boost::mutex is not re-entrant, a thread can only lock it once, otherwise it’s dead-locked. The following code snippet demonstrates it:
#include "boost/thread/mutex.hpp"
#include <iostream>
boost::mutex mtx;
void bar(){
boost::mutex::scoped_lock lLock(mtx);//!! dead-locked here
std::cout << "bar" << std::endl;
}
void foo() {
boost::mutex::scoped_lock lLock(mtx);
std::cout << "foo" << std::endl;
bar();
}
int _tmain(int argc, _TCHAR* argv[]){
foo();
return 0;
}
If you need re-entrant mutex, the boost::recursive_mutex is the choice. The following code snippet demonstrates it:
#include "boost/thread/recursive_mutex.hpp"
#include <iostream>
boost::recursive_mutex r_mtx;
void bar(){
boost::recursive_mutex::scoped_lock lLock(r_mtx);//no problem for a thread to lock more than once
std::cout << "bar" << std::endl;
}
void foo() {
boost::recursive_mutex::scoped_lock lLock(r_mtx);
std::cout << "foo" << std::endl;
bar();
}
int _tmain(int argc, _TCHAR* argv[]){
foo();
return 0;
}
I also did a benchmark on my PC. For 1 locking operation, the result is approximately: boost::mutex: 0.043 micro second, boost::recursive_mutex: 0.068 micro second.
Re-entrant mutex is the default in Java and C#. Generally speaking, if a mutex is shared by many modules/classes, it’s recommended to use boost::recursive_mutex; while if it’s only used by a single module/class and no re-entrant feature needed, it’s recommended to use boost::mutex.