Re: [boost] [test] Buglet in floating point comparison code.

From: Gennaro Prota (gennaro_prota_at_[hidden])
Date: 2006-08-03 16:08:23

On Thu, 3 Aug 2006 19:14:19 +0100, "John Maddock"
<john_at_[hidden]> wrote:

>Apparently this was not quite as simple as I thought it was, because we 
>still need to handle the 0/0 case, the following patch does the trick:

Sorry to jump in at the latest moment. I wasn't following this thread
(it's interesting, but time is limited :-()

>cvs diff: Diffing boost/test
>Index: boost/test/floating_point_comparison.hpp
>[...]
>     }
>+    else if(f1 == 0)
>+    {
>+       // Avoid 0 / 0, since we're calculating relative error

Shouldn't the comment read 0/x or 0/anything?

>+       // the result really is zero here no matter what the denominator:

Unless the denominator is zero, in which case returning zero is a
convention.

>+       return 0;
>+    }
>+    else if(f2 == 0)
>+    {
>+       // We have an infinite relative error, in practice any
>+       // big number returned here will cause our test to fail:
>+       return 1000000; // any really big number will do.

Could you please clarify the context a bit? Would this, in general, be
called also when numeric_limits is not specialized? When a
specialization exists I would return (std::numeric_limits<>::max)();
when it doesn't, how can one be sure that an FPT can be constructed
from the integer 1000000? Sorry if this is just noise; I just thought
better *I* raise some noise now than some executable somewhere will do
it later :-)

--
[ Gennaro Prota, C++ developer for hire ]