SemaChecking.cpp revision 263508
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements extra semantic analysis beyond what is enforced 11// by the C type system. 12// 13//===----------------------------------------------------------------------===// 14 15#include "clang/Sema/SemaInternal.h" 16#include "clang/AST/ASTContext.h" 17#include "clang/AST/CharUnits.h" 18#include "clang/AST/DeclCXX.h" 19#include "clang/AST/DeclObjC.h" 20#include "clang/AST/EvaluatedExprVisitor.h" 21#include "clang/AST/Expr.h" 22#include "clang/AST/ExprCXX.h" 23#include "clang/AST/ExprObjC.h" 24#include "clang/AST/StmtCXX.h" 25#include "clang/AST/StmtObjC.h" 26#include "clang/Analysis/Analyses/FormatString.h" 27#include "clang/Basic/CharInfo.h" 28#include "clang/Basic/TargetBuiltins.h" 29#include "clang/Basic/TargetInfo.h" 30#include "clang/Lex/Preprocessor.h" 31#include "clang/Sema/Initialization.h" 32#include "clang/Sema/Lookup.h" 33#include "clang/Sema/ScopeInfo.h" 34#include "clang/Sema/Sema.h" 35#include "llvm/ADT/SmallBitVector.h" 36#include "llvm/ADT/SmallString.h" 37#include "llvm/ADT/STLExtras.h" 38#include "llvm/Support/ConvertUTF.h" 39#include "llvm/Support/raw_ostream.h" 40#include <limits> 41using namespace clang; 42using namespace sema; 43 44SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 45 unsigned ByteNo) const { 46 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 47 PP.getLangOpts(), PP.getTargetInfo()); 48} 49 50/// Checks that a call expression's argument count is the desired number. 51/// This is useful when doing custom type-checking. Returns true on error. 52static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 53 unsigned argCount = call->getNumArgs(); 54 if (argCount == desiredArgCount) return false; 55 56 if (argCount < desiredArgCount) 57 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 58 << 0 /*function call*/ << desiredArgCount << argCount 59 << call->getSourceRange(); 60 61 // Highlight all the excess arguments. 62 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 63 call->getArg(argCount - 1)->getLocEnd()); 64 65 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 66 << 0 /*function call*/ << desiredArgCount << argCount 67 << call->getArg(1)->getSourceRange(); 68} 69 70/// Check that the first argument to __builtin_annotation is an integer 71/// and the second argument is a non-wide string literal. 72static bool SemaBuiltinAnnotation(Sema &S, CallExpr *TheCall) { 73 if (checkArgCount(S, TheCall, 2)) 74 return true; 75 76 // First argument should be an integer. 77 Expr *ValArg = TheCall->getArg(0); 78 QualType Ty = ValArg->getType(); 79 if (!Ty->isIntegerType()) { 80 S.Diag(ValArg->getLocStart(), diag::err_builtin_annotation_first_arg) 81 << ValArg->getSourceRange(); 82 return true; 83 } 84 85 // Second argument should be a constant string. 86 Expr *StrArg = TheCall->getArg(1)->IgnoreParenCasts(); 87 StringLiteral *Literal = dyn_cast<StringLiteral>(StrArg); 88 if (!Literal || !Literal->isAscii()) { 89 S.Diag(StrArg->getLocStart(), diag::err_builtin_annotation_second_arg) 90 << StrArg->getSourceRange(); 91 return true; 92 } 93 94 TheCall->setType(Ty); 95 return false; 96} 97 98/// Check that the argument to __builtin_addressof is a glvalue, and set the 99/// result type to the corresponding pointer type. 100static bool SemaBuiltinAddressof(Sema &S, CallExpr *TheCall) { 101 if (checkArgCount(S, TheCall, 1)) 102 return true; 103 104 ExprResult Arg(S.Owned(TheCall->getArg(0))); 105 QualType ResultType = S.CheckAddressOfOperand(Arg, TheCall->getLocStart()); 106 if (ResultType.isNull()) 107 return true; 108 109 TheCall->setArg(0, Arg.take()); 110 TheCall->setType(ResultType); 111 return false; 112} 113 114ExprResult 115Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 116 ExprResult TheCallResult(Owned(TheCall)); 117 118 // Find out if any arguments are required to be integer constant expressions. 119 unsigned ICEArguments = 0; 120 ASTContext::GetBuiltinTypeError Error; 121 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 122 if (Error != ASTContext::GE_None) 123 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 124 125 // If any arguments are required to be ICE's, check and diagnose. 126 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 127 // Skip arguments not required to be ICE's. 128 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 129 130 llvm::APSInt Result; 131 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 132 return true; 133 ICEArguments &= ~(1 << ArgNo); 134 } 135 136 switch (BuiltinID) { 137 case Builtin::BI__builtin___CFStringMakeConstantString: 138 assert(TheCall->getNumArgs() == 1 && 139 "Wrong # arguments to builtin CFStringMakeConstantString"); 140 if (CheckObjCString(TheCall->getArg(0))) 141 return ExprError(); 142 break; 143 case Builtin::BI__builtin_stdarg_start: 144 case Builtin::BI__builtin_va_start: 145 if (SemaBuiltinVAStart(TheCall)) 146 return ExprError(); 147 break; 148 case Builtin::BI__builtin_isgreater: 149 case Builtin::BI__builtin_isgreaterequal: 150 case Builtin::BI__builtin_isless: 151 case Builtin::BI__builtin_islessequal: 152 case Builtin::BI__builtin_islessgreater: 153 case Builtin::BI__builtin_isunordered: 154 if (SemaBuiltinUnorderedCompare(TheCall)) 155 return ExprError(); 156 break; 157 case Builtin::BI__builtin_fpclassify: 158 if (SemaBuiltinFPClassification(TheCall, 6)) 159 return ExprError(); 160 break; 161 case Builtin::BI__builtin_isfinite: 162 case Builtin::BI__builtin_isinf: 163 case Builtin::BI__builtin_isinf_sign: 164 case Builtin::BI__builtin_isnan: 165 case Builtin::BI__builtin_isnormal: 166 if (SemaBuiltinFPClassification(TheCall, 1)) 167 return ExprError(); 168 break; 169 case Builtin::BI__builtin_shufflevector: 170 return SemaBuiltinShuffleVector(TheCall); 171 // TheCall will be freed by the smart pointer here, but that's fine, since 172 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 173 case Builtin::BI__builtin_prefetch: 174 if (SemaBuiltinPrefetch(TheCall)) 175 return ExprError(); 176 break; 177 case Builtin::BI__builtin_object_size: 178 if (SemaBuiltinObjectSize(TheCall)) 179 return ExprError(); 180 break; 181 case Builtin::BI__builtin_longjmp: 182 if (SemaBuiltinLongjmp(TheCall)) 183 return ExprError(); 184 break; 185 186 case Builtin::BI__builtin_classify_type: 187 if (checkArgCount(*this, TheCall, 1)) return true; 188 TheCall->setType(Context.IntTy); 189 break; 190 case Builtin::BI__builtin_constant_p: 191 if (checkArgCount(*this, TheCall, 1)) return true; 192 TheCall->setType(Context.IntTy); 193 break; 194 case Builtin::BI__sync_fetch_and_add: 195 case Builtin::BI__sync_fetch_and_add_1: 196 case Builtin::BI__sync_fetch_and_add_2: 197 case Builtin::BI__sync_fetch_and_add_4: 198 case Builtin::BI__sync_fetch_and_add_8: 199 case Builtin::BI__sync_fetch_and_add_16: 200 case Builtin::BI__sync_fetch_and_sub: 201 case Builtin::BI__sync_fetch_and_sub_1: 202 case Builtin::BI__sync_fetch_and_sub_2: 203 case Builtin::BI__sync_fetch_and_sub_4: 204 case Builtin::BI__sync_fetch_and_sub_8: 205 case Builtin::BI__sync_fetch_and_sub_16: 206 case Builtin::BI__sync_fetch_and_or: 207 case Builtin::BI__sync_fetch_and_or_1: 208 case Builtin::BI__sync_fetch_and_or_2: 209 case Builtin::BI__sync_fetch_and_or_4: 210 case Builtin::BI__sync_fetch_and_or_8: 211 case Builtin::BI__sync_fetch_and_or_16: 212 case Builtin::BI__sync_fetch_and_and: 213 case Builtin::BI__sync_fetch_and_and_1: 214 case Builtin::BI__sync_fetch_and_and_2: 215 case Builtin::BI__sync_fetch_and_and_4: 216 case Builtin::BI__sync_fetch_and_and_8: 217 case Builtin::BI__sync_fetch_and_and_16: 218 case Builtin::BI__sync_fetch_and_xor: 219 case Builtin::BI__sync_fetch_and_xor_1: 220 case Builtin::BI__sync_fetch_and_xor_2: 221 case Builtin::BI__sync_fetch_and_xor_4: 222 case Builtin::BI__sync_fetch_and_xor_8: 223 case Builtin::BI__sync_fetch_and_xor_16: 224 case Builtin::BI__sync_add_and_fetch: 225 case Builtin::BI__sync_add_and_fetch_1: 226 case Builtin::BI__sync_add_and_fetch_2: 227 case Builtin::BI__sync_add_and_fetch_4: 228 case Builtin::BI__sync_add_and_fetch_8: 229 case Builtin::BI__sync_add_and_fetch_16: 230 case Builtin::BI__sync_sub_and_fetch: 231 case Builtin::BI__sync_sub_and_fetch_1: 232 case Builtin::BI__sync_sub_and_fetch_2: 233 case Builtin::BI__sync_sub_and_fetch_4: 234 case Builtin::BI__sync_sub_and_fetch_8: 235 case Builtin::BI__sync_sub_and_fetch_16: 236 case Builtin::BI__sync_and_and_fetch: 237 case Builtin::BI__sync_and_and_fetch_1: 238 case Builtin::BI__sync_and_and_fetch_2: 239 case Builtin::BI__sync_and_and_fetch_4: 240 case Builtin::BI__sync_and_and_fetch_8: 241 case Builtin::BI__sync_and_and_fetch_16: 242 case Builtin::BI__sync_or_and_fetch: 243 case Builtin::BI__sync_or_and_fetch_1: 244 case Builtin::BI__sync_or_and_fetch_2: 245 case Builtin::BI__sync_or_and_fetch_4: 246 case Builtin::BI__sync_or_and_fetch_8: 247 case Builtin::BI__sync_or_and_fetch_16: 248 case Builtin::BI__sync_xor_and_fetch: 249 case Builtin::BI__sync_xor_and_fetch_1: 250 case Builtin::BI__sync_xor_and_fetch_2: 251 case Builtin::BI__sync_xor_and_fetch_4: 252 case Builtin::BI__sync_xor_and_fetch_8: 253 case Builtin::BI__sync_xor_and_fetch_16: 254 case Builtin::BI__sync_val_compare_and_swap: 255 case Builtin::BI__sync_val_compare_and_swap_1: 256 case Builtin::BI__sync_val_compare_and_swap_2: 257 case Builtin::BI__sync_val_compare_and_swap_4: 258 case Builtin::BI__sync_val_compare_and_swap_8: 259 case Builtin::BI__sync_val_compare_and_swap_16: 260 case Builtin::BI__sync_bool_compare_and_swap: 261 case Builtin::BI__sync_bool_compare_and_swap_1: 262 case Builtin::BI__sync_bool_compare_and_swap_2: 263 case Builtin::BI__sync_bool_compare_and_swap_4: 264 case Builtin::BI__sync_bool_compare_and_swap_8: 265 case Builtin::BI__sync_bool_compare_and_swap_16: 266 case Builtin::BI__sync_lock_test_and_set: 267 case Builtin::BI__sync_lock_test_and_set_1: 268 case Builtin::BI__sync_lock_test_and_set_2: 269 case Builtin::BI__sync_lock_test_and_set_4: 270 case Builtin::BI__sync_lock_test_and_set_8: 271 case Builtin::BI__sync_lock_test_and_set_16: 272 case Builtin::BI__sync_lock_release: 273 case Builtin::BI__sync_lock_release_1: 274 case Builtin::BI__sync_lock_release_2: 275 case Builtin::BI__sync_lock_release_4: 276 case Builtin::BI__sync_lock_release_8: 277 case Builtin::BI__sync_lock_release_16: 278 case Builtin::BI__sync_swap: 279 case Builtin::BI__sync_swap_1: 280 case Builtin::BI__sync_swap_2: 281 case Builtin::BI__sync_swap_4: 282 case Builtin::BI__sync_swap_8: 283 case Builtin::BI__sync_swap_16: 284 return SemaBuiltinAtomicOverloaded(TheCallResult); 285#define BUILTIN(ID, TYPE, ATTRS) 286#define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \ 287 case Builtin::BI##ID: \ 288 return SemaAtomicOpsOverloaded(TheCallResult, AtomicExpr::AO##ID); 289#include "clang/Basic/Builtins.def" 290 case Builtin::BI__builtin_annotation: 291 if (SemaBuiltinAnnotation(*this, TheCall)) 292 return ExprError(); 293 break; 294 case Builtin::BI__builtin_addressof: 295 if (SemaBuiltinAddressof(*this, TheCall)) 296 return ExprError(); 297 break; 298 } 299 300 // Since the target specific builtins for each arch overlap, only check those 301 // of the arch we are compiling for. 302 if (BuiltinID >= Builtin::FirstTSBuiltin) { 303 switch (Context.getTargetInfo().getTriple().getArch()) { 304 case llvm::Triple::arm: 305 case llvm::Triple::thumb: 306 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 307 return ExprError(); 308 break; 309 case llvm::Triple::aarch64: 310 if (CheckAArch64BuiltinFunctionCall(BuiltinID, TheCall)) 311 return ExprError(); 312 break; 313 case llvm::Triple::mips: 314 case llvm::Triple::mipsel: 315 case llvm::Triple::mips64: 316 case llvm::Triple::mips64el: 317 if (CheckMipsBuiltinFunctionCall(BuiltinID, TheCall)) 318 return ExprError(); 319 break; 320 default: 321 break; 322 } 323 } 324 325 return TheCallResult; 326} 327 328// Get the valid immediate range for the specified NEON type code. 329static unsigned RFT(unsigned t, bool shift = false) { 330 NeonTypeFlags Type(t); 331 int IsQuad = Type.isQuad(); 332 switch (Type.getEltType()) { 333 case NeonTypeFlags::Int8: 334 case NeonTypeFlags::Poly8: 335 return shift ? 7 : (8 << IsQuad) - 1; 336 case NeonTypeFlags::Int16: 337 case NeonTypeFlags::Poly16: 338 return shift ? 15 : (4 << IsQuad) - 1; 339 case NeonTypeFlags::Int32: 340 return shift ? 31 : (2 << IsQuad) - 1; 341 case NeonTypeFlags::Int64: 342 case NeonTypeFlags::Poly64: 343 return shift ? 63 : (1 << IsQuad) - 1; 344 case NeonTypeFlags::Float16: 345 assert(!shift && "cannot shift float types!"); 346 return (4 << IsQuad) - 1; 347 case NeonTypeFlags::Float32: 348 assert(!shift && "cannot shift float types!"); 349 return (2 << IsQuad) - 1; 350 case NeonTypeFlags::Float64: 351 assert(!shift && "cannot shift float types!"); 352 return (1 << IsQuad) - 1; 353 } 354 llvm_unreachable("Invalid NeonTypeFlag!"); 355} 356 357/// getNeonEltType - Return the QualType corresponding to the elements of 358/// the vector type specified by the NeonTypeFlags. This is used to check 359/// the pointer arguments for Neon load/store intrinsics. 360static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context, 361 bool IsAArch64) { 362 switch (Flags.getEltType()) { 363 case NeonTypeFlags::Int8: 364 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 365 case NeonTypeFlags::Int16: 366 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 367 case NeonTypeFlags::Int32: 368 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 369 case NeonTypeFlags::Int64: 370 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 371 case NeonTypeFlags::Poly8: 372 return IsAArch64 ? Context.UnsignedCharTy : Context.SignedCharTy; 373 case NeonTypeFlags::Poly16: 374 return IsAArch64 ? Context.UnsignedShortTy : Context.ShortTy; 375 case NeonTypeFlags::Poly64: 376 return Context.UnsignedLongLongTy; 377 case NeonTypeFlags::Float16: 378 return Context.HalfTy; 379 case NeonTypeFlags::Float32: 380 return Context.FloatTy; 381 case NeonTypeFlags::Float64: 382 return Context.DoubleTy; 383 } 384 llvm_unreachable("Invalid NeonTypeFlag!"); 385} 386 387bool Sema::CheckAArch64BuiltinFunctionCall(unsigned BuiltinID, 388 CallExpr *TheCall) { 389 390 llvm::APSInt Result; 391 392 uint64_t mask = 0; 393 unsigned TV = 0; 394 int PtrArgNum = -1; 395 bool HasConstPtr = false; 396 switch (BuiltinID) { 397#define GET_NEON_AARCH64_OVERLOAD_CHECK 398#include "clang/Basic/arm_neon.inc" 399#undef GET_NEON_AARCH64_OVERLOAD_CHECK 400 } 401 402 // For NEON intrinsics which are overloaded on vector element type, validate 403 // the immediate which specifies which variant to emit. 404 unsigned ImmArg = TheCall->getNumArgs() - 1; 405 if (mask) { 406 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 407 return true; 408 409 TV = Result.getLimitedValue(64); 410 if ((TV > 63) || (mask & (1ULL << TV)) == 0) 411 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 412 << TheCall->getArg(ImmArg)->getSourceRange(); 413 } 414 415 if (PtrArgNum >= 0) { 416 // Check that pointer arguments have the specified type. 417 Expr *Arg = TheCall->getArg(PtrArgNum); 418 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 419 Arg = ICE->getSubExpr(); 420 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 421 QualType RHSTy = RHS.get()->getType(); 422 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context, true); 423 if (HasConstPtr) 424 EltTy = EltTy.withConst(); 425 QualType LHSTy = Context.getPointerType(EltTy); 426 AssignConvertType ConvTy; 427 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 428 if (RHS.isInvalid()) 429 return true; 430 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 431 RHS.get(), AA_Assigning)) 432 return true; 433 } 434 435 // For NEON intrinsics which take an immediate value as part of the 436 // instruction, range check them here. 437 unsigned i = 0, l = 0, u = 0; 438 switch (BuiltinID) { 439 default: 440 return false; 441#define GET_NEON_AARCH64_IMMEDIATE_CHECK 442#include "clang/Basic/arm_neon.inc" 443#undef GET_NEON_AARCH64_IMMEDIATE_CHECK 444 } 445 ; 446 447 // We can't check the value of a dependent argument. 448 if (TheCall->getArg(i)->isTypeDependent() || 449 TheCall->getArg(i)->isValueDependent()) 450 return false; 451 452 // Check that the immediate argument is actually a constant. 453 if (SemaBuiltinConstantArg(TheCall, i, Result)) 454 return true; 455 456 // Range check against the upper/lower values for this isntruction. 457 unsigned Val = Result.getZExtValue(); 458 if (Val < l || Val > (u + l)) 459 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 460 << l << u + l << TheCall->getArg(i)->getSourceRange(); 461 462 return false; 463} 464 465bool Sema::CheckARMBuiltinExclusiveCall(unsigned BuiltinID, CallExpr *TheCall) { 466 assert((BuiltinID == ARM::BI__builtin_arm_ldrex || 467 BuiltinID == ARM::BI__builtin_arm_strex) && 468 "unexpected ARM builtin"); 469 bool IsLdrex = BuiltinID == ARM::BI__builtin_arm_ldrex; 470 471 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 472 473 // Ensure that we have the proper number of arguments. 474 if (checkArgCount(*this, TheCall, IsLdrex ? 1 : 2)) 475 return true; 476 477 // Inspect the pointer argument of the atomic builtin. This should always be 478 // a pointer type, whose element is an integral scalar or pointer type. 479 // Because it is a pointer type, we don't have to worry about any implicit 480 // casts here. 481 Expr *PointerArg = TheCall->getArg(IsLdrex ? 0 : 1); 482 ExprResult PointerArgRes = DefaultFunctionArrayLvalueConversion(PointerArg); 483 if (PointerArgRes.isInvalid()) 484 return true; 485 PointerArg = PointerArgRes.take(); 486 487 const PointerType *pointerType = PointerArg->getType()->getAs<PointerType>(); 488 if (!pointerType) { 489 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 490 << PointerArg->getType() << PointerArg->getSourceRange(); 491 return true; 492 } 493 494 // ldrex takes a "const volatile T*" and strex takes a "volatile T*". Our next 495 // task is to insert the appropriate casts into the AST. First work out just 496 // what the appropriate type is. 497 QualType ValType = pointerType->getPointeeType(); 498 QualType AddrType = ValType.getUnqualifiedType().withVolatile(); 499 if (IsLdrex) 500 AddrType.addConst(); 501 502 // Issue a warning if the cast is dodgy. 503 CastKind CastNeeded = CK_NoOp; 504 if (!AddrType.isAtLeastAsQualifiedAs(ValType)) { 505 CastNeeded = CK_BitCast; 506 Diag(DRE->getLocStart(), diag::ext_typecheck_convert_discards_qualifiers) 507 << PointerArg->getType() 508 << Context.getPointerType(AddrType) 509 << AA_Passing << PointerArg->getSourceRange(); 510 } 511 512 // Finally, do the cast and replace the argument with the corrected version. 513 AddrType = Context.getPointerType(AddrType); 514 PointerArgRes = ImpCastExprToType(PointerArg, AddrType, CastNeeded); 515 if (PointerArgRes.isInvalid()) 516 return true; 517 PointerArg = PointerArgRes.take(); 518 519 TheCall->setArg(IsLdrex ? 0 : 1, PointerArg); 520 521 // In general, we allow ints, floats and pointers to be loaded and stored. 522 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 523 !ValType->isBlockPointerType() && !ValType->isFloatingType()) { 524 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intfltptr) 525 << PointerArg->getType() << PointerArg->getSourceRange(); 526 return true; 527 } 528 529 // But ARM doesn't have instructions to deal with 128-bit versions. 530 if (Context.getTypeSize(ValType) > 64) { 531 Diag(DRE->getLocStart(), diag::err_atomic_exclusive_builtin_pointer_size) 532 << PointerArg->getType() << PointerArg->getSourceRange(); 533 return true; 534 } 535 536 switch (ValType.getObjCLifetime()) { 537 case Qualifiers::OCL_None: 538 case Qualifiers::OCL_ExplicitNone: 539 // okay 540 break; 541 542 case Qualifiers::OCL_Weak: 543 case Qualifiers::OCL_Strong: 544 case Qualifiers::OCL_Autoreleasing: 545 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 546 << ValType << PointerArg->getSourceRange(); 547 return true; 548 } 549 550 551 if (IsLdrex) { 552 TheCall->setType(ValType); 553 return false; 554 } 555 556 // Initialize the argument to be stored. 557 ExprResult ValArg = TheCall->getArg(0); 558 InitializedEntity Entity = InitializedEntity::InitializeParameter( 559 Context, ValType, /*consume*/ false); 560 ValArg = PerformCopyInitialization(Entity, SourceLocation(), ValArg); 561 if (ValArg.isInvalid()) 562 return true; 563 TheCall->setArg(0, ValArg.get()); 564 565 // __builtin_arm_strex always returns an int. It's marked as such in the .def, 566 // but the custom checker bypasses all default analysis. 567 TheCall->setType(Context.IntTy); 568 return false; 569} 570 571bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 572 llvm::APSInt Result; 573 574 if (BuiltinID == ARM::BI__builtin_arm_ldrex || 575 BuiltinID == ARM::BI__builtin_arm_strex) { 576 return CheckARMBuiltinExclusiveCall(BuiltinID, TheCall); 577 } 578 579 uint64_t mask = 0; 580 unsigned TV = 0; 581 int PtrArgNum = -1; 582 bool HasConstPtr = false; 583 switch (BuiltinID) { 584#define GET_NEON_OVERLOAD_CHECK 585#include "clang/Basic/arm_neon.inc" 586#undef GET_NEON_OVERLOAD_CHECK 587 } 588 589 // For NEON intrinsics which are overloaded on vector element type, validate 590 // the immediate which specifies which variant to emit. 591 unsigned ImmArg = TheCall->getNumArgs()-1; 592 if (mask) { 593 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 594 return true; 595 596 TV = Result.getLimitedValue(64); 597 if ((TV > 63) || (mask & (1ULL << TV)) == 0) 598 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 599 << TheCall->getArg(ImmArg)->getSourceRange(); 600 } 601 602 if (PtrArgNum >= 0) { 603 // Check that pointer arguments have the specified type. 604 Expr *Arg = TheCall->getArg(PtrArgNum); 605 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 606 Arg = ICE->getSubExpr(); 607 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 608 QualType RHSTy = RHS.get()->getType(); 609 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context, false); 610 if (HasConstPtr) 611 EltTy = EltTy.withConst(); 612 QualType LHSTy = Context.getPointerType(EltTy); 613 AssignConvertType ConvTy; 614 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 615 if (RHS.isInvalid()) 616 return true; 617 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 618 RHS.get(), AA_Assigning)) 619 return true; 620 } 621 622 // For NEON intrinsics which take an immediate value as part of the 623 // instruction, range check them here. 624 unsigned i = 0, l = 0, u = 0; 625 switch (BuiltinID) { 626 default: return false; 627 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 628 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 629 case ARM::BI__builtin_arm_vcvtr_f: 630 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 631 case ARM::BI__builtin_arm_dmb: 632 case ARM::BI__builtin_arm_dsb: l = 0; u = 15; break; 633#define GET_NEON_IMMEDIATE_CHECK 634#include "clang/Basic/arm_neon.inc" 635#undef GET_NEON_IMMEDIATE_CHECK 636 }; 637 638 // We can't check the value of a dependent argument. 639 if (TheCall->getArg(i)->isTypeDependent() || 640 TheCall->getArg(i)->isValueDependent()) 641 return false; 642 643 // Check that the immediate argument is actually a constant. 644 if (SemaBuiltinConstantArg(TheCall, i, Result)) 645 return true; 646 647 // Range check against the upper/lower values for this isntruction. 648 unsigned Val = Result.getZExtValue(); 649 if (Val < l || Val > (u + l)) 650 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 651 << l << u+l << TheCall->getArg(i)->getSourceRange(); 652 653 // FIXME: VFP Intrinsics should error if VFP not present. 654 return false; 655} 656 657bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 658 unsigned i = 0, l = 0, u = 0; 659 switch (BuiltinID) { 660 default: return false; 661 case Mips::BI__builtin_mips_wrdsp: i = 1; l = 0; u = 63; break; 662 case Mips::BI__builtin_mips_rddsp: i = 0; l = 0; u = 63; break; 663 case Mips::BI__builtin_mips_append: i = 2; l = 0; u = 31; break; 664 case Mips::BI__builtin_mips_balign: i = 2; l = 0; u = 3; break; 665 case Mips::BI__builtin_mips_precr_sra_ph_w: i = 2; l = 0; u = 31; break; 666 case Mips::BI__builtin_mips_precr_sra_r_ph_w: i = 2; l = 0; u = 31; break; 667 case Mips::BI__builtin_mips_prepend: i = 2; l = 0; u = 31; break; 668 }; 669 670 // We can't check the value of a dependent argument. 671 if (TheCall->getArg(i)->isTypeDependent() || 672 TheCall->getArg(i)->isValueDependent()) 673 return false; 674 675 // Check that the immediate argument is actually a constant. 676 llvm::APSInt Result; 677 if (SemaBuiltinConstantArg(TheCall, i, Result)) 678 return true; 679 680 // Range check against the upper/lower values for this instruction. 681 unsigned Val = Result.getZExtValue(); 682 if (Val < l || Val > u) 683 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 684 << l << u << TheCall->getArg(i)->getSourceRange(); 685 686 return false; 687} 688 689/// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo 690/// parameter with the FormatAttr's correct format_idx and firstDataArg. 691/// Returns true when the format fits the function and the FormatStringInfo has 692/// been populated. 693bool Sema::getFormatStringInfo(const FormatAttr *Format, bool IsCXXMember, 694 FormatStringInfo *FSI) { 695 FSI->HasVAListArg = Format->getFirstArg() == 0; 696 FSI->FormatIdx = Format->getFormatIdx() - 1; 697 FSI->FirstDataArg = FSI->HasVAListArg ? 0 : Format->getFirstArg() - 1; 698 699 // The way the format attribute works in GCC, the implicit this argument 700 // of member functions is counted. However, it doesn't appear in our own 701 // lists, so decrement format_idx in that case. 702 if (IsCXXMember) { 703 if(FSI->FormatIdx == 0) 704 return false; 705 --FSI->FormatIdx; 706 if (FSI->FirstDataArg != 0) 707 --FSI->FirstDataArg; 708 } 709 return true; 710} 711 712/// Handles the checks for format strings, non-POD arguments to vararg 713/// functions, and NULL arguments passed to non-NULL parameters. 714void Sema::checkCall(NamedDecl *FDecl, 715 ArrayRef<const Expr *> Args, 716 unsigned NumProtoArgs, 717 bool IsMemberFunction, 718 SourceLocation Loc, 719 SourceRange Range, 720 VariadicCallType CallType) { 721 // FIXME: We should check as much as we can in the template definition. 722 if (CurContext->isDependentContext()) 723 return; 724 725 // Printf and scanf checking. 726 llvm::SmallBitVector CheckedVarArgs; 727 if (FDecl) { 728 for (specific_attr_iterator<FormatAttr> 729 I = FDecl->specific_attr_begin<FormatAttr>(), 730 E = FDecl->specific_attr_end<FormatAttr>(); 731 I != E; ++I) { 732 // Only create vector if there are format attributes. 733 CheckedVarArgs.resize(Args.size()); 734 735 CheckFormatArguments(*I, Args, IsMemberFunction, CallType, Loc, Range, 736 CheckedVarArgs); 737 } 738 } 739 740 // Refuse POD arguments that weren't caught by the format string 741 // checks above. 742 if (CallType != VariadicDoesNotApply) { 743 for (unsigned ArgIdx = NumProtoArgs; ArgIdx < Args.size(); ++ArgIdx) { 744 // Args[ArgIdx] can be null in malformed code. 745 if (const Expr *Arg = Args[ArgIdx]) { 746 if (CheckedVarArgs.empty() || !CheckedVarArgs[ArgIdx]) 747 checkVariadicArgument(Arg, CallType); 748 } 749 } 750 } 751 752 if (FDecl) { 753 for (specific_attr_iterator<NonNullAttr> 754 I = FDecl->specific_attr_begin<NonNullAttr>(), 755 E = FDecl->specific_attr_end<NonNullAttr>(); I != E; ++I) 756 CheckNonNullArguments(*I, Args.data(), Loc); 757 758 // Type safety checking. 759 for (specific_attr_iterator<ArgumentWithTypeTagAttr> 760 i = FDecl->specific_attr_begin<ArgumentWithTypeTagAttr>(), 761 e = FDecl->specific_attr_end<ArgumentWithTypeTagAttr>(); 762 i != e; ++i) { 763 CheckArgumentWithTypeTag(*i, Args.data()); 764 } 765 } 766} 767 768/// CheckConstructorCall - Check a constructor call for correctness and safety 769/// properties not enforced by the C type system. 770void Sema::CheckConstructorCall(FunctionDecl *FDecl, 771 ArrayRef<const Expr *> Args, 772 const FunctionProtoType *Proto, 773 SourceLocation Loc) { 774 VariadicCallType CallType = 775 Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply; 776 checkCall(FDecl, Args, Proto->getNumArgs(), 777 /*IsMemberFunction=*/true, Loc, SourceRange(), CallType); 778} 779 780/// CheckFunctionCall - Check a direct function call for various correctness 781/// and safety properties not strictly enforced by the C type system. 782bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall, 783 const FunctionProtoType *Proto) { 784 bool IsMemberOperatorCall = isa<CXXOperatorCallExpr>(TheCall) && 785 isa<CXXMethodDecl>(FDecl); 786 bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall) || 787 IsMemberOperatorCall; 788 VariadicCallType CallType = getVariadicCallType(FDecl, Proto, 789 TheCall->getCallee()); 790 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 791 Expr** Args = TheCall->getArgs(); 792 unsigned NumArgs = TheCall->getNumArgs(); 793 if (IsMemberOperatorCall) { 794 // If this is a call to a member operator, hide the first argument 795 // from checkCall. 796 // FIXME: Our choice of AST representation here is less than ideal. 797 ++Args; 798 --NumArgs; 799 } 800 checkCall(FDecl, llvm::makeArrayRef<const Expr *>(Args, NumArgs), 801 NumProtoArgs, 802 IsMemberFunction, TheCall->getRParenLoc(), 803 TheCall->getCallee()->getSourceRange(), CallType); 804 805 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 806 // None of the checks below are needed for functions that don't have 807 // simple names (e.g., C++ conversion functions). 808 if (!FnInfo) 809 return false; 810 811 unsigned CMId = FDecl->getMemoryFunctionKind(); 812 if (CMId == 0) 813 return false; 814 815 // Handle memory setting and copying functions. 816 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 817 CheckStrlcpycatArguments(TheCall, FnInfo); 818 else if (CMId == Builtin::BIstrncat) 819 CheckStrncatArguments(TheCall, FnInfo); 820 else 821 CheckMemaccessArguments(TheCall, CMId, FnInfo); 822 823 return false; 824} 825 826bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 827 ArrayRef<const Expr *> Args) { 828 VariadicCallType CallType = 829 Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply; 830 831 checkCall(Method, Args, Method->param_size(), 832 /*IsMemberFunction=*/false, 833 lbrac, Method->getSourceRange(), CallType); 834 835 return false; 836} 837 838bool Sema::CheckPointerCall(NamedDecl *NDecl, CallExpr *TheCall, 839 const FunctionProtoType *Proto) { 840 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 841 if (!V) 842 return false; 843 844 QualType Ty = V->getType(); 845 if (!Ty->isBlockPointerType() && !Ty->isFunctionPointerType()) 846 return false; 847 848 VariadicCallType CallType; 849 if (!Proto || !Proto->isVariadic()) { 850 CallType = VariadicDoesNotApply; 851 } else if (Ty->isBlockPointerType()) { 852 CallType = VariadicBlock; 853 } else { // Ty->isFunctionPointerType() 854 CallType = VariadicFunction; 855 } 856 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 857 858 checkCall(NDecl, 859 llvm::makeArrayRef<const Expr *>(TheCall->getArgs(), 860 TheCall->getNumArgs()), 861 NumProtoArgs, /*IsMemberFunction=*/false, 862 TheCall->getRParenLoc(), 863 TheCall->getCallee()->getSourceRange(), CallType); 864 865 return false; 866} 867 868/// Checks function calls when a FunctionDecl or a NamedDecl is not available, 869/// such as function pointers returned from functions. 870bool Sema::CheckOtherCall(CallExpr *TheCall, const FunctionProtoType *Proto) { 871 VariadicCallType CallType = getVariadicCallType(/*FDecl=*/0, Proto, 872 TheCall->getCallee()); 873 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 874 875 checkCall(/*FDecl=*/0, 876 llvm::makeArrayRef<const Expr *>(TheCall->getArgs(), 877 TheCall->getNumArgs()), 878 NumProtoArgs, /*IsMemberFunction=*/false, 879 TheCall->getRParenLoc(), 880 TheCall->getCallee()->getSourceRange(), CallType); 881 882 return false; 883} 884 885ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 886 AtomicExpr::AtomicOp Op) { 887 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 888 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 889 890 // All these operations take one of the following forms: 891 enum { 892 // C __c11_atomic_init(A *, C) 893 Init, 894 // C __c11_atomic_load(A *, int) 895 Load, 896 // void __atomic_load(A *, CP, int) 897 Copy, 898 // C __c11_atomic_add(A *, M, int) 899 Arithmetic, 900 // C __atomic_exchange_n(A *, CP, int) 901 Xchg, 902 // void __atomic_exchange(A *, C *, CP, int) 903 GNUXchg, 904 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 905 C11CmpXchg, 906 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 907 GNUCmpXchg 908 } Form = Init; 909 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 910 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 911 // where: 912 // C is an appropriate type, 913 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 914 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 915 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 916 // the int parameters are for orderings. 917 918 assert(AtomicExpr::AO__c11_atomic_init == 0 && 919 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 920 && "need to update code for modified C11 atomics"); 921 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 922 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 923 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 924 Op == AtomicExpr::AO__atomic_store_n || 925 Op == AtomicExpr::AO__atomic_exchange_n || 926 Op == AtomicExpr::AO__atomic_compare_exchange_n; 927 bool IsAddSub = false; 928 929 switch (Op) { 930 case AtomicExpr::AO__c11_atomic_init: 931 Form = Init; 932 break; 933 934 case AtomicExpr::AO__c11_atomic_load: 935 case AtomicExpr::AO__atomic_load_n: 936 Form = Load; 937 break; 938 939 case AtomicExpr::AO__c11_atomic_store: 940 case AtomicExpr::AO__atomic_load: 941 case AtomicExpr::AO__atomic_store: 942 case AtomicExpr::AO__atomic_store_n: 943 Form = Copy; 944 break; 945 946 case AtomicExpr::AO__c11_atomic_fetch_add: 947 case AtomicExpr::AO__c11_atomic_fetch_sub: 948 case AtomicExpr::AO__atomic_fetch_add: 949 case AtomicExpr::AO__atomic_fetch_sub: 950 case AtomicExpr::AO__atomic_add_fetch: 951 case AtomicExpr::AO__atomic_sub_fetch: 952 IsAddSub = true; 953 // Fall through. 954 case AtomicExpr::AO__c11_atomic_fetch_and: 955 case AtomicExpr::AO__c11_atomic_fetch_or: 956 case AtomicExpr::AO__c11_atomic_fetch_xor: 957 case AtomicExpr::AO__atomic_fetch_and: 958 case AtomicExpr::AO__atomic_fetch_or: 959 case AtomicExpr::AO__atomic_fetch_xor: 960 case AtomicExpr::AO__atomic_fetch_nand: 961 case AtomicExpr::AO__atomic_and_fetch: 962 case AtomicExpr::AO__atomic_or_fetch: 963 case AtomicExpr::AO__atomic_xor_fetch: 964 case AtomicExpr::AO__atomic_nand_fetch: 965 Form = Arithmetic; 966 break; 967 968 case AtomicExpr::AO__c11_atomic_exchange: 969 case AtomicExpr::AO__atomic_exchange_n: 970 Form = Xchg; 971 break; 972 973 case AtomicExpr::AO__atomic_exchange: 974 Form = GNUXchg; 975 break; 976 977 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 978 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 979 Form = C11CmpXchg; 980 break; 981 982 case AtomicExpr::AO__atomic_compare_exchange: 983 case AtomicExpr::AO__atomic_compare_exchange_n: 984 Form = GNUCmpXchg; 985 break; 986 } 987 988 // Check we have the right number of arguments. 989 if (TheCall->getNumArgs() < NumArgs[Form]) { 990 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 991 << 0 << NumArgs[Form] << TheCall->getNumArgs() 992 << TheCall->getCallee()->getSourceRange(); 993 return ExprError(); 994 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 995 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 996 diag::err_typecheck_call_too_many_args) 997 << 0 << NumArgs[Form] << TheCall->getNumArgs() 998 << TheCall->getCallee()->getSourceRange(); 999 return ExprError(); 1000 } 1001 1002 // Inspect the first argument of the atomic operation. 1003 Expr *Ptr = TheCall->getArg(0); 1004 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 1005 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 1006 if (!pointerType) { 1007 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 1008 << Ptr->getType() << Ptr->getSourceRange(); 1009 return ExprError(); 1010 } 1011 1012 // For a __c11 builtin, this should be a pointer to an _Atomic type. 1013 QualType AtomTy = pointerType->getPointeeType(); // 'A' 1014 QualType ValType = AtomTy; // 'C' 1015 if (IsC11) { 1016 if (!AtomTy->isAtomicType()) { 1017 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 1018 << Ptr->getType() << Ptr->getSourceRange(); 1019 return ExprError(); 1020 } 1021 if (AtomTy.isConstQualified()) { 1022 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_non_const_atomic) 1023 << Ptr->getType() << Ptr->getSourceRange(); 1024 return ExprError(); 1025 } 1026 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 1027 } 1028 1029 // For an arithmetic operation, the implied arithmetic must be well-formed. 1030 if (Form == Arithmetic) { 1031 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 1032 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 1033 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 1034 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 1035 return ExprError(); 1036 } 1037 if (!IsAddSub && !ValType->isIntegerType()) { 1038 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 1039 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 1040 return ExprError(); 1041 } 1042 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 1043 // For __atomic_*_n operations, the value type must be a scalar integral or 1044 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 1045 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 1046 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 1047 return ExprError(); 1048 } 1049 1050 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context) && 1051 !AtomTy->isScalarType()) { 1052 // For GNU atomics, require a trivially-copyable type. This is not part of 1053 // the GNU atomics specification, but we enforce it for sanity. 1054 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 1055 << Ptr->getType() << Ptr->getSourceRange(); 1056 return ExprError(); 1057 } 1058 1059 // FIXME: For any builtin other than a load, the ValType must not be 1060 // const-qualified. 1061 1062 switch (ValType.getObjCLifetime()) { 1063 case Qualifiers::OCL_None: 1064 case Qualifiers::OCL_ExplicitNone: 1065 // okay 1066 break; 1067 1068 case Qualifiers::OCL_Weak: 1069 case Qualifiers::OCL_Strong: 1070 case Qualifiers::OCL_Autoreleasing: 1071 // FIXME: Can this happen? By this point, ValType should be known 1072 // to be trivially copyable. 1073 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1074 << ValType << Ptr->getSourceRange(); 1075 return ExprError(); 1076 } 1077 1078 QualType ResultType = ValType; 1079 if (Form == Copy || Form == GNUXchg || Form == Init) 1080 ResultType = Context.VoidTy; 1081 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 1082 ResultType = Context.BoolTy; 1083 1084 // The type of a parameter passed 'by value'. In the GNU atomics, such 1085 // arguments are actually passed as pointers. 1086 QualType ByValType = ValType; // 'CP' 1087 if (!IsC11 && !IsN) 1088 ByValType = Ptr->getType(); 1089 1090 // The first argument --- the pointer --- has a fixed type; we 1091 // deduce the types of the rest of the arguments accordingly. Walk 1092 // the remaining arguments, converting them to the deduced value type. 1093 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 1094 QualType Ty; 1095 if (i < NumVals[Form] + 1) { 1096 switch (i) { 1097 case 1: 1098 // The second argument is the non-atomic operand. For arithmetic, this 1099 // is always passed by value, and for a compare_exchange it is always 1100 // passed by address. For the rest, GNU uses by-address and C11 uses 1101 // by-value. 1102 assert(Form != Load); 1103 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 1104 Ty = ValType; 1105 else if (Form == Copy || Form == Xchg) 1106 Ty = ByValType; 1107 else if (Form == Arithmetic) 1108 Ty = Context.getPointerDiffType(); 1109 else 1110 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 1111 break; 1112 case 2: 1113 // The third argument to compare_exchange / GNU exchange is a 1114 // (pointer to a) desired value. 1115 Ty = ByValType; 1116 break; 1117 case 3: 1118 // The fourth argument to GNU compare_exchange is a 'weak' flag. 1119 Ty = Context.BoolTy; 1120 break; 1121 } 1122 } else { 1123 // The order(s) are always converted to int. 1124 Ty = Context.IntTy; 1125 } 1126 1127 InitializedEntity Entity = 1128 InitializedEntity::InitializeParameter(Context, Ty, false); 1129 ExprResult Arg = TheCall->getArg(i); 1130 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1131 if (Arg.isInvalid()) 1132 return true; 1133 TheCall->setArg(i, Arg.get()); 1134 } 1135 1136 // Permute the arguments into a 'consistent' order. 1137 SmallVector<Expr*, 5> SubExprs; 1138 SubExprs.push_back(Ptr); 1139 switch (Form) { 1140 case Init: 1141 // Note, AtomicExpr::getVal1() has a special case for this atomic. 1142 SubExprs.push_back(TheCall->getArg(1)); // Val1 1143 break; 1144 case Load: 1145 SubExprs.push_back(TheCall->getArg(1)); // Order 1146 break; 1147 case Copy: 1148 case Arithmetic: 1149 case Xchg: 1150 SubExprs.push_back(TheCall->getArg(2)); // Order 1151 SubExprs.push_back(TheCall->getArg(1)); // Val1 1152 break; 1153 case GNUXchg: 1154 // Note, AtomicExpr::getVal2() has a special case for this atomic. 1155 SubExprs.push_back(TheCall->getArg(3)); // Order 1156 SubExprs.push_back(TheCall->getArg(1)); // Val1 1157 SubExprs.push_back(TheCall->getArg(2)); // Val2 1158 break; 1159 case C11CmpXchg: 1160 SubExprs.push_back(TheCall->getArg(3)); // Order 1161 SubExprs.push_back(TheCall->getArg(1)); // Val1 1162 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 1163 SubExprs.push_back(TheCall->getArg(2)); // Val2 1164 break; 1165 case GNUCmpXchg: 1166 SubExprs.push_back(TheCall->getArg(4)); // Order 1167 SubExprs.push_back(TheCall->getArg(1)); // Val1 1168 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 1169 SubExprs.push_back(TheCall->getArg(2)); // Val2 1170 SubExprs.push_back(TheCall->getArg(3)); // Weak 1171 break; 1172 } 1173 1174 AtomicExpr *AE = new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 1175 SubExprs, ResultType, Op, 1176 TheCall->getRParenLoc()); 1177 1178 if ((Op == AtomicExpr::AO__c11_atomic_load || 1179 (Op == AtomicExpr::AO__c11_atomic_store)) && 1180 Context.AtomicUsesUnsupportedLibcall(AE)) 1181 Diag(AE->getLocStart(), diag::err_atomic_load_store_uses_lib) << 1182 ((Op == AtomicExpr::AO__c11_atomic_load) ? 0 : 1); 1183 1184 return Owned(AE); 1185} 1186 1187 1188/// checkBuiltinArgument - Given a call to a builtin function, perform 1189/// normal type-checking on the given argument, updating the call in 1190/// place. This is useful when a builtin function requires custom 1191/// type-checking for some of its arguments but not necessarily all of 1192/// them. 1193/// 1194/// Returns true on error. 1195static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 1196 FunctionDecl *Fn = E->getDirectCallee(); 1197 assert(Fn && "builtin call without direct callee!"); 1198 1199 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 1200 InitializedEntity Entity = 1201 InitializedEntity::InitializeParameter(S.Context, Param); 1202 1203 ExprResult Arg = E->getArg(0); 1204 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 1205 if (Arg.isInvalid()) 1206 return true; 1207 1208 E->setArg(ArgIndex, Arg.take()); 1209 return false; 1210} 1211 1212/// SemaBuiltinAtomicOverloaded - We have a call to a function like 1213/// __sync_fetch_and_add, which is an overloaded function based on the pointer 1214/// type of its first argument. The main ActOnCallExpr routines have already 1215/// promoted the types of arguments because all of these calls are prototyped as 1216/// void(...). 1217/// 1218/// This function goes through and does final semantic checking for these 1219/// builtins, 1220ExprResult 1221Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 1222 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 1223 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1224 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1225 1226 // Ensure that we have at least one argument to do type inference from. 1227 if (TheCall->getNumArgs() < 1) { 1228 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1229 << 0 << 1 << TheCall->getNumArgs() 1230 << TheCall->getCallee()->getSourceRange(); 1231 return ExprError(); 1232 } 1233 1234 // Inspect the first argument of the atomic builtin. This should always be 1235 // a pointer type, whose element is an integral scalar or pointer type. 1236 // Because it is a pointer type, we don't have to worry about any implicit 1237 // casts here. 1238 // FIXME: We don't allow floating point scalars as input. 1239 Expr *FirstArg = TheCall->getArg(0); 1240 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 1241 if (FirstArgResult.isInvalid()) 1242 return ExprError(); 1243 FirstArg = FirstArgResult.take(); 1244 TheCall->setArg(0, FirstArg); 1245 1246 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 1247 if (!pointerType) { 1248 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 1249 << FirstArg->getType() << FirstArg->getSourceRange(); 1250 return ExprError(); 1251 } 1252 1253 QualType ValType = pointerType->getPointeeType(); 1254 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 1255 !ValType->isBlockPointerType()) { 1256 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 1257 << FirstArg->getType() << FirstArg->getSourceRange(); 1258 return ExprError(); 1259 } 1260 1261 switch (ValType.getObjCLifetime()) { 1262 case Qualifiers::OCL_None: 1263 case Qualifiers::OCL_ExplicitNone: 1264 // okay 1265 break; 1266 1267 case Qualifiers::OCL_Weak: 1268 case Qualifiers::OCL_Strong: 1269 case Qualifiers::OCL_Autoreleasing: 1270 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1271 << ValType << FirstArg->getSourceRange(); 1272 return ExprError(); 1273 } 1274 1275 // Strip any qualifiers off ValType. 1276 ValType = ValType.getUnqualifiedType(); 1277 1278 // The majority of builtins return a value, but a few have special return 1279 // types, so allow them to override appropriately below. 1280 QualType ResultType = ValType; 1281 1282 // We need to figure out which concrete builtin this maps onto. For example, 1283 // __sync_fetch_and_add with a 2 byte object turns into 1284 // __sync_fetch_and_add_2. 1285#define BUILTIN_ROW(x) \ 1286 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 1287 Builtin::BI##x##_8, Builtin::BI##x##_16 } 1288 1289 static const unsigned BuiltinIndices[][5] = { 1290 BUILTIN_ROW(__sync_fetch_and_add), 1291 BUILTIN_ROW(__sync_fetch_and_sub), 1292 BUILTIN_ROW(__sync_fetch_and_or), 1293 BUILTIN_ROW(__sync_fetch_and_and), 1294 BUILTIN_ROW(__sync_fetch_and_xor), 1295 1296 BUILTIN_ROW(__sync_add_and_fetch), 1297 BUILTIN_ROW(__sync_sub_and_fetch), 1298 BUILTIN_ROW(__sync_and_and_fetch), 1299 BUILTIN_ROW(__sync_or_and_fetch), 1300 BUILTIN_ROW(__sync_xor_and_fetch), 1301 1302 BUILTIN_ROW(__sync_val_compare_and_swap), 1303 BUILTIN_ROW(__sync_bool_compare_and_swap), 1304 BUILTIN_ROW(__sync_lock_test_and_set), 1305 BUILTIN_ROW(__sync_lock_release), 1306 BUILTIN_ROW(__sync_swap) 1307 }; 1308#undef BUILTIN_ROW 1309 1310 // Determine the index of the size. 1311 unsigned SizeIndex; 1312 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 1313 case 1: SizeIndex = 0; break; 1314 case 2: SizeIndex = 1; break; 1315 case 4: SizeIndex = 2; break; 1316 case 8: SizeIndex = 3; break; 1317 case 16: SizeIndex = 4; break; 1318 default: 1319 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 1320 << FirstArg->getType() << FirstArg->getSourceRange(); 1321 return ExprError(); 1322 } 1323 1324 // Each of these builtins has one pointer argument, followed by some number of 1325 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 1326 // that we ignore. Find out which row of BuiltinIndices to read from as well 1327 // as the number of fixed args. 1328 unsigned BuiltinID = FDecl->getBuiltinID(); 1329 unsigned BuiltinIndex, NumFixed = 1; 1330 switch (BuiltinID) { 1331 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 1332 case Builtin::BI__sync_fetch_and_add: 1333 case Builtin::BI__sync_fetch_and_add_1: 1334 case Builtin::BI__sync_fetch_and_add_2: 1335 case Builtin::BI__sync_fetch_and_add_4: 1336 case Builtin::BI__sync_fetch_and_add_8: 1337 case Builtin::BI__sync_fetch_and_add_16: 1338 BuiltinIndex = 0; 1339 break; 1340 1341 case Builtin::BI__sync_fetch_and_sub: 1342 case Builtin::BI__sync_fetch_and_sub_1: 1343 case Builtin::BI__sync_fetch_and_sub_2: 1344 case Builtin::BI__sync_fetch_and_sub_4: 1345 case Builtin::BI__sync_fetch_and_sub_8: 1346 case Builtin::BI__sync_fetch_and_sub_16: 1347 BuiltinIndex = 1; 1348 break; 1349 1350 case Builtin::BI__sync_fetch_and_or: 1351 case Builtin::BI__sync_fetch_and_or_1: 1352 case Builtin::BI__sync_fetch_and_or_2: 1353 case Builtin::BI__sync_fetch_and_or_4: 1354 case Builtin::BI__sync_fetch_and_or_8: 1355 case Builtin::BI__sync_fetch_and_or_16: 1356 BuiltinIndex = 2; 1357 break; 1358 1359 case Builtin::BI__sync_fetch_and_and: 1360 case Builtin::BI__sync_fetch_and_and_1: 1361 case Builtin::BI__sync_fetch_and_and_2: 1362 case Builtin::BI__sync_fetch_and_and_4: 1363 case Builtin::BI__sync_fetch_and_and_8: 1364 case Builtin::BI__sync_fetch_and_and_16: 1365 BuiltinIndex = 3; 1366 break; 1367 1368 case Builtin::BI__sync_fetch_and_xor: 1369 case Builtin::BI__sync_fetch_and_xor_1: 1370 case Builtin::BI__sync_fetch_and_xor_2: 1371 case Builtin::BI__sync_fetch_and_xor_4: 1372 case Builtin::BI__sync_fetch_and_xor_8: 1373 case Builtin::BI__sync_fetch_and_xor_16: 1374 BuiltinIndex = 4; 1375 break; 1376 1377 case Builtin::BI__sync_add_and_fetch: 1378 case Builtin::BI__sync_add_and_fetch_1: 1379 case Builtin::BI__sync_add_and_fetch_2: 1380 case Builtin::BI__sync_add_and_fetch_4: 1381 case Builtin::BI__sync_add_and_fetch_8: 1382 case Builtin::BI__sync_add_and_fetch_16: 1383 BuiltinIndex = 5; 1384 break; 1385 1386 case Builtin::BI__sync_sub_and_fetch: 1387 case Builtin::BI__sync_sub_and_fetch_1: 1388 case Builtin::BI__sync_sub_and_fetch_2: 1389 case Builtin::BI__sync_sub_and_fetch_4: 1390 case Builtin::BI__sync_sub_and_fetch_8: 1391 case Builtin::BI__sync_sub_and_fetch_16: 1392 BuiltinIndex = 6; 1393 break; 1394 1395 case Builtin::BI__sync_and_and_fetch: 1396 case Builtin::BI__sync_and_and_fetch_1: 1397 case Builtin::BI__sync_and_and_fetch_2: 1398 case Builtin::BI__sync_and_and_fetch_4: 1399 case Builtin::BI__sync_and_and_fetch_8: 1400 case Builtin::BI__sync_and_and_fetch_16: 1401 BuiltinIndex = 7; 1402 break; 1403 1404 case Builtin::BI__sync_or_and_fetch: 1405 case Builtin::BI__sync_or_and_fetch_1: 1406 case Builtin::BI__sync_or_and_fetch_2: 1407 case Builtin::BI__sync_or_and_fetch_4: 1408 case Builtin::BI__sync_or_and_fetch_8: 1409 case Builtin::BI__sync_or_and_fetch_16: 1410 BuiltinIndex = 8; 1411 break; 1412 1413 case Builtin::BI__sync_xor_and_fetch: 1414 case Builtin::BI__sync_xor_and_fetch_1: 1415 case Builtin::BI__sync_xor_and_fetch_2: 1416 case Builtin::BI__sync_xor_and_fetch_4: 1417 case Builtin::BI__sync_xor_and_fetch_8: 1418 case Builtin::BI__sync_xor_and_fetch_16: 1419 BuiltinIndex = 9; 1420 break; 1421 1422 case Builtin::BI__sync_val_compare_and_swap: 1423 case Builtin::BI__sync_val_compare_and_swap_1: 1424 case Builtin::BI__sync_val_compare_and_swap_2: 1425 case Builtin::BI__sync_val_compare_and_swap_4: 1426 case Builtin::BI__sync_val_compare_and_swap_8: 1427 case Builtin::BI__sync_val_compare_and_swap_16: 1428 BuiltinIndex = 10; 1429 NumFixed = 2; 1430 break; 1431 1432 case Builtin::BI__sync_bool_compare_and_swap: 1433 case Builtin::BI__sync_bool_compare_and_swap_1: 1434 case Builtin::BI__sync_bool_compare_and_swap_2: 1435 case Builtin::BI__sync_bool_compare_and_swap_4: 1436 case Builtin::BI__sync_bool_compare_and_swap_8: 1437 case Builtin::BI__sync_bool_compare_and_swap_16: 1438 BuiltinIndex = 11; 1439 NumFixed = 2; 1440 ResultType = Context.BoolTy; 1441 break; 1442 1443 case Builtin::BI__sync_lock_test_and_set: 1444 case Builtin::BI__sync_lock_test_and_set_1: 1445 case Builtin::BI__sync_lock_test_and_set_2: 1446 case Builtin::BI__sync_lock_test_and_set_4: 1447 case Builtin::BI__sync_lock_test_and_set_8: 1448 case Builtin::BI__sync_lock_test_and_set_16: 1449 BuiltinIndex = 12; 1450 break; 1451 1452 case Builtin::BI__sync_lock_release: 1453 case Builtin::BI__sync_lock_release_1: 1454 case Builtin::BI__sync_lock_release_2: 1455 case Builtin::BI__sync_lock_release_4: 1456 case Builtin::BI__sync_lock_release_8: 1457 case Builtin::BI__sync_lock_release_16: 1458 BuiltinIndex = 13; 1459 NumFixed = 0; 1460 ResultType = Context.VoidTy; 1461 break; 1462 1463 case Builtin::BI__sync_swap: 1464 case Builtin::BI__sync_swap_1: 1465 case Builtin::BI__sync_swap_2: 1466 case Builtin::BI__sync_swap_4: 1467 case Builtin::BI__sync_swap_8: 1468 case Builtin::BI__sync_swap_16: 1469 BuiltinIndex = 14; 1470 break; 1471 } 1472 1473 // Now that we know how many fixed arguments we expect, first check that we 1474 // have at least that many. 1475 if (TheCall->getNumArgs() < 1+NumFixed) { 1476 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1477 << 0 << 1+NumFixed << TheCall->getNumArgs() 1478 << TheCall->getCallee()->getSourceRange(); 1479 return ExprError(); 1480 } 1481 1482 // Get the decl for the concrete builtin from this, we can tell what the 1483 // concrete integer type we should convert to is. 1484 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1485 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1486 FunctionDecl *NewBuiltinDecl; 1487 if (NewBuiltinID == BuiltinID) 1488 NewBuiltinDecl = FDecl; 1489 else { 1490 // Perform builtin lookup to avoid redeclaring it. 1491 DeclarationName DN(&Context.Idents.get(NewBuiltinName)); 1492 LookupResult Res(*this, DN, DRE->getLocStart(), LookupOrdinaryName); 1493 LookupName(Res, TUScope, /*AllowBuiltinCreation=*/true); 1494 assert(Res.getFoundDecl()); 1495 NewBuiltinDecl = dyn_cast<FunctionDecl>(Res.getFoundDecl()); 1496 if (NewBuiltinDecl == 0) 1497 return ExprError(); 1498 } 1499 1500 // The first argument --- the pointer --- has a fixed type; we 1501 // deduce the types of the rest of the arguments accordingly. Walk 1502 // the remaining arguments, converting them to the deduced value type. 1503 for (unsigned i = 0; i != NumFixed; ++i) { 1504 ExprResult Arg = TheCall->getArg(i+1); 1505 1506 // GCC does an implicit conversion to the pointer or integer ValType. This 1507 // can fail in some cases (1i -> int**), check for this error case now. 1508 // Initialize the argument. 1509 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1510 ValType, /*consume*/ false); 1511 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1512 if (Arg.isInvalid()) 1513 return ExprError(); 1514 1515 // Okay, we have something that *can* be converted to the right type. Check 1516 // to see if there is a potentially weird extension going on here. This can 1517 // happen when you do an atomic operation on something like an char* and 1518 // pass in 42. The 42 gets converted to char. This is even more strange 1519 // for things like 45.123 -> char, etc. 1520 // FIXME: Do this check. 1521 TheCall->setArg(i+1, Arg.take()); 1522 } 1523 1524 ASTContext& Context = this->getASTContext(); 1525 1526 // Create a new DeclRefExpr to refer to the new decl. 1527 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1528 Context, 1529 DRE->getQualifierLoc(), 1530 SourceLocation(), 1531 NewBuiltinDecl, 1532 /*enclosing*/ false, 1533 DRE->getLocation(), 1534 Context.BuiltinFnTy, 1535 DRE->getValueKind()); 1536 1537 // Set the callee in the CallExpr. 1538 // FIXME: This loses syntactic information. 1539 QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType()); 1540 ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy, 1541 CK_BuiltinFnToFnPtr); 1542 TheCall->setCallee(PromotedCall.take()); 1543 1544 // Change the result type of the call to match the original value type. This 1545 // is arbitrary, but the codegen for these builtins ins design to handle it 1546 // gracefully. 1547 TheCall->setType(ResultType); 1548 1549 return TheCallResult; 1550} 1551 1552/// CheckObjCString - Checks that the argument to the builtin 1553/// CFString constructor is correct 1554/// Note: It might also make sense to do the UTF-16 conversion here (would 1555/// simplify the backend). 1556bool Sema::CheckObjCString(Expr *Arg) { 1557 Arg = Arg->IgnoreParenCasts(); 1558 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1559 1560 if (!Literal || !Literal->isAscii()) { 1561 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1562 << Arg->getSourceRange(); 1563 return true; 1564 } 1565 1566 if (Literal->containsNonAsciiOrNull()) { 1567 StringRef String = Literal->getString(); 1568 unsigned NumBytes = String.size(); 1569 SmallVector<UTF16, 128> ToBuf(NumBytes); 1570 const UTF8 *FromPtr = (const UTF8 *)String.data(); 1571 UTF16 *ToPtr = &ToBuf[0]; 1572 1573 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1574 &ToPtr, ToPtr + NumBytes, 1575 strictConversion); 1576 // Check for conversion failure. 1577 if (Result != conversionOK) 1578 Diag(Arg->getLocStart(), 1579 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1580 } 1581 return false; 1582} 1583 1584/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1585/// Emit an error and return true on failure, return false on success. 1586bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1587 Expr *Fn = TheCall->getCallee(); 1588 if (TheCall->getNumArgs() > 2) { 1589 Diag(TheCall->getArg(2)->getLocStart(), 1590 diag::err_typecheck_call_too_many_args) 1591 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1592 << Fn->getSourceRange() 1593 << SourceRange(TheCall->getArg(2)->getLocStart(), 1594 (*(TheCall->arg_end()-1))->getLocEnd()); 1595 return true; 1596 } 1597 1598 if (TheCall->getNumArgs() < 2) { 1599 return Diag(TheCall->getLocEnd(), 1600 diag::err_typecheck_call_too_few_args_at_least) 1601 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1602 } 1603 1604 // Type-check the first argument normally. 1605 if (checkBuiltinArgument(*this, TheCall, 0)) 1606 return true; 1607 1608 // Determine whether the current function is variadic or not. 1609 BlockScopeInfo *CurBlock = getCurBlock(); 1610 bool isVariadic; 1611 if (CurBlock) 1612 isVariadic = CurBlock->TheDecl->isVariadic(); 1613 else if (FunctionDecl *FD = getCurFunctionDecl()) 1614 isVariadic = FD->isVariadic(); 1615 else 1616 isVariadic = getCurMethodDecl()->isVariadic(); 1617 1618 if (!isVariadic) { 1619 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1620 return true; 1621 } 1622 1623 // Verify that the second argument to the builtin is the last argument of the 1624 // current function or method. 1625 bool SecondArgIsLastNamedArgument = false; 1626 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1627 1628 // These are valid if SecondArgIsLastNamedArgument is false after the next 1629 // block. 1630 QualType Type; 1631 SourceLocation ParamLoc; 1632 1633 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1634 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1635 // FIXME: This isn't correct for methods (results in bogus warning). 1636 // Get the last formal in the current function. 1637 const ParmVarDecl *LastArg; 1638 if (CurBlock) 1639 LastArg = *(CurBlock->TheDecl->param_end()-1); 1640 else if (FunctionDecl *FD = getCurFunctionDecl()) 1641 LastArg = *(FD->param_end()-1); 1642 else 1643 LastArg = *(getCurMethodDecl()->param_end()-1); 1644 SecondArgIsLastNamedArgument = PV == LastArg; 1645 1646 Type = PV->getType(); 1647 ParamLoc = PV->getLocation(); 1648 } 1649 } 1650 1651 if (!SecondArgIsLastNamedArgument) 1652 Diag(TheCall->getArg(1)->getLocStart(), 1653 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1654 else if (Type->isReferenceType()) { 1655 Diag(Arg->getLocStart(), 1656 diag::warn_va_start_of_reference_type_is_undefined); 1657 Diag(ParamLoc, diag::note_parameter_type) << Type; 1658 } 1659 1660 TheCall->setType(Context.VoidTy); 1661 return false; 1662} 1663 1664/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1665/// friends. This is declared to take (...), so we have to check everything. 1666bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1667 if (TheCall->getNumArgs() < 2) 1668 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1669 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1670 if (TheCall->getNumArgs() > 2) 1671 return Diag(TheCall->getArg(2)->getLocStart(), 1672 diag::err_typecheck_call_too_many_args) 1673 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1674 << SourceRange(TheCall->getArg(2)->getLocStart(), 1675 (*(TheCall->arg_end()-1))->getLocEnd()); 1676 1677 ExprResult OrigArg0 = TheCall->getArg(0); 1678 ExprResult OrigArg1 = TheCall->getArg(1); 1679 1680 // Do standard promotions between the two arguments, returning their common 1681 // type. 1682 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1683 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1684 return true; 1685 1686 // Make sure any conversions are pushed back into the call; this is 1687 // type safe since unordered compare builtins are declared as "_Bool 1688 // foo(...)". 1689 TheCall->setArg(0, OrigArg0.get()); 1690 TheCall->setArg(1, OrigArg1.get()); 1691 1692 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1693 return false; 1694 1695 // If the common type isn't a real floating type, then the arguments were 1696 // invalid for this operation. 1697 if (Res.isNull() || !Res->isRealFloatingType()) 1698 return Diag(OrigArg0.get()->getLocStart(), 1699 diag::err_typecheck_call_invalid_ordered_compare) 1700 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1701 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1702 1703 return false; 1704} 1705 1706/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1707/// __builtin_isnan and friends. This is declared to take (...), so we have 1708/// to check everything. We expect the last argument to be a floating point 1709/// value. 1710bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1711 if (TheCall->getNumArgs() < NumArgs) 1712 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1713 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1714 if (TheCall->getNumArgs() > NumArgs) 1715 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1716 diag::err_typecheck_call_too_many_args) 1717 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1718 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1719 (*(TheCall->arg_end()-1))->getLocEnd()); 1720 1721 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1722 1723 if (OrigArg->isTypeDependent()) 1724 return false; 1725 1726 // This operation requires a non-_Complex floating-point number. 1727 if (!OrigArg->getType()->isRealFloatingType()) 1728 return Diag(OrigArg->getLocStart(), 1729 diag::err_typecheck_call_invalid_unary_fp) 1730 << OrigArg->getType() << OrigArg->getSourceRange(); 1731 1732 // If this is an implicit conversion from float -> double, remove it. 1733 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1734 Expr *CastArg = Cast->getSubExpr(); 1735 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1736 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1737 "promotion from float to double is the only expected cast here"); 1738 Cast->setSubExpr(0); 1739 TheCall->setArg(NumArgs-1, CastArg); 1740 } 1741 } 1742 1743 return false; 1744} 1745 1746/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1747// This is declared to take (...), so we have to check everything. 1748ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1749 if (TheCall->getNumArgs() < 2) 1750 return ExprError(Diag(TheCall->getLocEnd(), 1751 diag::err_typecheck_call_too_few_args_at_least) 1752 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1753 << TheCall->getSourceRange()); 1754 1755 // Determine which of the following types of shufflevector we're checking: 1756 // 1) unary, vector mask: (lhs, mask) 1757 // 2) binary, vector mask: (lhs, rhs, mask) 1758 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1759 QualType resType = TheCall->getArg(0)->getType(); 1760 unsigned numElements = 0; 1761 1762 if (!TheCall->getArg(0)->isTypeDependent() && 1763 !TheCall->getArg(1)->isTypeDependent()) { 1764 QualType LHSType = TheCall->getArg(0)->getType(); 1765 QualType RHSType = TheCall->getArg(1)->getType(); 1766 1767 if (!LHSType->isVectorType() || !RHSType->isVectorType()) 1768 return ExprError(Diag(TheCall->getLocStart(), 1769 diag::err_shufflevector_non_vector) 1770 << SourceRange(TheCall->getArg(0)->getLocStart(), 1771 TheCall->getArg(1)->getLocEnd())); 1772 1773 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1774 unsigned numResElements = TheCall->getNumArgs() - 2; 1775 1776 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1777 // with mask. If so, verify that RHS is an integer vector type with the 1778 // same number of elts as lhs. 1779 if (TheCall->getNumArgs() == 2) { 1780 if (!RHSType->hasIntegerRepresentation() || 1781 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1782 return ExprError(Diag(TheCall->getLocStart(), 1783 diag::err_shufflevector_incompatible_vector) 1784 << SourceRange(TheCall->getArg(1)->getLocStart(), 1785 TheCall->getArg(1)->getLocEnd())); 1786 } else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1787 return ExprError(Diag(TheCall->getLocStart(), 1788 diag::err_shufflevector_incompatible_vector) 1789 << SourceRange(TheCall->getArg(0)->getLocStart(), 1790 TheCall->getArg(1)->getLocEnd())); 1791 } else if (numElements != numResElements) { 1792 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1793 resType = Context.getVectorType(eltType, numResElements, 1794 VectorType::GenericVector); 1795 } 1796 } 1797 1798 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1799 if (TheCall->getArg(i)->isTypeDependent() || 1800 TheCall->getArg(i)->isValueDependent()) 1801 continue; 1802 1803 llvm::APSInt Result(32); 1804 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1805 return ExprError(Diag(TheCall->getLocStart(), 1806 diag::err_shufflevector_nonconstant_argument) 1807 << TheCall->getArg(i)->getSourceRange()); 1808 1809 // Allow -1 which will be translated to undef in the IR. 1810 if (Result.isSigned() && Result.isAllOnesValue()) 1811 continue; 1812 1813 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1814 return ExprError(Diag(TheCall->getLocStart(), 1815 diag::err_shufflevector_argument_too_large) 1816 << TheCall->getArg(i)->getSourceRange()); 1817 } 1818 1819 SmallVector<Expr*, 32> exprs; 1820 1821 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1822 exprs.push_back(TheCall->getArg(i)); 1823 TheCall->setArg(i, 0); 1824 } 1825 1826 return Owned(new (Context) ShuffleVectorExpr(Context, exprs, resType, 1827 TheCall->getCallee()->getLocStart(), 1828 TheCall->getRParenLoc())); 1829} 1830 1831/// SemaConvertVectorExpr - Handle __builtin_convertvector 1832ExprResult Sema::SemaConvertVectorExpr(Expr *E, TypeSourceInfo *TInfo, 1833 SourceLocation BuiltinLoc, 1834 SourceLocation RParenLoc) { 1835 ExprValueKind VK = VK_RValue; 1836 ExprObjectKind OK = OK_Ordinary; 1837 QualType DstTy = TInfo->getType(); 1838 QualType SrcTy = E->getType(); 1839 1840 if (!SrcTy->isVectorType() && !SrcTy->isDependentType()) 1841 return ExprError(Diag(BuiltinLoc, 1842 diag::err_convertvector_non_vector) 1843 << E->getSourceRange()); 1844 if (!DstTy->isVectorType() && !DstTy->isDependentType()) 1845 return ExprError(Diag(BuiltinLoc, 1846 diag::err_convertvector_non_vector_type)); 1847 1848 if (!SrcTy->isDependentType() && !DstTy->isDependentType()) { 1849 unsigned SrcElts = SrcTy->getAs<VectorType>()->getNumElements(); 1850 unsigned DstElts = DstTy->getAs<VectorType>()->getNumElements(); 1851 if (SrcElts != DstElts) 1852 return ExprError(Diag(BuiltinLoc, 1853 diag::err_convertvector_incompatible_vector) 1854 << E->getSourceRange()); 1855 } 1856 1857 return Owned(new (Context) ConvertVectorExpr(E, TInfo, DstTy, VK, OK, 1858 BuiltinLoc, RParenLoc)); 1859 1860} 1861 1862/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1863// This is declared to take (const void*, ...) and can take two 1864// optional constant int args. 1865bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1866 unsigned NumArgs = TheCall->getNumArgs(); 1867 1868 if (NumArgs > 3) 1869 return Diag(TheCall->getLocEnd(), 1870 diag::err_typecheck_call_too_many_args_at_most) 1871 << 0 /*function call*/ << 3 << NumArgs 1872 << TheCall->getSourceRange(); 1873 1874 // Argument 0 is checked for us and the remaining arguments must be 1875 // constant integers. 1876 for (unsigned i = 1; i != NumArgs; ++i) { 1877 Expr *Arg = TheCall->getArg(i); 1878 1879 // We can't check the value of a dependent argument. 1880 if (Arg->isTypeDependent() || Arg->isValueDependent()) 1881 continue; 1882 1883 llvm::APSInt Result; 1884 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1885 return true; 1886 1887 // FIXME: gcc issues a warning and rewrites these to 0. These 1888 // seems especially odd for the third argument since the default 1889 // is 3. 1890 if (i == 1) { 1891 if (Result.getLimitedValue() > 1) 1892 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1893 << "0" << "1" << Arg->getSourceRange(); 1894 } else { 1895 if (Result.getLimitedValue() > 3) 1896 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1897 << "0" << "3" << Arg->getSourceRange(); 1898 } 1899 } 1900 1901 return false; 1902} 1903 1904/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1905/// TheCall is a constant expression. 1906bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1907 llvm::APSInt &Result) { 1908 Expr *Arg = TheCall->getArg(ArgNum); 1909 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1910 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1911 1912 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1913 1914 if (!Arg->isIntegerConstantExpr(Result, Context)) 1915 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1916 << FDecl->getDeclName() << Arg->getSourceRange(); 1917 1918 return false; 1919} 1920 1921/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1922/// int type). This simply type checks that type is one of the defined 1923/// constants (0-3). 1924// For compatibility check 0-3, llvm only handles 0 and 2. 1925bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1926 llvm::APSInt Result; 1927 1928 // We can't check the value of a dependent argument. 1929 if (TheCall->getArg(1)->isTypeDependent() || 1930 TheCall->getArg(1)->isValueDependent()) 1931 return false; 1932 1933 // Check constant-ness first. 1934 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1935 return true; 1936 1937 Expr *Arg = TheCall->getArg(1); 1938 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1939 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1940 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1941 } 1942 1943 return false; 1944} 1945 1946/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1947/// This checks that val is a constant 1. 1948bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1949 Expr *Arg = TheCall->getArg(1); 1950 llvm::APSInt Result; 1951 1952 // TODO: This is less than ideal. Overload this to take a value. 1953 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1954 return true; 1955 1956 if (Result != 1) 1957 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1958 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1959 1960 return false; 1961} 1962 1963namespace { 1964enum StringLiteralCheckType { 1965 SLCT_NotALiteral, 1966 SLCT_UncheckedLiteral, 1967 SLCT_CheckedLiteral 1968}; 1969} 1970 1971// Determine if an expression is a string literal or constant string. 1972// If this function returns false on the arguments to a function expecting a 1973// format string, we will usually need to emit a warning. 1974// True string literals are then checked by CheckFormatString. 1975static StringLiteralCheckType 1976checkFormatStringExpr(Sema &S, const Expr *E, ArrayRef<const Expr *> Args, 1977 bool HasVAListArg, unsigned format_idx, 1978 unsigned firstDataArg, Sema::FormatStringType Type, 1979 Sema::VariadicCallType CallType, bool InFunctionCall, 1980 llvm::SmallBitVector &CheckedVarArgs) { 1981 tryAgain: 1982 if (E->isTypeDependent() || E->isValueDependent()) 1983 return SLCT_NotALiteral; 1984 1985 E = E->IgnoreParenCasts(); 1986 1987 if (E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull)) 1988 // Technically -Wformat-nonliteral does not warn about this case. 1989 // The behavior of printf and friends in this case is implementation 1990 // dependent. Ideally if the format string cannot be null then 1991 // it should have a 'nonnull' attribute in the function prototype. 1992 return SLCT_UncheckedLiteral; 1993 1994 switch (E->getStmtClass()) { 1995 case Stmt::BinaryConditionalOperatorClass: 1996 case Stmt::ConditionalOperatorClass: { 1997 // The expression is a literal if both sub-expressions were, and it was 1998 // completely checked only if both sub-expressions were checked. 1999 const AbstractConditionalOperator *C = 2000 cast<AbstractConditionalOperator>(E); 2001 StringLiteralCheckType Left = 2002 checkFormatStringExpr(S, C->getTrueExpr(), Args, 2003 HasVAListArg, format_idx, firstDataArg, 2004 Type, CallType, InFunctionCall, CheckedVarArgs); 2005 if (Left == SLCT_NotALiteral) 2006 return SLCT_NotALiteral; 2007 StringLiteralCheckType Right = 2008 checkFormatStringExpr(S, C->getFalseExpr(), Args, 2009 HasVAListArg, format_idx, firstDataArg, 2010 Type, CallType, InFunctionCall, CheckedVarArgs); 2011 return Left < Right ? Left : Right; 2012 } 2013 2014 case Stmt::ImplicitCastExprClass: { 2015 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 2016 goto tryAgain; 2017 } 2018 2019 case Stmt::OpaqueValueExprClass: 2020 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 2021 E = src; 2022 goto tryAgain; 2023 } 2024 return SLCT_NotALiteral; 2025 2026 case Stmt::PredefinedExprClass: 2027 // While __func__, etc., are technically not string literals, they 2028 // cannot contain format specifiers and thus are not a security 2029 // liability. 2030 return SLCT_UncheckedLiteral; 2031 2032 case Stmt::DeclRefExprClass: { 2033 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 2034 2035 // As an exception, do not flag errors for variables binding to 2036 // const string literals. 2037 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 2038 bool isConstant = false; 2039 QualType T = DR->getType(); 2040 2041 if (const ArrayType *AT = S.Context.getAsArrayType(T)) { 2042 isConstant = AT->getElementType().isConstant(S.Context); 2043 } else if (const PointerType *PT = T->getAs<PointerType>()) { 2044 isConstant = T.isConstant(S.Context) && 2045 PT->getPointeeType().isConstant(S.Context); 2046 } else if (T->isObjCObjectPointerType()) { 2047 // In ObjC, there is usually no "const ObjectPointer" type, 2048 // so don't check if the pointee type is constant. 2049 isConstant = T.isConstant(S.Context); 2050 } 2051 2052 if (isConstant) { 2053 if (const Expr *Init = VD->getAnyInitializer()) { 2054 // Look through initializers like const char c[] = { "foo" } 2055 if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) { 2056 if (InitList->isStringLiteralInit()) 2057 Init = InitList->getInit(0)->IgnoreParenImpCasts(); 2058 } 2059 return checkFormatStringExpr(S, Init, Args, 2060 HasVAListArg, format_idx, 2061 firstDataArg, Type, CallType, 2062 /*InFunctionCall*/false, CheckedVarArgs); 2063 } 2064 } 2065 2066 // For vprintf* functions (i.e., HasVAListArg==true), we add a 2067 // special check to see if the format string is a function parameter 2068 // of the function calling the printf function. If the function 2069 // has an attribute indicating it is a printf-like function, then we 2070 // should suppress warnings concerning non-literals being used in a call 2071 // to a vprintf function. For example: 2072 // 2073 // void 2074 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 2075 // va_list ap; 2076 // va_start(ap, fmt); 2077 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 2078 // ... 2079 // } 2080 if (HasVAListArg) { 2081 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 2082 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 2083 int PVIndex = PV->getFunctionScopeIndex() + 1; 2084 for (specific_attr_iterator<FormatAttr> 2085 i = ND->specific_attr_begin<FormatAttr>(), 2086 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 2087 FormatAttr *PVFormat = *i; 2088 // adjust for implicit parameter 2089 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 2090 if (MD->isInstance()) 2091 ++PVIndex; 2092 // We also check if the formats are compatible. 2093 // We can't pass a 'scanf' string to a 'printf' function. 2094 if (PVIndex == PVFormat->getFormatIdx() && 2095 Type == S.GetFormatStringType(PVFormat)) 2096 return SLCT_UncheckedLiteral; 2097 } 2098 } 2099 } 2100 } 2101 } 2102 2103 return SLCT_NotALiteral; 2104 } 2105 2106 case Stmt::CallExprClass: 2107 case Stmt::CXXMemberCallExprClass: { 2108 const CallExpr *CE = cast<CallExpr>(E); 2109 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 2110 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 2111 unsigned ArgIndex = FA->getFormatIdx(); 2112 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 2113 if (MD->isInstance()) 2114 --ArgIndex; 2115 const Expr *Arg = CE->getArg(ArgIndex - 1); 2116 2117 return checkFormatStringExpr(S, Arg, Args, 2118 HasVAListArg, format_idx, firstDataArg, 2119 Type, CallType, InFunctionCall, 2120 CheckedVarArgs); 2121 } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) { 2122 unsigned BuiltinID = FD->getBuiltinID(); 2123 if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString || 2124 BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) { 2125 const Expr *Arg = CE->getArg(0); 2126 return checkFormatStringExpr(S, Arg, Args, 2127 HasVAListArg, format_idx, 2128 firstDataArg, Type, CallType, 2129 InFunctionCall, CheckedVarArgs); 2130 } 2131 } 2132 } 2133 2134 return SLCT_NotALiteral; 2135 } 2136 2137 case Stmt::ObjCMessageExprClass: { 2138 const ObjCMessageExpr *ME = cast<ObjCMessageExpr>(E); 2139 if (const ObjCMethodDecl *MDecl = ME->getMethodDecl()) { 2140 if (const NamedDecl *ND = dyn_cast<NamedDecl>(MDecl)) { 2141 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 2142 unsigned ArgIndex = FA->getFormatIdx(); 2143 if (ArgIndex <= ME->getNumArgs()) { 2144 const Expr *Arg = ME->getArg(ArgIndex-1); 2145 return checkFormatStringExpr(S, Arg, Args, 2146 HasVAListArg, format_idx, 2147 firstDataArg, Type, CallType, 2148 InFunctionCall, CheckedVarArgs); 2149 } 2150 } 2151 } 2152 } 2153 2154 return SLCT_NotALiteral; 2155 } 2156 2157 case Stmt::ObjCStringLiteralClass: 2158 case Stmt::StringLiteralClass: { 2159 const StringLiteral *StrE = NULL; 2160 2161 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 2162 StrE = ObjCFExpr->getString(); 2163 else 2164 StrE = cast<StringLiteral>(E); 2165 2166 if (StrE) { 2167 S.CheckFormatString(StrE, E, Args, HasVAListArg, format_idx, firstDataArg, 2168 Type, InFunctionCall, CallType, CheckedVarArgs); 2169 return SLCT_CheckedLiteral; 2170 } 2171 2172 return SLCT_NotALiteral; 2173 } 2174 2175 default: 2176 return SLCT_NotALiteral; 2177 } 2178} 2179 2180void 2181Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 2182 const Expr * const *ExprArgs, 2183 SourceLocation CallSiteLoc) { 2184 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 2185 e = NonNull->args_end(); 2186 i != e; ++i) { 2187 const Expr *ArgExpr = ExprArgs[*i]; 2188 2189 // As a special case, transparent unions initialized with zero are 2190 // considered null for the purposes of the nonnull attribute. 2191 if (const RecordType *UT = ArgExpr->getType()->getAsUnionType()) { 2192 if (UT->getDecl()->hasAttr<TransparentUnionAttr>()) 2193 if (const CompoundLiteralExpr *CLE = 2194 dyn_cast<CompoundLiteralExpr>(ArgExpr)) 2195 if (const InitListExpr *ILE = 2196 dyn_cast<InitListExpr>(CLE->getInitializer())) 2197 ArgExpr = ILE->getInit(0); 2198 } 2199 2200 bool Result; 2201 if (ArgExpr->EvaluateAsBooleanCondition(Result, Context) && !Result) 2202 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 2203 } 2204} 2205 2206Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 2207 return llvm::StringSwitch<FormatStringType>(Format->getType()->getName()) 2208 .Case("scanf", FST_Scanf) 2209 .Cases("printf", "printf0", FST_Printf) 2210 .Cases("NSString", "CFString", FST_NSString) 2211 .Case("strftime", FST_Strftime) 2212 .Case("strfmon", FST_Strfmon) 2213 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 2214 .Default(FST_Unknown); 2215} 2216 2217/// CheckFormatArguments - Check calls to printf and scanf (and similar 2218/// functions) for correct use of format strings. 2219/// Returns true if a format string has been fully checked. 2220bool Sema::CheckFormatArguments(const FormatAttr *Format, 2221 ArrayRef<const Expr *> Args, 2222 bool IsCXXMember, 2223 VariadicCallType CallType, 2224 SourceLocation Loc, SourceRange Range, 2225 llvm::SmallBitVector &CheckedVarArgs) { 2226 FormatStringInfo FSI; 2227 if (getFormatStringInfo(Format, IsCXXMember, &FSI)) 2228 return CheckFormatArguments(Args, FSI.HasVAListArg, FSI.FormatIdx, 2229 FSI.FirstDataArg, GetFormatStringType(Format), 2230 CallType, Loc, Range, CheckedVarArgs); 2231 return false; 2232} 2233 2234bool Sema::CheckFormatArguments(ArrayRef<const Expr *> Args, 2235 bool HasVAListArg, unsigned format_idx, 2236 unsigned firstDataArg, FormatStringType Type, 2237 VariadicCallType CallType, 2238 SourceLocation Loc, SourceRange Range, 2239 llvm::SmallBitVector &CheckedVarArgs) { 2240 // CHECK: printf/scanf-like function is called with no format string. 2241 if (format_idx >= Args.size()) { 2242 Diag(Loc, diag::warn_missing_format_string) << Range; 2243 return false; 2244 } 2245 2246 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 2247 2248 // CHECK: format string is not a string literal. 2249 // 2250 // Dynamically generated format strings are difficult to 2251 // automatically vet at compile time. Requiring that format strings 2252 // are string literals: (1) permits the checking of format strings by 2253 // the compiler and thereby (2) can practically remove the source of 2254 // many format string exploits. 2255 2256 // Format string can be either ObjC string (e.g. @"%d") or 2257 // C string (e.g. "%d") 2258 // ObjC string uses the same format specifiers as C string, so we can use 2259 // the same format string checking logic for both ObjC and C strings. 2260 StringLiteralCheckType CT = 2261 checkFormatStringExpr(*this, OrigFormatExpr, Args, HasVAListArg, 2262 format_idx, firstDataArg, Type, CallType, 2263 /*IsFunctionCall*/true, CheckedVarArgs); 2264 if (CT != SLCT_NotALiteral) 2265 // Literal format string found, check done! 2266 return CT == SLCT_CheckedLiteral; 2267 2268 // Strftime is particular as it always uses a single 'time' argument, 2269 // so it is safe to pass a non-literal string. 2270 if (Type == FST_Strftime) 2271 return false; 2272 2273 // Do not emit diag when the string param is a macro expansion and the 2274 // format is either NSString or CFString. This is a hack to prevent 2275 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 2276 // which are usually used in place of NS and CF string literals. 2277 if (Type == FST_NSString && 2278 SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart())) 2279 return false; 2280 2281 // If there are no arguments specified, warn with -Wformat-security, otherwise 2282 // warn only with -Wformat-nonliteral. 2283 if (Args.size() == firstDataArg) 2284 Diag(Args[format_idx]->getLocStart(), 2285 diag::warn_format_nonliteral_noargs) 2286 << OrigFormatExpr->getSourceRange(); 2287 else 2288 Diag(Args[format_idx]->getLocStart(), 2289 diag::warn_format_nonliteral) 2290 << OrigFormatExpr->getSourceRange(); 2291 return false; 2292} 2293 2294namespace { 2295class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 2296protected: 2297 Sema &S; 2298 const StringLiteral *FExpr; 2299 const Expr *OrigFormatExpr; 2300 const unsigned FirstDataArg; 2301 const unsigned NumDataArgs; 2302 const char *Beg; // Start of format string. 2303 const bool HasVAListArg; 2304 ArrayRef<const Expr *> Args; 2305 unsigned FormatIdx; 2306 llvm::SmallBitVector CoveredArgs; 2307 bool usesPositionalArgs; 2308 bool atFirstArg; 2309 bool inFunctionCall; 2310 Sema::VariadicCallType CallType; 2311 llvm::SmallBitVector &CheckedVarArgs; 2312public: 2313 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 2314 const Expr *origFormatExpr, unsigned firstDataArg, 2315 unsigned numDataArgs, const char *beg, bool hasVAListArg, 2316 ArrayRef<const Expr *> Args, 2317 unsigned formatIdx, bool inFunctionCall, 2318 Sema::VariadicCallType callType, 2319 llvm::SmallBitVector &CheckedVarArgs) 2320 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 2321 FirstDataArg(firstDataArg), NumDataArgs(numDataArgs), 2322 Beg(beg), HasVAListArg(hasVAListArg), 2323 Args(Args), FormatIdx(formatIdx), 2324 usesPositionalArgs(false), atFirstArg(true), 2325 inFunctionCall(inFunctionCall), CallType(callType), 2326 CheckedVarArgs(CheckedVarArgs) { 2327 CoveredArgs.resize(numDataArgs); 2328 CoveredArgs.reset(); 2329 } 2330 2331 void DoneProcessing(); 2332 2333 void HandleIncompleteSpecifier(const char *startSpecifier, 2334 unsigned specifierLen); 2335 2336 void HandleInvalidLengthModifier( 2337 const analyze_format_string::FormatSpecifier &FS, 2338 const analyze_format_string::ConversionSpecifier &CS, 2339 const char *startSpecifier, unsigned specifierLen, unsigned DiagID); 2340 2341 void HandleNonStandardLengthModifier( 2342 const analyze_format_string::FormatSpecifier &FS, 2343 const char *startSpecifier, unsigned specifierLen); 2344 2345 void HandleNonStandardConversionSpecifier( 2346 const analyze_format_string::ConversionSpecifier &CS, 2347 const char *startSpecifier, unsigned specifierLen); 2348 2349 virtual void HandlePosition(const char *startPos, unsigned posLen); 2350 2351 virtual void HandleInvalidPosition(const char *startSpecifier, 2352 unsigned specifierLen, 2353 analyze_format_string::PositionContext p); 2354 2355 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 2356 2357 void HandleNullChar(const char *nullCharacter); 2358 2359 template <typename Range> 2360 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 2361 const Expr *ArgumentExpr, 2362 PartialDiagnostic PDiag, 2363 SourceLocation StringLoc, 2364 bool IsStringLocation, Range StringRange, 2365 ArrayRef<FixItHint> Fixit = None); 2366 2367protected: 2368 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 2369 const char *startSpec, 2370 unsigned specifierLen, 2371 const char *csStart, unsigned csLen); 2372 2373 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 2374 const char *startSpec, 2375 unsigned specifierLen); 2376 2377 SourceRange getFormatStringRange(); 2378 CharSourceRange getSpecifierRange(const char *startSpecifier, 2379 unsigned specifierLen); 2380 SourceLocation getLocationOfByte(const char *x); 2381 2382 const Expr *getDataArg(unsigned i) const; 2383 2384 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 2385 const analyze_format_string::ConversionSpecifier &CS, 2386 const char *startSpecifier, unsigned specifierLen, 2387 unsigned argIndex); 2388 2389 template <typename Range> 2390 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 2391 bool IsStringLocation, Range StringRange, 2392 ArrayRef<FixItHint> Fixit = None); 2393 2394 void CheckPositionalAndNonpositionalArgs( 2395 const analyze_format_string::FormatSpecifier *FS); 2396}; 2397} 2398 2399SourceRange CheckFormatHandler::getFormatStringRange() { 2400 return OrigFormatExpr->getSourceRange(); 2401} 2402 2403CharSourceRange CheckFormatHandler:: 2404getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 2405 SourceLocation Start = getLocationOfByte(startSpecifier); 2406 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 2407 2408 // Advance the end SourceLocation by one due to half-open ranges. 2409 End = End.getLocWithOffset(1); 2410 2411 return CharSourceRange::getCharRange(Start, End); 2412} 2413 2414SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 2415 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 2416} 2417 2418void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 2419 unsigned specifierLen){ 2420 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 2421 getLocationOfByte(startSpecifier), 2422 /*IsStringLocation*/true, 2423 getSpecifierRange(startSpecifier, specifierLen)); 2424} 2425 2426void CheckFormatHandler::HandleInvalidLengthModifier( 2427 const analyze_format_string::FormatSpecifier &FS, 2428 const analyze_format_string::ConversionSpecifier &CS, 2429 const char *startSpecifier, unsigned specifierLen, unsigned DiagID) { 2430 using namespace analyze_format_string; 2431 2432 const LengthModifier &LM = FS.getLengthModifier(); 2433 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2434 2435 // See if we know how to fix this length modifier. 2436 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2437 if (FixedLM) { 2438 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2439 getLocationOfByte(LM.getStart()), 2440 /*IsStringLocation*/true, 2441 getSpecifierRange(startSpecifier, specifierLen)); 2442 2443 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2444 << FixedLM->toString() 2445 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2446 2447 } else { 2448 FixItHint Hint; 2449 if (DiagID == diag::warn_format_nonsensical_length) 2450 Hint = FixItHint::CreateRemoval(LMRange); 2451 2452 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2453 getLocationOfByte(LM.getStart()), 2454 /*IsStringLocation*/true, 2455 getSpecifierRange(startSpecifier, specifierLen), 2456 Hint); 2457 } 2458} 2459 2460void CheckFormatHandler::HandleNonStandardLengthModifier( 2461 const analyze_format_string::FormatSpecifier &FS, 2462 const char *startSpecifier, unsigned specifierLen) { 2463 using namespace analyze_format_string; 2464 2465 const LengthModifier &LM = FS.getLengthModifier(); 2466 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2467 2468 // See if we know how to fix this length modifier. 2469 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2470 if (FixedLM) { 2471 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2472 << LM.toString() << 0, 2473 getLocationOfByte(LM.getStart()), 2474 /*IsStringLocation*/true, 2475 getSpecifierRange(startSpecifier, specifierLen)); 2476 2477 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2478 << FixedLM->toString() 2479 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2480 2481 } else { 2482 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2483 << LM.toString() << 0, 2484 getLocationOfByte(LM.getStart()), 2485 /*IsStringLocation*/true, 2486 getSpecifierRange(startSpecifier, specifierLen)); 2487 } 2488} 2489 2490void CheckFormatHandler::HandleNonStandardConversionSpecifier( 2491 const analyze_format_string::ConversionSpecifier &CS, 2492 const char *startSpecifier, unsigned specifierLen) { 2493 using namespace analyze_format_string; 2494 2495 // See if we know how to fix this conversion specifier. 2496 Optional<ConversionSpecifier> FixedCS = CS.getStandardSpecifier(); 2497 if (FixedCS) { 2498 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2499 << CS.toString() << /*conversion specifier*/1, 2500 getLocationOfByte(CS.getStart()), 2501 /*IsStringLocation*/true, 2502 getSpecifierRange(startSpecifier, specifierLen)); 2503 2504 CharSourceRange CSRange = getSpecifierRange(CS.getStart(), CS.getLength()); 2505 S.Diag(getLocationOfByte(CS.getStart()), diag::note_format_fix_specifier) 2506 << FixedCS->toString() 2507 << FixItHint::CreateReplacement(CSRange, FixedCS->toString()); 2508 } else { 2509 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2510 << CS.toString() << /*conversion specifier*/1, 2511 getLocationOfByte(CS.getStart()), 2512 /*IsStringLocation*/true, 2513 getSpecifierRange(startSpecifier, specifierLen)); 2514 } 2515} 2516 2517void CheckFormatHandler::HandlePosition(const char *startPos, 2518 unsigned posLen) { 2519 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 2520 getLocationOfByte(startPos), 2521 /*IsStringLocation*/true, 2522 getSpecifierRange(startPos, posLen)); 2523} 2524 2525void 2526CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 2527 analyze_format_string::PositionContext p) { 2528 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 2529 << (unsigned) p, 2530 getLocationOfByte(startPos), /*IsStringLocation*/true, 2531 getSpecifierRange(startPos, posLen)); 2532} 2533 2534void CheckFormatHandler::HandleZeroPosition(const char *startPos, 2535 unsigned posLen) { 2536 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 2537 getLocationOfByte(startPos), 2538 /*IsStringLocation*/true, 2539 getSpecifierRange(startPos, posLen)); 2540} 2541 2542void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 2543 if (!isa<ObjCStringLiteral>(OrigFormatExpr)) { 2544 // The presence of a null character is likely an error. 2545 EmitFormatDiagnostic( 2546 S.PDiag(diag::warn_printf_format_string_contains_null_char), 2547 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 2548 getFormatStringRange()); 2549 } 2550} 2551 2552// Note that this may return NULL if there was an error parsing or building 2553// one of the argument expressions. 2554const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 2555 return Args[FirstDataArg + i]; 2556} 2557 2558void CheckFormatHandler::DoneProcessing() { 2559 // Does the number of data arguments exceed the number of 2560 // format conversions in the format string? 2561 if (!HasVAListArg) { 2562 // Find any arguments that weren't covered. 2563 CoveredArgs.flip(); 2564 signed notCoveredArg = CoveredArgs.find_first(); 2565 if (notCoveredArg >= 0) { 2566 assert((unsigned)notCoveredArg < NumDataArgs); 2567 if (const Expr *E = getDataArg((unsigned) notCoveredArg)) { 2568 SourceLocation Loc = E->getLocStart(); 2569 if (!S.getSourceManager().isInSystemMacro(Loc)) { 2570 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 2571 Loc, /*IsStringLocation*/false, 2572 getFormatStringRange()); 2573 } 2574 } 2575 } 2576 } 2577} 2578 2579bool 2580CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 2581 SourceLocation Loc, 2582 const char *startSpec, 2583 unsigned specifierLen, 2584 const char *csStart, 2585 unsigned csLen) { 2586 2587 bool keepGoing = true; 2588 if (argIndex < NumDataArgs) { 2589 // Consider the argument coverered, even though the specifier doesn't 2590 // make sense. 2591 CoveredArgs.set(argIndex); 2592 } 2593 else { 2594 // If argIndex exceeds the number of data arguments we 2595 // don't issue a warning because that is just a cascade of warnings (and 2596 // they may have intended '%%' anyway). We don't want to continue processing 2597 // the format string after this point, however, as we will like just get 2598 // gibberish when trying to match arguments. 2599 keepGoing = false; 2600 } 2601 2602 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2603 << StringRef(csStart, csLen), 2604 Loc, /*IsStringLocation*/true, 2605 getSpecifierRange(startSpec, specifierLen)); 2606 2607 return keepGoing; 2608} 2609 2610void 2611CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2612 const char *startSpec, 2613 unsigned specifierLen) { 2614 EmitFormatDiagnostic( 2615 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2616 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2617} 2618 2619bool 2620CheckFormatHandler::CheckNumArgs( 2621 const analyze_format_string::FormatSpecifier &FS, 2622 const analyze_format_string::ConversionSpecifier &CS, 2623 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2624 2625 if (argIndex >= NumDataArgs) { 2626 PartialDiagnostic PDiag = FS.usesPositionalArg() 2627 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2628 << (argIndex+1) << NumDataArgs) 2629 : S.PDiag(diag::warn_printf_insufficient_data_args); 2630 EmitFormatDiagnostic( 2631 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2632 getSpecifierRange(startSpecifier, specifierLen)); 2633 return false; 2634 } 2635 return true; 2636} 2637 2638template<typename Range> 2639void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2640 SourceLocation Loc, 2641 bool IsStringLocation, 2642 Range StringRange, 2643 ArrayRef<FixItHint> FixIt) { 2644 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2645 Loc, IsStringLocation, StringRange, FixIt); 2646} 2647 2648/// \brief If the format string is not within the funcion call, emit a note 2649/// so that the function call and string are in diagnostic messages. 2650/// 2651/// \param InFunctionCall if true, the format string is within the function 2652/// call and only one diagnostic message will be produced. Otherwise, an 2653/// extra note will be emitted pointing to location of the format string. 2654/// 2655/// \param ArgumentExpr the expression that is passed as the format string 2656/// argument in the function call. Used for getting locations when two 2657/// diagnostics are emitted. 2658/// 2659/// \param PDiag the callee should already have provided any strings for the 2660/// diagnostic message. This function only adds locations and fixits 2661/// to diagnostics. 2662/// 2663/// \param Loc primary location for diagnostic. If two diagnostics are 2664/// required, one will be at Loc and a new SourceLocation will be created for 2665/// the other one. 2666/// 2667/// \param IsStringLocation if true, Loc points to the format string should be 2668/// used for the note. Otherwise, Loc points to the argument list and will 2669/// be used with PDiag. 2670/// 2671/// \param StringRange some or all of the string to highlight. This is 2672/// templated so it can accept either a CharSourceRange or a SourceRange. 2673/// 2674/// \param FixIt optional fix it hint for the format string. 2675template<typename Range> 2676void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2677 const Expr *ArgumentExpr, 2678 PartialDiagnostic PDiag, 2679 SourceLocation Loc, 2680 bool IsStringLocation, 2681 Range StringRange, 2682 ArrayRef<FixItHint> FixIt) { 2683 if (InFunctionCall) { 2684 const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag); 2685 D << StringRange; 2686 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2687 I != E; ++I) { 2688 D << *I; 2689 } 2690 } else { 2691 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2692 << ArgumentExpr->getSourceRange(); 2693 2694 const Sema::SemaDiagnosticBuilder &Note = 2695 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2696 diag::note_format_string_defined); 2697 2698 Note << StringRange; 2699 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2700 I != E; ++I) { 2701 Note << *I; 2702 } 2703 } 2704} 2705 2706//===--- CHECK: Printf format string checking ------------------------------===// 2707 2708namespace { 2709class CheckPrintfHandler : public CheckFormatHandler { 2710 bool ObjCContext; 2711public: 2712 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2713 const Expr *origFormatExpr, unsigned firstDataArg, 2714 unsigned numDataArgs, bool isObjC, 2715 const char *beg, bool hasVAListArg, 2716 ArrayRef<const Expr *> Args, 2717 unsigned formatIdx, bool inFunctionCall, 2718 Sema::VariadicCallType CallType, 2719 llvm::SmallBitVector &CheckedVarArgs) 2720 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2721 numDataArgs, beg, hasVAListArg, Args, 2722 formatIdx, inFunctionCall, CallType, CheckedVarArgs), 2723 ObjCContext(isObjC) 2724 {} 2725 2726 2727 bool HandleInvalidPrintfConversionSpecifier( 2728 const analyze_printf::PrintfSpecifier &FS, 2729 const char *startSpecifier, 2730 unsigned specifierLen); 2731 2732 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2733 const char *startSpecifier, 2734 unsigned specifierLen); 2735 bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2736 const char *StartSpecifier, 2737 unsigned SpecifierLen, 2738 const Expr *E); 2739 2740 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2741 const char *startSpecifier, unsigned specifierLen); 2742 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2743 const analyze_printf::OptionalAmount &Amt, 2744 unsigned type, 2745 const char *startSpecifier, unsigned specifierLen); 2746 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2747 const analyze_printf::OptionalFlag &flag, 2748 const char *startSpecifier, unsigned specifierLen); 2749 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2750 const analyze_printf::OptionalFlag &ignoredFlag, 2751 const analyze_printf::OptionalFlag &flag, 2752 const char *startSpecifier, unsigned specifierLen); 2753 bool checkForCStrMembers(const analyze_printf::ArgType &AT, 2754 const Expr *E, const CharSourceRange &CSR); 2755 2756}; 2757} 2758 2759bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2760 const analyze_printf::PrintfSpecifier &FS, 2761 const char *startSpecifier, 2762 unsigned specifierLen) { 2763 const analyze_printf::PrintfConversionSpecifier &CS = 2764 FS.getConversionSpecifier(); 2765 2766 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2767 getLocationOfByte(CS.getStart()), 2768 startSpecifier, specifierLen, 2769 CS.getStart(), CS.getLength()); 2770} 2771 2772bool CheckPrintfHandler::HandleAmount( 2773 const analyze_format_string::OptionalAmount &Amt, 2774 unsigned k, const char *startSpecifier, 2775 unsigned specifierLen) { 2776 2777 if (Amt.hasDataArgument()) { 2778 if (!HasVAListArg) { 2779 unsigned argIndex = Amt.getArgIndex(); 2780 if (argIndex >= NumDataArgs) { 2781 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2782 << k, 2783 getLocationOfByte(Amt.getStart()), 2784 /*IsStringLocation*/true, 2785 getSpecifierRange(startSpecifier, specifierLen)); 2786 // Don't do any more checking. We will just emit 2787 // spurious errors. 2788 return false; 2789 } 2790 2791 // Type check the data argument. It should be an 'int'. 2792 // Although not in conformance with C99, we also allow the argument to be 2793 // an 'unsigned int' as that is a reasonably safe case. GCC also 2794 // doesn't emit a warning for that case. 2795 CoveredArgs.set(argIndex); 2796 const Expr *Arg = getDataArg(argIndex); 2797 if (!Arg) 2798 return false; 2799 2800 QualType T = Arg->getType(); 2801 2802 const analyze_printf::ArgType &AT = Amt.getArgType(S.Context); 2803 assert(AT.isValid()); 2804 2805 if (!AT.matchesType(S.Context, T)) { 2806 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2807 << k << AT.getRepresentativeTypeName(S.Context) 2808 << T << Arg->getSourceRange(), 2809 getLocationOfByte(Amt.getStart()), 2810 /*IsStringLocation*/true, 2811 getSpecifierRange(startSpecifier, specifierLen)); 2812 // Don't do any more checking. We will just emit 2813 // spurious errors. 2814 return false; 2815 } 2816 } 2817 } 2818 return true; 2819} 2820 2821void CheckPrintfHandler::HandleInvalidAmount( 2822 const analyze_printf::PrintfSpecifier &FS, 2823 const analyze_printf::OptionalAmount &Amt, 2824 unsigned type, 2825 const char *startSpecifier, 2826 unsigned specifierLen) { 2827 const analyze_printf::PrintfConversionSpecifier &CS = 2828 FS.getConversionSpecifier(); 2829 2830 FixItHint fixit = 2831 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2832 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2833 Amt.getConstantLength())) 2834 : FixItHint(); 2835 2836 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2837 << type << CS.toString(), 2838 getLocationOfByte(Amt.getStart()), 2839 /*IsStringLocation*/true, 2840 getSpecifierRange(startSpecifier, specifierLen), 2841 fixit); 2842} 2843 2844void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2845 const analyze_printf::OptionalFlag &flag, 2846 const char *startSpecifier, 2847 unsigned specifierLen) { 2848 // Warn about pointless flag with a fixit removal. 2849 const analyze_printf::PrintfConversionSpecifier &CS = 2850 FS.getConversionSpecifier(); 2851 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2852 << flag.toString() << CS.toString(), 2853 getLocationOfByte(flag.getPosition()), 2854 /*IsStringLocation*/true, 2855 getSpecifierRange(startSpecifier, specifierLen), 2856 FixItHint::CreateRemoval( 2857 getSpecifierRange(flag.getPosition(), 1))); 2858} 2859 2860void CheckPrintfHandler::HandleIgnoredFlag( 2861 const analyze_printf::PrintfSpecifier &FS, 2862 const analyze_printf::OptionalFlag &ignoredFlag, 2863 const analyze_printf::OptionalFlag &flag, 2864 const char *startSpecifier, 2865 unsigned specifierLen) { 2866 // Warn about ignored flag with a fixit removal. 2867 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2868 << ignoredFlag.toString() << flag.toString(), 2869 getLocationOfByte(ignoredFlag.getPosition()), 2870 /*IsStringLocation*/true, 2871 getSpecifierRange(startSpecifier, specifierLen), 2872 FixItHint::CreateRemoval( 2873 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2874} 2875 2876// Determines if the specified is a C++ class or struct containing 2877// a member with the specified name and kind (e.g. a CXXMethodDecl named 2878// "c_str()"). 2879template<typename MemberKind> 2880static llvm::SmallPtrSet<MemberKind*, 1> 2881CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) { 2882 const RecordType *RT = Ty->getAs<RecordType>(); 2883 llvm::SmallPtrSet<MemberKind*, 1> Results; 2884 2885 if (!RT) 2886 return Results; 2887 const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl()); 2888 if (!RD) 2889 return Results; 2890 2891 LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(), 2892 Sema::LookupMemberName); 2893 2894 // We just need to include all members of the right kind turned up by the 2895 // filter, at this point. 2896 if (S.LookupQualifiedName(R, RT->getDecl())) 2897 for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) { 2898 NamedDecl *decl = (*I)->getUnderlyingDecl(); 2899 if (MemberKind *FK = dyn_cast<MemberKind>(decl)) 2900 Results.insert(FK); 2901 } 2902 return Results; 2903} 2904 2905// Check if a (w)string was passed when a (w)char* was needed, and offer a 2906// better diagnostic if so. AT is assumed to be valid. 2907// Returns true when a c_str() conversion method is found. 2908bool CheckPrintfHandler::checkForCStrMembers( 2909 const analyze_printf::ArgType &AT, const Expr *E, 2910 const CharSourceRange &CSR) { 2911 typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet; 2912 2913 MethodSet Results = 2914 CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType()); 2915 2916 for (MethodSet::iterator MI = Results.begin(), ME = Results.end(); 2917 MI != ME; ++MI) { 2918 const CXXMethodDecl *Method = *MI; 2919 if (Method->getNumParams() == 0 && 2920 AT.matchesType(S.Context, Method->getResultType())) { 2921 // FIXME: Suggest parens if the expression needs them. 2922 SourceLocation EndLoc = 2923 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()); 2924 S.Diag(E->getLocStart(), diag::note_printf_c_str) 2925 << "c_str()" 2926 << FixItHint::CreateInsertion(EndLoc, ".c_str()"); 2927 return true; 2928 } 2929 } 2930 2931 return false; 2932} 2933 2934bool 2935CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2936 &FS, 2937 const char *startSpecifier, 2938 unsigned specifierLen) { 2939 2940 using namespace analyze_format_string; 2941 using namespace analyze_printf; 2942 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2943 2944 if (FS.consumesDataArgument()) { 2945 if (atFirstArg) { 2946 atFirstArg = false; 2947 usesPositionalArgs = FS.usesPositionalArg(); 2948 } 2949 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2950 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2951 startSpecifier, specifierLen); 2952 return false; 2953 } 2954 } 2955 2956 // First check if the field width, precision, and conversion specifier 2957 // have matching data arguments. 2958 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2959 startSpecifier, specifierLen)) { 2960 return false; 2961 } 2962 2963 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2964 startSpecifier, specifierLen)) { 2965 return false; 2966 } 2967 2968 if (!CS.consumesDataArgument()) { 2969 // FIXME: Technically specifying a precision or field width here 2970 // makes no sense. Worth issuing a warning at some point. 2971 return true; 2972 } 2973 2974 // Consume the argument. 2975 unsigned argIndex = FS.getArgIndex(); 2976 if (argIndex < NumDataArgs) { 2977 // The check to see if the argIndex is valid will come later. 2978 // We set the bit here because we may exit early from this 2979 // function if we encounter some other error. 2980 CoveredArgs.set(argIndex); 2981 } 2982 2983 // FreeBSD extensions 2984 if (CS.getKind() == ConversionSpecifier::FreeBSDbArg || 2985 CS.getKind() == ConversionSpecifier::FreeBSDDArg) { 2986 // claim the second argument 2987 CoveredArgs.set(argIndex + 1); 2988 2989 // Now type check the data expression that matches the 2990 // format specifier. 2991 const Expr *Ex = getDataArg(argIndex); 2992 const analyze_printf::ArgType &AT = 2993 (CS.getKind() == ConversionSpecifier::FreeBSDbArg) ? 2994 ArgType(S.Context.IntTy) : ArgType::CStrTy; 2995 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) 2996 S.Diag(getLocationOfByte(CS.getStart()), 2997 diag::warn_printf_conversion_argument_type_mismatch) 2998 << AT.getRepresentativeType(S.Context) << Ex->getType() 2999 << getSpecifierRange(startSpecifier, specifierLen) 3000 << Ex->getSourceRange(); 3001 3002 // Now type check the data expression that matches the 3003 // format specifier. 3004 Ex = getDataArg(argIndex + 1); 3005 const analyze_printf::ArgType &AT2 = ArgType::CStrTy; 3006 if (AT2.isValid() && !AT2.matchesType(S.Context, Ex->getType())) 3007 S.Diag(getLocationOfByte(CS.getStart()), 3008 diag::warn_printf_conversion_argument_type_mismatch) 3009 << AT2.getRepresentativeType(S.Context) << Ex->getType() 3010 << getSpecifierRange(startSpecifier, specifierLen) 3011 << Ex->getSourceRange(); 3012 3013 return true; 3014 } 3015 // END OF FREEBSD EXTENSIONS 3016 3017 // Check for using an Objective-C specific conversion specifier 3018 // in a non-ObjC literal. 3019 if (!ObjCContext && CS.isObjCArg()) { 3020 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 3021 specifierLen); 3022 } 3023 3024 // Check for invalid use of field width 3025 if (!FS.hasValidFieldWidth()) { 3026 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 3027 startSpecifier, specifierLen); 3028 } 3029 3030 // Check for invalid use of precision 3031 if (!FS.hasValidPrecision()) { 3032 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 3033 startSpecifier, specifierLen); 3034 } 3035 3036 // Check each flag does not conflict with any other component. 3037 if (!FS.hasValidThousandsGroupingPrefix()) 3038 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 3039 if (!FS.hasValidLeadingZeros()) 3040 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 3041 if (!FS.hasValidPlusPrefix()) 3042 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 3043 if (!FS.hasValidSpacePrefix()) 3044 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 3045 if (!FS.hasValidAlternativeForm()) 3046 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 3047 if (!FS.hasValidLeftJustified()) 3048 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 3049 3050 // Check that flags are not ignored by another flag 3051 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 3052 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 3053 startSpecifier, specifierLen); 3054 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 3055 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 3056 startSpecifier, specifierLen); 3057 3058 // Check the length modifier is valid with the given conversion specifier. 3059 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3060 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3061 diag::warn_format_nonsensical_length); 3062 else if (!FS.hasStandardLengthModifier()) 3063 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3064 else if (!FS.hasStandardLengthConversionCombination()) 3065 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3066 diag::warn_format_non_standard_conversion_spec); 3067 3068 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3069 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3070 3071 // The remaining checks depend on the data arguments. 3072 if (HasVAListArg) 3073 return true; 3074 3075 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3076 return false; 3077 3078 const Expr *Arg = getDataArg(argIndex); 3079 if (!Arg) 3080 return true; 3081 3082 return checkFormatExpr(FS, startSpecifier, specifierLen, Arg); 3083} 3084 3085static bool requiresParensToAddCast(const Expr *E) { 3086 // FIXME: We should have a general way to reason about operator 3087 // precedence and whether parens are actually needed here. 3088 // Take care of a few common cases where they aren't. 3089 const Expr *Inside = E->IgnoreImpCasts(); 3090 if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside)) 3091 Inside = POE->getSyntacticForm()->IgnoreImpCasts(); 3092 3093 switch (Inside->getStmtClass()) { 3094 case Stmt::ArraySubscriptExprClass: 3095 case Stmt::CallExprClass: 3096 case Stmt::CharacterLiteralClass: 3097 case Stmt::CXXBoolLiteralExprClass: 3098 case Stmt::DeclRefExprClass: 3099 case Stmt::FloatingLiteralClass: 3100 case Stmt::IntegerLiteralClass: 3101 case Stmt::MemberExprClass: 3102 case Stmt::ObjCArrayLiteralClass: 3103 case Stmt::ObjCBoolLiteralExprClass: 3104 case Stmt::ObjCBoxedExprClass: 3105 case Stmt::ObjCDictionaryLiteralClass: 3106 case Stmt::ObjCEncodeExprClass: 3107 case Stmt::ObjCIvarRefExprClass: 3108 case Stmt::ObjCMessageExprClass: 3109 case Stmt::ObjCPropertyRefExprClass: 3110 case Stmt::ObjCStringLiteralClass: 3111 case Stmt::ObjCSubscriptRefExprClass: 3112 case Stmt::ParenExprClass: 3113 case Stmt::StringLiteralClass: 3114 case Stmt::UnaryOperatorClass: 3115 return false; 3116 default: 3117 return true; 3118 } 3119} 3120 3121bool 3122CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 3123 const char *StartSpecifier, 3124 unsigned SpecifierLen, 3125 const Expr *E) { 3126 using namespace analyze_format_string; 3127 using namespace analyze_printf; 3128 // Now type check the data expression that matches the 3129 // format specifier. 3130 const analyze_printf::ArgType &AT = FS.getArgType(S.Context, 3131 ObjCContext); 3132 if (!AT.isValid()) 3133 return true; 3134 3135 QualType ExprTy = E->getType(); 3136 while (const TypeOfExprType *TET = dyn_cast<TypeOfExprType>(ExprTy)) { 3137 ExprTy = TET->getUnderlyingExpr()->getType(); 3138 } 3139 3140 if (AT.matchesType(S.Context, ExprTy)) 3141 return true; 3142 3143 // Look through argument promotions for our error message's reported type. 3144 // This includes the integral and floating promotions, but excludes array 3145 // and function pointer decay; seeing that an argument intended to be a 3146 // string has type 'char [6]' is probably more confusing than 'char *'. 3147 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3148 if (ICE->getCastKind() == CK_IntegralCast || 3149 ICE->getCastKind() == CK_FloatingCast) { 3150 E = ICE->getSubExpr(); 3151 ExprTy = E->getType(); 3152 3153 // Check if we didn't match because of an implicit cast from a 'char' 3154 // or 'short' to an 'int'. This is done because printf is a varargs 3155 // function. 3156 if (ICE->getType() == S.Context.IntTy || 3157 ICE->getType() == S.Context.UnsignedIntTy) { 3158 // All further checking is done on the subexpression. 3159 if (AT.matchesType(S.Context, ExprTy)) 3160 return true; 3161 } 3162 } 3163 } else if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E)) { 3164 // Special case for 'a', which has type 'int' in C. 3165 // Note, however, that we do /not/ want to treat multibyte constants like 3166 // 'MooV' as characters! This form is deprecated but still exists. 3167 if (ExprTy == S.Context.IntTy) 3168 if (llvm::isUIntN(S.Context.getCharWidth(), CL->getValue())) 3169 ExprTy = S.Context.CharTy; 3170 } 3171 3172 // %C in an Objective-C context prints a unichar, not a wchar_t. 3173 // If the argument is an integer of some kind, believe the %C and suggest 3174 // a cast instead of changing the conversion specifier. 3175 QualType IntendedTy = ExprTy; 3176 if (ObjCContext && 3177 FS.getConversionSpecifier().getKind() == ConversionSpecifier::CArg) { 3178 if (ExprTy->isIntegralOrUnscopedEnumerationType() && 3179 !ExprTy->isCharType()) { 3180 // 'unichar' is defined as a typedef of unsigned short, but we should 3181 // prefer using the typedef if it is visible. 3182 IntendedTy = S.Context.UnsignedShortTy; 3183 3184 // While we are here, check if the value is an IntegerLiteral that happens 3185 // to be within the valid range. 3186 if (const IntegerLiteral *IL = dyn_cast<IntegerLiteral>(E)) { 3187 const llvm::APInt &V = IL->getValue(); 3188 if (V.getActiveBits() <= S.Context.getTypeSize(IntendedTy)) 3189 return true; 3190 } 3191 3192 LookupResult Result(S, &S.Context.Idents.get("unichar"), E->getLocStart(), 3193 Sema::LookupOrdinaryName); 3194 if (S.LookupName(Result, S.getCurScope())) { 3195 NamedDecl *ND = Result.getFoundDecl(); 3196 if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(ND)) 3197 if (TD->getUnderlyingType() == IntendedTy) 3198 IntendedTy = S.Context.getTypedefType(TD); 3199 } 3200 } 3201 } 3202 3203 // Special-case some of Darwin's platform-independence types by suggesting 3204 // casts to primitive types that are known to be large enough. 3205 bool ShouldNotPrintDirectly = false; 3206 if (S.Context.getTargetInfo().getTriple().isOSDarwin()) { 3207 // Use a 'while' to peel off layers of typedefs. 3208 QualType TyTy = IntendedTy; 3209 while (const TypedefType *UserTy = TyTy->getAs<TypedefType>()) { 3210 StringRef Name = UserTy->getDecl()->getName(); 3211 QualType CastTy = llvm::StringSwitch<QualType>(Name) 3212 .Case("NSInteger", S.Context.LongTy) 3213 .Case("NSUInteger", S.Context.UnsignedLongTy) 3214 .Case("SInt32", S.Context.IntTy) 3215 .Case("UInt32", S.Context.UnsignedIntTy) 3216 .Default(QualType()); 3217 3218 if (!CastTy.isNull()) { 3219 ShouldNotPrintDirectly = true; 3220 IntendedTy = CastTy; 3221 break; 3222 } 3223 TyTy = UserTy->desugar(); 3224 } 3225 } 3226 3227 // We may be able to offer a FixItHint if it is a supported type. 3228 PrintfSpecifier fixedFS = FS; 3229 bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(), 3230 S.Context, ObjCContext); 3231 3232 if (success) { 3233 // Get the fix string from the fixed format specifier 3234 SmallString<16> buf; 3235 llvm::raw_svector_ostream os(buf); 3236 fixedFS.toString(os); 3237 3238 CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen); 3239 3240 if (IntendedTy == ExprTy) { 3241 // In this case, the specifier is wrong and should be changed to match 3242 // the argument. 3243 EmitFormatDiagnostic( 3244 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3245 << AT.getRepresentativeTypeName(S.Context) << IntendedTy 3246 << E->getSourceRange(), 3247 E->getLocStart(), 3248 /*IsStringLocation*/false, 3249 SpecRange, 3250 FixItHint::CreateReplacement(SpecRange, os.str())); 3251 3252 } else { 3253 // The canonical type for formatting this value is different from the 3254 // actual type of the expression. (This occurs, for example, with Darwin's 3255 // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but 3256 // should be printed as 'long' for 64-bit compatibility.) 3257 // Rather than emitting a normal format/argument mismatch, we want to 3258 // add a cast to the recommended type (and correct the format string 3259 // if necessary). 3260 SmallString<16> CastBuf; 3261 llvm::raw_svector_ostream CastFix(CastBuf); 3262 CastFix << "("; 3263 IntendedTy.print(CastFix, S.Context.getPrintingPolicy()); 3264 CastFix << ")"; 3265 3266 SmallVector<FixItHint,4> Hints; 3267 if (!AT.matchesType(S.Context, IntendedTy)) 3268 Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str())); 3269 3270 if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) { 3271 // If there's already a cast present, just replace it. 3272 SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc()); 3273 Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str())); 3274 3275 } else if (!requiresParensToAddCast(E)) { 3276 // If the expression has high enough precedence, 3277 // just write the C-style cast. 3278 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 3279 CastFix.str())); 3280 } else { 3281 // Otherwise, add parens around the expression as well as the cast. 3282 CastFix << "("; 3283 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 3284 CastFix.str())); 3285 3286 SourceLocation After = S.PP.getLocForEndOfToken(E->getLocEnd()); 3287 Hints.push_back(FixItHint::CreateInsertion(After, ")")); 3288 } 3289 3290 if (ShouldNotPrintDirectly) { 3291 // The expression has a type that should not be printed directly. 3292 // We extract the name from the typedef because we don't want to show 3293 // the underlying type in the diagnostic. 3294 StringRef Name = cast<TypedefType>(ExprTy)->getDecl()->getName(); 3295 3296 EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast) 3297 << Name << IntendedTy 3298 << E->getSourceRange(), 3299 E->getLocStart(), /*IsStringLocation=*/false, 3300 SpecRange, Hints); 3301 } else { 3302 // In this case, the expression could be printed using a different 3303 // specifier, but we've decided that the specifier is probably correct 3304 // and we should cast instead. Just use the normal warning message. 3305 EmitFormatDiagnostic( 3306 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3307 << AT.getRepresentativeTypeName(S.Context) << ExprTy 3308 << E->getSourceRange(), 3309 E->getLocStart(), /*IsStringLocation*/false, 3310 SpecRange, Hints); 3311 } 3312 } 3313 } else { 3314 const CharSourceRange &CSR = getSpecifierRange(StartSpecifier, 3315 SpecifierLen); 3316 // Since the warning for passing non-POD types to variadic functions 3317 // was deferred until now, we emit a warning for non-POD 3318 // arguments here. 3319 switch (S.isValidVarArgType(ExprTy)) { 3320 case Sema::VAK_Valid: 3321 case Sema::VAK_ValidInCXX11: 3322 EmitFormatDiagnostic( 3323 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3324 << AT.getRepresentativeTypeName(S.Context) << ExprTy 3325 << CSR 3326 << E->getSourceRange(), 3327 E->getLocStart(), /*IsStringLocation*/false, CSR); 3328 break; 3329 3330 case Sema::VAK_Undefined: 3331 EmitFormatDiagnostic( 3332 S.PDiag(diag::warn_non_pod_vararg_with_format_string) 3333 << S.getLangOpts().CPlusPlus11 3334 << ExprTy 3335 << CallType 3336 << AT.getRepresentativeTypeName(S.Context) 3337 << CSR 3338 << E->getSourceRange(), 3339 E->getLocStart(), /*IsStringLocation*/false, CSR); 3340 checkForCStrMembers(AT, E, CSR); 3341 break; 3342 3343 case Sema::VAK_Invalid: 3344 if (ExprTy->isObjCObjectType()) 3345 EmitFormatDiagnostic( 3346 S.PDiag(diag::err_cannot_pass_objc_interface_to_vararg_format) 3347 << S.getLangOpts().CPlusPlus11 3348 << ExprTy 3349 << CallType 3350 << AT.getRepresentativeTypeName(S.Context) 3351 << CSR 3352 << E->getSourceRange(), 3353 E->getLocStart(), /*IsStringLocation*/false, CSR); 3354 else 3355 // FIXME: If this is an initializer list, suggest removing the braces 3356 // or inserting a cast to the target type. 3357 S.Diag(E->getLocStart(), diag::err_cannot_pass_to_vararg_format) 3358 << isa<InitListExpr>(E) << ExprTy << CallType 3359 << AT.getRepresentativeTypeName(S.Context) 3360 << E->getSourceRange(); 3361 break; 3362 } 3363 3364 assert(FirstDataArg + FS.getArgIndex() < CheckedVarArgs.size() && 3365 "format string specifier index out of range"); 3366 CheckedVarArgs[FirstDataArg + FS.getArgIndex()] = true; 3367 } 3368 3369 return true; 3370} 3371 3372//===--- CHECK: Scanf format string checking ------------------------------===// 3373 3374namespace { 3375class CheckScanfHandler : public CheckFormatHandler { 3376public: 3377 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 3378 const Expr *origFormatExpr, unsigned firstDataArg, 3379 unsigned numDataArgs, const char *beg, bool hasVAListArg, 3380 ArrayRef<const Expr *> Args, 3381 unsigned formatIdx, bool inFunctionCall, 3382 Sema::VariadicCallType CallType, 3383 llvm::SmallBitVector &CheckedVarArgs) 3384 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 3385 numDataArgs, beg, hasVAListArg, 3386 Args, formatIdx, inFunctionCall, CallType, 3387 CheckedVarArgs) 3388 {} 3389 3390 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 3391 const char *startSpecifier, 3392 unsigned specifierLen); 3393 3394 bool HandleInvalidScanfConversionSpecifier( 3395 const analyze_scanf::ScanfSpecifier &FS, 3396 const char *startSpecifier, 3397 unsigned specifierLen); 3398 3399 void HandleIncompleteScanList(const char *start, const char *end); 3400}; 3401} 3402 3403void CheckScanfHandler::HandleIncompleteScanList(const char *start, 3404 const char *end) { 3405 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 3406 getLocationOfByte(end), /*IsStringLocation*/true, 3407 getSpecifierRange(start, end - start)); 3408} 3409 3410bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 3411 const analyze_scanf::ScanfSpecifier &FS, 3412 const char *startSpecifier, 3413 unsigned specifierLen) { 3414 3415 const analyze_scanf::ScanfConversionSpecifier &CS = 3416 FS.getConversionSpecifier(); 3417 3418 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 3419 getLocationOfByte(CS.getStart()), 3420 startSpecifier, specifierLen, 3421 CS.getStart(), CS.getLength()); 3422} 3423 3424bool CheckScanfHandler::HandleScanfSpecifier( 3425 const analyze_scanf::ScanfSpecifier &FS, 3426 const char *startSpecifier, 3427 unsigned specifierLen) { 3428 3429 using namespace analyze_scanf; 3430 using namespace analyze_format_string; 3431 3432 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 3433 3434 // Handle case where '%' and '*' don't consume an argument. These shouldn't 3435 // be used to decide if we are using positional arguments consistently. 3436 if (FS.consumesDataArgument()) { 3437 if (atFirstArg) { 3438 atFirstArg = false; 3439 usesPositionalArgs = FS.usesPositionalArg(); 3440 } 3441 else if (usesPositionalArgs != FS.usesPositionalArg()) { 3442 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 3443 startSpecifier, specifierLen); 3444 return false; 3445 } 3446 } 3447 3448 // Check if the field with is non-zero. 3449 const OptionalAmount &Amt = FS.getFieldWidth(); 3450 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 3451 if (Amt.getConstantAmount() == 0) { 3452 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 3453 Amt.getConstantLength()); 3454 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 3455 getLocationOfByte(Amt.getStart()), 3456 /*IsStringLocation*/true, R, 3457 FixItHint::CreateRemoval(R)); 3458 } 3459 } 3460 3461 if (!FS.consumesDataArgument()) { 3462 // FIXME: Technically specifying a precision or field width here 3463 // makes no sense. Worth issuing a warning at some point. 3464 return true; 3465 } 3466 3467 // Consume the argument. 3468 unsigned argIndex = FS.getArgIndex(); 3469 if (argIndex < NumDataArgs) { 3470 // The check to see if the argIndex is valid will come later. 3471 // We set the bit here because we may exit early from this 3472 // function if we encounter some other error. 3473 CoveredArgs.set(argIndex); 3474 } 3475 3476 // Check the length modifier is valid with the given conversion specifier. 3477 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3478 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3479 diag::warn_format_nonsensical_length); 3480 else if (!FS.hasStandardLengthModifier()) 3481 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3482 else if (!FS.hasStandardLengthConversionCombination()) 3483 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3484 diag::warn_format_non_standard_conversion_spec); 3485 3486 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3487 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3488 3489 // The remaining checks depend on the data arguments. 3490 if (HasVAListArg) 3491 return true; 3492 3493 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3494 return false; 3495 3496 // Check that the argument type matches the format specifier. 3497 const Expr *Ex = getDataArg(argIndex); 3498 if (!Ex) 3499 return true; 3500 3501 const analyze_format_string::ArgType &AT = FS.getArgType(S.Context); 3502 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) { 3503 ScanfSpecifier fixedFS = FS; 3504 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 3505 S.Context); 3506 3507 if (success) { 3508 // Get the fix string from the fixed format specifier. 3509 SmallString<128> buf; 3510 llvm::raw_svector_ostream os(buf); 3511 fixedFS.toString(os); 3512 3513 EmitFormatDiagnostic( 3514 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3515 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3516 << Ex->getSourceRange(), 3517 Ex->getLocStart(), 3518 /*IsStringLocation*/false, 3519 getSpecifierRange(startSpecifier, specifierLen), 3520 FixItHint::CreateReplacement( 3521 getSpecifierRange(startSpecifier, specifierLen), 3522 os.str())); 3523 } else { 3524 EmitFormatDiagnostic( 3525 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3526 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3527 << Ex->getSourceRange(), 3528 Ex->getLocStart(), 3529 /*IsStringLocation*/false, 3530 getSpecifierRange(startSpecifier, specifierLen)); 3531 } 3532 } 3533 3534 return true; 3535} 3536 3537void Sema::CheckFormatString(const StringLiteral *FExpr, 3538 const Expr *OrigFormatExpr, 3539 ArrayRef<const Expr *> Args, 3540 bool HasVAListArg, unsigned format_idx, 3541 unsigned firstDataArg, FormatStringType Type, 3542 bool inFunctionCall, VariadicCallType CallType, 3543 llvm::SmallBitVector &CheckedVarArgs) { 3544 3545 // CHECK: is the format string a wide literal? 3546 if (!FExpr->isAscii() && !FExpr->isUTF8()) { 3547 CheckFormatHandler::EmitFormatDiagnostic( 3548 *this, inFunctionCall, Args[format_idx], 3549 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 3550 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3551 return; 3552 } 3553 3554 // Str - The format string. NOTE: this is NOT null-terminated! 3555 StringRef StrRef = FExpr->getString(); 3556 const char *Str = StrRef.data(); 3557 unsigned StrLen = StrRef.size(); 3558 const unsigned numDataArgs = Args.size() - firstDataArg; 3559 3560 // CHECK: empty format string? 3561 if (StrLen == 0 && numDataArgs > 0) { 3562 CheckFormatHandler::EmitFormatDiagnostic( 3563 *this, inFunctionCall, Args[format_idx], 3564 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 3565 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3566 return; 3567 } 3568 3569 if (Type == FST_Printf || Type == FST_NSString) { 3570 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 3571 numDataArgs, (Type == FST_NSString), 3572 Str, HasVAListArg, Args, format_idx, 3573 inFunctionCall, CallType, CheckedVarArgs); 3574 3575 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 3576 getLangOpts(), 3577 Context.getTargetInfo())) 3578 H.DoneProcessing(); 3579 } else if (Type == FST_Scanf) { 3580 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs, 3581 Str, HasVAListArg, Args, format_idx, 3582 inFunctionCall, CallType, CheckedVarArgs); 3583 3584 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 3585 getLangOpts(), 3586 Context.getTargetInfo())) 3587 H.DoneProcessing(); 3588 } // TODO: handle other formats 3589} 3590 3591//===--- CHECK: Standard memory functions ---------------------------------===// 3592 3593/// \brief Determine whether the given type is a dynamic class type (e.g., 3594/// whether it has a vtable). 3595static bool isDynamicClassType(QualType T) { 3596 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 3597 if (CXXRecordDecl *Definition = Record->getDefinition()) 3598 if (Definition->isDynamicClass()) 3599 return true; 3600 3601 return false; 3602} 3603 3604/// \brief If E is a sizeof expression, returns its argument expression, 3605/// otherwise returns NULL. 3606static const Expr *getSizeOfExprArg(const Expr* E) { 3607 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3608 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3609 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 3610 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 3611 3612 return 0; 3613} 3614 3615/// \brief If E is a sizeof expression, returns its argument type. 3616static QualType getSizeOfArgType(const Expr* E) { 3617 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3618 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3619 if (SizeOf->getKind() == clang::UETT_SizeOf) 3620 return SizeOf->getTypeOfArgument(); 3621 3622 return QualType(); 3623} 3624 3625/// \brief Check for dangerous or invalid arguments to memset(). 3626/// 3627/// This issues warnings on known problematic, dangerous or unspecified 3628/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 3629/// function calls. 3630/// 3631/// \param Call The call expression to diagnose. 3632void Sema::CheckMemaccessArguments(const CallExpr *Call, 3633 unsigned BId, 3634 IdentifierInfo *FnName) { 3635 assert(BId != 0); 3636 3637 // It is possible to have a non-standard definition of memset. Validate 3638 // we have enough arguments, and if not, abort further checking. 3639 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 3640 if (Call->getNumArgs() < ExpectedNumArgs) 3641 return; 3642 3643 unsigned LastArg = (BId == Builtin::BImemset || 3644 BId == Builtin::BIstrndup ? 1 : 2); 3645 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 3646 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 3647 3648 // We have special checking when the length is a sizeof expression. 3649 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 3650 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 3651 llvm::FoldingSetNodeID SizeOfArgID; 3652 3653 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 3654 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 3655 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 3656 3657 QualType DestTy = Dest->getType(); 3658 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 3659 QualType PointeeTy = DestPtrTy->getPointeeType(); 3660 3661 // Never warn about void type pointers. This can be used to suppress 3662 // false positives. 3663 if (PointeeTy->isVoidType()) 3664 continue; 3665 3666 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 3667 // actually comparing the expressions for equality. Because computing the 3668 // expression IDs can be expensive, we only do this if the diagnostic is 3669 // enabled. 3670 if (SizeOfArg && 3671 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 3672 SizeOfArg->getExprLoc())) { 3673 // We only compute IDs for expressions if the warning is enabled, and 3674 // cache the sizeof arg's ID. 3675 if (SizeOfArgID == llvm::FoldingSetNodeID()) 3676 SizeOfArg->Profile(SizeOfArgID, Context, true); 3677 llvm::FoldingSetNodeID DestID; 3678 Dest->Profile(DestID, Context, true); 3679 if (DestID == SizeOfArgID) { 3680 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 3681 // over sizeof(src) as well. 3682 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 3683 StringRef ReadableName = FnName->getName(); 3684 3685 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 3686 if (UnaryOp->getOpcode() == UO_AddrOf) 3687 ActionIdx = 1; // If its an address-of operator, just remove it. 3688 if (!PointeeTy->isIncompleteType() && 3689 (Context.getTypeSize(PointeeTy) == Context.getCharWidth())) 3690 ActionIdx = 2; // If the pointee's size is sizeof(char), 3691 // suggest an explicit length. 3692 3693 // If the function is defined as a builtin macro, do not show macro 3694 // expansion. 3695 SourceLocation SL = SizeOfArg->getExprLoc(); 3696 SourceRange DSR = Dest->getSourceRange(); 3697 SourceRange SSR = SizeOfArg->getSourceRange(); 3698 SourceManager &SM = PP.getSourceManager(); 3699 3700 if (SM.isMacroArgExpansion(SL)) { 3701 ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts); 3702 SL = SM.getSpellingLoc(SL); 3703 DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()), 3704 SM.getSpellingLoc(DSR.getEnd())); 3705 SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()), 3706 SM.getSpellingLoc(SSR.getEnd())); 3707 } 3708 3709 DiagRuntimeBehavior(SL, SizeOfArg, 3710 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 3711 << ReadableName 3712 << PointeeTy 3713 << DestTy 3714 << DSR 3715 << SSR); 3716 DiagRuntimeBehavior(SL, SizeOfArg, 3717 PDiag(diag::warn_sizeof_pointer_expr_memaccess_note) 3718 << ActionIdx 3719 << SSR); 3720 3721 break; 3722 } 3723 } 3724 3725 // Also check for cases where the sizeof argument is the exact same 3726 // type as the memory argument, and where it points to a user-defined 3727 // record type. 3728 if (SizeOfArgTy != QualType()) { 3729 if (PointeeTy->isRecordType() && 3730 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 3731 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 3732 PDiag(diag::warn_sizeof_pointer_type_memaccess) 3733 << FnName << SizeOfArgTy << ArgIdx 3734 << PointeeTy << Dest->getSourceRange() 3735 << LenExpr->getSourceRange()); 3736 break; 3737 } 3738 } 3739 3740 // Always complain about dynamic classes. 3741 if (isDynamicClassType(PointeeTy)) { 3742 3743 unsigned OperationType = 0; 3744 // "overwritten" if we're warning about the destination for any call 3745 // but memcmp; otherwise a verb appropriate to the call. 3746 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 3747 if (BId == Builtin::BImemcpy) 3748 OperationType = 1; 3749 else if(BId == Builtin::BImemmove) 3750 OperationType = 2; 3751 else if (BId == Builtin::BImemcmp) 3752 OperationType = 3; 3753 } 3754 3755 DiagRuntimeBehavior( 3756 Dest->getExprLoc(), Dest, 3757 PDiag(diag::warn_dyn_class_memaccess) 3758 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 3759 << FnName << PointeeTy 3760 << OperationType 3761 << Call->getCallee()->getSourceRange()); 3762 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 3763 BId != Builtin::BImemset) 3764 DiagRuntimeBehavior( 3765 Dest->getExprLoc(), Dest, 3766 PDiag(diag::warn_arc_object_memaccess) 3767 << ArgIdx << FnName << PointeeTy 3768 << Call->getCallee()->getSourceRange()); 3769 else 3770 continue; 3771 3772 DiagRuntimeBehavior( 3773 Dest->getExprLoc(), Dest, 3774 PDiag(diag::note_bad_memaccess_silence) 3775 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 3776 break; 3777 } 3778 } 3779} 3780 3781// A little helper routine: ignore addition and subtraction of integer literals. 3782// This intentionally does not ignore all integer constant expressions because 3783// we don't want to remove sizeof(). 3784static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 3785 Ex = Ex->IgnoreParenCasts(); 3786 3787 for (;;) { 3788 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 3789 if (!BO || !BO->isAdditiveOp()) 3790 break; 3791 3792 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 3793 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 3794 3795 if (isa<IntegerLiteral>(RHS)) 3796 Ex = LHS; 3797 else if (isa<IntegerLiteral>(LHS)) 3798 Ex = RHS; 3799 else 3800 break; 3801 } 3802 3803 return Ex; 3804} 3805 3806static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty, 3807 ASTContext &Context) { 3808 // Only handle constant-sized or VLAs, but not flexible members. 3809 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) { 3810 // Only issue the FIXIT for arrays of size > 1. 3811 if (CAT->getSize().getSExtValue() <= 1) 3812 return false; 3813 } else if (!Ty->isVariableArrayType()) { 3814 return false; 3815 } 3816 return true; 3817} 3818 3819// Warn if the user has made the 'size' argument to strlcpy or strlcat 3820// be the size of the source, instead of the destination. 3821void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 3822 IdentifierInfo *FnName) { 3823 3824 // Don't crash if the user has the wrong number of arguments 3825 if (Call->getNumArgs() != 3) 3826 return; 3827 3828 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 3829 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 3830 const Expr *CompareWithSrc = NULL; 3831 3832 // Look for 'strlcpy(dst, x, sizeof(x))' 3833 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 3834 CompareWithSrc = Ex; 3835 else { 3836 // Look for 'strlcpy(dst, x, strlen(x))' 3837 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 3838 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 3839 && SizeCall->getNumArgs() == 1) 3840 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 3841 } 3842 } 3843 3844 if (!CompareWithSrc) 3845 return; 3846 3847 // Determine if the argument to sizeof/strlen is equal to the source 3848 // argument. In principle there's all kinds of things you could do 3849 // here, for instance creating an == expression and evaluating it with 3850 // EvaluateAsBooleanCondition, but this uses a more direct technique: 3851 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 3852 if (!SrcArgDRE) 3853 return; 3854 3855 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 3856 if (!CompareWithSrcDRE || 3857 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 3858 return; 3859 3860 const Expr *OriginalSizeArg = Call->getArg(2); 3861 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 3862 << OriginalSizeArg->getSourceRange() << FnName; 3863 3864 // Output a FIXIT hint if the destination is an array (rather than a 3865 // pointer to an array). This could be enhanced to handle some 3866 // pointers if we know the actual size, like if DstArg is 'array+2' 3867 // we could say 'sizeof(array)-2'. 3868 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 3869 if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context)) 3870 return; 3871 3872 SmallString<128> sizeString; 3873 llvm::raw_svector_ostream OS(sizeString); 3874 OS << "sizeof("; 3875 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3876 OS << ")"; 3877 3878 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 3879 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 3880 OS.str()); 3881} 3882 3883/// Check if two expressions refer to the same declaration. 3884static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 3885 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 3886 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 3887 return D1->getDecl() == D2->getDecl(); 3888 return false; 3889} 3890 3891static const Expr *getStrlenExprArg(const Expr *E) { 3892 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 3893 const FunctionDecl *FD = CE->getDirectCallee(); 3894 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 3895 return 0; 3896 return CE->getArg(0)->IgnoreParenCasts(); 3897 } 3898 return 0; 3899} 3900 3901// Warn on anti-patterns as the 'size' argument to strncat. 3902// The correct size argument should look like following: 3903// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 3904void Sema::CheckStrncatArguments(const CallExpr *CE, 3905 IdentifierInfo *FnName) { 3906 // Don't crash if the user has the wrong number of arguments. 3907 if (CE->getNumArgs() < 3) 3908 return; 3909 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 3910 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 3911 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 3912 3913 // Identify common expressions, which are wrongly used as the size argument 3914 // to strncat and may lead to buffer overflows. 3915 unsigned PatternType = 0; 3916 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 3917 // - sizeof(dst) 3918 if (referToTheSameDecl(SizeOfArg, DstArg)) 3919 PatternType = 1; 3920 // - sizeof(src) 3921 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 3922 PatternType = 2; 3923 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 3924 if (BE->getOpcode() == BO_Sub) { 3925 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 3926 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 3927 // - sizeof(dst) - strlen(dst) 3928 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 3929 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 3930 PatternType = 1; 3931 // - sizeof(src) - (anything) 3932 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 3933 PatternType = 2; 3934 } 3935 } 3936 3937 if (PatternType == 0) 3938 return; 3939 3940 // Generate the diagnostic. 3941 SourceLocation SL = LenArg->getLocStart(); 3942 SourceRange SR = LenArg->getSourceRange(); 3943 SourceManager &SM = PP.getSourceManager(); 3944 3945 // If the function is defined as a builtin macro, do not show macro expansion. 3946 if (SM.isMacroArgExpansion(SL)) { 3947 SL = SM.getSpellingLoc(SL); 3948 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 3949 SM.getSpellingLoc(SR.getEnd())); 3950 } 3951 3952 // Check if the destination is an array (rather than a pointer to an array). 3953 QualType DstTy = DstArg->getType(); 3954 bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy, 3955 Context); 3956 if (!isKnownSizeArray) { 3957 if (PatternType == 1) 3958 Diag(SL, diag::warn_strncat_wrong_size) << SR; 3959 else 3960 Diag(SL, diag::warn_strncat_src_size) << SR; 3961 return; 3962 } 3963 3964 if (PatternType == 1) 3965 Diag(SL, diag::warn_strncat_large_size) << SR; 3966 else 3967 Diag(SL, diag::warn_strncat_src_size) << SR; 3968 3969 SmallString<128> sizeString; 3970 llvm::raw_svector_ostream OS(sizeString); 3971 OS << "sizeof("; 3972 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3973 OS << ") - "; 3974 OS << "strlen("; 3975 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3976 OS << ") - 1"; 3977 3978 Diag(SL, diag::note_strncat_wrong_size) 3979 << FixItHint::CreateReplacement(SR, OS.str()); 3980} 3981 3982//===--- CHECK: Return Address of Stack Variable --------------------------===// 3983 3984static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3985 Decl *ParentDecl); 3986static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3987 Decl *ParentDecl); 3988 3989/// CheckReturnStackAddr - Check if a return statement returns the address 3990/// of a stack variable. 3991void 3992Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3993 SourceLocation ReturnLoc) { 3994 3995 Expr *stackE = 0; 3996 SmallVector<DeclRefExpr *, 8> refVars; 3997 3998 // Perform checking for returned stack addresses, local blocks, 3999 // label addresses or references to temporaries. 4000 if (lhsType->isPointerType() || 4001 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 4002 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 4003 } else if (lhsType->isReferenceType()) { 4004 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 4005 } 4006 4007 if (stackE == 0) 4008 return; // Nothing suspicious was found. 4009 4010 SourceLocation diagLoc; 4011 SourceRange diagRange; 4012 if (refVars.empty()) { 4013 diagLoc = stackE->getLocStart(); 4014 diagRange = stackE->getSourceRange(); 4015 } else { 4016 // We followed through a reference variable. 'stackE' contains the 4017 // problematic expression but we will warn at the return statement pointing 4018 // at the reference variable. We will later display the "trail" of 4019 // reference variables using notes. 4020 diagLoc = refVars[0]->getLocStart(); 4021 diagRange = refVars[0]->getSourceRange(); 4022 } 4023 4024 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 4025 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 4026 : diag::warn_ret_stack_addr) 4027 << DR->getDecl()->getDeclName() << diagRange; 4028 } else if (isa<BlockExpr>(stackE)) { // local block. 4029 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 4030 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 4031 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 4032 } else { // local temporary. 4033 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 4034 : diag::warn_ret_local_temp_addr) 4035 << diagRange; 4036 } 4037 4038 // Display the "trail" of reference variables that we followed until we 4039 // found the problematic expression using notes. 4040 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 4041 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 4042 // If this var binds to another reference var, show the range of the next 4043 // var, otherwise the var binds to the problematic expression, in which case 4044 // show the range of the expression. 4045 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 4046 : stackE->getSourceRange(); 4047 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 4048 << VD->getDeclName() << range; 4049 } 4050} 4051 4052/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 4053/// check if the expression in a return statement evaluates to an address 4054/// to a location on the stack, a local block, an address of a label, or a 4055/// reference to local temporary. The recursion is used to traverse the 4056/// AST of the return expression, with recursion backtracking when we 4057/// encounter a subexpression that (1) clearly does not lead to one of the 4058/// above problematic expressions (2) is something we cannot determine leads to 4059/// a problematic expression based on such local checking. 4060/// 4061/// Both EvalAddr and EvalVal follow through reference variables to evaluate 4062/// the expression that they point to. Such variables are added to the 4063/// 'refVars' vector so that we know what the reference variable "trail" was. 4064/// 4065/// EvalAddr processes expressions that are pointers that are used as 4066/// references (and not L-values). EvalVal handles all other values. 4067/// At the base case of the recursion is a check for the above problematic 4068/// expressions. 4069/// 4070/// This implementation handles: 4071/// 4072/// * pointer-to-pointer casts 4073/// * implicit conversions from array references to pointers 4074/// * taking the address of fields 4075/// * arbitrary interplay between "&" and "*" operators 4076/// * pointer arithmetic from an address of a stack variable 4077/// * taking the address of an array element where the array is on the stack 4078static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 4079 Decl *ParentDecl) { 4080 if (E->isTypeDependent()) 4081 return NULL; 4082 4083 // We should only be called for evaluating pointer expressions. 4084 assert((E->getType()->isAnyPointerType() || 4085 E->getType()->isBlockPointerType() || 4086 E->getType()->isObjCQualifiedIdType()) && 4087 "EvalAddr only works on pointers"); 4088 4089 E = E->IgnoreParens(); 4090 4091 // Our "symbolic interpreter" is just a dispatch off the currently 4092 // viewed AST node. We then recursively traverse the AST by calling 4093 // EvalAddr and EvalVal appropriately. 4094 switch (E->getStmtClass()) { 4095 case Stmt::DeclRefExprClass: { 4096 DeclRefExpr *DR = cast<DeclRefExpr>(E); 4097 4098 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 4099 // If this is a reference variable, follow through to the expression that 4100 // it points to. 4101 if (V->hasLocalStorage() && 4102 V->getType()->isReferenceType() && V->hasInit()) { 4103 // Add the reference variable to the "trail". 4104 refVars.push_back(DR); 4105 return EvalAddr(V->getInit(), refVars, ParentDecl); 4106 } 4107 4108 return NULL; 4109 } 4110 4111 case Stmt::UnaryOperatorClass: { 4112 // The only unary operator that make sense to handle here 4113 // is AddrOf. All others don't make sense as pointers. 4114 UnaryOperator *U = cast<UnaryOperator>(E); 4115 4116 if (U->getOpcode() == UO_AddrOf) 4117 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 4118 else 4119 return NULL; 4120 } 4121 4122 case Stmt::BinaryOperatorClass: { 4123 // Handle pointer arithmetic. All other binary operators are not valid 4124 // in this context. 4125 BinaryOperator *B = cast<BinaryOperator>(E); 4126 BinaryOperatorKind op = B->getOpcode(); 4127 4128 if (op != BO_Add && op != BO_Sub) 4129 return NULL; 4130 4131 Expr *Base = B->getLHS(); 4132 4133 // Determine which argument is the real pointer base. It could be 4134 // the RHS argument instead of the LHS. 4135 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 4136 4137 assert (Base->getType()->isPointerType()); 4138 return EvalAddr(Base, refVars, ParentDecl); 4139 } 4140 4141 // For conditional operators we need to see if either the LHS or RHS are 4142 // valid DeclRefExpr*s. If one of them is valid, we return it. 4143 case Stmt::ConditionalOperatorClass: { 4144 ConditionalOperator *C = cast<ConditionalOperator>(E); 4145 4146 // Handle the GNU extension for missing LHS. 4147 if (Expr *lhsExpr = C->getLHS()) { 4148 // In C++, we can have a throw-expression, which has 'void' type. 4149 if (!lhsExpr->getType()->isVoidType()) 4150 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 4151 return LHS; 4152 } 4153 4154 // In C++, we can have a throw-expression, which has 'void' type. 4155 if (C->getRHS()->getType()->isVoidType()) 4156 return NULL; 4157 4158 return EvalAddr(C->getRHS(), refVars, ParentDecl); 4159 } 4160 4161 case Stmt::BlockExprClass: 4162 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 4163 return E; // local block. 4164 return NULL; 4165 4166 case Stmt::AddrLabelExprClass: 4167 return E; // address of label. 4168 4169 case Stmt::ExprWithCleanupsClass: 4170 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 4171 ParentDecl); 4172 4173 // For casts, we need to handle conversions from arrays to 4174 // pointer values, and pointer-to-pointer conversions. 4175 case Stmt::ImplicitCastExprClass: 4176 case Stmt::CStyleCastExprClass: 4177 case Stmt::CXXFunctionalCastExprClass: 4178 case Stmt::ObjCBridgedCastExprClass: 4179 case Stmt::CXXStaticCastExprClass: 4180 case Stmt::CXXDynamicCastExprClass: 4181 case Stmt::CXXConstCastExprClass: 4182 case Stmt::CXXReinterpretCastExprClass: { 4183 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 4184 switch (cast<CastExpr>(E)->getCastKind()) { 4185 case CK_BitCast: 4186 case CK_LValueToRValue: 4187 case CK_NoOp: 4188 case CK_BaseToDerived: 4189 case CK_DerivedToBase: 4190 case CK_UncheckedDerivedToBase: 4191 case CK_Dynamic: 4192 case CK_CPointerToObjCPointerCast: 4193 case CK_BlockPointerToObjCPointerCast: 4194 case CK_AnyPointerToBlockPointerCast: 4195 return EvalAddr(SubExpr, refVars, ParentDecl); 4196 4197 case CK_ArrayToPointerDecay: 4198 return EvalVal(SubExpr, refVars, ParentDecl); 4199 4200 default: 4201 return 0; 4202 } 4203 } 4204 4205 case Stmt::MaterializeTemporaryExprClass: 4206 if (Expr *Result = EvalAddr( 4207 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 4208 refVars, ParentDecl)) 4209 return Result; 4210 4211 return E; 4212 4213 // Everything else: we simply don't reason about them. 4214 default: 4215 return NULL; 4216 } 4217} 4218 4219 4220/// EvalVal - This function is complements EvalAddr in the mutual recursion. 4221/// See the comments for EvalAddr for more details. 4222static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 4223 Decl *ParentDecl) { 4224do { 4225 // We should only be called for evaluating non-pointer expressions, or 4226 // expressions with a pointer type that are not used as references but instead 4227 // are l-values (e.g., DeclRefExpr with a pointer type). 4228 4229 // Our "symbolic interpreter" is just a dispatch off the currently 4230 // viewed AST node. We then recursively traverse the AST by calling 4231 // EvalAddr and EvalVal appropriately. 4232 4233 E = E->IgnoreParens(); 4234 switch (E->getStmtClass()) { 4235 case Stmt::ImplicitCastExprClass: { 4236 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 4237 if (IE->getValueKind() == VK_LValue) { 4238 E = IE->getSubExpr(); 4239 continue; 4240 } 4241 return NULL; 4242 } 4243 4244 case Stmt::ExprWithCleanupsClass: 4245 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 4246 4247 case Stmt::DeclRefExprClass: { 4248 // When we hit a DeclRefExpr we are looking at code that refers to a 4249 // variable's name. If it's not a reference variable we check if it has 4250 // local storage within the function, and if so, return the expression. 4251 DeclRefExpr *DR = cast<DeclRefExpr>(E); 4252 4253 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 4254 // Check if it refers to itself, e.g. "int& i = i;". 4255 if (V == ParentDecl) 4256 return DR; 4257 4258 if (V->hasLocalStorage()) { 4259 if (!V->getType()->isReferenceType()) 4260 return DR; 4261 4262 // Reference variable, follow through to the expression that 4263 // it points to. 4264 if (V->hasInit()) { 4265 // Add the reference variable to the "trail". 4266 refVars.push_back(DR); 4267 return EvalVal(V->getInit(), refVars, V); 4268 } 4269 } 4270 } 4271 4272 return NULL; 4273 } 4274 4275 case Stmt::UnaryOperatorClass: { 4276 // The only unary operator that make sense to handle here 4277 // is Deref. All others don't resolve to a "name." This includes 4278 // handling all sorts of rvalues passed to a unary operator. 4279 UnaryOperator *U = cast<UnaryOperator>(E); 4280 4281 if (U->getOpcode() == UO_Deref) 4282 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 4283 4284 return NULL; 4285 } 4286 4287 case Stmt::ArraySubscriptExprClass: { 4288 // Array subscripts are potential references to data on the stack. We 4289 // retrieve the DeclRefExpr* for the array variable if it indeed 4290 // has local storage. 4291 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 4292 } 4293 4294 case Stmt::ConditionalOperatorClass: { 4295 // For conditional operators we need to see if either the LHS or RHS are 4296 // non-NULL Expr's. If one is non-NULL, we return it. 4297 ConditionalOperator *C = cast<ConditionalOperator>(E); 4298 4299 // Handle the GNU extension for missing LHS. 4300 if (Expr *lhsExpr = C->getLHS()) 4301 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 4302 return LHS; 4303 4304 return EvalVal(C->getRHS(), refVars, ParentDecl); 4305 } 4306 4307 // Accesses to members are potential references to data on the stack. 4308 case Stmt::MemberExprClass: { 4309 MemberExpr *M = cast<MemberExpr>(E); 4310 4311 // Check for indirect access. We only want direct field accesses. 4312 if (M->isArrow()) 4313 return NULL; 4314 4315 // Check whether the member type is itself a reference, in which case 4316 // we're not going to refer to the member, but to what the member refers to. 4317 if (M->getMemberDecl()->getType()->isReferenceType()) 4318 return NULL; 4319 4320 return EvalVal(M->getBase(), refVars, ParentDecl); 4321 } 4322 4323 case Stmt::MaterializeTemporaryExprClass: 4324 if (Expr *Result = EvalVal( 4325 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 4326 refVars, ParentDecl)) 4327 return Result; 4328 4329 return E; 4330 4331 default: 4332 // Check that we don't return or take the address of a reference to a 4333 // temporary. This is only useful in C++. 4334 if (!E->isTypeDependent() && E->isRValue()) 4335 return E; 4336 4337 // Everything else: we simply don't reason about them. 4338 return NULL; 4339 } 4340} while (true); 4341} 4342 4343//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 4344 4345/// Check for comparisons of floating point operands using != and ==. 4346/// Issue a warning if these are no self-comparisons, as they are not likely 4347/// to do what the programmer intended. 4348void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 4349 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 4350 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 4351 4352 // Special case: check for x == x (which is OK). 4353 // Do not emit warnings for such cases. 4354 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 4355 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 4356 if (DRL->getDecl() == DRR->getDecl()) 4357 return; 4358 4359 4360 // Special case: check for comparisons against literals that can be exactly 4361 // represented by APFloat. In such cases, do not emit a warning. This 4362 // is a heuristic: often comparison against such literals are used to 4363 // detect if a value in a variable has not changed. This clearly can 4364 // lead to false negatives. 4365 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 4366 if (FLL->isExact()) 4367 return; 4368 } else 4369 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)) 4370 if (FLR->isExact()) 4371 return; 4372 4373 // Check for comparisons with builtin types. 4374 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 4375 if (CL->isBuiltinCall()) 4376 return; 4377 4378 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 4379 if (CR->isBuiltinCall()) 4380 return; 4381 4382 // Emit the diagnostic. 4383 Diag(Loc, diag::warn_floatingpoint_eq) 4384 << LHS->getSourceRange() << RHS->getSourceRange(); 4385} 4386 4387//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 4388//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 4389 4390namespace { 4391 4392/// Structure recording the 'active' range of an integer-valued 4393/// expression. 4394struct IntRange { 4395 /// The number of bits active in the int. 4396 unsigned Width; 4397 4398 /// True if the int is known not to have negative values. 4399 bool NonNegative; 4400 4401 IntRange(unsigned Width, bool NonNegative) 4402 : Width(Width), NonNegative(NonNegative) 4403 {} 4404 4405 /// Returns the range of the bool type. 4406 static IntRange forBoolType() { 4407 return IntRange(1, true); 4408 } 4409 4410 /// Returns the range of an opaque value of the given integral type. 4411 static IntRange forValueOfType(ASTContext &C, QualType T) { 4412 return forValueOfCanonicalType(C, 4413 T->getCanonicalTypeInternal().getTypePtr()); 4414 } 4415 4416 /// Returns the range of an opaque value of a canonical integral type. 4417 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 4418 assert(T->isCanonicalUnqualified()); 4419 4420 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4421 T = VT->getElementType().getTypePtr(); 4422 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4423 T = CT->getElementType().getTypePtr(); 4424 4425 // For enum types, use the known bit width of the enumerators. 4426 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 4427 EnumDecl *Enum = ET->getDecl(); 4428 if (!Enum->isCompleteDefinition()) 4429 return IntRange(C.getIntWidth(QualType(T, 0)), false); 4430 4431 unsigned NumPositive = Enum->getNumPositiveBits(); 4432 unsigned NumNegative = Enum->getNumNegativeBits(); 4433 4434 if (NumNegative == 0) 4435 return IntRange(NumPositive, true/*NonNegative*/); 4436 else 4437 return IntRange(std::max(NumPositive + 1, NumNegative), 4438 false/*NonNegative*/); 4439 } 4440 4441 const BuiltinType *BT = cast<BuiltinType>(T); 4442 assert(BT->isInteger()); 4443 4444 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4445 } 4446 4447 /// Returns the "target" range of a canonical integral type, i.e. 4448 /// the range of values expressible in the type. 4449 /// 4450 /// This matches forValueOfCanonicalType except that enums have the 4451 /// full range of their type, not the range of their enumerators. 4452 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 4453 assert(T->isCanonicalUnqualified()); 4454 4455 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4456 T = VT->getElementType().getTypePtr(); 4457 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4458 T = CT->getElementType().getTypePtr(); 4459 if (const EnumType *ET = dyn_cast<EnumType>(T)) 4460 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 4461 4462 const BuiltinType *BT = cast<BuiltinType>(T); 4463 assert(BT->isInteger()); 4464 4465 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4466 } 4467 4468 /// Returns the supremum of two ranges: i.e. their conservative merge. 4469 static IntRange join(IntRange L, IntRange R) { 4470 return IntRange(std::max(L.Width, R.Width), 4471 L.NonNegative && R.NonNegative); 4472 } 4473 4474 /// Returns the infinum of two ranges: i.e. their aggressive merge. 4475 static IntRange meet(IntRange L, IntRange R) { 4476 return IntRange(std::min(L.Width, R.Width), 4477 L.NonNegative || R.NonNegative); 4478 } 4479}; 4480 4481static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 4482 unsigned MaxWidth) { 4483 if (value.isSigned() && value.isNegative()) 4484 return IntRange(value.getMinSignedBits(), false); 4485 4486 if (value.getBitWidth() > MaxWidth) 4487 value = value.trunc(MaxWidth); 4488 4489 // isNonNegative() just checks the sign bit without considering 4490 // signedness. 4491 return IntRange(value.getActiveBits(), true); 4492} 4493 4494static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 4495 unsigned MaxWidth) { 4496 if (result.isInt()) 4497 return GetValueRange(C, result.getInt(), MaxWidth); 4498 4499 if (result.isVector()) { 4500 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 4501 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 4502 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 4503 R = IntRange::join(R, El); 4504 } 4505 return R; 4506 } 4507 4508 if (result.isComplexInt()) { 4509 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 4510 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 4511 return IntRange::join(R, I); 4512 } 4513 4514 // This can happen with lossless casts to intptr_t of "based" lvalues. 4515 // Assume it might use arbitrary bits. 4516 // FIXME: The only reason we need to pass the type in here is to get 4517 // the sign right on this one case. It would be nice if APValue 4518 // preserved this. 4519 assert(result.isLValue() || result.isAddrLabelDiff()); 4520 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 4521} 4522 4523static QualType GetExprType(Expr *E) { 4524 QualType Ty = E->getType(); 4525 if (const AtomicType *AtomicRHS = Ty->getAs<AtomicType>()) 4526 Ty = AtomicRHS->getValueType(); 4527 return Ty; 4528} 4529 4530/// Pseudo-evaluate the given integer expression, estimating the 4531/// range of values it might take. 4532/// 4533/// \param MaxWidth - the width to which the value will be truncated 4534static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 4535 E = E->IgnoreParens(); 4536 4537 // Try a full evaluation first. 4538 Expr::EvalResult result; 4539 if (E->EvaluateAsRValue(result, C)) 4540 return GetValueRange(C, result.Val, GetExprType(E), MaxWidth); 4541 4542 // I think we only want to look through implicit casts here; if the 4543 // user has an explicit widening cast, we should treat the value as 4544 // being of the new, wider type. 4545 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 4546 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 4547 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 4548 4549 IntRange OutputTypeRange = IntRange::forValueOfType(C, GetExprType(CE)); 4550 4551 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 4552 4553 // Assume that non-integer casts can span the full range of the type. 4554 if (!isIntegerCast) 4555 return OutputTypeRange; 4556 4557 IntRange SubRange 4558 = GetExprRange(C, CE->getSubExpr(), 4559 std::min(MaxWidth, OutputTypeRange.Width)); 4560 4561 // Bail out if the subexpr's range is as wide as the cast type. 4562 if (SubRange.Width >= OutputTypeRange.Width) 4563 return OutputTypeRange; 4564 4565 // Otherwise, we take the smaller width, and we're non-negative if 4566 // either the output type or the subexpr is. 4567 return IntRange(SubRange.Width, 4568 SubRange.NonNegative || OutputTypeRange.NonNegative); 4569 } 4570 4571 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 4572 // If we can fold the condition, just take that operand. 4573 bool CondResult; 4574 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 4575 return GetExprRange(C, CondResult ? CO->getTrueExpr() 4576 : CO->getFalseExpr(), 4577 MaxWidth); 4578 4579 // Otherwise, conservatively merge. 4580 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 4581 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 4582 return IntRange::join(L, R); 4583 } 4584 4585 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4586 switch (BO->getOpcode()) { 4587 4588 // Boolean-valued operations are single-bit and positive. 4589 case BO_LAnd: 4590 case BO_LOr: 4591 case BO_LT: 4592 case BO_GT: 4593 case BO_LE: 4594 case BO_GE: 4595 case BO_EQ: 4596 case BO_NE: 4597 return IntRange::forBoolType(); 4598 4599 // The type of the assignments is the type of the LHS, so the RHS 4600 // is not necessarily the same type. 4601 case BO_MulAssign: 4602 case BO_DivAssign: 4603 case BO_RemAssign: 4604 case BO_AddAssign: 4605 case BO_SubAssign: 4606 case BO_XorAssign: 4607 case BO_OrAssign: 4608 // TODO: bitfields? 4609 return IntRange::forValueOfType(C, GetExprType(E)); 4610 4611 // Simple assignments just pass through the RHS, which will have 4612 // been coerced to the LHS type. 4613 case BO_Assign: 4614 // TODO: bitfields? 4615 return GetExprRange(C, BO->getRHS(), MaxWidth); 4616 4617 // Operations with opaque sources are black-listed. 4618 case BO_PtrMemD: 4619 case BO_PtrMemI: 4620 return IntRange::forValueOfType(C, GetExprType(E)); 4621 4622 // Bitwise-and uses the *infinum* of the two source ranges. 4623 case BO_And: 4624 case BO_AndAssign: 4625 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 4626 GetExprRange(C, BO->getRHS(), MaxWidth)); 4627 4628 // Left shift gets black-listed based on a judgement call. 4629 case BO_Shl: 4630 // ...except that we want to treat '1 << (blah)' as logically 4631 // positive. It's an important idiom. 4632 if (IntegerLiteral *I 4633 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 4634 if (I->getValue() == 1) { 4635 IntRange R = IntRange::forValueOfType(C, GetExprType(E)); 4636 return IntRange(R.Width, /*NonNegative*/ true); 4637 } 4638 } 4639 // fallthrough 4640 4641 case BO_ShlAssign: 4642 return IntRange::forValueOfType(C, GetExprType(E)); 4643 4644 // Right shift by a constant can narrow its left argument. 4645 case BO_Shr: 4646 case BO_ShrAssign: { 4647 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4648 4649 // If the shift amount is a positive constant, drop the width by 4650 // that much. 4651 llvm::APSInt shift; 4652 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 4653 shift.isNonNegative()) { 4654 unsigned zext = shift.getZExtValue(); 4655 if (zext >= L.Width) 4656 L.Width = (L.NonNegative ? 0 : 1); 4657 else 4658 L.Width -= zext; 4659 } 4660 4661 return L; 4662 } 4663 4664 // Comma acts as its right operand. 4665 case BO_Comma: 4666 return GetExprRange(C, BO->getRHS(), MaxWidth); 4667 4668 // Black-list pointer subtractions. 4669 case BO_Sub: 4670 if (BO->getLHS()->getType()->isPointerType()) 4671 return IntRange::forValueOfType(C, GetExprType(E)); 4672 break; 4673 4674 // The width of a division result is mostly determined by the size 4675 // of the LHS. 4676 case BO_Div: { 4677 // Don't 'pre-truncate' the operands. 4678 unsigned opWidth = C.getIntWidth(GetExprType(E)); 4679 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4680 4681 // If the divisor is constant, use that. 4682 llvm::APSInt divisor; 4683 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 4684 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 4685 if (log2 >= L.Width) 4686 L.Width = (L.NonNegative ? 0 : 1); 4687 else 4688 L.Width = std::min(L.Width - log2, MaxWidth); 4689 return L; 4690 } 4691 4692 // Otherwise, just use the LHS's width. 4693 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4694 return IntRange(L.Width, L.NonNegative && R.NonNegative); 4695 } 4696 4697 // The result of a remainder can't be larger than the result of 4698 // either side. 4699 case BO_Rem: { 4700 // Don't 'pre-truncate' the operands. 4701 unsigned opWidth = C.getIntWidth(GetExprType(E)); 4702 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4703 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4704 4705 IntRange meet = IntRange::meet(L, R); 4706 meet.Width = std::min(meet.Width, MaxWidth); 4707 return meet; 4708 } 4709 4710 // The default behavior is okay for these. 4711 case BO_Mul: 4712 case BO_Add: 4713 case BO_Xor: 4714 case BO_Or: 4715 break; 4716 } 4717 4718 // The default case is to treat the operation as if it were closed 4719 // on the narrowest type that encompasses both operands. 4720 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4721 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 4722 return IntRange::join(L, R); 4723 } 4724 4725 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 4726 switch (UO->getOpcode()) { 4727 // Boolean-valued operations are white-listed. 4728 case UO_LNot: 4729 return IntRange::forBoolType(); 4730 4731 // Operations with opaque sources are black-listed. 4732 case UO_Deref: 4733 case UO_AddrOf: // should be impossible 4734 return IntRange::forValueOfType(C, GetExprType(E)); 4735 4736 default: 4737 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 4738 } 4739 } 4740 4741 if (OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E)) 4742 return GetExprRange(C, OVE->getSourceExpr(), MaxWidth); 4743 4744 if (FieldDecl *BitField = E->getSourceBitField()) 4745 return IntRange(BitField->getBitWidthValue(C), 4746 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 4747 4748 return IntRange::forValueOfType(C, GetExprType(E)); 4749} 4750 4751static IntRange GetExprRange(ASTContext &C, Expr *E) { 4752 return GetExprRange(C, E, C.getIntWidth(GetExprType(E))); 4753} 4754 4755/// Checks whether the given value, which currently has the given 4756/// source semantics, has the same value when coerced through the 4757/// target semantics. 4758static bool IsSameFloatAfterCast(const llvm::APFloat &value, 4759 const llvm::fltSemantics &Src, 4760 const llvm::fltSemantics &Tgt) { 4761 llvm::APFloat truncated = value; 4762 4763 bool ignored; 4764 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 4765 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 4766 4767 return truncated.bitwiseIsEqual(value); 4768} 4769 4770/// Checks whether the given value, which currently has the given 4771/// source semantics, has the same value when coerced through the 4772/// target semantics. 4773/// 4774/// The value might be a vector of floats (or a complex number). 4775static bool IsSameFloatAfterCast(const APValue &value, 4776 const llvm::fltSemantics &Src, 4777 const llvm::fltSemantics &Tgt) { 4778 if (value.isFloat()) 4779 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 4780 4781 if (value.isVector()) { 4782 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 4783 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 4784 return false; 4785 return true; 4786 } 4787 4788 assert(value.isComplexFloat()); 4789 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 4790 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 4791} 4792 4793static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 4794 4795static bool IsZero(Sema &S, Expr *E) { 4796 // Suppress cases where we are comparing against an enum constant. 4797 if (const DeclRefExpr *DR = 4798 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 4799 if (isa<EnumConstantDecl>(DR->getDecl())) 4800 return false; 4801 4802 // Suppress cases where the '0' value is expanded from a macro. 4803 if (E->getLocStart().isMacroID()) 4804 return false; 4805 4806 llvm::APSInt Value; 4807 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 4808} 4809 4810static bool HasEnumType(Expr *E) { 4811 // Strip off implicit integral promotions. 4812 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 4813 if (ICE->getCastKind() != CK_IntegralCast && 4814 ICE->getCastKind() != CK_NoOp) 4815 break; 4816 E = ICE->getSubExpr(); 4817 } 4818 4819 return E->getType()->isEnumeralType(); 4820} 4821 4822static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 4823 // Disable warning in template instantiations. 4824 if (!S.ActiveTemplateInstantiations.empty()) 4825 return; 4826 4827 BinaryOperatorKind op = E->getOpcode(); 4828 if (E->isValueDependent()) 4829 return; 4830 4831 if (op == BO_LT && IsZero(S, E->getRHS())) { 4832 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4833 << "< 0" << "false" << HasEnumType(E->getLHS()) 4834 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4835 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 4836 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4837 << ">= 0" << "true" << HasEnumType(E->getLHS()) 4838 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4839 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 4840 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4841 << "0 >" << "false" << HasEnumType(E->getRHS()) 4842 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4843 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 4844 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4845 << "0 <=" << "true" << HasEnumType(E->getRHS()) 4846 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4847 } 4848} 4849 4850static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E, 4851 Expr *Constant, Expr *Other, 4852 llvm::APSInt Value, 4853 bool RhsConstant) { 4854 // Disable warning in template instantiations. 4855 if (!S.ActiveTemplateInstantiations.empty()) 4856 return; 4857 4858 // 0 values are handled later by CheckTrivialUnsignedComparison(). 4859 if (Value == 0) 4860 return; 4861 4862 BinaryOperatorKind op = E->getOpcode(); 4863 QualType OtherT = Other->getType(); 4864 QualType ConstantT = Constant->getType(); 4865 QualType CommonT = E->getLHS()->getType(); 4866 if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT)) 4867 return; 4868 assert((OtherT->isIntegerType() && ConstantT->isIntegerType()) 4869 && "comparison with non-integer type"); 4870 4871 bool ConstantSigned = ConstantT->isSignedIntegerType(); 4872 bool CommonSigned = CommonT->isSignedIntegerType(); 4873 4874 bool EqualityOnly = false; 4875 4876 // TODO: Investigate using GetExprRange() to get tighter bounds on 4877 // on the bit ranges. 4878 IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT); 4879 unsigned OtherWidth = OtherRange.Width; 4880 4881 if (CommonSigned) { 4882 // The common type is signed, therefore no signed to unsigned conversion. 4883 if (!OtherRange.NonNegative) { 4884 // Check that the constant is representable in type OtherT. 4885 if (ConstantSigned) { 4886 if (OtherWidth >= Value.getMinSignedBits()) 4887 return; 4888 } else { // !ConstantSigned 4889 if (OtherWidth >= Value.getActiveBits() + 1) 4890 return; 4891 } 4892 } else { // !OtherSigned 4893 // Check that the constant is representable in type OtherT. 4894 // Negative values are out of range. 4895 if (ConstantSigned) { 4896 if (Value.isNonNegative() && OtherWidth >= Value.getActiveBits()) 4897 return; 4898 } else { // !ConstantSigned 4899 if (OtherWidth >= Value.getActiveBits()) 4900 return; 4901 } 4902 } 4903 } else { // !CommonSigned 4904 if (OtherRange.NonNegative) { 4905 if (OtherWidth >= Value.getActiveBits()) 4906 return; 4907 } else if (!OtherRange.NonNegative && !ConstantSigned) { 4908 // Check to see if the constant is representable in OtherT. 4909 if (OtherWidth > Value.getActiveBits()) 4910 return; 4911 // Check to see if the constant is equivalent to a negative value 4912 // cast to CommonT. 4913 if (S.Context.getIntWidth(ConstantT) == S.Context.getIntWidth(CommonT) && 4914 Value.isNegative() && Value.getMinSignedBits() <= OtherWidth) 4915 return; 4916 // The constant value rests between values that OtherT can represent after 4917 // conversion. Relational comparison still works, but equality 4918 // comparisons will be tautological. 4919 EqualityOnly = true; 4920 } else { // OtherSigned && ConstantSigned 4921 assert(0 && "Two signed types converted to unsigned types."); 4922 } 4923 } 4924 4925 bool PositiveConstant = !ConstantSigned || Value.isNonNegative(); 4926 4927 bool IsTrue = true; 4928 if (op == BO_EQ || op == BO_NE) { 4929 IsTrue = op == BO_NE; 4930 } else if (EqualityOnly) { 4931 return; 4932 } else if (RhsConstant) { 4933 if (op == BO_GT || op == BO_GE) 4934 IsTrue = !PositiveConstant; 4935 else // op == BO_LT || op == BO_LE 4936 IsTrue = PositiveConstant; 4937 } else { 4938 if (op == BO_LT || op == BO_LE) 4939 IsTrue = !PositiveConstant; 4940 else // op == BO_GT || op == BO_GE 4941 IsTrue = PositiveConstant; 4942 } 4943 4944 // If this is a comparison to an enum constant, include that 4945 // constant in the diagnostic. 4946 const EnumConstantDecl *ED = 0; 4947 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Constant)) 4948 ED = dyn_cast<EnumConstantDecl>(DR->getDecl()); 4949 4950 SmallString<64> PrettySourceValue; 4951 llvm::raw_svector_ostream OS(PrettySourceValue); 4952 if (ED) 4953 OS << '\'' << *ED << "' (" << Value << ")"; 4954 else 4955 OS << Value; 4956 4957 S.Diag(E->getOperatorLoc(), diag::warn_out_of_range_compare) 4958 << OS.str() << OtherT << IsTrue 4959 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4960} 4961 4962/// Analyze the operands of the given comparison. Implements the 4963/// fallback case from AnalyzeComparison. 4964static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 4965 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4966 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4967} 4968 4969/// \brief Implements -Wsign-compare. 4970/// 4971/// \param E the binary operator to check for warnings 4972static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 4973 // The type the comparison is being performed in. 4974 QualType T = E->getLHS()->getType(); 4975 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 4976 && "comparison with mismatched types"); 4977 if (E->isValueDependent()) 4978 return AnalyzeImpConvsInComparison(S, E); 4979 4980 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 4981 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 4982 4983 bool IsComparisonConstant = false; 4984 4985 // Check whether an integer constant comparison results in a value 4986 // of 'true' or 'false'. 4987 if (T->isIntegralType(S.Context)) { 4988 llvm::APSInt RHSValue; 4989 bool IsRHSIntegralLiteral = 4990 RHS->isIntegerConstantExpr(RHSValue, S.Context); 4991 llvm::APSInt LHSValue; 4992 bool IsLHSIntegralLiteral = 4993 LHS->isIntegerConstantExpr(LHSValue, S.Context); 4994 if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral) 4995 DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true); 4996 else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral) 4997 DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false); 4998 else 4999 IsComparisonConstant = 5000 (IsRHSIntegralLiteral && IsLHSIntegralLiteral); 5001 } else if (!T->hasUnsignedIntegerRepresentation()) 5002 IsComparisonConstant = E->isIntegerConstantExpr(S.Context); 5003 5004 // We don't do anything special if this isn't an unsigned integral 5005 // comparison: we're only interested in integral comparisons, and 5006 // signed comparisons only happen in cases we don't care to warn about. 5007 // 5008 // We also don't care about value-dependent expressions or expressions 5009 // whose result is a constant. 5010 if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant) 5011 return AnalyzeImpConvsInComparison(S, E); 5012 5013 // Check to see if one of the (unmodified) operands is of different 5014 // signedness. 5015 Expr *signedOperand, *unsignedOperand; 5016 if (LHS->getType()->hasSignedIntegerRepresentation()) { 5017 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 5018 "unsigned comparison between two signed integer expressions?"); 5019 signedOperand = LHS; 5020 unsignedOperand = RHS; 5021 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 5022 signedOperand = RHS; 5023 unsignedOperand = LHS; 5024 } else { 5025 CheckTrivialUnsignedComparison(S, E); 5026 return AnalyzeImpConvsInComparison(S, E); 5027 } 5028 5029 // Otherwise, calculate the effective range of the signed operand. 5030 IntRange signedRange = GetExprRange(S.Context, signedOperand); 5031 5032 // Go ahead and analyze implicit conversions in the operands. Note 5033 // that we skip the implicit conversions on both sides. 5034 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 5035 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 5036 5037 // If the signed range is non-negative, -Wsign-compare won't fire, 5038 // but we should still check for comparisons which are always true 5039 // or false. 5040 if (signedRange.NonNegative) 5041 return CheckTrivialUnsignedComparison(S, E); 5042 5043 // For (in)equality comparisons, if the unsigned operand is a 5044 // constant which cannot collide with a overflowed signed operand, 5045 // then reinterpreting the signed operand as unsigned will not 5046 // change the result of the comparison. 5047 if (E->isEqualityOp()) { 5048 unsigned comparisonWidth = S.Context.getIntWidth(T); 5049 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 5050 5051 // We should never be unable to prove that the unsigned operand is 5052 // non-negative. 5053 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 5054 5055 if (unsignedRange.Width < comparisonWidth) 5056 return; 5057 } 5058 5059 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 5060 S.PDiag(diag::warn_mixed_sign_comparison) 5061 << LHS->getType() << RHS->getType() 5062 << LHS->getSourceRange() << RHS->getSourceRange()); 5063} 5064 5065/// Analyzes an attempt to assign the given value to a bitfield. 5066/// 5067/// Returns true if there was something fishy about the attempt. 5068static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 5069 SourceLocation InitLoc) { 5070 assert(Bitfield->isBitField()); 5071 if (Bitfield->isInvalidDecl()) 5072 return false; 5073 5074 // White-list bool bitfields. 5075 if (Bitfield->getType()->isBooleanType()) 5076 return false; 5077 5078 // Ignore value- or type-dependent expressions. 5079 if (Bitfield->getBitWidth()->isValueDependent() || 5080 Bitfield->getBitWidth()->isTypeDependent() || 5081 Init->isValueDependent() || 5082 Init->isTypeDependent()) 5083 return false; 5084 5085 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 5086 5087 llvm::APSInt Value; 5088 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 5089 return false; 5090 5091 unsigned OriginalWidth = Value.getBitWidth(); 5092 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 5093 5094 if (OriginalWidth <= FieldWidth) 5095 return false; 5096 5097 // Compute the value which the bitfield will contain. 5098 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 5099 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 5100 5101 // Check whether the stored value is equal to the original value. 5102 TruncatedValue = TruncatedValue.extend(OriginalWidth); 5103 if (llvm::APSInt::isSameValue(Value, TruncatedValue)) 5104 return false; 5105 5106 // Special-case bitfields of width 1: booleans are naturally 0/1, and 5107 // therefore don't strictly fit into a signed bitfield of width 1. 5108 if (FieldWidth == 1 && Value == 1) 5109 return false; 5110 5111 std::string PrettyValue = Value.toString(10); 5112 std::string PrettyTrunc = TruncatedValue.toString(10); 5113 5114 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 5115 << PrettyValue << PrettyTrunc << OriginalInit->getType() 5116 << Init->getSourceRange(); 5117 5118 return true; 5119} 5120 5121/// Analyze the given simple or compound assignment for warning-worthy 5122/// operations. 5123static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 5124 // Just recurse on the LHS. 5125 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 5126 5127 // We want to recurse on the RHS as normal unless we're assigning to 5128 // a bitfield. 5129 if (FieldDecl *Bitfield = E->getLHS()->getSourceBitField()) { 5130 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 5131 E->getOperatorLoc())) { 5132 // Recurse, ignoring any implicit conversions on the RHS. 5133 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 5134 E->getOperatorLoc()); 5135 } 5136 } 5137 5138 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 5139} 5140 5141/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 5142static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 5143 SourceLocation CContext, unsigned diag, 5144 bool pruneControlFlow = false) { 5145 if (pruneControlFlow) { 5146 S.DiagRuntimeBehavior(E->getExprLoc(), E, 5147 S.PDiag(diag) 5148 << SourceType << T << E->getSourceRange() 5149 << SourceRange(CContext)); 5150 return; 5151 } 5152 S.Diag(E->getExprLoc(), diag) 5153 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 5154} 5155 5156/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 5157static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 5158 SourceLocation CContext, unsigned diag, 5159 bool pruneControlFlow = false) { 5160 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 5161} 5162 5163/// Diagnose an implicit cast from a literal expression. Does not warn when the 5164/// cast wouldn't lose information. 5165void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 5166 SourceLocation CContext) { 5167 // Try to convert the literal exactly to an integer. If we can, don't warn. 5168 bool isExact = false; 5169 const llvm::APFloat &Value = FL->getValue(); 5170 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 5171 T->hasUnsignedIntegerRepresentation()); 5172 if (Value.convertToInteger(IntegerValue, 5173 llvm::APFloat::rmTowardZero, &isExact) 5174 == llvm::APFloat::opOK && isExact) 5175 return; 5176 5177 // FIXME: Force the precision of the source value down so we don't print 5178 // digits which are usually useless (we don't really care here if we 5179 // truncate a digit by accident in edge cases). Ideally, APFloat::toString 5180 // would automatically print the shortest representation, but it's a bit 5181 // tricky to implement. 5182 SmallString<16> PrettySourceValue; 5183 unsigned precision = llvm::APFloat::semanticsPrecision(Value.getSemantics()); 5184 precision = (precision * 59 + 195) / 196; 5185 Value.toString(PrettySourceValue, precision); 5186 5187 SmallString<16> PrettyTargetValue; 5188 if (T->isSpecificBuiltinType(BuiltinType::Bool)) 5189 PrettyTargetValue = IntegerValue == 0 ? "false" : "true"; 5190 else 5191 IntegerValue.toString(PrettyTargetValue); 5192 5193 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 5194 << FL->getType() << T.getUnqualifiedType() << PrettySourceValue 5195 << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext); 5196} 5197 5198std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 5199 if (!Range.Width) return "0"; 5200 5201 llvm::APSInt ValueInRange = Value; 5202 ValueInRange.setIsSigned(!Range.NonNegative); 5203 ValueInRange = ValueInRange.trunc(Range.Width); 5204 return ValueInRange.toString(10); 5205} 5206 5207static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) { 5208 if (!isa<ImplicitCastExpr>(Ex)) 5209 return false; 5210 5211 Expr *InnerE = Ex->IgnoreParenImpCasts(); 5212 const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr(); 5213 const Type *Source = 5214 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 5215 if (Target->isDependentType()) 5216 return false; 5217 5218 const BuiltinType *FloatCandidateBT = 5219 dyn_cast<BuiltinType>(ToBool ? Source : Target); 5220 const Type *BoolCandidateType = ToBool ? Target : Source; 5221 5222 return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) && 5223 FloatCandidateBT && (FloatCandidateBT->isFloatingPoint())); 5224} 5225 5226void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall, 5227 SourceLocation CC) { 5228 unsigned NumArgs = TheCall->getNumArgs(); 5229 for (unsigned i = 0; i < NumArgs; ++i) { 5230 Expr *CurrA = TheCall->getArg(i); 5231 if (!IsImplicitBoolFloatConversion(S, CurrA, true)) 5232 continue; 5233 5234 bool IsSwapped = ((i > 0) && 5235 IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false)); 5236 IsSwapped |= ((i < (NumArgs - 1)) && 5237 IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false)); 5238 if (IsSwapped) { 5239 // Warn on this floating-point to bool conversion. 5240 DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(), 5241 CurrA->getType(), CC, 5242 diag::warn_impcast_floating_point_to_bool); 5243 } 5244 } 5245} 5246 5247void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 5248 SourceLocation CC, bool *ICContext = 0) { 5249 if (E->isTypeDependent() || E->isValueDependent()) return; 5250 5251 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 5252 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 5253 if (Source == Target) return; 5254 if (Target->isDependentType()) return; 5255 5256 // If the conversion context location is invalid don't complain. We also 5257 // don't want to emit a warning if the issue occurs from the expansion of 5258 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 5259 // delay this check as long as possible. Once we detect we are in that 5260 // scenario, we just return. 5261 if (CC.isInvalid()) 5262 return; 5263 5264 // Diagnose implicit casts to bool. 5265 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 5266 if (isa<StringLiteral>(E)) 5267 // Warn on string literal to bool. Checks for string literals in logical 5268 // expressions, for instances, assert(0 && "error here"), is prevented 5269 // by a check in AnalyzeImplicitConversions(). 5270 return DiagnoseImpCast(S, E, T, CC, 5271 diag::warn_impcast_string_literal_to_bool); 5272 if (Source->isFunctionType()) { 5273 // Warn on function to bool. Checks free functions and static member 5274 // functions. Weakly imported functions are excluded from the check, 5275 // since it's common to test their value to check whether the linker 5276 // found a definition for them. 5277 ValueDecl *D = 0; 5278 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 5279 D = R->getDecl(); 5280 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 5281 D = M->getMemberDecl(); 5282 } 5283 5284 if (D && !D->isWeak()) { 5285 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 5286 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 5287 << F << E->getSourceRange() << SourceRange(CC); 5288 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 5289 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 5290 QualType ReturnType; 5291 UnresolvedSet<4> NonTemplateOverloads; 5292 S.tryExprAsCall(*E, ReturnType, NonTemplateOverloads); 5293 if (!ReturnType.isNull() 5294 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 5295 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 5296 << FixItHint::CreateInsertion( 5297 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 5298 return; 5299 } 5300 } 5301 } 5302 } 5303 5304 // Strip vector types. 5305 if (isa<VectorType>(Source)) { 5306 if (!isa<VectorType>(Target)) { 5307 if (S.SourceMgr.isInSystemMacro(CC)) 5308 return; 5309 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 5310 } 5311 5312 // If the vector cast is cast between two vectors of the same size, it is 5313 // a bitcast, not a conversion. 5314 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 5315 return; 5316 5317 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 5318 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 5319 } 5320 5321 // Strip complex types. 5322 if (isa<ComplexType>(Source)) { 5323 if (!isa<ComplexType>(Target)) { 5324 if (S.SourceMgr.isInSystemMacro(CC)) 5325 return; 5326 5327 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 5328 } 5329 5330 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 5331 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 5332 } 5333 5334 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 5335 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 5336 5337 // If the source is floating point... 5338 if (SourceBT && SourceBT->isFloatingPoint()) { 5339 // ...and the target is floating point... 5340 if (TargetBT && TargetBT->isFloatingPoint()) { 5341 // ...then warn if we're dropping FP rank. 5342 5343 // Builtin FP kinds are ordered by increasing FP rank. 5344 if (SourceBT->getKind() > TargetBT->getKind()) { 5345 // Don't warn about float constants that are precisely 5346 // representable in the target type. 5347 Expr::EvalResult result; 5348 if (E->EvaluateAsRValue(result, S.Context)) { 5349 // Value might be a float, a float vector, or a float complex. 5350 if (IsSameFloatAfterCast(result.Val, 5351 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 5352 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 5353 return; 5354 } 5355 5356 if (S.SourceMgr.isInSystemMacro(CC)) 5357 return; 5358 5359 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 5360 } 5361 return; 5362 } 5363 5364 // If the target is integral, always warn. 5365 if (TargetBT && TargetBT->isInteger()) { 5366 if (S.SourceMgr.isInSystemMacro(CC)) 5367 return; 5368 5369 Expr *InnerE = E->IgnoreParenImpCasts(); 5370 // We also want to warn on, e.g., "int i = -1.234" 5371 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 5372 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 5373 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 5374 5375 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 5376 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 5377 } else { 5378 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 5379 } 5380 } 5381 5382 // If the target is bool, warn if expr is a function or method call. 5383 if (Target->isSpecificBuiltinType(BuiltinType::Bool) && 5384 isa<CallExpr>(E)) { 5385 // Check last argument of function call to see if it is an 5386 // implicit cast from a type matching the type the result 5387 // is being cast to. 5388 CallExpr *CEx = cast<CallExpr>(E); 5389 unsigned NumArgs = CEx->getNumArgs(); 5390 if (NumArgs > 0) { 5391 Expr *LastA = CEx->getArg(NumArgs - 1); 5392 Expr *InnerE = LastA->IgnoreParenImpCasts(); 5393 const Type *InnerType = 5394 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 5395 if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) { 5396 // Warn on this floating-point to bool conversion 5397 DiagnoseImpCast(S, E, T, CC, 5398 diag::warn_impcast_floating_point_to_bool); 5399 } 5400 } 5401 } 5402 return; 5403 } 5404 5405 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 5406 == Expr::NPCK_GNUNull) && !Target->isAnyPointerType() 5407 && !Target->isBlockPointerType() && !Target->isMemberPointerType() 5408 && Target->isScalarType() && !Target->isNullPtrType()) { 5409 SourceLocation Loc = E->getSourceRange().getBegin(); 5410 if (Loc.isMacroID()) 5411 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 5412 if (!Loc.isMacroID() || CC.isMacroID()) 5413 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 5414 << T << clang::SourceRange(CC) 5415 << FixItHint::CreateReplacement(Loc, 5416 S.getFixItZeroLiteralForType(T, Loc)); 5417 } 5418 5419 if (!Source->isIntegerType() || !Target->isIntegerType()) 5420 return; 5421 5422 // TODO: remove this early return once the false positives for constant->bool 5423 // in templates, macros, etc, are reduced or removed. 5424 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 5425 return; 5426 5427 IntRange SourceRange = GetExprRange(S.Context, E); 5428 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 5429 5430 if (SourceRange.Width > TargetRange.Width) { 5431 // If the source is a constant, use a default-on diagnostic. 5432 // TODO: this should happen for bitfield stores, too. 5433 llvm::APSInt Value(32); 5434 if (E->isIntegerConstantExpr(Value, S.Context)) { 5435 if (S.SourceMgr.isInSystemMacro(CC)) 5436 return; 5437 5438 std::string PrettySourceValue = Value.toString(10); 5439 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 5440 5441 S.DiagRuntimeBehavior(E->getExprLoc(), E, 5442 S.PDiag(diag::warn_impcast_integer_precision_constant) 5443 << PrettySourceValue << PrettyTargetValue 5444 << E->getType() << T << E->getSourceRange() 5445 << clang::SourceRange(CC)); 5446 return; 5447 } 5448 5449 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 5450 if (S.SourceMgr.isInSystemMacro(CC)) 5451 return; 5452 5453 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 5454 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 5455 /* pruneControlFlow */ true); 5456 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 5457 } 5458 5459 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 5460 (!TargetRange.NonNegative && SourceRange.NonNegative && 5461 SourceRange.Width == TargetRange.Width)) { 5462 5463 if (S.SourceMgr.isInSystemMacro(CC)) 5464 return; 5465 5466 unsigned DiagID = diag::warn_impcast_integer_sign; 5467 5468 // Traditionally, gcc has warned about this under -Wsign-compare. 5469 // We also want to warn about it in -Wconversion. 5470 // So if -Wconversion is off, use a completely identical diagnostic 5471 // in the sign-compare group. 5472 // The conditional-checking code will 5473 if (ICContext) { 5474 DiagID = diag::warn_impcast_integer_sign_conditional; 5475 *ICContext = true; 5476 } 5477 5478 return DiagnoseImpCast(S, E, T, CC, DiagID); 5479 } 5480 5481 // Diagnose conversions between different enumeration types. 5482 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 5483 // type, to give us better diagnostics. 5484 QualType SourceType = E->getType(); 5485 if (!S.getLangOpts().CPlusPlus) { 5486 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5487 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 5488 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 5489 SourceType = S.Context.getTypeDeclType(Enum); 5490 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 5491 } 5492 } 5493 5494 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 5495 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 5496 if (SourceEnum->getDecl()->hasNameForLinkage() && 5497 TargetEnum->getDecl()->hasNameForLinkage() && 5498 SourceEnum != TargetEnum) { 5499 if (S.SourceMgr.isInSystemMacro(CC)) 5500 return; 5501 5502 return DiagnoseImpCast(S, E, SourceType, T, CC, 5503 diag::warn_impcast_different_enum_types); 5504 } 5505 5506 return; 5507} 5508 5509void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5510 SourceLocation CC, QualType T); 5511 5512void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 5513 SourceLocation CC, bool &ICContext) { 5514 E = E->IgnoreParenImpCasts(); 5515 5516 if (isa<ConditionalOperator>(E)) 5517 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T); 5518 5519 AnalyzeImplicitConversions(S, E, CC); 5520 if (E->getType() != T) 5521 return CheckImplicitConversion(S, E, T, CC, &ICContext); 5522 return; 5523} 5524 5525void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5526 SourceLocation CC, QualType T) { 5527 AnalyzeImplicitConversions(S, E->getCond(), CC); 5528 5529 bool Suspicious = false; 5530 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 5531 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 5532 5533 // If -Wconversion would have warned about either of the candidates 5534 // for a signedness conversion to the context type... 5535 if (!Suspicious) return; 5536 5537 // ...but it's currently ignored... 5538 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 5539 CC)) 5540 return; 5541 5542 // ...then check whether it would have warned about either of the 5543 // candidates for a signedness conversion to the condition type. 5544 if (E->getType() == T) return; 5545 5546 Suspicious = false; 5547 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 5548 E->getType(), CC, &Suspicious); 5549 if (!Suspicious) 5550 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 5551 E->getType(), CC, &Suspicious); 5552} 5553 5554/// AnalyzeImplicitConversions - Find and report any interesting 5555/// implicit conversions in the given expression. There are a couple 5556/// of competing diagnostics here, -Wconversion and -Wsign-compare. 5557void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 5558 QualType T = OrigE->getType(); 5559 Expr *E = OrigE->IgnoreParenImpCasts(); 5560 5561 if (E->isTypeDependent() || E->isValueDependent()) 5562 return; 5563 5564 // For conditional operators, we analyze the arguments as if they 5565 // were being fed directly into the output. 5566 if (isa<ConditionalOperator>(E)) { 5567 ConditionalOperator *CO = cast<ConditionalOperator>(E); 5568 CheckConditionalOperator(S, CO, CC, T); 5569 return; 5570 } 5571 5572 // Check implicit argument conversions for function calls. 5573 if (CallExpr *Call = dyn_cast<CallExpr>(E)) 5574 CheckImplicitArgumentConversions(S, Call, CC); 5575 5576 // Go ahead and check any implicit conversions we might have skipped. 5577 // The non-canonical typecheck is just an optimization; 5578 // CheckImplicitConversion will filter out dead implicit conversions. 5579 if (E->getType() != T) 5580 CheckImplicitConversion(S, E, T, CC); 5581 5582 // Now continue drilling into this expression. 5583 5584 if (PseudoObjectExpr * POE = dyn_cast<PseudoObjectExpr>(E)) { 5585 if (POE->getResultExpr()) 5586 E = POE->getResultExpr(); 5587 } 5588 5589 if (const OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E)) 5590 return AnalyzeImplicitConversions(S, OVE->getSourceExpr(), CC); 5591 5592 // Skip past explicit casts. 5593 if (isa<ExplicitCastExpr>(E)) { 5594 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 5595 return AnalyzeImplicitConversions(S, E, CC); 5596 } 5597 5598 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5599 // Do a somewhat different check with comparison operators. 5600 if (BO->isComparisonOp()) 5601 return AnalyzeComparison(S, BO); 5602 5603 // And with simple assignments. 5604 if (BO->getOpcode() == BO_Assign) 5605 return AnalyzeAssignment(S, BO); 5606 } 5607 5608 // These break the otherwise-useful invariant below. Fortunately, 5609 // we don't really need to recurse into them, because any internal 5610 // expressions should have been analyzed already when they were 5611 // built into statements. 5612 if (isa<StmtExpr>(E)) return; 5613 5614 // Don't descend into unevaluated contexts. 5615 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 5616 5617 // Now just recurse over the expression's children. 5618 CC = E->getExprLoc(); 5619 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 5620 bool IsLogicalOperator = BO && BO->isLogicalOp(); 5621 for (Stmt::child_range I = E->children(); I; ++I) { 5622 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 5623 if (!ChildExpr) 5624 continue; 5625 5626 if (IsLogicalOperator && 5627 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 5628 // Ignore checking string literals that are in logical operators. 5629 continue; 5630 AnalyzeImplicitConversions(S, ChildExpr, CC); 5631 } 5632} 5633 5634} // end anonymous namespace 5635 5636/// Diagnoses "dangerous" implicit conversions within the given 5637/// expression (which is a full expression). Implements -Wconversion 5638/// and -Wsign-compare. 5639/// 5640/// \param CC the "context" location of the implicit conversion, i.e. 5641/// the most location of the syntactic entity requiring the implicit 5642/// conversion 5643void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 5644 // Don't diagnose in unevaluated contexts. 5645 if (isUnevaluatedContext()) 5646 return; 5647 5648 // Don't diagnose for value- or type-dependent expressions. 5649 if (E->isTypeDependent() || E->isValueDependent()) 5650 return; 5651 5652 // Check for array bounds violations in cases where the check isn't triggered 5653 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 5654 // ArraySubscriptExpr is on the RHS of a variable initialization. 5655 CheckArrayAccess(E); 5656 5657 // This is not the right CC for (e.g.) a variable initialization. 5658 AnalyzeImplicitConversions(*this, E, CC); 5659} 5660 5661/// Diagnose when expression is an integer constant expression and its evaluation 5662/// results in integer overflow 5663void Sema::CheckForIntOverflow (Expr *E) { 5664 if (isa<BinaryOperator>(E->IgnoreParens())) 5665 E->EvaluateForOverflow(Context); 5666} 5667 5668namespace { 5669/// \brief Visitor for expressions which looks for unsequenced operations on the 5670/// same object. 5671class SequenceChecker : public EvaluatedExprVisitor<SequenceChecker> { 5672 typedef EvaluatedExprVisitor<SequenceChecker> Base; 5673 5674 /// \brief A tree of sequenced regions within an expression. Two regions are 5675 /// unsequenced if one is an ancestor or a descendent of the other. When we 5676 /// finish processing an expression with sequencing, such as a comma 5677 /// expression, we fold its tree nodes into its parent, since they are 5678 /// unsequenced with respect to nodes we will visit later. 5679 class SequenceTree { 5680 struct Value { 5681 explicit Value(unsigned Parent) : Parent(Parent), Merged(false) {} 5682 unsigned Parent : 31; 5683 bool Merged : 1; 5684 }; 5685 SmallVector<Value, 8> Values; 5686 5687 public: 5688 /// \brief A region within an expression which may be sequenced with respect 5689 /// to some other region. 5690 class Seq { 5691 explicit Seq(unsigned N) : Index(N) {} 5692 unsigned Index; 5693 friend class SequenceTree; 5694 public: 5695 Seq() : Index(0) {} 5696 }; 5697 5698 SequenceTree() { Values.push_back(Value(0)); } 5699 Seq root() const { return Seq(0); } 5700 5701 /// \brief Create a new sequence of operations, which is an unsequenced 5702 /// subset of \p Parent. This sequence of operations is sequenced with 5703 /// respect to other children of \p Parent. 5704 Seq allocate(Seq Parent) { 5705 Values.push_back(Value(Parent.Index)); 5706 return Seq(Values.size() - 1); 5707 } 5708 5709 /// \brief Merge a sequence of operations into its parent. 5710 void merge(Seq S) { 5711 Values[S.Index].Merged = true; 5712 } 5713 5714 /// \brief Determine whether two operations are unsequenced. This operation 5715 /// is asymmetric: \p Cur should be the more recent sequence, and \p Old 5716 /// should have been merged into its parent as appropriate. 5717 bool isUnsequenced(Seq Cur, Seq Old) { 5718 unsigned C = representative(Cur.Index); 5719 unsigned Target = representative(Old.Index); 5720 while (C >= Target) { 5721 if (C == Target) 5722 return true; 5723 C = Values[C].Parent; 5724 } 5725 return false; 5726 } 5727 5728 private: 5729 /// \brief Pick a representative for a sequence. 5730 unsigned representative(unsigned K) { 5731 if (Values[K].Merged) 5732 // Perform path compression as we go. 5733 return Values[K].Parent = representative(Values[K].Parent); 5734 return K; 5735 } 5736 }; 5737 5738 /// An object for which we can track unsequenced uses. 5739 typedef NamedDecl *Object; 5740 5741 /// Different flavors of object usage which we track. We only track the 5742 /// least-sequenced usage of each kind. 5743 enum UsageKind { 5744 /// A read of an object. Multiple unsequenced reads are OK. 5745 UK_Use, 5746 /// A modification of an object which is sequenced before the value 5747 /// computation of the expression, such as ++n in C++. 5748 UK_ModAsValue, 5749 /// A modification of an object which is not sequenced before the value 5750 /// computation of the expression, such as n++. 5751 UK_ModAsSideEffect, 5752 5753 UK_Count = UK_ModAsSideEffect + 1 5754 }; 5755 5756 struct Usage { 5757 Usage() : Use(0), Seq() {} 5758 Expr *Use; 5759 SequenceTree::Seq Seq; 5760 }; 5761 5762 struct UsageInfo { 5763 UsageInfo() : Diagnosed(false) {} 5764 Usage Uses[UK_Count]; 5765 /// Have we issued a diagnostic for this variable already? 5766 bool Diagnosed; 5767 }; 5768 typedef llvm::SmallDenseMap<Object, UsageInfo, 16> UsageInfoMap; 5769 5770 Sema &SemaRef; 5771 /// Sequenced regions within the expression. 5772 SequenceTree Tree; 5773 /// Declaration modifications and references which we have seen. 5774 UsageInfoMap UsageMap; 5775 /// The region we are currently within. 5776 SequenceTree::Seq Region; 5777 /// Filled in with declarations which were modified as a side-effect 5778 /// (that is, post-increment operations). 5779 SmallVectorImpl<std::pair<Object, Usage> > *ModAsSideEffect; 5780 /// Expressions to check later. We defer checking these to reduce 5781 /// stack usage. 5782 SmallVectorImpl<Expr *> &WorkList; 5783 5784 /// RAII object wrapping the visitation of a sequenced subexpression of an 5785 /// expression. At the end of this process, the side-effects of the evaluation 5786 /// become sequenced with respect to the value computation of the result, so 5787 /// we downgrade any UK_ModAsSideEffect within the evaluation to 5788 /// UK_ModAsValue. 5789 struct SequencedSubexpression { 5790 SequencedSubexpression(SequenceChecker &Self) 5791 : Self(Self), OldModAsSideEffect(Self.ModAsSideEffect) { 5792 Self.ModAsSideEffect = &ModAsSideEffect; 5793 } 5794 ~SequencedSubexpression() { 5795 for (unsigned I = 0, E = ModAsSideEffect.size(); I != E; ++I) { 5796 UsageInfo &U = Self.UsageMap[ModAsSideEffect[I].first]; 5797 U.Uses[UK_ModAsSideEffect] = ModAsSideEffect[I].second; 5798 Self.addUsage(U, ModAsSideEffect[I].first, 5799 ModAsSideEffect[I].second.Use, UK_ModAsValue); 5800 } 5801 Self.ModAsSideEffect = OldModAsSideEffect; 5802 } 5803 5804 SequenceChecker &Self; 5805 SmallVector<std::pair<Object, Usage>, 4> ModAsSideEffect; 5806 SmallVectorImpl<std::pair<Object, Usage> > *OldModAsSideEffect; 5807 }; 5808 5809 /// RAII object wrapping the visitation of a subexpression which we might 5810 /// choose to evaluate as a constant. If any subexpression is evaluated and 5811 /// found to be non-constant, this allows us to suppress the evaluation of 5812 /// the outer expression. 5813 class EvaluationTracker { 5814 public: 5815 EvaluationTracker(SequenceChecker &Self) 5816 : Self(Self), Prev(Self.EvalTracker), EvalOK(true) { 5817 Self.EvalTracker = this; 5818 } 5819 ~EvaluationTracker() { 5820 Self.EvalTracker = Prev; 5821 if (Prev) 5822 Prev->EvalOK &= EvalOK; 5823 } 5824 5825 bool evaluate(const Expr *E, bool &Result) { 5826 if (!EvalOK || E->isValueDependent()) 5827 return false; 5828 EvalOK = E->EvaluateAsBooleanCondition(Result, Self.SemaRef.Context); 5829 return EvalOK; 5830 } 5831 5832 private: 5833 SequenceChecker &Self; 5834 EvaluationTracker *Prev; 5835 bool EvalOK; 5836 } *EvalTracker; 5837 5838 /// \brief Find the object which is produced by the specified expression, 5839 /// if any. 5840 Object getObject(Expr *E, bool Mod) const { 5841 E = E->IgnoreParenCasts(); 5842 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 5843 if (Mod && (UO->getOpcode() == UO_PreInc || UO->getOpcode() == UO_PreDec)) 5844 return getObject(UO->getSubExpr(), Mod); 5845 } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5846 if (BO->getOpcode() == BO_Comma) 5847 return getObject(BO->getRHS(), Mod); 5848 if (Mod && BO->isAssignmentOp()) 5849 return getObject(BO->getLHS(), Mod); 5850 } else if (MemberExpr *ME = dyn_cast<MemberExpr>(E)) { 5851 // FIXME: Check for more interesting cases, like "x.n = ++x.n". 5852 if (isa<CXXThisExpr>(ME->getBase()->IgnoreParenCasts())) 5853 return ME->getMemberDecl(); 5854 } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5855 // FIXME: If this is a reference, map through to its value. 5856 return DRE->getDecl(); 5857 return 0; 5858 } 5859 5860 /// \brief Note that an object was modified or used by an expression. 5861 void addUsage(UsageInfo &UI, Object O, Expr *Ref, UsageKind UK) { 5862 Usage &U = UI.Uses[UK]; 5863 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) { 5864 if (UK == UK_ModAsSideEffect && ModAsSideEffect) 5865 ModAsSideEffect->push_back(std::make_pair(O, U)); 5866 U.Use = Ref; 5867 U.Seq = Region; 5868 } 5869 } 5870 /// \brief Check whether a modification or use conflicts with a prior usage. 5871 void checkUsage(Object O, UsageInfo &UI, Expr *Ref, UsageKind OtherKind, 5872 bool IsModMod) { 5873 if (UI.Diagnosed) 5874 return; 5875 5876 const Usage &U = UI.Uses[OtherKind]; 5877 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) 5878 return; 5879 5880 Expr *Mod = U.Use; 5881 Expr *ModOrUse = Ref; 5882 if (OtherKind == UK_Use) 5883 std::swap(Mod, ModOrUse); 5884 5885 SemaRef.Diag(Mod->getExprLoc(), 5886 IsModMod ? diag::warn_unsequenced_mod_mod 5887 : diag::warn_unsequenced_mod_use) 5888 << O << SourceRange(ModOrUse->getExprLoc()); 5889 UI.Diagnosed = true; 5890 } 5891 5892 void notePreUse(Object O, Expr *Use) { 5893 UsageInfo &U = UsageMap[O]; 5894 // Uses conflict with other modifications. 5895 checkUsage(O, U, Use, UK_ModAsValue, false); 5896 } 5897 void notePostUse(Object O, Expr *Use) { 5898 UsageInfo &U = UsageMap[O]; 5899 checkUsage(O, U, Use, UK_ModAsSideEffect, false); 5900 addUsage(U, O, Use, UK_Use); 5901 } 5902 5903 void notePreMod(Object O, Expr *Mod) { 5904 UsageInfo &U = UsageMap[O]; 5905 // Modifications conflict with other modifications and with uses. 5906 checkUsage(O, U, Mod, UK_ModAsValue, true); 5907 checkUsage(O, U, Mod, UK_Use, false); 5908 } 5909 void notePostMod(Object O, Expr *Use, UsageKind UK) { 5910 UsageInfo &U = UsageMap[O]; 5911 checkUsage(O, U, Use, UK_ModAsSideEffect, true); 5912 addUsage(U, O, Use, UK); 5913 } 5914 5915public: 5916 SequenceChecker(Sema &S, Expr *E, SmallVectorImpl<Expr *> &WorkList) 5917 : Base(S.Context), SemaRef(S), Region(Tree.root()), ModAsSideEffect(0), 5918 WorkList(WorkList), EvalTracker(0) { 5919 Visit(E); 5920 } 5921 5922 void VisitStmt(Stmt *S) { 5923 // Skip all statements which aren't expressions for now. 5924 } 5925 5926 void VisitExpr(Expr *E) { 5927 // By default, just recurse to evaluated subexpressions. 5928 Base::VisitStmt(E); 5929 } 5930 5931 void VisitCastExpr(CastExpr *E) { 5932 Object O = Object(); 5933 if (E->getCastKind() == CK_LValueToRValue) 5934 O = getObject(E->getSubExpr(), false); 5935 5936 if (O) 5937 notePreUse(O, E); 5938 VisitExpr(E); 5939 if (O) 5940 notePostUse(O, E); 5941 } 5942 5943 void VisitBinComma(BinaryOperator *BO) { 5944 // C++11 [expr.comma]p1: 5945 // Every value computation and side effect associated with the left 5946 // expression is sequenced before every value computation and side 5947 // effect associated with the right expression. 5948 SequenceTree::Seq LHS = Tree.allocate(Region); 5949 SequenceTree::Seq RHS = Tree.allocate(Region); 5950 SequenceTree::Seq OldRegion = Region; 5951 5952 { 5953 SequencedSubexpression SeqLHS(*this); 5954 Region = LHS; 5955 Visit(BO->getLHS()); 5956 } 5957 5958 Region = RHS; 5959 Visit(BO->getRHS()); 5960 5961 Region = OldRegion; 5962 5963 // Forget that LHS and RHS are sequenced. They are both unsequenced 5964 // with respect to other stuff. 5965 Tree.merge(LHS); 5966 Tree.merge(RHS); 5967 } 5968 5969 void VisitBinAssign(BinaryOperator *BO) { 5970 // The modification is sequenced after the value computation of the LHS 5971 // and RHS, so check it before inspecting the operands and update the 5972 // map afterwards. 5973 Object O = getObject(BO->getLHS(), true); 5974 if (!O) 5975 return VisitExpr(BO); 5976 5977 notePreMod(O, BO); 5978 5979 // C++11 [expr.ass]p7: 5980 // E1 op= E2 is equivalent to E1 = E1 op E2, except that E1 is evaluated 5981 // only once. 5982 // 5983 // Therefore, for a compound assignment operator, O is considered used 5984 // everywhere except within the evaluation of E1 itself. 5985 if (isa<CompoundAssignOperator>(BO)) 5986 notePreUse(O, BO); 5987 5988 Visit(BO->getLHS()); 5989 5990 if (isa<CompoundAssignOperator>(BO)) 5991 notePostUse(O, BO); 5992 5993 Visit(BO->getRHS()); 5994 5995 // C++11 [expr.ass]p1: 5996 // the assignment is sequenced [...] before the value computation of the 5997 // assignment expression. 5998 // C11 6.5.16/3 has no such rule. 5999 notePostMod(O, BO, SemaRef.getLangOpts().CPlusPlus ? UK_ModAsValue 6000 : UK_ModAsSideEffect); 6001 } 6002 void VisitCompoundAssignOperator(CompoundAssignOperator *CAO) { 6003 VisitBinAssign(CAO); 6004 } 6005 6006 void VisitUnaryPreInc(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 6007 void VisitUnaryPreDec(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 6008 void VisitUnaryPreIncDec(UnaryOperator *UO) { 6009 Object O = getObject(UO->getSubExpr(), true); 6010 if (!O) 6011 return VisitExpr(UO); 6012 6013 notePreMod(O, UO); 6014 Visit(UO->getSubExpr()); 6015 // C++11 [expr.pre.incr]p1: 6016 // the expression ++x is equivalent to x+=1 6017 notePostMod(O, UO, SemaRef.getLangOpts().CPlusPlus ? UK_ModAsValue 6018 : UK_ModAsSideEffect); 6019 } 6020 6021 void VisitUnaryPostInc(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 6022 void VisitUnaryPostDec(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 6023 void VisitUnaryPostIncDec(UnaryOperator *UO) { 6024 Object O = getObject(UO->getSubExpr(), true); 6025 if (!O) 6026 return VisitExpr(UO); 6027 6028 notePreMod(O, UO); 6029 Visit(UO->getSubExpr()); 6030 notePostMod(O, UO, UK_ModAsSideEffect); 6031 } 6032 6033 /// Don't visit the RHS of '&&' or '||' if it might not be evaluated. 6034 void VisitBinLOr(BinaryOperator *BO) { 6035 // The side-effects of the LHS of an '&&' are sequenced before the 6036 // value computation of the RHS, and hence before the value computation 6037 // of the '&&' itself, unless the LHS evaluates to zero. We treat them 6038 // as if they were unconditionally sequenced. 6039 EvaluationTracker Eval(*this); 6040 { 6041 SequencedSubexpression Sequenced(*this); 6042 Visit(BO->getLHS()); 6043 } 6044 6045 bool Result; 6046 if (Eval.evaluate(BO->getLHS(), Result)) { 6047 if (!Result) 6048 Visit(BO->getRHS()); 6049 } else { 6050 // Check for unsequenced operations in the RHS, treating it as an 6051 // entirely separate evaluation. 6052 // 6053 // FIXME: If there are operations in the RHS which are unsequenced 6054 // with respect to operations outside the RHS, and those operations 6055 // are unconditionally evaluated, diagnose them. 6056 WorkList.push_back(BO->getRHS()); 6057 } 6058 } 6059 void VisitBinLAnd(BinaryOperator *BO) { 6060 EvaluationTracker Eval(*this); 6061 { 6062 SequencedSubexpression Sequenced(*this); 6063 Visit(BO->getLHS()); 6064 } 6065 6066 bool Result; 6067 if (Eval.evaluate(BO->getLHS(), Result)) { 6068 if (Result) 6069 Visit(BO->getRHS()); 6070 } else { 6071 WorkList.push_back(BO->getRHS()); 6072 } 6073 } 6074 6075 // Only visit the condition, unless we can be sure which subexpression will 6076 // be chosen. 6077 void VisitAbstractConditionalOperator(AbstractConditionalOperator *CO) { 6078 EvaluationTracker Eval(*this); 6079 { 6080 SequencedSubexpression Sequenced(*this); 6081 Visit(CO->getCond()); 6082 } 6083 6084 bool Result; 6085 if (Eval.evaluate(CO->getCond(), Result)) 6086 Visit(Result ? CO->getTrueExpr() : CO->getFalseExpr()); 6087 else { 6088 WorkList.push_back(CO->getTrueExpr()); 6089 WorkList.push_back(CO->getFalseExpr()); 6090 } 6091 } 6092 6093 void VisitCallExpr(CallExpr *CE) { 6094 // C++11 [intro.execution]p15: 6095 // When calling a function [...], every value computation and side effect 6096 // associated with any argument expression, or with the postfix expression 6097 // designating the called function, is sequenced before execution of every 6098 // expression or statement in the body of the function [and thus before 6099 // the value computation of its result]. 6100 SequencedSubexpression Sequenced(*this); 6101 Base::VisitCallExpr(CE); 6102 6103 // FIXME: CXXNewExpr and CXXDeleteExpr implicitly call functions. 6104 } 6105 6106 void VisitCXXConstructExpr(CXXConstructExpr *CCE) { 6107 // This is a call, so all subexpressions are sequenced before the result. 6108 SequencedSubexpression Sequenced(*this); 6109 6110 if (!CCE->isListInitialization()) 6111 return VisitExpr(CCE); 6112 6113 // In C++11, list initializations are sequenced. 6114 SmallVector<SequenceTree::Seq, 32> Elts; 6115 SequenceTree::Seq Parent = Region; 6116 for (CXXConstructExpr::arg_iterator I = CCE->arg_begin(), 6117 E = CCE->arg_end(); 6118 I != E; ++I) { 6119 Region = Tree.allocate(Parent); 6120 Elts.push_back(Region); 6121 Visit(*I); 6122 } 6123 6124 // Forget that the initializers are sequenced. 6125 Region = Parent; 6126 for (unsigned I = 0; I < Elts.size(); ++I) 6127 Tree.merge(Elts[I]); 6128 } 6129 6130 void VisitInitListExpr(InitListExpr *ILE) { 6131 if (!SemaRef.getLangOpts().CPlusPlus11) 6132 return VisitExpr(ILE); 6133 6134 // In C++11, list initializations are sequenced. 6135 SmallVector<SequenceTree::Seq, 32> Elts; 6136 SequenceTree::Seq Parent = Region; 6137 for (unsigned I = 0; I < ILE->getNumInits(); ++I) { 6138 Expr *E = ILE->getInit(I); 6139 if (!E) continue; 6140 Region = Tree.allocate(Parent); 6141 Elts.push_back(Region); 6142 Visit(E); 6143 } 6144 6145 // Forget that the initializers are sequenced. 6146 Region = Parent; 6147 for (unsigned I = 0; I < Elts.size(); ++I) 6148 Tree.merge(Elts[I]); 6149 } 6150}; 6151} 6152 6153void Sema::CheckUnsequencedOperations(Expr *E) { 6154 SmallVector<Expr *, 8> WorkList; 6155 WorkList.push_back(E); 6156 while (!WorkList.empty()) { 6157 Expr *Item = WorkList.pop_back_val(); 6158 SequenceChecker(*this, Item, WorkList); 6159 } 6160} 6161 6162void Sema::CheckCompletedExpr(Expr *E, SourceLocation CheckLoc, 6163 bool IsConstexpr) { 6164 CheckImplicitConversions(E, CheckLoc); 6165 CheckUnsequencedOperations(E); 6166 if (!IsConstexpr && !E->isValueDependent()) 6167 CheckForIntOverflow(E); 6168} 6169 6170void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 6171 FieldDecl *BitField, 6172 Expr *Init) { 6173 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 6174} 6175 6176/// CheckParmsForFunctionDef - Check that the parameters of the given 6177/// function are appropriate for the definition of a function. This 6178/// takes care of any checks that cannot be performed on the 6179/// declaration itself, e.g., that the types of each of the function 6180/// parameters are complete. 6181bool Sema::CheckParmsForFunctionDef(ParmVarDecl *const *P, 6182 ParmVarDecl *const *PEnd, 6183 bool CheckParameterNames) { 6184 bool HasInvalidParm = false; 6185 for (; P != PEnd; ++P) { 6186 ParmVarDecl *Param = *P; 6187 6188 // C99 6.7.5.3p4: the parameters in a parameter type list in a 6189 // function declarator that is part of a function definition of 6190 // that function shall not have incomplete type. 6191 // 6192 // This is also C++ [dcl.fct]p6. 6193 if (!Param->isInvalidDecl() && 6194 RequireCompleteType(Param->getLocation(), Param->getType(), 6195 diag::err_typecheck_decl_incomplete_type)) { 6196 Param->setInvalidDecl(); 6197 HasInvalidParm = true; 6198 } 6199 6200 // C99 6.9.1p5: If the declarator includes a parameter type list, the 6201 // declaration of each parameter shall include an identifier. 6202 if (CheckParameterNames && 6203 Param->getIdentifier() == 0 && 6204 !Param->isImplicit() && 6205 !getLangOpts().CPlusPlus) 6206 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 6207 6208 // C99 6.7.5.3p12: 6209 // If the function declarator is not part of a definition of that 6210 // function, parameters may have incomplete type and may use the [*] 6211 // notation in their sequences of declarator specifiers to specify 6212 // variable length array types. 6213 QualType PType = Param->getOriginalType(); 6214 while (const ArrayType *AT = Context.getAsArrayType(PType)) { 6215 if (AT->getSizeModifier() == ArrayType::Star) { 6216 // FIXME: This diagnostic should point the '[*]' if source-location 6217 // information is added for it. 6218 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 6219 break; 6220 } 6221 PType= AT->getElementType(); 6222 } 6223 6224 // MSVC destroys objects passed by value in the callee. Therefore a 6225 // function definition which takes such a parameter must be able to call the 6226 // object's destructor. 6227 if (getLangOpts().CPlusPlus && 6228 Context.getTargetInfo().getCXXABI().isArgumentDestroyedByCallee()) { 6229 if (const RecordType *RT = Param->getType()->getAs<RecordType>()) 6230 FinalizeVarWithDestructor(Param, RT); 6231 } 6232 } 6233 6234 return HasInvalidParm; 6235} 6236 6237/// CheckCastAlign - Implements -Wcast-align, which warns when a 6238/// pointer cast increases the alignment requirements. 6239void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 6240 // This is actually a lot of work to potentially be doing on every 6241 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 6242 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 6243 TRange.getBegin()) 6244 == DiagnosticsEngine::Ignored) 6245 return; 6246 6247 // Ignore dependent types. 6248 if (T->isDependentType() || Op->getType()->isDependentType()) 6249 return; 6250 6251 // Require that the destination be a pointer type. 6252 const PointerType *DestPtr = T->getAs<PointerType>(); 6253 if (!DestPtr) return; 6254 6255 // If the destination has alignment 1, we're done. 6256 QualType DestPointee = DestPtr->getPointeeType(); 6257 if (DestPointee->isIncompleteType()) return; 6258 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 6259 if (DestAlign.isOne()) return; 6260 6261 // Require that the source be a pointer type. 6262 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 6263 if (!SrcPtr) return; 6264 QualType SrcPointee = SrcPtr->getPointeeType(); 6265 6266 // Whitelist casts from cv void*. We already implicitly 6267 // whitelisted casts to cv void*, since they have alignment 1. 6268 // Also whitelist casts involving incomplete types, which implicitly 6269 // includes 'void'. 6270 if (SrcPointee->isIncompleteType()) return; 6271 6272 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 6273 if (SrcAlign >= DestAlign) return; 6274 6275 Diag(TRange.getBegin(), diag::warn_cast_align) 6276 << Op->getType() << T 6277 << static_cast<unsigned>(SrcAlign.getQuantity()) 6278 << static_cast<unsigned>(DestAlign.getQuantity()) 6279 << TRange << Op->getSourceRange(); 6280} 6281 6282static const Type* getElementType(const Expr *BaseExpr) { 6283 const Type* EltType = BaseExpr->getType().getTypePtr(); 6284 if (EltType->isAnyPointerType()) 6285 return EltType->getPointeeType().getTypePtr(); 6286 else if (EltType->isArrayType()) 6287 return EltType->getBaseElementTypeUnsafe(); 6288 return EltType; 6289} 6290 6291/// \brief Check whether this array fits the idiom of a size-one tail padded 6292/// array member of a struct. 6293/// 6294/// We avoid emitting out-of-bounds access warnings for such arrays as they are 6295/// commonly used to emulate flexible arrays in C89 code. 6296static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 6297 const NamedDecl *ND) { 6298 if (Size != 1 || !ND) return false; 6299 6300 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 6301 if (!FD) return false; 6302 6303 // Don't consider sizes resulting from macro expansions or template argument 6304 // substitution to form C89 tail-padded arrays. 6305 6306 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 6307 while (TInfo) { 6308 TypeLoc TL = TInfo->getTypeLoc(); 6309 // Look through typedefs. 6310 if (TypedefTypeLoc TTL = TL.getAs<TypedefTypeLoc>()) { 6311 const TypedefNameDecl *TDL = TTL.getTypedefNameDecl(); 6312 TInfo = TDL->getTypeSourceInfo(); 6313 continue; 6314 } 6315 if (ConstantArrayTypeLoc CTL = TL.getAs<ConstantArrayTypeLoc>()) { 6316 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 6317 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 6318 return false; 6319 } 6320 break; 6321 } 6322 6323 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 6324 if (!RD) return false; 6325 if (RD->isUnion()) return false; 6326 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 6327 if (!CRD->isStandardLayout()) return false; 6328 } 6329 6330 // See if this is the last field decl in the record. 6331 const Decl *D = FD; 6332 while ((D = D->getNextDeclInContext())) 6333 if (isa<FieldDecl>(D)) 6334 return false; 6335 return true; 6336} 6337 6338void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 6339 const ArraySubscriptExpr *ASE, 6340 bool AllowOnePastEnd, bool IndexNegated) { 6341 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 6342 if (IndexExpr->isValueDependent()) 6343 return; 6344 6345 const Type *EffectiveType = getElementType(BaseExpr); 6346 BaseExpr = BaseExpr->IgnoreParenCasts(); 6347 const ConstantArrayType *ArrayTy = 6348 Context.getAsConstantArrayType(BaseExpr->getType()); 6349 if (!ArrayTy) 6350 return; 6351 6352 llvm::APSInt index; 6353 if (!IndexExpr->EvaluateAsInt(index, Context)) 6354 return; 6355 if (IndexNegated) 6356 index = -index; 6357 6358 const NamedDecl *ND = NULL; 6359 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 6360 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 6361 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 6362 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 6363 6364 if (index.isUnsigned() || !index.isNegative()) { 6365 llvm::APInt size = ArrayTy->getSize(); 6366 if (!size.isStrictlyPositive()) 6367 return; 6368 6369 const Type* BaseType = getElementType(BaseExpr); 6370 if (BaseType != EffectiveType) { 6371 // Make sure we're comparing apples to apples when comparing index to size 6372 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 6373 uint64_t array_typesize = Context.getTypeSize(BaseType); 6374 // Handle ptrarith_typesize being zero, such as when casting to void* 6375 if (!ptrarith_typesize) ptrarith_typesize = 1; 6376 if (ptrarith_typesize != array_typesize) { 6377 // There's a cast to a different size type involved 6378 uint64_t ratio = array_typesize / ptrarith_typesize; 6379 // TODO: Be smarter about handling cases where array_typesize is not a 6380 // multiple of ptrarith_typesize 6381 if (ptrarith_typesize * ratio == array_typesize) 6382 size *= llvm::APInt(size.getBitWidth(), ratio); 6383 } 6384 } 6385 6386 if (size.getBitWidth() > index.getBitWidth()) 6387 index = index.zext(size.getBitWidth()); 6388 else if (size.getBitWidth() < index.getBitWidth()) 6389 size = size.zext(index.getBitWidth()); 6390 6391 // For array subscripting the index must be less than size, but for pointer 6392 // arithmetic also allow the index (offset) to be equal to size since 6393 // computing the next address after the end of the array is legal and 6394 // commonly done e.g. in C++ iterators and range-based for loops. 6395 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 6396 return; 6397 6398 // Also don't warn for arrays of size 1 which are members of some 6399 // structure. These are often used to approximate flexible arrays in C89 6400 // code. 6401 if (IsTailPaddedMemberArray(*this, size, ND)) 6402 return; 6403 6404 // Suppress the warning if the subscript expression (as identified by the 6405 // ']' location) and the index expression are both from macro expansions 6406 // within a system header. 6407 if (ASE) { 6408 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 6409 ASE->getRBracketLoc()); 6410 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 6411 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 6412 IndexExpr->getLocStart()); 6413 if (SourceMgr.isWrittenInSameFile(RBracketLoc, IndexLoc)) 6414 return; 6415 } 6416 } 6417 6418 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 6419 if (ASE) 6420 DiagID = diag::warn_array_index_exceeds_bounds; 6421 6422 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 6423 PDiag(DiagID) << index.toString(10, true) 6424 << size.toString(10, true) 6425 << (unsigned)size.getLimitedValue(~0U) 6426 << IndexExpr->getSourceRange()); 6427 } else { 6428 unsigned DiagID = diag::warn_array_index_precedes_bounds; 6429 if (!ASE) { 6430 DiagID = diag::warn_ptr_arith_precedes_bounds; 6431 if (index.isNegative()) index = -index; 6432 } 6433 6434 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 6435 PDiag(DiagID) << index.toString(10, true) 6436 << IndexExpr->getSourceRange()); 6437 } 6438 6439 if (!ND) { 6440 // Try harder to find a NamedDecl to point at in the note. 6441 while (const ArraySubscriptExpr *ASE = 6442 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 6443 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 6444 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 6445 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 6446 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 6447 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 6448 } 6449 6450 if (ND) 6451 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 6452 PDiag(diag::note_array_index_out_of_bounds) 6453 << ND->getDeclName()); 6454} 6455 6456void Sema::CheckArrayAccess(const Expr *expr) { 6457 int AllowOnePastEnd = 0; 6458 while (expr) { 6459 expr = expr->IgnoreParenImpCasts(); 6460 switch (expr->getStmtClass()) { 6461 case Stmt::ArraySubscriptExprClass: { 6462 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 6463 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 6464 AllowOnePastEnd > 0); 6465 return; 6466 } 6467 case Stmt::UnaryOperatorClass: { 6468 // Only unwrap the * and & unary operators 6469 const UnaryOperator *UO = cast<UnaryOperator>(expr); 6470 expr = UO->getSubExpr(); 6471 switch (UO->getOpcode()) { 6472 case UO_AddrOf: 6473 AllowOnePastEnd++; 6474 break; 6475 case UO_Deref: 6476 AllowOnePastEnd--; 6477 break; 6478 default: 6479 return; 6480 } 6481 break; 6482 } 6483 case Stmt::ConditionalOperatorClass: { 6484 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 6485 if (const Expr *lhs = cond->getLHS()) 6486 CheckArrayAccess(lhs); 6487 if (const Expr *rhs = cond->getRHS()) 6488 CheckArrayAccess(rhs); 6489 return; 6490 } 6491 default: 6492 return; 6493 } 6494 } 6495} 6496 6497//===--- CHECK: Objective-C retain cycles ----------------------------------// 6498 6499namespace { 6500 struct RetainCycleOwner { 6501 RetainCycleOwner() : Variable(0), Indirect(false) {} 6502 VarDecl *Variable; 6503 SourceRange Range; 6504 SourceLocation Loc; 6505 bool Indirect; 6506 6507 void setLocsFrom(Expr *e) { 6508 Loc = e->getExprLoc(); 6509 Range = e->getSourceRange(); 6510 } 6511 }; 6512} 6513 6514/// Consider whether capturing the given variable can possibly lead to 6515/// a retain cycle. 6516static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 6517 // In ARC, it's captured strongly iff the variable has __strong 6518 // lifetime. In MRR, it's captured strongly if the variable is 6519 // __block and has an appropriate type. 6520 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 6521 return false; 6522 6523 owner.Variable = var; 6524 if (ref) 6525 owner.setLocsFrom(ref); 6526 return true; 6527} 6528 6529static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 6530 while (true) { 6531 e = e->IgnoreParens(); 6532 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 6533 switch (cast->getCastKind()) { 6534 case CK_BitCast: 6535 case CK_LValueBitCast: 6536 case CK_LValueToRValue: 6537 case CK_ARCReclaimReturnedObject: 6538 e = cast->getSubExpr(); 6539 continue; 6540 6541 default: 6542 return false; 6543 } 6544 } 6545 6546 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 6547 ObjCIvarDecl *ivar = ref->getDecl(); 6548 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 6549 return false; 6550 6551 // Try to find a retain cycle in the base. 6552 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 6553 return false; 6554 6555 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 6556 owner.Indirect = true; 6557 return true; 6558 } 6559 6560 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 6561 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 6562 if (!var) return false; 6563 return considerVariable(var, ref, owner); 6564 } 6565 6566 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 6567 if (member->isArrow()) return false; 6568 6569 // Don't count this as an indirect ownership. 6570 e = member->getBase(); 6571 continue; 6572 } 6573 6574 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 6575 // Only pay attention to pseudo-objects on property references. 6576 ObjCPropertyRefExpr *pre 6577 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 6578 ->IgnoreParens()); 6579 if (!pre) return false; 6580 if (pre->isImplicitProperty()) return false; 6581 ObjCPropertyDecl *property = pre->getExplicitProperty(); 6582 if (!property->isRetaining() && 6583 !(property->getPropertyIvarDecl() && 6584 property->getPropertyIvarDecl()->getType() 6585 .getObjCLifetime() == Qualifiers::OCL_Strong)) 6586 return false; 6587 6588 owner.Indirect = true; 6589 if (pre->isSuperReceiver()) { 6590 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 6591 if (!owner.Variable) 6592 return false; 6593 owner.Loc = pre->getLocation(); 6594 owner.Range = pre->getSourceRange(); 6595 return true; 6596 } 6597 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 6598 ->getSourceExpr()); 6599 continue; 6600 } 6601 6602 // Array ivars? 6603 6604 return false; 6605 } 6606} 6607 6608namespace { 6609 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 6610 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 6611 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 6612 Variable(variable), Capturer(0) {} 6613 6614 VarDecl *Variable; 6615 Expr *Capturer; 6616 6617 void VisitDeclRefExpr(DeclRefExpr *ref) { 6618 if (ref->getDecl() == Variable && !Capturer) 6619 Capturer = ref; 6620 } 6621 6622 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 6623 if (Capturer) return; 6624 Visit(ref->getBase()); 6625 if (Capturer && ref->isFreeIvar()) 6626 Capturer = ref; 6627 } 6628 6629 void VisitBlockExpr(BlockExpr *block) { 6630 // Look inside nested blocks 6631 if (block->getBlockDecl()->capturesVariable(Variable)) 6632 Visit(block->getBlockDecl()->getBody()); 6633 } 6634 6635 void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) { 6636 if (Capturer) return; 6637 if (OVE->getSourceExpr()) 6638 Visit(OVE->getSourceExpr()); 6639 } 6640 }; 6641} 6642 6643/// Check whether the given argument is a block which captures a 6644/// variable. 6645static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 6646 assert(owner.Variable && owner.Loc.isValid()); 6647 6648 e = e->IgnoreParenCasts(); 6649 6650 // Look through [^{...} copy] and Block_copy(^{...}). 6651 if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) { 6652 Selector Cmd = ME->getSelector(); 6653 if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") { 6654 e = ME->getInstanceReceiver(); 6655 if (!e) 6656 return 0; 6657 e = e->IgnoreParenCasts(); 6658 } 6659 } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) { 6660 if (CE->getNumArgs() == 1) { 6661 FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl()); 6662 if (Fn) { 6663 const IdentifierInfo *FnI = Fn->getIdentifier(); 6664 if (FnI && FnI->isStr("_Block_copy")) { 6665 e = CE->getArg(0)->IgnoreParenCasts(); 6666 } 6667 } 6668 } 6669 } 6670 6671 BlockExpr *block = dyn_cast<BlockExpr>(e); 6672 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 6673 return 0; 6674 6675 FindCaptureVisitor visitor(S.Context, owner.Variable); 6676 visitor.Visit(block->getBlockDecl()->getBody()); 6677 return visitor.Capturer; 6678} 6679 6680static void diagnoseRetainCycle(Sema &S, Expr *capturer, 6681 RetainCycleOwner &owner) { 6682 assert(capturer); 6683 assert(owner.Variable && owner.Loc.isValid()); 6684 6685 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 6686 << owner.Variable << capturer->getSourceRange(); 6687 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 6688 << owner.Indirect << owner.Range; 6689} 6690 6691/// Check for a keyword selector that starts with the word 'add' or 6692/// 'set'. 6693static bool isSetterLikeSelector(Selector sel) { 6694 if (sel.isUnarySelector()) return false; 6695 6696 StringRef str = sel.getNameForSlot(0); 6697 while (!str.empty() && str.front() == '_') str = str.substr(1); 6698 if (str.startswith("set")) 6699 str = str.substr(3); 6700 else if (str.startswith("add")) { 6701 // Specially whitelist 'addOperationWithBlock:'. 6702 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 6703 return false; 6704 str = str.substr(3); 6705 } 6706 else 6707 return false; 6708 6709 if (str.empty()) return true; 6710 return !isLowercase(str.front()); 6711} 6712 6713/// Check a message send to see if it's likely to cause a retain cycle. 6714void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 6715 // Only check instance methods whose selector looks like a setter. 6716 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 6717 return; 6718 6719 // Try to find a variable that the receiver is strongly owned by. 6720 RetainCycleOwner owner; 6721 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 6722 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 6723 return; 6724 } else { 6725 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 6726 owner.Variable = getCurMethodDecl()->getSelfDecl(); 6727 owner.Loc = msg->getSuperLoc(); 6728 owner.Range = msg->getSuperLoc(); 6729 } 6730 6731 // Check whether the receiver is captured by any of the arguments. 6732 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 6733 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 6734 return diagnoseRetainCycle(*this, capturer, owner); 6735} 6736 6737/// Check a property assign to see if it's likely to cause a retain cycle. 6738void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 6739 RetainCycleOwner owner; 6740 if (!findRetainCycleOwner(*this, receiver, owner)) 6741 return; 6742 6743 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 6744 diagnoseRetainCycle(*this, capturer, owner); 6745} 6746 6747void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) { 6748 RetainCycleOwner Owner; 6749 if (!considerVariable(Var, /*DeclRefExpr=*/0, Owner)) 6750 return; 6751 6752 // Because we don't have an expression for the variable, we have to set the 6753 // location explicitly here. 6754 Owner.Loc = Var->getLocation(); 6755 Owner.Range = Var->getSourceRange(); 6756 6757 if (Expr *Capturer = findCapturingExpr(*this, Init, Owner)) 6758 diagnoseRetainCycle(*this, Capturer, Owner); 6759} 6760 6761static bool checkUnsafeAssignLiteral(Sema &S, SourceLocation Loc, 6762 Expr *RHS, bool isProperty) { 6763 // Check if RHS is an Objective-C object literal, which also can get 6764 // immediately zapped in a weak reference. Note that we explicitly 6765 // allow ObjCStringLiterals, since those are designed to never really die. 6766 RHS = RHS->IgnoreParenImpCasts(); 6767 6768 // This enum needs to match with the 'select' in 6769 // warn_objc_arc_literal_assign (off-by-1). 6770 Sema::ObjCLiteralKind Kind = S.CheckLiteralKind(RHS); 6771 if (Kind == Sema::LK_String || Kind == Sema::LK_None) 6772 return false; 6773 6774 S.Diag(Loc, diag::warn_arc_literal_assign) 6775 << (unsigned) Kind 6776 << (isProperty ? 0 : 1) 6777 << RHS->getSourceRange(); 6778 6779 return true; 6780} 6781 6782static bool checkUnsafeAssignObject(Sema &S, SourceLocation Loc, 6783 Qualifiers::ObjCLifetime LT, 6784 Expr *RHS, bool isProperty) { 6785 // Strip off any implicit cast added to get to the one ARC-specific. 6786 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6787 if (cast->getCastKind() == CK_ARCConsumeObject) { 6788 S.Diag(Loc, diag::warn_arc_retained_assign) 6789 << (LT == Qualifiers::OCL_ExplicitNone) 6790 << (isProperty ? 0 : 1) 6791 << RHS->getSourceRange(); 6792 return true; 6793 } 6794 RHS = cast->getSubExpr(); 6795 } 6796 6797 if (LT == Qualifiers::OCL_Weak && 6798 checkUnsafeAssignLiteral(S, Loc, RHS, isProperty)) 6799 return true; 6800 6801 return false; 6802} 6803 6804bool Sema::checkUnsafeAssigns(SourceLocation Loc, 6805 QualType LHS, Expr *RHS) { 6806 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 6807 6808 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 6809 return false; 6810 6811 if (checkUnsafeAssignObject(*this, Loc, LT, RHS, false)) 6812 return true; 6813 6814 return false; 6815} 6816 6817void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 6818 Expr *LHS, Expr *RHS) { 6819 QualType LHSType; 6820 // PropertyRef on LHS type need be directly obtained from 6821 // its declaration as it has a PsuedoType. 6822 ObjCPropertyRefExpr *PRE 6823 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 6824 if (PRE && !PRE->isImplicitProperty()) { 6825 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6826 if (PD) 6827 LHSType = PD->getType(); 6828 } 6829 6830 if (LHSType.isNull()) 6831 LHSType = LHS->getType(); 6832 6833 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 6834 6835 if (LT == Qualifiers::OCL_Weak) { 6836 DiagnosticsEngine::Level Level = 6837 Diags.getDiagnosticLevel(diag::warn_arc_repeated_use_of_weak, Loc); 6838 if (Level != DiagnosticsEngine::Ignored) 6839 getCurFunction()->markSafeWeakUse(LHS); 6840 } 6841 6842 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 6843 return; 6844 6845 // FIXME. Check for other life times. 6846 if (LT != Qualifiers::OCL_None) 6847 return; 6848 6849 if (PRE) { 6850 if (PRE->isImplicitProperty()) 6851 return; 6852 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6853 if (!PD) 6854 return; 6855 6856 unsigned Attributes = PD->getPropertyAttributes(); 6857 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 6858 // when 'assign' attribute was not explicitly specified 6859 // by user, ignore it and rely on property type itself 6860 // for lifetime info. 6861 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 6862 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 6863 LHSType->isObjCRetainableType()) 6864 return; 6865 6866 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6867 if (cast->getCastKind() == CK_ARCConsumeObject) { 6868 Diag(Loc, diag::warn_arc_retained_property_assign) 6869 << RHS->getSourceRange(); 6870 return; 6871 } 6872 RHS = cast->getSubExpr(); 6873 } 6874 } 6875 else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) { 6876 if (checkUnsafeAssignObject(*this, Loc, Qualifiers::OCL_Weak, RHS, true)) 6877 return; 6878 } 6879 } 6880} 6881 6882//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 6883 6884namespace { 6885bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 6886 SourceLocation StmtLoc, 6887 const NullStmt *Body) { 6888 // Do not warn if the body is a macro that expands to nothing, e.g: 6889 // 6890 // #define CALL(x) 6891 // if (condition) 6892 // CALL(0); 6893 // 6894 if (Body->hasLeadingEmptyMacro()) 6895 return false; 6896 6897 // Get line numbers of statement and body. 6898 bool StmtLineInvalid; 6899 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 6900 &StmtLineInvalid); 6901 if (StmtLineInvalid) 6902 return false; 6903 6904 bool BodyLineInvalid; 6905 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 6906 &BodyLineInvalid); 6907 if (BodyLineInvalid) 6908 return false; 6909 6910 // Warn if null statement and body are on the same line. 6911 if (StmtLine != BodyLine) 6912 return false; 6913 6914 return true; 6915} 6916} // Unnamed namespace 6917 6918void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 6919 const Stmt *Body, 6920 unsigned DiagID) { 6921 // Since this is a syntactic check, don't emit diagnostic for template 6922 // instantiations, this just adds noise. 6923 if (CurrentInstantiationScope) 6924 return; 6925 6926 // The body should be a null statement. 6927 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6928 if (!NBody) 6929 return; 6930 6931 // Do the usual checks. 6932 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6933 return; 6934 6935 Diag(NBody->getSemiLoc(), DiagID); 6936 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6937} 6938 6939void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 6940 const Stmt *PossibleBody) { 6941 assert(!CurrentInstantiationScope); // Ensured by caller 6942 6943 SourceLocation StmtLoc; 6944 const Stmt *Body; 6945 unsigned DiagID; 6946 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 6947 StmtLoc = FS->getRParenLoc(); 6948 Body = FS->getBody(); 6949 DiagID = diag::warn_empty_for_body; 6950 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 6951 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 6952 Body = WS->getBody(); 6953 DiagID = diag::warn_empty_while_body; 6954 } else 6955 return; // Neither `for' nor `while'. 6956 6957 // The body should be a null statement. 6958 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6959 if (!NBody) 6960 return; 6961 6962 // Skip expensive checks if diagnostic is disabled. 6963 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 6964 DiagnosticsEngine::Ignored) 6965 return; 6966 6967 // Do the usual checks. 6968 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6969 return; 6970 6971 // `for(...);' and `while(...);' are popular idioms, so in order to keep 6972 // noise level low, emit diagnostics only if for/while is followed by a 6973 // CompoundStmt, e.g.: 6974 // for (int i = 0; i < n; i++); 6975 // { 6976 // a(i); 6977 // } 6978 // or if for/while is followed by a statement with more indentation 6979 // than for/while itself: 6980 // for (int i = 0; i < n; i++); 6981 // a(i); 6982 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 6983 if (!ProbableTypo) { 6984 bool BodyColInvalid; 6985 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 6986 PossibleBody->getLocStart(), 6987 &BodyColInvalid); 6988 if (BodyColInvalid) 6989 return; 6990 6991 bool StmtColInvalid; 6992 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 6993 S->getLocStart(), 6994 &StmtColInvalid); 6995 if (StmtColInvalid) 6996 return; 6997 6998 if (BodyCol > StmtCol) 6999 ProbableTypo = true; 7000 } 7001 7002 if (ProbableTypo) { 7003 Diag(NBody->getSemiLoc(), DiagID); 7004 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 7005 } 7006} 7007 7008//===--- Layout compatibility ----------------------------------------------// 7009 7010namespace { 7011 7012bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2); 7013 7014/// \brief Check if two enumeration types are layout-compatible. 7015bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) { 7016 // C++11 [dcl.enum] p8: 7017 // Two enumeration types are layout-compatible if they have the same 7018 // underlying type. 7019 return ED1->isComplete() && ED2->isComplete() && 7020 C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType()); 7021} 7022 7023/// \brief Check if two fields are layout-compatible. 7024bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) { 7025 if (!isLayoutCompatible(C, Field1->getType(), Field2->getType())) 7026 return false; 7027 7028 if (Field1->isBitField() != Field2->isBitField()) 7029 return false; 7030 7031 if (Field1->isBitField()) { 7032 // Make sure that the bit-fields are the same length. 7033 unsigned Bits1 = Field1->getBitWidthValue(C); 7034 unsigned Bits2 = Field2->getBitWidthValue(C); 7035 7036 if (Bits1 != Bits2) 7037 return false; 7038 } 7039 7040 return true; 7041} 7042 7043/// \brief Check if two standard-layout structs are layout-compatible. 7044/// (C++11 [class.mem] p17) 7045bool isLayoutCompatibleStruct(ASTContext &C, 7046 RecordDecl *RD1, 7047 RecordDecl *RD2) { 7048 // If both records are C++ classes, check that base classes match. 7049 if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) { 7050 // If one of records is a CXXRecordDecl we are in C++ mode, 7051 // thus the other one is a CXXRecordDecl, too. 7052 const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2); 7053 // Check number of base classes. 7054 if (D1CXX->getNumBases() != D2CXX->getNumBases()) 7055 return false; 7056 7057 // Check the base classes. 7058 for (CXXRecordDecl::base_class_const_iterator 7059 Base1 = D1CXX->bases_begin(), 7060 BaseEnd1 = D1CXX->bases_end(), 7061 Base2 = D2CXX->bases_begin(); 7062 Base1 != BaseEnd1; 7063 ++Base1, ++Base2) { 7064 if (!isLayoutCompatible(C, Base1->getType(), Base2->getType())) 7065 return false; 7066 } 7067 } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) { 7068 // If only RD2 is a C++ class, it should have zero base classes. 7069 if (D2CXX->getNumBases() > 0) 7070 return false; 7071 } 7072 7073 // Check the fields. 7074 RecordDecl::field_iterator Field2 = RD2->field_begin(), 7075 Field2End = RD2->field_end(), 7076 Field1 = RD1->field_begin(), 7077 Field1End = RD1->field_end(); 7078 for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) { 7079 if (!isLayoutCompatible(C, *Field1, *Field2)) 7080 return false; 7081 } 7082 if (Field1 != Field1End || Field2 != Field2End) 7083 return false; 7084 7085 return true; 7086} 7087 7088/// \brief Check if two standard-layout unions are layout-compatible. 7089/// (C++11 [class.mem] p18) 7090bool isLayoutCompatibleUnion(ASTContext &C, 7091 RecordDecl *RD1, 7092 RecordDecl *RD2) { 7093 llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields; 7094 for (RecordDecl::field_iterator Field2 = RD2->field_begin(), 7095 Field2End = RD2->field_end(); 7096 Field2 != Field2End; ++Field2) { 7097 UnmatchedFields.insert(*Field2); 7098 } 7099 7100 for (RecordDecl::field_iterator Field1 = RD1->field_begin(), 7101 Field1End = RD1->field_end(); 7102 Field1 != Field1End; ++Field1) { 7103 llvm::SmallPtrSet<FieldDecl *, 8>::iterator 7104 I = UnmatchedFields.begin(), 7105 E = UnmatchedFields.end(); 7106 7107 for ( ; I != E; ++I) { 7108 if (isLayoutCompatible(C, *Field1, *I)) { 7109 bool Result = UnmatchedFields.erase(*I); 7110 (void) Result; 7111 assert(Result); 7112 break; 7113 } 7114 } 7115 if (I == E) 7116 return false; 7117 } 7118 7119 return UnmatchedFields.empty(); 7120} 7121 7122bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) { 7123 if (RD1->isUnion() != RD2->isUnion()) 7124 return false; 7125 7126 if (RD1->isUnion()) 7127 return isLayoutCompatibleUnion(C, RD1, RD2); 7128 else 7129 return isLayoutCompatibleStruct(C, RD1, RD2); 7130} 7131 7132/// \brief Check if two types are layout-compatible in C++11 sense. 7133bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) { 7134 if (T1.isNull() || T2.isNull()) 7135 return false; 7136 7137 // C++11 [basic.types] p11: 7138 // If two types T1 and T2 are the same type, then T1 and T2 are 7139 // layout-compatible types. 7140 if (C.hasSameType(T1, T2)) 7141 return true; 7142 7143 T1 = T1.getCanonicalType().getUnqualifiedType(); 7144 T2 = T2.getCanonicalType().getUnqualifiedType(); 7145 7146 const Type::TypeClass TC1 = T1->getTypeClass(); 7147 const Type::TypeClass TC2 = T2->getTypeClass(); 7148 7149 if (TC1 != TC2) 7150 return false; 7151 7152 if (TC1 == Type::Enum) { 7153 return isLayoutCompatible(C, 7154 cast<EnumType>(T1)->getDecl(), 7155 cast<EnumType>(T2)->getDecl()); 7156 } else if (TC1 == Type::Record) { 7157 if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType()) 7158 return false; 7159 7160 return isLayoutCompatible(C, 7161 cast<RecordType>(T1)->getDecl(), 7162 cast<RecordType>(T2)->getDecl()); 7163 } 7164 7165 return false; 7166} 7167} 7168 7169//===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----// 7170 7171namespace { 7172/// \brief Given a type tag expression find the type tag itself. 7173/// 7174/// \param TypeExpr Type tag expression, as it appears in user's code. 7175/// 7176/// \param VD Declaration of an identifier that appears in a type tag. 7177/// 7178/// \param MagicValue Type tag magic value. 7179bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx, 7180 const ValueDecl **VD, uint64_t *MagicValue) { 7181 while(true) { 7182 if (!TypeExpr) 7183 return false; 7184 7185 TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts(); 7186 7187 switch (TypeExpr->getStmtClass()) { 7188 case Stmt::UnaryOperatorClass: { 7189 const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr); 7190 if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) { 7191 TypeExpr = UO->getSubExpr(); 7192 continue; 7193 } 7194 return false; 7195 } 7196 7197 case Stmt::DeclRefExprClass: { 7198 const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr); 7199 *VD = DRE->getDecl(); 7200 return true; 7201 } 7202 7203 case Stmt::IntegerLiteralClass: { 7204 const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr); 7205 llvm::APInt MagicValueAPInt = IL->getValue(); 7206 if (MagicValueAPInt.getActiveBits() <= 64) { 7207 *MagicValue = MagicValueAPInt.getZExtValue(); 7208 return true; 7209 } else 7210 return false; 7211 } 7212 7213 case Stmt::BinaryConditionalOperatorClass: 7214 case Stmt::ConditionalOperatorClass: { 7215 const AbstractConditionalOperator *ACO = 7216 cast<AbstractConditionalOperator>(TypeExpr); 7217 bool Result; 7218 if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) { 7219 if (Result) 7220 TypeExpr = ACO->getTrueExpr(); 7221 else 7222 TypeExpr = ACO->getFalseExpr(); 7223 continue; 7224 } 7225 return false; 7226 } 7227 7228 case Stmt::BinaryOperatorClass: { 7229 const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr); 7230 if (BO->getOpcode() == BO_Comma) { 7231 TypeExpr = BO->getRHS(); 7232 continue; 7233 } 7234 return false; 7235 } 7236 7237 default: 7238 return false; 7239 } 7240 } 7241} 7242 7243/// \brief Retrieve the C type corresponding to type tag TypeExpr. 7244/// 7245/// \param TypeExpr Expression that specifies a type tag. 7246/// 7247/// \param MagicValues Registered magic values. 7248/// 7249/// \param FoundWrongKind Set to true if a type tag was found, but of a wrong 7250/// kind. 7251/// 7252/// \param TypeInfo Information about the corresponding C type. 7253/// 7254/// \returns true if the corresponding C type was found. 7255bool GetMatchingCType( 7256 const IdentifierInfo *ArgumentKind, 7257 const Expr *TypeExpr, const ASTContext &Ctx, 7258 const llvm::DenseMap<Sema::TypeTagMagicValue, 7259 Sema::TypeTagData> *MagicValues, 7260 bool &FoundWrongKind, 7261 Sema::TypeTagData &TypeInfo) { 7262 FoundWrongKind = false; 7263 7264 // Variable declaration that has type_tag_for_datatype attribute. 7265 const ValueDecl *VD = NULL; 7266 7267 uint64_t MagicValue; 7268 7269 if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue)) 7270 return false; 7271 7272 if (VD) { 7273 for (specific_attr_iterator<TypeTagForDatatypeAttr> 7274 I = VD->specific_attr_begin<TypeTagForDatatypeAttr>(), 7275 E = VD->specific_attr_end<TypeTagForDatatypeAttr>(); 7276 I != E; ++I) { 7277 if (I->getArgumentKind() != ArgumentKind) { 7278 FoundWrongKind = true; 7279 return false; 7280 } 7281 TypeInfo.Type = I->getMatchingCType(); 7282 TypeInfo.LayoutCompatible = I->getLayoutCompatible(); 7283 TypeInfo.MustBeNull = I->getMustBeNull(); 7284 return true; 7285 } 7286 return false; 7287 } 7288 7289 if (!MagicValues) 7290 return false; 7291 7292 llvm::DenseMap<Sema::TypeTagMagicValue, 7293 Sema::TypeTagData>::const_iterator I = 7294 MagicValues->find(std::make_pair(ArgumentKind, MagicValue)); 7295 if (I == MagicValues->end()) 7296 return false; 7297 7298 TypeInfo = I->second; 7299 return true; 7300} 7301} // unnamed namespace 7302 7303void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind, 7304 uint64_t MagicValue, QualType Type, 7305 bool LayoutCompatible, 7306 bool MustBeNull) { 7307 if (!TypeTagForDatatypeMagicValues) 7308 TypeTagForDatatypeMagicValues.reset( 7309 new llvm::DenseMap<TypeTagMagicValue, TypeTagData>); 7310 7311 TypeTagMagicValue Magic(ArgumentKind, MagicValue); 7312 (*TypeTagForDatatypeMagicValues)[Magic] = 7313 TypeTagData(Type, LayoutCompatible, MustBeNull); 7314} 7315 7316namespace { 7317bool IsSameCharType(QualType T1, QualType T2) { 7318 const BuiltinType *BT1 = T1->getAs<BuiltinType>(); 7319 if (!BT1) 7320 return false; 7321 7322 const BuiltinType *BT2 = T2->getAs<BuiltinType>(); 7323 if (!BT2) 7324 return false; 7325 7326 BuiltinType::Kind T1Kind = BT1->getKind(); 7327 BuiltinType::Kind T2Kind = BT2->getKind(); 7328 7329 return (T1Kind == BuiltinType::SChar && T2Kind == BuiltinType::Char_S) || 7330 (T1Kind == BuiltinType::UChar && T2Kind == BuiltinType::Char_U) || 7331 (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) || 7332 (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar); 7333} 7334} // unnamed namespace 7335 7336void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr, 7337 const Expr * const *ExprArgs) { 7338 const IdentifierInfo *ArgumentKind = Attr->getArgumentKind(); 7339 bool IsPointerAttr = Attr->getIsPointer(); 7340 7341 const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()]; 7342 bool FoundWrongKind; 7343 TypeTagData TypeInfo; 7344 if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context, 7345 TypeTagForDatatypeMagicValues.get(), 7346 FoundWrongKind, TypeInfo)) { 7347 if (FoundWrongKind) 7348 Diag(TypeTagExpr->getExprLoc(), 7349 diag::warn_type_tag_for_datatype_wrong_kind) 7350 << TypeTagExpr->getSourceRange(); 7351 return; 7352 } 7353 7354 const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()]; 7355 if (IsPointerAttr) { 7356 // Skip implicit cast of pointer to `void *' (as a function argument). 7357 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr)) 7358 if (ICE->getType()->isVoidPointerType() && 7359 ICE->getCastKind() == CK_BitCast) 7360 ArgumentExpr = ICE->getSubExpr(); 7361 } 7362 QualType ArgumentType = ArgumentExpr->getType(); 7363 7364 // Passing a `void*' pointer shouldn't trigger a warning. 7365 if (IsPointerAttr && ArgumentType->isVoidPointerType()) 7366 return; 7367 7368 if (TypeInfo.MustBeNull) { 7369 // Type tag with matching void type requires a null pointer. 7370 if (!ArgumentExpr->isNullPointerConstant(Context, 7371 Expr::NPC_ValueDependentIsNotNull)) { 7372 Diag(ArgumentExpr->getExprLoc(), 7373 diag::warn_type_safety_null_pointer_required) 7374 << ArgumentKind->getName() 7375 << ArgumentExpr->getSourceRange() 7376 << TypeTagExpr->getSourceRange(); 7377 } 7378 return; 7379 } 7380 7381 QualType RequiredType = TypeInfo.Type; 7382 if (IsPointerAttr) 7383 RequiredType = Context.getPointerType(RequiredType); 7384 7385 bool mismatch = false; 7386 if (!TypeInfo.LayoutCompatible) { 7387 mismatch = !Context.hasSameType(ArgumentType, RequiredType); 7388 7389 // C++11 [basic.fundamental] p1: 7390 // Plain char, signed char, and unsigned char are three distinct types. 7391 // 7392 // But we treat plain `char' as equivalent to `signed char' or `unsigned 7393 // char' depending on the current char signedness mode. 7394 if (mismatch) 7395 if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(), 7396 RequiredType->getPointeeType())) || 7397 (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType))) 7398 mismatch = false; 7399 } else 7400 if (IsPointerAttr) 7401 mismatch = !isLayoutCompatible(Context, 7402 ArgumentType->getPointeeType(), 7403 RequiredType->getPointeeType()); 7404 else 7405 mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType); 7406 7407 if (mismatch) 7408 Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch) 7409 << ArgumentType << ArgumentKind->getName() 7410 << TypeInfo.LayoutCompatible << RequiredType 7411 << ArgumentExpr->getSourceRange() 7412 << TypeTagExpr->getSourceRange(); 7413} 7414