module property with a free-function getter and member setter crashes at module load #1519

class_<T>::property() has one overload per getter/setter combination (member function, const member function, or free function taking T*). The overload taking a free-function getter and a member-function setter never returns *this:

template <typename PROP>
self& property( const char* name_, PROP (*GetMethod)(Class*), void (Class::*SetMethod)(PROP), const char* docstring = 0 ){
    AddProperty(
        name_,
        new CppProperty_GetPointer_SetMethod<Class,PROP>(GetMethod, SetMethod, docstring)
    ) ;
}

Flowing off the end of a non-void function is undefined behavior, and the result is used by the next call in the class_<T> builder chain. With Apple clang the compiler emits a trap instruction there, so any module that uses this overload kills the R session as soon as the module is loaded. The compiler does warn:

inst/include/Rcpp/module/Module_Add_Property.h:86:2: warning: non-void function does not return a value [-Wreturn-type]

but sourceCpp() and package builds don't surface warnings by default. The code dates back to 2010; none of the free-function property() overloads are exercised by the test suite.

Reproducible example

repro.cpp:

#include <Rcpp.h>

using namespace Rcpp;

class Counter {
public:
    Counter() : n(0) {}

    void set_n(int n_) {
        n = n_;
    }

    int n;
};

// getter as a free function taking the object pointer
int get_n(Counter* c) {
    return c->n;
}

RCPP_MODULE(counter_module) {
    class_<Counter>("Counter")
        .constructor()
        // free-function getter + member-function setter
        .property("n", &get_n, &Counter::set_n);
}

repro.R:

Rcpp::sourceCpp("repro.cpp")
cat("module loaded\n")

counter <- new(Counter)
counter$n <- 5L
print(counter$n)

Output with R 4.6.1 (macOS 27, Apple clang 21, Rcpp 1.1.2.6 / current master):

Trace/BPT trap: 5          Rscript repro.R

Rscript exits with code 133 (SIGTRAP) while sourceCpp() loads the module; module loaded is never printed. Other compilers may not trap, but the behavior is still undefined.

I'll follow up with a PR that adds the missing return *this; and adds tests covering every property() overload.