C#•5mo ago

✅ Positional record with alternate constructor

Say we have this record:

public record Point(float x, float y);

public record Point(float x, float y);

And we wanna provide a way for it to be initialized from doubles too:

public record Point(float x, float y) {
    public Point(double x, double y) {
        this.x = (float) x;
        this.y = (float) y;
    }
}

public record Point(float x, float y) {
    public Point(double x, double y) {
        this.x = (float) x;
        this.y = (float) y;
    }
}

But the compiler complains "A constructor declared in 'record' with parameter list must have 'this' constructor" It wants me to do something like this:

public record Point(float x, float y) {
    public Point(double x, double y) : this(0, 0) { // The numbers here can be anything
        this.x = (float) x;
        this.y = (float) y;
    }
}

public record Point(float x, float y) {
    public Point(double x, double y) : this(0, 0) { // The numbers here can be anything
        this.x = (float) x;
        this.y = (float) y;
    }
}

This doesn't feel right to me, seems like I'm going against the language and doesn't look like idiomatic C#. Because we're setting the fields then setting them again. But I don't see why this can't work, it should be possible to specify a constructor with some other parameters than the primary constructor, assuming all the fields get initialized. I'm looking for ways to achieve this result with idiomatic C# that isn't a lot of code, like removing the positional parameters and writing those properties manually. Thank you.

13 Replies

333fred•5mo ago

Just do:

public record Point(float x, float y) {
    public Point(double x, double y) : this((float)x, (float)y) {}
}

public record Point(float x, float y) {
    public Point(double x, double y) : this((float)x, (float)y) {}
}

Though to be honest, this cast and second constructor is an anti-pattern, and you shouldn't do it Either use double as the types of the properties, or make the user do the cast

hadeedjiOP•5mo ago

Yeah, got it what you're saying This was an example, in general shouldn't it be possible to have a second constructor for a record that takes some other parameter and initializes it? In some situations that would be a good idea. Say you have a color record with rgb bytes, and want to initialize it with a float vector3. I don't get why it isn't allowed What's really throwing me off is that the compiler isn't allowing something, but by adding : this(0, 0) it lets me do it. But I don't wanna add that in, looks like an anti pattern or code smell. What are your thoughts? Maybe when an anti pattern makes the thing you want to do work, the thing you want to do isn't a good idea?

333fred•5mo ago

Point.x and Point.y are in scope throughout the rest of the type. They must be initialized before anything else happens Consider as an example:

public record Person(string FullName)
{
    public string FirstName { get; } = FullName.Split(' ')[0];
}

public record Person(string FullName)
{
    public string FirstName { get; } = FullName.Split(' ')[0];
}

That property initializer will run before the body of your alternate constructor, and FullName must have a valid value or it will null ref

hadeedjiOP•5mo ago

Got it. Final question, how do you think you'd handle a color record struct and want to provide a constructor that inits it with a float vec3

333fred•5mo ago

Say you have a color record with rgb bytes, and want to initialize it with a float vector3

This would be perfectly fine.

public record struct RGB(float R, float G, float B)
{
    public RGG(Vector3 vector) : this(vector.X, vector.Y, vector.Z) {}
}

public record struct RGB(float R, float G, float B)
{
    public RGG(Vector3 vector) : this(vector.X, vector.Y, vector.Z) {}
}

hadeedjiOP•5mo ago

That makes sense. Thanks a lot!

333fred•5mo ago

You're welcome. Please $close the thread if you didn't have any other questions 🙂

MODiX•5mo ago

If you have no further questions, please use /close to mark the forum thread as answered

jcotton42•5mo ago

Another approach here might also be a FromVector static method

public record struct RGB(float R, float G, float B)
{
    public static RGB FromVector(Vector3 vector) => new(vector.X, vector.Y, vector.Z);
}

public record struct RGB(float R, float G, float B)
{
    public static RGB FromVector(Vector3 vector) => new(vector.X, vector.Y, vector.Z);
}

I personally am liking static methods as "named constructors" more and more lately @hadeedji.

333fred•5mo ago

I mean, you certainly could, but I'm not sure I would do that here

hadeedjiOP•5mo ago

Thanks, this is what some would call a Factory, right?

333fred•5mo ago

Factory method, perhaps. An actual factory is usually a separate type entirely

Angius•5mo ago

In some cases, you could also use generic math to take any floating-point type, be it float or double.

public record Point<T>(T X, T Y) where T : IFloatingPoint<T>;

public record Point<T>(T X, T Y) where T : IFloatingPoint<T>;

https://learn.microsoft.com/en-us/dotnet/standard/generics/math

Gaming

Programming

✅ Positional record with alternate constructor