[Solved] How to improve the uniformly distributed of a given random function to generate uniformly distributed numbers? [closed]

Question

what can be improved on a given random function to make it more random or for a bigger range or something else?

and therefore no SRANDOM can be used up to now.

How to improve the randomness of the fuction above, if possible ?

Sooo write your own SRANDOM with your own semantics. Ex:

srandom() {
   # take random number from /dev/urandom
   # we take only just 4 bytes - one 2^32 number
   printf "%d\n" "0x$(
       dd if=/dev/urandom of=/dev/stdout bs=4 count=1 status=none |
       xxd -p)"
}

and then:

normalize_value(){
   ...
   rnd=$(srandom)
   rnd_count=$((rnd / ...))
}

Accepting a wider range of numbers

If you are not happy with the way shell arithmetic expansion works, then… use a different tool. bc calculator has unlimited range.

rnd_count=$(echo "
     # see https://superuser.com/questions/31445/gnu-bc-modulo-with-scale-other-than-0
     scale=0; 
     # super big random number from three 2^32 numbers
     rnd = $(srandom) * 2^(32*2) +  $(srandom) * 2^32 + $(srandom)
     rnd % ($max - $min + 1) + $min
     " | bc)

You can write your own C program with getrandom() and compile it on the fly echo "int main() { stuff(); }" | gcc -xc - && ./a.out; rm ./a.out basically granting you any semantics you want. There are also other scripting languages, like perl, python, ruby, all most probably with their own big-number libraries and urandom number generation implementations. Sky the limit.

Every improvement should use bash only.

Is from my perspective a pointless limitation – overall, I am paid for results, not really “how” I solve problems. Anyway, you could, giving you a bunch of ideas how to proceed:

First write a function that would read from /dev/urandom and convert the bytes into a number.
- I have no good idea how to do it in pure bash while keeping the randomness at sane levels. I suspect the input will drain fast.
- You could read one byte from urandom. You’ll have to ignore read exit status, cause the byte may be zero byte or newline.
- Then check if that byte is a digit. If it’s not, repeat previous step.
- Treat such algorithm treat as a generator of random number within the range of 0-9. Build bigger numbers from those digits.
Then develop your own big-number library using arithmetic expansion as a “backend”, written in bash.
- Seems pretty pointless, because bc is commonly available.
- This would work as usual big number libraries do.
- I suggest to store the number as an array of number at max 2^16. For inspiration, research similar libraries written in C and C++ languages, and convert it to bash.

Accepted Answer