rpavlik / luabind Goto Github PK

Hello, I'm having some trouble std::shared_ptr smart-pointer support in luabind. It seems luabind uses luabind::detail::has_get_pointer to figure out if a type is a smart-pointer, and this does not work natively with std::shared_ptr (since that does not define a get_pointer function, at least not in VC2010). Prior to boost version 1.53.0, I worked around that by defining get_pointer in the namespaces of my pointee objects (since it isn't allowed to add anything to std::). Since boost 1.53.0 however, this no longer compiles. Boost now defines get_pointer in boost:: for std::shared_ptr, and it seems like luabind::detail::has_get_pointer either finds that one and the implementation in my pointee's namespace or none at all. In the first case it doesn't compile, in the second, if I remove my implementation, luabind seems to treat the shared_ptr as a base class which isn't registered and gives me a runtime-error.
The only fix I found was to add a partial specialization like this:

template<class T>
struct has_get_pointer<std::shared_ptr<T>> : boost::mpl::true_ {};

Is this a proper fix, or am I misunderstanding something?

luabind call_function throw exception on destructor, but on C++ 11 this is default no throw.So you cannot catch exception.

Hi,

I'd like to have the ability to do the following, ideally:

1. call c++ from lua. This seems not so hard, eg this library probably does that. So does swig
2. override c++ classes from lua, and have c++ call back into those. Swig doesnt do this. I heard that luabind does?
3. not have to install boost. it's easy to install on linux, not so easy on windows
4. be able to build using only c++0x and visual studio 2010. visual studio 2010 is the current version used by python, that I'm also linking with. c++0x is still the highest standard in many commonly available gcc versions

After compiling this version on Windows 8.1 VS2013, x64, I got seemingly working luabind.

However, 2 or 3 tests does fail. Maybe you will be able to locate the problems?
I use the library only to compile existing source code like OSRM, but can do any testing on my system if needed.

Boost 1.55, Lua 5.2.3 from https://github.com/LuaDist/lua , built with cmake in Visual Studio 2013 x64 command prompt.

Here is the detailed test log (ctest -V):
https://gist.github.com/alex85k/10016791#file-luabind_failing_tests-log

Seen as part of this commit: ThomasNelson@1ca4157

Wrapping the invoke-Call, including the try-catch block, inside the entry_point-method of function_object_impl with another not-inlined wrapper-function, seems to fix exception issues encountered on x86 builds for win32.

I was wondering whether the current development of luabind is still supporting LuaJIT? As of version 2.0.2, it provides Lua 5.1 headers and library.

The branch build_improvements has a file FindLua52.cmake which suggests that support for 5.1 has been dropped?

I have added LuaJIT to the CMake module (see https://github.com/fhoefling/luaponte/tree/build-improvements). It compiles with Clang 3.3, but all test except for one fail with a segfault.

Hi,

Is this project abandoned?
The link on front page is 404, and last commit was 2015..

I think the container policy misses an overload to match for by_value
and more importantly, the loop that fetches the table elements from the stack has the table at position -2 (ie. index-1). I fixed that in my fork, however I don't know if this only applies to my reworked policy system.

The policy "out_value" seems to be broken (for a long time?), it is also not covered by the tests (test_policies.cpp has "pure_out_value" only). The following example is copied from the luabind documentation, GCC and Clang generate an error about calling a non-static member function:

.../luabind/out_value_policy.hpp:126:71: error: cannot call member function ‘int luabind::native_converter_base::match(lua_State*, luabind::detail::by_value, int) [with T = float; Derived = luabind::default_converter; lua_State = lua_State]’ without object
             return converter::match(L, LUABIND_DECORATE_TYPE(T), index);

Example:

#include "test.hpp"

#include 
#include 

void f(float& val) { val = val + 10.f; }

void test_main(lua_State* L)
{
    using namespace luabind;

    module(L)
    [
         def("f", &f, out_value(_1)) // compiles with "pure_out_value"
    ];

    // out_value
    DOSTRING(L,
        "a = f(5)\n"
        "assert(a == 15)");
}

"error LNK2001: unresolved external symbol "private: static unsigned int const luabind::detail::class_id_map::local_id_base" (?local_id_base@class_id_map@detail@luabind@@0ib) "

I get this when trying to compile my project using Luabind. How do I fix this? =S

Thanks!

Luabind causes a shared_ptr corruption when you do this in lua:

local gSharedObject = nil;
function Testfunction()
    gSharedObject = SharedObject.Create();
    return gSharedObject;
end

where the C++ export looks like that:

class SharedObject {
public:
    SharedObject() {}

    static boost::shared_ptr<SharedObject> Create() {
        return boost::make_shared<SharedObject>();
    }
}

luabind::module(L) [
    luabind::class_<SharedObject, boost::shared_ptr<SharedObject>>("SharedObject")
        .scope [ luabind::def("Create", &SharedObject::Create) ]
];

Now when you do the following you will get a use_count == 1 where it should actually be 2 because lua should hold a shared_ptr and your code should:

boost::shared_ptr<SharedObject> obj = luabind::call_function<boost::shared_ptr<SharedObject>>(L, "Testfunction");
std::cout << obj.use_count() << std::endl;
// prints 1 but should actually be 2, one from the current scope and one in lua globals, because the lua state is still alive... ?!?

This is a huge problem when your GC runs because it always thinks the shared_ptr is the only one in use and it will always release objects created like that, not matter if a ptr is used by the c++ code atm.

Example:

boost::shared_ptr<SharedObject> obj = luabind::call_function<boost::shared_ptr<SharedObject>>(L, "Testfunction");
std::cout << obj.use_count() << std::endl;

lua_close(L); // invokes ~SharedObject -_-

std::cout << obj.use_count() << std::endl; // still 1

The problem exists in lua 5.1 and 5.2, everywhere I checked

Is it planned to bring up future releases of luabind, bundeling the progress of all the recent development efforts? And if so, what is preventing the next release?

I would like to base my other software on luabind again (and abandon the luaponte fork). Some kind of stable release would be very helpful here.

BTW, a reasonable versioning policy call "semantic versioning" has been suggested here: http://semver.org/

I think the username (rpavlik) set in the submodule for jenkins-ctest-plugin is requiring authentication and login through Github. If this was switched over to using just using the plain HTTP(S) method, it might fix this.

https://github.com/rpavlik/luabind.git

https://github.com/fhoefling/luaponte

Hello everyone.

I have build luabind on Window 10 with bjam, toolset msvc, 32bit latest VC++, boost 1.59.0 and Lua 5.2. When running the following code:

#include <lua.hpp>
#include <luabind/luabind.hpp>

int main()
{
    lua_State* L = luaL_newstate();

    try
    {
        luabind::open(L);
    }
    catch(std::runtime_error& e)
    {
        printf("%s\n", e.what());
    }

    return 0;
}

I get the following error:

luabind::open() must be called with the main thread lua_State*

luabind::open() calls the following function in open.cpp, line 115:

  LUABIND_API void open(lua_State* L)
    {
        bool is_main_thread = lua_pushthread(L) == 1;
        lua_pop(L, 1);

        if (!is_main_thread)
        {
            throw std::runtime_error(
                "luabind::open() must be called with the main thread "
                "lua_State*"
            );
        }

        createGarbageCollectedRegistryUserdata<detail::class_registry>(L, "__luabind_classes", L);
        createGarbageCollectedRegistryUserdata<detail::class_id_map>(L, "__luabind_class_id_map");
        createGarbageCollectedRegistryUserdata<detail::cast_graph>(L, "__luabind_cast_graph");
        createGarbageCollectedRegistryUserdata<detail::class_map>(L, "__luabind_class_map");

        // add functions (class, cast etc...)
        lua_pushcclosure(L, detail::create_class::stage1, 0);
        lua_setglobal(L, "class");

        lua_pushcclosure(L, &make_property, 0);
        lua_setglobal(L, "property");

        lua_pushlightuserdata(L, &main_thread_tag);
        lua_pushlightuserdata(L, L);
        lua_rawset(L, LUA_REGISTRYINDEX);

        lua_pushcclosure(L, &deprecated_super, 0);
        lua_setglobal(L, "super");

        //set_package_preload(L, "luabind.function_introspection", &bind_function_introspection);
    }

Might there be an issue with the following line?

bool is_main_thread = lua_pushthread(L) == 1;

I can't find anything on Google at all, that is why I am asking it here.

this issue is probably related to the way homebrew makes lua51 and lua(52) live together.

So the cmake-bundled FindLua51.cmake finds lua.h and (incorrectly) sets LUA_INCLUDE_DIR to /usr/local/include, and LUA_LIBRARY to /usr/local/lib/liblua5.1.dylib. This produces all sorts of Undefined symbols for architecture... during the build phase.

Given that homebrew is (one of) Mac's preferred package manager(s) this may be an issue for lots of Mac users. Until the solution (likely making a custom FindLua51) is found, maybe we can generate a warning if both versions are found on a Mac?

Trying to convert from luabind(the boost version) to this one, getting error at startup.

const char* path = "..\code\script?.lua;"
luabind::call_function(_State, "set_path",path);

It seems to be trying to find the class_id for "const char_", not finding anything, and then reporting that const char_ is not a registered class.

" Trying to use unregistered class: char const *"

Trying to build with clang 3.2
I am getting this error message

In file included from ../test/test_exception_handlers.cpp:7:
In file included from ../luabind/luabind.hpp:27:
In file included from ../luabind/class.hpp:89:
../luabind/function.hpp:28:43: error: no matching function for call to 'deduce_signature'
          object fn = make_function(L, f, deduce_signature(f), policies);
                                          ^~~~~~~~~~~~~~~~
../luabind/function.hpp:48:13: note: in instantiation of member function 'luabind::detail::function_registration<void (*)() __attribute__((noreturn)), luabind::detail::null_type>::register_' requested here
        new detail::function_registration<F, Policies>(name, f, policies)));
            ^
../luabind/function.hpp:54:12: note: in instantiation of function template specialization 'luabind::def<void (*)() __attribute__((noreturn)), luabind::detail::null_type>' requested here
    return def(name, f, detail::null_type());
           ^
../test/test_exception_handlers.cpp:50:9: note: in instantiation of function template specialization 'luabind::def<void (*)() __attribute__((noreturn))>' requested here
        def("raise", &raise_my_exception),
        ^
../luabind/detail/deduce_signature.hpp:47:5: note: candidate template ignored: disabled by 'enable_if' [with F = void (*)() __attribute__((noreturn))]
    is_function<F>,
    ^
../luabind/detail/deduce_signature.hpp:60:1: note: candidate function template not viable: requires 2 arguments, but 1 was provided
deduce_signature(F, Wrapped*)
^
1 error generated.

Hi,

There are numerous forks from the luabind repository. An I do believe this one seems to be the reference from other development done around.

Who is in charge now of taking care of pull requests, releases, etc?
Is there anybody left on the boat? :)

Would there be a possibility to gather the multiple efforts done there?

Best regards,