I have built the following example in order to demonstrate my concern. In
reality I am obtaining an object such as the example mymat with an attached
vector as an attribute from Douglas Bates' 1998 C code for spline_basis in the
splines package. It seems that back in the days of stone knives and clubs
(before Rcpp) this was indeed a unique way of passing a list of objects back to
R.
Some may recall that I have ripped splines.c successfully into a new package
project, keeping the original code intact, complete with its identity.
My desire is to build calling code without any need to go back to the R
interpreter , hoping for optimal performance. As can be seen, by grabbing the
attr() function using what I call the "environment facility" of Rcpp to extract
the attached vector violates my best intentions.
I am hoping for a pure C/C++ way of accessing the vector that returns as an
attribute.
One possibility might be a C function that takes the SEXP object with the
attribute, and returns just the attrubute as a separate SEXP object. Alas,
this is too far over my head at the moment.
src<- '
Rcpp::NumericMatrix m_rcpp(arg1);
arma::mat m_arma = Rcpp::as<arma::mat>(arg1);
Environment base("package:base");
Function getAttrib = base["attr"];
Rcpp::NumericVector v_rcpp(4);
v_rcpp = getAttrib(m_rcpp,"attachment");
arma::vec v_arma = Rcpp::as<arma::vec>(v_rcpp);
return(Rcpp::List::create(Rcpp::Named("origMatrix")=wrap(m_arma),
Rcpp::Named("origVector")=wrap(v_arma)));
'
attr_sep<-cxxfunction(signature(arg1="numeric"),
body = src, plugin = "RcppArmadillo")
## R test code:
mymat<-matrix(1:12,3,4)
myvec<-rep(5,4)
attr(mymat,"attachment")<-myvec
return_list<-attr_sep(mymat)
_______________________________________________
Rcpp-devel mailing list
Rcpp-devel@lists.r-forge.r-project.org
https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel
