Re: How to overwrite/implement '__len'?

[Clemens Hintze <ml-lua@...>]

Doug Rogers schrieb:
> Clemens Hintze wrote:

Thanks for your answer,

(...)

> Yes, like the __gc metamethod (but for different reasons - see
> http://lua-users.org/lists/lua-l/2006-01/msg00172.html), __len can only
> be overridden by userdatum types. Experience has taught me that there
> will be a good reason for such an apparent oversight. It may just be
> performance; see http://lua-users.org/lists/lua-l/2006-03/msg00700.html.

I have already seen those both. I had only hoped that there would be any 
other way than taking the table implementation of Lua and reintroduce it 
as loadable extension with type() e.g. "object" only to get to a 
functional __len(). Pity that! :-(

I like orthogonality and consistency. And Lua was one of the most 
consistent so far (the other could be Tcl, IMO).

> I don't see an easy way to avoid the cost of walking the metatable tree
> for the case where __len is not defined.

You're right. Beside to attach something like 'meta flags' to a table 
that could indicate, that there is a __len that should be taken, I would 
also have no idea, how to avoid those costs :-(

But as it is only a small detail, I can probably live with it I mean :-)

Either I re-take the table implementation for implementing objects or I 
will assign a 'size()' method to my table-build objects at Lua level. 
The only confusing remaining thing would be, that one had to remember 
that she should *not* take '#' to get the size of my rbtree instance.

And there will be no way to disable '#' for my rbtree ... *gnarz* :-/

> But perhaps there are other issues. I was unable to find any further
> discussion in the archives, but I seem to remember more than what I found.

Its okay, thank you anyway. It was at least a hint, that I had not 
overseen something. Thanks again!


Ciao,
Clemens.