Python.NET performs some conversions between .NET and Python automatically. For example, when Python calls this C# method:

void Foo(int bar) { ... }

via Foo(42), Python value 42 of type int will be automatically converted to .NET type System.Int32. Another way to invoke those conversions is to call dotNetObject.ToPython() (available as an extension method) or pyObject.As<T>() to convert PyObject to .NET.

An incomplete list of Python types, that are converted between Python and .NET automatically: most numeric types, bool, string, Nullable<T> to its Value or None and back, etc.

A custom conversion (Codec) can be defined by implementing one of the (or both) interfaces:

• Python.Runtime.IPyObjectDecoder to marshal Python objects to .NET

interface IPyObjectDecoder {
  bool CanDecode(PyObject objectType, System.Type targetType);
  bool TryDecode<T>(PyObject pyObj, out T value);
}

• Python.Runtime.IPyObjectEncoder to marshal .NET objects to Python

interface IPyObjectEncoder {
  bool CanEncode(System.Type);
  PyObject TryEncode(System.Object);
}

Once implemented, instances have to be registered with Python.Runtime.PyObjectConversions.RegisterEncoder/-Decoder. One can override some of the default conversions by registering new codecs.

When multiple codecs are registered, the runtime will first try the ones, that were registered earlier. If you need to have some grouping of codecs by priority, create and expose Python.Runtime.Codecs.EncoderGroup/-.DecoderGroup. For example:

public static EncoderGroup HighPriorityEncoders{ get; } = new EncoderGroup();

void Init() {
  PyObjectConversions.RegisterEncoder(HighPriorityEncoders);
  var lowPriorityEncoder = new SomeEncoder();
  PyObjectConversions.RegisterEncoder(lowPriorityEncoder);
}

... some time later

HighPriorityEncoders.Add(new SomeOtherEncoder());