CODEWITHSUNDEEP
randomlua
Wookai
Wookai

Reputation: 21773

Why is the first random number always the same on some platforms in lua?

Consider the following lua code snippet :

local time = os.time()
for _= 1, 10 do
    time = time + 1
    print('Seeding with ' .. time)
    math.randomseed(time)
    for i = 1, 5 do
        print('\t' .. math.random(100))
    end
end

On a Linux machine, the result is, as expected, random numbers. But it seems that at least on Mac OS X, the first random number after changing the seed is always the same !

I guess this is related to the fact that Lua relies on the C rand() function for generating random numbers, but does anybody have an explanation ?

EDIT: here is an extract of the output of the above code on a linux machine (ie the output is as expected) :

$ lua test.lua
Seeding with 1232472273
    69
    30
    83
    59
    84
Seeding with 1232472274
    5
    21
    63
    91
    27
[...]

On an OS X machine, the first number after "Seeding with ..." was always 66.

Upvotes: 5

Views: 5252

Answers (3)

orip
orip

Reputation: 75477

Lua's random used to use C's rand(3) and srand(3) functions (see here). UPDATE: newer Lua versions use random(3) where available.

Both the C90 standard and POSIX suggest an cross-platform implementation of rand and srand that isn't the best. It especially lacks randomness in the lower bits.

Some platforms like Linux moved off of the standard recommendation to a better implementation (e.g. random(3)).

OS/X remains true to the classic rand implementation, and Lua inherits it.

Upvotes: 4

Alexander Gladysh
Alexander Gladysh

Reputation: 41433

As others noted, Lua intentionally uses C90 random generator for portability sake -- and C90 RNG is not very good.

If you need good random numbers, use some Lua module to get it. For example, here is Mersenne Twister RNG binding by one of Lua authors.

Upvotes: 1

Gerald
Gerald

Reputation: 23499

It's generally a bad idea to call srand multiple times with seeds that are numerically close (and especially bad to do so with time values). In many cases, the variance of the first random number is similar to the variance of the seeds. When dealing with a scripting language that has to convert number representations, it can be even more so.

Does the same thing occur if you change your seed value by a larger amount?

Upvotes: 2

Related Questions