--+QahgC5+KEYLbs62 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Sat, Nov 05, 2011 at 03:11:55PM +0900, Yuki Sonoda (Yugui) wrote: > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > Hi, > > == Background > I have been maintaining ruby_1_9_1 and ruby_1_9_2 branch. > Ruby 1.9 is now stable as Ruby 1.8 was, but it has not yet reached to > the bugless nirvana. So Ruby 1.9 needs continuous maintenance. > > But actually ruby_1_9_1 is no longer well-maintained. My backports to > ruby_1_9_? tend to be late. There is a bottle-neck. > > The bottle-neck is review process for commits in trunk. > A few bug is branch-specific. Most of bugs in Ruby 1.9.x reproduce > with trunk too. So I rarely need to write a branch specific patch. > Just backporting from trunk fixes most of bugs in Ruby 1.9.x. > But I need to read a commit carefully and run unit tests before > backport the commit. This review process takes really really long. > That's why patch level releases have been late. > > == Proposal > Let's parallelize the bottle-neck. > Review and tests are necessary for stability and compatibility of > released branches but these processes can be parallelized. > I propose the following process: > > * A committer who fixed a bug in trunk should also check if the bug > reproduces with other active branches. If reproduces, (s)he should > create a backport request on the Redmine. > * Or anyone who want us to backport a commit in trunk can create a > backport request. > * Another committer review the request. This reviewer checks if this > commit is good enough and backport it to the older branch. I think this is good. If a committer requests to backport something, and another committer approves the backport request, shouldn't the original requester actually perform the commit? > * Any committer who thinks the backport breaks compatiblity can revert > it. > * Eventually the maintainer of the branch decides to revert or not. > > CI and a new Redmine plugin can help this process. > * Automatic request triggered by commit message? > > And the branch maintainer can have time to plan the next patch-level > release. I think this is very good. I don't like overwhelming you with backport requests. It seems like too much work for one person, especially when you take stdlib in to account. -- Aaron Patterson http://tenderlovemaking.com/ --+QahgC5+KEYLbs62 Content-Type: application/pgp-signature -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.11 (Darwin) iQEcBAEBAgAGBQJOt1DlAAoJEJUxcLy0/6/GsngH/jwkMq9wF0cbuh9HwbW6DFl9 ALKqgaqPvlsBV71+NEzndtsp00+gWgMpqjsdgx7/rdts3HXdfcLIUOPeQq74WU/Q sYxAghars/xasSemIpl91sqYlhrSh/EFhmFTR5Ac3IX3UlH0vfyNPjM4++Dx7g6E iPIMtsYlO7xNOZtxDsvZRl1je/fbz3cBrvt2eaXgglN6RYqMpC6PO61XpaAMVj7o xHwK9tHVj3lTtzhywc+nPt6A/6AXWtz/1IzCqq4cykDDut8NKEmY6EjRTDY9aHdS pkX/B/5wJc/BL292jrXC+xRskoFddEYEQSQUzT/eJmVaRcRQO6X0NptGjaGoGX4 tM -----END PGP SIGNATURE----- --+QahgC5+KEYLbs62--