A c++ dlopen/PV interface test summary.

Here the test/, test_noabst, test_samedota, and test_novirt
subdirectories test different combinations of
c++ features and dynamic loading.

Note that trying to 'fake it' by doing a C cast or static_cast
from a base (vPort) to a derived (sPort) will simply not compile
under ISO c++ (which is what we're stuck with).

test/		dynamic loading and pure virtual interfaces
		make clean test # demo the way c++ abi standard requires
		make testno # demo the way private dlopen fails c++.

test_novirt/	demonstrates that you cannot eliminate the 
		virtual keyword and still have polymorphism,
		which is according to the c++ standard.
		dynamic_cast fails compile and static_cast is
		unsafe. If using static_cast to convert Port to
		a specific port, might as well have CCA define
		Port as void* rather than a base type. Good luck
		ever debugging either one.

test_noabst/	demonstrates that you can make pure virtual (classic)
		interfaces such that all the functions have default
		implementations [exit(1)] but that this does not
		solve the dynamic loading problem.
		make clean test # gives the same result as test/
		make testno # gives the same fail as test/.

test_samedota/	demonstrates the same thing as test_noabst above,
		but under the condition that the .so are built
		against the same .a. Both .so end up with a copy
		of what's in the .a, and everything is the same at runtime.

Conclusion:
	The best way to get c++ polymorphism and uses/provides
ala the cca is exactly as the 1999 classic and as scirun
have it: pure virtual, [virtual int fcn()=0;].

	Putting in function body dummies doesn't solve
the dynamic loading problem and misleads new implementors
into thinking they've successfully implemented a Port
without doing anything but inheriting that port.

	All Ports (and any virtual interface types that
are in the function arguments of those ports) must be linked
statically in main at build type or dynamically loaded with
a global flag somehow specified.