Bug#884795: Bug#884795: patman: FTBFS on x32: final link failed: Nonrepresentable section

Andreas Tille tille at debian.org
Thu Dec 21 08:04:47 UTC 2017


Hi for the reproducible builds and QA team,

this discussion was triggered by bug #866137 which was causes by the
fact that an arch:all package had different content depending whether it
was build together with the arch any packages of htslib or
independently.  IMHO the verification that arch all packages are identical
no matter what build method was used.

On Thu, Dec 21, 2017 at 08:43:37AM +0200, Graham Inggs wrote:
> > Hmmmm, I see.  Its a bit dangarous when arch all packages end up
> > differently when beeing build separately from those who are build only
> > for arch all.  I had enother example with this difference and it always
> > leads to problems.  May be there should be a general test on the Debian
> > package pool spotting those differences ...
> 
> We would probably be to told we need test this ourselves by doing our
> own arch-indep only builds, because that's the way it is done on the
> buildds.

Well, there are QA tests whether packages build at all and build
reproducibly.  Isn't it a sensible test to check whether arch-indep
packages are the same?  While I think in a *prefect* world maintainers
should check this but as this example shows maintainers don't (and I
need to admit even now I'm not really motivated to do this work
manually.)

> Interestingly, Ubuntu build the arch:amd64 and arch:all packages together.

But there is no real right or wrong here.  The right thing would be to
implement automatic verification that under each build method the result
will be the same.

------ end of general QA discussion here ------
 
> I noticed htslib 1.6-3 failed its autopkgtests again [1] :-/
> However, it passed the missing cram headers and now fails because of
> missing win/rand.c, a new file introduced in 1.6 [2].
> I will take care of this today.

Our mails seem to have crossed.  I noticed this as well and reopened
the bug.  Thanks a lot that you will care.

Kind regards

        Andreas.

> [1] https://ci.debian.net/packages/h/htslib/
> [2] https://anonscm.debian.org/cgit/debian-med/htslib.git/diff/win/rand.c?id=0b1efffb5e9c447681992014800392e378e67a5f

-- 
http://fam-tille.de



More information about the Reproducible-builds mailing list