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

Re: recommended way to fix Bug#492814 ?



On Tue, Jul 29, 2008 at 02:29:18PM -0700, Ivan Kohler wrote:
> > Ok; apparently I misread the control file.

> >   Build-Depends: [...] libxcrypt-dev (>= 2.4), libxcrypt-dev (< 3.0)

> > Apparently, my question was meant to be: why was a package uploaded to
> > unstable that build-depends on a version of libxcrypt that's only in
> > testing (since libxcrypt-dev 3.0-2 was uploaded over a month before your
> > libpam-unix2 upload)?

> Because that's the version it needs to be functional.  libpam-unix2 2.4 
> needs libxcrypt before 3.0.  It doesn't work with libxcrypt 3.0.

> >  I'm not seeing how you expected this to work, so I'm
> > not sure what to recommend as a way out of this mess.

> I expected (wrongly, obviously) it would pick up the build dependency 
> and build against the correct packages...

Packages uploaded to unstable have to be buildable in unstable.

> > > To fix the libxcrypt build problem in unstable, I'll need to revert to 
> > > the previous version.

> > Why can't you *fix* the bug?  It's a trivial build failure to fix - just
> > stop building with -Werror.

...

> > > 3.0 currently in unstable is unsuitable for release.
> > 
> > For reasons other than the build failure?

> Yes.  libpam-unix2 needs the previous version to build and work 
> correctly.

> > That is certainly not evident from the BTS. 

> http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=487487

There's nothing in bug #487487 that explains why libpam-unix2 can't use the
newer version of libxcrypt.  I had to figure this out for myself by
downloading the libpam-unix2 2.4.1 package, to see that it uses a function
from libxcrypt1 that's no longer present in libxcrypt2.

> > If the build failure is the only reason, I don't see why you
> > wouldn't just fix it.

> The libxcrypt build failure is not really relevant.  It could have built 
> fine and there would still be the same exact problem.

Which problem?  The problems I can see are:

- libpam-unix2 2.5.0 was broken, because it was built against the old
  version of libxcrypt which it's not compatible with
- libxcrypt is failing to build
- libpam-unix2 1:2.4.1-1 is failing to build because it has a versioned
  build-dependency on a previous version of libxcrypt

>From what I see, fixing libxcrypt so that it builds would have also
addressed the first problem because libpam-unix2 2.5.0 could have been
rebuilt against the new libxcrypt; but now you've uploded an older version
of libpam-unix2 which introduces a new incompatibility, so you would have to
both fix libxcrypt, and upload libpam-unix 2.5.0 again.

I still think that's the best option, unless there's some *other*
undocumented reason why libxcrypt2 is unsuitable for release.  I think it's
the best option because it avoids epoching a library, and because it would
continue to use the libpam-unix2 2.5.0 package which, aside from not being
able to do password changing at present, appears to have at least received
some testing.

> > > How do you suggest I revert to libxcrypt 2.4 to fix the build problems 
> > > in unstable *without* using a library epoch?

> > I don't.

> Okay, so you are confirming that the library epoch, distasteful though 
> it may be, is still the best/correct way to fix the problem.  I will 
> prepare and upload corrected packages as per my original suggestion:

> - libxcrypt version 1:2.4-2 (no changes from 2.4-2 currently in 
>   testing)
> - libpam-unix2 1:2.4.1-2 version of libpam-unix2 (no changes except to 
>   update the build-dep)

The problem with this is that it brings in a "new" upstream version of
libpam-unix2, 2.4.1, which has before now not been in unstable or testing.
That should not normally qualify for a freeze exception, AFAICS; I'll let
someone else on the release team make the decision whether it should get
that freeze exception, but I believe the alternate approach I outlined above
allows for better QA by comparison.

-- 
Steve Langasek                   Give me a lever long enough and a Free OS
Debian Developer                   to set it on, and I can move the world.
Ubuntu Developer                                    http://www.debian.org/
slangasek@ubuntu.com                                     vorlon@debian.org


Reply to: