From inside of awk, I want to generate a string of X alphanumeric characters reasonably random (i.e., random but not cryptographic) on demand and rapidly.

In Ruby, I could do this:

ruby -e '
def rand_string(len, min=48, max=123, pattern=/[[:alnum:]]/)
    rtr=""
    while rtr.length<len do
        rtr+=(0..len).map { (min + rand(max-min)).chr }.
            select{|e| e[pattern] }.join
    end                     # falls out when min length achieved 
    rtr[0..len]
end

(0..5).each{|_| puts rand_string(20)}'  

Prints:

7Ntz5NF5juUL7tGmYQhsc
kaOzO1aIxkW5rmJ9CaKtD
49SpdFTibXR1WPWV7li6c
PT862YZQd0dOIaFOIY0d1
vYktRXkdsj38iH3s2WKI
3nQZ7cCVEXvoaOZvm6mTR

For a time comparison, the Ruby can produce 1,000,000 unique strings (no duplicates) in roughly 9 seconds.

Taking that, I tried in awk:

awk -v r=$RANDOM '
# the r value will only be a new seed each invocation -- not each f call
function rand_string(i) {
    s=""
    min=48
    max=123
    srand(r)
    while (length(s)<i) {
        c=sprintf("%c", int(min+rand()*(max-min+1)))
        if (c~/[[:alnum:]]/) s=s c
    }
    return s
}
BEGIN{ for (i=1; i<=5; i++) {print rand_string(20)}}'

That does not work — same seed, same string result. Prints:

D65CsI55zTsk5otzSoJI
D65CsI55zTsk5otzSoJI
D65CsI55zTsk5otzSoJI
D65CsI55zTsk5otzSoJI
D65CsI55zTsk5otzSoJI

Now try reading /dev/urandom with od:

awk '
function rand_string(i) {
    arg=i*4
    cmd="od -A n -t u1 -N " arg " /dev/urandom"  # this is POSIX
    #             ^  ^                unsigned character
    #                   ^  ^          count of i*4 bytes
    s=""
    min=48
    max=123
    while (length(s)<i) {
        while((cmd | getline line)>0) {
            split(line, la)
            for (e in la) {
                if (la[e]<min || la[e]>max) continue
                c=sprintf("%c", la[e])
                if (c~/[[:alnum:]]/) s=s c
            }
        }
        close(cmd)
    }
    return substr(s,1,i)
}
BEGIN {for(i=1;i<=5;i++) print rand_string(20) }'

This works as desired. Prints:

sYY195x6fFQdYMrOn1OS
9mv7KwtgdUu2DgslQByo
LyVvVauEBZU2Ad6kVY9q
WFsJXvw8YWYmySIP87Nz
AMcZY2hKNzBhN1ByX7LW

But now the problem is with the pipe od -A n -t u1 -N " arg " /dev/urandom is is really slow — unusable except for a trivial number of strings.

Any idea how I can modify one of those awks so that it:

1. Runs on most platforms (i.e., default POSIX kit);
2. Can produce reasonably random strings of X length rapidly.

This question has been asked a few times:

1. How can I replace a string with a random alphanumeric string 48 characters long using awk where the answer is use external tools — too slow;
2. Substitute given pattern with a random one with awk but that is a random int and does not use srand;
3. Execute a command (to generate random strings) inside awk but again uses shell pipe (too slow) and Linux only.