Kernel stability underestimated due to RC entries in changelog

[HastD]

Looking at the changelog for Fedora's kernel RPM package, we see many entries for .rc* tags, which are for kernel release candidates. On non-Rawhide versions of Fedora, I don't think these ever correspond to stable releases. However, they're still present in the changelogs, which means they count against the package in chunkah's stability calculation.

I'm not sure how difficult it would be to address this in chunkah, but would it be possible to somehow filter these release tags out from the stability calculation and only count changelog entries that correspond to actual releases? Of course, it would probably have to be configurable, since those entries should be counted if the image is based on Fedora Rawhide. There may also be other packages with a similar issue; I'm not aware of any other than the kernel, but I haven't looked very hard.

[jlebon]

One thing to note is that this isn't specific to the kernel. While Rawhide in general churns more, all packages carry their Rawhide changelog items into stable releases, which hopefully means stability is roughly uniformly impacted for all of them (which in turn dampens the effect).

But yes, some packages in Rawhide churn more than others (esp. those hooked up to some CI setup that build on push). If there's an elegant way to try to taper down this bias in a generic way, I'm open to just having that live in the code. But overall I'm unsure how much squeeze there is to be had there in terms of final effect on packing performance. I think also exponential decay weighing would help here too.