OK, but this should really be fixed "soon". We might want to consider
dropping this from 2.0 and adding a better version later. In any case, I
will submit a patch documenting current behavior.
One more thing that shows up in this method and elsewhere is poorly defined
behavior on integer overflows. Shouldn't we be checking for these and
throwing ArithmeticExceptions? If not, we should document (maybe at the
class level) that Integer arithmetic is being performed everywhere and
overflows will result in spurious values. My vote would be to add the
checks and throw ArithmeticExceptions on overflow. I will add this as a
separate patch.
Phil
 Stephen Colebourne <scolebourne@btopenworld.com> wrote:
> Bear in mind that this came from a contributer. I don't know the maths of
> it!
>
> I added the 25 limit, to avoid infinite recursion. The test cases
> demostrate
> that the method returns the correct result for many, many cases. But I'm
> sure it could be improved. For 2.0 I would suggest documenting behaviour
> rather than adding a new method to control the value '25'.
>
> Stephen
>
>
>  Original Message 
> From: "Phil Steitz" <phil@steitz.com>
> To: "Jakarta Commons Developers List" <commonsdev@jakarta.apache.org>
> Sent: Monday, August 11, 2003 4:24 AM
> Subject: Re: [lang] Fraction.getFraction(double) uses magic numbers
>
>
> > Brian S O'Neill wrote:
> > > I don't understand why the continued fraction implementation exists
> at
> all.
> > > Why not just get the bits from the double floating point number
> directly
> > > rather than introduce error? The floating point number is already a
> > > fraction, just encoded specially.
> >
> > The point as I see it is to get the best rational approximation of a
> > double value with bounded denominator. The continued fraction
> > decomposition will do this better and more efficiently than just
> > reducing the fraction implied by the decimal or binary representation
> of
> > the number. Consider, for example, the number 0.66666, to be
> > represented by a fraction with denominator <=10,000. The direct
> > approach using the decimal representation would give 6667/10000, which
> > is not as good as 2/3, which you would get by continued fractions.
> >
> > Phil
> >
> > >
> > >  Original Message 
> > > From: "Phil Steitz" <phil@steitz.com>
> > > To: "Jakarta Commons Developers List"
> <commonsdev@jakarta.apache.org>
> > > Sent: Sunday, August 10, 2003 02:48 P
> > > Subject: [lang] Fraction.getFraction(double) uses magic numbers
> > >
> > >
> > >
> > >>o.a.c.l.math.Fraction includes a getFraction factory method that
> takes a
> > >>double and uses continued fractions to find a fractional
> approximation
> > >>of the input. The continued fraction implementation has a hardcoded
> > >>maximum number of iterations (25) and maximum denominator (1000).
> These
> > >>should be documented (and the ArithmeticException if maximum
> iterations
> > >>is reached before convergence). Better (IMHO) would be to add
> another
> > >>version that takes these as parameters, possibly even replacing the
> > >>current method (I think this is new in 2.0, so there would be no
> problem
> > >>with backward compatability).
> > >>
> > >>If there are no objections, I will submit a patch that clarifies
> current
> > >>behavior and adds another method that takes maximum iterations and
> > >>maximum denominator as additional parameters.
> > >>
> > >>I would also like to improve the implementation, but this can wait
> until
> > >>after 2.0.
> > >>
> > >>Phil
> > >>
> > >>
> > >>
> > >>To unsubscribe, email: commonsdevunsubscribe@jakarta.apache.org
> > >>For additional commands, email: commonsdevhelp@jakarta.apache.org
> > >>
> > >
> > >
> > >
> > > 
> > > To unsubscribe, email: commonsdevunsubscribe@jakarta.apache.org
> > > For additional commands, email: commonsdevhelp@jakarta.apache.org
> > >
> >
> >
> >
> >
> > 
> > To unsubscribe, email: commonsdevunsubscribe@jakarta.apache.org
> > For additional commands, email: commonsdevhelp@jakarta.apache.org
> >
>
>
> 
> To unsubscribe, email: commonsdevunsubscribe@jakarta.apache.org
> For additional commands, email: commonsdevhelp@jakarta.apache.org
>
__________________________________
Do you Yahoo!?
Yahoo! SiteBuilder  Free, easytouse web site design software
http://sitebuilder.yahoo.com
