wfRandomString: Correct bias toward digits 1-7

Two problems with a single line of code:

* dechex() does not include leading zeros, so the first digit
  generated in each iteration cannot be zero unless the return
  value of mt_rand() is 0.

  This also means wfRandomString() is extremely unlikely to
  start a string with '0'.

* mt_rand() does not actually uniformly distribute values over
  [0,2^32-1]; it actually right-shifts one of the bits off (just
  because "[...] the previous php_rand only returns 31 at most"),
  so the maximum value is 0x7fffffff, not 0xffffffff.

  This means wfRandomString() will never start a string with
  any of the digits 8-f.

Including leading zeros and using only the 28 LSBs to form seven
hex digits at a time is the simplest fix.

Change-Id: Ic19b5b97c582485780b24fd35ffef2111cc8b3ca

diff --git a/RELEASE-NOTES-1.22 b/RELEASE-NOTES-1.22
index ec424d3..f0ec6bb 100644 (file)
--- a/RELEASE-NOTES-1.22
+++ b/RELEASE-NOTES-1.22
@@ -63,6 +63,8 @@ production.
 * (bug 46768) Usernames of blocking users now display correctly, even if numeric.
 * (bug 39590) {{PAGESIZE}} for the current page and self-transclusions now
   show the most up to date result always instead of being a revision behind.
+* A bias in wfRandomString() toward digits 1-7 has been corrected. Generated
+  strings now can start with digits 0 and 8-f.
 
 === API changes in 1.22 ===
 * (bug 46626) xmldoublequote parameter was removed. Because of a bug, the

diff --git a/includes/GlobalFunctions.php b/includes/GlobalFunctions.php
index 50287b3..43caf7a 100644 (file)
--- a/includes/GlobalFunctions.php
+++ b/includes/GlobalFunctions.php
@@ -322,8 +322,8 @@ function wfRandom() {
  */
 function wfRandomString( $length = 32 ) {
        $str = '';
-       while ( strlen( $str ) < $length ) {
-               $str .= dechex( mt_rand() );
+       for ( $n = 0; $n < $length; $n += 7 ) {
+               $str .= sprintf( '%07x', mt_rand() & 0xfffffff );
        }
        return substr( $str, 0, $length );
 }