Changeset 43156 in webkit for trunk/JavaScriptCore/runtime/JSGlobalObjectFunctions.cpp

Timestamp:

May 3, 2009, 9:49:35 AM (16 years ago)

Author:

Darin Adler

Message:

2009-05-02 Darin Adler <Darin Adler>

Reviewed by Maciej Stachowiak.

Bug 25519: streamline lexer by handling BOMs differently
​https://bugs.webkit.org/show_bug.cgi?id=25519

Roughly 1% faster SunSpider.

• parser/Grammar.y: Tweak formatting a bit.

• parser/Lexer.cpp: (JSC::Lexer::Lexer): Remove unnnecessary initialization of data members that are set up by setCode. (JSC::Lexer::currentOffset): Added. Used where the old code would look at m_currentOffset. (JSC::Lexer::shift1): Replaces the old shift function. No longer does anything to handle BOM characters. (JSC::Lexer::shift2): Ditto. (JSC::Lexer::shift3): Ditto. (JSC::Lexer::shift4): Ditto. (JSC::Lexer::setCode): Updated for name change from yylineno to m_line. Removed now-unused m_eatNextIdentifier, m_stackToken, and m_restrKeyword. Replaced m_skipLF and m_skipCR with m_skipLineEnd. Replaced the old m_length with m_codeEnd and m_currentOffset with m_codeStart. Added code to scan for a BOM character and call copyCodeWithoutBOMs() if we find any. (JSC::Lexer::copyCodeWithoutBOMs): Added. (JSC::Lexer::nextLine): Updated for name change from yylineno to m_line. (JSC::Lexer::makeIdentifier): Moved up higher in the file. (JSC::Lexer::matchPunctuator): Moved up higher in the file and changed to use a switch statement instead of just if statements. (JSC::Lexer::isLineTerminator): Moved up higher in the file and changed to have fewer branches. (JSC::Lexer::lastTokenWasRestrKeyword): Added. This replaces the old m_restrKeyword boolean. (JSC::Lexer::isIdentStart): Moved up higher in the file. Changed to use fewer branches in the ASCII but not identifier case. (JSC::Lexer::isIdentPart): Ditto. (JSC::Lexer::singleEscape): Moved up higher in the file. (JSC::Lexer::convertOctal): Moved up higher in the file. (JSC::Lexer::convertHex): Moved up higher in the file. Changed to use toASCIIHexValue instead of rolling our own here. (JSC::Lexer::convertUnicode): Ditto. (JSC::Lexer::record8): Moved up higher in the file. (JSC::Lexer::record16): Moved up higher in the file. (JSC::Lexer::lex): Changed type of stringType to int. Replaced m_skipLF and m_skipCR with m_skipLineEnd, which requires fewer branches in the main lexer loop. Use currentOffset instead of m_currentOffset. Removed unneeded m_stackToken. Use isASCIIDigit instead of isDecimalDigit. Split out the two cases for InIdentifierOrKeyword and InIdentifier. Added special case tight loops for identifiers and other simple states. Removed a branch from the code that sets m_atLineStart to false using goto. Streamlined the number-handling code so we don't check for the same types twice for non-numeric cases and don't add a null to m_buffer8 when it's not being used. Removed m_eatNextIdentifier, which wasn't working anyway, and m_restrKeyword, which is redundant with m_lastToken. Set the m_delimited flag without using a branch. (JSC::Lexer::scanRegExp): Tweaked style a bit. (JSC::Lexer::clear): Clear m_codeWithoutBOMs so we don't use memory after parsing. Clear out UString objects in the more conventional way. (JSC::Lexer::sourceCode): Made this no-longer inline since it has more work to do in the case where we stripped BOMs.

• parser/Lexer.h: Renamed yylineno to m_lineNumber. Removed convertHex function, which is the same as toASCIIHexValue. Removed isHexDigit function, which is the same as isASCIIHedDigit. Replaced shift with four separate shift functions. Removed isWhiteSpace function that passes m_current, instead just passing m_current explicitly. Removed isOctalDigit, which is the same as isASCIIOctalDigit. Eliminated unused arguments from matchPunctuator. Added copyCoodeWithoutBOMs and currentOffset. Moved the makeIdentifier function out of the header. Added lastTokenWasRestrKeyword function. Added new constants for m_skipLineEnd. Removed unused yycolumn, m_restrKeyword, m_skipLF, m_skipCR, m_eatNextIdentifier, m_stackToken, m_position, m_length, m_currentOffset, m_nextOffset1, m_nextOffset2, m_nextOffset3. Added m_skipLineEnd, m_codeStart, m_codeEnd, and m_codeWithoutBOMs.

• parser/SourceProvider.h: Added hasBOMs function. In the future this can be used to tell the lexer about strings known not to have BOMs.

• runtime/JSGlobalObjectFunctions.cpp: (JSC::globalFuncUnescape): Changed to use isASCIIHexDigit.

• wtf/ASCIICType.h: Added using statements to match the design of the other WTF headers.

File:

• trunk/JavaScriptCore/runtime/JSGlobalObjectFunctions.cpp (modified) (1 diff)

trunk/JavaScriptCore/runtime/JSGlobalObjectFunctions.cpp

@@ -405 +405 @@
         UChar u;
         if (c[0] == '%' && k <= len - 6 && c[1] == 'u') {
-            if (Lexer::isHexDigit(c[2]) && Lexer::isHexDigit(c[3]) && Lexer::isHexDigit(c[4]) && Lexer::isHexDigit(c[5])) {
+            if (isASCIIHexDigit(c[2]) && isASCIIHexDigit(c[3]) && isASCIIHexDigit(c[4]) && isASCIIHexDigit(c[5])) {
                 u = Lexer::convertUnicode(c[2], c[3], c[4], c[5]);
                 c = &u;
                 k += 5;
             }
-        } else if (c[0] == '%' && k <= len - 3 && Lexer::isHexDigit(c[1]) && Lexer::isHexDigit(c[2])) {
+        } else if (c[0] == '%' && k <= len - 3 && isASCIIHexDigit(c[1]) && isASCIIHexDigit(c[2])) {
             u = UChar(Lexer::convertHex(c[1], c[2]));
             c = &u;