Re: linking tables to C++ objects?

[Graham Wakefield <lists@...>]

Thanks.

Wouldn't it be cheaper & more efficient to use the registry rather  
than a metatable subtable?

callback(cpp_class *c)
{
      lua_State *L = c->lua();
      lua_pushlightuserdata(L, c);
      lua_gettable(L, LUA_REGISTRYINDEX);
      //either nil or your userdata will no be on the stack
}

If I need to type-check, I can compare the metatable of the userdata  
to the class name:

bool typecheck(lua_State * L)
{
	lua_getfield(L, LUA_REGISTRYINDEX, c::classname);
	if (lua_getmetatable(L, -1) && lua_rawequal(L, -1, -2))
	{
		lua_pop(L, 2);	// remove both metatables
		return true;
	}
	lua_pop(L, 1);	// remove class table
	return false;
}


On Feb 26, 2007, at 5:23 PM, Wesley Smith wrote:

> I think, as suggested by Rici, that you need a table mapping C++
> pointers to userdata in the Lua userdata's metatable.  somehow you'll
> need to get the lua_State.  In this case, I assume a pointer is cached
> in the cpp_class and that the table of cpp pointers to userdata is in
> the metatable's "instances" field;
>
> pseudo code
>
> callback(cpp_class *c)
> {
>    lua_State *L = c->lua();
>       luaL_getmetatable(L, cpp_class::metatable_name);
>       lua_pushstring(L, "instances");
>       lua_gettable(L, -2)
>       lua_pushinteger(L, (int)c);
>       lua_gettable(L, -2)
>       //either nil or your userdata will no be on the stack
> }
>
>
> Also I would make the "instances" table a weak table with __mode = "v"
>
> wes