/*%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Copyright (c) Non, Inc. 1997 -- All Rights Reserved PROJECT: C Compiler MODULE: Parser FILE: stdc.g AUTHOR: John D. Mitchell (john@non.net), Jul 12, 1997 REVISION HISTORY: Name Date Description ---- ---- ----------- JDM 97.07.12 Initial version. JTC 97.11.18 Declaration vs declarator & misc. hacking. JDM 97.11.20 Fixed: declaration vs funcDef, parenthesized expressions, declarator iteration, varargs recognition, empty source file recognition, and some typos. DESCRIPTION: This grammar supports the Standard C language. Note clearly that this grammar does *NOT* deal with preprocessor functionality (including things like trigraphs) Nor does this grammar deal with multi-byte characters nor strings containing multi-byte characters [these constructs are "exercises for the reader" as it were :-)]. Please refer to the ISO/ANSI C Language Standard if you believe this grammar to be in error. Please cite chapter and verse in any correspondence to the author to back up your claim. TODO: - typedefName is commented out, needs a symbol table to resolve ambiguity. - trees %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%*/ header { package com.jogamp.gluegen.cgram; import java.io.*; import antlr.CommonAST; import antlr.DumpASTVisitor; } class StdCParser extends Parser; options { k = 2; exportVocab = STDC; buildAST = true; ASTLabelType = "TNode"; // Copied following options from java grammar. codeGenMakeSwitchThreshold = 2; codeGenBitsetTestThreshold = 3; } { // Suppport C++-style single-line comments? public static boolean CPPComments = true; // access to symbol table public CSymbolTable symbolTable = new CSymbolTable(); // source for names to unnamed scopes protected int unnamedScopeCounter = 0; public boolean isTypedefName(String name) { boolean returnValue = false; TNode node = symbolTable.lookupNameInCurrentScope(name); for (; node != null; node = (TNode) node.getNextSibling() ) { if(node.getType() == LITERAL_typedef) { returnValue = true; break; } } return returnValue; } public String getAScopeName() { return "" + (unnamedScopeCounter++); } public void pushScope(String scopeName) { symbolTable.pushScope(scopeName); } public void popScope() { symbolTable.popScope(); } int traceDepth = 0; public void reportError(RecognitionException ex) { try { System.err.println("ANTLR Parsing Error: "+ex + " token name:" + tokenNames[LA(1)]); ex.printStackTrace(System.err); } catch (TokenStreamException e) { System.err.println("ANTLR Parsing Error: "+ex); ex.printStackTrace(System.err); } } public void reportError(String s) { System.err.println("ANTLR Parsing Error from String: " + s); } public void reportWarning(String s) { System.err.println("ANTLR Parsing Warning from String: " + s); } public void match(int t) throws MismatchedTokenException { boolean debugging = false; if ( debugging ) { for (int x=0; x0)?" [inputState.guessing "+ inputState.guessing + "]":"")); } } try { if ( LA(1)!=t ) { if ( debugging ){ for (int x=0; x "+rname+"; LA(1)==("+ tokenNames[LT(1).getType()] + ") " + LT(1).getText() + " [inputState.guessing "+ inputState.guessing + "]"); } catch (TokenStreamException e) { } } public void traceOut(String rname) { for (int x=0; x declaration | functionDef | asm_expr ; asm_expr : "asm"^ ("volatile")? LCURLY! expr RCURLY! SEMI! ; declaration { AST ds1 = null; } : ds:declSpecifiers { ds1 = astFactory.dupList(#ds); } ( initDeclList[ds1] )? SEMI! { ## = #( #[NDeclaration], ##); } ; declSpecifiers { int specCount=0; } : ( options { // this loop properly aborts when // it finds a non-typedefName ID MBZ warnWhenFollowAmbig = false; } : s:storageClassSpecifier | typeQualifier | ( "struct" | "union" | "enum" | typeSpecifier[specCount] )=> specCount = typeSpecifier[specCount] )+ ; storageClassSpecifier : "auto" | "register" | "typedef" | functionStorageClassSpecifier ; functionStorageClassSpecifier : "extern" | "static" ; typeQualifier : "const" | "volatile" ; typeSpecifier [int specCount] returns [int retSpecCount] { retSpecCount = specCount + 1; } : ( "void" | "char" | "short" | "int" | "long" | "float" | "double" | "signed" | "unsigned" | "int8_t" | "uint8_t" | "int16_t" | "uint16_t" | "__int32" | "int32_t" | "wchar_t" | "uint32_t" | "__int64" | "int64_t" | "uint64_t" | "ptrdiff_t" | "intptr_t" | "size_t" | "uintptr_t" | structOrUnionSpecifier | enumSpecifier | { specCount == 0 }? typedefName ) ; typedefName : { isTypedefName ( LT(1).getText() ) }? i:ID { ## = #(#[NTypedefName], #i); } ; structOrUnionSpecifier { String scopeName; } : sou:structOrUnion! ( ( ID LCURLY )=> i:ID l:LCURLY { scopeName = #sou.getText() + " " + #i.getText(); #l.setText(scopeName); pushScope(scopeName); } structDeclarationList { popScope();} RCURLY! | l1:LCURLY { scopeName = getAScopeName(); #l1.setText(scopeName); pushScope(scopeName); } structDeclarationList { popScope(); } RCURLY! | ID ) { ## = #( #sou, ## ); } ; structOrUnion : "struct" | "union" ; structDeclarationList : ( structDeclaration )+ ; structDeclaration : specifierQualifierList structDeclaratorList ( SEMI! )+ ; specifierQualifierList { int specCount = 0; } : ( options { // this loop properly aborts when // it finds a non-typedefName ID MBZ warnWhenFollowAmbig = false; } : ( "struct" | "union" | "enum" | typeSpecifier[specCount] )=> specCount = typeSpecifier[specCount] | typeQualifier )+ ; structDeclaratorList : structDeclarator ( COMMA! structDeclarator )* ; structDeclarator : ( COLON constExpr | declarator[false] ( COLON constExpr )? ) { ## = #( #[NStructDeclarator], ##); } ; enumSpecifier : "enum"^ ( ( ID LCURLY )=> i:ID LCURLY enumList[i.getText()] RCURLY! | LCURLY enumList["anonymous"] RCURLY! | ID ) ; enumList[String enumName] : enumerator[enumName] ( COMMA! enumerator[enumName] )* ; enumerator[String enumName] : i:ID { symbolTable.add( i.getText(), #( null, #[LITERAL_enum, "enum"], #[ ID, enumName] ) ); } (ASSIGN constExpr)? ; initDeclList[AST declarationSpecifiers] : initDecl[declarationSpecifiers] ( COMMA! initDecl[declarationSpecifiers] )* ; initDecl[AST declarationSpecifiers] { String declName = ""; } : declName = d:declarator[false] { AST ds1, d1; ds1 = astFactory.dupList(declarationSpecifiers); d1 = astFactory.dupList(#d); symbolTable.add(declName, #(null, ds1, d1) ); } ( ASSIGN initializer | COLON expr )? { ## = #( #[NInitDecl], ## ); } ; pointerGroup : ( STAR ( typeQualifier )* )+ { ## = #( #[NPointerGroup], ##); } ; idList : ID ( COMMA! ID )* ; initializer : ( assignExpr | LCURLY initializerList ( COMMA! )? RCURLY! ) { ## = #( #[NInitializer], ## ); } ; initializerList : initializer ( COMMA! initializer )* ; declarator[boolean isFunctionDefinition] returns [String declName] { declName = ""; } : ( pointerGroup )? ( id:ID { declName = id.getText(); } | LPAREN declName = declarator[false] RPAREN ) ( ! LPAREN { if (isFunctionDefinition) { pushScope(declName); } else { pushScope("!"+declName); } } ( (declSpecifiers)=> p:parameterTypeList { ## = #( null, ##, #( #[NParameterTypeList], #p ) ); } | (i:idList)? { ## = #( null, ##, #( #[NParameterTypeList], #i ) ); } ) { popScope(); } RPAREN | LBRACKET ( constExpr )? RBRACKET )* { ## = #( #[NDeclarator], ## ); } ; parameterTypeList : parameterDeclaration ( options { warnWhenFollowAmbig = false; } : COMMA! parameterDeclaration )* ( COMMA! VARARGS )? ; parameterDeclaration { String declName; } : ds:declSpecifiers ( ( declarator[false] )=> declName = d:declarator[false] { AST d2, ds2; d2 = astFactory.dupList(#d); ds2 = astFactory.dupList(#ds); symbolTable.add(declName, #(null, ds2, d2)); } | nonemptyAbstractDeclarator )? { ## = #( #[NParameterDeclaration], ## ); } ; /* JTC: * This handles both new and old style functions. * see declarator rule to see differences in parameters * and here (declaration SEMI)* is the param type decls for the * old style. may want to do some checking to check for illegal * combinations (but I assume all parsed code will be legal?) */ functionDef { String declName; } : ( (functionDeclSpecifiers)=> ds:functionDeclSpecifiers | //epsilon ) declName = d:declarator[true] { AST d2, ds2; d2 = astFactory.dupList(#d); ds2 = astFactory.dupList(#ds); symbolTable.add(declName, #(null, ds2, d2)); pushScope(declName); } ( declaration )* (VARARGS)? ( SEMI! )* { popScope(); } compoundStatement[declName] { ## = #( #[NFunctionDef], ## );} ; functionDeclSpecifiers { int specCount = 0; } : ( options { // this loop properly aborts when // it finds a non-typedefName ID MBZ warnWhenFollowAmbig = false; } : functionStorageClassSpecifier | typeQualifier | ( "struct" | "union" | "enum" | typeSpecifier[specCount] )=> specCount = typeSpecifier[specCount] )+ ; declarationList : ( options { // this loop properly aborts when // it finds a non-typedefName ID MBZ warnWhenFollowAmbig = false; } : ( declarationPredictor )=> declaration )+ ; declarationPredictor : (options { //only want to look at declaration if I don't see typedef warnWhenFollowAmbig = false; }: "typedef" | declaration ) ; compoundStatement[String scopeName] : LCURLY! { pushScope(scopeName); } ( ( declarationPredictor)=> declarationList )? ( statementList )? { popScope(); } RCURLY! { ## = #( #[NCompoundStatement, scopeName], ##); } ; statementList : ( statement )+ ; statement : SEMI // Empty statements | compoundStatement[getAScopeName()] // Group of statements | expr SEMI! { ## = #( #[NStatementExpr], ## ); } // Expressions // Iteration statements: | "while"^ LPAREN! expr RPAREN! statement | "do"^ statement "while"! LPAREN! expr RPAREN! SEMI! |! "for" LPAREN ( e1:expr )? SEMI ( e2:expr )? SEMI ( e3:expr )? RPAREN s:statement { if ( #e1 == null) { #e1 = (TNode) #[ NEmptyExpression ]; } if ( #e2 == null) { #e2 = (TNode) #[ NEmptyExpression ]; } if ( #e3 == null) { #e3 = (TNode) #[ NEmptyExpression ]; } ## = #( #[LITERAL_for, "for"], #e1, #e2, #e3, #s ); } // Jump statements: | "goto"^ ID SEMI! | "continue" SEMI! | "break" SEMI! | "return"^ ( expr )? SEMI! // Labeled statements: | ID COLON! (options {warnWhenFollowAmbig=false;}:statement)? { ## = #( #[NLabel], ## ); } | "case"^ constExpr COLON! statement | "default"^ COLON! statement // Selection statements: | "if"^ LPAREN! expr RPAREN! statement ( //standard if-else ambiguity options { warnWhenFollowAmbig = false; } : "else" statement )? | "switch"^ LPAREN! expr RPAREN! statement ; expr : assignExpr (options { /* MBZ: COMMA is ambiguous between comma expressions and argument lists. argExprList should get priority, and it does by being deeper in the expr rule tree and using (COMMA assignExpr)* */ warnWhenFollowAmbig = false; } : c:COMMA^ { #c.setType(NCommaExpr); } assignExpr )* ; assignExpr : conditionalExpr ( a:assignOperator! assignExpr { ## = #( #a, ## );} )? ; assignOperator : ASSIGN | DIV_ASSIGN | PLUS_ASSIGN | MINUS_ASSIGN | STAR_ASSIGN | MOD_ASSIGN | RSHIFT_ASSIGN | LSHIFT_ASSIGN | BAND_ASSIGN | BOR_ASSIGN | BXOR_ASSIGN ; conditionalExpr : logicalOrExpr ( QUESTION^ expr COLON! conditionalExpr )? ; constExpr : conditionalExpr ; logicalOrExpr : logicalAndExpr ( LOR^ logicalAndExpr )* ; logicalAndExpr : inclusiveOrExpr ( LAND^ inclusiveOrExpr )* ; inclusiveOrExpr : exclusiveOrExpr ( BOR^ exclusiveOrExpr )* ; exclusiveOrExpr : bitAndExpr ( BXOR^ bitAndExpr )* ; bitAndExpr : equalityExpr ( BAND^ equalityExpr )* ; equalityExpr : relationalExpr ( ( EQUAL^ | NOT_EQUAL^ ) relationalExpr )* ; relationalExpr : shiftExpr ( ( LT^ | LTE^ | GT^ | GTE^ ) shiftExpr )* ; shiftExpr : additiveExpr ( ( LSHIFT^ | RSHIFT^ ) additiveExpr )* ; additiveExpr : multExpr ( ( PLUS^ | MINUS^ ) multExpr )* ; multExpr : castExpr ( ( STAR^ | DIV^ | MOD^ ) castExpr )* ; castExpr : ( LPAREN typeName RPAREN )=> LPAREN! typeName RPAREN! ( castExpr ) { ## = #( #[NCast, "("], ## ); } | unaryExpr ; typeName : specifierQualifierList (nonemptyAbstractDeclarator)? ; nonemptyAbstractDeclarator : ( pointerGroup ( (LPAREN ( nonemptyAbstractDeclarator | parameterTypeList )? RPAREN) | (LBRACKET (expr)? RBRACKET) )* | ( (LPAREN ( nonemptyAbstractDeclarator | parameterTypeList )? RPAREN) | (LBRACKET (expr)? RBRACKET) )+ ) { ## = #( #[NNonemptyAbstractDeclarator], ## ); } ; /* JTC: LR rules: abstractDeclarator : nonemptyAbstractDeclarator | // null ; nonemptyAbstractDeclarator : LPAREN nonemptyAbstractDeclarator RPAREN | abstractDeclarator LPAREN RPAREN | abstractDeclarator (LBRACKET (expr)? RBRACKET) | STAR abstractDeclarator ; */ unaryExpr : postfixExpr | INC^ unaryExpr | DEC^ unaryExpr | u:unaryOperator castExpr { ## = #( #[NUnaryExpr], ## ); } | "sizeof"^ ( ( LPAREN typeName )=> LPAREN typeName RPAREN | unaryExpr ) ; unaryOperator : BAND | STAR | PLUS | MINUS | BNOT | LNOT ; postfixExpr : primaryExpr ( postfixSuffix {## = #( #[NPostfixExpr], ## );} )? ; postfixSuffix : ( PTR ID | DOT ID | functionCall | LBRACKET expr RBRACKET | INC | DEC )+ ; functionCall : LPAREN^ (a:argExprList)? RPAREN { ##.setType( NFunctionCallArgs ); } ; primaryExpr : ID | charConst | intConst | floatConst | stringConst // JTC: // ID should catch the enumerator // leaving it in gives ambiguous err // | enumerator | LPAREN! expr RPAREN! { ## = #( #[NExpressionGroup, "("], ## ); } ; argExprList : assignExpr ( COMMA! assignExpr )* ; protected charConst : CharLiteral ; protected stringConst : (StringLiteral)+ { ## = #(#[NStringSeq], ##); } ; protected intConst : IntOctalConst | LongOctalConst | UnsignedOctalConst | IntIntConst | LongIntConst | UnsignedIntConst | IntHexConst | LongHexConst | UnsignedHexConst ; protected floatConst : FloatDoubleConst | DoubleDoubleConst | LongDoubleConst ; dummy : NTypedefName | NInitDecl | NDeclarator | NStructDeclarator | NDeclaration | NCast | NPointerGroup | NExpressionGroup | NFunctionCallArgs | NNonemptyAbstractDeclarator | NInitializer | NStatementExpr | NEmptyExpression | NParameterTypeList | NFunctionDef | NCompoundStatement | NParameterDeclaration | NCommaExpr | NUnaryExpr | NLabel | NPostfixExpr | NRangeExpr | NStringSeq | NInitializerElementLabel | NLcurlyInitializer | NAsmAttribute | NGnuAsmExpr | NTypeMissing ; { import java.io.*; import antlr.*; } class StdCLexer extends Lexer; options { k = 3; exportVocab = STDC; testLiterals = false; } { LineObject lineObject = new LineObject(); String originalSource = ""; PreprocessorInfoChannel preprocessorInfoChannel = new PreprocessorInfoChannel(); int tokenNumber = 0; boolean countingTokens = true; int deferredLineCount = 0; public void setCountingTokens(boolean ct) { countingTokens = ct; if ( countingTokens ) { tokenNumber = 0; } else { tokenNumber = 1; } } public void setOriginalSource(String src) { originalSource = src; lineObject.setSource(src); } public void setSource(String src) { lineObject.setSource(src); } public PreprocessorInfoChannel getPreprocessorInfoChannel() { return preprocessorInfoChannel; } public void setPreprocessingDirective(String pre) { preprocessorInfoChannel.addLineForTokenNumber( pre, new Integer(tokenNumber) ); } public void addDefine(String name, String value) { } protected Token makeToken(int t) { if ( t != Token.SKIP && countingTokens) { tokenNumber++; } CToken tok = (CToken) super.makeToken(t); tok.setLine(lineObject.line); tok.setSource(lineObject.source); tok.setTokenNumber(tokenNumber); lineObject.line += deferredLineCount; deferredLineCount = 0; return tok; } public void deferredNewline() { deferredLineCount++; } public void newline() { lineObject.newline(); } } protected Vocabulary : '\3'..'\377' ; /* Operators: */ ASSIGN : '=' ; COLON : ':' ; COMMA : ',' ; QUESTION : '?' ; SEMI : ';' ; PTR : "->" ; // DOT & VARARGS are commented out since they are generated as part of // the Number rule below due to some bizarre lexical ambiguity shme. // DOT : '.' ; protected DOT:; // VARARGS : "..." ; protected VARARGS:; LPAREN : '(' ; RPAREN : ')' ; LBRACKET : '[' ; RBRACKET : ']' ; LCURLY : '{' ; RCURLY : '}' ; EQUAL : "==" ; NOT_EQUAL : "!=" ; LTE : "<=" ; LT : "<" ; GTE : ">=" ; GT : ">" ; DIV : '/' ; DIV_ASSIGN : "/=" ; PLUS : '+' ; PLUS_ASSIGN : "+=" ; INC : "++" ; MINUS : '-' ; MINUS_ASSIGN : "-=" ; DEC : "--" ; STAR : '*' ; STAR_ASSIGN : "*=" ; MOD : '%' ; MOD_ASSIGN : "%=" ; RSHIFT : ">>" ; RSHIFT_ASSIGN : ">>=" ; LSHIFT : "<<" ; LSHIFT_ASSIGN : "<<=" ; LAND : "&&" ; LNOT : '!' ; LOR : "||" ; BAND : '&' ; BAND_ASSIGN : "&=" ; BNOT : '~' ; BOR : '|' ; BOR_ASSIGN : "|=" ; BXOR : '^' ; BXOR_ASSIGN : "^=" ; Whitespace : ( ( '\003'..'\010' | '\t' | '\013' | '\f' | '\016'.. '\037' | '\177'..'\377' | ' ' ) | "\r\n" { newline(); } | ( '\n' | '\r' ) { newline(); } ) { _ttype = Token.SKIP; } ; Comment : "/*" ( { LA(2) != '/' }? '*' | "\r\n" { deferredNewline(); } | ( '\r' | '\n' ) { deferredNewline(); } | ~( '*'| '\r' | '\n' ) )* "*/" { _ttype = Token.SKIP; } ; CPPComment : "//" ( ~('\n') )* { _ttype = Token.SKIP; } ; protected NonWhitespace : (~('\r' | '\n'))* ; PREPROC_DIRECTIVE options { paraphrase = "a line directive"; } : '#' ( ( "line" || (( ' ' | '\t' | '\014')+ '0'..'9')) => LineDirective | ( (Space)* "define" (Space)* i:ID (Space)* (n:DefineExpr)? nw:NonWhitespace ("\r\n" | "\r" | "\n") ) { if (n != null) { // System.out.println("addDefine: #define " + i.getText() + " " + n.getText()+" @ "+lineObject.getSource()+":"+(lineObject.line+deferredLineCount)); addDefine(i.getText(), n.getText()); } else { setPreprocessingDirective("#define " + i.getText() + " " + nw.getText()); } deferredNewline(); } | (~'\n')* { setPreprocessingDirective(getText()); } ) { _ttype = Token.SKIP; } ; DefineExpr: ((LPAREN) (Space)* (DefineExpr2) (Space)* (RPAREN)) | (DefineExpr2) ; DefineExpr2: (Number) ((Space)* (LSHIFT | RSHIFT | PLUS | MINUS | STAR | DIV | MOD) (Space)* (DefineExpr))? ; protected Space: ( ' ' | '\t' | '\014') ; protected LineDirective { boolean oldCountingTokens = countingTokens; countingTokens = false; } : { lineObject = new LineObject(); deferredLineCount = 0; } ("line")? //this would be for if the directive started "#line", but not there for GNU directives (Space)+ n:Number { lineObject.setLine(Integer.parseInt(n.getText())); } (Space)+ ( fn:StringLiteral { try { final String newSource = fn.getText().substring(1,fn.getText().length()-1); // System.out.println("line: "+lineObject.getSource()+" -> "+newSource+", line "+(lineObject.line+deferredLineCount)); lineObject.setSource(newSource); } catch (StringIndexOutOfBoundsException e) { /*not possible*/ } } | fi:ID { lineObject.setSource(fi.getText()); } )? (Space)* ("1" { lineObject.setEnteringFile(true); } )? (Space)* ("2" { lineObject.setReturningToFile(true); } )? (Space)* ("3" { lineObject.setSystemHeader(true); } )? (Space)* ("4" { lineObject.setTreatAsC(true); } )? (~('\r' | '\n'))* ("\r\n" | "\r" | "\n") { preprocessorInfoChannel.addLineForTokenNumber(new LineObject(lineObject), new Integer(tokenNumber)); countingTokens = oldCountingTokens; } ; /* Literals: */ /* * Note that we do NOT handle tri-graphs nor multi-byte sequences. */ /* * Note that we can't have empty character constants (even though we * can have empty strings :-). */ CharLiteral : '\'' ( Escape | ~( '\'' ) ) '\'' ; /* * Can't have raw imbedded newlines in string constants. Strict reading of * the standard gives odd dichotomy between newlines & carriage returns. * Go figure. */ StringLiteral : '"' ( Escape | ( '\r' { deferredNewline(); } | '\n' { deferredNewline(); _ttype = BadStringLiteral; } | '\\' '\n' { deferredNewline(); } ) | ~( '"' | '\r' | '\n' | '\\' ) )* '"' ; protected BadStringLiteral : // Imaginary token. ; /* * Handle the various escape sequences. * * Note carefully that these numeric escape *sequences* are *not* of the * same form as the C language numeric *constants*. * * There is no such thing as a binary numeric escape sequence. * * Octal escape sequences are either 1, 2, or 3 octal digits exactly. * * There is no such thing as a decimal escape sequence. * * Hexadecimal escape sequences are begun with a leading \x and continue * until a non-hexadecimal character is found. * * No real handling of tri-graph sequences, yet. */ protected Escape : '\\' ( options{warnWhenFollowAmbig=false;}: 'a' | 'b' | 'f' | 'n' | 'r' | 't' | 'v' | '"' | '\'' | '\\' | '?' | ('0'..'3') ( options{warnWhenFollowAmbig=false;}: Digit ( options{warnWhenFollowAmbig=false;}: Digit )? )? | ('4'..'7') ( options{warnWhenFollowAmbig=false;}: Digit )? | 'x' ( options{warnWhenFollowAmbig=false;}: Digit | 'a'..'f' | 'A'..'F' )+ ) ; /* Numeric Constants: */ protected Digit : '0'..'9' ; protected LongSuffix : 'l' | 'L' ; protected UnsignedSuffix : 'u' | 'U' ; protected FloatSuffix : 'f' | 'F' ; protected Exponent : ( 'e' | 'E' ) ( '+' | '-' )? ( Digit )+ ; protected DoubleDoubleConst:; protected FloatDoubleConst:; protected LongDoubleConst:; protected IntOctalConst:; protected LongOctalConst:; protected UnsignedOctalConst:; protected IntIntConst:; protected LongIntConst:; protected UnsignedIntConst:; protected IntHexConst:; protected LongHexConst:; protected UnsignedHexConst:; Number : ( ( Digit )+ ( '.' | 'e' | 'E' ) )=> ( Digit )+ ( '.' ( Digit )* ( Exponent )? | Exponent ) { _ttype = DoubleDoubleConst; } ( FloatSuffix { _ttype = FloatDoubleConst; } | LongSuffix { _ttype = LongDoubleConst; } )? | ( "..." )=> "..." { _ttype = VARARGS; } | '.' { _ttype = DOT; } ( ( Digit )+ ( Exponent )? { _ttype = DoubleDoubleConst; } ( FloatSuffix { _ttype = FloatDoubleConst; } | LongSuffix { _ttype = LongDoubleConst; } )? )? | '0' ( '0'..'7' )* { _ttype = IntOctalConst; } ( LongSuffix { _ttype = LongOctalConst; } | UnsignedSuffix { _ttype = UnsignedOctalConst; } )? | '1'..'9' ( Digit )* { _ttype = IntIntConst; } ( LongSuffix { _ttype = LongIntConst; } | UnsignedSuffix { _ttype = UnsignedIntConst; } )? | '0' ( 'x' | 'X' ) ( 'a'..'f' | 'A'..'F' | Digit )+ { _ttype = IntHexConst; } ( LongSuffix { _ttype = LongHexConst; } | UnsignedSuffix { _ttype = UnsignedHexConst; } )? ; ID options { testLiterals = true; } : ( 'a'..'z' | 'A'..'Z' | '_' ) ( 'a'..'z' | 'A'..'Z' | '_' | '0'..'9' )* ;