What's new in C# 7

C# 7 adds a number of new features to the C# language:

  • out variables:
    • You can declare out values inline as arguments to the method where they are used.
  • Tuples
    • You can create lightweight, unnamed types that contain multiple public fields. Compilers and IDE tools understand the semantics of these types.
  • Pattern Matching
    • You can create branching logic based on arbitrary types and values of the members of those types.
  • ref locals and returns
    • Method arguments and local variables can be references to other storage.
  • Local Functions
    • You can nest functions inside other functions to limit their scope and visibility.
  • More expression-bodied members
    • The list of members that can be authored using expressions has grown.
  • throw Expressions
    • You can throw exceptions in code constructs that previously were not allowed because throw was a statement.
  • Generalized async return types
    • Methods declared with the async modifier can return other types in addition to Task and Task<T>.
  • Numeric literal syntax improvements
    • New tokens improve readability for numeric constants.

The remainder of this topic discusses each of the features. For each feature, you'll learn the reasoning behind it. You'll learn the syntax. You'll see some sample scenarios where using the new feature will make you more productive as a developer.

out variables

The existing syntax that supports out parameters has been improved in this version.

Previously, you would need to separate the declaration of the out variable and its initialization into two different statements:

int numericResult;
if (int.TryParse(input, out numericResult))
    WriteLine(numericResult);
else
    WriteLine("Could not parse input");

You can now declare out variables in the argument list of a method call, rather than writing a separate declaration statement:

if (int.TryParse(input, out int result))
    WriteLine(result);
else
    WriteLine("Could not parse input");

You may want to specify the type of the out variable for clarity, as shown above. However, the language does support using an implicitly typed local variable:

if (int.TryParse(input, out var answer))
    WriteLine(answer);
else
    WriteLine("Could not parse input");
  • The code is easier to read.
    • You declare the out variable where you use it, not on another line above.
  • No need to assign an initial value.
    • By declaring the out variable where it is used in a method call, you can't accidentally use it before it is assigned.

The most common use for this feature will be the Try pattern. In this pattern, a method returns a bool indicating success or failure and an out variable that provides the result if the method succeeds.

Tuples

C# provides a rich syntax for classes and structs that is used to explain your design intent. But sometimes that rich syntax requires extra work with minimal benefit. You may often write methods that need a simple structure containing more than one data element. To support these scenarios tuples were added to C#. Tuples are lightweight data structures that contain multiple fields to represent the data members. The fields are not validated, and you cannot define your own methods

Note

Tuples were available before C# 7 as an API, but had many limitations. Most importantly, the members of these tuples were named Item1, Item2 and so on. The language support enables semantic names for the fields of a Tuple.

You can create a tuple by assigning each member to a value:

var letters = ("a", "b");

That assignment creates a tuple whose members are Item1 and Item2, following the existing Tuple syntax. You can modify that assignment to create a tuple that provides semantic names to each of the members of the tuple:

(string Alpha, string Beta) namedLetters = ("a", "b");
Note

The new tuples features require the System.ValueTuple type. For Visual Studio 15 Preview 5 and earlier preview releases, you must add the NuGet package "System.ValueTuple", available in the pre-release stream.

The namedLetters tuple contains fields referred to as Alpha and Beta. In a tuple assignment, you can also specify the names of the fields on the right-hand side of the assignment:

var alphabetStart = (Alpha: "a", Beta: "b");

The language allows you to specify names for the fields on both the left and right-hand side of the assignment:

(string First, string Second) firstLetters = (Alpha: "a", Beta: "b");

The line above generates a warning, CS8123, telling you that the names on the right side of the assignment, Alpha and Beta are ignored because they conflict with the names on the left side, First and Second.

The examples above show the basic syntax to declare tuples. Tuples are most useful as return types for private and internal methods. Tuples provide a simple syntax for those methods to return multiple discrete values: You save the work of authoring a class or a struct that defines the type returned. There is no need for creating a new type.

Creating a tuple is more efficient and more productive. It is a simpler, lightweight syntax to define a data structure that carries more than one value. The example method below returns the minimum and maximum values found in a sequence of integers:

private static (int Max, int Min) Range(IEnumerable<int> numbers)
{
    int min = int.MaxValue;
    int max = int.MinValue;
    foreach(var n in numbers)
    {
        min = (n < min) ? n : min;
        max = (n > max) ? n : max;
    }
    return (max, min);
}

Using tuples in this way offers several advantages:

  • You save the work of authoring a class or a struct that defines the type returned.
  • You do not need to create new type.
  • The language enhancements removes the need to call the Create<T1>(T1) methods.

The declaration for the method provides the names for the fields of the tuple that is returned. When you call the method, the return value is a tuple whose fields are Max and Min:

var range = Range(numbers);

There may be times when you want to unpackage the members of a tuple that were returned from a method. You can do that by declaring separate variables for each of the values in the tuple. This is called deconstructing the tuple:

(int max, int min) = Range(numbers);

You can also provide a similar deconstruction for any type in .NET. This is done by writing a Deconstruct method as a member of the class. That Deconstruct method provides a set of out arguments for each of the properties you want to extract. Consider this Point class that provides a deconstructor method that extracts the X and Y coordinates:

public class Point
{
    public Point(double x, double y)
    {
        this.X = x;
        this.Y = y;
    }

    public double X { get; }
    public double Y { get; }

    public void Deconstruct(out double x, out double y)
    {
        x = this.X;
        y = this.Y;
    }
}

You can extract the individual fields by assigning a tuple to a Point:

var p = new Point(3.14, 2.71);
(double X, double Y) = p;

You are not bound by the names defined in the Deconstruct method. You can rename the extract variables as part of the assignment:

(double horizontalDistance, double verticalDistance) = p;

You can learn more in depth about tuples in the tuples topic.

Pattern matching

Pattern matching is a feature that allows you to implement method dispatch on properties other than the type of an object. You're probably already familiar with method dispatch based on the type of an object. In Object Oriented programming, virtual and override methods provide language syntax to implement method dispatching based on an object's type. Base and Derived classes provide different implementations. Pattern matching expressions extend this concept so that you can easily implement similar dispatch patterns for types and data elements that are not related through an inheritance hierarchy.

Pattern matching supports is expressions and switch expressions. Each enables inspecting an object and its properties to determine if that object satisfies the sought pattern. You use the when keyword to specify additional rules to the pattern.

is expressions

The is pattern expressions extend the familiar is operator to query a object beyond its type.

Let's start with a simple scenario. We'll add capabilities to this scenario that demonstrate how pattern matching expressions make algorithms that work with unrelated types easy. We'll start with a method that computes the sum of a number of die rolls:

public static int DiceSum(IEnumerable<int> values)
{
    return values.Sum();
}

You might quickly find that you need to find the sum of die rolls where some of the rolls are made with more than one die. Part of the input sequence may be multiple results instead of a single number:

public static int DiceSum2(IEnumerable<object> values)
{
    var sum = 0;
    foreach(var item in values)
    {
        if (item is int val)
            sum += val;
        else if (item is IEnumerable<object> subList)
            sum += DiceSum2(subList);
    }
    return sum;
}

The is pattern expression works quite well in this scenario. As part of checking the type, you write a variable initialization. This creates a new variable of the validated runtime type.

As you keep extending these scenarios, you may find that you build more if and else if statements. Once that becomes unwieldy, you'll likely want to switch to switch pattern expressions.

switch expressions

The switch pattern expression has a familiar syntax, based on the switch statement already part of the C# language. Let's translate the existing code to use a switch expression before adding new cases:

public static int DiceSum3(IEnumerable<object> values)
{
    var sum = 0;
    foreach (var item in values)
    {
        switch (item)
        {
            case int val:
                sum += val;
                break;
            case IEnumerable<object> subList:
                sum += DiceSum3(subList);
                break;
        }
    }
    return sum;
}

The switch expressions have a slightly different syntax than the is expressions, where you declare the type and variable at the beginning of the case expression.

The switch expressions also support constants. This can save time by factoring out simple cases:

public static int DiceSum4(IEnumerable<object> values)
{
    var sum = 0;
    foreach (var item in values)
    {
        switch (item)
        {
            case 0:
                break;
            case int val:
                sum += val;
                break;
            case IEnumerable<object> subList when subList.Any():
                sum += DiceSum4(subList);
                break;
            case IEnumerable<object> subList:
                break;
            case null:
                break;
            default:
                throw new InvalidOperationException("unknown item type");
        }
    }
    return sum;
}

The code above adds cases for 0 as a special case of int, and null as a special case when there is no input. This demonstrates one important new feature in switch pattern expressions: the order of the case expressions now matters. The 0 case must appear before the general int case. Otherwise, the first pattern to match would be the int case, even when the value is 0.

This same behavior enables the special case for an empty input sequence. You can see that the case for an IEnumerable item that has elements must appear before the general IEnumerable case.

This version has also added a default case. The default case is always evaluated last, regardless of the order it appears in the source. For that reason, convention is to put the default case last.

Finally, let's add one last case for a new style of die. Some games use percentile dice to represent larger ranges of numbers.

Note

Two 10-sided percentile dice can represent every number from 0 through 99. One die has sides labelled 00, 10, 20, ... 90. The other die has sides labeled 0, 1, 2, ... 9. Add the two die values together and you can get every number from 0 through 99.

To add this kind of die to your collection, first define a type to represent the percentile die:

public struct PercentileDie
{
    public int Value { get; }
    public int Multiplier { get; }

    public PercentileDie(int multiplier, int value)
    {
        this.Value = value;
        this.Multiplier = multiplier;
    }
}

Then, add a case expression for the new type:

public static int DiceSum5(IEnumerable<object> values)
{
    var sum = 0;
    foreach (var item in values)
    {
        switch (item)
        {
            case 0:
                break;
            case int val:
                sum += val;
                break;
            case PercentileDie die:
                sum += die.Multiplier * die.Value;
                break;
            case IEnumerable<object> subList when subList.Any():
                sum += DiceSum5(subList);
                break;
            case IEnumerable<object> subList:
                break;
            case null:
                break;
            default:
                throw new InvalidOperationException("unknown item type");
        }
    }
    return sum;
}

The new syntax for pattern matching expressions makes it easier to create dispatch algorithms based on an object's type, or other properties, using a clear and concise syntax. Pattern matching expressions enable these constructs on data types that are unrelated by inheritance.

You can learn more about pattern matching in the topic dedicated to pattern matching in C#.

Ref locals and returns

This feature enables algorithms that use and return references to variables defined elsewhere. One example is working with large matrices, and finding a single location with certain characteristics. One method would return the two indices for a single location in the matrix:

public static (int i, int j) Find(int[,] matrix, Func<int, bool> predicate)
{
    for (int i = 0; i < matrix.GetLength(0); i++)
        for (int j = 0; j < matrix.GetLength(1); j++)
            if (predicate(matrix[i, j]))
                return (i, j);
    return (-1, -1); // Not found
}

There are many issues with this code. First of all, it's a public method that's returning a tuple. The language supports this, but user defined types (either classes or structs) are preferred for public APIs.

Second, this method is returning the indices to the item in the matrix. That leads callers to write code that uses those indices to dereference the matrix and modify a single element:

var indices = MatrixSearch.Find(matrix, (val) => val == 42);
Console.WriteLine(indices);
matrix[indices.i, indices.j] = 24;

You'd rather write a method that returns a reference to the element of the matrix that you want to change. You could only accomplish this by using unsafe code and returning a pointer to an int in previous versions.

Let's walk through a series of changes to demonstrate the ref local feature and show how to create a method that returns a reference to internal storage. Along the way, you'll learn the rules of the ref return and ref local feature that protects you from accidentally misusing it.

Start by modifying the Find method declaration so that it returns a ref int instead of a tuple. Then, modify the return statement so it returns the value stored in the matrix instead of the two indices:

// Note that this won't compile. 
// Method declaration indicates ref return,
// but return statement specifies a value return.
public static ref int Find2(int[,] matrix, Func<int, bool> predicate)
{
    for (int i = 0; i < matrix.GetLength(0); i++)
        for (int j = 0; j < matrix.GetLength(1); j++)
            if (predicate(matrix[i, j]))
                return matrix[i, j];
    throw new InvalidOperationException("Not found");
}

When you declare that a method returns a ref variable, you must also add the ref keyword to each return statement. That indicates return by reference, and helps developers reading the code later remember that the method returns by reference:

public static ref int Find3(int[,] matrix, Func<int, bool> predicate)
{
    for (int i = 0; i < matrix.GetLength(0); i++)
        for (int j = 0; j < matrix.GetLength(1); j++)
            if (predicate(matrix[i, j]))
                return ref matrix[i, j];
    throw new InvalidOperationException("Not found");
}

Now that the method returns a reference to the integer value in the matrix, you need to modify where it's called. The var declaration means that valItem is now an int rather than a tuple:

var valItem = MatrixSearch.Find3(matrix, (val) => val == 42);
Console.WriteLine(valItem);
valItem = 24;
Console.WriteLine(matrix[4, 2]);

The second WriteLine statement in the example above prints out the value 42, not 24. The variable valItem is an int, not a ref int. The var keyword enables the compiler to specify the type, but will not implicitly add the ref modifier. Instead, the value referred to by the ref return is copied to the variable on the left-hand side of the assignment. The variable is not a ref local.

In order to get the result you want, you need to add the ref modifier to the local variable declaration to make the variable a reference when the return value is a reference:

ref var item = ref MatrixSearch.Find3(matrix, (val) => val == 42);
Console.WriteLine(item);
item = 24;
Console.WriteLine(matrix[4, 2]);

Now, the second WriteLine statement in the example above will print out the value 24, indicating that the storage in the matrix has been modified. The local variable has been declared with the ref modifier, and it will take a ref return. You must initialize a ref variable when it is declared, you cannot split the declaration and the initialization.

The C# language has two other rules that protect you from misusing the ref locals and returns:

  • You cannot assign a value to a ref variable.
    • That disallows statements like ref int i = sequence.Count();
  • You cannot return a ref to a variable whose lifetime does not extend beyond the execution of the method.
    • That means you cannot return a reference to a local variable, or similar scope.

These rules ensure that you cannot accidentally mix value variables and reference variables. They also ensure that you cannot have a reference variable refer to storage that is a candidate for garbage collection.

The addition of ref locals and ref returns enable algorithms that are more efficient by avoiding copying values, or performing dereferencing operations multiple times.

Local functions

Many designs for classes include methods that are called from only one location. These additional private methods keep each method small and focused. However, they can make it harder to understand a class when reading it the first time. These methods must be understood outside of the context of the single calling location.

For those designs, local functions enable you to declare methods inside the context of another method. This makes it easier for readers of the class to see that the local method is only called from the context in which is it declared.

There are two very common use cases for local functions: public iterator methods and public async methods. Both types of methods generate code that reports errors later than programmers might expect. In the case of iterator methods, any exceptions are observed only when calling code that enumerates the returned sequence. In the case of async methods, any exceptions are only observed when the returned Task is awaited.

Let's start with an iterator method:

public static IEnumerable<char> AlphabetSubset(char start, char end)
{
    if ((start < 'a') || (start > 'z'))
        throw new ArgumentOutOfRangeException(paramName: nameof(start), message: "start must be a letter");
    if ((end < 'a') || (end > 'z'))
        throw new ArgumentOutOfRangeException(paramName: nameof(end), message: "end must be a letter");

    if (end <= start)
        throw new ArgumentException($"{nameof(end)} must be greater than {nameof(start)}");
    for (var c = start; c < end; c++)
        yield return c;
}

Examine the code below that calls the iterator method incorrectly:

var resultSet = Iterator.AlphabetSubset('f', 'a');
Console.WriteLine("iterator created");
foreach (var thing in resultSet)
    Console.Write($"{thing}, ");

The exception is thrown when resultSet is iterated, not when resultSet is created. In this contained example, most developers could quickly diagnose the problem. However, in larger codebases, the code that creates an iterator often isn't as close to the code that enumerates the result. You can refactor the code so that the public method validates all arguments, and a private method generates the enumeration:

public static IEnumerable<char> AlphabetSubset2(char start, char end)
{
    if ((start < 'a') || (start > 'z'))
        throw new ArgumentOutOfRangeException(paramName: nameof(start), message: "start must be a letter");
    if ((end < 'a') || (end > 'z'))
        throw new ArgumentOutOfRangeException(paramName: nameof(end), message: "end must be a letter");

    if (end <= start)
        throw new ArgumentException($"{nameof(end)} must be greater than {nameof(start)}");
    return alphabetSubsetImplementation(start, end);
}

private static IEnumerable<char> alphabetSubsetImplementation(char start, char end)
{ 
    for (var c = start; c < end; c++)
        yield return c;
}

This refactored version will throw exceptions immediately because the public method is not an iterator method; only the private method uses the yield return syntax. However, there are potential problems with this refactoring. The private method should only be called from the public interface method, because otherwise all argument validation is skipped. Readers of the class must discover this fact by reading the entire class and searching for any other references to the alphabetSubsetImplementation method.

You can make that design intent more clear by declaring the alphabetSubsetImplementation as a local function inside the public API method:

public static IEnumerable<char> AlphabetSubset3(char start, char end)
{
    if ((start < 'a') || (start > 'z'))
        throw new ArgumentOutOfRangeException(paramName: nameof(start), message: "start must be a letter");
    if ((end < 'a') || (end > 'z'))
        throw new ArgumentOutOfRangeException(paramName: nameof(end), message: "end must be a letter");

    if (end <= start)
        throw new ArgumentException($"{nameof(end)} must be greater than {nameof(start)}");

    return alphabetSubsetImplementation();

    IEnumerable<char> alphabetSubsetImplementation()
    {
        for (var c = start; c < end; c++)
            yield return c;
    }
}

The version above makes it clear that the local method is referenced only in the context of the outer method. The rules for local functions also ensure that a developer can't accidentally call the local function from another location in the class and bypass the argument validation.

The same technique can be employed with async methods to ensure that exceptions arising from argument validation are thrown before the asynchronous work begins:

public Task<string> PerformLongRunningWork(string address, int index, string name)
{
    if (string.IsNullOrWhiteSpace(address))
        throw new ArgumentException(message: "An address is required", paramName: nameof(address));
    if (index < 0)
        throw new ArgumentOutOfRangeException(paramName: nameof(index), message: "The index must be non-negative");
    if (string.IsNullOrWhiteSpace(name))
        throw new ArgumentException(message: "You must supply a name", paramName: nameof(name));

    return longRunningWorkImplementation();

    async Task<string> longRunningWorkImplementation()
    {
        var interimResult = await FirstWork(address);
        var secondResult = await SecondStep(index, name);
        return $"The results are {interimResult} and {secondResult}. Enjoy.";
    }
}
Note

Some of the designs that are supported by local functions could also be accomplished using lambda expressions. Those interested can read more about the differences

More expression-bodied members

C# 6 introduced expression-bodied members for member functions, and read-only properties. C# 7 expands the allowed members that can be implemented as expressions. In C# 7, you can implement constructors, finalizers, and get and set accessors on properties and indexers. The following code shows examples of each:

// Expression-bodied constructor
public ExpressionMembersExample(string label) => this.Label = label;

// Expression-bodied finalizer
~ExpressionMembersExample() => Console.Error.WriteLine("Finalized!");

private string label;

// Expression-bodied get / set accessors.
public string Label
{
    get => label;
    set => this.label = value ?? "Default label";
}
Note

This example does not need a finalizer, but it is shown to demonstrate the syntax. You should not implement a finalizer in your class unless it is necessary to release unmanaged resources. You should also consider using the SafeHandle class instead of managing unmanaged resources directly.

These new locations for expression-bodied members represent an important milestone for the C# language: These features were implemented by community members working on the open-source Roslyn project.

Throw expressions

In C#, throw has always been a statement. Because throw is a statement, not an expression, there were C# constructs where you could not use it. These included conditional expressions, null coalescing expressions, and some lambda expressions. The addition of expression-bodied members adds more locations where throw expressions would be useful. So that you can write any of these constructs, C# 7 introduces throw expressions.

The syntax is the same as you've always used for throw statements. The only difference is that now you can place them in new locations, such as in a conditional expression:

public string Name
{
    get => name;
    set => name = value ?? 
        throw new ArgumentNullException(paramName: nameof(value), message: "New name must not be null");
}

This features enables using throw expressions in initialization expressions:

private ConfigResource loadedConfig = LoadConfigResourceOrDefault() ?? 
    throw new InvalidOperationException("Could not load config");

Previously, those initializations would need to be in a constructor, with the throw statements in the body of the constructor:

public ApplicationOptions()
{
    loadedConfig = LoadConfigResourceOrDefault();
    if (loadedConfig == null)
        throw new InvalidOperationException("Could not load config");

}
Note

Both of the preceding constructs will cause exceptions to be thrown during the construction of an object. Those are often difficult to recover from. For that reason, designs that throw exceptions during construction are discouraged.

Generalized async return types

Returning a Task object from async methods can introduce performance bottlenecks in certain paths. Task is a reference type, so using it means allocating an object. In cases where a method declared with the async modifier returns a cached result, or completes synchronously, the extra allocations can become a significant time cost in performance critical sections of code. It can become very costly if those allocations occur in tight loops.

The new language feature means that async methods may return other types in addition to Task, Task<T> and void. The returned type must still satisfy the async pattern, meaning a GetAwaiter method must be accessible. As one concrete example, the ValueTask type has been added to the .NET framework to make use of this new language feature:

public async ValueTask<int> Func()
{
    await Task.Delay(100);
    return 5;
}
Note

You need to add the pre-release NuGet package System.Threading.Tasks.Extensions in order to use ValueTask in Visual Studio 15 Preview 5.

A simple optimization would be to use ValueTask in places where Task would be used before. However, if you want to perform extra optimizations by hand, you can cache results from async work and reuse the result in subsequent calls. The ValueTask struct has a constructor with a Task parameter so that you can construct a ValueTask from the return value of any existing async method:

public ValueTask<int> CachedFunc()
{
    return (cache) ? new ValueTask<int>(cacheResult) : new ValueTask<int>(loadCache());
}
private bool cache = false;
private int cacheResult;
private async Task<int> loadCache()
{
    // simulate async work:
    await Task.Delay(100);
    cache = true;
    cacheResult = 100;
    return cacheResult;
}

As with all performance recommendations, you should benchmark both versions before making large scale changes to your code.

Numeric literal syntax improvements

Misreading numeric constants can make it harder to understand code when reading it for the first time. This often occurs when those numbers are used as bit masks or other symbolic rather than numeric values. C# 7 includes two new features to make it easier to write numbers in the most readable fashion for the intended use: binary literals, and digit separators.

For those times when you are creating bit masks, or whenever a binary representation of a number makes the most readable code, write that number in binary:

public const int One =  0b0001;
public const int Two =  0b0010;
public const int Four = 0b0100;
public const int Eight = 0b1000;

The 0b at the beginning of the constant indicates that the number is written as a binary number.

Binary numbers can get very long, so it's often easier to see the bit patterns by introducing the _ as a digit separator:

public const int Sixteen =   0b0001_0000;
public const int ThirtyTwo = 0b0010_0000;
public const int SixtyFour = 0b0100_0000;
public const int OneHundredTwentyEight = 0b1000_0000;

The digit separator can appear anywhere in the constant. For base 10 numbers, it would be common to use it as a thousands separator:

public const long BillionsAndBillions = 100_000_000_000;

The digit separator can be used with decimal, float and double types as well:

public const double AvogadroConstant = 6.022_140_857_747_474e23;
public const decimal GoldenRatio = 1.618_033_988_749_894_848_204_586_834_365_638_117_720_309_179M;

Taken together, you can declare numeric constants with much more readability.