Re: [boost] Rational library update (at last!)

From: Paul Moore (gustav_at_[hidden])
Date: 2001-02-07 15:27:04

On 5 Feb 2001, at 17:01, Beman Dawes wrote:

> At 08:28 PM 2/5/2001 +0000, Paul Moore wrote:
> 
>  >I've uploaded a new version of the rational numbers library to the
>  >boost vault (folder rational, file rational_2001-02-05.zip).
> 
> I ran the Win32 regression tests after Jens updated the CVS.  Borland, 
> Metrowerks, Microsoft, and Microsoft/STLport compilers worked on both 
> rational_test and rational_example.
> 
> GCC fails twice, and Intel fails once.  See logs below

Jens already picked up on, and fixed, one of the gcc failures (the 
library supplied with 2.95.2 doesn't have ios_base). The other is a 
gcc library issue again, affecting the test suite - there is no 
stringstream, so the IO tests don't work. They can be hacked to 
work with sstreams, which is what I did temporarily while I was 
testing. Hmm, I suppose a longer-term solution may be to include 
either sstream or stringstream, and typedef the one we have 
available.

Is there a config flag saying whether stringstreams exist? There 
isn't one in the 1.20.1 release that I could find...

I'll have to look at the intel one...

> ** Intel C++ 5.0
> c:/boost/site/libs/rational/rational_example.cpp(66) : error: no suitable 
> conversion function from "boost::rational<int>" to "int" exists
>        assert(abs(minus_half) == half);
>        ^

This implies that abs() isn't finding the version which has a rational 
parameter, implying a Koenig lookup problem. That would probably 
be related to the "if _MSVC_VER" (or is it BOOST_MSVC in your 
version?) hack at the top of rational_example.cpp. There's a new 
config flag which says whether the compiler supports Koenig 
lookup - please change to use this and see if that helps. I can't use 
it as it's not in 1.20.1 and I don't have the CVS cersion.

> c:/boost/site/boost/rational.hpp(442) : error: class "boost::bad_rational" 
> has no suitable copy constructor
>            throw bad_rational();

Doesn't this say that bad_rational() isn't copyable? But it's derived 
from domain error, so that implies that domain error is *also* not 
copyable.

This seems like a library/compiler bug. I can't think how I would 
work around it...

Hope this helps, thanks for the test reports,
Paul.