Re: [lmi] Problem of the week: testing a testing tool

[Greg Chicares]

On 2007-1-8 19:56 UTC, Ericksberg, Richard wrote:
[...]
>   double const w = std::min(std::log10(v), static_cast<double>(ULONG_MAX));
>   unsigned long int const x = static_cast<unsigned long int>(w);
>   double const volatile y = std::pow(10.0, static_cast<double>(x));
>   unsigned long int const z = static_cast<unsigned long int>(y);
[...]
> The erroneous output happens when the result of std::log10(v)
> is negative and then cast to an unsigned long int.
> Example [test using 0.001 for max_seconds_ and no mpatrol]:
> 
> [Overhead]
>           v: 0.0501144
>  trial_time: 0.0199543
> max_seconds: 0.001
>   frequency: 3192150000
> 
>   ULONG_MAX: 4294967295
>           v: 0.0501144
>    log10(v): -1.30004
>           w: -1.30004
>   cast of w: 4294967295
>           x: 4294967295
>           y: 1.#INF
>           z: 0
> 
> In this case, when the negative result of std::log10(v) is
> cast as an unsigned long int, it's getting the largest value
> of an unsigned long int [ULONG_MAX]. This happens when the
> integer value is one or greater. When the integer value is
> zero, you get 0 from the cast.

We've found at least one problem in a four-line block of code.

That block serves a single purpose, and it's kind of tricky to
get it right.

Above, you're exploring it by adding print statements--and
manually changing 'input_test.cpp' in order to create the
conditions you want to test. That demonstrates the problem at
hand. I think it'd be easier to fix it--and make sure it stays
fixed--if that block were a function with its own tests, so
I've made such a change in cvs. (Please review it--did I
reshuffle everything correctly?)

Now that we've isolated that aspect of the problem, how should
we fix it?