README.md
In bbolker/revdepcheck: Reverse Dependency Checks

revdepcheck

An attempt to bring together some of the home-grown tools I've cobbled together to do reverse-dependency checking. (At the moment this README file is somewhere between a proper README and a set of stream-of-consciousness notes/grumbles about what I've tried so far and what has(n't) worked.)

The two existing frameworks that I'm aware of for reverse-dependency checking are tools::check_packages_in_dir and devtools::revdep(). I should probably try to roll this code, or the features of this code, into one of those frameworks. The main reason I've written my own is that I started before these frameworks were available. The main reason I haven't given up and gone with one of the others (I did try) is that so far I haven't been able to get either of them to work/do everything I want. (It is of course possible that some of these desiderata can be achieved with one of the existing frameworks, I've just failed to figure out how.)

start from a completely clean sandbox environment (check_packages_in_dir doesn't seem to install the dependencies of the reverse dependencies; revdep_check does).
allow caching of downloaded tarballs and installed packages (revdep_check does this), and of checks (for when you want to re-run the checks again after having fixed issues with some packages, or with an updated version of a package)
allow packages to be drawn from a variety of sites (CRAN, r-forge, bioconductor, local tarballs ... Github???) in order to allow for development/updated versions of packages
ignore installed site library (this one is a huge pain, because in order to shut off inspection of the site library you have to set R_LIBS_SITE and R_LIBS_USER -- and then you don't have access to the package itself! Also, R CMD CHECK itself uses a different mechanism (.R/check.Renviron) and ignores the environment variables ...
use virtual frame buffers to allow testing in a non-interactive session (I think check_packages_in_dir does this, I haven't seen how to make revdeps_check do this)
make nice output/ports, including automatic mailing of results to maintainers? (there is some capability in both of the existing tools, but I haven't looked at them too closely -- I like my versions)

Current frustrations:

if I put this material in a package, I can't use it with restricted site libraries unless I first install it in the restricted site libraries! Perhaps this means I should just not be trying to use this on any existing system and should only be working on the problem inside of some sort of sensible (Rocker) sandbox where I first install
or maybe I should give up on using restrictions ... the main frustration is really that .libPaths() isn't sufficiently flexible -- there isn't any way to turn off the site library inside a running R session ... note Gavin Simpson comment (ADD URL) about R ignoring the contents of .check.Renviron -- need to set R_LIBS explicitly (in the same shell environment) instead?

My code is extremely organic/redundant and needs to be refactored.

how much of the internal complexity can be reduced by using devtools machinery where possible?
it would be nice to avoid installing the focal packages to test every time. Is there an easy way to make sure dependencies are installed without installing the package itself?
It would be nice to be able to parse the DESCRIPTION file within a tarball for dependencies (in particular if dependencies have changed since the CRAN version)
wishlist: use make, or some better tool, to automate the caching/dependency network (make won't work perfectly because ideally one would combine checking the dates and/or versions of package versions on the web, not just timestamps of local files)
better reporting on installation failure
have genReport report on NOTEs, not just WARNING/ERROR ?