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 //===----------------------------------------------------------------------===//
15 #include "clang/Sema/Initialization.h"
16 #include "clang/Sema/Sema.h"
17 #include "clang/Sema/SemaInternal.h"
18 #include "clang/Sema/Initialization.h"
19 #include "clang/Sema/ScopeInfo.h"
20 #include "clang/Analysis/Analyses/FormatString.h"
21 #include "clang/AST/ASTContext.h"
22 #include "clang/AST/CharUnits.h"
23 #include "clang/AST/DeclCXX.h"
24 #include "clang/AST/DeclObjC.h"
25 #include "clang/AST/ExprCXX.h"
26 #include "clang/AST/ExprObjC.h"
27 #include "clang/AST/EvaluatedExprVisitor.h"
28 #include "clang/AST/DeclObjC.h"
29 #include "clang/AST/StmtCXX.h"
30 #include "clang/AST/StmtObjC.h"
31 #include "clang/Lex/Preprocessor.h"
32 #include "llvm/ADT/BitVector.h"
33 #include "llvm/ADT/STLExtras.h"
34 #include "llvm/Support/raw_ostream.h"
35 #include "clang/Basic/TargetBuiltins.h"
36 #include "clang/Basic/TargetInfo.h"
37 #include "clang/Basic/ConvertUTF.h"
39 using namespace clang;
42 SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL,
43 unsigned ByteNo) const {
44 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(),
45 PP.getLangOptions(), PP.getTargetInfo());
49 /// CheckablePrintfAttr - does a function call have a "printf" attribute
50 /// and arguments that merit checking?
51 bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) {
52 if (Format->getType() == "printf") return true;
53 if (Format->getType() == "printf0") {
54 // printf0 allows null "format" string; if so don't check format/args
55 unsigned format_idx = Format->getFormatIdx() - 1;
56 // Does the index refer to the implicit object argument?
57 if (isa<CXXMemberCallExpr>(TheCall)) {
62 if (format_idx < TheCall->getNumArgs()) {
63 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts();
64 if (!Format->isNullPointerConstant(Context,
65 Expr::NPC_ValueDependentIsNull))
72 /// Checks that a call expression's argument count is the desired number.
73 /// This is useful when doing custom type-checking. Returns true on error.
74 static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) {
75 unsigned argCount = call->getNumArgs();
76 if (argCount == desiredArgCount) return false;
78 if (argCount < desiredArgCount)
79 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args)
80 << 0 /*function call*/ << desiredArgCount << argCount
81 << call->getSourceRange();
83 // Highlight all the excess arguments.
84 SourceRange range(call->getArg(desiredArgCount)->getLocStart(),
85 call->getArg(argCount - 1)->getLocEnd());
87 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args)
88 << 0 /*function call*/ << desiredArgCount << argCount
89 << call->getArg(1)->getSourceRange();
92 /// CheckBuiltinAnnotationString - Checks that string argument to the builtin
93 /// annotation is a non wide string literal.
94 static bool CheckBuiltinAnnotationString(Sema &S, Expr *Arg) {
95 Arg = Arg->IgnoreParenCasts();
96 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
97 if (!Literal || !Literal->isAscii()) {
98 S.Diag(Arg->getLocStart(), diag::err_builtin_annotation_not_string_constant)
99 << Arg->getSourceRange();
106 Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
107 ExprResult TheCallResult(Owned(TheCall));
109 // Find out if any arguments are required to be integer constant expressions.
110 unsigned ICEArguments = 0;
111 ASTContext::GetBuiltinTypeError Error;
112 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments);
113 if (Error != ASTContext::GE_None)
114 ICEArguments = 0; // Don't diagnose previously diagnosed errors.
116 // If any arguments are required to be ICE's, check and diagnose.
117 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) {
118 // Skip arguments not required to be ICE's.
119 if ((ICEArguments & (1 << ArgNo)) == 0) continue;
122 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
124 ICEArguments &= ~(1 << ArgNo);
128 case Builtin::BI__builtin___CFStringMakeConstantString:
129 assert(TheCall->getNumArgs() == 1 &&
130 "Wrong # arguments to builtin CFStringMakeConstantString");
131 if (CheckObjCString(TheCall->getArg(0)))
134 case Builtin::BI__builtin_stdarg_start:
135 case Builtin::BI__builtin_va_start:
136 if (SemaBuiltinVAStart(TheCall))
139 case Builtin::BI__builtin_isgreater:
140 case Builtin::BI__builtin_isgreaterequal:
141 case Builtin::BI__builtin_isless:
142 case Builtin::BI__builtin_islessequal:
143 case Builtin::BI__builtin_islessgreater:
144 case Builtin::BI__builtin_isunordered:
145 if (SemaBuiltinUnorderedCompare(TheCall))
148 case Builtin::BI__builtin_fpclassify:
149 if (SemaBuiltinFPClassification(TheCall, 6))
152 case Builtin::BI__builtin_isfinite:
153 case Builtin::BI__builtin_isinf:
154 case Builtin::BI__builtin_isinf_sign:
155 case Builtin::BI__builtin_isnan:
156 case Builtin::BI__builtin_isnormal:
157 if (SemaBuiltinFPClassification(TheCall, 1))
160 case Builtin::BI__builtin_shufflevector:
161 return SemaBuiltinShuffleVector(TheCall);
162 // TheCall will be freed by the smart pointer here, but that's fine, since
163 // SemaBuiltinShuffleVector guts it, but then doesn't release it.
164 case Builtin::BI__builtin_prefetch:
165 if (SemaBuiltinPrefetch(TheCall))
168 case Builtin::BI__builtin_object_size:
169 if (SemaBuiltinObjectSize(TheCall))
172 case Builtin::BI__builtin_longjmp:
173 if (SemaBuiltinLongjmp(TheCall))
177 case Builtin::BI__builtin_classify_type:
178 if (checkArgCount(*this, TheCall, 1)) return true;
179 TheCall->setType(Context.IntTy);
181 case Builtin::BI__builtin_constant_p:
182 if (checkArgCount(*this, TheCall, 1)) return true;
183 TheCall->setType(Context.IntTy);
185 case Builtin::BI__sync_fetch_and_add:
186 case Builtin::BI__sync_fetch_and_sub:
187 case Builtin::BI__sync_fetch_and_or:
188 case Builtin::BI__sync_fetch_and_and:
189 case Builtin::BI__sync_fetch_and_xor:
190 case Builtin::BI__sync_add_and_fetch:
191 case Builtin::BI__sync_sub_and_fetch:
192 case Builtin::BI__sync_and_and_fetch:
193 case Builtin::BI__sync_or_and_fetch:
194 case Builtin::BI__sync_xor_and_fetch:
195 case Builtin::BI__sync_val_compare_and_swap:
196 case Builtin::BI__sync_bool_compare_and_swap:
197 case Builtin::BI__sync_lock_test_and_set:
198 case Builtin::BI__sync_lock_release:
199 case Builtin::BI__sync_swap:
200 return SemaBuiltinAtomicOverloaded(move(TheCallResult));
201 case Builtin::BI__atomic_load:
202 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Load);
203 case Builtin::BI__atomic_store:
204 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Store);
205 case Builtin::BI__atomic_exchange:
206 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xchg);
207 case Builtin::BI__atomic_compare_exchange_strong:
208 return SemaAtomicOpsOverloaded(move(TheCallResult),
209 AtomicExpr::CmpXchgStrong);
210 case Builtin::BI__atomic_compare_exchange_weak:
211 return SemaAtomicOpsOverloaded(move(TheCallResult),
212 AtomicExpr::CmpXchgWeak);
213 case Builtin::BI__atomic_fetch_add:
214 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Add);
215 case Builtin::BI__atomic_fetch_sub:
216 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Sub);
217 case Builtin::BI__atomic_fetch_and:
218 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::And);
219 case Builtin::BI__atomic_fetch_or:
220 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Or);
221 case Builtin::BI__atomic_fetch_xor:
222 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xor);
223 case Builtin::BI__builtin_annotation:
224 if (CheckBuiltinAnnotationString(*this, TheCall->getArg(1)))
229 // Since the target specific builtins for each arch overlap, only check those
230 // of the arch we are compiling for.
231 if (BuiltinID >= Builtin::FirstTSBuiltin) {
232 switch (Context.getTargetInfo().getTriple().getArch()) {
233 case llvm::Triple::arm:
234 case llvm::Triple::thumb:
235 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall))
243 return move(TheCallResult);
246 // Get the valid immediate range for the specified NEON type code.
247 static unsigned RFT(unsigned t, bool shift = false) {
248 bool quad = t & 0x10;
252 return shift ? 7 : (8 << (int)quad) - 1;
254 return shift ? 15 : (4 << (int)quad) - 1;
256 return shift ? 31 : (2 << (int)quad) - 1;
258 return shift ? 63 : (1 << (int)quad) - 1;
260 assert(!shift && "cannot shift float types!");
261 return (2 << (int)quad) - 1;
263 return shift ? 7 : (8 << (int)quad) - 1;
265 return shift ? 15 : (4 << (int)quad) - 1;
267 assert(!shift && "cannot shift float types!");
268 return (4 << (int)quad) - 1;
273 bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
279 #define GET_NEON_OVERLOAD_CHECK
280 #include "clang/Basic/arm_neon.inc"
281 #undef GET_NEON_OVERLOAD_CHECK
284 // For NEON intrinsics which are overloaded on vector element type, validate
285 // the immediate which specifies which variant to emit.
287 unsigned ArgNo = TheCall->getNumArgs()-1;
288 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
291 TV = Result.getLimitedValue(32);
292 if ((TV > 31) || (mask & (1 << TV)) == 0)
293 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code)
294 << TheCall->getArg(ArgNo)->getSourceRange();
297 // For NEON intrinsics which take an immediate value as part of the
298 // instruction, range check them here.
299 unsigned i = 0, l = 0, u = 0;
301 default: return false;
302 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break;
303 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break;
304 case ARM::BI__builtin_arm_vcvtr_f:
305 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break;
306 #define GET_NEON_IMMEDIATE_CHECK
307 #include "clang/Basic/arm_neon.inc"
308 #undef GET_NEON_IMMEDIATE_CHECK
311 // Check that the immediate argument is actually a constant.
312 if (SemaBuiltinConstantArg(TheCall, i, Result))
315 // Range check against the upper/lower values for this isntruction.
316 unsigned Val = Result.getZExtValue();
317 if (Val < l || Val > (u + l))
318 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
319 << l << u+l << TheCall->getArg(i)->getSourceRange();
321 // FIXME: VFP Intrinsics should error if VFP not present.
325 /// CheckFunctionCall - Check a direct function call for various correctness
326 /// and safety properties not strictly enforced by the C type system.
327 bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) {
328 // Get the IdentifierInfo* for the called function.
329 IdentifierInfo *FnInfo = FDecl->getIdentifier();
331 // None of the checks below are needed for functions that don't have
332 // simple names (e.g., C++ conversion functions).
336 // FIXME: This mechanism should be abstracted to be less fragile and
337 // more efficient. For example, just map function ids to custom
340 // Printf and scanf checking.
341 for (specific_attr_iterator<FormatAttr>
342 i = FDecl->specific_attr_begin<FormatAttr>(),
343 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) {
345 const FormatAttr *Format = *i;
346 const bool b = Format->getType() == "scanf";
347 if (b || CheckablePrintfAttr(Format, TheCall)) {
348 bool HasVAListArg = Format->getFirstArg() == 0;
349 CheckPrintfScanfArguments(TheCall, HasVAListArg,
350 Format->getFormatIdx() - 1,
351 HasVAListArg ? 0 : Format->getFirstArg() - 1,
356 for (specific_attr_iterator<NonNullAttr>
357 i = FDecl->specific_attr_begin<NonNullAttr>(),
358 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) {
359 CheckNonNullArguments(*i, TheCall->getArgs(),
360 TheCall->getCallee()->getLocStart());
365 switch (FDecl->getBuiltinID()) {
366 case Builtin::BI__builtin_memset:
367 case Builtin::BI__builtin___memset_chk:
368 case Builtin::BImemset:
372 case Builtin::BI__builtin_memcpy:
373 case Builtin::BI__builtin___memcpy_chk:
374 case Builtin::BImemcpy:
378 case Builtin::BI__builtin_memmove:
379 case Builtin::BI__builtin___memmove_chk:
380 case Builtin::BImemmove:
384 case Builtin::BIstrlcpy:
385 case Builtin::BIstrlcat:
386 CheckStrlcpycatArguments(TheCall, FnInfo);
389 case Builtin::BI__builtin_memcmp:
393 case Builtin::BI__builtin_strncpy:
394 case Builtin::BI__builtin___strncpy_chk:
395 case Builtin::BIstrncpy:
399 case Builtin::BI__builtin_strncmp:
403 case Builtin::BI__builtin_strncasecmp:
404 CMF = CMF_Strncasecmp;
407 case Builtin::BI__builtin_strncat:
408 case Builtin::BIstrncat:
412 case Builtin::BI__builtin_strndup:
413 case Builtin::BIstrndup:
418 if (FDecl->getLinkage() == ExternalLinkage &&
419 (!getLangOptions().CPlusPlus || FDecl->isExternC())) {
420 if (FnInfo->isStr("memset"))
422 else if (FnInfo->isStr("memcpy"))
424 else if (FnInfo->isStr("memmove"))
426 else if (FnInfo->isStr("memcmp"))
428 else if (FnInfo->isStr("strncpy"))
430 else if (FnInfo->isStr("strncmp"))
432 else if (FnInfo->isStr("strncasecmp"))
433 CMF = CMF_Strncasecmp;
434 else if (FnInfo->isStr("strncat"))
436 else if (FnInfo->isStr("strndup"))
442 // Memset/memcpy/memmove handling
444 CheckMemaccessArguments(TheCall, CheckedMemoryFunction(CMF), FnInfo);
449 bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) {
451 const FormatAttr *Format = NDecl->getAttr<FormatAttr>();
455 const VarDecl *V = dyn_cast<VarDecl>(NDecl);
459 QualType Ty = V->getType();
460 if (!Ty->isBlockPointerType())
463 const bool b = Format->getType() == "scanf";
464 if (!b && !CheckablePrintfAttr(Format, TheCall))
467 bool HasVAListArg = Format->getFirstArg() == 0;
468 CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1,
469 HasVAListArg ? 0 : Format->getFirstArg() - 1, !b);
475 Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, AtomicExpr::AtomicOp Op) {
476 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get());
477 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
479 // All these operations take one of the following four forms:
480 // T __atomic_load(_Atomic(T)*, int) (loads)
481 // T* __atomic_add(_Atomic(T*)*, ptrdiff_t, int) (pointer add/sub)
482 // int __atomic_compare_exchange_strong(_Atomic(T)*, T*, T, int, int)
484 // T __atomic_exchange(_Atomic(T)*, T, int) (everything else)
485 // where T is an appropriate type, and the int paremeterss are for orderings.
486 unsigned NumVals = 1;
487 unsigned NumOrders = 1;
488 if (Op == AtomicExpr::Load) {
490 } else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) {
495 if (TheCall->getNumArgs() < NumVals+NumOrders+1) {
496 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
497 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs()
498 << TheCall->getCallee()->getSourceRange();
500 } else if (TheCall->getNumArgs() > NumVals+NumOrders+1) {
501 Diag(TheCall->getArg(NumVals+NumOrders+1)->getLocStart(),
502 diag::err_typecheck_call_too_many_args)
503 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs()
504 << TheCall->getCallee()->getSourceRange();
508 // Inspect the first argument of the atomic operation. This should always be
509 // a pointer to an _Atomic type.
510 Expr *Ptr = TheCall->getArg(0);
511 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get();
512 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>();
514 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic)
515 << Ptr->getType() << Ptr->getSourceRange();
519 QualType AtomTy = pointerType->getPointeeType();
520 if (!AtomTy->isAtomicType()) {
521 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic)
522 << Ptr->getType() << Ptr->getSourceRange();
525 QualType ValType = AtomTy->getAs<AtomicType>()->getValueType();
527 if ((Op == AtomicExpr::Add || Op == AtomicExpr::Sub) &&
528 !ValType->isIntegerType() && !ValType->isPointerType()) {
529 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr)
530 << Ptr->getType() << Ptr->getSourceRange();
534 if (!ValType->isIntegerType() &&
535 (Op == AtomicExpr::And || Op == AtomicExpr::Or || Op == AtomicExpr::Xor)){
536 Diag(DRE->getLocStart(), diag::err_atomic_op_logical_needs_atomic_int)
537 << Ptr->getType() << Ptr->getSourceRange();
541 switch (ValType.getObjCLifetime()) {
542 case Qualifiers::OCL_None:
543 case Qualifiers::OCL_ExplicitNone:
547 case Qualifiers::OCL_Weak:
548 case Qualifiers::OCL_Strong:
549 case Qualifiers::OCL_Autoreleasing:
550 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
551 << ValType << Ptr->getSourceRange();
555 QualType ResultType = ValType;
556 if (Op == AtomicExpr::Store)
557 ResultType = Context.VoidTy;
558 else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong)
559 ResultType = Context.BoolTy;
561 // The first argument --- the pointer --- has a fixed type; we
562 // deduce the types of the rest of the arguments accordingly. Walk
563 // the remaining arguments, converting them to the deduced value type.
564 for (unsigned i = 1; i != NumVals+NumOrders+1; ++i) {
565 ExprResult Arg = TheCall->getArg(i);
568 // The second argument to a cmpxchg is a pointer to the data which will
569 // be exchanged. The second argument to a pointer add/subtract is the
570 // amount to add/subtract, which must be a ptrdiff_t. The third
571 // argument to a cmpxchg and the second argument in all other cases
572 // is the type of the value.
573 if (i == 1 && (Op == AtomicExpr::CmpXchgWeak ||
574 Op == AtomicExpr::CmpXchgStrong))
575 Ty = Context.getPointerType(ValType.getUnqualifiedType());
576 else if (!ValType->isIntegerType() &&
577 (Op == AtomicExpr::Add || Op == AtomicExpr::Sub))
578 Ty = Context.getPointerDiffType();
582 // The order(s) are always converted to int.
585 InitializedEntity Entity =
586 InitializedEntity::InitializeParameter(Context, Ty, false);
587 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
590 TheCall->setArg(i, Arg.get());
593 SmallVector<Expr*, 5> SubExprs;
594 SubExprs.push_back(Ptr);
595 if (Op == AtomicExpr::Load) {
596 SubExprs.push_back(TheCall->getArg(1)); // Order
597 } else if (Op != AtomicExpr::CmpXchgWeak && Op != AtomicExpr::CmpXchgStrong) {
598 SubExprs.push_back(TheCall->getArg(2)); // Order
599 SubExprs.push_back(TheCall->getArg(1)); // Val1
601 SubExprs.push_back(TheCall->getArg(3)); // Order
602 SubExprs.push_back(TheCall->getArg(1)); // Val1
603 SubExprs.push_back(TheCall->getArg(2)); // Val2
604 SubExprs.push_back(TheCall->getArg(4)); // OrderFail
607 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(),
608 SubExprs.data(), SubExprs.size(),
610 TheCall->getRParenLoc()));
614 /// checkBuiltinArgument - Given a call to a builtin function, perform
615 /// normal type-checking on the given argument, updating the call in
616 /// place. This is useful when a builtin function requires custom
617 /// type-checking for some of its arguments but not necessarily all of
620 /// Returns true on error.
621 static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) {
622 FunctionDecl *Fn = E->getDirectCallee();
623 assert(Fn && "builtin call without direct callee!");
625 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex);
626 InitializedEntity Entity =
627 InitializedEntity::InitializeParameter(S.Context, Param);
629 ExprResult Arg = E->getArg(0);
630 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg);
634 E->setArg(ArgIndex, Arg.take());
638 /// SemaBuiltinAtomicOverloaded - We have a call to a function like
639 /// __sync_fetch_and_add, which is an overloaded function based on the pointer
640 /// type of its first argument. The main ActOnCallExpr routines have already
641 /// promoted the types of arguments because all of these calls are prototyped as
644 /// This function goes through and does final semantic checking for these
647 Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
648 CallExpr *TheCall = (CallExpr *)TheCallResult.get();
649 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
650 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
652 // Ensure that we have at least one argument to do type inference from.
653 if (TheCall->getNumArgs() < 1) {
654 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
655 << 0 << 1 << TheCall->getNumArgs()
656 << TheCall->getCallee()->getSourceRange();
660 // Inspect the first argument of the atomic builtin. This should always be
661 // a pointer type, whose element is an integral scalar or pointer type.
662 // Because it is a pointer type, we don't have to worry about any implicit
664 // FIXME: We don't allow floating point scalars as input.
665 Expr *FirstArg = TheCall->getArg(0);
666 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>();
668 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
669 << FirstArg->getType() << FirstArg->getSourceRange();
673 QualType ValType = pointerType->getPointeeType();
674 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
675 !ValType->isBlockPointerType()) {
676 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
677 << FirstArg->getType() << FirstArg->getSourceRange();
681 switch (ValType.getObjCLifetime()) {
682 case Qualifiers::OCL_None:
683 case Qualifiers::OCL_ExplicitNone:
687 case Qualifiers::OCL_Weak:
688 case Qualifiers::OCL_Strong:
689 case Qualifiers::OCL_Autoreleasing:
690 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
691 << ValType << FirstArg->getSourceRange();
695 // Strip any qualifiers off ValType.
696 ValType = ValType.getUnqualifiedType();
698 // The majority of builtins return a value, but a few have special return
699 // types, so allow them to override appropriately below.
700 QualType ResultType = ValType;
702 // We need to figure out which concrete builtin this maps onto. For example,
703 // __sync_fetch_and_add with a 2 byte object turns into
704 // __sync_fetch_and_add_2.
705 #define BUILTIN_ROW(x) \
706 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
707 Builtin::BI##x##_8, Builtin::BI##x##_16 }
709 static const unsigned BuiltinIndices[][5] = {
710 BUILTIN_ROW(__sync_fetch_and_add),
711 BUILTIN_ROW(__sync_fetch_and_sub),
712 BUILTIN_ROW(__sync_fetch_and_or),
713 BUILTIN_ROW(__sync_fetch_and_and),
714 BUILTIN_ROW(__sync_fetch_and_xor),
716 BUILTIN_ROW(__sync_add_and_fetch),
717 BUILTIN_ROW(__sync_sub_and_fetch),
718 BUILTIN_ROW(__sync_and_and_fetch),
719 BUILTIN_ROW(__sync_or_and_fetch),
720 BUILTIN_ROW(__sync_xor_and_fetch),
722 BUILTIN_ROW(__sync_val_compare_and_swap),
723 BUILTIN_ROW(__sync_bool_compare_and_swap),
724 BUILTIN_ROW(__sync_lock_test_and_set),
725 BUILTIN_ROW(__sync_lock_release),
726 BUILTIN_ROW(__sync_swap)
730 // Determine the index of the size.
732 switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
733 case 1: SizeIndex = 0; break;
734 case 2: SizeIndex = 1; break;
735 case 4: SizeIndex = 2; break;
736 case 8: SizeIndex = 3; break;
737 case 16: SizeIndex = 4; break;
739 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
740 << FirstArg->getType() << FirstArg->getSourceRange();
744 // Each of these builtins has one pointer argument, followed by some number of
745 // values (0, 1 or 2) followed by a potentially empty varags list of stuff
746 // that we ignore. Find out which row of BuiltinIndices to read from as well
747 // as the number of fixed args.
748 unsigned BuiltinID = FDecl->getBuiltinID();
749 unsigned BuiltinIndex, NumFixed = 1;
751 default: llvm_unreachable("Unknown overloaded atomic builtin!");
752 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break;
753 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break;
754 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break;
755 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break;
756 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break;
758 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break;
759 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break;
760 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break;
761 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break;
762 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break;
764 case Builtin::BI__sync_val_compare_and_swap:
768 case Builtin::BI__sync_bool_compare_and_swap:
771 ResultType = Context.BoolTy;
773 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break;
774 case Builtin::BI__sync_lock_release:
777 ResultType = Context.VoidTy;
779 case Builtin::BI__sync_swap: BuiltinIndex = 14; break;
782 // Now that we know how many fixed arguments we expect, first check that we
783 // have at least that many.
784 if (TheCall->getNumArgs() < 1+NumFixed) {
785 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
786 << 0 << 1+NumFixed << TheCall->getNumArgs()
787 << TheCall->getCallee()->getSourceRange();
791 // Get the decl for the concrete builtin from this, we can tell what the
792 // concrete integer type we should convert to is.
793 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
794 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
795 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
796 FunctionDecl *NewBuiltinDecl =
797 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
798 TUScope, false, DRE->getLocStart()));
800 // The first argument --- the pointer --- has a fixed type; we
801 // deduce the types of the rest of the arguments accordingly. Walk
802 // the remaining arguments, converting them to the deduced value type.
803 for (unsigned i = 0; i != NumFixed; ++i) {
804 ExprResult Arg = TheCall->getArg(i+1);
806 // If the argument is an implicit cast, then there was a promotion due to
807 // "...", just remove it now.
808 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg.get())) {
809 Arg = ICE->getSubExpr();
811 TheCall->setArg(i+1, Arg.get());
814 // GCC does an implicit conversion to the pointer or integer ValType. This
815 // can fail in some cases (1i -> int**), check for this error case now.
816 // Initialize the argument.
817 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context,
818 ValType, /*consume*/ false);
819 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
823 // Okay, we have something that *can* be converted to the right type. Check
824 // to see if there is a potentially weird extension going on here. This can
825 // happen when you do an atomic operation on something like an char* and
826 // pass in 42. The 42 gets converted to char. This is even more strange
827 // for things like 45.123 -> char, etc.
828 // FIXME: Do this check.
829 TheCall->setArg(i+1, Arg.take());
832 ASTContext& Context = this->getASTContext();
834 // Create a new DeclRefExpr to refer to the new decl.
835 DeclRefExpr* NewDRE = DeclRefExpr::Create(
837 DRE->getQualifierLoc(),
840 NewBuiltinDecl->getType(),
841 DRE->getValueKind());
843 // Set the callee in the CallExpr.
844 // FIXME: This leaks the original parens and implicit casts.
845 ExprResult PromotedCall = UsualUnaryConversions(NewDRE);
846 if (PromotedCall.isInvalid())
848 TheCall->setCallee(PromotedCall.take());
850 // Change the result type of the call to match the original value type. This
851 // is arbitrary, but the codegen for these builtins ins design to handle it
853 TheCall->setType(ResultType);
855 return move(TheCallResult);
858 /// CheckObjCString - Checks that the argument to the builtin
859 /// CFString constructor is correct
860 /// Note: It might also make sense to do the UTF-16 conversion here (would
861 /// simplify the backend).
862 bool Sema::CheckObjCString(Expr *Arg) {
863 Arg = Arg->IgnoreParenCasts();
864 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
866 if (!Literal || !Literal->isAscii()) {
867 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
868 << Arg->getSourceRange();
872 if (Literal->containsNonAsciiOrNull()) {
873 StringRef String = Literal->getString();
874 unsigned NumBytes = String.size();
875 SmallVector<UTF16, 128> ToBuf(NumBytes);
876 const UTF8 *FromPtr = (UTF8 *)String.data();
877 UTF16 *ToPtr = &ToBuf[0];
879 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes,
880 &ToPtr, ToPtr + NumBytes,
882 // Check for conversion failure.
883 if (Result != conversionOK)
884 Diag(Arg->getLocStart(),
885 diag::warn_cfstring_truncated) << Arg->getSourceRange();
890 /// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
891 /// Emit an error and return true on failure, return false on success.
892 bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
893 Expr *Fn = TheCall->getCallee();
894 if (TheCall->getNumArgs() > 2) {
895 Diag(TheCall->getArg(2)->getLocStart(),
896 diag::err_typecheck_call_too_many_args)
897 << 0 /*function call*/ << 2 << TheCall->getNumArgs()
898 << Fn->getSourceRange()
899 << SourceRange(TheCall->getArg(2)->getLocStart(),
900 (*(TheCall->arg_end()-1))->getLocEnd());
904 if (TheCall->getNumArgs() < 2) {
905 return Diag(TheCall->getLocEnd(),
906 diag::err_typecheck_call_too_few_args_at_least)
907 << 0 /*function call*/ << 2 << TheCall->getNumArgs();
910 // Type-check the first argument normally.
911 if (checkBuiltinArgument(*this, TheCall, 0))
914 // Determine whether the current function is variadic or not.
915 BlockScopeInfo *CurBlock = getCurBlock();
918 isVariadic = CurBlock->TheDecl->isVariadic();
919 else if (FunctionDecl *FD = getCurFunctionDecl())
920 isVariadic = FD->isVariadic();
922 isVariadic = getCurMethodDecl()->isVariadic();
925 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
929 // Verify that the second argument to the builtin is the last argument of the
930 // current function or method.
931 bool SecondArgIsLastNamedArgument = false;
932 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
934 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
935 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
936 // FIXME: This isn't correct for methods (results in bogus warning).
937 // Get the last formal in the current function.
938 const ParmVarDecl *LastArg;
940 LastArg = *(CurBlock->TheDecl->param_end()-1);
941 else if (FunctionDecl *FD = getCurFunctionDecl())
942 LastArg = *(FD->param_end()-1);
944 LastArg = *(getCurMethodDecl()->param_end()-1);
945 SecondArgIsLastNamedArgument = PV == LastArg;
949 if (!SecondArgIsLastNamedArgument)
950 Diag(TheCall->getArg(1)->getLocStart(),
951 diag::warn_second_parameter_of_va_start_not_last_named_argument);
955 /// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
956 /// friends. This is declared to take (...), so we have to check everything.
957 bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
958 if (TheCall->getNumArgs() < 2)
959 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
960 << 0 << 2 << TheCall->getNumArgs()/*function call*/;
961 if (TheCall->getNumArgs() > 2)
962 return Diag(TheCall->getArg(2)->getLocStart(),
963 diag::err_typecheck_call_too_many_args)
964 << 0 /*function call*/ << 2 << TheCall->getNumArgs()
965 << SourceRange(TheCall->getArg(2)->getLocStart(),
966 (*(TheCall->arg_end()-1))->getLocEnd());
968 ExprResult OrigArg0 = TheCall->getArg(0);
969 ExprResult OrigArg1 = TheCall->getArg(1);
971 // Do standard promotions between the two arguments, returning their common
973 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
974 if (OrigArg0.isInvalid() || OrigArg1.isInvalid())
977 // Make sure any conversions are pushed back into the call; this is
978 // type safe since unordered compare builtins are declared as "_Bool
980 TheCall->setArg(0, OrigArg0.get());
981 TheCall->setArg(1, OrigArg1.get());
983 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent())
986 // If the common type isn't a real floating type, then the arguments were
987 // invalid for this operation.
988 if (!Res->isRealFloatingType())
989 return Diag(OrigArg0.get()->getLocStart(),
990 diag::err_typecheck_call_invalid_ordered_compare)
991 << OrigArg0.get()->getType() << OrigArg1.get()->getType()
992 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd());
997 /// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
998 /// __builtin_isnan and friends. This is declared to take (...), so we have
999 /// to check everything. We expect the last argument to be a floating point
1001 bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
1002 if (TheCall->getNumArgs() < NumArgs)
1003 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
1004 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
1005 if (TheCall->getNumArgs() > NumArgs)
1006 return Diag(TheCall->getArg(NumArgs)->getLocStart(),
1007 diag::err_typecheck_call_too_many_args)
1008 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
1009 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
1010 (*(TheCall->arg_end()-1))->getLocEnd());
1012 Expr *OrigArg = TheCall->getArg(NumArgs-1);
1014 if (OrigArg->isTypeDependent())
1017 // This operation requires a non-_Complex floating-point number.
1018 if (!OrigArg->getType()->isRealFloatingType())
1019 return Diag(OrigArg->getLocStart(),
1020 diag::err_typecheck_call_invalid_unary_fp)
1021 << OrigArg->getType() << OrigArg->getSourceRange();
1023 // If this is an implicit conversion from float -> double, remove it.
1024 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
1025 Expr *CastArg = Cast->getSubExpr();
1026 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
1027 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
1028 "promotion from float to double is the only expected cast here");
1029 Cast->setSubExpr(0);
1030 TheCall->setArg(NumArgs-1, CastArg);
1038 /// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
1039 // This is declared to take (...), so we have to check everything.
1040 ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
1041 if (TheCall->getNumArgs() < 2)
1042 return ExprError(Diag(TheCall->getLocEnd(),
1043 diag::err_typecheck_call_too_few_args_at_least)
1044 << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1045 << TheCall->getSourceRange());
1047 // Determine which of the following types of shufflevector we're checking:
1048 // 1) unary, vector mask: (lhs, mask)
1049 // 2) binary, vector mask: (lhs, rhs, mask)
1050 // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
1051 QualType resType = TheCall->getArg(0)->getType();
1052 unsigned numElements = 0;
1054 if (!TheCall->getArg(0)->isTypeDependent() &&
1055 !TheCall->getArg(1)->isTypeDependent()) {
1056 QualType LHSType = TheCall->getArg(0)->getType();
1057 QualType RHSType = TheCall->getArg(1)->getType();
1059 if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
1060 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
1061 << SourceRange(TheCall->getArg(0)->getLocStart(),
1062 TheCall->getArg(1)->getLocEnd());
1066 numElements = LHSType->getAs<VectorType>()->getNumElements();
1067 unsigned numResElements = TheCall->getNumArgs() - 2;
1069 // Check to see if we have a call with 2 vector arguments, the unary shuffle
1070 // with mask. If so, verify that RHS is an integer vector type with the
1071 // same number of elts as lhs.
1072 if (TheCall->getNumArgs() == 2) {
1073 if (!RHSType->hasIntegerRepresentation() ||
1074 RHSType->getAs<VectorType>()->getNumElements() != numElements)
1075 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
1076 << SourceRange(TheCall->getArg(1)->getLocStart(),
1077 TheCall->getArg(1)->getLocEnd());
1078 numResElements = numElements;
1080 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
1081 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
1082 << SourceRange(TheCall->getArg(0)->getLocStart(),
1083 TheCall->getArg(1)->getLocEnd());
1085 } else if (numElements != numResElements) {
1086 QualType eltType = LHSType->getAs<VectorType>()->getElementType();
1087 resType = Context.getVectorType(eltType, numResElements,
1088 VectorType::GenericVector);
1092 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
1093 if (TheCall->getArg(i)->isTypeDependent() ||
1094 TheCall->getArg(i)->isValueDependent())
1097 llvm::APSInt Result(32);
1098 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
1099 return ExprError(Diag(TheCall->getLocStart(),
1100 diag::err_shufflevector_nonconstant_argument)
1101 << TheCall->getArg(i)->getSourceRange());
1103 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
1104 return ExprError(Diag(TheCall->getLocStart(),
1105 diag::err_shufflevector_argument_too_large)
1106 << TheCall->getArg(i)->getSourceRange());
1109 SmallVector<Expr*, 32> exprs;
1111 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
1112 exprs.push_back(TheCall->getArg(i));
1113 TheCall->setArg(i, 0);
1116 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(),
1117 exprs.size(), resType,
1118 TheCall->getCallee()->getLocStart(),
1119 TheCall->getRParenLoc()));
1122 /// SemaBuiltinPrefetch - Handle __builtin_prefetch.
1123 // This is declared to take (const void*, ...) and can take two
1124 // optional constant int args.
1125 bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
1126 unsigned NumArgs = TheCall->getNumArgs();
1129 return Diag(TheCall->getLocEnd(),
1130 diag::err_typecheck_call_too_many_args_at_most)
1131 << 0 /*function call*/ << 3 << NumArgs
1132 << TheCall->getSourceRange();
1134 // Argument 0 is checked for us and the remaining arguments must be
1135 // constant integers.
1136 for (unsigned i = 1; i != NumArgs; ++i) {
1137 Expr *Arg = TheCall->getArg(i);
1139 llvm::APSInt Result;
1140 if (SemaBuiltinConstantArg(TheCall, i, Result))
1143 // FIXME: gcc issues a warning and rewrites these to 0. These
1144 // seems especially odd for the third argument since the default
1147 if (Result.getLimitedValue() > 1)
1148 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1149 << "0" << "1" << Arg->getSourceRange();
1151 if (Result.getLimitedValue() > 3)
1152 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1153 << "0" << "3" << Arg->getSourceRange();
1160 /// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
1161 /// TheCall is a constant expression.
1162 bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
1163 llvm::APSInt &Result) {
1164 Expr *Arg = TheCall->getArg(ArgNum);
1165 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
1166 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
1168 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
1170 if (!Arg->isIntegerConstantExpr(Result, Context))
1171 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
1172 << FDecl->getDeclName() << Arg->getSourceRange();
1177 /// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
1178 /// int type). This simply type checks that type is one of the defined
1179 /// constants (0-3).
1180 // For compatibility check 0-3, llvm only handles 0 and 2.
1181 bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
1182 llvm::APSInt Result;
1184 // Check constant-ness first.
1185 if (SemaBuiltinConstantArg(TheCall, 1, Result))
1188 Expr *Arg = TheCall->getArg(1);
1189 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
1190 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1191 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1197 /// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
1198 /// This checks that val is a constant 1.
1199 bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
1200 Expr *Arg = TheCall->getArg(1);
1201 llvm::APSInt Result;
1203 // TODO: This is less than ideal. Overload this to take a value.
1204 if (SemaBuiltinConstantArg(TheCall, 1, Result))
1208 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
1209 << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1214 // Handle i > 1 ? "x" : "y", recursively.
1215 bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall,
1217 unsigned format_idx, unsigned firstDataArg,
1220 if (E->isTypeDependent() || E->isValueDependent())
1223 E = E->IgnoreParens();
1225 switch (E->getStmtClass()) {
1226 case Stmt::BinaryConditionalOperatorClass:
1227 case Stmt::ConditionalOperatorClass: {
1228 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E);
1229 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg,
1230 format_idx, firstDataArg, isPrintf)
1231 && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg,
1232 format_idx, firstDataArg, isPrintf);
1235 case Stmt::IntegerLiteralClass:
1236 // Technically -Wformat-nonliteral does not warn about this case.
1237 // The behavior of printf and friends in this case is implementation
1238 // dependent. Ideally if the format string cannot be null then
1239 // it should have a 'nonnull' attribute in the function prototype.
1242 case Stmt::ImplicitCastExprClass: {
1243 E = cast<ImplicitCastExpr>(E)->getSubExpr();
1247 case Stmt::OpaqueValueExprClass:
1248 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) {
1254 case Stmt::PredefinedExprClass:
1255 // While __func__, etc., are technically not string literals, they
1256 // cannot contain format specifiers and thus are not a security
1260 case Stmt::DeclRefExprClass: {
1261 const DeclRefExpr *DR = cast<DeclRefExpr>(E);
1263 // As an exception, do not flag errors for variables binding to
1264 // const string literals.
1265 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
1266 bool isConstant = false;
1267 QualType T = DR->getType();
1269 if (const ArrayType *AT = Context.getAsArrayType(T)) {
1270 isConstant = AT->getElementType().isConstant(Context);
1271 } else if (const PointerType *PT = T->getAs<PointerType>()) {
1272 isConstant = T.isConstant(Context) &&
1273 PT->getPointeeType().isConstant(Context);
1277 if (const Expr *Init = VD->getAnyInitializer())
1278 return SemaCheckStringLiteral(Init, TheCall,
1279 HasVAListArg, format_idx, firstDataArg,
1283 // For vprintf* functions (i.e., HasVAListArg==true), we add a
1284 // special check to see if the format string is a function parameter
1285 // of the function calling the printf function. If the function
1286 // has an attribute indicating it is a printf-like function, then we
1287 // should suppress warnings concerning non-literals being used in a call
1288 // to a vprintf function. For example:
1291 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
1293 // va_start(ap, fmt);
1294 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt".
1298 // FIXME: We don't have full attribute support yet, so just check to see
1299 // if the argument is a DeclRefExpr that references a parameter. We'll
1300 // add proper support for checking the attribute later.
1302 if (isa<ParmVarDecl>(VD))
1309 case Stmt::CallExprClass: {
1310 const CallExpr *CE = cast<CallExpr>(E);
1311 if (const ImplicitCastExpr *ICE
1312 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) {
1313 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) {
1314 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) {
1315 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) {
1316 unsigned ArgIndex = FA->getFormatIdx();
1317 const Expr *Arg = CE->getArg(ArgIndex - 1);
1319 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg,
1320 format_idx, firstDataArg, isPrintf);
1328 case Stmt::ObjCStringLiteralClass:
1329 case Stmt::StringLiteralClass: {
1330 const StringLiteral *StrE = NULL;
1332 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1333 StrE = ObjCFExpr->getString();
1335 StrE = cast<StringLiteral>(E);
1338 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx,
1339 firstDataArg, isPrintf);
1352 Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1353 const Expr * const *ExprArgs,
1354 SourceLocation CallSiteLoc) {
1355 for (NonNullAttr::args_iterator i = NonNull->args_begin(),
1356 e = NonNull->args_end();
1358 const Expr *ArgExpr = ExprArgs[*i];
1359 if (ArgExpr->isNullPointerConstant(Context,
1360 Expr::NPC_ValueDependentIsNotNull))
1361 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
1365 /// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar
1366 /// functions) for correct use of format strings.
1368 Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg,
1369 unsigned format_idx, unsigned firstDataArg,
1372 const Expr *Fn = TheCall->getCallee();
1374 // The way the format attribute works in GCC, the implicit this argument
1375 // of member functions is counted. However, it doesn't appear in our own
1376 // lists, so decrement format_idx in that case.
1377 if (isa<CXXMemberCallExpr>(TheCall)) {
1378 const CXXMethodDecl *method_decl =
1379 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl());
1380 if (method_decl && method_decl->isInstance()) {
1381 // Catch a format attribute mistakenly referring to the object argument.
1382 if (format_idx == 0)
1385 if(firstDataArg != 0)
1390 // CHECK: printf/scanf-like function is called with no format string.
1391 if (format_idx >= TheCall->getNumArgs()) {
1392 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string)
1393 << Fn->getSourceRange();
1397 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts();
1399 // CHECK: format string is not a string literal.
1401 // Dynamically generated format strings are difficult to
1402 // automatically vet at compile time. Requiring that format strings
1403 // are string literals: (1) permits the checking of format strings by
1404 // the compiler and thereby (2) can practically remove the source of
1405 // many format string exploits.
1407 // Format string can be either ObjC string (e.g. @"%d") or
1408 // C string (e.g. "%d")
1409 // ObjC string uses the same format specifiers as C string, so we can use
1410 // the same format string checking logic for both ObjC and C strings.
1411 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx,
1412 firstDataArg, isPrintf))
1413 return; // Literal format string found, check done!
1415 // If there are no arguments specified, warn with -Wformat-security, otherwise
1416 // warn only with -Wformat-nonliteral.
1417 if (TheCall->getNumArgs() == format_idx+1)
1418 Diag(TheCall->getArg(format_idx)->getLocStart(),
1419 diag::warn_format_nonliteral_noargs)
1420 << OrigFormatExpr->getSourceRange();
1422 Diag(TheCall->getArg(format_idx)->getLocStart(),
1423 diag::warn_format_nonliteral)
1424 << OrigFormatExpr->getSourceRange();
1428 class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1431 const StringLiteral *FExpr;
1432 const Expr *OrigFormatExpr;
1433 const unsigned FirstDataArg;
1434 const unsigned NumDataArgs;
1435 const bool IsObjCLiteral;
1436 const char *Beg; // Start of format string.
1437 const bool HasVAListArg;
1438 const CallExpr *TheCall;
1440 llvm::BitVector CoveredArgs;
1441 bool usesPositionalArgs;
1444 CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1445 const Expr *origFormatExpr, unsigned firstDataArg,
1446 unsigned numDataArgs, bool isObjCLiteral,
1447 const char *beg, bool hasVAListArg,
1448 const CallExpr *theCall, unsigned formatIdx)
1449 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1450 FirstDataArg(firstDataArg),
1451 NumDataArgs(numDataArgs),
1452 IsObjCLiteral(isObjCLiteral), Beg(beg),
1453 HasVAListArg(hasVAListArg),
1454 TheCall(theCall), FormatIdx(formatIdx),
1455 usesPositionalArgs(false), atFirstArg(true) {
1456 CoveredArgs.resize(numDataArgs);
1457 CoveredArgs.reset();
1460 void DoneProcessing();
1462 void HandleIncompleteSpecifier(const char *startSpecifier,
1463 unsigned specifierLen);
1465 virtual void HandleInvalidPosition(const char *startSpecifier,
1466 unsigned specifierLen,
1467 analyze_format_string::PositionContext p);
1469 virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
1471 void HandleNullChar(const char *nullCharacter);
1474 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
1475 const char *startSpec,
1476 unsigned specifierLen,
1477 const char *csStart, unsigned csLen);
1479 SourceRange getFormatStringRange();
1480 CharSourceRange getSpecifierRange(const char *startSpecifier,
1481 unsigned specifierLen);
1482 SourceLocation getLocationOfByte(const char *x);
1484 const Expr *getDataArg(unsigned i) const;
1486 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
1487 const analyze_format_string::ConversionSpecifier &CS,
1488 const char *startSpecifier, unsigned specifierLen,
1493 SourceRange CheckFormatHandler::getFormatStringRange() {
1494 return OrigFormatExpr->getSourceRange();
1497 CharSourceRange CheckFormatHandler::
1498 getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
1499 SourceLocation Start = getLocationOfByte(startSpecifier);
1500 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1);
1502 // Advance the end SourceLocation by one due to half-open ranges.
1503 End = End.getLocWithOffset(1);
1505 return CharSourceRange::getCharRange(Start, End);
1508 SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
1509 return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
1512 void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
1513 unsigned specifierLen){
1514 SourceLocation Loc = getLocationOfByte(startSpecifier);
1515 S.Diag(Loc, diag::warn_printf_incomplete_specifier)
1516 << getSpecifierRange(startSpecifier, specifierLen);
1520 CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
1521 analyze_format_string::PositionContext p) {
1522 SourceLocation Loc = getLocationOfByte(startPos);
1523 S.Diag(Loc, diag::warn_format_invalid_positional_specifier)
1524 << (unsigned) p << getSpecifierRange(startPos, posLen);
1527 void CheckFormatHandler::HandleZeroPosition(const char *startPos,
1529 SourceLocation Loc = getLocationOfByte(startPos);
1530 S.Diag(Loc, diag::warn_format_zero_positional_specifier)
1531 << getSpecifierRange(startPos, posLen);
1534 void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
1535 if (!IsObjCLiteral) {
1536 // The presence of a null character is likely an error.
1537 S.Diag(getLocationOfByte(nullCharacter),
1538 diag::warn_printf_format_string_contains_null_char)
1539 << getFormatStringRange();
1543 const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
1544 return TheCall->getArg(FirstDataArg + i);
1547 void CheckFormatHandler::DoneProcessing() {
1548 // Does the number of data arguments exceed the number of
1549 // format conversions in the format string?
1550 if (!HasVAListArg) {
1551 // Find any arguments that weren't covered.
1553 signed notCoveredArg = CoveredArgs.find_first();
1554 if (notCoveredArg >= 0) {
1555 assert((unsigned)notCoveredArg < NumDataArgs);
1556 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(),
1557 diag::warn_printf_data_arg_not_used)
1558 << getFormatStringRange();
1564 CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
1566 const char *startSpec,
1567 unsigned specifierLen,
1568 const char *csStart,
1571 bool keepGoing = true;
1572 if (argIndex < NumDataArgs) {
1573 // Consider the argument coverered, even though the specifier doesn't
1575 CoveredArgs.set(argIndex);
1578 // If argIndex exceeds the number of data arguments we
1579 // don't issue a warning because that is just a cascade of warnings (and
1580 // they may have intended '%%' anyway). We don't want to continue processing
1581 // the format string after this point, however, as we will like just get
1582 // gibberish when trying to match arguments.
1586 S.Diag(Loc, diag::warn_format_invalid_conversion)
1587 << StringRef(csStart, csLen)
1588 << getSpecifierRange(startSpec, specifierLen);
1594 CheckFormatHandler::CheckNumArgs(
1595 const analyze_format_string::FormatSpecifier &FS,
1596 const analyze_format_string::ConversionSpecifier &CS,
1597 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
1599 if (argIndex >= NumDataArgs) {
1600 if (FS.usesPositionalArg()) {
1601 S.Diag(getLocationOfByte(CS.getStart()),
1602 diag::warn_printf_positional_arg_exceeds_data_args)
1603 << (argIndex+1) << NumDataArgs
1604 << getSpecifierRange(startSpecifier, specifierLen);
1607 S.Diag(getLocationOfByte(CS.getStart()),
1608 diag::warn_printf_insufficient_data_args)
1609 << getSpecifierRange(startSpecifier, specifierLen);
1617 //===--- CHECK: Printf format string checking ------------------------------===//
1620 class CheckPrintfHandler : public CheckFormatHandler {
1622 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
1623 const Expr *origFormatExpr, unsigned firstDataArg,
1624 unsigned numDataArgs, bool isObjCLiteral,
1625 const char *beg, bool hasVAListArg,
1626 const CallExpr *theCall, unsigned formatIdx)
1627 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1628 numDataArgs, isObjCLiteral, beg, hasVAListArg,
1629 theCall, formatIdx) {}
1632 bool HandleInvalidPrintfConversionSpecifier(
1633 const analyze_printf::PrintfSpecifier &FS,
1634 const char *startSpecifier,
1635 unsigned specifierLen);
1637 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
1638 const char *startSpecifier,
1639 unsigned specifierLen);
1641 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
1642 const char *startSpecifier, unsigned specifierLen);
1643 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
1644 const analyze_printf::OptionalAmount &Amt,
1646 const char *startSpecifier, unsigned specifierLen);
1647 void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1648 const analyze_printf::OptionalFlag &flag,
1649 const char *startSpecifier, unsigned specifierLen);
1650 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
1651 const analyze_printf::OptionalFlag &ignoredFlag,
1652 const analyze_printf::OptionalFlag &flag,
1653 const char *startSpecifier, unsigned specifierLen);
1657 bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
1658 const analyze_printf::PrintfSpecifier &FS,
1659 const char *startSpecifier,
1660 unsigned specifierLen) {
1661 const analyze_printf::PrintfConversionSpecifier &CS =
1662 FS.getConversionSpecifier();
1664 return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1665 getLocationOfByte(CS.getStart()),
1666 startSpecifier, specifierLen,
1667 CS.getStart(), CS.getLength());
1670 bool CheckPrintfHandler::HandleAmount(
1671 const analyze_format_string::OptionalAmount &Amt,
1672 unsigned k, const char *startSpecifier,
1673 unsigned specifierLen) {
1675 if (Amt.hasDataArgument()) {
1676 if (!HasVAListArg) {
1677 unsigned argIndex = Amt.getArgIndex();
1678 if (argIndex >= NumDataArgs) {
1679 S.Diag(getLocationOfByte(Amt.getStart()),
1680 diag::warn_printf_asterisk_missing_arg)
1681 << k << getSpecifierRange(startSpecifier, specifierLen);
1682 // Don't do any more checking. We will just emit
1687 // Type check the data argument. It should be an 'int'.
1688 // Although not in conformance with C99, we also allow the argument to be
1689 // an 'unsigned int' as that is a reasonably safe case. GCC also
1690 // doesn't emit a warning for that case.
1691 CoveredArgs.set(argIndex);
1692 const Expr *Arg = getDataArg(argIndex);
1693 QualType T = Arg->getType();
1695 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context);
1696 assert(ATR.isValid());
1698 if (!ATR.matchesType(S.Context, T)) {
1699 S.Diag(getLocationOfByte(Amt.getStart()),
1700 diag::warn_printf_asterisk_wrong_type)
1702 << ATR.getRepresentativeType(S.Context) << T
1703 << getSpecifierRange(startSpecifier, specifierLen)
1704 << Arg->getSourceRange();
1705 // Don't do any more checking. We will just emit
1714 void CheckPrintfHandler::HandleInvalidAmount(
1715 const analyze_printf::PrintfSpecifier &FS,
1716 const analyze_printf::OptionalAmount &Amt,
1718 const char *startSpecifier,
1719 unsigned specifierLen) {
1720 const analyze_printf::PrintfConversionSpecifier &CS =
1721 FS.getConversionSpecifier();
1722 switch (Amt.getHowSpecified()) {
1723 case analyze_printf::OptionalAmount::Constant:
1724 S.Diag(getLocationOfByte(Amt.getStart()),
1725 diag::warn_printf_nonsensical_optional_amount)
1728 << getSpecifierRange(startSpecifier, specifierLen)
1729 << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
1730 Amt.getConstantLength()));
1734 S.Diag(getLocationOfByte(Amt.getStart()),
1735 diag::warn_printf_nonsensical_optional_amount)
1738 << getSpecifierRange(startSpecifier, specifierLen);
1743 void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1744 const analyze_printf::OptionalFlag &flag,
1745 const char *startSpecifier,
1746 unsigned specifierLen) {
1747 // Warn about pointless flag with a fixit removal.
1748 const analyze_printf::PrintfConversionSpecifier &CS =
1749 FS.getConversionSpecifier();
1750 S.Diag(getLocationOfByte(flag.getPosition()),
1751 diag::warn_printf_nonsensical_flag)
1752 << flag.toString() << CS.toString()
1753 << getSpecifierRange(startSpecifier, specifierLen)
1754 << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1));
1757 void CheckPrintfHandler::HandleIgnoredFlag(
1758 const analyze_printf::PrintfSpecifier &FS,
1759 const analyze_printf::OptionalFlag &ignoredFlag,
1760 const analyze_printf::OptionalFlag &flag,
1761 const char *startSpecifier,
1762 unsigned specifierLen) {
1763 // Warn about ignored flag with a fixit removal.
1764 S.Diag(getLocationOfByte(ignoredFlag.getPosition()),
1765 diag::warn_printf_ignored_flag)
1766 << ignoredFlag.toString() << flag.toString()
1767 << getSpecifierRange(startSpecifier, specifierLen)
1768 << FixItHint::CreateRemoval(getSpecifierRange(
1769 ignoredFlag.getPosition(), 1));
1773 CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
1775 const char *startSpecifier,
1776 unsigned specifierLen) {
1778 using namespace analyze_format_string;
1779 using namespace analyze_printf;
1780 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
1782 if (FS.consumesDataArgument()) {
1785 usesPositionalArgs = FS.usesPositionalArg();
1787 else if (usesPositionalArgs != FS.usesPositionalArg()) {
1788 // Cannot mix-and-match positional and non-positional arguments.
1789 S.Diag(getLocationOfByte(CS.getStart()),
1790 diag::warn_format_mix_positional_nonpositional_args)
1791 << getSpecifierRange(startSpecifier, specifierLen);
1796 // First check if the field width, precision, and conversion specifier
1797 // have matching data arguments.
1798 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
1799 startSpecifier, specifierLen)) {
1803 if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
1804 startSpecifier, specifierLen)) {
1808 if (!CS.consumesDataArgument()) {
1809 // FIXME: Technically specifying a precision or field width here
1810 // makes no sense. Worth issuing a warning at some point.
1814 // Consume the argument.
1815 unsigned argIndex = FS.getArgIndex();
1816 if (argIndex < NumDataArgs) {
1817 // The check to see if the argIndex is valid will come later.
1818 // We set the bit here because we may exit early from this
1819 // function if we encounter some other error.
1820 CoveredArgs.set(argIndex);
1823 // FreeBSD extensions
1824 if (CS.getKind() == ConversionSpecifier::bArg || CS.getKind() == ConversionSpecifier::DArg) {
1825 // claim the second argument
1826 CoveredArgs.set(argIndex + 1);
1828 // Now type check the data expression that matches the
1829 // format specifier.
1830 const Expr *Ex = getDataArg(argIndex);
1831 const analyze_printf::ArgTypeResult &ATR =
1832 (CS.getKind() == ConversionSpecifier::bArg) ?
1833 ArgTypeResult(S.Context.IntTy) : ArgTypeResult::CStrTy;
1834 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType()))
1835 S.Diag(getLocationOfByte(CS.getStart()),
1836 diag::warn_printf_conversion_argument_type_mismatch)
1837 << ATR.getRepresentativeType(S.Context) << Ex->getType()
1838 << getSpecifierRange(startSpecifier, specifierLen)
1839 << Ex->getSourceRange();
1841 // Now type check the data expression that matches the
1842 // format specifier.
1843 Ex = getDataArg(argIndex + 1);
1844 const analyze_printf::ArgTypeResult &ATR2 = ArgTypeResult::CStrTy;
1845 if (ATR2.isValid() && !ATR2.matchesType(S.Context, Ex->getType()))
1846 S.Diag(getLocationOfByte(CS.getStart()),
1847 diag::warn_printf_conversion_argument_type_mismatch)
1848 << ATR2.getRepresentativeType(S.Context) << Ex->getType()
1849 << getSpecifierRange(startSpecifier, specifierLen)
1850 << Ex->getSourceRange();
1854 // END OF FREEBSD EXTENSIONS
1856 // Check for using an Objective-C specific conversion specifier
1857 // in a non-ObjC literal.
1858 if (!IsObjCLiteral && CS.isObjCArg()) {
1859 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
1863 // Check for invalid use of field width
1864 if (!FS.hasValidFieldWidth()) {
1865 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
1866 startSpecifier, specifierLen);
1869 // Check for invalid use of precision
1870 if (!FS.hasValidPrecision()) {
1871 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
1872 startSpecifier, specifierLen);
1875 // Check each flag does not conflict with any other component.
1876 if (!FS.hasValidThousandsGroupingPrefix())
1877 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen);
1878 if (!FS.hasValidLeadingZeros())
1879 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
1880 if (!FS.hasValidPlusPrefix())
1881 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
1882 if (!FS.hasValidSpacePrefix())
1883 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
1884 if (!FS.hasValidAlternativeForm())
1885 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
1886 if (!FS.hasValidLeftJustified())
1887 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
1889 // Check that flags are not ignored by another flag
1890 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
1891 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
1892 startSpecifier, specifierLen);
1893 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
1894 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
1895 startSpecifier, specifierLen);
1897 // Check the length modifier is valid with the given conversion specifier.
1898 const LengthModifier &LM = FS.getLengthModifier();
1899 if (!FS.hasValidLengthModifier())
1900 S.Diag(getLocationOfByte(LM.getStart()),
1901 diag::warn_format_nonsensical_length)
1902 << LM.toString() << CS.toString()
1903 << getSpecifierRange(startSpecifier, specifierLen)
1904 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1907 // Are we using '%n'?
1908 if (CS.getKind() == ConversionSpecifier::nArg) {
1909 // Issue a warning about this being a possible security issue.
1910 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back)
1911 << getSpecifierRange(startSpecifier, specifierLen);
1912 // Continue checking the other format specifiers.
1916 // The remaining checks depend on the data arguments.
1920 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1923 // Now type check the data expression that matches the
1924 // format specifier.
1925 const Expr *Ex = getDataArg(argIndex);
1926 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context);
1927 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) {
1928 // Check if we didn't match because of an implicit cast from a 'char'
1929 // or 'short' to an 'int'. This is done because printf is a varargs
1931 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex))
1932 if (ICE->getType() == S.Context.IntTy) {
1933 // All further checking is done on the subexpression.
1934 Ex = ICE->getSubExpr();
1935 if (ATR.matchesType(S.Context, Ex->getType()))
1939 // We may be able to offer a FixItHint if it is a supported type.
1940 PrintfSpecifier fixedFS = FS;
1941 bool success = fixedFS.fixType(Ex->getType());
1944 // Get the fix string from the fixed format specifier
1945 llvm::SmallString<128> buf;
1946 llvm::raw_svector_ostream os(buf);
1947 fixedFS.toString(os);
1949 // FIXME: getRepresentativeType() perhaps should return a string
1950 // instead of a QualType to better handle when the representative
1951 // type is 'wint_t' (which is defined in the system headers).
1952 S.Diag(getLocationOfByte(CS.getStart()),
1953 diag::warn_printf_conversion_argument_type_mismatch)
1954 << ATR.getRepresentativeType(S.Context) << Ex->getType()
1955 << getSpecifierRange(startSpecifier, specifierLen)
1956 << Ex->getSourceRange()
1957 << FixItHint::CreateReplacement(
1958 getSpecifierRange(startSpecifier, specifierLen),
1962 S.Diag(getLocationOfByte(CS.getStart()),
1963 diag::warn_printf_conversion_argument_type_mismatch)
1964 << ATR.getRepresentativeType(S.Context) << Ex->getType()
1965 << getSpecifierRange(startSpecifier, specifierLen)
1966 << Ex->getSourceRange();
1973 //===--- CHECK: Scanf format string checking ------------------------------===//
1976 class CheckScanfHandler : public CheckFormatHandler {
1978 CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
1979 const Expr *origFormatExpr, unsigned firstDataArg,
1980 unsigned numDataArgs, bool isObjCLiteral,
1981 const char *beg, bool hasVAListArg,
1982 const CallExpr *theCall, unsigned formatIdx)
1983 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1984 numDataArgs, isObjCLiteral, beg, hasVAListArg,
1985 theCall, formatIdx) {}
1987 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
1988 const char *startSpecifier,
1989 unsigned specifierLen);
1991 bool HandleInvalidScanfConversionSpecifier(
1992 const analyze_scanf::ScanfSpecifier &FS,
1993 const char *startSpecifier,
1994 unsigned specifierLen);
1996 void HandleIncompleteScanList(const char *start, const char *end);
2000 void CheckScanfHandler::HandleIncompleteScanList(const char *start,
2002 S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete)
2003 << getSpecifierRange(start, end - start);
2006 bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
2007 const analyze_scanf::ScanfSpecifier &FS,
2008 const char *startSpecifier,
2009 unsigned specifierLen) {
2011 const analyze_scanf::ScanfConversionSpecifier &CS =
2012 FS.getConversionSpecifier();
2014 return HandleInvalidConversionSpecifier(FS.getArgIndex(),
2015 getLocationOfByte(CS.getStart()),
2016 startSpecifier, specifierLen,
2017 CS.getStart(), CS.getLength());
2020 bool CheckScanfHandler::HandleScanfSpecifier(
2021 const analyze_scanf::ScanfSpecifier &FS,
2022 const char *startSpecifier,
2023 unsigned specifierLen) {
2025 using namespace analyze_scanf;
2026 using namespace analyze_format_string;
2028 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
2030 // Handle case where '%' and '*' don't consume an argument. These shouldn't
2031 // be used to decide if we are using positional arguments consistently.
2032 if (FS.consumesDataArgument()) {
2035 usesPositionalArgs = FS.usesPositionalArg();
2037 else if (usesPositionalArgs != FS.usesPositionalArg()) {
2038 // Cannot mix-and-match positional and non-positional arguments.
2039 S.Diag(getLocationOfByte(CS.getStart()),
2040 diag::warn_format_mix_positional_nonpositional_args)
2041 << getSpecifierRange(startSpecifier, specifierLen);
2046 // Check if the field with is non-zero.
2047 const OptionalAmount &Amt = FS.getFieldWidth();
2048 if (Amt.getHowSpecified() == OptionalAmount::Constant) {
2049 if (Amt.getConstantAmount() == 0) {
2050 const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
2051 Amt.getConstantLength());
2052 S.Diag(getLocationOfByte(Amt.getStart()),
2053 diag::warn_scanf_nonzero_width)
2054 << R << FixItHint::CreateRemoval(R);
2058 if (!FS.consumesDataArgument()) {
2059 // FIXME: Technically specifying a precision or field width here
2060 // makes no sense. Worth issuing a warning at some point.
2064 // Consume the argument.
2065 unsigned argIndex = FS.getArgIndex();
2066 if (argIndex < NumDataArgs) {
2067 // The check to see if the argIndex is valid will come later.
2068 // We set the bit here because we may exit early from this
2069 // function if we encounter some other error.
2070 CoveredArgs.set(argIndex);
2073 // Check the length modifier is valid with the given conversion specifier.
2074 const LengthModifier &LM = FS.getLengthModifier();
2075 if (!FS.hasValidLengthModifier()) {
2076 S.Diag(getLocationOfByte(LM.getStart()),
2077 diag::warn_format_nonsensical_length)
2078 << LM.toString() << CS.toString()
2079 << getSpecifierRange(startSpecifier, specifierLen)
2080 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
2084 // The remaining checks depend on the data arguments.
2088 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
2091 // FIXME: Check that the argument type matches the format specifier.
2096 void Sema::CheckFormatString(const StringLiteral *FExpr,
2097 const Expr *OrigFormatExpr,
2098 const CallExpr *TheCall, bool HasVAListArg,
2099 unsigned format_idx, unsigned firstDataArg,
2102 // CHECK: is the format string a wide literal?
2103 if (!FExpr->isAscii()) {
2104 Diag(FExpr->getLocStart(),
2105 diag::warn_format_string_is_wide_literal)
2106 << OrigFormatExpr->getSourceRange();
2110 // Str - The format string. NOTE: this is NOT null-terminated!
2111 StringRef StrRef = FExpr->getString();
2112 const char *Str = StrRef.data();
2113 unsigned StrLen = StrRef.size();
2114 const unsigned numDataArgs = TheCall->getNumArgs() - firstDataArg;
2116 // CHECK: empty format string?
2117 if (StrLen == 0 && numDataArgs > 0) {
2118 Diag(FExpr->getLocStart(), diag::warn_empty_format_string)
2119 << OrigFormatExpr->getSourceRange();
2124 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
2125 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr),
2126 Str, HasVAListArg, TheCall, format_idx);
2128 bool FormatExtensions = getLangOptions().FormatExtensions;
2129 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen,
2134 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
2135 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr),
2136 Str, HasVAListArg, TheCall, format_idx);
2138 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen))
2143 //===--- CHECK: Standard memory functions ---------------------------------===//
2145 /// \brief Determine whether the given type is a dynamic class type (e.g.,
2146 /// whether it has a vtable).
2147 static bool isDynamicClassType(QualType T) {
2148 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl())
2149 if (CXXRecordDecl *Definition = Record->getDefinition())
2150 if (Definition->isDynamicClass())
2156 /// \brief If E is a sizeof expression, returns its argument expression,
2157 /// otherwise returns NULL.
2158 static const Expr *getSizeOfExprArg(const Expr* E) {
2159 if (const UnaryExprOrTypeTraitExpr *SizeOf =
2160 dyn_cast<UnaryExprOrTypeTraitExpr>(E))
2161 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType())
2162 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts();
2167 /// \brief If E is a sizeof expression, returns its argument type.
2168 static QualType getSizeOfArgType(const Expr* E) {
2169 if (const UnaryExprOrTypeTraitExpr *SizeOf =
2170 dyn_cast<UnaryExprOrTypeTraitExpr>(E))
2171 if (SizeOf->getKind() == clang::UETT_SizeOf)
2172 return SizeOf->getTypeOfArgument();
2177 /// \brief Check for dangerous or invalid arguments to memset().
2179 /// This issues warnings on known problematic, dangerous or unspecified
2180 /// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp'
2183 /// \param Call The call expression to diagnose.
2184 void Sema::CheckMemaccessArguments(const CallExpr *Call,
2185 CheckedMemoryFunction CMF,
2186 IdentifierInfo *FnName) {
2187 // It is possible to have a non-standard definition of memset. Validate
2188 // we have enough arguments, and if not, abort further checking.
2189 unsigned ExpectedNumArgs = (CMF == CMF_Strndup ? 2 : 3);
2190 if (Call->getNumArgs() < ExpectedNumArgs)
2193 unsigned LastArg = (CMF == CMF_Memset || CMF == CMF_Strndup ? 1 : 2);
2194 unsigned LenArg = (CMF == CMF_Strndup ? 1 : 2);
2195 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts();
2197 // We have special checking when the length is a sizeof expression.
2198 QualType SizeOfArgTy = getSizeOfArgType(LenExpr);
2199 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr);
2200 llvm::FoldingSetNodeID SizeOfArgID;
2202 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) {
2203 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts();
2204 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange();
2206 QualType DestTy = Dest->getType();
2207 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) {
2208 QualType PointeeTy = DestPtrTy->getPointeeType();
2210 // Never warn about void type pointers. This can be used to suppress
2212 if (PointeeTy->isVoidType())
2215 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by
2216 // actually comparing the expressions for equality. Because computing the
2217 // expression IDs can be expensive, we only do this if the diagnostic is
2220 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess,
2221 SizeOfArg->getExprLoc())) {
2222 // We only compute IDs for expressions if the warning is enabled, and
2223 // cache the sizeof arg's ID.
2224 if (SizeOfArgID == llvm::FoldingSetNodeID())
2225 SizeOfArg->Profile(SizeOfArgID, Context, true);
2226 llvm::FoldingSetNodeID DestID;
2227 Dest->Profile(DestID, Context, true);
2228 if (DestID == SizeOfArgID) {
2229 // TODO: For strncpy() and friends, this could suggest sizeof(dst)
2230 // over sizeof(src) as well.
2231 unsigned ActionIdx = 0; // Default is to suggest dereferencing.
2232 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest))
2233 if (UnaryOp->getOpcode() == UO_AddrOf)
2234 ActionIdx = 1; // If its an address-of operator, just remove it.
2235 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth())
2236 ActionIdx = 2; // If the pointee's size is sizeof(char),
2237 // suggest an explicit length.
2238 unsigned DestSrcSelect = (CMF == CMF_Strndup ? 1 : ArgIdx);
2239 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest,
2240 PDiag(diag::warn_sizeof_pointer_expr_memaccess)
2241 << FnName << DestSrcSelect << ActionIdx
2242 << Dest->getSourceRange()
2243 << SizeOfArg->getSourceRange());
2248 // Also check for cases where the sizeof argument is the exact same
2249 // type as the memory argument, and where it points to a user-defined
2251 if (SizeOfArgTy != QualType()) {
2252 if (PointeeTy->isRecordType() &&
2253 Context.typesAreCompatible(SizeOfArgTy, DestTy)) {
2254 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest,
2255 PDiag(diag::warn_sizeof_pointer_type_memaccess)
2256 << FnName << SizeOfArgTy << ArgIdx
2257 << PointeeTy << Dest->getSourceRange()
2258 << LenExpr->getSourceRange());
2263 // Always complain about dynamic classes.
2264 if (isDynamicClassType(PointeeTy))
2265 DiagRuntimeBehavior(
2266 Dest->getExprLoc(), Dest,
2267 PDiag(diag::warn_dyn_class_memaccess)
2268 << (CMF == CMF_Memcmp ? ArgIdx + 2 : ArgIdx) << FnName << PointeeTy
2269 // "overwritten" if we're warning about the destination for any call
2270 // but memcmp; otherwise a verb appropriate to the call.
2271 << (ArgIdx == 0 && CMF != CMF_Memcmp ? 0 : (unsigned)CMF)
2272 << Call->getCallee()->getSourceRange());
2273 else if (PointeeTy.hasNonTrivialObjCLifetime() && CMF != CMF_Memset)
2274 DiagRuntimeBehavior(
2275 Dest->getExprLoc(), Dest,
2276 PDiag(diag::warn_arc_object_memaccess)
2277 << ArgIdx << FnName << PointeeTy
2278 << Call->getCallee()->getSourceRange());
2282 DiagRuntimeBehavior(
2283 Dest->getExprLoc(), Dest,
2284 PDiag(diag::note_bad_memaccess_silence)
2285 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
2291 // A little helper routine: ignore addition and subtraction of integer literals.
2292 // This intentionally does not ignore all integer constant expressions because
2293 // we don't want to remove sizeof().
2294 static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) {
2295 Ex = Ex->IgnoreParenCasts();
2298 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex);
2299 if (!BO || !BO->isAdditiveOp())
2302 const Expr *RHS = BO->getRHS()->IgnoreParenCasts();
2303 const Expr *LHS = BO->getLHS()->IgnoreParenCasts();
2305 if (isa<IntegerLiteral>(RHS))
2307 else if (isa<IntegerLiteral>(LHS))
2316 // Warn if the user has made the 'size' argument to strlcpy or strlcat
2317 // be the size of the source, instead of the destination.
2318 void Sema::CheckStrlcpycatArguments(const CallExpr *Call,
2319 IdentifierInfo *FnName) {
2321 // Don't crash if the user has the wrong number of arguments
2322 if (Call->getNumArgs() != 3)
2325 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context);
2326 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context);
2327 const Expr *CompareWithSrc = NULL;
2329 // Look for 'strlcpy(dst, x, sizeof(x))'
2330 if (const Expr *Ex = getSizeOfExprArg(SizeArg))
2331 CompareWithSrc = Ex;
2333 // Look for 'strlcpy(dst, x, strlen(x))'
2334 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) {
2335 if (SizeCall->isBuiltinCall(Context) == Builtin::BIstrlen
2336 && SizeCall->getNumArgs() == 1)
2337 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context);
2341 if (!CompareWithSrc)
2344 // Determine if the argument to sizeof/strlen is equal to the source
2345 // argument. In principle there's all kinds of things you could do
2346 // here, for instance creating an == expression and evaluating it with
2347 // EvaluateAsBooleanCondition, but this uses a more direct technique:
2348 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg);
2352 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc);
2353 if (!CompareWithSrcDRE ||
2354 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl())
2357 const Expr *OriginalSizeArg = Call->getArg(2);
2358 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size)
2359 << OriginalSizeArg->getSourceRange() << FnName;
2361 // Output a FIXIT hint if the destination is an array (rather than a
2362 // pointer to an array). This could be enhanced to handle some
2363 // pointers if we know the actual size, like if DstArg is 'array+2'
2364 // we could say 'sizeof(array)-2'.
2365 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts();
2366 QualType DstArgTy = DstArg->getType();
2368 // Only handle constant-sized or VLAs, but not flexible members.
2369 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) {
2370 // Only issue the FIXIT for arrays of size > 1.
2371 if (CAT->getSize().getSExtValue() <= 1)
2373 } else if (!DstArgTy->isVariableArrayType()) {
2377 llvm::SmallString<128> sizeString;
2378 llvm::raw_svector_ostream OS(sizeString);
2380 DstArg->printPretty(OS, Context, 0, getPrintingPolicy());
2383 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size)
2384 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(),
2388 //===--- CHECK: Return Address of Stack Variable --------------------------===//
2390 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars);
2391 static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars);
2393 /// CheckReturnStackAddr - Check if a return statement returns the address
2394 /// of a stack variable.
2396 Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
2397 SourceLocation ReturnLoc) {
2400 SmallVector<DeclRefExpr *, 8> refVars;
2402 // Perform checking for returned stack addresses, local blocks,
2403 // label addresses or references to temporaries.
2404 if (lhsType->isPointerType() ||
2405 (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) {
2406 stackE = EvalAddr(RetValExp, refVars);
2407 } else if (lhsType->isReferenceType()) {
2408 stackE = EvalVal(RetValExp, refVars);
2412 return; // Nothing suspicious was found.
2414 SourceLocation diagLoc;
2415 SourceRange diagRange;
2416 if (refVars.empty()) {
2417 diagLoc = stackE->getLocStart();
2418 diagRange = stackE->getSourceRange();
2420 // We followed through a reference variable. 'stackE' contains the
2421 // problematic expression but we will warn at the return statement pointing
2422 // at the reference variable. We will later display the "trail" of
2423 // reference variables using notes.
2424 diagLoc = refVars[0]->getLocStart();
2425 diagRange = refVars[0]->getSourceRange();
2428 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var.
2429 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref
2430 : diag::warn_ret_stack_addr)
2431 << DR->getDecl()->getDeclName() << diagRange;
2432 } else if (isa<BlockExpr>(stackE)) { // local block.
2433 Diag(diagLoc, diag::err_ret_local_block) << diagRange;
2434 } else if (isa<AddrLabelExpr>(stackE)) { // address of label.
2435 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange;
2436 } else { // local temporary.
2437 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref
2438 : diag::warn_ret_local_temp_addr)
2442 // Display the "trail" of reference variables that we followed until we
2443 // found the problematic expression using notes.
2444 for (unsigned i = 0, e = refVars.size(); i != e; ++i) {
2445 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl());
2446 // If this var binds to another reference var, show the range of the next
2447 // var, otherwise the var binds to the problematic expression, in which case
2448 // show the range of the expression.
2449 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange()
2450 : stackE->getSourceRange();
2451 Diag(VD->getLocation(), diag::note_ref_var_local_bind)
2452 << VD->getDeclName() << range;
2456 /// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
2457 /// check if the expression in a return statement evaluates to an address
2458 /// to a location on the stack, a local block, an address of a label, or a
2459 /// reference to local temporary. The recursion is used to traverse the
2460 /// AST of the return expression, with recursion backtracking when we
2461 /// encounter a subexpression that (1) clearly does not lead to one of the
2462 /// above problematic expressions (2) is something we cannot determine leads to
2463 /// a problematic expression based on such local checking.
2465 /// Both EvalAddr and EvalVal follow through reference variables to evaluate
2466 /// the expression that they point to. Such variables are added to the
2467 /// 'refVars' vector so that we know what the reference variable "trail" was.
2469 /// EvalAddr processes expressions that are pointers that are used as
2470 /// references (and not L-values). EvalVal handles all other values.
2471 /// At the base case of the recursion is a check for the above problematic
2474 /// This implementation handles:
2476 /// * pointer-to-pointer casts
2477 /// * implicit conversions from array references to pointers
2478 /// * taking the address of fields
2479 /// * arbitrary interplay between "&" and "*" operators
2480 /// * pointer arithmetic from an address of a stack variable
2481 /// * taking the address of an array element where the array is on the stack
2482 static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) {
2483 if (E->isTypeDependent())
2486 // We should only be called for evaluating pointer expressions.
2487 assert((E->getType()->isAnyPointerType() ||
2488 E->getType()->isBlockPointerType() ||
2489 E->getType()->isObjCQualifiedIdType()) &&
2490 "EvalAddr only works on pointers");
2492 E = E->IgnoreParens();
2494 // Our "symbolic interpreter" is just a dispatch off the currently
2495 // viewed AST node. We then recursively traverse the AST by calling
2496 // EvalAddr and EvalVal appropriately.
2497 switch (E->getStmtClass()) {
2498 case Stmt::DeclRefExprClass: {
2499 DeclRefExpr *DR = cast<DeclRefExpr>(E);
2501 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
2502 // If this is a reference variable, follow through to the expression that
2504 if (V->hasLocalStorage() &&
2505 V->getType()->isReferenceType() && V->hasInit()) {
2506 // Add the reference variable to the "trail".
2507 refVars.push_back(DR);
2508 return EvalAddr(V->getInit(), refVars);
2514 case Stmt::UnaryOperatorClass: {
2515 // The only unary operator that make sense to handle here
2516 // is AddrOf. All others don't make sense as pointers.
2517 UnaryOperator *U = cast<UnaryOperator>(E);
2519 if (U->getOpcode() == UO_AddrOf)
2520 return EvalVal(U->getSubExpr(), refVars);
2525 case Stmt::BinaryOperatorClass: {
2526 // Handle pointer arithmetic. All other binary operators are not valid
2528 BinaryOperator *B = cast<BinaryOperator>(E);
2529 BinaryOperatorKind op = B->getOpcode();
2531 if (op != BO_Add && op != BO_Sub)
2534 Expr *Base = B->getLHS();
2536 // Determine which argument is the real pointer base. It could be
2537 // the RHS argument instead of the LHS.
2538 if (!Base->getType()->isPointerType()) Base = B->getRHS();
2540 assert (Base->getType()->isPointerType());
2541 return EvalAddr(Base, refVars);
2544 // For conditional operators we need to see if either the LHS or RHS are
2545 // valid DeclRefExpr*s. If one of them is valid, we return it.
2546 case Stmt::ConditionalOperatorClass: {
2547 ConditionalOperator *C = cast<ConditionalOperator>(E);
2549 // Handle the GNU extension for missing LHS.
2550 if (Expr *lhsExpr = C->getLHS()) {
2551 // In C++, we can have a throw-expression, which has 'void' type.
2552 if (!lhsExpr->getType()->isVoidType())
2553 if (Expr* LHS = EvalAddr(lhsExpr, refVars))
2557 // In C++, we can have a throw-expression, which has 'void' type.
2558 if (C->getRHS()->getType()->isVoidType())
2561 return EvalAddr(C->getRHS(), refVars);
2564 case Stmt::BlockExprClass:
2565 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures())
2566 return E; // local block.
2569 case Stmt::AddrLabelExprClass:
2570 return E; // address of label.
2572 // For casts, we need to handle conversions from arrays to
2573 // pointer values, and pointer-to-pointer conversions.
2574 case Stmt::ImplicitCastExprClass:
2575 case Stmt::CStyleCastExprClass:
2576 case Stmt::CXXFunctionalCastExprClass:
2577 case Stmt::ObjCBridgedCastExprClass: {
2578 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
2579 QualType T = SubExpr->getType();
2581 if (SubExpr->getType()->isPointerType() ||
2582 SubExpr->getType()->isBlockPointerType() ||
2583 SubExpr->getType()->isObjCQualifiedIdType())
2584 return EvalAddr(SubExpr, refVars);
2585 else if (T->isArrayType())
2586 return EvalVal(SubExpr, refVars);
2591 // C++ casts. For dynamic casts, static casts, and const casts, we
2592 // are always converting from a pointer-to-pointer, so we just blow
2593 // through the cast. In the case the dynamic cast doesn't fail (and
2594 // return NULL), we take the conservative route and report cases
2595 // where we return the address of a stack variable. For Reinterpre
2596 // FIXME: The comment about is wrong; we're not always converting
2597 // from pointer to pointer. I'm guessing that this code should also
2598 // handle references to objects.
2599 case Stmt::CXXStaticCastExprClass:
2600 case Stmt::CXXDynamicCastExprClass:
2601 case Stmt::CXXConstCastExprClass:
2602 case Stmt::CXXReinterpretCastExprClass: {
2603 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
2604 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
2605 return EvalAddr(S, refVars);
2610 case Stmt::MaterializeTemporaryExprClass:
2611 if (Expr *Result = EvalAddr(
2612 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
2618 // Everything else: we simply don't reason about them.
2625 /// EvalVal - This function is complements EvalAddr in the mutual recursion.
2626 /// See the comments for EvalAddr for more details.
2627 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) {
2629 // We should only be called for evaluating non-pointer expressions, or
2630 // expressions with a pointer type that are not used as references but instead
2631 // are l-values (e.g., DeclRefExpr with a pointer type).
2633 // Our "symbolic interpreter" is just a dispatch off the currently
2634 // viewed AST node. We then recursively traverse the AST by calling
2635 // EvalAddr and EvalVal appropriately.
2637 E = E->IgnoreParens();
2638 switch (E->getStmtClass()) {
2639 case Stmt::ImplicitCastExprClass: {
2640 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E);
2641 if (IE->getValueKind() == VK_LValue) {
2642 E = IE->getSubExpr();
2648 case Stmt::DeclRefExprClass: {
2649 // When we hit a DeclRefExpr we are looking at code that refers to a
2650 // variable's name. If it's not a reference variable we check if it has
2651 // local storage within the function, and if so, return the expression.
2652 DeclRefExpr *DR = cast<DeclRefExpr>(E);
2654 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
2655 if (V->hasLocalStorage()) {
2656 if (!V->getType()->isReferenceType())
2659 // Reference variable, follow through to the expression that
2662 // Add the reference variable to the "trail".
2663 refVars.push_back(DR);
2664 return EvalVal(V->getInit(), refVars);
2671 case Stmt::UnaryOperatorClass: {
2672 // The only unary operator that make sense to handle here
2673 // is Deref. All others don't resolve to a "name." This includes
2674 // handling all sorts of rvalues passed to a unary operator.
2675 UnaryOperator *U = cast<UnaryOperator>(E);
2677 if (U->getOpcode() == UO_Deref)
2678 return EvalAddr(U->getSubExpr(), refVars);
2683 case Stmt::ArraySubscriptExprClass: {
2684 // Array subscripts are potential references to data on the stack. We
2685 // retrieve the DeclRefExpr* for the array variable if it indeed
2686 // has local storage.
2687 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars);
2690 case Stmt::ConditionalOperatorClass: {
2691 // For conditional operators we need to see if either the LHS or RHS are
2692 // non-NULL Expr's. If one is non-NULL, we return it.
2693 ConditionalOperator *C = cast<ConditionalOperator>(E);
2695 // Handle the GNU extension for missing LHS.
2696 if (Expr *lhsExpr = C->getLHS())
2697 if (Expr *LHS = EvalVal(lhsExpr, refVars))
2700 return EvalVal(C->getRHS(), refVars);
2703 // Accesses to members are potential references to data on the stack.
2704 case Stmt::MemberExprClass: {
2705 MemberExpr *M = cast<MemberExpr>(E);
2707 // Check for indirect access. We only want direct field accesses.
2711 // Check whether the member type is itself a reference, in which case
2712 // we're not going to refer to the member, but to what the member refers to.
2713 if (M->getMemberDecl()->getType()->isReferenceType())
2716 return EvalVal(M->getBase(), refVars);
2719 case Stmt::MaterializeTemporaryExprClass:
2720 if (Expr *Result = EvalVal(
2721 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
2728 // Check that we don't return or take the address of a reference to a
2729 // temporary. This is only useful in C++.
2730 if (!E->isTypeDependent() && E->isRValue())
2733 // Everything else: we simply don't reason about them.
2739 //===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
2741 /// Check for comparisons of floating point operands using != and ==.
2742 /// Issue a warning if these are no self-comparisons, as they are not likely
2743 /// to do what the programmer intended.
2744 void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) {
2745 bool EmitWarning = true;
2747 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts();
2748 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts();
2750 // Special case: check for x == x (which is OK).
2751 // Do not emit warnings for such cases.
2752 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
2753 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
2754 if (DRL->getDecl() == DRR->getDecl())
2755 EmitWarning = false;
2758 // Special case: check for comparisons against literals that can be exactly
2759 // represented by APFloat. In such cases, do not emit a warning. This
2760 // is a heuristic: often comparison against such literals are used to
2761 // detect if a value in a variable has not changed. This clearly can
2762 // lead to false negatives.
2764 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
2766 EmitWarning = false;
2768 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
2770 EmitWarning = false;
2774 // Check for comparisons with builtin types.
2776 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
2777 if (CL->isBuiltinCall(Context))
2778 EmitWarning = false;
2781 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
2782 if (CR->isBuiltinCall(Context))
2783 EmitWarning = false;
2785 // Emit the diagnostic.
2787 Diag(Loc, diag::warn_floatingpoint_eq)
2788 << LHS->getSourceRange() << RHS->getSourceRange();
2791 //===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
2792 //===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
2796 /// Structure recording the 'active' range of an integer-valued
2799 /// The number of bits active in the int.
2802 /// True if the int is known not to have negative values.
2805 IntRange(unsigned Width, bool NonNegative)
2806 : Width(Width), NonNegative(NonNegative)
2809 /// Returns the range of the bool type.
2810 static IntRange forBoolType() {
2811 return IntRange(1, true);
2814 /// Returns the range of an opaque value of the given integral type.
2815 static IntRange forValueOfType(ASTContext &C, QualType T) {
2816 return forValueOfCanonicalType(C,
2817 T->getCanonicalTypeInternal().getTypePtr());
2820 /// Returns the range of an opaque value of a canonical integral type.
2821 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) {
2822 assert(T->isCanonicalUnqualified());
2824 if (const VectorType *VT = dyn_cast<VectorType>(T))
2825 T = VT->getElementType().getTypePtr();
2826 if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2827 T = CT->getElementType().getTypePtr();
2829 // For enum types, use the known bit width of the enumerators.
2830 if (const EnumType *ET = dyn_cast<EnumType>(T)) {
2831 EnumDecl *Enum = ET->getDecl();
2832 if (!Enum->isCompleteDefinition())
2833 return IntRange(C.getIntWidth(QualType(T, 0)), false);
2835 unsigned NumPositive = Enum->getNumPositiveBits();
2836 unsigned NumNegative = Enum->getNumNegativeBits();
2838 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0);
2841 const BuiltinType *BT = cast<BuiltinType>(T);
2842 assert(BT->isInteger());
2844 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2847 /// Returns the "target" range of a canonical integral type, i.e.
2848 /// the range of values expressible in the type.
2850 /// This matches forValueOfCanonicalType except that enums have the
2851 /// full range of their type, not the range of their enumerators.
2852 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) {
2853 assert(T->isCanonicalUnqualified());
2855 if (const VectorType *VT = dyn_cast<VectorType>(T))
2856 T = VT->getElementType().getTypePtr();
2857 if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2858 T = CT->getElementType().getTypePtr();
2859 if (const EnumType *ET = dyn_cast<EnumType>(T))
2860 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr();
2862 const BuiltinType *BT = cast<BuiltinType>(T);
2863 assert(BT->isInteger());
2865 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2868 /// Returns the supremum of two ranges: i.e. their conservative merge.
2869 static IntRange join(IntRange L, IntRange R) {
2870 return IntRange(std::max(L.Width, R.Width),
2871 L.NonNegative && R.NonNegative);
2874 /// Returns the infinum of two ranges: i.e. their aggressive merge.
2875 static IntRange meet(IntRange L, IntRange R) {
2876 return IntRange(std::min(L.Width, R.Width),
2877 L.NonNegative || R.NonNegative);
2881 IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) {
2882 if (value.isSigned() && value.isNegative())
2883 return IntRange(value.getMinSignedBits(), false);
2885 if (value.getBitWidth() > MaxWidth)
2886 value = value.trunc(MaxWidth);
2888 // isNonNegative() just checks the sign bit without considering
2890 return IntRange(value.getActiveBits(), true);
2893 IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
2894 unsigned MaxWidth) {
2896 return GetValueRange(C, result.getInt(), MaxWidth);
2898 if (result.isVector()) {
2899 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
2900 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
2901 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
2902 R = IntRange::join(R, El);
2907 if (result.isComplexInt()) {
2908 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
2909 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
2910 return IntRange::join(R, I);
2913 // This can happen with lossless casts to intptr_t of "based" lvalues.
2914 // Assume it might use arbitrary bits.
2915 // FIXME: The only reason we need to pass the type in here is to get
2916 // the sign right on this one case. It would be nice if APValue
2918 assert(result.isLValue());
2919 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType());
2922 /// Pseudo-evaluate the given integer expression, estimating the
2923 /// range of values it might take.
2925 /// \param MaxWidth - the width to which the value will be truncated
2926 IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
2927 E = E->IgnoreParens();
2929 // Try a full evaluation first.
2930 Expr::EvalResult result;
2931 if (E->Evaluate(result, C))
2932 return GetValueRange(C, result.Val, E->getType(), MaxWidth);
2934 // I think we only want to look through implicit casts here; if the
2935 // user has an explicit widening cast, we should treat the value as
2936 // being of the new, wider type.
2937 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
2938 if (CE->getCastKind() == CK_NoOp)
2939 return GetExprRange(C, CE->getSubExpr(), MaxWidth);
2941 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType());
2943 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast);
2945 // Assume that non-integer casts can span the full range of the type.
2947 return OutputTypeRange;
2950 = GetExprRange(C, CE->getSubExpr(),
2951 std::min(MaxWidth, OutputTypeRange.Width));
2953 // Bail out if the subexpr's range is as wide as the cast type.
2954 if (SubRange.Width >= OutputTypeRange.Width)
2955 return OutputTypeRange;
2957 // Otherwise, we take the smaller width, and we're non-negative if
2958 // either the output type or the subexpr is.
2959 return IntRange(SubRange.Width,
2960 SubRange.NonNegative || OutputTypeRange.NonNegative);
2963 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
2964 // If we can fold the condition, just take that operand.
2966 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
2967 return GetExprRange(C, CondResult ? CO->getTrueExpr()
2968 : CO->getFalseExpr(),
2971 // Otherwise, conservatively merge.
2972 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
2973 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
2974 return IntRange::join(L, R);
2977 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
2978 switch (BO->getOpcode()) {
2980 // Boolean-valued operations are single-bit and positive.
2989 return IntRange::forBoolType();
2991 // The type of the assignments is the type of the LHS, so the RHS
2992 // is not necessarily the same type.
3001 return IntRange::forValueOfType(C, E->getType());
3003 // Simple assignments just pass through the RHS, which will have
3004 // been coerced to the LHS type.
3007 return GetExprRange(C, BO->getRHS(), MaxWidth);
3009 // Operations with opaque sources are black-listed.
3012 return IntRange::forValueOfType(C, E->getType());
3014 // Bitwise-and uses the *infinum* of the two source ranges.
3017 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
3018 GetExprRange(C, BO->getRHS(), MaxWidth));
3020 // Left shift gets black-listed based on a judgement call.
3022 // ...except that we want to treat '1 << (blah)' as logically
3023 // positive. It's an important idiom.
3024 if (IntegerLiteral *I
3025 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
3026 if (I->getValue() == 1) {
3027 IntRange R = IntRange::forValueOfType(C, E->getType());
3028 return IntRange(R.Width, /*NonNegative*/ true);
3034 return IntRange::forValueOfType(C, E->getType());
3036 // Right shift by a constant can narrow its left argument.
3038 case BO_ShrAssign: {
3039 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
3041 // If the shift amount is a positive constant, drop the width by
3044 if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
3045 shift.isNonNegative()) {
3046 unsigned zext = shift.getZExtValue();
3047 if (zext >= L.Width)
3048 L.Width = (L.NonNegative ? 0 : 1);
3056 // Comma acts as its right operand.
3058 return GetExprRange(C, BO->getRHS(), MaxWidth);
3060 // Black-list pointer subtractions.
3062 if (BO->getLHS()->getType()->isPointerType())
3063 return IntRange::forValueOfType(C, E->getType());
3066 // The width of a division result is mostly determined by the size
3069 // Don't 'pre-truncate' the operands.
3070 unsigned opWidth = C.getIntWidth(E->getType());
3071 IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
3073 // If the divisor is constant, use that.
3074 llvm::APSInt divisor;
3075 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) {
3076 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor))
3077 if (log2 >= L.Width)
3078 L.Width = (L.NonNegative ? 0 : 1);
3080 L.Width = std::min(L.Width - log2, MaxWidth);
3084 // Otherwise, just use the LHS's width.
3085 IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
3086 return IntRange(L.Width, L.NonNegative && R.NonNegative);
3089 // The result of a remainder can't be larger than the result of
3092 // Don't 'pre-truncate' the operands.
3093 unsigned opWidth = C.getIntWidth(E->getType());
3094 IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
3095 IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
3097 IntRange meet = IntRange::meet(L, R);
3098 meet.Width = std::min(meet.Width, MaxWidth);
3102 // The default behavior is okay for these.
3110 // The default case is to treat the operation as if it were closed
3111 // on the narrowest type that encompasses both operands.
3112 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
3113 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
3114 return IntRange::join(L, R);
3117 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
3118 switch (UO->getOpcode()) {
3119 // Boolean-valued operations are white-listed.
3121 return IntRange::forBoolType();
3123 // Operations with opaque sources are black-listed.
3125 case UO_AddrOf: // should be impossible
3126 return IntRange::forValueOfType(C, E->getType());
3129 return GetExprRange(C, UO->getSubExpr(), MaxWidth);
3133 if (dyn_cast<OffsetOfExpr>(E)) {
3134 IntRange::forValueOfType(C, E->getType());
3137 if (FieldDecl *BitField = E->getBitField())
3138 return IntRange(BitField->getBitWidthValue(C),
3139 BitField->getType()->isUnsignedIntegerOrEnumerationType());
3141 return IntRange::forValueOfType(C, E->getType());
3144 IntRange GetExprRange(ASTContext &C, Expr *E) {
3145 return GetExprRange(C, E, C.getIntWidth(E->getType()));
3148 /// Checks whether the given value, which currently has the given
3149 /// source semantics, has the same value when coerced through the
3150 /// target semantics.
3151 bool IsSameFloatAfterCast(const llvm::APFloat &value,
3152 const llvm::fltSemantics &Src,
3153 const llvm::fltSemantics &Tgt) {
3154 llvm::APFloat truncated = value;
3157 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
3158 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
3160 return truncated.bitwiseIsEqual(value);
3163 /// Checks whether the given value, which currently has the given
3164 /// source semantics, has the same value when coerced through the
3165 /// target semantics.
3167 /// The value might be a vector of floats (or a complex number).
3168 bool IsSameFloatAfterCast(const APValue &value,
3169 const llvm::fltSemantics &Src,
3170 const llvm::fltSemantics &Tgt) {
3171 if (value.isFloat())
3172 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
3174 if (value.isVector()) {
3175 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
3176 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
3181 assert(value.isComplexFloat());
3182 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
3183 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
3186 void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC);
3188 static bool IsZero(Sema &S, Expr *E) {
3189 // Suppress cases where we are comparing against an enum constant.
3190 if (const DeclRefExpr *DR =
3191 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts()))
3192 if (isa<EnumConstantDecl>(DR->getDecl()))
3195 // Suppress cases where the '0' value is expanded from a macro.
3196 if (E->getLocStart().isMacroID())
3200 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
3203 static bool HasEnumType(Expr *E) {
3204 // Strip off implicit integral promotions.
3205 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
3206 if (ICE->getCastKind() != CK_IntegralCast &&
3207 ICE->getCastKind() != CK_NoOp)
3209 E = ICE->getSubExpr();
3212 return E->getType()->isEnumeralType();
3215 void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
3216 BinaryOperatorKind op = E->getOpcode();
3217 if (E->isValueDependent())
3220 if (op == BO_LT && IsZero(S, E->getRHS())) {
3221 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
3222 << "< 0" << "false" << HasEnumType(E->getLHS())
3223 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3224 } else if (op == BO_GE && IsZero(S, E->getRHS())) {
3225 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
3226 << ">= 0" << "true" << HasEnumType(E->getLHS())
3227 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3228 } else if (op == BO_GT && IsZero(S, E->getLHS())) {
3229 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
3230 << "0 >" << "false" << HasEnumType(E->getRHS())
3231 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3232 } else if (op == BO_LE && IsZero(S, E->getLHS())) {
3233 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
3234 << "0 <=" << "true" << HasEnumType(E->getRHS())
3235 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3239 /// Analyze the operands of the given comparison. Implements the
3240 /// fallback case from AnalyzeComparison.
3241 void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
3242 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
3243 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
3246 /// \brief Implements -Wsign-compare.
3248 /// \param E the binary operator to check for warnings
3249 void AnalyzeComparison(Sema &S, BinaryOperator *E) {
3250 // The type the comparison is being performed in.
3251 QualType T = E->getLHS()->getType();
3252 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
3253 && "comparison with mismatched types");
3255 // We don't do anything special if this isn't an unsigned integral
3256 // comparison: we're only interested in integral comparisons, and
3257 // signed comparisons only happen in cases we don't care to warn about.
3259 // We also don't care about value-dependent expressions or expressions
3260 // whose result is a constant.
3261 if (!T->hasUnsignedIntegerRepresentation()
3262 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context))
3263 return AnalyzeImpConvsInComparison(S, E);
3265 Expr *LHS = E->getLHS()->IgnoreParenImpCasts();
3266 Expr *RHS = E->getRHS()->IgnoreParenImpCasts();
3268 // Check to see if one of the (unmodified) operands is of different
3270 Expr *signedOperand, *unsignedOperand;
3271 if (LHS->getType()->hasSignedIntegerRepresentation()) {
3272 assert(!RHS->getType()->hasSignedIntegerRepresentation() &&
3273 "unsigned comparison between two signed integer expressions?");
3274 signedOperand = LHS;
3275 unsignedOperand = RHS;
3276 } else if (RHS->getType()->hasSignedIntegerRepresentation()) {
3277 signedOperand = RHS;
3278 unsignedOperand = LHS;
3280 CheckTrivialUnsignedComparison(S, E);
3281 return AnalyzeImpConvsInComparison(S, E);
3284 // Otherwise, calculate the effective range of the signed operand.
3285 IntRange signedRange = GetExprRange(S.Context, signedOperand);
3287 // Go ahead and analyze implicit conversions in the operands. Note
3288 // that we skip the implicit conversions on both sides.
3289 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc());
3290 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc());
3292 // If the signed range is non-negative, -Wsign-compare won't fire,
3293 // but we should still check for comparisons which are always true
3295 if (signedRange.NonNegative)
3296 return CheckTrivialUnsignedComparison(S, E);
3298 // For (in)equality comparisons, if the unsigned operand is a
3299 // constant which cannot collide with a overflowed signed operand,
3300 // then reinterpreting the signed operand as unsigned will not
3301 // change the result of the comparison.
3302 if (E->isEqualityOp()) {
3303 unsigned comparisonWidth = S.Context.getIntWidth(T);
3304 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
3306 // We should never be unable to prove that the unsigned operand is
3308 assert(unsignedRange.NonNegative && "unsigned range includes negative?");
3310 if (unsignedRange.Width < comparisonWidth)
3314 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison)
3315 << LHS->getType() << RHS->getType()
3316 << LHS->getSourceRange() << RHS->getSourceRange();
3319 /// Analyzes an attempt to assign the given value to a bitfield.
3321 /// Returns true if there was something fishy about the attempt.
3322 bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
3323 SourceLocation InitLoc) {
3324 assert(Bitfield->isBitField());
3325 if (Bitfield->isInvalidDecl())
3328 // White-list bool bitfields.
3329 if (Bitfield->getType()->isBooleanType())
3332 // Ignore value- or type-dependent expressions.
3333 if (Bitfield->getBitWidth()->isValueDependent() ||
3334 Bitfield->getBitWidth()->isTypeDependent() ||
3335 Init->isValueDependent() ||
3336 Init->isTypeDependent())
3339 Expr *OriginalInit = Init->IgnoreParenImpCasts();
3341 Expr::EvalResult InitValue;
3342 if (!OriginalInit->Evaluate(InitValue, S.Context) ||
3343 !InitValue.Val.isInt())
3346 const llvm::APSInt &Value = InitValue.Val.getInt();
3347 unsigned OriginalWidth = Value.getBitWidth();
3348 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context);
3350 if (OriginalWidth <= FieldWidth)
3353 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth);
3355 // It's fairly common to write values into signed bitfields
3356 // that, if sign-extended, would end up becoming a different
3357 // value. We don't want to warn about that.
3358 if (Value.isSigned() && Value.isNegative())
3359 TruncatedValue = TruncatedValue.sext(OriginalWidth);
3361 TruncatedValue = TruncatedValue.zext(OriginalWidth);
3363 if (Value == TruncatedValue)
3366 std::string PrettyValue = Value.toString(10);
3367 std::string PrettyTrunc = TruncatedValue.toString(10);
3369 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant)
3370 << PrettyValue << PrettyTrunc << OriginalInit->getType()
3371 << Init->getSourceRange();
3376 /// Analyze the given simple or compound assignment for warning-worthy
3378 void AnalyzeAssignment(Sema &S, BinaryOperator *E) {
3379 // Just recurse on the LHS.
3380 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
3382 // We want to recurse on the RHS as normal unless we're assigning to
3384 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) {
3385 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(),
3386 E->getOperatorLoc())) {
3387 // Recurse, ignoring any implicit conversions on the RHS.
3388 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(),
3389 E->getOperatorLoc());
3393 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
3396 /// Diagnose an implicit cast; purely a helper for CheckImplicitConversion.
3397 void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T,
3398 SourceLocation CContext, unsigned diag) {
3399 S.Diag(E->getExprLoc(), diag)
3400 << SourceType << T << E->getSourceRange() << SourceRange(CContext);
3403 /// Diagnose an implicit cast; purely a helper for CheckImplicitConversion.
3404 void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext,
3406 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag);
3409 /// Diagnose an implicit cast from a literal expression. Does not warn when the
3410 /// cast wouldn't lose information.
3411 void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T,
3412 SourceLocation CContext) {
3413 // Try to convert the literal exactly to an integer. If we can, don't warn.
3414 bool isExact = false;
3415 const llvm::APFloat &Value = FL->getValue();
3416 llvm::APSInt IntegerValue(S.Context.getIntWidth(T),
3417 T->hasUnsignedIntegerRepresentation());
3418 if (Value.convertToInteger(IntegerValue,
3419 llvm::APFloat::rmTowardZero, &isExact)
3420 == llvm::APFloat::opOK && isExact)
3423 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer)
3424 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext);
3427 std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) {
3428 if (!Range.Width) return "0";
3430 llvm::APSInt ValueInRange = Value;
3431 ValueInRange.setIsSigned(!Range.NonNegative);
3432 ValueInRange = ValueInRange.trunc(Range.Width);
3433 return ValueInRange.toString(10);
3436 static bool isFromSystemMacro(Sema &S, SourceLocation loc) {
3437 SourceManager &smgr = S.Context.getSourceManager();
3438 return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc));
3441 void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
3442 SourceLocation CC, bool *ICContext = 0) {
3443 if (E->isTypeDependent() || E->isValueDependent()) return;
3445 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
3446 const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
3447 if (Source == Target) return;
3448 if (Target->isDependentType()) return;
3450 // If the conversion context location is invalid don't complain. We also
3451 // don't want to emit a warning if the issue occurs from the expansion of
3452 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we
3453 // delay this check as long as possible. Once we detect we are in that
3454 // scenario, we just return.
3458 // Diagnose implicit casts to bool.
3459 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) {
3460 if (isa<StringLiteral>(E))
3461 // Warn on string literal to bool. Checks for string literals in logical
3462 // expressions, for instances, assert(0 && "error here"), is prevented
3463 // by a check in AnalyzeImplicitConversions().
3464 return DiagnoseImpCast(S, E, T, CC,
3465 diag::warn_impcast_string_literal_to_bool);
3466 return; // Other casts to bool are not checked.
3469 // Strip vector types.
3470 if (isa<VectorType>(Source)) {
3471 if (!isa<VectorType>(Target)) {
3472 if (isFromSystemMacro(S, CC))
3474 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar);
3477 // If the vector cast is cast between two vectors of the same size, it is
3478 // a bitcast, not a conversion.
3479 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target))
3482 Source = cast<VectorType>(Source)->getElementType().getTypePtr();
3483 Target = cast<VectorType>(Target)->getElementType().getTypePtr();
3486 // Strip complex types.
3487 if (isa<ComplexType>(Source)) {
3488 if (!isa<ComplexType>(Target)) {
3489 if (isFromSystemMacro(S, CC))
3492 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar);
3495 Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
3496 Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
3499 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
3500 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
3502 // If the source is floating point...
3503 if (SourceBT && SourceBT->isFloatingPoint()) {
3504 // ...and the target is floating point...
3505 if (TargetBT && TargetBT->isFloatingPoint()) {
3506 // ...then warn if we're dropping FP rank.
3508 // Builtin FP kinds are ordered by increasing FP rank.
3509 if (SourceBT->getKind() > TargetBT->getKind()) {
3510 // Don't warn about float constants that are precisely
3511 // representable in the target type.
3512 Expr::EvalResult result;
3513 if (E->Evaluate(result, S.Context)) {
3514 // Value might be a float, a float vector, or a float complex.
3515 if (IsSameFloatAfterCast(result.Val,
3516 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
3517 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
3521 if (isFromSystemMacro(S, CC))
3524 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision);
3529 // If the target is integral, always warn.
3530 if ((TargetBT && TargetBT->isInteger())) {
3531 if (isFromSystemMacro(S, CC))
3534 Expr *InnerE = E->IgnoreParenImpCasts();
3535 // We also want to warn on, e.g., "int i = -1.234"
3536 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE))
3537 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus)
3538 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts();
3540 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) {
3541 DiagnoseFloatingLiteralImpCast(S, FL, T, CC);
3543 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer);
3550 if (!Source->isIntegerType() || !Target->isIntegerType())
3553 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull)
3554 == Expr::NPCK_GNUNull) && Target->isIntegerType()) {
3555 S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer)
3556 << E->getSourceRange() << clang::SourceRange(CC);
3560 IntRange SourceRange = GetExprRange(S.Context, E);
3561 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target);
3563 if (SourceRange.Width > TargetRange.Width) {
3564 // If the source is a constant, use a default-on diagnostic.
3565 // TODO: this should happen for bitfield stores, too.
3566 llvm::APSInt Value(32);
3567 if (E->isIntegerConstantExpr(Value, S.Context)) {
3568 if (isFromSystemMacro(S, CC))
3571 std::string PrettySourceValue = Value.toString(10);
3572 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange);
3574 S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant)
3575 << PrettySourceValue << PrettyTargetValue
3576 << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC);
3580 // People want to build with -Wshorten-64-to-32 and not -Wconversion.
3581 if (isFromSystemMacro(S, CC))
3584 if (SourceRange.Width == 64 && TargetRange.Width == 32)
3585 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32);
3586 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision);
3589 if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
3590 (!TargetRange.NonNegative && SourceRange.NonNegative &&
3591 SourceRange.Width == TargetRange.Width)) {
3593 if (isFromSystemMacro(S, CC))
3596 unsigned DiagID = diag::warn_impcast_integer_sign;
3598 // Traditionally, gcc has warned about this under -Wsign-compare.
3599 // We also want to warn about it in -Wconversion.
3600 // So if -Wconversion is off, use a completely identical diagnostic
3601 // in the sign-compare group.
3602 // The conditional-checking code will
3604 DiagID = diag::warn_impcast_integer_sign_conditional;
3608 return DiagnoseImpCast(S, E, T, CC, DiagID);
3611 // Diagnose conversions between different enumeration types.
3612 // In C, we pretend that the type of an EnumConstantDecl is its enumeration
3613 // type, to give us better diagnostics.
3614 QualType SourceType = E->getType();
3615 if (!S.getLangOptions().CPlusPlus) {
3616 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
3617 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) {
3618 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext());
3619 SourceType = S.Context.getTypeDeclType(Enum);
3620 Source = S.Context.getCanonicalType(SourceType).getTypePtr();
3624 if (const EnumType *SourceEnum = Source->getAs<EnumType>())
3625 if (const EnumType *TargetEnum = Target->getAs<EnumType>())
3626 if ((SourceEnum->getDecl()->getIdentifier() ||
3627 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) &&
3628 (TargetEnum->getDecl()->getIdentifier() ||
3629 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) &&
3630 SourceEnum != TargetEnum) {
3631 if (isFromSystemMacro(S, CC))
3634 return DiagnoseImpCast(S, E, SourceType, T, CC,
3635 diag::warn_impcast_different_enum_types);
3641 void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T);
3643 void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
3644 SourceLocation CC, bool &ICContext) {
3645 E = E->IgnoreParenImpCasts();
3647 if (isa<ConditionalOperator>(E))
3648 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T);
3650 AnalyzeImplicitConversions(S, E, CC);
3651 if (E->getType() != T)
3652 return CheckImplicitConversion(S, E, T, CC, &ICContext);
3656 void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) {
3657 SourceLocation CC = E->getQuestionLoc();
3659 AnalyzeImplicitConversions(S, E->getCond(), CC);
3661 bool Suspicious = false;
3662 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious);
3663 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious);
3665 // If -Wconversion would have warned about either of the candidates
3666 // for a signedness conversion to the context type...
3667 if (!Suspicious) return;
3669 // ...but it's currently ignored...
3670 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional,
3674 // ...then check whether it would have warned about either of the
3675 // candidates for a signedness conversion to the condition type.
3676 if (E->getType() == T) return;
3679 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
3680 E->getType(), CC, &Suspicious);
3682 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
3683 E->getType(), CC, &Suspicious);
3686 /// AnalyzeImplicitConversions - Find and report any interesting
3687 /// implicit conversions in the given expression. There are a couple
3688 /// of competing diagnostics here, -Wconversion and -Wsign-compare.
3689 void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
3690 QualType T = OrigE->getType();
3691 Expr *E = OrigE->IgnoreParenImpCasts();
3693 if (E->isTypeDependent() || E->isValueDependent())
3696 // For conditional operators, we analyze the arguments as if they
3697 // were being fed directly into the output.
3698 if (isa<ConditionalOperator>(E)) {
3699 ConditionalOperator *CO = cast<ConditionalOperator>(E);
3700 CheckConditionalOperator(S, CO, T);
3704 // Go ahead and check any implicit conversions we might have skipped.
3705 // The non-canonical typecheck is just an optimization;
3706 // CheckImplicitConversion will filter out dead implicit conversions.
3707 if (E->getType() != T)
3708 CheckImplicitConversion(S, E, T, CC);
3710 // Now continue drilling into this expression.
3712 // Skip past explicit casts.
3713 if (isa<ExplicitCastExpr>(E)) {
3714 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
3715 return AnalyzeImplicitConversions(S, E, CC);
3718 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
3719 // Do a somewhat different check with comparison operators.
3720 if (BO->isComparisonOp())
3721 return AnalyzeComparison(S, BO);
3723 // And with assignments and compound assignments.
3724 if (BO->isAssignmentOp())
3725 return AnalyzeAssignment(S, BO);
3728 // These break the otherwise-useful invariant below. Fortunately,
3729 // we don't really need to recurse into them, because any internal
3730 // expressions should have been analyzed already when they were
3731 // built into statements.
3732 if (isa<StmtExpr>(E)) return;
3734 // Don't descend into unevaluated contexts.
3735 if (isa<UnaryExprOrTypeTraitExpr>(E)) return;
3737 // Now just recurse over the expression's children.
3738 CC = E->getExprLoc();
3739 BinaryOperator *BO = dyn_cast<BinaryOperator>(E);
3740 bool IsLogicalOperator = BO && BO->isLogicalOp();
3741 for (Stmt::child_range I = E->children(); I; ++I) {
3742 Expr *ChildExpr = cast<Expr>(*I);
3743 if (IsLogicalOperator &&
3744 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts()))
3745 // Ignore checking string literals that are in logical operators.
3747 AnalyzeImplicitConversions(S, ChildExpr, CC);
3751 } // end anonymous namespace
3753 /// Diagnoses "dangerous" implicit conversions within the given
3754 /// expression (which is a full expression). Implements -Wconversion
3755 /// and -Wsign-compare.
3757 /// \param CC the "context" location of the implicit conversion, i.e.
3758 /// the most location of the syntactic entity requiring the implicit
3760 void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) {
3761 // Don't diagnose in unevaluated contexts.
3762 if (ExprEvalContexts.back().Context == Sema::Unevaluated)
3765 // Don't diagnose for value- or type-dependent expressions.
3766 if (E->isTypeDependent() || E->isValueDependent())
3769 // Check for array bounds violations in cases where the check isn't triggered
3770 // elsewhere for other Expr types (like BinaryOperators), e.g. when an
3771 // ArraySubscriptExpr is on the RHS of a variable initialization.
3772 CheckArrayAccess(E);
3774 // This is not the right CC for (e.g.) a variable initialization.
3775 AnalyzeImplicitConversions(*this, E, CC);
3778 void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
3779 FieldDecl *BitField,
3781 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
3784 /// CheckParmsForFunctionDef - Check that the parameters of the given
3785 /// function are appropriate for the definition of a function. This
3786 /// takes care of any checks that cannot be performed on the
3787 /// declaration itself, e.g., that the types of each of the function
3788 /// parameters are complete.
3789 bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd,
3790 bool CheckParameterNames) {
3791 bool HasInvalidParm = false;
3792 for (; P != PEnd; ++P) {
3793 ParmVarDecl *Param = *P;
3795 // C99 6.7.5.3p4: the parameters in a parameter type list in a
3796 // function declarator that is part of a function definition of
3797 // that function shall not have incomplete type.
3799 // This is also C++ [dcl.fct]p6.
3800 if (!Param->isInvalidDecl() &&
3801 RequireCompleteType(Param->getLocation(), Param->getType(),
3802 diag::err_typecheck_decl_incomplete_type)) {
3803 Param->setInvalidDecl();
3804 HasInvalidParm = true;
3807 // C99 6.9.1p5: If the declarator includes a parameter type list, the
3808 // declaration of each parameter shall include an identifier.
3809 if (CheckParameterNames &&
3810 Param->getIdentifier() == 0 &&
3811 !Param->isImplicit() &&
3812 !getLangOptions().CPlusPlus)
3813 Diag(Param->getLocation(), diag::err_parameter_name_omitted);
3816 // If the function declarator is not part of a definition of that
3817 // function, parameters may have incomplete type and may use the [*]
3818 // notation in their sequences of declarator specifiers to specify
3819 // variable length array types.
3820 QualType PType = Param->getOriginalType();
3821 if (const ArrayType *AT = Context.getAsArrayType(PType)) {
3822 if (AT->getSizeModifier() == ArrayType::Star) {
3823 // FIXME: This diagnosic should point the the '[*]' if source-location
3824 // information is added for it.
3825 Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
3830 return HasInvalidParm;
3833 /// CheckCastAlign - Implements -Wcast-align, which warns when a
3834 /// pointer cast increases the alignment requirements.
3835 void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) {
3836 // This is actually a lot of work to potentially be doing on every
3837 // cast; don't do it if we're ignoring -Wcast_align (as is the default).
3838 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align,
3840 == DiagnosticsEngine::Ignored)
3843 // Ignore dependent types.
3844 if (T->isDependentType() || Op->getType()->isDependentType())
3847 // Require that the destination be a pointer type.
3848 const PointerType *DestPtr = T->getAs<PointerType>();
3849 if (!DestPtr) return;
3851 // If the destination has alignment 1, we're done.
3852 QualType DestPointee = DestPtr->getPointeeType();
3853 if (DestPointee->isIncompleteType()) return;
3854 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee);
3855 if (DestAlign.isOne()) return;
3857 // Require that the source be a pointer type.
3858 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>();
3859 if (!SrcPtr) return;
3860 QualType SrcPointee = SrcPtr->getPointeeType();
3862 // Whitelist casts from cv void*. We already implicitly
3863 // whitelisted casts to cv void*, since they have alignment 1.
3864 // Also whitelist casts involving incomplete types, which implicitly
3866 if (SrcPointee->isIncompleteType()) return;
3868 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee);
3869 if (SrcAlign >= DestAlign) return;
3871 Diag(TRange.getBegin(), diag::warn_cast_align)
3872 << Op->getType() << T
3873 << static_cast<unsigned>(SrcAlign.getQuantity())
3874 << static_cast<unsigned>(DestAlign.getQuantity())
3875 << TRange << Op->getSourceRange();
3878 static const Type* getElementType(const Expr *BaseExpr) {
3879 const Type* EltType = BaseExpr->getType().getTypePtr();
3880 if (EltType->isAnyPointerType())
3881 return EltType->getPointeeType().getTypePtr();
3882 else if (EltType->isArrayType())
3883 return EltType->getBaseElementTypeUnsafe();
3887 /// \brief Check whether this array fits the idiom of a size-one tail padded
3888 /// array member of a struct.
3890 /// We avoid emitting out-of-bounds access warnings for such arrays as they are
3891 /// commonly used to emulate flexible arrays in C89 code.
3892 static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size,
3893 const NamedDecl *ND) {
3894 if (Size != 1 || !ND) return false;
3896 const FieldDecl *FD = dyn_cast<FieldDecl>(ND);
3897 if (!FD) return false;
3899 // Don't consider sizes resulting from macro expansions or template argument
3900 // substitution to form C89 tail-padded arrays.
3901 ConstantArrayTypeLoc TL =
3902 cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc());
3903 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr());
3904 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID())
3907 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext());
3908 if (!RD || !RD->isStruct())
3911 // See if this is the last field decl in the record.
3913 while ((D = D->getNextDeclInContext()))
3914 if (isa<FieldDecl>(D))
3919 void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr,
3920 bool isSubscript, bool AllowOnePastEnd) {
3921 const Type* EffectiveType = getElementType(BaseExpr);
3922 BaseExpr = BaseExpr->IgnoreParenCasts();
3923 IndexExpr = IndexExpr->IgnoreParenCasts();
3925 const ConstantArrayType *ArrayTy =
3926 Context.getAsConstantArrayType(BaseExpr->getType());
3930 if (IndexExpr->isValueDependent())
3933 if (!IndexExpr->isIntegerConstantExpr(index, Context))
3936 const NamedDecl *ND = NULL;
3937 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
3938 ND = dyn_cast<NamedDecl>(DRE->getDecl());
3939 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
3940 ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
3942 if (index.isUnsigned() || !index.isNegative()) {
3943 llvm::APInt size = ArrayTy->getSize();
3944 if (!size.isStrictlyPositive())
3947 const Type* BaseType = getElementType(BaseExpr);
3948 if (BaseType != EffectiveType) {
3949 // Make sure we're comparing apples to apples when comparing index to size
3950 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType);
3951 uint64_t array_typesize = Context.getTypeSize(BaseType);
3952 // Handle ptrarith_typesize being zero, such as when casting to void*
3953 if (!ptrarith_typesize) ptrarith_typesize = 1;
3954 if (ptrarith_typesize != array_typesize) {
3955 // There's a cast to a different size type involved
3956 uint64_t ratio = array_typesize / ptrarith_typesize;
3957 // TODO: Be smarter about handling cases where array_typesize is not a
3958 // multiple of ptrarith_typesize
3959 if (ptrarith_typesize * ratio == array_typesize)
3960 size *= llvm::APInt(size.getBitWidth(), ratio);
3964 if (size.getBitWidth() > index.getBitWidth())
3965 index = index.sext(size.getBitWidth());
3966 else if (size.getBitWidth() < index.getBitWidth())
3967 size = size.sext(index.getBitWidth());
3969 // For array subscripting the index must be less than size, but for pointer
3970 // arithmetic also allow the index (offset) to be equal to size since
3971 // computing the next address after the end of the array is legal and
3972 // commonly done e.g. in C++ iterators and range-based for loops.
3973 if (AllowOnePastEnd ? index.sle(size) : index.slt(size))
3976 // Also don't warn for arrays of size 1 which are members of some
3977 // structure. These are often used to approximate flexible arrays in C89
3979 if (IsTailPaddedMemberArray(*this, size, ND))
3982 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds;
3984 DiagID = diag::warn_array_index_exceeds_bounds;
3986 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
3987 PDiag(DiagID) << index.toString(10, true)
3988 << size.toString(10, true)
3989 << (unsigned)size.getLimitedValue(~0U)
3990 << IndexExpr->getSourceRange());
3992 unsigned DiagID = diag::warn_array_index_precedes_bounds;
3994 DiagID = diag::warn_ptr_arith_precedes_bounds;
3995 if (index.isNegative()) index = -index;
3998 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
3999 PDiag(DiagID) << index.toString(10, true)
4000 << IndexExpr->getSourceRange());
4004 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr,
4005 PDiag(diag::note_array_index_out_of_bounds)
4006 << ND->getDeclName());
4009 void Sema::CheckArrayAccess(const Expr *expr) {
4010 int AllowOnePastEnd = 0;
4012 expr = expr->IgnoreParenImpCasts();
4013 switch (expr->getStmtClass()) {
4014 case Stmt::ArraySubscriptExprClass: {
4015 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr);
4016 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), true,
4017 AllowOnePastEnd > 0);
4020 case Stmt::UnaryOperatorClass: {
4021 // Only unwrap the * and & unary operators
4022 const UnaryOperator *UO = cast<UnaryOperator>(expr);
4023 expr = UO->getSubExpr();
4024 switch (UO->getOpcode()) {
4036 case Stmt::ConditionalOperatorClass: {
4037 const ConditionalOperator *cond = cast<ConditionalOperator>(expr);
4038 if (const Expr *lhs = cond->getLHS())
4039 CheckArrayAccess(lhs);
4040 if (const Expr *rhs = cond->getRHS())
4041 CheckArrayAccess(rhs);
4050 //===--- CHECK: Objective-C retain cycles ----------------------------------//
4053 struct RetainCycleOwner {
4054 RetainCycleOwner() : Variable(0), Indirect(false) {}
4060 void setLocsFrom(Expr *e) {
4061 Loc = e->getExprLoc();
4062 Range = e->getSourceRange();
4067 /// Consider whether capturing the given variable can possibly lead to
4069 static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) {
4070 // In ARC, it's captured strongly iff the variable has __strong
4071 // lifetime. In MRR, it's captured strongly if the variable is
4072 // __block and has an appropriate type.
4073 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
4076 owner.Variable = var;
4077 owner.setLocsFrom(ref);
4081 static bool findRetainCycleOwner(Expr *e, RetainCycleOwner &owner) {
4083 e = e->IgnoreParens();
4084 if (CastExpr *cast = dyn_cast<CastExpr>(e)) {
4085 switch (cast->getCastKind()) {
4087 case CK_LValueBitCast:
4088 case CK_LValueToRValue:
4089 case CK_ARCReclaimReturnedObject:
4090 e = cast->getSubExpr();
4093 case CK_GetObjCProperty: {
4094 // Bail out if this isn't a strong explicit property.
4095 const ObjCPropertyRefExpr *pre = cast->getSubExpr()->getObjCProperty();
4096 if (pre->isImplicitProperty()) return false;
4097 ObjCPropertyDecl *property = pre->getExplicitProperty();
4098 if (!property->isRetaining() &&
4099 !(property->getPropertyIvarDecl() &&
4100 property->getPropertyIvarDecl()->getType()
4101 .getObjCLifetime() == Qualifiers::OCL_Strong))
4104 owner.Indirect = true;
4105 e = const_cast<Expr*>(pre->getBase());
4114 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) {
4115 ObjCIvarDecl *ivar = ref->getDecl();
4116 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
4119 // Try to find a retain cycle in the base.
4120 if (!findRetainCycleOwner(ref->getBase(), owner))
4123 if (ref->isFreeIvar()) owner.setLocsFrom(ref);
4124 owner.Indirect = true;
4128 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) {
4129 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl());
4130 if (!var) return false;
4131 return considerVariable(var, ref, owner);
4134 if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) {
4135 owner.Variable = ref->getDecl();
4136 owner.setLocsFrom(ref);
4140 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) {
4141 if (member->isArrow()) return false;
4143 // Don't count this as an indirect ownership.
4144 e = member->getBase();
4155 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> {
4156 FindCaptureVisitor(ASTContext &Context, VarDecl *variable)
4157 : EvaluatedExprVisitor<FindCaptureVisitor>(Context),
4158 Variable(variable), Capturer(0) {}
4163 void VisitDeclRefExpr(DeclRefExpr *ref) {
4164 if (ref->getDecl() == Variable && !Capturer)
4168 void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) {
4169 if (ref->getDecl() == Variable && !Capturer)
4173 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) {
4174 if (Capturer) return;
4175 Visit(ref->getBase());
4176 if (Capturer && ref->isFreeIvar())
4180 void VisitBlockExpr(BlockExpr *block) {
4181 // Look inside nested blocks
4182 if (block->getBlockDecl()->capturesVariable(Variable))
4183 Visit(block->getBlockDecl()->getBody());
4188 /// Check whether the given argument is a block which captures a
4190 static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) {
4191 assert(owner.Variable && owner.Loc.isValid());
4193 e = e->IgnoreParenCasts();
4194 BlockExpr *block = dyn_cast<BlockExpr>(e);
4195 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable))
4198 FindCaptureVisitor visitor(S.Context, owner.Variable);
4199 visitor.Visit(block->getBlockDecl()->getBody());
4200 return visitor.Capturer;
4203 static void diagnoseRetainCycle(Sema &S, Expr *capturer,
4204 RetainCycleOwner &owner) {
4206 assert(owner.Variable && owner.Loc.isValid());
4208 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle)
4209 << owner.Variable << capturer->getSourceRange();
4210 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner)
4211 << owner.Indirect << owner.Range;
4214 /// Check for a keyword selector that starts with the word 'add' or
4216 static bool isSetterLikeSelector(Selector sel) {
4217 if (sel.isUnarySelector()) return false;
4219 StringRef str = sel.getNameForSlot(0);
4220 while (!str.empty() && str.front() == '_') str = str.substr(1);
4221 if (str.startswith("set") || str.startswith("add"))
4222 str = str.substr(3);
4226 if (str.empty()) return true;
4227 return !islower(str.front());
4230 /// Check a message send to see if it's likely to cause a retain cycle.
4231 void Sema::checkRetainCycles(ObjCMessageExpr *msg) {
4232 // Only check instance methods whose selector looks like a setter.
4233 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector()))
4236 // Try to find a variable that the receiver is strongly owned by.
4237 RetainCycleOwner owner;
4238 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) {
4239 if (!findRetainCycleOwner(msg->getInstanceReceiver(), owner))
4242 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance);
4243 owner.Variable = getCurMethodDecl()->getSelfDecl();
4244 owner.Loc = msg->getSuperLoc();
4245 owner.Range = msg->getSuperLoc();
4248 // Check whether the receiver is captured by any of the arguments.
4249 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i)
4250 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner))
4251 return diagnoseRetainCycle(*this, capturer, owner);
4254 /// Check a property assign to see if it's likely to cause a retain cycle.
4255 void Sema::checkRetainCycles(Expr *receiver, Expr *argument) {
4256 RetainCycleOwner owner;
4257 if (!findRetainCycleOwner(receiver, owner))
4260 if (Expr *capturer = findCapturingExpr(*this, argument, owner))
4261 diagnoseRetainCycle(*this, capturer, owner);
4264 bool Sema::checkUnsafeAssigns(SourceLocation Loc,
4265 QualType LHS, Expr *RHS) {
4266 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime();
4267 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone)
4269 // strip off any implicit cast added to get to the one arc-specific
4270 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
4271 if (cast->getCastKind() == CK_ARCConsumeObject) {
4272 Diag(Loc, diag::warn_arc_retained_assign)
4273 << (LT == Qualifiers::OCL_ExplicitNone)
4274 << RHS->getSourceRange();
4277 RHS = cast->getSubExpr();
4282 void Sema::checkUnsafeExprAssigns(SourceLocation Loc,
4283 Expr *LHS, Expr *RHS) {
4284 QualType LHSType = LHS->getType();
4285 if (checkUnsafeAssigns(Loc, LHSType, RHS))
4287 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime();
4288 // FIXME. Check for other life times.
4289 if (LT != Qualifiers::OCL_None)
4292 if (ObjCPropertyRefExpr *PRE = dyn_cast<ObjCPropertyRefExpr>(LHS)) {
4293 if (PRE->isImplicitProperty())
4295 const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
4299 unsigned Attributes = PD->getPropertyAttributes();
4300 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign)
4301 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
4302 if (cast->getCastKind() == CK_ARCConsumeObject) {
4303 Diag(Loc, diag::warn_arc_retained_property_assign)
4304 << RHS->getSourceRange();
4307 RHS = cast->getSubExpr();