php.net |  support |  documentation |  report a bug |  advanced search |  search howto |  statistics |  random bug |  login
Bug #63174 Magic negative number generated by rand
Submitted: 2012-09-27 16:38 UTC Modified: 2017-10-24 09:41 UTC
Votes:3
Avg. Score:4.0 ± 0.8
Reproduced:2 of 2 (100.0%)
Same Version:0 (0.0%)
Same OS:0 (0.0%)
From: ub dot x7b8 at gmail dot com Assigned: nikic (profile)
Status: Closed Package: *Math Functions
PHP Version: Irrelevant OS: 32bit
Private report: No CVE-ID: None
View Add Comment Developer Edit
Anyone can comment on a bug. Have a simpler test case? Does it work for you on a different platform? Let us know!
Just going to say 'Me too!'? Don't clutter the database with that please !
Your email address:
MUST BE VALID
Solve the problem:
23 + 17 = ?
Subscribe to this entry?

 
 [2012-09-27 16:38 UTC] ub dot x7b8 at gmail dot com
Description:
------------
rand ( $min, getrandmax());
if $min < 0 then function generates number = getrandmax () - abs ( $min ) + 1 with very high probability.

REPRODUCED: PHP 5.2.6-1 Debian 5.0.10 i386 under VmWare
REPRODUCED: PHP 5.2.6-1 Debian 5.0.5 i386
REPRODUCED: PHP 5.2.3, 5.2.8, 5.2.9, 5.4.7 Debian 6.0.4 i386 under VirtualBox
NOT REPRODUCED: PHP 5.2.1 Solaris 5.10 sparc
NOT REPRODUCED: PHP 5.2.9 Solaris 5.10 sparc
NOT REPRODUCED: PHP 4.4.4-8 Debian 4.0 i386 under VmWare

Test script:
---------------
<?php
$rand_min = -getrandmax (); # 2137483647
$rand_max = getrandmax ();
$c = 1000;
for ( $min = $rand_min; $min < 0; $min+=10000000 ) {
	$diff = $rand_max - abs ( $min ) + 1;
	$cnt = 0;
	for ( $i = 0; $i < $c; $i++ ) {
		$super_random_value = rand ( $min, $rand_max );
		if ( $super_random_value == $diff ) $cnt++;
	}
	$per = $cnt * 100 / $c;
	echo "Number '$diff' was generated $cnt times, " . $per . "%, minimum = $min\n";
} 
?>

Expected result:
----------------
Number '1' was generated 0 times, 0%, minimum = -2147483647
Number '10000001' was generated 0 times, 0%, minimum = -2137483647
Number '20000001' was generated 0 times, 0%, minimum = -2127483647
Number '30000001' was generated 0 times, 0%, minimum = -2117483647
Number '40000001' was generated 0 times, 0%, minimum = -2107483647
...skipped....
Number '2100000001' was generated 0 times, 0%, minimum = -47483647
Number '2110000001' was generated 0 times, 0%, minimum = -37483647
Number '2120000001' was generated 0 times, 0%, minimum = -27483647
Number '2130000001' was generated 0 times, 0%, minimum = -17483647
Number '2140000001' was generated 0 times, 0%, minimum = -7483647

Actual result:
--------------
Number '1' was generated 514 times, 51.4%, minimum = -2147483647
Number '10000001' was generated 507 times, 50.7%, minimum = -2137483647
Number '20000001' was generated 522 times, 52.2%, minimum = -2127483647
Number '30000001' was generated 520 times, 52%, minimum = -2117483647
Number '40000001' was generated 502 times, 50.2%, minimum = -2107483647
...skipped...
Number '2100000001' was generated 12 times, 1.2%, minimum = -47483647
Number '2110000001' was generated 14 times, 1.4%, minimum = -37483647
Number '2120000001' was generated 10 times, 1%, minimum = -27483647
Number '2130000001' was generated 11 times, 1.1%, minimum = -17483647
Number '2140000001' was generated 2 times, 0.2%, minimum = -7483647

Patches

rand-range-double (last revision 2015-07-15 22:32 UTC by cmb@php.net)

Add a Patch

Pull Requests

Add a Pull Request

History

AllCommentsChangesGit/SVN commitsRelated reports
 [2015-07-08 21:50 UTC] yohgaki@php.net
It seems working well on 3v4l.
http://3v4l.org/BqqSr
 [2015-07-09 15:52 UTC] bishop@php.net
@yohgaki That's because 3v4l runs 64-bit: http://3v4l.org/a8KaY

This problem only manifests when there's overflow on 32-bit platforms.
 [2015-07-09 15:58 UTC] bishop@php.net
-Assigned To: +Assigned To: bishop
 [2015-07-09 16:12 UTC] bishop@php.net
Duplicated by #70003.
 [2015-07-09 20:56 UTC] yohgaki@php.net
-Operating System: +Operating System: 32bit
 [2015-07-13 19:52 UTC] bishop@php.net
-Status: Assigned +Status: Verified
 [2015-07-15 22:32 UTC] cmb@php.net
The following patch has been added/updated:

Patch Name: rand-range-double
Revision:   1436999569
URL:        https://bugs.php.net/patch-display.php?bug=63174&patch=rand-range-double&revision=1436999569
 [2015-07-15 22:49 UTC] cmb@php.net
> This problem only manifests when there's overflow on 32-bit
> platforms.

ACK. However, on 64bit platforms there's a closely related problem
if $max-$min > PHP_INT_MAX, in which case rand() unevenly
distributes, e.g. running

    <?php
    for ($i = 0; $i < 1000000; $i++) {
            if (rand(PHP_INT_MIN, PHP_INT_MAX) > 0) {
                    echo "Never\n";
                    break;
            }
    }
    ?>
    
never prints "Never". The problem is in RAND_RANGE() where only
the offset is calculated as double, instead of the whole
expression (see the attached patch "rand-range-double"). This
change would also fix the behavior on 32bit architectures.

However, changing the results of (mt_)rand() for $max-$min >
PHP_INT_MAX and with regard to inexact double to integer
conversion on 64bit architectures would be a BC break, so I'm not
sure whether we should apply that for PHP 5. IMHO, changing PHP 7
in this regard would be acceptable even though it's in feature
freeze.

Anyway, I would suggest to better document the behavior for non
default ranges.
 [2017-10-24 05:46 UTC] kalle@php.net
-Status: Verified +Status: Assigned
 [2017-10-24 06:03 UTC] kalle@php.net
-Status: Assigned +Status: Open -Assigned To: bishop +Assigned To:
 [2017-10-24 09:41 UTC] nikic@php.net
-Status: Open +Status: Closed -Assigned To: +Assigned To: nikic
 [2017-10-24 09:41 UTC] nikic@php.net
Biases in mt_rand() have been fixed in PHP 7.1, with a remaining issue fixed in PHP 7.2. As such, this is now resolved.
 
PHP Copyright © 2001-2024 The PHP Group
All rights reserved.
Last updated: Fri Mar 29 02:01:30 2024 UTC