2014-08-20 13:30 GMT+01:00 slowfranklin <[email protected]>: > It seems there wasn't any other technical argument for that other then > "that's what Debian does". > > So short of technical arguments I found it interesting to get a picture what > others might find an appropriate amount of time. My gut feeling was 2 weeks > it too short.
There is probably a tradeoff involved. if the delay is too long, users of the testing catalog (probably most of our users) will get an unnecessary delay in getting the benefit of new packages. This also applies to security fixes, unless they are pushed directly to testing by the maintainer. This is what Yann did with OpenSSL previously, but not now: recently updated OpenSSL packages still have 3 days to go. If the delay is too short, users won't have enough time to spot a problem with a new package. But we don't know how short is too short. The problem is we don't have enough information to evaluate that tradeoff. If nobody uses unstable, the delay value doesn't matter. Ideally users would run unstable on some of their machines, and testing or stable on the rest. But we don't know if they do or not. In general, if you want to optimize the delay, I'd suggest getting more information rather than relying on the gut feeling. Maybe educating users about the relation between unstable and testing is more important than any particular value of the delay? Maciej
