Re: C++ PATCH: Remove old ABI mangling

[Per Bothner <per at bothner dot com>]

Mark Mitchell <mark@codesourcery.com> writes:

> I knew there would be issues with [Java] layout, so I didn't plan on
> touching those yet, but what's the dependency with mangling?

Libgjc needs to reference to the the Class object for certain types,
i.e. the Java version of the rrti object of a class.  Unfortunately,
we don't have any clean way of doing it, so we refer to the mangled
name directly.  For example in natObject.cc:

        #define CloneableClass _CL_Q34java4lang9Cloneable
        extern java::lang::Class CloneableClass;

This is of course really bad, but nobody has gotten around to
implementing a cleaner mechanism.  CNI (the Compiled Native Interface)
needs to be extended to provide a standard mechanism to reference
the run-time Class for a given compile-time class.  This has been
discussed a few times on the java-discuss mailing list.

The nicest solution would be to merge Java Classes with C++ type_info
objects (basically. make Class be a subtype of type_info).  That means
we could use (java::lang::Class*)typeid(java::lang::Cloneable).  It
also improves cross-language exception handling.  However, it would
be a non-trivial amount of work - and it might break the new C++ ABI.

An easier implementation would be to add a __builtin_java_class
primitive to G++.  Even easier might be to add appropriate declarations
to gcjh.  If the Class is defined as a static member of its class,
with a conventional name, when we just need to make sure gcjh emits
a declaration for the static member.  We are almost there:  Gcj does
define such a static member, named "class".  This makes sense from
a Java point of view (where 'Xxx.class' is valid syntax that returns
the Class for Xxx), but it would break the C++ parser.
-- 
	--Per Bothner
per@bothner.com   http://www.bothner.com/~per/