Re: Wrong reproducibility reported for libdevel-cover-perl? (i.e. I suspect a false negative)
On Thu, Apr 21, 2016 at 11:30:19PM +0200, Axel Beckert wrote:
> today I was debugging strange autopkgtest failures of
> libdevel-cover-perl inside (!) adt-run (locally, not
> yet on ci.debian.net):
> t/e2e/change.t ..............
> Can't mkdir /home/abe/pkg-perl/libdevel-cover-perl/t/e2e/cover_db_complex_change/: No such file or directory at /usr/lib/x86_64-linux-gnu/perl5/5.22/Devel/Cover/Test.pm line 88.
> Dubious, test returned 2 (wstat 512, 0x200)
> But that path there was just my local working copy where I built the
> package to test it. The adt-run directory was at
> /tmp/adt-run.bV4vwb/adttmp/smokeLkOAX4/. How the fuck can my path pop
> up in there?
> Well, libdevel-cover-perl puts the full build path into
> /usr/lib/*/perl5/5*/Devel/Cover/Inc.pm and installs that file
> our $Base = '/build/libdevel-cover-perl-ht7Bko/libdevel-cover-perl-1.21';
> (This code example is taken from 1.21-1+b1 as currently in Unstable
> and Testing.)
> So I was sure that this package can't be built reproducible. But the
> sun on
> still shines.
> So either there is an according bug somewhere in diffoscope (which I
> doubt :-) for a very long time or we should also make the second build
> differ in the build directory name to catch such issues.
> In case the latter is already done, I'm out of ideas why that package
> counts as reproducibly buildable.
The build path is part of the specification for the overall environment,
if I'm not mistaken - so the build is allowed to vary with changes to it.
I suppose embedding build paths is too widespread to make eliminating
that feasible, and making the build path the same is easy enough?