Re: gcc 3.2 is now the default compiler in unstable
On Wed, Jan 08, 2003 at 10:38:18PM +0100, "Martin v. L?wis" wrote:
> I haven't checked, but I would expect that migration of gcc-defaults
> itself would render a number of packages in testing uninstallable,
> because they happen to depend on earlier gcc versions (on purpose or by
> mistake). Then, the updating mechanism won't migrate gcc-defaults into
> testing. So I expect that magic is needed to override the conservative
> updating procedure.
The gcc-defaults change doesn't seem to me like the kind of change that
causes problems for testing. Such problems occur when packages become
uninstallable; what has become uninstallable due to the new
Given that gcc-2.95 and gcc-3.2 etc. are coinstallable, your suggestion
would require packages to have less-than versioned dependencies or
equal-to versioned dependencies on one of gcc-defaults' binary packages,
as I read it. There are no such packages currently in unstable.
> As for needing 6 months for the migration even in unstable: Many
> release-critical bugs will be filed against various packages, as they
> fail to compile or work with g++ 3.2. Fixing those problems will take
> real manpower, delaying the transition by many weeks, per stratum (from
There is only likely to be a problem if a large number of those
release-critical bugs end up against the core compiler and libraries,
preventing g++-3.2 (>= 1:3.2.2-0pre2) from itself being eligible for
testing soon or once glibc is fixed. If they're just on leaf packages,
that's OK; their old versions currently in testing are still installable
after gcc-defaults is upgraded. My system didn't disappear when I
performed the upgrade here.
For better or worse, testing doesn't currently track build-dependencies.
If it did, then this transition would probably be more difficult. As it
is, the fact that libstdc++ has changed soname and is coinstallable with
the old versions is a feature that many painful transitions don't share.
Colin Watson [email@example.com]