evolution

[bearophile]

Andrei Alexandrescu:
> I plan to overhaul a number of modules in Phobos2, such as 
> std.algorithm, std.stdio, and std.random.

Regarding std.stdio: the current function writeln() has a large amount of faults. I have fixed about 20 of them in my put()/putr() and str()/repr() functions, but not all them. You can copy part of the code of them from my libs, and I can list some things in this newsgroup.

-------------

Regarding std.random: it has to reconcile a wide variety of purposes and needs, often one against each other:
1) It has to allow to write short & quick code;
2) It has to give random numbers very quickly;
3) It has to give high quality random numbers;
4) It has to be used in multi-threaded code.
5) It must be able to perform the most handy and common functions.

I think that can be solved adding 3 RND generators and various helper functions. Three different random generators can cover most usage cases: a very fast one, a good one fast enough, and a very good one.
- The very fast one is thread-safe, can be based on R250-521.
- The default one isn't thread-safe and it's designed for most cases. Not being thread-safe you don't need to create a struct/class to use it, so it allows a short and handy syntax. It can be Kiss used by Tango.
- The precise one can be thread-safe, a new version of Mersenne Twister is okay.

The helper functions can generate: random integer in closed interval, random integer in interval open on the right, uniform real value in interval, Gaussian real value, random choice among given items, in-place shuffling of sequence, not in-place shuffling, sampling with and without replacement from a given sequence of items. Many more things are possible, but those cover a lot of cases.

My d.random module is based on this design, even if few bits aren't finished yet (the R250-521 isn't integrated yet, there's a fault I have recently found in the randInt()).

Bye,
bearophile