# 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
# http://www.gnu.org/copyleft/gpl.html
+
/**
*
- * @package MediaWiki
- * @subpackage SpecialPage
+ * @addtogroup SpecialPage
*/
-
class WikiExporter {
var $list_authors = false ; # Return distinct author list (when not returning full history)
var $author_list = "" ;
-
+
const FULL = 0;
const CURRENT = 1;
* main query is still running.
*
* @param Database $db
- * @param mixed $history one of WikiExporter::FULL or WikiExporter::CURRENT, or an
+ * @param mixed $history one of WikiExporter::FULL or WikiExporter::CURRENT, or an
* associative array:
* offset: non-inclusive offset at which to start the query
* limit: maximum number of rows to return
* dir: "asc" or "desc" timestamp order
* @param int $buffer one of WikiExporter::BUFFER or WikiExporter::STREAM
*/
- function WikiExporter( &$db, $history = WikiExporter::CURRENT,
+ function __construct( &$db, $history = WikiExporter::CURRENT,
$buffer = WikiExporter::BUFFER, $text = WikiExporter::TEXT ) {
$this->db =& $db;
$this->history = $history;
$fname = "do_list_authors" ;
wfProfileIn( $fname );
$this->author_list = "<contributors>";
- $sql = "SELECT DISTINCT rev_user_text,rev_user FROM {$page},{$revision} WHERE page_id=rev_page AND " . $cond ;
+ //rev_deleted
+ $nothidden = '(rev_deleted & '.Revision::DELETED_USER.') = 0';
+
+ $sql = "SELECT DISTINCT rev_user_text,rev_user FROM {$page},{$revision} WHERE page_id=rev_page AND $nothidden AND " . $cond ;
$result = $this->db->query( $sql, $fname );
$resultset = $this->db->resultObject( $result );
while( $row = $resultset->fetchObject() ) {
$page = $this->db->tableName( 'page' );
$revision = $this->db->tableName( 'revision' );
$text = $this->db->tableName( 'text' );
-
+
$order = 'ORDER BY page_id';
$limit = '';
-
+
if( $this->history == WikiExporter::FULL ) {
$join = 'page_id=rev_page';
} elseif( $this->history == WikiExporter::CURRENT ) {
$order .= ', rev_timestamp DESC';
}
if ( !empty( $this->history['offset'] ) ) {
- $join .= " AND rev_timestamp $op " . $this->db->addQuotes(
+ $join .= " AND rev_timestamp $op " . $this->db->addQuotes(
$this->db->timestamp( $this->history['offset'] ) );
}
if ( !empty( $this->history['limit'] ) ) {
$result = $this->db->query( $sql, $fname );
$wrapper = $this->db->resultObject( $result );
$this->outputStream( $wrapper );
-
+
if ( $this->list_authors ) {
$this->outputStream( $wrapper );
}
}
}
+/**
+ * @addtogroup Dump
+ */
class XmlDumpWriter {
/**
}
function homelink() {
- $page = Title::newFromText( wfMsgForContent( 'mainpage' ) );
- return wfElement( 'base', array(), $page->getFullUrl() );
+ return wfElement( 'base', array(), Title::newMainPage()->getFullUrl() );
}
function caseSetting() {
/**
* Base class for output stream; prints to stdout or buffer or whereever.
+ * @addtogroup Dump
*/
class DumpOutput {
function writeOpenStream( $string ) {
/**
* Stream outputter to send data to a file.
+ * @addtogroup Dump
*/
class DumpFileOutput extends DumpOutput {
var $handle;
* Stream outputter to send data to a file via some filter program.
* Even if compression is available in a library, using a separate
* program can allow us to make use of a multi-processor system.
+ * @addtogroup Dump
*/
class DumpPipeOutput extends DumpFileOutput {
function DumpPipeOutput( $command, $file = null ) {
/**
* Sends dump output via the gzip compressor.
+ * @addtogroup Dump
*/
class DumpGZipOutput extends DumpPipeOutput {
function DumpGZipOutput( $file ) {
/**
* Sends dump output via the bgzip2 compressor.
+ * @addtogroup Dump
*/
class DumpBZip2Output extends DumpPipeOutput {
function DumpBZip2Output( $file ) {
/**
* Sends dump output via the p7zip compressor.
+ * @addtogroup Dump
*/
class Dump7ZipOutput extends DumpPipeOutput {
function Dump7ZipOutput( $file ) {
$command = "7za a -bd -si " . wfEscapeShellArg( $file );
// Suppress annoying useless crap from p7zip
// Unfortunately this could suppress real error messages too
- $command .= " >/dev/null 2>&1";
+ $command .= ' >' . wfGetNull() . ' 2>&1';
parent::DumpPipeOutput( $command );
}
}
* Dump output filter class.
* This just does output filtering and streaming; XML formatting is done
* higher up, so be careful in what you do.
+ * @addtogroup Dump
*/
class DumpFilter {
function DumpFilter( &$sink ) {
/**
* Simple dump output filter to exclude all talk pages.
+ * @addtogroup Dump
*/
class DumpNotalkFilter extends DumpFilter {
function pass( $page ) {
/**
* Dump output filter to include or exclude pages in a given set of namespaces.
+ * @addtogroup Dump
*/
class DumpNamespaceFilter extends DumpFilter {
var $invert = false;
/**
* Dump output filter to include only the last revision in each page sequence.
+ * @addtogroup Dump
*/
class DumpLatestFilter extends DumpFilter {
var $page, $pageString, $rev, $revString;
/**
* Base class for output stream; prints to stdout or buffer or whereever.
+ * @addtogroup Dump
*/
class DumpMultiWriter {
function DumpMultiWriter( $sinks ) {
return $string;
}
-?>
+