I’ll try to summarize interesting notes from several talks but for me the Modularity related talks - Langdon’s Modularity, Ralph’s Future of tooling and the Modularity workshop were the most important ones. Most sessions were recorded and they should soon pop up on Fedora youtube channel ...
First talk that I’ve attended was Christian’s overview of plans for Fedora Workstation. The focus was on flatpak aka xdg-apps and the vision of integrating 3rd party repos with flatpak apps. Christian also mentioned work on http://www.fwupd.org/ - a service that helps vendors host firmware updates for Linux systems and client tooling to make updating firmware on Linux automatic and painless. Christian also mentioned that Wayland seems to be gaining lost of stability, the issues with nvidia drivers seems to be resolved and the plan is to have Wayland default in Fedora 25.
Petr Viktorin did great overview of Python work in Fedora. The most interesting part was the status of Python3 in Fedora. Half of the packages have been already ported and these seems to be the package user download and use most.
Containers as we use them today are very developer focused. Dan Walsh talked about why developer environment shouldn’t be replicated into production and what are potential risks of doing so. Dan described why COW filesystems are not the right fit for production and why shared storage is a better solution. Dan showed a demo with Atomic registry and images shared over NFS. In such environment docker pull is no longer needed and it doesn’t need internal network bandwith necessary to update images on all nodes of production environment. Dan also mentioned the ongoing discussion between systemd folks and docker upstream. Docker still doesn’t have any priorities and concept of run time dependencies between containers. Also he introduced ocid, tool based on runc that ensures containers are immutable, use shared storage etc. I hope to see more from Dan about this tool.
A great talk that I recommend to take a look at was done by Sinny Kummari and Dodji Seketeli. Ensuring that there’s no ABI breakage between two different version of libraries is fairly important for application developers and Dodji talked about generic concept of ABI and why it is not that easy to detect ABI breakage or make decisions based on ABI changes. Certain symbols and structures can be internal only, can be used in the way that any change doesn’t matter etc The library that he has been working with Sinny on - libabigail - and the set of tools around it provide analysis of C and C++ applications. In Fedora rpmgrill integrates libabigail and informs maintainers of whitelisted components about abi changes. There are still limitations to be solved and memory constraints that doesn’t allow to review bigger components.
Visiting the city in the evening with tour guide was great and I was surprised how nice and friendly Krakow is!