On 07/12/2022 14:04, Stefan Ram wrote: > r...@zedat.fu-berlin.de (Stefan Ram) writes: >> So, in this case, careful code reviews might be better than >> tests. For example, assuming, random.intrange( 0, 2 ) works >> as advertised, we can be pretty sure that >> 0 if random.randint( 0, 2 ) else 1 >> fulfills the requirement. > > In practice, tests should still be done. When such a function > returns the same result a hundred times, it does make sense > to become suspicious and investigate it.
Whatever it does a hundred times is not enough. Code review is definitely not enough. The design of PRNGs should be good in theory --- designed with good mathematical foundations --- and in practice by passing all tests in the best-designed batteries. As far as I know, the state-of-the-art in statistical tests against PRNGs is the TestU01 library, available at http://simul.iro.umontreal.ca/testu01/tu01.html -- https://mail.python.org/mailman/listinfo/python-list
