Wikimedia Bugzilla is closed!

Created attachment 10420 [details]
Workaround.

MediaWiki 1.18.1
SemanticMedia Wiki 1.7.1

I have never seen this issue when working with MediaWiki interactively, but running maintenance/runJobs.php script often crashes:

Fatal error: Call to a member function getUser() on a non-object in /var/www/oc.su/Extensions/SemanticMediaWiki-1.7.1/includes/SMW_ParseData.php on line 218

Call Stack:
    0.0004     668472   1. {main}() /var/www/oc.su/MediaWiki-1.18.1/maintenance/runJobs.php:0
    0.0029    1239944   2. require_once('/var/www/oc.su/MediaWiki-1.18.1/maintenance/doMaintenance.php') /var/www/oc.su/MediaWiki-1.18.1/maintenance/runJobs.php:108
    0.1163   18504696   3. RunJobs->execute() /var/www/oc.su/MediaWiki-1.18.1/maintenance/doMaintenance.php:105
    0.1813   19744368   4. RefreshLinksJob2->run() /var/www/oc.su/MediaWiki-1.18.1/maintenance/runJobs.php:78
    0.7043   39067088   5. LinksUpdate->__construct() /var/www/oc.su/MediaWiki-1.18.1/includes/job/RefreshLinksJob.php:119
    0.7048   39069280   6. wfRunHooks() /var/www/oc.su/MediaWiki-1.18.1/includes/LinksUpdate.php:98
    0.7048   39069280   7. Hooks::run() /var/www/oc.su/MediaWiki-1.18.1/includes/GlobalFunctions.php:3631
    0.7048   39084176   8. call_user_func_array() /var/www/oc.su/MediaWiki-1.18.1/includes/Hooks.php:216
    0.7048   39084512   9. SMWParseData::onLinksUpdateConstructed() /var/www/oc.su/MediaWiki-1.18.1/includes/Hooks.php:216
    0.7049   39084560  10. SMWParseData::storeData() /var/www/oc.su/Extensions/SemanticMediaWiki-1.7.1/includes/SMW_ParseData.php:481

(reported line numbers may be not fully correct because I added bunch of trace statements to the code to track down the problem.)

The problem appears in SMW_ParseData.php:210:

case '_LEDT' :
	$revision = Revision::newFromId( $title->getLatestRevID() );
	$user = User::newFromId( $revision->getUser() );
	$value = SMWDIWikiPage::newFromTitle( $user->getUserPage() );
	break;

For unknown reason, getLatestRevID() returns zero for valid title of *existing* page, $revision set to null, getUser() failed.

I am not sure if it is a bug in Title, LinkCache, BacklinkCache, RefreshLinksJob2 classes, or in runJobs.php script. 

Roughly, the scenario is:

runJobs.php calls RunJobs->execute() which calls RefreshLinksJob2->run() which calls BacklinkCache->getLinks(). The latter gets list of pages, but only titls, namespaces, and ids (no lastest rev id, no redirection, etc):

// @todo FIXME: Make this a function?
if ( !isset( $this->fullResultCache[$table] ) ) {
	wfDebug( __METHOD__ . ": from DB\n" );
	$res = $this->getDB()->select(
		array( $table, 'page' ),
		array( 'page_namespace', 'page_title', 'page_id' ), 
                // ===>>> NOTE: If I add 'page_latest' to the list, 
                // ===>>> the problem will disappear.
		$this->getConditions( $table ),
		__METHOD__,
		array(
			'STRAIGHT_JOIN',
			'ORDER BY' => $fromField,
		) );
	$this->fullResultCache[$table] = $res;
}

This information is cached in LinkCache, and later returned by $title->getLatestRevID(). I think it is obviously a bug (but do not know where exactly).

Meanwhile, Semantic MediaWiki can be fixed to workaround it:

case '_LEDT' :
	// Do *not* use
	//     $revision = Revision::newFromId( $title->getLatestRevID() );
	// being run from maintenance/runJobs.php it causes exceptions because 
	// `$title->getLatestRevID()' returns zero for *existing* page.
	// Not sure whether it is a MediaWiki bug or not, but using
	// `Revision::newFromTitle' helps to avid this problem.
	$revision = Revision::newFromTitle( $title );
	$user = User::newFromId( $revision->getUser() );
	$value = SMWDIWikiPage::newFromTitle( $user->getUserPage() );
	break;

Revision::newFromTitle() does not rely on cache, but gets last revision id directly from database. I should be less efficient, but works and does not crash.