From: Jeff Paquette (paquette_at_[hidden])
Date: 2000-11-20 11:34:11
PMFJI, but did anyone else notice the fix went from (stylistic arguments
aside):
if (den == 0)
to
if (r.num == 0)
?
Shouldn't the correct fix be:
if (r.den == 0) throw bad_rational()
--
Jeff Paquette
paquette at mediaone.net
http://www.atnetsend.net
> -----Original Message-----
> From: Beman Dawes [mailto:beman_at_[hidden]]
> Sent: Sunday, November 19, 2000 5:36 PM
> To: boost_at_[hidden]; boost_at_[hidden]
> Subject: RE: [boost] Bug in boost/rational.hpp
>
>
> At 09:57 PM 11/19/2000 +0000, Paul Moore wrote:
>
> >Good point. I missed this case. And thanks for the clear and
> detailed bug
> >report.
> >
> >I'd rather add a check as
> >
> >template <typename IntType>
> >rational<IntType>& rational<IntType>::operator/= (const
> rational<IntType>&
> >r)
> >{
> > // Trap division by zero
> > if (r.num == 0) throw bad_rational();
> > // Avoid overflow and preserve normalization
> > IntType gcd1 = gcd<IntType>(num, r.num);
> > IntType gcd2 = gcd<IntType>(r.den, den);
> > num = (num/gcd1) * (r.den/gcd2);
> > den = (den/gcd2) * (r.num/gcd1);
> > return *this;
> >}
> >
> >I.e., test at the top, on the pinciple of avoiding unnecessary work. Why
> >calculate two GCDs when you won't use them?
> >
> >A question - should this throw bad_rational() or is there a standard
> divide
> >by zero exception? (Couldn't find one in a brief glance in the
> standard...)
> >
> >Beman, could you add this or would you like me to send a complete
> >replacement rational.hpp?
>
> CVS changed. bad_rational() seems OK to me. Paul, if you want write
> access to CVS, let me know.
>
> It's always great when boost developers can respond promptly to bug
> reports.
>
> Thanks,
>
> --Beman
>
>
>
>