Just last week, I released
drake version 2.1.0 to GitHub and CRAN! This is a minor update: important, but fully back-compatible with the last release (2.0.0). There are three major changes:
A new parallel computing option:
parallel::parLapply(), now default on Windows. From now on, Windows users can enjoy the same single-R-session parallelism already afforded to non-Windows users via
parallel::mclapply(). On non-Windows systems,
mclapply() remains the default mode of parallelism because of its lower overhead and back-compatibility with version 2.0.0.
tracked() function. It is possible to trick
drake into missing dependencies in strange edge cases, so be sure to use
tracked() to list all the targets and imports that are reproducibly tracked in your project. To see the complete dependency map, see the functions
read_graph (already in
drake version 2.0.0).
The "caution" vignette (latest version on GitHub). There are several ways to trick drake into failing or making incorrect decisions, so I decided to compile and explain some of the most likely pitfalls in a vignette.