Over the past half year, we've made two releases of the DebianInstaller that included full CD sets of the whole Debian testing archive. These d-i releases had minor problems but have on the whole been quite successful and have been used by many Debian users, to the point that most user queries about installing stable on the debian-user list are now met with suggestions to install sarge with the DebianInstaller instead.
Given this success, the proposal is to just try to keep testing in a releasable state at all times, and cut releases from it using the same techniques whenever a major new version of the installer, or of some other major subsystem, like Gnome, is ready. Such a release would not necessarily have to be "the" stable Debian release, or we could only bless occasional releases that worked really well as "stable" while keeping the other releases as testing releases. But these releases would be promoted and supported as full Debian releases, and would be forked in the archive as a release that users could install instead of installing the ever-changing testing.
Pros:
Seems to work in practice, see above tooting of my DebianInstaller horn.
- Release frequency drops to quarterly or lower.
Forking the debian repository at the release would solve one problem that has plagued the DebianInstaller releases, that of changes in testing after the release breaking some installation methods.
Cons:
- This proposal probably entails lowering our expectations about what consistutes a Debian release. These new releases would have more RC bugs.
- There would need to be security support for testing for this to work. The nasceant testing security team hopes to provide this, but it's untried and will be a lot of work.
This works better when testing is in a close to releasable state. Currently we've had a rather poor track record of keeping testing in such a state when we're not in the run-up to a major release. It can be argued that the existing DebianInstaller releases have worked as well as they have because most of them have been made when testing was in a near-releasable state.
Opinions:
This is a slightly undeveloped and perhaps strange proposal, but if we don't do anything else, we may end up with a lesser version of this by default. As I said, our user community is already using and recommending these releases to others to a certain extent, and if releases drag on for years, they will more and more. -- JoeyHess
Proposed by: JoeyHess
seconded by: DanConnolly
See ReleaseProposals for alternatives.