There are several goals that Modular approach is trying to achieve. First, a module is bundle that is managed as a single unit. The concept even though right now it is using solely RPMs can eventually be used with different types of content. The module as external API that doesn’t change within minor updates. And finally several modules can form a module stack where dependencies are defined by the API and not by RPMs dependencies as we use them today.
Langdon showed couple demos. First replacing existing kernel, than installing different apache modules and example of module stack with different version of PHP. Also described a option to define profiles for different modules with postgresql database. All has been done using DNF plugin, a build pipeline which can already compose modules and create repositories.
It is still a very new approach with several rough edges. The discussion about Modularity continued even after the presentation and Langdon invited people to join the Modularity Working Group tuesday IRC meeting as well as to review existing documents on the Modularity wiki.
Ralph did a good follow up after lunch for Langdon’s talk describing Factory 2.0. First he showed a long list of thing that Factory 2.0 is not! Truth is that our tooling needs major overhaul but it will never happen all together but rather as iterative steps. Also the list showing what Factory 2.0 is not showed that there are several issues with the infrastructure and build pipeline where people are asking for better solution.
I’ve also joined Patrick Uiterwijk’s presentation about Fedora Atomic as Workstation. Patrick build his own ostree and starting experimenting with the right content for that tree. The aim was to provide basic workstation configured to easily use docker, flatpak apps and other tools that Patrick needs. Patrick was hitting some issues with latest ostree in rawhide but other than that his tree keeps regenerating and he’s been using it for some time. Also Patrick mentioned ongoing issue with garbage collections which forces him to manually clean local package cache every now and then.
Last talk of that day that I’ve attended was Jan Blazek and OpenQA. The talk was focused on testing anaconda using OpenQA screenshoting feature and comparison of displaying elements. To be honest I was slightly disappointed by the talk. OpenQA is nice framework that Fedora QA has been using for porting of their work but it felt like they also have to maintain yet another framework, they can’t use or share tests across distributions and it is not used as a gating mechanism whether certain update can hit bodhi or not. I would love to see a common testing approach where developers can become test contributors and where Fedora QA focuses on tooling and workflow instead of writing tests by themselves.
Great day finished with a boat party on river Visla and nice walk thru the city center around midnight :-)