This is what we have done in ripple (and webworks) master - always stable current shipping code next - always 'stable' next release. Expectation that code has been tested / run before merged into this branch. feature branches - branched off of next and merged into next when stable / done. Not expected to be stable or runnable until merge time.
On Wed, Jan 2, 2013 at 2:32 PM, Filip Maj <f...@adobe.com> wrote: > Am I correct when I say that, with this approach, master becomes a series > of merge commits coming from dev, then ? > > A couple questions to follow up: > > - "features get forked from stable" - forked from master, yes? > - "features, when ready, tested against dev branch" - what does this mean? > Does this mean, you would merge feature branch into dev branch (locally) > then run tests to make sure things work? > > On 1/2/13 11:19 AM, "Joe Bowser" <bows...@gmail.com> wrote: > > >OK, Let's rethink this: > > > >After talking with Brian on the 21st, I think we agree on this: > > > > * Master remains stable and sits at the most recent released code > >(i.e. 2.3.0 once we get 2.3.0 done) (Stable Channel) > > * Dev happens on branches for the releases (i.e. 2.4.0) (Dev Channel) > > * In the case of a point release, dev happens in the branch of the > >major release (i.e. 2.3.1 would happen in the 2.3.0 branch, not > >master) (Testing Channel) > > * Features get forked on stable then once the feature is ready, > >tested against the dev branch. If they work with stable, they SHOULD > >work with 2.4.0. If they don't, the tickets get added to 2.4.0 to > >make it work with that release. That way things are more predictable > >as far as new features are concerned. (You will burn your face > >channel). > > > >Does that make sense? Working on master for things causes us pain and > >we should use git conventions to make it simpler for people who expect > >our master to work all the time. I don't think this will speed up the > >release as much as automating tagging of RCs so that when the JS is > >tagged, everything else is tagged. The week it takes to tag an RC is > >way too long. > > > > > >On Wed, Jan 2, 2013 at 11:08 AM, Filip Maj <f...@adobe.com> wrote: > >> Bumping this thread. I'd like Joe to clarify as well. > >> > >> On 12/20/12 12:26 PM, "Brian LeRoux" <b...@brian.io> wrote: > >> > >>>Ok, I want to understand this, let me take a stab. > >>> > >>>You describe three long-lived branches like this: > >>> > >>>- Master: This is stable and frozen on the last tagged release. > >>>- Dev: the next release to be tagged. Feature branches merged from > >>>master when confident. > >>>- Unstable: the current working branch for a particular tag. Feature > >>>branches merged as needed for collaboration. > >>> > >>>Everyone works from local feature branch rebasing and committing to > >>>master. When that feature branch is considered good enough, it is > >>>merged into dev, and work continues. Whatever date we happen to pick > >>>for a release that is what dev becomes, we tag, and move that sha to > >>>stable if its not an RC. > >>> > >>>? > >>> > >>> > >>> > >>> > >>>On Thu, Dec 20, 2012 at 10:52 AM, Joe Bowser <bows...@gmail.com> wrote: > >>>> I'm OK with this, but I think your example is off: > >>>> > >>>> Where n is the current released piece of the software: > >>>> > >>>> n.x.x = Stable > >>>> n+1.x.x = Dev > >>>> master = Unstable, can have things merged in from feature branches > >>>> > >>>> This fully uncouples features from release planning, which is good > >>>> because it means the release will land in the version when it's ready, > >>>> and not for any other reason. I also propose that we keep using the > >>>> same RC tags and that for a final release we tag it x.x.xFinal. We > >>>> still need to tag an RC and re-tag it. > >>>> > >>>> Release Process: > >>>> 1. Tag the dev tree > >>>> 2. merge the dev tree back into master > >>>> 3. Create 2.5.0 branch > >>>> 4. File issues from 2.5.0 in JIRA > >>>> > >>>> I also propose that we automate the tagging. If an RC is broken, we > >>>> just cut another RC. A lot of our retagging is done to get around the > >>>> pain of having to do another RC. The biggest part of the delay is > >>>> waiting for every single platform maintainer to tag their platform > >>>> after the JS was tagged. For example, I tagged rc2 for the JS and for > >>>> Android on Monday last week from my hotel room, and the release wasn't > >>>> fully tagged until this week. I'm fine with RCs going up to 10 as > >>>> long as we can release early, release often and release when we want > >>>> to and not run out of time and have to delay. > >>>> > >>>> On Thu, Dec 20, 2012 at 10:33 AM, Brian LeRoux <b...@brian.io> wrote: > >>>>> Truth. Though lets not get hung up on the past and just focus on the > >>>>> present. We've done a really good job getting where we are. > >>>>> > >>>>> So, Joe, are you saying you like the idea of three long lived > >>>>>branches > >>>>> and merges happen from local feature branches? > >>>>> > >>>>> > >>>>> On Thu, Dec 20, 2012 at 10:22 AM, Joe Bowser <bows...@gmail.com> > >>>>>wrote: > >>>>>> We are totally doing something wrong with the way that we do > >>>>>>releases. > >>>>>> I personally think that we're not using git right, and here's why: > >>>>>> > >>>>>> Currently, when we do a release, we tag the RC, and we test the RC. > >>>>>> There's nothing preventing us from putting things after that tag and > >>>>>> if we don't want to those things in the release branching off that > >>>>>> tag. We've done it before and other than the problem with CoHo, it > >>>>>> worked really well. I propose that instead of tagging the release, > >>>>>>we > >>>>>> branch when we want to do a release, and we do all the bug fixes on > >>>>>> that branch. Once that branch is ready to roll, we merge it back > >>>>>>into > >>>>>> master. In fact, nobody should be working on master except to do > >>>>>> merges. The way we're doing this now feels dirty and wrong. > >>>>>> > >>>>>> I honestly feel that this is a much faster way of working, and that > >>>>>> we're missing the point if we have to tell everyone to jump out of > >>>>>>the > >>>>>> pool every time we do an RC. I know that we could be working on our > >>>>>> branches, but that work is almost entirely invisible to the rest of > >>>>>> the project until it's time to merge it back in, which takes > >>>>>>forever. > >>>>>> > >>>>>> > >>>>>> On Thu, Dec 20, 2012 at 10:07 AM, Michal Mocny <mmo...@chromium.org > > > >>>>>>wrote: > >>>>>>> So there is something to be said about having devs shift focus from > >>>>>>>dev to > >>>>>>> testing during an RC. However, as the team grows, not all of us > >>>>>>>are > >>>>>>>really > >>>>>>> being responsible for cutting releases. Maybe that means we need > >>>>>>>to > >>>>>>>train > >>>>>>> the entire team to change current behavior, but that doesn't feel > >>>>>>> necessary/scalable. > >>>>>>> > >>>>>>> With growing external contributions, I would have to say that a > >>>>>>>code > >>>>>>>freeze > >>>>>>> on trunk doesn't seem to make as much sense. > >>>>>>> > >>>>>>> -Michal > >>>>>>> > >>>>>>> > >>>>>>> On Thu, Dec 20, 2012 at 9:47 AM, Andrew Grieve > >>>>>>><agri...@chromium.org> wrote: > >>>>>>> > >>>>>>>> I definitely think we'd get more done if we didn't have such a > >>>>>>>>long > >>>>>>>> code-freeze. I'm not sure this is the same as what you were > >>>>>>>>suggesting, but > >>>>>>>> have a script/tool to branch all of the platforms into an rc > >>>>>>>>branch. Then, > >>>>>>>> each platform can fix themselves up a bit and tag their RC. > >>>>>>>>Meanwhile, dev > >>>>>>>> can continue to happen on edge. > >>>>>>>> > >>>>>>>> My main concern with our current approach is just that the > >>>>>>>>code-freeze time > >>>>>>>> is super long. > >>>>>>>> > >>>>>>>> > >>>>>>>> On Wed, Dec 19, 2012 at 3:36 PM, Marcel Kinard > >>>>>>>><cmarc...@gmail.com> > >>>>>>>>wrote: > >>>>>>>> > >>>>>>>> > One of the things that strikes me here is the difference between > >>>>>>>>calendar > >>>>>>>> > time and effort time. (This assumes folks already concurred that > >>>>>>>>the rc > >>>>>>>> is > >>>>>>>> > ready to release.) Based on my reading of > >>>>>>>>http://wiki.apache.org/** > >>>>>>>> > cordova/CuttingReleases > >>>>>>>><http://wiki.apache.org/cordova/CuttingReleases>there > >>>>>>>> isn't a lot of effort time involved to cut a release. It seems > >>>>>>>>like > >>>>>>>>a > >>>>>>>> > good chunk of the calendar time is getting folks to tag their > >>>>>>>>platform. > >>>>>>>> > Ideally the promotion from rc to final should take very little > >>>>>>>>effort > >>>>>>>> time. > >>>>>>>> > > >>>>>>>> > What I like about the rc is that it provides a settling > >>>>>>>>mechanism > >>>>>>>>for the > >>>>>>>> > churn to calm down, run tests across more integration, and see > >>>>>>>>the bigger > >>>>>>>> > picture to assess release readiness. I would expect that the > >>>>>>>>promotion > >>>>>>>> from > >>>>>>>> > edge to rc should take a decent amount of effort time, but not > >>>>>>>>because of > >>>>>>>> > the "cut" activities. > >>>>>>>> > > >>>>>>>> > So when we are at rc and don't find any surprises, why does it > >>>>>>>>take a > >>>>>>>> week > >>>>>>>> > to promote to final? If we spend a week in rc1, another week in > >>>>>>>>rc2, and > >>>>>>>> > another week to cut final, that leaves only 1 week in a 4-week > >>>>>>>>cycle for > >>>>>>>> > active dev work? > >>>>>>>> > > >>>>>>>> > I like the ideal of a channel/stream/branch/whatever where there > >>>>>>>>is a > >>>>>>>> > place for the rc to settle without necessarily blocking commits > >>>>>>>>to edge. > >>>>>>>> > Where I'm going with this is that if there is an area where > >>>>>>>>commits to > >>>>>>>> the > >>>>>>>> > rc are carefully controlled, then perhaps one person (i.e, Steve > >>>>>>>>G) could > >>>>>>>> > cut the release for ALL platforms using scripts. This may > >>>>>>>>involve > >>>>>>>>that > >>>>>>>> one > >>>>>>>> > person tagging/branching/whatever across multiple platforms. > >>>>>>>> > > >>>>>>>> > I also like putting the "how to cut" magic in each platform. > >>>>>>>>Then > >>>>>>>>perhaps > >>>>>>>> > a good chunk of coho is tests to make sure that the platform > >>>>>>>>magic > >>>>>>>> > delivered the correct format to it. > >>>>>>>> > > >>>>>>>> > -- Marcel Kinard > >>>>>>>> > > >>>>>>>> > >> > >
