Re: std::thread...too little, too late?
me <noone@all.net> wrote in news:B_ykw.832074$Ub6.8258@fx20.iad:
I was looking at what made its way into the C++11 standard in the way
of threads, and it seems very braindead. I mean third party libraries
have been giving us working threading for years, and "THAT"was all
they came up with for the standard?
I fully understand the implications of adding features that need to be
portable and that give a good least common denominator, but
std::thread seems too little, too late.
I thought it would be easy to at least build upon the std::thread
class and to something that more closely resembles the common
threading utility available in a variety of libraries: something with
delayed thread starting, and the ability to make the process OO with a
virtual run() method...no such luck.
Consider the following example that uses std::thread as a class trait
and then subclasses a UsefulThread class, adding the requested
functionality:
class UsefulThread {
thread* stdThread;
public:
UsefulThread();
virtual ~UsefulThread() { delete stdThread; }
virtual void run()=0;
void start() { stdThread=new thread(run); }
void join() { stdThread->join(); }
};
class WorkerThread: public UsefulThread {
int _id;
BufferObject& r;
public:
WorkerThread(int id, BufferObject& b): UsefulThread(), _id(id),
r(b)
{}
virtual void run() { for (auto i=0; i<_id; i++) r << _id; }
// share r object is made threadsafe with an internal mutex
};
int main(int argc, char** argv) {
using namespace std;
list<int> v ={1, 2, 3, 4, 5, 6, 7, 8, 9};
BufferObject b;
list<WorkerThread> threadList;
for (auto i: v) { WorkerThread t(i, b); threadList.push_back(t);
}; for (auto i: threadList) i.start();
for (auto i: threadList) i.join();
cout << b.str() << endl;
return 0;
}
This example mostly demonstrates antipatterns and is probably inspired by
some Java-originating ideas that everything must be OO and derived from
some common base class.
I have not used std::thread but I gather it is similar to boost::thread
so I use that instead as a reference.
The data in the WorkerThread object remains in shared use so the access
to it must be protected. With boost::thread the functor object will be
copied over to another thread before the actual thread start, ensuring
that the user does not have to worry about how to get the data safely
into the new thread.
The delayed start() method complicates the usage and implementation and
is not needed for anything. If the thread needs to wait before performing
any actual tasks it could just wait on some condition in the start of the
thread function.
The run() function is called only once when the thread is created. At the
thread creation time the exact class which is created is of course known
so there is no need to have run() as a virtual function. In the above
example it is virtual only because somebody wanted to call it from a
useless non-template base class.
when using a normal thread object instead of a thread*
the resulting WorkerThread cannot be put into any sort of std
container, and gives compile errors about (use of deleted function)
from the constructor.
This is because copying of threads does not make sense (should it spawn a
new thread or what?), and all STL containers are based on value copying.
Use some pointers or smartpointers instead for holding the thread objects
in a container.
So the idea behind boost::thread and std::thread is to provide the most
simple building block, without any unnecessary kludges like forcing use
of virtual functions and without forcing use of shared data. It can be
used quite easily in simple scenarios. If a more complex scenario is
needed, one can build on top of it (like WorkerThread does, albeit in ill
fashion). For example, to avoid functor copying and keep data in shared
use one can define a simple template (probably boost has it already
defined by some other name) which can be used for all such threads:
template<typename T>
struct FunctorForwarder {
T& p_;
FunctorForwarder(T& p): p_(p) {}
void operator()() {p_();}
};
class TimerMgr {
public:
TimerMgr():
thread_(FunctorForwarder<TimerMgr>(*this))
{}
void operator()() {
// the thread function
}
// ...
private:
boost::thread thread_;
// ... shared data
};
hth
Paavo