On Thu, Feb 7, 2013 at 9:24 AM, John W. Eaton wrote:
> On 02/07/2013 09:15 AM, Mike Miller wrote:
>
>> The tests that are failing for me are chop, axis, betainv, normpdf, and
>> dec2base. I looked into them enough that I believe all of them are
>> failing because the value returned by std::pow is different when Java is
>> running.
EDIT: I contradicted myself here. I believe the chop, betainv, and
normpdf tests are failing due to std::pow. Dec2base fails because of
the subtraction example I showed above truncating the number of
significant bits. I am not sure about axis.
Might this be related to the problem reported previously with std::pow in libstdc++ when compiled C++11 mode? The issue is that overloads with integer exponent have been removed and the computation is always done as std::pow(double,double), which is subject to round-off issues.
Michael.