Yes, i know that. But XS is actually needed since i want to write a OO interface to a (non trivial) Win32 C API.
(and for sure, i know how to walk! :) )
so for now i have the following code

Foo-Bar/typemap

TYPEMAP

PBar         T_PTROBJ
[download]

Foo-Bar/Bar.xs

typedef struct __Bar {
// ...
} Bar, *PBar;

MODULE = Foo::Bar        PACKAGE = Foo::Bar

PBar
new(class);
    const char* class
 PREINIT:
     PBar        retval;
 CODE:

     New(0, retval, 1, Bar);

     RETVAL = retval;
 OUTPUT:
     RETVAL
[download]

Foo-Bar-Builder/typemap

TYPEMAP
PBarBuilder         T_PTROBJ
PBar         T_PTROBJ
[download]

Foo-Bar-Builder/Builder.xs

typedef struct __BarBuilder {
// ...
} BarBuilder, *PBarBuilder;


MODULE = Foo::Bar::Builder        PACKAGE = Foo::Bar::Builder

PBarBuilder
new(class);
    const char* class
 PREINIT:
     PBarBuilder retval;
 CODE:

     New(0, retval, 1, BarBuilder);

     RETVAL = retval;
 OUTPUT:
     RETVAL

MODULE = Foo::Bar::Builder        PACKAGE = Foo::Bar::BuilderPtr

PBar
create(void);
 PREINIT:
     PBar retval;
 CODE:

    RETVAL = retval;
 OUTPUT:
     RETVAL
[download]

so, the question is: do i have to call Foo::Bar::new inside Foo::Bar::Builder::create? or do i have to reimplement the new function of Foo::Bar::new inside Foo::Bar::Builder, or what?
so, the Foo::Bar::new is now defined inside the Foo/Bar/Bar.dll, and the C compiler does not know, where to find it. I guess it does not help to link against Foo\Bar\Bar.lib, since the dll just exports

1    0 00001000 _boot_Foo__Bar
2    1 00001000 boot_Foo__Bar

Get the point?