On Fri, 10 Jan 2014, Sven Schreiber wrote:
Hi,
an existing script of mine stopped working at the line:
set RNG SFMT
I suspect this is related to these changelog entries:
" Enable use of arrays in the SFMT random number generator"
" Internals: add support for multiple, independent PRNGs"
But is it a bug, or a deliberate change?
It's a deliberate change, but I should have added (and now will) an entry
to the backward-incompatibilities log. In revision 1.103 of
lib/src/random.c (2013-12-01) I removed the choice of the old GLib-based
version of the Mersenne Twister versus SFMT: the SFMT is now obligatory.
Yes, this is related to the changelog entries you mention. The multiple
RNGs thing -- see
http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/DC/dc.html -- is not yet
exposed to userspace because it's associated with parallelization work
that's not yet ready, but when I added dcmt I didn't want to have another,
old layer of optionality hanging around. (When we introduced SFMT we
initially kept GLib as an option for replication purposes, but that was
back in 2011 and I didn't think we needed it any more.)
Allin