In .NET, using "foreach" to iterate an instance of IEnumerable will create a copy? So should I prefer to use "for" instead of "foreach"?

Question

In .NET, using "foreach" to iterate an instance of IEnumerable will create a copy? So should I prefer to use "for" instead of "foreach"?

I wrote some code to testify this:

struct ValueTypeWithOneField
{
    private Int64 field1;
}

struct ValueTypeWithFiveField
{
    private Int64 field1;
    private Int64 field2;
    private Int64 field3;
    private Int64 field4;
    private Int64 field5;
}

public class Program
{
    static void Main(string[] args)
    {
        Console.WriteLine("one field");
        Test<ValueTypeWithOneField>();

        Console.WriteLine("-----------");

        Console.WriteLine("Five field");
        Test<ValueTypeWithFiveField>();

        Console.ReadLine();
    }

    static void Test<T>()
    {
        var test = new List<T>();
        for (int i = 0; i < 5000000; i++)
        {
            test.Add(default(T));
        }

        Stopwatch sw = new Stopwatch();

        for (int i = 0; i < 5; i++)
        {
            sw.Start();

            foreach (var item in test)
            {

            }

            sw.Stop();
            Console.WriteLine("foreach " + sw.ElapsedMilliseconds);
            sw.Restart();

            for (int j = 0; j < test.Count; j++)
            {
                T temp = test[j];
            }

            sw.Stop();
            Console.WriteLine("for " + sw.ElapsedMilliseconds);
            sw.Reset();
        }
    }}

And this is the result that I got after I ran the code:

    one field
    foreach 68
    for 72
    foreach 68
    for 72
    foreach 67
    for 72
    foreach 64
    for 73
    foreach 68
    for 72
    -----------
    Five field
    foreach 272
    for 193
    foreach 273
    for 191
    foreach 272
    for 190
    foreach 271
    for 190
    foreach 275
    for 188

As we can see in the result, "foreach" always takes more time than "for".

So should I prefer to use "for" instead of "foreach" when iterating through a generic collection of value type?

Note: thanks for the reminder, I edited the code and result. but still, foreach is running slower than for.

Answer 1

Your question is way, way too complex. Break it down.

Does using “foreach” to iterate a sequence of value types create a copy of the sequence?

No.

Does using "foreach" to iterate a sequence of value types create a copy of each value?

Yes.

Does using "for" to do an equivalent iteration of an indexed sequence of value types create a copy of each value?

Usually, yes. There are things you can do to avoid the copying if you know special things about the collection, like for instance that it is an array. But in the general case of indexed collections, indexing the sequence returns a copy of the value in the sequence, not a reference to a storage location containing the value.

Does doing anything to a value type make a copy of the value?

Just about. Value types are copied by value. That's why they're called value types. The only things that you do to value types that do not make a copy are calls to methods on the value type, and passing a value type variable using "out" or "ref". Value types are copied constantly; that's why value types are often slower than reference types.

Does using "foreach" or "for" to iterate a sequence of reference type copy the reference?

Yes. The value of an expression of reference type is a reference. That reference is copied whenever it is used.

So what's the difference between value types and reference types as far as their copying behaviour is concerned?

Value types are copied by value. Reference types copy the reference but not the thing being referred to. A 16-byte value type copies 16 bytes every time you use it. A 16 byte reference type copies the 4 (or 8) byte reference every time you use it.

Is the foreach loop slower than the for loop?

Often it is. The foreach loop is often doing more work, in that it is creating an enumerator and calling methods on the enumerator, instead of just incrementing an integer. Integer increments are extremely fast. Also don't forget that the enumerator in a foreach loop has to be disposed, and that can take time as well.

Should I use the for loop instead of the foreach loop because the for loop is sometimes a few microseconds faster?

No. That's dumb. You should make smart engineering decisions based on customer-focussed empirical data. The extra burden of a foreach loop is tiny. The customer will probably never notice. What you should do is:

• Set performance goals based on customer input
• Measure to see if you've met your goals
• If you have not, find the slowest thing using a profiler
• Fix it
• Repeat until you've met your goals

Odds are extremely good that if you have a performance problem, changing a foreach loop to a for loop will make no difference whatsoever to your problem. Write the code the way it looks clear and understandable first.

Answer 2

Your test is not accurate; in the foreach version, you're actually spinning up the enumerator and retrieving each value from the list (even though you aren't using it). In the for version, you aren't doing anything with the list at all, other than looking at its Count property. You're essentially testing the performance of an enumerator traversing a collection compared to incrementing an integer variable an equivalent number of times.

To create parity, you'd need to declare a temporary variable and assign it in each iteration of the for loop.

That being said, the answer to your question is yes. A copy of the value will be created with every assignment or return statement.

Performance

This pseudocode breakdown should explain why foreach is somewhat slower than using for in this particular instance:

foreach:

try
{
    var en = test.GetEnumerator(); //creates a ListEnumerator
    T item;

    while(en.MoveNext()) // MoveNext increments the current index and returns
                         // true if the new index is valid, or false if it's
                         // beyond the end of the list. If it returns true,
                         // it retrieves the value at that index and holds it 
                         // in an instance variable
    {
        item = en.Current; // Current retrieves the value of the current instance
                           // variable
    }
}
finally { }

for:

int index = -1;
T item;

while(++index < test.Count)
{
    item = test[index];
}

As you can see, there's simply less code in the for implementation, and foreach has a layer of abstraction (the enumerator) on top of the for. I wrote the for using a while loop to show the two versions in a similar representation.

With all that said...

You're talking about a trivial difference in execution time. Use the loop that makes the code clearer and smaller, and in this circumstance that looks like foreach.

Answer 3

You're not resetting the "stopwatch" after the "for" test, so the time taken in the 'for' test is being added to the subsequent 'foreach' test. Also, as correctly specified, you should do an assignment inside the 'for' to mimic the exact behaviour of the foreach.

sw.Start();

foreach (var item in test)
{

}

sw.Stop();
Console.WriteLine("foreach " + sw.ElapsedMilliseconds);
sw.Restart();

for (int j = 0; j < test.Count; j++)
{
    T temp = test[j];
}

sw.Stop();
Console.WriteLine("for " + sw.ElapsedMilliseconds);
sw.Reset(); // -- This bit is missing!

Answer 4

In your for cycle, I don't see you actually accessing items from the test. If you add var x = test[i]; into the for cycle, you'll see that the performance will be (virtually) the same.

Every access to a value-type property creates a copy, either with foreach or using indexer on the list in a for cycle.

Answer 5

here's a discussion on the topic Why should I use foreach instead of for (int i=0; i<length; i++) in loops?

Answer 6

Your test is not fair. Consider how the foreach loop operates. You have the following code:

foreach (var item in test)
{

}

This creates a variable item, and on each iteration fetches the next object from the collection, and assigns it to item. This fetch and assign shouldn't create a copy, but it does take time to access the underlying collection and assign the correct value to the variable.

Then you have this code:

for (int j = 0; j < test.Count; j++)
{

}

This does not access the underlying collection at all. It does not read and assign a variable on each iteration. It simply increments an integer test.Count times, so of course it is faster. And if the compiler is smart, it will see that no operation happens in the loop and just optimize the whole thing away.

A fair comparison would replace that second bit of code with something like:

var item;
for (int j = 0; j < test.Count; j++)
{
    item = test.get(j);
} 

That is more comparable to what your foreach loop is doing.

As for which to use, it's really a matter of personal preference and coding style. I generally feel that foreach is more clear than for(...) from a readability standpoint.

Answer 7

I think that foreach provides an abstract way of looping through but it is technically slower than the for loop, a good article on the differences between the for loop and foreach can be found here

Answer 8

I found only one case when it matters - Developing for Windows Phone 7. There are two reason why one should change

foreach(var item in colletion)
{
}

To

int length = array.Length;
for(int i = 0; i < length; ++i)
{
}

in XNA game if collections are big or it is called often(f.e. Update method).

• It is a bit faster
• less garbage

and garbage is critical, since Compact Framework GC fires every 1MB allocation, as a result, it may causes annoying freezes.