Testing for NaN

SylvainGa over 1 year ago

Hi, how does one test for NaN? The following if statement is skipped, although 'rad' is NaN

Lang.NaN doesn't work either (doesn't compile). "try/catch" doesn't catch it either.

Thanks.

Top Replies

Travis.ConnectIQ 9 months ago +2 suggested

I think this should work.

const FLT_MAX = 3.4028235e38f;

function isnan(x as Float) as Boolean {
    return x != x;
}

function isinf(x as Float) as Boolean {
    return (x < -FLT_MAX || FLT_MAX…

FlowState over 1 year ago in reply to SylvainGa +1
That's bizarre but I think I found a workaround for you.

System.println(Math.acos(45d).toFloat().equals(Math.acos(45))); // true
```
class MathIsFun {
    static const negativeNan as Float = Math.acos…
```
SylvainGa over 1 year ago in reply to FlowState +1

Yep, that did the trick, thanks again :-)

All Replies

0 FlowState over 1 year ago

rad.equals(NaN) works for me.

e.g.

var rad = NaN;
System.println("rad == NaN: " + (rad == NaN)); // false
System.println("rad.equals(NaN): " + rad.equals(NaN)); // true

Seems to be an exception to the Monkey C rule that you should be able to use "==" with "primitive" types.

0 SylvainGa over 1 year ago in reply to FlowState

LOL, now try this:

If I manually set it to NaN, it works with your code, but when it's returned through a Math function, nope! WTH!

0 FlowState over 1 year ago in reply to SylvainGa

lmao that's terrible

0 SylvainGa over 1 year ago in reply to FlowState

I guess it's time to report a bug...

+1 FlowState over 1 year ago in reply to SylvainGa

Looks like Math.acos(45) actually returns -NaN.

System.println("Math.acos(45): " + Math.acos(45)); // "Math.acos(45): -nan(ind)"System.println(Math.acos(45).equals(-NaN)); // "false" (not that I expect this to work since NaN * x => NaN, therefore -NaN => NaN)

However:

System.println(Math.acos(45).equals(Math.acos(45))); // "true"

This suggests you can test for -NaN by comparing any number to the result of Math.acos(45) lol. This also suggests an implementation for a function that seems to be sorely needed:public function isFinite(x as Number or Long or Float or Double) as Boolean { return !x.equals(NaN) && !x.equals(Math.acos(45));
}

EDIT: isFinite() is probably the wrong name for this function, but I don't have a better one :/

EDIT2: This answer is bad in the sense that it probably doesn't cover all possible forms of NaN that can be encountered in Monkey C (the IEEE 754 spec reserves ~16 million bit patterns for NaN).

See Travis.ConnectIQ's comment for the correct answer:

[https://forums.garmin.com/developer/connect-iq/f/discussion/338071/testing-for-nan/1777041#1777041]

0 SylvainGa over 1 year ago in reply to FlowState

That would mean I would need a 'finite' function for all the Math functions I should test against NaN :-O

Edit: Oh, I see what you mean, I'll try it,

0 SylvainGa over 1 year ago in reply to FlowState

Yep, it worked! Thanks for your help :-)

0 SylvainGa over 1 year ago in reply to FlowState

Hi, to increase the accuracy of my results, I've converted all floats to doubles but by doing so, it broke the isFinite function :-( I tried a few variations but non returned false when NaN.double was returned. What would be the magic equation for doubles? Thanks,

BTW, Had to convert from float to double because of funky precision in plain arithmetic.

0 FlowState over 1 year ago in reply to SylvainGa

What's the function and input(s) that return NaN.double (and/or -Nan.double)? You must have at least one example, since you're seeing Nan.double return from somewhere.

Store the result of that function call in a variable and use it with your isFinite() function. It's the same concept behind using Math.acos(45) to get -NaN (float).

0 FlowState over 1 year ago in reply to FlowState

You may also have to handle Infinity/-Infinity, although I'm not sure how to derive those values in Monkey C since dividing by zero results in an exception IIRC.