%NFACT bug

[David Hayden]

It appears to me that the System RPL entry %NFACT on my 50g doesn't work properly. This computes the factorial of a real number. Note that it's different from %FACT which computes gamma(x+1) and works correctly.

%NFACT appears to return IP(ABS(x))! There are 2 bugs

1. It doesn't honor system flag 22 when the result overflows. It returns 9.99999999999E499, even if the flag is clear
2. If x is between -1 and 1 (exclusive of the endpoints), it return 9.99999999999E499 instead of 1.
   

Hoping that someone can confirm or explain to me why I'm wrong.

Thanks,
Dave

[Han]

(01-08-2014 03:01 AM)David Hayden Wrote:  It appears to me that the System RPL entry %NFACT on my 50g doesn't work properly. This computes the factorial of a real number. Note that it's different from %FACT which computes gamma(x+1) and works correctly.

%NFACT appears to return IP(ABS(x))! There are 2 bugs

1. It doesn't honor system flag 22 when the result overflows. It returns 9.99999999999E499, even if the flag is clear
2. If x is between -1 and 1 (exclusive of the endpoints), it return 9.99999999999E499 instead of 1.
   

Hoping that someone can confirm or explain to me why I'm wrong.

Thanks,
Dave

Sounds like this bug I reported a while back: http://bugs.hpcalc.org/show_bug.cgi?id=255

This bug affects pretty much ANY command that uses argument checking and auto-converts DOZINT types to DOREAL -- which is to say a lot of commands are affected by it. I found the bug while porting Jazz for the HP48 series to the HP50G.

[David Hayden]

(01-08-2014 03:14 AM)Han Wrote:  Sounds like this bug I reported a while back: http://bugs.hpcalc.org/show_bug.cgi?id=255

I don't think so. %NFACT is a primitive code object and operates on real numbers only. It doesn't call CK&DISPATCH1

[Han]

If I am not mistaken, is flag 22 is for cases when you truly have infinity (such as division by 0), whereas flag 21 is for overflow. %NFACT does respect flag 21.

As for the second issue, it has to do with the fact at %NFACT (and pretty much any other SysRPL command) does not do any argument checking. That it works for almost any decimal value is a mere bonus due to the algorithm implemented. Basically the algorithm checks a few special cases and then proceeds with the general algorithm, which fails for the case when the input \( x \) satisfies \( x\in (-1,1) \). Perhaps you are expecting it to return the factorial of any real number, whereas the program is designed to return the factorial of an DOREAL integer.

[David Hayden]

(01-08-2014 06:12 PM)Han Wrote:  If I am not mistaken, is flag 22 is for cases when you truly have infinity (such as division by 0), whereas flag 21 is for overflow. %NFACT does respect flag 21.

Ah! That's it. Thanks so much.

(01-08-2014 06:12 PM)Han Wrote:  Perhaps you are expecting it to return the factorial of any real number, whereas the program is designed to return the factorial of an DOREAL integer.

Yes I think that's fair. A good way to look at it is that %NFACT expects a DOREAL integer argument greater than zero. The fact that it returns something sort-of meaningful for DOREAL integers less than -1 is just coincidence.

Thanks for your help!
Dave