Re: Beyond GCC 3.0: Summing Up

[Andreas Jaeger <aj at suse dot de>]

kenner@vlsi1.ultra.nyu.edu (Richard Kenner) writes:

>     I can tell that right now, it is not an easy job to do. It takes more
>     resources you can image. Why not let Linux/xxxBSD distribution vendors
>     do it for us. The only thing we have to do is to make frequent bug fix
>     releases to fix the regressions they have found. But if it takes a few
>     months for a bug fix, I don't think any Linux/xxxBSD distribution
>     vendors will try the new gcc to build their whole distribution. They
>     need a responsive compiler provider.
> 
> I don't follow you.  They would be doing this *before* a release, on a
> tree that is frozen as part of the release process.  We would not make a
> release until they confirm that it works.

I can recompile a whole distribution with a new compiler but I see the
following problems:
- New compilers need changes to programs.  With GCC 3.0 the C++
  and libstdc++ changes might break a lot of non-standard conforming
  C++ programs.  It does take a lot of time to check whether we have a
  compiler problem or a broken program.
- Some programs come with self tests but most don't - and no testsuite
  is complete.  Testing 2000+ packages by hand is not possible.  After
  a recompile of a whole distribution, you'll have a number of ICEs
  but only few miscompilations since finding them is not that easy.
  Or should I just send a list of packages and tell you: These are
  broken - without a detailed analysis?

If people think this is something that would be really helpfull, I
could discuss it within SuSE and we might do such a test for one or
two platforms, e.g. ia32 and PowerPC, on a limited timeframe.

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj