I understand these reasons, and they certainly make sense when a package has a big/complicated src/ directory. Perhaps one day more developers will move the building and checking to cloud services (e.g. I have been using Travis CI), so nobody cares about the building/checking time spent on local machines any more.
For now, I think either `R CMD check --build` (must build before checking) or `R CMD check --force` (definitely check without building) sounds like a reasonable solution. Regards, Yihui -- Yihui Xie <xieyi...@gmail.com> Web: http://yihui.name Department of Statistics, Iowa State University 2215 Snedecor Hall, Ames, IA On Mon, Oct 28, 2013 at 12:40 PM, Martin Maechler <maech...@stat.math.ethz.ch> wrote: >>>>>> Duncan Murdoch <murdoch.dun...@gmail.com> >>>>>> on Sun, 27 Oct 2013 08:56:31 -0400 writes: > > > On 13-10-26 9:49 PM, Simon Urbanek wrote: > >> On Oct 25, 2013, at 12:12 PM, Yihui Xie wrote: > >> > >>> This has been asked soooo many times that I think it may > >>> be a good idea for R CMD check to just stop when the > >>> user passes a directory instead of a tar ball to it, or > >>> automatically run R CMD build before moving on. In my > >>> opinion, sometimes an FAQ and a bug are not entirely > >>> different. > >>> > >> > >> +1 -- and I'd do the same for R CMD INSTALL. If someone > >> insists, there could be --force or something like that > >> for those that really want to work on directories despite > >> all the issues with that, but IMHO the default should be > >> for both INSTALL and check to bail out if not presented > >> with a file -- it would save a lot of people a lot of > >> time spent in chasing ghost issues. > > > That seems like a reasonable suggestion. I wouldn't want > > to lose the ability to install or check a directory; for > > development of packages like rgl which have a lot of > > compiled code, installing from a tarball takes a lot > > longer than installing when all of the code has already > > been compiled. > > Even more extreme for Matrix, and as you mention other packages > with a considerable portion of compiled code. > > R CMD build also does some checks that then are done again, when > you R CMD check the tarball. > It really adds a considerable amount of time/work to have to do > both and I'd strongl oppose to disable checking of the source > directory of a package. > > Actually, I do think that the 'build/check' R code could and should be > modularized so that checking a directory with an 'Authors@R' in DESCRIPTION > ((and no 'Maintainer'/'Authors' as they are auto-created)) would > still work fine... > > > > On the other hand, it isn't all that hard to put together > > an R function or shell script to do it (the hardest part > > is figuring out the name of the tarball). For example: > > > installpkgdir <- function(dir) { > > x <- system(paste("R CMD build", dir), intern=TRUE) > > .... > > } > > > I haven't tested this much, and it doesn't offer options > > to the build or install steps, but it could be the basis > > of a simple function that always builds a tarball before > > installing a source directory. > > well, there have been other ways to bundle "build + check", of > course, but for some developers, package building really is not fast, > notably for people like me with a large library of R packages, > where the all dependent packages incl those in 'Suggests' are > sought and checked for the correct version, etc. > Also, by default the vignette rebuilding; manuals etc take time, > and it used to need other manual manipulations if you disabled those, > and then wanted to check the tarball. > > I'd be very much inconvenienced if I'd need to build packages > every time I want to check them during development. Things are > different shortly before a release. > > Martin ______________________________________________ R-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel
