fix for PR 4447: is this really correct?

[Joe Buck]

I wrote:
> > I'd prefer to have neither the ICE nor the ABI breakage, but I'd
> > prefer the latter to the former.

However, it turns out that the strange mangling is exactly what the ABI
specifies, according to Kriang, so the issue is moot.

> > News flash: 3.1 will have a couple of minor ABI bug fixes
> 
> :-(  Such is life.
> 
> > so it seems that we're already in a position to break the ABI,
> 
> We don't just doom 3.1 to breaking, but some random future version of
> the compiler, with luck, it will just be 3.1.

We are converging to a written spec, which is supposed to be supported
by multiple compilers.  If we leave the 3.0 deviations from this spec,
we don't keep our word about implementing the spec.  We had an argument
about whether we should have fixed the first error discovered in 3.0.1,
and we decided to wait until 3.1.  Just as well, since one more was
found.  In practice, the main 3.0 -> 3.1 incompatibility will be 
in the standard library.