Incorrect broadcasting with uninitialized recieving vector (with suitable capacity)

[GiovaGa]

Consider the following (with suitable includes and initializations for MPI):

const int LPF_MPI_AUTO_INITIALIZE = 0;

void grbProgram( const size_t &n, bool &success ) {
	grb::RC rc = grb::SUCCESS;
	const size_t s = grb::spmd<>::pid();
	grb::Vector< int, grb::reference > v ( 10 );
	rc = rc ? rc : grb::resize( v, 10 );
	if( s == 0 ){
		rc = rc ? rc : grb::set( v, 1 );
		assert( grb::nnz( v ) == grb::size( v ));
	}

	rc = rc ? rc : grb::internal::broadcast( v, 0 );
	assert( rc == grb::SUCCESS );
	std::cerr << "Process " << s << ": " << grb::nnz( v ) << " == " << grb::size( v ) << std::endl;
	assert( grb::nnz( v ) == grb::size( v ) );
}


int main( int argc, char ** argv ) {
	// init things...
	grb::Launcher< grb::FROM_MPI > launcher( MPI_COMM_WORLD );

	// note: this exec passes pointers within a single process
	const grb::RC rc = launcher.exec( &grbProgram, n, success, true );
	if( rc != grb::SUCCESS ) {
		std::cerr << "grb::Launcher< FROM_MPI >::exec returns with non-SUCCESS "
			<< "exit code " << grb::toString(rc) << std::endl;
	}
	// ...

This program crashes at the last assert in grbProgram: assert( grb::nnz( v ) == grb::size( v ) );.

The issue seems to be that even though all return codes are grb::SUCCESS the values in v are not correctly written to all the processes. Doing a grb::set( v, 0 ) in all processes resolves the issue (and the values are correctly communicated).