[Rcpp-devel] named and default arguments in exposed C++ classes
Dirk Eddelbuettel
edd at debian.org
Thu Mar 6 16:18:12 CET 2014
On 6 March 2014 at 15:52, Andreas Prescher wrote:
| Thanks a lot for your quick response,
|
| I got a working version, also inspired by
| your comments from
| http://stackoverflow.com/questions/12405655/returning-a-custom-object-from-a-wrapped-method-in-rcpp.
|
| I did something like this in my simple example
| (maybe some typos, I tried with my real package..)
|
| for Foo:
|
| template <> Foo* as( SEXP obj)
| {
| Rcpp::Environment e(obj);
| Rcpp::XPtr<Foo> exPointer( (SEXP) e.get(".pointer") );
| return (Foo*) exPointer;
| }
|
| RCPP_EXPOSED_CLASS(Foo)
|
| SEXP createFoo() {
| Rcpp::XPtr<Foo> exPointer( new Foo(), true ) ;
| Function maker=Environment::Rcpp_namespace()[ "cpp_object_maker"];
| return maker ( typeid(Foo).name() , exPointer );
| }
| int bar(Foo* f,int x, int y) {return f->bar(x,y);}
|
| RCPP_MODULE(PsiMin) {
| function( "createFoo", &createFoo);
| function( "bar", &bar, List::create(_["f"], _["x"], _["y"] = 2));
| }
| in R:
| > f <- createFoo()
| > bar(f = f, x = 3);
|
| Is this OK, are there some memory issues I've overlooked?
|
| What I'm now thinking of (because in my case there are
| normally only 1-16 objects created, depending on the cpu cores),
| is to store the objects on the C++ side,
| so a call of
| bar(f = f, x = 3);
| would be
| bar(3) //calling first object Foo::f inside package
|
| But what is with the allocated memory?
If it is within C++ scope, you get C++ scope behaviour, ie default destructors
will be invoked.
| Of course I could write a cleanup method,
| but this seems not very elegant.
Hard to say based on the toy example. You could always proceed and the
(memory) profile later...
| Is there a safe way, some kind of "hooked method",
| that will be always called when the module
| is unloaded. I only found the finalize method
| in the modules/class context.
That's the one.
Dirk
|
| (Maybe a question for a new thread)
|
|
| Andreas
|
|
|
|
|
| Am 05.03.2014 16:36, schrieb Dirk Eddelbuettel:
| >
| > On 5 March 2014 at 16:26, Andreas Prescher wrote:
| > | Hello,
| > |
| > | I wrote a R package using Rcpp (0.10.4) modules
| > | to expose my C++ classes.
| > |
| > | All works fine, but for convenient using from R,
| > | it would be great to use the class methods with
| > | named arguments and default arguments.
| > |
| > | At the moment I have to use all class methods as
| > | declared in the corresponding C++ header files.
| > |
| > | something like this:
| > | RCPP_MODULE(PsiMin) {
| > |
| > | class_<Foo>( "Foo" )
| > | ..method( "bar", &Foo::bar); }
| > |
| > | class Foo {
| > |
| > | int bar(int x, int y = 2) {return x + y};
| > | }
| > |
| > | using from R
| > | > f <- new(Foo)
| > | > f$bar(x = 1) => 3
| > | > f$bar(x = 1, y = 1) => 2
| > |
| > | I only can call my methods like this:
| > | > f$bar(3,4) => 7
| > |
| > | Is there a workaround to make named/default
| > | arguments working when using classes?
| > |
| > | E.g. make some "mix" with 'function'
| > | and 'class_'.
| > |
| > | Any suggestions would be great.
| >
| > In RcppCNPy I use free functions exposed via Rcpp Modules and that works:
| >
| > R> library(RcppCNPy)
| > R> npyLoad
| > internal C++ function <0x19721a0>
| > docstring : read an npy file into a numeric or integer vector or matrix
| > signature : Rcpp::RObject npyLoad(std::string, std::string, bool)
| > R> args(npyLoad)
| > function (filename, type = "numeric", dotranspose = TRUE)
| > NULL
| > R> str(npyLoad(file="/home/edd/git/rcppcnpy/tests/fvec.npy"))
| > num [1:5] 0 1.1 2.2 3.3 4.4
| > R>
| >
| > If it doesn't work that way for autoexported member functions you may need to
| > a) live with it, b) but a wrapper function around it or c) patch Rcpp Modules.
| >
| > Dirk
| >
|
|
| _______________________________________________
| Rcpp-devel mailing list
| Rcpp-devel at lists.r-forge.r-project.org
| https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel
--
Dirk Eddelbuettel | edd at debian.org | http://dirk.eddelbuettel.com
More information about the Rcpp-devel
mailing list