> On 2010-11-02 at 19:10:20 [+0100], Adrien Destugues <adestugu@xxxxxxxxx> > > wrote: > > > Following up on the R1/A3 talks [1], > > > one of the major hurdles seems to be the lack of a release branch > > > manager. > > > > > > Would it be possible to enforce a very strict merging-to-branch > > > policy? > > In theory yes, in practice this might turn out to be more > complicated. E.g. > there are at least a few components (ShowImage, ext2, the Gutenprint > stuff, > Locale Kit/ReadOnlyBootPrompt) for which I couldn't say what state > they are > in ATM. If any of those still requires significant work to get them > into an > acceptable state for the release, we would already have exceptions to > the > policy. ReadOnlyBootPrompt has trouble with line breaking chinese text, but works fine otherwise. This can only be solved properly by using ICU layouting engine for linebreaking the text. I hope I'll have some time to do it, but I can't be sure. > > > > This should greatly reduce the workload of the branch manager. As > > > the > > > branch > > > should be seeing less commits, it should make it easier to perform > > > QA > > > and spot any regressions. > > Yep, generally that's a reasonable approach. > > > I believe the 'alpha' selection should not be made on commits, but > > on > > tickets in trac. If a commit fixes a ticket attached to the alpha > > milestone, it goes in the branch. If not, it stays out. > > Sounds rather impractical. If someone would fix a small issue for > which no > ticket exists, he'd either have to sneak one in or the fix wouldn't > make it > to the release branch. Right. I think it depends on how we handle the branch. Do we need a coordinator and super-strict rules, or would guidelines and letting everyone commit things to the branch work ? I think no one would want to mess with the branch anyway, but individual changeset tracking more or less prevented a lot of people working on the branch. While doing it at ticket assignation stage allows everyone to work on it. The rule would be something like : * If a ticket exists and is assigned to R1a3, go ahead and commit in the branch (or do we want an extra quality review here ?) * If a ticket exists and is assigned to R1, dont commit in the branch * If no ticket exists, dont commit in the branch yet, state it in the commit message and the case will be handled on haiku-commits mailing list. This way it does filter out most stuff, but we still need to take decisions for these ticket-less commits. They can be done on vote, or a branch manager can do it. -- Adrien.