On 16 November 2016 at 11:30, Rainer Müller wrote: > On 2016-11-16 04:09, Ryan Schmidt wrote: >> >>> On Nov 15, 2016, at 6:06 AM, rod <r...@pu-gh.com> wrote: >>> >>> (Apologies if this has been brought up before but I couldn't find a >>> way to search the mailing list archives...) >>> >>> Travis CI has tight integration with Github for testing PRs, and >>> may offer what you need... >>> >>> https://docs.travis-ci.com/user/osx-ci-environment/ >>> https://docs.travis-ci.com/user/customizing-the-build >> >> We rejected Travis CI as unsuitable when we set up our Buildbot >> system to build ports after commits to master. Why do you think the >> same reasons for rejection would not apply to building ports after a >> pull request? > > I cannot remember previous discussion or its arguments, but Travis CI > offers only recent versions of OS X and we have no control over the > installation, such as which Xcode would be installed or when new > versions are available. It also has time limits that would not work for > all ports [1]. Therefore it is indeed not suitable to produce binary > archives. > > However, as a quick check to see if a Portfile will work at all on at > least some version of macOS, Travis CI should be sufficient for the > majority of ports. Just running a 'port lint' check should even work for > all.
I totally agree with Rainer. Travis is useless as our main platform, but there might be a slim chance that it could be used for the most basic testing of pull requests (if we could avoid interference with HB and if we could make it work at all – to start with). I have no clue how difficult it would be to set everything up. There is a chance that we would have to install MacPorts from scratch each time we would want to test something. Maybe there is a chance to ask them for some basic support of MacPorts, ideally to the same extent as HB is supported. I doubt this would be accepted, but you never know. Even if only checksums would be tested (or "port -v patch" or some other basic stuff), that would be more than nothing. If one could in fact install the full port, even if just for some pull requests and even if just on a single OS, that would be more that we could have wished for. Sure, ideally we would test the PR on all build slaves, but that's currently impossible and will remain impossible for quite some time. Mojca
