Hi Paul, Super-briefly as I am at work:
On 8 November 2013 at 13:12, Paul Gilbert wrote: | I may have misunderstood the first question in this thread, but now that | I have a vague idea what Travis-CI is about, perhaps someone could | answer some questions (apologies if I have not yet grokked the proper | terminology): A lot of stuff got discussed in the 'Issue' tickets on GH. You can browse them there. For better or worse, kids these days don't use mailing lists ... | | -An R package would often require other R packages in order to build and | test. Would the strategy for making other packages available be to use | "install", or "before_script"? Potentially this settup will take very | much more time than the actual testing, or is there a way to easy this | load? Rock, meet hard place. Until R and key packages are in a pre-build image used by Travis, that time has to be spent. It is a tough job to contribute to global warming and burning cpu cycles and disk spins, but someone has got to do it ... | -Are you thinking of all packages being available via apt-get, or will | this support "R CMD INSTALL"? A mix, because there is no complete CRAN archive accessible via apt-get on an Ubuntu platform. | -How do the virtual machine snapshots get set up? Is that what is being | done in the R+Travis project, or what would be needed to merge this into | travis as a first-class citizen? The intent is to do so much good work in the r-travis project that Travis-CI notices and promotes R to a first-class language. We have a good shot at that. | -Is there a mechanism to contribute worker boxes for specific platforms? | Is it possible to restrict them to be available only for certain | purposes? (e.g. R package builds but not C++ project testing.) I do not fully understand the question. r-travis specific questions are kinda misplaced here. Maybe create an Issue ticket at Github. | -Is this thread on R-sig-debian only because the original question was | about installing on Ubuntu? I have the impression that the Travis-CI | infrastructure is more general than debian, or am I wrong about that?. I think Kirill was lost in an apt-get attempt and asked here which makes perfect sense as this is the place for R on Debian/Ubuntu deployment questions. | WRT my original misunderstanding: I do think you need to think of R and | recommended packages as one combined compiler system, not try to test | newer recommended packages against older R versions. I don't think | mixing new recommended packages with old R versions should be expected | to work very well, and (I hope) very few people would think of using a Agreed, and most of us disagree with what Kirill tried there. Arguably, he was just trying for a creative shortcut, but the better solution would be to have several Ubuntu release (say, current plus LTS plus maybe the upcoming one for bleeding edge) under Travis. A shortcut remains a shortcut, and is ill-advised in something designed to be "the" QA and Testing framework. | mixed system. Even installing from source you get them together. OTOH, | testing new versions of other packages against old versions of R (with | the corresponding recommended packages) makes sense if the package | maintainer has not indicated that a newer R version is needed. Dirk -- Dirk Eddelbuettel | [email protected] | http://dirk.eddelbuettel.com _______________________________________________ R-SIG-Debian mailing list [email protected] https://stat.ethz.ch/mailman/listinfo/r-sig-debian
