Stefan Teleman <Stefan.Teleman at Sun.COM> writes: > Ian Lance Taylor wrote: > >> The book is incorrect. > > And, apparently so is everyone else.
I'm not sure who "everyone else" is, so I can't reasonably respond. > One thing i am not going to do is make a far reaching architectural > decision about GCC's future C++ ABI compatibility, based on mailing > list statements, in absolute disregard for significant and documented > C++ ABI compatibility breakage, within GCC Major Version 4, and in > direct contradiction of your claimed compatibility statements. > > In this particular case, past performance is indicative of future results. > > Barring an official commitment statement from the GCC developers to > the effect of "GCC will never break its C++ ABI on Solaris in an > incompatible way, within the boundaries of a Major GCC Release > Version", sent to this list, and made an integral part of the ARC Case > Materials for GCC4, GCC4's C++ ABI on Solaris will maintain its "C++ > ABI subject to change in an incompatible way, and without notice in a > future release" status. > > Solaris' tolerance for ABI breakage is zero. We don't do "sorry we > broke the ABI" statements. You obviously must make your own decisions. I simply wanted to correct your statement about C++ ABI incompatibilities in GCC. We understand that ABI breakages are a very bad idea, and we no longer do them. It's certainly true that we did do them before the 3.4 release, for what we thought were good reasons. In the 3.4 release we rewrote the C++ parser, and committed to a single ABI (documented at http://codesourcery.com/cxx-abi/ -- we use it for all targets, not just for Itanium). If you choose to act as though GCC will break the C++ ABI with every minor release, then you will certainly be safe. I don't know what, if anything, that decision will cost you. Ian
