gc protection problem in dplyr code

[spkal]

The assignment below (several instances of which exist in dplyr.cpp) has a gc protection problem: the object returned by shallow_copy(data) is unprotected upon return, while the DataFrame_Impl constructor (highlighted with __ below) creating the copy object will do an allocation in the default Vector constructor (from which DataFrame_Impl inherits) that will collect the unprotected result of shallow_copy(data) before it can be protected by the set__(x) call in the DataFrame_Impl constructor.

// -- dplyr.cpp
    DataFrame copy = shallow_copy(data) ;

// -- DataFrame.h
    template <template <class> class StoragePolicy>
    class DataFrame_Impl : __public Vector<VECSXP, StoragePolicy>__ {
    public:
        typedef Vector<VECSXP, StoragePolicy> Parent ;

        DataFrame_Impl() : Parent( internal::empty_data_frame() ){}
        DataFrame_Impl(SEXP x) {
            set__(x);
        }

// -- Vector.h  (in Rcpp)
    Vector() {
      Storage::set__( Rf_allocVector(RTYPE, 0 ) );
      init() ;
    }  

[hadley]

@romainfrancois is this now fixed?

[romainfrancois]

Note sure. I'll have a look.

[romainfrancois]

This looks like a problem in Rcpp. I have:

• put a workaround for dplyr for now.
• submitted a PR to Rcpp for an upstream fix of the problem