wiki.techinc.nl/includes/resourceloader/ResourceLoader.php

1649 lines
52 KiB
PHP
Raw Normal View History

<?php
/**
* Base class for resource loading system.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
* http://www.gnu.org/copyleft/gpl.html
*
2010-09-05 13:31:34 +00:00
* @file
* @author Roan Kattouw
* @author Trevor Parscal
*/
use Psr\Log\LoggerAwareInterface;
use Psr\Log\LoggerInterface;
use Psr\Log\NullLogger;
use WrappedString\WrappedString;
2010-09-05 13:31:34 +00:00
/**
* Dynamic JavaScript and CSS resource loading system.
*
* Most of the documentation is on the MediaWiki documentation wiki starting at:
* https://www.mediawiki.org/wiki/ResourceLoader
*/
class ResourceLoader implements LoggerAwareInterface {
/** @var int */
protected static $filterCacheVersion = 7;
resourceloader: Implement "skip function" feature A module can be registered with a skip function. Such function, if provided, will be invoked by the client when a module is queued for loading. If the function returns true, the client will bypass any further loading action and mark the module as 'ready'. This can be used to implement a feature test for a module providing a shim or polyfill. * Change visibility of method ResourceLoader::filter to public. So that it can be invoked by ResourceLoaderStartupModule. * Add option to suppress the cache key report in ResourceLoader::filter. We usually only call the minifier once on an entire request reponse (because it's all concatenated javascript or embedded javascript in various different closures, still valid as one large script) and only add a little bottom line for the cache key. When embedding the skip function we have to run the minifier on them separately as they're output as strings (not actual functions). These strings are typically quite small and blowing up the response with loads of cache keys is not desirable in production. * Add method to clear the static cache of ResourceLoader::inDebugMode. Global static state is evil but, as long as we have it, we at least need to clear it after switching contexts in the test suite. Also: * Remove obsolete setting of 'debug=true' in the FauxRequest in ResourceLoaderTestCase. It already sets global wgResourceLoaderDebug in the setUp() method. Bug: 66390 Change-Id: I87a0ea888d791ad39f114380c42e2daeca470961
2014-04-30 21:06:51 +00:00
/** @var bool */
protected static $debugMode = null;
/** @var array */
private static $lessVars = null;
/**
* Module name/ResourceLoaderModule object pairs
* @var array
*/
protected $modules = array();
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
/**
* Associative array mapping module name to info associative array
* @var array
*/
* Made Resources.php return a pure-data array instead of an ugly mix of data and code. This allows the class code to be lazy-loaded with the autoloader, for a performance advantage especially on non-APC installs. And using the convention where if the class is omitted, ResourceLoaderFileModule is assumed, the registration code becomes shorter and simpler. * Modified ResourceLoader to lazy-initialise module objects, for a further performance advantage. * Deleted ResourceLoader::getModules(), provided getModuleNames() instead. Although the startup module needs this functionality, it's slow to generate, so to avoid misuse, it's better to provide a foolproof fast interface and let the startup module do the slow thing itself. * Modified ResourceLoader::register() to optionally accept an info array instead of an object. * Added $wgResourceModules, allowing extensions to efficiently define their own resource loader modules. The trouble with hooks is that they contain code, and code is slow. We've been through all this before with i18n. Hooks are useful as a performance tool only if you call them very rarely. * Moved ResourceLoader settings to their own section in DefaultSettings.php * Added options to ResourceLoaderFileModule equivalent to the $localBasePath and $remoteBasePath parameters, to allow it to be instantiated via the new array style. Also added remoteExtPath, which allows modules to be registered before $wgExtensionAssetsPath is known. * Added OutputPage::getResourceLoader(), mostly for debugging. * The time saving at the moment is about 5ms per request with no extensions, which is significant already with 6 load.php requests for a cold cache page view. This is a much more scalable interface; the relative saving will grow as more extensions are added which use this interface, especially for non-APC installs. Although the interface is backwards compatible, extension updates will follow in a subsequent commit.
2010-11-19 10:41:06 +00:00
protected $moduleInfos = array();
/** @var Config $config */
private $config;
/**
* Associative array mapping framework ids to a list of names of test suite modules
* like array( 'qunit' => array( 'mediawiki.tests.qunit.suites', 'ext.foo.tests', .. ), .. )
* @var array
*/
protected $testModuleNames = array();
2010-09-04 12:53:01 +00:00
/**
* E.g. array( 'source-id' => 'http://.../load.php' )
* @var array
*/
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
protected $sources = array();
/**
* Errors accumulated during current respond() call.
* @var array
*/
protected $errors = array();
/**
* @var MessageBlobStore
*/
protected $blobStore;
/**
* @var LoggerInterface
*/
private $logger;
/**
* Load information stored in the database about modules.
*
* This method grabs modules dependencies from the database and updates modules
* objects.
*
* This is not inside the module code because it is much faster to
* request all of the information at once than it is to have each module
* requests its own information. This sacrifice of modularity yields a substantial
* performance improvement.
*
* @param array $modules List of module names to preload information for
* @param ResourceLoaderContext $context Context to load the information within
*/
public function preloadModuleInfo( array $modules, ResourceLoaderContext $context ) {
if ( !count( $modules ) ) {
// Or else Database*::select() will explode, plus it's cheaper!
return;
}
$dbr = wfGetDB( DB_SLAVE );
$skin = $context->getSkin();
$lang = $context->getLanguage();
// Get file dependency information
$res = $dbr->select( 'module_deps', array( 'md_module', 'md_deps' ), array(
'md_module' => $modules,
'md_skin' => $skin
), __METHOD__
);
// Set modules' dependencies
$modulesWithDeps = array();
foreach ( $res as $row ) {
$module = $this->getModule( $row->md_module );
if ( $module ) {
$module->setFileDependencies( $skin, FormatJson::decode( $row->md_deps, true ) );
$modulesWithDeps[] = $row->md_module;
}
}
// Register the absence of a dependency row too
foreach ( array_diff( $modules, $modulesWithDeps ) as $name ) {
$module = $this->getModule( $name );
if ( $module ) {
$this->getModule( $name )->setFileDependencies( $skin, array() );
}
}
// Get message blob mtimes. Only do this for modules with messages
$modulesWithMessages = array();
foreach ( $modules as $name ) {
$module = $this->getModule( $name );
if ( $module && count( $module->getMessages() ) ) {
$modulesWithMessages[] = $name;
}
}
$modulesWithoutMessages = array_flip( $modules ); // Will be trimmed down by the loop below
if ( count( $modulesWithMessages ) ) {
$res = $dbr->select( 'msg_resource', array( 'mr_resource', 'mr_timestamp' ), array(
'mr_resource' => $modulesWithMessages,
'mr_lang' => $lang
), __METHOD__
);
foreach ( $res as $row ) {
$module = $this->getModule( $row->mr_resource );
if ( $module ) {
$module->setMsgBlobMtime( $lang, wfTimestamp( TS_UNIX, $row->mr_timestamp ) );
unset( $modulesWithoutMessages[$row->mr_resource] );
}
}
}
foreach ( array_keys( $modulesWithoutMessages ) as $name ) {
$module = $this->getModule( $name );
if ( $module ) {
$module->setMsgBlobMtime( $lang, 1 );
}
}
}
/**
* Run JavaScript or CSS data through a filter, caching the filtered result for future calls.
*
* Available filters are:
*
* - minify-js \see JavaScriptMinifier::minify
* - minify-css \see CSSMin::minify
*
* If $data is empty, only contains whitespace or the filter was unknown,
* $data is returned unmodified.
*
* @param string $filter Name of filter to run
* @param string $data Text to filter, such as JavaScript or CSS text
* @param array $options For back-compat, can also be the boolean value for "cacheReport". Keys:
* - (bool) cache: Whether to allow caching this data. Default: true.
* - (bool) cacheReport: Whether to include the "cache key" report comment. Default: false.
* @return string Filtered data, or a comment containing an error message
*/
public function filter( $filter, $data, $options = array() ) {
// Back-compat
if ( is_bool( $options ) ) {
$options = array( 'cacheReport' => $options );
}
// Defaults
$options += array( 'cache' => true, 'cacheReport' => false );
$stats = RequestContext::getMain()->getStats();
2010-09-04 12:53:01 +00:00
// Don't filter empty content
if ( trim( $data ) === '' ) {
return $data;
}
2010-09-04 12:53:01 +00:00
if ( !in_array( $filter, array( 'minify-js', 'minify-css' ) ) ) {
$this->logger->warning( 'Invalid filter {filter}', array(
'filter' => $filter
) );
return $data;
}
2010-09-04 12:53:01 +00:00
if ( !$options['cache'] ) {
$result = self::applyFilter( $filter, $data, $this->config );
} else {
$key = wfGlobalCacheKey( 'resourceloader', 'filter', $filter, self::$filterCacheVersion, md5( $data ) );
$cache = wfGetCache( wfIsHHVM() ? CACHE_ACCEL : CACHE_ANYTHING );
$cacheEntry = $cache->get( $key );
if ( is_string( $cacheEntry ) ) {
$stats->increment( "resourceloader_cache.$filter.hit" );
return $cacheEntry;
}
$result = '';
try {
$statStart = microtime( true );
$result = self::applyFilter( $filter, $data, $this->config );
$statTiming = microtime( true ) - $statStart;
$stats->increment( "resourceloader_cache.$filter.miss" );
$stats->timing( "resourceloader_cache.$filter.timing", 1000 * $statTiming );
if ( $options['cacheReport'] ) {
$result .= "\n/* cache key: $key */";
}
// Set a TTL since HHVM's APC doesn't have any limitation or eviction logic.
$cache->set( $key, $result, 24 * 3600 );
} catch ( Exception $e ) {
MWExceptionHandler::logException( $e );
$this->logger->warning( 'Minification failed: {exception}', array(
'exception' => $e
) );
$this->errors[] = self::formatExceptionNoComment( $e );
}
}
return $result;
}
private static function applyFilter( $filter, $data, Config $config ) {
switch ( $filter ) {
case 'minify-js':
return JavaScriptMinifier::minify( $data,
$config->get( 'ResourceLoaderMinifierStatementsOnOwnLine' ),
$config->get( 'ResourceLoaderMinifierMaxLineLength' )
);
case 'minify-css':
return CSSMin::minify( $data );
}
return $data;
}
2010-09-04 12:53:01 +00:00
/* Methods */
/**
* Register core modules and runs registration hooks.
* @param Config|null $config
*/
public function __construct( Config $config = null, LoggerInterface $logger = null ) {
global $IP;
if ( !$logger ) {
$logger = new NullLogger();
}
$this->setLogger( $logger );
if ( !$config ) {
$this->logger->debug( __METHOD__ . ' was called without providing a Config instance' );
$config = ConfigFactory::getDefaultInstance()->makeConfig( 'main' );
}
$this->config = $config;
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
// Add 'local' source first
$this->addSource( 'local', wfScript( 'load' ) );
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
// Add other sources
$this->addSource( $config->get( 'ResourceLoaderSources' ) );
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
// Register core modules
$this->register( include "$IP/resources/Resources.php" );
$this->register( include "$IP/resources/ResourcesOOUI.php" );
// Register extension modules
Hooks::run( 'ResourceLoaderRegisterModules', array( &$this ) );
$this->register( $config->get( 'ResourceModules' ) );
if ( $config->get( 'EnableJavaScriptTest' ) === true ) {
$this->registerTestModules();
}
$this->setMessageBlobStore( new MessageBlobStore() );
}
2010-09-04 12:53:01 +00:00
/**
* @return Config
*/
public function getConfig() {
return $this->config;
}
public function setLogger( LoggerInterface $logger ) {
$this->logger = $logger;
}
/**
* @since 1.26
* @return MessageBlobStore
*/
public function getMessageBlobStore() {
return $this->blobStore;
}
/**
* @since 1.25
* @param MessageBlobStore $blobStore
*/
public function setMessageBlobStore( MessageBlobStore $blobStore ) {
$this->blobStore = $blobStore;
}
/**
* Register a module with the ResourceLoader system.
*
* @param mixed $name Name of module as a string or List of name/object pairs as an array
* @param array $info Module info array. For backwards compatibility with 1.17alpha,
* this may also be a ResourceLoaderModule object. Optional when using
* Made Resources.php return a pure-data array instead of an ugly mix of data and code. This allows the class code to be lazy-loaded with the autoloader, for a performance advantage especially on non-APC installs. And using the convention where if the class is omitted, ResourceLoaderFileModule is assumed, the registration code becomes shorter and simpler. * Modified ResourceLoader to lazy-initialise module objects, for a further performance advantage. * Deleted ResourceLoader::getModules(), provided getModuleNames() instead. Although the startup module needs this functionality, it's slow to generate, so to avoid misuse, it's better to provide a foolproof fast interface and let the startup module do the slow thing itself. * Modified ResourceLoader::register() to optionally accept an info array instead of an object. * Added $wgResourceModules, allowing extensions to efficiently define their own resource loader modules. The trouble with hooks is that they contain code, and code is slow. We've been through all this before with i18n. Hooks are useful as a performance tool only if you call them very rarely. * Moved ResourceLoader settings to their own section in DefaultSettings.php * Added options to ResourceLoaderFileModule equivalent to the $localBasePath and $remoteBasePath parameters, to allow it to be instantiated via the new array style. Also added remoteExtPath, which allows modules to be registered before $wgExtensionAssetsPath is known. * Added OutputPage::getResourceLoader(), mostly for debugging. * The time saving at the moment is about 5ms per request with no extensions, which is significant already with 6 load.php requests for a cold cache page view. This is a much more scalable interface; the relative saving will grow as more extensions are added which use this interface, especially for non-APC installs. Although the interface is backwards compatible, extension updates will follow in a subsequent commit.
2010-11-19 10:41:06 +00:00
* multiple-registration calling style.
* @throws MWException If a duplicate module registration is attempted
* @throws MWException If a module name contains illegal characters (pipes or commas)
* @throws MWException If something other than a ResourceLoaderModule is being registered
* @return bool False if there were any errors, in which case one or more modules were
* not registered
*/
* Made Resources.php return a pure-data array instead of an ugly mix of data and code. This allows the class code to be lazy-loaded with the autoloader, for a performance advantage especially on non-APC installs. And using the convention where if the class is omitted, ResourceLoaderFileModule is assumed, the registration code becomes shorter and simpler. * Modified ResourceLoader to lazy-initialise module objects, for a further performance advantage. * Deleted ResourceLoader::getModules(), provided getModuleNames() instead. Although the startup module needs this functionality, it's slow to generate, so to avoid misuse, it's better to provide a foolproof fast interface and let the startup module do the slow thing itself. * Modified ResourceLoader::register() to optionally accept an info array instead of an object. * Added $wgResourceModules, allowing extensions to efficiently define their own resource loader modules. The trouble with hooks is that they contain code, and code is slow. We've been through all this before with i18n. Hooks are useful as a performance tool only if you call them very rarely. * Moved ResourceLoader settings to their own section in DefaultSettings.php * Added options to ResourceLoaderFileModule equivalent to the $localBasePath and $remoteBasePath parameters, to allow it to be instantiated via the new array style. Also added remoteExtPath, which allows modules to be registered before $wgExtensionAssetsPath is known. * Added OutputPage::getResourceLoader(), mostly for debugging. * The time saving at the moment is about 5ms per request with no extensions, which is significant already with 6 load.php requests for a cold cache page view. This is a much more scalable interface; the relative saving will grow as more extensions are added which use this interface, especially for non-APC installs. Although the interface is backwards compatible, extension updates will follow in a subsequent commit.
2010-11-19 10:41:06 +00:00
public function register( $name, $info = null ) {
// Allow multiple modules to be registered in one call
$registrations = is_array( $name ) ? $name : array( $name => $info );
foreach ( $registrations as $name => $info ) {
// Disallow duplicate registrations
if ( isset( $this->moduleInfos[$name] ) ) {
// A module has already been registered by this name
throw new MWException(
'ResourceLoader duplicate registration error. ' .
'Another module has already been registered as ' . $name
);
}
// Check $name for validity
if ( !self::isValidModuleName( $name ) ) {
throw new MWException( "ResourceLoader module name '$name' is invalid, "
. "see ResourceLoader::isValidModuleName()" );
* Made Resources.php return a pure-data array instead of an ugly mix of data and code. This allows the class code to be lazy-loaded with the autoloader, for a performance advantage especially on non-APC installs. And using the convention where if the class is omitted, ResourceLoaderFileModule is assumed, the registration code becomes shorter and simpler. * Modified ResourceLoader to lazy-initialise module objects, for a further performance advantage. * Deleted ResourceLoader::getModules(), provided getModuleNames() instead. Although the startup module needs this functionality, it's slow to generate, so to avoid misuse, it's better to provide a foolproof fast interface and let the startup module do the slow thing itself. * Modified ResourceLoader::register() to optionally accept an info array instead of an object. * Added $wgResourceModules, allowing extensions to efficiently define their own resource loader modules. The trouble with hooks is that they contain code, and code is slow. We've been through all this before with i18n. Hooks are useful as a performance tool only if you call them very rarely. * Moved ResourceLoader settings to their own section in DefaultSettings.php * Added options to ResourceLoaderFileModule equivalent to the $localBasePath and $remoteBasePath parameters, to allow it to be instantiated via the new array style. Also added remoteExtPath, which allows modules to be registered before $wgExtensionAssetsPath is known. * Added OutputPage::getResourceLoader(), mostly for debugging. * The time saving at the moment is about 5ms per request with no extensions, which is significant already with 6 load.php requests for a cold cache page view. This is a much more scalable interface; the relative saving will grow as more extensions are added which use this interface, especially for non-APC installs. Although the interface is backwards compatible, extension updates will follow in a subsequent commit.
2010-11-19 10:41:06 +00:00
}
// Attach module
if ( $info instanceof ResourceLoaderModule ) {
$this->moduleInfos[$name] = array( 'object' => $info );
$info->setName( $name );
$this->modules[$name] = $info;
} elseif ( is_array( $info ) ) {
// New calling convention
$this->moduleInfos[$name] = $info;
} else {
throw new MWException(
'ResourceLoader module info type error for module \'' . $name .
'\': expected ResourceLoaderModule or array (got: ' . gettype( $info ) . ')'
);
}
resourceloader: Allow skins to provide additional styles for any module The newly introduced $wgResourceModuleSkinStyles global enables skins to provide additional stylesheets to existing ResourceLoader module. This both makes it easier (or at all possible) to override default styles and lowers the style footprint by making it possible not to load styles unused on most pages. ---- Example: Use the file 'foo-styles.css' for the 'mediawiki.foo' module when using the MySkin skin: $wgResourceModuleSkinStyles['myskin'] = array( 'mediawiki.foo' => 'foo-styles.css', 'remoteSkinPath' => 'MySkin', 'localBasePath' => __DIR__, ); For detailed documentation, see the doc comment in DefaultSettings.php. For a practical usage example, see Vector.php. ---- Implementation notes: * The values defined in $wgResourceModuleSkinStyles are embedded into the modules as late as possible (in ResourceLoader::register()). * Only plain file modules are supported, setting module skin styles for other module types has no effect. * ResourceLoader and ResourceLoaderFileModule now support loading files from arbitrary paths to make this possible, defined using ResourceLoaderFilePath objects. * This required some adjustments in seemingly unrelated places for code which didn't handle the paths fully correctly before. * ResourceLoader and ResourceLoaderFileModule are now a bit more tightly coupled than before :( * Included a tiny example change for the Vector skin, a lot more of similar cleanup is possible and planned for the future. * Many of the non-essential mediawiki.* modules defined in Resources.php should be using `'skinStyles' => array( 'default' => … )` instead of `'styles' => …` to allow more customizations, this is also planned for the future after auditing which ones would actually benefit from this. Change-Id: Ica4ff9696b490e35f60288d7ce1295766c427e87
2014-06-26 14:29:31 +00:00
// Last-minute changes
// Apply custom skin-defined styles to existing modules.
if ( $this->isFileModule( $name ) ) {
foreach ( $this->config->get( 'ResourceModuleSkinStyles' ) as $skinName => $skinStyles ) {
resourceloader: Allow skins to provide additional styles for any module The newly introduced $wgResourceModuleSkinStyles global enables skins to provide additional stylesheets to existing ResourceLoader module. This both makes it easier (or at all possible) to override default styles and lowers the style footprint by making it possible not to load styles unused on most pages. ---- Example: Use the file 'foo-styles.css' for the 'mediawiki.foo' module when using the MySkin skin: $wgResourceModuleSkinStyles['myskin'] = array( 'mediawiki.foo' => 'foo-styles.css', 'remoteSkinPath' => 'MySkin', 'localBasePath' => __DIR__, ); For detailed documentation, see the doc comment in DefaultSettings.php. For a practical usage example, see Vector.php. ---- Implementation notes: * The values defined in $wgResourceModuleSkinStyles are embedded into the modules as late as possible (in ResourceLoader::register()). * Only plain file modules are supported, setting module skin styles for other module types has no effect. * ResourceLoader and ResourceLoaderFileModule now support loading files from arbitrary paths to make this possible, defined using ResourceLoaderFilePath objects. * This required some adjustments in seemingly unrelated places for code which didn't handle the paths fully correctly before. * ResourceLoader and ResourceLoaderFileModule are now a bit more tightly coupled than before :( * Included a tiny example change for the Vector skin, a lot more of similar cleanup is possible and planned for the future. * Many of the non-essential mediawiki.* modules defined in Resources.php should be using `'skinStyles' => array( 'default' => … )` instead of `'styles' => …` to allow more customizations, this is also planned for the future after auditing which ones would actually benefit from this. Change-Id: Ica4ff9696b490e35f60288d7ce1295766c427e87
2014-06-26 14:29:31 +00:00
// If this module already defines skinStyles for this skin, ignore $wgResourceModuleSkinStyles.
if ( isset( $this->moduleInfos[$name]['skinStyles'][$skinName] ) ) {
continue;
}
// If $name is preceded with a '+', the defined style files will be added to 'default'
// skinStyles, otherwise 'default' will be ignored as it normally would be.
if ( isset( $skinStyles[$name] ) ) {
$paths = (array)$skinStyles[$name];
resourceloader: Allow skins to provide additional styles for any module The newly introduced $wgResourceModuleSkinStyles global enables skins to provide additional stylesheets to existing ResourceLoader module. This both makes it easier (or at all possible) to override default styles and lowers the style footprint by making it possible not to load styles unused on most pages. ---- Example: Use the file 'foo-styles.css' for the 'mediawiki.foo' module when using the MySkin skin: $wgResourceModuleSkinStyles['myskin'] = array( 'mediawiki.foo' => 'foo-styles.css', 'remoteSkinPath' => 'MySkin', 'localBasePath' => __DIR__, ); For detailed documentation, see the doc comment in DefaultSettings.php. For a practical usage example, see Vector.php. ---- Implementation notes: * The values defined in $wgResourceModuleSkinStyles are embedded into the modules as late as possible (in ResourceLoader::register()). * Only plain file modules are supported, setting module skin styles for other module types has no effect. * ResourceLoader and ResourceLoaderFileModule now support loading files from arbitrary paths to make this possible, defined using ResourceLoaderFilePath objects. * This required some adjustments in seemingly unrelated places for code which didn't handle the paths fully correctly before. * ResourceLoader and ResourceLoaderFileModule are now a bit more tightly coupled than before :( * Included a tiny example change for the Vector skin, a lot more of similar cleanup is possible and planned for the future. * Many of the non-essential mediawiki.* modules defined in Resources.php should be using `'skinStyles' => array( 'default' => … )` instead of `'styles' => …` to allow more customizations, this is also planned for the future after auditing which ones would actually benefit from this. Change-Id: Ica4ff9696b490e35f60288d7ce1295766c427e87
2014-06-26 14:29:31 +00:00
$styleFiles = array();
} elseif ( isset( $skinStyles['+' . $name] ) ) {
$paths = (array)$skinStyles['+' . $name];
resourceloader: Allow skins to provide additional styles for any module The newly introduced $wgResourceModuleSkinStyles global enables skins to provide additional stylesheets to existing ResourceLoader module. This both makes it easier (or at all possible) to override default styles and lowers the style footprint by making it possible not to load styles unused on most pages. ---- Example: Use the file 'foo-styles.css' for the 'mediawiki.foo' module when using the MySkin skin: $wgResourceModuleSkinStyles['myskin'] = array( 'mediawiki.foo' => 'foo-styles.css', 'remoteSkinPath' => 'MySkin', 'localBasePath' => __DIR__, ); For detailed documentation, see the doc comment in DefaultSettings.php. For a practical usage example, see Vector.php. ---- Implementation notes: * The values defined in $wgResourceModuleSkinStyles are embedded into the modules as late as possible (in ResourceLoader::register()). * Only plain file modules are supported, setting module skin styles for other module types has no effect. * ResourceLoader and ResourceLoaderFileModule now support loading files from arbitrary paths to make this possible, defined using ResourceLoaderFilePath objects. * This required some adjustments in seemingly unrelated places for code which didn't handle the paths fully correctly before. * ResourceLoader and ResourceLoaderFileModule are now a bit more tightly coupled than before :( * Included a tiny example change for the Vector skin, a lot more of similar cleanup is possible and planned for the future. * Many of the non-essential mediawiki.* modules defined in Resources.php should be using `'skinStyles' => array( 'default' => … )` instead of `'styles' => …` to allow more customizations, this is also planned for the future after auditing which ones would actually benefit from this. Change-Id: Ica4ff9696b490e35f60288d7ce1295766c427e87
2014-06-26 14:29:31 +00:00
$styleFiles = isset( $this->moduleInfos[$name]['skinStyles']['default'] ) ?
(array)$this->moduleInfos[$name]['skinStyles']['default'] :
resourceloader: Allow skins to provide additional styles for any module The newly introduced $wgResourceModuleSkinStyles global enables skins to provide additional stylesheets to existing ResourceLoader module. This both makes it easier (or at all possible) to override default styles and lowers the style footprint by making it possible not to load styles unused on most pages. ---- Example: Use the file 'foo-styles.css' for the 'mediawiki.foo' module when using the MySkin skin: $wgResourceModuleSkinStyles['myskin'] = array( 'mediawiki.foo' => 'foo-styles.css', 'remoteSkinPath' => 'MySkin', 'localBasePath' => __DIR__, ); For detailed documentation, see the doc comment in DefaultSettings.php. For a practical usage example, see Vector.php. ---- Implementation notes: * The values defined in $wgResourceModuleSkinStyles are embedded into the modules as late as possible (in ResourceLoader::register()). * Only plain file modules are supported, setting module skin styles for other module types has no effect. * ResourceLoader and ResourceLoaderFileModule now support loading files from arbitrary paths to make this possible, defined using ResourceLoaderFilePath objects. * This required some adjustments in seemingly unrelated places for code which didn't handle the paths fully correctly before. * ResourceLoader and ResourceLoaderFileModule are now a bit more tightly coupled than before :( * Included a tiny example change for the Vector skin, a lot more of similar cleanup is possible and planned for the future. * Many of the non-essential mediawiki.* modules defined in Resources.php should be using `'skinStyles' => array( 'default' => … )` instead of `'styles' => …` to allow more customizations, this is also planned for the future after auditing which ones would actually benefit from this. Change-Id: Ica4ff9696b490e35f60288d7ce1295766c427e87
2014-06-26 14:29:31 +00:00
array();
} else {
continue;
}
// Add new file paths, remapping them to refer to our directories and not use settings
// from the module we're modifying, which come from the base definition.
resourceloader: Allow skins to provide additional styles for any module The newly introduced $wgResourceModuleSkinStyles global enables skins to provide additional stylesheets to existing ResourceLoader module. This both makes it easier (or at all possible) to override default styles and lowers the style footprint by making it possible not to load styles unused on most pages. ---- Example: Use the file 'foo-styles.css' for the 'mediawiki.foo' module when using the MySkin skin: $wgResourceModuleSkinStyles['myskin'] = array( 'mediawiki.foo' => 'foo-styles.css', 'remoteSkinPath' => 'MySkin', 'localBasePath' => __DIR__, ); For detailed documentation, see the doc comment in DefaultSettings.php. For a practical usage example, see Vector.php. ---- Implementation notes: * The values defined in $wgResourceModuleSkinStyles are embedded into the modules as late as possible (in ResourceLoader::register()). * Only plain file modules are supported, setting module skin styles for other module types has no effect. * ResourceLoader and ResourceLoaderFileModule now support loading files from arbitrary paths to make this possible, defined using ResourceLoaderFilePath objects. * This required some adjustments in seemingly unrelated places for code which didn't handle the paths fully correctly before. * ResourceLoader and ResourceLoaderFileModule are now a bit more tightly coupled than before :( * Included a tiny example change for the Vector skin, a lot more of similar cleanup is possible and planned for the future. * Many of the non-essential mediawiki.* modules defined in Resources.php should be using `'skinStyles' => array( 'default' => … )` instead of `'styles' => …` to allow more customizations, this is also planned for the future after auditing which ones would actually benefit from this. Change-Id: Ica4ff9696b490e35f60288d7ce1295766c427e87
2014-06-26 14:29:31 +00:00
list( $localBasePath, $remoteBasePath ) =
ResourceLoaderFileModule::extractBasePaths( $skinStyles );
foreach ( $paths as $path ) {
$styleFiles[] = new ResourceLoaderFilePath( $path, $localBasePath, $remoteBasePath );
}
$this->moduleInfos[$name]['skinStyles'][$skinName] = $styleFiles;
}
}
}
}
2010-09-04 12:53:01 +00:00
/**
*/
public function registerTestModules() {
global $IP;
if ( $this->config->get( 'EnableJavaScriptTest' ) !== true ) {
throw new MWException( 'Attempt to register JavaScript test modules '
. 'but <code>$wgEnableJavaScriptTest</code> is false. '
. 'Edit your <code>LocalSettings.php</code> to enable it.' );
}
// Get core test suites
$testModules = array();
$testModules['qunit'] = array();
// Get other test suites (e.g. from extensions)
Hooks::run( 'ResourceLoaderTestModules', array( &$testModules, &$this ) );
// Add the testrunner (which configures QUnit) to the dependencies.
// Since it must be ready before any of the test suites are executed.
foreach ( $testModules['qunit'] as &$module ) {
// Make sure all test modules are top-loading so that when QUnit starts
// on document-ready, it will run once and finish. If some tests arrive
// later (possibly after QUnit has already finished) they will be ignored.
$module['position'] = 'top';
$module['dependencies'][] = 'test.mediawiki.qunit.testrunner';
}
$testModules['qunit'] =
( include "$IP/tests/qunit/QUnitTestResources.php" ) + $testModules['qunit'];
foreach ( $testModules as $id => $names ) {
// Register test modules
$this->register( $testModules[$id] );
// Keep track of their names so that they can be loaded together
$this->testModuleNames[$id] = array_keys( $testModules[$id] );
}
}
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
/**
* Add a foreign source of modules.
*
* @param array|string $id Source ID (string), or array( id1 => loadUrl, id2 => loadUrl, ... )
* @param string|array $loadUrl load.php url (string), or array with loadUrl key for
* backwards-compatibility.
* @throws MWException
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
*/
public function addSource( $id, $loadUrl = null ) {
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
// Allow multiple sources to be registered in one call
if ( is_array( $id ) ) {
foreach ( $id as $key => $value ) {
$this->addSource( $key, $value );
}
return;
}
// Disallow duplicates
if ( isset( $this->sources[$id] ) ) {
throw new MWException(
'ResourceLoader duplicate source addition error. ' .
'Another source has already been registered as ' . $id
);
}
// Pre 1.24 backwards-compatibility
if ( is_array( $loadUrl ) ) {
if ( !isset( $loadUrl['loadScript'] ) ) {
throw new MWException(
__METHOD__ . ' was passed an array with no "loadScript" key.'
);
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
}
$loadUrl = $loadUrl['loadScript'];
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
}
$this->sources[$id] = $loadUrl;
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
}
/**
* Get a list of module names.
*
* @return array List of module names
*/
* Made Resources.php return a pure-data array instead of an ugly mix of data and code. This allows the class code to be lazy-loaded with the autoloader, for a performance advantage especially on non-APC installs. And using the convention where if the class is omitted, ResourceLoaderFileModule is assumed, the registration code becomes shorter and simpler. * Modified ResourceLoader to lazy-initialise module objects, for a further performance advantage. * Deleted ResourceLoader::getModules(), provided getModuleNames() instead. Although the startup module needs this functionality, it's slow to generate, so to avoid misuse, it's better to provide a foolproof fast interface and let the startup module do the slow thing itself. * Modified ResourceLoader::register() to optionally accept an info array instead of an object. * Added $wgResourceModules, allowing extensions to efficiently define their own resource loader modules. The trouble with hooks is that they contain code, and code is slow. We've been through all this before with i18n. Hooks are useful as a performance tool only if you call them very rarely. * Moved ResourceLoader settings to their own section in DefaultSettings.php * Added options to ResourceLoaderFileModule equivalent to the $localBasePath and $remoteBasePath parameters, to allow it to be instantiated via the new array style. Also added remoteExtPath, which allows modules to be registered before $wgExtensionAssetsPath is known. * Added OutputPage::getResourceLoader(), mostly for debugging. * The time saving at the moment is about 5ms per request with no extensions, which is significant already with 6 load.php requests for a cold cache page view. This is a much more scalable interface; the relative saving will grow as more extensions are added which use this interface, especially for non-APC installs. Although the interface is backwards compatible, extension updates will follow in a subsequent commit.
2010-11-19 10:41:06 +00:00
public function getModuleNames() {
return array_keys( $this->moduleInfos );
}
/**
* Get a list of test module names for one (or all) frameworks.
*
* If the given framework id is unknkown, or if the in-object variable is not an array,
* then it will return an empty array.
*
* @param string $framework Get only the test module names for one
* particular framework (optional)
* @return array
*/
public function getTestModuleNames( $framework = 'all' ) {
/** @todo api siteinfo prop testmodulenames modulenames */
if ( $framework == 'all' ) {
return $this->testModuleNames;
} elseif ( isset( $this->testModuleNames[$framework] )
&& is_array( $this->testModuleNames[$framework] )
) {
return $this->testModuleNames[$framework];
} else {
return array();
}
}
2010-09-04 12:53:01 +00:00
/**
* Check whether a ResourceLoader module is registered
*
* @since 1.25
* @param string $name
* @return bool
*/
public function isModuleRegistered( $name ) {
return isset( $this->moduleInfos[$name] );
}
/**
* Get the ResourceLoaderModule object for a given module name.
2010-09-05 13:31:34 +00:00
*
* If an array of module parameters exists but a ResourceLoaderModule object has not
* yet been instantiated, this method will instantiate and cache that object such that
* subsequent calls simply return the same object.
*
* @param string $name Module name
* @return ResourceLoaderModule|null If module has been registered, return a
* ResourceLoaderModule instance. Otherwise, return null.
*/
public function getModule( $name ) {
* Made Resources.php return a pure-data array instead of an ugly mix of data and code. This allows the class code to be lazy-loaded with the autoloader, for a performance advantage especially on non-APC installs. And using the convention where if the class is omitted, ResourceLoaderFileModule is assumed, the registration code becomes shorter and simpler. * Modified ResourceLoader to lazy-initialise module objects, for a further performance advantage. * Deleted ResourceLoader::getModules(), provided getModuleNames() instead. Although the startup module needs this functionality, it's slow to generate, so to avoid misuse, it's better to provide a foolproof fast interface and let the startup module do the slow thing itself. * Modified ResourceLoader::register() to optionally accept an info array instead of an object. * Added $wgResourceModules, allowing extensions to efficiently define their own resource loader modules. The trouble with hooks is that they contain code, and code is slow. We've been through all this before with i18n. Hooks are useful as a performance tool only if you call them very rarely. * Moved ResourceLoader settings to their own section in DefaultSettings.php * Added options to ResourceLoaderFileModule equivalent to the $localBasePath and $remoteBasePath parameters, to allow it to be instantiated via the new array style. Also added remoteExtPath, which allows modules to be registered before $wgExtensionAssetsPath is known. * Added OutputPage::getResourceLoader(), mostly for debugging. * The time saving at the moment is about 5ms per request with no extensions, which is significant already with 6 load.php requests for a cold cache page view. This is a much more scalable interface; the relative saving will grow as more extensions are added which use this interface, especially for non-APC installs. Although the interface is backwards compatible, extension updates will follow in a subsequent commit.
2010-11-19 10:41:06 +00:00
if ( !isset( $this->modules[$name] ) ) {
if ( !isset( $this->moduleInfos[$name] ) ) {
// No such module
return null;
}
// Construct the requested object
$info = $this->moduleInfos[$name];
/** @var ResourceLoaderModule $object */
* Made Resources.php return a pure-data array instead of an ugly mix of data and code. This allows the class code to be lazy-loaded with the autoloader, for a performance advantage especially on non-APC installs. And using the convention where if the class is omitted, ResourceLoaderFileModule is assumed, the registration code becomes shorter and simpler. * Modified ResourceLoader to lazy-initialise module objects, for a further performance advantage. * Deleted ResourceLoader::getModules(), provided getModuleNames() instead. Although the startup module needs this functionality, it's slow to generate, so to avoid misuse, it's better to provide a foolproof fast interface and let the startup module do the slow thing itself. * Modified ResourceLoader::register() to optionally accept an info array instead of an object. * Added $wgResourceModules, allowing extensions to efficiently define their own resource loader modules. The trouble with hooks is that they contain code, and code is slow. We've been through all this before with i18n. Hooks are useful as a performance tool only if you call them very rarely. * Moved ResourceLoader settings to their own section in DefaultSettings.php * Added options to ResourceLoaderFileModule equivalent to the $localBasePath and $remoteBasePath parameters, to allow it to be instantiated via the new array style. Also added remoteExtPath, which allows modules to be registered before $wgExtensionAssetsPath is known. * Added OutputPage::getResourceLoader(), mostly for debugging. * The time saving at the moment is about 5ms per request with no extensions, which is significant already with 6 load.php requests for a cold cache page view. This is a much more scalable interface; the relative saving will grow as more extensions are added which use this interface, especially for non-APC installs. Although the interface is backwards compatible, extension updates will follow in a subsequent commit.
2010-11-19 10:41:06 +00:00
if ( isset( $info['object'] ) ) {
// Object given in info array
$object = $info['object'];
} else {
if ( !isset( $info['class'] ) ) {
$class = 'ResourceLoaderFileModule';
} else {
$class = $info['class'];
}
/** @var ResourceLoaderModule $object */
* Made Resources.php return a pure-data array instead of an ugly mix of data and code. This allows the class code to be lazy-loaded with the autoloader, for a performance advantage especially on non-APC installs. And using the convention where if the class is omitted, ResourceLoaderFileModule is assumed, the registration code becomes shorter and simpler. * Modified ResourceLoader to lazy-initialise module objects, for a further performance advantage. * Deleted ResourceLoader::getModules(), provided getModuleNames() instead. Although the startup module needs this functionality, it's slow to generate, so to avoid misuse, it's better to provide a foolproof fast interface and let the startup module do the slow thing itself. * Modified ResourceLoader::register() to optionally accept an info array instead of an object. * Added $wgResourceModules, allowing extensions to efficiently define their own resource loader modules. The trouble with hooks is that they contain code, and code is slow. We've been through all this before with i18n. Hooks are useful as a performance tool only if you call them very rarely. * Moved ResourceLoader settings to their own section in DefaultSettings.php * Added options to ResourceLoaderFileModule equivalent to the $localBasePath and $remoteBasePath parameters, to allow it to be instantiated via the new array style. Also added remoteExtPath, which allows modules to be registered before $wgExtensionAssetsPath is known. * Added OutputPage::getResourceLoader(), mostly for debugging. * The time saving at the moment is about 5ms per request with no extensions, which is significant already with 6 load.php requests for a cold cache page view. This is a much more scalable interface; the relative saving will grow as more extensions are added which use this interface, especially for non-APC installs. Although the interface is backwards compatible, extension updates will follow in a subsequent commit.
2010-11-19 10:41:06 +00:00
$object = new $class( $info );
$object->setConfig( $this->getConfig() );
* Made Resources.php return a pure-data array instead of an ugly mix of data and code. This allows the class code to be lazy-loaded with the autoloader, for a performance advantage especially on non-APC installs. And using the convention where if the class is omitted, ResourceLoaderFileModule is assumed, the registration code becomes shorter and simpler. * Modified ResourceLoader to lazy-initialise module objects, for a further performance advantage. * Deleted ResourceLoader::getModules(), provided getModuleNames() instead. Although the startup module needs this functionality, it's slow to generate, so to avoid misuse, it's better to provide a foolproof fast interface and let the startup module do the slow thing itself. * Modified ResourceLoader::register() to optionally accept an info array instead of an object. * Added $wgResourceModules, allowing extensions to efficiently define their own resource loader modules. The trouble with hooks is that they contain code, and code is slow. We've been through all this before with i18n. Hooks are useful as a performance tool only if you call them very rarely. * Moved ResourceLoader settings to their own section in DefaultSettings.php * Added options to ResourceLoaderFileModule equivalent to the $localBasePath and $remoteBasePath parameters, to allow it to be instantiated via the new array style. Also added remoteExtPath, which allows modules to be registered before $wgExtensionAssetsPath is known. * Added OutputPage::getResourceLoader(), mostly for debugging. * The time saving at the moment is about 5ms per request with no extensions, which is significant already with 6 load.php requests for a cold cache page view. This is a much more scalable interface; the relative saving will grow as more extensions are added which use this interface, especially for non-APC installs. Although the interface is backwards compatible, extension updates will follow in a subsequent commit.
2010-11-19 10:41:06 +00:00
}
$object->setName( $name );
$this->modules[$name] = $object;
}
return $this->modules[$name];
}
2010-09-04 12:53:01 +00:00
resourceloader: Allow skins to provide additional styles for any module The newly introduced $wgResourceModuleSkinStyles global enables skins to provide additional stylesheets to existing ResourceLoader module. This both makes it easier (or at all possible) to override default styles and lowers the style footprint by making it possible not to load styles unused on most pages. ---- Example: Use the file 'foo-styles.css' for the 'mediawiki.foo' module when using the MySkin skin: $wgResourceModuleSkinStyles['myskin'] = array( 'mediawiki.foo' => 'foo-styles.css', 'remoteSkinPath' => 'MySkin', 'localBasePath' => __DIR__, ); For detailed documentation, see the doc comment in DefaultSettings.php. For a practical usage example, see Vector.php. ---- Implementation notes: * The values defined in $wgResourceModuleSkinStyles are embedded into the modules as late as possible (in ResourceLoader::register()). * Only plain file modules are supported, setting module skin styles for other module types has no effect. * ResourceLoader and ResourceLoaderFileModule now support loading files from arbitrary paths to make this possible, defined using ResourceLoaderFilePath objects. * This required some adjustments in seemingly unrelated places for code which didn't handle the paths fully correctly before. * ResourceLoader and ResourceLoaderFileModule are now a bit more tightly coupled than before :( * Included a tiny example change for the Vector skin, a lot more of similar cleanup is possible and planned for the future. * Many of the non-essential mediawiki.* modules defined in Resources.php should be using `'skinStyles' => array( 'default' => … )` instead of `'styles' => …` to allow more customizations, this is also planned for the future after auditing which ones would actually benefit from this. Change-Id: Ica4ff9696b490e35f60288d7ce1295766c427e87
2014-06-26 14:29:31 +00:00
/**
* Return whether the definition of a module corresponds to a simple ResourceLoaderFileModule.
*
* @param string $name Module name
* @return bool
resourceloader: Allow skins to provide additional styles for any module The newly introduced $wgResourceModuleSkinStyles global enables skins to provide additional stylesheets to existing ResourceLoader module. This both makes it easier (or at all possible) to override default styles and lowers the style footprint by making it possible not to load styles unused on most pages. ---- Example: Use the file 'foo-styles.css' for the 'mediawiki.foo' module when using the MySkin skin: $wgResourceModuleSkinStyles['myskin'] = array( 'mediawiki.foo' => 'foo-styles.css', 'remoteSkinPath' => 'MySkin', 'localBasePath' => __DIR__, ); For detailed documentation, see the doc comment in DefaultSettings.php. For a practical usage example, see Vector.php. ---- Implementation notes: * The values defined in $wgResourceModuleSkinStyles are embedded into the modules as late as possible (in ResourceLoader::register()). * Only plain file modules are supported, setting module skin styles for other module types has no effect. * ResourceLoader and ResourceLoaderFileModule now support loading files from arbitrary paths to make this possible, defined using ResourceLoaderFilePath objects. * This required some adjustments in seemingly unrelated places for code which didn't handle the paths fully correctly before. * ResourceLoader and ResourceLoaderFileModule are now a bit more tightly coupled than before :( * Included a tiny example change for the Vector skin, a lot more of similar cleanup is possible and planned for the future. * Many of the non-essential mediawiki.* modules defined in Resources.php should be using `'skinStyles' => array( 'default' => … )` instead of `'styles' => …` to allow more customizations, this is also planned for the future after auditing which ones would actually benefit from this. Change-Id: Ica4ff9696b490e35f60288d7ce1295766c427e87
2014-06-26 14:29:31 +00:00
*/
protected function isFileModule( $name ) {
if ( !isset( $this->moduleInfos[$name] ) ) {
return false;
}
$info = $this->moduleInfos[$name];
if ( isset( $info['object'] ) || isset( $info['class'] ) ) {
return false;
}
return true;
}
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
/**
* Get the list of sources.
*
* @return array Like array( id => load.php url, .. )
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
*/
public function getSources() {
return $this->sources;
}
/**
* Get the URL to the load.php endpoint for the given
* ResourceLoader source
*
* @since 1.24
* @param string $source
* @throws MWException On an invalid $source name
* @return string
*/
public function getLoadScript( $source ) {
if ( !isset( $this->sources[$source] ) ) {
throw new MWException( "The $source source was never registered in ResourceLoader." );
}
return $this->sources[$source];
}
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
/**
* @since 1.26
* @param string $value
* @return string Hash
*/
public static function makeHash( $value ) {
// Use base64 to output more entropy in a more compact string (default hex is only base16).
// The first 8 chars of a base64 encoded digest represent the same binary as
// the first 12 chars of a hex encoded digest.
return substr( base64_encode( sha1( $value, true ) ), 0, 8 );
}
/**
* Helper method to get and combine versions of multiple modules.
*
* @since 1.26
* @param ResourceLoaderContext $context
* @param array $modules List of ResourceLoaderModule objects
* @return string Hash
*/
public function getCombinedVersion( ResourceLoaderContext $context, Array $modules ) {
if ( !$modules ) {
return '';
}
// Support: PHP 5.3 ("$this" for anonymous functions was added in PHP 5.4.0)
// http://php.net/functions.anonymous
$rl = $this;
$hashes = array_map( function ( $module ) use ( $rl, $context ) {
return $rl->getModule( $module )->getVersionHash( $context );
}, $modules );
return self::makeHash( implode( $hashes ) );
}
2010-09-05 13:31:34 +00:00
/**
* Output a response to a load request, including the content-type header.
*
* @param ResourceLoaderContext $context Context in which a response should be formed
*/
public function respond( ResourceLoaderContext $context ) {
// Buffer output to catch warnings. Normally we'd use ob_clean() on the
// top-level output buffer to clear warnings, but that breaks when ob_gzhandler
// is used: ob_clean() will clear the GZIP header in that case and it won't come
// back for subsequent output, resulting in invalid GZIP. So we have to wrap
// the whole thing in our own output buffer to be sure the active buffer
// doesn't use ob_gzhandler.
// See http://bugs.php.net/bug.php?id=36514
ob_start();
// Find out which modules are missing and instantiate the others
$modules = array();
$missing = array();
foreach ( $context->getModules() as $name ) {
$module = $this->getModule( $name );
if ( $module ) {
* (bug 34212) ApiBlock/ApiUnblock allow action to take place without a token parameter present * (bug 35317) CSRF in Special:Upload Revert r56793, which removed the CSRF check for Special:Upload for normal file uploads. Cross-site posting of file uploads without user interaction has been possible since at least as early as Chrome 8 (late 2010) and Firefox 6 (mid 2011). Commonist has used api.php since version 0.4.0 (April 2010), and the API already requires an edit token, so Commonist 0.4.0+ is not affected by this change. * (bug 34907) Fix for CSRF vulnerability due to mw.user.tokens. Patch by Roan Kattouw and Tim Starling. * Filter out private modules early in ResourceLoader::makeResponse() and just pretend they weren't specified. This means these modules cannot be loaded through load.php . This filtering must not happen in makeModuleResponse(), because that would break inlining. * Force inlining of private modules in OutputPage::makeResourceLoaderLink(), disregarding $wgResourceLoaderInlinePrivateModules * Remove $wgResourceLoaderInlinePrivateModules * Remove special treatment of private modules ($private) in ResourceLoader::makeResponse() and sendResponseHeaders(), because we're not allowing private modules to be loaded through here any more * Remove identity checks in ResourceLoaderUserOptionsModule and ResourceLoaderUserCSSPrefsModule, they didn't make a lot of sense before but they're certainly useless now. * Factored out error comment construction in ResourceLoader.php and stripped comment terminations from exception messages. I didn't find an XSS vulnerability but it looked scary. Patchset2: Removes whitespace error that prevented automatic merge by Gerrit: includes/resourceloader/ResourceLoaderUserOptionsModule.php Change-Id: I2dec8b8caf9db3c64919763865cc10cccdd6a1a3
2012-03-22 19:52:37 +00:00
// Do not allow private modules to be loaded from the web.
// This is a security issue, see bug 34907.
if ( $module->getGroup() === 'private' ) {
$this->logger->debug( "Request for private module '$name' denied" );
$this->errors[] = "Cannot show private module \"$name\"";
* (bug 34212) ApiBlock/ApiUnblock allow action to take place without a token parameter present * (bug 35317) CSRF in Special:Upload Revert r56793, which removed the CSRF check for Special:Upload for normal file uploads. Cross-site posting of file uploads without user interaction has been possible since at least as early as Chrome 8 (late 2010) and Firefox 6 (mid 2011). Commonist has used api.php since version 0.4.0 (April 2010), and the API already requires an edit token, so Commonist 0.4.0+ is not affected by this change. * (bug 34907) Fix for CSRF vulnerability due to mw.user.tokens. Patch by Roan Kattouw and Tim Starling. * Filter out private modules early in ResourceLoader::makeResponse() and just pretend they weren't specified. This means these modules cannot be loaded through load.php . This filtering must not happen in makeModuleResponse(), because that would break inlining. * Force inlining of private modules in OutputPage::makeResourceLoaderLink(), disregarding $wgResourceLoaderInlinePrivateModules * Remove $wgResourceLoaderInlinePrivateModules * Remove special treatment of private modules ($private) in ResourceLoader::makeResponse() and sendResponseHeaders(), because we're not allowing private modules to be loaded through here any more * Remove identity checks in ResourceLoaderUserOptionsModule and ResourceLoaderUserCSSPrefsModule, they didn't make a lot of sense before but they're certainly useless now. * Factored out error comment construction in ResourceLoader.php and stripped comment terminations from exception messages. I didn't find an XSS vulnerability but it looked scary. Patchset2: Removes whitespace error that prevented automatic merge by Gerrit: includes/resourceloader/ResourceLoaderUserOptionsModule.php Change-Id: I2dec8b8caf9db3c64919763865cc10cccdd6a1a3
2012-03-22 19:52:37 +00:00
continue;
}
$modules[$name] = $module;
} else {
$missing[] = $name;
}
}
2010-09-04 12:53:01 +00:00
try {
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
// Preload for getCombinedVersion()
$this->preloadModuleInfo( array_keys( $modules ), $context );
} catch ( Exception $e ) {
MWExceptionHandler::logException( $e );
$this->logger->warning( 'Preloading module info failed: {exception}', array(
'exception' => $e
) );
$this->errors[] = self::formatExceptionNoComment( $e );
}
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
// Combine versions to propagate cache invalidation
$versionHash = '';
try {
$versionHash = $this->getCombinedVersion( $context, array_keys( $modules ) );
} catch ( Exception $e ) {
MWExceptionHandler::logException( $e );
$this->logger->warning( 'Calculating version hash failed: {exception}', array(
'exception' => $e
) );
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
$this->errors[] = self::formatExceptionNoComment( $e );
}
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
// See RFC 2616 § 3.11 Entity Tags
// http://www.w3.org/Protocols/rfc2616/rfc2616-sec3.html#sec3.11
$etag = 'W/"' . $versionHash . '"';
// Try the client-side cache first
if ( $this->tryRespondNotModified( $context, $etag ) ) {
return; // output handled (buffers cleared)
}
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
// Use file cache if enabled and available...
if ( $this->config->get( 'UseFileCache' ) ) {
$fileCache = ResourceFileCache::newFromContext( $context );
if ( $this->tryRespondFromFileCache( $fileCache, $context, $etag ) ) {
return; // output handled
}
}
// Generate a response
$response = $this->makeModuleResponse( $context, $modules, $missing );
// Capture any PHP warnings from the output buffer and append them to the
// error list if we're in debug mode.
if ( $context->getDebug() && strlen( $warnings = ob_get_contents() ) ) {
$this->errors[] = $warnings;
}
* (bug 34212) ApiBlock/ApiUnblock allow action to take place without a token parameter present * (bug 35317) CSRF in Special:Upload Revert r56793, which removed the CSRF check for Special:Upload for normal file uploads. Cross-site posting of file uploads without user interaction has been possible since at least as early as Chrome 8 (late 2010) and Firefox 6 (mid 2011). Commonist has used api.php since version 0.4.0 (April 2010), and the API already requires an edit token, so Commonist 0.4.0+ is not affected by this change. * (bug 34907) Fix for CSRF vulnerability due to mw.user.tokens. Patch by Roan Kattouw and Tim Starling. * Filter out private modules early in ResourceLoader::makeResponse() and just pretend they weren't specified. This means these modules cannot be loaded through load.php . This filtering must not happen in makeModuleResponse(), because that would break inlining. * Force inlining of private modules in OutputPage::makeResourceLoaderLink(), disregarding $wgResourceLoaderInlinePrivateModules * Remove $wgResourceLoaderInlinePrivateModules * Remove special treatment of private modules ($private) in ResourceLoader::makeResponse() and sendResponseHeaders(), because we're not allowing private modules to be loaded through here any more * Remove identity checks in ResourceLoaderUserOptionsModule and ResourceLoaderUserCSSPrefsModule, they didn't make a lot of sense before but they're certainly useless now. * Factored out error comment construction in ResourceLoader.php and stripped comment terminations from exception messages. I didn't find an XSS vulnerability but it looked scary. Patchset2: Removes whitespace error that prevented automatic merge by Gerrit: includes/resourceloader/ResourceLoaderUserOptionsModule.php Change-Id: I2dec8b8caf9db3c64919763865cc10cccdd6a1a3
2012-03-22 19:52:37 +00:00
// Save response to file cache unless there are errors
if ( isset( $fileCache ) && !$this->errors && !count( $missing ) ) {
// Cache single modules and images...and other requests if there are enough hits
if ( ResourceFileCache::useFileCache( $context ) ) {
if ( $fileCache->isCacheWorthy() ) {
$fileCache->saveText( $response );
} else {
$fileCache->incrMissesRecent( $context->getRequest() );
}
}
}
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
$this->sendResponseHeaders( $context, $etag, (bool)$this->errors );
// Remove the output buffer and output the response
ob_end_clean();
if ( $context->getImageObj() && $this->errors ) {
// We can't show both the error messages and the response when it's an image.
$errorText = '';
foreach ( $this->errors as $error ) {
$errorText .= $error . "\n";
}
$response = $errorText;
} elseif ( $this->errors ) {
// Prepend comments indicating errors
$errorText = '';
foreach ( $this->errors as $error ) {
$errorText .= self::makeComment( $error );
}
$response = $errorText . $response;
}
$this->errors = array();
echo $response;
}
/**
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
* Send main response headers to the client.
*
* Deals with Content-Type, CORS (for stylesheets), and caching.
*
* @param ResourceLoaderContext $context
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
* @param string $etag ETag header value
* @param bool $errors Whether there are errors in the response
* @return void
*/
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
protected function sendResponseHeaders( ResourceLoaderContext $context, $etag, $errors ) {
$rlMaxage = $this->config->get( 'ResourceLoaderMaxage' );
// If a version wasn't specified we need a shorter expiry time for updates
// to propagate to clients quickly
// If there were errors, we also need a shorter expiry time so we can recover quickly
if ( is_null( $context->getVersion() ) || $errors ) {
$maxage = $rlMaxage['unversioned']['client'];
$smaxage = $rlMaxage['unversioned']['server'];
// If a version was specified we can use a longer expiry time since changing
// version numbers causes cache misses
} else {
$maxage = $rlMaxage['versioned']['client'];
$smaxage = $rlMaxage['versioned']['server'];
}
if ( $context->getImageObj() ) {
// Output different headers if we're outputting textual errors.
if ( $errors ) {
header( 'Content-Type: text/plain; charset=utf-8' );
} else {
$context->getImageObj()->sendResponseHeaders( $context );
}
} elseif ( $context->getOnly() === 'styles' ) {
header( 'Content-Type: text/css; charset=utf-8' );
header( 'Access-Control-Allow-Origin: *' );
} else {
header( 'Content-Type: text/javascript; charset=utf-8' );
}
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
// See RFC 2616 § 14.19 ETag
// http://www.w3.org/Protocols/rfc2616/rfc2616-sec14.html#sec14.19
header( 'ETag: ' . $etag );
if ( $context->getDebug() ) {
// Do not cache debug responses
header( 'Cache-Control: private, no-cache, must-revalidate' );
header( 'Pragma: no-cache' );
} else {
header( "Cache-Control: public, max-age=$maxage, s-maxage=$smaxage" );
* (bug 34212) ApiBlock/ApiUnblock allow action to take place without a token parameter present * (bug 35317) CSRF in Special:Upload Revert r56793, which removed the CSRF check for Special:Upload for normal file uploads. Cross-site posting of file uploads without user interaction has been possible since at least as early as Chrome 8 (late 2010) and Firefox 6 (mid 2011). Commonist has used api.php since version 0.4.0 (April 2010), and the API already requires an edit token, so Commonist 0.4.0+ is not affected by this change. * (bug 34907) Fix for CSRF vulnerability due to mw.user.tokens. Patch by Roan Kattouw and Tim Starling. * Filter out private modules early in ResourceLoader::makeResponse() and just pretend they weren't specified. This means these modules cannot be loaded through load.php . This filtering must not happen in makeModuleResponse(), because that would break inlining. * Force inlining of private modules in OutputPage::makeResourceLoaderLink(), disregarding $wgResourceLoaderInlinePrivateModules * Remove $wgResourceLoaderInlinePrivateModules * Remove special treatment of private modules ($private) in ResourceLoader::makeResponse() and sendResponseHeaders(), because we're not allowing private modules to be loaded through here any more * Remove identity checks in ResourceLoaderUserOptionsModule and ResourceLoaderUserCSSPrefsModule, they didn't make a lot of sense before but they're certainly useless now. * Factored out error comment construction in ResourceLoader.php and stripped comment terminations from exception messages. I didn't find an XSS vulnerability but it looked scary. Patchset2: Removes whitespace error that prevented automatic merge by Gerrit: includes/resourceloader/ResourceLoaderUserOptionsModule.php Change-Id: I2dec8b8caf9db3c64919763865cc10cccdd6a1a3
2012-03-22 19:52:37 +00:00
$exp = min( $maxage, $smaxage );
header( 'Expires: ' . wfTimestamp( TS_RFC2822, $exp + time() ) );
}
// Send the current time expressed as fractional seconds since epoch,
// with microsecond precision. This helps distinguish hits from misses
// in edge caches.
header( 'MediaWiki-Timestamp: ' . microtime( true ) );
}
2010-09-04 12:53:01 +00:00
/**
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
* Respond with HTTP 304 Not Modified if appropiate.
*
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
* If there's an If-None-Match header, respond with a 304 appropriately
* and clear out the output buffer. If the client cache is too old then do nothing.
*
* @param ResourceLoaderContext $context
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
* @param string $etag ETag header value
* @return bool True if HTTP 304 was sent and output handled
*/
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
protected function tryRespondNotModified( ResourceLoaderContext $context, $etag ) {
// See RFC 2616 § 14.26 If-None-Match
// http://www.w3.org/Protocols/rfc2616/rfc2616-sec14.html#sec14.26
$clientKeys = $context->getRequest()->getHeader( 'If-None-Match', WebRequest::GETHEADER_LIST );
// Never send 304s in debug mode
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
if ( $clientKeys !== false && !$context->getDebug() && in_array( $etag, $clientKeys ) ) {
// There's another bug in ob_gzhandler (see also the comment at
// the top of this function) that causes it to gzip even empty
// responses, meaning it's impossible to produce a truly empty
// response (because the gzip header is always there). This is
// a problem because 304 responses have to be completely empty
// per the HTTP spec, and Firefox behaves buggily when they're not.
// See also http://bugs.php.net/bug.php?id=51579
// To work around this, we tear down all output buffering before
// sending the 304.
wfResetOutputBuffers( /* $resetGzipEncoding = */ true );
HttpStatus::header( 304 );
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
$this->sendResponseHeaders( $context, $etag, false );
return true;
}
return false;
}
/**
* Send out code for a response from file cache if possible.
*
* @param ResourceFileCache $fileCache Cache object for this request URL
* @param ResourceLoaderContext $context Context in which to generate a response
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
* @param string $etag ETag header value
* @return bool If this found a cache file and handled the response
*/
protected function tryRespondFromFileCache(
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
ResourceFileCache $fileCache,
ResourceLoaderContext $context,
$etag
) {
$rlMaxage = $this->config->get( 'ResourceLoaderMaxage' );
// Buffer output to catch warnings.
ob_start();
// Get the maximum age the cache can be
$maxage = is_null( $context->getVersion() )
? $rlMaxage['unversioned']['server']
: $rlMaxage['versioned']['server'];
// Minimum timestamp the cache file must have
$good = $fileCache->isCacheGood( wfTimestamp( TS_MW, time() - $maxage ) );
if ( !$good ) {
try { // RL always hits the DB on file cache miss...
wfGetDB( DB_SLAVE );
} catch ( DBConnectionError $e ) { // ...check if we need to fallback to cache
$good = $fileCache->isCacheGood(); // cache existence check
}
}
if ( $good ) {
$ts = $fileCache->cacheTimestamp();
// Send content type and cache headers
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
$this->sendResponseHeaders( $context, $etag, false );
$response = $fileCache->fetchText();
// Capture any PHP warnings from the output buffer and append them to the
// response in a comment if we're in debug mode.
if ( $context->getDebug() && strlen( $warnings = ob_get_contents() ) ) {
$response = self::makeComment( $warnings ) . $response;
}
// Remove the output buffer and output the response
ob_end_clean();
echo $response . "\n/* Cached {$ts} */";
return true; // cache hit
}
// Clear buffer
ob_end_clean();
return false; // cache miss
}
/**
* Generate a CSS or JS comment block.
*
* Only use this for public data, not error message details.
*
* @param string $text
* @return string
*/
public static function makeComment( $text ) {
* (bug 34212) ApiBlock/ApiUnblock allow action to take place without a token parameter present * (bug 35317) CSRF in Special:Upload Revert r56793, which removed the CSRF check for Special:Upload for normal file uploads. Cross-site posting of file uploads without user interaction has been possible since at least as early as Chrome 8 (late 2010) and Firefox 6 (mid 2011). Commonist has used api.php since version 0.4.0 (April 2010), and the API already requires an edit token, so Commonist 0.4.0+ is not affected by this change. * (bug 34907) Fix for CSRF vulnerability due to mw.user.tokens. Patch by Roan Kattouw and Tim Starling. * Filter out private modules early in ResourceLoader::makeResponse() and just pretend they weren't specified. This means these modules cannot be loaded through load.php . This filtering must not happen in makeModuleResponse(), because that would break inlining. * Force inlining of private modules in OutputPage::makeResourceLoaderLink(), disregarding $wgResourceLoaderInlinePrivateModules * Remove $wgResourceLoaderInlinePrivateModules * Remove special treatment of private modules ($private) in ResourceLoader::makeResponse() and sendResponseHeaders(), because we're not allowing private modules to be loaded through here any more * Remove identity checks in ResourceLoaderUserOptionsModule and ResourceLoaderUserCSSPrefsModule, they didn't make a lot of sense before but they're certainly useless now. * Factored out error comment construction in ResourceLoader.php and stripped comment terminations from exception messages. I didn't find an XSS vulnerability but it looked scary. Patchset2: Removes whitespace error that prevented automatic merge by Gerrit: includes/resourceloader/ResourceLoaderUserOptionsModule.php Change-Id: I2dec8b8caf9db3c64919763865cc10cccdd6a1a3
2012-03-22 19:52:37 +00:00
$encText = str_replace( '*/', '* /', $text );
return "/*\n$encText\n*/\n";
}
/**
* Handle exception display.
*
* @param Exception $e Exception to be shown to the user
* @return string Sanitized text in a CSS/JS comment that can be returned to the user
*/
public static function formatException( $e ) {
return self::makeComment( self::formatExceptionNoComment( $e ) );
}
/**
* Handle exception display.
*
* @since 1.25
* @param Exception $e Exception to be shown to the user
* @return string Sanitized text that can be returned to the user
*/
protected static function formatExceptionNoComment( $e ) {
global $wgShowExceptionDetails;
if ( !$wgShowExceptionDetails ) {
return 'Internal error';
}
return $e->__toString();
}
/**
* Generate code for a response.
*
* @param ResourceLoaderContext $context Context in which to generate a response
* @param array $modules List of module objects keyed by module name
* @param array $missing List of requested module names that are unregistered (optional)
* @return string Response data
*/
public function makeModuleResponse( ResourceLoaderContext $context,
array $modules, array $missing = array()
) {
$out = '';
$states = array();
if ( !count( $modules ) && !count( $missing ) ) {
return <<<MESSAGE
/* This file is the Web entry point for MediaWiki's ResourceLoader:
<https://www.mediawiki.org/wiki/ResourceLoader>. In this request,
no modules were requested. Max made me put this here. */
MESSAGE;
}
$image = $context->getImageObj();
if ( $image ) {
$data = $image->getImageData( $context );
if ( $data === false ) {
$data = '';
$this->errors[] = 'Image generation failed';
}
return $data;
}
// Pre-fetch blobs
if ( $context->shouldIncludeMessages() ) {
try {
$this->blobStore->get( $this, $modules, $context->getLanguage() );
} catch ( Exception $e ) {
MWExceptionHandler::logException( $e );
$this->logger->warning( 'Prefetching MessageBlobStore failed: {exception}', array(
'exception' => $e
) );
$this->errors[] = self::formatExceptionNoComment( $e );
}
}
2010-09-04 12:53:01 +00:00
foreach ( $missing as $name ) {
$states[$name] = 'missing';
}
// Generate output
$isRaw = false;
foreach ( $modules as $name => $module ) {
try {
$content = $module->getModuleContent( $context );
// Append output
switch ( $context->getOnly() ) {
case 'scripts':
$scripts = $content['scripts'];
if ( is_string( $scripts ) ) {
// Load scripts raw...
$out .= $scripts;
} elseif ( is_array( $scripts ) ) {
// ...except when $scripts is an array of URLs
$out .= self::makeLoaderImplementScript( $name, $scripts, array(), array(), array() );
}
break;
case 'styles':
$styles = $content['styles'];
ResourceLoader: Refactor style loading Fixes: * bug 31676: Work around IE stylesheet limit. * bug 35562: @import styles broken in modules that combine multiple stylesheets. * bug 40498: Don't output empty "@media print { }" blocks. * bug 40500: Don't ignore media-type for urls in debug mode. Approach: * Re-use the same <style> tag so that we stay under the 31 stylesheet limit in IE. Unless the to-be-added css text from the being-loaded module contains @import, in which case we do create a new <style> tag and then re-use that one from that point on (bug 31676). * Return stylesheets as arrays, instead of a concatenated string. This fixes bug 35562, because @import only works when at the top of a stylesheet. By not unconditionally concatenating files within a module on the server side already, @import will work in e.g. module 'site' that contains 2 wiki pages. This is normalized in ResourceLoader::makeCombinedStyles(), so far only ResourceLoaderWikiModule makes use of this. Misc. clean up and bug fixes: * Reducing usage of jQuery() and mw.html.element() where native DOM would be very simple and faster. Aside from simplicity and speed, this is also working towards a more stand-alone ResourceLoader. * Trim server output a little bit more - Redundant new line after minify-css (it is now an array, so no need to keep space afterwards) - Redundant semi-colon after minify-js if it ends in a colon * Allow space in styleTest.css.php * Clean up and extend unit tests to cover for these features and bug fixes. * Don't set styleEl.rel = 'stylesheet'; that has no business on a <style> tag. * Fix bug in mw.loader's addStyleTag(). It turns out IE6 has an odd security measure that does not allow manipulation of elements (at least style tags) that are created by a different script (even if that script was served from the same domain/origin etc.). We didn't ran into this before because we only created new style tags, never appended to them. Now that we do, this came up. Took a while to figure out because it was created by mediawiki.js but it calls jQuery which did the actual dom insertion. Odd thing is, we load jquery.js and mediawiki.js in the same request even... Without this all css-url related mw.loader tests would fail in IE6. * mediawiki.js and mediawiki.test.js now pass jshint again. Tested (and passing qunit/?module=mediawiki; 123 of 123): * Chrome 14, 21 * Firefox 3.0, 3.6, 4, 7, 14, 15, 16beta * IE 6, 7, 8, 9 * Safari 4.0, 5.0, 5.1 * Opera 10.0, 11.1, 11.5, 11.6, 12.0, 12.5beta * iPhone 3GS / iOS 3.0 / Mobile Safari 4.0 iPhone 4 / iOS 4.0.1 / Mobile Safari 4.0.5 iPhone 4S / iOS 6.0 Beta / Mobile Safari 6.0 Change-Id: I3e8227ddb87fd9441071ca935439fc6467751dab
2012-07-25 21:20:21 +00:00
// We no longer seperate into media, they are all combined now with
// custom media type groups into @media .. {} sections as part of the css string.
// Module returns either an empty array or a numerical array with css strings.
$out .= isset( $styles['css'] ) ? implode( '', $styles['css'] ) : '';
break;
default:
$out .= self::makeLoaderImplementScript(
$name,
isset( $content['scripts'] ) ? $content['scripts'] : '',
isset( $content['styles'] ) ? $content['styles'] : array(),
isset( $content['messagesBlob'] ) ? new XmlJsCode( $content['messagesBlob'] ) : array(),
isset( $content['templates'] ) ? $content['templates'] : array()
);
break;
}
} catch ( Exception $e ) {
MWExceptionHandler::logException( $e );
$this->logger->warning( 'Generating module package failed: {exception}', array(
'exception' => $e
) );
$this->errors[] = self::formatExceptionNoComment( $e );
// Respond to client with error-state instead of module implementation
$states[$name] = 'error';
unset( $modules[$name] );
}
$isRaw |= $module->isRaw();
}
2010-09-04 12:53:01 +00:00
// Update module states
if ( $context->shouldIncludeScripts() && !$context->getRaw() && !$isRaw ) {
if ( count( $modules ) && $context->getOnly() === 'scripts' ) {
// Set the state of modules loaded as only scripts to ready as
// they don't have an mw.loader.implement wrapper that sets the state
foreach ( $modules as $name => $module ) {
$states[$name] = 'ready';
}
}
// Set the state of modules we didn't respond to with mw.loader.implement
if ( count( $states ) ) {
$out .= self::makeLoaderStateScript( $states );
}
} else {
if ( count( $states ) ) {
$this->errors[] = 'Problematic modules: ' .
FormatJson::encode( $states, ResourceLoader::inDebugMode() );
}
}
2010-09-04 12:53:01 +00:00
$enableFilterCache = true;
if ( count( $modules ) === 1 && reset( $modules ) instanceof ResourceLoaderUserTokensModule ) {
// If we're building the embedded user.tokens, don't cache (T84960)
$enableFilterCache = false;
}
if ( !$context->getDebug() ) {
if ( $context->getOnly() === 'styles' ) {
$out = $this->filter( 'minify-css', $out );
} else {
$out = $this->filter( 'minify-js', $out, array(
'cache' => $enableFilterCache
) );
}
}
return $out;
}
/* Static Methods */
/**
* Return JS code that calls mw.loader.implement with given module properties.
*
* @param string $name Module name
* @param mixed $scripts List of URLs to JavaScript files or String of JavaScript code
* @param mixed $styles Array of CSS strings keyed by media type, or an array of lists of URLs
* to CSS files keyed by media type
* @param mixed $messages List of messages associated with this module. May either be an
* associative array mapping message key to value, or a JSON-encoded message blob containing
* the same data, wrapped in an XmlJsCode object.
* @param array $templates Keys are name of templates and values are the source of
* the template.
* @throws MWException
* @return string
*/
public static function makeLoaderImplementScript(
$name, $scripts, $styles, $messages, $templates
) {
if ( is_string( $scripts ) ) {
// Site and user module are a legacy scripts that run in the global scope (no closure).
// Transportation as string instructs mw.loader.implement to use globalEval.
if ( $name === 'site' || $name === 'user' ) {
// Minify manually because the general makeModuleResponse() minification won't be
// effective here due to the script being a string instead of a function. (T107377)
if ( !ResourceLoader::inDebugMode() ) {
$scripts = self::applyFilter( 'minify-js', $scripts,
ConfigFactory::getDefaultInstance()->makeConfig( 'main' ) );
}
} else {
$scripts = new XmlJsCode( "function ( $, jQuery ) {\n{$scripts}\n}" );
}
} elseif ( !is_array( $scripts ) ) {
throw new MWException( 'Invalid scripts error. Array of URLs or string of code expected.' );
}
// mw.loader.implement requires 'styles', 'messages' and 'templates' to be objects (not
// arrays). json_encode considers empty arrays to be numerical and outputs "[]" instead
// of "{}". Force them to objects.
$module = array(
$name,
$scripts,
(object)$styles,
(object)$messages,
(object)$templates,
);
self::trimArray( $module );
return Xml::encodeJsCall( 'mw.loader.implement', $module, ResourceLoader::inDebugMode() );
}
/**
* Returns JS code which, when called, will register a given list of messages.
*
* @param mixed $messages Either an associative array mapping message key to value, or a
* JSON-encoded message blob containing the same data, wrapped in an XmlJsCode object.
* @return string
*/
public static function makeMessageSetScript( $messages ) {
return Xml::encodeJsCall(
'mw.messages.set',
array( (object)$messages ),
ResourceLoader::inDebugMode()
);
}
/**
* Combines an associative array mapping media type to CSS into a
* single stylesheet with "@media" blocks.
*
* @param array $stylePairs Array keyed by media type containing (arrays of) CSS strings
* @return array
*/
public static function makeCombinedStyles( array $stylePairs ) {
ResourceLoader: Refactor style loading Fixes: * bug 31676: Work around IE stylesheet limit. * bug 35562: @import styles broken in modules that combine multiple stylesheets. * bug 40498: Don't output empty "@media print { }" blocks. * bug 40500: Don't ignore media-type for urls in debug mode. Approach: * Re-use the same <style> tag so that we stay under the 31 stylesheet limit in IE. Unless the to-be-added css text from the being-loaded module contains @import, in which case we do create a new <style> tag and then re-use that one from that point on (bug 31676). * Return stylesheets as arrays, instead of a concatenated string. This fixes bug 35562, because @import only works when at the top of a stylesheet. By not unconditionally concatenating files within a module on the server side already, @import will work in e.g. module 'site' that contains 2 wiki pages. This is normalized in ResourceLoader::makeCombinedStyles(), so far only ResourceLoaderWikiModule makes use of this. Misc. clean up and bug fixes: * Reducing usage of jQuery() and mw.html.element() where native DOM would be very simple and faster. Aside from simplicity and speed, this is also working towards a more stand-alone ResourceLoader. * Trim server output a little bit more - Redundant new line after minify-css (it is now an array, so no need to keep space afterwards) - Redundant semi-colon after minify-js if it ends in a colon * Allow space in styleTest.css.php * Clean up and extend unit tests to cover for these features and bug fixes. * Don't set styleEl.rel = 'stylesheet'; that has no business on a <style> tag. * Fix bug in mw.loader's addStyleTag(). It turns out IE6 has an odd security measure that does not allow manipulation of elements (at least style tags) that are created by a different script (even if that script was served from the same domain/origin etc.). We didn't ran into this before because we only created new style tags, never appended to them. Now that we do, this came up. Took a while to figure out because it was created by mediawiki.js but it calls jQuery which did the actual dom insertion. Odd thing is, we load jquery.js and mediawiki.js in the same request even... Without this all css-url related mw.loader tests would fail in IE6. * mediawiki.js and mediawiki.test.js now pass jshint again. Tested (and passing qunit/?module=mediawiki; 123 of 123): * Chrome 14, 21 * Firefox 3.0, 3.6, 4, 7, 14, 15, 16beta * IE 6, 7, 8, 9 * Safari 4.0, 5.0, 5.1 * Opera 10.0, 11.1, 11.5, 11.6, 12.0, 12.5beta * iPhone 3GS / iOS 3.0 / Mobile Safari 4.0 iPhone 4 / iOS 4.0.1 / Mobile Safari 4.0.5 iPhone 4S / iOS 6.0 Beta / Mobile Safari 6.0 Change-Id: I3e8227ddb87fd9441071ca935439fc6467751dab
2012-07-25 21:20:21 +00:00
$out = array();
foreach ( $stylePairs as $media => $styles ) {
// ResourceLoaderFileModule::getStyle can return the styles
// as a string or an array of strings. This is to allow separation in
// the front-end.
$styles = (array)$styles;
ResourceLoader: Refactor style loading Fixes: * bug 31676: Work around IE stylesheet limit. * bug 35562: @import styles broken in modules that combine multiple stylesheets. * bug 40498: Don't output empty "@media print { }" blocks. * bug 40500: Don't ignore media-type for urls in debug mode. Approach: * Re-use the same <style> tag so that we stay under the 31 stylesheet limit in IE. Unless the to-be-added css text from the being-loaded module contains @import, in which case we do create a new <style> tag and then re-use that one from that point on (bug 31676). * Return stylesheets as arrays, instead of a concatenated string. This fixes bug 35562, because @import only works when at the top of a stylesheet. By not unconditionally concatenating files within a module on the server side already, @import will work in e.g. module 'site' that contains 2 wiki pages. This is normalized in ResourceLoader::makeCombinedStyles(), so far only ResourceLoaderWikiModule makes use of this. Misc. clean up and bug fixes: * Reducing usage of jQuery() and mw.html.element() where native DOM would be very simple and faster. Aside from simplicity and speed, this is also working towards a more stand-alone ResourceLoader. * Trim server output a little bit more - Redundant new line after minify-css (it is now an array, so no need to keep space afterwards) - Redundant semi-colon after minify-js if it ends in a colon * Allow space in styleTest.css.php * Clean up and extend unit tests to cover for these features and bug fixes. * Don't set styleEl.rel = 'stylesheet'; that has no business on a <style> tag. * Fix bug in mw.loader's addStyleTag(). It turns out IE6 has an odd security measure that does not allow manipulation of elements (at least style tags) that are created by a different script (even if that script was served from the same domain/origin etc.). We didn't ran into this before because we only created new style tags, never appended to them. Now that we do, this came up. Took a while to figure out because it was created by mediawiki.js but it calls jQuery which did the actual dom insertion. Odd thing is, we load jquery.js and mediawiki.js in the same request even... Without this all css-url related mw.loader tests would fail in IE6. * mediawiki.js and mediawiki.test.js now pass jshint again. Tested (and passing qunit/?module=mediawiki; 123 of 123): * Chrome 14, 21 * Firefox 3.0, 3.6, 4, 7, 14, 15, 16beta * IE 6, 7, 8, 9 * Safari 4.0, 5.0, 5.1 * Opera 10.0, 11.1, 11.5, 11.6, 12.0, 12.5beta * iPhone 3GS / iOS 3.0 / Mobile Safari 4.0 iPhone 4 / iOS 4.0.1 / Mobile Safari 4.0.5 iPhone 4S / iOS 6.0 Beta / Mobile Safari 6.0 Change-Id: I3e8227ddb87fd9441071ca935439fc6467751dab
2012-07-25 21:20:21 +00:00
foreach ( $styles as $style ) {
$style = trim( $style );
// Don't output an empty "@media print { }" block (bug 40498)
if ( $style !== '' ) {
// Transform the media type based on request params and config
// The way that this relies on $wgRequest to propagate request params is slightly evil
$media = OutputPage::transformCssMedia( $media );
if ( $media === '' || $media == 'all' ) {
$out[] = $style;
} elseif ( is_string( $media ) ) {
ResourceLoader: Refactor style loading Fixes: * bug 31676: Work around IE stylesheet limit. * bug 35562: @import styles broken in modules that combine multiple stylesheets. * bug 40498: Don't output empty "@media print { }" blocks. * bug 40500: Don't ignore media-type for urls in debug mode. Approach: * Re-use the same <style> tag so that we stay under the 31 stylesheet limit in IE. Unless the to-be-added css text from the being-loaded module contains @import, in which case we do create a new <style> tag and then re-use that one from that point on (bug 31676). * Return stylesheets as arrays, instead of a concatenated string. This fixes bug 35562, because @import only works when at the top of a stylesheet. By not unconditionally concatenating files within a module on the server side already, @import will work in e.g. module 'site' that contains 2 wiki pages. This is normalized in ResourceLoader::makeCombinedStyles(), so far only ResourceLoaderWikiModule makes use of this. Misc. clean up and bug fixes: * Reducing usage of jQuery() and mw.html.element() where native DOM would be very simple and faster. Aside from simplicity and speed, this is also working towards a more stand-alone ResourceLoader. * Trim server output a little bit more - Redundant new line after minify-css (it is now an array, so no need to keep space afterwards) - Redundant semi-colon after minify-js if it ends in a colon * Allow space in styleTest.css.php * Clean up and extend unit tests to cover for these features and bug fixes. * Don't set styleEl.rel = 'stylesheet'; that has no business on a <style> tag. * Fix bug in mw.loader's addStyleTag(). It turns out IE6 has an odd security measure that does not allow manipulation of elements (at least style tags) that are created by a different script (even if that script was served from the same domain/origin etc.). We didn't ran into this before because we only created new style tags, never appended to them. Now that we do, this came up. Took a while to figure out because it was created by mediawiki.js but it calls jQuery which did the actual dom insertion. Odd thing is, we load jquery.js and mediawiki.js in the same request even... Without this all css-url related mw.loader tests would fail in IE6. * mediawiki.js and mediawiki.test.js now pass jshint again. Tested (and passing qunit/?module=mediawiki; 123 of 123): * Chrome 14, 21 * Firefox 3.0, 3.6, 4, 7, 14, 15, 16beta * IE 6, 7, 8, 9 * Safari 4.0, 5.0, 5.1 * Opera 10.0, 11.1, 11.5, 11.6, 12.0, 12.5beta * iPhone 3GS / iOS 3.0 / Mobile Safari 4.0 iPhone 4 / iOS 4.0.1 / Mobile Safari 4.0.5 iPhone 4S / iOS 6.0 Beta / Mobile Safari 6.0 Change-Id: I3e8227ddb87fd9441071ca935439fc6467751dab
2012-07-25 21:20:21 +00:00
$out[] = "@media $media {\n" . str_replace( "\n", "\n\t", "\t" . $style ) . "}";
}
// else: skip
}
}
}
return $out;
}
/**
* Returns a JS call to mw.loader.state, which sets the state of a
* module or modules to a given value. Has two calling conventions:
*
* - ResourceLoader::makeLoaderStateScript( $name, $state ):
* Set the state of a single module called $name to $state
*
* - ResourceLoader::makeLoaderStateScript( array( $name => $state, ... ) ):
* Set the state of modules with the given names to the given states
*
* @param string $name
* @param string $state
* @return string
*/
public static function makeLoaderStateScript( $name, $state = null ) {
if ( is_array( $name ) ) {
return Xml::encodeJsCall(
'mw.loader.state',
array( $name ),
ResourceLoader::inDebugMode()
);
} else {
return Xml::encodeJsCall(
'mw.loader.state',
array( $name, $state ),
ResourceLoader::inDebugMode()
);
}
}
/**
* Returns JS code which calls the script given by $script. The script will
* be called with local variables name, version, dependencies and group,
* which will have values corresponding to $name, $version, $dependencies
* and $group as supplied.
*
* @param string $name Module name
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
* @param string $version Module version hash
* @param array $dependencies List of module names on which this module depends
* @param string $group Group which the module is in.
* @param string $source Source of the module, or 'local' if not foreign.
* @param string $script JavaScript code
* @return string
*/
public static function makeCustomLoaderScript( $name, $version, $dependencies,
$group, $source, $script
) {
$script = str_replace( "\n", "\n\t", trim( $script ) );
return Xml::encodeJsCall(
"( function ( name, version, dependencies, group, source ) {\n\t$script\n} )",
array( $name, $version, $dependencies, $group, $source ),
ResourceLoader::inDebugMode()
);
}
private static function isEmptyObject( stdClass $obj ) {
foreach ( $obj as $key => &$value ) {
return false;
}
return true;
}
/**
* Remove empty values from the end of an array.
*
* Values considered empty:
*
* - null
* - array()
* - new XmlJsCode( '{}' )
* - new stdClass() // (object) array()
*
* @param Array $array
*/
private static function trimArray( Array &$array ) {
$i = count( $array );
while ( $i-- ) {
if ( $array[$i] === null
|| $array[$i] === array()
|| ( $array[$i] instanceof XmlJsCode && $array[$i]->value === '{}' )
|| ( $array[$i] instanceof stdClass && self::isEmptyObject( $array[$i] ) )
) {
unset( $array[$i] );
} else {
break;
}
}
}
/**
* Returns JS code which calls mw.loader.register with the given
* parameters. Has three calling conventions:
*
* - ResourceLoader::makeLoaderRegisterScript( $name, $version,
* $dependencies, $group, $source, $skip
* ):
* Register a single module.
*
* - ResourceLoader::makeLoaderRegisterScript( array( $name1, $name2 ) ):
* Register modules with the given names.
*
* - ResourceLoader::makeLoaderRegisterScript( array(
resourceloader: Implement "skip function" feature A module can be registered with a skip function. Such function, if provided, will be invoked by the client when a module is queued for loading. If the function returns true, the client will bypass any further loading action and mark the module as 'ready'. This can be used to implement a feature test for a module providing a shim or polyfill. * Change visibility of method ResourceLoader::filter to public. So that it can be invoked by ResourceLoaderStartupModule. * Add option to suppress the cache key report in ResourceLoader::filter. We usually only call the minifier once on an entire request reponse (because it's all concatenated javascript or embedded javascript in various different closures, still valid as one large script) and only add a little bottom line for the cache key. When embedding the skip function we have to run the minifier on them separately as they're output as strings (not actual functions). These strings are typically quite small and blowing up the response with loads of cache keys is not desirable in production. * Add method to clear the static cache of ResourceLoader::inDebugMode. Global static state is evil but, as long as we have it, we at least need to clear it after switching contexts in the test suite. Also: * Remove obsolete setting of 'debug=true' in the FauxRequest in ResourceLoaderTestCase. It already sets global wgResourceLoaderDebug in the setUp() method. Bug: 66390 Change-Id: I87a0ea888d791ad39f114380c42e2daeca470961
2014-04-30 21:06:51 +00:00
* array( $name1, $version1, $dependencies1, $group1, $source1, $skip1 ),
* array( $name2, $version2, $dependencies1, $group2, $source2, $skip2 ),
* ...
* ) ):
* Registers modules with the given names and parameters.
*
* @param string $name Module name
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
* @param string $version Module version hash
* @param array $dependencies List of module names on which this module depends
* @param string $group Group which the module is in
* @param string $source Source of the module, or 'local' if not foreign
resourceloader: Implement "skip function" feature A module can be registered with a skip function. Such function, if provided, will be invoked by the client when a module is queued for loading. If the function returns true, the client will bypass any further loading action and mark the module as 'ready'. This can be used to implement a feature test for a module providing a shim or polyfill. * Change visibility of method ResourceLoader::filter to public. So that it can be invoked by ResourceLoaderStartupModule. * Add option to suppress the cache key report in ResourceLoader::filter. We usually only call the minifier once on an entire request reponse (because it's all concatenated javascript or embedded javascript in various different closures, still valid as one large script) and only add a little bottom line for the cache key. When embedding the skip function we have to run the minifier on them separately as they're output as strings (not actual functions). These strings are typically quite small and blowing up the response with loads of cache keys is not desirable in production. * Add method to clear the static cache of ResourceLoader::inDebugMode. Global static state is evil but, as long as we have it, we at least need to clear it after switching contexts in the test suite. Also: * Remove obsolete setting of 'debug=true' in the FauxRequest in ResourceLoaderTestCase. It already sets global wgResourceLoaderDebug in the setUp() method. Bug: 66390 Change-Id: I87a0ea888d791ad39f114380c42e2daeca470961
2014-04-30 21:06:51 +00:00
* @param string $skip Script body of the skip function
* @return string
*/
public static function makeLoaderRegisterScript( $name, $version = null,
resourceloader: Implement "skip function" feature A module can be registered with a skip function. Such function, if provided, will be invoked by the client when a module is queued for loading. If the function returns true, the client will bypass any further loading action and mark the module as 'ready'. This can be used to implement a feature test for a module providing a shim or polyfill. * Change visibility of method ResourceLoader::filter to public. So that it can be invoked by ResourceLoaderStartupModule. * Add option to suppress the cache key report in ResourceLoader::filter. We usually only call the minifier once on an entire request reponse (because it's all concatenated javascript or embedded javascript in various different closures, still valid as one large script) and only add a little bottom line for the cache key. When embedding the skip function we have to run the minifier on them separately as they're output as strings (not actual functions). These strings are typically quite small and blowing up the response with loads of cache keys is not desirable in production. * Add method to clear the static cache of ResourceLoader::inDebugMode. Global static state is evil but, as long as we have it, we at least need to clear it after switching contexts in the test suite. Also: * Remove obsolete setting of 'debug=true' in the FauxRequest in ResourceLoaderTestCase. It already sets global wgResourceLoaderDebug in the setUp() method. Bug: 66390 Change-Id: I87a0ea888d791ad39f114380c42e2daeca470961
2014-04-30 21:06:51 +00:00
$dependencies = null, $group = null, $source = null, $skip = null
) {
if ( is_array( $name ) ) {
// Build module name index
$index = array();
foreach ( $name as $i => &$module ) {
$index[$module[0]] = $i;
}
// Transform dependency names into indexes when possible, they will be resolved by
// mw.loader.register on the other end
foreach ( $name as &$module ) {
if ( isset( $module[2] ) ) {
foreach ( $module[2] as &$dependency ) {
if ( isset( $index[$dependency] ) ) {
$dependency = $index[$dependency];
}
}
}
}
array_walk( $name, array( 'self', 'trimArray' ) );
return Xml::encodeJsCall(
'mw.loader.register',
array( $name ),
ResourceLoader::inDebugMode()
);
} else {
$registration = array( $name, $version, $dependencies, $group, $source, $skip );
self::trimArray( $registration );
return Xml::encodeJsCall(
'mw.loader.register',
$registration,
ResourceLoader::inDebugMode()
);
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
}
}
/**
* Returns JS code which calls mw.loader.addSource() with the given
* parameters. Has two calling conventions:
*
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
* - ResourceLoader::makeLoaderSourcesScript( $id, $properties ):
* Register a single source
*
* - ResourceLoader::makeLoaderSourcesScript( array( $id1 => $loadUrl, $id2 => $loadUrl, ... ) );
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
* Register sources with the given IDs and properties.
*
* @param string $id Source ID
* @param array $properties Source properties (see addSource())
* @return string
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
*/
public static function makeLoaderSourcesScript( $id, $properties = null ) {
if ( is_array( $id ) ) {
return Xml::encodeJsCall(
'mw.loader.addSource',
array( $id ),
ResourceLoader::inDebugMode()
);
[ResourceLoader 2]: Add support for multiple loadScript sources Front-end: * New mw.loader method: addSource(). Call with two arguments or an object as first argument for multiple registrations * New property in module registry: "source". Optional for local modules (falls back to 'local'). When loading/using one or more modules, the worker will group the request by source and make separate requests to the sources as needed. * Re-arranging object properties in mw.loader.register to match the same order all other code parts use. * Adding documentation for 'source' and where missing updating it to include 'group' as well. * Refactor of mw.loader.work() by Roan Kattouw and Timo Tijhof:' -- Additional splitting layer by source (in addition to splitting by group), renamed 'groups' to 'splits' -- Clean up of the loop, and removing a no longer needed loop after the for-in-loop -- Much more function documentation in mw.loader.work() -- Moved caching of wgResourceLoaderMaxQueryLength out of the loop and renamed 'limit' to 'maxQueryLength Back-end changed provided through patch by Roan Kattouw (to avoid broken code between commits): * New method in ResourceLoader: addSource(). During construction of ResourceLoader this will be called by default for 'local' with loadScript property set to $wgLoadScript. Additional sources can be registered through $wgResourceLoaderSources (empty array by default) * Calling mw.loader.addSource from the startup module * Passing source to mw.loader.register from startup module * Some new static helper methods Use: * By default nothing should change in core, all modules simply default to 'local'. This info originates from the getSource()-method of the ResourceLoaderModule class, which is inherited to all core ResourceLoaderModule-implementations (none override it) * Third-party users and/or extensions can create new classes extending ResourceLoaderModule, re-implementing the getSource-method to return something else. Basic example: $wgResourceLoaderSources['mywiki'] = array( 'loadScript' => 'http://example.org/w/load.php' ); class MyCentralWikiModule extends ResourceLoaderModule { function getSource(){ return 'mywiki'; } } $wgResourceModules['cool.stuff'] => array( 'class' => 'MyCentralWikiModule' ); More complicated example // imagine some stuff with a ForeignGadgetRepo class, putting stuff in $wgResourceLoaderSources in the __construct() method class ForeignGadgetRepoGadget extends ResourceLoaderModule { function getSource(){ return $this->source; } } Loading: Loading is completely transparent, stuff like $wgOut->addModules() or mw.loader.loader/using both take it as any other module and load from the right source accordingly. -- This commit is part of the ResourceLoader 2 project.
2011-07-26 21:10:34 +00:00
} else {
return Xml::encodeJsCall(
'mw.loader.addSource',
array( $id, $properties ),
ResourceLoader::inDebugMode()
);
}
}
/**
* Returns JS code which runs given JS code if the client-side framework is
* present.
*
* @deprecated since 1.25; use makeInlineScript instead
* @param string $script JavaScript code
* @return string
*/
public static function makeLoaderConditionalScript( $script ) {
return "window.RLQ = window.RLQ || []; window.RLQ.push( function () {\n" . trim( $script ) . "\n} );";
}
/**
* Construct an inline script tag with given JS code.
*
* The code will be wrapped in a closure, and it will be executed by ResourceLoader
* only if the client has adequate support for MediaWiki JavaScript code.
*
* @param string $script JavaScript code
* @return WrappedString HTML
*/
public static function makeInlineScript( $script ) {
$js = self::makeLoaderConditionalScript( $script );
return new WrappedString(
Html::inlineScript( $js ),
"<script>window.RLQ = window.RLQ || []; window.RLQ.push( function () {\n",
"\n} );</script>"
);
}
/**
* Returns JS code which will set the MediaWiki configuration array to
* the given value.
*
* @param array $configuration List of configuration values keyed by variable name
* @return string
*/
public static function makeConfigSetScript( array $configuration ) {
if ( ResourceLoader::inDebugMode() ) {
return Xml::encodeJsCall( 'mw.config.set', array( $configuration ), true );
}
$config = RequestContext::getMain()->getConfig();
$js = Xml::encodeJsCall( 'mw.config.set', array( $configuration ), false );
return self::applyFilter( 'minify-js', $js, $config );
}
/**
* Convert an array of module names to a packed query string.
*
* For example, array( 'foo.bar', 'foo.baz', 'bar.baz', 'bar.quux' )
* becomes 'foo.bar,baz|bar.baz,quux'
* @param array $modules List of module names (strings)
* @return string Packed query string
*/
public static function makePackedModulesString( $modules ) {
$groups = array(); // array( prefix => array( suffixes ) )
foreach ( $modules as $module ) {
$pos = strrpos( $module, '.' );
$prefix = $pos === false ? '' : substr( $module, 0, $pos );
$suffix = $pos === false ? $module : substr( $module, $pos + 1 );
$groups[$prefix][] = $suffix;
}
$arr = array();
foreach ( $groups as $prefix => $suffixes ) {
$p = $prefix === '' ? '' : $prefix . '.';
$arr[] = $p . implode( ',', $suffixes );
}
$str = implode( '|', $arr );
return $str;
}
/**
* Determine whether debug mode was requested
* Order of priority is 1) request param, 2) cookie, 3) $wg setting
* @return bool
*/
public static function inDebugMode() {
resourceloader: Implement "skip function" feature A module can be registered with a skip function. Such function, if provided, will be invoked by the client when a module is queued for loading. If the function returns true, the client will bypass any further loading action and mark the module as 'ready'. This can be used to implement a feature test for a module providing a shim or polyfill. * Change visibility of method ResourceLoader::filter to public. So that it can be invoked by ResourceLoaderStartupModule. * Add option to suppress the cache key report in ResourceLoader::filter. We usually only call the minifier once on an entire request reponse (because it's all concatenated javascript or embedded javascript in various different closures, still valid as one large script) and only add a little bottom line for the cache key. When embedding the skip function we have to run the minifier on them separately as they're output as strings (not actual functions). These strings are typically quite small and blowing up the response with loads of cache keys is not desirable in production. * Add method to clear the static cache of ResourceLoader::inDebugMode. Global static state is evil but, as long as we have it, we at least need to clear it after switching contexts in the test suite. Also: * Remove obsolete setting of 'debug=true' in the FauxRequest in ResourceLoaderTestCase. It already sets global wgResourceLoaderDebug in the setUp() method. Bug: 66390 Change-Id: I87a0ea888d791ad39f114380c42e2daeca470961
2014-04-30 21:06:51 +00:00
if ( self::$debugMode === null ) {
global $wgRequest, $wgResourceLoaderDebug;
self::$debugMode = $wgRequest->getFuzzyBool( 'debug',
$wgRequest->getCookie( 'resourceLoaderDebug', '', $wgResourceLoaderDebug )
);
}
resourceloader: Implement "skip function" feature A module can be registered with a skip function. Such function, if provided, will be invoked by the client when a module is queued for loading. If the function returns true, the client will bypass any further loading action and mark the module as 'ready'. This can be used to implement a feature test for a module providing a shim or polyfill. * Change visibility of method ResourceLoader::filter to public. So that it can be invoked by ResourceLoaderStartupModule. * Add option to suppress the cache key report in ResourceLoader::filter. We usually only call the minifier once on an entire request reponse (because it's all concatenated javascript or embedded javascript in various different closures, still valid as one large script) and only add a little bottom line for the cache key. When embedding the skip function we have to run the minifier on them separately as they're output as strings (not actual functions). These strings are typically quite small and blowing up the response with loads of cache keys is not desirable in production. * Add method to clear the static cache of ResourceLoader::inDebugMode. Global static state is evil but, as long as we have it, we at least need to clear it after switching contexts in the test suite. Also: * Remove obsolete setting of 'debug=true' in the FauxRequest in ResourceLoaderTestCase. It already sets global wgResourceLoaderDebug in the setUp() method. Bug: 66390 Change-Id: I87a0ea888d791ad39f114380c42e2daeca470961
2014-04-30 21:06:51 +00:00
return self::$debugMode;
}
/**
* Reset static members used for caching.
*
* Global state and $wgRequest are evil, but we're using it right
* now and sometimes we need to be able to force ResourceLoader to
* re-evaluate the context because it has changed (e.g. in the test suite).
*/
public static function clearCache() {
self::$debugMode = null;
}
/**
* Build a load.php URL
*
* @since 1.24
* @param string $source Name of the ResourceLoader source
* @param ResourceLoaderContext $context
* @param array $extraQuery
* @return string URL to load.php. May be protocol-relative if $wgLoadScript is, too.
*/
public function createLoaderURL( $source, ResourceLoaderContext $context,
$extraQuery = array()
) {
$query = self::createLoaderQuery( $context, $extraQuery );
$script = $this->getLoadScript( $source );
return wfAppendQuery( $script, $query );
}
/**
* Build a load.php URL
resourceloader: Replace timestamp system with version hashing Modules now track their version via getVersionHash() instead of getModifiedTime(). == Background == While some resources have observeable timestamps (e.g. files stored on disk), many other resources do not. E.g. config variables, and module definitions. For static file modules, one can e.g. revert one of more files in a module to a previous version and not affect the max timestamp. Wiki modules include pages only if they exist. The user module supports common.js and skin.js. By default neither exists. If a user has both, and then the less-recently modified one is deleted, the max-timestamp remains unchanged. For client-side caching, batch requests use "Math.max" on the relevant timestamps. Again, if a module changes but another module is more recent (e.g. out-of-order deployment, or out-of-order discovery), the change would not result in a cache miss. More scenarios can be found in the associated Phabricator tasks. == Version hash == Previously we virtually mapped these variables to a timestamp by storing the current time alongside a hash of the value in ObjectCache. Considering the number of possible request contexts (wikis * modules * users * skins * languages) this doesn't work well. It results in needless cache invalidation when the first time observation is purged due to LRU algorithms. It also has other minor bugs leading to fewer cache hits. All modules automatically get the benefits of version hashing with this change. The old getDefinitionMtime() and getHashMtime() have been replaced with dummies that return 1. These functions are often called from getModifiedTime() in subclasses. For backward-compatibility, their respective values (definition summary and hash) are now included in getVersionHash directly. As examples, the following modules have been updated to use getVersionHash directly. Other modules still work fine and can be updated later. * ResourceLoaderFileModule * ResourceLoaderEditToolbarModule * ResourceLoaderStartUpModule * ResourceLoaderWikiModule The presence of hashes in place of timestamps increases the startup module size on a default MediaWiki install from 4.4k to 5.8k (after gzip and minification). == ETag == Since timestamps are no longer tracked, we need a different way to implement caching for cache proxies (e.g. Varnish) and web browsers. Previously we used the Last-Modified header (in combination with Cache-Control and Expires). Instead of Last-Modified (and If-Modified-Since), we use ETag (and If-None-Match). Entity tags (new in HTTP/1.1) are much stricter than Last-Modified by default. They instruct browsers to allow usage of partial Range requests. Since our responses are dynamically generated, we need to use the Weak version of ETag. While this sounds bad, it's no different than Last-Modified. As reassured by RFC 2616 <http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.html#sec13.3.3> the specified behaviour behind Last-Modified follows the same "Weak" caching logic as Entity tags. It's just that entity tags are capable of a stricter mode (whereas Last-Modified is inherently weak). == File cache == If $wgUseFileCache is enabled, ResourceLoader uses ResourceFileCache to cache load.php responses. While the blind TTL handling (during the allowed expiry period) is still maxage/timestamp based, tryRespondNotModified() now requires the caller to know the expected ETag. For this to work, the FileCache handling had to be moved from the top of ResoureLoader::respond() to after the expected ETag is computed. This also allows us to remove the duplicate tryRespondNotModified() handling since that's is already handled by ResourceLoader::respond() meanwhile. == Misc == * Remove redundant modifiedTime cache in ResourceLoaderFileModule. * Change bugzilla references to Phabricator. * Centralised inclusion of wgCacheEpoch using getDefinitionSummary. Previously this logic was duplicated in each place the modified timestamp was used. * It's easy to forget calling the parent class in getDefinitionSummary(). Previously this method only tracked 'class' by default. As such, various extensions hardcoded that one value instead of calling the parent and extending the array. To better prevent this in the future, getVersionHash() now asserts that the '_cacheEpoch' property made it through. * tests: Don't use getDefinitionSummary() as an API. Fix ResourceLoaderWikiModuleTest to call getPages properly. * In tests, the default timestamp used to be 1388534400000 (which is the unix time of 20140101000000; the unit tests' CacheEpoch). The new version hash of these modules is "XyCC+PSK", which is the base64 encoded prefix of the SHA1 digest of: '{"_class":"ResourceLoaderTestModule","_cacheEpoch":"20140101000000"}' * Add sha1.js library for client-side hash generation. Compared various different implementations for code size (after minfication/gzip), and speed (when used for short hexidecimal strings). https://jsperf.com/sha1-implementations - CryptoJS <https://code.google.com/p/crypto-js/#SHA-1> (min+gzip: 2.5k) http://crypto-js.googlecode.com/svn/tags/3.1.2/build/rollups/sha1.js Chrome: 45k, Firefox: 89k, Safari: 92k - jsSHA <https://github.com/Caligatio/jsSHA> https://github.com/Caligatio/jsSHA/blob/3c1d4f2e/src/sha1.js (min+gzip: 1.8k) Chrome: 65k, Firefox: 53k, Safari: 69k - phpjs-sha1 <https://github.com/kvz/phpjs> (RL min+gzip: 0.8k) https://github.com/kvz/phpjs/blob/1eaab15d/functions/strings/sha1.js Chrome: 200k, Firefox: 280k, Safari: 78k Modern browsers implement the HTML5 Crypto API. However, this API is asynchronous, only enabled when on HTTPS in Chromium, and is quite low-level. It requires boilerplate code to actually use with TextEncoder, ArrayBuffer and Uint32Array. Due this being needed in the module loader, we'd have to load the fallback regardless. Considering this is not used in a critical path for performance, it's not worth shipping two implementations for this optimisation. May also resolve: * T44094 * T90411 * T94810 Bug: T94074 Change-Id: Ibb292d2416839327d1807a66c78fd96dac0637d0
2015-04-29 22:53:24 +00:00
* @deprecated since 1.24 Use createLoaderURL() instead
* @param array $modules Array of module names (strings)
* @param string $lang Language code
* @param string $skin Skin name
* @param string|null $user User name. If null, the &user= parameter is omitted
* @param string|null $version Versioning timestamp
* @param bool $debug Whether the request should be in debug mode
* @param string|null $only &only= parameter
* @param bool $printable Printable mode
* @param bool $handheld Handheld mode
* @param array $extraQuery Extra query parameters to add
* @return string URL to load.php. May be protocol-relative if $wgLoadScript is, too.
*/
public static function makeLoaderURL( $modules, $lang, $skin, $user = null,
$version = null, $debug = false, $only = null, $printable = false,
$handheld = false, $extraQuery = array()
) {
global $wgLoadScript;
$query = self::makeLoaderQuery( $modules, $lang, $skin, $user, $version, $debug,
$only, $printable, $handheld, $extraQuery
);
return wfAppendQuery( $wgLoadScript, $query );
}
/**
* Helper for createLoaderURL()
*
* @since 1.24
* @see makeLoaderQuery
* @param ResourceLoaderContext $context
* @param array $extraQuery
* @return array
*/
public static function createLoaderQuery( ResourceLoaderContext $context, $extraQuery = array() ) {
return self::makeLoaderQuery(
$context->getModules(),
$context->getLanguage(),
$context->getSkin(),
$context->getUser(),
$context->getVersion(),
$context->getDebug(),
$context->getOnly(),
$context->getRequest()->getBool( 'printable' ),
$context->getRequest()->getBool( 'handheld' ),
$extraQuery
);
}
/**
* Build a query array (array representation of query string) for load.php. Helper
* function for makeLoaderURL().
*
* @param array $modules
* @param string $lang
* @param string $skin
* @param string $user
* @param string $version
* @param bool $debug
* @param string $only
* @param bool $printable
* @param bool $handheld
* @param array $extraQuery
*
* @return array
*/
public static function makeLoaderQuery( $modules, $lang, $skin, $user = null,
$version = null, $debug = false, $only = null, $printable = false,
$handheld = false, $extraQuery = array()
) {
$query = array(
'modules' => self::makePackedModulesString( $modules ),
'lang' => $lang,
'skin' => $skin,
'debug' => $debug ? 'true' : 'false',
);
if ( $user !== null ) {
$query['user'] = $user;
}
if ( $version !== null ) {
$query['version'] = $version;
}
if ( $only !== null ) {
$query['only'] = $only;
}
if ( $printable ) {
$query['printable'] = 1;
}
if ( $handheld ) {
$query['handheld'] = 1;
}
$query += $extraQuery;
// Make queries uniform in order
ksort( $query );
return $query;
}
/**
* Check a module name for validity.
*
* Module names may not contain pipes (|), commas (,) or exclamation marks (!) and can be
* at most 255 bytes.
*
* @param string $moduleName Module name to check
* @return bool Whether $moduleName is a valid module name
*/
public static function isValidModuleName( $moduleName ) {
return !preg_match( '/[|,!]/', $moduleName ) && strlen( $moduleName ) <= 255;
}
/**
* Returns LESS compiler set up for use with MediaWiki
*
* @param Config $config
* @throws MWException
* @since 1.22
* @return lessc
*/
public static function getLessCompiler( Config $config ) {
// When called from the installer, it is possible that a required PHP extension
// is missing (at least for now; see bug 47564). If this is the case, throw an
// exception (caught by the installer) to prevent a fatal error later on.
if ( !class_exists( 'lessc' ) ) {
throw new MWException( 'MediaWiki requires the lessphp compiler' );
}
if ( !function_exists( 'ctype_digit' ) ) {
throw new MWException( 'lessc requires the Ctype extension' );
}
$less = new lessc();
$less->setPreserveComments( true );
$less->setVariables( self::getLessVars( $config ) );
$less->setImportDir( $config->get( 'ResourceLoaderLESSImportPaths' ) );
return $less;
}
/**
* Get global LESS variables.
*
* @param Config $config
* @since 1.22
* @return array Map of variable names to string CSS values.
*/
public static function getLessVars( Config $config ) {
if ( !self::$lessVars ) {
$lessVars = $config->get( 'ResourceLoaderLESSVars' );
Hooks::run( 'ResourceLoaderGetLessVars', array( &$lessVars ) );
// Sort by key to ensure consistent hashing for cache lookups.
ksort( $lessVars );
self::$lessVars = $lessVars;
}
return self::$lessVars;
}
}