Re: Cons, tuple and __slots__ -- a simple mechanism to implement them in Lua

[Roberto Ierusalimschy <roberto@...>]

> I'm very sad that nobody reply my last post[1], even Roberto didn't.
> [...]

As you want some comment, I will give a brief comment regarding your
implementation. I guess there is a little "bug" in the garbage
collection of your userdata. Consider your patch:

*** 248,254 ****
      case LUA_TUSERDATA: {
        Table *mt = gco2u(o)->metatable;
        markobject(g, mt);
!       markobject(g, gco2u(o)->env);
        gray2black(o);  /* all pointers marked */
        return;
      }
--- 248,259 ----
      case LUA_TUSERDATA: {
        Table *mt = gco2u(o)->metatable;
        markobject(g, mt);
!       if (gco2u(o)->valuecount != 0)
!       {
!         int i, len = gco2u(o)->valuecount;
!         for (i = 0; i < len; ++i)
!           markobject(g, &gco2u(o)->values[i]);    <<<<<
!       }
        gray2black(o);  /* all pointers marked */
        return;
      }

In the original version, the call to 'markobject' was always over
a table, and so we knew the recursion would stop at that level
(the 'reallymarkobject' for tables only queues the table for later
traversal). In your version, the call to 'markobject' (see '<<<<') may
be marking any object, in particular another userdata. So, it seems
that a long linked list of userdata could force an unlimited depth of
recursive calls to 'reallymarkobject', eventually crashing the C stack.

(BTW, as you can now have any value as userdata, it seems that you
should use 'markvalue' instead of 'markobject' in the line marked
with '<<<<'.)

-- Roberto