the release problem

In trying to manage and release d-i, I time and time again run into problems that we have in overall release management in Debian, in miniature. While we've avoided whole classes of problems by being a smaller group, and by having different takes on who can do what with packages (which has also brought with it its own unique problems), the similarities are often striking, and none have been as obvious as in the last week.

Last weekend we decided to try for a release at the end of the month, and to have a string freeze this week. If that was a real goal, the sane thing to do would be to stop making changes except when absolutely necessary, to talk to others before making changes, to avoid committing lots of unpolished strings on the day the string freeze began, and so on. But instead we've gotten a lot of commits of new code and new strings. Some people managed to do this without breaking anything or causing any trouble; most people didn't. We're now already behind schedule. I was happy with every single day's build all last week; so far this week today's build is the first one that seemed anywhere close to releasable.

Of course this is all just human nature, and part of me is glad to see an uptick in work on d-i, even if we had to set a deadline to kickstart it. Many of the changes that are being made are indeed important, and were being put off. And it can't help that it's been more than 2 months since the last release, so we have increasing pressure to make a release, and increasing pressure to get every conceivable improvement into this one.

When this happens in Debian proper, we find it very hard to not let the schedule slip again and again, as new versions of gnome and kde, new installers, and so on are put into the distribution, and as we work to find and stamp out all the resulting bugs. Our users have very specific expectations about Debian releases, and while they don't expect speed from us anymore, they do expect a certian quality, they expect as much as possible to be fully up-to-date at release time, they expect it to be supportable and supported for a long period of time. And we can't afford, or haven't the guts to try deflating those expectations for the sake of a quicker release.

I think that with d-i, we have the luxury of taking different approaches. The pressure for a release is there, so we will have a release, and soon. The quality of that release will depend on what people do between now and then, if it's worse than the last one then some lessons may be learned.