[prev in list] [next in list] [prev in thread] [next in thread] 

List:       kde-devel
Subject:    Re: [despammed] Re: Open Bugs in 3.2
From:       Nicolas Goutte <nicolasg () snafu ! de>
Date:       2004-01-12 18:26:28
Message-ID: 200401121926.29596.nicolasg () snafu ! de
[Download RAW message or body]

On Monday 12 January 2004 17:58, Esben Mose Hansen wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On Monday 12 January 2004 00:32, Joachim Eibl wrote:
> > Check this on your machine.
> >   long double a=49.84;       // double constant
> >   printf("%2.30llf\n",a);        // 49.840000000000003410605131648481
> >   long double a1=49.84L;   // long double constant
> >   printf("%2.30llf\n",a1);      // 49.840000000000000000138777878078
> >
> > This just illustrates how tricky it can be.
> > BTW: There is no real general solution. You might just settle for a
> > compromise like showing fewer digits. But it won't help you when you
> > start subtracting almost equal numbers like: 1000.0000 - 1000.0001 etc.
>
> The general solution is: don't convert to binary floating point. Keep it in
> decimal floating point (as vector<char> (decimals)+ long (decimal point) or
> whatever) and do the calculations there. Not something I would propose for
> KCalc at this stage, though. There would probably be (L)GPL libraries for
> this anyway.

Someone is planning to do a GnuMP plugin for KCalc. Then KCalc would calculate 
in decimal. There is no need to reinvent the wheel, especially as GnuMP seems 
to be optimized (special algorithms and assembler codes.)

>
> Why not simply limit KCalc to e.g. 13 significant digits? That's what
> calculators used to do, and it works fine 99.999% of the time ;-)
>
> - --
> regards, Esben
>
> Homepage: http://www.mosehansen.dk
> Signature fingerprint at http://www.mosehansen.dk/about
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.2.3 (GNU/Linux)
>
> iD8DBQFAAtY1rfnftt13wXIRAlmRAJ9u++Msy+iScvJBoPmdk7D68EhPmQCeMKZF
> nrjpN6NY5t8KPqKgiPLktzk=
> =AHZE
> -----END PGP SIGNATURE-----
>
> >> Visit http://mail.kde.org/mailman/listinfo/kde-devel#unsub to
> >> unsubscribe <<

 
>> Visit http://mail.kde.org/mailman/listinfo/kde-devel#unsub to unsubscribe <<
[prev in list] [next in list] [prev in thread] [next in thread]