RFC: rand() change, second try

Hi, [Impact and relevance: all functions that use randomness (including array_shuffle&co)] After the previous discussion, I've been thinking, and I think I've got a way which has only a neglectible BC problem (unlike my first proposal), while still keeping the following in mind: Probably 99% or more people who use rand() & co, simply want as random numbers as possible. In _this_ context, you can say that algorithms are exchangable. Most people don't know/don't want to know about seeding pseudo-random-generators, which algorithms to use, and that kind of stuff, they simply want random numbers. For the <1% people that want to exploit the property that you get the same "random" numbers after feeding the same seed, _do_ know about that. This proposal has these advantages: - Technical and syntactical possibility to add other sources of randomness: real random numbers (by means of device or internet-source)/other algorithms. No new functions needed in such cases. - Very easy to work with random numbers, this allows quick and easy coding with randomness without needing to worry about how random it is: PHP takes care of it, tunable by ini-entries. - Usually no need for srand(). - Still regression tests possible, with more relevant syntax - Seeding specific numbers and the choice of algorithm also influence functions like array_shuffle. - ? Disadvantages: - Minor break of BC, only if you AND expect a certain number after a certain seed AND you use various random-based functions through eachother. - A very small extra startuptime for each script-execution - Mixed use of algorithms a bit troublesome (no problem in case of serial mixed use) - ? Below I give some examples in stead of some technical explaination. Please let me know what you think of it, Jeroen -------------------------------------------------- I think I can illustrate my proposal best with some examples: [note: simplified sometimes, it's about functionality, not implementation] ========= [ini] random_generator = algorithm-x ; typical 'mt' (default) ======== // for simply random numbers: <?php file://seed the random number generator "algorithm-x" with a 'random' seed (1e6*microtime()) srand(); // (no parameters) echo rand(); // use "algorithm-x" ?> ======== // for simply random numbers: <!- on script start, implicit call to srand() (noargs) -> <?php echo rand(); // use "algorithm-x" ?> These two previous scripts are equivalent with: algx_srand( (float) microtime() * 1e6); agx_rand(); ======== // for a specific seed: <?php // state that you want to use MT with seed 1234 in all next calls to any random function srand(1234, RAND_MT ); echo rand(1,6); // throw with a dice, using mt. shuffle($array); // this also is affected ?> ======== // mixed <?php // state that you want to use the system rand() with seed 1234 in all next calls to any random function srand(1234, RAND_SYSTEM ); echo rand(1,6); // throw with a dice, using system rand(). shuffle($array); // this also is affected srand(); // back to default behaviour echo rand(1,6); // algx, seeded with 1e6*microtime() on previous line ?> ======== // also impact on other random functions <?php srand(456, RAND_MT ); array_shuffle($array); // This shuffles the same way each time! ?> ====== If you seed a specific seed, that's possible, but you need to specify the algorithm. (which can be extended!). This is simply because it wouldn't make much sense otherwise. Calling srand() (no-args) while a real-random source is selected, will have no effect. [mt_]srand($arg1) with one argument is deprecated. mt_* are aliasses for their equivelents, so no double set of functions anymore. In PHP 5, all this deprecated stuff should be removed. ====== // BC: goes OK srand(123); // equivalent to: srand(123,RAND_SYSTEM); echo rand(); ====== // BC: goes OK mt_srand(321); // equivalent to: srand(321,RAND_MT); echo mt_rand(); // alias of rand() ====== // BC: changed behaviour, but this is extremely unlikely // You don't mix up mt and system rand... mt_srand(321); // equivalent to: srand(321,RAND_MT); echo rand(); // will use MT, seeded with 321 srand(123); // equivalent to: srand(123,RAND_SYSTEM); echo mt_rand(); // will use SYSTEM, seeded with 123 ====== // BC: changed behaviour, but this is extremely unlikely mt_srand(321); // equivalent to: srand(321,RAND_MT); array_shuffle($array); // will eat up some random numbers echo mt_rand(); // not the first in the batch, since they have been eaten up ======