[Rcpp-devel] Using Armadillo objects as arguments in calls to R

[Davor Cubranic]

I think x, y, and weights are in my case not necessarily SEXPs, but can be (i.e., some of them are pure Armadillo colvecs, while others are created from a NumericVector without copying its contents).

Davor


On 2010-05-11, at 1:25 PM, Romain Francois wrote:

> 
> Thanks for letting us know of the problem ... and the solution
> 
> Not sure what the problem is. The _["x"] =  is intended for implicit calls to wrap, but it should not trouble you if you make them explicit. maybe we did not handle properly the special case where the rhs is already a SEXP.
> 
> Will have a look when I get a chance.
> 
> Thanks.
> 
> Romain
> 
> Le 11/05/10 22:17, Davor Cubranic a écrit :
>> 
>> Hmm, once I stopped using 'wrap' and just passed Armadillo objects using 'Named', I'm able to run unit tests with no errors.
>> 
>> I.e., I now use:
>> 
>>   List data = List::create(_["x"] = x,
>> 			   _["y"] = y);
>>   const NumericVector out_r = predict_fn(loess_fn(formula_fn("y~x"),
>> 						  data,
>> 						  _["weights"] = weights,
>> 						  _["span"] = span));
>> 
>> Could the issue really have been caused my using 'wrap'? What does 'wrap' do that would lead to a segfault, that I now avoid by using 'Named'?
>> 
>> Davor
>> (R 2.10.1, Rcpp 0.7.11, Armadillo 0.9.4)
>> 
>> 
>> On 2010-05-11, at 12:35 PM, Davor Cubranic wrote:
>> 
>>> Can anyone tell me about issues I have to look out for when dealing with Rcpp Functions and passing Armadillo objects as arguments? I'm getting "not compatible with INTSXP" and segfault errors when calling the following Rcpp equivalent of R's 'predict(loess(y~x, weights=w, span=span))'. The errors usually only manifest only after several dozen calls.
>>> 
>>> Davor
>>> 
>>> 
>>> const vec predict(const vec&  x, const vec&  y,
>>> 			const vec&  weights, const double span) {
>>>  Function formula_fn("formula");
>>>  Function loess_fn("loess");
>>>  Function predict_fn("predict");
>>> 
>>>  List data = List::create(_["x"] = wrap(x),
>>> 			   _["y"] = wrap(y));
>>>  NumericVector weights_r = wrap(weights);
>>>  const NumericVector out_r = predict_fn(loess_fn(formula_fn("y~x"),
>>> 						  data,
>>> 						  weights_r,
>>> 						  Named("span") = span));
>>>  const vec out = vec(out_r.begin(), out_r.size());
>>>  return out;
>>> }
> 
> 
> 
> -- 
> Romain Francois
> Professional R Enthusiast
> +33(0) 6 28 91 30 30
> http://romainfrancois.blog.free.fr
> |- http://bit.ly/9aKDM9 : embed images in Rd documents
> |- http://tr.im/OIXN : raster images and RImageJ
> |- http://tr.im/OcQe : Rcpp 0.7.7
>