Comparing NaN constant(s) with buggy results!!

[QFang]

Hey NI Forum people!

I know I should be very carefull when making comparisons on floating point numbers due to rounding and representation issues.

That said, I got completely caught off guard when I wanted a 'true' boolean if a number was NOT equal to 'NaN'.. I wasn't thinking about it too much, just slinging wires for a quick conditional for loop terminal for some testing I was doing in a temporary VI, and implemented this with a 'not equal to' primitive (instead of the 'Not a Number/path/refnum' primitive.

The result caught me by surprise, and when I dug into it, it became even weirder... turns out comparing a single constant to itself will yield unexpected results too!! 

Is this how it has always been or is this NEW in LV 2015?? 

I included an LV 2011 version if anyone wants to try on an older version.. I ran this on a Windows 7 PC.

[Sam_Sharp · Accepted solution]

Whilst it might not make sense, it is documented behaviour though.

Documentation:

---

Equal? Function

If you compare two inputs with the value not a number, NaN, or one input with the value NaN to an input with another value, this function always returns FALSE. Use the Not a Number/Path/Refnum? function to compare one or more inputs with a value of NaN.

---

 (same applies to Not Equal?)

[QFang]

Damn, you know, checking the documentation for THAT primitive just honestly didn't cross my mind... I "knew" very well what it was doing.. or so I thought!

[James.M]

There's a forum topic on here somewhere about caveats to LabVIEW that some people could get tripped up by, but I can't seem to find it... I think Jack Dunaway started it.

Anyways, this was posted in there a while ago.

Here's an NI page about it at least.

[shansen1]

Isn't this part of the IEEE 754 floating point standard? There is a binary representation for NaN, and then the second part of the definition is that Value != Value if and only if Value == NaN. With C-code I often use Value != Value to check for NaN.

[JackDunaway]

---

@shansen1 wrote:

Isn't this part of the IEEE 754 floating point standard

---

Yes. Since the original source is frustratingly behind a paywall, here's a link that discusses this issue: https://en.wikipedia.org/wiki/IEEE_floating_point

---

@James.M wrote:

There's a forum topic on here somewhere about caveats to LabVIEW that some people could get tripped up by

---

 http://forums.ni.com/t5/BreakPoint/LabVIEW-Minutiae-that-may-bite-you-someday/m-p/1122234/highlight/...

[James.M]

There it is! Thanks.

(I swear, you must have a script that pings you every time your name is mentioned...)

[tst]

---

@James.M wrote:

(I swear, you must have a script that pings you every time your name is mentioned...)

---

The forums allow you to configure a search subscription, which should do this if your name is unique enough. Haven't tried it myself. Maybe I'll try it now, but I expect I will mainly get noise from threads I participate in.

[Hooovahh]

And the fact that some of us would need to subscribe to many mis-spellings of our name, Hooovahh, hoovah, hooovah, hoovahh, then there are our actual names.  Sounds like a lot of work, just read every post on the forums instead. 

[James.M]

Hey, that's neat! Looks like the following query works for me. It even shows a post from Bob Schor that I missed where he called me out a while ago.

james.morris OR jamesmorris OR "james morris" -"James.Morris wrote"

I expect "tst" will show up everywhere. And yeah, Hooovahhhhhh is easy to get wrong when summoning you...