Strange behaviour in C++...

[Jeffrey A Law]

  > Mike Stump writes:
  > > Forget about talk, talk is cheap.  What I have _never_ seen, are real
  > > performance numbers for real applications and/or micro benchmarks,
  > > that show what the numbers are. 
  > 
  > I'd like to second Mike's comments.  What you may be overlooking, Jeff, is
  > that the vast, vast majority of C++ objects are plain single inheritance,
  > which means that no thunks are needed and the non-thunks implementation
  > needs twice as much data space for the virtual function tables.  So it
  > simply does not matter if thunks are slower, since thunks are rarely
  > called.  It is a time-space tradeoff where the excess time cost is rare.
I understand completely, and unfortunately I left out a few important details.
Most importantly that there is a group working on a C++ ABI which (from my
understanding) will not include thunks (due to the performance concerns).

I don't think we want to burn a lot of time on thunks if there's a reasonable
chance the C++ ABI will take a different direction.

Hopefully the people involved in this project are reading this list & thread 
and
will chime in, particularly if I've mis-stated what's happening and why.

jeff