blob: 3e70db4c480ba40053b16316609e6cb60a64cefc [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
377 // Combine the LHS and RHS into the LHS (e.g. build AST).
378 if (TernaryMiddle.isInvalid)
379 LHS = Actions.ParseBinOp(OpToken.getLocation(), OpToken.getKind(),
380 LHS.Val, RHS.Val);
381 else
382 LHS = Actions.ParseConditionalOp(OpToken.getLocation(), ColonLoc,
383 LHS.Val, TernaryMiddle.Val, RHS.Val);
Chris Lattnerdbd583c2007-08-31 04:58:34 +0000384 if (LHS.isInvalid)
385 return LHS;
Reid Spencer5f016e22007-07-11 17:01:13 +0000386 }
387}
388
389/// ParseCastExpression - Parse a cast-expression, or, if isUnaryExpression is
390/// true, parse a unary-expression.
391///
392/// cast-expression: [C99 6.5.4]
393/// unary-expression
394/// '(' type-name ')' cast-expression
395///
396/// unary-expression: [C99 6.5.3]
397/// postfix-expression
398/// '++' unary-expression
399/// '--' unary-expression
400/// unary-operator cast-expression
401/// 'sizeof' unary-expression
402/// 'sizeof' '(' type-name ')'
403/// [GNU] '__alignof' unary-expression
404/// [GNU] '__alignof' '(' type-name ')'
405/// [GNU] '&&' identifier
406///
407/// unary-operator: one of
408/// '&' '*' '+' '-' '~' '!'
409/// [GNU] '__extension__' '__real' '__imag'
410///
411/// primary-expression: [C99 6.5.1]
412/// identifier
413/// constant
414/// string-literal
415/// [C++] boolean-literal [C++ 2.13.5]
416/// '(' expression ')'
417/// '__func__' [C99 6.4.2.2]
418/// [GNU] '__FUNCTION__'
419/// [GNU] '__PRETTY_FUNCTION__'
420/// [GNU] '(' compound-statement ')'
421/// [GNU] '__builtin_va_arg' '(' assignment-expression ',' type-name ')'
422/// [GNU] '__builtin_offsetof' '(' type-name ',' offsetof-member-designator')'
423/// [GNU] '__builtin_choose_expr' '(' assign-expr ',' assign-expr ','
424/// assign-expr ')'
425/// [GNU] '__builtin_types_compatible_p' '(' type-name ',' type-name ')'
426/// [OBC] '[' objc-receiver objc-message-args ']' [TODO]
427/// [OBC] '@selector' '(' objc-selector-arg ')' [TODO]
428/// [OBC] '@protocol' '(' identifier ')' [TODO]
429/// [OBC] '@encode' '(' type-name ')' [TODO]
Anders Carlsson5dac7f62007-08-21 19:18:49 +0000430/// [OBC] objc-string-literal
Reid Spencer5f016e22007-07-11 17:01:13 +0000431/// [C++] 'const_cast' '<' type-name '>' '(' expression ')' [C++ 5.2p1]
432/// [C++] 'dynamic_cast' '<' type-name '>' '(' expression ')' [C++ 5.2p1]
433/// [C++] 'reinterpret_cast' '<' type-name '>' '(' expression ')' [C++ 5.2p1]
434/// [C++] 'static_cast' '<' type-name '>' '(' expression ')' [C++ 5.2p1]
435///
436/// constant: [C99 6.4.4]
437/// integer-constant
438/// floating-constant
439/// enumeration-constant -> identifier
440/// character-constant
441///
442Parser::ExprResult Parser::ParseCastExpression(bool isUnaryExpression) {
443 ExprResult Res;
444 tok::TokenKind SavedKind = Tok.getKind();
445
446 // This handles all of cast-expression, unary-expression, postfix-expression,
447 // and primary-expression. We handle them together like this for efficiency
448 // and to simplify handling of an expression starting with a '(' token: which
449 // may be one of a parenthesized expression, cast-expression, compound literal
450 // expression, or statement expression.
451 //
452 // If the parsed tokens consist of a primary-expression, the cases below
453 // call ParsePostfixExpressionSuffix to handle the postfix expression
454 // suffixes. Cases that cannot be followed by postfix exprs should
455 // return without invoking ParsePostfixExpressionSuffix.
456 switch (SavedKind) {
457 case tok::l_paren: {
458 // If this expression is limited to being a unary-expression, the parent can
459 // not start a cast expression.
460 ParenParseOption ParenExprType =
461 isUnaryExpression ? CompoundLiteral : CastExpr;
462 TypeTy *CastTy;
463 SourceLocation LParenLoc = Tok.getLocation();
464 SourceLocation RParenLoc;
465 Res = ParseParenExpression(ParenExprType, CastTy, RParenLoc);
466 if (Res.isInvalid) return Res;
467
468 switch (ParenExprType) {
469 case SimpleExpr: break; // Nothing else to do.
470 case CompoundStmt: break; // Nothing else to do.
471 case CompoundLiteral:
472 // We parsed '(' type-name ')' '{' ... '}'. If any suffixes of
473 // postfix-expression exist, parse them now.
474 break;
475 case CastExpr:
476 // We parsed '(' type-name ')' and the thing after it wasn't a '{'. Parse
477 // the cast-expression that follows it next.
478 // TODO: For cast expression with CastTy.
479 Res = ParseCastExpression(false);
480 if (!Res.isInvalid)
481 Res = Actions.ParseCastExpr(LParenLoc, CastTy, RParenLoc, Res.Val);
482 return Res;
483 }
484
485 // These can be followed by postfix-expr pieces.
486 return ParsePostfixExpressionSuffix(Res);
487 }
488
489 // primary-expression
490 case tok::numeric_constant:
491 // constant: integer-constant
492 // constant: floating-constant
493
494 Res = Actions.ParseNumericConstant(Tok);
495 ConsumeToken();
496
497 // These can be followed by postfix-expr pieces.
498 return ParsePostfixExpressionSuffix(Res);
499
500 case tok::kw_true:
501 case tok::kw_false:
502 return ParseCXXBoolLiteral();
503
504 case tok::identifier: { // primary-expression: identifier
505 // constant: enumeration-constant
506 // Consume the identifier so that we can see if it is followed by a '('.
507 // Function designators are allowed to be undeclared (C99 6.5.1p2), so we
508 // need to know whether or not this identifier is a function designator or
509 // not.
510 IdentifierInfo &II = *Tok.getIdentifierInfo();
511 SourceLocation L = ConsumeToken();
512 Res = Actions.ParseIdentifierExpr(CurScope, L, II,
513 Tok.getKind() == tok::l_paren);
514 // These can be followed by postfix-expr pieces.
515 return ParsePostfixExpressionSuffix(Res);
516 }
517 case tok::char_constant: // constant: character-constant
518 Res = Actions.ParseCharacterConstant(Tok);
519 ConsumeToken();
520 // These can be followed by postfix-expr pieces.
521 return ParsePostfixExpressionSuffix(Res);
522 case tok::kw___func__: // primary-expression: __func__ [C99 6.4.2.2]
523 case tok::kw___FUNCTION__: // primary-expression: __FUNCTION__ [GNU]
524 case tok::kw___PRETTY_FUNCTION__: // primary-expression: __P..Y_F..N__ [GNU]
Anders Carlsson22742662007-07-21 05:21:51 +0000525 Res = Actions.ParsePreDefinedExpr(Tok.getLocation(), SavedKind);
Reid Spencer5f016e22007-07-11 17:01:13 +0000526 ConsumeToken();
527 // These can be followed by postfix-expr pieces.
528 return ParsePostfixExpressionSuffix(Res);
529 case tok::string_literal: // primary-expression: string-literal
530 case tok::wide_string_literal:
531 Res = ParseStringLiteralExpression();
532 if (Res.isInvalid) return Res;
533 // This can be followed by postfix-expr pieces (e.g. "foo"[1]).
534 return ParsePostfixExpressionSuffix(Res);
535 case tok::kw___builtin_va_arg:
536 case tok::kw___builtin_offsetof:
537 case tok::kw___builtin_choose_expr:
538 case tok::kw___builtin_types_compatible_p:
539 return ParseBuiltinPrimaryExpression();
540 case tok::plusplus: // unary-expression: '++' unary-expression
541 case tok::minusminus: { // unary-expression: '--' unary-expression
542 SourceLocation SavedLoc = ConsumeToken();
543 Res = ParseCastExpression(true);
544 if (!Res.isInvalid)
545 Res = Actions.ParseUnaryOp(SavedLoc, SavedKind, Res.Val);
546 return Res;
547 }
548 case tok::amp: // unary-expression: '&' cast-expression
549 case tok::star: // unary-expression: '*' cast-expression
550 case tok::plus: // unary-expression: '+' cast-expression
551 case tok::minus: // unary-expression: '-' cast-expression
552 case tok::tilde: // unary-expression: '~' cast-expression
553 case tok::exclaim: // unary-expression: '!' cast-expression
554 case tok::kw___real: // unary-expression: '__real' cast-expression [GNU]
555 case tok::kw___imag: // unary-expression: '__imag' cast-expression [GNU]
556 case tok::kw___extension__:{//unary-expression:'__extension__' cast-expr [GNU]
Chris Lattnerab18c4c2007-07-24 16:58:17 +0000557 // FIXME: Extension should silence extwarns in subexpressions.
Reid Spencer5f016e22007-07-11 17:01:13 +0000558 SourceLocation SavedLoc = ConsumeToken();
559 Res = ParseCastExpression(false);
560 if (!Res.isInvalid)
561 Res = Actions.ParseUnaryOp(SavedLoc, SavedKind, Res.Val);
562 return Res;
563 }
564 case tok::kw_sizeof: // unary-expression: 'sizeof' unary-expression
565 // unary-expression: 'sizeof' '(' type-name ')'
566 case tok::kw___alignof: // unary-expression: '__alignof' unary-expression
567 // unary-expression: '__alignof' '(' type-name ')'
568 return ParseSizeofAlignofExpression();
569 case tok::ampamp: { // unary-expression: '&&' identifier
570 SourceLocation AmpAmpLoc = ConsumeToken();
571 if (Tok.getKind() != tok::identifier) {
572 Diag(Tok, diag::err_expected_ident);
573 return ExprResult(true);
574 }
575
576 Diag(AmpAmpLoc, diag::ext_gnu_address_of_label);
577 Res = Actions.ParseAddrLabel(AmpAmpLoc, Tok.getLocation(),
578 Tok.getIdentifierInfo());
579 ConsumeToken();
580 return Res;
581 }
582 case tok::kw_const_cast:
583 case tok::kw_dynamic_cast:
584 case tok::kw_reinterpret_cast:
585 case tok::kw_static_cast:
586 return ParseCXXCasts();
Anders Carlsson55085182007-08-21 17:43:55 +0000587 case tok::at:
588 return ParseObjCExpression();
Reid Spencer5f016e22007-07-11 17:01:13 +0000589 default:
590 Diag(Tok, diag::err_expected_expression);
591 return ExprResult(true);
592 }
593
594 // unreachable.
595 abort();
596}
597
598/// ParsePostfixExpressionSuffix - Once the leading part of a postfix-expression
599/// is parsed, this method parses any suffixes that apply.
600///
601/// postfix-expression: [C99 6.5.2]
602/// primary-expression
603/// postfix-expression '[' expression ']'
604/// postfix-expression '(' argument-expression-list[opt] ')'
605/// postfix-expression '.' identifier
606/// postfix-expression '->' identifier
607/// postfix-expression '++'
608/// postfix-expression '--'
609/// '(' type-name ')' '{' initializer-list '}'
610/// '(' type-name ')' '{' initializer-list ',' '}'
611///
612/// argument-expression-list: [C99 6.5.2]
613/// argument-expression
614/// argument-expression-list ',' assignment-expression
615///
616Parser::ExprResult Parser::ParsePostfixExpressionSuffix(ExprResult LHS) {
617
618 // Now that the primary-expression piece of the postfix-expression has been
619 // parsed, see if there are any postfix-expression pieces here.
620 SourceLocation Loc;
621 while (1) {
622 switch (Tok.getKind()) {
623 default: // Not a postfix-expression suffix.
624 return LHS;
625 case tok::l_square: { // postfix-expression: p-e '[' expression ']'
626 Loc = ConsumeBracket();
627 ExprResult Idx = ParseExpression();
628
629 SourceLocation RLoc = Tok.getLocation();
630
631 if (!LHS.isInvalid && !Idx.isInvalid && Tok.getKind() == tok::r_square)
632 LHS = Actions.ParseArraySubscriptExpr(LHS.Val, Loc, Idx.Val, RLoc);
633 else
634 LHS = ExprResult(true);
635
636 // Match the ']'.
637 MatchRHSPunctuation(tok::r_square, Loc);
638 break;
639 }
640
641 case tok::l_paren: { // p-e: p-e '(' argument-expression-list[opt] ')'
642 llvm::SmallVector<ExprTy*, 8> ArgExprs;
643 llvm::SmallVector<SourceLocation, 8> CommaLocs;
Reid Spencer5f016e22007-07-11 17:01:13 +0000644
645 Loc = ConsumeParen();
646
647 if (Tok.getKind() != tok::r_paren) {
648 while (1) {
649 ExprResult ArgExpr = ParseAssignmentExpression();
650 if (ArgExpr.isInvalid) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000651 SkipUntil(tok::r_paren);
Chris Lattner2ff54262007-07-21 05:18:12 +0000652 return ExprResult(true);
Reid Spencer5f016e22007-07-11 17:01:13 +0000653 } else
654 ArgExprs.push_back(ArgExpr.Val);
655
656 if (Tok.getKind() != tok::comma)
657 break;
658 // Move to the next argument, remember where the comma was.
659 CommaLocs.push_back(ConsumeToken());
660 }
661 }
662
663 // Match the ')'.
Chris Lattner2ff54262007-07-21 05:18:12 +0000664 if (!LHS.isInvalid && Tok.getKind() == tok::r_paren) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000665 assert((ArgExprs.size() == 0 || ArgExprs.size()-1 == CommaLocs.size())&&
666 "Unexpected number of commas!");
667 LHS = Actions.ParseCallExpr(LHS.Val, Loc, &ArgExprs[0], ArgExprs.size(),
668 &CommaLocs[0], Tok.getLocation());
669 }
670
Chris Lattner2ff54262007-07-21 05:18:12 +0000671 MatchRHSPunctuation(tok::r_paren, Loc);
Reid Spencer5f016e22007-07-11 17:01:13 +0000672 break;
673 }
674 case tok::arrow: // postfix-expression: p-e '->' identifier
675 case tok::period: { // postfix-expression: p-e '.' identifier
676 tok::TokenKind OpKind = Tok.getKind();
677 SourceLocation OpLoc = ConsumeToken(); // Eat the "." or "->" token.
678
679 if (Tok.getKind() != tok::identifier) {
680 Diag(Tok, diag::err_expected_ident);
681 return ExprResult(true);
682 }
683
684 if (!LHS.isInvalid)
685 LHS = Actions.ParseMemberReferenceExpr(LHS.Val, OpLoc, OpKind,
686 Tok.getLocation(),
687 *Tok.getIdentifierInfo());
688 ConsumeToken();
689 break;
690 }
691 case tok::plusplus: // postfix-expression: postfix-expression '++'
692 case tok::minusminus: // postfix-expression: postfix-expression '--'
693 if (!LHS.isInvalid)
694 LHS = Actions.ParsePostfixUnaryOp(Tok.getLocation(), Tok.getKind(),
695 LHS.Val);
696 ConsumeToken();
697 break;
698 }
699 }
700}
701
702
703/// ParseSizeofAlignofExpression - Parse a sizeof or alignof expression.
704/// unary-expression: [C99 6.5.3]
705/// 'sizeof' unary-expression
706/// 'sizeof' '(' type-name ')'
707/// [GNU] '__alignof' unary-expression
708/// [GNU] '__alignof' '(' type-name ')'
709Parser::ExprResult Parser::ParseSizeofAlignofExpression() {
710 assert((Tok.getKind() == tok::kw_sizeof ||
711 Tok.getKind() == tok::kw___alignof) &&
712 "Not a sizeof/alignof expression!");
Chris Lattnerd2177732007-07-20 16:59:19 +0000713 Token OpTok = Tok;
Reid Spencer5f016e22007-07-11 17:01:13 +0000714 ConsumeToken();
715
716 // If the operand doesn't start with an '(', it must be an expression.
717 ExprResult Operand;
718 if (Tok.getKind() != 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.ParseSizeOfAlignOfTypeExpr(OpTok.getLocation(),
734 OpTok.getKind() == tok::kw_sizeof,
735 LParenLoc, CastTy, RParenLoc);
736 }
737 }
738
739 // If we get here, the operand to the sizeof/alignof was an expresion.
740 if (!Operand.isInvalid)
741 Operand = Actions.ParseUnaryOp(OpTok.getLocation(), OpTok.getKind(),
742 Operand.Val);
743 return Operand;
744}
745
746/// ParseBuiltinPrimaryExpression
747///
748/// primary-expression: [C99 6.5.1]
749/// [GNU] '__builtin_va_arg' '(' assignment-expression ',' type-name ')'
750/// [GNU] '__builtin_offsetof' '(' type-name ',' offsetof-member-designator')'
751/// [GNU] '__builtin_choose_expr' '(' assign-expr ',' assign-expr ','
752/// assign-expr ')'
753/// [GNU] '__builtin_types_compatible_p' '(' type-name ',' type-name ')'
754///
755/// [GNU] offsetof-member-designator:
756/// [GNU] identifier
757/// [GNU] offsetof-member-designator '.' identifier
758/// [GNU] offsetof-member-designator '[' expression ']'
759///
760Parser::ExprResult Parser::ParseBuiltinPrimaryExpression() {
761 ExprResult Res(false);
762 const IdentifierInfo *BuiltinII = Tok.getIdentifierInfo();
763
764 tok::TokenKind T = Tok.getKind();
765 SourceLocation StartLoc = ConsumeToken(); // Eat the builtin identifier.
766
767 // All of these start with an open paren.
768 if (Tok.getKind() != tok::l_paren) {
769 Diag(Tok, diag::err_expected_lparen_after, BuiltinII->getName());
770 return ExprResult(true);
771 }
772
773 SourceLocation LParenLoc = ConsumeParen();
774 // TODO: Build AST.
775
776 switch (T) {
777 default: assert(0 && "Not a builtin primary expression!");
778 case tok::kw___builtin_va_arg:
779 Res = ParseAssignmentExpression();
780 if (Res.isInvalid) {
781 SkipUntil(tok::r_paren);
782 return Res;
783 }
784
785 if (ExpectAndConsume(tok::comma, diag::err_expected_comma, "",tok::r_paren))
786 return ExprResult(true);
787
788 ParseTypeName();
Chris Lattner6eb21092007-08-30 15:52:49 +0000789
790 MatchRHSPunctuation(tok::r_paren, LParenLoc);
Reid Spencer5f016e22007-07-11 17:01:13 +0000791 break;
792
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000793 case tok::kw___builtin_offsetof: {
Chris Lattner9fddf0a2007-08-30 17:08:45 +0000794 SourceLocation TypeLoc = Tok.getLocation();
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000795 TypeTy *Ty = ParseTypeName();
Reid Spencer5f016e22007-07-11 17:01:13 +0000796
797 if (ExpectAndConsume(tok::comma, diag::err_expected_comma, "",tok::r_paren))
798 return ExprResult(true);
799
800 // We must have at least one identifier here.
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000801 if (Tok.getKind() != tok::identifier) {
802 Diag(Tok, diag::err_expected_ident);
803 SkipUntil(tok::r_paren);
804 return true;
805 }
806
807 // Keep track of the various subcomponents we see.
808 llvm::SmallVector<Action::OffsetOfComponent, 4> Comps;
809
810 Comps.push_back(Action::OffsetOfComponent());
811 Comps.back().isBrackets = false;
812 Comps.back().U.IdentInfo = Tok.getIdentifierInfo();
813 Comps.back().LocStart = Comps.back().LocEnd = ConsumeToken();
Reid Spencer5f016e22007-07-11 17:01:13 +0000814
815 while (1) {
816 if (Tok.getKind() == tok::period) {
817 // offsetof-member-designator: offsetof-member-designator '.' identifier
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000818 Comps.push_back(Action::OffsetOfComponent());
819 Comps.back().isBrackets = false;
820 Comps.back().LocStart = ConsumeToken();
Reid Spencer5f016e22007-07-11 17:01:13 +0000821
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000822 if (Tok.getKind() != tok::identifier) {
823 Diag(Tok, diag::err_expected_ident);
824 SkipUntil(tok::r_paren);
825 return true;
826 }
827 Comps.back().U.IdentInfo = Tok.getIdentifierInfo();
828 Comps.back().LocEnd = ConsumeToken();
829
Reid Spencer5f016e22007-07-11 17:01:13 +0000830 } else if (Tok.getKind() == tok::l_square) {
831 // offsetof-member-designator: offsetof-member-design '[' expression ']'
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000832 Comps.push_back(Action::OffsetOfComponent());
833 Comps.back().isBrackets = true;
834 Comps.back().LocStart = ConsumeBracket();
Reid Spencer5f016e22007-07-11 17:01:13 +0000835 Res = ParseExpression();
836 if (Res.isInvalid) {
837 SkipUntil(tok::r_paren);
838 return Res;
839 }
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000840 Comps.back().U.E = Res.Val;
Reid Spencer5f016e22007-07-11 17:01:13 +0000841
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000842 Comps.back().LocEnd =
843 MatchRHSPunctuation(tok::r_square, Comps.back().LocStart);
844 } else if (Tok.getKind() == tok::r_paren) {
Chris Lattner9fddf0a2007-08-30 17:08:45 +0000845 Res = Actions.ParseBuiltinOffsetOf(StartLoc, TypeLoc, Ty, &Comps[0],
Chris Lattner6eb21092007-08-30 15:52:49 +0000846 Comps.size(), ConsumeParen());
847 break;
Reid Spencer5f016e22007-07-11 17:01:13 +0000848 } else {
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000849 // Error occurred.
850 return ExprResult(true);
Reid Spencer5f016e22007-07-11 17:01:13 +0000851 }
852 }
853 break;
Chris Lattnerf9aa3cb2007-08-30 15:51:11 +0000854 }
Steve Naroffd04fdd52007-08-03 21:21:27 +0000855 case tok::kw___builtin_choose_expr: {
856 ExprResult Cond = ParseAssignmentExpression();
857 if (Cond.isInvalid) {
858 SkipUntil(tok::r_paren);
859 return Cond;
860 }
Reid Spencer5f016e22007-07-11 17:01:13 +0000861 if (ExpectAndConsume(tok::comma, diag::err_expected_comma, "",tok::r_paren))
862 return ExprResult(true);
863
Steve Naroffd04fdd52007-08-03 21:21:27 +0000864 ExprResult Expr1 = ParseAssignmentExpression();
865 if (Expr1.isInvalid) {
866 SkipUntil(tok::r_paren);
867 return Expr1;
868 }
Reid Spencer5f016e22007-07-11 17:01:13 +0000869 if (ExpectAndConsume(tok::comma, diag::err_expected_comma, "",tok::r_paren))
870 return ExprResult(true);
871
Steve Naroffd04fdd52007-08-03 21:21:27 +0000872 ExprResult Expr2 = ParseAssignmentExpression();
873 if (Expr2.isInvalid) {
874 SkipUntil(tok::r_paren);
875 return Expr2;
876 }
877 if (Tok.getKind() != tok::r_paren) {
878 Diag(Tok, diag::err_expected_rparen);
879 return ExprResult(true);
880 }
Chris Lattner6eb21092007-08-30 15:52:49 +0000881 Res = Actions.ParseChooseExpr(StartLoc, Cond.Val, Expr1.Val, Expr2.Val,
882 ConsumeParen());
883 break;
Steve Naroffd04fdd52007-08-03 21:21:27 +0000884 }
Reid Spencer5f016e22007-07-11 17:01:13 +0000885 case tok::kw___builtin_types_compatible_p:
Steve Naroff363bcff2007-08-01 23:45:51 +0000886 TypeTy *Ty1 = ParseTypeName();
Reid Spencer5f016e22007-07-11 17:01:13 +0000887
888 if (ExpectAndConsume(tok::comma, diag::err_expected_comma, "",tok::r_paren))
889 return ExprResult(true);
890
Steve Naroff363bcff2007-08-01 23:45:51 +0000891 TypeTy *Ty2 = ParseTypeName();
892
893 if (Tok.getKind() != tok::r_paren) {
894 Diag(Tok, diag::err_expected_rparen);
895 return ExprResult(true);
896 }
Chris Lattner6eb21092007-08-30 15:52:49 +0000897 Res = Actions.ParseTypesCompatibleExpr(StartLoc, Ty1, Ty2, ConsumeParen());
898 break;
Reid Spencer5f016e22007-07-11 17:01:13 +0000899 }
900
Reid Spencer5f016e22007-07-11 17:01:13 +0000901 // These can be followed by postfix-expr pieces because they are
902 // primary-expressions.
903 return ParsePostfixExpressionSuffix(Res);
904}
905
906/// ParseParenExpression - This parses the unit that starts with a '(' token,
907/// based on what is allowed by ExprType. The actual thing parsed is returned
908/// in ExprType.
909///
910/// primary-expression: [C99 6.5.1]
911/// '(' expression ')'
912/// [GNU] '(' compound-statement ')' (if !ParenExprOnly)
913/// postfix-expression: [C99 6.5.2]
914/// '(' type-name ')' '{' initializer-list '}'
915/// '(' type-name ')' '{' initializer-list ',' '}'
916/// cast-expression: [C99 6.5.4]
917/// '(' type-name ')' cast-expression
918///
919Parser::ExprResult Parser::ParseParenExpression(ParenParseOption &ExprType,
920 TypeTy *&CastTy,
921 SourceLocation &RParenLoc) {
922 assert(Tok.getKind() == tok::l_paren && "Not a paren expr!");
923 SourceLocation OpenLoc = ConsumeParen();
Chris Lattnerab18c4c2007-07-24 16:58:17 +0000924 ExprResult Result(true);
Reid Spencer5f016e22007-07-11 17:01:13 +0000925 CastTy = 0;
926
927 if (ExprType >= CompoundStmt && Tok.getKind() == tok::l_brace &&
928 !getLang().NoExtensions) {
929 Diag(Tok, diag::ext_gnu_statement_expr);
Chris Lattnerab18c4c2007-07-24 16:58:17 +0000930 Parser::StmtResult Stmt = ParseCompoundStatement();
Reid Spencer5f016e22007-07-11 17:01:13 +0000931 ExprType = CompoundStmt;
Chris Lattnerab18c4c2007-07-24 16:58:17 +0000932
933 // If the substmt parsed correctly, build the AST node.
934 if (!Stmt.isInvalid && Tok.getKind() == tok::r_paren)
935 Result = Actions.ParseStmtExpr(OpenLoc, Stmt.Val, Tok.getLocation());
936
Reid Spencer5f016e22007-07-11 17:01:13 +0000937 } else if (ExprType >= CompoundLiteral && isTypeSpecifierQualifier()) {
938 // Otherwise, this is a compound literal expression or cast expression.
939 TypeTy *Ty = ParseTypeName();
940
941 // Match the ')'.
942 if (Tok.getKind() == tok::r_paren)
943 RParenLoc = ConsumeParen();
944 else
945 MatchRHSPunctuation(tok::r_paren, OpenLoc);
946
947 if (Tok.getKind() == tok::l_brace) {
948 if (!getLang().C99) // Compound literals don't exist in C90.
949 Diag(OpenLoc, diag::ext_c99_compound_literal);
950 Result = ParseInitializer();
951 ExprType = CompoundLiteral;
Steve Naroff4aa88f82007-07-19 01:06:55 +0000952 if (!Result.isInvalid)
953 return Actions.ParseCompoundLiteral(OpenLoc, Ty, RParenLoc, Result.Val);
Reid Spencer5f016e22007-07-11 17:01:13 +0000954 } else if (ExprType == CastExpr) {
955 // Note that this doesn't parse the subsequence cast-expression, it just
956 // returns the parsed type to the callee.
957 ExprType = CastExpr;
958 CastTy = Ty;
959 return ExprResult(false);
960 } else {
961 Diag(Tok, diag::err_expected_lbrace_in_compound_literal);
962 return ExprResult(true);
963 }
964 return Result;
965 } else {
966 Result = ParseExpression();
967 ExprType = SimpleExpr;
968 if (!Result.isInvalid && Tok.getKind() == tok::r_paren)
969 Result = Actions.ParseParenExpr(OpenLoc, Tok.getLocation(), Result.Val);
970 }
971
972 // Match the ')'.
973 if (Result.isInvalid)
974 SkipUntil(tok::r_paren);
975 else {
976 if (Tok.getKind() == tok::r_paren)
977 RParenLoc = ConsumeParen();
978 else
979 MatchRHSPunctuation(tok::r_paren, OpenLoc);
980 }
981
982 return Result;
983}
984
985/// ParseStringLiteralExpression - This handles the various token types that
986/// form string literals, and also handles string concatenation [C99 5.1.1.2,
987/// translation phase #6].
988///
989/// primary-expression: [C99 6.5.1]
990/// string-literal
991Parser::ExprResult Parser::ParseStringLiteralExpression() {
992 assert(isTokenStringLiteral() && "Not a string literal!");
993
994 // String concat. Note that keywords like __func__ and __FUNCTION__ are not
995 // considered to be strings for concatenation purposes.
Chris Lattnerd2177732007-07-20 16:59:19 +0000996 llvm::SmallVector<Token, 4> StringToks;
Reid Spencer5f016e22007-07-11 17:01:13 +0000997
998 do {
999 StringToks.push_back(Tok);
1000 ConsumeStringToken();
1001 } while (isTokenStringLiteral());
1002
1003 // Pass the set of string tokens, ready for concatenation, to the actions.
1004 return Actions.ParseStringLiteral(&StringToks[0], StringToks.size());
1005}