みんなのRcpp 35.4 (仮)
Rcppパッケージで外部C++ライブラリを使う
Using external C++ libraries in Rcpp packages

https://heavywatal.github.io/slides/japanr2018/

Rcpp: a bridge between R and C++

r_for = function(n) {
  s = 0; for (i in seq_len(n)) {s = s + 1 / i}; s
}
r_vec = function(n) sum(1 / seq_len(n))

Rcpp::cppFunction("double rcpp(int n) {
  double s = 0; for (int i = 1; i <= n; ++i) {s += 1.0 / i;} return s;
}")  # Compilation takes a few seconds here

n = 1000000L
rbenchmark::benchmark(r_for(n), r_vec(n), rcpp(n))[,1:4]

      test replications elapsed relative
1 r_for(n)          100   3.968   29.835
2 r_vec(n)          100   0.473    3.556
3  rcpp(n)          100   0.133    1.000

⚡ Rcpp helps improving performance.
📦 ~1500 CRAN packages depend on Rcpp.
🎉 Happy 10th birthday and version 1.0.0 (2018-11-05).

Online resources for developing Rcpp packages

• Rcpp-package.pdf by Dirk Eddelbuettel and Romain François
• Compiled code (src/) - R packages by Hadley Wickham
• https://heavywatal.github.io/rstats/devtools
  https://heavywatal.github.io/rstats/rcpp (404)
• みんなのRcpp and Rcpp for everyone by 津駄@teuder さん

But lacking “how to use external libraries” 😢

Necessity is the mother of invention. Let’s do it 💪

Structure of a typical C/C++ library

Source repository:

evolution/
├── README.md
└── src/
    ├── individual.hpp      # header file for declaration
    ├── individual.cpp      # source file for definition
    ├── population.hpp
    └── population.cpp

Compiled and installed:

${PREFIX}/                  # e.g., /usr/local/
├── include/                # header files
│   └── evolution/
│       ├── individual.hpp
│       └── population.hpp
└── lib/
    └── libevolution.a      # shared library (.dylib, .dll)

Libraries can be header-only, e.g., Eigen, nlohmann/json.

Basic steps to incorporate C/C++ libraries

• Write src/Makevars to instruct compilers and linkers:

  CXX_STD=CXX11
  PKG_CPPFLAGS=-DSTRICT_R_HEADERS -I/usr/local/include
  PKG_LIBS=-L/usr/local/lib -Wl,-rpath,/usr/local/lib -levolution
  

• Use configure script to

  • Generate src/Makevars from a template src/Makevars.in.
  • Inspect build environments: OS, compilers, etc.
  • Detect installed libraries.
  • Do anything you want.
    e.g., git, cmake, etc.

Various styles to incorporate C/C++ libraries

• Find and use installed libraries. Fail if not found.
  • 😍 Simple. 😨 Requires users’ knowledge and effort.
  • e.g., xml2, magick, RcppGSL
• Keep source code copies in inst/include/ and src/.
  • 😍 Robust. Developers control everything. 😨 Redundant.
  • e.g., BH, RcppArmadillo, RcppEigen, RcppParallel, RcppTOML
• Fetch and track source codes with Git submodule
  • 🙂 Less redundant. 🙁 Foreign materials in package structure.
  • e.g., rigraph, rstan (StanHeaders)
• Find and use installed libraries. Install if not found.
  • 😍 Lean. Not requires users’ effort.
  • e.g., tumopp

Let’s discuss later 🍻

• Any other ways to handle external C++ libraries?
• Where to share these knowledge?
  • My own website
  • みんなのRcpp
  • Rcpp official documentation
• How are Windows users doing?

Thank you for your attention 🤘