[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: GCC 4.2 transition

Andreas Barth a écrit :
> * Aurelien Jarno (aurelien@aurel32.net) [070721 00:33]:
>> On Fri, Jul 20, 2007 at 10:48:32PM +0200, Andreas Barth wrote:
>>> * Aurelien Jarno (aurelien@aurel32.net) [070720 21:15]:
>>>> We (glibc maintainers) plan to do a change in glibc first. We will drop
>>>> libc6-sparcv9 and change the optimizations of libc6 to SPARC v9 from
>>>> SPARC v8.
>>>> Doing it on the glibc first have the advantage that we can put a check
>>>> in the preinst script to stop the installation on a SPARC v8 system. As
>>>> libc6 is installed on all systems, that should prevent system breakages
>>>> with SIGILL on random packages.
>>>> This is already implemented in the SVN, and we plan to do the upload on
>>>> Sunday.
>>> I hope this doesn't yet include another shlib bump (though it would be
>>> good if the sparc v9 binaries have one).
>> Given the current situation wrt the transitions, we won't bump the shlib, 
>> though strictly speaking that should be necessary.
> Actually, I would like to see an shlib bump before gcc is changed (so
> that gcc and all binaries built by the new gcc depend on the new glibc).
> However, I agree that it is currently a unfortunate time for such a
> change.
> So, some ideas come to my mind:
> 1. delay glibc upload until glib transitioned to testing
> 2. do another upload with shlib bump after glib transitioned to testing,
> and wait with gcc to that upload.
> (there might be better combinations, but none comes currently to my
> mind)

We really wants to do an upload of the glibc soon to fix various things.
What about the following combination:
1. Upload glibc without the shlib bump, but with the switch to SPARC V9
optimizations + check in the preinst script.
2. Upload glibc with shlib bump when the release team think it is the
good time to do that.
3. Upload gcc with the switch to SPARC V9 optimizations.

We may repeat the step 1 a few times. Ideally there should be almost now
difference between the upload in step 1 and in step 2 (except the shlib
bump) to ensure the version could go very quickly to testing.

  .''`.  Aurelien Jarno	            | GPG: 1024D/F1BCDB73
 : :' :  Debian developer           | Electrical Engineer
 `. `'   aurel32@debian.org         | aurelien@aurel32.net
   `-    people.debian.org/~aurel32 | www.aurel32.net

Reply to: