List Commands Library for 50g

[pier4r]

Hmm. If you are referring to this https://groups.google.com/d/topic/comp.s...discussion discussion, I read it.

While JH Meyers is more often than providing great info, I learned the hard way that is not bad sometimes to see the data of the analysis (not only the conclusion, as one may see the data a bit differently than others). In any case, seeing the data that led to the conclusion helps to understand the conclusion themselves (in the worst case, one can disprove the conclusions).

Anyway what I got from that post is:
- there are better LCRNG than the one used in the RPL systems. Ok.
- the 15C has a better LCRNG. Ok.
- if one uses the rightmost digits, one may have very short periods. And you confirmed it. Confirming claims is healthy.
- if one uses IP after random, one _might_ - due to how IP works with digits - get sometimes a rounding errors due to the guard digits sneaking values that shouldn't be there. On this I am a bit skeptical. I may just run 16 RAND * IP until I see a 16 slipping in. If it is after some X million times, it doesn't matter much in a calculator.
- TL;DR: RAND is not random enough for cryptography applications. Ok. For all the rest, it seems good enough.

My perspective is:
- I take RAND as good enough. As selected by HP and used by thousands of people in thousands of calculators maybe millions of times. It can be that millions of people are wrong, but in this case it seems that RAND was good enough for them.
- I set to build a command to shuffle a list that, compared to a shuffle based on RAND, is mostly impossible to discern for practical purpose. For this I may go after a super refined metric, but since I don't want to encrypt anything to let the NSA out, I set a couple of metrics that should capture an idea of the difference between a list shuffled with RAND and one built with my routine. I would say that comparing the number of attempts against a goal list is not a bad start. (see my test code for example)
- If the metric somehow shows close values on a large amount of simulations, that's it. Until someone provides a better insight.

It is like life, one settles on a solution that doesn't look bad until a better one is found.

Then again, that is how I would approach the problem.