On Fri, Aug 21, 2009 at 5:17 AM, Dustin <[email protected]> wrote: > > > On Aug 20, 12:19 pm, Mark Atwood <[email protected]> wrote: > > Also discussed is eliminating the "even/odd" release numbering system. > > I don't have a huge opinion here. I just want it to be obvious > which releases are newer than others (feature-wise). > > > I propose instead that we keep the main tree usable and stable and > > passing all tests at all times (like Drizzle does), and development > > and expermental features be in their own independent branches, and > > they get merged into the main branch as they stablize. > > We work really hard to ensure that master is always well-tested > code. We don't push code there that doesn't build on all of our first- > class supported builders (having some problems with OpenBSD, but that > is hopefully to be first-class soon). >
To be honest I only care a little bit, and neither choice would affect me much, but I do not prefer the even/odd system. I think it promotes larger gaps between releases, and much harder to get quick patches in. My preferred method when working on projects like this... is to look at the todo list... pick the top number of items that would take about 2 weeks to do, and assign those to the next potential release... When those items are done (even if it takes more than the estimated 2 weeks), release them and any other (critical or not) patches that happened to come in during that time (and were tested). Test the release as a whole, and release it. Rinse and repeat. I'm hardly active in the memcached development though, so my opinion has very little value. :) PS: I am a Release Manager in real life. :) -- "Be excellent to each other"
