> You should read these:
be assured: i did ... not all of them to the full extent, but most of them multiple times ... i do! have a understanding of fp-math,
> you can't understand why numeric calculations behave as they do,
be assured: i can differentiate 'behave as they do' between 'as they actually do in M$ Excel', 'as they actually do in LO Calc', 'as they actually do in gnumeric' and partly 'as they could behave in pure-fp' and partly 'as they could behave in pure-fp with meaningful corrections'. i'm quite sure that 'crossover' between roundup and rounddown can be avoided and that 0,30000000000000004 can be rounded to 16-digit 0,3 with IEEE 754 rounding mode 'ties to even'.
the base of the cited documentation is about 35 years old now, and it's not! pointing out or prooving that it's impossible to gain better results with meaningful calculations, but explaining which accuracy is reached with standard-fp-math. it's repeated and repeated and repeated again for decades, and time for something new to come.
i'd like to use gnumeric to develop, test and show that, as it's originally already targetted and aiming for precise calculations and provides good access to the real values. my question is about the improvements i would like to propose in accordance with the customs in the code of gnumeric to avoid unnecessary inconveniences. e.g. what to do in goffice, what in gnumeric, which system standards and constants to use and such ...
best regards,
b.
Gesendet: Sonntag, 04. Juli 2021 um 14:42 Uhr
Von: "Steven D'Aprano" <steve pearwood info>
An: gnumeric-list gnome org
Betreff: Re: deco-Math project, step 00_a: exact bin and dec 'ranges' (in gnumeric).
