Unexpected randomseed results (was [ANN] uuid: pure Lua uuid generator)

> I'm looking at the code, and I think that seeding the random number

> generator is *not* something it should do---that should be left to the

> user

> to do.

You are completely right. It's something I added, but was bugging me already.

> -spc (What? Not version 3 or 5 UUIDs? [4])

Nope. I needed a simple library, to generate the uuids randomly, no standards compliance, but pure Lua.

I'll make sure to add a note to the documentation.

But to make matters worse, the library currently creates the exact same sequence of uuids every time I start it.

The seed used is `socket.gettime()*10000` which is far better than `os.time()` __unless you notice__ that math.randomseed [1] casts the number to an int. Which means that the former gets reduced to either 0 or 1 (on my system) and all randomness is completely gone. (I only found out by accident)

Here's some test output, left to right;

1) uuid,

2) seed used (doubled every next iteration),

3) return value of math.randomseed (see below),

4) socket.gettime()*10000

C:\Users\Thijs\Dropbox\Lua projects\uuid>c:\users\public\lua\5.1\bin\lua.exe seedtest.lua

008cd403-87d6-42a0-c22f-29dcbc07559a 4 4 13682539533420

00dc02f2-c951-4345-c36e-24de700c0bb8 8 8 13682539533430

007b5ed2-4e47-468e-c6ee-1be1d81676f3 16 16 13682539533440

01bb1792-5731-4b20-cbee-08e6a72a4c6a 32 32 13682539533450

013a8a12-6a07-4544-c5ed-e3f14653f859 64 64 13682539533480

03386e12-8fb2-498c-c9eb-990783a45036 128 128 13682539533560

06353611-da09-421d-c2e7-0533fe4600f0 256 256 13682539533670

0d2dc610-6fb7-433f-c3de-de8bf48b6164 512 512 13682539533730

1a1fe70d-9912-4482-c4ce-8e3bdf14214d 1024 1024 13682539533910

34022808-eec7-4807-c8ae-ef9cb627a31d 2048 2048 13682539533960

68c9abfd-9733-4f13-cf6c-b05d644ca5bf 4096 4096 13682539534100

d156b0e7-e90a-4e2a-ceea-33dec097aa02 8192 8192 13682539534150

a270bbbb-8eb9-4b59-cbe6-39e1792bb588 16384 16384 13682539534210

44a4d163-d817-44b5-c4dd-45e7e955ca93 32768 32768 13682539534260

880cfdb3-6bd1-486f-c8cc-5df3c9a9f3aa 65536 65536 13682539534420

10dc5554-9147-4fe2-cfa9-8d0b895046d9 131072 131072 13682539534490

207c0595-dd32-4dc9-cd64-ed3b0a9ded36 262144 262144 13682539534580

40bc6418-7509-4896-c8d9-ae9a0b3939f0 524288 524288 13682539534650

7f3d221d-a6b7-4f30-cfc4-2e590e71d364 1048576 1048576 13682539534690

ff3e9f27-0812-4e64-ce99-2ed714e0054c 2097152 2097152 13682539534760

fe40983b-cbc8-4bcc-cb43-30d41fbe6a1c 4194304 4194304 13682539534860

fd458a64-5235-469d-c698-32cc357b35bd 8388608 8388608 13682539534890

fa4e6fb5-5f0e-4b3e-cb41-37bd62f4cafe 16777216 16777216 13682539534910

f4603957-79c0-4580-c593-419fbce6f480 33554432 33554432 13682539534940

e884cd9b-ad24-4904-c937-556370ca4884 67108864 67108864 13682539534960

d0ccf523-15ec-410c-c17f-7debd892f08c 134217728 134217728 13682539534980

a05c4533-e57c-411c-c10f-cdfba822409c 268435456 268435456 13682539535010

407ce553-859c-413c-c12f-6d1b4842e0bc 536870912 536870912 13682539535030

80bc2593-c5dc-417c-c16f-ad5b888220fc 1073741824 1073741824 13682539535050

003ca513-455c-41fc-c1ef-2ddb0802a07c 2147483648 -2147483648 13682539535070

003ca513-455c-41fc-c1ef-2ddb0802a07c 4294967296 0 13682539535100

Truncation (2nd and 3rd column unequal) occurs already at a far lower number (2,147,483,648) than the socket-seed created (13,682,539,535,070) and from there on randomness is gone (same uuids are generated).

Regarding return value of math.randomseed(); I temporarily modified Lua source to return the seed set;

static int math_randomseed (lua_State *L) {
srand(luaL_checkint(L, 1));
//return 0;
lua_pushinteger(L, luaL_checkint(L, 1)); // return seed to make truncation visible
return 1;
}

I looked math.randomseed() up in the manual [2] which only states: “Sets x as the "seed" for the pseudo-random generator: equal seeds produce equal sequences of numbers.”

I think a note would be worthwhile here, indicating that it the seed gets cast into a (system specific size) integer, and if the seed is outside the range of the system specific integer, the actual seed used is a system dependent value that probably has the opposite effect of what one is trying to achieve when setting the seed (lack of randomness when randomness is expected)

Possibly including the change above in math_randomseed in the Lua sources could be used to make it visible from the Lua side whether the seed was actually properly set in these cases.

Thijs

PS. Is there a way to detect the maximum int size used from Lua?

[1] http://www.lua.org/source/5.1/lmathlib.c.html#math_randomseed

[2] http://www.lua.org/manual/5.1/manual.html#pdf-math.randomseed