Re: Some small requests for the next release

[Rici Lake <lua@...>]

On 2-Mar-05, at 12:28 PM, Roberto Ierusalimschy wrote:

> > > What is wrong with the following code?
> > >
> > >   math.nan = 0/0
> > >   math.inf = 1/0
> >
> > If you've compiled Lua with integer Numbers rather than floating point
> > ones, this will crash and burn, probably.
>
> Is there any reason why anyone would use such library in that case?

If they haven't, then the assignment will crash and burn, which is not 
desirable either.

However, I do tend to use "a" math library (called "math") even in 
integer installs; it contains compatible integer implementations of the 
non-floating point dependent math library standard functions (random, 
min, max, mod, and a few others).

It is not clear to me under what circumstances you might actually want 
to use a math.nan constant since you cannot actually compare it to 
anything, but it would be useful to have the math.inf constants, even 
with integer arithmetic. (They're convenient initial values for max and 
min, for example).

With that sort of use case in mind, one could argue that math.infinity 
(and possibly math.minusinfinity) might be useful additions to the 
(specification of) a standard math library interface.

math.nan is rather a different category, though, since technically 
math.nan ~= math.nan, which renders the whole thing somewhat academic. 
On the other hand, if a nan had a boolean coercion to false, you could 
write, for example:

a = assert(math.sqrt(x), "Negatives only have round roots")