Other ways to deal with “loop initialization” in C

Question

To start with I'll say that I agree that goto statements are largely made irrelevant by higher level constructs in modern programming languages and shouldn't be used when a suitable substitute is available.

I was re-reading an original edition of Steve McConnell's Code Complete recently and had forgotten about his suggestion for a common coding problem. I had read it years ago when I was first getting started and don't think I realized how useful the recipe would be. The coding problem is the following: when executing a loop you often need to execute part of the loop to initialize state and then execute the loop with some other logic and ending each loop with the same initialization logic. A concrete example is implementing String.Join(delimiter, array) method.

I think everybody's first take on the problem is this. Assume the append method is defined to add the argument to your return value.

bool isFirst = true;
foreach (var element in array)
{
  if (!isFirst)
  {
     append(delimiter);
  }
  else
  {
    isFirst = false;
  }

  append(element);
}

Note: A slight optimization to this is to remove the else and put it at the end of the loop. An assignment usually being a single instruction and equivalent to an else and decreases the number of basic blocks by 1 and increases the basic block size of the main part. The result being that execute a condition in each loop to determine if you should add the delimiter or not.

I've also seen and used other takes on dealing with this common loop problem. You can execute the initial element code first outside the loop, then perform your loop from the second element to the end. You can also change the logic to always append the element then the delimiter and once the loop is completed you can simply remove the last delimiter you added.

The latter solution tends to be the one that I prefer only because it doesn't duplicate any code. If the logic of the initialization sequence ever changes, you don't have to remember to fix it in two places. It does however require extra "work" to do something and then undo it, causing at least extra cpu cycles and in many cases such as our String.Join example requires extra memory as well.

I was excited then to read this construct

var enumerator = array.GetEnumerator();
if (enumerator.MoveNext())
{
  goto start;
  do {
    append(delimiter);

  start:
    append(enumerator.Current);
  } while (enumerator.MoveNext());
}

The benefit here being that you get no duplicated code and you get no additional work. You start your loop half way into the execution of your first loop and that is your initialization. You are limited to simulating other loops with the do while construct but the translation is easy and reading it is not difficult.

So, now the question. I happily went to try adding this to some code I was working on and found it didn't work. Works great in C, C++, Basic but it turns out in C# you can't jump to a label inside a different lexical scope that is not a parent scope. I was very disappointed. So I was left wondering, what is the best way to deal with this very common coding problem (I see it mostly in string generation) in C#?

To perhaps be more specific with requirements:

• Don't duplicate code
• Don't do unnecessary work
• Don't be more than 2 or 3 times slower than other code
• Be readable

I think readability is the only thing that might arguably suffer with the recipe I stated. However it doesn't work in C# so what's the next best thing?

* Edit * I changed my performance criteria because of some of the discussion. Performance is generally not a limiting factor here, so the goal more correctly should be to not be unreasonable, not to be the fastest ever.

The reason I dislike the alternate implementations I suggest is because they either duplicate code which leaves room for changing one part and not the other or for the one I generally choose it requires "undoing" the operation which requires extra thought and time to undo the thing that you just did. With string manipulation in particular this usually leaves you open for off by one errors or failing to account for an empty array and trying to undo something that didn't happen.

Answer 1

Why don't move dealing with first element outside a loop ?

StringBuilder sb = new StrindBuilder()
sb.append(array.first)
foreach (var elem in array.skip(1)) {
  sb.append(",")
  sb.append(elem)
}

Answer 2

I prefer first variable method. It is probably not cleanest but most efficient way. Alternatively you could use Length of the thing you appending to and compare it to zero. Works well with StringBuilder.

Answer 3

If you want to go the functional route, you could define String.Join like LINQ construct that is reusable across types.

Personally, I would almost always go for code clarity over saving a few opcode executions.

EG:

namespace Play
{
    public static class LinqExtensions {
        public static U JoinElements<T, U>(this IEnumerable<T> list, Func<T, U> initializer, Func<U, T, U> joiner)
        {
            U joined = default(U);
            bool first = true;
            foreach (var item in list)
            {
                if (first)
                {
                    joined = initializer(item);
                    first = false;
                }
                else
                {
                    joined = joiner(joined, item);
                }
            }
            return joined;
        }
    }

    class Program
    {

        static void Main(string[] args)
        {
            List<int> nums = new List<int>() { 1, 2, 3 };
            var sum = nums.JoinElements(a => a, (a, b) => a + b);
            Console.WriteLine(sum); // outputs 6

            List<string> words = new List<string>() { "a", "b", "c" };
            var buffer = words.JoinElements(
                a => new StringBuilder(a), 
                (a, b) => a.Append(",").Append(b)
                );

            Console.WriteLine(buffer); // outputs "a,b,c"

            Console.ReadKey();
        }

    }
}