConstantOfInterestToThisClass1:
        //  Do something defined by this constant.
        break;

    case aConstantOfInterestToThisClass2:
        //  Do something defined by this constant.
        break;

    default:
        //  It's not a valid value, raise an exception
        RaiseException(ERROR);
        break;
    }
}

This code will work fine. So what's wrong with it?

How I'd have coded it

class CBClass
{
public:
    enum constsOfInterestToThisClass
    {
        bConstantOfInterestToThisClass1 = 0,
        bConstantOfInterestToThisClass2,  // The value will be the 'next' 
                                          // value
    };

    CBClass(constsOfInterestToThisClass constValue,
            LPCTSTR someOtherParameter);

    void DoSomething();

private:
    constsOfInterestToThisClass m_const;
    CString m_parameter;
};

and the implementation...

CBClass::CBClass(constsOfInterestToThisClass constValue,
                 LPCTSTR someOtherParameter)
{
    m_const = constValue;
    m_parameter = someOtherParameter;
}
    
void CBClass::DoSomething()
{
    switch (m_const)
    {
    case bConstantOfInterestToThisClass1:
        //  Do something defined by this constant.
        break;

    case bConstantOfInterestToThisClass2:
        //  Do something defined by this constant.
        break;

    default:
        //  It's not a valid value, raise an exception
        //  We should never get here...
        RaiseException(ERROR);
        break;
    }
}

There's almost no difference. Any good c++ compiler would compile identical code for both classes.

So what's the difference?

CAClass(UINT constValue, LPCTSTR someOtherParameter);

CBClass(constsOfInterestToThisClass constValue, LPCTSTR someOtherParameter);

CAClass can accept any valid UINT value. That's over 4 billion possible values, only 2 of which are of any possible interest to the class. Any valid UINT value outside of 0 or 1 will cause the DoSomething function to raise an exception that other code within your application must handle.

CBClass in contrast will accept only one of the two enum values. Try and pass any invalid constant and the compiler will (should) complain with an error or warning message.

Contrast this

CAClass obj(1000, _T("This is a string"));

obj.DoSomething();

with this

CBClass obj(1000, _T("This is a string"));

obj.DoSomething();

The first example CAClass obj(1000, _T("This is a string")); will compile and throw an exception at runtime when it calls obj.DoSomething(). The second example CBClass obj(1000, _T("This is a string)); will at the very least throw up an error message in your compiler, at compile time. A good implementation will fail to produce an executable file until you've corrected the error and provided a valid value. VC++ flags a warning but produces an executable if you've set error level to 3 and not checked 'warnings as errors'. I always compile my code at error level 4 and 'warnings as errors'.

The CBClass constructor expects a first parameter of type constsOfInterestToThisClass. This may be either bConstantOfInterestToThisClass1 or bConstantOfInterestToThisClass2 or a variable of type constsOfInterestToThisClass. The compiler will let you define a variable of type constsOfInterestToThisClass but will only let you assign values from the enum values you define.

CBClass::constsOfInterestToThisClass var;

var = CBClass::bConstantOfInterestToThisClass1;  // OK

var = 47;  // Error

Another issue

From reading the foregoing it's tempting to conclude that the final default: case in CBClass::DoSomething() is superfluous. You might even have thought I left it in by mistake. After all, if you've used an enum correctly the default: should never occur. That's true today. But what if you add a new enum constant sometime down the track and forget to add code to the DoSomething() function to handle it? If your switch statement silently ignores enum values it doesn't know about you run the risk of incurring all kinds of unexpected (and difficult to trace) behaviour. Leaving the default: case in place greatly increases your chances of catching such oversights during program development and testing.

Casts

As one or two readers have pointed out it's possible to defeat the whole point of this article by using casts.  For example one could code my error example from above thusly:

CBClass::constsOfInterestToThisClass var;

var = CBClass::bConstantOfInterestToThisClass1;  // OK

var = (CBClass::constsOfInterestToThisClass) 47; // Compiles

and the compiler will happily compile your code.  Of course it won't run as you expected but if you left the default: code in the switch statement at least you'll get an exception at runtime and hopefully during program testing.  What you're doing here of course is saying to the compiler, in effect, 'you know and I know that 47 isn't a valid constant here but I know better than you do so just go ahead and compile it for me'.  Once you've asserted your superior knowledge to the compiler all bets are off.

Interestingly, the compiler considers some casts to be so extreme that it still won't compile them without an intermediate step.  Ie, cast something to something else, then cast that something else to the final type.

Conclusion