Non-word characters don't terminate tag names.
[mediawiki.git] / includes / HistoryBlob.php
blob31aa0f875a5cbee2f0b842642dc0fede56c134ad
1 <?php
2 /**
3 * Efficient concatenated text storage.
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.
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.
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
20 * @file
23 /**
24 * Base class for general text storage via the "object" flag in old_flags, or
25 * two-part external storage URLs. Used for represent efficient concatenated
26 * storage, and migration-related pointer objects.
28 interface HistoryBlob
30 /**
31 * Adds an item of text, returns a stub object which points to the item.
32 * You must call setLocation() on the stub object before storing it to the
33 * database
35 * @param $text string
37 * @return String: the key for getItem()
39 function addItem( $text );
41 /**
42 * Get item by key, or false if the key is not present
44 * @param $key string
46 * @return String or false
48 function getItem( $key );
50 /**
51 * Set the "default text"
52 * This concept is an odd property of the current DB schema, whereby each text item has a revision
53 * associated with it. The default text is the text of the associated revision. There may, however,
54 * be other revisions in the same object.
56 * Default text is not required for two-part external storage URLs.
58 * @param $text string
60 function setText( $text );
62 /**
63 * Get default text. This is called from Revision::getRevisionText()
65 * @return String
67 function getText();
70 /**
71 * Concatenated gzip (CGZ) storage
72 * Improves compression ratio by concatenating like objects before gzipping
74 class ConcatenatedGzipHistoryBlob implements HistoryBlob
76 public $mVersion = 0, $mCompressed = false, $mItems = array(), $mDefaultHash = '';
77 public $mSize = 0;
78 public $mMaxSize = 10000000;
79 public $mMaxCount = 100;
81 /** Constructor */
82 public function __construct() {
83 if ( !function_exists( 'gzdeflate' ) ) {
84 throw new MWException( "Need zlib support to read or write this kind of history object (ConcatenatedGzipHistoryBlob)\n" );
88 /**
89 * @param $text string
90 * @return string
92 public function addItem( $text ) {
93 $this->uncompress();
94 $hash = md5( $text );
95 if ( !isset( $this->mItems[$hash] ) ) {
96 $this->mItems[$hash] = $text;
97 $this->mSize += strlen( $text );
99 return $hash;
103 * @param $hash string
104 * @return array|bool
106 public function getItem( $hash ) {
107 $this->uncompress();
108 if ( array_key_exists( $hash, $this->mItems ) ) {
109 return $this->mItems[$hash];
110 } else {
111 return false;
116 * @param $text string
117 * @return void
119 public function setText( $text ) {
120 $this->uncompress();
121 $this->mDefaultHash = $this->addItem( $text );
125 * @return array|bool
127 public function getText() {
128 $this->uncompress();
129 return $this->getItem( $this->mDefaultHash );
133 * Remove an item
135 * @param $hash string
137 public function removeItem( $hash ) {
138 $this->mSize -= strlen( $this->mItems[$hash] );
139 unset( $this->mItems[$hash] );
143 * Compress the bulk data in the object
145 public function compress() {
146 if ( !$this->mCompressed ) {
147 $this->mItems = gzdeflate( serialize( $this->mItems ) );
148 $this->mCompressed = true;
153 * Uncompress bulk data
155 public function uncompress() {
156 if ( $this->mCompressed ) {
157 $this->mItems = unserialize( gzinflate( $this->mItems ) );
158 $this->mCompressed = false;
163 * @return array
165 function __sleep() {
166 $this->compress();
167 return array( 'mVersion', 'mCompressed', 'mItems', 'mDefaultHash' );
170 function __wakeup() {
171 $this->uncompress();
175 * Helper function for compression jobs
176 * Returns true until the object is "full" and ready to be committed
178 * @return bool
180 public function isHappy() {
181 return $this->mSize < $this->mMaxSize
182 && count( $this->mItems ) < $this->mMaxCount;
187 * Pointer object for an item within a CGZ blob stored in the text table.
189 class HistoryBlobStub {
191 * One-step cache variable to hold base blobs; operations that
192 * pull multiple revisions may often pull multiple times from
193 * the same blob. By keeping the last-used one open, we avoid
194 * redundant unserialization and decompression overhead.
196 protected static $blobCache = array();
198 var $mOldId, $mHash, $mRef;
201 * @param string $hash the content hash of the text
202 * @param $oldid Integer the old_id for the CGZ object
204 function __construct( $hash = '', $oldid = 0 ) {
205 $this->mHash = $hash;
209 * Sets the location (old_id) of the main object to which this object
210 * points
212 function setLocation( $id ) {
213 $this->mOldId = $id;
217 * Sets the location (old_id) of the referring object
219 function setReferrer( $id ) {
220 $this->mRef = $id;
224 * Gets the location of the referring object
226 function getReferrer() {
227 return $this->mRef;
231 * @return string
233 function getText() {
234 if ( isset( self::$blobCache[$this->mOldId] ) ) {
235 $obj = self::$blobCache[$this->mOldId];
236 } else {
237 $dbr = wfGetDB( DB_SLAVE );
238 $row = $dbr->selectRow( 'text', array( 'old_flags', 'old_text' ), array( 'old_id' => $this->mOldId ) );
239 if ( !$row ) {
240 return false;
242 $flags = explode( ',', $row->old_flags );
243 if ( in_array( 'external', $flags ) ) {
244 $url = $row->old_text;
245 $parts = explode( '://', $url, 2 );
246 if ( !isset( $parts[1] ) || $parts[1] == '' ) {
247 return false;
249 $row->old_text = ExternalStore::fetchFromUrl( $url );
252 if ( !in_array( 'object', $flags ) ) {
253 return false;
256 if ( in_array( 'gzip', $flags ) ) {
257 // This shouldn't happen, but a bug in the compress script
258 // may at times gzip-compress a HistoryBlob object row.
259 $obj = unserialize( gzinflate( $row->old_text ) );
260 } else {
261 $obj = unserialize( $row->old_text );
264 if ( !is_object( $obj ) ) {
265 // Correct for old double-serialization bug.
266 $obj = unserialize( $obj );
269 // Save this item for reference; if pulling many
270 // items in a row we'll likely use it again.
271 $obj->uncompress();
272 self::$blobCache = array( $this->mOldId => $obj );
274 return $obj->getItem( $this->mHash );
278 * Get the content hash
280 * @return string
282 function getHash() {
283 return $this->mHash;
288 * To speed up conversion from 1.4 to 1.5 schema, text rows can refer to the
289 * leftover cur table as the backend. This avoids expensively copying hundreds
290 * of megabytes of data during the conversion downtime.
292 * Serialized HistoryBlobCurStub objects will be inserted into the text table
293 * on conversion if $wgFastSchemaUpgrades is set to true.
295 class HistoryBlobCurStub {
296 var $mCurId;
299 * @param $curid Integer: the cur_id pointed to
301 function __construct( $curid = 0 ) {
302 $this->mCurId = $curid;
306 * Sets the location (cur_id) of the main object to which this object
307 * points
309 * @param $id int
311 function setLocation( $id ) {
312 $this->mCurId = $id;
316 * @return string|bool
318 function getText() {
319 $dbr = wfGetDB( DB_SLAVE );
320 $row = $dbr->selectRow( 'cur', array( 'cur_text' ), array( 'cur_id' => $this->mCurId ) );
321 if ( !$row ) {
322 return false;
324 return $row->cur_text;
329 * Diff-based history compression
330 * Requires xdiff 1.5+ and zlib
332 class DiffHistoryBlob implements HistoryBlob {
333 /** Uncompressed item cache */
334 var $mItems = array();
336 /** Total uncompressed size */
337 var $mSize = 0;
340 * Array of diffs. If a diff D from A to B is notated D = B - A, and Z is
341 * an empty string:
343 * { item[map[i]] - item[map[i-1]] where i > 0
344 * diff[i] = {
345 * { item[map[i]] - Z where i = 0
347 var $mDiffs;
349 /** The diff map, see above */
350 var $mDiffMap;
353 * The key for getText()
355 var $mDefaultKey;
358 * Compressed storage
360 var $mCompressed;
363 * True if the object is locked against further writes
365 var $mFrozen = false;
368 * The maximum uncompressed size before the object becomes sad
369 * Should be less than max_allowed_packet
371 var $mMaxSize = 10000000;
374 * The maximum number of text items before the object becomes sad
376 var $mMaxCount = 100;
378 /** Constants from xdiff.h */
379 const XDL_BDOP_INS = 1;
380 const XDL_BDOP_CPY = 2;
381 const XDL_BDOP_INSB = 3;
383 function __construct() {
384 if ( !function_exists( 'gzdeflate' ) ) {
385 throw new MWException( "Need zlib support to read or write DiffHistoryBlob\n" );
390 * @throws MWException
391 * @param $text string
392 * @return int
394 function addItem( $text ) {
395 if ( $this->mFrozen ) {
396 throw new MWException( __METHOD__ . ": Cannot add more items after sleep/wakeup" );
399 $this->mItems[] = $text;
400 $this->mSize += strlen( $text );
401 $this->mDiffs = null; // later
402 return count( $this->mItems ) - 1;
406 * @param $key string
407 * @return string
409 function getItem( $key ) {
410 return $this->mItems[$key];
414 * @param $text string
416 function setText( $text ) {
417 $this->mDefaultKey = $this->addItem( $text );
421 * @return string
423 function getText() {
424 return $this->getItem( $this->mDefaultKey );
428 * @throws MWException
430 function compress() {
431 if ( !function_exists( 'xdiff_string_rabdiff' ) ) {
432 throw new MWException( "Need xdiff 1.5+ support to write DiffHistoryBlob\n" );
434 if ( isset( $this->mDiffs ) ) {
435 // Already compressed
436 return;
438 if ( !count( $this->mItems ) ) {
439 // Empty
440 return;
443 // Create two diff sequences: one for main text and one for small text
444 $sequences = array(
445 'small' => array(
446 'tail' => '',
447 'diffs' => array(),
448 'map' => array(),
450 'main' => array(
451 'tail' => '',
452 'diffs' => array(),
453 'map' => array(),
456 $smallFactor = 0.5;
458 for ( $i = 0; $i < count( $this->mItems ); $i++ ) {
459 $text = $this->mItems[$i];
460 if ( $i == 0 ) {
461 $seqName = 'main';
462 } else {
463 $mainTail = $sequences['main']['tail'];
464 if ( strlen( $text ) < strlen( $mainTail ) * $smallFactor ) {
465 $seqName = 'small';
466 } else {
467 $seqName = 'main';
470 $seq =& $sequences[$seqName];
471 $tail = $seq['tail'];
472 $diff = $this->diff( $tail, $text );
473 $seq['diffs'][] = $diff;
474 $seq['map'][] = $i;
475 $seq['tail'] = $text;
477 unset( $seq ); // unlink dangerous alias
479 // Knit the sequences together
480 $tail = '';
481 $this->mDiffs = array();
482 $this->mDiffMap = array();
483 foreach ( $sequences as $seq ) {
484 if ( !count( $seq['diffs'] ) ) {
485 continue;
487 if ( $tail === '' ) {
488 $this->mDiffs[] = $seq['diffs'][0];
489 } else {
490 $head = $this->patch( '', $seq['diffs'][0] );
491 $this->mDiffs[] = $this->diff( $tail, $head );
493 $this->mDiffMap[] = $seq['map'][0];
494 for ( $i = 1; $i < count( $seq['diffs'] ); $i++ ) {
495 $this->mDiffs[] = $seq['diffs'][$i];
496 $this->mDiffMap[] = $seq['map'][$i];
498 $tail = $seq['tail'];
503 * @param $t1
504 * @param $t2
505 * @return string
507 function diff( $t1, $t2 ) {
508 # Need to do a null concatenation with warnings off, due to bugs in the current version of xdiff
509 # "String is not zero-terminated"
510 wfSuppressWarnings();
511 $diff = xdiff_string_rabdiff( $t1, $t2 ) . '';
512 wfRestoreWarnings();
513 return $diff;
517 * @param $base
518 * @param $diff
519 * @return bool|string
521 function patch( $base, $diff ) {
522 if ( function_exists( 'xdiff_string_bpatch' ) ) {
523 wfSuppressWarnings();
524 $text = xdiff_string_bpatch( $base, $diff ) . '';
525 wfRestoreWarnings();
526 return $text;
529 # Pure PHP implementation
531 $header = unpack( 'Vofp/Vcsize', substr( $diff, 0, 8 ) );
533 # Check the checksum if hash extension is available
534 $ofp = $this->xdiffAdler32( $base );
535 if ( $ofp !== false && $ofp !== substr( $diff, 0, 4 ) ) {
536 wfDebug( __METHOD__ . ": incorrect base checksum\n" );
537 return false;
539 if ( $header['csize'] != strlen( $base ) ) {
540 wfDebug( __METHOD__ . ": incorrect base length\n" );
541 return false;
544 $p = 8;
545 $out = '';
546 while ( $p < strlen( $diff ) ) {
547 $x = unpack( 'Cop', substr( $diff, $p, 1 ) );
548 $op = $x['op'];
549 ++$p;
550 switch ( $op ) {
551 case self::XDL_BDOP_INS:
552 $x = unpack( 'Csize', substr( $diff, $p, 1 ) );
553 $p++;
554 $out .= substr( $diff, $p, $x['size'] );
555 $p += $x['size'];
556 break;
557 case self::XDL_BDOP_INSB:
558 $x = unpack( 'Vcsize', substr( $diff, $p, 4 ) );
559 $p += 4;
560 $out .= substr( $diff, $p, $x['csize'] );
561 $p += $x['csize'];
562 break;
563 case self::XDL_BDOP_CPY:
564 $x = unpack( 'Voff/Vcsize', substr( $diff, $p, 8 ) );
565 $p += 8;
566 $out .= substr( $base, $x['off'], $x['csize'] );
567 break;
568 default:
569 wfDebug( __METHOD__ . ": invalid op\n" );
570 return false;
573 return $out;
577 * Compute a binary "Adler-32" checksum as defined by LibXDiff, i.e. with
578 * the bytes backwards and initialised with 0 instead of 1. See bug 34428.
580 * @param string $s
581 * @return string|bool: false if the hash extension is not available
583 function xdiffAdler32( $s ) {
584 if ( !function_exists( 'hash' ) ) {
585 return false;
588 static $init;
589 if ( $init === null ) {
590 $init = str_repeat( "\xf0", 205 ) . "\xee" . str_repeat( "\xf0", 67 ) . "\x02";
593 // The real Adler-32 checksum of $init is zero, so it initialises the
594 // state to zero, as it is at the start of LibXDiff's checksum
595 // algorithm. Appending the subject string then simulates LibXDiff.
596 return strrev( hash( 'adler32', $init . $s, true ) );
599 function uncompress() {
600 if ( !$this->mDiffs ) {
601 return;
603 $tail = '';
604 for ( $diffKey = 0; $diffKey < count( $this->mDiffs ); $diffKey++ ) {
605 $textKey = $this->mDiffMap[$diffKey];
606 $text = $this->patch( $tail, $this->mDiffs[$diffKey] );
607 $this->mItems[$textKey] = $text;
608 $tail = $text;
613 * @return array
615 function __sleep() {
616 $this->compress();
617 if ( !count( $this->mItems ) ) {
618 // Empty object
619 $info = false;
620 } else {
621 // Take forward differences to improve the compression ratio for sequences
622 $map = '';
623 $prev = 0;
624 foreach ( $this->mDiffMap as $i ) {
625 if ( $map !== '' ) {
626 $map .= ',';
628 $map .= $i - $prev;
629 $prev = $i;
631 $info = array(
632 'diffs' => $this->mDiffs,
633 'map' => $map
636 if ( isset( $this->mDefaultKey ) ) {
637 $info['default'] = $this->mDefaultKey;
639 $this->mCompressed = gzdeflate( serialize( $info ) );
640 return array( 'mCompressed' );
643 function __wakeup() {
644 // addItem() doesn't work if mItems is partially filled from mDiffs
645 $this->mFrozen = true;
646 $info = unserialize( gzinflate( $this->mCompressed ) );
647 unset( $this->mCompressed );
649 if ( !$info ) {
650 // Empty object
651 return;
654 if ( isset( $info['default'] ) ) {
655 $this->mDefaultKey = $info['default'];
657 $this->mDiffs = $info['diffs'];
658 if ( isset( $info['base'] ) ) {
659 // Old format
660 $this->mDiffMap = range( 0, count( $this->mDiffs ) - 1 );
661 array_unshift( $this->mDiffs,
662 pack( 'VVCV', 0, 0, self::XDL_BDOP_INSB, strlen( $info['base'] ) ) .
663 $info['base'] );
664 } else {
665 // New format
666 $map = explode( ',', $info['map'] );
667 $cur = 0;
668 $this->mDiffMap = array();
669 foreach ( $map as $i ) {
670 $cur += $i;
671 $this->mDiffMap[] = $cur;
674 $this->uncompress();
678 * Helper function for compression jobs
679 * Returns true until the object is "full" and ready to be committed
681 * @return bool
683 function isHappy() {
684 return $this->mSize < $this->mMaxSize
685 && count( $this->mItems ) < $this->mMaxCount;