What are the limits to code generation from XML Schema in C#?

Question

I\'ve seen several questions regarding problems with generating classes from XML Schema using xsd.exe, along with suggestions for how to pre-process the schema (oft

Answer 1

Interesting question. Not too long ago, I was wondering about exactly the same thing.

I will show a couple examples of how far I got. My demonstration will not be complete (considering that the XML Schema specification is fairly comprehensive), but it should suffice to show...

• that you can do better than xsd.exe (if you're willing to adhere to certain patterns when you write your XML Schema); and
• that XML Schema allows type declarations that cannot be expressed in C#. This should not come as a big surprise, considering that XML and C# are very different languages with quite different purposes.

Declaring an interface in XML Schema

C# interfaces can be defined in XML Schema with complex types. For example:

corresponds fairly well to:

interface IFoo
{
    string Bar { get; set; }
    int?   Baz { get; set; }
}

The pattern here is that abstract and named (non-anonymous) complex types are basically the XML Schema equivalent of interfaces in C#.

Note some problems with the mapping:

• C# access modifiers such as public, internal etc. cannot be rendered in XML Schema.

• You have no way of expressing the difference between a C# field and a property in XML Schema.

• You cannot define methods in XML Schema.

• You also have no way of expressing the difference between a C# struct and class. (There's simply types in XML Schema, which roughly correspond to .NET value types; but they're much more restricted in XML Schema than complex types.)

• The usage of usage="optional" can be used to map nullable types. In XML Schema, you could define a string attribute as optional. Crossing over to C#, some loss in translation occurs: Since string is a reference type, it cannot be declared as nullable (since it's already nullable by default).

• XML Schema also allows usage="prohibited". This is again something that cannot be expressed in C#, or at least in a nice fashion (AFAIK).

• From my experiments, it appears that xsd.exe will never generate C# interfaces from abstract complex types; it will stay with abstract classes instead. (I'm guessing that this is to keep the translation logic reasonably simple.)

Declaring abstract classes

Abstract classes can be done very similarly to interfaces:

  
    ...
  

Here, you define an element with the abstract attribute set to true, and embed an anonymous complex type inside it.

This corresponds to the following type declaration in C#:

abstract class FooBase { ... }

Declaring classes

As above, but omit the abstract="true".

Declaring classes that implement an interface

  ...

This maps to:

interface IFoo { ... }

class Foo : IFoo { ... }

That is, you define both a named, abstract complex type (the interface), and a named element with that type.

• Note that the C# code snippet above contains ... twice, while the XML Schema snippet has only one .... How come?

  Because you cannot define methods (code), and because you also cannot specify access modifiers, you don't need to "implement" a complex type with the element in XML Schema. The "implementation" of the complex type would be identical to the original declaration. If the complex type defines some attributes, these simply get mapped to auto-properties in a C# interface implementation.

Expressing inheritance relationships in XML Schema

Class and interface inheritance in XML Schema can be defined through a combination of type extensions and element substitution groups:

This maps to:

class Base
{
    bool Foo { get; set; }
}

class Derived : Base
{
    string Bar { get; set; }
}

Note:

• We're again using named complex types. But this time, they're not defined abstract="true", since we're not defining any C# interface type.

• Note the references: Element Derived is in Base's substitution group; at the same time, complex type derived is an extension of complex type base. Derived has type derived, Base has type base.

• Named complex types that are not abstract have no direct counterpart in C#. They're not classes, since they cannot be instantiated (in XML, elements, not types, have roughly the same function as value constructors in F# or object instantiation in C#); neither are they truly interfaces, since they are not declared abstract.

Some things that I haven't covered in my answer

• Showing how one would declare, in XML Schema, a C# class type that implements several interfaces.

• Showing how complex content in XML Schema maps to C# (my first guess it that there's no correspondence in C# at all; at least not in the general case).

• enums. (They are realised in XML Schema by restricting a simple type via enumeration, btw.)

• const fields in a class (these would possibly map to attributes with a fixed value).

• How to map xsd:choice, xsd:sequence to C#; How to correctly map IEnumerable, ICollection, IList, IDictionary to XML Schema?

• XML Schema simple types, which sound like they're the corresponding concept of .NET value types; but are far more restricted and have a different purpose.

There's many many more things that I haven't shown, but by now you can probably see the basic patterns behind my examples.

To do all this correctly, one would have to systematically go through the XML Schema specification and see how each concept mentioned there maps best to C#. (There's perhaps no single best solution, but several alternatives.) But I explicitly meant to show only a couple of interesting examples. I hope that was still informative enough!