Changes in R by

We present important changes in the development version of R (referred to as R-devel, to become R 4.2) and give a summary of the new search engine interfaced by RSiteSearch(). Some statistics on bug tracking activities in 2021 are also provided. R-devel selected changes R 4.2.0 is due to be released around April 2022. The following gives a selection of the most important changes in R-devel, which are likely to appear in the new release. Native UTF-8 support and other changes on Windows R on Windows now uses UTF-8 as the native encoding. This feature requires recent Windows 10 or newer (or Windows Server 2022 or newer). On older systems, a (non-Unicode) system locale encoding will be used as in earlier versions of R. With this feature, it is now possible to work with characters not representable in the locale encoding (e.g., with Asian characters on European locales). Previously, such characters could only be used with considerable care needed to prevent their mis-representation or undesirable substitution. It is now possible to use Unicode characters even in Rterm, the console front-end for R. To make this possible, R switched to the Universal C Runtime (UCRT), which is the new C library on Windows and has to be installed manually on Windows 8.1 and older. The switch required a new toolchain targeting UCRT. All code linked statically to R or R packages has to be rebuilt. Therefore, a new toolchain bundle, Rtools42, has been created which includes a recent GCC 10 compiler toolchain targeting 64-bit UCRT and a set of pre-compiled static libraries for R packages. R and CRAN use this new toolchain for R-devel (to become R 4.2.0). Older versions of R will still use older toolchains. As from 4.2, R on Windows will no longer support 32-bit builds. Rtools42, containing only the 64-bit toolchain, is one step simpler to install for users than the earlier toolchain bundle. The change so far required updates of over 100 CRAN packages and several of their Bioconductor dependencies. As these packages have a very large number of reverse dependencies (packages depending recursively on them), R gained support for automated installation-time patching of packages, so that packages can be quickly patched and their reverse dependencies tested, giving package authors more time to incorporate the updates. This feature is experimental and may be removed in the future. R allows package authors to maintain the same package sources for R 4.2 (Rtools42) and R 4.1 (Rtools40) by supporting ‘Makevars.ucrt’ and other make/configuration files with extension ‘.ucrt’ which are used by R 4.2 in preference of their existing ‘.win’ variants, but ignored by older versions of R. Both toolchain bundles can coexist on the same machine. The work on the toolchain and on testing CRAN packages has lead to the discovery of new bugs in GCC: invalid unwind tables causing crashes (GCC PR#103274), inconsistency in option handling related to unwind tables (GCC PR#103465) and lack of support for UCRT/C99 format strings (GCC PR#95130). Additional bugs were found that turned out to be fixed already in later versions of GCC, but required a back-port (GCC PR#101238, GCC PR#100402). Thanks to MinGW-W64 developer Martin Storsjo and GCC developers Eric Botcazou and Martin Liska for their help with identifying and resolving the issues. The Rtools42 toolchain bundle includes patches for these and other, smaller, issues. Following the philosophy that disruptive changes for users and package authors should be rare, this seemed a good time to change also the default personal library location. Now it is a subdirectory of the Local Application Data directory (usually a hidden directory C:\Users\username\ AppData\Local). This is to follow Windows conventions, but also to avoid problems users experienced with various cloud backup/syncing services enabled by default for the personal directory (usually C:\Users\username\Documents). For the very same reason, the default installation location for useronly installation has been changed to C:\Users\username\AppData\Local\Programs. Additional bug fixes (e.g., for handling previously untested code paths involving characters not representable in system locale encoding) and improvements (e.g., removal of workarounds no longer needed with UCRT) are being added following testing and reports from package authors and are to appear in R 4.2. More details on the changes in R for Windows and on what is required from package authors are available in Tomas Kalibera et al. blog post and material linked from there. The R Journal Vol. 13/2, December 2021 ISSN 2073-4859