Sorry should have provided the link to this page: https://travis-ci.org/ . If you scroll down a bit on that page there's a Pull Request flow section, it's the flow I'd be most in favour of. There's also a decent (but rushed) demo here: https://www.youtube.com/watch?v=Uft5KBimzyk . We actually don't need to do a lot of the work that he demos, i.e. no node or gulp configuration. Our setup is close enough to default a default java project that we just have to tell it to build java 8 and then it runs maven properly.
Using a CI server would have some aspects that are similar to the branching document you mention, and some benefits that are a bit orthogonal. Most of these benefits have to do with unit testing, which isn't covered in the doc. First the orthogonal benefits: The main benefit we would get from using CI is that we guarantee code in our repo is never broken. That is to say tests always pass and it always builds correctly. CI servers are really useful to prevent problems where one developer may have everything working properly on his/her machine, but when they later realize it's not working on another devs machine. A good example of this is the class-based-lm-test we pushed recently. It works fine for me locally but it would fail for anyone without kenlm.so. There are many other examples (javadoc errors, code style, etc) but what will happen in these cases is we'll see a big obvious 'The build has problems' message in the PR page on Github. If the CI server runs of all of our code quality checks and finds that everything is good we'll get a big 'This PR is ready to merge' message. Now to the part that overlaps a bit with branching. There are various branching strategies that we could adopt for the project. The master / dev branch one is a possibility. I'd suggest we try commit code strictly in PRs rather than pushing to git. This would be the equivalent of feature branching from your link. The reason I'd suggest that approach is that from what I've seen it'll be dead simple to get working with Github and Travis, and it gives us the same goal of having a stable master branch. If you'd like we can walk through setting this up together on a forked version of our Github repo. We could do a quick example of how code would be pushed and merged. I should be available for a google hangout some time this week if that works for you? -Kellen On Mon, Jul 11, 2016 at 10:51 PM, Mattmann, Chris A (3980) < chris.a.mattm...@jpl.nasa.gov> wrote: > CI = continuous integration :) > > ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ > Chris Mattmann, Ph.D. > Chief Architect > Instrument Software and Science Data Systems Section (398) > NASA Jet Propulsion Laboratory Pasadena, CA 91109 USA > Office: 168-519, Mailstop: 168-527 > Email: chris.a.mattm...@nasa.gov > WWW: http://sunset.usc.edu/~mattmann/ > ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ > Director, Information Retrieval and Data Science Group (IRDS) > Adjunct Associate Professor, Computer Science Department > University of Southern California, Los Angeles, CA 90089 USA > WWW: http://irds.usc.edu/ > ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ > > > > > > > > > > > On 7/11/16, 4:50 PM, "Matt Post" <p...@cs.jhu.edu> wrote: > > >This sounds fine to me. What does CI stand for? > > > >Another thing we should do, which might be complementary to this, is just > be more formal about our process. I had been using this method for a while: > > > > http://nvie.com/posts/a-successful-git-branching-model/ > > > >Sort of informally, but that could be a good approach (I think someone > suggested it a while ago). In short: > > > >- "master" is always stable and records official releases > >- development takes place on "develop" > >- if you need to make an important fix, you branch off master, fix it, > then merge that into both "master" (as a point release) and "develop" > > > >I was using "release" for releases and "master" for develop, but we could > adopt anything. > > > >Kellen, how does this fit with CI? It seems like we could set it up to do > testing on "master" and "develop" branches --- the first as a sanity check, > and the second as a test for when we could merge into master? > > > >matt > > > > > >> On Jul 11, 2016, at 8:17 AM, kellen sunderland < > kellen.sunderl...@gmail.com> wrote: > >> > >> We've made a lot of progress on moving the project over to Apache + > Maven. > >> I was wondering if now would be a good time to consider re-thinking how > we > >> merge changes into master. The main goal would be to make sure we have > a > >> stable master branch that everyone can pull from. > >> > >> What I'd suggest is that we only merge into master once CI has completed > >> testing. This way we can codify style rules, best practices, and make > sure > >> builds succeed and tests pass. We can develop new features create PRs > as > >> normal, and then get quick feedback if those PRs are mergable. I'd also > >> suggest we dis-allow manual pushing to the master branch. > >> > >> I'm not sure how much effort this would be with the existing CI server, > but > >> I could investigate this if someone could grant me admin permissions. > If > >> it's a Jenkins server I'm sure it's possible. > >> > >> Another option is to use Travis CI. I have taken a quick look at > Travis CI > >> and it seems like a quite polished solution. It's free to use for open > >> source projects. It supports automatically building + testing PRs. The > >> interface is really clean. It has email notifications and group > >> administration support. It's got support for multiple (programming) > >> languages so we could in theory build kenlm as a build step and run > those > >> tests. > >> > >> Here's some more info on what the workflow with Travis-CI and PRs would > be > >> https://docs.travis-ci.com/user/pull-requests > >> > >> What do you guys think? Is there a strong preference for using Jenkins > >> from the Apache community? Would everyone be ok with avoiding direct > >> pushes to master? > >> > >> -Kellen > > >
