Library does not work with more than 38 digits.

[blackjetrock]

No description provided.

[blackjetrock]

I would like to use this library with 200 digits, 100 before the decimal point and 100 after. That would be equivalent to 1E99 to 1E-99 and allows me to use it for calculators with that range of numbers. At the moment it doesn't work above 38 digits, and there is a comment to that effect in the code.
Is it known what the reason for the 38 digit limit is?

[edwig]

Have a look at bcd.h at lines 52-56. There are two constants that control this behaviour:
bcdDigits : The number of digits per integer (here 8)
bcdLength: The number of integers in the mantissa (here 5)
Currently the values are so that the maximum (unrounded) number of mantissa = 8 * 5 = 40 digits (38 rounded)
To increase the mantissa considder to increase the bcdLenght to more integers. For 200 that would be 200/8 = 25.
And then you should be able to use mantissa's of 200 digits.

[blackjetrock]

Hello,

I have done this and unfortunately the code crashes. I have ported the code to the RP Pico and above 38 digits I get an error that I don't understand yet. I also expected it to work for any number of digits. The comment that mentions a rash is actually one that I put in the code. the problem is in DividePositive but isn't a divide by zero. I think it's some form of overflow of the digits outside the mdata array bounds. Maybe.

[edwig]

Hello blackjetrock,

The proof of the pudding is in the eating.
So…..
I checked in an extra class “bcd200” with a 200 digits mantissa.
Pull the latest from https://github.com/edwig/bcd
The unit tests of bcd are copied to UnitTestsBCD200.

Besides the bcdLength the following changes were made

• Static PI() -> Larger precision
• Static LN2() -> Larger precision
• Static LN10() -> Larger precision
• The unit tests ‘expected’ variables now all have a higher precision

All unit tests of bcd200 run in the green!

Have fun!