Code Coverage |
||||||||||
Lines |
Functions and Methods |
Classes and Traits |
||||||||
| Total | |
0.00% |
0 / 127 |
|
0.00% |
0 / 6 |
CRAP | |
0.00% |
0 / 1 |
| LinkRecommendationUpdater | |
0.00% |
0 / 127 |
|
0.00% |
0 / 6 |
1056 | |
0.00% |
0 / 1 |
| __construct | |
0.00% |
0 / 10 |
|
0.00% |
0 / 1 |
2 | |||
| processCandidate | |
0.00% |
0 / 41 |
|
0.00% |
0 / 1 |
42 | |||
| evaluateTitle | |
0.00% |
0 / 43 |
|
0.00% |
0 / 1 |
182 | |||
| evaluateRecommendation | |
0.00% |
0 / 20 |
|
0.00% |
0 / 1 |
56 | |||
| getLinkRecommendationTaskType | |
0.00% |
0 / 12 |
|
0.00% |
0 / 1 |
20 | |||
| failure | |
0.00% |
0 / 1 |
|
0.00% |
0 / 1 |
2 | |||
| 1 | <?php |
| 2 | |
| 3 | namespace GrowthExperiments\NewcomerTasks\AddLink; |
| 4 | |
| 5 | use ChangeTags; |
| 6 | use Exception; |
| 7 | use GrowthExperiments\NewcomerTasks\ConfigurationLoader\ConfigurationLoader; |
| 8 | use GrowthExperiments\NewcomerTasks\TaskType\LinkRecommendationTaskType; |
| 9 | use GrowthExperiments\NewcomerTasks\TaskType\LinkRecommendationTaskTypeHandler; |
| 10 | use GrowthExperiments\WikiConfigException; |
| 11 | use MediaWiki\ChangeTags\ChangeTagsStore; |
| 12 | use MediaWiki\Content\WikitextContent; |
| 13 | use MediaWiki\Language\RawMessage; |
| 14 | use MediaWiki\Page\PageIdentityValue; |
| 15 | use MediaWiki\Page\PageProps; |
| 16 | use MediaWiki\Revision\RevisionRecord; |
| 17 | use MediaWiki\Revision\RevisionStore; |
| 18 | use MediaWiki\Revision\SlotRecord; |
| 19 | use MediaWiki\Status\Status; |
| 20 | use MediaWiki\Storage\NameTableStore; |
| 21 | use MediaWiki\Title\Title; |
| 22 | use MediaWiki\Utils\MWTimestamp; |
| 23 | use Psr\Log\LoggerInterface; |
| 24 | use StatusValue; |
| 25 | use Wikimedia\Rdbms\DBReadOnlyError; |
| 26 | use Wikimedia\Rdbms\IConnectionProvider; |
| 27 | use Wikimedia\Rdbms\IDatabase; |
| 28 | use Wikimedia\Rdbms\IDBAccessObject; |
| 29 | |
| 30 | /** |
| 31 | * Handles creating or updating a link recommendation entry. |
| 32 | * This includes fetching a recommendation from the service, validating it, and updating |
| 33 | * the database and the search index. |
| 34 | */ |
| 35 | class LinkRecommendationUpdater { |
| 36 | |
| 37 | private LoggerInterface $logger; |
| 38 | private IConnectionProvider $connectionProvider; |
| 39 | private RevisionStore $revisionStore; |
| 40 | private NameTableStore $changeDefNameTableStore; |
| 41 | private PageProps $pageProps; |
| 42 | private ConfigurationLoader $configurationLoader; |
| 43 | private ChangeTagsStore $changeTagsStore; |
| 44 | /** |
| 45 | * @var callable returning {@link \CirrusSearch\WeightedTagsUpdater} |
| 46 | */ |
| 47 | private $weightedTagsUpdaterProvider; |
| 48 | private LinkRecommendationStore $linkRecommendationStore; |
| 49 | private LinkRecommendationProvider $linkRecommendationProvider; |
| 50 | private ?LinkRecommendationTaskType $linkRecommendationTaskType = null; |
| 51 | |
| 52 | /** |
| 53 | * @param LoggerInterface $logger |
| 54 | * @param IConnectionProvider $connectionProvider |
| 55 | * @param RevisionStore $revisionStore |
| 56 | * @param NameTableStore $changeDefNameTableStore |
| 57 | * @param PageProps $pageProps |
| 58 | * @param ChangeTagsStore $changeTagsStore |
| 59 | * @param ConfigurationLoader $configurationLoader |
| 60 | * @param callable(): \CirrusSearch\WeightedTagsUpdater $weightedTagsUpdaterProvider |
| 61 | * @param LinkRecommendationProvider $linkRecommendationProvider Note that this needs to be |
| 62 | * the uncached provider, as caching is done by LinkRecommendationUpdater. |
| 63 | * @param LinkRecommendationStore $linkRecommendationStore |
| 64 | */ |
| 65 | public function __construct( |
| 66 | LoggerInterface $logger, |
| 67 | IConnectionProvider $connectionProvider, |
| 68 | RevisionStore $revisionStore, |
| 69 | NameTableStore $changeDefNameTableStore, |
| 70 | PageProps $pageProps, |
| 71 | ChangeTagsStore $changeTagsStore, |
| 72 | ConfigurationLoader $configurationLoader, |
| 73 | callable $weightedTagsUpdaterProvider, |
| 74 | LinkRecommendationProvider $linkRecommendationProvider, |
| 75 | LinkRecommendationStore $linkRecommendationStore |
| 76 | ) { |
| 77 | $this->logger = $logger; |
| 78 | $this->connectionProvider = $connectionProvider; |
| 79 | $this->revisionStore = $revisionStore; |
| 80 | $this->changeDefNameTableStore = $changeDefNameTableStore; |
| 81 | $this->pageProps = $pageProps; |
| 82 | $this->changeTagsStore = $changeTagsStore; |
| 83 | |
| 84 | $this->configurationLoader = $configurationLoader; |
| 85 | $this->weightedTagsUpdaterProvider = $weightedTagsUpdaterProvider; |
| 86 | $this->linkRecommendationStore = $linkRecommendationStore; |
| 87 | $this->linkRecommendationProvider = $linkRecommendationProvider; |
| 88 | } |
| 89 | |
| 90 | /** |
| 91 | * Evaluate a task candidate and generate the task if the candidate is viable. |
| 92 | * If a link recommendation task already exists for the given page, it will be overwritten. |
| 93 | * @param Title $title |
| 94 | * @param bool $force Ignore all failed conditions that can be safely ignored. |
| 95 | * @return StatusValue Success status. Note that the error messages are not intended |
| 96 | * for users (and as such not localizable). |
| 97 | * @throws WikiConfigException if the task type is not properly configured. |
| 98 | * @throws DBReadOnlyError |
| 99 | */ |
| 100 | public function processCandidate( Title $title, bool $force = false ): StatusValue { |
| 101 | $lastRevision = $this->revisionStore->getRevisionByTitle( $title ); |
| 102 | $status = $this->evaluateTitle( $title, $lastRevision, $force ); |
| 103 | if ( !$status->isOK() ) { |
| 104 | return $status; |
| 105 | } |
| 106 | |
| 107 | // Prevent infinite loop. Cirrus updates are not realtime so pages we have |
| 108 | // just created recommendations for will be included again in the next batch. |
| 109 | // Skip them to ensure $recommendationsFound is only nonzero then we have |
| 110 | // actually added a new recommendation. |
| 111 | // FIXME there is probably a better way to do this via search offsets. |
| 112 | if ( $this->linkRecommendationStore->getByRevId( $lastRevision->getId(), |
| 113 | IDBAccessObject::READ_LATEST ) |
| 114 | ) { |
| 115 | return $this->failure( 'link recommendation already stored' ); |
| 116 | } |
| 117 | |
| 118 | $recommendation = $this->linkRecommendationProvider->get( $title, |
| 119 | $this->getLinkRecommendationTaskType() ); |
| 120 | if ( $recommendation instanceof StatusValue ) { |
| 121 | // Returning a StatusValue is always an error for the provider. When returning it |
| 122 | // from this class, it isn't necessarily interpreted that way. |
| 123 | $recommendation->setOK( false ); |
| 124 | return $recommendation; |
| 125 | } |
| 126 | $status = $this->evaluateRecommendation( $recommendation, $lastRevision, $force ); |
| 127 | if ( !$status->isOK() ) { |
| 128 | return $status; |
| 129 | } |
| 130 | |
| 131 | // If an error happens later, uncommitted DB writes get discarded, while |
| 132 | // updateCirrusSearchIndex() is immediate. Minimize the likelihood of the DB |
| 133 | // and the search index getting out of sync by wrapping the insert into a |
| 134 | // transaction (in general start/endAtomic doesn't guarantee that but this method |
| 135 | // will usually be called from maintenance scripts). |
| 136 | $db = $this->linkRecommendationStore->getDB( DB_PRIMARY ); |
| 137 | $db->startAtomic( __METHOD__, IDatabase::ATOMIC_CANCELABLE ); |
| 138 | $this->linkRecommendationStore->insert( $recommendation ); |
| 139 | |
| 140 | $pageIdentity = new PageIdentityValue( |
| 141 | $lastRevision->getPageId( $lastRevision->getWikiId() ), |
| 142 | $lastRevision->getPage()->getNamespace(), |
| 143 | $lastRevision->getPage()->getDBkey(), |
| 144 | $lastRevision->getWikiId() |
| 145 | ); |
| 146 | |
| 147 | try { |
| 148 | ( $this->weightedTagsUpdaterProvider )()->updateWeightedTags( |
| 149 | $pageIdentity, |
| 150 | LinkRecommendationTaskTypeHandler::WEIGHTED_TAG_PREFIX |
| 151 | ); |
| 152 | } catch ( Exception $e ) { |
| 153 | $db->cancelAtomic( __METHOD__ ); |
| 154 | |
| 155 | $this->logger->error( __METHOD__ . ' failed to update weighted tags', [ |
| 156 | 'exception' => $e, |
| 157 | 'pageTitle' => $title->getPrefixedText(), |
| 158 | ] ); |
| 159 | return Status::newFatal( |
| 160 | 'Failed to request weighted tags update', |
| 161 | LinkRecommendationTaskTypeHandler::WEIGHTED_TAG_PREFIX, |
| 162 | (string)$e |
| 163 | ); |
| 164 | } |
| 165 | |
| 166 | $db->endAtomic( __METHOD__ ); |
| 167 | return StatusValue::newGood(); |
| 168 | } |
| 169 | |
| 170 | /** |
| 171 | * Check all conditions which are not related to the recommendation. |
| 172 | * @param Title $title The title for which a recommendation is being requested. |
| 173 | * @param RevisionRecord|null $revision The current revision of the title. |
| 174 | * @param bool $force Ignore all failed conditions that can be safely ignored. |
| 175 | * @return StatusValue Success status. Note that the error messages are not intended |
| 176 | * for users (and as such not localizable). |
| 177 | */ |
| 178 | private function evaluateTitle( Title $title, ?RevisionRecord $revision, bool $force ): StatusValue { |
| 179 | // 1. the revision must exist and the mwaddlink service must be able to interpret it. |
| 180 | if ( $revision === null ) { |
| 181 | // Maybe the article has just been deleted and the search index is behind? |
| 182 | return $this->failure( 'page not found' ); |
| 183 | } |
| 184 | $content = $revision->getContent( SlotRecord::MAIN ); |
| 185 | if ( !$content instanceof WikitextContent ) { |
| 186 | return $this->failure( 'content not found' ); |
| 187 | } |
| 188 | |
| 189 | if ( $force ) { |
| 190 | return StatusValue::newGood(); |
| 191 | } |
| 192 | |
| 193 | // 2. the article must match size conditions. |
| 194 | $wordCount = preg_match_all( '/\w+/', $content->getText() ); |
| 195 | if ( $wordCount < $this->getLinkRecommendationTaskType()->getMinimumWordCount() ) { |
| 196 | return $this->failure( "word count too small ($wordCount)" ); |
| 197 | } elseif ( $wordCount > $this->getLinkRecommendationTaskType()->getMaximumWordCount() ) { |
| 198 | return $this->failure( "word count too large ($wordCount)" ); |
| 199 | } |
| 200 | |
| 201 | // 3. exclude articles which have been edited very recently. |
| 202 | $revisionTime = (int)MWTimestamp::convert( TS_UNIX, $revision->getTimestamp() ); |
| 203 | if ( time() - $revisionTime < $this->getLinkRecommendationTaskType()->getMinimumTimeSinceLastEdit() ) { |
| 204 | return $this->failure( 'minimum time since last edit did not pass' ); |
| 205 | } |
| 206 | |
| 207 | // 4. exclude disambiguation pages. |
| 208 | if ( $this->pageProps->getProperties( $title, 'disambiguation' ) ) { |
| 209 | return $this->failure( 'disambiguation page' ); |
| 210 | } |
| 211 | |
| 212 | // 5. exclude pages where the last edit is a link recommendation edit or its revert. |
| 213 | $dbr = $this->connectionProvider->getReplicaDatabase(); |
| 214 | $tags = $this->changeTagsStore->getTagsWithData( $dbr, null, $revision->getId() ); |
| 215 | if ( array_key_exists( LinkRecommendationTaskTypeHandler::CHANGE_TAG, $tags ) ) { |
| 216 | return $this->failure( 'last edit is a link recommendation' ); |
| 217 | } |
| 218 | $revertTagData = null; |
| 219 | foreach ( ChangeTags::REVERT_TAGS as $revertTagName ) { |
| 220 | if ( !empty( $tags[$revertTagName] ) ) { |
| 221 | $revertTagData = json_decode( $tags[$revertTagName], true ); |
| 222 | break; |
| 223 | } |
| 224 | } |
| 225 | if ( is_array( $revertTagData ) ) { |
| 226 | $linkRecommendationChangeTagId = $this->changeDefNameTableStore |
| 227 | ->acquireId( LinkRecommendationTaskTypeHandler::CHANGE_TAG ); |
| 228 | $revertedAddLinkEditCount = $dbr->newSelectQueryBuilder() |
| 229 | ->from( 'revision' ) |
| 230 | ->join( 'change_tag', null, [ 'rev_id = ct_rev_id' ] ) |
| 231 | ->where( [ |
| 232 | 'rev_page' => $title->getArticleID(), |
| 233 | $dbr->expr( 'rev_id', '<=', (int)$revertTagData['newestRevertedRevId'] ), |
| 234 | $dbr->expr( 'rev_id', '>=', (int)$revertTagData['oldestRevertedRevId'] ), |
| 235 | 'ct_tag_id' => $linkRecommendationChangeTagId, |
| 236 | ] ) |
| 237 | ->caller( __METHOD__ ) |
| 238 | ->fetchRowCount(); |
| 239 | if ( $revertedAddLinkEditCount > 0 ) { |
| 240 | return $this->failure( 'last edit reverts a link recommendation edit' ); |
| 241 | } |
| 242 | } |
| 243 | return StatusValue::newGood(); |
| 244 | } |
| 245 | |
| 246 | /** |
| 247 | * Validate a recommendation against the criteria in the task type and safety checks. |
| 248 | * @param LinkRecommendation $recommendation |
| 249 | * @param RevisionRecord $revision The current revision of the page the recommendation is for. |
| 250 | * @param bool $force Ignore all failed conditions that can be safely ignored. |
| 251 | * @return StatusValue Success status. Note that the error messages are not intended |
| 252 | * for users (and as such not localizable). |
| 253 | */ |
| 254 | private function evaluateRecommendation( |
| 255 | LinkRecommendation $recommendation, |
| 256 | RevisionRecord $revision, |
| 257 | bool $force |
| 258 | ): StatusValue { |
| 259 | if ( $recommendation->getRevisionId() !== $revision->getId() ) { |
| 260 | // Some kind of race condition? Generating another task is easy so just discard this. |
| 261 | return $this->failure( 'revision ID mismatch' ); |
| 262 | } |
| 263 | |
| 264 | // T291253 |
| 265 | if ( !$force && $this->linkRecommendationStore->hasSubmission( $recommendation, |
| 266 | IDBAccessObject::READ_LATEST ) |
| 267 | ) { |
| 268 | return $this->failure( 'submission already exists for revision ' . $revision->getId() ); |
| 269 | } |
| 270 | |
| 271 | // We could check here for more race conditions, ie. whether the revision in the |
| 272 | // recommendation matches the live revision. But there are plenty of other ways for race |
| 273 | // conditions to happen, so we'll have to deal with them on the client side anyway. No |
| 274 | // point in getting a primary database connection just for that. |
| 275 | |
| 276 | $goodLinks = array_filter( $recommendation->getLinks(), function ( LinkRecommendationLink $link ) { |
| 277 | return $link->getScore() >= $this->getLinkRecommendationTaskType()->getMinimumLinkScore(); |
| 278 | } ); |
| 279 | $recommendation = new LinkRecommendation( |
| 280 | $recommendation->getTitle(), |
| 281 | $recommendation->getPageId(), |
| 282 | $recommendation->getRevisionId(), |
| 283 | $goodLinks, |
| 284 | $recommendation->getMetadata() |
| 285 | ); |
| 286 | $goodLinkCount = count( $recommendation->getLinks() ); |
| 287 | if ( $goodLinkCount === 0 |
| 288 | || ( !$force && $goodLinkCount < $this->getLinkRecommendationTaskType()->getMinimumLinksPerTask() ) |
| 289 | ) { |
| 290 | return $this->failure( "number of good links too small ($goodLinkCount)" ); |
| 291 | } |
| 292 | |
| 293 | return StatusValue::newGood(); |
| 294 | } |
| 295 | |
| 296 | /** |
| 297 | * Internal helper for loading the Add Link task type. Due to the involvement of on-wiki |
| 298 | * configuration, this is not available at setup time so it cannot be dependency-injected. |
| 299 | * @return LinkRecommendationTaskType |
| 300 | * @throws WikiConfigException if the task type is not properly configured. |
| 301 | */ |
| 302 | private function getLinkRecommendationTaskType(): LinkRecommendationTaskType { |
| 303 | if ( !$this->linkRecommendationTaskType ) { |
| 304 | $taskTypes = $this->configurationLoader->loadTaskTypes(); |
| 305 | if ( $taskTypes instanceof StatusValue ) { |
| 306 | throw new WikiConfigException( 'Could not load task types: ' . |
| 307 | Status::wrap( $taskTypes )->getWikiText( false, false, 'en' ) ); |
| 308 | } |
| 309 | $taskTypes = $this->configurationLoader->getTaskTypes() + |
| 310 | $this->configurationLoader->getDisabledTaskTypes(); |
| 311 | $taskType = $taskTypes[LinkRecommendationTaskTypeHandler::TASK_TYPE_ID] ?? null; |
| 312 | if ( !( $taskType instanceof LinkRecommendationTaskType ) ) { |
| 313 | throw new WikiConfigException( 'Could not load link recommendation task type' ); |
| 314 | } |
| 315 | $this->linkRecommendationTaskType = $taskType; |
| 316 | } |
| 317 | return $this->linkRecommendationTaskType; |
| 318 | } |
| 319 | |
| 320 | /** |
| 321 | * Convenience shortcut for making StatusValue objects with non-localized messages. |
| 322 | * @param string $error |
| 323 | * @return StatusValue |
| 324 | */ |
| 325 | private function failure( string $error ): StatusValue { |
| 326 | return StatusValue::newFatal( new RawMessage( $error ) ); |
| 327 | } |
| 328 | |
| 329 | } |