On Sun, Feb 16, 2003 at 20:57:29 -0800, Kris Kennaway wrote: > On Sun, Feb 16, 2003 at 07:52:35PM -0800, Andrey A. Chernov wrote: > > > So, monotonically increased seed->first value correlation problem remains... > > I think we should commit this patch (to -current) and fix all the > problems that pop up. For example, it's used in awk (which started > this set of changes), and in some of the XFree86 libraries.
I agree. (diff is for old rand.c, but idea is clear in anycase). I have no fancy ideas how to fix correlation problem AND keep rand_r() compatibility at the same time. All linear mod-type generators share this problem with monotonic seeding and usual solution is shuffling array (more complex code than I use), but it will be incompatible with rand_r() again. If we back out most of changes and return to very first formulae with overflow, we 1) Make monotonic increasing for seed->first_value less visible. 2) Not fix seed->first_value correlation (taking some parts of bits will show repeated pattern). 3) Make distribution and lower bits bad. -- Andrey A. Chernov http://ache.pp.ru/
msg52528/pgp00000.pgp
Description: PGP signature
