2008-11-20 Sam Weinig <sam@webkit.org>

2008-11-20  Sam Weinig  <sam@webkit.org>

        Reviewed by Darin Adler.

        Patch for https://bugs.webkit.org/show_bug.cgi?id=22385
        <rdar://problem/6390179>
        Lazily reparse FunctionBodyNodes on first execution.

        - Saves 57MB on Membuster head.

        * bytecompiler/BytecodeGenerator.cpp:
        (JSC::BytecodeGenerator::generate): Remove vector shrinking since this is now
        handled by destroying the ScopeNodeData after generation.

        * parser/Grammar.y: Add alternate NoNode version of the grammar
        that does not create nodes.  This is used to lazily create FunctionBodyNodes
        on first execution.

        * parser/Lexer.cpp:
        (JSC::Lexer::setCode): Fix bug where on reparse, the Lexer was confused about
        what position and length meant. Position is the current position in the original
        data buffer (important for getting correct line/column information) and length
        the end offset in the original buffer.
        * parser/Lexer.h:
        (JSC::Lexer::sourceCode): Positions are relative to the beginning of the buffer.

        * parser/Nodes.cpp:
        (JSC::ScopeNodeData::ScopeNodeData): Move initialization of ScopeNode data here.
        (JSC::ScopeNode::ScopeNode): Add constructor that only sets the JSGlobalData
        for FunctionBodyNode stubs.
        (JSC::ScopeNode::~ScopeNode): Release m_children now that we don't inherit from
        BlockNode.
        (JSC::ScopeNode::releaseNodes): Ditto.
        (JSC::EvalNode::generateBytecode): Only shrink m_children, as we need to keep around
        the rest of the data.
        (JSC::FunctionBodyNode::FunctionBodyNode): Add constructor that only sets the
        JSGlobalData.
        (JSC::FunctionBodyNode::create): Ditto.
        (JSC::FunctionBodyNode::generateBytecode): If we don't have the data, do a reparse
        to construct it. Then after generation, destroy the data.
        (JSC::ProgramNode::generateBytecode): After generation, destroy the AST data.
        * parser/Nodes.h:
        (JSC::ExpressionNode::): Add isFuncExprNode for FunctionConstructor.
        (JSC::StatementNode::): Add isExprStatementNode for FunctionConstructor.
        (JSC::ExprStatementNode::): Ditto.
        (JSC::ExprStatementNode::expr): Add accessor for FunctionConstructor.
        (JSC::FuncExprNode::): Add isFuncExprNode for FunctionConstructor

        (JSC::ScopeNode::adoptData): Adopts a ScopeNodeData.
        (JSC::ScopeNode::data): Accessor for ScopeNodeData.
        (JSC::ScopeNode::destroyData): Deletes the ScopeNodeData.
        (JSC::ScopeNode::setFeatures): Added.
        (JSC::ScopeNode::varStack): Added assert.
        (JSC::ScopeNode::functionStack): Ditto.
        (JSC::ScopeNode::children): Ditto.
        (JSC::ScopeNode::neededConstants): Ditto.
        Factor m_varStack, m_functionStack, m_children and m_numConstants into ScopeNodeData.

        * parser/Parser.cpp:
        (JSC::Parser::reparse): Reparse the SourceCode in the FunctionBodyNode and set
        set up the ScopeNodeData for it.
        * parser/Parser.h:

        * parser/SourceCode.h:
        (JSC::SourceCode::endOffset): Added for use in the lexer.

        * runtime/FunctionConstructor.cpp:
        (JSC::getFunctionBody): Assuming a ProgramNode with one FunctionExpression in it,
        get the FunctionBodyNode.  Any issues signifies a parse failure in constructFunction.
        (JSC::constructFunction): Make parsing functions in the form new Function(""), easier
        by concatenating the strings together (with some glue) and parsing the function expression
        as a ProgramNode from which we can receive the FunctionBodyNode. This has the added benefit
        of not having special parsing code for the arguments and lazily constructing the
        FunctionBodyNode's AST on first execution.

        * runtime/Identifier.h:
        (JSC::operator!=): Added.

git-svn-id: https://svn.webkit.org/repository/webkit/trunk@38635 268f45cc-cd09-0410-ab3c-d52691b4dbfc