Install to custom (user) directory

[gaborcsardi]

• Windows
• macOS
• Linux

Windows

On Windows, this seems easy. It is possible that we'll need to replace Rscript.exe, see below at macOS.

macOS

• On macOS we can set DYLD_LIBRARY_PATH in the R startup script:

  # export DYLD_PRINT_LIBRARIES=1
  R_HOME_DIR=/Users/gaborcsardi/R/R-4.5.3
  R_HOME_DIR="$(cd "$(dirname "$(readlink -f "$0" 2>/dev/null || realpath "$0" 2>/dev/null || echo "$0")")/.." && pwd)"
  export DYLD_LIBRARY_PATH="${R_HOME_DIR}/lib"
  

  Positron can't handle this (yet), so temporarily we'll set R_HOME_DIR to a constant first, because this is what Positron parses from the R file.

• We'll also need to update the hardcoded paths in some script files.

• The most hacky part is working around the install path being hardcoded into the Rscript binary. We'll need to replace that with a shell script that sets RHOME, which Rscript uses to override the R home, and then calls the original Rscript binary.

AFAICT there is no need to patch the binaries with install_name_tool, because DYLD_LIBRARY_PATH takes precedence even over absolute paths. Using install_name_tool and otool is not great, anyway, because they are not part of a vanilla macOS installation.

Similarly, AFAICT there is no need to patch CRAN binary packages after install.packages(), either. CRAN binaries work OOTB, without patching.

Linux

• The difficulty here is the system packages. We need to use the local package manage (i.e. apt, dpkg, etc) to download system packages and install them into the R installation. Based on some quick tests, this is possible. We'll need to create some symlinks for BLAS/LAPACK, otherwise it is pretty straightforward, at least on Debian/Ubuntu.

• We'll need to update the hardcoded paths in scripts.

• We need to replace Rscript with a script, as discussed above for macOS.