TypeUtil.ToCSharpString(this Type) does not handle nested generic types correctly

[stakx]

Migrated from castleproject/Core#365.

---

@pmg23 wrote:

The useful extension method TypeUtil.ToCSharpString(this Type) does not correctly deal with types nested within generic types [as of Core 4.2.1]. For example:

internal class A<T1>
{
    internal class B {}
    internal class C<T2> {}
}

Assert.That(typeof(A<string>.B).ToCSharpString(),
        Is.EqualTo("A<String>.B")); // FAIL: actually "A<·T1·>.B<String>"
Assert.That(typeof(A<string>.C<int>).ToCSharpString(),
        Is.EqualTo("A<String>.C<Int32>")); // FAIL: actually "A<·T1·>.C<String, Int32>"

---

@stakx replied:

(Shameless advertisement:)

Type name formatting is surprisingly complex because there are quite a few corner cases to deal with. I happen to have written a small library recently (stakx/TypeNameFormatter) that would solve this problem. It's available as a source code NuGet package.

---

@stakx replied:

That being said, I cannot seem to find a method TypeUtil.ToCSharpString(this Type) in Castle Core (this repository). Did you mean this method in Castle Windsor (castleproject/Windsor)?

[pmg23]

Sorry, I should have given the full name: Castle.Core.Internal.TypeUtil.ToCSharpString(this Type); which (as @stakx kindly points out) is part of Windsor. We observe this behaviour in Windsor 4.1.0.

Yep, looks like unbound generic types are being passed recursively through the AppendGenericParameters method here.

[stakx]

Not quite. The thing is this:

• Under the hood, nested types "inherit" the generic type parameters of their enclosing type(s). This explains why the type C from A<String>.C<Int32> ends up having two generic type arguments in the formatted output. (It's perhaps easier to think of C as an independent type instead of somehow being connected to A to understand why. Note though that its IL name is still C`1 despite having two parameters.)

• Reflecting over the enclosing type will return the type definition A<T1>, i.e. the type argument String is lost during that call. This explains why you end up with A<T1>... instead of A<String>.

To fix the formatting of nested generic types, you'd have to:

• Figure out how many own parameters the nested type declares (let's call this number n). For CLS-compliant types, this is equal to the number of generic type args/params minus that of the enclosing type.

• Only use the last n generic params/args for the nested type. The remaining (leading) args need to be passed recursively to the formatting of the enclosing type(s).