Prerequisites¶

Apart from Lua 5.1 or (recommended) 5.2, Luabind depends on a number of Boost libraries. It also depends on CMake to build the library and run the tests.

Windows¶

If CMake has problems finding Lua, LUA_DIR needs to be set to point to a directory containing the Lua include directory and built libraries.

The same applies to Boost and the BOOST_ROOT environment variable.

Linux and other *nix flavors¶

If your system already has Lua installed, it is very likely that the build system will automatically find it and just work. If you have Lua installed in a non-standard location, you may need to set LUA_DIR to point to the installation prefix.

BOOST_ROOT can be set to a Boost installation directory. If left unset, the build system will try to use boost headers from the standard include path.

Building and testing¶

Building the default variant of the library, which is a static release library, is simply done by invoking cmake in the luabind root directory and then running your native build tools on the generated files (e.g. make on Unix or nmake/msbuild ALL_BUILD.vcxproj for MSVC).

NOTE:

To run the unit tests, invoke your build tool with the test target, e.g. make test on Unix or nmake test/msbuild RUN_TESTS.vcxproj for MSVC. This run the (previously built) unit tests in the current variant. A clean test run output should end with something like:

100% tests passed, 0 tests failed out of 51
Total Test time (real) =   1.23 sec

A failed run would end with something like:

98% tests passed, 1 tests failed out of 51
Total Test time (real) =   1.23 sec
The following tests FAILED:


    1 - abstract_base (Failed)

Hello world¶

#include <iostream>
#include <luabind/luabind.hpp>
void greet()
{


    std::cout << "hello world!\n";
}
extern "C" int init(lua_State* L)
{


    using namespace luabind;


    open(L);


    module(L)


    [


        def("greet", &greet)


    ];


    return 0;
}

Lua 5.0  Copyright (C) 1994-2003 Tecgraf, PUC-Rio
> loadlib('hello_world.dll', 'init')()
> greet()
Hello world!
>

Overloaded functions¶

If you have more than one function with the same name, and want to register them in Lua, you have to explicitly give the signature. This is to let C++ know which function you refer to. For example, if you have two functions, int f(const char*) and void f(int).

module(L)
[


    def("f", (int(*)(const char*)) &f),


    def("f", (void(*)(int)) &f)
];

Signature matching¶

luabind will generate code that checks the Lua stack to see if the values there can match your functions' signatures. It will handle implicit typecasts between derived classes, and it will prefer matches with the least number of implicit casts. In a function call, if the function is overloaded and there's no overload that match the parameters better than the other, you have an ambiguity. This will spawn a run-time error, stating that the function call is ambiguous. A simple example of this is to register one function that takes an int and one that takes a float. Since Lua doesn't distinguish between floats and integers, both will always match.

Since all overloads are tested, it will always find the best match (not the first match). This also means that it can handle situations where the only difference in the signature is that one member function is const and the other isn't.

Ownership transfer¶

To correctly handle ownership transfer, create_a() would need an adopt return value policy. More on this in the part-policies section.

For example, if the following function and class is registered:

struct A
{


    void f();


    void f() const;
};
const A* create_a();
struct B: A {};
struct C: B {};
void g(A*);
void g(B*);

And the following Lua code is executed:

a1 = create_a()
a1:f() -- the const version is called
a2 = A()
a2:f() -- the non-const version is called
a = A()
b = B()
c = C()
g(a) -- calls g(A*)
g(b) -- calls g(B*)
g(c) -- calls g(B*)

Calling Lua functions¶

To call a Lua function, you can either use call_function() or an object.

template<class Ret>
Ret call_function(lua_State* L, const char* name, ...)
template<class Ret>
Ret call_function(object const& obj, ...)

There are two overloads of the call_function function, one that calls a function given its name, and one that takes an object that should be a Lua value that can be called as a function.

The overload that takes a name can only call global Lua functions. The ... represents a variable number of parameters that are sent to the Lua function. This function call may throw luabind::error if the function call fails.

The return value isn't actually Ret (the template parameter), but a proxy object that will do the function call. This enables you to give policies to the call. You do this with the operator[]. You give the policies within the brackets, like this:

int ret = call_function<int>(


    L


  , "a_lua_function"


  , new complex_class()
)[ adopt(_1) ];

If you want to pass a parameter as a reference, you have to wrap it with the Boost.Ref.

Like this:

int ret = call_function(L, "fun", boost::ref(val));

If you want to use a custom error handler for the function call, see set_pcall_callback under sec-pcall-errorfunc.

Using Lua threads¶

To start a Lua thread, you have to call lua_resume(), this means that you cannot use the previous function call_function() to start a thread. You have to use

template<class Ret>
Ret resume_function(lua_State* L, const char* name, ...)
template<class Ret>
Ret resume_function(object const& obj, ...)

and

template<class Ret>
Ret resume(lua_State* L, ...)

The first time you start the thread, you have to give it a function to execute. i.e. you have to use resume_function, when the Lua function yields, it will return the first value passed in to lua_yield(). When you want to continue the execution, you just call resume() on your lua_State, since it's already executing a function, you don't pass it one. The parameters to resume() will be returned by yield() on the Lua side.

For yielding C++-functions (without the support of passing data back and forth between the Lua side and the c++ side), you can use the policy-yield policy.

With the overload of resume_function that takes an part-object, it is important that the object was constructed with the thread as its lua_State*. Like this:

lua_State* thread = lua_newthread(L);
object fun = get_global(thread)["my_thread_fun"];
resume_function(fun);

Binding function objects with explicit signatures¶

Using luabind::tag_function<> it is possible to export function objects from which luabind can't automatically deduce a signature. This can be used to slightly alter the signature of a bound function, or even to bind stateful function objects.

Synopsis:

template <class Signature, class F>
implementation-defined tag_function(F f);

Where Signature is a function type describing the signature of F. It can be used like this:

int f(int x);
// alter the signature so that the return value is ignored
def("f", tag_function<void(int)>(f));
struct plus
{


    plus(int x)


      : x(x)


    {}


    int operator()(int y) const


    {


        return x + y;


    }
};
// bind a stateful function object
def("plus3", tag_function<int(int)>(plus(3)));

Overloaded member functions¶

When binding more than one overloads of a member function, or just binding one overload of an overloaded member function, you have to disambiguate the member function pointer you pass to def. To do this, you can use an ordinary C-style cast, to cast it to the right overload. To do this, you have to know how to express member function types in C++, here's a short tutorial (for more info, refer to your favorite book on C++).

The syntax for member function pointer follows:

return-value (class-name::*)(arg1-type, arg2-type, ...)

Here's an example illlustrating this:

struct A
{


    void f(int);


    void f(int, int);
};

class_<A>()


    .def("f", (void(A::*)(int))&A::f)

This selects the first overload of the function f to bind. The second overload is not bound.

Properties¶

To register a global data member with a class is easily done. Consider the following class:

struct A
{


    int a;
};

This class is registered like this:

module(L)
[


    class_<A>("A")


        .def_readwrite("a", &A::a)
];

This gives read and write access to the member variable A::a. It is also possible to register attributes with read-only access:

module(L)
[


    class_<A>("A")


        .def_readonly("a", &A::a)
];

When binding members that are a non-primitive type, the auto generated getter function will return a reference to it. This is to allow chained .-operators. For example, when having a struct containing another struct. Like this:

struct A { int m; };
struct B { A a; };

When binding B to lua, the following expression code should work:

b = B()
b.a.m = 1
assert(b.a.m == 1)

This requires the first lookup (on a) to return a reference to A, and not a copy. In that case, luabind will automatically use the dependency policy to make the return value dependent on the object in which it is stored. So, if the returned reference lives longer than all references to the object (b in this case) it will keep the object alive, to avoid being a dangling pointer.

You can also register getter and setter functions and make them look as if they were a public data member. Consider the following class:

class A
{
public:


    void set_a(int x) { a = x; }


    int get_a() const { return a; }
private:


    int a;
};

It can be registered as if it had a public data member a like this:

class_<A>("A")


    .property("a", &A::get_a, &A::set_a)

This way the get_a() and set_a() functions will be called instead of just writing to the data member. If you want to make it read only you can just omit the last parameter. Please note that the get function has to be const, otherwise it won't compile. This seems to be a common source of errors.

Enums¶

If your class contains enumerated constants (enums), you can register them as well to make them available in Lua. Note that they will not be type safe, all enums are integers in Lua, and all functions that takes an enum, will accept any integer. You register them like this:

module(L)
[


    class_<A>("A")


        .enum_("constants")


        [


            value("my_enum", 4),


            value("my_2nd_enum", 7),


            value("another_enum", 6)


        ]
];

In Lua they are accessed like any data member, except that they are read-only and reached on the class itself rather than on an instance of the class.

Lua 5.0  Copyright (C) 1994-2003 Tecgraf, PUC-Rio
> print(A.my_enum)
4
> print(A.another_enum)
6

Operators¶

To bind operators you have to include <luabind/operator.hpp>.

The mechanism for registering operators on your class is pretty simple. You use a global name luabind::self to refer to the class itself and then you just write the operator expression inside the def() call. This class:

struct vec
{


    vec operator+(int s);
};

Is registered like this:

module(L)
[


    class_<vec>("vec")


        .def(self + int())
];

This will work regardless if your plus operator is defined inside your class or as a free function.

If your operator is const (or, when defined as a free function, takes a const reference to the class itself) you have to use const_self instead of self. Like this:

module(L)
[


    class_<vec>("vec")


        .def(const_self + int())
];

The operators supported are those available in Lua:

+    -    *    /    ==    <    <=    %

This means, no in-place operators.

Default implementations (described below) are provided for == and the special __tostring pseudo-operator. If any other operator is called, Luabind will trigger an error ("[const] class <type>: no <metamethod name> defined.", e.g. "class vec: no __div operator defined.").

The equality operator (==) has a little hitch; it will not be called if the references are equal. This means that the == operator has to do pretty much what's it's expected to do.

For Luabind's default == operator, two objects are equal only if they are both objects of Luabind-exported classes and have the same addresses, after casting both to a common base if necessary. If they do not have a common base (and are not of the same type), they compare unequal.

Lua does not support operators such as !=, > or >=. That's why you can only register the operators listed above. When you invoke one of the mentioned operators, lua will define it in terms of one of the available operators.

In the above example the other operand type is instantiated by writing int(). If the operand type is a complex type that cannot easily be instantiated you can wrap the type in a class called other<>. For example:

To register this class, we don't want to instantiate a string just to register the operator.

struct vec
{


    vec operator+(std::string);
};

Instead we use the other<> wrapper like this:

module(L)
[


    class_<vec>("vec")


        .def(self + other<std::string>())
];

To register an application (function call-) operator:

module(L)
[


    class_<vec>("vec")


        .def( self(int()) )
];

There's one special operator. In Lua it's called __tostring, it's not really an operator. It is used for converting objects to strings in a standard way in Lua. If you register this functionality, you will be able to use the lua standard function tostring() for converting your object to a string.

To implement this operator in C++ you should supply an operator<< for std::ostream. Like this example:

class number {};
std::ostream& operator<<(std::ostream&, number&);
...
module(L)
[


    class_<number>("number")


        .def(tostring(self))
];

If you do not define a __tostring operator, Luabind supplies a default which result in strings of the form [const] <type> object: <address>, i.e. const is prepended if the object is const, <type> will be the string you supplied to class_ (or a string derived from std::type_info::name for unnamed classes) and <address> will be the address of the C++ object. (Note that in multiple inheritance scenarios where the same object is pushed as multiple different base types, the addresses returned for the same most-derived object will differ).

Nested scopes and static functions¶

It is possible to add nested scopes to a class. This is useful when you need to wrap a nested class, or a static function.

class_<foo>("foo")


    .def(constructor<>())


    .scope


    [


        class_<inner>("nested"),


        def("f", &f)


    ];

In this example, f will behave like a static member function of the class foo, and the class nested will behave like a nested class of foo.

It's also possible to add namespaces to classes using the same syntax.

Derived classes¶

If you want to register classes that derives from other classes, you can specify a template parameter bases<> to the class_ instantiation. The following hierarchy:

struct A {};
struct B : A {};

Would be registered like this:

module(L)
[


    class_<A>("A"),


    class_<B, A>("B")
];

If you have multiple inheritance you can specify more than one base. If B would also derive from a class C, it would be registered like this:

module(L)
[


    class_<B, bases<A, C> >("B")
];

Note that you can omit bases<> when using single inheritance.

NOTE:

Smart pointers¶

When registering a class you can tell luabind to hold all instances explicitly created in Lua in a specific smart pointer type, rather than the default raw pointer. This is done by passing an additional template parameter to class_:

class_<X, P>(…)

Where the requirements of P are:

where:

get_pointer() overloads are provided for the smart pointers in Boost, and std::auto_ptr<>. Should you need to provide your own overload, note that it is called unqualified and is expected to be found by argument dependent lookup. Thus it should be defined in the same namespace as the pointer type it operates on.

For example:

class_<X, boost::scoped_ptr<X> >("X")


  .def(constructor<>())

Will cause luabind to hold any instance created on the Lua side in a boost::scoped_ptr<X>. Note that this doesn't mean all instances will be held by a boost::scoped_ptr<X>. If, for example, you register a function:

std::auto_ptr<X> make_X();

the instance returned by that will be held in std::auto_ptr<X>. This is handled automatically for all smart pointers that implement a get_pointer() overload.

IMPORTANT:

IMPORTANT:

bool is_non_null(std::auto_ptr<X> const& p)
{


    return p.get();
}
…
def("is_non_null", &is_non_null)

When registering a hierarchy of classes, where all instances are to be held by a smart pointer, all the classes should have the baseclass' holder type. Like this:

module(L)
[


    class_<base, boost::shared_ptr<base> >("base")


        .def(constructor<>()),


    class_<derived, base, boost::shared_ptr<base> >("derived")


        .def(constructor<>())
];

Internally, luabind will do the necessary conversions on the raw pointers, which are first extracted from the holder type.

This means that for Luabind a smart_ptr<derived> is not related to a smart_ptr<base>, but derived* and base* are, as are smart_ptr<derived> and base*. In other words, upcasting does not work for smart pointers as target types, but as source types.

Additional support for shared_ptr and intrusive_ptr¶

This limitation cannot be removed for all smart pointers in a generic way, but luabind has special support for

You should include the header(s) you need in the cpp files which register functions that accept the corresponding smart pointer types, to get automatic conversions from smart_ptr<X> to smart_ptr<Y>, whenever Luabind would convert X* to Y*, removing the limitation mentioned above.

However, the shared_ptr converters might not behave exactly as you would expect:

shared_ptr<RequestedT>(raw->shared_from_this(), raw)

1. is as you should have expected and 2. is special behavior introduced to avoid 3. when possible. If you cannot derive your (root) classes from enable_shared_from_this (which is the recommended way of providing a shared_from_this method) you must be careful not to close the lua_State when you still have a shared_ptr obtained by 3.

There are three functions provided to support this (in namespace luabind):

bool is_state_unreferenced(lua_State* L);
typedef void(*state_unreferenced_fun)(lua_State* L);
void set_state_unreferenced_callback(lua_State* L, state_unreferenced_fun cb);
state_unreferenced_fun get_state_unreferenced_callback(lua_State* L);

is_state_unreferenced will return false if closing L would make existing shared_ptrs dangling and true if it safe (in this respect) to call lua_close(L).

The cb argument passed to set_state_unreferenced_callback will be called whenever the return value of is_state_unreferenced(L) would change from false to true.

get_state_unreferenced_callback returns the current state_unreferenced_fun for L.

A typical use of this functions would be to replace

lua_close(L);

with

if (luabind::is_state_unreferenced(L))


    lua_close(L);
else


    luabind::set_state_unreferenced_callback(L, &lua_close);

(lua_close happens to be a valid state_unreferenced_fun.)

Unnamed classes¶

You can register unnamed classes by not passing a name to class_:

class_<X>()

This does not export the class object itself to Lua, meaning that constructors cannot be called and enums are only accessible via objects of this class' type.

This is useful e.g. for registering multiple instantiations of a class template, and construct a matching instance using a factory function, like boost::make_shared of for hiding intermediate classes in inheritance hierarchies.

Splitting class registrations¶

In some situations it may be desirable to split a registration of a class across different compilation units. Partly to save rebuild time when changing in one part of the binding, and in some cases compiler limits may force you to split it. To do this is very simple. Consider the following sample code:

void register_part1(class_<X>& x)
{


    x.def(/*...*/);
}
void register_part2(class_<X>& x)
{


    x.def(/*...*/);
}
void register_(lua_State* L)
{


    class_<X> x("x");


    register_part1(x);


    register_part2(x);


    module(L) [ x ];
}

Here, the class X is registered in two steps. The two functions register_part1 and register_part2 may be put in separate compilation units.

To separate the module registration and the classes to be registered, see part-split-registration.

Synopsis¶

class object
{
public:


    template<class T>


    object(lua_State*, T const& value);


    object(from_stack const&);


    object(object const&);


    object();


    ~object();


    lua_State* interpreter() const;


    void push() const;


    bool is_valid() const;


    operator safe_bool_type () const;


    template<class Key>


    implementation-defined operator[](Key const&);


    template<class T>


    object& operator=(T const&);


    object& operator=(object const&);


    bool operator==(object const&) const;


    bool operator<(object const&) const;


    bool operator<=(object const&) const;


    bool operator>(object const&) const;


    bool operator>=(object const&) const;


    bool operator!=(object const&) const;


    template <class T>


    implementation-defined operator[](T const& key) const


    void swap(object&);


    implementation-defined operator()();


    template<class A0>


    implementation-defined operator()(A0 const& a0);


    template<class A0, class A1>


    implementation-defined operator()(A0 const& a0, A1 const& a1);


    /* ... */
};

When you have a Lua object, you can assign it a new value with the assignment operator (=). When you do this, the default_policy will be used to make the conversion from C++ value to Lua. If your luabind::object is a table you can access its members through the operator[] or the Iterators. The value returned from the operator[] is a proxy object that can be used both for reading and writing values into the table (using operator=).

Note that it is impossible to know if a Lua value is indexable or not (lua_gettable doesn't fail, it succeeds or crashes). This means that if you're trying to index something that cannot be indexed, you're on your own. Lua will call its panic() function. See sec-lua-panic.

There are also free functions that can be used for indexing the table, see Related functions.

The constructor that takes a from_stack object is used when you want to initialize the object with a value from the lua stack. The from_stack type has the following constructor:

from_stack(lua_State* L, int index);

The index is an ordinary lua stack index, negative values are indexed from the top of the stack. You use it like this:

object o(from_stack(L, -1));

This will create the object o and copy the value from the top of the lua stack.

The interpreter() function returns the Lua state where this object is stored. If you want to manipulate the object with Lua functions directly you can push it onto the Lua stack by calling push().

The operator== will call lua_compare() on the operands and return its result.

The is_valid() function tells you whether the object has been initialized or not. When created with its default constructor, objects are invalid. To make an object valid, you can assign it a value. If you want to invalidate an object you can simply assign it an invalid object.

The operator safe_bool_type() is equivalent to is_valid(). This means that these snippets are equivalent:

object o;
// ...
if (o)
{


    // ...
}
...
object o;
// ...
if (o.is_valid())
{


    // ...
}

The application operator will call the value as if it was a function. You can give it any number of parameters (currently the default_policy will be used for the conversion). The returned object refers to the return value (currently only one return value is supported). This operator may throw luabind::error if the function call fails. If you want to specify policies to your function call, you can use index-operator (operator[]) on the function call, and give the policies within the [ and ]. Like this:

my_function_object(


    2


  , 8


  , new my_complex_structure(6)
) [ adopt(_3) ];

This tells luabind to make Lua adopt the ownership and responsibility for the pointer passed in to the lua-function.

It's important that all instances of object have been destructed by the time the Lua state is closed. The object will keep a pointer to the lua state and release its Lua object in its destructor.

Here's an example of how a function can use a table:

void my_function(object const& table)
{


    if (type(table) == LUA_TTABLE)


    {


        table["time"] = std::clock();


        table["name"] = std::rand() < 500 ? "unusual" : "usual";


        std::cout << object_cast<std::string>(table[5]) << "\n";


    }
}

If you take a luabind::object as a parameter to a function, any Lua value will match that parameter. That's why we have to make sure it's a table before we index into it.

std::ostream& operator<<(std::ostream&, object const&);

There's a stream operator that makes it possible to print objects or use boost::lexical_cast to convert it to a string. This will use lua's string conversion function. So if you convert a C++ object with a tostring operator, the stream operator for that type will be used.

Iterators¶

There are two kinds of iterators. The normal iterator that will use the metamethod of the object (if there is any) when the value is retrieved. This iterator is simply called luabind::iterator. The other iterator is called luabind::raw_iterator and will bypass the metamethod and give the true contents of the table. They have identical interfaces, which implements the ForwardIterator concept. Apart from the members of standard iterators, they have the following members and constructors:

class iterator
{


    iterator();


    iterator(object const&);


    object key() const;


    standard iterator members
};

The constructor that takes a luabind::object is actually a template that can be used with object. Passing an object as the parameter to the iterator will construct the iterator to refer to the first element in the object.

The default constructor will initialize the iterator to the one-past-end iterator. This is used to test for the end of the sequence.

The value type of the iterator is an implementation defined proxy type which supports the same operations as luabind::object. Which means that in most cases you can just treat it as an ordinary object. The difference is that any assignments to this proxy will result in the value being inserted at the iterators position, in the table.

The key() member returns the key used by the iterator when indexing the associated Lua table.

An example using iterators:

for (iterator i(globals(L)["a"]), end; i != end; ++i)
{


  *i = 1;
}

The iterator named end will be constructed using the default constructor and hence refer to the end of the sequence. This example will simply iterate over the entries in the global table a and set all its values to 1.

Related functions¶

There are a couple of functions related to objects and tables.

int type(object const&);

This function will return the lua type index of the given object. i.e. LUA_TNIL, LUA_TNUMBER etc.

template<class T, class K>
void settable(object const& o, K const& key, T const& value);
template<class K>
object gettable(object const& o, K const& key);
template<class T, class K>
void rawset(object const& o, K const& key, T const& value);
template<class K>
object rawget(object const& o, K const& key);

These functions are used for indexing into tables. settable and gettable translates into calls to lua_settable and lua_gettable respectively. Which means that you could just as well use the index operator of the object.

rawset and rawget will translate into calls to lua_rawset and lua_rawget respectively. So they will bypass any metamethod and give you the true value of the table entry.

template<class T>
T object_cast<T>(object const&);
template<class T, class Policies>
T object_cast<T>(object const&, Policies);
template<class T>
boost::optional<T> object_cast_nothrow<T>(object const&);
template<class T, class Policies>
boost::optional<T> object_cast_nothrow<T>(object const&, Policies);

The object_cast function casts the value of an object to a C++ value. You can supply a policy to handle the conversion from lua to C++. If the cast cannot be made a cast_failed exception will be thrown. If you have defined LUABIND_NO_ERROR_CHECKING (see part-build-options) no checking will occur, and if the cast is invalid the application may very well crash. The nothrow versions will return an uninitialized boost::optional<T> object, to indicate that the cast could not be performed.

The function signatures of all of the above functions are really templates for the object parameter, but the intention is that you should only pass objects in there, that's why it's left out of the documentation.

object globals(lua_State*);
object registry(lua_State*);

These functions return the global environment table and the registry table respectively.

object newtable(lua_State*);

This function creates a new table and returns it as an object.

object getmetatable(object const& obj);
void setmetatable(object const& obj, object const& metatable);

These functions get and set the metatable of a Lua object.

lua_CFunction tocfunction(object const& value);
template <class T> T* touserdata(object const& value)

These extract values from the object at a lower level than object_cast().

object getupvalue(object const& function, int index);
void setupvalue(object const& function, int index, object const& value);

These get and set the upvalues of function.

Assigning nil¶

To set a table entry to nil, you can use luabind::nil. It will avoid having to take the detour by first assigning nil to an object and then assign that to the table entry. It will simply result in a lua_pushnil() call, instead of copying an object.

Example:

using luabind;
object table = newtable(L);
table["foo"] = "bar";
// now, clear the "foo"-field
table["foo"] = nil;

Deriving in lua¶

It is also possible to derive Lua classes from C++ classes, and override virtual functions with Lua functions. To do this we have to create a wrapper class for our C++ base class. This is the class that will hold the Lua object when we instantiate a Lua class.

class base
{
public:


    base(const char* s)


    { std::cout << s << "\n"; }


    virtual void f(int a)


    { std::cout << "f(" << a << ")\n"; }
};
struct base_wrapper : base, luabind::wrap_base
{


    base_wrapper(const char* s)


        : base(s)


    {}


    virtual void f(int a)


    {


        call<void>("f", a);


    }


    static void default_f(base* ptr, int a)


    {


        return ptr->base::f(a);


    }
};
// ...
module(L)
[


    class_<base, base_wrapper>("base")


        .def(constructor<const char*>())


        .def("f", &base::f, &base_wrapper::default_f)
];

IMPORTANT:

Note that if you have both base classes and a base class wrapper, you must give both bases and the base class wrapper type as template parameter to class_ (as done in the example above). The order in which you specify them is not important. You must also register both the static version and the virtual version of the function from the wrapper, this is necessary in order to allow luabind to use both dynamic and static dispatch when calling the function.

IMPORTANT:

If we didn't have a class wrapper, it would not be possible to pass a Lua class back to C++. Since the entry points of the virtual functions would still point to the C++ base class, and not to the functions defined in Lua. That's why we need one function that calls the base class' real function (used if the lua class doesn't redefine it) and one virtual function that dispatches the call into luabind, to allow it to select if a Lua function should be called, or if the original function should be called. If you don't intend to derive from a C++ class, or if it doesn't have any virtual member functions, you can register it without a class wrapper.

You don't need to have a class wrapper in order to derive from a class, but if it has virtual functions you may have silent errors.

The wrappers must derive from luabind::wrap_base, it contains a Lua reference that will hold the Lua instance of the object to make it possible to dispatch virtual function calls into Lua. This is done through an overloaded member function:

template<class Ret>
Ret call(char const* name, ...)

Its used in a similar way as call_function, with the exception that it doesn't take a lua_State pointer, and the name is a member function in the Lua class.

WARNING:

NOTE:

class "D" (B)
function D:__init() B.__init(self) end
function D:virtual_function() ... end

b = B()
function b:virtual_function() ... end

Object identity¶

When a pointer or reference to a registered class with a wrapper is passed to Lua, luabind will query for it's dynamic type. If the dynamic type inherits from wrap_base, object identity is preserved.

struct A { .. };
struct A_wrap : A, wrap_base { .. };
A* f(A* ptr) { return ptr; }
module(L)
[


    class_<A, A_wrap>("A"),


    def("f", &f)
];

> class 'B' (A)
> x = B()
> assert(x == f(x)) -- object identity is preserved when object is


                    -- passed through C++

This functionality relies on RTTI being enabled (that LUABIND_NO_RTTI is not defined).

Overloading operators¶

You can overload most operators in Lua for your classes. You do this by simply declaring a member function with the same name as an operator (the name of the metamethods in Lua). The operators you can overload are:

__tostring isn't really an operator, but it's the metamethod that is called by the standard library's tostring() function. There's one strange behavior regarding binary operators. You are not guaranteed that the self pointer you get actually refers to an instance of your class. This is because Lua doesn't distinguish the two cases where you get the other operand as left hand value or right hand value. Consider the following examples:

class 'my_class'


  function my_class:__init(v)


      self.val = v


  end


  function my_class:__sub(v)


      return my_class(self.val - v.val)


  end


  function my_class:__tostring()


      return self.val


  end

This will work well as long as you only subtracts instances of my_class with each other. But If you want to be able to subtract ordinary numbers from your class too, you have to manually check the type of both operands, including the self object.

function my_class:__sub(v)


    if (type(self) == 'number') then


        return my_class(self - v.val)


    elseif (type(v) == 'number') then


        return my_class(self.val - v)


    else


        -- assume both operands are instances of my_class


        return my_class(self.val - v.val)


    end
end

The reason why __sub is used as an example is because subtraction is not commutative (the order of the operands matters). That's why luabind cannot change order of the operands to make the self reference always refer to the actual class instance.

If you have two different Lua classes with an overloaded operator, the operator of the right hand side type will be called. If the other operand is a C++ class with the same operator overloaded, it will be prioritized over the Lua class' operator. If none of the C++ overloads matches, the Lua class operator will be called.

Finalizers¶

If an object needs to perform actions when it's collected we provide a __finalize function that can be overridden in lua-classes. The __finalize functions will be called on all classes in the inheritance chain, starting with the most derived type.

...
function lua_testclass:__finalize()


    -- called when the an object is collected
end

Slicing¶

If your lua C++ classes don't have wrappers (see Deriving in lua) and you derive from them in lua, they may be sliced. Meaning, if an object is passed into C++ as a pointer to its base class, the lua part will be separated from the C++ base part. This means that if you call virtual functions on that C++ object, they will not be dispatched to the lua class. It also means that if you adopt the object, the lua part will be garbage collected.

+--------------------+
| C++ object         |    <- ownership of this part is transferred
|                    |       to c++ when adopted
+--------------------+
| lua class instance |    <- this part is garbage collected when
| and lua members    |       instance is adopted, since it cannot
+--------------------+       be held by c++.

The problem can be illustrated by this example:

struct A {};
A* filter_a(A* a) { return a; }
void adopt_a(A* a) { delete a; }

using namespace luabind;
module(L)
[


    class_<A>("A"),


    def("filter_a", &filter_a),


    def("adopt_a", &adopt_a, adopt(_1))
]

In lua:

a = A()
b = filter_a(a)
adopt_a(b)

In this example, lua cannot know that b actually is the same object as a, and it will therefore consider the object to be owned by the C++ side. When the b pointer then is adopted, a runtime error will be raised because an object not owned by lua is being adopted to C++.

If you have a wrapper for your class, none of this will happen, see Object identity.

Policies currently implemented¶

adopt¶

Motivation¶

Used to transfer ownership across language boundaries.

Defined in¶

#include <luabind/adopt_policy.hpp>

Synopsis¶

adopt(index)

Parameters¶

Example¶

X* create()
{


    return new X;
}

module(L)
[


    def("create", &create, adopt(result))
];

dependency¶

Motivation¶

The dependency policy is used to create life-time dependencies between values. This is needed for example when returning internal references to some class.

Defined in¶

#include <luabind/dependency_policy.hpp>

Synopsis¶

dependency(nurse_index, patient_index)

Parameters¶

Example¶

struct X
{


    B member;


    B& get() { return member; }
};

module(L)
[


    class_<X>("X")


        .def("get", &X::get, dependency(result, _1))
];

out_value¶

Motivation¶

This policy makes it possible to wrap functions that take non-const references or pointer to non-const as it's parameters with the intention to write return values to them. Since it's impossible to pass references to primitive types in lua, this policy will add another return value with the value after the call. If the function already has one return value, one instance of this policy will add another return value (read about multiple return values in the lua manual).

Defined in¶

#include <luabind/out_value_policy.hpp>

Synopsis¶

out_value(index, policies = none)

Parameters¶

Example¶

void f1(float& val) { val = val + 10.f; }
void f2(float* val) { *val = *val + 10.f; }

module(L)
[


    def("f", &f, out_value(_1))
];

Lua 5.0  Copyright (C) 1994-2003 Tecgraf, PUC-Rio
> print(f1(10))
20
> print(f2(10))
20

pure_out_value¶

Motivation¶

This works exactly like out_value, except that it will pass a default constructed object instead of converting an argument from Lua. This means that the parameter will be removed from the lua signature.

Defined in¶

#include <luabind/out_value_policy.hpp>

Synopsis¶

pure_out_value(index, policies = none)

Parameters¶

Example¶

Note that no values are passed to the calls to f1 and f2.

void f1(float& val) { val = 10.f; }
void f2(float\* val) { \*val = 10.f; }

module(L)
[


    def("f", &f, pure_out_value(_1))
];

Lua 5.0  Copyright (C) 1994-2003 Tecgraf, PUC-Rio
> print(f1())
10
> print(f2())
10

return_reference_to¶

Motivation¶

It is very common to return references to arguments or the this-pointer to allow for chaining in C++.

struct A
{


    float val;


    A& set(float v)


    {


        val = v;


        return *this;


    }
};

When luabind generates code for this, it will create a new object for the return-value, pointing to the self-object. This isn't a problem, but could be a bit inefficient. When using the return_reference_to-policy we have the ability to tell luabind that the return-value is already on the lua stack.

Defined in¶

#include <luabind/return_reference_to_policy.hpp>

Synopsis¶

return_reference_to(index)

Parameters¶

Example¶

struct A
{


    float val;


    A& set(float v)


    {


        val = v;


        return *this;


    }
};

module(L)
[


    class_<A>("A")


        .def(constructor<>())


        .def("set", &A::set, return_reference_to(_1))
];

WARNING:

copy¶

Motivation¶

This will make a copy of the parameter. This is the default behavior when passing parameters by-value. Note that this can only be used when passing from C++ to Lua. This policy requires that the parameter type has an accessible copy constructor.

Defined in¶

#include <luabind/copy_policy.hpp>

Synopsis¶

copy(index)

Parameters¶

Example¶

X* get()
{


    static X instance;


    return &instance;
}

module(L)
[


    def("create", &create, copy(result))
];

discard_result¶

Motivation¶

This is a very simple policy which makes it possible to throw away the value returned by a C++ function, instead of converting it to Lua.

Defined in¶

#include <luabind/discard_result_policy.hpp>

Synopsis¶

discard_result

Example¶

struct X
{


    X& set(T n)


    {


        ...


        return *this;


    }
};

module(L)
[


    class_<X>("X")


        .def("set", &simple::set, discard_result)
];

return_stl_iterator¶

Motivation¶

This policy converts an STL container to a generator function that can be used in lua to iterate over the container. It works on any container that defines begin() and end() member functions (they have to return iterators).

Defined in¶

#include <luabind/iterator_policy.hpp>

Synopsis¶

return_stl_iterator

Example¶

struct X
{


    std::vector<std::string> names;
};

module(L)
[


    class_<A>("A")


        .def_readwrite("names", &X::names, return_stl_iterator)
];

> a = A()
> for name in a.names do
>  print(name)
> end

raw¶

NOTE:

Motivation¶

This converter policy will pass through the lua_State* unmodified. This can be useful for example when binding functions that need to return a luabind::object. The parameter will be removed from the function signature, decreasing the function arity by one.

Defined in¶

#include <luabind/raw_policy.hpp>

Synopsis¶

raw(index)

Parameters¶

Example¶

void greet(lua_State* L)
{


    lua_pushstring(L, "hello");
}

module(L)
[


    def("greet", &greet, raw(_1))
];

> print(greet())
hello

yield¶

Motivation¶

Makes a C++ function yield when returning.

Defined in¶

#include <luabind/yield_policy.hpp>

Synopsis¶

yield

Example¶

void do_thing_that_takes_time()
{


    // ...
}

module(L)
[


    def("do_thing_that_takes_time", &do_thing_that_takes_time, yield)
];

pcall errorfunc¶

As mentioned in the Lua documentation, it is possible to pass an error handler function to lua_pcall(). Luabind makes use of lua_pcall() internally when calling member functions and free functions. It is possible to set the error handler function that Luabind will use globally:

typedef int(*pcall_callback_fun)(lua_State*);
void set_pcall_callback(pcall_callback_fun fn);

This is primarily useful for adding more information to the error message returned by a failed protected call. For more information on how to use the pcall_callback function, see errfunc under the pcall section of the lua manual.

For more information on how to retrieve debugging information from lua, see the debug section of the lua manual.

The message returned by the pcall_callback is accessible as the top lua value on the stack. For example, if you would like to access it as a luabind object, you could do like this:

catch(error& e)
{


    object error_msg(from_stack(e.state(), -1));


    std::cout << error_msg << std::endl;
}

file and line numbers¶

If you want to add file name and line number to the error messages generated by luabind you can define your own pcall errorfunc. You may want to modify this callback to better suit your needs, but the basic functionality could be implemented like this:

int add_file_and_line(lua_State* L)
{


   lua_Debug d;


   lua_getstack(L, 1, &d);


   lua_getinfo(L, "Sln", &d);


   std::string err = lua_tostring(L, -1);


   lua_pop(L, 1);


   std::stringstream msg;


   msg << d.short_src << ":" << d.currentline;


   if (d.name != 0)


   {


      msg << "(" << d.namewhat << " " << d.name << ")";


   }


   msg << " " << err;


   lua_pushstring(L, msg.str().c_str());


   return 1;
}

For more information about what kind of information you can add to the error message, see the debug section of the lua manual.

Note that the callback set by set_pcall_callback() will only be used when luabind executes lua code. Anytime when you call lua_pcall yourself, you have to supply your function if you want error messages translated.

lua panic¶

When lua encounters a fatal error caused by a bug from the C/C++ side, it will call its internal panic function. This can happen, for example, when you call lua_gettable on a value that isn't a table. If you do the same thing from within lua, it will of course just fail with an error message.

The default panic function will exit() the application. If you want to handle this case without terminating your application, you can define your own panic function using lua_atpanic. The best way to continue from the panic function is to make sure lua is compiled as C++ and throw an exception from the panic function. Throwing an exception instead of using setjmp and longjmp will make sure the stack is correctly unwound.

When the panic function is called, the lua state is invalid, and the only allowed operation on it is to close it.

For more information, see the lua manual section 3.19.

structured exceptions (MSVC)¶

Since lua is generally built as a C library, any callbacks called from lua cannot under any circumstance throw an exception. Because of that, luabind has to catch all exceptions and translate them into proper lua errors (by calling lua_error()). This means we have a catch(...) {} in there.

In Visual Studio, catch (...) will not only catch C++ exceptions, it will also catch structured exceptions, such as segmentation fault. This means that if your function, that gets called from luabind, makes an invalid memory addressing, you won't notice it. All that will happen is that lua will return an error message saying "unknown exception".

To remedy this, you can create your own exception translator:

void straight_to_debugger(unsigned int, _EXCEPTION_POINTERS*)
{ throw; }
#ifdef _MSC_VER


   ::_set_se_translator(straight_to_debugger);
#endif

This will make structured exceptions, like segmentation fault, to actually get caught by the debugger.

Error messages¶

These are the error messages that can be generated by luabind, with a more in-depth explanation.

the attribute 'class-name.attribute-name' is read only

the attribute 'class-name.attribute-name' is of type: (class-name) and does not match (class_name)

class-name() threw an exception, class-name:function-name() threw an exception

no overload of 'class-name:function-name' matched the arguments (parameter-types)
no match for function call 'function-name' with the parameters (parameter-types)
no constructor of class-name matched the arguments (parameter-types)
no operator operator-name matched the arguments (parameter-types)

call of overloaded 'class-name:function-name*(*parameter-types)' is ambiguous
ambiguous match for function call 'function-name' with the parameters (parameter-types)
call of overloaded constructor 'class-name*(*parameter-types)' is ambiguous
call of overloaded operator operator-name (parameter-types) is ambiguous

cannot derive from C++ class 'class-name'. It does not have a wrapped type.

CMake options¶

The following options can be given to CMake when building luabind. You can either specify them directly on the commandline using -Doption=value or let CMake ask you for each of them by invoking it with the -i flag. On Windows, you will usually prefer cmake-gui.

All options are booleans, unless specified otherwise.

The detail namespace and undocumented features¶

Inside the luabind namespace, there’s another namespace called detail. Everything cotained therein should not be used by user code and is subject to change or vanish even between patch versions. There also exist no explicit tests or documentation for this namespace’s contents.

Classes and functions which reside directly in the luabind namespace but are not documented should also rather not be used. However, if you have to depend on undocumented features, these are still better than the detail ones.