Re: upcoming changes in Lua 5.2 [was Re: Location of a package]

["Alex Davies" <alex.mania@...>]

As you say, __index is supposed to be called for non-existent keys. ipairs
though, like a lot of the table library, does not respect __index even if 
the key
is non-existent. An oversight imo, but I guess it must have some advantages 
(?)...

The main difference is that in letting __ipairs be overriden you can change 
the
order from 1, 2, n to 3, 8, 5, 1 or whatever else tickles your fancy. That 
will
break any existing code which assumes ipairs always starts from 1 and counts 
up
(I've been guilty of making this assumption before), but again I guess it 
must
have some advantages =P.

- Alex

----- Original Message ----- 
From: "Ignacio Burgueño" <ignaciob@inconcertcc.com>
To: "Lua list" <lua@bazar2.conectiva.com.br>
Sent: Thursday, February 21, 2008 11:05 PM
Subject: Re: upcoming changes in Lua 5.2 [was Re: Location of a package]


> Jérôme Vuarand wrote:
> First, I think that the above would be a very welcomed addition to Lua.
> Second, in my opinion, __index and __ipairs (whatever the final semantics 
> may be) are not comparable. __index is triggered on access to non-existent 
> keys and __ipairs would be used to redefine the way an array-like table is 
> iterated, right?
>
> Also, using the same metamethod for both cases will make it hard to 
> distinguish when you're indexing a table and when you're going to iterate 
> it.
>
> Regards,
> Ignacio Burgueño