I think it would be better to handle all this in Record.translate. It's these lines that will write a struct declaration inline. When the definition is already know, don't execute those lines.

I am not sure I understand. Lines you have linked get called from writeRecord and are responsible for writing the fields (CXCursor_FieldDecl). Struct declarations get written from the return value of writeRecord (return context.data).

The lines I've linked above are the lines responsible for translating the actual type declaration. The lines below are for the field declaration.

This is the flow of for translating a struct:

1. All top level nodes are iterated and translated, here
2. When a struct is seen, it will start translating that, here
3. writeRecord is called which will translate the identifier of the struct, here
4. writeRecord then calls the delegate which will start iterating all the children nodes of the struct, here
5. When a field is seen it will check if the type is not exposed and has a declaration, here
6. If the type is not exposed and has a declaration it will translate the declaration of the type here
7. If the declaration is a struct it will start from the beginning at step 2

This code, at step 6, is there to handle a struct and field in the same declaration, i.e.

struct Foo
{
    struct Bar
    {
        int x;
    } bar;
};

Ok, thanks, I have missed step 6 where it went in recursion and was confused by seemingly broken iteration over struct members.

Please use camel casing.

Updated