TITLE: Illegal initialization of virtual bases


PROBLEM: b91926@fnclub.fnal.gov (David Sachs), 2 Jun 93
	 Fermi National Accelerator Laboratory


Markku Sakkinen, in his recent article in "c++ Report", pointed out one
case in which the ARM rules absolutely cannot be followed. Suppose you
have a virtual base, which has a non-virtual base: e.g.

struct a ....

struct b : a ....

struct c : virtual b ....

The wording of the ARM would appear to require (for initializing a c
object) that the b subobject be initialized before ITS a subobject.

Personally, I would prefer, that virtual bases be initialized the FIRST
time they are encountered in the normal initialization sequence. This would
mean that a virtual base that occurs only once would be initialized as if
it were not virtual. This would also tend to minimize the code generated,
as only classes that actually specify the virtual base would have to be
able to initialize it.

This area of the standard is a can of worms, that might best be left unopened.