[buildd-tools-devel] Some Debian package upgrades are corrupting rsync "quick check" backups

Aurelien Jarno aurelien at aurel32.net
Mon Aug 28 10:42:53 UTC 2017


On 2017-08-28 12:33, Aurelien Jarno wrote:
> On 2017-08-28 18:06, Adam Warner wrote:
> > On Sat, 2017-05-13 at 22:48 +0200, Aurelien Jarno wrote:
> > > On 2017-05-13 21:34, Aurelien Jarno wrote:
> > > > On 2017-05-13 17:52, Mattia Rizzolo wrote:
> > > > > On Sat, May 13, 2017 at 03:44:57PM +0100, Chris Lamb wrote:
> > > > > >  a) Has anything changed in the meantime?
> > > > > 
> > > > > Yes: sbuild stopped repeating the changelog time taking it from
> > > > > the last
> > > > > entry, and will instead generate a new timestamp based on the
> > > > > current
> > > > > time:
> > > > > 
> > > > >   * For binNMUs, instead of copying the timestamp of the last
> > > > > changelog entry,
> > > > >     generate a new one (closes: #843773)
> > > > > 
> > > > > In version 0.73.0-1.
> > > > 
> > > > And I am glad that after all that months with people talking about
> > > > the
> > > > issue, I finally got a detailed description of the issue and a
> > > > pointer
> > > > to the commit to backport. I'll work on that.
> > > 
> > > The above change should now be deployed on most jessie based buildds,
> > > it's only missing on the buildds that are currently down.
> > 
> > Original thread author here reporting to beware that some rsync data
> > corruption can still become apparent after all this time.
> > 
> > It's been a while since I did a full rsync checksum test. Decided to do
> > one after a recent upgrade that includes clang 3.9 related files. These
> > are Debian systems that default to unstable BUT include all debian apt
> > sources including experimental/unstable/testing/stable/oldstable.
> > 
> > I found these four corrupted files in my rsync backups:
> > 
> > var/lib/dpkg/info/clang-3.9.md5sums
> > var/lib/dpkg/info/libclang-common-3.9-dev.md5sums
> > var/lib/dpkg/info/libclang1-3.9:amd64.md5sums
> > var/lib/dpkg/info/libllvm3.9:amd64.md5sums
> > 
> > The latest packages were installed from this repository:
> > 
> > Get:17 https://cdn-aws.deb.debian.org/debian unstable/main amd64 clang-3.9 amd64 1:3.9.1-11 [37.3 MB]
> > Get:18 https://cdn-aws.deb.debian.org/debian unstable/main amd64 libclang-common-3.9-dev amd64 1:3.9.1-11 [2,587 kB]
> > Get:19 https://cdn-aws.deb.debian.org/debian unstable/main amd64 libllvm3.9 amd64 1:3.9.1-11 [11.4 MB]
> > Get:20 https://cdn-aws.deb.debian.org/debian unstable/main amd64 libclang1-3.9 amd64 1:3.9.1-11 [5,896 kB]                                                                  
> 
> These files haven't been built on a build daemon, but instead have
> been uploaded by the maintainer [1]. This is therefore not a buildd
> issue, the issue has been fixed there already with the upgrade to
> stretch.

More precisely the two latest changelog entries have the same date:

| llvm-toolchain-3.9 (1:3.9.1-11) unstable; urgency=medium
| 
|   [ Sylvestre Ledru ]
|   * Remove the --no-discard-stderr option from help2man calls
|   * Also add a missing include in ftfbs-gcc.diff to fix a ftbfs
|     with gcc 7
|   * clang was producing unusable binaries on armv5tel (Closes #873304)
|     Thanks to Adrian Bunk for the patch
|   * Disable -gsplit-dwarf when using gcc 7 for causing a linking issue
|     See https://bugs.llvm.org/show_bug.cgi?id=34140 (Closes: #853524)
| 
|   [ Gianfranco Costamagna, John Paul Adrian Glaubitz ]
|   * Add powerpcspe to latomic archs 
| 
|   [ Katsuhiko Nishimra ]
|   * Ensure /usr/bin/g++-$(GCC_VERSION) exists (Closes: #871591)
| 
|  -- Sylvestre Ledru <sylvestre at debian.org>  Sun, 18 Jun 2017 19:12:15 +0200
| 
| llvm-toolchain-3.9 (1:3.9.1-10) unstable; urgency=medium
| 
|   * Now that strech has been released, upload in unstable!
|     This is necessary for rust in unstable
|   * Try to fix some PATH_MAX on hurd
|   * Enable the verbose mode when trying to build libfuzzer
|     to detect potential issues in the path search
| 
|  -- Sylvestre Ledru <sylvestre at debian.org>  Sun, 18 Jun 2017 19:12:15 +0200

That's the reason why the files ended-up with the same date but
different content.

Aurelien

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien at aurel32.net                 http://www.aurel32.net
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <http://lists.alioth.debian.org/pipermail/reproducible-builds/attachments/20170828/0fc21271/attachment.sig>


More information about the Reproducible-builds mailing list