Question

I am trying to generate random floats using nothing but bytes I get from /dev/urandom. Currently my best idea is to get the platform precision doing something like:

$maximumPrecision = strlen('' . 1/3) - 2;

and then construct a string of 0-9 in a loop the number of times $maximumPrecision tells us. For examle, if the precision is 12, I would generate 12 random numbers and concatenate them. I think it’s an ugly idea.

Update: Does this make sense?

$bytes =getRandomBytes(7); // Just a function that returns random bytes.
$bytes[6] = $bytes[6] & chr(15); // Get rid off the other half
$bytes .= chr(0); // Add a null byte

$parts = unpack('V2', $bytes);

$number = $parts[1] + pow(2.0, 32) * $parts[2];
$number /= pow(2.0, 52);

Answer 1

PHP’s float type typically is implemented as IEEE double. This format has 52 bit precision of mantissa, so in principle it should be able to generate 2⁵² different uniform numbers in [0, 1).

Therefore, you could extract 52 bits from the /dev/urandom, interpret as an integer, and divide by 2⁵². For example:

// assume we have 52 bits of data, little endian.
$bytes = "\x12\x34\x56\x78\x9a\xbc\x0d\x00";
//                                  ^   ^^ 12 bits of padding.

$parts = unpack('V2', $bytes);

$theNumber = $parts[1] + pow(2.0, 32) * $parts[2];  // <-- this is precise.
$theNumber /= pow(2.0, 52);                         // <-- this is precise.