[boost] Re: [rational] Announcing new maintainer -- call for feature requests

From: Søren Lassen (s.lassen_at_[hidden])
Date: 2005-04-06 11:47:25

Hi!

Just a couple of words to introduce myself: I am from Denmark, but  
currently living in Nicaragua, where my wife is employed at the Danish  
embassy. I suppose this means that my current job title is "spouse".  
Before moving, I made a living as a Data Warehouse programmer/analyst.  
Having more time on my hands, I decided to spend some of it learning C++.  
In my experience, one of the best ways to learn a programming language is  
by looking at (and possibly trying to improve) existing code, preferably  
of good quality. Which is why I am now showing up on the boost mailing  
list.

I have spent some time playing around with the rational.hpp file, and have  
made some improvements. Most important is probably operator<, where my new  
version is completely safe from overflow and (I think) considerably  
faster. Other features are: a member template for converting between  
different types of rationals, functions inverse() and power().

Should I put my version of rational.hpp on the sandbox CVS, or mail it to  
somebody - or is it not relevant at this time?

Anyway, considering the call for new features: Maybe the class should have  
a representation of +-infinity and NAN (the latter follows logically from  
the former, because infinity*0->NAN). There may be quite a lot of cases  
where it is not the best solution to throw exceptions for divisions by  
zero, e.g. a spreadsheet or calculator application, where the user may do  
all kinds of funny things - when a division by zero occurs, we just want  
to mark the affected variable as infinite, and write "DIV/0" or something  
similar in the relevant space in the application UI. The easiest way to  
accomplish this is to implement it as a selectable feature in the rational  
class, rather than checking for zero before each division, or setting up  
try...catch blocks around all the division operations. Of course, the  
introduction of NAN rationals makes the class only partially ordered (if  
we follow the floating-point standard, where all comparison operations  
involving NAN types return false).

---
Søren
P.S. Have I posted this reply to the call for features in the right way?  
If not, what is the right way, using Opera for newsgroups and mail?