WebCore:
Reviewed by Darin.
Make page loads go fast.
http://bugs.webkit.org/show_bug.cgi?id=17480
- Implement speculative preloading. When a script load blocks the main parser, use a side
parser to pick up more resources.
- Implement per-host load queues, prioritize scripts and stylesheets over images.
Depending on content and network latency this may speed things up quite a bit.
* WebCore.xcodeproj/project.pbxproj:
* dom/Document.cpp:
(WebCore::Document::implicitClose):
Clear the preloads after laoding completes.
* html/HTMLLinkElement.cpp:
(WebCore::HTMLLinkElement::parseMappedAttribute):
(WebCore::HTMLLinkElement::tokenizeRelAttribute):
* html/HTMLLinkElement.h:
Make tokenizeRelAttribute() public static so it can be used from elsewhere.
Eliminate a pointless bitfield so I can get references.
* html/HTMLTokenizer.cpp:
(WebCore::HTMLTokenizer::scriptHandler):
(WebCore::HTMLTokenizer::scriptExecution):
(WebCore::HTMLTokenizer::write):
* html/HTMLTokenizer.h:
Spin up the preload scanner whenever a script load blocks the parser. One scanner tracks the end of
the document while temporary ones are created as needed to scan document.write() output.
* html/PreloadScanner.cpp: Added.
(WebCore::PreloadScanner::PreloadScanner):
(WebCore::PreloadScanner::~PreloadScanner):
(WebCore::PreloadScanner::begin):
(WebCore::PreloadScanner::end):
(WebCore::PreloadScanner::reset):
(WebCore::PreloadScanner::write):
(WebCore::isWhitespace):
(WebCore::PreloadScanner::clearLastCharacters):
(WebCore::PreloadScanner::rememberCharacter):
(WebCore::PreloadScanner::lastCharactersMatch):
(WebCore::legalEntityFor):
(WebCore::PreloadScanner::consumeEntity):
(WebCore::PreloadScanner::tokenize):
(WebCore::PreloadScanner::processAttribute):
(WebCore::PreloadScanner::emitCharacter):
(WebCore::PreloadScanner::tokenizeCSS):
(WebCore::PreloadScanner::emitTag):
(WebCore::PreloadScanner::emitCSSRule):
* html/PreloadScanner.h: Added.
(WebCore::PreloadScanner::inProgress):
(WebCore::PreloadScanner::):
HTML5 tokenization plus some glue code. Fake CSS parsing thrown in just for fun.
* loader/Cache.cpp:
(WebCore::Cache::pruneDeadResources):
Preloads have zero refcount, avoid kicking them out too early.
* loader/CachedResource.cpp:
(WebCore::CachedResource::CachedResource):
(WebCore::CachedResource::ref):
* loader/CachedResource.h:
(WebCore::CachedResource::):
(WebCore::CachedResource::preloadResult):
(WebCore::CachedResource::setRequestedFromNetworkingLayer):
(WebCore::CachedResource::canDelete):
(WebCore::CachedResource::isPreloaded):
(WebCore::CachedResource::increasePreloadCount):
(WebCore::CachedResource::decreasePreloadCount):
Keep track which resources are preloads. Avoid deleting them. Track
at which point of the loading preloads get utilized to enable some interesting
statistics.
* loader/DocLoader.cpp:
(WebCore::DocLoader::~DocLoader):
(WebCore::DocLoader::checkForReload):
(WebCore::DocLoader::registerPreload):
(WebCore::DocLoader::clearPreloads):
(WebCore::DocLoader::printPreloadStats):
* loader/DocLoader.h:
Ensure we utilize preloaded resources during reloads.
Keep a list of all preloads in the document. Clear the preloads after
parsing is complete. Some debug statistics.
* loader/DocumentLoader.cpp:
(WebCore::DocumentLoader::isLoadingInAPISense):
Avoid signaling that loading is complete too early.
* loader/loader.cpp:
(WebCore::Loader::Loader):
(WebCore::Loader::~Loader):
(WebCore::Loader::determinePriority):
(WebCore::Loader::load):
(WebCore::Loader::scheduleServePendingRequests):
(WebCore::Loader::requestTimerFired):
(WebCore::Loader::servePendingRequests):
(WebCore::Loader::cancelRequests):
(WebCore::Loader::Host::Host):
(WebCore::Loader::Host::~Host):
(WebCore::Loader::Host::addRequest):
(WebCore::Loader::Host::hasRequests):
(WebCore::Loader::Host::servePendingRequests):
(WebCore::Loader::Host::didFinishLoading):
(WebCore::Loader::Host::didFail):
(WebCore::Loader::Host::didReceiveResponse):
(WebCore::Loader::Host::didReceiveData):
(WebCore::Loader::Host::cancelPendingRequests):
(WebCore::Loader::Host::cancelRequests):
* loader/loader.h:
(WebCore::Loader::):
Distribute load requests to per-host priority queues. Limit the number of loads issued to the
networking layer so we have better changes of getting important requests through first.
Prioritize scripts > stylesheets > images.
WebKit/mac:
Reviewed by Darin.
* ForwardingHeaders/wtf/Deque.h: Added.
git-svn-id: http://svn.webkit.org/repository/webkit/trunk@31038 268f45cc-cd09-0410-ab3c-d52691b4dbfc
19 files changed
