Re: __iter metamethod?

[David Given <dg@...>]

steve donovan wrote:
[...]
> for i,v in t do

In this context, 'for' expects t to be a callable thing. Normally this 
is a function, as the first parameter of a tuple:

   for k, v in ipairs(table) do
=> for k, v in fn, table, 0 do

'for' then calls fn repeatedly to fetch the values, passing in the rest 
of the tuple (which gets modified during the iteration).

The problem is, the rest of the tuple is *optional*. It's perfectly 
possible to use an iterator function without them.

  i = 0
  function iterator()
    i = i + 1
    return i
  end

  for a in iterator do
    print(a)
  end

And, of course, tables can be overridden using the __call metatable 
entry to look like functions.

The upshot of this is, if you add an __iter metatable entry to allow 
iteration on tables, you also add a nasty ambiguity as to what happens 
if the table has both a __call *and* an __iter. That's not to say it 
can't be done... just that ambiguities like this are best avoided if you 
wish to retain your sanity.

(And implementing iteration using __call is tricky because you don't get 
told when the iteration starts or stops. Note that my iterator function 
above uses *global* state, not a per-invocation state.)

--
David Given
dg@cowlark.com