Re: Use of php_mt_rand() rather than php_rand()

Hi Yasuo, Some times ago I mailed my idea about refactoring random number generator API with min BC breaks but I didn't create an RFC right now. see http://marc.info/?l=php-internals&m=137772363015217 Marc On 16.07.2014 07:13, Yasuo Ohgaki wrote:

Hi all, There are few places that uses php_rand() currently. https://bugs.php.net/bug.php?id=66718 http://lxr.php.net/search?q=php_rand&defs=&refs=&path=&hist=&project=PHP_5_5 These functions could use php_mt_rand() instead of php_rand(). php_rand() uses several rand functions 62PHPAPI long php_rand(TSRMLS_D) 63{

64    long ret;

65

66    if (!BG(rand_is_seeded)) {
67        php_srand(GENERATE_SEED() TSRMLS_CC);
68    }

69 70#ifdef ZTS

71    ret = php_rand_r(&BG(rand_seed));

72#else 73# if defined(HAVE_RANDOM)

74    ret = random();

75# elif defined(HAVE_LRAND48)

76    ret = lrand48();

77# else

78    ret = rand();

79# endif 80#endif 81

82    return ret;

83} Most systems use random() which has cycle of 16 * ((2^31) - 1), while MT rand has 2^19937-1. php_mt_rand() could be used where php_rand() is used. Unlike php_rand(), php_mt_rand() does not check if it is seeded or not. It should be changed to check seed status. The only BC issue I can think of is game that expects the same pseudo random sequences. These apps may use mt_srand() to get fixed random sequences and adjust their apps if it is needed. This is acceptable BC for new releases. IMO. It would be good idea to support 64bit version of MT on 64bit platforms, too. http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/emt64.html Any comments? -- Yasuo Ohgaki yohgaki@ohgaki.net