Re: [boost] [math] RFC: comparison function objects

From: Andy Little (andy_at_[hidden])
Date: 2006-04-24 04:18:23

"Alberto Ganesh Barbati" <abarbati_at_[hidden]> wrote in message 
news:e2h7lb$mtt$1_at_sea.gmane.org...
> Andy Little ha scritto:
>> "Alberto Ganesh Barbati" wrote
>>
>>>  bool operator()(T x, T y) const { return abs(x - y) < eps; }
>>
>> Maybe should be:
>>
>>   bool operator()(T x, T y) const { return abs(x - y) < abs(eps); }
>>
>> Because I have been caught out when eps was negative ;-)
>> Negaitive eps is easy to do if it is some function of  x y or both, which is
>> common e.g percentage etc.
>>
>
> Thanks for pointing that out. I would prefer getting the absolute value
> of eps once and for all in the constructor, in order to avoid repeated
> calls to abs() if the functor is invoked several times. eps could be
> made private to improve data hiding.
>
> However, I am inclined to require eps to be positive as a precondition,
> leaving to the programmer the responsibility to take the abs() if he
> deems it necessary. I bet most of the times eps will be positive if not
> even a compile time constant, so this design would fit in the "don't pay
> for what you don't use" paradigm.

 Many times the "float error less than epsilon" test  is used to terminate a 
successive approximation routine, which will be an infinite loop if eps is 
negative. Does a negative eps trigger an exception?

regards
Andy little