[clang] [CodeGen] Mark weak vtables unnamed_addr on duplicable-vtable targets (PR #205930)

[Chuanqi Xu via cfe-commits]

ChuanqiXu9 wrote:

> Honestly the ever increasing edge cases make me wonder if we should consider removing the exact final subclass optimization, the complexity of the fallout continues to cause problems, or gate it on a different flag.

It might be bad to revert an optimization landed a lot of years... I think given the current status, it may be fine to land this as a workaround. For the work in  longer term, I think it may be better to rethink/redesign the Itanium-Apple ABI if possible.

https://github.com/llvm/llvm-project/pull/205930