Re: rename LUA_RIDX_CPCALL to LUA_RIDX_CCALL? (5.2.0-work2)

[Drake Wilson <drake@...>]

Quoth Luiz Henrique de Figueiredo <lhf@tecgraf.puc-rio.br>, on 2010-03-29 08:57:21 -0300:
> > The dance required to get strings and boxed pointers passed to lua
> > callbacks is borderline ridiculous. :)
> 
> What do you mean exactly?

I'm not the OP, but the problem as I see it is that you may have
C-level unwind requirements during data transfer from some other
source into Lua space.

E.g.:

  char const *const str = some_library_get_string();
  lua_pushstring(L, str);  /* Memory leak if this fails! */
  some_library_free_string(str);

and in this case you can't even easily use (5.1) lua_cpcall:

  static int
  actually_push_string(lua_State *L)
  {
      /* Wait a minute, there's no good way to get the string back
         if the return values go away. */
      lua_pushstring(L, (char const *)lua_touserdata(L, 1));
      return 1;
  }

so instead you have to do something like:

  /* luaL_checkstack(L, 2, ""); */
  lua_pushcfunction(L, &actually_push_string);
  char const *const str = some_library_get_string();
  lua_pushlightuserdata(L, str);
  int const err = lua_pcall(L, 1, 1, 0);
  some_library_free_string(str);
  if (err != 0) lua_error(L);

There are sometimes ways to simplify this, but it's very tricky to
work with and get right every time.

(Incidentally, in a Lua bolt-on I'm currently working on privately,
I've bandied about the idea of providing C-level unwind-protect
registration of some kind for this exact reason, which might or might
not say anything about the topic.)

   ---> Drake Wilson