ParseExpr.cpp revision 5dcc6ce4a68b3bd4b89c5697c9728e1533e71e03
1//===--- ParseExpr.cpp - Expression Parsing -------------------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements the Expression parsing implementation. Expressions in 11// C99 basically consist of a bunch of binary operators with unary operators and 12// other random stuff at the leaves. 13// 14// In the C99 grammar, these unary operators bind tightest and are represented 15// as the 'cast-expression' production. Everything else is either a binary 16// operator (e.g. '/') or a ternary operator ("?:"). The unary leaves are 17// handled by ParseCastExpression, the higher level pieces are handled by 18// ParseBinaryExpression. 19// 20//===----------------------------------------------------------------------===// 21 22#include "clang/Parse/Parser.h" 23#include "clang/Basic/Diagnostic.h" 24#include "llvm/ADT/SmallVector.h" 25#include "llvm/ADT/SmallString.h" 26using namespace clang; 27 28/// PrecedenceLevels - These are precedences for the binary/ternary operators in 29/// the C99 grammar. These have been named to relate with the C99 grammar 30/// productions. Low precedences numbers bind more weakly than high numbers. 31namespace prec { 32 enum Level { 33 Unknown = 0, // Not binary operator. 34 Comma = 1, // , 35 Assignment = 2, // =, *=, /=, %=, +=, -=, <<=, >>=, &=, ^=, |= 36 Conditional = 3, // ? 37 LogicalOr = 4, // || 38 LogicalAnd = 5, // && 39 InclusiveOr = 6, // | 40 ExclusiveOr = 7, // ^ 41 And = 8, // & 42 Equality = 9, // ==, != 43 Relational = 10, // >=, <=, >, < 44 Shift = 11, // <<, >> 45 Additive = 12, // -, + 46 Multiplicative = 13 // *, /, % 47 }; 48} 49 50 51/// getBinOpPrecedence - Return the precedence of the specified binary operator 52/// token. This returns: 53/// 54static prec::Level getBinOpPrecedence(tok::TokenKind Kind) { 55 switch (Kind) { 56 default: return prec::Unknown; 57 case tok::comma: return prec::Comma; 58 case tok::equal: 59 case tok::starequal: 60 case tok::slashequal: 61 case tok::percentequal: 62 case tok::plusequal: 63 case tok::minusequal: 64 case tok::lesslessequal: 65 case tok::greatergreaterequal: 66 case tok::ampequal: 67 case tok::caretequal: 68 case tok::pipeequal: return prec::Assignment; 69 case tok::question: return prec::Conditional; 70 case tok::pipepipe: return prec::LogicalOr; 71 case tok::ampamp: return prec::LogicalAnd; 72 case tok::pipe: return prec::InclusiveOr; 73 case tok::caret: return prec::ExclusiveOr; 74 case tok::amp: return prec::And; 75 case tok::exclaimequal: 76 case tok::equalequal: return prec::Equality; 77 case tok::lessequal: 78 case tok::less: 79 case tok::greaterequal: 80 case tok::greater: return prec::Relational; 81 case tok::lessless: 82 case tok::greatergreater: return prec::Shift; 83 case tok::plus: 84 case tok::minus: return prec::Additive; 85 case tok::percent: 86 case tok::slash: 87 case tok::star: return prec::Multiplicative; 88 } 89} 90 91 92/// ParseExpression - Simple precedence-based parser for binary/ternary 93/// operators. 94/// 95/// Note: we diverge from the C99 grammar when parsing the assignment-expression 96/// production. C99 specifies that the LHS of an assignment operator should be 97/// parsed as a unary-expression, but consistency dictates that it be a 98/// conditional-expession. In practice, the important thing here is that the 99/// LHS of an assignment has to be an l-value, which productions between 100/// unary-expression and conditional-expression don't produce. Because we want 101/// consistency, we parse the LHS as a conditional-expression, then check for 102/// l-value-ness in semantic analysis stages. 103/// 104/// multiplicative-expression: [C99 6.5.5] 105/// cast-expression 106/// multiplicative-expression '*' cast-expression 107/// multiplicative-expression '/' cast-expression 108/// multiplicative-expression '%' cast-expression 109/// 110/// additive-expression: [C99 6.5.6] 111/// multiplicative-expression 112/// additive-expression '+' multiplicative-expression 113/// additive-expression '-' multiplicative-expression 114/// 115/// shift-expression: [C99 6.5.7] 116/// additive-expression 117/// shift-expression '<<' additive-expression 118/// shift-expression '>>' additive-expression 119/// 120/// relational-expression: [C99 6.5.8] 121/// shift-expression 122/// relational-expression '<' shift-expression 123/// relational-expression '>' shift-expression 124/// relational-expression '<=' shift-expression 125/// relational-expression '>=' shift-expression 126/// 127/// equality-expression: [C99 6.5.9] 128/// relational-expression 129/// equality-expression '==' relational-expression 130/// equality-expression '!=' relational-expression 131/// 132/// AND-expression: [C99 6.5.10] 133/// equality-expression 134/// AND-expression '&' equality-expression 135/// 136/// exclusive-OR-expression: [C99 6.5.11] 137/// AND-expression 138/// exclusive-OR-expression '^' AND-expression 139/// 140/// inclusive-OR-expression: [C99 6.5.12] 141/// exclusive-OR-expression 142/// inclusive-OR-expression '|' exclusive-OR-expression 143/// 144/// logical-AND-expression: [C99 6.5.13] 145/// inclusive-OR-expression 146/// logical-AND-expression '&&' inclusive-OR-expression 147/// 148/// logical-OR-expression: [C99 6.5.14] 149/// logical-AND-expression 150/// logical-OR-expression '||' logical-AND-expression 151/// 152/// conditional-expression: [C99 6.5.15] 153/// logical-OR-expression 154/// logical-OR-expression '?' expression ':' conditional-expression 155/// [GNU] logical-OR-expression '?' ':' conditional-expression 156/// 157/// assignment-expression: [C99 6.5.16] 158/// conditional-expression 159/// unary-expression assignment-operator assignment-expression 160/// [C++] throw-expression [C++ 15] 161/// 162/// assignment-operator: one of 163/// = *= /= %= += -= <<= >>= &= ^= |= 164/// 165/// expression: [C99 6.5.17] 166/// assignment-expression 167/// expression ',' assignment-expression 168/// 169Parser::ExprResult Parser::ParseExpression() { 170 if (Tok.is(tok::kw_throw)) 171 return ParseThrowExpression(); 172 173 ExprResult LHS = ParseCastExpression(false); 174 if (LHS.isInvalid) return LHS; 175 176 return ParseRHSOfBinaryExpression(LHS, prec::Comma); 177} 178 179/// This routine is called when the '@' is seen and consumed. 180/// Current token is an Identifier and is not a 'try'. This 181/// routine is necessary to disambiguate @try-statement from, 182/// for example, @encode-expression. 183/// 184Parser::ExprResult Parser::ParseExpressionWithLeadingAt(SourceLocation AtLoc) { 185 ExprResult LHS = ParseObjCAtExpression(AtLoc); 186 if (LHS.isInvalid) return LHS; 187 188 return ParseRHSOfBinaryExpression(LHS, prec::Comma); 189} 190 191/// ParseAssignmentExpression - Parse an expr that doesn't include commas. 192/// 193Parser::ExprResult Parser::ParseAssignmentExpression() { 194 if (Tok.is(tok::kw_throw)) 195 return ParseThrowExpression(); 196 197 ExprResult LHS = ParseCastExpression(false); 198 if (LHS.isInvalid) return LHS; 199 200 return ParseRHSOfBinaryExpression(LHS, prec::Assignment); 201} 202 203Parser::ExprResult Parser::ParseConstantExpression() { 204 ExprResult LHS = ParseCastExpression(false); 205 if (LHS.isInvalid) return LHS; 206 207 return ParseRHSOfBinaryExpression(LHS, prec::Conditional); 208} 209 210/// ParseExpressionWithLeadingIdentifier - This special purpose method is used 211/// in contexts where we have already consumed an identifier (which we saved in 212/// 'IdTok'), then discovered that the identifier was really the leading token 213/// of part of an expression. For example, in "A[1]+B", we consumed "A" (which 214/// is now in 'IdTok') and the current token is "[". 215Parser::ExprResult Parser:: 216ParseExpressionWithLeadingIdentifier(const Token &IdTok) { 217 // We know that 'IdTok' must correspond to this production: 218 // primary-expression: identifier 219 220 // Let the actions module handle the identifier. 221 ExprResult Res = Actions.ActOnIdentifierExpr(CurScope, IdTok.getLocation(), 222 *IdTok.getIdentifierInfo(), 223 Tok.is(tok::l_paren)); 224 225 // Because we have to parse an entire cast-expression before starting the 226 // ParseRHSOfBinaryExpression method (which parses any trailing binops), we 227 // need to handle the 'postfix-expression' rules. We do this by invoking 228 // ParsePostfixExpressionSuffix to consume any postfix-expression suffixes: 229 Res = ParsePostfixExpressionSuffix(Res); 230 if (Res.isInvalid) return Res; 231 232 // At this point, the "A[1]" part of "A[1]+B" has been consumed. Once this is 233 // done, we know we don't have to do anything for cast-expression, because the 234 // only non-postfix-expression production starts with a '(' token, and we know 235 // we have an identifier. As such, we can invoke ParseRHSOfBinaryExpression 236 // to consume any trailing operators (e.g. "+" in this example) and connected 237 // chunks of the expression. 238 return ParseRHSOfBinaryExpression(Res, prec::Comma); 239} 240 241/// ParseExpressionWithLeadingIdentifier - This special purpose method is used 242/// in contexts where we have already consumed an identifier (which we saved in 243/// 'IdTok'), then discovered that the identifier was really the leading token 244/// of part of an assignment-expression. For example, in "A[1]+B", we consumed 245/// "A" (which is now in 'IdTok') and the current token is "[". 246Parser::ExprResult Parser:: 247ParseAssignmentExprWithLeadingIdentifier(const Token &IdTok) { 248 // We know that 'IdTok' must correspond to this production: 249 // primary-expression: identifier 250 251 // Let the actions module handle the identifier. 252 ExprResult Res = Actions.ActOnIdentifierExpr(CurScope, IdTok.getLocation(), 253 *IdTok.getIdentifierInfo(), 254 Tok.is(tok::l_paren)); 255 256 // Because we have to parse an entire cast-expression before starting the 257 // ParseRHSOfBinaryExpression method (which parses any trailing binops), we 258 // need to handle the 'postfix-expression' rules. We do this by invoking 259 // ParsePostfixExpressionSuffix to consume any postfix-expression suffixes: 260 Res = ParsePostfixExpressionSuffix(Res); 261 if (Res.isInvalid) return Res; 262 263 // At this point, the "A[1]" part of "A[1]+B" has been consumed. Once this is 264 // done, we know we don't have to do anything for cast-expression, because the 265 // only non-postfix-expression production starts with a '(' token, and we know 266 // we have an identifier. As such, we can invoke ParseRHSOfBinaryExpression 267 // to consume any trailing operators (e.g. "+" in this example) and connected 268 // chunks of the expression. 269 return ParseRHSOfBinaryExpression(Res, prec::Assignment); 270} 271 272 273/// ParseRHSOfBinaryExpression - Parse a binary expression that starts with 274/// LHS and has a precedence of at least MinPrec. 275Parser::ExprResult 276Parser::ParseRHSOfBinaryExpression(ExprResult LHS, unsigned MinPrec) { 277 unsigned NextTokPrec = getBinOpPrecedence(Tok.getKind()); 278 SourceLocation ColonLoc; 279 280 while (1) { 281 // If this token has a lower precedence than we are allowed to parse (e.g. 282 // because we are called recursively, or because the token is not a binop), 283 // then we are done! 284 if (NextTokPrec < MinPrec) 285 return LHS; 286 287 // Consume the operator, saving the operator token for error reporting. 288 Token OpToken = Tok; 289 ConsumeToken(); 290 291 // Special case handling for the ternary operator. 292 ExprResult TernaryMiddle(true); 293 if (NextTokPrec == prec::Conditional) { 294 if (Tok.isNot(tok::colon)) { 295 // Handle this production specially: 296 // logical-OR-expression '?' expression ':' conditional-expression 297 // In particular, the RHS of the '?' is 'expression', not 298 // 'logical-OR-expression' as we might expect. 299 TernaryMiddle = ParseExpression(); 300 if (TernaryMiddle.isInvalid) { 301 Actions.DeleteExpr(LHS.Val); 302 return TernaryMiddle; 303 } 304 } else { 305 // Special case handling of "X ? Y : Z" where Y is empty: 306 // logical-OR-expression '?' ':' conditional-expression [GNU] 307 TernaryMiddle = ExprResult(false); 308 Diag(Tok, diag::ext_gnu_conditional_expr); 309 } 310 311 if (Tok.isNot(tok::colon)) { 312 Diag(Tok, diag::err_expected_colon); 313 Diag(OpToken, diag::err_matching, "?"); 314 Actions.DeleteExpr(LHS.Val); 315 Actions.DeleteExpr(TernaryMiddle.Val); 316 return ExprResult(true); 317 } 318 319 // Eat the colon. 320 ColonLoc = ConsumeToken(); 321 } 322 323 // Parse another leaf here for the RHS of the operator. 324 ExprResult RHS = ParseCastExpression(false); 325 if (RHS.isInvalid) { 326 Actions.DeleteExpr(LHS.Val); 327 Actions.DeleteExpr(TernaryMiddle.Val); 328 return RHS; 329 } 330 331 // Remember the precedence of this operator and get the precedence of the 332 // operator immediately to the right of the RHS. 333 unsigned ThisPrec = NextTokPrec; 334 NextTokPrec = getBinOpPrecedence(Tok.getKind()); 335 336 // Assignment and conditional expressions are right-associative. 337 bool isRightAssoc = ThisPrec == prec::Conditional || 338 ThisPrec == prec::Assignment; 339 340 // Get the precedence of the operator to the right of the RHS. If it binds 341 // more tightly with RHS than we do, evaluate it completely first. 342 if (ThisPrec < NextTokPrec || 343 (ThisPrec == NextTokPrec && isRightAssoc)) { 344 // If this is left-associative, only parse things on the RHS that bind 345 // more tightly than the current operator. If it is left-associative, it 346 // is okay, to bind exactly as tightly. For example, compile A=B=C=D as 347 // A=(B=(C=D)), where each paren is a level of recursion here. 348 RHS = ParseRHSOfBinaryExpression(RHS, ThisPrec + !isRightAssoc); 349 if (RHS.isInvalid) { 350 Actions.DeleteExpr(LHS.Val); 351 Actions.DeleteExpr(TernaryMiddle.Val); 352 return RHS; 353 } 354 355 NextTokPrec = getBinOpPrecedence(Tok.getKind()); 356 } 357 assert(NextTokPrec <= ThisPrec && "Recursion didn't work!"); 358 359 if (!LHS.isInvalid) { 360 // Combine the LHS and RHS into the LHS (e.g. build AST). 361 if (TernaryMiddle.isInvalid) 362 LHS = Actions.ActOnBinOp(OpToken.getLocation(), OpToken.getKind(), 363 LHS.Val, RHS.Val); 364 else 365 LHS = Actions.ActOnConditionalOp(OpToken.getLocation(), ColonLoc, 366 LHS.Val, TernaryMiddle.Val, RHS.Val); 367 } else { 368 // We had a semantic error on the LHS. Just free the RHS and continue. 369 Actions.DeleteExpr(TernaryMiddle.Val); 370 Actions.DeleteExpr(RHS.Val); 371 } 372 } 373} 374 375/// ParseCastExpression - Parse a cast-expression, or, if isUnaryExpression is 376/// true, parse a unary-expression. 377/// 378/// cast-expression: [C99 6.5.4] 379/// unary-expression 380/// '(' type-name ')' cast-expression 381/// 382/// unary-expression: [C99 6.5.3] 383/// postfix-expression 384/// '++' unary-expression 385/// '--' unary-expression 386/// unary-operator cast-expression 387/// 'sizeof' unary-expression 388/// 'sizeof' '(' type-name ')' 389/// [GNU] '__alignof' unary-expression 390/// [GNU] '__alignof' '(' type-name ')' 391/// [GNU] '&&' identifier 392/// 393/// unary-operator: one of 394/// '&' '*' '+' '-' '~' '!' 395/// [GNU] '__extension__' '__real' '__imag' 396/// 397/// primary-expression: [C99 6.5.1] 398/// identifier 399/// constant 400/// string-literal 401/// [C++] boolean-literal [C++ 2.13.5] 402/// '(' expression ')' 403/// '__func__' [C99 6.4.2.2] 404/// [GNU] '__FUNCTION__' 405/// [GNU] '__PRETTY_FUNCTION__' 406/// [GNU] '(' compound-statement ')' 407/// [GNU] '__builtin_va_arg' '(' assignment-expression ',' type-name ')' 408/// [GNU] '__builtin_offsetof' '(' type-name ',' offsetof-member-designator')' 409/// [GNU] '__builtin_choose_expr' '(' assign-expr ',' assign-expr ',' 410/// assign-expr ')' 411/// [GNU] '__builtin_types_compatible_p' '(' type-name ',' type-name ')' 412/// [OBJC] '[' objc-message-expr ']' 413/// [OBJC] '@selector' '(' objc-selector-arg ')' 414/// [OBJC] '@protocol' '(' identifier ')' 415/// [OBJC] '@encode' '(' type-name ')' 416/// [OBJC] objc-string-literal 417/// [C++] 'const_cast' '<' type-name '>' '(' expression ')' [C++ 5.2p1] 418/// [C++] 'dynamic_cast' '<' type-name '>' '(' expression ')' [C++ 5.2p1] 419/// [C++] 'reinterpret_cast' '<' type-name '>' '(' expression ')' [C++ 5.2p1] 420/// [C++] 'static_cast' '<' type-name '>' '(' expression ')' [C++ 5.2p1] 421/// 422/// constant: [C99 6.4.4] 423/// integer-constant 424/// floating-constant 425/// enumeration-constant -> identifier 426/// character-constant 427/// 428Parser::ExprResult Parser::ParseCastExpression(bool isUnaryExpression) { 429 ExprResult Res; 430 tok::TokenKind SavedKind = Tok.getKind(); 431 432 // This handles all of cast-expression, unary-expression, postfix-expression, 433 // and primary-expression. We handle them together like this for efficiency 434 // and to simplify handling of an expression starting with a '(' token: which 435 // may be one of a parenthesized expression, cast-expression, compound literal 436 // expression, or statement expression. 437 // 438 // If the parsed tokens consist of a primary-expression, the cases below 439 // call ParsePostfixExpressionSuffix to handle the postfix expression 440 // suffixes. Cases that cannot be followed by postfix exprs should 441 // return without invoking ParsePostfixExpressionSuffix. 442 switch (SavedKind) { 443 case tok::l_paren: { 444 // If this expression is limited to being a unary-expression, the parent can 445 // not start a cast expression. 446 ParenParseOption ParenExprType = 447 isUnaryExpression ? CompoundLiteral : CastExpr; 448 TypeTy *CastTy; 449 SourceLocation LParenLoc = Tok.getLocation(); 450 SourceLocation RParenLoc; 451 Res = ParseParenExpression(ParenExprType, CastTy, RParenLoc); 452 if (Res.isInvalid) return Res; 453 454 switch (ParenExprType) { 455 case SimpleExpr: break; // Nothing else to do. 456 case CompoundStmt: break; // Nothing else to do. 457 case CompoundLiteral: 458 // We parsed '(' type-name ')' '{' ... '}'. If any suffixes of 459 // postfix-expression exist, parse them now. 460 break; 461 case CastExpr: 462 // We parsed '(' type-name ')' and the thing after it wasn't a '{'. Parse 463 // the cast-expression that follows it next. 464 // TODO: For cast expression with CastTy. 465 Res = ParseCastExpression(false); 466 if (!Res.isInvalid) 467 Res = Actions.ActOnCastExpr(LParenLoc, CastTy, RParenLoc, Res.Val); 468 return Res; 469 } 470 471 // These can be followed by postfix-expr pieces. 472 return ParsePostfixExpressionSuffix(Res); 473 } 474 475 // primary-expression 476 case tok::numeric_constant: 477 // constant: integer-constant 478 // constant: floating-constant 479 480 Res = Actions.ActOnNumericConstant(Tok); 481 ConsumeToken(); 482 483 // These can be followed by postfix-expr pieces. 484 return ParsePostfixExpressionSuffix(Res); 485 486 case tok::kw_true: 487 case tok::kw_false: 488 return ParseCXXBoolLiteral(); 489 490 case tok::identifier: { // primary-expression: identifier 491 // constant: enumeration-constant 492 // Consume the identifier so that we can see if it is followed by a '('. 493 // Function designators are allowed to be undeclared (C99 6.5.1p2), so we 494 // need to know whether or not this identifier is a function designator or 495 // not. 496 IdentifierInfo &II = *Tok.getIdentifierInfo(); 497 SourceLocation L = ConsumeToken(); 498 Res = Actions.ActOnIdentifierExpr(CurScope, L, II, Tok.is(tok::l_paren)); 499 // These can be followed by postfix-expr pieces. 500 return ParsePostfixExpressionSuffix(Res); 501 } 502 case tok::char_constant: // constant: character-constant 503 Res = Actions.ActOnCharacterConstant(Tok); 504 ConsumeToken(); 505 // These can be followed by postfix-expr pieces. 506 return ParsePostfixExpressionSuffix(Res); 507 case tok::kw___func__: // primary-expression: __func__ [C99 6.4.2.2] 508 case tok::kw___FUNCTION__: // primary-expression: __FUNCTION__ [GNU] 509 case tok::kw___PRETTY_FUNCTION__: // primary-expression: __P..Y_F..N__ [GNU] 510 Res = Actions.ActOnPreDefinedExpr(Tok.getLocation(), SavedKind); 511 ConsumeToken(); 512 // These can be followed by postfix-expr pieces. 513 return ParsePostfixExpressionSuffix(Res); 514 case tok::string_literal: // primary-expression: string-literal 515 case tok::wide_string_literal: 516 Res = ParseStringLiteralExpression(); 517 if (Res.isInvalid) return Res; 518 // This can be followed by postfix-expr pieces (e.g. "foo"[1]). 519 return ParsePostfixExpressionSuffix(Res); 520 case tok::kw___builtin_va_arg: 521 case tok::kw___builtin_offsetof: 522 case tok::kw___builtin_choose_expr: 523 case tok::kw___builtin_overload: 524 case tok::kw___builtin_types_compatible_p: 525 return ParseBuiltinPrimaryExpression(); 526 case tok::plusplus: // unary-expression: '++' unary-expression 527 case tok::minusminus: { // unary-expression: '--' unary-expression 528 SourceLocation SavedLoc = ConsumeToken(); 529 Res = ParseCastExpression(true); 530 if (!Res.isInvalid) 531 Res = Actions.ActOnUnaryOp(SavedLoc, SavedKind, Res.Val); 532 return Res; 533 } 534 case tok::amp: // unary-expression: '&' cast-expression 535 case tok::star: // unary-expression: '*' cast-expression 536 case tok::plus: // unary-expression: '+' cast-expression 537 case tok::minus: // unary-expression: '-' cast-expression 538 case tok::tilde: // unary-expression: '~' cast-expression 539 case tok::exclaim: // unary-expression: '!' cast-expression 540 case tok::kw___real: // unary-expression: '__real' cast-expression [GNU] 541 case tok::kw___imag: { // unary-expression: '__imag' cast-expression [GNU] 542 SourceLocation SavedLoc = ConsumeToken(); 543 Res = ParseCastExpression(false); 544 if (!Res.isInvalid) 545 Res = Actions.ActOnUnaryOp(SavedLoc, SavedKind, Res.Val); 546 return Res; 547 } 548 549 case tok::kw___extension__:{//unary-expression:'__extension__' cast-expr [GNU] 550 // __extension__ silences extension warnings in the subexpression. 551 bool SavedExtWarn = Diags.getWarnOnExtensions(); 552 Diags.setWarnOnExtensions(false); 553 SourceLocation SavedLoc = ConsumeToken(); 554 Res = ParseCastExpression(false); 555 if (!Res.isInvalid) 556 Res = Actions.ActOnUnaryOp(SavedLoc, SavedKind, Res.Val); 557 Diags.setWarnOnExtensions(SavedExtWarn); 558 return Res; 559 } 560 case tok::kw_sizeof: // unary-expression: 'sizeof' unary-expression 561 // unary-expression: 'sizeof' '(' type-name ')' 562 case tok::kw___alignof: // unary-expression: '__alignof' unary-expression 563 // unary-expression: '__alignof' '(' type-name ')' 564 return ParseSizeofAlignofExpression(); 565 case tok::ampamp: { // unary-expression: '&&' identifier 566 SourceLocation AmpAmpLoc = ConsumeToken(); 567 if (Tok.isNot(tok::identifier)) { 568 Diag(Tok, diag::err_expected_ident); 569 return ExprResult(true); 570 } 571 572 Diag(AmpAmpLoc, diag::ext_gnu_address_of_label); 573 Res = Actions.ActOnAddrLabel(AmpAmpLoc, Tok.getLocation(), 574 Tok.getIdentifierInfo()); 575 ConsumeToken(); 576 return Res; 577 } 578 case tok::kw_const_cast: 579 case tok::kw_dynamic_cast: 580 case tok::kw_reinterpret_cast: 581 case tok::kw_static_cast: 582 return ParseCXXCasts(); 583 case tok::at: { 584 SourceLocation AtLoc = ConsumeToken(); 585 return ParseObjCAtExpression(AtLoc); 586 } 587 case tok::l_square: 588 // These can be followed by postfix-expr pieces. 589 return ParsePostfixExpressionSuffix(ParseObjCMessageExpression()); 590 default: 591 Diag(Tok, diag::err_expected_expression); 592 return ExprResult(true); 593 } 594 595 // unreachable. 596 abort(); 597} 598 599/// ParsePostfixExpressionSuffix - Once the leading part of a postfix-expression 600/// is parsed, this method parses any suffixes that apply. 601/// 602/// postfix-expression: [C99 6.5.2] 603/// primary-expression 604/// postfix-expression '[' expression ']' 605/// postfix-expression '(' argument-expression-list[opt] ')' 606/// postfix-expression '.' identifier 607/// postfix-expression '->' identifier 608/// postfix-expression '++' 609/// postfix-expression '--' 610/// '(' type-name ')' '{' initializer-list '}' 611/// '(' type-name ')' '{' initializer-list ',' '}' 612/// 613/// argument-expression-list: [C99 6.5.2] 614/// argument-expression 615/// argument-expression-list ',' assignment-expression 616/// 617Parser::ExprResult Parser::ParsePostfixExpressionSuffix(ExprResult LHS) { 618 619 // Now that the primary-expression piece of the postfix-expression has been 620 // parsed, see if there are any postfix-expression pieces here. 621 SourceLocation Loc; 622 while (1) { 623 switch (Tok.getKind()) { 624 default: // Not a postfix-expression suffix. 625 return LHS; 626 case tok::l_square: { // postfix-expression: p-e '[' expression ']' 627 Loc = ConsumeBracket(); 628 ExprResult Idx = ParseExpression(); 629 630 SourceLocation RLoc = Tok.getLocation(); 631 632 if (!LHS.isInvalid && !Idx.isInvalid && Tok.is(tok::r_square)) 633 LHS = Actions.ActOnArraySubscriptExpr(LHS.Val, Loc, Idx.Val, RLoc); 634 else 635 LHS = ExprResult(true); 636 637 // Match the ']'. 638 MatchRHSPunctuation(tok::r_square, Loc); 639 break; 640 } 641 642 case tok::l_paren: { // p-e: p-e '(' argument-expression-list[opt] ')' 643 llvm::SmallVector<ExprTy*, 8> ArgExprs; 644 llvm::SmallVector<SourceLocation, 8> CommaLocs; 645 646 Loc = ConsumeParen(); 647 648 if (Tok.isNot(tok::r_paren)) { 649 while (1) { 650 ExprResult ArgExpr = ParseAssignmentExpression(); 651 if (ArgExpr.isInvalid) { 652 SkipUntil(tok::r_paren); 653 return ExprResult(true); 654 } else 655 ArgExprs.push_back(ArgExpr.Val); 656 657 if (Tok.isNot(tok::comma)) 658 break; 659 // Move to the next argument, remember where the comma was. 660 CommaLocs.push_back(ConsumeToken()); 661 } 662 } 663 664 // Match the ')'. 665 if (!LHS.isInvalid && Tok.is(tok::r_paren)) { 666 assert((ArgExprs.size() == 0 || ArgExprs.size()-1 == CommaLocs.size())&& 667 "Unexpected number of commas!"); 668 LHS = Actions.ActOnCallExpr(LHS.Val, Loc, &ArgExprs[0], ArgExprs.size(), 669 &CommaLocs[0], Tok.getLocation()); 670 } 671 672 MatchRHSPunctuation(tok::r_paren, Loc); 673 break; 674 } 675 case tok::arrow: // postfix-expression: p-e '->' identifier 676 case tok::period: { // postfix-expression: p-e '.' identifier 677 tok::TokenKind OpKind = Tok.getKind(); 678 SourceLocation OpLoc = ConsumeToken(); // Eat the "." or "->" token. 679 680 if (Tok.isNot(tok::identifier)) { 681 Diag(Tok, diag::err_expected_ident); 682 return ExprResult(true); 683 } 684 685 if (!LHS.isInvalid) 686 LHS = Actions.ActOnMemberReferenceExpr(LHS.Val, OpLoc, OpKind, 687 Tok.getLocation(), 688 *Tok.getIdentifierInfo()); 689 ConsumeToken(); 690 break; 691 } 692 case tok::plusplus: // postfix-expression: postfix-expression '++' 693 case tok::minusminus: // postfix-expression: postfix-expression '--' 694 if (!LHS.isInvalid) 695 LHS = Actions.ActOnPostfixUnaryOp(Tok.getLocation(), Tok.getKind(), 696 LHS.Val); 697 ConsumeToken(); 698 break; 699 } 700 } 701} 702 703 704/// ParseSizeofAlignofExpression - Parse a sizeof or alignof expression. 705/// unary-expression: [C99 6.5.3] 706/// 'sizeof' unary-expression 707/// 'sizeof' '(' type-name ')' 708/// [GNU] '__alignof' unary-expression 709/// [GNU] '__alignof' '(' type-name ')' 710Parser::ExprResult Parser::ParseSizeofAlignofExpression() { 711 assert((Tok.is(tok::kw_sizeof) || Tok.is(tok::kw___alignof)) && 712 "Not a sizeof/alignof expression!"); 713 Token OpTok = Tok; 714 ConsumeToken(); 715 716 // If the operand doesn't start with an '(', it must be an expression. 717 ExprResult Operand; 718 if (Tok.isNot(tok::l_paren)) { 719 Operand = ParseCastExpression(true); 720 } else { 721 // If it starts with a '(', we know that it is either a parenthesized 722 // type-name, or it is a unary-expression that starts with a compound 723 // literal, or starts with a primary-expression that is a parenthesized 724 // expression. 725 ParenParseOption ExprType = CastExpr; 726 TypeTy *CastTy; 727 SourceLocation LParenLoc = Tok.getLocation(), RParenLoc; 728 Operand = ParseParenExpression(ExprType, CastTy, RParenLoc); 729 730 // If ParseParenExpression parsed a '(typename)' sequence only, the this is 731 // sizeof/alignof a type. Otherwise, it is sizeof/alignof an expression. 732 if (ExprType == CastExpr) 733 return Actions.ActOnSizeOfAlignOfTypeExpr(OpTok.getLocation(), 734 OpTok.is(tok::kw_sizeof), 735 LParenLoc, CastTy, RParenLoc); 736 737 // If this is a parenthesized expression, it is the start of a 738 // unary-expression, but doesn't include any postfix pieces. Parse these 739 // now if present. 740 Operand = ParsePostfixExpressionSuffix(Operand); 741 } 742 743 // If we get here, the operand to the sizeof/alignof was an expresion. 744 if (!Operand.isInvalid) 745 Operand = Actions.ActOnUnaryOp(OpTok.getLocation(), OpTok.getKind(), 746 Operand.Val); 747 return Operand; 748} 749 750/// ParseBuiltinPrimaryExpression 751/// 752/// primary-expression: [C99 6.5.1] 753/// [GNU] '__builtin_va_arg' '(' assignment-expression ',' type-name ')' 754/// [GNU] '__builtin_offsetof' '(' type-name ',' offsetof-member-designator')' 755/// [GNU] '__builtin_choose_expr' '(' assign-expr ',' assign-expr ',' 756/// assign-expr ')' 757/// [GNU] '__builtin_types_compatible_p' '(' type-name ',' type-name ')' 758/// [CLANG] '__builtin_overload' '(' expr (',' expr)* ')' 759/// 760/// [GNU] offsetof-member-designator: 761/// [GNU] identifier 762/// [GNU] offsetof-member-designator '.' identifier 763/// [GNU] offsetof-member-designator '[' expression ']' 764/// 765Parser::ExprResult Parser::ParseBuiltinPrimaryExpression() { 766 ExprResult Res(false); 767 const IdentifierInfo *BuiltinII = Tok.getIdentifierInfo(); 768 769 tok::TokenKind T = Tok.getKind(); 770 SourceLocation StartLoc = ConsumeToken(); // Eat the builtin identifier. 771 772 // All of these start with an open paren. 773 if (Tok.isNot(tok::l_paren)) { 774 Diag(Tok, diag::err_expected_lparen_after, BuiltinII->getName()); 775 return ExprResult(true); 776 } 777 778 SourceLocation LParenLoc = ConsumeParen(); 779 // TODO: Build AST. 780 781 switch (T) { 782 default: assert(0 && "Not a builtin primary expression!"); 783 case tok::kw___builtin_va_arg: { 784 ExprResult Expr = ParseAssignmentExpression(); 785 if (Expr.isInvalid) { 786 SkipUntil(tok::r_paren); 787 return Res; 788 } 789 790 if (ExpectAndConsume(tok::comma, diag::err_expected_comma, "",tok::r_paren)) 791 return ExprResult(true); 792 793 TypeTy *Ty = ParseTypeName(); 794 795 if (Tok.isNot(tok::r_paren)) { 796 Diag(Tok, diag::err_expected_rparen); 797 return ExprResult(true); 798 } 799 Res = Actions.ActOnVAArg(StartLoc, Expr.Val, Ty, ConsumeParen()); 800 break; 801 } 802 case tok::kw___builtin_offsetof: { 803 SourceLocation TypeLoc = Tok.getLocation(); 804 TypeTy *Ty = ParseTypeName(); 805 806 if (ExpectAndConsume(tok::comma, diag::err_expected_comma, "",tok::r_paren)) 807 return ExprResult(true); 808 809 // We must have at least one identifier here. 810 if (Tok.isNot(tok::identifier)) { 811 Diag(Tok, diag::err_expected_ident); 812 SkipUntil(tok::r_paren); 813 return true; 814 } 815 816 // Keep track of the various subcomponents we see. 817 llvm::SmallVector<Action::OffsetOfComponent, 4> Comps; 818 819 Comps.push_back(Action::OffsetOfComponent()); 820 Comps.back().isBrackets = false; 821 Comps.back().U.IdentInfo = Tok.getIdentifierInfo(); 822 Comps.back().LocStart = Comps.back().LocEnd = ConsumeToken(); 823 824 while (1) { 825 if (Tok.is(tok::period)) { 826 // offsetof-member-designator: offsetof-member-designator '.' identifier 827 Comps.push_back(Action::OffsetOfComponent()); 828 Comps.back().isBrackets = false; 829 Comps.back().LocStart = ConsumeToken(); 830 831 if (Tok.isNot(tok::identifier)) { 832 Diag(Tok, diag::err_expected_ident); 833 SkipUntil(tok::r_paren); 834 return true; 835 } 836 Comps.back().U.IdentInfo = Tok.getIdentifierInfo(); 837 Comps.back().LocEnd = ConsumeToken(); 838 839 } else if (Tok.is(tok::l_square)) { 840 // offsetof-member-designator: offsetof-member-design '[' expression ']' 841 Comps.push_back(Action::OffsetOfComponent()); 842 Comps.back().isBrackets = true; 843 Comps.back().LocStart = ConsumeBracket(); 844 Res = ParseExpression(); 845 if (Res.isInvalid) { 846 SkipUntil(tok::r_paren); 847 return Res; 848 } 849 Comps.back().U.E = Res.Val; 850 851 Comps.back().LocEnd = 852 MatchRHSPunctuation(tok::r_square, Comps.back().LocStart); 853 } else if (Tok.is(tok::r_paren)) { 854 Res = Actions.ActOnBuiltinOffsetOf(StartLoc, TypeLoc, Ty, &Comps[0], 855 Comps.size(), ConsumeParen()); 856 break; 857 } else { 858 // Error occurred. 859 return ExprResult(true); 860 } 861 } 862 break; 863 } 864 case tok::kw___builtin_choose_expr: { 865 ExprResult Cond = ParseAssignmentExpression(); 866 if (Cond.isInvalid) { 867 SkipUntil(tok::r_paren); 868 return Cond; 869 } 870 if (ExpectAndConsume(tok::comma, diag::err_expected_comma, "",tok::r_paren)) 871 return ExprResult(true); 872 873 ExprResult Expr1 = ParseAssignmentExpression(); 874 if (Expr1.isInvalid) { 875 SkipUntil(tok::r_paren); 876 return Expr1; 877 } 878 if (ExpectAndConsume(tok::comma, diag::err_expected_comma, "",tok::r_paren)) 879 return ExprResult(true); 880 881 ExprResult Expr2 = ParseAssignmentExpression(); 882 if (Expr2.isInvalid) { 883 SkipUntil(tok::r_paren); 884 return Expr2; 885 } 886 if (Tok.isNot(tok::r_paren)) { 887 Diag(Tok, diag::err_expected_rparen); 888 return ExprResult(true); 889 } 890 Res = Actions.ActOnChooseExpr(StartLoc, Cond.Val, Expr1.Val, Expr2.Val, 891 ConsumeParen()); 892 break; 893 } 894 case tok::kw___builtin_overload: { 895 llvm::SmallVector<ExprTy*, 8> ArgExprs; 896 llvm::SmallVector<SourceLocation, 8> CommaLocs; 897 898 // For each iteration through the loop look for assign-expr followed by a 899 // comma. If there is no comma, break and attempt to match r-paren. 900 if (Tok.isNot(tok::r_paren)) { 901 while (1) { 902 ExprResult ArgExpr = ParseAssignmentExpression(); 903 if (ArgExpr.isInvalid) { 904 SkipUntil(tok::r_paren); 905 return ExprResult(true); 906 } else 907 ArgExprs.push_back(ArgExpr.Val); 908 909 if (Tok.isNot(tok::comma)) 910 break; 911 // Move to the next argument, remember where the comma was. 912 CommaLocs.push_back(ConsumeToken()); 913 } 914 } 915 916 // Attempt to consume the r-paren 917 if (Tok.isNot(tok::r_paren)) { 918 Diag(Tok, diag::err_expected_rparen); 919 SkipUntil(tok::r_paren); 920 return ExprResult(true); 921 } 922 Res = Actions.ActOnOverloadExpr(&ArgExprs[0], ArgExprs.size(), 923 &CommaLocs[0], StartLoc, ConsumeParen()); 924 break; 925 } 926 case tok::kw___builtin_types_compatible_p: 927 TypeTy *Ty1 = ParseTypeName(); 928 929 if (ExpectAndConsume(tok::comma, diag::err_expected_comma, "",tok::r_paren)) 930 return ExprResult(true); 931 932 TypeTy *Ty2 = ParseTypeName(); 933 934 if (Tok.isNot(tok::r_paren)) { 935 Diag(Tok, diag::err_expected_rparen); 936 return ExprResult(true); 937 } 938 Res = Actions.ActOnTypesCompatibleExpr(StartLoc, Ty1, Ty2, ConsumeParen()); 939 break; 940 } 941 942 // These can be followed by postfix-expr pieces because they are 943 // primary-expressions. 944 return ParsePostfixExpressionSuffix(Res); 945} 946 947/// ParseParenExpression - This parses the unit that starts with a '(' token, 948/// based on what is allowed by ExprType. The actual thing parsed is returned 949/// in ExprType. 950/// 951/// primary-expression: [C99 6.5.1] 952/// '(' expression ')' 953/// [GNU] '(' compound-statement ')' (if !ParenExprOnly) 954/// postfix-expression: [C99 6.5.2] 955/// '(' type-name ')' '{' initializer-list '}' 956/// '(' type-name ')' '{' initializer-list ',' '}' 957/// cast-expression: [C99 6.5.4] 958/// '(' type-name ')' cast-expression 959/// 960Parser::ExprResult Parser::ParseParenExpression(ParenParseOption &ExprType, 961 TypeTy *&CastTy, 962 SourceLocation &RParenLoc) { 963 assert(Tok.is(tok::l_paren) && "Not a paren expr!"); 964 SourceLocation OpenLoc = ConsumeParen(); 965 ExprResult Result(true); 966 CastTy = 0; 967 968 if (ExprType >= CompoundStmt && Tok.is(tok::l_brace)) { 969 Diag(Tok, diag::ext_gnu_statement_expr); 970 Parser::StmtResult Stmt = ParseCompoundStatement(true); 971 ExprType = CompoundStmt; 972 973 // If the substmt parsed correctly, build the AST node. 974 if (!Stmt.isInvalid && Tok.is(tok::r_paren)) 975 Result = Actions.ActOnStmtExpr(OpenLoc, Stmt.Val, Tok.getLocation()); 976 977 } else if (ExprType >= CompoundLiteral && isTypeSpecifierQualifier()) { 978 // Otherwise, this is a compound literal expression or cast expression. 979 TypeTy *Ty = ParseTypeName(); 980 981 // Match the ')'. 982 if (Tok.is(tok::r_paren)) 983 RParenLoc = ConsumeParen(); 984 else 985 MatchRHSPunctuation(tok::r_paren, OpenLoc); 986 987 if (Tok.is(tok::l_brace)) { 988 if (!getLang().C99) // Compound literals don't exist in C90. 989 Diag(OpenLoc, diag::ext_c99_compound_literal); 990 Result = ParseInitializer(); 991 ExprType = CompoundLiteral; 992 if (!Result.isInvalid) 993 return Actions.ActOnCompoundLiteral(OpenLoc, Ty, RParenLoc, Result.Val); 994 } else if (ExprType == CastExpr) { 995 // Note that this doesn't parse the subsequence cast-expression, it just 996 // returns the parsed type to the callee. 997 ExprType = CastExpr; 998 CastTy = Ty; 999 return ExprResult(false); 1000 } else { 1001 Diag(Tok, diag::err_expected_lbrace_in_compound_literal); 1002 return ExprResult(true); 1003 } 1004 return Result; 1005 } else { 1006 Result = ParseExpression(); 1007 ExprType = SimpleExpr; 1008 if (!Result.isInvalid && Tok.is(tok::r_paren)) 1009 Result = Actions.ActOnParenExpr(OpenLoc, Tok.getLocation(), Result.Val); 1010 } 1011 1012 // Match the ')'. 1013 if (Result.isInvalid) 1014 SkipUntil(tok::r_paren); 1015 else { 1016 if (Tok.is(tok::r_paren)) 1017 RParenLoc = ConsumeParen(); 1018 else 1019 MatchRHSPunctuation(tok::r_paren, OpenLoc); 1020 } 1021 1022 return Result; 1023} 1024 1025/// ParseStringLiteralExpression - This handles the various token types that 1026/// form string literals, and also handles string concatenation [C99 5.1.1.2, 1027/// translation phase #6]. 1028/// 1029/// primary-expression: [C99 6.5.1] 1030/// string-literal 1031Parser::ExprResult Parser::ParseStringLiteralExpression() { 1032 assert(isTokenStringLiteral() && "Not a string literal!"); 1033 1034 // String concat. Note that keywords like __func__ and __FUNCTION__ are not 1035 // considered to be strings for concatenation purposes. 1036 llvm::SmallVector<Token, 4> StringToks; 1037 1038 do { 1039 StringToks.push_back(Tok); 1040 ConsumeStringToken(); 1041 } while (isTokenStringLiteral()); 1042 1043 // Pass the set of string tokens, ready for concatenation, to the actions. 1044 return Actions.ActOnStringLiteral(&StringToks[0], StringToks.size()); 1045} 1046