C# 9.0 on the record

It’s official: C# 9.0 is out! Back in May I blogged about the C# 9.0 plans, and the following is an updated version of that post to match what we actually ended up shipping.

With every new version of C# we strive for greater clarity and simplicity in common coding scenarios, and C# 9.0 is no exception. One particular focus this time is supporting terse and immutable representation of data shapes.

Init-only properties

Object initializers are pretty awesome. They give the client of a type a very flexible and readable format for creating an object, and they are especially great for nested object creation where a whole tree of objects is created in one go. Here’s a simple one:

var person = new Person { FirstName = "Mads", LastName = "Torgersen" };

Object initializers also free the type author from writing a lot of construction boilerplate – all they have to do is write some properties!

public class Person{    public string? FirstName { get; set; }    public string? LastName { get; set; }}

The one big limitation today is that the properties have to bemutable for object initializers to work: They function by first calling the object’s constructor (the default, parameterless one in this case) and then assigning to the property setters.Init-only properties fix that! They introduce aninit accessor that is a variant of theset accessor which can only be called during object initialization:

public class Person{    public string? FirstName { get; init; }    public string? LastName { get; init; }}

With this declaration, the client code above is still legal, but any subsequent assignment to theFirstName andLastName properties is an error:

var person = new Person { FirstName = "Mads", LastName = "Nielsen" }; // OKperson.LastName = "Torgersen"; // ERROR!

Thus, init-only properties protect the state of the object from mutation once initialization is finished.

Init accessors and readonly fields

Becauseinit accessors can only be called during initialization, they are allowed to mutatereadonly fields of the enclosing class, just like you can in a constructor.

public class Person{    private readonly string firstName = "<unknown>";    private readonly string lastName = "<unknown>";        public string FirstName     {         get => firstName;         init => firstName = (value ?? throw new ArgumentNullException(nameof(FirstName)));    }    public string LastName     {         get => lastName;         init => lastName = (value ?? throw new ArgumentNullException(nameof(LastName)));    }}

Records

At the core of classic object-oriented programming is the idea that an object has strong identity and encapsulates mutable state that evolves over time. C# has always worked great for that, But sometimes you want pretty much the exact opposite, and here C#’s defaults have tended to get in the way, making things very laborious.

If you find yourself wanting the whole object to be immutable and behave like a value, then you should consider declaring it as arecord:

public record Person{    public string? FirstName { get; init; }    public string? LastName { get; init; }}

A record is still a class, but therecord keyword imbues it with several additional value-like behaviors. Generally speaking, records are defined by their contents, not their identity. In this regard, records are much closer to structs, but records are still reference types.

While records can be mutable, they are primarily built for better supporting immutable data models.

With-expressions

When working with immutable data, a common pattern is to create new values from existing ones to represent a new state. For instance, if our person were to change their last name we would represent it as a new object that’s a copy of the old one, except with a different last name. This technique is often referred to asnon-destructive mutation. Instead of representing the personover time, the record represents the person’s stateat a given time. To help with this style of programming, records allow for a new kind of expression; thewith-expression:

var person = new Person { FirstName = "Mads", LastName = "Nielsen" };var otherPerson = person with { LastName = "Torgersen" };

With-expressions use object initializer syntax to state what’s different in the new object from the old object. You can specify multiple properties.

The with-expression works by actually copying the full state of the old object into a new one, then mutating it according to the object initializer. This means that properties must have aninit orset accessor to be changed in a with-expression.

Value-based equality

All objects inherit a virtualEquals(object) method from theobject class. This is used as the basis for theObject.Equals(object, object) static method when both parameters are non-null. Structs override this to have "value-based equality", comparing each field of the struct by callingEquals on them recursively. Records do the same. This means that in accordance with their "value-ness" two record objects can be equal to one another without being thesame object. For instance if we modify the last name of the modified person back again:

var originalPerson = otherPerson with { LastName = "Nielsen" };

We would now haveReferenceEquals(person, originalPerson) = false (they aren’t the same object) butEquals(person, originalPerson) = true (they have the same value). Along with the value-basedEquals there’s also a value-basedGetHashCode() override to go along with it. Additionally, records implementIEquatable<T> and overload the== and!= operators, so that the value-based behavior shows up consistently across all those different equality mechanisms.

Value equality and mutability don’t always mesh well. One problem is that changing values could cause the result ofGetHashCode to change over time, which is unfortunate if the object is stored in a hash table! We don’t disallow mutable records, but we discourage them unless you have thought through the consequences!

Inheritance

Records can inherit from other records:

public record Student : Person{    public int ID;}

With-expressions and value equality work well with record inheritance, in that they take the whole runtime object into account, not just the type that it’s statically known by. Say that I create aStudent but store it in aPerson variable:

Person student = new Student { FirstName = "Mads", LastName = "Nielsen", ID = 129 };

A with-expression will still copy the whole object and keep the runtime type:

var otherStudent = student with { LastName = "Torgersen" };WriteLine(otherStudent is Student); // true

In the same manner, value equality makes sure the two objects have the same runtime type, and then compares all their state:

Person similarStudent = new Student { FirstName = "Mads", LastName = "Nielsen", ID = 130 };WriteLine(student != similarStudent); //true, since ID's are different

Positional records

Sometimes it’s useful to have a more positional approach to a record, where its contents are given via constructor arguments, and can be extracted with positional deconstruction. It’s perfectly possible to specify your own constructor and deconstructor in a record:

public record Person {     public string FirstName { get; init; }     public string LastName { get; init; }    public Person(string firstName, string lastName)       => (FirstName, LastName) = (firstName, lastName);    public void Deconstruct(out string firstName, out string lastName)       => (firstName, lastName) = (FirstName, LastName);}

But there’s a much shorter syntax for expressing exactly the same thing (modulo casing of parameter names):

public record Person(string FirstName, string LastName);

This declares the public init-only auto-propertiesand the constructorand the deconstructor, so that you can write:

var person = new Person("Mads", "Torgersen"); // positional constructionvar (f, l) = person;                        // positional deconstruction

If you don’t like the generated auto-property you can define your own property of the same name instead, and the generated constructor and deconstructor will just use that one. In this case, the parameter is in scope for you to use for initialization. Say, for instance, that you’d rather have theFirstName be a protected property:

public record Person(string FirstName, string LastName){    protected string FirstName { get; init; } = FirstName; }

A positional record can call a base constructor like this:

public record Student(string FirstName, string LastName, int ID) : Person(FirstName, LastName);

Top-level programs

Writing a simple program in C# requires a remarkable amount of boilerplate code:

using System;class Program{    static void Main()    {        Console.WriteLine("Hello World!");    }}

This is not only overwhelming for language beginners, but clutters up the code and adds levels of indentation. In C# 9.0 you can just write your main program at the top level instead:

using System;Console.WriteLine("Hello World!");

Any statement is allowed. The program has to occur after theusings and before any type or namespace declarations in the file, and you can only do this in one file, just as you can have only oneMain method today. If you want toreturn a status code you can do that. If you want toawait things you can do that. And if you want to access command line arguments,args is available as a "magic" parameter.

using static System.Console;using System.Threading.Tasks;WriteLine(args[0]);await Task.Delay(1000);return 0;

Local functions are a form of statement and are also allowed in the top level program. It is an error to call them from anywhere outside of the top level statement section.

Improved pattern matching