Re: [RFC][VOTE] RNG fixes

On 09.07.2016 at 10:49, Pierre Joye wrote: > On Jul 9, 2016 3:19 PM, "Leigh" <leight@gmail.com> wrote: >> >> On Sat, 9 Jul 2016 at 08:48 Pierre Joye <pierre.php@gmail.com> wrote: >>> >>> So, I voted no then as it is clear that you do not see a problem to >>> break codes without a single warning or time to adapt before. >>> >>> The other sections are fine and voted yes. >> >> For the part where you voted no. Still nobody has presented (publicly >> available) source that makes legitimate use of mt_srand (yes it's mt_srand >> that is "broken" here, not mt_rand) for deterministic streams of random >> numbers. I can only assume by this that almost nobody does. However, for >> those that do, they can still use the old algorithm. > > I am sorry but this PR possibly breaks BC and cases have been clearly > explained how and why. Asking to show production code publically is not > something that you should request. ACK. However, it appears to me that it has not been sufficiently verified that the random distribution of the current mt_rand() implementation is as good as the original algorithm (apparently, there are only some demos and quick investigations available). Therefore fixing this in a minor version with the option to enforce the old behavior looks good to me. -- Christoph M. Becker