RE: [boost] Boost Test Library: Bug in floating-point comparison?

From: Rozental, Gennadiy (gennadiy.rozental_at_[hidden])
Date: 2004-07-27 15:30:03

> Hi,
> 
> I want to use the boost::test floating-point comparison 
> algorithms (Boost 
> 1.31.0) and wrote the following simple lines of code:
> 
> double d1 = 0.2;
> double d2 = 0.8;
> double d3 = d2-d1-d1-d1-d1;  // d3 should be zero
>         
> bool b = boost::test_toolbox::check_is_closed( d3, 0., 1e-3 );
> 
> To my surprise, the return value b is FALSE, although d3 
> should be _very_ 
> close to zero (and is 5.55112e-17 on my machine).
> 
> Am I missing something or is this a bug in the test library?
> 
> Thanks,
>  Michael
> 
> -- 
> Michael Kettner
> kettner_at_[hidden] 

Read algorithm description. The algorithm compare "relative errors". In your
case it's 1 and infinity. Definitely more then 1e-03. It you want to compare
absolute values (which is not recommended in real life applications) you
either could write you own predicate or wait a bit, while I am considering
adding BOOST_IS_SMALL tool that will be responsible namely for this kind of
test.

Gennadiy.