So, another pretty quiet week overall. There was a soname bump in libpng, but it was untagged and never landed in a rawhide compose. The maintainer is going to create a compat package for the old version first, then update the main package so things can slowly move to the new version over time instead of breaking things over many days. I’m glad people are pushing back on these kinds of things to keep rawhide more generally useful. There are of course still unannounced soname updates that do break things, but I get the impression we are slowly getting better on this.
Another related nice change that landed this week is that rpm by default is only going to look for dependencies in .so files that also start in ‘lib’ or ‘ld’ and not just anything ending in .so. So far I’ve not seen any breakage due to this change and it should nicely reduce the number of deps carried in the collection.
Finally, something to note about rawhide testing that I am not sure I have any kind of answer for, but I think is worth exploring: When running rawhide it’s generally a good idea to reboot frequently. New kernels arrive almost daily, and it’s good to check that things all still boot normally (this exercizes at least: dracut, kernel, grub2 or other bootloader, systemd, other stuff that starts only on boot, etc). However, this also means that there’s likely not very many people running rawhide that test longer uptimes. This would exersize things like: updates of long running processes, systemd, memory leaks, logrotation, and other things that collect up and don’t run at boot). I’ve hit several odd systemd issues on upgrade, and I’m sure there’s more out there. Perhaps we should just say that this is what branched is better suited for (as kernels and reasons for update are less frequent there), or perhaps there’s some artificial way to test this, or a vm or the like could be left running to do so. Just something to consider.