Re: Proxy tables

[RLake@...]

> Could rigorous use of a (fictitious) meta method __pairs help
> to solve this?  Then pairs(object) can always be used to iterate over
> object.  (I think __pairs reflects "iteration overriding" better than a 
more
> low level __next?)

I agree. Whether rigorous use of __pairs would solve all iteration 
problems
is an open question, but it is certainly at least as general as using
a __next metamethod, because __pairs could be implemented by default as

  return getmetatable(t).__next, t

So __pairs is at least as powerful as __next, but it could do other things
as well.

(Leaving out a lot of details about checking for metatable existence and 
all that.)

The huge advantage of __pairs, other than the fact that it gets to look at 
the object as a whole rather than in little pieces, is that it only gets 
called once per loop, instead of once per loop iteration. That is a big 
performance gain. Unfortunately, in the (deprecated) syntax:

  for k, v in t do ... end

What actually happens is

  for k, v in next, t do ... end

Rather than

  for k, v in pairs(t) do ... end

which would arguably have been easier to override.

In answer to Mark's question, what do you need to override:

1) You don't need to use __newindex for a lazy copy, just __index. But if 
it is a pure proxy, you would need to override whatever metamethods your 
object implemented; in the case of tables, that is just __index and 
__newindex.

2) You don't *need* to override pairs; all it does is return <next, t>. 
The only reason you would have to override pairs is if you had an 
iteration problem you couldn't solve with a custom next function.

3) If you need ordered iteration, you need to override table.foreachi and 
maybe ipairs as well as table.foreach. In addition, you would probably 
need to override a number of other table. functions which implement 
vectors-as-tables.

4) Lua does not lend itself to complex metaobject protocol implementations 
because you cannot stack MOPs. However, that may be considered a feature 
if you believe that complex 

R.