Flock to Fedora 2025

As during every Flock, we'll hold a session with FESCo members to introduce themselves and say a few words about our plans for the next year. The majority of time will be spent on questions from the audience.

The effort to make Fedora package builds fully reproducible is progressing. For Fedora 42 we introduced a tool to resolve common sources of build irreproducibility into the set of postprocessing tools that are executed at the end of every package build. We also reported bugs or opened pull requests to fix individual issues in many packages. Together, this has increased the fraction of reproducible builds to about 90%.

For Fedora 43 we want to start treating irreproducible builds as packaging bugs. We introduced a portal for packagers to view rebuild reports for their packages. Bugzilla tickets are being opened when a build irreproducibility is found. Our rough goal is to have 99% of package builds reproducible.

In this session, we want to describe what is happening and why, what resources are available to packagers, where we the next steps, and answer any questions packagers or users might have.

A few years ago we got rid of ldconfig scriptlets. With sysusers support in rpm, in Fedora 42 & 43 we got rid of scriptlets calling getent/useradd/groupadd. Can we get rid of most other scriplets? One of the most pervasive are the scriptlets for systemd unit enablement. Let's discuss options how to make those unnecessary and simplify packaging. What about the rest? It'd be great to remove the calls to update-alternatives and make scriptlets the rare exception, not the norm.

Another source of complexity in packaging is arch-specific dependencies (%ifarch foo, BuildRequires:bar-devel, %endif). We could replace those by rich dependencies, making srpms arch-independent. This would make querying package dependencies much easier, but also help with reproducible builds. What is the best design for this? Some syntactic sugar in rpm would make this easier to implement.