Re: Java name-mangling

[Per Bothner <per at bothner dot com>]

Mark Mitchell <mark@codesourcery.com> writes:

> I think I misunderstood the interfaces point.  I thought that Java
> (via CNI) actually wanted to be able to implement a Java class
> implementing an interface as a C++ class derived from a virtual base.

Well, ideally we would like to.  What we *need* to is implement a Java
class implementing an interface as a C++ class singly derived
from a non-virtual base.  It would be nice if C++ could know
about the implemented Java interfaces, and treat them as virtual
bases classes using a different ABI than the normal one.  However,
at this point that seems to be more pain that it is worth.

> If that were so, then the Java front-end would have to expect the
> layout of the C++ object to match its own layout of the same class if
> it were written entire in Java.

Yes, but only single non-virtual inheritance is important.
What we should make sure is that the ABI Java uses for interfaces
does not *conflict* with the C++ ABI.  I don't think is a problem.

> If, however, you're not going to actually implement Java classes
> implementing interfaces by coding them as C++ classes deriving from
> virtual bases, then this matters not at all.

No we're not.  As I said, I don't think we can.

> As long as the only compatibility has to be with C++ classes using
> single non-virtual inheritance, there shouldn't be any unduly severe
> problems.  Vptrs in the class are now located in a different place,
> and the contents of the vtable are slightly different: the RTTI and
> offset-to-top fields are at negative offsets from the point where the
> vptr points.

That doesn't sound difficult.
-- 
	--Per Bothner
per@bothner.com   http://www.bothner.com/~per/