-Wabi is misdesigned (answers the wrong question)

[Joe Buck]

On Fri, Oct 24, 2003 at 10:59:52AM -0700, Mark Mitchell wrote:
> On Wed, 2003-10-22 at 17:20, law@redhat.com wrote:
> > In message <20031022165429.A20170@synopsys.com>, Joe Buck writes:
> >  >It's not too complicated to implement, as the compiler already knows how
> >  >to compute both layouts.
> > But it may be difficult to compute one layout without side effects which
> > may make computing the second layout exceedingly difficult.
> 
> Yes, that's the case.
> 
> It's probably possible to do what Joe wants, but it's not trivial.  One
> would have to be very careful not to have the first layout algorithm
> bias the results of the second; after running the first, you'd want to
> clear out the layout entirely and do it again.

Another alternative is the following: if we had a dump format that
would contain a description of the layout of each class that is encountered,
then we could simply run the compiler twice, once with the 3.2 ABI,
once with the 3.4 ABI, and then diff the dumps.