From: Moore, Paul (paul.moore_at_[hidden])
Date: 2002-03-12 04:13:27
From: Beman Dawes <bdawes_at_[hidden]>
> Let me try to answer this to make sure I understand Paul's
> position (so I can repeat it to the LWG).
> Yes, rational is parameterized on the integer type, and so
> can be used with any build-in integer or user-supplied class
> which meets the requirements.
> However, because of subtle rounding, precision, and other issues
> not well-known to most programmers, not all integer types are
> really suitable for most applications. In fact, the most
> suitable integer type would often be a unlimited-precision integer,
> yet neither Boost nor the standard currently have such a type.
> Thus even though rational is itself ready for standardization,
> some people might feel it would be more appropriate to wait
> until there is also an unlimited-precision integer class available
> for standardization.
> Paul, please correct me if I've got this wrong:-)
That's exactly right. The only point I'd add (as background) is that I don't
feel that there is any significant real-world experience to indicate how
likely the issues with using limited-precision integers would be to cause
problems in practice. This makes assessing the risks difficult.
Thanks for taking the time to get this all clear,
Boost list run by bdawes at acm.org, david.abrahams at rcn.com, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk