1 //===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements extra semantic analysis beyond what is enforced
11 // by the C type system.
13 //===----------------------------------------------------------------------===//
16 #include "clang/AST/ASTContext.h"
17 #include "clang/AST/DeclObjC.h"
18 #include "clang/AST/ExprCXX.h"
19 #include "clang/AST/ExprObjC.h"
20 #include "clang/Lex/LiteralSupport.h"
21 #include "clang/Lex/Preprocessor.h"
23 using namespace clang;
25 /// getLocationOfStringLiteralByte - Return a source location that points to the
26 /// specified byte of the specified string literal.
28 /// Strings are amazingly complex. They can be formed from multiple tokens and
29 /// can have escape sequences in them in addition to the usual trigraph and
30 /// escaped newline business. This routine handles this complexity.
32 SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL,
33 unsigned ByteNo) const {
34 assert(!SL->isWide() && "This doesn't work for wide strings yet");
36 // Loop over all of the tokens in this string until we find the one that
37 // contains the byte we're looking for.
40 assert(TokNo < SL->getNumConcatenated() && "Invalid byte number!");
41 SourceLocation StrTokLoc = SL->getStrTokenLoc(TokNo);
43 // Get the spelling of the string so that we can get the data that makes up
44 // the string literal, not the identifier for the macro it is potentially
46 SourceLocation StrTokSpellingLoc = SourceMgr.getSpellingLoc(StrTokLoc);
48 // Re-lex the token to get its length and original spelling.
49 std::pair<FileID, unsigned> LocInfo =
50 SourceMgr.getDecomposedLoc(StrTokSpellingLoc);
51 std::pair<const char *,const char *> Buffer =
52 SourceMgr.getBufferData(LocInfo.first);
53 const char *StrData = Buffer.first+LocInfo.second;
55 // Create a langops struct and enable trigraphs. This is sufficient for
58 LangOpts.Trigraphs = true;
60 // Create a lexer starting at the beginning of this token.
61 Lexer TheLexer(StrTokSpellingLoc, LangOpts, Buffer.first, StrData,
64 TheLexer.LexFromRawLexer(TheTok);
66 // Use the StringLiteralParser to compute the length of the string in bytes.
67 StringLiteralParser SLP(&TheTok, 1, PP);
68 unsigned TokNumBytes = SLP.GetStringLength();
70 // If the byte is in this token, return the location of the byte.
71 if (ByteNo < TokNumBytes ||
72 (ByteNo == TokNumBytes && TokNo == SL->getNumConcatenated())) {
74 StringLiteralParser::getOffsetOfStringByte(TheTok, ByteNo, PP);
76 // Now that we know the offset of the token in the spelling, use the
77 // preprocessor to get the offset in the original source.
78 return PP.AdvanceToTokenCharacter(StrTokLoc, Offset);
81 // Move to the next string token.
83 ByteNo -= TokNumBytes;
87 /// CheckablePrintfAttr - does a function call have a "printf" attribute
88 /// and arguments that merit checking?
89 bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) {
90 if (Format->getType() == "printf") return true;
91 if (Format->getType() == "printf0") {
92 // printf0 allows null "format" string; if so don't check format/args
93 unsigned format_idx = Format->getFormatIdx() - 1;
94 // Does the index refer to the implicit object argument?
95 if (isa<CXXMemberCallExpr>(TheCall)) {
100 if (format_idx < TheCall->getNumArgs()) {
101 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts();
102 if (!Format->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNull))
109 Action::OwningExprResult
110 Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
111 OwningExprResult TheCallResult(Owned(TheCall));
114 case Builtin::BI__builtin___CFStringMakeConstantString:
115 assert(TheCall->getNumArgs() == 1 &&
116 "Wrong # arguments to builtin CFStringMakeConstantString");
117 if (CheckObjCString(TheCall->getArg(0)))
120 case Builtin::BI__builtin_stdarg_start:
121 case Builtin::BI__builtin_va_start:
122 if (SemaBuiltinVAStart(TheCall))
125 case Builtin::BI__builtin_isgreater:
126 case Builtin::BI__builtin_isgreaterequal:
127 case Builtin::BI__builtin_isless:
128 case Builtin::BI__builtin_islessequal:
129 case Builtin::BI__builtin_islessgreater:
130 case Builtin::BI__builtin_isunordered:
131 if (SemaBuiltinUnorderedCompare(TheCall))
134 case Builtin::BI__builtin_isfinite:
135 case Builtin::BI__builtin_isinf:
136 case Builtin::BI__builtin_isinf_sign:
137 case Builtin::BI__builtin_isnan:
138 case Builtin::BI__builtin_isnormal:
139 if (SemaBuiltinUnaryFP(TheCall))
142 case Builtin::BI__builtin_return_address:
143 case Builtin::BI__builtin_frame_address:
144 if (SemaBuiltinStackAddress(TheCall))
147 case Builtin::BI__builtin_eh_return_data_regno:
148 if (SemaBuiltinEHReturnDataRegNo(TheCall))
151 case Builtin::BI__builtin_shufflevector:
152 return SemaBuiltinShuffleVector(TheCall);
153 // TheCall will be freed by the smart pointer here, but that's fine, since
154 // SemaBuiltinShuffleVector guts it, but then doesn't release it.
155 case Builtin::BI__builtin_prefetch:
156 if (SemaBuiltinPrefetch(TheCall))
159 case Builtin::BI__builtin_object_size:
160 if (SemaBuiltinObjectSize(TheCall))
163 case Builtin::BI__builtin_longjmp:
164 if (SemaBuiltinLongjmp(TheCall))
167 case Builtin::BI__sync_fetch_and_add:
168 case Builtin::BI__sync_fetch_and_sub:
169 case Builtin::BI__sync_fetch_and_or:
170 case Builtin::BI__sync_fetch_and_and:
171 case Builtin::BI__sync_fetch_and_xor:
172 case Builtin::BI__sync_fetch_and_nand:
173 case Builtin::BI__sync_add_and_fetch:
174 case Builtin::BI__sync_sub_and_fetch:
175 case Builtin::BI__sync_and_and_fetch:
176 case Builtin::BI__sync_or_and_fetch:
177 case Builtin::BI__sync_xor_and_fetch:
178 case Builtin::BI__sync_nand_and_fetch:
179 case Builtin::BI__sync_val_compare_and_swap:
180 case Builtin::BI__sync_bool_compare_and_swap:
181 case Builtin::BI__sync_lock_test_and_set:
182 case Builtin::BI__sync_lock_release:
183 if (SemaBuiltinAtomicOverloaded(TheCall))
188 return move(TheCallResult);
191 /// CheckFunctionCall - Check a direct function call for various correctness
192 /// and safety properties not strictly enforced by the C type system.
193 bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) {
194 // Get the IdentifierInfo* for the called function.
195 IdentifierInfo *FnInfo = FDecl->getIdentifier();
197 // None of the checks below are needed for functions that don't have
198 // simple names (e.g., C++ conversion functions).
202 // FIXME: This mechanism should be abstracted to be less fragile and
203 // more efficient. For example, just map function ids to custom
207 if (const FormatAttr *Format = FDecl->getAttr<FormatAttr>()) {
208 if (CheckablePrintfAttr(Format, TheCall)) {
209 bool HasVAListArg = Format->getFirstArg() == 0;
211 if (const FunctionProtoType *Proto
212 = FDecl->getType()->getAs<FunctionProtoType>())
213 HasVAListArg = !Proto->isVariadic();
215 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1,
216 HasVAListArg ? 0 : Format->getFirstArg() - 1);
220 for (const NonNullAttr *NonNull = FDecl->getAttr<NonNullAttr>(); NonNull;
221 NonNull = NonNull->getNext<NonNullAttr>())
222 CheckNonNullArguments(NonNull, TheCall);
227 bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) {
229 const FormatAttr *Format = NDecl->getAttr<FormatAttr>();
233 const VarDecl *V = dyn_cast<VarDecl>(NDecl);
237 QualType Ty = V->getType();
238 if (!Ty->isBlockPointerType())
241 if (!CheckablePrintfAttr(Format, TheCall))
244 bool HasVAListArg = Format->getFirstArg() == 0;
246 const FunctionType *FT =
247 Ty->getAs<BlockPointerType>()->getPointeeType()->getAs<FunctionType>();
248 if (const FunctionProtoType *Proto = dyn_cast<FunctionProtoType>(FT))
249 HasVAListArg = !Proto->isVariadic();
251 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1,
252 HasVAListArg ? 0 : Format->getFirstArg() - 1);
257 /// SemaBuiltinAtomicOverloaded - We have a call to a function like
258 /// __sync_fetch_and_add, which is an overloaded function based on the pointer
259 /// type of its first argument. The main ActOnCallExpr routines have already
260 /// promoted the types of arguments because all of these calls are prototyped as
263 /// This function goes through and does final semantic checking for these
265 bool Sema::SemaBuiltinAtomicOverloaded(CallExpr *TheCall) {
266 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
267 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
269 // Ensure that we have at least one argument to do type inference from.
270 if (TheCall->getNumArgs() < 1)
271 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
272 << 0 << TheCall->getCallee()->getSourceRange();
274 // Inspect the first argument of the atomic builtin. This should always be
275 // a pointer type, whose element is an integral scalar or pointer type.
276 // Because it is a pointer type, we don't have to worry about any implicit
278 Expr *FirstArg = TheCall->getArg(0);
279 if (!FirstArg->getType()->isPointerType())
280 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
281 << FirstArg->getType() << FirstArg->getSourceRange();
283 QualType ValType = FirstArg->getType()->getAs<PointerType>()->getPointeeType();
284 if (!ValType->isIntegerType() && !ValType->isPointerType() &&
285 !ValType->isBlockPointerType())
286 return Diag(DRE->getLocStart(),
287 diag::err_atomic_builtin_must_be_pointer_intptr)
288 << FirstArg->getType() << FirstArg->getSourceRange();
290 // We need to figure out which concrete builtin this maps onto. For example,
291 // __sync_fetch_and_add with a 2 byte object turns into
292 // __sync_fetch_and_add_2.
293 #define BUILTIN_ROW(x) \
294 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
295 Builtin::BI##x##_8, Builtin::BI##x##_16 }
297 static const unsigned BuiltinIndices[][5] = {
298 BUILTIN_ROW(__sync_fetch_and_add),
299 BUILTIN_ROW(__sync_fetch_and_sub),
300 BUILTIN_ROW(__sync_fetch_and_or),
301 BUILTIN_ROW(__sync_fetch_and_and),
302 BUILTIN_ROW(__sync_fetch_and_xor),
303 BUILTIN_ROW(__sync_fetch_and_nand),
305 BUILTIN_ROW(__sync_add_and_fetch),
306 BUILTIN_ROW(__sync_sub_and_fetch),
307 BUILTIN_ROW(__sync_and_and_fetch),
308 BUILTIN_ROW(__sync_or_and_fetch),
309 BUILTIN_ROW(__sync_xor_and_fetch),
310 BUILTIN_ROW(__sync_nand_and_fetch),
312 BUILTIN_ROW(__sync_val_compare_and_swap),
313 BUILTIN_ROW(__sync_bool_compare_and_swap),
314 BUILTIN_ROW(__sync_lock_test_and_set),
315 BUILTIN_ROW(__sync_lock_release)
319 // Determine the index of the size.
321 switch (Context.getTypeSize(ValType)/8) {
322 case 1: SizeIndex = 0; break;
323 case 2: SizeIndex = 1; break;
324 case 4: SizeIndex = 2; break;
325 case 8: SizeIndex = 3; break;
326 case 16: SizeIndex = 4; break;
328 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
329 << FirstArg->getType() << FirstArg->getSourceRange();
332 // Each of these builtins has one pointer argument, followed by some number of
333 // values (0, 1 or 2) followed by a potentially empty varags list of stuff
334 // that we ignore. Find out which row of BuiltinIndices to read from as well
335 // as the number of fixed args.
336 unsigned BuiltinID = FDecl->getBuiltinID();
337 unsigned BuiltinIndex, NumFixed = 1;
339 default: assert(0 && "Unknown overloaded atomic builtin!");
340 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break;
341 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break;
342 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break;
343 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break;
344 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break;
345 case Builtin::BI__sync_fetch_and_nand:BuiltinIndex = 5; break;
347 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 6; break;
348 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 7; break;
349 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 8; break;
350 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 9; break;
351 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex =10; break;
352 case Builtin::BI__sync_nand_and_fetch:BuiltinIndex =11; break;
354 case Builtin::BI__sync_val_compare_and_swap:
358 case Builtin::BI__sync_bool_compare_and_swap:
362 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 14; break;
363 case Builtin::BI__sync_lock_release:
369 // Now that we know how many fixed arguments we expect, first check that we
370 // have at least that many.
371 if (TheCall->getNumArgs() < 1+NumFixed)
372 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
373 << 0 << TheCall->getCallee()->getSourceRange();
376 // Get the decl for the concrete builtin from this, we can tell what the
377 // concrete integer type we should convert to is.
378 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
379 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
380 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
381 FunctionDecl *NewBuiltinDecl =
382 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
383 TUScope, false, DRE->getLocStart()));
384 const FunctionProtoType *BuiltinFT =
385 NewBuiltinDecl->getType()->getAs<FunctionProtoType>();
386 ValType = BuiltinFT->getArgType(0)->getAs<PointerType>()->getPointeeType();
388 // If the first type needs to be converted (e.g. void** -> int*), do it now.
389 if (BuiltinFT->getArgType(0) != FirstArg->getType()) {
390 ImpCastExprToType(FirstArg, BuiltinFT->getArgType(0), CastExpr::CK_BitCast);
391 TheCall->setArg(0, FirstArg);
394 // Next, walk the valid ones promoting to the right type.
395 for (unsigned i = 0; i != NumFixed; ++i) {
396 Expr *Arg = TheCall->getArg(i+1);
398 // If the argument is an implicit cast, then there was a promotion due to
399 // "...", just remove it now.
400 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) {
401 Arg = ICE->getSubExpr();
403 ICE->Destroy(Context);
404 TheCall->setArg(i+1, Arg);
407 // GCC does an implicit conversion to the pointer or integer ValType. This
408 // can fail in some cases (1i -> int**), check for this error case now.
409 CastExpr::CastKind Kind = CastExpr::CK_Unknown;
410 CXXMethodDecl *ConversionDecl = 0;
411 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind,
415 // Okay, we have something that *can* be converted to the right type. Check
416 // to see if there is a potentially weird extension going on here. This can
417 // happen when you do an atomic operation on something like an char* and
418 // pass in 42. The 42 gets converted to char. This is even more strange
419 // for things like 45.123 -> char, etc.
420 // FIXME: Do this check.
421 ImpCastExprToType(Arg, ValType, Kind, /*isLvalue=*/false);
422 TheCall->setArg(i+1, Arg);
425 // Switch the DeclRefExpr to refer to the new decl.
426 DRE->setDecl(NewBuiltinDecl);
427 DRE->setType(NewBuiltinDecl->getType());
429 // Set the callee in the CallExpr.
430 // FIXME: This leaks the original parens and implicit casts.
431 Expr *PromotedCall = DRE;
432 UsualUnaryConversions(PromotedCall);
433 TheCall->setCallee(PromotedCall);
436 // Change the result type of the call to match the result type of the decl.
437 TheCall->setType(NewBuiltinDecl->getResultType());
442 /// CheckObjCString - Checks that the argument to the builtin
443 /// CFString constructor is correct
444 /// FIXME: GCC currently emits the following warning:
445 /// "warning: input conversion stopped due to an input byte that does not
446 /// belong to the input codeset UTF-8"
447 /// Note: It might also make sense to do the UTF-16 conversion here (would
448 /// simplify the backend).
449 bool Sema::CheckObjCString(Expr *Arg) {
450 Arg = Arg->IgnoreParenCasts();
451 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
453 if (!Literal || Literal->isWide()) {
454 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
455 << Arg->getSourceRange();
459 const char *Data = Literal->getStrData();
460 unsigned Length = Literal->getByteLength();
462 for (unsigned i = 0; i < Length; ++i) {
464 Diag(getLocationOfStringLiteralByte(Literal, i),
465 diag::warn_cfstring_literal_contains_nul_character)
466 << Arg->getSourceRange();
474 /// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
475 /// Emit an error and return true on failure, return false on success.
476 bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
477 Expr *Fn = TheCall->getCallee();
478 if (TheCall->getNumArgs() > 2) {
479 Diag(TheCall->getArg(2)->getLocStart(),
480 diag::err_typecheck_call_too_many_args)
481 << 0 /*function call*/ << Fn->getSourceRange()
482 << SourceRange(TheCall->getArg(2)->getLocStart(),
483 (*(TheCall->arg_end()-1))->getLocEnd());
487 if (TheCall->getNumArgs() < 2) {
488 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
489 << 0 /*function call*/;
492 // Determine whether the current function is variadic or not.
495 isVariadic = CurBlock->isVariadic;
496 else if (getCurFunctionDecl()) {
497 if (FunctionProtoType* FTP =
498 dyn_cast<FunctionProtoType>(getCurFunctionDecl()->getType()))
499 isVariadic = FTP->isVariadic();
503 isVariadic = getCurMethodDecl()->isVariadic();
507 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
511 // Verify that the second argument to the builtin is the last argument of the
512 // current function or method.
513 bool SecondArgIsLastNamedArgument = false;
514 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
516 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
517 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
518 // FIXME: This isn't correct for methods (results in bogus warning).
519 // Get the last formal in the current function.
520 const ParmVarDecl *LastArg;
522 LastArg = *(CurBlock->TheDecl->param_end()-1);
523 else if (FunctionDecl *FD = getCurFunctionDecl())
524 LastArg = *(FD->param_end()-1);
526 LastArg = *(getCurMethodDecl()->param_end()-1);
527 SecondArgIsLastNamedArgument = PV == LastArg;
531 if (!SecondArgIsLastNamedArgument)
532 Diag(TheCall->getArg(1)->getLocStart(),
533 diag::warn_second_parameter_of_va_start_not_last_named_argument);
537 /// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
538 /// friends. This is declared to take (...), so we have to check everything.
539 bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
540 if (TheCall->getNumArgs() < 2)
541 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
542 << 0 /*function call*/;
543 if (TheCall->getNumArgs() > 2)
544 return Diag(TheCall->getArg(2)->getLocStart(),
545 diag::err_typecheck_call_too_many_args)
546 << 0 /*function call*/
547 << SourceRange(TheCall->getArg(2)->getLocStart(),
548 (*(TheCall->arg_end()-1))->getLocEnd());
550 Expr *OrigArg0 = TheCall->getArg(0);
551 Expr *OrigArg1 = TheCall->getArg(1);
553 // Do standard promotions between the two arguments, returning their common
555 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
557 // Make sure any conversions are pushed back into the call; this is
558 // type safe since unordered compare builtins are declared as "_Bool
560 TheCall->setArg(0, OrigArg0);
561 TheCall->setArg(1, OrigArg1);
563 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent())
566 // If the common type isn't a real floating type, then the arguments were
567 // invalid for this operation.
568 if (!Res->isRealFloatingType())
569 return Diag(OrigArg0->getLocStart(),
570 diag::err_typecheck_call_invalid_ordered_compare)
571 << OrigArg0->getType() << OrigArg1->getType()
572 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd());
577 /// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isnan and
578 /// friends. This is declared to take (...), so we have to check everything.
579 bool Sema::SemaBuiltinUnaryFP(CallExpr *TheCall) {
580 if (TheCall->getNumArgs() < 1)
581 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
582 << 0 /*function call*/;
583 if (TheCall->getNumArgs() > 1)
584 return Diag(TheCall->getArg(1)->getLocStart(),
585 diag::err_typecheck_call_too_many_args)
586 << 0 /*function call*/
587 << SourceRange(TheCall->getArg(1)->getLocStart(),
588 (*(TheCall->arg_end()-1))->getLocEnd());
590 Expr *OrigArg = TheCall->getArg(0);
592 if (OrigArg->isTypeDependent())
595 // This operation requires a floating-point number
596 if (!OrigArg->getType()->isRealFloatingType())
597 return Diag(OrigArg->getLocStart(),
598 diag::err_typecheck_call_invalid_unary_fp)
599 << OrigArg->getType() << OrigArg->getSourceRange();
604 bool Sema::SemaBuiltinStackAddress(CallExpr *TheCall) {
605 // The signature for these builtins is exact; the only thing we need
606 // to check is that the argument is a constant.
608 if (!TheCall->getArg(0)->isTypeDependent() &&
609 !TheCall->getArg(0)->isValueDependent() &&
610 !TheCall->getArg(0)->isIntegerConstantExpr(Context, &Loc))
611 return Diag(Loc, diag::err_stack_const_level) << TheCall->getSourceRange();
616 /// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
617 // This is declared to take (...), so we have to check everything.
618 Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
619 if (TheCall->getNumArgs() < 3)
620 return ExprError(Diag(TheCall->getLocEnd(),
621 diag::err_typecheck_call_too_few_args)
622 << 0 /*function call*/ << TheCall->getSourceRange());
624 unsigned numElements = std::numeric_limits<unsigned>::max();
625 if (!TheCall->getArg(0)->isTypeDependent() &&
626 !TheCall->getArg(1)->isTypeDependent()) {
627 QualType FAType = TheCall->getArg(0)->getType();
628 QualType SAType = TheCall->getArg(1)->getType();
630 if (!FAType->isVectorType() || !SAType->isVectorType()) {
631 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
632 << SourceRange(TheCall->getArg(0)->getLocStart(),
633 TheCall->getArg(1)->getLocEnd());
637 if (!Context.hasSameUnqualifiedType(FAType, SAType)) {
638 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
639 << SourceRange(TheCall->getArg(0)->getLocStart(),
640 TheCall->getArg(1)->getLocEnd());
644 numElements = FAType->getAs<VectorType>()->getNumElements();
645 if (TheCall->getNumArgs() != numElements+2) {
646 if (TheCall->getNumArgs() < numElements+2)
647 return ExprError(Diag(TheCall->getLocEnd(),
648 diag::err_typecheck_call_too_few_args)
649 << 0 /*function call*/ << TheCall->getSourceRange());
650 return ExprError(Diag(TheCall->getLocEnd(),
651 diag::err_typecheck_call_too_many_args)
652 << 0 /*function call*/ << TheCall->getSourceRange());
656 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
657 if (TheCall->getArg(i)->isTypeDependent() ||
658 TheCall->getArg(i)->isValueDependent())
661 llvm::APSInt Result(32);
662 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
663 return ExprError(Diag(TheCall->getLocStart(),
664 diag::err_shufflevector_nonconstant_argument)
665 << TheCall->getArg(i)->getSourceRange());
667 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
668 return ExprError(Diag(TheCall->getLocStart(),
669 diag::err_shufflevector_argument_too_large)
670 << TheCall->getArg(i)->getSourceRange());
673 llvm::SmallVector<Expr*, 32> exprs;
675 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
676 exprs.push_back(TheCall->getArg(i));
677 TheCall->setArg(i, 0);
680 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(),
681 exprs.size(), exprs[0]->getType(),
682 TheCall->getCallee()->getLocStart(),
683 TheCall->getRParenLoc()));
686 /// SemaBuiltinPrefetch - Handle __builtin_prefetch.
687 // This is declared to take (const void*, ...) and can take two
688 // optional constant int args.
689 bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
690 unsigned NumArgs = TheCall->getNumArgs();
693 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_many_args)
694 << 0 /*function call*/ << TheCall->getSourceRange();
696 // Argument 0 is checked for us and the remaining arguments must be
697 // constant integers.
698 for (unsigned i = 1; i != NumArgs; ++i) {
699 Expr *Arg = TheCall->getArg(i);
700 if (Arg->isTypeDependent())
703 if (!Arg->getType()->isIntegralType())
704 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_arg_type)
705 << Arg->getSourceRange();
707 ImpCastExprToType(Arg, Context.IntTy, CastExpr::CK_IntegralCast);
708 TheCall->setArg(i, Arg);
710 if (Arg->isValueDependent())
714 if (!Arg->isIntegerConstantExpr(Result, Context))
715 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_arg_ice)
716 << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
718 // FIXME: gcc issues a warning and rewrites these to 0. These
719 // seems especially odd for the third argument since the default
722 if (Result.getLimitedValue() > 1)
723 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
724 << "0" << "1" << Arg->getSourceRange();
726 if (Result.getLimitedValue() > 3)
727 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
728 << "0" << "3" << Arg->getSourceRange();
735 /// SemaBuiltinEHReturnDataRegNo - Handle __builtin_eh_return_data_regno, the
736 /// operand must be an integer constant.
737 bool Sema::SemaBuiltinEHReturnDataRegNo(CallExpr *TheCall) {
739 if (!TheCall->getArg(0)->isIntegerConstantExpr(Result, Context))
740 return Diag(TheCall->getLocStart(), diag::err_expr_not_ice)
741 << TheCall->getArg(0)->getSourceRange();
747 /// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
748 /// int type). This simply type checks that type is one of the defined
750 // For compatability check 0-3, llvm only handles 0 and 2.
751 bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
752 Expr *Arg = TheCall->getArg(1);
753 if (Arg->isTypeDependent())
756 QualType ArgType = Arg->getType();
757 const BuiltinType *BT = ArgType->getAs<BuiltinType>();
758 llvm::APSInt Result(32);
759 if (!BT || BT->getKind() != BuiltinType::Int)
760 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument)
761 << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
763 if (Arg->isValueDependent())
766 if (!Arg->isIntegerConstantExpr(Result, Context)) {
767 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument)
768 << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
771 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
772 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
773 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
779 /// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
780 /// This checks that val is a constant 1.
781 bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
782 Expr *Arg = TheCall->getArg(1);
783 if (Arg->isTypeDependent() || Arg->isValueDependent())
786 llvm::APSInt Result(32);
787 if (!Arg->isIntegerConstantExpr(Result, Context) || Result != 1)
788 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
789 << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
794 // Handle i > 1 ? "x" : "y", recursivelly
795 bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall,
797 unsigned format_idx, unsigned firstDataArg) {
798 if (E->isTypeDependent() || E->isValueDependent())
801 switch (E->getStmtClass()) {
802 case Stmt::ConditionalOperatorClass: {
803 const ConditionalOperator *C = cast<ConditionalOperator>(E);
804 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall,
805 HasVAListArg, format_idx, firstDataArg)
806 && SemaCheckStringLiteral(C->getRHS(), TheCall,
807 HasVAListArg, format_idx, firstDataArg);
810 case Stmt::ImplicitCastExprClass: {
811 const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E);
812 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg,
813 format_idx, firstDataArg);
816 case Stmt::ParenExprClass: {
817 const ParenExpr *Expr = cast<ParenExpr>(E);
818 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg,
819 format_idx, firstDataArg);
822 case Stmt::DeclRefExprClass: {
823 const DeclRefExpr *DR = cast<DeclRefExpr>(E);
825 // As an exception, do not flag errors for variables binding to
826 // const string literals.
827 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
828 bool isConstant = false;
829 QualType T = DR->getType();
831 if (const ArrayType *AT = Context.getAsArrayType(T)) {
832 isConstant = AT->getElementType().isConstant(Context);
833 } else if (const PointerType *PT = T->getAs<PointerType>()) {
834 isConstant = T.isConstant(Context) &&
835 PT->getPointeeType().isConstant(Context);
839 const VarDecl *Def = 0;
840 if (const Expr *Init = VD->getDefinition(Def))
841 return SemaCheckStringLiteral(Init, TheCall,
842 HasVAListArg, format_idx, firstDataArg);
845 // For vprintf* functions (i.e., HasVAListArg==true), we add a
846 // special check to see if the format string is a function parameter
847 // of the function calling the printf function. If the function
848 // has an attribute indicating it is a printf-like function, then we
849 // should suppress warnings concerning non-literals being used in a call
850 // to a vprintf function. For example:
853 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
855 // va_start(ap, fmt);
856 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt".
860 // FIXME: We don't have full attribute support yet, so just check to see
861 // if the argument is a DeclRefExpr that references a parameter. We'll
862 // add proper support for checking the attribute later.
864 if (isa<ParmVarDecl>(VD))
871 case Stmt::CallExprClass: {
872 const CallExpr *CE = cast<CallExpr>(E);
873 if (const ImplicitCastExpr *ICE
874 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) {
875 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) {
876 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) {
877 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) {
878 unsigned ArgIndex = FA->getFormatIdx();
879 const Expr *Arg = CE->getArg(ArgIndex - 1);
881 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg,
882 format_idx, firstDataArg);
890 case Stmt::ObjCStringLiteralClass:
891 case Stmt::StringLiteralClass: {
892 const StringLiteral *StrE = NULL;
894 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
895 StrE = ObjCFExpr->getString();
897 StrE = cast<StringLiteral>(E);
900 CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx,
914 Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
915 const CallExpr *TheCall) {
916 for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end();
918 const Expr *ArgExpr = TheCall->getArg(*i);
919 if (ArgExpr->isNullPointerConstant(Context,
920 Expr::NPC_ValueDependentIsNotNull))
921 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg)
922 << ArgExpr->getSourceRange();
926 /// CheckPrintfArguments - Check calls to printf (and similar functions) for
927 /// correct use of format strings.
929 /// HasVAListArg - A predicate indicating whether the printf-like
930 /// function is passed an explicit va_arg argument (e.g., vprintf)
932 /// format_idx - The index into Args for the format string.
934 /// Improper format strings to functions in the printf family can be
935 /// the source of bizarre bugs and very serious security holes. A
936 /// good source of information is available in the following paper
937 /// (which includes additional references):
939 /// FormatGuard: Automatic Protection From printf Format String
940 /// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001.
942 /// Functionality implemented:
944 /// We can statically check the following properties for string
945 /// literal format strings for non v.*printf functions (where the
946 /// arguments are passed directly):
948 /// (1) Are the number of format conversions equal to the number of
951 /// (2) Does each format conversion correctly match the type of the
952 /// corresponding data argument? (TODO)
954 /// Moreover, for all printf functions we can:
956 /// (3) Check for a missing format string (when not caught by type checking).
958 /// (4) Check for no-operation flags; e.g. using "#" with format
959 /// conversion 'c' (TODO)
961 /// (5) Check the use of '%n', a major source of security holes.
963 /// (6) Check for malformed format conversions that don't specify anything.
965 /// (7) Check for empty format strings. e.g: printf("");
967 /// (8) Check that the format string is a wide literal.
969 /// (9) Also check the arguments of functions with the __format__ attribute.
972 /// All of these checks can be done by parsing the format string.
974 /// For now, we ONLY do (1), (3), (5), (6), (7), and (8).
976 Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg,
977 unsigned format_idx, unsigned firstDataArg) {
978 const Expr *Fn = TheCall->getCallee();
980 // The way the format attribute works in GCC, the implicit this argument
981 // of member functions is counted. However, it doesn't appear in our own
982 // lists, so decrement format_idx in that case.
983 if (isa<CXXMemberCallExpr>(TheCall)) {
984 // Catch a format attribute mistakenly referring to the object argument.
988 if(firstDataArg != 0)
992 // CHECK: printf-like function is called with no format string.
993 if (format_idx >= TheCall->getNumArgs()) {
994 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string)
995 << Fn->getSourceRange();
999 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts();
1001 // CHECK: format string is not a string literal.
1003 // Dynamically generated format strings are difficult to
1004 // automatically vet at compile time. Requiring that format strings
1005 // are string literals: (1) permits the checking of format strings by
1006 // the compiler and thereby (2) can practically remove the source of
1007 // many format string exploits.
1009 // Format string can be either ObjC string (e.g. @"%d") or
1010 // C string (e.g. "%d")
1011 // ObjC string uses the same format specifiers as C string, so we can use
1012 // the same format string checking logic for both ObjC and C strings.
1013 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx,
1015 return; // Literal format string found, check done!
1017 // If there are no arguments specified, warn with -Wformat-security, otherwise
1018 // warn only with -Wformat-nonliteral.
1019 if (TheCall->getNumArgs() == format_idx+1)
1020 Diag(TheCall->getArg(format_idx)->getLocStart(),
1021 diag::warn_printf_nonliteral_noargs)
1022 << OrigFormatExpr->getSourceRange();
1024 Diag(TheCall->getArg(format_idx)->getLocStart(),
1025 diag::warn_printf_nonliteral)
1026 << OrigFormatExpr->getSourceRange();
1029 void Sema::CheckPrintfString(const StringLiteral *FExpr,
1030 const Expr *OrigFormatExpr,
1031 const CallExpr *TheCall, bool HasVAListArg,
1032 unsigned format_idx, unsigned firstDataArg) {
1034 const ObjCStringLiteral *ObjCFExpr =
1035 dyn_cast<ObjCStringLiteral>(OrigFormatExpr);
1037 // CHECK: is the format string a wide literal?
1038 if (FExpr->isWide()) {
1039 Diag(FExpr->getLocStart(),
1040 diag::warn_printf_format_string_is_wide_literal)
1041 << OrigFormatExpr->getSourceRange();
1045 // Str - The format string. NOTE: this is NOT null-terminated!
1046 const char *Str = FExpr->getStrData();
1048 // CHECK: empty format string?
1049 unsigned StrLen = FExpr->getByteLength();
1052 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string)
1053 << OrigFormatExpr->getSourceRange();
1057 // We process the format string using a binary state machine. The
1058 // current state is stored in CurrentState.
1062 } CurrentState = state_OrdChr;
1064 // numConversions - The number of conversions seen so far. This is
1065 // incremented as we traverse the format string.
1066 unsigned numConversions = 0;
1068 // numDataArgs - The number of data arguments after the format
1069 // string. This can only be determined for non vprintf-like
1070 // functions. For those functions, this value is 1 (the sole
1071 // va_arg argument).
1072 unsigned numDataArgs = TheCall->getNumArgs()-firstDataArg;
1074 // Inspect the format string.
1075 unsigned StrIdx = 0;
1077 // LastConversionIdx - Index within the format string where we last saw
1078 // a '%' character that starts a new format conversion.
1079 unsigned LastConversionIdx = 0;
1081 for (; StrIdx < StrLen; ++StrIdx) {
1083 // Is the number of detected conversion conversions greater than
1084 // the number of matching data arguments? If so, stop.
1085 if (!HasVAListArg && numConversions > numDataArgs) break;
1088 if (Str[StrIdx] == '\0') {
1089 // The string returned by getStrData() is not null-terminated,
1090 // so the presence of a null character is likely an error.
1091 Diag(getLocationOfStringLiteralByte(FExpr, StrIdx),
1092 diag::warn_printf_format_string_contains_null_char)
1093 << OrigFormatExpr->getSourceRange();
1097 // Ordinary characters (not processing a format conversion).
1098 if (CurrentState == state_OrdChr) {
1099 if (Str[StrIdx] == '%') {
1100 CurrentState = state_Conversion;
1101 LastConversionIdx = StrIdx;
1106 // Seen '%'. Now processing a format conversion.
1107 switch (Str[StrIdx]) {
1108 // Handle dynamic precision or width specifier.
1112 if (!HasVAListArg) {
1113 if (numConversions > numDataArgs) {
1114 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx);
1116 if (Str[StrIdx-1] == '.')
1117 Diag(Loc, diag::warn_printf_asterisk_precision_missing_arg)
1118 << OrigFormatExpr->getSourceRange();
1120 Diag(Loc, diag::warn_printf_asterisk_width_missing_arg)
1121 << OrigFormatExpr->getSourceRange();
1123 // Don't do any more checking. We'll just emit spurious errors.
1127 // Perform type checking on width/precision specifier.
1128 const Expr *E = TheCall->getArg(format_idx+numConversions);
1129 if (const BuiltinType *BT = E->getType()->getAs<BuiltinType>())
1130 if (BT->getKind() == BuiltinType::Int)
1133 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx);
1135 if (Str[StrIdx-1] == '.')
1136 Diag(Loc, diag::warn_printf_asterisk_precision_wrong_type)
1137 << E->getType() << E->getSourceRange();
1139 Diag(Loc, diag::warn_printf_asterisk_width_wrong_type)
1140 << E->getType() << E->getSourceRange();
1146 // Characters which can terminate a format conversion
1147 // (e.g. "%d"). Characters that specify length modifiers or
1148 // other flags are handled by the default case below.
1150 // FIXME: additional checks will go into the following cases.
1174 CurrentState = state_OrdChr;
1178 // FIXME: Warn in situations where this isn't supported!
1179 CurrentState = state_OrdChr;
1182 // CHECK: Are we using "%n"? Issue a warning.
1185 CurrentState = state_OrdChr;
1186 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr,
1189 Diag(Loc, diag::warn_printf_write_back)<<OrigFormatExpr->getSourceRange();
1195 // %@ is allowed in ObjC format strings only.
1196 if (ObjCFExpr != NULL)
1197 CurrentState = state_OrdChr;
1199 // Issue a warning: invalid format conversion.
1200 SourceLocation Loc =
1201 getLocationOfStringLiteralByte(FExpr, LastConversionIdx);
1203 Diag(Loc, diag::warn_printf_invalid_conversion)
1204 << std::string(Str+LastConversionIdx,
1205 Str+std::min(LastConversionIdx+2, StrLen))
1206 << OrigFormatExpr->getSourceRange();
1213 // Sanity check: Was the first "%" character the previous one?
1214 // If not, we will assume that we have a malformed format
1215 // conversion, and that the current "%" character is the start
1216 // of a new conversion.
1217 if (StrIdx - LastConversionIdx == 1)
1218 CurrentState = state_OrdChr;
1220 // Issue a warning: invalid format conversion.
1221 SourceLocation Loc =
1222 getLocationOfStringLiteralByte(FExpr, LastConversionIdx);
1224 Diag(Loc, diag::warn_printf_invalid_conversion)
1225 << std::string(Str+LastConversionIdx, Str+StrIdx)
1226 << OrigFormatExpr->getSourceRange();
1228 // This conversion is broken. Advance to the next format
1230 LastConversionIdx = StrIdx;
1236 // This case catches all other characters: flags, widths, etc.
1237 // We should eventually process those as well.
1242 if (CurrentState == state_Conversion) {
1243 // Issue a warning: invalid format conversion.
1244 SourceLocation Loc =
1245 getLocationOfStringLiteralByte(FExpr, LastConversionIdx);
1247 Diag(Loc, diag::warn_printf_invalid_conversion)
1248 << std::string(Str+LastConversionIdx,
1249 Str+std::min(LastConversionIdx+2, StrLen))
1250 << OrigFormatExpr->getSourceRange();
1254 if (!HasVAListArg) {
1255 // CHECK: Does the number of format conversions exceed the number
1256 // of data arguments?
1257 if (numConversions > numDataArgs) {
1258 SourceLocation Loc =
1259 getLocationOfStringLiteralByte(FExpr, LastConversionIdx);
1261 Diag(Loc, diag::warn_printf_insufficient_data_args)
1262 << OrigFormatExpr->getSourceRange();
1264 // CHECK: Does the number of data arguments exceed the number of
1265 // format conversions in the format string?
1266 else if (numConversions < numDataArgs)
1267 Diag(TheCall->getArg(format_idx+numConversions+1)->getLocStart(),
1268 diag::warn_printf_too_many_data_args)
1269 << OrigFormatExpr->getSourceRange();
1273 //===--- CHECK: Return Address of Stack Variable --------------------------===//
1275 static DeclRefExpr* EvalVal(Expr *E);
1276 static DeclRefExpr* EvalAddr(Expr* E);
1278 /// CheckReturnStackAddr - Check if a return statement returns the address
1279 /// of a stack variable.
1281 Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
1282 SourceLocation ReturnLoc) {
1284 // Perform checking for returned stack addresses.
1285 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) {
1286 if (DeclRefExpr *DR = EvalAddr(RetValExp))
1287 Diag(DR->getLocStart(), diag::warn_ret_stack_addr)
1288 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange();
1290 // Skip over implicit cast expressions when checking for block expressions.
1291 RetValExp = RetValExp->IgnoreParenCasts();
1293 if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp))
1294 if (C->hasBlockDeclRefExprs())
1295 Diag(C->getLocStart(), diag::err_ret_local_block)
1296 << C->getSourceRange();
1298 if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp))
1299 Diag(ALE->getLocStart(), diag::warn_ret_addr_label)
1300 << ALE->getSourceRange();
1302 } else if (lhsType->isReferenceType()) {
1303 // Perform checking for stack values returned by reference.
1304 // Check for a reference to the stack
1305 if (DeclRefExpr *DR = EvalVal(RetValExp))
1306 Diag(DR->getLocStart(), diag::warn_ret_stack_ref)
1307 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange();
1311 /// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
1312 /// check if the expression in a return statement evaluates to an address
1313 /// to a location on the stack. The recursion is used to traverse the
1314 /// AST of the return expression, with recursion backtracking when we
1315 /// encounter a subexpression that (1) clearly does not lead to the address
1316 /// of a stack variable or (2) is something we cannot determine leads to
1317 /// the address of a stack variable based on such local checking.
1319 /// EvalAddr processes expressions that are pointers that are used as
1320 /// references (and not L-values). EvalVal handles all other values.
1321 /// At the base case of the recursion is a check for a DeclRefExpr* in
1322 /// the refers to a stack variable.
1324 /// This implementation handles:
1326 /// * pointer-to-pointer casts
1327 /// * implicit conversions from array references to pointers
1328 /// * taking the address of fields
1329 /// * arbitrary interplay between "&" and "*" operators
1330 /// * pointer arithmetic from an address of a stack variable
1331 /// * taking the address of an array element where the array is on the stack
1332 static DeclRefExpr* EvalAddr(Expr *E) {
1333 // We should only be called for evaluating pointer expressions.
1334 assert((E->getType()->isAnyPointerType() ||
1335 E->getType()->isBlockPointerType() ||
1336 E->getType()->isObjCQualifiedIdType()) &&
1337 "EvalAddr only works on pointers");
1339 // Our "symbolic interpreter" is just a dispatch off the currently
1340 // viewed AST node. We then recursively traverse the AST by calling
1341 // EvalAddr and EvalVal appropriately.
1342 switch (E->getStmtClass()) {
1343 case Stmt::ParenExprClass:
1344 // Ignore parentheses.
1345 return EvalAddr(cast<ParenExpr>(E)->getSubExpr());
1347 case Stmt::UnaryOperatorClass: {
1348 // The only unary operator that make sense to handle here
1349 // is AddrOf. All others don't make sense as pointers.
1350 UnaryOperator *U = cast<UnaryOperator>(E);
1352 if (U->getOpcode() == UnaryOperator::AddrOf)
1353 return EvalVal(U->getSubExpr());
1358 case Stmt::BinaryOperatorClass: {
1359 // Handle pointer arithmetic. All other binary operators are not valid
1361 BinaryOperator *B = cast<BinaryOperator>(E);
1362 BinaryOperator::Opcode op = B->getOpcode();
1364 if (op != BinaryOperator::Add && op != BinaryOperator::Sub)
1367 Expr *Base = B->getLHS();
1369 // Determine which argument is the real pointer base. It could be
1370 // the RHS argument instead of the LHS.
1371 if (!Base->getType()->isPointerType()) Base = B->getRHS();
1373 assert (Base->getType()->isPointerType());
1374 return EvalAddr(Base);
1377 // For conditional operators we need to see if either the LHS or RHS are
1378 // valid DeclRefExpr*s. If one of them is valid, we return it.
1379 case Stmt::ConditionalOperatorClass: {
1380 ConditionalOperator *C = cast<ConditionalOperator>(E);
1382 // Handle the GNU extension for missing LHS.
1383 if (Expr *lhsExpr = C->getLHS())
1384 if (DeclRefExpr* LHS = EvalAddr(lhsExpr))
1387 return EvalAddr(C->getRHS());
1390 // For casts, we need to handle conversions from arrays to
1391 // pointer values, and pointer-to-pointer conversions.
1392 case Stmt::ImplicitCastExprClass:
1393 case Stmt::CStyleCastExprClass:
1394 case Stmt::CXXFunctionalCastExprClass: {
1395 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
1396 QualType T = SubExpr->getType();
1398 if (SubExpr->getType()->isPointerType() ||
1399 SubExpr->getType()->isBlockPointerType() ||
1400 SubExpr->getType()->isObjCQualifiedIdType())
1401 return EvalAddr(SubExpr);
1402 else if (T->isArrayType())
1403 return EvalVal(SubExpr);
1408 // C++ casts. For dynamic casts, static casts, and const casts, we
1409 // are always converting from a pointer-to-pointer, so we just blow
1410 // through the cast. In the case the dynamic cast doesn't fail (and
1411 // return NULL), we take the conservative route and report cases
1412 // where we return the address of a stack variable. For Reinterpre
1413 // FIXME: The comment about is wrong; we're not always converting
1414 // from pointer to pointer. I'm guessing that this code should also
1415 // handle references to objects.
1416 case Stmt::CXXStaticCastExprClass:
1417 case Stmt::CXXDynamicCastExprClass:
1418 case Stmt::CXXConstCastExprClass:
1419 case Stmt::CXXReinterpretCastExprClass: {
1420 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
1421 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
1427 // Everything else: we simply don't reason about them.
1434 /// EvalVal - This function is complements EvalAddr in the mutual recursion.
1435 /// See the comments for EvalAddr for more details.
1436 static DeclRefExpr* EvalVal(Expr *E) {
1438 // We should only be called for evaluating non-pointer expressions, or
1439 // expressions with a pointer type that are not used as references but instead
1440 // are l-values (e.g., DeclRefExpr with a pointer type).
1442 // Our "symbolic interpreter" is just a dispatch off the currently
1443 // viewed AST node. We then recursively traverse the AST by calling
1444 // EvalAddr and EvalVal appropriately.
1445 switch (E->getStmtClass()) {
1446 case Stmt::DeclRefExprClass: {
1447 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking
1448 // at code that refers to a variable's name. We check if it has local
1449 // storage within the function, and if so, return the expression.
1450 DeclRefExpr *DR = cast<DeclRefExpr>(E);
1452 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
1453 if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR;
1458 case Stmt::ParenExprClass:
1459 // Ignore parentheses.
1460 return EvalVal(cast<ParenExpr>(E)->getSubExpr());
1462 case Stmt::UnaryOperatorClass: {
1463 // The only unary operator that make sense to handle here
1464 // is Deref. All others don't resolve to a "name." This includes
1465 // handling all sorts of rvalues passed to a unary operator.
1466 UnaryOperator *U = cast<UnaryOperator>(E);
1468 if (U->getOpcode() == UnaryOperator::Deref)
1469 return EvalAddr(U->getSubExpr());
1474 case Stmt::ArraySubscriptExprClass: {
1475 // Array subscripts are potential references to data on the stack. We
1476 // retrieve the DeclRefExpr* for the array variable if it indeed
1477 // has local storage.
1478 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase());
1481 case Stmt::ConditionalOperatorClass: {
1482 // For conditional operators we need to see if either the LHS or RHS are
1483 // non-NULL DeclRefExpr's. If one is non-NULL, we return it.
1484 ConditionalOperator *C = cast<ConditionalOperator>(E);
1486 // Handle the GNU extension for missing LHS.
1487 if (Expr *lhsExpr = C->getLHS())
1488 if (DeclRefExpr *LHS = EvalVal(lhsExpr))
1491 return EvalVal(C->getRHS());
1494 // Accesses to members are potential references to data on the stack.
1495 case Stmt::MemberExprClass: {
1496 MemberExpr *M = cast<MemberExpr>(E);
1498 // Check for indirect access. We only want direct field accesses.
1500 return EvalVal(M->getBase());
1505 // Everything else: we simply don't reason about them.
1511 //===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
1513 /// Check for comparisons of floating point operands using != and ==.
1514 /// Issue a warning if these are no self-comparisons, as they are not likely
1515 /// to do what the programmer intended.
1516 void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) {
1517 bool EmitWarning = true;
1519 Expr* LeftExprSansParen = lex->IgnoreParens();
1520 Expr* RightExprSansParen = rex->IgnoreParens();
1522 // Special case: check for x == x (which is OK).
1523 // Do not emit warnings for such cases.
1524 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
1525 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
1526 if (DRL->getDecl() == DRR->getDecl())
1527 EmitWarning = false;
1530 // Special case: check for comparisons against literals that can be exactly
1531 // represented by APFloat. In such cases, do not emit a warning. This
1532 // is a heuristic: often comparison against such literals are used to
1533 // detect if a value in a variable has not changed. This clearly can
1534 // lead to false negatives.
1536 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
1538 EmitWarning = false;
1540 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
1542 EmitWarning = false;
1546 // Check for comparisons with builtin types.
1548 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
1549 if (CL->isBuiltinCall(Context))
1550 EmitWarning = false;
1553 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
1554 if (CR->isBuiltinCall(Context))
1555 EmitWarning = false;
1557 // Emit the diagnostic.
1559 Diag(loc, diag::warn_floatingpoint_eq)
1560 << lex->getSourceRange() << rex->getSourceRange();