Reputation: 825
I have the following code for C++, in a templated class that represents a point. I would like to translate it into C#:
template <class T>
class Point
{
public:
T x;
T y;
T z;
template<typename U> explicit Point(const Point<U> &p)
: x((T)p.x), y((T)p.y), z((T)p.z)
{
}
}
This code enables a point of a given type to be explicitly cast into a point of another type. For example, you may use it something like (admittedly I am not 100% sure on the syntax here, but I get the concept):
Point<float> p;
Point<int> q = (Point<int>)p;
How could I enable the equivalent to this in C#? So far I have:
public class Point<T>
{
public T X { get; set; }
public T Y { get; set; }
public T Z { get; set; }
// Constructors exist that accept X, Y, and Z as parameters
public static explicit operator Point<U>(Point<T> p)
{
}
}
This gives an error, however, saying "U" is undefined. This makes sense... but how/where do I define U? Is my approach incorrect?
The difference between my question and the one here is that I am simply changing the underlaying type of the generic class via a cast... not trying to change one generic class into a different generic class with the same underlaying type.
Upvotes: 4
Views: 4102
Reputation: 5444
You cannot define additional generic type constraints, but you can do something like this, using operators and methods.
public class Point<T>
{
public T X { get; set; }
public T Y { get; set; }
public T Z { get; set; }
public static explicit operator Point<T>(Point<int> v)
{
return v.As<T>();
}
public static explicit operator Point<T>(Point<double> v)
{
return v.As<T>();
}
public static explicit operator Point<T>(Point<float> v)
{
return v.As<T>();
}
public Point<TU> As<TU>()
{
return new Point<TU>()
{
X = ConvertTo<TU>(X),
Y = ConvertTo<TU>(Y),
Z = ConvertTo<TU>(Z)
};
}
private static TU ConvertTo<TU>(T t)
{
return (TU) Convert.ChangeType(t, typeof(TU));
}
}
Usage:
Point<double> d = new Point<double>()
{
X = 10d, Y = 10d, Z = 10d
};
Point<int> i = (Point<int>) d;
Point<float> f = (Point<float>) i;
d = (Point<double>) f;
Upvotes: 1
Reputation: 1196
You cannot declare operators with additional generic type arguments, but you can declare ones to or from specific generic types like Point<int>
. C# will also not let you perform arbitrary conversions by casting from or to T
.
The least boilerplate heavy option that maintains a modicum of type safety would be to constrain the T
parameter to IConvertible
:
public class Point<T> where T : IConvertible
{
// ...
public static explicit operator Point<int>(Point<T> point)
{
// The IFormatProvider parameter has no effect on purely numeric conversions
return new Point<int>(point.X.ToInt32(null), point.Y.ToInt32(null), point.Y.ToInt32(null));
}
}
However, this will not prevent users from declaring nonsensical, unsupported types such as Point<DateTime>
which will then throw at runtime when attempting a conversion.
Upvotes: 1
Reputation: 21709
Similar to Kevin's answer, but without dynamic
is to use a double cast:
public Point<U> To<U>()
{
return new Point<U>((U)(object)X, (U)(object)Y, (U)(object)Z);
}
Both of our answers don't catch any issues at compile-time.
Upvotes: 3
Reputation: 35594
I think the best you can get is this:
public class Point<T>
{
public T X { get; set; }
public T Y { get; set; }
public T Z { get; set; }
public Point<U> As<U>()
{
return new Point<U>()
{
X = Convert<U>(X),
Y = Convert<U>(Y),
Z = Convert<U>(Z)
};
}
static U Convert<U>(T t) => (U)System.Convert.ChangeType(t, typeof(U));
}
You cannot define a generic conversion operator, so you need it to be an explicit function. Moreover, a simple cast (U)t
won't work, so you need Convert.ChangeType
(which will work if your types are numeric).
Usage:
var p1 = new Point<int> { X = 1, Y = 2, Z = 3 };
var p2 = p1.As<double>();
Upvotes: 3
Reputation: 39007
As far as I know, this kind of generic cast is only allowed in C# if there's some kind of inheritance relationship between T
and U
.
The closest equivalent would be to define a generic method for the conversion:
public Point<U> To<U>()
{
dynamic p = this;
return new Point<U>((U)p.X, (U)p.Y, (U)p.Z);
}
You cannot convert directly T
to U
as the compiler has no way to know whether it'll be safe. I use the dynamic
keyword to bypass that restriction.
Upvotes: 3