Re: Bug#324550: Needs dependency fix [linux-image-2.6.12-1-686: install fails with "cannot stat `(0xffffe000)': No such file or directory"]
At Wed, 24 Aug 2005 12:46:34 +0900,
> On Tue, Aug 23, 2005 at 08:17:58AM -0400, Theodore Ts'o wrote:
> > On Tue, Aug 23, 2005 at 07:49:23PM +0900, Simon Horman [Horms] wrote:
> > > long time no see. It seems that the problem is indeed fixed if you get
> > > the sarge (or later) versions of e2fsprogs and glibc. However, some
> > > people don't have that, and its causing some breakage for those people.
> > > Would it be possible for you to add the conflicts that Ted suggested to
> > > the next glibc release. This would seem like a nice way to make the
> > > problem go away for everyone.
> > Hmm, OK. This is how I understand the problem.
> > If you are using the sarge versions of e2fsprogs (1.37-2sarge1) and
> > libc6 (2.3.2), you're fine.
> > If you are using the latest unstable versions of e2fsprogs (1.38-1 or
> > the just uploaded 1.38-2) and glibc (2.3.5) you're also fine.
> > The problem comes if you are using the sarge (1.37-2sarge1) version of
> > e2fsprogs (or any version of e2fsprogs before 1.37-5) and an unstable
> > version of glibc which is newer than 2.3.4, due to the change in what
> > ldd outputs (the linux-gate.so entry).
> > Since we can't retroactively fix e2fsprogs (although I suppose I am
> > trying to get an updated 1.37-2sarge2 into the next stable update, I
> > could try to convince the release managers to let me get an additional
> > conflicts added, or to get the linux-gate.so filtering added to
> > -2sarge2), the only way to fix this is to add the conflicts line to
> > libc6.
Thanks to Simon and Ted with the detailed explanation, I understand
what the problem is.
> > On the other hand, do we have to support these kinds of wierd
> > cross-release dependencies? I have in the past updated to an unstable
> > version of libc on a stable system, for various sordid reasons, but I
> > always considered it something hazardous that might break things. I
> > don't know that supporting a mix-and-match between stable and unstable
> > is something we have to do, and if it means adding extra hair into
> > libc6's dependencies that in practice may not get removed for a long
> > time, is it worth doing?
> I'm not sure that kind of mixing and matching is really
> supported, in the sense that if a new version exists, the
> recommended solution is always to upgrade.
> I think your idea is worthwile, as people do mix and match,
> but I'll also understand if Goto-san doesn't wan the
> extra cruft in control - it will become irrelevant over time.
> I'm going to reassign this bug to libc6, Goto-san can close
> it from there in whatever way he sees fit.
Thinking about this issue, I agree with Ted's opinion - we don't
probably need to consider about cross-release. However, actually this
problem was caused by glibc, and such change made old e2fsprogs
unusable. It's a bit hard to decide, but in this case I choose to add
"Conflicts: e2fsprogs (<= 1.37-2sarge1)". Ted, is this version <=
1.37-2sarge1 correct value?