Changeset 62233 in webkit

Timestamp:

Jul 1, 2010 12:39:52 AM (14 years ago)

Author:

abarth@webkit.org

Message:

2010-07-01 Adam Barth <​abarth@webkit.org>

Reviewed by Eric Seidel.

BogusCommentState should come in from the cold
​https://bugs.webkit.org/show_bug.cgi?id=41439

Test progression => :)

• html5lib/runner-expected.txt:
• html5lib/webkit-resumer-expected.txt:

2010-07-01 Adam Barth <​abarth@webkit.org>

Reviewed by Eric Seidel.

BogusCommentState should come in from the cold
​https://bugs.webkit.org/show_bug.cgi?id=41439

The BogusCommentState has always been wrong. The proximate issue is
that it didn't handle resuming correctly when parsing a partial input
stream. Now that we have EOF working properly, we can actually
implement this state correctly.

We need to distinguish when we enter this state from when we continue
in this state. We could do that with a branch for each character, but
it seemed easier to split the state in two, even though that leaves us
with one more state in our tokenizer than we have in the HTML5 spec.

• html/HTMLTokenizer.cpp: (WebCore::HTMLTokenizer::nextToken):
• html/HTMLTokenizer.h: (WebCore::HTMLTokenizer::):

Location:

trunk

Files:

• LayoutTests/ChangeLog (modified) (1 diff)
• LayoutTests/html5lib/runner-expected.txt (modified) (1 diff)
• LayoutTests/html5lib/webkit-resumer-expected.txt (modified) (1 diff)
• WebCore/ChangeLog (modified) (1 diff)
• WebCore/html/HTMLTokenizer.cpp (modified) (1 diff)
• WebCore/html/HTMLTokenizer.h (modified) (1 diff)

trunk/LayoutTests/ChangeLog

@@ +1 @@
+2010-07-01  Adam Barth  <abarth@webkit.org>
+
+        Reviewed by Eric Seidel.
+
+        BogusCommentState should come in from the cold
+        https://bugs.webkit.org/show_bug.cgi?id=41439
+
+        Test progression => :)
+
+        * html5lib/runner-expected.txt:
+        * html5lib/webkit-resumer-expected.txt:
+
 2010-07-01  Eric Seidel  <eric@webkit.org>
 

trunk/LayoutTests/html5lib/runner-expected.txt

@@ -10 +10 @@
 33
 34
-41
 50
 51

trunk/LayoutTests/html5lib/webkit-resumer-expected.txt

@@ -158 +158 @@
 CONSOLE MESSAGE: line 2: FOO<span>BAR</span>BAZ
 resources/webkit01.dat:
-519.3
-520.4
-521.5
-522.6
-523.7
 593.1
 594.2

trunk/WebCore/ChangeLog

@@ +1 @@
+2010-07-01  Adam Barth  <abarth@webkit.org>
+
+        Reviewed by Eric Seidel.
+
+        BogusCommentState should come in from the cold
+        https://bugs.webkit.org/show_bug.cgi?id=41439
+
+        The BogusCommentState has always been wrong.  The proximate issue is
+        that it didn't handle resuming correctly when parsing a partial input
+        stream.  Now that we have EOF working properly, we can actually
+        implement this state correctly.
+
+        We need to distinguish when we enter this state from when we continue
+        in this state.  We could do that with a branch for each character, but
+        it seemed easier to split the state in two, even though that leaves us
+        with one more state in our tokenizer than we have in the HTML5 spec.
+
+        * html/HTMLTokenizer.cpp:
+        (WebCore::HTMLTokenizer::nextToken):
+        * html/HTMLTokenizer.h:
+        (WebCore::HTMLTokenizer::):
+
 2010-06-30  Adam Barth  <abarth@webkit.org>
 

trunk/WebCore/html/HTMLTokenizer.cpp

@@ -1058 +1058 @@
 
     BEGIN_STATE(BogusCommentState) {
-        // FIXME: This state isn't correct because we'll terminate the
-        // comment early if we don't have the whole input stream available.
         m_token->beginComment();
-        while (!source.isEmpty()) {
-            cc = m_inputStreamPreprocessor.nextInputCharacter();
-            if (cc == '>')
-                return emitAndResumeIn(source, DataState);
+        RECONSUME_IN(ContinueBogusCommentState);
+    }
+    END_STATE()
+
+    BEGIN_STATE(ContinueBogusCommentState) {
+        if (cc == '>')
+            return emitAndResumeIn(source, DataState);
+        else if (cc == InputStreamPreprocessor::endOfFileMarker)
+            return emitAndReconsumeIn(source, DataState);
+        else {
             m_token->appendToComment(cc);
-            m_inputStreamPreprocessor.advance(source, m_lineNumber);
-            // We ignore the return value (which indicates that |source| is
-            // empty) because it's checked by the loop condition above.
-        }
-        m_state = DataState;
-        return true;
-        // FIXME: Handle EOF properly.
+            ADVANCE_TO(ContinueBogusCommentState);
+        }
     }
     END_STATE()

trunk/WebCore/html/HTMLTokenizer.h

@@ -84 +84 @@
         SelfClosingStartTagState,
         BogusCommentState,
+        // The ContinueBogusCommentState is not in the HTML5 spec, but we use
+        // it internally to keep track of whether we've started the bogus
+        // comment token yet.
+        ContinueBogusCommentState,
         MarkupDeclarationOpenState,
         CommentStartState,