Frank Ch. Eigler (fche@xxxxxxxxxx) said: > > Clearly, bohdi/bugzilla/pk interaction is not good enough to collect > > the kind of feedback needed for the karma system to work. And bohdi > > should get smarter about identifying packages that need this > > feedback. Critical path is a good first approximation but what would > > really help is some heuristic about how much breakage a bad package > > can cause : how many other packages depend on it (dependency > > metrics), how long is has lived (has it been in Fedora for years of > > imported the week before), was it even in the default install for > > some people, etc. [...] > > +1. Extra tooling should help here. > > Consider also the possibility of adding in download statistics into > the 'slow-down-cowboy' heuristics: use records that particular > update-testing RPMs have been downloaded as evidence that they are OK. Unfortunately, this is one space where our awesome distributed mirror system hurts us. For any users using the default configurations for repositories (including updates-testing), we only have the requests for the repository itself - the file requests are all on various mirrors' logs, which we don't have access to. Bill -- devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxxx https://admin.fedoraproject.org/mailman/listinfo/devel