Added PHP port of CDB, with abstraction layer. Tested for correctness with a differen...
authorTim Starling <tstarling@users.mediawiki.org>
Sat, 20 Jun 2009 15:59:56 +0000 (15:59 +0000)
committerTim Starling <tstarling@users.mediawiki.org>
Sat, 20 Jun 2009 15:59:56 +0000 (15:59 +0000)
Ported existing uses of CDB to the new abstraction layer.

RELEASE-NOTES
includes/AutoLoader.php
includes/Cdb.php [new file with mode: 0644]
includes/Cdb_PHP.php [new file with mode: 0644]
includes/Interwiki.php
maintenance/dumpInterwiki.inc
maintenance/dumpInterwiki.php

index 67badd7..1ac0005 100644 (file)
@@ -89,6 +89,9 @@ it from source control: http://www.mediawiki.org/wiki/Download_from_SVN
 * Add $wgRevokePermissions as a means of restricting a group's rights. The syntax is
   identical to $wgGroupPermissions, but users in these groups will have these rights
   stripped from them.
+* Added a PHP port of CDB (constant database), for improved local caching when
+  the DBA extension is not available.
+
 
 === Bug fixes in 1.16 ===
 
index ce86237..482c34b 100644 (file)
@@ -27,6 +27,13 @@ $wgAutoloadLocalClasses = array(
        'Categoryfinder' => 'includes/Categoryfinder.php',
        'CategoryPage' => 'includes/CategoryPage.php',
        'CategoryViewer' => 'includes/CategoryPage.php',
+       'CdbFunctions' => 'includes/Cdb_PHP.php',
+       'CdbReader' => 'includes/Cdb.php',
+       'CdbReader_DBA' => 'includes/Cdb.php',
+       'CdbReader_PHP' => 'includes/Cdb_PHP.php',
+       'CdbWriter' => 'includes/Cdb.php',
+       'CdbWriter_DBA' => 'includes/Cdb.php',
+       'CdbWriter_PHP' => 'includes/Cdb_PHP.php',
        'ChangesList' => 'includes/ChangesList.php',
        'ChangesFeed' => 'includes/ChangesFeed.php',
        'ChangeTags' => 'includes/ChangeTags.php',
diff --git a/includes/Cdb.php b/includes/Cdb.php
new file mode 100644 (file)
index 0000000..20cb7e3
--- /dev/null
@@ -0,0 +1,147 @@
+<?php
+
+/**
+ * Read from a CDB file.
+ * Native and pure PHP implementations are provided.
+ * http://cr.yp.to/cdb.html
+ */
+abstract class CdbReader {
+       /**
+        * Open a file and return a subclass instance
+        */
+       public static function open( $fileName ) {
+               if ( self::haveExtension() ) {
+                       return new CdbReader_DBA( $fileName );
+               } else {
+                       wfDebug( 'Warning: no dba extension found, using emulation.' );
+                       return new CdbReader_PHP( $fileName );
+               }
+       }
+
+       /**
+        * Returns true if the native extension is available
+        */
+       public static function haveExtension() {
+               if ( !function_exists( 'dba_handlers' ) ) {
+                       return false;
+               }
+               $handlers = dba_handlers();
+               if ( !in_array( 'cdb', $handlers ) || !in_array( 'cdb_make', $handlers ) ) {
+                       return false;
+               }
+               return true;
+       }
+
+       /**
+        * Construct the object and open the file
+        */
+       abstract function __construct( $fileName );
+
+       /**
+        * Close the file. Optional, you can just let the variable go out of scope.
+        */
+       abstract function close();
+
+       /**
+        * Get a value with a given key. Only string values are supported.
+        */
+       abstract public function get( $key );
+}
+
+/**
+ * Write to a CDB file.
+ * Native and pure PHP implementations are provided.
+ */
+abstract class CdbWriter {
+       /**
+        * Open a writer and return a subclass instance.
+        * The user must have write access to the directory, for temporary file creation.
+        */
+       public static function open( $fileName ) {
+               if ( CdbReader::haveExtension() ) {
+                       return new CdbWriter_DBA( $fileName );
+               } else {
+                       wfDebug( 'Warning: no dba extension found, using emulation.' );
+                       return new CdbWriter_PHP( $fileName );
+               }
+       }
+
+       /**
+        * Create the object and open the file
+        */
+       abstract function __construct( $fileName );
+
+       /**
+        * Set a key to a given value. The value will be converted to string.
+        */
+       abstract public function set( $key, $value );
+
+       /**
+        * Close the writer object. You should call this function before the object
+        * goes out of scope, to write out the final hashtables.
+        */
+       abstract public function close();
+}
+
+
+/**
+ * Reader class which uses the DBA extension
+ */
+class CdbReader_DBA {
+       var $handle;
+
+       function __construct( $fileName ) {
+               $this->handle = dba_open( $fileName, 'r-', 'cdb' );
+               if ( !$this->handle ) {
+                       throw new MWException( 'Unable to open DB file "' . $fileName . '"' );
+               }
+       }
+
+       function close() {
+               dba_close( $this->handle );
+               unset( $this->handle );
+       }
+
+       function get( $key ) {
+               return dba_fetch( $key, $this->handle );
+       }
+}
+
+
+/**
+ * Writer class which uses the DBA extension
+ */
+class CdbWriter_DBA {
+       var $handle, $realFileName, $tmpFileName;
+
+       function __construct( $fileName ) {
+               $this->realFileName = $fileName;
+               $this->tmpFileName = $fileName . '.tmp.' . mt_rand( 0, 0x7fffffff );
+               $this->handle = dba_open( $this->tmpFileName, 'n', 'cdb_make' );
+               if ( !$this->handle ) {
+                       throw new MWException( 'Unable to open DB file for write "' . $fileName . '"' );
+               }
+       }
+
+       function set( $key, $value ) {
+               return dba_insert( $key, $value, $this->handle );
+       }
+
+       function close() {
+               dba_close( $this->handle );
+               if ( wfIsWindows() ) {
+                       unlink( $this->realFileName );
+               }
+               if ( !rename( $this->tmpFileName, $this->realFileName ) ) {
+                       throw new MWException( 'Unable to move the new CDB file into place.' );
+               }
+               unset( $this->handle );
+       }
+
+       function __destruct() {
+               if ( isset( $this->handle ) ) {
+                       $this->close();
+               }
+       }
+}
+
diff --git a/includes/Cdb_PHP.php b/includes/Cdb_PHP.php
new file mode 100644 (file)
index 0000000..eb2d317
--- /dev/null
@@ -0,0 +1,388 @@
+<?php
+
+/**
+ * This is a port of D.J. Bernstein's CDB to PHP. It's based on the copy that 
+ * appears in PHP 5.3. Changes are:
+ *    * Error returns replaced with exceptions
+ *    * Exception thrown if sizes or offsets are between 2GB and 4GB
+ *    * Some variables renamed
+ */
+
+/**
+ * Common functions for readers and writers
+ */
+class CdbFunctions {
+       /**
+        * Do a sum of 32-bit signed integers with 2's complement overflow.
+        *
+        * PHP has broken plus and minus operators, but the bitwise operators 
+        * (&, |, ^, ~, <<, >>) are all implemented as a simple wrapper around the
+        * underlying C operator. The algorithm here uses a binary view of addition
+        * to simulate 32-bit addition using 31-bit registers.
+        */
+       public static function sumWithOverflow( $a, $b ) {
+               $sum = $a + $b;
+               if ( is_float( $sum ) ) {
+                       // Use the plus operator to do a sum of the lowest 30 bits to produce a 31-bit result
+                       $lowA = $a & 0x3fffffff;
+                       $lowB = $b & 0x3fffffff;
+                       $sum = $lowA + $lowB;
+
+                       // Strip off the carry bit
+                       $carry = ($sum & 0x40000000) >> 30;
+                       $sum = $sum & 0x3fffffff;
+
+                       // Get the last two bits
+                       $highA = self::unsignedShiftRight( $a, 30 );
+                       $highB = self::unsignedShiftRight( $b, 30 );
+
+                       // Add with carry
+                       $highSum = $carry + $highA + $highB;
+
+                       // Recombine
+                       $sum = $sum | ( $highSum << 30 );
+               }
+               return $sum;
+       }
+
+       /**
+        * Take a modulo of a signed integer as if it were an unsigned integer.
+        * $b must be less than 0x40000000 and greater than 0
+        */
+       public static function unsignedMod( $a, $b ) {
+               if ( $a < 0 ) {
+                       $m = ( $a & 0x7fffffff ) % $b + 2 * ( 0x40000000 % $b );
+                       return $m % $b;
+               } else {
+                       return $a % $b;
+               }
+       }
+       
+       /**
+        * Shift a signed integer right as if it were unsigned
+        */
+       public static function unsignedShiftRight( $a, $b ) {
+               if ( $b == 0 ) {
+                       return $a;
+               }
+               if ( $a < 0 ) {
+                       return ( ( $a & 0x7fffffff ) >> $b ) | ( 0x40000000 >> ( $b - 1 ) );
+               } else {
+                       return $a >> $b;
+               }
+       }
+
+       public static function hash( $s ) {
+               $h = 5381;
+               for ( $i = 0; $i < strlen( $s ); $i++ ) {
+                       $h = self::sumWithOverflow( $h, $h << 5 ) ^ ord( $s[$i] );
+               }
+               return $h;
+       }
+}
+
+/**
+ * CDB reader class
+ */
+class CdbReader_PHP extends CdbReader {
+       /** The file handle */
+       var $handle;
+
+       /* number of hash slots searched under this key */
+       var $loop;
+
+       /* initialized if loop is nonzero */
+       var $khash;
+
+       /* initialized if loop is nonzero */
+       var $kpos;
+
+       /* initialized if loop is nonzero */
+       var $hpos;
+
+       /* initialized if loop is nonzero */
+       var $hslots;
+
+       /* initialized if findNext() returns true */
+       var $dpos;
+
+       /* initialized if cdb_findnext() returns 1 */
+       var $dlen; 
+
+       function __construct( $fileName ) {
+               $this->handle = fopen( $fileName, 'rb' );
+               if ( !$this->handle ) {
+                       throw new MWException( 'Unable to open DB file "' . $fileName . '"' );
+               }
+               $this->findStart();
+       }
+
+       function close() {
+               fclose( $this->handle );
+               unset( $this->handle );
+       }
+
+       public function get( $key ) {
+               // strval is required
+               if ( $this->find( strval( $key ) ) ) {
+                       return $this->read( $this->dlen, $this->dpos );
+               } else {
+                       return false;
+               }
+       }
+
+       protected function match( $key, $pos ) {
+               $buf = $this->read( strlen( $key ), $pos );
+               return $buf === $key;
+       }
+
+       protected function findStart() {
+               $this->loop = 0;
+       }
+
+       protected function read( $length, $pos ) {
+               if ( fseek( $this->handle, $pos ) == -1 ) {
+                       // This can easily happen if the internal pointers are incorrect
+                       throw new MWException( __METHOD__.': seek failed, file may be corrupted.' );
+               }
+
+               if ( $length == 0 ) {
+                       return '';
+               }
+
+               $buf = fread( $this->handle, $length );
+               if ( $buf === false || strlen( $buf ) !== $length ) {
+                       throw new MWException( __METHOD__.': read from cdb file failed, file may be corrupted' );
+               }
+               return $buf;
+       }
+
+       /**
+        * Unpack an unsigned integer and throw an exception if it needs more than 31 bits
+        */
+       protected function unpack31( $s ) {
+               $data = unpack( 'V', $s );
+               if ( $data[1] > 0x7fffffff ) {
+                       throw new MWException( __METHOD__.': error in CDB file, integer too big' );
+               }
+               return $data[1];
+       }
+
+       /**
+        * Unpack a 32-bit signed integer
+        */
+       protected function unpackSigned( $s ) {
+               $data = unpack( 'va/vb', $s );
+               return $data['a'] | ( $data['b'] << 16 );
+       }
+
+       protected function findNext( $key ) {
+               if ( !$this->loop ) {
+                       $u = CdbFunctions::hash( $key );
+                       $buf = $this->read( 8, ( $u << 3 ) & 2047 );
+                       $this->hslots = $this->unpack31( substr( $buf, 4 ) );
+                       if ( !$this->hslots ) {
+                               return false;
+                       }
+                       $this->hpos = $this->unpack31( substr( $buf, 0, 4 ) );
+                       $this->khash = $u;
+                       $u = CdbFunctions::unsignedShiftRight( $u, 8 );
+                       $u = CdbFunctions::unsignedMod( $u, $this->hslots );
+                       $u <<= 3;
+                       $this->kpos = $this->hpos + $u;
+               }
+
+               while ( $this->loop < $this->hslots ) {
+                       $buf = $this->read( 8, $this->kpos );
+                       $pos = $this->unpack31( substr( $buf, 4 ) );
+                       if ( !$pos ) {
+                               return false;
+                       }
+                       $this->loop += 1;
+                       $this->kpos += 8;
+                       if ( $this->kpos == $this->hpos + ( $this->hslots << 3 ) ) {
+                               $this->kpos = $this->hpos;
+                       }
+                       $u = $this->unpackSigned( substr( $buf, 0, 4 ) );
+                       if ( $u === $this->khash ) {
+                               $buf = $this->read( 8, $pos );
+                               $keyLen = $this->unpack31( substr( $buf, 0, 4 ) );
+                               if ( $keyLen == strlen( $key ) && $this->match( $key, $pos + 8 ) ) {
+                                       // Found
+                                       $this->dlen = $this->unpack31( substr( $buf, 4 ) );
+                                       $this->dpos = $pos + 8 + $keyLen;
+                                       return true;
+                               }
+                       }
+               }
+               return false;
+       }
+
+       protected function find( $key ) {
+               $this->findStart();
+               return $this->findNext( $key );
+       }
+}
+
+/**
+ * CDB writer class
+ */
+class CdbWriter_PHP extends CdbWriter {
+       var $handle, $realFileName, $tmpFileName;
+
+       var $hplist;
+       var $numEntries, $pos;
+
+       function __construct( $fileName ) {
+               $this->realFileName = $fileName;
+               $this->tmpFileName = $fileName . '.tmp.' . mt_rand( 0, 0x7fffffff );
+               $this->handle = fopen( $this->tmpFileName, 'wb' );
+               if ( !$this->handle ) {
+                       throw new MWException( 'Unable to open DB file for write "' . $fileName . '"' );
+               }
+               $this->hplist = array();
+               $this->numentries = 0;
+               $this->pos = 2048; // leaving space for the pointer array, 256 * 8
+               if ( fseek( $this->handle, $this->pos ) == -1 ) {
+                       throw new MWException( __METHOD__.': fseek failed' );
+               }
+       }
+
+       function __destruct() {
+               if ( isset( $this->handle ) ) {
+                       $this->close();
+               }
+       }
+
+       public function set( $key, $value ) {
+               if ( strval( $key ) === '' ) {
+                       // DBA cross-check hack
+                       return;
+               }
+               $this->addbegin( strlen( $key ), strlen( $value ) );
+               $this->write( $key );
+               $this->write( $value );
+               $this->addend( strlen( $key ), strlen( $value ), CdbFunctions::hash( $key ) );
+       }
+
+       public function close() {
+               $this->finish();
+               fclose( $this->handle );
+               if ( wfIsWindows() ) {
+                       unlink( $this->realFileName );
+               }
+               if ( !rename( $this->tmpFileName, $this->realFileName ) ) {
+                       throw new MWException( 'Unable to move the new CDB file into place.' );
+               }
+               unset( $this->handle );
+       }
+
+       protected function write( $buf ) {
+               $len = fwrite( $this->handle, $buf );
+               if ( $len !== strlen( $buf ) ) {
+                       throw new MWException( 'Error writing to CDB file.' );
+               }
+       }
+
+       protected function posplus( $len ) {
+               $newpos = $this->pos + $len;
+               if ( $newpos > 0x7fffffff ) {
+                       throw new MWException( 'A value in the CDB file is too large' );
+               }
+               $this->pos = $newpos;
+       }
+
+       protected function addend( $keylen, $datalen, $h ) {
+               $this->hplist[] = array(
+                       'h' => $h,
+                       'p' => $this->pos
+               );
+
+               $this->numentries++;
+               $this->posplus( 8 );
+               $this->posplus( $keylen );
+               $this->posplus( $datalen );
+       }
+
+       protected function addbegin( $keylen, $datalen ) {
+               if ( $keylen > 0x7fffffff ) {
+                       throw new MWException( __METHOD__.': key length too long' );
+               }
+               if ( $datalen > 0x7fffffff ) {
+                       throw new MWException( __METHOD__.': data length too long' );
+               }
+               $buf = pack( 'VV', $keylen, $datalen );
+               $this->write( $buf );
+       }
+
+       protected function finish() {
+               // Hack for DBA cross-check
+               $this->hplist = array_reverse( $this->hplist );
+
+               // Calculate the number of items that will be in each hashtable
+               $counts = array_fill( 0, 256, 0 );
+               foreach ( $this->hplist as $item ) {
+                       ++ $counts[ 255 & $item['h'] ];
+               }
+
+               // Fill in $starts with the *end* indexes
+               $starts = array();
+               $pos = 0;
+               for ( $i = 0; $i < 256; ++$i ) {
+                       $pos += $counts[$i];
+                       $starts[$i] = $pos;
+               }
+
+               // Excessively clever and indulgent code to simultaneously fill $packedTables
+               // with the packed hashtables, and adjust the elements of $starts 
+               // to actually point to the starts instead of the ends.
+               $packedTables = array_fill( 0, $this->numentries, false );
+               foreach ( $this->hplist as $item ) {
+                       $packedTables[--$starts[255 & $item['h']]] = $item;
+               }
+
+               $final = '';
+               for ( $i = 0; $i < 256; ++$i ) {
+                       $count = $counts[$i];
+
+                       // The size of the hashtable will be double the item count.
+                       // The rest of the slots will be empty.
+                       $len = $count + $count;
+                       $final .= pack( 'VV', $this->pos, $len );
+
+                       $hashtable = array();
+                       for ( $u = 0; $u < $len; ++$u ) {
+                               $hashtable[$u] = array( 'h' => 0, 'p' => 0 );
+                       }
+
+                       // Fill the hashtable, using the next empty slot if the hashed slot
+                       // is taken.
+                       for ( $u = 0; $u < $count; ++$u ) {
+                               $hp = $packedTables[$starts[$i] + $u];
+                               $where = CdbFunctions::unsignedMod( 
+                                       CdbFunctions::unsignedShiftRight( $hp['h'], 8 ), $len );
+                               while ( $hashtable[$where]['p'] )
+                                       if ( ++$where == $len )
+                                               $where = 0;
+                               $hashtable[$where] = $hp;
+                       }
+
+                       // Write the hashtable
+                       for ( $u = 0; $u < $len; ++$u ) {
+                               $buf = pack( 'vvV', 
+                                       $hashtable[$u]['h'] & 0xffff,
+                                       CdbFunctions::unsignedShiftRight( $hashtable[$u]['h'], 16 ),
+                                       $hashtable[$u]['p'] );
+                               $this->write( $buf );
+                               $this->posplus( 8 );
+                       }
+               }
+
+               // Write the pointer array at the start of the file
+               rewind( $this->handle );
+               if ( ftell( $this->handle ) != 0 ) {
+                       throw new MWException( __METHOD__.': Error rewinding to start of file' );
+               }
+               $this->write( $final );
+       }
+}
index f42f7e3..5ca9262 100644 (file)
@@ -104,24 +104,24 @@ class Interwiki {
 
                wfDebug( __METHOD__ . "( $prefix )\n" );
                if( !$db ) {
-                       $db = dba_open( $wgInterwikiCache, 'r', 'cdb' );
+                       $db = CdbReader::open( $wgInterwikiCache );
                }
                /* Resolve site name */
                if( $wgInterwikiScopes>=3 && !$site ) {
-                       $site = dba_fetch( '__sites:' . wfWikiID(), $db );
+                       $site = $db->get( '__sites:' . wfWikiID() );
                        if ( $site == '' ) {
                                $site = $wgInterwikiFallbackSite;
                        }
                }
 
-               $value = dba_fetch( wfMemcKey( $prefix ), $db );
+               $value = $db->get( wfMemcKey( $prefix ) );
                // Site level
                if ( $value == '' && $wgInterwikiScopes >= 3 ) {
-                       $value = dba_fetch( "_{$site}:{$prefix}", $db );
+                       $value = $db->get( "_{$site}:{$prefix}" );
                }
                // Global Level
                if ( $value == '' && $wgInterwikiScopes >= 2 ) {
-                       $value = dba_fetch( "__global:{$prefix}", $db );
+                       $value = $db->get( "__global:{$prefix}" );
                }
                if ( $value == 'undef' )
                        $value = '';
index 481e21c..031d044 100644 (file)
@@ -201,7 +201,7 @@ function makeLink( $entry, $source ) {
                 array_key_exists($entry['iw_prefix'],$prefixRewrites[$source]))
                     $entry['iw_prefix'] = $prefixRewrites[$source][$entry['iw_prefix']];
         if ($dbFile)
-            dba_insert("{$source}:{$entry['iw_prefix']}", trim("{$entry['iw_local']} {$entry['iw_url']}"),$dbFile);
+            $dbFile->set( "{$source}:{$entry['iw_prefix']}", trim("{$entry['iw_local']} {$entry['iw_url']}") );
         else
             print "{$source}:{$entry['iw_prefix']} {$entry['iw_url']} {$entry['iw_local']}\n";
 
index f5f22f3..16bf4db 100644 (file)
@@ -19,7 +19,7 @@ chdir( $oldCwd );
 # Output
 if ( isset( $options['o'] ) ) {
     # To database specified with -o
-    $dbFile = dba_open( $options['o'], "n", "cdb_make" );
+    $dbFile = CdbWriter::open( $options['o'] );
 } 
 
 getRebuildInterwikiDump();