[rb-general] Report from the Summit
Bernhard M. Wiedemann
bernhardout2 at lsmod.de
Tue Dec 20 13:22:48 CET 2016
On 2016-12-17 21:31, Eric Myhre wrote:
> During the "defining reproducible builds" sessions, even though there's
> broad agreement on how to define reproducible *outputs*, this factoid
> was a useful example of how there's some room for discussion and very
> practical impacts of defining what constitutes reproducible *inputs*.
in general, there are a lot of implicit inputs, such as
locale+timezone environment variables (and files?)
compile host filesystem, mount options (e.g. noatime, xattrs)
compile host kernel version (influences fs, scheduler etc)
compile host CPU type (e.g. with SSE4, 8-cores)
In a perfectly reproducible software (e.g. hello-world.c)
none of those matter.
Some of them are easy to normalize using build tools that setup a clean
build environment and some aren't.
(e.g. I tried libfaketime but it did not help much)
More information about the rb-general