glibc 2.1.91 based distributions :-(.

[Andreas Jaeger]

>>>>> Ulrich Drepper writes:

Uli> Andreas Jaeger <aj@suse.de> writes:
>> glibc will definitly change in an incompatible way, e.g. the
>> RPC-IPv6 code has to be fixed.

Uli> As Jakub mentioned, the verson we released comes without the IPv6
Uli> stuff and is therefore copmatible with 2.1.  Jakub's patch is the one
Uli> I'll apply if I don't get a solution implementing IPv6 correctly in
Uli> the next few days.  Therefore there is really no compatibility
Uli> problem.

Uli> And yes, I would have wished if the public beta would have been later.
Uli> But so be it.  What I managed to do before was to add all the binary
Uli> incompatible changes before the release (and Bruno had the CTYPE
At least all you're aware of for now ;-)
Uli> changes also ready in time).  This beta was the reason I could spend
Uli> so much time on glibc lately (and believe, it was a lot).
I had that impression already and agree that's good ;-)

Uli> So, if you have anything you think is wrong let me know now.  My todo
Uli> list is almost empty.  There are not many solvable PRs left, I know of
Uli> no critical problems myself, there will be cleanups and various
Uli> profiling measurements.

AFAIK the only problems are RPC-IPv6 (which you took care of) and
getipnodebyaddr.

>> This has caused already in the past some problems, e.g. the usage
>> of an early test release of glibc 2.0 on PowerPC and Alpha (we changed
>> a serious bug and introduced an incompatibility in the final 2.0).

Uli> Yes, and this would be done here as well *if* it's necessary.  It's a
Uli> risk the people making the distribution here in house took.

Good to hear that those folks are aware of the problems.  So, the
only;-) problem are the users which aren't really informed about this.

>> I do know that this means lots of testers and testing - which I do
>> appreciate.

Uli> It especially means compatibility testing which hasn't been done by
Uli> anybody here so far in large scale.  Just assume there wouldn't be
Uli> this wide-scale call for testing and 2.2 comes out without binary
Uli> compatibility with 2.2.  It's better to do it this way.  I would wish
Uli> everybody would do these technology previews (similar to our Rawhide
Uli> stuff) and even test releases.  If you take a look at the announcement
Uli> you'll see that it contains a warning similar to the one we have in
Uli> configure.  It discourages people from using it on production
But that's about the distribution itself - and not that the
distribution uses alpha software.
Uli> machines.  I think this is as clear as it can get.

Thanks for your comments,
Andreas

P.S. RedHat is not the only one doing it - Caldera might also do this
(but the article was withdrawn soon afterwards, I don't know if
they're really up to this now).
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj