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/