When to use enums, and when to replace them with a class with static members?

Solution 1:

Enums are great for lightweight state information. For example, your color enum (excluding blue) would be good for querying the state of a traffic light. The true color along with the whole concept of color and all its baggage (alpha, color space, etc) don't matter, just which state is the light in. Also, changing your enum a little to represent the state of the traffic light:

[Flags()]
public enum LightColors
{
    unknown = 0,
    red = 1,
    yellow = 2,
    green = 4,
    green_arrow = 8
}

The current light state could be set as:

LightColors c = LightColors.red | LightColors.green_arrow;

And queried as:

if ((c & LightColors.red) == LightColors.red)
{
    //Don't drive
}
else if ((c & LightColors.green_arrow) == LightColors.green_arrow)
{
    //Turn
}

Static class color members would be able to support this multiple state without extra functionality.

However, static class members are wonderful for commonly used objects. The System.Drawing.Color members are great examples as they represent a known-name colors that have obscure constructors (unless you know your hex colors). If they were implemented as enums you would have to do something like this every time you wanted to use the value as a color:

colors c = colors.red;
switch (c)
{
    case colors.red:
        return System.Drawing.Color.FromArgb(255, 0, 0);
        break;
    case colors.green:
        return System.Drawing.Color.FromArgb(0,255,0);
        break;
}

So if you've got an enum and find that your constantly doing a switch/case/if/else/whatever to derive an object, you might want to use static class members. If you're only querying the state of something, I'd stick with enums. Also, if you have to pass data around in an unsafe fashion enums will probably survive better than a serialized version of your object.

Edit: @stakx, I think you stumbled on something important, too in response to @Anton's post and that is complexity or more importantly, who is it complex for?

From a consumer's standpoint, I would immensely prefer System.Drawing.Color static class members over having to write all of that. From a producer's standpoint, however, it would be a pain to have to write all of that. So if other people are going to be using your code you might be saving them a lot of trouble by using static class members even though it might take you 10 times as long to write/test/debug. However, if its just you you might find it easier to just use enums and cast/convert as needed.

Solution 2:

Some things I found in the meantime, if anyone else is interested:

• switch blocks won't work with an enum-as-class.

• User empi mentioned the similarity of the above enum-as-class sample to Java enums. It seems that in the Java world, there is a recognised pattern called the Typesafe Enum; apparently this pattern goes back to Joshua Bloch's book Effective Java.

Solution 3:

I have actually been struggling with something like this quite a bit at work.

It was based on an example that was posted by John B in Jon Skeet's blog article "Enhanced enums in C#".

What I did not get to work properly without a LOT of ugliness was extending an enumeration by deriving a base enumeration class and adding additional static fields of the deriving type.

If you check the basics that are posted in that blog you'll notice that deriving classes will share a common set of values which have serious typing issues and when a certain enumeration base class is derived in two different other classes their value sets will be in the same collection too.

I mean, you'd be hard pressed to create a variable of DerivedEnumClass1 and having to store a value from it's enumeration collection which is actually of type BaseEnumClass1 in that variable without workarounds.

Another problem was Xml Serialization which we use a lot. I got around that using two generic classes as datatypes on enumeration variables on our business objects: one that represented a single enumeration value and one that represented a set of enumeration values which I used to mimic flags enumeration behaviour. They handled xml serialization and deserialization of the actual values that were stored in the properties they were representing. A couple operator overloads were all that was necessary to recreate bitflag behaviour this way.

Those are some of the major issure I ran into.

All in all I'm not very pleased with the end result, some smelly stuff in there but it does the job for now.

Why our chief architect decided to still try and get it to work was the possibilities of having actual classes and extending them with all kinds of behaviour, either generic or specific. Thus far I have not seen a lot I wouldn't have been able to provide with extension methods, but we might run into something.

Don't have any of the code here and I'm not gonna be able to get to it for the weekend otherwise I could have shown where I had gone with it. Not very pretty though... :o

Solution 4:

I would use the class constants when I have to interop with some legacy code that wants some obscure bit flag values that might be or'd together. In this case the enum might look like

public enum Legacy : ushort
{
  SomeFlag1 = 0x0001,
  SomeFlag2 = 0x0002,
  // etc...
}

Then the marshalling at the P/Invoke is less readable because of the casting needed to translate the enum to the appropriate value.

If it were a const class variable a cast wouldn't be needed.