[Rcpp-devel] Broken conversion from R-type integer to uvec after update to newer Rcpp version

[Kevin Ushey]

Hi guys,

Here's what I think is going on. RcppArmadillo has the following
'InputParameter' class to handle 'as':

template <typename T, typename VEC, typename REF>
    class ArmaVec_InputParameter {
    public:
        ArmaVec_InputParameter( SEXP x_ ) : v(x_), vec(
reinterpret_cast<T*>( v.begin() ), v.size(), false ){}

        inline operator REF(){
            return vec ;
        }

    private:
        Rcpp::Vector< Rcpp::traits::r_sexptype_traits<T>::rtype > v ;
        VEC vec ;
    } ;

We are trying to re-use memory from an R vector, and here we are just
reinterpreting the 'int' contents of the R vector as 'unsigned int' without
any coercion. This won't work.

(Interestingly, in the 'base' case (ie, arma::uvec vs. const or
ref-qualified arma::uvec), we are just dispatching to an internal Rcpp as,
not any of the RcppArmadillo 'as'es. That's why we don't see it here --
that dispatch explicitly uses 'as')

To fix this, we need to coerce when necessary. I believe the
r_sexptype_traits class has a bit of 'does this fit in an R vector, or does
it need coercion?' in it, so maybe we can leverage that.

I have a fix that I will submit to RcppArmadillo as a pull request in a few
moments.

Thanks for the bug report!

Kevin


On Sun, Apr 20, 2014 at 11:36 AM, Dirk Eddelbuettel <edd at debian.org> wrote:

>
> On 20 April 2014 at 20:18, Mitov Venelin wrote:
> | Hi Dirk,
> |
> | Had a look at your example (function named venelin). It seems to me that
> the problem is not only the const qualifier but also the passing of an
> argument by reference.
> |
> | uvec touvec(uvec& v) {    //fails!
> |    return v;
> | }
> |
> | uvec touvec(uvec v) {     // works!
> |   return v;
> | }
> |
> | Don’t know if this helps….
>
> It seems the combination of either/or/both const and & AND use of uvec
> triggers this.  A base uvec vector seems to pass in fine, as does const ref
> use of a normal int vector.
>
> I think we are being tripped up by a partial conversion. We will try to get
> to it.  In the meantime, maybe just avoid 'uvec' which has no direct R
> counterpart.
>
> Dirk
>
> --
> Dirk Eddelbuettel | edd at debian.org | http://dirk.eddelbuettel.com
> _______________________________________________
> 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
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/rcpp-devel/attachments/20140420/2268e90d/attachment.html>