Linux vs. Windows: different behaviour [re rand()]

[Michele Dondi]

This spreaded from some tests I made in connection with the thread
"Fast random string generation":

# perl -le '$,="\t"; print unpack "C*", pack "L", rand 2**32 for
1..15'
30 155 146 154
155 211 255 102
39 119 228 119
204 29 168 240
3 100 77 39
159 11 214 76
5 105 14 31
167 198 182 196
247 107 241 244
194 246 30 146
226 199 219 59
97 168 246 154
185 204 221 142
178 77 149 4
70 10 119 34

C:\TEMP>perl -le "$,=qq|\t|; print unpack 'C*', pack 'L', rand 2**32
for 1..15"
0 0 246 13
0 0 174 202
0 0 104 241
0 0 116 246
0 0 10 76
0 0 4 44
0 0 64 61
0 0 244 123
0 0 252 86
0 0 78 181
0 0 124 45
0 0 100 103
0 0 188 62
0 0 100 100
0 0 196 191


Does the cmt at the end of 'perldoc -f rand' apply? Note: 5.8.4 in
both cases (AS under Windows).


Michele

 

[A. Sinan Unur]

This spreaded from some tests I made in connection with the thread
"Fast random string generation":

# perl -le '$,="\t"; print unpack "C*", pack "L", rand 2**32 for
1..15'

I am not sufficiently at ease with pack to see if something is wrong
above.

Does the cmt at the end of 'perldoc -f rand' apply? Note: 5.8.4 in
both cases (AS under Windows).

Same version here. You got me curious. I have:

C:\Home> perl -MConfig -e"print qq{$Config{randbits}}
15

I also checked stdlib.h for the free command line compiler they have:

C:\Home> cl
Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 13.10.3077 for
80x86

and it says

#define RAND_MAX 0x7fff

So those are consistent.

Then, I did:

use strict;
use warnings;


my @data;
push @data, rand for 1 .. 10_000;

open my $data, '>', 'data.txt' or die "data.txt: $!";
for (@data) { print $data "$_\n" }

__END__

Five times, and looked at the resulting emprical CDF in each case, they
were all straight lines from (0,0) to (100,100) so the output rand
generates is fairly uniformly distributed.

Dunno what you are observing but I don't think the bit about

If your rand function consistently returns numbers that
are too large or too small,

applies here.

Hope this helps.

Sinan.

 

[Sisyphus]

C:\Home> perl -MConfig -e"print qq{$Config{randbits}}
15

Or:
perl -V:randbits

On my Linux (Mandrake-9.1) I have 48 random bits. I don't know if this
accounts for the different behaviour. (I'm not real comfortable with
'pack' either

Cheers,
Rob

 

[Michele Dondi]

I am not sufficiently at ease with pack to see if something is wrong
above.

Neither am I, in the sense that I always have to check the docs, which
OTOH are clear enough. Well, at least for qw/C L/ I'm quite sure
nothing is consistently *wrong*.

Same version here. You got me curious. I have:

C:\Home> perl -MConfig -e"print qq{$Config{randbits}}
15

Me too! (verified with 'perl -V:randbits', as suggested by another
poster.) And 48 under Linux.

Dunno what you are observing but I don't think the bit about

If your rand function consistently returns numbers that
are too large or too small,

applies here.

Definitely I *think* it does, in fact, with referral to my other post:

| C:\TEMP>perl -le "$,=qq|\t|; print unpack 'C*', pack 'L', rand 2**32 for 1..15"
| 0 0 246 13
| 0 0 174 202
| 0 0 104 241
| 0 0 116 246
| 0 0 10 76
| 0 0 4 44
| 0 0 64 61
| 0 0 244 123
| 0 0 252 86
| 0 0 78 181
| 0 0 124 45
| 0 0 100 103
| 0 0 188 62
| 0 0 100 100
| 0 0 196 191

The high order bytes are always 0, and the third column values are
always even (of course this is *not* an artifact of the limited sample
- check for yourself!)


Michele

 

[A. Sinan Unur]

....

Definitely I *think* it does, in fact, with referral to my other post:

| C:\TEMP>perl -le "$,=qq|\t|; print unpack 'C*', pack 'L', rand 2**32
| for 1..15" 0 0 246 13
| 0 0 174 202
....

The high order bytes are always 0, and the third column values are
always even (of course this is *not* an artifact of the limited sample
- check for yourself!)

OK, so I did this:

printf "%8.8X\n", int rand(2**32) for 1 .. 10_000;

First with AS Perl:

C:\Home> perl -v

This is perl, v5.8.4 built for MSWin32-x86-multi-thread

Here is some representative output:

6F860000 EBF20000 0E140000 28600000 02C00000 75B00000
1D140000 93260000 25960000 643C0000 CACA0000 A2800000
20980000 A0820000 94860000 B61A0000 CDE40000 ...

On the other hand,

asu1@host ~
$ perl -v

This is perl, v5.8.5 built for cygwin-thread-multi-64int

And here is some output in that case as well:

8758274B 80717716 BBCCB917 26622DE8 05D10A8A 84B2174B
1BA68432 4F146432 59605227 88D18BA6 498D9C48 788227AD
F0892E9E 9FB09C8D D2E53F9F C56B388E 0D6A6B5D 0E8CCAA1
DF68A5F0 41E118A0 ...

I would say you are right.

So, would you like to file a bug report with ActiveState

Sinan.