Classic, Woodstock and Spice Emulators

[Willy R. Kunz]

(06-19-2016 07:59 PM)PANAMATIK Wrote:  

(06-19-2016 05:43 PM)Dieter Wrote:  Willy, are you out there? If it's you who wrote this, could you please explain why you think a "real HP67" is wrong here? Maybe I didn't understand you correctly.

Dieter

The HP-67 has 10 mantissa digits internally. The result of 1/(9 E+87) is
1.111111111000.... E -88
times 8 is
8.888888888000.... E -88

The comment assumes , that there are more internal mantissa digits present, which is obviously only true in the RPN-67 simulator. The original HP-67 is not wrong, but shows the limit of its internal precision.

Bernhard

Thank you, Bernhard, that's the correct explanation. If you're running my 67/97 products in what I call "HP-67 accuracy", I limit values to what the display can hold by essentially producing the 0's you show above. Only in high-accuracy mode the invisible digits are left in place and are used for rounding.
I found out the hard way that having only high accuracy would break many original HP programs. For example, one program would repeatedly multiply a value by 10 until enough 0's had been shifted in to return TRUE when testing for zero.

Willy

Addendum:

Interestingly, the HP-65 Diagnostic Program (STD-18A) creates the -8.888888889E-88 check number by actually computing the TAN of -5.092958178E-86°. The correct value is -8.888888887 (2471)E-88, so the last digit is indeed 7, not 8. The HP-67, on the other hand, computes TAN correctly, so HP had to take a different approach to produce the desired number. (TAN(-5.092958179E-86) returns -8.888888889E-88, so that's no help either.)

As an aside, Free42 returns -1.158028306E-34 for the tangent, although the sine (-8.888888887E-87) and the cosine (1.00000000) are correct.

And TAN(-5.092958178E-86°) simply crashes WP34s... (version 3.3, both on iPhone and iPad).