Attributes are a new feature of \pkg{Rcpp} version 0.10.0 \citep{CRAN:Rcpp,JSS:Rcpp} that provide infrastructure for seamless language bindings between \proglang{R} and \proglang{C++}. The motivation for attributes is several-fold:
The core concept is to add annotations to \proglang{C++} source files that provide the context required to automatically generate \proglang{R} bindings to \proglang{C++} functions. Attributes and their supporting functions include:
Rcpp::export
attribute to export a \proglang{C++} function
to \proglang{R}sourceCpp
function to source exported functions from a filecppFunction
and evalCpp
functions for inline
declarations and executionRcpp::depends
attribute for specifying additional build
dependencies for sourceCpp
Attributes can also be used for package development via the
compileAttributes
function, which automatically generates
extern "C"
and .Call
wrappers for \proglang{C++}
functions within packages.
Attributes are annotations that are added to C++ source files to provide additional information to the compiler. \pkg{Rcpp} supports attributes to indicate that C++ functions should be made available as R functions, as well as to optionally specify additional build dependencies for source files.
\proglang{C++11} specifies a standard syntax for attributes \citep{Maurer+Wong:2008:AttributesInC++}. Since this standard isn't yet fully supported across all compilers, \pkg{Rcpp} attributes are included in source files using specially formatted comments.
The sourceCpp
function parses a \proglang{C++} file and looks for
functions marked with the Rcpp::export
attribute. A shared
library is then built and its exported functions are made available as R
functions in the specified environment. For example, this source file
contains an implementation of convolve (note the Rcpp::export
attribute in the comment above the function):
```{Rcpp, eval = FALSE}
using namespace Rcpp;
// [[Rcpp::export]] NumericVector convolveCpp(NumericVector a, NumericVector b) {
int na = a.size(), nb = b.size(); int nab = na + nb - 1; NumericVector xab(nab); for (int i = 0; i < na; i++) for (int j = 0; j < nb; j++) xab[i + j] += a[i] * b[j]; return xab;
}
The addition of the export attribute allows us to do this from the \proglang{R} prompt: ```r sourceCpp("convolve.cpp") convolveCpp(x, y)
We can now write \proglang{C++} functions using built-in \proglang{C++} types and \pkg{Rcpp} wrapper types and then source them just as we would an \proglang{R} script.
The sourceCpp
function performs caching based on the last
modified date of the source file and it's local dependencies so as
long as the source does not change the compilation will occur only
once per R session.
If default argument values are provided in the C++ function definition then these defaults are also used for the exported R function. For example, the following C++ function:
```{Rcpp, eval = FALSE} DataFrame readData(CharacterVector file, CharacterVector colNames = CharacterVector::create(), std::string comment = "#", bool header = true)
Will be exported to R as: ```r function(file, colNames=character(), comment="#", header=TRUE)
Note that C++ rules for default arguments still apply: they must occur consecutively at the end of the function signature and (unlike R) can't rely on the values of other arguments.
Not all \proglang{C++} default argument values can be parsed into their \proglang{R} equivalents, however the most common cases are supported, including:
"foo"
)10
or 4.5
)true
, false
,
R_NilValue
, NA_STRING
, NA_INTEGER
,
NA_REAL
, and NA_LOGICAL
.CharacterVector
, IntegerVector
,
and NumericVector
) instantiated using the ::create
static member function.Matrix
types instantiated using the rows
,
cols
constructor.Within \proglang{R} code the stop
function is typically used to signal
errors. Within \proglang{R} extensions written in \proglang{C} the Rf_error
function is typically used. However, within \proglang{C++} code you cannot
safely use Rf_error
because it results in a longjmp
over
any \proglang{C++} destructors on the stack.
The correct way to signal errors within \proglang{C++} functions is to throw an Rcpp::exception
. For example:
```{Rcpp, eval = FALSE} if (unexpectedCondition) throw Rcpp::exception("Unexpected " "condition occurred");
There is also an `Rcpp::stop` function that is shorthand for throwing an `Rcpp::exception`. For example: ```{Rcpp, eval = FALSE} if (unexpectedCondition) Rcpp::stop("Unexpected condition occurred");
In both cases the \proglang{C++} exception will be caught by \pkg{Rcpp} prior to returning control to \proglang{R} and converted into the correct signal to \proglang{R} that execution should stop with the specified message.
You can similarly also signal warnings with the Rcpp::warning
function:
```{Rcpp, eval = FALSE} if (unexpectedCondition) Rcpp::warning("Unexpected condition occurred");
## Supporting User Interruption If your function may run for an extended period of time, users will appreciate the ability to interrupt it's processing and return to the REPL. This is handled automatically for R code (as R checks for user interrupts periodically during processing) however requires explicit accounting for in C and C++ extensions to R. To make computations interrupt-able, you should periodically call the `Rcpp::checkUserInterrupt` function, for example: ```{Rcpp, eval = FALSE} for (int i=0; i<1000000; i++) { // check for interrupt every 1000 iterations if (i % 1000 == 0) Rcpp::checkUserInterrupt(); // ...do some expensive work... }
A good guideline is to call Rcpp::checkUserInterrupt
every 1 or 2
seconds that your computation is running. In the above code, if the user
requests an interrupt then an exception is thrown and the attributes wrapper
code arranges for the user to be returned to the REPL.
Note that R provides a \proglang{C} API for the same purpose
(R_CheckUserInterrupt
) however this API is not safe to use in
\proglang{C++} code as it uses longjmp
to exit the current scope,
bypassing any C++ destructors on the stack. The Rcpp::checkUserInterrupt
function is provided as a safe alternative for \proglang{C++} code.
Typically \proglang{C++} and \proglang{R} code are kept in their own source
files. However, it's often convenient to bundle code from both languages into
a common source file that can be executed using single call to sourceCpp
.
To embed chunks of \proglang{R} code within a \proglang{C++}
source file you include the \proglang{R} code within a block comment that
has the prefix of /*** R
. For example:
```{Rcpp, eval = FALSE} /*** R
fibonacci(10)
*/
Multiple \proglang{R} code chunks can be included in a \proglang{C++} file. The `sourceCpp` function will first compile the \proglang{C++} code into a shared library and then source the embedded \proglang{R} code. ## Modifying Function Names You can change the name of an exported function as it appears to \proglang{R} by adding a name parameter to `Rcpp::export`. For example: ```{Rcpp, eval = FALSE} // [[Rcpp::export(name = ".convolveCpp")]] NumericVector convolveCpp(NumericVector a, NumericVector b)
Note that in this case since the specified name is prefaced by a \code{.} the exported R function will be hidden. You can also use this method to provide implementations of S3 methods (which wouldn't otherwise be possible because C++ functions can't contain a '.' in their name).
Typically, only void
-returning functions are wrapped by invisible()
in
RcppExports.R
. In some cases, however, it is preferred to return an object
invisibly. This can be done by adding an invisible parameter to Rcpp::export
.
For example:
```{Rcpp, eval = FALSE} // [[Rcpp::export(invisible = true)]] NumericVector convolveCpp(NumericVector a, NumericVector b)
Then the R wrapper of `convolveCpp` will return `invisible(.Call(...))` rather than `.Call(...)`. ## Function Requirements Functions marked with the `Rcpp::export` attribute must meet several requirements to be correctly handled: - Be defined in the global namespace (i.e. not within a C++ namespace declaration) - Have a return type that is either void or compatible with `Rcpp::wrap` and parameter types that are compatible with `Rcpp::as` (see sections 3.1 and 3.2 of the '\textsl{Rcpp-jss-2011}' vignette for more details). - Use fully qualified type names for the return value and all parameters. Rcpp types may however appear without a namespace qualifier (i.e. `DataFrame` is okay as a type name but `std::string` must be specified fully). ## Random Number Generation \proglang{R} functions implemented in \proglang{C} or \proglang{C++} need to be careful to surround use of internal random number generation routines (e.g. `unif_rand`) with calls to `GetRNGstate` and `PutRNGstate`. Within \pkg{Rcpp}, this is typically done using the `RNGScope` class. However, this is not necessary for \proglang{C++} functions exported using attributes because an `RNGScope` is established for them automatically. Note that \pkg{Rcpp} implements `RNGScope` using a counter, so it's still safe to execute code that may establish it's own `RNGScope` (such as the \pkg{Rcpp} sugar functions that deal with random number generation). The overhead associated with using `RNGScope` is negligible (only a couple of milliseconds) and it provides a guarantee that all C++ code will inter-operate correctly with R's random number generation. If you are certain that no C++ code will make use of random number generation and the 2ms of execution time is meaningful in your context, you can disable the automatic injection of `RNGScope` using the `rng` parameter of the `Rcpp::export` attribute. For example: ```{Rcpp, eval = FALSE} // [[Rcpp::export(rng = false)]] double myFunction(double input) { // ...code that never uses the // R random number generation... }
It's also possible to use the Rcpp::depends
attribute to declare
dependencies on other packages. For example:
```{Rcpp, eval = FALSE} // [[Rcpp::depends(RcppArmadillo)]]
using namespace Rcpp;
// [[Rcpp::export]] List fastLm(NumericVector yr, NumericMatrix Xr) {
int n = Xr.nrow(), k = Xr.ncol(); arma::mat X(Xr.begin(), n, k, false); arma::colvec y(yr.begin(), yr.size(), false); arma::colvec coef = arma::solve(X, y); arma::colvec rd = y - X*coef; double sig2 = arma::as_scalar(arma::trans(rd)*rd/(n-k)); arma::colvec sderr = arma::sqrt(sig2 * arma::diagvec(arma::inv(arma::trans(X)*X))); return List::create(Named("coef") = coef, Named("sderr")= sderr);
}
The inclusion of the `Rcpp::depends` attribute causes `sourceCpp` to configure the build environment to correctly compile and link against the \pkg{RcppArmadillo} package. Source files can declare more than one dependency either by using multiple `Rcpp::depends` attributes or with syntax like this: ```{Rcpp, eval = FALSE} // [[Rcpp::depends(Matrix, RcppArmadillo)]]
Dependencies are discovered both by scanning for package include directories and by invoking \pkg{inline} plugins if they are available for a package.
Note that while the Rcpp::depends
attribute establishes dependencies
for sourceCpp
, it's important to note that if you include the same
source file in an \proglang{R} package these dependencies must still be
listed in the Imports
and/or LinkingTo
fields of the package
DESCRIPTION
file.
The core use case for sourceCpp
is the compilation of a single
self-contained source file. Code within this file can import other C++ code
by using the Rcpp::depends
attribute as described above.
The recommended practice for sharing C++ code across many uses of
sourceCpp
is therefore to create an R package to wrap the C++
code. This has many benefits not the least of which is easy distribution of
shared code. More information on creating packages that contain C++ code
is included in the Package Development section below.
If you need to share a small amount of C++ code between source files
compiled with sourceCpp
and the option of creating a package
isn't practical, then you can also share code using local includes of C++
header files. To do this, create a header file with the definition of
shared functions, classes, enums, etc. For example:
```{Rcpp, eval = FALSE}
inline double timesTwo(double x) { return x * 2; }
Note the use of the `#ifndef` include guard, this is important to ensure that code is not included more than once in a source file. You should use an include guard and be sure to pick a unique name for the corresponding `#define`. Also note the use of the \code{inline} keyword preceding the function. This is important to ensure that there are not multiple definitions of functions included from header files. Classes fully defined in header files automatically have inline semantics so don't require this treatment. To use this code in a source file you'd just include it based on it's relative path (being sure to use `"` as the delimiter to indicate a local file reference). For example: ```{Rcpp, eval = FALSE} #include "shared/utilities.hpp" // [[Rcpp::export]] double transformValue(double x) { return timesTwo(x) * 10; }
When scanning for locally included header files \code{sourceCpp} also checks for a corresponding implementation file and automatically includes it in the compilation if it exists.
This enables you to break the shared code entirely into it's own source file. In terms of the above example, this would mean having only a function declaration in the header:
```{Rcpp, eval = FALSE}
double timesTwo(double x);
Then actually defining the function in a separate source file with the same base name as the header file but with a .cpp extension (in the above example this would be \code{utilities.cpp}): ```{Rcpp, eval = FALSE} #include "utilities.hpp" double timesTwo(double x) { return x * 2; }
It's also possible to use attributes to declare dependencies and exported functions within shared header and source files. This enables you to take a source file that is typically used standalone and include it when compiling another source file.
Note that since additional source files are processed as separate translation units the total compilation time will increase proportional to the number of files processed. From this standpoint it's often preferable to use shared header files with definitions fully inlined as demonstrated above.
Note also that embedded R code is only executed for the main source file not those referenced by local includes.
Maintaining C++ code in it's own source file provides several benefits including the ability to use \proglang{C++} aware text-editing tools and straightforward mapping of compilation errors to lines in the source file. However, it's also possible to do inline declaration and execution of C++ code.
There are several ways to accomplish this, including passing a code
string to sourceCpp
or using the shorter-form cppFunction
or evalCpp
functions. For example:
cppFunction(' int fibonacci(const int x) { if (x < 2) return x; else return (fibonacci(x-1)) + fibonacci(x-2); } ') evalCpp('std::numeric_limits<double>::max()')
You can also specify a depends parameter to cppFunction
or evalCpp
:
cppFunction(depends='RcppArmadillo', code='...')
One of the goals of \pkg{Rcpp} attributes is to simultaneously facilitate ad-hoc and interactive work with \proglang{C++} while also making it very easy to migrate that work into an \proglang{R} package. There are several benefits of moving code from a standalone \proglang{C++} source file to a package:
To create a package that is based on \pkg{Rcpp} you should follow the
guidelines in the '\textsl{Rcpp-package}' vignette. For a new package this
is most conveniently done using the Rcpp.package.skeleton
function.
To generate a new package with a simple hello, world function that uses attributes you can do the following:
Rcpp.package.skeleton("NewPackage", attributes = TRUE)
To generate a package based on \proglang{C++} files that you've been using
with sourceCpp
you can use the cpp_files
parameter:
Rcpp.package.skeleton("NewPackage", example_code = FALSE, cpp_files = c("convolve.cpp"))
Once you've migrated \proglang{C++} code into a package, the dependencies for
source files are derived from the Imports
and LinkingTo
fields
in the package DESCRIPTION
file rather than the Rcpp::depends
attribute. Some packages also require the addition of an entry to the package
NAMESPACE
file to ensure that the package's shared library is loaded
prior to callers using the package. For every package you import C++ code from
(including \pkg{Rcpp}) you need to add these entries.
Packages that provide only C++ header files (and no shared library) need only
be referred to using LinkingTo
. You should consult the documentation
for the package you are using for the requirements particular to that package.
For example, if your package depends on \pkg{Rcpp} you'd have the following
entries in the DESCRIPTION
file:
```{bash, eval = FALSE} Imports: Rcpp (>= 0.11.4) LinkingTo: Rcpp
And the following entry in your `NAMESPACE` file: ```{bash, eval = FALSE} importFrom(Rcpp, evalCpp)
If your package additionally depended on the \pkg{BH} (Boost headers) package
you'd just add an entry for \pkg{BH} to the LinkingTo
field since
\pkg{BH} is a header-only package:
```{bash, eval = FALSE} Imports: Rcpp (>= 0.11.4) LinkingTo: Rcpp, BH
## Exporting R Functions Within interactive sessions you call the `sourceCpp` function on individual files to export \proglang{C++} functions into the global environment. However, for packages you call a single utility function to export all \proglang{C++} functions within the package. The `compileAttributes` function scans the source files within a package for export attributes and generates code as required. For example, executing this from within the package working directory: ```r compileAttributes()
Results in the generation of the following two source files:
src/RcppExports.cpp
-- The extern "C"
wrappers required
to call exported \proglang{C++} functions within the package.R/RcppExports.R
-- The .Call
wrappers required to call
the extern "C"
functions defined in RcppExports.cpp
.You should re-run compileAttributes
whenever functions are added,
removed, or have their signatures changed. Note that if you are using either
RStudio or \pkg{devtools} to build your package then the
compileAttributes
function is called automatically whenever your
package is built.
The compileAttributes
function deals only with exporting
\proglang{C++} functions to \proglang{R}. If you want the functions to
additionally be publicly available from your package's namespace another
step may be required. Specifically, if your package NAMESPACE
file
does not use a pattern to export functions then you should add an explicit
entry to NAMESPACE
for each R function you want publicly available.
Rcpp attribute compilation will automatically generate a package R_init function that does native routine registration as described here: https://cran.r-project.org/doc/manuals/r-release/R-exts.html#Registering-native-routines.
You may however want to add additional C++ code to the package initialization sequence. To do this, you can add the [[Rcpp::init]]
attribute to functions within your package. For example:
```{Rcpp, eval = FALSE} // [[Rcpp::init]] void my_package_init(DllInfo *dll) { // initialization code here }
In this case, a call to `my_package_init()` will be added to the end of the automatically generated R_init function within RcppExports.cpp. For example: ```{Rcpp, eval = FALSE} void my_package_init(DllInfo *dll); RcppExport void R_init_pkgname(DllInfo *dll) { R_registerRoutines(dll, NULL, CallEntries, NULL, NULL); R_useDynamicSymbols(dll, FALSE); my_package_init(dll); }
In some cases the signatures of the C++ functions that are generated within
RcppExports.cpp
may have additional type requirements beyond the core
standard library and \pkg{Rcpp} types (e.g. CharacterVector
,
NumericVector
, etc.). Examples might include convenience typedefs,
as/wrap handlers for marshaling between custom types and SEXP, or types
wrapped by the Rcpp XPtr
template.
In this case, you can create a header file that contains these type definitions
(either defined inline or by including other headers) and have this header
file automatically included in RcppExports.cpp
. Headers named with
the convention pkgname_types
are automatically included along with
the generated C++ code. For example, if your package is named \pkg{fastcode}
then any of the following header files would be automatically included in
RcppExports.cpp
:
```{Rcpp, eval = FALSE} src/fastcode_types.h src/fastcode_types.hpp inst/include/fastcode_types.h inst/include/fastcode_types.hpp
There is one other mechanism for type visibility in `RcppExports.cpp`. If your package provides a master include file for consumption by C++ clients then this file will also be automatically included. For example, if the \pkg{fastcode} package had a C++ API and the following header file: ```{Rcpp, eval = FALSE} inst/include/fastcode.h
This header file will also automatically be included in
RcppExports.cpp
. Note that the convention of using .h
for
header files containing C++ code may seem unnatural, but this comes from the
recommended practices described in '\textsl{Writing R Extensions}'
\citep{R:Extensions}.
The \pkg{roxygen2} package \citep{CRAN:roxygen2} provides a facility for automatically generating \proglang{R} documentation files based on specially formatted comments in \proglang{R} source code.
If you include roxygen comments in your \proglang{C++} source file with a
//'
prefix then compileAttributes
will transpose them
into R roxygen comments within R/RcppExports.R
. For example the
following code in a \proglang{C++} source file:
```{Rcpp, eval = FALSE} //' The length of a string (in characters). //' //' @param str input character vector //' @return characters in each element of the vector // [[Rcpp::export]] NumericVector strLength(CharacterVector str)
Results in the following code in the generated \proglang{R} source file: ```r #' The length of a string (in characters). #' #' @param str input character vector #' @return characters in each element of the vector strLength <- function(str)
The interface exposed from \proglang{R} packages is most typically a set of
\proglang{R} functions. However, the \proglang{R} package system also provides
a mechanism to allow the exporting of \proglang{C} and \proglang{C++}
interfaces using package header files. This is based on the
R_RegisterCCallable
and R_GetCCallable
functions described in
'\textsl{Writing R Extensions}' \citep{R:Extensions}.
\proglang{C++} interfaces to a package are published within the
top level include
directory of the package (which within the package
source directory is located at inst/include
). The \proglang{R} build
system automatically adds the required include
directories for all
packages specified in the LinkingTo
field of the package
DESCRIPTION
file.
The Rcpp::interfaces
attribute can be used to automatically
generate a header-only interface to your \proglang{C++} functions
within the include
directory of your package.
The Rcpp::interfaces
attribute is specified on a per-source
file basis, and indicates which interfaces (\proglang{R}, \proglang{C++},
or both) should be provided for exported functions within the file.
For example, the following specifies that both R and \proglang{C++} interfaces should be generated for a source file:
```{Rcpp, eval = FALSE} // [[Rcpp::interfaces(r, cpp)]]
Note that the default behavior if an `Rcpp::interfaces` attribute is not included in a source file is to generate an R interface only. ### Generated Code If you request a `cpp` interface for a source file then `compileAttributes` generates the following header files (substituting \emph{Package} with the name of the package code is being generated for): ```{bash, eval = FALSE} inst/include/Package.h inst/include/Package_RcppExports.h
The Package_RcppExports.h
file has inline definitions for all
exported \proglang{C++} functions that enable calling them using the
R_GetCCallable
mechanism.
The Package.h
file does nothing other than include the
Package_RcppExports.h
header. This is done so
that package authors can replace the Package.h
header with
a custom one and still be able to include the automatically generated exports
(details on doing this are provided in the next section).
The exported functions are defined within a \proglang{C++} namespace that matches
the name of the package. For example, an exported \proglang{C++} function
bar
could be called from package MyPackage
as follows:
```{Rcpp, eval = FALSE} // [[Rcpp::depends(MyPackage)]]
void foo() { MyPackage::bar(); }
### Including Additional Code You might wish to use the `Rcpp::interfaces` attribute to generate a part of your package's \proglang{C++} interface but also provide additional custom \proglang{C++} code. In this case you should replace the generated `Package.h` file with one of your own. Note that the way \pkg{Rcpp} distinguishes user verses generated files is by checking for the presence a special token in the file (if it's present then it's known to be generated and thus safe to overwrite). You'll see this token at the top of the generated `Package.h` file, be sure to remove it if you want to provide a custom header. Once you've established a custom package header file, you need only include the `Package_RcppExports.h` file within your header to make available the automatically generated code alongside your own. If you need to include code from your custom header files within the compilation of your package source files, you will also need to add the following entry to `Makevars` and `Makevars.win` (both are in the `src` directory of your package): ```{bash, eval = FALSE} PKG_CPPFLAGS += -I../inst/include/
Note that the R package build system does not automatically force a rebuild
when headers in inst/include
change, so you should be sure to perform a
full rebuild of the package after making changes to these headers.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.