Optimized scanner to avoid virtual calls for every character read.

src/scanner-base.cc

Index: src/scanner-base.cc
diff --git a/src/scanner-base.cc b/src/scanner-base.cc
index 9e58c4e6371d746fe4ef523701c14015390be70f..e23d5a3c4a1c500f926be2671bd0b49ab59f7607 100644
--- a/src/scanner-base.cc
+++ b/src/scanner-base.cc
@@ -35,12 +35,6 @@ namespace v8 {
 namespace internal {
 
 // ----------------------------------------------------------------------------
-// UTF16Buffer
-
-UTF16Buffer::UTF16Buffer()
-    : pos_(0), end_(kNoEndPosition) { }
-
-// ----------------------------------------------------------------------------
 // LiteralCollector
 
 LiteralCollector::LiteralCollector()
@@ -92,7 +86,7 @@ bool ScannerConstants::IsIdentifier(unibrow::CharacterStream* buffer) {
 // ----------------------------------------------------------------------------
 // Scanner
 
-Scanner::Scanner() : source_(NULL) {}
+Scanner::Scanner() { }
 
 
 uc32 Scanner::ScanHexEscape(uc32 c, int length) {
@@ -142,8 +136,7 @@ uc32 Scanner::ScanOctalEscape(uc32 c, int length) {
 // ----------------------------------------------------------------------------
 // JavaScriptScanner
 
-JavaScriptScanner::JavaScriptScanner()
-    : has_line_terminator_before_next_(false) {}
+JavaScriptScanner::JavaScriptScanner() : Scanner() {}
 
 
 Token::Value JavaScriptScanner::Next() {
@@ -503,12 +496,21 @@ void JavaScriptScanner::Scan() {
 
 
 void JavaScriptScanner::SeekForward(int pos) {
-  source_->SeekForward(pos - 1);
-  Advance();
-  // This function is only called to seek to the location
-  // of the end of a function (at the "}" token). It doesn't matter
-  // whether there was a line terminator in the part we skip.
-  has_line_terminator_before_next_ = false;
+  // After this call, we will have the token at the given position as
+  // the "next" token. The "current" token will be invalid.
+  if (pos == next_.location.beg_pos) return;
+  int current_pos = source_pos();
+  ASSERT_EQ(next_.location.end_pos, current_pos);
+  // Positions inside the lookahead token isn't supported.

[Erik Corry, 2010/12/07 12:27:30]

Positions ... isn't -> Positions ... are not

[Lasse Reichstein, 2010/12/07 14:05:54]

Done.

+  ASSERT(pos >= current_pos);
+  if (pos != current_pos) {
+    source_->SeekForward(pos - source_->pos());
+    Advance();
+    // This function is only called to seek to the location
+    // of the end of a function (at the "}" token). It doesn't matter
+    // whether there was a line terminator in the part we skip.
+    has_line_terminator_before_next_ = false;

[Erik Corry, 2010/12/07 12:27:30]

Can't we assert something here to back up the comment?

[Lasse Reichstein, 2010/12/07 14:05:54]

We can't know whether there is a newline before, since we skip reading the
characters.
The only option is to check that the next token isn't one that cares about a
newline before it. But the requirement is generally about newlines after a token
that cares, and not recognizable on the next token.

I'll be very specific and require the next token to be '}'.

+  }
   Scan();
 }