Delicious Bookmark this on Delicious Share on Facebook SlashdotSlashdot It! Digg! Digg

PHP : Function Reference : Miscellaneous Functions : uniqid


Generate a unique ID (PHP 4, PHP 5)
string uniqid ( [string prefix [, bool more_entropy]] )

Example 1366. uniqid() Example

// no prefix
// works only in PHP 5 and later versions
$token = md5(uniqid());

// better, difficult to guess
$better_token = md5(uniqid(rand(), true));

Related Examples ( Source code ) » uniqid

Code Examples / Notes » uniqid


wooshoofoo, the reason mimec is calling mt_rand multiple times is because the largest number mt_rand can produce is 2^31 (2147483647, as reported by mt_getrandmax() on my server).  RFC 4122 requires a 128 bit value.
Also they are not "4 digit sequeces", but 4 digit hexadecimal numbers.  16^4 == 2^16.
mimec's limiting each random result to 2^16 avoids problem of PHP's 2^32 integer max (
If you want to call mt_rand fewer times:  mimec's version calls mt_rand 8 times ( 16 bits * 8 = 128 bits ).  You *could* call mt_rand 5 times ( 31 bits + 31 bits + 31 bits + 31 bits + 4 bits = 128 bits ).  But then you would have keep all your values as strings.
Something like:
* Another (ugly) "random or pseudo-random" version of RFC 4122
* This version calls mt_rand() the fewest possible times.
* if mt_getrandmax() == 2^31 then this will call mt_rand() 5 times YMMV
* Personally, I would use mimec's version
* To handle the large values, we'll keep everything as strings.
* @return string
function uuid() {    
    // Generate 128 bit random sequence
    $randmax_bits = strlen(base_convert(mt_getrandmax(), 10, 2));  // how many bits is mt_getrandmax()
    $x = '';
    while (strlen($x) < 128) {
        $maxbits = (128 - strlen($x) < $randmax_bits) ? 128 - strlen($x) :  $randmax_bits;
        $x .= str_pad(base_convert(mt_rand(0, pow(2,$maxbits)), 10, 2), $maxbits, "0", STR_PAD_LEFT);
    // break into fields
    $a = array();
    $a['time_low_part'] = substr($x, 0, 32);
    $a['time_mid'] = substr($x, 32, 16);
    $a['time_hi_and_version'] = substr($x, 48, 16);
    $a['clock_seq'] = substr($x, 64, 16);
    $a['node_part'] =  substr($x, 80, 48);
    // Apply bit masks for "random or pseudo-random" version per RFC
    $a['time_hi_and_version'] = substr_replace($a['time_hi_and_version'], '0100', 0, 4);
    $a['clock_seq'] = substr_replace($a['clock_seq'], '10', 0, 2);
   // Format output
   return sprintf('%s-%s-%s-%s-%s',
       str_pad(base_convert($a['time_low_part'], 2, 16), 8, "0", STR_PAD_LEFT),
       str_pad(base_convert($a['time_mid'], 2, 16), 4, "0", STR_PAD_LEFT),
       str_pad(base_convert($a['time_hi_and_version'], 2, 16), 4, "0", STR_PAD_LEFT),
       str_pad(base_convert($a['clock_seq'], 2, 16), 4, "0", STR_PAD_LEFT),
       str_pad(base_convert($a['node_part'], 2, 16), 12, "0", STR_PAD_LEFT));
However, I think mimec's version is much more elegant.


WARNING : I believe there are a couple of mistakes in the function provided just below by maciej dot strzelecki at gmail dot com. Namely, that in the two substr_replace() calls, the third parameters should respectively be 12 (instead of 11) and 6 (instead of 5).
Considering the importance of this function, I went to read RFC 4122 myself, and found the discrepancy. I therefore chose to write my own function, inspired by the previous one, but with a few enhancements detailed in the comments. On the downside, it might be slightly less easy to understand at first glance.
Please feel free to use it yourself. Thank you also in advance for any feedback at dholmes at .
* Generates a Universally Unique IDentifier, version 4.
* RFC 4122 ( defines a special type of Globally
* Unique IDentifiers (GUID), as well as several methods for producing them. One
* such method, described in section 4.4, is based on truly random or pseudo-random
* number generators, and is therefore implementable in a language like PHP.
* We choose to produce pseudo-random numbers with the Mersenne Twister, and to always
* limit single generated numbers to 16 bits (ie. the decimal value 65535). That is
* because, even on 32-bit systems, PHP's RAND_MAX will often be the maximum *signed*
* value, with only the equivalent of 31 significant bits. Producing two 16-bit random
* numbers to make up a 32-bit one is less efficient, but guarantees that all 32 bits
* are random.
* The algorithm for version 4 UUIDs (ie. those based on random number generators)
* states that all 128 bits separated into the various fields (32 bits, 16 bits, 16 bits,
* 8 bits and 8 bits, 48 bits) should be random, except : (a) the version number should
* be the last 4 bits in the 3rd field, and (b) bits 6 and 7 of the 4th field should
* be 01. We try to conform to that definition as efficiently as possible, generating
* smaller values where possible, and minimizing the number of base conversions.
* @copyright   Copyright (c) CFD Labs, 2006. This function may be used freely for
*              any purpose ; it is distributed without any form of warranty whatsoever.
* @author      David Holmes <>
* @return  string  A UUID, made up of 32 hex digits and 4 hyphens.
function uuid() {
   // The field names refer to RFC 4122 section 4.1.2
   return sprintf('%04x%04x-%04x-%03x4-%04x-%04x%04x%04x',
       mt_rand(0, 65535), mt_rand(0, 65535), // 32 bits for "time_low"
       mt_rand(0, 65535), // 16 bits for "time_mid"
       mt_rand(0, 4095),  // 12 bits before the 0100 of (version) 4 for "time_hi_and_version"
       bindec(substr_replace(sprintf('%016b', mt_rand(0, 65535)), '01', 6, 2)),
           // 8 bits, the last two of which (positions 6 and 7) are 01, for "clk_seq_hi_res"
           // (hence, the 2nd hex digit after the 3rd hyphen can only be 1, 5, 9 or d)
           // 8 bits for "clk_seq_low"
       mt_rand(0, 65535), mt_rand(0, 65535), mt_rand(0, 65535) // 48 bits for "node"  

maciej dot strzelecki

This is an implementation of version 4 UUID, which is generating UUIDs from truly-random numbers.
/* Copyright 2006 Maciej Strzelecki
  This program is free software; you can redistribute it and/or modify
  it under the terms of the GNU General Public License as published by
  the Free Software Foundation; either version 2 of the License, or
  (at your option) any later version.
  This program is distributed in the hope that it will be useful,
  but WITHOUT ANY WARRANTY; without even the implied warranty of
  GNU General Public License for more details.
  You should have received a copy of the GNU General Public License
  along with this program; if not, write to the Free Software
  Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA */
function uuid()
   // version 4 UUID
   return sprintf(
       mt_rand(0, 65535),
           sprintf('%016b', mt_rand(0, 65535)), '0100', 11, 4)
       bindec(substr_replace(sprintf('%08b', mt_rand(0, 255)), '01', 5, 2)),
       mt_rand(0, 255),


This function is painfully slow if you're using it to give images random names inside of a loop.  The following function will give you a random name *every* time and is much faster.
function nameImage($imgExtension)
   return time() . substr(md5(microtime()), 0, rand(5, 12)) . $imgExtension;


The example given in this document for a "better token" should be:
$better_token = uniqid(md5(rand()), true);
As it is now, the result isn't guaranteed to be unique, because MD5 has collisions.


Neither the pseudo-random number rand() nor the Mersenne Twister algorithms are cryptographically strong, and this is well known.  Simply combining non-cryptographically strong algorithms doesn't not make a cryptographically strong algorithm either.  Mersenne Twister is a fast algorithm with good k-distribution which will give you numbers for a long time before it repeats itself.  MT, rand(), and MD5 should NOT be used for encryption, or for cookies that that store a session ID that gives personal information.  A simple application where non-collision of session IDs is highly preferred but not critical, such as storing a user's shopping cart items for when they return to your site (but not their personal information), IS a good use for the MT, rand() MD5, uniqid() and combinations thereof.


Just to note this function is fairly slow, and can bring your script to a crawl if it is in a loop. Strangely if you run it as uniqid('', true) it runs much more quickly

kristoffer dot paro

In response to the notes about UUID generation added by mimec and lance_rushing at hotmail dot com.
Calling mt_rand the fewest possible times is not necessarily the fastest, if it heavily utilizes string handling routines. I did a quick benchmark between the two functions and discovered that lance's function (using only 5 mt_rands) was about 6.5 times _slower_ than mimec's on my system.

03-apr-2006 07:16

if you're generating UUIDs you should really check out first ('version 4' UUID, chapter 4.4), as they are not completely random.


If you can guarantee a connection to mysql when you need your UUID then you can wrap up MySQL's (v5+) function.
function uuid() {
return mysql_result(mysql_query('Select UUID()'),0);


I'm not too fond of the recommendation to use an MD5 of the unique ID for session IDs. It would be a better idea just to use uniqueid(rand(), true) without the MD5, because even though it's a rare circumstance, MD5 is a hash, not an encryption, which means it has collisions. Therefore you theoretically could have multiple users given the same session ID which could result in one user's ability to access another user's data.


I'm not sure the previous function by mimec is really all that random.  For one thing, generating 8 small random 4 digit sequeces != generating one 32 digit sequence.

dot dot dot dot dot alexander

I use this mangle currently:
( inserts the IP, uses time() and a prefix, aside the uniqid)
function newid($prefix = "user_"){
return ($prefix . uniqid( hash("md5", time()), TRUE ) . time() . @$_SERVER['REMOTE_ADDR']);
}/* endfunction */
}/* endif  */


I have been using mimecs version lately and do not think it's safe to think the results are always unqiue.
Although it could be just my bad programming, I found exactly 1 collission while debugging my code. It seems to me that if my code was incorrect it would have happened more than once.
I recommend anyone to include time as a factor of such an ID as to be a little more certain it is in fact unique.


Here is the correct version of a function generating a pseudo-random UUID according to RFC 4122:
function uuid()
   return sprintf( '%04x%04x-%04x-%04x-%04x-%04x%04x%04x',
       mt_rand( 0, 0xffff ), mt_rand( 0, 0xffff ), mt_rand( 0, 0xffff ),
       mt_rand( 0, 0x0fff ) | 0x4000,
       mt_rand( 0, 0x3fff ) | 0x8000,
       mt_rand( 0, 0xffff ), mt_rand( 0, 0xffff ), mt_rand( 0, 0xffff ) );
The version and variant is located at the MSB (most significant bits) of the time_hi_and_version and clock_seq_hi_and_reserved fields, not the LSB as in dholmes version.


Also you may use this if you like it.
function uniqid2() {
return dechex(time()).dechex(mt_rand(1,65535));

Change Language

Follow Navioo On Twitter
eXTReMe Tracker