Changeset 144407 in webkit

Timestamp:

Feb 28, 2013 6:12:54 PM (11 years ago)

Author:

abarth@webkit.org

Message:

The threaded HTML parser shouldn't need to invalidate the speculation buffer on every document.write
​https://bugs.webkit.org/show_bug.cgi?id=111130

Reviewed by Eric Seidel.

Previously, the threaded HTML parser always invalidated its speculation
buffer when it received a document.write. That means we performed
poorly on web sites that contained document.write calls early in the
page.

This patch teaches the HTMLDocumentParser that we don't need to discard
the speculation buffer in the common case of starting and ending in the
DataState.

• html/parser/BackgroundHTMLParser.cpp:

(WebCore::BackgroundHTMLParser::sendTokensToMainThread):

• html/parser/HTMLDocumentParser.cpp:

(WebCore::HTMLDocumentParser::checkForSpeculationFailure):
(WebCore::HTMLDocumentParser::didFailSpeculation):

• html/parser/HTMLDocumentParser.h:

(WebCore):
(ParsedChunk):

Location:

trunk/Source/WebCore

Files:

• ChangeLog (modified) (1 diff)
• html/parser/BackgroundHTMLParser.cpp (modified) (1 diff)
• html/parser/HTMLDocumentParser.cpp (modified) (2 diffs)
• html/parser/HTMLDocumentParser.h (modified) (3 diffs)

trunk/Source/WebCore/ChangeLog

@@ +1 @@
+2013-02-28  Adam Barth  <abarth@webkit.org>
+
+        The threaded HTML parser shouldn't need to invalidate the speculation buffer on every document.write
+        https://bugs.webkit.org/show_bug.cgi?id=111130
+
+        Reviewed by Eric Seidel.
+
+        Previously, the threaded HTML parser always invalidated its speculation
+        buffer when it received a document.write.  That means we performed
+        poorly on web sites that contained document.write calls early in the
+        page.
+
+        This patch teaches the HTMLDocumentParser that we don't need to discard
+        the speculation buffer in the common case of starting and ending in the
+        DataState.
+
+        * html/parser/BackgroundHTMLParser.cpp:
+        (WebCore::BackgroundHTMLParser::sendTokensToMainThread):
+        * html/parser/HTMLDocumentParser.cpp:
+        (WebCore::HTMLDocumentParser::checkForSpeculationFailure):
+        (WebCore::HTMLDocumentParser::didFailSpeculation):
+        * html/parser/HTMLDocumentParser.h:
+        (WebCore):
+        (ParsedChunk):
+
 2013-02-28  Eberhard Graether  <egraether@google.com>
 

trunk/Source/WebCore/html/parser/BackgroundHTMLParser.cpp

@@ -280 +280 @@
     chunk->tokens = m_pendingTokens.release();
     chunk->preloads.swap(m_pendingPreloads);
+    chunk->tokenizerState = m_tokenizer->state();
     chunk->inputCheckpoint = m_input.createCheckpoint();
     chunk->preloadScannerCheckpoint = m_preloadScanner->createCheckpoint();

trunk/Source/WebCore/html/parser/HTMLDocumentParser.cpp

@@ -333 +333 @@
     if (!m_tokenizer)
         return;
-    // FIXME: If the tokenizer is in the same state as when we started this function,
-    // then we haven't necessarily failed our speculation.
+    if (!m_currentChunk)
+        return;
+    // Currently we're only smart enough to reuse the speculation buffer if the tokenizer
+    // both starts and ends in the DataState. That state is simplest because the HTMLToken
+    // is always in the Uninitialized state. We should consider whether we can reuse the
+    // speculation buffer in other states, but we'd likely need to do something more
+    // sophisticated with the HTMLToken.
+    if (m_currentChunk->tokenizerState == HTMLTokenizer::DataState && m_tokenizer->state() == HTMLTokenizer::DataState && m_input.current().isEmpty()) {
+        ASSERT(m_token->isUninitialized());
+        m_tokenizer.clear();
+        m_token.clear();
+        return;
+    }
     didFailSpeculation(m_token.release(), m_tokenizer.release());
 }
@@ -340 +351 @@
 void HTMLDocumentParser::didFailSpeculation(PassOwnPtr<HTMLToken> token, PassOwnPtr<HTMLTokenizer> tokenizer)
 {
-    if (!m_currentChunk)
-        return;
-
     m_weakFactory.revokeAll();
     m_speculations.clear();

trunk/Source/WebCore/html/parser/HTMLDocumentParser.h

@@ -37 +37 @@
 #include "HTMLSourceTracker.h"
 #include "HTMLToken.h"
+#include "HTMLTokenizer.h"
 #include "ScriptableDocumentParser.h"
 #include "SegmentedString.h"
@@ -55 +56 @@
 class HTMLDocument;
 class HTMLParserScheduler;
-class HTMLTokenizer;
 class HTMLScriptRunner;
 class HTMLTreeBuilder;
@@ -90 +90 @@
         OwnPtr<CompactHTMLTokenStream> tokens;
         PreloadRequestStream preloads;
+        HTMLTokenizer::State tokenizerState;
         HTMLInputCheckpoint inputCheckpoint;
         TokenPreloadScannerCheckpoint preloadScannerCheckpoint;