Benjamin Mesing wrote:
However I have often heard complaints about broken dependencies and broken software in testing. From what I have heard, I would not like to go with testing for my system.
How on Earth would that be allowed into testing? I can imagine Serious bugs slipping trough (because most are reported as Normal, after all), but broken dependencies?
I think one of the reason why testing is doing bad compared to unstable is, because serious bugs usually get fixed in unstable pretty fast, however due to dependency problems the fixes often take a long time to propagate to testing. Nevertheless I think the stable/testing/unstable framework is a goodchoice because it offers a good way for preparing stable releases.
Buggy packages should not be allowed into Testing in the first place. And, as far as I know, they are not. If, however, a considerable amount of buggy packages do get into Testing, I would suggest reviewing what is considered buggy. Maybe not only Serious bugs, but also something like size-of-package/number-of-bugs quotient should be taken into account.
Another level between Stable and Testing could also be introduced. Something like beta-pre-releases, a more or less tested snapshots of Testing. Or make Testing public only after it is considered stable enough, like halfway to release date, and let people who really want to do the testing run Unstable.
There is infinite number of ways Debian could go, but I really have no facts to support the need for any of this.