% !Rnw driver = highlight::HighlightWeaveLatex()
\documentclass[10pt]{article}
%\VignetteIndexEntry{Rcpp-sugar}
%\VignetteEngine{highlight::highlight}
%\VignetteKeywords{Rcpp, syntactic sugar}
%\VignetteDepends{Rcpp}

\usepackage[USletter]{vmargin}
\setmargrb{0.75in}{0.75in}{0.75in}{0.75in}

\usepackage{color, alltt}
\usepackage[authoryear,round,longnamesfirst]{natbib}
\usepackage[colorlinks]{hyperref}
\definecolor{link}{rgb}{0,0,0.3}	%% next few lines courtesy of RJournal.sty
\hypersetup{
    colorlinks,%
    citecolor=link,%
    filecolor=link,%
    linkcolor=link,%
    urlcolor=link
}

\usepackage{microtype}                     %% cf http://www.khirevich.com/latex/microtype/
\usepackage[T1]{fontenc}		   %% cf http://www.khirevich.com/latex/font/
\usepackage[bitstream-charter]{mathdesign} %% cf http://www.khirevich.com/latex/font/

\newcommand{\proglang}[1]{\textsf{#1}}
\newcommand{\pkg}[1]{{\fontseries{b}\selectfont #1}}
\newcommand{\sugar}{\textsl{Rcpp sugar}}
\newcommand{\ith}{\textsl{i}-\textsuperscript{th}}
\newcommand{\code}[1]{\texttt{#1}}


%% This corresponds to setting boxes=TRUE for highlight
\newsavebox{\hlbox}
\definecolor{hlBg}{rgb}{0.949019607843137,0.949019607843137,0.949019607843137}
\definecolor{hlBd}{rgb}{0.9,0.9,0.9} % border
\renewenvironment{Hchunk}{\vspace{0.5em}\noindent\begin{lrbox}{\hlbox}\begin{minipage}[b]{.9\textwidth}}%
    {\end{minipage}\end{lrbox}\fcolorbox{hlBd}{hlBg}{\usebox{\hlbox}}\vspace{0.5em}}


<<echo=FALSE,print=FALSE>>=
prettyVersion <- packageDescription("Rcpp")$Version
prettyDate <- format(Sys.Date(), "%B %e, %Y")
@

\author{Dirk Eddelbuettel \and Romain Fran\c{c}ois}
\title{\pkg{Rcpp} syntactic sugar}
\date{\pkg{Rcpp} version \Sexpr{prettyVersion} as of \Sexpr{prettyDate}}

\begin{document}
\maketitle

\abstract{
  \noindent
  This note describes \sugar~which has been introduced in
  version 0.8.3 of \pkg{Rcpp} \citep{CRAN:Rcpp,JSS:Rcpp}. \sugar~
  brings a higher-level of abstraction to \proglang{C++} code
  written using the \pkg{Rcpp} API.

  \sugar~is based on expression templates
  \citep{Abrahams+Gurtovoy:2004:TemplateMetaprogramming,Vandevoorde+Josuttis:2003:Templates} and provides
  some 'syntactic sugar' facilities directly in \pkg{Rcpp}. This is similar
  to how \pkg{RcppArmadillo} \citep{CRAN:RcppArmadillo} offers
  linear algebra \proglang{C++} classes based on \pkg{Armadillo}
  \citep{Sanderson:2010:Armadillo}.
  % TODO: reference to armadillo, blitz, etc ...
}

\section{Motivation}

\pkg{Rcpp} facilitates development of internal compiled code in an \proglang{R}
package by abstracting low-level details of the \proglang{R} API \citep{R:Extensions}
into a consistent set of \proglang{C++} classes.

Code written using \pkg{Rcpp} classes is easier to read, write and maintain,
without loosing performance. Consider the following code example which
provides a function \texttt{foo} as a \proglang{C++} extension to
\proglang{R} by using the \pkg{Rcpp} API:

<<lang=cpp>>=
RcppExport SEXP foo( SEXP x, SEXP y){
    Rcpp::NumericVector xx(x), yy(y) ;
    int n = xx.size() ;
    Rcpp::NumericVector res( n ) ;
    double x_ = 0.0, y_ = 0.0 ;
    for( int i=0; i<n; i++){
        x_ = xx[i] ;
        y_ = yy[i] ;
        if( x_ < y_ ){
            res[i] = x_ * x_ ;
        } else {
            res[i] = -( y_ * y_)  ;
        }
    }
    return res ;
}
@

The goal of the function \texttt{foo} code is simple. Given two
\texttt{numeric} vectors, we create a third one. This is typical low-level
\proglang{C++} code that that could be written much more consicely in
\proglang{R} thanks to vectorisation as shown in the next example.

<<eval=FALSE>>=
foo <- function(x, y){
    ifelse( x < y, x*x, -(y*y) )
}
@

Put succinctly, the motivation of \sugar~is to bring a subset of the
high-level \proglang{R} syntax in \proglang{C++}. Hence, with \sugar, the
\proglang{C++} version of \texttt{foo} now becomes:

<<lang=cpp>>=
RcppExport SEXP foo( SEXP xs, SEXP ys){
    Rcpp::NumericVector x(xs) ;
    Rcpp::NumericVector y(ys) ;
    return Rcpp::wrap( ifelse( x < y, x*x, -(y*y) )) ;
}
@

Apart from the fact that we need to assign the two objects we obtain from
\proglang{R}---which is a simple statement each thanks to the template magic
in \pkg{Rcpp}---and the need for explicit \texttt{return} and
\texttt{Rcpp::wrap} statements, the code is now identical between
highly-vectorised \proglang{R} and \proglang{C++}.

\sugar~is written using expression templates and lazy evaluation techniques
\citep{Abrahams+Gurtovoy:2004:TemplateMetaprogramming,Vandevoorde+Josuttis:2003:Templates}.
This not only allows a much nicer high-level syntax, but also makes it
rather efficient (as we detail in section~\ref{sec:performance} below).

\section{Operators}

\sugar~takes advantage of \proglang{C++} operator overloading. The next few
sections discuss several examples.

\subsection{Binary arithmetic operators}

\sugar~defines the usual binary arithmetic operators : \texttt{+}, \texttt{-},
\texttt{*}, \texttt{/}.

<<lang=cpp>>=
// two numeric vectors of the same size
NumericVector x ;
NumericVector y ;

// expressions involving two vectors
NumericVector res = x + y ;
NumericVector res = x - y ;
NumericVector res = x * y ;
NumericVector res = x / y ;

// one vector, one single value
NumericVector res = x + 2.0 ;
NumericVector res = 2.0 - x;
NumericVector res = y * 2.0 ;
NumericVector res = 2.0 / y;

// two expressions
NumericVector res = x * y + y / 2.0 ;
NumericVector res = x * ( y - 2.0 ) ;
NumericVector res = x / ( y * y ) ;
@

The left hand side (lhs) and the right hand side (rhs) of each binary
arithmetic expression must be of the same type (for example they should be both
\texttt{numeric} expressions).

The lhs and the rhs can either have the same size or one of them could
be a primitive value of the appropriate type, for example adding a
\texttt{NumericVector} and a \texttt{double}.

\subsection{Binary logical operators}

Binary logical operators create a \texttt{logical} sugar expression
from either two sugar expressions of the same type or one sugar expression
and a primitive value of the associated type.

<<lang=cpp>>=
// two integer vectors of the same size
NumericVector x ;
NumericVector y ;

// expressions involving two vectors
LogicalVector res = x < y ;
LogicalVector res = x > y ;
LogicalVector res = x <= y ;
LogicalVector res = x >= y ;
LogicalVector res = x == y ;
LogicalVector res = x != y ;

// one vector, one single value
LogicalVector res = x < 2 ;
LogicalVector res = 2 > x;
LogicalVector res = y <= 2 ;
LogicalVector res = 2 != y;

// two expressions
LogicalVector res = ( x + y ) <  ( x*x ) ;
LogicalVector res = ( x + y ) >= ( x*x ) ;
LogicalVector res = ( x + y ) == ( x*x ) ;
@

\subsection{Unary operators}

The unary \texttt{operator-} can be used to negate a (numeric) sugar expression.
whereas the unary \texttt{operator!} negates a logical sugar expression:

<<lang=cpp>>=
// a numeric vector
NumericVector x ;

// negate x
NumericVector res = -x ;

// use it as part of a numerical expression
NumericVector res = -x * ( x + 2.0 ) ;

// two integer vectors of the same size
NumericVector y ;
NumericVector z ;

// negate the logical expression "y < z"
LogicalVector res = ! ( y < z );
@

\section{Functions}

\sugar~defines functions that closely match the behavior of \proglang{R}
functions of the same name.

\subsection{Functions producing a single logical result}

Given a logical sugar expression, the \texttt{all} function identifies if all
the elements are \texttt{TRUE}. Similarly, the \texttt{any} function
identifies if any the element is \texttt{TRUE} when
given a logical sugar expression.


<<lang=cpp>>=
IntegerVector x = seq_len( 1000 ) ;
all( x*x < 3 ) ;
any( x*x < 3 ) ;
@

Either call to \texttt{all} and \texttt{any} creates an object of a class
that has member functions \texttt{is\_true}, \texttt{is\_false},
\texttt{is\_na} and a conversion to \texttt{SEXP} operator.

One important thing to highlight is that \texttt{all} is lazy. Unlike
\proglang{R}, there is no need to fully evaluate the expression. In the
example above, the result of \texttt{all} is fully resolved after evaluating
only the two first indices of the expression \verb|x * x < 3|. \texttt{any}
is lazy too, so it will only need to resolve the first element of the example
above.

%\subsubsection{Conversion to bool}

One important thing to note concernc the conversion to the \texttt{bool}
type. In order to respect the concept of missing values (\texttt{NA}) in
\proglang{R}, expressions generated by \texttt{any} or \texttt{all} can not
be converted to \texttt{bool}. Instead one must use \texttt{is\_true},
\texttt{is\_false} or \texttt{is\_na}:

<<lang=cpp>>=
// wrong: will generate a compile error
bool res = any( x < y) ) ;

// ok
bool res = is_true( any( x < y ) )
bool res = is_false( any( x < y ) )
bool res = is_na( any( x < y ) )
@

% FIXME this may need some expanding the trivariate bool and how to use it


\subsection{Functions producing sugar expressions}

\subsubsection{is\_na}

Given a sugar expression of any type, \verb|is_na| (just like the other
functions in this section) produces a logical sugar expression of the same
length. Each element of the result expression evaluates to \texttt{TRUE} if
the corresponding input is a missing value, or \texttt{FALSE} otherwise.

<<lang=cpp>>=
IntegerVector x = IntegerVector::create( 0, 1, NA_INTEGER, 3 ) ;

is_na( x )
all( is_na( x ) )
any( ! is_na( x ) )
@

\subsubsection{seq\_along}

Given a sugar expression of any type, \texttt{seq\_along} creates an
integer sugar expression whose values go from 1 to the sire of the input.

<<lang=cpp>>=
IntegerVector x = IntegerVector::create( 0, 1, NA_INTEGER, 3 ) ;

seq_along( x )
seq_along( x * x * x * x * x * x * x )
@

This is the most lazy function, as it only needs to call the \texttt{size}
member function of the input expression. The input expression need not to be
resolved. The two examples above gives the same result with the same efficiency
at runtime. The compile time will be affected by the complexity of the
second expression, since the abstract syntax tree is built at compile time.

\subsubsection{seq\_len}

\texttt{seq\_len} creates an integer sugar expression whose \ith\
element expands to \texttt{i}. \texttt{seq\_len} is particularly useful in
conjunction with \texttt{sapply} and \texttt{lapply}.

<<lang=cpp>>=
// 1, 2, ..., 10
IntegerVector x = seq_len( 10 ) ;

lapply( seq_len(10), seq_len )
@

\subsubsection{pmin and pmax}

Given two sugar expressions of the same type and size, or one expression and
one primitive value of the appropriate type, \texttt{pmin} (\texttt{pmax})
generates a sugar expression of the same type whose \ith\ element expands to
the lowest (highest) value between the \ith\ element of the first expression
and the \ith element of the second expression.

<<lang=cpp>>=
IntegerVector x = seq_len( 10 ) ;

pmin( x, x*x )
pmin( x*x, 2 )

pmin( x, x*x )
pmin( x*x, 2 )
@

\subsubsection{ifelse}

Given a logical sugar expression and either :
\begin{itemize}
\item two compatible sugar expression (same type, same size)
\item one sugar expression and one compatible primitive
\end{itemize}
\texttt{ifelse} expands to a sugar expression whose \ith\
element is the \ith\ element of the first expression
if the \ith\ element of the condition expands to \texttt{TRUE}
or the \ith\ of the second expression if
the \ith\ element of the condition expands to \texttt{FALSE},
or the appropriate missing value otherwise.

<<lang=cpp>>=
IntegerVector x ;
IntegerVector y ;

ifelse( x < y, x, (x+y)*y )
ifelse( x > y, x, 2 )
@

\subsubsection{sapply}

\texttt{sapply} applies a \proglang{C++} function to each element
of the given expression to create a new expression. The type of the
resulting expression is deduced by the compiler from the result type of
the function.

The function can be a free \proglang{C++} function such as the overload
generated by the template function below:

<<lang=cpp>>=
template <typename T>
T square( const T& x){
    return x * x ;
}
sapply( seq_len(10), square<int> ) ;
@

Alternatively, the function can be a functor whose type has a nested type
called \texttt{result\_type}

<<lang=cpp>>=
template <typename T>
struct square : std::unary_function<T,T> {
    T operator()(const T& x){
        return x * x ;
    }
}
sapply( seq_len(10), square<int>() ) ;
@

\subsubsection{lapply}

\texttt{lapply} is similar to \texttt{sapply} except that the result is
allways an list expression (an expression of type \texttt{VECSXP}).

\subsubsection{sign}

Given a numeric or integer expression, \texttt{sign} expands to an expression
whose values are one of 1, 0, -1 or \texttt{NA}, depending on the sign
of the input expression.

<<lang=cpp>>=
IntegerVector xx;

sign( xx )
sign( xx * xx )
@

\subsubsection{diff}

The \ith\ element of the result of \texttt{diff} is
the difference between the $(i+1)$\textsuperscript{th} and the
\ith\ element of the input expression. Supported types are
integer and numeric.

<<lang=cpp>>=
IntegerVector xx;

diff( xx )
@

\subsection{Mathematical functions}

For the following set of functions, generally speaking, the \ith\ element of
the result of the given function (say, \texttt{abs}) is the result of
applying that function to this \ith\ element of the input expression.
Supported types are integer and numeric.

<<lang=cpp>>=
IntegerVector x;

abs( x )
exp( x )
floor( x )
ceil( x )
pow(x, z)     # x to the power of z
@

% log() and log10() maybe?  Or ln() ?

\subsection{The d/q/p/r statistical functions}

The framework provided by \sugar also permits easy and efficient access the
density, distribution function, quantile and random number generation
functions function by \proglang{R} in the \code{Rmath} library.

Currently, most of these functions are vectorised for the first element which
denote size. Consequently, these calls works in \proglang{C++} just as they
would in \proglang{R}:

<<lang=cpp>>=
x1 = dnorm(y1, 0, 1);  // density of y1 at m=0, sd=1
x2 = qnorm(y2, 0, 1);  // quantiles of y2
x3 = pnorm(y3, 0, 1);  // distribution function of y3
x4 = rnorm(n, 0, 1);   // 'n' RNG draws of N(0, 1)
@

Similar d/q/p/r functions are provided for the most common distributions:
beta, binom, cauchy, chisq, exp, f, gamma, geom, hyper, lnorm, logis, nbeta,
nbinom, nbinom\_mu, nchisq, nf, norm, nt, pois, t, unif, and weibull.

Note that the parameterization used in these sugar functions may differ between
the top-level functions exposed in an \proglang{R} session. For example, 
the internal \code{rexp} is parameterized by \code{scale},
whereas the R-level \code{stats::rexp} is parameterized by \code{rate}. 
Consult \href{http://cran.r-project.org/doc/manuals/r-release/R-exts.html#Distribution-functions}{Distribution Functions}
for more details on the parameterization used for these sugar functions.

One point to note is that the programmer using these functions needs to
initialize the state of the random number generator as detailed in Section
6.3 of the `Writing R Extensions' manual \citep{R:Extensions}.  A nice
\proglang{C++} solution for this is to use a \textsl{scoped} class that sets
the random number generatator on entry to a block and resets it on exit. We
offer the \code{RNGScope} class which allows code such as

<<lang=cpp>>=
RcppExport SEXP getRGamma() {
    RNGScope scope;
    NumericVector x = rgamma( 10, 1, 1 );
    return x;
}
@

As there is some computational overhead involved in using \code{RNGScope}, we
are not wrapping it around each inner function.  Rather, the user of these
functions (\textsl{i.e.} you) should place an \code{RNGScope} at the
appropriate level of your code.


\section{Performance}
\label{sec:performance}

TBD

\section{Implementation}

This section details some of the techniques used in the implementation of
\sugar. Note that the user need not to be familiar with the implementation
details in order to use \sugar, so this section can be skipped upon a first
read of the paper.

Writing \sugar~functions is fairly repetitive and follows a well-structured
pattern. So once the basic concepts are mastered (which may take time given
the inherent complexities in template programming), it should be possible to
extend the set of function further following the established pattern..

\subsection{The curiously recurring template pattern}

Expression templates such as those used by \sugar~use a technique
called the \emph{Curiously Recurring Template Pattern} (CRTP). The general
form of CRTP is:

<<lang=cpp>>=
// The Curiously Recurring Template Pattern (CRTP)
template <typename T>
struct base {
    // ...
};
struct derived : base<derived> {
    // ...
};
@

The \texttt{base} class is templated by the class that derives from it :
\texttt{derived}. This shifts the relationship between a base class and a
derived class as it allows the base class to access methods of the derived
class.

\subsection{The VectorBase class}

The CRTP is used as the basis for \sugar~with the \texttt{VectorBase}
class template. All sugar expression derive from one class generated by the
\texttt{VectorBase} template. The current definition of \texttt{VectorBase}
is given here:

<<lang=cpp>>=
template <int RTYPE, bool na, typename VECTOR>
class VectorBase {
public:
    struct r_type : traits::integral_constant<int,RTYPE>{} ;
    struct can_have_na : traits::integral_constant<bool,na>{} ;
    typedef typename traits::storage_type<RTYPE>::type stored_type ;

    VECTOR& get_ref(){
        return static_cast<VECTOR&>(*this) ;
    }

    inline stored_type operator[]( int i) const {
        return static_cast<const VECTOR*>(this)->operator[](i) ;
    }

    inline int size() const { return static_cast<const VECTOR*>(this)->size() ; }

    /* definition ommited here */
    class iterator ;

    inline iterator begin() const { return iterator(*this, 0) ; }
    inline iterator end() const { return iterator(*this, size() ) ; }
}
@

The \texttt{VectorBase} template has three parameters:

\begin{itemize}
\item \texttt{RTYPE}: This controls the type of expression (INTSXP, REALSXP,
  ...)
\item \texttt{na}: This embeds in the derived type information about whether
  instances may contain missing values. \pkg{Rcpp} vector types
  (\texttt{IntegerVector}, ...)  derive from \texttt{VectorBase} with this
  parameter set to \texttt{true} because there is no way to know at
  compile-time if the vector will contain missing values at run-time.
  However, this parameter is set to \texttt{false} for types that are
  generated by sugar expressions as these are guaranteed to produce
  expressions that are without missing values. An example is the
  \texttt{is\_na} function. This parameter is used in several places as part
  of the compile time dispatch to limit the occurence of redundant
  operations.
\item \texttt{VECTOR}: This parameter is the key of \sugar. This is the
  manifestation of CRTP. The indexing operator and the \texttt{size} method
  of \texttt{VectorBase} use a static cast of \texttt{this} to the
  \texttt{VECTOR} type to forward calls to the actual method of the derived
  class.
\end{itemize}

\subsection{Example : sapply}

As an example, the current implementation of \texttt{sapply}, supported by
the template class \texttt{Rcpp::sugar::Sapply} is given below:

<<lang=cpp>>=
template <int RTYPE, bool NA, typename T, typename Function>
class Sapply : public VectorBase<
    Rcpp::traits::r_sexptype_traits<
        typename ::Rcpp::traits::result_of<Function>::type
    >::rtype ,
    true ,
    Sapply<RTYPE,NA,T,Function>
> {
public:
    typedef typename ::Rcpp::traits::result_of<Function>::type  ;
    const static int RESULT_R_TYPE =
        Rcpp::traits::r_sexptype_traits<result_type>::rtype ;
    typedef Rcpp::VectorBase<RTYPE,NA,T> VEC ;
    typedef typename Rcpp::traits::r_vector_element_converter<RESULT_R_TYPE>::type
        converter_type ;
    typedef typename Rcpp::traits::storage_type<RESULT_R_TYPE>::type STORAGE ;

    Sapply( const VEC& vec_, Function fun_ ) : vec(vec_), fun(fun_){}

    inline STORAGE operator[]( int i ) const {
        return converter_type::get( fun( vec[i] ) );
    }
    inline int size() const { return vec.size() ; }

private:
    const VEC& vec ;
    Function fun ;
} ;

// sugar

template <int RTYPE, bool _NA_, typename T, typename Function >
inline sugar::Sapply<RTYPE,_NA_,T,Function>
sapply( const Rcpp::VectorBase<RTYPE,_NA_,T>& t, Function fun ){
    return sugar::Sapply<RTYPE,_NA_,T,Function>( t, fun ) ;
}
@

\subsubsection{The sapply function}

\texttt{sapply} is a template function that takes two arguments.
\begin{itemize}
\item The first argument
is a sugar expression, which we recognize because of the relationship with
the \texttt{VectorBase} class template.
\item The second argument is the function to apply.
\end{itemize}

The \texttt{sapply} function itself does not do anything, it is just used
to trigger compiler detection of the template parameters that will be used
in the \texttt{sugar::Sapply} template.

\subsubsection{Detection of return type of the function}

In order to decide which kind of expression is built, the \texttt{Sapply}
template class queries the template argument via the \texttt{Rcpp::traits::result\_of}
template.

<<lang=cpp>>=
typedef typename ::Rcpp::traits::result_of<Function>::type result_type ;
@

The \texttt{result\_of} type trait is implemented as such:

<<lang=cpp>>=
template <typename T>
struct result_of{
    typedef typename T::result_type type ;
} ;

template <typename RESULT_TYPE, typename INPUT_TYPE>
struct result_of< RESULT_TYPE (*)(INPUT_TYPE) >{
    typedef RESULT_TYPE type ;
} ;
@

The generic definition of \texttt{result\_of} targets functors
with a nested \texttt{result\_type} type.

The second definition is a partial specialization targetting
function pointers.

\subsubsection{Indentification of expression type}

Based on the result type of the function, the \texttt{r\_sexptype\_traits}
trait is used to identify the expression type.

<<lang=cpp>>=
const static int RESULT_R_TYPE =
    Rcpp::traits::r_sexptype_traits<result_type>::rtype ;
@

\subsubsection{Converter}

The \texttt{r\_vector\_element\_converter} class is used to convert an
object of the function's result type to the actual storage type suitable
for the sugar expression.

<<lang=cpp>>=
typedef typename Rcpp::traits::r_vector_element_converter<RESULT_R_TYPE>::type
    converter_type ;

@

\subsubsection{Storage type}

The \texttt{storage\_type} trait is used to get access to the storage type
associated with a sugar expression type. For example, the storage type
of a \texttt{REALSXP} expression is \texttt{double}.

<<lang=cpp>>=
typedef typename Rcpp::traits::storage_type<RESULT_R_TYPE>::type STORAGE ;
@

\subsubsection{Input expression base type}

The input expression --- the expression over which \texttt{sapply} runs --- is
also typedef'ed for convenience:

<<lang=cpp>>=
typedef Rcpp::VectorBase<RTYPE,NA,T> VEC ;
@

\subsubsection{Output expression base type}

In order to be part of the \sugar~system, the type generated by the
\texttt{Sapply} class template must inherit from \texttt{VectorBase}.

<<lang=cpp>>=
template <int RTYPE, bool NA, typename T, typename Function>
class Sapply : public VectorBase<
    Rcpp::traits::r_sexptype_traits<
        typename ::Rcpp::traits::result_of<Function>::type
    >::rtype ,
    true ,
    Sapply<RTYPE,NA,T,Function>
>
@

The expression built by \texttt{Sapply} depends on the result type
of the function, may contain missing values, and the third argument
is the manifestation of the \emph{CRTP}.

\subsubsection{Constructor}

The constructor of the \texttt{Sapply} class template is straightforward, it
simply consists of holding the reference to the input expression and the
function.

<<lang=cpp>>=
Sapply( const VEC& vec_, Function fun_ ) : vec(vec_), fun(fun_){}

private:
    const VEC& vec ;
    Function fun ;
@

\subsubsection{Implementation}

The indexing operator and the \texttt{size} member function is what
the \texttt{VectorBase} expects. The size of the result expression is
the same as the size of the input expression and the i\textsuperscript{th}
element of the result is simply retrieved by applying the function
and the converter. Both these methods are inline to maximize performance:

<<lang=cpp>>=
inline STORAGE operator[]( int i ) const {
	return converter_type::get( fun( vec[i] ) );
}
inline int size() const { return vec.size() ; }
@

\section{Summary}

TBD

\bibliographystyle{plainnat}
\bibliography{Rcpp}

\end{document}