Re: Why is nil not a valid table key?

[Rici Lake <lua@...>]

On 26-Nov-05, at 4:54 AM, Wim Couwenberg wrote:

> If nil is in the range, then surely *all* values are in the domain?
> ;-)

Yes, that was my point. Considering gettable as a function, the range 
is <all values> and the domain is all pairs t, k such that t is a table 
and k is any value. (Leaving aside things with metatables for now). But 
settable differs, in that the domain of the second argument does not 
include nil (or NaNs).

If we return to metatables, we'll see that __index(t, k) can receive 
nil as its second argument, but __newindex(t, k, v) cannot.

That is the sense in which I say that nil is not in the domain.

eg.

> foo = setmetatable({}, {__index = print, __newindex = print})
> =foo[nil]
table: 0x808c400        nil
nil
> foo[nil] = 3
stdin:1: table index is nil
stack traceback:
        stdin:1: in main chunk
        [C]: ?