I have questions remaining about efficient implementation of discard
for the random number engines. I would like a "10000" test for
independent_bits and for seed_seq. I'd like to take another look at
some of the log2 math in generate_canonical. But the system is quite
usable now and these can be tweaked as necessary in the tree.
I had some real fights with svn diff. Which eventually won -
probably by user error ;-). So here is a generic diff and a change
log.
Here's my stab a regenerating your patch.
I am testing this on darwin, results after lunch, and have finished
testing on x86_64. Looks good, one new fail:
testsuite/20_util/shared_ptr/thread/default_weaktoshared.cc
Uses -std=gnu++0x, but with the old mersenne_twister. Changing this to
mersenne_twister_engine gets into other issues, where number of
template parameters should be 12, not 14. Looking at N2798's
mersenne_twister_engine confirms. Thoughts?
-benjamin