What's the difference between System.ValueTuple and System.Tuple?

Steve Fan picture Steve Fan · Dec 11, 2016 · Viewed 44.4k times · Source

I decompiled some C# 7 libraries and saw ValueTuple generics being used. What are ValueTuples and why not Tuple instead?

Answer

Yuval Itzchakov picture Yuval Itzchakov · Dec 11, 2016

What are ValueTuples and why not Tuple instead?

A ValueTuple is a struct which reflects a tuple, same as the original System.Tuple class.

The main difference between Tuple and ValueTuple are:

  • System.ValueTuple is a value type (struct), while System.Tuple is a reference type (class). This is meaningful when talking about allocations and GC pressure.
  • System.ValueTuple isn't only a struct, it's a mutable one, and one has to be careful when using them as such. Think what happens when a class holds a System.ValueTuple as a field.
  • System.ValueTuple exposes its items via fields instead of properties.

Until C# 7, using tuples wasn't very convenient. Their field names are Item1, Item2, etc, and the language hadn't supplied syntax sugar for them like most other languages do (Python, Scala).

When the .NET language design team decided to incorporate tuples and add syntax sugar to them at the language level an important factor was performance. With ValueTuple being a value type, you can avoid GC pressure when using them because (as an implementation detail) they'll be allocated on the stack.

Additionally, a struct gets automatic (shallow) equality semantics by the runtime, where a class doesn't. Although the design team made sure there will be an even more optimized equality for tuples, hence implemented a custom equality for it.

Here is a paragraph from the design notes of Tuples:

Struct or Class:

As mentioned, I propose to make tuple types structs rather than classes, so that no allocation penalty is associated with them. They should be as lightweight as possible.

Arguably, structs can end up being more costly, because assignment copies a bigger value. So if they are assigned a lot more than they are created, then structs would be a bad choice.

In their very motivation, though, tuples are ephemeral. You would use them when the parts are more important than the whole. So the common pattern would be to construct, return and immediately deconstruct them. In this situation structs are clearly preferable.

Structs also have a number of other benefits, which will become obvious in the following.

Examples:

You can easily see that working with System.Tuple becomes ambiguous very quickly. For example, say we have a method which calculates a sum and a count of a List<Int>:

public Tuple<int, int> DoStuff(IEnumerable<int> values)
{
    var sum = 0;
    var count = 0;

    foreach (var value in values) { sum += value; count++; }

    return new Tuple(sum, count);
}

On the receiving end, we end up with:

Tuple<int, int> result = DoStuff(Enumerable.Range(0, 10));

// What is Item1 and what is Item2?
// Which one is the sum and which is the count?
Console.WriteLine(result.Item1);
Console.WriteLine(result.Item2);

The way you can deconstruct value tuples into named arguments is the real power of the feature:

public (int sum, int count) DoStuff(IEnumerable<int> values) 
{
    var res = (sum: 0, count: 0);
    foreach (var value in values) { res.sum += value; res.count++; }
    return res;
}

And on the receiving end:

var result = DoStuff(Enumerable.Range(0, 10));
Console.WriteLine($"Sum: {result.Sum}, Count: {result.Count}");

Or:

var (sum, count) = DoStuff(Enumerable.Range(0, 10));
Console.WriteLine($"Sum: {sum}, Count: {count}");

Compiler goodies:

If we look under the cover of our previous example, we can see exactly how the compiler is interpreting ValueTuple when we ask it to deconstruct:

[return: TupleElementNames(new string[] {
    "sum",
    "count"
})]
public ValueTuple<int, int> DoStuff(IEnumerable<int> values)
{
    ValueTuple<int, int> result;
    result..ctor(0, 0);
    foreach (int current in values)
    {
        result.Item1 += current;
        result.Item2++;
    }
    return result;
}

public void Foo()
{
    ValueTuple<int, int> expr_0E = this.DoStuff(Enumerable.Range(0, 10));
    int item = expr_0E.Item1;
    int arg_1A_0 = expr_0E.Item2;
}

Internally, the compiled code utilizes Item1 and Item2, but all of this is abstracted away from us since we work with a decomposed tuple. A tuple with named arguments gets annotated with the TupleElementNamesAttribute. If we use a single fresh variable instead of decomposing, we get:

public void Foo()
{
    ValueTuple<int, int> valueTuple = this.DoStuff(Enumerable.Range(0, 10));
    Console.WriteLine(string.Format("Sum: {0}, Count: {1})", valueTuple.Item1, valueTuple.Item2));
}

Note that the compiler still has to make some magic happen (via the attribute) when we debug our application, as it would be odd to see Item1, Item2.