Bug#823465: dpkg: Won't run at all on i586 Pentium MMX due to illegal instruction

2016-05-05 Thread Pierre Ynard
retitle 823465 Dropping i586 support unexpectedly breaks i586 systems on upgrade
stop

That's nice. That piece of information didn't get to me. I didn't read
that gcc changelog either because I haven't upgraded it on my system.

This is my production system and I run a lot of services on it. This
isn't a playful, pointless experiment. I'm not too happy to hear that
Debian is dropping support. That's not the idea I'd like to have about
Debian. If I wanted something that wasn't going to work on this system,
I would have installed  long ago.

So are there release notes somewhere? What's the recommended course of
actions? Am I supposed to go ahead with the upgrade, tail the logs and
eat popcorn while watching them fill with SIGILL?

-- 
Pierre Ynard
"Une âme dans un corps, c'est comme un dessin sur une feuille de papier."



Bug#823465: dpkg: Won't run at all on i586 Pentium MMX due to illegal instruction

2016-05-06 Thread Pierre Ynard
Yes, I run unstable in production. My stuff isn't business-critical.
But I can say the same thing about this making it impossible to run
unstable. That's not the idea I'd like to have about Debian. If I wanted
a distro where unstable is broken and unusable, I would have installed
 long ago.

I don't care too much that the change is "silent", since I'm here and
know now. That's nice to document the next release, but I was asking for
an answer to my problem now. What do I do with my i586 system running
unstable of last week? Drop it into a tub of water, just like i586
support is getting dropped? That's going to cause way more downtime than
simply running unstable in production.

My "complaint" against unstable is valid. Typical Debian bugs don't
(or at least shouldn't) wait for the stable release to get fixed. I
don't see why you retitled this copy of the bug, since it described the
situation accurately. i586 users running unstable are getting their
system broken, with no obvious way to handle it. Maybe I'm the only such
user and you don't care, then at least have the decency to wontfix me.

Please tell me, what do I do with it??

-- 
Pierre Ynard
"Une âme dans un corps, c'est comme un dessin sur une feuille de papier."



Bug#823465: dpkg: Won't run at all on i586 Pentium MMX due to illegal instruction

2016-05-07 Thread Pierre Ynard
> My recommendation would be going to jessie[1], it has whole four years
> of support left. Anything you need from unstable can be backported.

Hopefully the downgrade path would be workable.

> After those four years you can reconsider, in the unlikely case your
> machine will be still alive.

That's harsh. The hardware is almost 20 years old, and it's been running
the same Debian install for 11 years already. I don't see 4 more years
as too unlikely.

> That's four more years than ia64 guys got. Unlike 586's half the speed
> of first-gen RasPi, ia64 machines can be pretty beefy -- new ones even
> are still being manufactured.

I'm aware yes.

> What kind of solution would you propose? We can't exactly add preinst
> guards to every single package. The only package that's depended on by
> (almost) all compiled code is libc6, but because of symbols handling
> the dependency is usually libc6 (>= 2.15) or such rather than (>=
> 2.22-7).

I was thinking more along the lines of adding some central check in dpkg
maybe, that detects the lack of i686 support and errors out on new,
incompatible packages. Discriminating packages could be as simple as a
by-passable check on the build/release date. But then this is a bit late
to implement in advance.

-- 
Pierre Ynard
"Une âme dans un corps, c'est comme un dessin sur une feuille de papier."



Bug#823465: dpkg: Won't run at all on i586 Pentium MMX due to illegal instruction

2016-05-19 Thread Pierre Ynard
I've successfully downgraded most of my system from unstable to jessie,
so this seems to be workable.

Among the things I run, there's only MySQL that can't be downgraded just
like that from unstable's 5.6 to jessie's 5.5, and for which there is
no backport of 5.6; and also libstdc++6 that I can't downgrade from GCC
5 to GCC 4.9 because mysql binaries depend on it. An unsupported MySQL
server is a nice attack vector.

Is that kind of downgrade supposed to be supported? I encountered
configuration migration problems for apt and postfix, shall I file bugs
for these?

I still think this bug should be marked as wontfix because "enjoy 4
years of jessie" is not a real fix to the problem.

Thanks for the guidance,

-- 
Pierre Ynard
"Une âme dans un corps, c'est comme un dessin sur une feuille de papier."