Francesco Bochicchio: > Possibly to prepare the test data set you might need a > different - and already proven - implementation of > the algorithm.
Usually a brute force or slow but short algorithm is OK (beside some hard-coded input-output pairs). Sometimes you may use the first implementation of your code that's usually simpler, before becoming complex because of successive optimizations. But you have to be careful, because the first implementation too may be buggy. Some other times there are simpler algorithms to test if the output of another algorithm is correct (think of exponential algorithms with a polinomial test of the correct result), for example to test a fancy Introsort implementation you can use a very small and simple O(n^2) sorter, or better a simple linear loop that tests the result to be sorted. Here, beside unit tests, are also useful languages (or tools) that allow to add pre- and post-conditions, class invariants, etc. Bye, bearophile -- http://mail.python.org/mailman/listinfo/python-list
