Make MessageCache::load() require a language code
[lhc/web/wiklou.git] / includes / cache / MessageCache.php
1 <?php
2 /**
3 * Localisation messages cache.
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
18 * http://www.gnu.org/copyleft/gpl.html
19 *
20 * @file
21 * @ingroup Cache
22 */
23 use MediaWiki\MediaWikiServices;
24
25 /**
26 * MediaWiki message cache structure version.
27 * Bump this whenever the message cache format has changed.
28 */
29 define( 'MSG_CACHE_VERSION', 2 );
30
31 /**
32 * Message cache
33 * Performs various MediaWiki namespace-related functions
34 * @ingroup Cache
35 */
36 class MessageCache {
37 const FOR_UPDATE = 1; // force message reload
38
39 /** How long to wait for memcached locks */
40 const WAIT_SEC = 15;
41 /** How long memcached locks last */
42 const LOCK_TTL = 30;
43
44 /**
45 * Process local cache of loaded messages that are defined in
46 * MediaWiki namespace. First array level is a language code,
47 * second level is message key and the values are either message
48 * content prefixed with space, or !NONEXISTENT for negative
49 * caching.
50 * @var array $mCache
51 */
52 protected $mCache;
53
54 /**
55 * Should mean that database cannot be used, but check
56 * @var bool $mDisable
57 */
58 protected $mDisable;
59
60 /**
61 * Lifetime for cache, used by object caching.
62 * Set on construction, see __construct().
63 */
64 protected $mExpiry;
65
66 /**
67 * Message cache has its own parser which it uses to transform
68 * messages.
69 */
70 protected $mParserOptions, $mParser;
71
72 /**
73 * Variable for tracking which variables are already loaded
74 * @var array $mLoadedLanguages
75 */
76 protected $mLoadedLanguages = [];
77
78 /**
79 * @var bool $mInParser
80 */
81 protected $mInParser = false;
82
83 /** @var BagOStuff */
84 protected $mMemc;
85 /** @var WANObjectCache */
86 protected $wanCache;
87
88 /**
89 * Singleton instance
90 *
91 * @var MessageCache $instance
92 */
93 private static $instance;
94
95 /**
96 * Get the signleton instance of this class
97 *
98 * @since 1.18
99 * @return MessageCache
100 */
101 public static function singleton() {
102 if ( self::$instance === null ) {
103 global $wgUseDatabaseMessages, $wgMsgCacheExpiry;
104 self::$instance = new self(
105 wfGetMessageCacheStorage(),
106 $wgUseDatabaseMessages,
107 $wgMsgCacheExpiry
108 );
109 }
110
111 return self::$instance;
112 }
113
114 /**
115 * Destroy the singleton instance
116 *
117 * @since 1.18
118 */
119 public static function destroyInstance() {
120 self::$instance = null;
121 }
122
123 /**
124 * Normalize message key input
125 *
126 * @param string $key Input message key to be normalized
127 * @return string Normalized message key
128 */
129 public static function normalizeKey( $key ) {
130 global $wgContLang;
131 $lckey = strtr( $key, ' ', '_' );
132 if ( ord( $lckey ) < 128 ) {
133 $lckey[0] = strtolower( $lckey[0] );
134 } else {
135 $lckey = $wgContLang->lcfirst( $lckey );
136 }
137
138 return $lckey;
139 }
140
141 /**
142 * @param BagOStuff $memCached A cache instance. If none, fall back to CACHE_NONE.
143 * @param bool $useDB
144 * @param int $expiry Lifetime for cache. @see $mExpiry.
145 */
146 function __construct( $memCached, $useDB, $expiry ) {
147 global $wgUseLocalMessageCache;
148
149 if ( !$memCached ) {
150 $memCached = wfGetCache( CACHE_NONE );
151 }
152
153 $this->mMemc = $memCached;
154 $this->mDisable = !$useDB;
155 $this->mExpiry = $expiry;
156
157 if ( $wgUseLocalMessageCache ) {
158 $this->localCache = MediaWikiServices::getInstance()->getLocalServerObjectCache();
159 } else {
160 $this->localCache = new EmptyBagOStuff();
161 }
162
163 $this->wanCache = ObjectCache::getMainWANInstance();
164 }
165
166 /**
167 * ParserOptions is lazy initialised.
168 *
169 * @return ParserOptions
170 */
171 function getParserOptions() {
172 global $wgUser;
173
174 if ( !$this->mParserOptions ) {
175 if ( !$wgUser->isSafeToLoad() ) {
176 // $wgUser isn't unstubbable yet, so don't try to get a
177 // ParserOptions for it. And don't cache this ParserOptions
178 // either.
179 $po = ParserOptions::newFromAnon();
180 $po->setEditSection( false );
181 return $po;
182 }
183
184 $this->mParserOptions = new ParserOptions;
185 $this->mParserOptions->setEditSection( false );
186 }
187
188 return $this->mParserOptions;
189 }
190
191 /**
192 * Try to load the cache from APC.
193 *
194 * @param string $code Optional language code, see documenation of load().
195 * @return array|bool The cache array, or false if not in cache.
196 */
197 protected function getLocalCache( $code ) {
198 $cacheKey = wfMemcKey( __CLASS__, $code );
199
200 return $this->localCache->get( $cacheKey );
201 }
202
203 /**
204 * Save the cache to APC.
205 *
206 * @param string $code
207 * @param array $cache The cache array
208 */
209 protected function saveToLocalCache( $code, $cache ) {
210 $cacheKey = wfMemcKey( __CLASS__, $code );
211 $this->localCache->set( $cacheKey, $cache );
212 }
213
214 /**
215 * Loads messages from caches or from database in this order:
216 * (1) local message cache (if $wgUseLocalMessageCache is enabled)
217 * (2) memcached
218 * (3) from the database.
219 *
220 * When succesfully loading from (2) or (3), all higher level caches are
221 * updated for the newest version.
222 *
223 * Nothing is loaded if member variable mDisable is true, either manually
224 * set by calling code or if message loading fails (is this possible?).
225 *
226 * Returns true if cache is already populated or it was succesfully populated,
227 * or false if populating empty cache fails. Also returns true if MessageCache
228 * is disabled.
229 *
230 * @param string $code Language to which load messages
231 * @param integer $mode Use MessageCache::FOR_UPDATE to skip process cache [optional]
232 * @throws MWException
233 * @return bool
234 */
235 protected function load( $code, $mode = null ) {
236 if ( !is_string( $code ) ) {
237 throw new InvalidArgumentException( "Missing language code" );
238 }
239
240 # Don't do double loading...
241 if ( isset( $this->mLoadedLanguages[$code] ) && $mode != self::FOR_UPDATE ) {
242 return true;
243 }
244
245 # 8 lines of code just to say (once) that message cache is disabled
246 if ( $this->mDisable ) {
247 static $shownDisabled = false;
248 if ( !$shownDisabled ) {
249 wfDebug( __METHOD__ . ": disabled\n" );
250 $shownDisabled = true;
251 }
252
253 return true;
254 }
255
256 # Loading code starts
257 $success = false; # Keep track of success
258 $staleCache = false; # a cache array with expired data, or false if none has been loaded
259 $where = []; # Debug info, delayed to avoid spamming debug log too much
260
261 # Hash of the contents is stored in memcache, to detect if data-center cache
262 # or local cache goes out of date (e.g. due to replace() on some other server)
263 list( $hash, $hashVolatile ) = $this->getValidationHash( $code );
264
265 # Try the local cache and check against the cluster hash key...
266 $cache = $this->getLocalCache( $code );
267 if ( !$cache ) {
268 $where[] = 'local cache is empty';
269 } elseif ( !isset( $cache['HASH'] ) || $cache['HASH'] !== $hash ) {
270 $where[] = 'local cache has the wrong hash';
271 $staleCache = $cache;
272 } elseif ( $this->isCacheExpired( $cache ) ) {
273 $where[] = 'local cache is expired';
274 $staleCache = $cache;
275 } elseif ( $hashVolatile ) {
276 $where[] = 'local cache validation key is expired/volatile';
277 $staleCache = $cache;
278 } else {
279 $where[] = 'got from local cache';
280 $success = true;
281 $this->mCache[$code] = $cache;
282 }
283
284 if ( !$success ) {
285 $cacheKey = wfMemcKey( 'messages', $code ); # Key in memc for messages
286 # Try the global cache. If it is empty, try to acquire a lock. If
287 # the lock can't be acquired, wait for the other thread to finish
288 # and then try the global cache a second time.
289 for ( $failedAttempts = 0; $failedAttempts <= 1; $failedAttempts++ ) {
290 if ( $hashVolatile && $staleCache ) {
291 # Do not bother fetching the whole cache blob to avoid I/O.
292 # Instead, just try to get the non-blocking $statusKey lock
293 # below, and use the local stale value if it was not acquired.
294 $where[] = 'global cache is presumed expired';
295 } else {
296 $cache = $this->mMemc->get( $cacheKey );
297 if ( !$cache ) {
298 $where[] = 'global cache is empty';
299 } elseif ( $this->isCacheExpired( $cache ) ) {
300 $where[] = 'global cache is expired';
301 $staleCache = $cache;
302 } elseif ( $hashVolatile ) {
303 # DB results are replica DB lag prone until the holdoff TTL passes.
304 # By then, updates should be reflected in loadFromDBWithLock().
305 # One thread renerates the cache while others use old values.
306 $where[] = 'global cache is expired/volatile';
307 $staleCache = $cache;
308 } else {
309 $where[] = 'got from global cache';
310 $this->mCache[$code] = $cache;
311 $this->saveToCaches( $cache, 'local-only', $code );
312 $success = true;
313 }
314 }
315
316 if ( $success ) {
317 # Done, no need to retry
318 break;
319 }
320
321 # We need to call loadFromDB. Limit the concurrency to one process.
322 # This prevents the site from going down when the cache expires.
323 # Note that the DB slam protection lock here is non-blocking.
324 $loadStatus = $this->loadFromDBWithLock( $code, $where, $mode );
325 if ( $loadStatus === true ) {
326 $success = true;
327 break;
328 } elseif ( $staleCache ) {
329 # Use the stale cache while some other thread constructs the new one
330 $where[] = 'using stale cache';
331 $this->mCache[$code] = $staleCache;
332 $success = true;
333 break;
334 } elseif ( $failedAttempts > 0 ) {
335 # Already blocked once, so avoid another lock/unlock cycle.
336 # This case will typically be hit if memcached is down, or if
337 # loadFromDB() takes longer than LOCK_WAIT.
338 $where[] = "could not acquire status key.";
339 break;
340 } elseif ( $loadStatus === 'cantacquire' ) {
341 # Wait for the other thread to finish, then retry. Normally,
342 # the memcached get() will then yeild the other thread's result.
343 $where[] = 'waited for other thread to complete';
344 $this->getReentrantScopedLock( $cacheKey );
345 } else {
346 # Disable cache; $loadStatus is 'disabled'
347 break;
348 }
349 }
350 }
351
352 if ( !$success ) {
353 $where[] = 'loading FAILED - cache is disabled';
354 $this->mDisable = true;
355 $this->mCache = false;
356 wfDebugLog( 'MessageCacheError', __METHOD__ . ": Failed to load $code\n" );
357 # This used to throw an exception, but that led to nasty side effects like
358 # the whole wiki being instantly down if the memcached server died
359 } else {
360 # All good, just record the success
361 $this->mLoadedLanguages[$code] = true;
362 }
363
364 $info = implode( ', ', $where );
365 wfDebugLog( 'MessageCache', __METHOD__ . ": Loading $code... $info\n" );
366
367 return $success;
368 }
369
370 /**
371 * @param string $code
372 * @param array $where List of wfDebug() comments
373 * @param integer $mode Use MessageCache::FOR_UPDATE to use DB_MASTER
374 * @return bool|string True on success or one of ("cantacquire", "disabled")
375 */
376 protected function loadFromDBWithLock( $code, array &$where, $mode = null ) {
377 global $wgUseLocalMessageCache;
378
379 # If cache updates on all levels fail, give up on message overrides.
380 # This is to avoid easy site outages; see $saveSuccess comments below.
381 $statusKey = wfMemcKey( 'messages', $code, 'status' );
382 $status = $this->mMemc->get( $statusKey );
383 if ( $status === 'error' ) {
384 $where[] = "could not load; method is still globally disabled";
385 return 'disabled';
386 }
387
388 # Now let's regenerate
389 $where[] = 'loading from database';
390
391 # Lock the cache to prevent conflicting writes.
392 # This lock is non-blocking so stale cache can quickly be used.
393 # Note that load() will call a blocking getReentrantScopedLock()
394 # after this if it really need to wait for any current thread.
395 $cacheKey = wfMemcKey( 'messages', $code );
396 $scopedLock = $this->getReentrantScopedLock( $cacheKey, 0 );
397 if ( !$scopedLock ) {
398 $where[] = 'could not acquire main lock';
399 return 'cantacquire';
400 }
401
402 $cache = $this->loadFromDB( $code, $mode );
403 $this->mCache[$code] = $cache;
404 $saveSuccess = $this->saveToCaches( $cache, 'all', $code );
405
406 if ( !$saveSuccess ) {
407 /**
408 * Cache save has failed.
409 *
410 * There are two main scenarios where this could be a problem:
411 * - The cache is more than the maximum size (typically 1MB compressed).
412 * - Memcached has no space remaining in the relevant slab class. This is
413 * unlikely with recent versions of memcached.
414 *
415 * Either way, if there is a local cache, nothing bad will happen. If there
416 * is no local cache, disabling the message cache for all requests avoids
417 * incurring a loadFromDB() overhead on every request, and thus saves the
418 * wiki from complete downtime under moderate traffic conditions.
419 */
420 if ( !$wgUseLocalMessageCache ) {
421 $this->mMemc->set( $statusKey, 'error', 60 * 5 );
422 $where[] = 'could not save cache, disabled globally for 5 minutes';
423 } else {
424 $where[] = "could not save global cache";
425 }
426 }
427
428 return true;
429 }
430
431 /**
432 * Loads cacheable messages from the database. Messages bigger than
433 * $wgMaxMsgCacheEntrySize are assigned a special value, and are loaded
434 * on-demand from the database later.
435 *
436 * @param string $code Language code
437 * @param integer $mode Use MessageCache::FOR_UPDATE to skip process cache
438 * @return array Loaded messages for storing in caches
439 */
440 function loadFromDB( $code, $mode = null ) {
441 global $wgMaxMsgCacheEntrySize, $wgLanguageCode, $wgAdaptiveMessageCache;
442
443 $dbr = wfGetDB( ( $mode == self::FOR_UPDATE ) ? DB_MASTER : DB_REPLICA );
444
445 $cache = [];
446
447 # Common conditions
448 $conds = [
449 'page_is_redirect' => 0,
450 'page_namespace' => NS_MEDIAWIKI,
451 ];
452
453 $mostused = [];
454 if ( $wgAdaptiveMessageCache && $code !== $wgLanguageCode ) {
455 if ( !isset( $this->mCache[$wgLanguageCode] ) ) {
456 $this->load( $wgLanguageCode );
457 }
458 $mostused = array_keys( $this->mCache[$wgLanguageCode] );
459 foreach ( $mostused as $key => $value ) {
460 $mostused[$key] = "$value/$code";
461 }
462 }
463
464 if ( count( $mostused ) ) {
465 $conds['page_title'] = $mostused;
466 } elseif ( $code !== $wgLanguageCode ) {
467 $conds[] = 'page_title' . $dbr->buildLike( $dbr->anyString(), '/', $code );
468 } else {
469 # Effectively disallows use of '/' character in NS_MEDIAWIKI for uses
470 # other than language code.
471 $conds[] = 'page_title NOT' . $dbr->buildLike( $dbr->anyString(), '/', $dbr->anyString() );
472 }
473
474 # Conditions to fetch oversized pages to ignore them
475 $bigConds = $conds;
476 $bigConds[] = 'page_len > ' . intval( $wgMaxMsgCacheEntrySize );
477
478 # Load titles for all oversized pages in the MediaWiki namespace
479 $res = $dbr->select( 'page', 'page_title', $bigConds, __METHOD__ . "($code)-big" );
480 foreach ( $res as $row ) {
481 $cache[$row->page_title] = '!TOO BIG';
482 }
483
484 # Conditions to load the remaining pages with their contents
485 $smallConds = $conds;
486 $smallConds[] = 'page_latest=rev_id';
487 $smallConds[] = 'rev_text_id=old_id';
488 $smallConds[] = 'page_len <= ' . intval( $wgMaxMsgCacheEntrySize );
489
490 $res = $dbr->select(
491 [ 'page', 'revision', 'text' ],
492 [ 'page_title', 'old_text', 'old_flags' ],
493 $smallConds,
494 __METHOD__ . "($code)-small"
495 );
496
497 foreach ( $res as $row ) {
498 $text = Revision::getRevisionText( $row );
499 if ( $text === false ) {
500 // Failed to fetch data; possible ES errors?
501 // Store a marker to fetch on-demand as a workaround...
502 $entry = '!TOO BIG';
503 wfDebugLog(
504 'MessageCache',
505 __METHOD__
506 . ": failed to load message page text for {$row->page_title} ($code)"
507 );
508 } else {
509 $entry = ' ' . $text;
510 }
511 $cache[$row->page_title] = $entry;
512 }
513
514 $cache['VERSION'] = MSG_CACHE_VERSION;
515 ksort( $cache );
516 $cache['HASH'] = md5( serialize( $cache ) );
517 $cache['EXPIRY'] = wfTimestamp( TS_MW, time() + $this->mExpiry );
518
519 return $cache;
520 }
521
522 /**
523 * Updates cache as necessary when message page is changed
524 *
525 * @param string|bool $title Name of the page changed (false if deleted)
526 * @param mixed $text New contents of the page.
527 */
528 public function replace( $title, $text ) {
529 global $wgMaxMsgCacheEntrySize, $wgContLang, $wgLanguageCode;
530
531 if ( $this->mDisable ) {
532 return;
533 }
534
535 list( $msg, $code ) = $this->figureMessage( $title );
536 if ( strpos( $title, '/' ) !== false && $code === $wgLanguageCode ) {
537 // Content language overrides do not use the /<code> suffix
538 return;
539 }
540
541 // Note that if the cache is volatile, load() may trigger a DB fetch.
542 // In that case we reenter/reuse the existing cache key lock to avoid
543 // a self-deadlock. This is safe as no reads happen *directly* in this
544 // method between getReentrantScopedLock() and load() below. There is
545 // no risk of data "changing under our feet" for replace().
546 $cacheKey = wfMemcKey( 'messages', $code );
547 $scopedLock = $this->getReentrantScopedLock( $cacheKey );
548 $this->load( $code, self::FOR_UPDATE );
549
550 $titleKey = wfMemcKey( 'messages', 'individual', $title );
551 if ( $text === false ) {
552 // Article was deleted
553 $this->mCache[$code][$title] = '!NONEXISTENT';
554 $this->wanCache->delete( $titleKey );
555 } elseif ( strlen( $text ) > $wgMaxMsgCacheEntrySize ) {
556 // Check for size
557 $this->mCache[$code][$title] = '!TOO BIG';
558 $this->wanCache->set( $titleKey, ' ' . $text, $this->mExpiry );
559 } else {
560 $this->mCache[$code][$title] = ' ' . $text;
561 $this->wanCache->delete( $titleKey );
562 }
563
564 // Mark this cache as definitely "latest" (non-volatile) so
565 // load() calls do try to refresh the cache with replica DB data
566 $this->mCache[$code]['LATEST'] = time();
567
568 // Update caches if the lock was acquired
569 if ( $scopedLock ) {
570 $this->saveToCaches( $this->mCache[$code], 'all', $code );
571 }
572
573 ScopedCallback::consume( $scopedLock );
574 // Relay the purge to APC and other DCs
575 $this->wanCache->touchCheckKey( wfMemcKey( 'messages', $code ) );
576
577 // Also delete cached sidebar... just in case it is affected
578 $codes = [ $code ];
579 if ( $code === 'en' ) {
580 // Delete all sidebars, like for example on action=purge on the
581 // sidebar messages
582 $codes = array_keys( Language::fetchLanguageNames() );
583 }
584
585 foreach ( $codes as $code ) {
586 $sidebarKey = wfMemcKey( 'sidebar', $code );
587 $this->wanCache->delete( $sidebarKey );
588 }
589
590 // Update the message in the message blob store
591 $resourceloader = RequestContext::getMain()->getOutput()->getResourceLoader();
592 $blobStore = $resourceloader->getMessageBlobStore();
593 $blobStore->updateMessage( $wgContLang->lcfirst( $msg ) );
594
595 Hooks::run( 'MessageCacheReplace', [ $title, $text ] );
596 }
597
598 /**
599 * Is the given cache array expired due to time passing or a version change?
600 *
601 * @param array $cache
602 * @return bool
603 */
604 protected function isCacheExpired( $cache ) {
605 if ( !isset( $cache['VERSION'] ) || !isset( $cache['EXPIRY'] ) ) {
606 return true;
607 }
608 if ( $cache['VERSION'] != MSG_CACHE_VERSION ) {
609 return true;
610 }
611 if ( wfTimestampNow() >= $cache['EXPIRY'] ) {
612 return true;
613 }
614
615 return false;
616 }
617
618 /**
619 * Shortcut to update caches.
620 *
621 * @param array $cache Cached messages with a version.
622 * @param string $dest Either "local-only" to save to local caches only
623 * or "all" to save to all caches.
624 * @param string|bool $code Language code (default: false)
625 * @return bool
626 */
627 protected function saveToCaches( array $cache, $dest, $code = false ) {
628 if ( $dest === 'all' ) {
629 $cacheKey = wfMemcKey( 'messages', $code );
630 $success = $this->mMemc->set( $cacheKey, $cache );
631 } else {
632 $success = true;
633 }
634
635 $this->setValidationHash( $code, $cache );
636 $this->saveToLocalCache( $code, $cache );
637
638 return $success;
639 }
640
641 /**
642 * Get the md5 used to validate the local APC cache
643 *
644 * @param string $code
645 * @return array (hash or false, bool expiry/volatility status)
646 */
647 protected function getValidationHash( $code ) {
648 $curTTL = null;
649 $value = $this->wanCache->get(
650 wfMemcKey( 'messages', $code, 'hash', 'v1' ),
651 $curTTL,
652 [ wfMemcKey( 'messages', $code ) ]
653 );
654
655 if ( !$value ) {
656 // No hash found at all; cache must regenerate to be safe
657 $hash = false;
658 $expired = true;
659 } else {
660 $hash = $value['hash'];
661 if ( ( time() - $value['latest'] ) < WANObjectCache::HOLDOFF_TTL ) {
662 // Cache was recently updated via replace() and should be up-to-date
663 $expired = false;
664 } else {
665 // See if the "check" key was bumped after the hash was generated
666 $expired = ( $curTTL < 0 );
667 }
668 }
669
670 return [ $hash, $expired ];
671 }
672
673 /**
674 * Set the md5 used to validate the local disk cache
675 *
676 * If $cache has a 'LATEST' UNIX timestamp key, then the hash will not
677 * be treated as "volatile" by getValidationHash() for the next few seconds
678 *
679 * @param string $code
680 * @param array $cache Cached messages with a version
681 */
682 protected function setValidationHash( $code, array $cache ) {
683 $this->wanCache->set(
684 wfMemcKey( 'messages', $code, 'hash', 'v1' ),
685 [
686 'hash' => $cache['HASH'],
687 'latest' => isset( $cache['LATEST'] ) ? $cache['LATEST'] : 0
688 ],
689 WANObjectCache::TTL_INDEFINITE
690 );
691 }
692
693 /**
694 * @param string $key A language message cache key that stores blobs
695 * @param integer $timeout Wait timeout in seconds
696 * @return null|ScopedCallback
697 */
698 protected function getReentrantScopedLock( $key, $timeout = self::WAIT_SEC ) {
699 return $this->mMemc->getScopedLock( $key, $timeout, self::LOCK_TTL, __METHOD__ );
700 }
701
702 /**
703 * Get a message from either the content language or the user language.
704 *
705 * First, assemble a list of languages to attempt getting the message from. This
706 * chain begins with the requested language and its fallbacks and then continues with
707 * the content language and its fallbacks. For each language in the chain, the following
708 * process will occur (in this order):
709 * 1. If a language-specific override, i.e., [[MW:msg/lang]], is available, use that.
710 * Note: for the content language, there is no /lang subpage.
711 * 2. Fetch from the static CDB cache.
712 * 3. If available, check the database for fallback language overrides.
713 *
714 * This process provides a number of guarantees. When changing this code, make sure all
715 * of these guarantees are preserved.
716 * * If the requested language is *not* the content language, then the CDB cache for that
717 * specific language will take precedence over the root database page ([[MW:msg]]).
718 * * Fallbacks will be just that: fallbacks. A fallback language will never be reached if
719 * the message is available *anywhere* in the language for which it is a fallback.
720 *
721 * @param string $key The message key
722 * @param bool $useDB If true, look for the message in the DB, false
723 * to use only the compiled l10n cache.
724 * @param bool|string|object $langcode Code of the language to get the message for.
725 * - If string and a valid code, will create a standard language object
726 * - If string but not a valid code, will create a basic language object
727 * - If boolean and false, create object from the current users language
728 * - If boolean and true, create object from the wikis content language
729 * - If language object, use it as given
730 * @param bool $isFullKey Specifies whether $key is a two part key "msg/lang".
731 *
732 * @throws MWException When given an invalid key
733 * @return string|bool False if the message doesn't exist, otherwise the
734 * message (which can be empty)
735 */
736 function get( $key, $useDB = true, $langcode = true, $isFullKey = false ) {
737 if ( is_int( $key ) ) {
738 // Fix numerical strings that somehow become ints
739 // on their way here
740 $key = (string)$key;
741 } elseif ( !is_string( $key ) ) {
742 throw new MWException( 'Non-string key given' );
743 } elseif ( $key === '' ) {
744 // Shortcut: the empty key is always missing
745 return false;
746 }
747
748 // For full keys, get the language code from the key
749 $pos = strrpos( $key, '/' );
750 if ( $isFullKey && $pos !== false ) {
751 $langcode = substr( $key, $pos + 1 );
752 $key = substr( $key, 0, $pos );
753 }
754
755 // Normalise title-case input (with some inlining)
756 $lckey = MessageCache::normalizeKey( $key );
757
758 Hooks::run( 'MessageCache::get', [ &$lckey ] );
759
760 // Loop through each language in the fallback list until we find something useful
761 $lang = wfGetLangObj( $langcode );
762 $message = $this->getMessageFromFallbackChain(
763 $lang,
764 $lckey,
765 !$this->mDisable && $useDB
766 );
767
768 // If we still have no message, maybe the key was in fact a full key so try that
769 if ( $message === false ) {
770 $parts = explode( '/', $lckey );
771 // We may get calls for things that are http-urls from sidebar
772 // Let's not load nonexistent languages for those
773 // They usually have more than one slash.
774 if ( count( $parts ) == 2 && $parts[1] !== '' ) {
775 $message = Language::getMessageFor( $parts[0], $parts[1] );
776 if ( $message === null ) {
777 $message = false;
778 }
779 }
780 }
781
782 // Post-processing if the message exists
783 if ( $message !== false ) {
784 // Fix whitespace
785 $message = str_replace(
786 [
787 # Fix for trailing whitespace, removed by textarea
788 '&#32;',
789 # Fix for NBSP, converted to space by firefox
790 '&nbsp;',
791 '&#160;',
792 '&shy;'
793 ],
794 [
795 ' ',
796 "\xc2\xa0",
797 "\xc2\xa0",
798 "\xc2\xad"
799 ],
800 $message
801 );
802 }
803
804 return $message;
805 }
806
807 /**
808 * Given a language, try and fetch messages from that language.
809 *
810 * Will also consider fallbacks of that language, the site language, and fallbacks for
811 * the site language.
812 *
813 * @see MessageCache::get
814 * @param Language|StubObject $lang Preferred language
815 * @param string $lckey Lowercase key for the message (as for localisation cache)
816 * @param bool $useDB Whether to include messages from the wiki database
817 * @return string|bool The message, or false if not found
818 */
819 protected function getMessageFromFallbackChain( $lang, $lckey, $useDB ) {
820 global $wgContLang;
821
822 $alreadyTried = [];
823
824 // First try the requested language.
825 $message = $this->getMessageForLang( $lang, $lckey, $useDB, $alreadyTried );
826 if ( $message !== false ) {
827 return $message;
828 }
829
830 // Now try checking the site language.
831 $message = $this->getMessageForLang( $wgContLang, $lckey, $useDB, $alreadyTried );
832 return $message;
833 }
834
835 /**
836 * Given a language, try and fetch messages from that language and its fallbacks.
837 *
838 * @see MessageCache::get
839 * @param Language|StubObject $lang Preferred language
840 * @param string $lckey Lowercase key for the message (as for localisation cache)
841 * @param bool $useDB Whether to include messages from the wiki database
842 * @param bool[] $alreadyTried Contains true for each language that has been tried already
843 * @return string|bool The message, or false if not found
844 */
845 private function getMessageForLang( $lang, $lckey, $useDB, &$alreadyTried ) {
846 global $wgContLang;
847 $langcode = $lang->getCode();
848
849 // Try checking the database for the requested language
850 if ( $useDB ) {
851 $uckey = $wgContLang->ucfirst( $lckey );
852
853 if ( !isset( $alreadyTried[ $langcode ] ) ) {
854 $message = $this->getMsgFromNamespace(
855 $this->getMessagePageName( $langcode, $uckey ),
856 $langcode
857 );
858
859 if ( $message !== false ) {
860 return $message;
861 }
862 $alreadyTried[ $langcode ] = true;
863 }
864 } else {
865 $uckey = null;
866 }
867
868 // Check the CDB cache
869 $message = $lang->getMessage( $lckey );
870 if ( $message !== null ) {
871 return $message;
872 }
873
874 // Try checking the database for all of the fallback languages
875 if ( $useDB ) {
876 $fallbackChain = Language::getFallbacksFor( $langcode );
877
878 foreach ( $fallbackChain as $code ) {
879 if ( isset( $alreadyTried[ $code ] ) ) {
880 continue;
881 }
882
883 $message = $this->getMsgFromNamespace(
884 $this->getMessagePageName( $code, $uckey ), $code );
885
886 if ( $message !== false ) {
887 return $message;
888 }
889 $alreadyTried[ $code ] = true;
890 }
891 }
892
893 return false;
894 }
895
896 /**
897 * Get the message page name for a given language
898 *
899 * @param string $langcode
900 * @param string $uckey Uppercase key for the message
901 * @return string The page name
902 */
903 private function getMessagePageName( $langcode, $uckey ) {
904 global $wgLanguageCode;
905 if ( $langcode === $wgLanguageCode ) {
906 // Messages created in the content language will not have the /lang extension
907 return $uckey;
908 } else {
909 return "$uckey/$langcode";
910 }
911 }
912
913 /**
914 * Get a message from the MediaWiki namespace, with caching. The key must
915 * first be converted to two-part lang/msg form if necessary.
916 *
917 * Unlike self::get(), this function doesn't resolve fallback chains, and
918 * some callers require this behavior. LanguageConverter::parseCachedTable()
919 * and self::get() are some examples in core.
920 *
921 * @param string $title Message cache key with initial uppercase letter.
922 * @param string $code Code denoting the language to try.
923 * @return string|bool The message, or false if it does not exist or on error
924 */
925 public function getMsgFromNamespace( $title, $code ) {
926 $this->load( $code );
927 if ( isset( $this->mCache[$code][$title] ) ) {
928 $entry = $this->mCache[$code][$title];
929 if ( substr( $entry, 0, 1 ) === ' ' ) {
930 // The message exists, so make sure a string
931 // is returned.
932 return (string)substr( $entry, 1 );
933 } elseif ( $entry === '!NONEXISTENT' ) {
934 return false;
935 } elseif ( $entry === '!TOO BIG' ) {
936 // Fall through and try invididual message cache below
937 }
938 } else {
939 // XXX: This is not cached in process cache, should it?
940 $message = false;
941 Hooks::run( 'MessagesPreLoad', [ $title, &$message ] );
942 if ( $message !== false ) {
943 return $message;
944 }
945
946 return false;
947 }
948
949 # Try the individual message cache
950 $titleKey = wfMemcKey( 'messages', 'individual', $title );
951 $entry = $this->wanCache->get( $titleKey );
952 if ( $entry ) {
953 if ( substr( $entry, 0, 1 ) === ' ' ) {
954 $this->mCache[$code][$title] = $entry;
955
956 // The message exists, so make sure a string
957 // is returned.
958 return (string)substr( $entry, 1 );
959 } elseif ( $entry === '!NONEXISTENT' ) {
960 $this->mCache[$code][$title] = '!NONEXISTENT';
961
962 return false;
963 } else {
964 # Corrupt/obsolete entry, delete it
965 $this->wanCache->delete( $titleKey );
966 }
967 }
968
969 # Try loading it from the database
970 $revision = Revision::newFromTitle( Title::makeTitle( NS_MEDIAWIKI, $title ) );
971 if ( $revision ) {
972 $content = $revision->getContent();
973 if ( !$content ) {
974 // A possibly temporary loading failure.
975 wfDebugLog(
976 'MessageCache',
977 __METHOD__ . ": failed to load message page text for {$title} ($code)"
978 );
979 $message = null; // no negative caching
980 } else {
981 // XXX: Is this the right way to turn a Content object into a message?
982 // NOTE: $content is typically either WikitextContent, JavaScriptContent or
983 // CssContent. MessageContent is *not* used for storing messages, it's
984 // only used for wrapping them when needed.
985 $message = $content->getWikitextForTransclusion();
986
987 if ( $message === false || $message === null ) {
988 wfDebugLog(
989 'MessageCache',
990 __METHOD__ . ": message content doesn't provide wikitext "
991 . "(content model: " . $content->getModel() . ")"
992 );
993
994 $message = false; // negative caching
995 } else {
996 $this->mCache[$code][$title] = ' ' . $message;
997 $this->wanCache->set( $titleKey, ' ' . $message, $this->mExpiry );
998 }
999 }
1000 } else {
1001 $message = false; // negative caching
1002 }
1003
1004 if ( $message === false ) { // negative caching
1005 $this->mCache[$code][$title] = '!NONEXISTENT';
1006 $this->wanCache->set( $titleKey, '!NONEXISTENT', $this->mExpiry );
1007 }
1008
1009 return $message;
1010 }
1011
1012 /**
1013 * @param string $message
1014 * @param bool $interface
1015 * @param string $language Language code
1016 * @param Title $title
1017 * @return string
1018 */
1019 function transform( $message, $interface = false, $language = null, $title = null ) {
1020 // Avoid creating parser if nothing to transform
1021 if ( strpos( $message, '{{' ) === false ) {
1022 return $message;
1023 }
1024
1025 if ( $this->mInParser ) {
1026 return $message;
1027 }
1028
1029 $parser = $this->getParser();
1030 if ( $parser ) {
1031 $popts = $this->getParserOptions();
1032 $popts->setInterfaceMessage( $interface );
1033 $popts->setTargetLanguage( $language );
1034
1035 $userlang = $popts->setUserLang( $language );
1036 $this->mInParser = true;
1037 $message = $parser->transformMsg( $message, $popts, $title );
1038 $this->mInParser = false;
1039 $popts->setUserLang( $userlang );
1040 }
1041
1042 return $message;
1043 }
1044
1045 /**
1046 * @return Parser
1047 */
1048 function getParser() {
1049 global $wgParser, $wgParserConf;
1050 if ( !$this->mParser && isset( $wgParser ) ) {
1051 # Do some initialisation so that we don't have to do it twice
1052 $wgParser->firstCallInit();
1053 # Clone it and store it
1054 $class = $wgParserConf['class'];
1055 if ( $class == 'ParserDiffTest' ) {
1056 # Uncloneable
1057 $this->mParser = new $class( $wgParserConf );
1058 } else {
1059 $this->mParser = clone $wgParser;
1060 }
1061 }
1062
1063 return $this->mParser;
1064 }
1065
1066 /**
1067 * @param string $text
1068 * @param Title $title
1069 * @param bool $linestart Whether or not this is at the start of a line
1070 * @param bool $interface Whether this is an interface message
1071 * @param Language|string $language Language code
1072 * @return ParserOutput|string
1073 */
1074 public function parse( $text, $title = null, $linestart = true,
1075 $interface = false, $language = null
1076 ) {
1077 if ( $this->mInParser ) {
1078 return htmlspecialchars( $text );
1079 }
1080
1081 $parser = $this->getParser();
1082 $popts = $this->getParserOptions();
1083 $popts->setInterfaceMessage( $interface );
1084
1085 if ( is_string( $language ) ) {
1086 $language = Language::factory( $language );
1087 }
1088 $popts->setTargetLanguage( $language );
1089
1090 if ( !$title || !$title instanceof Title ) {
1091 global $wgTitle;
1092 wfDebugLog( 'GlobalTitleFail', __METHOD__ . ' called by ' .
1093 wfGetAllCallers( 6 ) . ' with no title set.' );
1094 $title = $wgTitle;
1095 }
1096 // Sometimes $wgTitle isn't set either...
1097 if ( !$title ) {
1098 # It's not uncommon having a null $wgTitle in scripts. See r80898
1099 # Create a ghost title in such case
1100 $title = Title::makeTitle( NS_SPECIAL, 'Badtitle/title not set in ' . __METHOD__ );
1101 }
1102
1103 $this->mInParser = true;
1104 $res = $parser->parse( $text, $title, $popts, $linestart );
1105 $this->mInParser = false;
1106
1107 return $res;
1108 }
1109
1110 function disable() {
1111 $this->mDisable = true;
1112 }
1113
1114 function enable() {
1115 $this->mDisable = false;
1116 }
1117
1118 /**
1119 * Whether DB/cache usage is disabled for determining messages
1120 *
1121 * If so, this typically indicates either:
1122 * - a) load() failed to find a cached copy nor query the DB
1123 * - b) we are in a special context or error mode that cannot use the DB
1124 * If the DB is ignored, any derived HTML output or cached objects may be wrong.
1125 * To avoid long-term cache pollution, TTLs can be adjusted accordingly.
1126 *
1127 * @return bool
1128 * @since 1.27
1129 */
1130 public function isDisabled() {
1131 return $this->mDisable;
1132 }
1133
1134 /**
1135 * Clear all stored messages. Mainly used after a mass rebuild.
1136 */
1137 function clear() {
1138 $langs = Language::fetchLanguageNames( null, 'mw' );
1139 foreach ( array_keys( $langs ) as $code ) {
1140 # Global and local caches
1141 $this->wanCache->touchCheckKey( wfMemcKey( 'messages', $code ) );
1142 }
1143
1144 $this->mLoadedLanguages = [];
1145 }
1146
1147 /**
1148 * @param string $key
1149 * @return array
1150 */
1151 public function figureMessage( $key ) {
1152 global $wgLanguageCode;
1153
1154 $pieces = explode( '/', $key );
1155 if ( count( $pieces ) < 2 ) {
1156 return [ $key, $wgLanguageCode ];
1157 }
1158
1159 $lang = array_pop( $pieces );
1160 if ( !Language::fetchLanguageName( $lang, null, 'mw' ) ) {
1161 return [ $key, $wgLanguageCode ];
1162 }
1163
1164 $message = implode( '/', $pieces );
1165
1166 return [ $message, $lang ];
1167 }
1168
1169 /**
1170 * Get all message keys stored in the message cache for a given language.
1171 * If $code is the content language code, this will return all message keys
1172 * for which MediaWiki:msgkey exists. If $code is another language code, this
1173 * will ONLY return message keys for which MediaWiki:msgkey/$code exists.
1174 * @param string $code Language code
1175 * @return array Array of message keys (strings)
1176 */
1177 public function getAllMessageKeys( $code ) {
1178 global $wgContLang;
1179 $this->load( $code );
1180 if ( !isset( $this->mCache[$code] ) ) {
1181 // Apparently load() failed
1182 return null;
1183 }
1184 // Remove administrative keys
1185 $cache = $this->mCache[$code];
1186 unset( $cache['VERSION'] );
1187 unset( $cache['EXPIRY'] );
1188 // Remove any !NONEXISTENT keys
1189 $cache = array_diff( $cache, [ '!NONEXISTENT' ] );
1190
1191 // Keys may appear with a capital first letter. lcfirst them.
1192 return array_map( [ $wgContLang, 'lcfirst' ], array_keys( $cache ) );
1193 }
1194 }