c++ - Boost.Python 和 Boost.Signals2 : Segmentation faults

我在将 boost.signals2 集成到我使用 boost.python 公开的现有 C++ 库时遇到问题。

我有一个通过 std::shared_ptr 暴露给 python 的类。这个类应该能够在某些事件上发出一些信号。因此，我公开了一个 connect_slot 函数，它接受一个 boost::python::object 作为参数。如果我在连接插槽后直接发出信号，一切正常，但如果类稍后发出信号，我会收到段错误。

我认为这可能与 c++ 库中的线程有关(它也使用 boos::asio 等)

下面是一些代码片段:

我的类.h:

public:
    typedef boost::signals2::signal<void (std::shared_ptr<int>)> signal_my_sig;
    void connect_slot(boost::python::object const & slot);

private:
    signal_my_sig    m_sig;

我的类.cpp:

void MyClass::connect_slot(boost::python::object const & slot) { 
    std::cout << "register shd" << std::endl;
    m_sig.connect(slot);

    m_sig(12345); // this works
}


void MyClass::some_later_event() {
    m_sig(654321); // this does not work

}

我在 python 中使用自定义 python 函数调用 MyClass::connect_slot 函数，如下所示:

def testfunc(some_int):
    print("slot called")

m = myext.MyClass()
m.connect_slot(testfunc)

MyClass::some_later_event 中引发的段错误的回溯(使用 gdb)如下所示:

[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
[New Thread 0x7ffff3c37700 (LWP 20634)]

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7ffff3c37700 (LWP 20634)]
0x00000000004f7480 in PyObject_Call ()
(gdb) 
(gdb) backtrace
#0  0x00000000004f7480 in PyObject_Call ()
#1  0x00000000004f7aa6 in PyEval_CallObjectWithKeywords ()
#2  0x000000000049bd84 in PyEval_CallFunction ()
#3  0x00007ffff5375d9f in boost::python::call<boost::python::api::object, int>
(callable=0x7ffff7ed4578, a0=@0x7ffff3c35b34: 5)
at /usr/local/boost_1_55_0/boost/python/call.hpp:66
#4  0x00007ffff5374b81 in boost::python::api::object_operators<boost::python::api::object>::operator()<int> (this=0x9e3bf0, a0=@0x7ffff3c35b34: 5)
at /usr/local/boost_1_55_0/boost/python/object_call.hpp:19
#5  0x00007ffff5373658 in boost::detail::function::void_function_obj_invoker1<boost::python::api::object, void, int>::invoke (function_obj_ptr=..., a0=5)
at /usr/local/boost_1_55_0/boost/function/function_template.hpp:153
#6  0x00007ffff5378a3c in boost::function1<void, int>::operator() (
this=0x9e3be8, a0=5)
at /usr/local/boost_1_55_0/boost/function/function_template.hpp:767
#7  0x00007ffff53781f9 in boost::signals2::detail::call_with_tuple_args<boost::signals2::detail::void_type>::m_invoke<boost::function<void (int)>, 0u, int&>(void*, boost::function<void (int)>&, boost::signals2::detail::unsigned_meta_array<0u>, std::tuple<int&>) const (this=0x7ffff3c35c7f, func=..., args=...)
at /usr/local/boost_1_55_0/boost/signals2/detail/variadic_slot_invoker.hpp:92

有什么想法吗？

最佳答案

如果 MyClass::some_later_event() 是从未明确管理 Global Interpreter Lock 的 C++ 线程调用的(GIL)，那么这可能会导致未定义的行为。

Python 和 C++ 线程。

让我们考虑 C++ 线程与 Python 交互的情况。例如，可以通过 MyClass.event_in(seconds, value) 将 C++ 线程设置为在一段时间后调用 MyClass 的信号。

这个例子可能会变得相当复杂，所以让我们从基础开始:Python 的 GIL。简而言之，GIL 是解释器周围的互斥量。如果线程正在做任何影响 python 托管对象的引用计数的事情，那么它需要获取 GIL。在 GDB 回溯中，Boost.Signals2 库可能试图在没有 GIL 的情况下调用 Python 对象，从而导致崩溃。虽然管理 GIL 相当简单，但它很快就会变得复杂。

首先，该模块需要让 Python 初始化 GIL 以进行线程处理。

BOOST_PYTHON_MODULE(example)
{
  PyEval_InitThreads(); // Initialize GIL to support non-python threads.
  // ...
}

为了方便起见，让我们创建一个简单的类来帮助通过作用域管理 GIL:

/// @brief RAII class used to lock and unlock the GIL.
class gil_lock
{
public:
  gil_lock()  { state_ = PyGILState_Ensure(); }
  ~gil_lock() { PyGILState_Release(state_);   }
private:
  PyGILState_STATE state_;
};

让我们确定 C++ 线程何时需要 GIL:

boost::signals2::signal 可以制作连接对象的额外拷贝，就像信号为 concurrently 时所做的那样。调用。
调用通过 boost::signals2::signal 连接的 Python 对象。回调肯定会影响 python 对象。例如，提供给 __call__ 方法的 self 参数将增加和减少对象的引用计数。

`MyClass` 类。

这是一个基于原始代码的基本模型类:

/// @brief Mockup class.
class MyClass
{
public:
  /// @brief Connect a slot to the signal.
  template <typename Slot>
  void connect_slot(const Slot& slot)
  {
    signal_.connect(slot);
  }

  /// @brief Send an event to the signal.
  void event(int value)
  {
    signal_(value);
  }

private:
  boost::signals2::signal<void(int)> signal_;
};

由于 C++ 线程可能会调用 MyClass 的信号，因此 MyClass 的生命周期必须至少与线程一样长。一个很好的实现方法是让 Boost.Python 使用 boost::shared_ptr 管理 MyClass。

BOOST_PYTHON_MODULE(example)
{
  PyEval_InitThreads(); // Initialize GIL to support non-python threads.

  namespace python = boost::python;
  python::class_<MyClass, boost::shared_ptr<MyClass>,
                 boost::noncopyable>("MyClass")
    .def("event", &MyClass::event)
    // ...
    ;
}

`boost::signals2::signal` 与 python 对象交互。

boost::signals2::signal 可以在调用时进行复制。此外，可能有 C++ 插槽连接到信号，因此最好只在调用 Python 插槽时锁定 GIL。但是，signal 不提供钩子(Hook)来让我们在创建插槽拷贝或调用插槽之前获取 GIL。

为了避免 signal 创建 boost::python::object 槽的拷贝，可以使用创建 boost 拷贝的包装类::python::object 以便引用计数保持准确，并通过 shared_ptr 管理拷贝。这允许 signal 自由创建 shared_ptr 的拷贝，而不是在没有 GIL 的情况下复制 boost::python::object。

这个 GIL 安全槽可以封装在辅助类中。

/// @brief Helepr type that will manage the GIL for a python slot.
///
/// @detail GIL management:
///           * Caller must own GIL when constructing py_slot, as 
///             the python::object will be copy-constructed (increment
///             reference to the object)
///           * The newly constructed python::object will be managed
///             by a shared_ptr.  Thus, it may be copied without owning
///             the GIL.  However, a custom deleter will acquire the
///             GIL during deletion.
///           * When py_slot is invoked (operator()), it will acquire
///             the GIL then delegate to the managed python::object.
struct py_slot
{
public:

  /// @brief Constructor that assumes the caller has the GIL locked.
  py_slot(const boost::python::object& object)
    : object_(
        new boost::python::object(object),  // GIL locked, so copy.
        [](boost::python::object* object)   // Delete needs GIL.
        {
          gil_lock lock;
          delete object;
        }
      )
  {}

  // Use default copy-constructor and assignment-operator.
  py_slot(const py_slot&) = default;
  py_slot& operator=(const py_slot&) = default;

  template <typename ...Args>
  void operator()(Args... args)
  {
    // Lock the GIL as the python object is going to be invoked.
    gil_lock lock;
    (*object_)(args...); 
  }

private:
  boost::shared_ptr<boost::python::object> object_;
};

辅助函数将暴露给 Python 以帮助调整类型。

/// @brief MyClass::connect_slot helper.
template <typename ...Args>
void MyClass_connect_slot(
  MyClass& self,
  boost::python::object object)
{
  py_slot slot(object); // Adapt object to a py_slot for GIL management.

  // Using a lambda here allows for the args to be expanded automatically.
  // If bind was used, the placeholders would need to be explicitly added.
  self.connect_slot([slot](Args... args) mutable { slot(args...); });
}

更新后的绑定(bind)公开了辅助函数:

python::class_<MyClass, boost::shared_ptr<MyClass>,
               boost::noncopyable>("MyClass")
  .def("connect_slot", &MyClass_connect_slot<int>)
  .def("event",        &MyClass::event)
  // ...
  ;

线程本身。

线程的功能是相当基本的:它休眠然后调用信号。但是，了解 GIL 的上下文很重要。

/// @brief Sleep then invoke an event on MyClass.
template <typename ...Args>
void MyClass_event_in_thread(
  boost::shared_ptr<MyClass> self,
  unsigned int seconds,
  Args... args)
{
  // Sleep without the GIl.
  std::this_thread::sleep_for(std::chrono::seconds(seconds));

  // We do not want to hold the GIL while invoking or copying 
  // C++-specific slots connected to the signal.  Thus, it is the 
  // responsibility of python slots to manage the GIL via the 
  // py_slot wrapper class.
  self->event(args...);
}

/// @brief Function that will be exposed to python that will create
///        a thread to call the signal.
template <typename ...Args>
void MyClass_event_in(
  boost::shared_ptr<MyClass> self,
  unsigned int seconds,
  Args... args)
{
  // The caller may or may not have the GIL.  Regardless, spawn off a 
  // thread that will sleep and then invoke an event on MyClass.  The
  // thread will not be joined so detach from it.  Additionally, as
  // shared_ptr is thread safe, copies of it can be made without the
  // GIL.
  std::thread(&MyClass_event_in_thread<Args...>, self, seconds, args...)
      .detach();
}

_{请注意，MyClass_event_in_thread 可以表示为 lambda，但在 lambda 中解压模板包在某些编译器上不起作用。}

并且更新了 MyClass 绑定(bind)。

python::class_<MyClass, boost::shared_ptr<MyClass>,
               boost::noncopyable>("MyClass")
  .def("connect_slot", &MyClass_connect_slot<int>)
  .def("event",        &MyClass::event)
  .def("event_in",     &MyClass_event_in<int>)
  ;

最终的解决方案是这样的:

#include <thread> // std::thread, std::chrono
#include <boost/python.hpp>
#include <boost/shared_ptr.hpp>
#include <boost/signals2/signal.hpp>

/// @brief Mockup class.
class MyClass
{
public:
  /// @brief Connect a slot to the signal.
  template <typename Slot>
  void connect_slot(const Slot& slot)
  {
    signal_.connect(slot);
  }

  /// @brief Send an event to the signal.
  void event(int value)
  {
    signal_(value);
  }

private:
  boost::signals2::signal<void(int)> signal_;
};

/// @brief RAII class used to lock and unlock the GIL.
class gil_lock
{
public:
  gil_lock()  { state_ = PyGILState_Ensure(); }
  ~gil_lock() { PyGILState_Release(state_);   }
private:
  PyGILState_STATE state_;
};    

/// @brief Helepr type that will manage the GIL for a python slot.
///
/// @detail GIL management:
///           * Caller must own GIL when constructing py_slot, as 
///             the python::object will be copy-constructed (increment
///             reference to the object)
///           * The newly constructed python::object will be managed
///             by a shared_ptr.  Thus, it may be copied without owning
///             the GIL.  However, a custom deleter will acquire the
///             GIL during deletion.
///           * When py_slot is invoked (operator()), it will acquire
///             the GIL then delegate to the managed python::object.
struct py_slot
{
public:

  /// @brief Constructor that assumes the caller has the GIL locked.
  py_slot(const boost::python::object& object)
    : object_(
        new boost::python::object(object),  // GIL locked, so copy.
        [](boost::python::object* object)   // Delete needs GIL.
        {
          gil_lock lock;
          delete object;
        }
      )
  {}

  // Use default copy-constructor and assignment-operator.
  py_slot(const py_slot&) = default;
  py_slot& operator=(const py_slot&) = default;

  template <typename ...Args>
  void operator()(Args... args)
  {
    // Lock the GIL as the python object is going to be invoked.
    gil_lock lock;
    (*object_)(args...); 
  }

private:
  boost::shared_ptr<boost::python::object> object_;
};

/// @brief MyClass::connect_slot helper.
template <typename ...Args>
void MyClass_connect_slot(
  MyClass& self,
  boost::python::object object)
{
  py_slot slot(object); // Adapt object to a py_slot for GIL management.

  // Using a lambda here allows for the args to be expanded automatically.
  // If bind was used, the placeholders would need to be explicitly added.
  self.connect_slot([slot](Args... args) mutable { slot(args...); });
}

/// @brief Sleep then invoke an event on MyClass.
template <typename ...Args>
void MyClass_event_in_thread(
  boost::shared_ptr<MyClass> self,
  unsigned int seconds,
  Args... args)
{
  // Sleep without the GIL.
  std::this_thread::sleep_for(std::chrono::seconds(seconds));

  // We do not want to hold the GIL while invoking or copying 
  // C++-specific slots connected to the signal.  Thus, it is the 
  // responsibility of python slots to manage the GIL via the 
  // py_slot wrapper class.
  self->event(args...);
}

/// @brief Function that will be exposed to python that will create
///        a thread to call the signal.
template <typename ...Args>
void MyClass_event_in(
  boost::shared_ptr<MyClass> self,
  unsigned int seconds,
  Args... args)
{
  // The caller may or may not have the GIL.  Regardless, spawn off a 
  // thread that will sleep and then invoke an event on MyClass.  The
  // thread will not be joined so detach from it.  Additionally, as
  // shared_ptr is thread safe, copies of it can be made without the
  // GIL.
  // Note: MyClass_event_in_thread could be expressed as a lambda,
  //       but unpacking a template pack within a lambda does not work
  //       on some compilers.
  std::thread(&MyClass_event_in_thread<Args...>, self, seconds, args...)
      .detach();
}

BOOST_PYTHON_MODULE(example)
{
  PyEval_InitThreads(); // Initialize GIL to support non-python threads.

  namespace python = boost::python;
  python::class_<MyClass, boost::shared_ptr<MyClass>,
                 boost::noncopyable>("MyClass")
    .def("connect_slot", &MyClass_connect_slot<int>)
    .def("event",        &MyClass::event)
    .def("event_in",     &MyClass_event_in<int>)
    ;
}

还有一个测试脚本:

from time import sleep
import example

def spam1(x):
  print "spam1: ", x

def spam2(x):
  print "spam2: ", x

c = example.MyClass()
c.connect_slot(spam1)
c.connect_slot(spam2)
c.event(123)
print "Sleeping"
c.event_in(3, 321)
sleep(5)
print "Done sleeping"

结果如下:

spam1:  123
spam2:  123
Sleeping
spam1:  321
spam2:  321
Done sleeping

关于c++ - Boost.Python 和 Boost.Signals2 : Segmentation faults，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/20799422/

c++ - Boost.Python 和 Boost.Signals2 : Segmentation faults

Python 和 C++ 线程。

`MyClass` 类。

`boost::signals2::signal` 与 python 对象交互。

线程本身。

上一篇：c++ - 如何在 C++ 中访问动态分配的矩阵？

下一篇：c++ - 将指针分配给临时对象——如何防止呢？

c++ - Boost.Python 和 Boost.Signals2 : Segmentation faults

Python 和 C++ 线程。

MyClass 类。

boost::signals2::signal 与 python 对象交互。

线程本身。

上一篇：c++ - 如何在 C++ 中访问动态分配的矩阵？

下一篇：c++ - 将指针分配给临时对象——如何防止呢？

`MyClass` 类。

`boost::signals2::signal` 与 python 对象交互。