[rb-general] reproducible anything

[Richard Purdie]

On Mon, 2018-10-08 at 17:27 +0200, Bernhard M. Wiedemann wrote:
> > And I want reproducible Debian *much* earlier, so I advocate for
> > doing
> > rebuilds of buster in those same pathes... I know this is "wrong",
> > packages should be reproducible in any path, but I want
> > reproducible
> > Debian *before* that is the case.
> 
> I'm completely with you there. I'm already using a highly normalized
> build-env for openSUSE tests (including constant build user, path,
> umask, timezone etc)
> to solve the other, more pressing issues first (i.e. indeterminism
> from timestamps, readdir, hashes, races, random, CPU-detection, ASLR,
> PGO).

FWIW in the Yocto Project we now have reproducible builds for arbitrary
paths and build users. This is only tested with our core-image-sato
level which is an X based graphical image but means our core system
components are reproducible and means we have a solid base further
improvements can be made from.

We did run into an issue recently where every architecture except IA
had debug symbols with build path leakage. This was coming from glibc
and we ended up having to patch glibc to work around the issue:

http://git.yoctoproject.org/cgit.cgi/poky/commit/?id=cd158dd197a3a41a91679107a94c6f213476921c

Obviously work like this needs to be discussed with upstream but is of
benefit to anyone else aiming for build path independence.

Cheers,

Richard