Include a binary mode option for the comparaison 'egal' and 'not egal'

[EricC.]

Because the 'NaN' value is very special, make avaible someting like a binary mode for the test egal and not egal.

So the test 'NaN' not egale to 'NaN' is False.

[JackDunaway]

I was reading up on some SQL the other day, and the author made a clear point that "variable = NULL" by design returns a different result than "variable IS NULL". LabVIEW may have been designed with the same concept - it's impossible to assign comparison equality or inequality to Not a Number. Comments from R&D?

Laura, if it is alright with Cthulhu, I would suggest the title of this post be changed to "Include a binary mode option for the comparison 'Equal' and 'Not Equal'" for better searching (it took me a while to realize egal=equal).

[tst]

> it's impossible to assign comparison equality or inequality to Not a Number

That's correct. I believe the IEEE standard LV uses specifies that comparing to NaN will always fail.

LabVIEW already has a primitive (called Not a Number/Path/Refnum?) which can be used to test if a number is NaN.

[smercurio_fc]

---

Cthulhu wrote:

Because the 'NaN' value is very special,

---

And that's precisely the point. As noted, NaN is not supposed to be used for equality operations. The IEEE standaard that tst alludes to is IEEE754. It states that NaN=NaN always returns false. Further, according to the spec, NaN can actually have a sign, but it has no meaning other than just diagnostics.

[altenbach]

Also note that a binary comparison of NaNs can easily fail, because the Mantissa can be arbitrary as long as it is not zero. This means we can have 2^52-1 different NaN and any comparison is meaningless.

Here's an illustrative example (from here) 

[EricC.]

I know the function "Is NaN/Path/Refnum", but when you have the value 'NaN' in a cluster or in any complex struct, it is not esay to use. So to make a test if the value change, you have only the choice to compare the Flattened to string data.

[smercurio_fc]

Well, even using Flatten To String is not guaranteed to work because NaN is so special. As noted by altenbach, and as detailed in the IEE754 spec I linked to earlier, NaN is only defined by the exponent being a specific pattern. The "fraction" part can be any non-zero value. Thus, a Flatten to String would give you 2^32 - 1 possible flattened values for SGL and 2^52 - 1 possible flattened values for DBL. If you need to do that kind of comparison then you should use Inf or -Inf since the bit pattern for those values is clearly defined.

[EricC.]

by experience say that the function Flatten To string work well.

When i use this function, i can test a copy of the memory used to record the data.

So i can test if a cluster or an array change betwenn the readed data and his local copy.

[Darren]

Any idea that has received less than 2 kudos within 2 years after posting will be automatically declined.