[Shootout-list] Stuff (benchmark design)

[Sebastien Loisel]

> > if the design of the benchmark allows something,
> > then this solution should be allowed to be the
> > ''real'' solution. it should not be judged
> > unfit, becuse it is different to how other languages

> Yes -- with the one exception of 'same way' tests
> that require the same data structures and algorithms.

For my numerical tests, I am generally not concerned whether the map
data structure is implemented using balanced binary trees, hash
tables, 2-3 trees or whatever. However, there are often significant
mathematical shortcuts that should not be taken.

For instance, it is possible to obtain 10 digits of precision for
spectral norm without going to a 10,000x10,000 matrix (in other words,
one can get the precision of 10,000x10,000 without actually doing
10,000x10,000.) This should not be allowed.

If the algorithm calls for gradient descent to compute the roots of
polynomials, it isn't okay to replace it with Durand-Kerner. Whether
your gradient descent is lazy or not, I don't care.

To avoid temptation, I try to write things that can't be easily coded
around, but this sometimes clashes with the desire for short programs
and so can't always be achieved.

Other than that, have a ball.

Sebastien Loisel