Undocumented change to Mersenne Twister breaks repeatability

[Brian Gough]

Jamie Lokier writes:
 > I noticed that between gsl-0.6 and gsl-0.7, the mt19937 generator was
 > changed to calculate the seed state differently and hence yield a
 > different random sequence from the same seed.  Furthermore, the change
 > isn't documented, either in NEWS, the ChangeLogs, or in the
 > documentation.

Not sure why you don't have them but there are entries for the updated
seeding procedure in the NEWS file and rng/ChangeLog of gsl-0.7, and
later versions.

 > I would have expected a new mt19937 generator to have a new name.  It's
 > too late now, but could the old generator be reinstated under a
 > different name, perhaps mt19937_1998?

I've made the old generator available in the current CVS version as
requested.

regards
Brian Gough