>>>>>>I am struggling implementing a code that would be able to add large numbers. Basically it is working, I am using the Math_unlimited.vcx from the download section. However once the result would be negative it gives wrong results. I would need to rewrite the code, but I was wondering if someone has a better code for this, or perhaps even an fll solution that could be used instead?
>>>>>
>>>>>How big of numbers do you need?
>>>>>
>>>>>I wrote an open source arbitrary precision math library that used MPFR several years ago in C/C++. It could be ported to an FLL with little difficulty. It provided basic trig functions, exponents, roots, etc.
>>>>>
>>>>>UPDATE: In fact, in thinking about it, I think I wrote a VFP wrapper at one point. Hmm...
>>>>
>>>>I have numbers that come close to 20 digits, no decimals, but could get negative. Actually it's for calculating hash codes for bank files, which is a relatively general task.
>>>
>>>I found the code. I don't remember what state it's in. You can look at it and test it. See attached.
>>>
>>>The way it works is you allocate a number inside the bignum.dll. It returns an id or handle. You use that handle to reference the number performing functions on it. When you're done, use the to_*() methods to retrieve the value.
>>>
>>>The maximum precision most computers can handle is from an 80-bit real, which is 17-18 decimals of precision in base-10. For ~20 you need to use a fixed integer most likely, and there probably are other libraries which are better for those kinds of calculations. As I recall, bignum would use the size you allocate in the new() method.
>>>
>>>new() -- Allocate a number
>>>free() -- Release it when you're done
>>>to_*() -- Retrieve the value into VFP
>>
>>Great, thanks a lot, I will review it and see if I can implement it.
>
>Ahem, are you certain vfp currency is not up to the job ? IIRC it is an 8-byte signend integer with an "arbitrary" set decimal point resulting in 4 decimals. So if you are certain you will not have any decimals, you can eliminate or "shift" the 4 decimals. And if it is only for hash codes, where probably the only questions asked are == and !=, why not implement in C and just save the result, either as 2 integers or currency or in this case my favorite varchar binary ? Last one can grow above 8 byte length if the collision area needs to be reduced by enlarging hash result abd function....
>
>curious
>
>thomas

Currency may be an idea, but I did not use a cursor where I could define the data type. So I suppose I could create a dummy cursor and then assign the variable based on that field. C would be probably the best solution, but I have not done that before. But since that is a simple requirement, it may be a good idea to use this as a starting point.