Proposed Process Changes For X Server 1.8
Peter Hutterer, who has been working on the X.Org input code for some time and is the developer behind MPX, late into the X.Org 7.5 release cycle decided to step up to the plate and personally get this important X.Org update out the door. Peter is not stopping after X Server 1.7, but has already made a proposal regarding the release and development changes going forward for X Server 1.8 and into the future.
Peter acknowledges the current key problems with the X.Org release schedule being unpredictable, Git master is unusable at times, and there is a late testing cycle. Mr. Hutterer is proposing three fundamental changes and they include the use of feature branches, three stages to the development cycle, and predictable time-based releases. With regard to feature branches, Peter proposes that all work that is large and potentially disruptive to the X Server stability should first be done in a branch of the X Server rather than in Git master. Once the branch is developed, tested, and fairly sane it could then be pushed along into the master branch. Peter cites his Multi-Pointer X development work last year as a good example of a feature branch working out well.
The three stages of the proposed development cycle for X.Org would be feature merge, bug-fix, and release freeze. The merge window for the X Server would be open for about three months at a time followed by two months of bug-fixing and then a one month period for the X Server to be frozen and well-tested prior to a release. This would allow consistent six-month releases of the X Server / X.Org. Albeit with different time lengths, this would put the X.Org development process much closer to that of the Linux kernel. Feature branches ready for the next release would be merged into Git master during that time, followed by bug-fixing afterwards, and then at the end it's time to cut the release and then branch the X Server. The X Server would not be branched in advance of a release, but rather afterwards and prior to the next release's merge window opening.
Lastly, Peter calls for predictable time-based releases and not to delay any releases based upon half-complete features. As part of getting X.Org released on time, he also calls for monthly snapshots of Git master for easier testing and then towards the end of the cycle to provide weekly or bi-weekly snapshots until the release candidates are pumped out. Having predictable time-based releases would please quite a lot of people, especially the distribution vendors that have been challenged in the past by deciding whether they can switch to the next X.Org / X Server in their development cycle or whether delays will jeopardize the move and postpone it to their next release.
Peter's mailing list message that outlines this proposal for X Server 1.8+ can be read on xorg-devel. As of right now there are no replies to his thread, but hopefully that will change and a version of this proposal will become policy for X.Org going into the future.