Re: NaNs etc. once again. Was: Re: 5.1 and vc6

[Rici Lake <lua@...>]

On 6-Mar-05, at 9:32 PM, Mark Hamburg wrote:

> I think one could separate the issues of whether NaN tests false in a
> conditional context from what it does in an assert.

Indeed. And one could use math.isnan (if there were such a thing) to 
test for NaNs (if the architecture had such things).

But somehow:

  x = math.sqrt(math.atan(theta))
  if not x then
    -- oops an error
  end

seems easier to explain and understand than either:

  if x ~= x then ..

or

  if math.isnan(x) then ...

And it seems slightly more consistent with other Lua functions which 
return false, <error> on error. There would be nothing stopping 
math.sqrt(-1) from returning:

  NaN, "square roots of negative numbers are a complex subject"

although that would be trickier with 0 / 0 and 0 * infinity