testing secnds.f

[Brooks Moses]

Dominique Dhumieres wrote:
> Tired to have intermittent failures with the testsuite secnds.f and
> secnds_1.f, I decided to have a look at them.  First, as it has already
> been reported, making the computation of dat1 and dat2 fully double
> precision (with _8 and real(*,8)) increase the number of failures.  Then I
> realized that SECNDS() somehow computes the difference between single
> precision arguments.  Thus it probably does not make sense to do
> comparisons with an accuracy better that the difference between the
> argument/result of SECNDS() and the nearest values.  So I ran the following
> modification of secnds.f for about half a day on a G5 under OSX 10.3.9
> without any failure (from 10PM to 11AM):

Thank you!

Please submit a formal patch for this, so I can approve it.  :)

- Brooks