Re: Testsuite ad RedHat 7

[Jakub Jelinek <jakub at redhat dot com>]

On Wed, Oct 11, 2000 at 10:37:48AM +0200, Marc Espie wrote:
> In article <200010101603.SAA04655@quatramaran.ens.fr> I wrote:
> >I think the Steering Committee made it pretty clear that shipping a 2.96
> >e.g., *development* snapshot, with an official `release' is a very stupid
> >decision.
> 
> Just to clarify, the Steering Committee only made it clear that 2.96 never was
> guaranteed to be compatible with anything. The inference that shipping it
> is a stupid decision is mine.
> 
> I do happen to think that shipping such a compiler with an actual release of
> a product is an idiotic move.  Especially with a wide-spread product such as
> RedHat. Because it will further the impression that gcc and C++ are unstable
> stuff.  Also because people will need to update again in a few months, when
> gcc 3.0 is out. And also because people will get burnt over this, as they 
> don't really expect things not to be binary incompatible.

I think you missed the point that even if we shipped g++ 2.95.2, it would be
binary incompatible with g++ 2.95.2 from other distributions (because we
ship with glibc 2.1.9x, soon to be 2.2 and libstdc++ soname contains the
glibc version number in it, so C++ binaries built with 2.95.2 on glibc 2.2
system would be linked against libstdc++-libc6.2-2.so.3 while binaries built
with 2.95.2 on glibc 2.1.[0-3] systems are linked against
libstdc++-libc6.1-2.so.3).

Also, if we had to chase the numerous bugs in 2.95.x, it would have smaller
value to the community than when we spent the time on stabilizing the
current compiler. Note that all bugs reported so far against the RH compiler
were reproduceable in vanilla CVS as well.

	Jakub