sqrt of integers returns a double NaN

[ltratt]

This program:

sqrter = (
    run = (
        -1 sqrt println.
    )
)

prints NaN which was a little surprising to me -- I wouldn't have expected a conversion to a double in such a case. What we've done in softdevteam/yksom#57 is to throw an error in such cases. Admittedly, on reflection, both the NaN and "thrown an error" behaviour are defensible, though I suspect the latter catches more genuine programmer errors. Maybe something to think about for the SOM spec!

[smarr]

Hi @ltratt:

Smalltalks classically try to move between numerical representations for such things.
I am not sure I like that. One problem is that it violates the typing dimension, and makes it value dependent. This leads to surprising results.

---> 4 sqrt
it = 2
---> 4 sqrt class
it = Integer
---> 4.0 sqrt class
it = Double
---> 1 sqrt class
it = Integer
---> 3 sqrt class
it = Double
---> 10 sqrt class
it = Double
---> -1 sqrt class
it = Double

The above behavior (note the 4, 1, and 3 case), is consistent with Pharo. Except for -1 raising an exception.

So, the conversion to a Double and NaN isn't surprising to me and seem consistent with the other behaviors.
Though, I agree that an error as in yksom or Pharo is likely the better result.

I did see the PR earlier, but didn't have an opinion on it before.
Is yksom consistent with sqrt results for 4, 1, and 3?