blob: c2b54828556c1d481ecc9f83b4689fda55807f0c [file] [log] [blame]
Reid Spencer5f016e22007-07-11 17:01:13 +00001//===--- ParseExpr.cpp - Expression Parsing -------------------------------===//
2//
3// The LLVM Compiler Infrastructure
4//
5// This file was developed by Chris Lattner and is distributed under
6// the University of Illinois Open Source 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///
161/// assignment-operator: one of
162/// = *= /= %= += -= <<= >>= &= ^= |=
163///
164/// expression: [C99 6.5.17]
165/// assignment-expression
166/// expression ',' assignment-expression
167///
168Parser::ExprResult Parser::ParseExpression() {
169 ExprResult LHS = ParseCastExpression(false);
170 if (LHS.isInvalid) return LHS;
171
172 return ParseRHSOfBinaryExpression(LHS, prec::Comma);
173}
174
175/// ParseAssignmentExpression - Parse an expr that doesn't include commas.
176///
177Parser::ExprResult Parser::ParseAssignmentExpression() {
178 ExprResult LHS = ParseCastExpression(false);
179 if (LHS.isInvalid) return LHS;
180
181 return ParseRHSOfBinaryExpression(LHS, prec::Assignment);
182}
183
184Parser::ExprResult Parser::ParseConstantExpression() {
185 ExprResult LHS = ParseCastExpression(false);
186 if (LHS.isInvalid) return LHS;
187
188 // TODO: Validate that this is a constant expr!
189 return ParseRHSOfBinaryExpression(LHS, prec::Conditional);
190}
191
192/// ParseExpressionWithLeadingIdentifier - This special purpose method is used
193/// in contexts where we have already consumed an identifier (which we saved in
194/// 'IdTok'), then discovered that the identifier was really the leading token
195/// of part of an expression. For example, in "A[1]+B", we consumed "A" (which
196/// is now in 'IdTok') and the current token is "[".
197Parser::ExprResult Parser::
Chris Lattnerd2177732007-07-20 16:59:19 +0000198ParseExpressionWithLeadingIdentifier(const Token &IdTok) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000199 // We know that 'IdTok' must correspond to this production:
200 // primary-expression: identifier
201
202 // Let the actions module handle the identifier.
203 ExprResult Res = Actions.ParseIdentifierExpr(CurScope, IdTok.getLocation(),
204 *IdTok.getIdentifierInfo(),
205 Tok.getKind() == tok::l_paren);
206
207 // Because we have to parse an entire cast-expression before starting the
208 // ParseRHSOfBinaryExpression method (which parses any trailing binops), we
209 // need to handle the 'postfix-expression' rules. We do this by invoking
210 // ParsePostfixExpressionSuffix to consume any postfix-expression suffixes:
211 Res = ParsePostfixExpressionSuffix(Res);
212 if (Res.isInvalid) return Res;
213
214 // At this point, the "A[1]" part of "A[1]+B" has been consumed. Once this is
215 // done, we know we don't have to do anything for cast-expression, because the
216 // only non-postfix-expression production starts with a '(' token, and we know
217 // we have an identifier. As such, we can invoke ParseRHSOfBinaryExpression
218 // to consume any trailing operators (e.g. "+" in this example) and connected
219 // chunks of the expression.
220 return ParseRHSOfBinaryExpression(Res, prec::Comma);
221}
222
223/// ParseExpressionWithLeadingIdentifier - This special purpose method is used
224/// in contexts where we have already consumed an identifier (which we saved in
225/// 'IdTok'), then discovered that the identifier was really the leading token
226/// of part of an assignment-expression. For example, in "A[1]+B", we consumed
227/// "A" (which is now in 'IdTok') and the current token is "[".
228Parser::ExprResult Parser::
Chris Lattnerd2177732007-07-20 16:59:19 +0000229ParseAssignmentExprWithLeadingIdentifier(const Token &IdTok) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000230 // We know that 'IdTok' must correspond to this production:
231 // primary-expression: identifier
232
233 // Let the actions module handle the identifier.
234 ExprResult Res = Actions.ParseIdentifierExpr(CurScope, IdTok.getLocation(),
235 *IdTok.getIdentifierInfo(),
236 Tok.getKind() == tok::l_paren);
237
238 // Because we have to parse an entire cast-expression before starting the
239 // ParseRHSOfBinaryExpression method (which parses any trailing binops), we
240 // need to handle the 'postfix-expression' rules. We do this by invoking
241 // ParsePostfixExpressionSuffix to consume any postfix-expression suffixes:
242 Res = ParsePostfixExpressionSuffix(Res);
243 if (Res.isInvalid) return Res;
244
245 // At this point, the "A[1]" part of "A[1]+B" has been consumed. Once this is
246 // done, we know we don't have to do anything for cast-expression, because the
247 // only non-postfix-expression production starts with a '(' token, and we know
248 // we have an identifier. As such, we can invoke ParseRHSOfBinaryExpression
249 // to consume any trailing operators (e.g. "+" in this example) and connected
250 // chunks of the expression.
251 return ParseRHSOfBinaryExpression(Res, prec::Assignment);
252}
253
254
255/// ParseAssignmentExpressionWithLeadingStar - This special purpose method is
256/// used in contexts where we have already consumed a '*' (which we saved in
257/// 'StarTok'), then discovered that the '*' was really the leading token of an
258/// expression. For example, in "*(int*)P+B", we consumed "*" (which is
259/// now in 'StarTok') and the current token is "(".
260Parser::ExprResult Parser::
Chris Lattnerd2177732007-07-20 16:59:19 +0000261ParseAssignmentExpressionWithLeadingStar(const Token &StarTok) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000262 // We know that 'StarTok' must correspond to this production:
263 // unary-expression: unary-operator cast-expression
264 // where 'unary-operator' is '*'.
265
266 // Parse the cast-expression that follows the '*'. This will parse the
267 // "*(int*)P" part of "*(int*)P+B".
268 ExprResult Res = ParseCastExpression(false);
269 if (Res.isInvalid) return Res;
270
271 // Combine StarTok + Res to get the new AST for the combined expression..
272 Res = Actions.ParseUnaryOp(StarTok.getLocation(), tok::star, Res.Val);
273 if (Res.isInvalid) return Res;
274
275
276 // We have to parse an entire cast-expression before starting the
277 // ParseRHSOfBinaryExpression method (which parses any trailing binops). Since
278 // we know that the only production above us is the cast-expression
279 // production, and because the only alternative productions start with a '('
280 // token (we know we had a '*'), there is no work to do to get a whole
281 // cast-expression.
282
283 // At this point, the "*(int*)P" part of "*(int*)P+B" has been consumed. Once
284 // this is done, we can invoke ParseRHSOfBinaryExpression to consume any
285 // trailing operators (e.g. "+" in this example) and connected chunks of the
286 // assignment-expression.
287 return ParseRHSOfBinaryExpression(Res, prec::Assignment);
288}
289
290
291/// ParseRHSOfBinaryExpression - Parse a binary expression that starts with
292/// LHS and has a precedence of at least MinPrec.
293Parser::ExprResult
294Parser::ParseRHSOfBinaryExpression(ExprResult LHS, unsigned MinPrec) {
295 unsigned NextTokPrec = getBinOpPrecedence(Tok.getKind());
296 SourceLocation ColonLoc;
297
298 while (1) {
299 // If this token has a lower precedence than we are allowed to parse (e.g.
300 // because we are called recursively, or because the token is not a binop),
301 // then we are done!
302 if (NextTokPrec < MinPrec)
303 return LHS;
304
305 // Consume the operator, saving the operator token for error reporting.
Chris Lattnerd2177732007-07-20 16:59:19 +0000306 Token OpToken = Tok;
Reid Spencer5f016e22007-07-11 17:01:13 +0000307 ConsumeToken();
308
309 // Special case handling for the ternary operator.
310 ExprResult TernaryMiddle(true);
311 if (NextTokPrec == prec::Conditional) {
312 if (Tok.getKind() != tok::colon) {
313 // Handle this production specially:
314 // logical-OR-expression '?' expression ':' conditional-expression
315 // In particular, the RHS of the '?' is 'expression', not
316 // 'logical-OR-expression' as we might expect.
317 TernaryMiddle = ParseExpression();
Chris Lattnerdbd583c2007-08-31 04:58:34 +0000318 if (TernaryMiddle.isInvalid) {
319 Actions.DeleteExpr(LHS.Val);
320 return TernaryMiddle;
321 }
Reid Spencer5f016e22007-07-11 17:01:13 +0000322 } else {
323 // Special case handling of "X ? Y : Z" where Y is empty:
324 // logical-OR-expression '?' ':' conditional-expression [GNU]
325 TernaryMiddle = ExprResult(false);
326 Diag(Tok, diag::ext_gnu_conditional_expr);
327 }
328
329 if (Tok.getKind() != tok::colon) {
330 Diag(Tok, diag::err_expected_colon);
331 Diag(OpToken, diag::err_matching, "?");
Chris Lattnerdbd583c2007-08-31 04:58:34 +0000332 Actions.DeleteExpr(LHS.Val);
333 Actions.DeleteExpr(TernaryMiddle.Val);
Reid Spencer5f016e22007-07-11 17:01:13 +0000334 return ExprResult(true);
335 }
336
337 // Eat the colon.
338 ColonLoc = ConsumeToken();
339 }
340
341 // Parse another leaf here for the RHS of the operator.
342 ExprResult RHS = ParseCastExpression(false);
Chris Lattnerdbd583c2007-08-31 04:58:34 +0000343 if (RHS.isInvalid) {
344 Actions.DeleteExpr(LHS.Val);
345 Actions.DeleteExpr(TernaryMiddle.Val);
346 return RHS;
347 }
Reid Spencer5f016e22007-07-11 17:01:13 +0000348
349 // Remember the precedence of this operator and get the precedence of the
350 // operator immediately to the right of the RHS.
351 unsigned ThisPrec = NextTokPrec;
352 NextTokPrec = getBinOpPrecedence(Tok.getKind());
353
354 // Assignment and conditional expressions are right-associative.
355 bool isRightAssoc = NextTokPrec == prec::Conditional ||
356 NextTokPrec == prec::Assignment;
357
358 // Get the precedence of the operator to the right of the RHS. If it binds
359 // more tightly with RHS than we do, evaluate it completely first.
360 if (ThisPrec < NextTokPrec ||
361 (ThisPrec == NextTokPrec && isRightAssoc)) {
362 // If this is left-associative, only parse things on the RHS that bind
363 // more tightly than the current operator. If it is left-associative, it
364 // is okay, to bind exactly as tightly. For example, compile A=B=C=D as
365 // A=(B=(C=D)), where each paren is a level of recursion here.
366 RHS = ParseRHSOfBinaryExpression(RHS, ThisPrec + !isRightAssoc);
Chris Lattnerdbd583c2007-08-31 04:58:34 +0000367 if (RHS.isInvalid) {
368 Actions.DeleteExpr(LHS.Val);
369 Actions.DeleteExpr(TernaryMiddle.Val);
370 return RHS;
371 }
Reid Spencer5f016e22007-07-11 17:01:13 +0000372
373 NextTokPrec = getBinOpPrecedence(Tok.getKind());
374 }
375 assert(NextTokPrec <= ThisPrec && "Recursion didn't work!");
376
Chris Lattnerd56d6b62007-08-31 05:01:50 +0000377 if (!LHS.isInvalid) {
378 // Combine the LHS and RHS into the LHS (e.g. build AST).
379 if (TernaryMiddle.isInvalid)
380 LHS = Actions.ParseBinOp(OpToken.getLocation(), OpToken.getKind(),
381 LHS.Val, RHS.Val);
382 else
383 LHS = Actions.ParseConditionalOp(OpToken.getLocation(), ColonLoc,
384 LHS.Val, TernaryMiddle.Val, RHS.Val);
385 } else {
386 // We had a semantic error on the LHS. Just free the RHS and continue.
387 Actions.DeleteExpr(TernaryMiddle.Val);
388 Actions.DeleteExpr(RHS.Val);
389 }
Reid Spencer5f016e22007-07-11 17:01:13 +0000390 }
391}
392
393/// ParseCastExpression - Parse a cast-expression, or, if isUnaryExpression is
394/// true, parse a unary-expression.
395///
396/// cast-expression: [C99 6.5.4]
397/// unary-expression
398/// '(' type-name ')' cast-expression
399///
400/// unary-expression: [C99 6.5.3]
401/// postfix-expression
402/// '++' unary-expression
403/// '--' unary-expression
404/// unary-operator cast-expression
405/// 'sizeof' unary-expression
406/// 'sizeof' '(' type-name ')'
407/// [GNU] '__alignof' unary-expression
408/// [GNU] '__alignof' '(' type-name ')'
409/// [GNU] '&&' identifier
410///
411/// unary-operator: one of
412/// '&' '*' '+' '-' '~' '!'
413/// [GNU] '__extension__' '__real' '__imag'
414///
415/// primary-expression: [C99 6.5.1]
416/// identifier
417/// constant
418/// string-literal
419/// [C++] boolean-literal [C++ 2.13.5]
420/// '(' expression ')'
421/// '__func__' [C99 6.4.2.2]
422/// [GNU] '__FUNCTION__'
423/// [GNU] '__PRETTY_FUNCTION__'
424/// [GNU] '(' compound-statement ')'
425/// [GNU] '__builtin_va_arg' '(' assignment-expression ',' type-name ')'
426/// [GNU] '__builtin_offsetof' '(' type-name ',' offsetof-member-designator')'
427/// [GNU] '__builtin_choose_expr' '(' assign-expr ',' assign-expr ','
428/// assign-expr ')'
429/// [GNU] '__builtin_types_compatible_p' '(' type-name ',' type-name ')'
430/// [OBC] '[' objc-receiver objc-message-args ']' [TODO]
431/// [OBC] '@selector' '(' objc-selector-arg ')' [TODO]
432/// [OBC] '@protocol' '(' identifier ')' [TODO]
433/// [OBC] '@encode' '(' type-name ')' [TODO]
Anders Carlsson5dac7f62007-08-21 19:18:49 +0000434/// [OBC] objc-string-literal
Reid Spencer5f016e22007-07-11 17:01:13 +0000435/// [C++] 'const_cast' '<' type-name '>' '(' expression ')' [C++ 5.2p1]
436/// [C++] 'dynamic_cast' '<' type-name '>' '(' expression ')' [C++ 5.2p1]
437/// [C++] 'reinterpret_cast' '<' type-name '>' '(' expression ')' [C++ 5.2p1]
438/// [C++] 'static_cast' '<' type-name '>' '(' expression ')' [C++ 5.2p1]
439///
440/// constant: [C99 6.4.4]
441/// integer-constant
442/// floating-constant
443/// enumeration-constant -> identifier
444/// character-constant
445///
446Parser::ExprResult Parser::ParseCastExpression(bool isUnaryExpression) {
447 ExprResult Res;
448 tok::TokenKind SavedKind = Tok.getKind();
449
450 // This handles all of cast-expression, unary-expression, postfix-expression,
451 // and primary-expression. We handle them together like this for efficiency
452 // and to simplify handling of an expression starting with a '(' token: which
453 // may be one of a parenthesized expression, cast-expression, compound literal
454 // expression, or statement expression.
455 //
456 // If the parsed tokens consist of a primary-expression, the cases below
457 // call ParsePostfixExpressionSuffix to handle the postfix expression
458 // suffixes. Cases that cannot be followed by postfix exprs should
459 // return without invoking ParsePostfixExpressionSuffix.
460 switch (SavedKind) {
461 case tok::l_paren: {
462 // If this expression is limited to being a unary-expression, the parent can
463 // not start a cast expression.
464 ParenParseOption ParenExprType =
465 isUnaryExpression ? CompoundLiteral : CastExpr;
466 TypeTy *CastTy;
467 SourceLocation LParenLoc = Tok.getLocation();
468 SourceLocation RParenLoc;
469 Res = ParseParenExpression(ParenExprType, CastTy, RParenLoc);
470 if (Res.isInvalid) return Res;
471
472 switch (ParenExprType) {
473 case SimpleExpr: break; // Nothing else to do.
474 case CompoundStmt: break; // Nothing else to do.
475 case CompoundLiteral:
476 // We parsed '(' type-name ')' '{' ... '}'. If any suffixes of
477 // postfix-expression exist, parse them now.
478 break;
479 case CastExpr:
480 // We parsed '(' type-name ')' and the thing after it wasn't a '{'. Parse
481 // the cast-expression that follows it next.
482 // TODO: For cast expression with CastTy.
483 Res = ParseCastExpression(false);
484 if (!Res.isInvalid)
485 Res = Actions.ParseCastExpr(LParenLoc, CastTy, RParenLoc, Res.Val);
486 return Res;
487 }
488
489 // These can be followed by postfix-expr pieces.
490 return ParsePostfixExpressionSuffix(Res);
491 }
492
493 // primary-expression
494 case tok::numeric_constant:
495 // constant: integer-constant
496 // constant: floating-constant
497
498 Res = Actions.ParseNumericConstant(Tok);
499 ConsumeToken();
500
501 // These can be followed by postfix-expr pieces.
502 return ParsePostfixExpressionSuffix(Res);
503
504 case tok::kw_true:
505 case tok::kw_false:
506 return ParseCXXBoolLiteral();
507
508 case tok::identifier: { // primary-expression: identifier
509 // constant: enumeration-constant
510 // Consume the identifier so that we can see if it is followed by a '('.
511 // Function designators are allowed to be undeclared (C99 6.5.1p2), so we
512 // need to know whether or not this identifier is a function designator or
513 // not.
514 IdentifierInfo &II = *Tok.getIdentifierInfo();
515 SourceLocation L = ConsumeToken();
516 Res = Actions.ParseIdentifierExpr(CurScope, L, II,
517 Tok.getKind() == tok::l_paren);
518 // These can be followed by postfix-expr pieces.
519 return ParsePostfixExpressionSuffix(Res);
520 }
521 case tok::char_constant: // constant: character-constant
522 Res = Actions.ParseCharacterConstant(Tok);
523 ConsumeToken();
524 // These can be followed by postfix-expr pieces.
525 return ParsePostfixExpressionSuffix(Res);
526 case tok::kw___func__: // primary-expression: __func__ [C99 6.4.2.2]
527 case tok::kw___FUNCTION__: // primary-expression: __FUNCTION__ [GNU]
528 case tok::kw___PRETTY_FUNCTION__: // primary-expression: __P..Y_F..N__ [GNU]
Anders Carlsson22742662007-07-21 05:21:51 +0000529 Res = Actions.ParsePreDefinedExpr(Tok.getLocation(), SavedKind);
Reid Spencer5f016e22007-07-11 17:01:13 +0000530 ConsumeToken();
531 // These can be followed by postfix-expr pieces.
532 return ParsePostfixExpressionSuffix(Res);
533 case tok::string_literal: // primary-expression: string-literal
534 case tok::wide_string_literal:
535 Res = ParseStringLiteralExpression();
536 if (Res.isInvalid) return Res;
537 // This can be followed by postfix-expr pieces (e.g. "foo"[1]).
538 return ParsePostfixExpressionSuffix(Res);
539 case tok::kw___builtin_va_arg:
540 case tok::kw___builtin_offsetof:
541 case tok::kw___builtin_choose_expr:
542 case tok::kw___builtin_types_compatible_p:
543 return ParseBuiltinPrimaryExpression();
544 case tok::plusplus: // unary-expression: '++' unary-expression
545 case tok::minusminus: { // unary-expression: '--' unary-expression
546 SourceLocation SavedLoc = ConsumeToken();
547 Res = ParseCastExpression(true);
548 if (!Res.isInvalid)
549 Res = Actions.ParseUnaryOp(SavedLoc, SavedKind, Res.Val);
550 return Res;
551 }
552 case tok::amp: // unary-expression: '&' cast-expression
553 case tok::star: // unary-expression: '*' cast-expression
554 case tok::plus: // unary-expression: '+' cast-expression
555 case tok::minus: // unary-expression: '-' cast-expression
556 case tok::tilde: // unary-expression: '~' cast-expression
557 case tok::exclaim: // unary-expression: '!' cast-expression
558 case tok::kw___real: // unary-expression: '__real' cast-expression [GNU]
559 case tok::kw___imag: // unary-expression: '__imag' cast-expression [GNU]
560 case tok::kw___extension__:{//unary-expression:'__extension__' cast-expr [GNU]
Chris Lattnerab18c4c2007-07-24 16:58:17 +0000561 // FIXME: Extension should silence extwarns in subexpressions.
Reid Spencer5f016e22007-07-11 17:01:13 +0000562 SourceLocation SavedLoc = ConsumeToken();
563 Res = ParseCastExpression(false);
564 if (!Res.isInvalid)
565 Res = Actions.ParseUnaryOp(SavedLoc, SavedKind, Res.Val);
566 return Res;
567 }
568 case tok::kw_sizeof: // unary-expression: 'sizeof' unary-expression
569 // unary-expression: 'sizeof' '(' type-name ')'
570 case tok::kw___alignof: // unary-expression: '__alignof' unary-expression
571 // unary-expression: '__alignof' '(' type-name ')'
572 return ParseSizeofAlignofExpression();
573 case tok::ampamp: { // unary-expression: '&&' identifier
574 SourceLocation AmpAmpLoc = ConsumeToken();
575 if (Tok.getKind() != tok::identifier) {
576 Diag(Tok, diag::err_expected_ident);
577 return ExprResult(true);
578 }
579
580 Diag(AmpAmpLoc, diag::ext_gnu_address_of_label);
581 Res = Actions.ParseAddrLabel(AmpAmpLoc, Tok.getLocation(),
582 Tok.getIdentifierInfo());
583 ConsumeToken();
584 return Res;
585 }
586 case tok::kw_const_cast:
587 case tok::kw_dynamic_cast:
588 case tok::kw_reinterpret_cast:
589 case tok::kw_static_cast:
590 return ParseCXXCasts();
Anders Carlsson55085182007-08-21 17:43:55 +0000591 case tok::at:
592 return ParseObjCExpression();
Reid Spencer5f016e22007-07-11 17:01:13 +0000593 default:
594 Diag(Tok, diag::err_expected_expression);
595 return ExprResult(true);
596 }
597
598 // unreachable.
599 abort();
600}
601
602/// ParsePostfixExpressionSuffix - Once the leading part of a postfix-expression
603/// is parsed, this method parses any suffixes that apply.
604///
605/// postfix-expression: [C99 6.5.2]
606/// primary-expression
607/// postfix-expression '[' expression ']'
608/// postfix-expression '(' argument-expression-list[opt] ')'
609/// postfix-expression '.' identifier
610/// postfix-expression '->' identifier
611/// postfix-expression '++'
612/// postfix-expression '--'
613/// '(' type-name ')' '{' initializer-list '}'
614/// '(' type-name ')' '{' initializer-list ',' '}'
615///
616/// argument-expression-list: [C99 6.5.2]
617/// argument-expression
618/// argument-expression-list ',' assignment-expression
619///
620Parser::ExprResult Parser::ParsePostfixExpressionSuffix(ExprResult LHS) {
621
622 // Now that the primary-expression piece of the postfix-expression has been
623 // parsed, see if there are any postfix-expression pieces here.
624 SourceLocation Loc;
625 while (1) {
626 switch (Tok.getKind()) {
627 default: // Not a postfix-expression suffix.
628 return LHS;
629 case tok::l_square: { // postfix-expression: p-e '[' expression ']'
630 Loc = ConsumeBracket();
631 ExprResult Idx = ParseExpression();
632
633 SourceLocation RLoc = Tok.getLocation();
634
635 if (!LHS.isInvalid && !Idx.isInvalid && Tok.getKind() == tok::r_square)
636 LHS = Actions.ParseArraySubscriptExpr(LHS.Val, Loc, Idx.Val, RLoc);
637 else
638 LHS = ExprResult(true);
639
640 // Match the ']'.
641 MatchRHSPunctuation(tok::r_square, Loc);
642 break;
643 }
644
645 case tok::l_paren: { // p-e: p-e '(' argument-expression-list[opt] ')'
646 llvm::SmallVector<ExprTy*, 8> ArgExprs;
647 llvm::SmallVector<SourceLocation, 8> CommaLocs;
Reid Spencer5f016e22007-07-11 17:01:13 +0000648
649 Loc = ConsumeParen();
650
651 if (Tok.getKind() != tok::r_paren) {
652 while (1) {
653 ExprResult ArgExpr = ParseAssignmentExpression();
654 if (ArgExpr.isInvalid) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000655 SkipUntil(tok::r_paren);
Chris Lattner2ff54262007-07-21 05:18:12 +0000656 return ExprResult(true);
Reid Spencer5f016e22007-07-11 17:01:13 +0000657 } else
658 ArgExprs.push_back(ArgExpr.Val);
659
660 if (Tok.getKind() != tok::comma)
661 break;
662 // Move to the next argument, remember where the comma was.
663 CommaLocs.push_back(ConsumeToken());
664 }
665 }
666
667 // Match the ')'.
Chris Lattner2ff54262007-07-21 05:18:12 +0000668 if (!LHS.isInvalid && Tok.getKind() == tok::r_paren) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000669 assert((ArgExprs.size() == 0 || ArgExprs.size()-1 == CommaLocs.size())&&
670 "Unexpected number of commas!");
671 LHS = Actions.ParseCallExpr(LHS.Val, Loc, &ArgExprs[0], ArgExprs.size(),
672 &CommaLocs[0], Tok.getLocation());
673 }
674
Chris Lattner2ff54262007-07-21 05:18:12 +0000675 MatchRHSPunctuation(tok::r_paren, Loc);
Reid Spencer5f016e22007-07-11 17:01:13 +0000676 break;
677 }
678 case tok::arrow: // postfix-expression: p-e '->' identifier
679 case tok::period: { // postfix-expression: p-e '.' identifier
680 tok::TokenKind OpKind = Tok.getKind();
681 SourceLocation OpLoc = ConsumeToken(); // Eat the "." or "->" token.
682
683 if (Tok.getKind() != tok::identifier) {
684 Diag(Tok, diag::err_expected_ident);
685 return ExprResult(true);
686 }
687
688 if (!LHS.isInvalid)
689 LHS = Actions.ParseMemberReferenceExpr(LHS.Val, OpLoc, OpKind,
690 Tok.getLocation(),
691 *Tok.getIdentifierInfo());
692 ConsumeToken();
693 break;
694 }
695 case tok::plusplus: // postfix-expression: postfix-expression '++'
696 case tok::minusminus: // postfix-expression: postfix-expression '--'
697 if (!LHS.isInvalid)
698 LHS = Actions.ParsePostfixUnaryOp(Tok.getLocation(), Tok.getKind(),
699 LHS.Val);
700 ConsumeToken();
701 break;
702 }
703 }
704}
705
706
707/// ParseSizeofAlignofExpression - Parse a sizeof or alignof expression.
708/// unary-expression: [C99 6.5.3]
709/// 'sizeof' unary-expression
710/// 'sizeof' '(' type-name ')'
711/// [GNU] '__alignof' unary-expression
712/// [GNU] '__alignof' '(' type-name ')'
713Parser::ExprResult Parser::ParseSizeofAlignofExpression() {
714 assert((Tok.getKind() == tok::kw_sizeof ||
715 Tok.getKind() == tok::kw___alignof) &&
716 "Not a sizeof/alignof expression!");
Chris Lattnerd2177732007-07-20 16:59:19 +0000717 Token OpTok = Tok;
Reid Spencer5f016e22007-07-11 17:01:13 +0000718 ConsumeToken();
719
720 // If the operand doesn't start with an '(', it must be an expression.
721 ExprResult Operand;
722 if (Tok.getKind() != tok::l_paren) {
723 Operand = ParseCastExpression(true);
724 } else {
725 // If it starts with a '(', we know that it is either a parenthesized
726 // type-name, or it is a unary-expression that starts with a compound
727 // literal, or starts with a primary-expression that is a parenthesized
728 // expression.
729 ParenParseOption ExprType = CastExpr;
730 TypeTy *CastTy;
731 SourceLocation LParenLoc = Tok.getLocation(), RParenLoc;
732 Operand = ParseParenExpression(ExprType, CastTy, RParenLoc);
733
734 // If ParseParenExpression parsed a '(typename)' sequence only, the this is
735 // sizeof/alignof a type. Otherwise, it is sizeof/alignof an expression.
736 if (ExprType == CastExpr) {
737 return Actions.ParseSizeOfAlignOfTypeExpr(OpTok.getLocation(),
738 OpTok.getKind() == tok::kw_sizeof,
739 LParenLoc, CastTy, RParenLoc);
740 }
741 }
742
743 // If we get here, the operand to the sizeof/alignof was an expresion.
744 if (!Operand.isInvalid)
745 Operand = Actions.ParseUnaryOp(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///
759/// [GNU] offsetof-member-designator:
760/// [GNU] identifier
761/// [GNU] offsetof-member-designator '.' identifier
762/// [GNU] offsetof-member-designator '[' expression ']'
763///
764Parser::ExprResult Parser::ParseBuiltinPrimaryExpression() {
765 ExprResult Res(false);
766 const IdentifierInfo *BuiltinII = Tok.getIdentifierInfo();
767
768 tok::TokenKind T = Tok.getKind();
769 SourceLocation StartLoc = ConsumeToken(); // Eat the builtin identifier.
770
771 // All of these start with an open paren.
772 if (Tok.getKind() != tok::l_paren) {
773 Diag(Tok, diag::err_expected_lparen_after, BuiltinII->getName());
774 return ExprResult(true);
775 }
776
777 SourceLocation LParenLoc = ConsumeParen();
778 // TODO: Build AST.
779
780 switch (T) {
781 default: assert(0 && "Not a builtin primary expression!");
782 case tok::kw___builtin_va_arg:
783 Res = ParseAssignmentExpression();
784 if (Res.isInvalid) {
785 SkipUntil(tok::r_paren);
786 return Res;
787 }
788
789 if (ExpectAndConsume(tok::comma, diag::err_expected_comma, "",tok::r_paren))
790 return ExprResult(true);
791
792 ParseTypeName();
Chris Lattner6eb21092007-08-30 15:52:49 +0000793
794 MatchRHSPunctuation(tok::r_paren, LParenLoc);
Reid Spencer5f016e22007-07-11 17:01:13 +0000795 break;
796
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000797 case tok::kw___builtin_offsetof: {
Chris Lattner9fddf0a2007-08-30 17:08:45 +0000798 SourceLocation TypeLoc = Tok.getLocation();
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000799 TypeTy *Ty = ParseTypeName();
Reid Spencer5f016e22007-07-11 17:01:13 +0000800
801 if (ExpectAndConsume(tok::comma, diag::err_expected_comma, "",tok::r_paren))
802 return ExprResult(true);
803
804 // We must have at least one identifier here.
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000805 if (Tok.getKind() != tok::identifier) {
806 Diag(Tok, diag::err_expected_ident);
807 SkipUntil(tok::r_paren);
808 return true;
809 }
810
811 // Keep track of the various subcomponents we see.
812 llvm::SmallVector<Action::OffsetOfComponent, 4> Comps;
813
814 Comps.push_back(Action::OffsetOfComponent());
815 Comps.back().isBrackets = false;
816 Comps.back().U.IdentInfo = Tok.getIdentifierInfo();
817 Comps.back().LocStart = Comps.back().LocEnd = ConsumeToken();
Reid Spencer5f016e22007-07-11 17:01:13 +0000818
819 while (1) {
820 if (Tok.getKind() == tok::period) {
821 // offsetof-member-designator: offsetof-member-designator '.' identifier
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000822 Comps.push_back(Action::OffsetOfComponent());
823 Comps.back().isBrackets = false;
824 Comps.back().LocStart = ConsumeToken();
Reid Spencer5f016e22007-07-11 17:01:13 +0000825
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000826 if (Tok.getKind() != tok::identifier) {
827 Diag(Tok, diag::err_expected_ident);
828 SkipUntil(tok::r_paren);
829 return true;
830 }
831 Comps.back().U.IdentInfo = Tok.getIdentifierInfo();
832 Comps.back().LocEnd = ConsumeToken();
833
Reid Spencer5f016e22007-07-11 17:01:13 +0000834 } else if (Tok.getKind() == tok::l_square) {
835 // offsetof-member-designator: offsetof-member-design '[' expression ']'
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000836 Comps.push_back(Action::OffsetOfComponent());
837 Comps.back().isBrackets = true;
838 Comps.back().LocStart = ConsumeBracket();
Reid Spencer5f016e22007-07-11 17:01:13 +0000839 Res = ParseExpression();
840 if (Res.isInvalid) {
841 SkipUntil(tok::r_paren);
842 return Res;
843 }
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000844 Comps.back().U.E = Res.Val;
Reid Spencer5f016e22007-07-11 17:01:13 +0000845
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000846 Comps.back().LocEnd =
847 MatchRHSPunctuation(tok::r_square, Comps.back().LocStart);
848 } else if (Tok.getKind() == tok::r_paren) {
Chris Lattner9fddf0a2007-08-30 17:08:45 +0000849 Res = Actions.ParseBuiltinOffsetOf(StartLoc, TypeLoc, Ty, &Comps[0],
Chris Lattner6eb21092007-08-30 15:52:49 +0000850 Comps.size(), ConsumeParen());
851 break;
Reid Spencer5f016e22007-07-11 17:01:13 +0000852 } else {
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000853 // Error occurred.
854 return ExprResult(true);
Reid Spencer5f016e22007-07-11 17:01:13 +0000855 }
856 }
857 break;
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000858 }
Steve Naroffd04fdd52007-08-03 21:21:27 +0000859 case tok::kw___builtin_choose_expr: {
860 ExprResult Cond = ParseAssignmentExpression();
861 if (Cond.isInvalid) {
862 SkipUntil(tok::r_paren);
863 return Cond;
864 }
Reid Spencer5f016e22007-07-11 17:01:13 +0000865 if (ExpectAndConsume(tok::comma, diag::err_expected_comma, "",tok::r_paren))
866 return ExprResult(true);
867
Steve Naroffd04fdd52007-08-03 21:21:27 +0000868 ExprResult Expr1 = ParseAssignmentExpression();
869 if (Expr1.isInvalid) {
870 SkipUntil(tok::r_paren);
871 return Expr1;
872 }
Reid Spencer5f016e22007-07-11 17:01:13 +0000873 if (ExpectAndConsume(tok::comma, diag::err_expected_comma, "",tok::r_paren))
874 return ExprResult(true);
875
Steve Naroffd04fdd52007-08-03 21:21:27 +0000876 ExprResult Expr2 = ParseAssignmentExpression();
877 if (Expr2.isInvalid) {
878 SkipUntil(tok::r_paren);
879 return Expr2;
880 }
881 if (Tok.getKind() != tok::r_paren) {
882 Diag(Tok, diag::err_expected_rparen);
883 return ExprResult(true);
884 }
Chris Lattner6eb21092007-08-30 15:52:49 +0000885 Res = Actions.ParseChooseExpr(StartLoc, Cond.Val, Expr1.Val, Expr2.Val,
886 ConsumeParen());
887 break;
Steve Naroffd04fdd52007-08-03 21:21:27 +0000888 }
Reid Spencer5f016e22007-07-11 17:01:13 +0000889 case tok::kw___builtin_types_compatible_p:
Steve Naroff363bcff2007-08-01 23:45:51 +0000890 TypeTy *Ty1 = ParseTypeName();
Reid Spencer5f016e22007-07-11 17:01:13 +0000891
892 if (ExpectAndConsume(tok::comma, diag::err_expected_comma, "",tok::r_paren))
893 return ExprResult(true);
894
Steve Naroff363bcff2007-08-01 23:45:51 +0000895 TypeTy *Ty2 = ParseTypeName();
896
897 if (Tok.getKind() != tok::r_paren) {
898 Diag(Tok, diag::err_expected_rparen);
899 return ExprResult(true);
900 }
Chris Lattner6eb21092007-08-30 15:52:49 +0000901 Res = Actions.ParseTypesCompatibleExpr(StartLoc, Ty1, Ty2, ConsumeParen());
902 break;
Reid Spencer5f016e22007-07-11 17:01:13 +0000903 }
904
Reid Spencer5f016e22007-07-11 17:01:13 +0000905 // These can be followed by postfix-expr pieces because they are
906 // primary-expressions.
907 return ParsePostfixExpressionSuffix(Res);
908}
909
910/// ParseParenExpression - This parses the unit that starts with a '(' token,
911/// based on what is allowed by ExprType. The actual thing parsed is returned
912/// in ExprType.
913///
914/// primary-expression: [C99 6.5.1]
915/// '(' expression ')'
916/// [GNU] '(' compound-statement ')' (if !ParenExprOnly)
917/// postfix-expression: [C99 6.5.2]
918/// '(' type-name ')' '{' initializer-list '}'
919/// '(' type-name ')' '{' initializer-list ',' '}'
920/// cast-expression: [C99 6.5.4]
921/// '(' type-name ')' cast-expression
922///
923Parser::ExprResult Parser::ParseParenExpression(ParenParseOption &ExprType,
924 TypeTy *&CastTy,
925 SourceLocation &RParenLoc) {
926 assert(Tok.getKind() == tok::l_paren && "Not a paren expr!");
927 SourceLocation OpenLoc = ConsumeParen();
Chris Lattnerab18c4c2007-07-24 16:58:17 +0000928 ExprResult Result(true);
Reid Spencer5f016e22007-07-11 17:01:13 +0000929 CastTy = 0;
930
Chris Lattner98414c12007-08-31 21:49:55 +0000931 if (ExprType >= CompoundStmt && Tok.getKind() == tok::l_brace) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000932 Diag(Tok, diag::ext_gnu_statement_expr);
Chris Lattner98414c12007-08-31 21:49:55 +0000933 Parser::StmtResult Stmt = ParseCompoundStatement(true);
Reid Spencer5f016e22007-07-11 17:01:13 +0000934 ExprType = CompoundStmt;
Chris Lattnerab18c4c2007-07-24 16:58:17 +0000935
936 // If the substmt parsed correctly, build the AST node.
937 if (!Stmt.isInvalid && Tok.getKind() == tok::r_paren)
938 Result = Actions.ParseStmtExpr(OpenLoc, Stmt.Val, Tok.getLocation());
939
Reid Spencer5f016e22007-07-11 17:01:13 +0000940 } else if (ExprType >= CompoundLiteral && isTypeSpecifierQualifier()) {
941 // Otherwise, this is a compound literal expression or cast expression.
942 TypeTy *Ty = ParseTypeName();
943
944 // Match the ')'.
945 if (Tok.getKind() == tok::r_paren)
946 RParenLoc = ConsumeParen();
947 else
948 MatchRHSPunctuation(tok::r_paren, OpenLoc);
949
950 if (Tok.getKind() == tok::l_brace) {
951 if (!getLang().C99) // Compound literals don't exist in C90.
952 Diag(OpenLoc, diag::ext_c99_compound_literal);
953 Result = ParseInitializer();
954 ExprType = CompoundLiteral;
Steve Naroff4aa88f82007-07-19 01:06:55 +0000955 if (!Result.isInvalid)
956 return Actions.ParseCompoundLiteral(OpenLoc, Ty, RParenLoc, Result.Val);
Reid Spencer5f016e22007-07-11 17:01:13 +0000957 } else if (ExprType == CastExpr) {
958 // Note that this doesn't parse the subsequence cast-expression, it just
959 // returns the parsed type to the callee.
960 ExprType = CastExpr;
961 CastTy = Ty;
962 return ExprResult(false);
963 } else {
964 Diag(Tok, diag::err_expected_lbrace_in_compound_literal);
965 return ExprResult(true);
966 }
967 return Result;
968 } else {
969 Result = ParseExpression();
970 ExprType = SimpleExpr;
971 if (!Result.isInvalid && Tok.getKind() == tok::r_paren)
972 Result = Actions.ParseParenExpr(OpenLoc, Tok.getLocation(), Result.Val);
973 }
974
975 // Match the ')'.
976 if (Result.isInvalid)
977 SkipUntil(tok::r_paren);
978 else {
979 if (Tok.getKind() == tok::r_paren)
980 RParenLoc = ConsumeParen();
981 else
982 MatchRHSPunctuation(tok::r_paren, OpenLoc);
983 }
984
985 return Result;
986}
987
988/// ParseStringLiteralExpression - This handles the various token types that
989/// form string literals, and also handles string concatenation [C99 5.1.1.2,
990/// translation phase #6].
991///
992/// primary-expression: [C99 6.5.1]
993/// string-literal
994Parser::ExprResult Parser::ParseStringLiteralExpression() {
995 assert(isTokenStringLiteral() && "Not a string literal!");
996
997 // String concat. Note that keywords like __func__ and __FUNCTION__ are not
998 // considered to be strings for concatenation purposes.
Chris Lattnerd2177732007-07-20 16:59:19 +0000999 llvm::SmallVector<Token, 4> StringToks;
Reid Spencer5f016e22007-07-11 17:01:13 +00001000
1001 do {
1002 StringToks.push_back(Tok);
1003 ConsumeStringToken();
1004 } while (isTokenStringLiteral());
1005
1006 // Pass the set of string tokens, ready for concatenation, to the actions.
1007 return Actions.ParseStringLiteral(&StringToks[0], StringToks.size());
1008}