Re: Alternative (better?) __index implementation

["Thomas Lefort" <thomas.lefort@...>]

On Nov 27, 2007 2:50 AM, alex.mania@iinet.net.au
<alex.mania@iinet.net.au> wrote:
> Hey,
>
> First time poster here.. be gentle.
>
> For a particular oo system I was experimenting with, I found the current __index
> metamethod to not work quite as desired. The difference is best explained through
> an example:
>
>  local proxy = setmetatable({}, {__index = function(tab) return tab.greeting end})
>  local test = setmetatable({ greeting = "hello" }, {__index = proxy})
>  print(test.unknownkey)
>
> The current implementation would print nil,

Or maybe:

lua: tmp/index.lua:1: C stack overflow
stack traceback:
        tmp/index.lua:1: in function <tmp/index.lua:1>
        tmp/index.lua:1: in function <tmp/index.lua:1>
        tmp/index.lua:1: in function <tmp/index.lua:1>
        tmp/index.lua:1: in function <tmp/index.lua:1>
        tmp/index.lua:1: in function <tmp/index.lua:1>
        tmp/index.lua:1: in function <tmp/index.lua:1>
        tmp/index.lua:1: in function <tmp/index.lua:1>
        tmp/index.lua:1: in function <tmp/index.lua:1>
        tmp/index.lua:1: in function <tmp/index.lua:1>
        tmp/index.lua:1: in function <tmp/index.lua:1>
        ...

You can guess why. Next time test your code before you post ;-)

> I propose it would be more meaningful
> for the end __index metamethod to be called with the table that instigated the
> __index loop, rather than table that holds the function. The performance
> difference is negligible, and the change can be implemented with one line of code
> and one changed variable name.

Yes, I found myself wishing for this at some point too, but it can
still be emulated with adequate metamethods. Even if this can get a
bit heavy (very relative when doing Lua anyway) I believe it should be
preferred over a semantic change in the language core.

-- 
-- Thomas