Re: RFA: add support for GNU V3 C++ ABI

[Daniel Berlin <dan@cgsoftware.com>]

Jim Blandy <jimb@zwingli.cygnus.com> writes:

> This is a first cut at supporting the new V3 C++ ABI.  It provides all
> the functions currently abstracted in the cp-abi.h interface,
> including RTTI and virtual function calls.
> 
> It doesn't handle virtual base classes, since there is no support for
> that in cp-abi.h; hopefully, that'll be addressed soon.
> 
The patch is approved (the part i can approve, anyway. The minsym
stuff isn't mine to approve, but I can at least vouch for it's
correctness. There is no case where a v3 mangled symbol doesn't start
with "_Z". We default to the v2 abi, so we don't need to check for
that ABI and switch.)

And if someone from HP doesn't get back to me in about 1 week, tops, i'm
just going to say "screw it", and if the baseclass_offset stuff breaks
the HP aCC stuff, then when someone knowledgable enough to fix it
comes along, they'll fix it.

This is actually what is holding me up.  I'm trying hard to do the
right thing here, and make sure we didn't break it, but without access
to an HP machine with HP aCC and a lot of time to reverse engineer all
the cases, and without anyone who will answer the C++ ABI related
questions, I can't maintain or support it. 

I'd normally be happy to let it sit there bit rot, but we can't in
this case (because moving baseclass_offset to a common area requires
changing it's args to be a superset of find_rt_vbase_offset and
baseclass_offset, and find_rt_vbase_offset wants an argument that we
can't always give it), and since it's holding up support for an ABI for a
new version of the GNU compiler, which is, IMHO, more important than
code nobody seems willing to even answer questions about the
algorithms it's using, ...

It's also likely it's not incorrect, we just can't prove it without
knowing how to do it ourselves.

If anyone else on the list knows how to calculate baseclass offset
addresses for the HP aCC ABI, and thus, could tell me if a small piece
of code is correct, or explain the algorithm in english, i'd greatly appreciate it.

--Dan