I was wondering how could I make some Rcpp code use automatic unwind protection in all Rcpp object creations.
For example, suppose I have some code like this:
#include <stdint.h>
#include <Rcpp.h>
class MyObj {
public:
int val;
MyObj(int val) : val(val) {};
~MyObj() {
std::cout << "I' being destructed - value was: " << val << std::endl;
}
};
// [[Rcpp::export]]
Rcpp::NumericVector crashme(unsigned int seed)
{
srand(seed);
MyObj obj1(rand());
Rcpp::NumericVector out(INT64_MAX-1, 100.);
return out;
}
When I call crashme, obj1 doesn't get destructed before the function ends, due to R's long jumps which I want to protect against.
I see there is a function Rcpp::unwindProtect, but it's implemented as something that takes a callback.
I'm not 100% sure if I'm doing it right, but I managed to add unwind protection like this:
#include <stdint.h>
#include <Rcpp.h>
#include <Rcpp/unwindProtect.h>
// [[Rcpp::plugins(unwindProtect)]]
class MyObj {
public:
int val;
MyObj(int val) : val(val) {};
~MyObj() {
std::cout << "I' being destructed - value was: " << val << std::endl;
}
};
struct NumVecArgs {
size_t size;
double fillwith;
};
SEXP alloc_NumVec(void *data)
{
NumVecArgs *args = (NumVecArgs*)data;
return Rcpp::NumericVector(args->size, args->fillwith);
}
// [[Rcpp::export]]
Rcpp::NumericVector crashme(unsigned int seed)
{
srand(seed);
MyObj obj1(rand());
NumVecArgs args = {INT64_MAX-1, 100.};
Rcpp::NumericVector out = Rcpp::unwindProtect(alloc_NumVec, (void*)&args);
return out;
}
Now calling crashme will successfully destruct obj1 and print the destructor message.
But this is very inconvenient, since I have a series of different Rcpp object allocations taking different constructor types, which would imply either defining a different struct and callback for each one of them, or translating all the calls to lengthy lambda functions.
Is there any way to automatically make all calls to constructors of e.g. Rcpp::NumericVector and Rcpp::IntegerVector have unwind protection?
Related
I am trying to construct a std::thread with a member function that takes no arguments and returns void. I can't figure out any syntax that works - the compiler complains no matter what. What is the correct way to implement spawn() so that it returns a std::thread that executes test()?
#include <thread>
class blub {
void test() {
}
public:
std::thread spawn() {
return { test };
}
};
#include <thread>
#include <iostream>
class bar {
public:
void foo() {
std::cout << "hello from member function" << std::endl;
}
};
int main()
{
std::thread t(&bar::foo, bar());
t.join();
}
EDIT:
Accounting your edit, you have to do it like this:
std::thread spawn() {
return std::thread(&blub::test, this);
}
UPDATE: I want to explain some more points, some of them have also been discussed in the comments.
The syntax described above is defined in terms of the INVOKE definition (§20.8.2.1):
Define INVOKE (f, t1, t2, ..., tN) as follows:
(t1.*f)(t2, ..., tN) when f is a pointer to a member function of a class T and t1 is an object of type T or a reference to an object of
type T or a reference to an object of a type derived from T;
((*t1).*f)(t2, ..., tN) when f is a pointer to a member function of a class T and t1 is not one of the types described in the previous
item;
t1.*f when N == 1 and f is a pointer to member data of a class T and t 1 is an object of type T or a
reference to an object of type T or a reference to an object of a
type derived from T;
(*t1).*f when N == 1 and f is a pointer to member data of a class T and t 1 is not one of the types described in the previous item;
f(t1, t2, ..., tN) in all other cases.
Another general fact which I want to point out is that by default the thread constructor will copy all arguments passed to it. The reason for this is that the arguments may need to outlive the calling thread, copying the arguments guarantees that. Instead, if you want to really pass a reference, you can use a std::reference_wrapper created by std::ref.
std::thread (foo, std::ref(arg1));
By doing this, you are promising that you will take care of guaranteeing that the arguments will still exist when the thread operates on them.
Note that all the things mentioned above can also be applied to std::async and std::bind.
Since you are using C++11, lambda-expression is a nice&clean solution.
class blub {
void test() {}
public:
std::thread spawn() {
return std::thread( [this] { this->test(); } );
}
};
since this-> can be omitted, it could be shorten to:
std::thread( [this] { test(); } )
or just (deprecated)
std::thread( [=] { test(); } )
Here is a complete example
#include <thread>
#include <iostream>
class Wrapper {
public:
void member1() {
std::cout << "i am member1" << std::endl;
}
void member2(const char *arg1, unsigned arg2) {
std::cout << "i am member2 and my first arg is (" << arg1 << ") and second arg is (" << arg2 << ")" << std::endl;
}
std::thread member1Thread() {
return std::thread([=] { member1(); });
}
std::thread member2Thread(const char *arg1, unsigned arg2) {
return std::thread([=] { member2(arg1, arg2); });
}
};
int main(int argc, char **argv) {
Wrapper *w = new Wrapper();
std::thread tw1 = w->member1Thread();
std::thread tw2 = w->member2Thread("hello", 100);
tw1.join();
tw2.join();
return 0;
}
Compiling with g++ produces the following result
g++ -Wall -std=c++11 hello.cc -o hello -pthread
i am member1
i am member2 and my first arg is (hello) and second arg is (100)
#hop5 and #RnMss suggested to use C++11 lambdas, but if you deal with pointers, you can use them directly:
#include <thread>
#include <iostream>
class CFoo {
public:
int m_i = 0;
void bar() {
++m_i;
}
};
int main() {
CFoo foo;
std::thread t1(&CFoo::bar, &foo);
t1.join();
std::thread t2(&CFoo::bar, &foo);
t2.join();
std::cout << foo.m_i << std::endl;
return 0;
}
outputs
2
Rewritten sample from this answer would be then:
#include <thread>
#include <iostream>
class Wrapper {
public:
void member1() {
std::cout << "i am member1" << std::endl;
}
void member2(const char *arg1, unsigned arg2) {
std::cout << "i am member2 and my first arg is (" << arg1 << ") and second arg is (" << arg2 << ")" << std::endl;
}
std::thread member1Thread() {
return std::thread(&Wrapper::member1, this);
}
std::thread member2Thread(const char *arg1, unsigned arg2) {
return std::thread(&Wrapper::member2, this, arg1, arg2);
}
};
int main() {
Wrapper *w = new Wrapper();
std::thread tw1 = w->member1Thread();
tw1.join();
std::thread tw2 = w->member2Thread("hello", 100);
tw2.join();
return 0;
}
Some users have already given their answer and explained it very well.
I would like to add few more things related to thread.
How to work with functor and thread.
Please refer to below example.
The thread will make its own copy of the object while passing the object.
#include<thread>
#include<Windows.h>
#include<iostream>
using namespace std;
class CB
{
public:
CB()
{
cout << "this=" << this << endl;
}
void operator()();
};
void CB::operator()()
{
cout << "this=" << this << endl;
for (int i = 0; i < 5; i++)
{
cout << "CB()=" << i << endl;
Sleep(1000);
}
}
void main()
{
CB obj; // please note the address of obj.
thread t(obj); // here obj will be passed by value
//i.e. thread will make it own local copy of it.
// we can confirm it by matching the address of
//object printed in the constructor
// and address of the obj printed in the function
t.join();
}
Another way of achieving the same thing is like:
void main()
{
thread t((CB()));
t.join();
}
But if you want to pass the object by reference then use the below syntax:
void main()
{
CB obj;
//thread t(obj);
thread t(std::ref(obj));
t.join();
}
I am trying to put the condition inside a function but it is throwing confusing compile time error . While if I write it in lambda function like this []{ retur i == k;} it is showing k is unidentified . Can anybody tell How to solve this problem .
#include <iostream>
#include <mutex>
#include <sstream>
#include <thread>
#include <chrono>
#include <condition_variable>
using namespace std;
condition_variable cv;
mutex m;
int i;
bool check_func(int i,int k)
{
return i == k;
}
void print(int k)
{
unique_lock<mutex> lk(m);
cv.wait(lk,check_func(i,k)); // Line 33
cout<<"Thread no. "<<this_thread::get_id()<<" parameter "<<k<<"\n";
i++;
return;
}
int main()
{
thread threads[10];
for(int i = 0; i < 10; i++)
threads[i] = thread(print,i);
for(auto &t : threads)
t.join();
return 0;
}
Compiler Error:
In file included from 6:0:
/usr/include/c++/4.9/condition_variable: In instantiation of 'void std::condition_variable::wait(std::unique_lock<std::mutex>&, _Predicate) [with _Predicate = bool]':
33:30: required from here
/usr/include/c++/4.9/condition_variable:97:14: error: '__p' cannot be used as a function
while (!__p())
^
wait() takes a predicate, which is a callable unary function returning bool. wait() uses that predicate like so:
while (!pred()) {
wait(lock);
}
check_func(i,k) is a bool. It's not callable and it's a constant - which defeats the purpose. You're waiting on something that can change. You need to wrap it in something that can be repeatedly callable - like a lambda:
cv.wait(lk, [&]{ return check_func(i,k); });
Consider a container of type T elements, e.g. std::vector<T> input, and an excessively time-consuming function T f(const T&, const T&). I wish to apply f on input until one single element remains - in parallel.
I am new to C++ so I adapted an existing implementation for a Thread pool pattern in C++11 to my needs, in particular this example.
Code for discussion:
#include <thread>
#include <iostream>
#include <algorithm>
#include <vector>
#include <queue>
#include <memory>
#include <future>
#include <functional>
#include <stdexcept>
#include "ThreadPool.h"
template<class T>
T ParallelReduction(ThreadPool& pool, const std::vector<T>& input, const std::function<T(T,T)>& func)
{
std::queue<std::future<T>> futureResults;
std::for_each(input.begin(), input.end(), [&futureResults](T e)
{
// T -> future<T>
futureResults.push(std::async([e]{return e;}));
});
for (;;)
{
if (futureResults.size() > 1)
{
// Get two "old" partial results, forward them to func(T,T)
T a = futureResults.front().get();
futureResults.pop();
T b = futureResults.front().get();
futureResults.pop();
futureResults.push(pool.enqueue([a, b, func]{return func(a,b);}));
}
else
{
// all input and partial result elements have been processed once, return the final result
return futureResults.front().get();
}
}
}
int main(int argc, char *argv[])
{
ThreadPool pool(4);
std::vector<int> input(12);
int n = 0;
std::generate(input.begin(), input.end(), [&n]{ return ++n; });
auto sum = ParallelReduction<int>(pool, input, [](int a, int b){return a + b;});
std::cout << sum << "\n";
auto fibonacci = ParallelReduction<int>(pool, input, [](int a, int b){return a * b;});
std::cout << fibonacci << "\n";
return 0;
}
My questions:
Is it possible to implement this without introducing futureResults, i.e. a solution that works "inline" on std::vector<T>& input?
If not, can we optimize the way we assign values T to std::future<T> in the first part?
Is it more efficient to forward std::future<T> to our function func instead of T, i.e. let the child thread wait and access the value of a and b and not the main thread?
The parts inside the loop look awful - any ideas?
Where can I optimize the code?
I have some homework, and I have troubles understanding, (probably) how passing parameters to std::thread constructor works.
Assume following code (I deleted unneeded parts)
template<typename T, typename Task>
class Scheduler
{
private:
typedef std::unordered_map<std::size_t, T> Results;
class Solver
{
public:
Solver(Task&& task) : m_thread(&Solver::thread_function, std::move(task))
{
m_thread.detach();
}
Solver(Solver&& solver) = default; // required for vector::emplace_back
~Solver() = default;
private:
void thread_function(Task&& task)
{
task();
}
std::thread m_thread;
};
public:
Scheduler() = default;
~Scheduler() = default;
void add_task(Task&& task)
{
m_solvers.emplace_back(std::move(task));
}
private:
std::vector<Solver> m_solvers;
};
template<typename T>
struct Ftor
{
explicit Ftor(const T& t) : data(t) { }
T operator()() { std::cout << "Computed" << std::endl; return data; }
T data;
};
int main()
{
Scheduler<int, Ftor<int>> scheduler_ftor;
Scheduler<int, std::function<int(void)>> scheduler_lambda;
Ftor<int> s(5);
scheduler_ftor.add_task(std::move(s));
scheduler_lambda.add_task([](){ std::cout << "Computed" << std::endl; return 1; });
}
Why it doesn't compile?
MVS2015 is complaining about
functional(1195): error C2064: term does not evaluate to a function taking 1 arguments
functional(1195): note: class does not define an 'operator()' or a user defined conversion operator to a pointer-to-function or reference-to-function that takes appropriate number of arguments
note: while compiling class template member function 'Scheduler<int,Ftor<int> >::Solver::Solver(Task &&)'
While G++ 4.9.2
functional: In instantiation of ‘struct std::_Bind_simple<std::_Mem_fn<void (Scheduler<int, Ftor<int> >::Solver::*)(Ftor<int>&&)>(Ftor<int>)>’:
required from ‘void Scheduler<T, Task>::add_task(Task&&) [with T = int; Task = Ftor<int>]’
functional:1665:61: error: no type named ‘type’ in ‘class std::result_of<std::_Mem_fn<void (Scheduler<int, Ftor<int> >::Solver::*)(Ftor<int>&&)>(Ftor<int>)>’
typedef typename result_of<_Callable(_Args...)>::type result_type;
I suppose there are some problems with std::moving to std::thread.
If you use member function as first thread argument, second argument supposed to be this pointer, pointing to the object to which member function could be called to
UPDATE
Good discussion here
Start thread with member function
I don't follow your code, but addressing the question, a extrapolated answer will be( most of the code is psuedocode)
lets assume that there is a function int test(int name).
thread t0;
t0 = thread(test,32);
thread t1(test,43);
Passing a argument to function.
int temp = 0;
int testfunc(int& q)
{
cout<<q;
}
thread t1;
t1 = thread(testfunc,ref(temp));
In short, you just pass the name of the function that must be run in the thread as the first argument, and the functions parameters follow it in same order as they are in the function definition, for passing by reference you can use the ref() wrapper.See the below example.
#include <iostream>
#include <thread>
#include <string>
using namespace std;
void test(int a,int &a,string test)
{
\\do something
}
int main()
{
int test1 = 0;
string tt = "hello";
thread t1;
t1 = thread(23,&test1,tt);
t1.detach();
return 0;
}
if you are wondering about the use of join() and detach(), refer to this thread: When should I use std::thread::detach?, refer to my answer post in that thread.
I'm trying to write some code that can execute a class member function in a separate thread, but with some added code for checking odds and ends before and after the thread has executed.
For this I'm using the std::bind and std::thread functionality of c++11 in VS2012.
All this works well if I fix the arguments the class member function can have to e.g. void*, but if I try to template that, I get an error.
Overview of Code
I have a simple class containing two functions, which will be executed in a thread. They differ in arguments and name.
The class I'm creating has a templated constructor and a function, which executes the functions of the previous class, but with the ability to perform it's own checks/notifications if the thread does/does not finish.
Finally the main function is simply to test the code.
The Code
Include Part:
#include <iostream>
#include <memory>
#include <thread>
#include <functional>
using namespace std;
Class to be executed:
class k1
{
public:
k1( int nVal = 0 ) : _val(nVal){};
~k1(){};
void doFunc( void * pParam ){
cout << "Val in class = " << _val << ", param = " << pParam << "\n";
}
void doFunc2( float pParam ){
cout << "Val in class = " << _val << ", param = " << pParam << "\n";
}
int _val;
};
typedef shared_ptr<k1> PK;
Thread Handle Class
class H
{
public:
void controlFunction( std::function<void(void)> callRef ){
cout << "Before calling\n";
callRef();
cout << "After calling\n";
}
template<class T, typename ParamType> // Constructor for any type of class function - void * parameter as only input
H( void(T::* pFunction)(ParamType *), T * pClass, ParamType pParam ){
std::function<void(void)> _call = std::bind( pFunction, pClass, pParam );
_thread = shared_ptr<std::thread>( new thread( &H::controlFunction, this, _call ));
}
~H( void ){
_thread->join();
}
shared_ptr<thread> _thread;
};
typedef shared_ptr<H> PH;
Main Function:
int main(int argc, char* argv[])
{
PK k = make_shared<k1>( 12 );
int i1 = 2;
float f1 = 1.0f;
PH p = PH( new H(&k1::doFunc, k.get(), &i1 ));
PH p2 = PH( new H(&k1::doFunc2, k.get(), f1 ));
return 0;
}
The error that comes out is:
error C2660: 'H::H' : function does not take 3 arguments
Thanks in advance!
/Henrik
I'm guessing that you get the error on the second line where you create a H object taking k1::doFunc2 as argument. The probable reason for the error is because k1::doFunc2 doesn't take a pointer for argument, while the member function pointer argument in the H constructor expects it to.
There is also some problems with the first line, when you declare the p variable. This is because then the ParamType template type can be deduced to be either int or int*. The reason for this is because the member function pointer have ParamType *, while the third argument to the constructor uses non-pointer ParamType but you pass a pointer here (&i1).