On Mon, Oct 12, 2015 at 11:11:19AM +0300, Niko Tyni wrote:
> On Sun, Oct 11, 2015 at 07:50:49PM +0200, Holger Levsen wrote:
> > On Sonntag, 11. Oktober 2015, Niko Tyni wrote:
> > > It's indeed an unfortunate interaction of dh_installdocs and disorderfs
> > > when installing directory hierarchies.
> > we don't use disorderfs since weeks:
> Ah, I'm pretty sure it was still on the list of variations on the
> reproducible.d.n front page yesterday when I checked that. So I thought
> the issues had been fixed. Never mind & thanks for fixing the list.
> In that case, I think something else has introduced file ordering
> variation between build1 and build2 lately, highlighting this issue
> (and others in for instance perl and libapache2-mod-perl2).
> This is a good thing, of course.
> I still think the directory timestamp issue happens because dh_installdocs
> uses find + cp --parents -p, which generates different results depending
> on the order of the find results.
Indeed. Just replace "disorderfs" by "different filesystem ordering"
and we still have a likely reason for the differences.
Not using disorderfs does not automatically make filesystem ordering
to be deterministic.
Reproducible-builds mailing list