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/STLExtras.h" 36 #include "llvm/ADT/SmallBitVector.h" 37 #include "llvm/ADT/SmallString.h" 38 #include "llvm/Support/ConvertUTF.h" 39 #include "llvm/Support/raw_ostream.h" 40 #include <limits> 41 using namespace clang; 42 using namespace sema; 43 44 SourceLocation 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. 52 static 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. 72 static 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. 100 static 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 } 113 114 ExprResult 115 Sema::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 case Builtin::BI__va_start: 146 if (SemaBuiltinVAStart(TheCall)) 147 return ExprError(); 148 break; 149 case Builtin::BI__builtin_isgreater: 150 case Builtin::BI__builtin_isgreaterequal: 151 case Builtin::BI__builtin_isless: 152 case Builtin::BI__builtin_islessequal: 153 case Builtin::BI__builtin_islessgreater: 154 case Builtin::BI__builtin_isunordered: 155 if (SemaBuiltinUnorderedCompare(TheCall)) 156 return ExprError(); 157 break; 158 case Builtin::BI__builtin_fpclassify: 159 if (SemaBuiltinFPClassification(TheCall, 6)) 160 return ExprError(); 161 break; 162 case Builtin::BI__builtin_isfinite: 163 case Builtin::BI__builtin_isinf: 164 case Builtin::BI__builtin_isinf_sign: 165 case Builtin::BI__builtin_isnan: 166 case Builtin::BI__builtin_isnormal: 167 if (SemaBuiltinFPClassification(TheCall, 1)) 168 return ExprError(); 169 break; 170 case Builtin::BI__builtin_shufflevector: 171 return SemaBuiltinShuffleVector(TheCall); 172 // TheCall will be freed by the smart pointer here, but that's fine, since 173 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 174 case Builtin::BI__builtin_prefetch: 175 if (SemaBuiltinPrefetch(TheCall)) 176 return ExprError(); 177 break; 178 case Builtin::BI__builtin_object_size: 179 if (SemaBuiltinObjectSize(TheCall)) 180 return ExprError(); 181 break; 182 case Builtin::BI__builtin_longjmp: 183 if (SemaBuiltinLongjmp(TheCall)) 184 return ExprError(); 185 break; 186 187 case Builtin::BI__builtin_classify_type: 188 if (checkArgCount(*this, TheCall, 1)) return true; 189 TheCall->setType(Context.IntTy); 190 break; 191 case Builtin::BI__builtin_constant_p: 192 if (checkArgCount(*this, TheCall, 1)) return true; 193 TheCall->setType(Context.IntTy); 194 break; 195 case Builtin::BI__sync_fetch_and_add: 196 case Builtin::BI__sync_fetch_and_add_1: 197 case Builtin::BI__sync_fetch_and_add_2: 198 case Builtin::BI__sync_fetch_and_add_4: 199 case Builtin::BI__sync_fetch_and_add_8: 200 case Builtin::BI__sync_fetch_and_add_16: 201 case Builtin::BI__sync_fetch_and_sub: 202 case Builtin::BI__sync_fetch_and_sub_1: 203 case Builtin::BI__sync_fetch_and_sub_2: 204 case Builtin::BI__sync_fetch_and_sub_4: 205 case Builtin::BI__sync_fetch_and_sub_8: 206 case Builtin::BI__sync_fetch_and_sub_16: 207 case Builtin::BI__sync_fetch_and_or: 208 case Builtin::BI__sync_fetch_and_or_1: 209 case Builtin::BI__sync_fetch_and_or_2: 210 case Builtin::BI__sync_fetch_and_or_4: 211 case Builtin::BI__sync_fetch_and_or_8: 212 case Builtin::BI__sync_fetch_and_or_16: 213 case Builtin::BI__sync_fetch_and_and: 214 case Builtin::BI__sync_fetch_and_and_1: 215 case Builtin::BI__sync_fetch_and_and_2: 216 case Builtin::BI__sync_fetch_and_and_4: 217 case Builtin::BI__sync_fetch_and_and_8: 218 case Builtin::BI__sync_fetch_and_and_16: 219 case Builtin::BI__sync_fetch_and_xor: 220 case Builtin::BI__sync_fetch_and_xor_1: 221 case Builtin::BI__sync_fetch_and_xor_2: 222 case Builtin::BI__sync_fetch_and_xor_4: 223 case Builtin::BI__sync_fetch_and_xor_8: 224 case Builtin::BI__sync_fetch_and_xor_16: 225 case Builtin::BI__sync_add_and_fetch: 226 case Builtin::BI__sync_add_and_fetch_1: 227 case Builtin::BI__sync_add_and_fetch_2: 228 case Builtin::BI__sync_add_and_fetch_4: 229 case Builtin::BI__sync_add_and_fetch_8: 230 case Builtin::BI__sync_add_and_fetch_16: 231 case Builtin::BI__sync_sub_and_fetch: 232 case Builtin::BI__sync_sub_and_fetch_1: 233 case Builtin::BI__sync_sub_and_fetch_2: 234 case Builtin::BI__sync_sub_and_fetch_4: 235 case Builtin::BI__sync_sub_and_fetch_8: 236 case Builtin::BI__sync_sub_and_fetch_16: 237 case Builtin::BI__sync_and_and_fetch: 238 case Builtin::BI__sync_and_and_fetch_1: 239 case Builtin::BI__sync_and_and_fetch_2: 240 case Builtin::BI__sync_and_and_fetch_4: 241 case Builtin::BI__sync_and_and_fetch_8: 242 case Builtin::BI__sync_and_and_fetch_16: 243 case Builtin::BI__sync_or_and_fetch: 244 case Builtin::BI__sync_or_and_fetch_1: 245 case Builtin::BI__sync_or_and_fetch_2: 246 case Builtin::BI__sync_or_and_fetch_4: 247 case Builtin::BI__sync_or_and_fetch_8: 248 case Builtin::BI__sync_or_and_fetch_16: 249 case Builtin::BI__sync_xor_and_fetch: 250 case Builtin::BI__sync_xor_and_fetch_1: 251 case Builtin::BI__sync_xor_and_fetch_2: 252 case Builtin::BI__sync_xor_and_fetch_4: 253 case Builtin::BI__sync_xor_and_fetch_8: 254 case Builtin::BI__sync_xor_and_fetch_16: 255 case Builtin::BI__sync_val_compare_and_swap: 256 case Builtin::BI__sync_val_compare_and_swap_1: 257 case Builtin::BI__sync_val_compare_and_swap_2: 258 case Builtin::BI__sync_val_compare_and_swap_4: 259 case Builtin::BI__sync_val_compare_and_swap_8: 260 case Builtin::BI__sync_val_compare_and_swap_16: 261 case Builtin::BI__sync_bool_compare_and_swap: 262 case Builtin::BI__sync_bool_compare_and_swap_1: 263 case Builtin::BI__sync_bool_compare_and_swap_2: 264 case Builtin::BI__sync_bool_compare_and_swap_4: 265 case Builtin::BI__sync_bool_compare_and_swap_8: 266 case Builtin::BI__sync_bool_compare_and_swap_16: 267 case Builtin::BI__sync_lock_test_and_set: 268 case Builtin::BI__sync_lock_test_and_set_1: 269 case Builtin::BI__sync_lock_test_and_set_2: 270 case Builtin::BI__sync_lock_test_and_set_4: 271 case Builtin::BI__sync_lock_test_and_set_8: 272 case Builtin::BI__sync_lock_test_and_set_16: 273 case Builtin::BI__sync_lock_release: 274 case Builtin::BI__sync_lock_release_1: 275 case Builtin::BI__sync_lock_release_2: 276 case Builtin::BI__sync_lock_release_4: 277 case Builtin::BI__sync_lock_release_8: 278 case Builtin::BI__sync_lock_release_16: 279 case Builtin::BI__sync_swap: 280 case Builtin::BI__sync_swap_1: 281 case Builtin::BI__sync_swap_2: 282 case Builtin::BI__sync_swap_4: 283 case Builtin::BI__sync_swap_8: 284 case Builtin::BI__sync_swap_16: 285 return SemaBuiltinAtomicOverloaded(TheCallResult); 286 #define BUILTIN(ID, TYPE, ATTRS) 287 #define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \ 288 case Builtin::BI##ID: \ 289 return SemaAtomicOpsOverloaded(TheCallResult, AtomicExpr::AO##ID); 290 #include "clang/Basic/Builtins.def" 291 case Builtin::BI__builtin_annotation: 292 if (SemaBuiltinAnnotation(*this, TheCall)) 293 return ExprError(); 294 break; 295 case Builtin::BI__builtin_addressof: 296 if (SemaBuiltinAddressof(*this, TheCall)) 297 return ExprError(); 298 break; 299 } 300 301 // Since the target specific builtins for each arch overlap, only check those 302 // of the arch we are compiling for. 303 if (BuiltinID >= Builtin::FirstTSBuiltin) { 304 switch (Context.getTargetInfo().getTriple().getArch()) { 305 case llvm::Triple::arm: 306 case llvm::Triple::armeb: 307 case llvm::Triple::thumb: 308 case llvm::Triple::thumbeb: 309 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 310 return ExprError(); 311 break; 312 case llvm::Triple::aarch64: 313 case llvm::Triple::aarch64_be: 314 if (CheckAArch64BuiltinFunctionCall(BuiltinID, TheCall)) 315 return ExprError(); 316 break; 317 case llvm::Triple::mips: 318 case llvm::Triple::mipsel: 319 case llvm::Triple::mips64: 320 case llvm::Triple::mips64el: 321 if (CheckMipsBuiltinFunctionCall(BuiltinID, TheCall)) 322 return ExprError(); 323 break; 324 case llvm::Triple::x86: 325 case llvm::Triple::x86_64: 326 if (CheckX86BuiltinFunctionCall(BuiltinID, TheCall)) 327 return ExprError(); 328 break; 329 default: 330 break; 331 } 332 } 333 334 return TheCallResult; 335 } 336 337 // Get the valid immediate range for the specified NEON type code. 338 static unsigned RFT(unsigned t, bool shift = false, bool ForceQuad = false) { 339 NeonTypeFlags Type(t); 340 int IsQuad = ForceQuad ? true : Type.isQuad(); 341 switch (Type.getEltType()) { 342 case NeonTypeFlags::Int8: 343 case NeonTypeFlags::Poly8: 344 return shift ? 7 : (8 << IsQuad) - 1; 345 case NeonTypeFlags::Int16: 346 case NeonTypeFlags::Poly16: 347 return shift ? 15 : (4 << IsQuad) - 1; 348 case NeonTypeFlags::Int32: 349 return shift ? 31 : (2 << IsQuad) - 1; 350 case NeonTypeFlags::Int64: 351 case NeonTypeFlags::Poly64: 352 return shift ? 63 : (1 << IsQuad) - 1; 353 case NeonTypeFlags::Poly128: 354 return shift ? 127 : (1 << IsQuad) - 1; 355 case NeonTypeFlags::Float16: 356 assert(!shift && "cannot shift float types!"); 357 return (4 << IsQuad) - 1; 358 case NeonTypeFlags::Float32: 359 assert(!shift && "cannot shift float types!"); 360 return (2 << IsQuad) - 1; 361 case NeonTypeFlags::Float64: 362 assert(!shift && "cannot shift float types!"); 363 return (1 << IsQuad) - 1; 364 } 365 llvm_unreachable("Invalid NeonTypeFlag!"); 366 } 367 368 /// getNeonEltType - Return the QualType corresponding to the elements of 369 /// the vector type specified by the NeonTypeFlags. This is used to check 370 /// the pointer arguments for Neon load/store intrinsics. 371 static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context, 372 bool IsAArch64) { 373 switch (Flags.getEltType()) { 374 case NeonTypeFlags::Int8: 375 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 376 case NeonTypeFlags::Int16: 377 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 378 case NeonTypeFlags::Int32: 379 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 380 case NeonTypeFlags::Int64: 381 if (IsAArch64) 382 return Flags.isUnsigned() ? Context.UnsignedLongTy : Context.LongTy; 383 else 384 return Flags.isUnsigned() ? Context.UnsignedLongLongTy 385 : Context.LongLongTy; 386 case NeonTypeFlags::Poly8: 387 return IsAArch64 ? Context.UnsignedCharTy : Context.SignedCharTy; 388 case NeonTypeFlags::Poly16: 389 return IsAArch64 ? Context.UnsignedShortTy : Context.ShortTy; 390 case NeonTypeFlags::Poly64: 391 return Context.UnsignedLongTy; 392 case NeonTypeFlags::Poly128: 393 break; 394 case NeonTypeFlags::Float16: 395 return Context.HalfTy; 396 case NeonTypeFlags::Float32: 397 return Context.FloatTy; 398 case NeonTypeFlags::Float64: 399 return Context.DoubleTy; 400 } 401 llvm_unreachable("Invalid NeonTypeFlag!"); 402 } 403 404 bool Sema::CheckNeonBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 405 llvm::APSInt Result; 406 uint64_t mask = 0; 407 unsigned TV = 0; 408 int PtrArgNum = -1; 409 bool HasConstPtr = false; 410 switch (BuiltinID) { 411 #define GET_NEON_OVERLOAD_CHECK 412 #include "clang/Basic/arm_neon.inc" 413 #undef GET_NEON_OVERLOAD_CHECK 414 } 415 416 // For NEON intrinsics which are overloaded on vector element type, validate 417 // the immediate which specifies which variant to emit. 418 unsigned ImmArg = TheCall->getNumArgs()-1; 419 if (mask) { 420 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 421 return true; 422 423 TV = Result.getLimitedValue(64); 424 if ((TV > 63) || (mask & (1ULL << TV)) == 0) 425 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 426 << TheCall->getArg(ImmArg)->getSourceRange(); 427 } 428 429 if (PtrArgNum >= 0) { 430 // Check that pointer arguments have the specified type. 431 Expr *Arg = TheCall->getArg(PtrArgNum); 432 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 433 Arg = ICE->getSubExpr(); 434 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 435 QualType RHSTy = RHS.get()->getType(); 436 437 bool IsAArch64 = 438 Context.getTargetInfo().getTriple().getArch() == llvm::Triple::aarch64; 439 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context, IsAArch64); 440 if (HasConstPtr) 441 EltTy = EltTy.withConst(); 442 QualType LHSTy = Context.getPointerType(EltTy); 443 AssignConvertType ConvTy; 444 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 445 if (RHS.isInvalid()) 446 return true; 447 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 448 RHS.get(), AA_Assigning)) 449 return true; 450 } 451 452 // For NEON intrinsics which take an immediate value as part of the 453 // instruction, range check them here. 454 unsigned i = 0, l = 0, u = 0; 455 switch (BuiltinID) { 456 default: 457 return false; 458 #define GET_NEON_IMMEDIATE_CHECK 459 #include "clang/Basic/arm_neon.inc" 460 #undef GET_NEON_IMMEDIATE_CHECK 461 } 462 ; 463 464 // We can't check the value of a dependent argument. 465 if (TheCall->getArg(i)->isTypeDependent() || 466 TheCall->getArg(i)->isValueDependent()) 467 return false; 468 469 // Check that the immediate argument is actually a constant. 470 if (SemaBuiltinConstantArg(TheCall, i, Result)) 471 return true; 472 473 // Range check against the upper/lower values for this isntruction. 474 unsigned Val = Result.getZExtValue(); 475 if (Val < l || Val > (u + l)) 476 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 477 << l << u + l << TheCall->getArg(i)->getSourceRange(); 478 479 return false; 480 } 481 482 bool Sema::CheckAArch64BuiltinFunctionCall(unsigned BuiltinID, 483 CallExpr *TheCall) { 484 if (CheckNeonBuiltinFunctionCall(BuiltinID, TheCall)) 485 return true; 486 487 return false; 488 } 489 490 bool Sema::CheckARMBuiltinExclusiveCall(unsigned BuiltinID, CallExpr *TheCall) { 491 assert((BuiltinID == ARM::BI__builtin_arm_ldrex || 492 BuiltinID == ARM::BI__builtin_arm_strex) && 493 "unexpected ARM builtin"); 494 bool IsLdrex = BuiltinID == ARM::BI__builtin_arm_ldrex; 495 496 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 497 498 // Ensure that we have the proper number of arguments. 499 if (checkArgCount(*this, TheCall, IsLdrex ? 1 : 2)) 500 return true; 501 502 // Inspect the pointer argument of the atomic builtin. This should always be 503 // a pointer type, whose element is an integral scalar or pointer type. 504 // Because it is a pointer type, we don't have to worry about any implicit 505 // casts here. 506 Expr *PointerArg = TheCall->getArg(IsLdrex ? 0 : 1); 507 ExprResult PointerArgRes = DefaultFunctionArrayLvalueConversion(PointerArg); 508 if (PointerArgRes.isInvalid()) 509 return true; 510 PointerArg = PointerArgRes.take(); 511 512 const PointerType *pointerType = PointerArg->getType()->getAs<PointerType>(); 513 if (!pointerType) { 514 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 515 << PointerArg->getType() << PointerArg->getSourceRange(); 516 return true; 517 } 518 519 // ldrex takes a "const volatile T*" and strex takes a "volatile T*". Our next 520 // task is to insert the appropriate casts into the AST. First work out just 521 // what the appropriate type is. 522 QualType ValType = pointerType->getPointeeType(); 523 QualType AddrType = ValType.getUnqualifiedType().withVolatile(); 524 if (IsLdrex) 525 AddrType.addConst(); 526 527 // Issue a warning if the cast is dodgy. 528 CastKind CastNeeded = CK_NoOp; 529 if (!AddrType.isAtLeastAsQualifiedAs(ValType)) { 530 CastNeeded = CK_BitCast; 531 Diag(DRE->getLocStart(), diag::ext_typecheck_convert_discards_qualifiers) 532 << PointerArg->getType() 533 << Context.getPointerType(AddrType) 534 << AA_Passing << PointerArg->getSourceRange(); 535 } 536 537 // Finally, do the cast and replace the argument with the corrected version. 538 AddrType = Context.getPointerType(AddrType); 539 PointerArgRes = ImpCastExprToType(PointerArg, AddrType, CastNeeded); 540 if (PointerArgRes.isInvalid()) 541 return true; 542 PointerArg = PointerArgRes.take(); 543 544 TheCall->setArg(IsLdrex ? 0 : 1, PointerArg); 545 546 // In general, we allow ints, floats and pointers to be loaded and stored. 547 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 548 !ValType->isBlockPointerType() && !ValType->isFloatingType()) { 549 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intfltptr) 550 << PointerArg->getType() << PointerArg->getSourceRange(); 551 return true; 552 } 553 554 // But ARM doesn't have instructions to deal with 128-bit versions. 555 if (Context.getTypeSize(ValType) > 64) { 556 Diag(DRE->getLocStart(), diag::err_atomic_exclusive_builtin_pointer_size) 557 << PointerArg->getType() << PointerArg->getSourceRange(); 558 return true; 559 } 560 561 switch (ValType.getObjCLifetime()) { 562 case Qualifiers::OCL_None: 563 case Qualifiers::OCL_ExplicitNone: 564 // okay 565 break; 566 567 case Qualifiers::OCL_Weak: 568 case Qualifiers::OCL_Strong: 569 case Qualifiers::OCL_Autoreleasing: 570 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 571 << ValType << PointerArg->getSourceRange(); 572 return true; 573 } 574 575 576 if (IsLdrex) { 577 TheCall->setType(ValType); 578 return false; 579 } 580 581 // Initialize the argument to be stored. 582 ExprResult ValArg = TheCall->getArg(0); 583 InitializedEntity Entity = InitializedEntity::InitializeParameter( 584 Context, ValType, /*consume*/ false); 585 ValArg = PerformCopyInitialization(Entity, SourceLocation(), ValArg); 586 if (ValArg.isInvalid()) 587 return true; 588 TheCall->setArg(0, ValArg.get()); 589 590 // __builtin_arm_strex always returns an int. It's marked as such in the .def, 591 // but the custom checker bypasses all default analysis. 592 TheCall->setType(Context.IntTy); 593 return false; 594 } 595 596 bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 597 llvm::APSInt Result; 598 599 if (BuiltinID == ARM::BI__builtin_arm_ldrex || 600 BuiltinID == ARM::BI__builtin_arm_strex) { 601 return CheckARMBuiltinExclusiveCall(BuiltinID, TheCall); 602 } 603 604 if (CheckNeonBuiltinFunctionCall(BuiltinID, TheCall)) 605 return true; 606 607 // For NEON intrinsics which take an immediate value as part of the 608 // instruction, range check them here. 609 unsigned i = 0, l = 0, u = 0; 610 switch (BuiltinID) { 611 default: return false; 612 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 613 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 614 case ARM::BI__builtin_arm_vcvtr_f: 615 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 616 case ARM::BI__builtin_arm_dmb: 617 case ARM::BI__builtin_arm_dsb: l = 0; u = 15; break; 618 }; 619 620 // We can't check the value of a dependent argument. 621 if (TheCall->getArg(i)->isTypeDependent() || 622 TheCall->getArg(i)->isValueDependent()) 623 return false; 624 625 // Check that the immediate argument is actually a constant. 626 if (SemaBuiltinConstantArg(TheCall, i, Result)) 627 return true; 628 629 // Range check against the upper/lower values for this isntruction. 630 unsigned Val = Result.getZExtValue(); 631 if (Val < l || Val > (u + l)) 632 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 633 << l << u+l << TheCall->getArg(i)->getSourceRange(); 634 635 // FIXME: VFP Intrinsics should error if VFP not present. 636 return false; 637 } 638 639 bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 640 unsigned i = 0, l = 0, u = 0; 641 switch (BuiltinID) { 642 default: return false; 643 case Mips::BI__builtin_mips_wrdsp: i = 1; l = 0; u = 63; break; 644 case Mips::BI__builtin_mips_rddsp: i = 0; l = 0; u = 63; break; 645 case Mips::BI__builtin_mips_append: i = 2; l = 0; u = 31; break; 646 case Mips::BI__builtin_mips_balign: i = 2; l = 0; u = 3; break; 647 case Mips::BI__builtin_mips_precr_sra_ph_w: i = 2; l = 0; u = 31; break; 648 case Mips::BI__builtin_mips_precr_sra_r_ph_w: i = 2; l = 0; u = 31; break; 649 case Mips::BI__builtin_mips_prepend: i = 2; l = 0; u = 31; break; 650 }; 651 652 // We can't check the value of a dependent argument. 653 if (TheCall->getArg(i)->isTypeDependent() || 654 TheCall->getArg(i)->isValueDependent()) 655 return false; 656 657 // Check that the immediate argument is actually a constant. 658 llvm::APSInt Result; 659 if (SemaBuiltinConstantArg(TheCall, i, Result)) 660 return true; 661 662 // Range check against the upper/lower values for this instruction. 663 unsigned Val = Result.getZExtValue(); 664 if (Val < l || Val > u) 665 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 666 << l << u << TheCall->getArg(i)->getSourceRange(); 667 668 return false; 669 } 670 671 bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 672 switch (BuiltinID) { 673 case X86::BI_mm_prefetch: 674 return SemaBuiltinMMPrefetch(TheCall); 675 } 676 return false; 677 } 678 679 /// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo 680 /// parameter with the FormatAttr's correct format_idx and firstDataArg. 681 /// Returns true when the format fits the function and the FormatStringInfo has 682 /// been populated. 683 bool Sema::getFormatStringInfo(const FormatAttr *Format, bool IsCXXMember, 684 FormatStringInfo *FSI) { 685 FSI->HasVAListArg = Format->getFirstArg() == 0; 686 FSI->FormatIdx = Format->getFormatIdx() - 1; 687 FSI->FirstDataArg = FSI->HasVAListArg ? 0 : Format->getFirstArg() - 1; 688 689 // The way the format attribute works in GCC, the implicit this argument 690 // of member functions is counted. However, it doesn't appear in our own 691 // lists, so decrement format_idx in that case. 692 if (IsCXXMember) { 693 if(FSI->FormatIdx == 0) 694 return false; 695 --FSI->FormatIdx; 696 if (FSI->FirstDataArg != 0) 697 --FSI->FirstDataArg; 698 } 699 return true; 700 } 701 702 /// Checks if a the given expression evaluates to null. 703 /// 704 /// \brief Returns true if the value evaluates to null. 705 static bool CheckNonNullExpr(Sema &S, 706 const Expr *Expr) { 707 // As a special case, transparent unions initialized with zero are 708 // considered null for the purposes of the nonnull attribute. 709 if (const RecordType *UT = Expr->getType()->getAsUnionType()) { 710 if (UT->getDecl()->hasAttr<TransparentUnionAttr>()) 711 if (const CompoundLiteralExpr *CLE = 712 dyn_cast<CompoundLiteralExpr>(Expr)) 713 if (const InitListExpr *ILE = 714 dyn_cast<InitListExpr>(CLE->getInitializer())) 715 Expr = ILE->getInit(0); 716 } 717 718 bool Result; 719 return (!Expr->isValueDependent() && 720 Expr->EvaluateAsBooleanCondition(Result, S.Context) && 721 !Result); 722 } 723 724 static void CheckNonNullArgument(Sema &S, 725 const Expr *ArgExpr, 726 SourceLocation CallSiteLoc) { 727 if (CheckNonNullExpr(S, ArgExpr)) 728 S.Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 729 } 730 731 static void CheckNonNullArguments(Sema &S, 732 const NamedDecl *FDecl, 733 const Expr * const *ExprArgs, 734 SourceLocation CallSiteLoc) { 735 // Check the attributes attached to the method/function itself. 736 for (const auto *NonNull : FDecl->specific_attrs<NonNullAttr>()) { 737 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 738 e = NonNull->args_end(); 739 i != e; ++i) { 740 CheckNonNullArgument(S, ExprArgs[*i], CallSiteLoc); 741 } 742 } 743 744 // Check the attributes on the parameters. 745 ArrayRef<ParmVarDecl*> parms; 746 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(FDecl)) 747 parms = FD->parameters(); 748 else if (const ObjCMethodDecl *MD = dyn_cast<ObjCMethodDecl>(FDecl)) 749 parms = MD->parameters(); 750 751 unsigned argIndex = 0; 752 for (ArrayRef<ParmVarDecl*>::iterator I = parms.begin(), E = parms.end(); 753 I != E; ++I, ++argIndex) { 754 const ParmVarDecl *PVD = *I; 755 if (PVD->hasAttr<NonNullAttr>()) 756 CheckNonNullArgument(S, ExprArgs[argIndex], CallSiteLoc); 757 } 758 } 759 760 /// Handles the checks for format strings, non-POD arguments to vararg 761 /// functions, and NULL arguments passed to non-NULL parameters. 762 void Sema::checkCall(NamedDecl *FDecl, ArrayRef<const Expr *> Args, 763 unsigned NumParams, bool IsMemberFunction, 764 SourceLocation Loc, SourceRange Range, 765 VariadicCallType CallType) { 766 // FIXME: We should check as much as we can in the template definition. 767 if (CurContext->isDependentContext()) 768 return; 769 770 // Printf and scanf checking. 771 llvm::SmallBitVector CheckedVarArgs; 772 if (FDecl) { 773 for (const auto *I : FDecl->specific_attrs<FormatAttr>()) { 774 // Only create vector if there are format attributes. 775 CheckedVarArgs.resize(Args.size()); 776 777 CheckFormatArguments(I, Args, IsMemberFunction, CallType, Loc, Range, 778 CheckedVarArgs); 779 } 780 } 781 782 // Refuse POD arguments that weren't caught by the format string 783 // checks above. 784 if (CallType != VariadicDoesNotApply) { 785 for (unsigned ArgIdx = NumParams; ArgIdx < Args.size(); ++ArgIdx) { 786 // Args[ArgIdx] can be null in malformed code. 787 if (const Expr *Arg = Args[ArgIdx]) { 788 if (CheckedVarArgs.empty() || !CheckedVarArgs[ArgIdx]) 789 checkVariadicArgument(Arg, CallType); 790 } 791 } 792 } 793 794 if (FDecl) { 795 CheckNonNullArguments(*this, FDecl, Args.data(), Loc); 796 797 // Type safety checking. 798 for (const auto *I : FDecl->specific_attrs<ArgumentWithTypeTagAttr>()) 799 CheckArgumentWithTypeTag(I, Args.data()); 800 } 801 } 802 803 /// CheckConstructorCall - Check a constructor call for correctness and safety 804 /// properties not enforced by the C type system. 805 void Sema::CheckConstructorCall(FunctionDecl *FDecl, 806 ArrayRef<const Expr *> Args, 807 const FunctionProtoType *Proto, 808 SourceLocation Loc) { 809 VariadicCallType CallType = 810 Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply; 811 checkCall(FDecl, Args, Proto->getNumParams(), 812 /*IsMemberFunction=*/true, Loc, SourceRange(), CallType); 813 } 814 815 /// CheckFunctionCall - Check a direct function call for various correctness 816 /// and safety properties not strictly enforced by the C type system. 817 bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall, 818 const FunctionProtoType *Proto) { 819 bool IsMemberOperatorCall = isa<CXXOperatorCallExpr>(TheCall) && 820 isa<CXXMethodDecl>(FDecl); 821 bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall) || 822 IsMemberOperatorCall; 823 VariadicCallType CallType = getVariadicCallType(FDecl, Proto, 824 TheCall->getCallee()); 825 unsigned NumParams = Proto ? Proto->getNumParams() : 0; 826 Expr** Args = TheCall->getArgs(); 827 unsigned NumArgs = TheCall->getNumArgs(); 828 if (IsMemberOperatorCall) { 829 // If this is a call to a member operator, hide the first argument 830 // from checkCall. 831 // FIXME: Our choice of AST representation here is less than ideal. 832 ++Args; 833 --NumArgs; 834 } 835 checkCall(FDecl, llvm::makeArrayRef<const Expr *>(Args, NumArgs), NumParams, 836 IsMemberFunction, TheCall->getRParenLoc(), 837 TheCall->getCallee()->getSourceRange(), CallType); 838 839 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 840 // None of the checks below are needed for functions that don't have 841 // simple names (e.g., C++ conversion functions). 842 if (!FnInfo) 843 return false; 844 845 CheckAbsoluteValueFunction(TheCall, FDecl, FnInfo); 846 847 unsigned CMId = FDecl->getMemoryFunctionKind(); 848 if (CMId == 0) 849 return false; 850 851 // Handle memory setting and copying functions. 852 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 853 CheckStrlcpycatArguments(TheCall, FnInfo); 854 else if (CMId == Builtin::BIstrncat) 855 CheckStrncatArguments(TheCall, FnInfo); 856 else 857 CheckMemaccessArguments(TheCall, CMId, FnInfo); 858 859 return false; 860 } 861 862 bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 863 ArrayRef<const Expr *> Args) { 864 VariadicCallType CallType = 865 Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply; 866 867 checkCall(Method, Args, Method->param_size(), 868 /*IsMemberFunction=*/false, 869 lbrac, Method->getSourceRange(), CallType); 870 871 return false; 872 } 873 874 bool Sema::CheckPointerCall(NamedDecl *NDecl, CallExpr *TheCall, 875 const FunctionProtoType *Proto) { 876 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 877 if (!V) 878 return false; 879 880 QualType Ty = V->getType(); 881 if (!Ty->isBlockPointerType() && !Ty->isFunctionPointerType()) 882 return false; 883 884 VariadicCallType CallType; 885 if (!Proto || !Proto->isVariadic()) { 886 CallType = VariadicDoesNotApply; 887 } else if (Ty->isBlockPointerType()) { 888 CallType = VariadicBlock; 889 } else { // Ty->isFunctionPointerType() 890 CallType = VariadicFunction; 891 } 892 unsigned NumParams = Proto ? Proto->getNumParams() : 0; 893 894 checkCall(NDecl, llvm::makeArrayRef<const Expr *>(TheCall->getArgs(), 895 TheCall->getNumArgs()), 896 NumParams, /*IsMemberFunction=*/false, TheCall->getRParenLoc(), 897 TheCall->getCallee()->getSourceRange(), CallType); 898 899 return false; 900 } 901 902 /// Checks function calls when a FunctionDecl or a NamedDecl is not available, 903 /// such as function pointers returned from functions. 904 bool Sema::CheckOtherCall(CallExpr *TheCall, const FunctionProtoType *Proto) { 905 VariadicCallType CallType = getVariadicCallType(/*FDecl=*/0, Proto, 906 TheCall->getCallee()); 907 unsigned NumParams = Proto ? Proto->getNumParams() : 0; 908 909 checkCall(/*FDecl=*/0, llvm::makeArrayRef<const Expr *>( 910 TheCall->getArgs(), TheCall->getNumArgs()), 911 NumParams, /*IsMemberFunction=*/false, TheCall->getRParenLoc(), 912 TheCall->getCallee()->getSourceRange(), CallType); 913 914 return false; 915 } 916 917 static bool isValidOrderingForOp(int64_t Ordering, AtomicExpr::AtomicOp Op) { 918 if (Ordering < AtomicExpr::AO_ABI_memory_order_relaxed || 919 Ordering > AtomicExpr::AO_ABI_memory_order_seq_cst) 920 return false; 921 922 switch (Op) { 923 case AtomicExpr::AO__c11_atomic_init: 924 llvm_unreachable("There is no ordering argument for an init"); 925 926 case AtomicExpr::AO__c11_atomic_load: 927 case AtomicExpr::AO__atomic_load_n: 928 case AtomicExpr::AO__atomic_load: 929 return Ordering != AtomicExpr::AO_ABI_memory_order_release && 930 Ordering != AtomicExpr::AO_ABI_memory_order_acq_rel; 931 932 case AtomicExpr::AO__c11_atomic_store: 933 case AtomicExpr::AO__atomic_store: 934 case AtomicExpr::AO__atomic_store_n: 935 return Ordering != AtomicExpr::AO_ABI_memory_order_consume && 936 Ordering != AtomicExpr::AO_ABI_memory_order_acquire && 937 Ordering != AtomicExpr::AO_ABI_memory_order_acq_rel; 938 939 default: 940 return true; 941 } 942 } 943 944 ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 945 AtomicExpr::AtomicOp Op) { 946 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 947 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 948 949 // All these operations take one of the following forms: 950 enum { 951 // C __c11_atomic_init(A *, C) 952 Init, 953 // C __c11_atomic_load(A *, int) 954 Load, 955 // void __atomic_load(A *, CP, int) 956 Copy, 957 // C __c11_atomic_add(A *, M, int) 958 Arithmetic, 959 // C __atomic_exchange_n(A *, CP, int) 960 Xchg, 961 // void __atomic_exchange(A *, C *, CP, int) 962 GNUXchg, 963 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 964 C11CmpXchg, 965 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 966 GNUCmpXchg 967 } Form = Init; 968 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 969 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 970 // where: 971 // C is an appropriate type, 972 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 973 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 974 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 975 // the int parameters are for orderings. 976 977 assert(AtomicExpr::AO__c11_atomic_init == 0 && 978 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 979 && "need to update code for modified C11 atomics"); 980 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 981 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 982 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 983 Op == AtomicExpr::AO__atomic_store_n || 984 Op == AtomicExpr::AO__atomic_exchange_n || 985 Op == AtomicExpr::AO__atomic_compare_exchange_n; 986 bool IsAddSub = false; 987 988 switch (Op) { 989 case AtomicExpr::AO__c11_atomic_init: 990 Form = Init; 991 break; 992 993 case AtomicExpr::AO__c11_atomic_load: 994 case AtomicExpr::AO__atomic_load_n: 995 Form = Load; 996 break; 997 998 case AtomicExpr::AO__c11_atomic_store: 999 case AtomicExpr::AO__atomic_load: 1000 case AtomicExpr::AO__atomic_store: 1001 case AtomicExpr::AO__atomic_store_n: 1002 Form = Copy; 1003 break; 1004 1005 case AtomicExpr::AO__c11_atomic_fetch_add: 1006 case AtomicExpr::AO__c11_atomic_fetch_sub: 1007 case AtomicExpr::AO__atomic_fetch_add: 1008 case AtomicExpr::AO__atomic_fetch_sub: 1009 case AtomicExpr::AO__atomic_add_fetch: 1010 case AtomicExpr::AO__atomic_sub_fetch: 1011 IsAddSub = true; 1012 // Fall through. 1013 case AtomicExpr::AO__c11_atomic_fetch_and: 1014 case AtomicExpr::AO__c11_atomic_fetch_or: 1015 case AtomicExpr::AO__c11_atomic_fetch_xor: 1016 case AtomicExpr::AO__atomic_fetch_and: 1017 case AtomicExpr::AO__atomic_fetch_or: 1018 case AtomicExpr::AO__atomic_fetch_xor: 1019 case AtomicExpr::AO__atomic_fetch_nand: 1020 case AtomicExpr::AO__atomic_and_fetch: 1021 case AtomicExpr::AO__atomic_or_fetch: 1022 case AtomicExpr::AO__atomic_xor_fetch: 1023 case AtomicExpr::AO__atomic_nand_fetch: 1024 Form = Arithmetic; 1025 break; 1026 1027 case AtomicExpr::AO__c11_atomic_exchange: 1028 case AtomicExpr::AO__atomic_exchange_n: 1029 Form = Xchg; 1030 break; 1031 1032 case AtomicExpr::AO__atomic_exchange: 1033 Form = GNUXchg; 1034 break; 1035 1036 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 1037 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 1038 Form = C11CmpXchg; 1039 break; 1040 1041 case AtomicExpr::AO__atomic_compare_exchange: 1042 case AtomicExpr::AO__atomic_compare_exchange_n: 1043 Form = GNUCmpXchg; 1044 break; 1045 } 1046 1047 // Check we have the right number of arguments. 1048 if (TheCall->getNumArgs() < NumArgs[Form]) { 1049 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1050 << 0 << NumArgs[Form] << TheCall->getNumArgs() 1051 << TheCall->getCallee()->getSourceRange(); 1052 return ExprError(); 1053 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 1054 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 1055 diag::err_typecheck_call_too_many_args) 1056 << 0 << NumArgs[Form] << TheCall->getNumArgs() 1057 << TheCall->getCallee()->getSourceRange(); 1058 return ExprError(); 1059 } 1060 1061 // Inspect the first argument of the atomic operation. 1062 Expr *Ptr = TheCall->getArg(0); 1063 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 1064 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 1065 if (!pointerType) { 1066 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 1067 << Ptr->getType() << Ptr->getSourceRange(); 1068 return ExprError(); 1069 } 1070 1071 // For a __c11 builtin, this should be a pointer to an _Atomic type. 1072 QualType AtomTy = pointerType->getPointeeType(); // 'A' 1073 QualType ValType = AtomTy; // 'C' 1074 if (IsC11) { 1075 if (!AtomTy->isAtomicType()) { 1076 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 1077 << Ptr->getType() << Ptr->getSourceRange(); 1078 return ExprError(); 1079 } 1080 if (AtomTy.isConstQualified()) { 1081 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_non_const_atomic) 1082 << Ptr->getType() << Ptr->getSourceRange(); 1083 return ExprError(); 1084 } 1085 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 1086 } 1087 1088 // For an arithmetic operation, the implied arithmetic must be well-formed. 1089 if (Form == Arithmetic) { 1090 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 1091 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 1092 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 1093 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 1094 return ExprError(); 1095 } 1096 if (!IsAddSub && !ValType->isIntegerType()) { 1097 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 1098 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 1099 return ExprError(); 1100 } 1101 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 1102 // For __atomic_*_n operations, the value type must be a scalar integral or 1103 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 1104 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 1105 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 1106 return ExprError(); 1107 } 1108 1109 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context) && 1110 !AtomTy->isScalarType()) { 1111 // For GNU atomics, require a trivially-copyable type. This is not part of 1112 // the GNU atomics specification, but we enforce it for sanity. 1113 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 1114 << Ptr->getType() << Ptr->getSourceRange(); 1115 return ExprError(); 1116 } 1117 1118 // FIXME: For any builtin other than a load, the ValType must not be 1119 // const-qualified. 1120 1121 switch (ValType.getObjCLifetime()) { 1122 case Qualifiers::OCL_None: 1123 case Qualifiers::OCL_ExplicitNone: 1124 // okay 1125 break; 1126 1127 case Qualifiers::OCL_Weak: 1128 case Qualifiers::OCL_Strong: 1129 case Qualifiers::OCL_Autoreleasing: 1130 // FIXME: Can this happen? By this point, ValType should be known 1131 // to be trivially copyable. 1132 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1133 << ValType << Ptr->getSourceRange(); 1134 return ExprError(); 1135 } 1136 1137 QualType ResultType = ValType; 1138 if (Form == Copy || Form == GNUXchg || Form == Init) 1139 ResultType = Context.VoidTy; 1140 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 1141 ResultType = Context.BoolTy; 1142 1143 // The type of a parameter passed 'by value'. In the GNU atomics, such 1144 // arguments are actually passed as pointers. 1145 QualType ByValType = ValType; // 'CP' 1146 if (!IsC11 && !IsN) 1147 ByValType = Ptr->getType(); 1148 1149 // The first argument --- the pointer --- has a fixed type; we 1150 // deduce the types of the rest of the arguments accordingly. Walk 1151 // the remaining arguments, converting them to the deduced value type. 1152 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 1153 QualType Ty; 1154 if (i < NumVals[Form] + 1) { 1155 switch (i) { 1156 case 1: 1157 // The second argument is the non-atomic operand. For arithmetic, this 1158 // is always passed by value, and for a compare_exchange it is always 1159 // passed by address. For the rest, GNU uses by-address and C11 uses 1160 // by-value. 1161 assert(Form != Load); 1162 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 1163 Ty = ValType; 1164 else if (Form == Copy || Form == Xchg) 1165 Ty = ByValType; 1166 else if (Form == Arithmetic) 1167 Ty = Context.getPointerDiffType(); 1168 else 1169 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 1170 break; 1171 case 2: 1172 // The third argument to compare_exchange / GNU exchange is a 1173 // (pointer to a) desired value. 1174 Ty = ByValType; 1175 break; 1176 case 3: 1177 // The fourth argument to GNU compare_exchange is a 'weak' flag. 1178 Ty = Context.BoolTy; 1179 break; 1180 } 1181 } else { 1182 // The order(s) are always converted to int. 1183 Ty = Context.IntTy; 1184 } 1185 1186 InitializedEntity Entity = 1187 InitializedEntity::InitializeParameter(Context, Ty, false); 1188 ExprResult Arg = TheCall->getArg(i); 1189 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1190 if (Arg.isInvalid()) 1191 return true; 1192 TheCall->setArg(i, Arg.get()); 1193 } 1194 1195 // Permute the arguments into a 'consistent' order. 1196 SmallVector<Expr*, 5> SubExprs; 1197 SubExprs.push_back(Ptr); 1198 switch (Form) { 1199 case Init: 1200 // Note, AtomicExpr::getVal1() has a special case for this atomic. 1201 SubExprs.push_back(TheCall->getArg(1)); // Val1 1202 break; 1203 case Load: 1204 SubExprs.push_back(TheCall->getArg(1)); // Order 1205 break; 1206 case Copy: 1207 case Arithmetic: 1208 case Xchg: 1209 SubExprs.push_back(TheCall->getArg(2)); // Order 1210 SubExprs.push_back(TheCall->getArg(1)); // Val1 1211 break; 1212 case GNUXchg: 1213 // Note, AtomicExpr::getVal2() has a special case for this atomic. 1214 SubExprs.push_back(TheCall->getArg(3)); // Order 1215 SubExprs.push_back(TheCall->getArg(1)); // Val1 1216 SubExprs.push_back(TheCall->getArg(2)); // Val2 1217 break; 1218 case C11CmpXchg: 1219 SubExprs.push_back(TheCall->getArg(3)); // Order 1220 SubExprs.push_back(TheCall->getArg(1)); // Val1 1221 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 1222 SubExprs.push_back(TheCall->getArg(2)); // Val2 1223 break; 1224 case GNUCmpXchg: 1225 SubExprs.push_back(TheCall->getArg(4)); // Order 1226 SubExprs.push_back(TheCall->getArg(1)); // Val1 1227 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 1228 SubExprs.push_back(TheCall->getArg(2)); // Val2 1229 SubExprs.push_back(TheCall->getArg(3)); // Weak 1230 break; 1231 } 1232 1233 if (SubExprs.size() >= 2 && Form != Init) { 1234 llvm::APSInt Result(32); 1235 if (SubExprs[1]->isIntegerConstantExpr(Result, Context) && 1236 !isValidOrderingForOp(Result.getSExtValue(), Op)) 1237 Diag(SubExprs[1]->getLocStart(), 1238 diag::warn_atomic_op_has_invalid_memory_order) 1239 << SubExprs[1]->getSourceRange(); 1240 } 1241 1242 AtomicExpr *AE = new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 1243 SubExprs, ResultType, Op, 1244 TheCall->getRParenLoc()); 1245 1246 if ((Op == AtomicExpr::AO__c11_atomic_load || 1247 (Op == AtomicExpr::AO__c11_atomic_store)) && 1248 Context.AtomicUsesUnsupportedLibcall(AE)) 1249 Diag(AE->getLocStart(), diag::err_atomic_load_store_uses_lib) << 1250 ((Op == AtomicExpr::AO__c11_atomic_load) ? 0 : 1); 1251 1252 return Owned(AE); 1253 } 1254 1255 1256 /// checkBuiltinArgument - Given a call to a builtin function, perform 1257 /// normal type-checking on the given argument, updating the call in 1258 /// place. This is useful when a builtin function requires custom 1259 /// type-checking for some of its arguments but not necessarily all of 1260 /// them. 1261 /// 1262 /// Returns true on error. 1263 static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 1264 FunctionDecl *Fn = E->getDirectCallee(); 1265 assert(Fn && "builtin call without direct callee!"); 1266 1267 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 1268 InitializedEntity Entity = 1269 InitializedEntity::InitializeParameter(S.Context, Param); 1270 1271 ExprResult Arg = E->getArg(0); 1272 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 1273 if (Arg.isInvalid()) 1274 return true; 1275 1276 E->setArg(ArgIndex, Arg.take()); 1277 return false; 1278 } 1279 1280 /// SemaBuiltinAtomicOverloaded - We have a call to a function like 1281 /// __sync_fetch_and_add, which is an overloaded function based on the pointer 1282 /// type of its first argument. The main ActOnCallExpr routines have already 1283 /// promoted the types of arguments because all of these calls are prototyped as 1284 /// void(...). 1285 /// 1286 /// This function goes through and does final semantic checking for these 1287 /// builtins, 1288 ExprResult 1289 Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 1290 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 1291 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1292 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1293 1294 // Ensure that we have at least one argument to do type inference from. 1295 if (TheCall->getNumArgs() < 1) { 1296 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1297 << 0 << 1 << TheCall->getNumArgs() 1298 << TheCall->getCallee()->getSourceRange(); 1299 return ExprError(); 1300 } 1301 1302 // Inspect the first argument of the atomic builtin. This should always be 1303 // a pointer type, whose element is an integral scalar or pointer type. 1304 // Because it is a pointer type, we don't have to worry about any implicit 1305 // casts here. 1306 // FIXME: We don't allow floating point scalars as input. 1307 Expr *FirstArg = TheCall->getArg(0); 1308 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 1309 if (FirstArgResult.isInvalid()) 1310 return ExprError(); 1311 FirstArg = FirstArgResult.take(); 1312 TheCall->setArg(0, FirstArg); 1313 1314 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 1315 if (!pointerType) { 1316 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 1317 << FirstArg->getType() << FirstArg->getSourceRange(); 1318 return ExprError(); 1319 } 1320 1321 QualType ValType = pointerType->getPointeeType(); 1322 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 1323 !ValType->isBlockPointerType()) { 1324 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 1325 << FirstArg->getType() << FirstArg->getSourceRange(); 1326 return ExprError(); 1327 } 1328 1329 switch (ValType.getObjCLifetime()) { 1330 case Qualifiers::OCL_None: 1331 case Qualifiers::OCL_ExplicitNone: 1332 // okay 1333 break; 1334 1335 case Qualifiers::OCL_Weak: 1336 case Qualifiers::OCL_Strong: 1337 case Qualifiers::OCL_Autoreleasing: 1338 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1339 << ValType << FirstArg->getSourceRange(); 1340 return ExprError(); 1341 } 1342 1343 // Strip any qualifiers off ValType. 1344 ValType = ValType.getUnqualifiedType(); 1345 1346 // The majority of builtins return a value, but a few have special return 1347 // types, so allow them to override appropriately below. 1348 QualType ResultType = ValType; 1349 1350 // We need to figure out which concrete builtin this maps onto. For example, 1351 // __sync_fetch_and_add with a 2 byte object turns into 1352 // __sync_fetch_and_add_2. 1353 #define BUILTIN_ROW(x) \ 1354 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 1355 Builtin::BI##x##_8, Builtin::BI##x##_16 } 1356 1357 static const unsigned BuiltinIndices[][5] = { 1358 BUILTIN_ROW(__sync_fetch_and_add), 1359 BUILTIN_ROW(__sync_fetch_and_sub), 1360 BUILTIN_ROW(__sync_fetch_and_or), 1361 BUILTIN_ROW(__sync_fetch_and_and), 1362 BUILTIN_ROW(__sync_fetch_and_xor), 1363 1364 BUILTIN_ROW(__sync_add_and_fetch), 1365 BUILTIN_ROW(__sync_sub_and_fetch), 1366 BUILTIN_ROW(__sync_and_and_fetch), 1367 BUILTIN_ROW(__sync_or_and_fetch), 1368 BUILTIN_ROW(__sync_xor_and_fetch), 1369 1370 BUILTIN_ROW(__sync_val_compare_and_swap), 1371 BUILTIN_ROW(__sync_bool_compare_and_swap), 1372 BUILTIN_ROW(__sync_lock_test_and_set), 1373 BUILTIN_ROW(__sync_lock_release), 1374 BUILTIN_ROW(__sync_swap) 1375 }; 1376 #undef BUILTIN_ROW 1377 1378 // Determine the index of the size. 1379 unsigned SizeIndex; 1380 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 1381 case 1: SizeIndex = 0; break; 1382 case 2: SizeIndex = 1; break; 1383 case 4: SizeIndex = 2; break; 1384 case 8: SizeIndex = 3; break; 1385 case 16: SizeIndex = 4; break; 1386 default: 1387 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 1388 << FirstArg->getType() << FirstArg->getSourceRange(); 1389 return ExprError(); 1390 } 1391 1392 // Each of these builtins has one pointer argument, followed by some number of 1393 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 1394 // that we ignore. Find out which row of BuiltinIndices to read from as well 1395 // as the number of fixed args. 1396 unsigned BuiltinID = FDecl->getBuiltinID(); 1397 unsigned BuiltinIndex, NumFixed = 1; 1398 switch (BuiltinID) { 1399 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 1400 case Builtin::BI__sync_fetch_and_add: 1401 case Builtin::BI__sync_fetch_and_add_1: 1402 case Builtin::BI__sync_fetch_and_add_2: 1403 case Builtin::BI__sync_fetch_and_add_4: 1404 case Builtin::BI__sync_fetch_and_add_8: 1405 case Builtin::BI__sync_fetch_and_add_16: 1406 BuiltinIndex = 0; 1407 break; 1408 1409 case Builtin::BI__sync_fetch_and_sub: 1410 case Builtin::BI__sync_fetch_and_sub_1: 1411 case Builtin::BI__sync_fetch_and_sub_2: 1412 case Builtin::BI__sync_fetch_and_sub_4: 1413 case Builtin::BI__sync_fetch_and_sub_8: 1414 case Builtin::BI__sync_fetch_and_sub_16: 1415 BuiltinIndex = 1; 1416 break; 1417 1418 case Builtin::BI__sync_fetch_and_or: 1419 case Builtin::BI__sync_fetch_and_or_1: 1420 case Builtin::BI__sync_fetch_and_or_2: 1421 case Builtin::BI__sync_fetch_and_or_4: 1422 case Builtin::BI__sync_fetch_and_or_8: 1423 case Builtin::BI__sync_fetch_and_or_16: 1424 BuiltinIndex = 2; 1425 break; 1426 1427 case Builtin::BI__sync_fetch_and_and: 1428 case Builtin::BI__sync_fetch_and_and_1: 1429 case Builtin::BI__sync_fetch_and_and_2: 1430 case Builtin::BI__sync_fetch_and_and_4: 1431 case Builtin::BI__sync_fetch_and_and_8: 1432 case Builtin::BI__sync_fetch_and_and_16: 1433 BuiltinIndex = 3; 1434 break; 1435 1436 case Builtin::BI__sync_fetch_and_xor: 1437 case Builtin::BI__sync_fetch_and_xor_1: 1438 case Builtin::BI__sync_fetch_and_xor_2: 1439 case Builtin::BI__sync_fetch_and_xor_4: 1440 case Builtin::BI__sync_fetch_and_xor_8: 1441 case Builtin::BI__sync_fetch_and_xor_16: 1442 BuiltinIndex = 4; 1443 break; 1444 1445 case Builtin::BI__sync_add_and_fetch: 1446 case Builtin::BI__sync_add_and_fetch_1: 1447 case Builtin::BI__sync_add_and_fetch_2: 1448 case Builtin::BI__sync_add_and_fetch_4: 1449 case Builtin::BI__sync_add_and_fetch_8: 1450 case Builtin::BI__sync_add_and_fetch_16: 1451 BuiltinIndex = 5; 1452 break; 1453 1454 case Builtin::BI__sync_sub_and_fetch: 1455 case Builtin::BI__sync_sub_and_fetch_1: 1456 case Builtin::BI__sync_sub_and_fetch_2: 1457 case Builtin::BI__sync_sub_and_fetch_4: 1458 case Builtin::BI__sync_sub_and_fetch_8: 1459 case Builtin::BI__sync_sub_and_fetch_16: 1460 BuiltinIndex = 6; 1461 break; 1462 1463 case Builtin::BI__sync_and_and_fetch: 1464 case Builtin::BI__sync_and_and_fetch_1: 1465 case Builtin::BI__sync_and_and_fetch_2: 1466 case Builtin::BI__sync_and_and_fetch_4: 1467 case Builtin::BI__sync_and_and_fetch_8: 1468 case Builtin::BI__sync_and_and_fetch_16: 1469 BuiltinIndex = 7; 1470 break; 1471 1472 case Builtin::BI__sync_or_and_fetch: 1473 case Builtin::BI__sync_or_and_fetch_1: 1474 case Builtin::BI__sync_or_and_fetch_2: 1475 case Builtin::BI__sync_or_and_fetch_4: 1476 case Builtin::BI__sync_or_and_fetch_8: 1477 case Builtin::BI__sync_or_and_fetch_16: 1478 BuiltinIndex = 8; 1479 break; 1480 1481 case Builtin::BI__sync_xor_and_fetch: 1482 case Builtin::BI__sync_xor_and_fetch_1: 1483 case Builtin::BI__sync_xor_and_fetch_2: 1484 case Builtin::BI__sync_xor_and_fetch_4: 1485 case Builtin::BI__sync_xor_and_fetch_8: 1486 case Builtin::BI__sync_xor_and_fetch_16: 1487 BuiltinIndex = 9; 1488 break; 1489 1490 case Builtin::BI__sync_val_compare_and_swap: 1491 case Builtin::BI__sync_val_compare_and_swap_1: 1492 case Builtin::BI__sync_val_compare_and_swap_2: 1493 case Builtin::BI__sync_val_compare_and_swap_4: 1494 case Builtin::BI__sync_val_compare_and_swap_8: 1495 case Builtin::BI__sync_val_compare_and_swap_16: 1496 BuiltinIndex = 10; 1497 NumFixed = 2; 1498 break; 1499 1500 case Builtin::BI__sync_bool_compare_and_swap: 1501 case Builtin::BI__sync_bool_compare_and_swap_1: 1502 case Builtin::BI__sync_bool_compare_and_swap_2: 1503 case Builtin::BI__sync_bool_compare_and_swap_4: 1504 case Builtin::BI__sync_bool_compare_and_swap_8: 1505 case Builtin::BI__sync_bool_compare_and_swap_16: 1506 BuiltinIndex = 11; 1507 NumFixed = 2; 1508 ResultType = Context.BoolTy; 1509 break; 1510 1511 case Builtin::BI__sync_lock_test_and_set: 1512 case Builtin::BI__sync_lock_test_and_set_1: 1513 case Builtin::BI__sync_lock_test_and_set_2: 1514 case Builtin::BI__sync_lock_test_and_set_4: 1515 case Builtin::BI__sync_lock_test_and_set_8: 1516 case Builtin::BI__sync_lock_test_and_set_16: 1517 BuiltinIndex = 12; 1518 break; 1519 1520 case Builtin::BI__sync_lock_release: 1521 case Builtin::BI__sync_lock_release_1: 1522 case Builtin::BI__sync_lock_release_2: 1523 case Builtin::BI__sync_lock_release_4: 1524 case Builtin::BI__sync_lock_release_8: 1525 case Builtin::BI__sync_lock_release_16: 1526 BuiltinIndex = 13; 1527 NumFixed = 0; 1528 ResultType = Context.VoidTy; 1529 break; 1530 1531 case Builtin::BI__sync_swap: 1532 case Builtin::BI__sync_swap_1: 1533 case Builtin::BI__sync_swap_2: 1534 case Builtin::BI__sync_swap_4: 1535 case Builtin::BI__sync_swap_8: 1536 case Builtin::BI__sync_swap_16: 1537 BuiltinIndex = 14; 1538 break; 1539 } 1540 1541 // Now that we know how many fixed arguments we expect, first check that we 1542 // have at least that many. 1543 if (TheCall->getNumArgs() < 1+NumFixed) { 1544 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1545 << 0 << 1+NumFixed << TheCall->getNumArgs() 1546 << TheCall->getCallee()->getSourceRange(); 1547 return ExprError(); 1548 } 1549 1550 // Get the decl for the concrete builtin from this, we can tell what the 1551 // concrete integer type we should convert to is. 1552 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1553 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1554 FunctionDecl *NewBuiltinDecl; 1555 if (NewBuiltinID == BuiltinID) 1556 NewBuiltinDecl = FDecl; 1557 else { 1558 // Perform builtin lookup to avoid redeclaring it. 1559 DeclarationName DN(&Context.Idents.get(NewBuiltinName)); 1560 LookupResult Res(*this, DN, DRE->getLocStart(), LookupOrdinaryName); 1561 LookupName(Res, TUScope, /*AllowBuiltinCreation=*/true); 1562 assert(Res.getFoundDecl()); 1563 NewBuiltinDecl = dyn_cast<FunctionDecl>(Res.getFoundDecl()); 1564 if (NewBuiltinDecl == 0) 1565 return ExprError(); 1566 } 1567 1568 // The first argument --- the pointer --- has a fixed type; we 1569 // deduce the types of the rest of the arguments accordingly. Walk 1570 // the remaining arguments, converting them to the deduced value type. 1571 for (unsigned i = 0; i != NumFixed; ++i) { 1572 ExprResult Arg = TheCall->getArg(i+1); 1573 1574 // GCC does an implicit conversion to the pointer or integer ValType. This 1575 // can fail in some cases (1i -> int**), check for this error case now. 1576 // Initialize the argument. 1577 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1578 ValType, /*consume*/ false); 1579 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1580 if (Arg.isInvalid()) 1581 return ExprError(); 1582 1583 // Okay, we have something that *can* be converted to the right type. Check 1584 // to see if there is a potentially weird extension going on here. This can 1585 // happen when you do an atomic operation on something like an char* and 1586 // pass in 42. The 42 gets converted to char. This is even more strange 1587 // for things like 45.123 -> char, etc. 1588 // FIXME: Do this check. 1589 TheCall->setArg(i+1, Arg.take()); 1590 } 1591 1592 ASTContext& Context = this->getASTContext(); 1593 1594 // Create a new DeclRefExpr to refer to the new decl. 1595 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1596 Context, 1597 DRE->getQualifierLoc(), 1598 SourceLocation(), 1599 NewBuiltinDecl, 1600 /*enclosing*/ false, 1601 DRE->getLocation(), 1602 Context.BuiltinFnTy, 1603 DRE->getValueKind()); 1604 1605 // Set the callee in the CallExpr. 1606 // FIXME: This loses syntactic information. 1607 QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType()); 1608 ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy, 1609 CK_BuiltinFnToFnPtr); 1610 TheCall->setCallee(PromotedCall.take()); 1611 1612 // Change the result type of the call to match the original value type. This 1613 // is arbitrary, but the codegen for these builtins ins design to handle it 1614 // gracefully. 1615 TheCall->setType(ResultType); 1616 1617 return TheCallResult; 1618 } 1619 1620 /// CheckObjCString - Checks that the argument to the builtin 1621 /// CFString constructor is correct 1622 /// Note: It might also make sense to do the UTF-16 conversion here (would 1623 /// simplify the backend). 1624 bool Sema::CheckObjCString(Expr *Arg) { 1625 Arg = Arg->IgnoreParenCasts(); 1626 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1627 1628 if (!Literal || !Literal->isAscii()) { 1629 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1630 << Arg->getSourceRange(); 1631 return true; 1632 } 1633 1634 if (Literal->containsNonAsciiOrNull()) { 1635 StringRef String = Literal->getString(); 1636 unsigned NumBytes = String.size(); 1637 SmallVector<UTF16, 128> ToBuf(NumBytes); 1638 const UTF8 *FromPtr = (const UTF8 *)String.data(); 1639 UTF16 *ToPtr = &ToBuf[0]; 1640 1641 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1642 &ToPtr, ToPtr + NumBytes, 1643 strictConversion); 1644 // Check for conversion failure. 1645 if (Result != conversionOK) 1646 Diag(Arg->getLocStart(), 1647 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1648 } 1649 return false; 1650 } 1651 1652 /// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1653 /// Emit an error and return true on failure, return false on success. 1654 bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1655 Expr *Fn = TheCall->getCallee(); 1656 if (TheCall->getNumArgs() > 2) { 1657 Diag(TheCall->getArg(2)->getLocStart(), 1658 diag::err_typecheck_call_too_many_args) 1659 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1660 << Fn->getSourceRange() 1661 << SourceRange(TheCall->getArg(2)->getLocStart(), 1662 (*(TheCall->arg_end()-1))->getLocEnd()); 1663 return true; 1664 } 1665 1666 if (TheCall->getNumArgs() < 2) { 1667 return Diag(TheCall->getLocEnd(), 1668 diag::err_typecheck_call_too_few_args_at_least) 1669 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1670 } 1671 1672 // Type-check the first argument normally. 1673 if (checkBuiltinArgument(*this, TheCall, 0)) 1674 return true; 1675 1676 // Determine whether the current function is variadic or not. 1677 BlockScopeInfo *CurBlock = getCurBlock(); 1678 bool isVariadic; 1679 if (CurBlock) 1680 isVariadic = CurBlock->TheDecl->isVariadic(); 1681 else if (FunctionDecl *FD = getCurFunctionDecl()) 1682 isVariadic = FD->isVariadic(); 1683 else 1684 isVariadic = getCurMethodDecl()->isVariadic(); 1685 1686 if (!isVariadic) { 1687 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1688 return true; 1689 } 1690 1691 // Verify that the second argument to the builtin is the last argument of the 1692 // current function or method. 1693 bool SecondArgIsLastNamedArgument = false; 1694 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1695 1696 // These are valid if SecondArgIsLastNamedArgument is false after the next 1697 // block. 1698 QualType Type; 1699 SourceLocation ParamLoc; 1700 1701 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1702 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1703 // FIXME: This isn't correct for methods (results in bogus warning). 1704 // Get the last formal in the current function. 1705 const ParmVarDecl *LastArg; 1706 if (CurBlock) 1707 LastArg = *(CurBlock->TheDecl->param_end()-1); 1708 else if (FunctionDecl *FD = getCurFunctionDecl()) 1709 LastArg = *(FD->param_end()-1); 1710 else 1711 LastArg = *(getCurMethodDecl()->param_end()-1); 1712 SecondArgIsLastNamedArgument = PV == LastArg; 1713 1714 Type = PV->getType(); 1715 ParamLoc = PV->getLocation(); 1716 } 1717 } 1718 1719 if (!SecondArgIsLastNamedArgument) 1720 Diag(TheCall->getArg(1)->getLocStart(), 1721 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1722 else if (Type->isReferenceType()) { 1723 Diag(Arg->getLocStart(), 1724 diag::warn_va_start_of_reference_type_is_undefined); 1725 Diag(ParamLoc, diag::note_parameter_type) << Type; 1726 } 1727 1728 TheCall->setType(Context.VoidTy); 1729 return false; 1730 } 1731 1732 /// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1733 /// friends. This is declared to take (...), so we have to check everything. 1734 bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1735 if (TheCall->getNumArgs() < 2) 1736 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1737 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1738 if (TheCall->getNumArgs() > 2) 1739 return Diag(TheCall->getArg(2)->getLocStart(), 1740 diag::err_typecheck_call_too_many_args) 1741 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1742 << SourceRange(TheCall->getArg(2)->getLocStart(), 1743 (*(TheCall->arg_end()-1))->getLocEnd()); 1744 1745 ExprResult OrigArg0 = TheCall->getArg(0); 1746 ExprResult OrigArg1 = TheCall->getArg(1); 1747 1748 // Do standard promotions between the two arguments, returning their common 1749 // type. 1750 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1751 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1752 return true; 1753 1754 // Make sure any conversions are pushed back into the call; this is 1755 // type safe since unordered compare builtins are declared as "_Bool 1756 // foo(...)". 1757 TheCall->setArg(0, OrigArg0.get()); 1758 TheCall->setArg(1, OrigArg1.get()); 1759 1760 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1761 return false; 1762 1763 // If the common type isn't a real floating type, then the arguments were 1764 // invalid for this operation. 1765 if (Res.isNull() || !Res->isRealFloatingType()) 1766 return Diag(OrigArg0.get()->getLocStart(), 1767 diag::err_typecheck_call_invalid_ordered_compare) 1768 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1769 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1770 1771 return false; 1772 } 1773 1774 /// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1775 /// __builtin_isnan and friends. This is declared to take (...), so we have 1776 /// to check everything. We expect the last argument to be a floating point 1777 /// value. 1778 bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1779 if (TheCall->getNumArgs() < NumArgs) 1780 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1781 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1782 if (TheCall->getNumArgs() > NumArgs) 1783 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1784 diag::err_typecheck_call_too_many_args) 1785 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1786 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1787 (*(TheCall->arg_end()-1))->getLocEnd()); 1788 1789 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1790 1791 if (OrigArg->isTypeDependent()) 1792 return false; 1793 1794 // This operation requires a non-_Complex floating-point number. 1795 if (!OrigArg->getType()->isRealFloatingType()) 1796 return Diag(OrigArg->getLocStart(), 1797 diag::err_typecheck_call_invalid_unary_fp) 1798 << OrigArg->getType() << OrigArg->getSourceRange(); 1799 1800 // If this is an implicit conversion from float -> double, remove it. 1801 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1802 Expr *CastArg = Cast->getSubExpr(); 1803 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1804 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1805 "promotion from float to double is the only expected cast here"); 1806 Cast->setSubExpr(0); 1807 TheCall->setArg(NumArgs-1, CastArg); 1808 } 1809 } 1810 1811 return false; 1812 } 1813 1814 /// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1815 // This is declared to take (...), so we have to check everything. 1816 ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1817 if (TheCall->getNumArgs() < 2) 1818 return ExprError(Diag(TheCall->getLocEnd(), 1819 diag::err_typecheck_call_too_few_args_at_least) 1820 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1821 << TheCall->getSourceRange()); 1822 1823 // Determine which of the following types of shufflevector we're checking: 1824 // 1) unary, vector mask: (lhs, mask) 1825 // 2) binary, vector mask: (lhs, rhs, mask) 1826 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1827 QualType resType = TheCall->getArg(0)->getType(); 1828 unsigned numElements = 0; 1829 1830 if (!TheCall->getArg(0)->isTypeDependent() && 1831 !TheCall->getArg(1)->isTypeDependent()) { 1832 QualType LHSType = TheCall->getArg(0)->getType(); 1833 QualType RHSType = TheCall->getArg(1)->getType(); 1834 1835 if (!LHSType->isVectorType() || !RHSType->isVectorType()) 1836 return ExprError(Diag(TheCall->getLocStart(), 1837 diag::err_shufflevector_non_vector) 1838 << SourceRange(TheCall->getArg(0)->getLocStart(), 1839 TheCall->getArg(1)->getLocEnd())); 1840 1841 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1842 unsigned numResElements = TheCall->getNumArgs() - 2; 1843 1844 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1845 // with mask. If so, verify that RHS is an integer vector type with the 1846 // same number of elts as lhs. 1847 if (TheCall->getNumArgs() == 2) { 1848 if (!RHSType->hasIntegerRepresentation() || 1849 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1850 return ExprError(Diag(TheCall->getLocStart(), 1851 diag::err_shufflevector_incompatible_vector) 1852 << SourceRange(TheCall->getArg(1)->getLocStart(), 1853 TheCall->getArg(1)->getLocEnd())); 1854 } else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1855 return ExprError(Diag(TheCall->getLocStart(), 1856 diag::err_shufflevector_incompatible_vector) 1857 << SourceRange(TheCall->getArg(0)->getLocStart(), 1858 TheCall->getArg(1)->getLocEnd())); 1859 } else if (numElements != numResElements) { 1860 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1861 resType = Context.getVectorType(eltType, numResElements, 1862 VectorType::GenericVector); 1863 } 1864 } 1865 1866 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1867 if (TheCall->getArg(i)->isTypeDependent() || 1868 TheCall->getArg(i)->isValueDependent()) 1869 continue; 1870 1871 llvm::APSInt Result(32); 1872 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1873 return ExprError(Diag(TheCall->getLocStart(), 1874 diag::err_shufflevector_nonconstant_argument) 1875 << TheCall->getArg(i)->getSourceRange()); 1876 1877 // Allow -1 which will be translated to undef in the IR. 1878 if (Result.isSigned() && Result.isAllOnesValue()) 1879 continue; 1880 1881 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1882 return ExprError(Diag(TheCall->getLocStart(), 1883 diag::err_shufflevector_argument_too_large) 1884 << TheCall->getArg(i)->getSourceRange()); 1885 } 1886 1887 SmallVector<Expr*, 32> exprs; 1888 1889 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1890 exprs.push_back(TheCall->getArg(i)); 1891 TheCall->setArg(i, 0); 1892 } 1893 1894 return Owned(new (Context) ShuffleVectorExpr(Context, exprs, resType, 1895 TheCall->getCallee()->getLocStart(), 1896 TheCall->getRParenLoc())); 1897 } 1898 1899 /// SemaConvertVectorExpr - Handle __builtin_convertvector 1900 ExprResult Sema::SemaConvertVectorExpr(Expr *E, TypeSourceInfo *TInfo, 1901 SourceLocation BuiltinLoc, 1902 SourceLocation RParenLoc) { 1903 ExprValueKind VK = VK_RValue; 1904 ExprObjectKind OK = OK_Ordinary; 1905 QualType DstTy = TInfo->getType(); 1906 QualType SrcTy = E->getType(); 1907 1908 if (!SrcTy->isVectorType() && !SrcTy->isDependentType()) 1909 return ExprError(Diag(BuiltinLoc, 1910 diag::err_convertvector_non_vector) 1911 << E->getSourceRange()); 1912 if (!DstTy->isVectorType() && !DstTy->isDependentType()) 1913 return ExprError(Diag(BuiltinLoc, 1914 diag::err_convertvector_non_vector_type)); 1915 1916 if (!SrcTy->isDependentType() && !DstTy->isDependentType()) { 1917 unsigned SrcElts = SrcTy->getAs<VectorType>()->getNumElements(); 1918 unsigned DstElts = DstTy->getAs<VectorType>()->getNumElements(); 1919 if (SrcElts != DstElts) 1920 return ExprError(Diag(BuiltinLoc, 1921 diag::err_convertvector_incompatible_vector) 1922 << E->getSourceRange()); 1923 } 1924 1925 return Owned(new (Context) ConvertVectorExpr(E, TInfo, DstTy, VK, OK, 1926 BuiltinLoc, RParenLoc)); 1927 1928 } 1929 1930 /// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1931 // This is declared to take (const void*, ...) and can take two 1932 // optional constant int args. 1933 bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1934 unsigned NumArgs = TheCall->getNumArgs(); 1935 1936 if (NumArgs > 3) 1937 return Diag(TheCall->getLocEnd(), 1938 diag::err_typecheck_call_too_many_args_at_most) 1939 << 0 /*function call*/ << 3 << NumArgs 1940 << TheCall->getSourceRange(); 1941 1942 // Argument 0 is checked for us and the remaining arguments must be 1943 // constant integers. 1944 for (unsigned i = 1; i != NumArgs; ++i) { 1945 Expr *Arg = TheCall->getArg(i); 1946 1947 // We can't check the value of a dependent argument. 1948 if (Arg->isTypeDependent() || Arg->isValueDependent()) 1949 continue; 1950 1951 llvm::APSInt Result; 1952 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1953 return true; 1954 1955 // FIXME: gcc issues a warning and rewrites these to 0. These 1956 // seems especially odd for the third argument since the default 1957 // is 3. 1958 if (i == 1) { 1959 if (Result.getLimitedValue() > 1) 1960 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1961 << "0" << "1" << Arg->getSourceRange(); 1962 } else { 1963 if (Result.getLimitedValue() > 3) 1964 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1965 << "0" << "3" << Arg->getSourceRange(); 1966 } 1967 } 1968 1969 return false; 1970 } 1971 1972 /// SemaBuiltinMMPrefetch - Handle _mm_prefetch. 1973 // This is declared to take (const char*, int) 1974 bool Sema::SemaBuiltinMMPrefetch(CallExpr *TheCall) { 1975 Expr *Arg = TheCall->getArg(1); 1976 1977 // We can't check the value of a dependent argument. 1978 if (Arg->isTypeDependent() || Arg->isValueDependent()) 1979 return false; 1980 1981 llvm::APSInt Result; 1982 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1983 return true; 1984 1985 if (Result.getLimitedValue() > 3) 1986 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1987 << "0" << "3" << Arg->getSourceRange(); 1988 1989 return false; 1990 } 1991 1992 /// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1993 /// TheCall is a constant expression. 1994 bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1995 llvm::APSInt &Result) { 1996 Expr *Arg = TheCall->getArg(ArgNum); 1997 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1998 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1999 2000 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 2001 2002 if (!Arg->isIntegerConstantExpr(Result, Context)) 2003 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 2004 << FDecl->getDeclName() << Arg->getSourceRange(); 2005 2006 return false; 2007 } 2008 2009 /// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 2010 /// int type). This simply type checks that type is one of the defined 2011 /// constants (0-3). 2012 // For compatibility check 0-3, llvm only handles 0 and 2. 2013 bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 2014 llvm::APSInt Result; 2015 2016 // We can't check the value of a dependent argument. 2017 if (TheCall->getArg(1)->isTypeDependent() || 2018 TheCall->getArg(1)->isValueDependent()) 2019 return false; 2020 2021 // Check constant-ness first. 2022 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 2023 return true; 2024 2025 Expr *Arg = TheCall->getArg(1); 2026 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 2027 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 2028 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 2029 } 2030 2031 return false; 2032 } 2033 2034 /// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 2035 /// This checks that val is a constant 1. 2036 bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 2037 Expr *Arg = TheCall->getArg(1); 2038 llvm::APSInt Result; 2039 2040 // TODO: This is less than ideal. Overload this to take a value. 2041 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 2042 return true; 2043 2044 if (Result != 1) 2045 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 2046 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 2047 2048 return false; 2049 } 2050 2051 namespace { 2052 enum StringLiteralCheckType { 2053 SLCT_NotALiteral, 2054 SLCT_UncheckedLiteral, 2055 SLCT_CheckedLiteral 2056 }; 2057 } 2058 2059 // Determine if an expression is a string literal or constant string. 2060 // If this function returns false on the arguments to a function expecting a 2061 // format string, we will usually need to emit a warning. 2062 // True string literals are then checked by CheckFormatString. 2063 static StringLiteralCheckType 2064 checkFormatStringExpr(Sema &S, const Expr *E, ArrayRef<const Expr *> Args, 2065 bool HasVAListArg, unsigned format_idx, 2066 unsigned firstDataArg, Sema::FormatStringType Type, 2067 Sema::VariadicCallType CallType, bool InFunctionCall, 2068 llvm::SmallBitVector &CheckedVarArgs) { 2069 tryAgain: 2070 if (E->isTypeDependent() || E->isValueDependent()) 2071 return SLCT_NotALiteral; 2072 2073 E = E->IgnoreParenCasts(); 2074 2075 if (E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull)) 2076 // Technically -Wformat-nonliteral does not warn about this case. 2077 // The behavior of printf and friends in this case is implementation 2078 // dependent. Ideally if the format string cannot be null then 2079 // it should have a 'nonnull' attribute in the function prototype. 2080 return SLCT_UncheckedLiteral; 2081 2082 switch (E->getStmtClass()) { 2083 case Stmt::BinaryConditionalOperatorClass: 2084 case Stmt::ConditionalOperatorClass: { 2085 // The expression is a literal if both sub-expressions were, and it was 2086 // completely checked only if both sub-expressions were checked. 2087 const AbstractConditionalOperator *C = 2088 cast<AbstractConditionalOperator>(E); 2089 StringLiteralCheckType Left = 2090 checkFormatStringExpr(S, C->getTrueExpr(), Args, 2091 HasVAListArg, format_idx, firstDataArg, 2092 Type, CallType, InFunctionCall, CheckedVarArgs); 2093 if (Left == SLCT_NotALiteral) 2094 return SLCT_NotALiteral; 2095 StringLiteralCheckType Right = 2096 checkFormatStringExpr(S, C->getFalseExpr(), Args, 2097 HasVAListArg, format_idx, firstDataArg, 2098 Type, CallType, InFunctionCall, CheckedVarArgs); 2099 return Left < Right ? Left : Right; 2100 } 2101 2102 case Stmt::ImplicitCastExprClass: { 2103 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 2104 goto tryAgain; 2105 } 2106 2107 case Stmt::OpaqueValueExprClass: 2108 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 2109 E = src; 2110 goto tryAgain; 2111 } 2112 return SLCT_NotALiteral; 2113 2114 case Stmt::PredefinedExprClass: 2115 // While __func__, etc., are technically not string literals, they 2116 // cannot contain format specifiers and thus are not a security 2117 // liability. 2118 return SLCT_UncheckedLiteral; 2119 2120 case Stmt::DeclRefExprClass: { 2121 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 2122 2123 // As an exception, do not flag errors for variables binding to 2124 // const string literals. 2125 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 2126 bool isConstant = false; 2127 QualType T = DR->getType(); 2128 2129 if (const ArrayType *AT = S.Context.getAsArrayType(T)) { 2130 isConstant = AT->getElementType().isConstant(S.Context); 2131 } else if (const PointerType *PT = T->getAs<PointerType>()) { 2132 isConstant = T.isConstant(S.Context) && 2133 PT->getPointeeType().isConstant(S.Context); 2134 } else if (T->isObjCObjectPointerType()) { 2135 // In ObjC, there is usually no "const ObjectPointer" type, 2136 // so don't check if the pointee type is constant. 2137 isConstant = T.isConstant(S.Context); 2138 } 2139 2140 if (isConstant) { 2141 if (const Expr *Init = VD->getAnyInitializer()) { 2142 // Look through initializers like const char c[] = { "foo" } 2143 if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) { 2144 if (InitList->isStringLiteralInit()) 2145 Init = InitList->getInit(0)->IgnoreParenImpCasts(); 2146 } 2147 return checkFormatStringExpr(S, Init, Args, 2148 HasVAListArg, format_idx, 2149 firstDataArg, Type, CallType, 2150 /*InFunctionCall*/false, CheckedVarArgs); 2151 } 2152 } 2153 2154 // For vprintf* functions (i.e., HasVAListArg==true), we add a 2155 // special check to see if the format string is a function parameter 2156 // of the function calling the printf function. If the function 2157 // has an attribute indicating it is a printf-like function, then we 2158 // should suppress warnings concerning non-literals being used in a call 2159 // to a vprintf function. For example: 2160 // 2161 // void 2162 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 2163 // va_list ap; 2164 // va_start(ap, fmt); 2165 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 2166 // ... 2167 // } 2168 if (HasVAListArg) { 2169 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 2170 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 2171 int PVIndex = PV->getFunctionScopeIndex() + 1; 2172 for (const auto *PVFormat : ND->specific_attrs<FormatAttr>()) { 2173 // adjust for implicit parameter 2174 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 2175 if (MD->isInstance()) 2176 ++PVIndex; 2177 // We also check if the formats are compatible. 2178 // We can't pass a 'scanf' string to a 'printf' function. 2179 if (PVIndex == PVFormat->getFormatIdx() && 2180 Type == S.GetFormatStringType(PVFormat)) 2181 return SLCT_UncheckedLiteral; 2182 } 2183 } 2184 } 2185 } 2186 } 2187 2188 return SLCT_NotALiteral; 2189 } 2190 2191 case Stmt::CallExprClass: 2192 case Stmt::CXXMemberCallExprClass: { 2193 const CallExpr *CE = cast<CallExpr>(E); 2194 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 2195 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 2196 unsigned ArgIndex = FA->getFormatIdx(); 2197 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 2198 if (MD->isInstance()) 2199 --ArgIndex; 2200 const Expr *Arg = CE->getArg(ArgIndex - 1); 2201 2202 return checkFormatStringExpr(S, Arg, Args, 2203 HasVAListArg, format_idx, firstDataArg, 2204 Type, CallType, InFunctionCall, 2205 CheckedVarArgs); 2206 } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) { 2207 unsigned BuiltinID = FD->getBuiltinID(); 2208 if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString || 2209 BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) { 2210 const Expr *Arg = CE->getArg(0); 2211 return checkFormatStringExpr(S, Arg, Args, 2212 HasVAListArg, format_idx, 2213 firstDataArg, Type, CallType, 2214 InFunctionCall, CheckedVarArgs); 2215 } 2216 } 2217 } 2218 2219 return SLCT_NotALiteral; 2220 } 2221 case Stmt::ObjCStringLiteralClass: 2222 case Stmt::StringLiteralClass: { 2223 const StringLiteral *StrE = NULL; 2224 2225 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 2226 StrE = ObjCFExpr->getString(); 2227 else 2228 StrE = cast<StringLiteral>(E); 2229 2230 if (StrE) { 2231 S.CheckFormatString(StrE, E, Args, HasVAListArg, format_idx, firstDataArg, 2232 Type, InFunctionCall, CallType, CheckedVarArgs); 2233 return SLCT_CheckedLiteral; 2234 } 2235 2236 return SLCT_NotALiteral; 2237 } 2238 2239 default: 2240 return SLCT_NotALiteral; 2241 } 2242 } 2243 2244 Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 2245 return llvm::StringSwitch<FormatStringType>(Format->getType()->getName()) 2246 .Case("scanf", FST_Scanf) 2247 .Cases("printf", "printf0", FST_Printf) 2248 .Cases("NSString", "CFString", FST_NSString) 2249 .Case("strftime", FST_Strftime) 2250 .Case("strfmon", FST_Strfmon) 2251 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 2252 .Default(FST_Unknown); 2253 } 2254 2255 /// CheckFormatArguments - Check calls to printf and scanf (and similar 2256 /// functions) for correct use of format strings. 2257 /// Returns true if a format string has been fully checked. 2258 bool Sema::CheckFormatArguments(const FormatAttr *Format, 2259 ArrayRef<const Expr *> Args, 2260 bool IsCXXMember, 2261 VariadicCallType CallType, 2262 SourceLocation Loc, SourceRange Range, 2263 llvm::SmallBitVector &CheckedVarArgs) { 2264 FormatStringInfo FSI; 2265 if (getFormatStringInfo(Format, IsCXXMember, &FSI)) 2266 return CheckFormatArguments(Args, FSI.HasVAListArg, FSI.FormatIdx, 2267 FSI.FirstDataArg, GetFormatStringType(Format), 2268 CallType, Loc, Range, CheckedVarArgs); 2269 return false; 2270 } 2271 2272 bool Sema::CheckFormatArguments(ArrayRef<const Expr *> Args, 2273 bool HasVAListArg, unsigned format_idx, 2274 unsigned firstDataArg, FormatStringType Type, 2275 VariadicCallType CallType, 2276 SourceLocation Loc, SourceRange Range, 2277 llvm::SmallBitVector &CheckedVarArgs) { 2278 // CHECK: printf/scanf-like function is called with no format string. 2279 if (format_idx >= Args.size()) { 2280 Diag(Loc, diag::warn_missing_format_string) << Range; 2281 return false; 2282 } 2283 2284 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 2285 2286 // CHECK: format string is not a string literal. 2287 // 2288 // Dynamically generated format strings are difficult to 2289 // automatically vet at compile time. Requiring that format strings 2290 // are string literals: (1) permits the checking of format strings by 2291 // the compiler and thereby (2) can practically remove the source of 2292 // many format string exploits. 2293 2294 // Format string can be either ObjC string (e.g. @"%d") or 2295 // C string (e.g. "%d") 2296 // ObjC string uses the same format specifiers as C string, so we can use 2297 // the same format string checking logic for both ObjC and C strings. 2298 StringLiteralCheckType CT = 2299 checkFormatStringExpr(*this, OrigFormatExpr, Args, HasVAListArg, 2300 format_idx, firstDataArg, Type, CallType, 2301 /*IsFunctionCall*/true, CheckedVarArgs); 2302 if (CT != SLCT_NotALiteral) 2303 // Literal format string found, check done! 2304 return CT == SLCT_CheckedLiteral; 2305 2306 // Strftime is particular as it always uses a single 'time' argument, 2307 // so it is safe to pass a non-literal string. 2308 if (Type == FST_Strftime) 2309 return false; 2310 2311 // Do not emit diag when the string param is a macro expansion and the 2312 // format is either NSString or CFString. This is a hack to prevent 2313 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 2314 // which are usually used in place of NS and CF string literals. 2315 if (Type == FST_NSString && 2316 SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart())) 2317 return false; 2318 2319 // If there are no arguments specified, warn with -Wformat-security, otherwise 2320 // warn only with -Wformat-nonliteral. 2321 if (Args.size() == firstDataArg) 2322 Diag(Args[format_idx]->getLocStart(), 2323 diag::warn_format_nonliteral_noargs) 2324 << OrigFormatExpr->getSourceRange(); 2325 else 2326 Diag(Args[format_idx]->getLocStart(), 2327 diag::warn_format_nonliteral) 2328 << OrigFormatExpr->getSourceRange(); 2329 return false; 2330 } 2331 2332 namespace { 2333 class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 2334 protected: 2335 Sema &S; 2336 const StringLiteral *FExpr; 2337 const Expr *OrigFormatExpr; 2338 const unsigned FirstDataArg; 2339 const unsigned NumDataArgs; 2340 const char *Beg; // Start of format string. 2341 const bool HasVAListArg; 2342 ArrayRef<const Expr *> Args; 2343 unsigned FormatIdx; 2344 llvm::SmallBitVector CoveredArgs; 2345 bool usesPositionalArgs; 2346 bool atFirstArg; 2347 bool inFunctionCall; 2348 Sema::VariadicCallType CallType; 2349 llvm::SmallBitVector &CheckedVarArgs; 2350 public: 2351 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 2352 const Expr *origFormatExpr, unsigned firstDataArg, 2353 unsigned numDataArgs, const char *beg, bool hasVAListArg, 2354 ArrayRef<const Expr *> Args, 2355 unsigned formatIdx, bool inFunctionCall, 2356 Sema::VariadicCallType callType, 2357 llvm::SmallBitVector &CheckedVarArgs) 2358 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 2359 FirstDataArg(firstDataArg), NumDataArgs(numDataArgs), 2360 Beg(beg), HasVAListArg(hasVAListArg), 2361 Args(Args), FormatIdx(formatIdx), 2362 usesPositionalArgs(false), atFirstArg(true), 2363 inFunctionCall(inFunctionCall), CallType(callType), 2364 CheckedVarArgs(CheckedVarArgs) { 2365 CoveredArgs.resize(numDataArgs); 2366 CoveredArgs.reset(); 2367 } 2368 2369 void DoneProcessing(); 2370 2371 void HandleIncompleteSpecifier(const char *startSpecifier, 2372 unsigned specifierLen) override; 2373 2374 void HandleInvalidLengthModifier( 2375 const analyze_format_string::FormatSpecifier &FS, 2376 const analyze_format_string::ConversionSpecifier &CS, 2377 const char *startSpecifier, unsigned specifierLen, 2378 unsigned DiagID); 2379 2380 void HandleNonStandardLengthModifier( 2381 const analyze_format_string::FormatSpecifier &FS, 2382 const char *startSpecifier, unsigned specifierLen); 2383 2384 void HandleNonStandardConversionSpecifier( 2385 const analyze_format_string::ConversionSpecifier &CS, 2386 const char *startSpecifier, unsigned specifierLen); 2387 2388 void HandlePosition(const char *startPos, unsigned posLen) override; 2389 2390 void HandleInvalidPosition(const char *startSpecifier, 2391 unsigned specifierLen, 2392 analyze_format_string::PositionContext p) override; 2393 2394 void HandleZeroPosition(const char *startPos, unsigned posLen) override; 2395 2396 void HandleNullChar(const char *nullCharacter) override; 2397 2398 template <typename Range> 2399 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 2400 const Expr *ArgumentExpr, 2401 PartialDiagnostic PDiag, 2402 SourceLocation StringLoc, 2403 bool IsStringLocation, Range StringRange, 2404 ArrayRef<FixItHint> Fixit = None); 2405 2406 protected: 2407 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 2408 const char *startSpec, 2409 unsigned specifierLen, 2410 const char *csStart, unsigned csLen); 2411 2412 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 2413 const char *startSpec, 2414 unsigned specifierLen); 2415 2416 SourceRange getFormatStringRange(); 2417 CharSourceRange getSpecifierRange(const char *startSpecifier, 2418 unsigned specifierLen); 2419 SourceLocation getLocationOfByte(const char *x); 2420 2421 const Expr *getDataArg(unsigned i) const; 2422 2423 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 2424 const analyze_format_string::ConversionSpecifier &CS, 2425 const char *startSpecifier, unsigned specifierLen, 2426 unsigned argIndex); 2427 2428 template <typename Range> 2429 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 2430 bool IsStringLocation, Range StringRange, 2431 ArrayRef<FixItHint> Fixit = None); 2432 }; 2433 } 2434 2435 SourceRange CheckFormatHandler::getFormatStringRange() { 2436 return OrigFormatExpr->getSourceRange(); 2437 } 2438 2439 CharSourceRange CheckFormatHandler:: 2440 getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 2441 SourceLocation Start = getLocationOfByte(startSpecifier); 2442 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 2443 2444 // Advance the end SourceLocation by one due to half-open ranges. 2445 End = End.getLocWithOffset(1); 2446 2447 return CharSourceRange::getCharRange(Start, End); 2448 } 2449 2450 SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 2451 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 2452 } 2453 2454 void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 2455 unsigned specifierLen){ 2456 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 2457 getLocationOfByte(startSpecifier), 2458 /*IsStringLocation*/true, 2459 getSpecifierRange(startSpecifier, specifierLen)); 2460 } 2461 2462 void CheckFormatHandler::HandleInvalidLengthModifier( 2463 const analyze_format_string::FormatSpecifier &FS, 2464 const analyze_format_string::ConversionSpecifier &CS, 2465 const char *startSpecifier, unsigned specifierLen, unsigned DiagID) { 2466 using namespace analyze_format_string; 2467 2468 const LengthModifier &LM = FS.getLengthModifier(); 2469 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2470 2471 // See if we know how to fix this length modifier. 2472 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2473 if (FixedLM) { 2474 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2475 getLocationOfByte(LM.getStart()), 2476 /*IsStringLocation*/true, 2477 getSpecifierRange(startSpecifier, specifierLen)); 2478 2479 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2480 << FixedLM->toString() 2481 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2482 2483 } else { 2484 FixItHint Hint; 2485 if (DiagID == diag::warn_format_nonsensical_length) 2486 Hint = FixItHint::CreateRemoval(LMRange); 2487 2488 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2489 getLocationOfByte(LM.getStart()), 2490 /*IsStringLocation*/true, 2491 getSpecifierRange(startSpecifier, specifierLen), 2492 Hint); 2493 } 2494 } 2495 2496 void CheckFormatHandler::HandleNonStandardLengthModifier( 2497 const analyze_format_string::FormatSpecifier &FS, 2498 const char *startSpecifier, unsigned specifierLen) { 2499 using namespace analyze_format_string; 2500 2501 const LengthModifier &LM = FS.getLengthModifier(); 2502 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2503 2504 // See if we know how to fix this length modifier. 2505 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2506 if (FixedLM) { 2507 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2508 << LM.toString() << 0, 2509 getLocationOfByte(LM.getStart()), 2510 /*IsStringLocation*/true, 2511 getSpecifierRange(startSpecifier, specifierLen)); 2512 2513 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2514 << FixedLM->toString() 2515 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2516 2517 } else { 2518 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2519 << LM.toString() << 0, 2520 getLocationOfByte(LM.getStart()), 2521 /*IsStringLocation*/true, 2522 getSpecifierRange(startSpecifier, specifierLen)); 2523 } 2524 } 2525 2526 void CheckFormatHandler::HandleNonStandardConversionSpecifier( 2527 const analyze_format_string::ConversionSpecifier &CS, 2528 const char *startSpecifier, unsigned specifierLen) { 2529 using namespace analyze_format_string; 2530 2531 // See if we know how to fix this conversion specifier. 2532 Optional<ConversionSpecifier> FixedCS = CS.getStandardSpecifier(); 2533 if (FixedCS) { 2534 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2535 << CS.toString() << /*conversion specifier*/1, 2536 getLocationOfByte(CS.getStart()), 2537 /*IsStringLocation*/true, 2538 getSpecifierRange(startSpecifier, specifierLen)); 2539 2540 CharSourceRange CSRange = getSpecifierRange(CS.getStart(), CS.getLength()); 2541 S.Diag(getLocationOfByte(CS.getStart()), diag::note_format_fix_specifier) 2542 << FixedCS->toString() 2543 << FixItHint::CreateReplacement(CSRange, FixedCS->toString()); 2544 } else { 2545 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2546 << CS.toString() << /*conversion specifier*/1, 2547 getLocationOfByte(CS.getStart()), 2548 /*IsStringLocation*/true, 2549 getSpecifierRange(startSpecifier, specifierLen)); 2550 } 2551 } 2552 2553 void CheckFormatHandler::HandlePosition(const char *startPos, 2554 unsigned posLen) { 2555 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 2556 getLocationOfByte(startPos), 2557 /*IsStringLocation*/true, 2558 getSpecifierRange(startPos, posLen)); 2559 } 2560 2561 void 2562 CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 2563 analyze_format_string::PositionContext p) { 2564 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 2565 << (unsigned) p, 2566 getLocationOfByte(startPos), /*IsStringLocation*/true, 2567 getSpecifierRange(startPos, posLen)); 2568 } 2569 2570 void CheckFormatHandler::HandleZeroPosition(const char *startPos, 2571 unsigned posLen) { 2572 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 2573 getLocationOfByte(startPos), 2574 /*IsStringLocation*/true, 2575 getSpecifierRange(startPos, posLen)); 2576 } 2577 2578 void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 2579 if (!isa<ObjCStringLiteral>(OrigFormatExpr)) { 2580 // The presence of a null character is likely an error. 2581 EmitFormatDiagnostic( 2582 S.PDiag(diag::warn_printf_format_string_contains_null_char), 2583 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 2584 getFormatStringRange()); 2585 } 2586 } 2587 2588 // Note that this may return NULL if there was an error parsing or building 2589 // one of the argument expressions. 2590 const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 2591 return Args[FirstDataArg + i]; 2592 } 2593 2594 void CheckFormatHandler::DoneProcessing() { 2595 // Does the number of data arguments exceed the number of 2596 // format conversions in the format string? 2597 if (!HasVAListArg) { 2598 // Find any arguments that weren't covered. 2599 CoveredArgs.flip(); 2600 signed notCoveredArg = CoveredArgs.find_first(); 2601 if (notCoveredArg >= 0) { 2602 assert((unsigned)notCoveredArg < NumDataArgs); 2603 if (const Expr *E = getDataArg((unsigned) notCoveredArg)) { 2604 SourceLocation Loc = E->getLocStart(); 2605 if (!S.getSourceManager().isInSystemMacro(Loc)) { 2606 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 2607 Loc, /*IsStringLocation*/false, 2608 getFormatStringRange()); 2609 } 2610 } 2611 } 2612 } 2613 } 2614 2615 bool 2616 CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 2617 SourceLocation Loc, 2618 const char *startSpec, 2619 unsigned specifierLen, 2620 const char *csStart, 2621 unsigned csLen) { 2622 2623 bool keepGoing = true; 2624 if (argIndex < NumDataArgs) { 2625 // Consider the argument coverered, even though the specifier doesn't 2626 // make sense. 2627 CoveredArgs.set(argIndex); 2628 } 2629 else { 2630 // If argIndex exceeds the number of data arguments we 2631 // don't issue a warning because that is just a cascade of warnings (and 2632 // they may have intended '%%' anyway). We don't want to continue processing 2633 // the format string after this point, however, as we will like just get 2634 // gibberish when trying to match arguments. 2635 keepGoing = false; 2636 } 2637 2638 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2639 << StringRef(csStart, csLen), 2640 Loc, /*IsStringLocation*/true, 2641 getSpecifierRange(startSpec, specifierLen)); 2642 2643 return keepGoing; 2644 } 2645 2646 void 2647 CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2648 const char *startSpec, 2649 unsigned specifierLen) { 2650 EmitFormatDiagnostic( 2651 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2652 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2653 } 2654 2655 bool 2656 CheckFormatHandler::CheckNumArgs( 2657 const analyze_format_string::FormatSpecifier &FS, 2658 const analyze_format_string::ConversionSpecifier &CS, 2659 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2660 2661 if (argIndex >= NumDataArgs) { 2662 PartialDiagnostic PDiag = FS.usesPositionalArg() 2663 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2664 << (argIndex+1) << NumDataArgs) 2665 : S.PDiag(diag::warn_printf_insufficient_data_args); 2666 EmitFormatDiagnostic( 2667 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2668 getSpecifierRange(startSpecifier, specifierLen)); 2669 return false; 2670 } 2671 return true; 2672 } 2673 2674 template<typename Range> 2675 void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2676 SourceLocation Loc, 2677 bool IsStringLocation, 2678 Range StringRange, 2679 ArrayRef<FixItHint> FixIt) { 2680 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2681 Loc, IsStringLocation, StringRange, FixIt); 2682 } 2683 2684 /// \brief If the format string is not within the funcion call, emit a note 2685 /// so that the function call and string are in diagnostic messages. 2686 /// 2687 /// \param InFunctionCall if true, the format string is within the function 2688 /// call and only one diagnostic message will be produced. Otherwise, an 2689 /// extra note will be emitted pointing to location of the format string. 2690 /// 2691 /// \param ArgumentExpr the expression that is passed as the format string 2692 /// argument in the function call. Used for getting locations when two 2693 /// diagnostics are emitted. 2694 /// 2695 /// \param PDiag the callee should already have provided any strings for the 2696 /// diagnostic message. This function only adds locations and fixits 2697 /// to diagnostics. 2698 /// 2699 /// \param Loc primary location for diagnostic. If two diagnostics are 2700 /// required, one will be at Loc and a new SourceLocation will be created for 2701 /// the other one. 2702 /// 2703 /// \param IsStringLocation if true, Loc points to the format string should be 2704 /// used for the note. Otherwise, Loc points to the argument list and will 2705 /// be used with PDiag. 2706 /// 2707 /// \param StringRange some or all of the string to highlight. This is 2708 /// templated so it can accept either a CharSourceRange or a SourceRange. 2709 /// 2710 /// \param FixIt optional fix it hint for the format string. 2711 template<typename Range> 2712 void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2713 const Expr *ArgumentExpr, 2714 PartialDiagnostic PDiag, 2715 SourceLocation Loc, 2716 bool IsStringLocation, 2717 Range StringRange, 2718 ArrayRef<FixItHint> FixIt) { 2719 if (InFunctionCall) { 2720 const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag); 2721 D << StringRange; 2722 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2723 I != E; ++I) { 2724 D << *I; 2725 } 2726 } else { 2727 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2728 << ArgumentExpr->getSourceRange(); 2729 2730 const Sema::SemaDiagnosticBuilder &Note = 2731 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2732 diag::note_format_string_defined); 2733 2734 Note << StringRange; 2735 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2736 I != E; ++I) { 2737 Note << *I; 2738 } 2739 } 2740 } 2741 2742 //===--- CHECK: Printf format string checking ------------------------------===// 2743 2744 namespace { 2745 class CheckPrintfHandler : public CheckFormatHandler { 2746 bool ObjCContext; 2747 public: 2748 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2749 const Expr *origFormatExpr, unsigned firstDataArg, 2750 unsigned numDataArgs, bool isObjC, 2751 const char *beg, bool hasVAListArg, 2752 ArrayRef<const Expr *> Args, 2753 unsigned formatIdx, bool inFunctionCall, 2754 Sema::VariadicCallType CallType, 2755 llvm::SmallBitVector &CheckedVarArgs) 2756 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2757 numDataArgs, beg, hasVAListArg, Args, 2758 formatIdx, inFunctionCall, CallType, CheckedVarArgs), 2759 ObjCContext(isObjC) 2760 {} 2761 2762 2763 bool HandleInvalidPrintfConversionSpecifier( 2764 const analyze_printf::PrintfSpecifier &FS, 2765 const char *startSpecifier, 2766 unsigned specifierLen) override; 2767 2768 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2769 const char *startSpecifier, 2770 unsigned specifierLen) override; 2771 bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2772 const char *StartSpecifier, 2773 unsigned SpecifierLen, 2774 const Expr *E); 2775 2776 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2777 const char *startSpecifier, unsigned specifierLen); 2778 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2779 const analyze_printf::OptionalAmount &Amt, 2780 unsigned type, 2781 const char *startSpecifier, unsigned specifierLen); 2782 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2783 const analyze_printf::OptionalFlag &flag, 2784 const char *startSpecifier, unsigned specifierLen); 2785 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2786 const analyze_printf::OptionalFlag &ignoredFlag, 2787 const analyze_printf::OptionalFlag &flag, 2788 const char *startSpecifier, unsigned specifierLen); 2789 bool checkForCStrMembers(const analyze_printf::ArgType &AT, 2790 const Expr *E); 2791 2792 }; 2793 } 2794 2795 bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2796 const analyze_printf::PrintfSpecifier &FS, 2797 const char *startSpecifier, 2798 unsigned specifierLen) { 2799 const analyze_printf::PrintfConversionSpecifier &CS = 2800 FS.getConversionSpecifier(); 2801 2802 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2803 getLocationOfByte(CS.getStart()), 2804 startSpecifier, specifierLen, 2805 CS.getStart(), CS.getLength()); 2806 } 2807 2808 bool CheckPrintfHandler::HandleAmount( 2809 const analyze_format_string::OptionalAmount &Amt, 2810 unsigned k, const char *startSpecifier, 2811 unsigned specifierLen) { 2812 2813 if (Amt.hasDataArgument()) { 2814 if (!HasVAListArg) { 2815 unsigned argIndex = Amt.getArgIndex(); 2816 if (argIndex >= NumDataArgs) { 2817 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2818 << k, 2819 getLocationOfByte(Amt.getStart()), 2820 /*IsStringLocation*/true, 2821 getSpecifierRange(startSpecifier, specifierLen)); 2822 // Don't do any more checking. We will just emit 2823 // spurious errors. 2824 return false; 2825 } 2826 2827 // Type check the data argument. It should be an 'int'. 2828 // Although not in conformance with C99, we also allow the argument to be 2829 // an 'unsigned int' as that is a reasonably safe case. GCC also 2830 // doesn't emit a warning for that case. 2831 CoveredArgs.set(argIndex); 2832 const Expr *Arg = getDataArg(argIndex); 2833 if (!Arg) 2834 return false; 2835 2836 QualType T = Arg->getType(); 2837 2838 const analyze_printf::ArgType &AT = Amt.getArgType(S.Context); 2839 assert(AT.isValid()); 2840 2841 if (!AT.matchesType(S.Context, T)) { 2842 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2843 << k << AT.getRepresentativeTypeName(S.Context) 2844 << T << Arg->getSourceRange(), 2845 getLocationOfByte(Amt.getStart()), 2846 /*IsStringLocation*/true, 2847 getSpecifierRange(startSpecifier, specifierLen)); 2848 // Don't do any more checking. We will just emit 2849 // spurious errors. 2850 return false; 2851 } 2852 } 2853 } 2854 return true; 2855 } 2856 2857 void CheckPrintfHandler::HandleInvalidAmount( 2858 const analyze_printf::PrintfSpecifier &FS, 2859 const analyze_printf::OptionalAmount &Amt, 2860 unsigned type, 2861 const char *startSpecifier, 2862 unsigned specifierLen) { 2863 const analyze_printf::PrintfConversionSpecifier &CS = 2864 FS.getConversionSpecifier(); 2865 2866 FixItHint fixit = 2867 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2868 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2869 Amt.getConstantLength())) 2870 : FixItHint(); 2871 2872 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2873 << type << CS.toString(), 2874 getLocationOfByte(Amt.getStart()), 2875 /*IsStringLocation*/true, 2876 getSpecifierRange(startSpecifier, specifierLen), 2877 fixit); 2878 } 2879 2880 void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2881 const analyze_printf::OptionalFlag &flag, 2882 const char *startSpecifier, 2883 unsigned specifierLen) { 2884 // Warn about pointless flag with a fixit removal. 2885 const analyze_printf::PrintfConversionSpecifier &CS = 2886 FS.getConversionSpecifier(); 2887 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2888 << flag.toString() << CS.toString(), 2889 getLocationOfByte(flag.getPosition()), 2890 /*IsStringLocation*/true, 2891 getSpecifierRange(startSpecifier, specifierLen), 2892 FixItHint::CreateRemoval( 2893 getSpecifierRange(flag.getPosition(), 1))); 2894 } 2895 2896 void CheckPrintfHandler::HandleIgnoredFlag( 2897 const analyze_printf::PrintfSpecifier &FS, 2898 const analyze_printf::OptionalFlag &ignoredFlag, 2899 const analyze_printf::OptionalFlag &flag, 2900 const char *startSpecifier, 2901 unsigned specifierLen) { 2902 // Warn about ignored flag with a fixit removal. 2903 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2904 << ignoredFlag.toString() << flag.toString(), 2905 getLocationOfByte(ignoredFlag.getPosition()), 2906 /*IsStringLocation*/true, 2907 getSpecifierRange(startSpecifier, specifierLen), 2908 FixItHint::CreateRemoval( 2909 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2910 } 2911 2912 // Determines if the specified is a C++ class or struct containing 2913 // a member with the specified name and kind (e.g. a CXXMethodDecl named 2914 // "c_str()"). 2915 template<typename MemberKind> 2916 static llvm::SmallPtrSet<MemberKind*, 1> 2917 CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) { 2918 const RecordType *RT = Ty->getAs<RecordType>(); 2919 llvm::SmallPtrSet<MemberKind*, 1> Results; 2920 2921 if (!RT) 2922 return Results; 2923 const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl()); 2924 if (!RD || !RD->getDefinition()) 2925 return Results; 2926 2927 LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(), 2928 Sema::LookupMemberName); 2929 R.suppressDiagnostics(); 2930 2931 // We just need to include all members of the right kind turned up by the 2932 // filter, at this point. 2933 if (S.LookupQualifiedName(R, RT->getDecl())) 2934 for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) { 2935 NamedDecl *decl = (*I)->getUnderlyingDecl(); 2936 if (MemberKind *FK = dyn_cast<MemberKind>(decl)) 2937 Results.insert(FK); 2938 } 2939 return Results; 2940 } 2941 2942 /// Check if we could call '.c_str()' on an object. 2943 /// 2944 /// FIXME: This returns the wrong results in some cases (if cv-qualifiers don't 2945 /// allow the call, or if it would be ambiguous). 2946 bool Sema::hasCStrMethod(const Expr *E) { 2947 typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet; 2948 MethodSet Results = 2949 CXXRecordMembersNamed<CXXMethodDecl>("c_str", *this, E->getType()); 2950 for (MethodSet::iterator MI = Results.begin(), ME = Results.end(); 2951 MI != ME; ++MI) 2952 if ((*MI)->getMinRequiredArguments() == 0) 2953 return true; 2954 return false; 2955 } 2956 2957 // Check if a (w)string was passed when a (w)char* was needed, and offer a 2958 // better diagnostic if so. AT is assumed to be valid. 2959 // Returns true when a c_str() conversion method is found. 2960 bool CheckPrintfHandler::checkForCStrMembers( 2961 const analyze_printf::ArgType &AT, const Expr *E) { 2962 typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet; 2963 2964 MethodSet Results = 2965 CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType()); 2966 2967 for (MethodSet::iterator MI = Results.begin(), ME = Results.end(); 2968 MI != ME; ++MI) { 2969 const CXXMethodDecl *Method = *MI; 2970 if (Method->getMinRequiredArguments() == 0 && 2971 AT.matchesType(S.Context, Method->getReturnType())) { 2972 // FIXME: Suggest parens if the expression needs them. 2973 SourceLocation EndLoc = 2974 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()); 2975 S.Diag(E->getLocStart(), diag::note_printf_c_str) 2976 << "c_str()" 2977 << FixItHint::CreateInsertion(EndLoc, ".c_str()"); 2978 return true; 2979 } 2980 } 2981 2982 return false; 2983 } 2984 2985 bool 2986 CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2987 &FS, 2988 const char *startSpecifier, 2989 unsigned specifierLen) { 2990 2991 using namespace analyze_format_string; 2992 using namespace analyze_printf; 2993 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2994 2995 if (FS.consumesDataArgument()) { 2996 if (atFirstArg) { 2997 atFirstArg = false; 2998 usesPositionalArgs = FS.usesPositionalArg(); 2999 } 3000 else if (usesPositionalArgs != FS.usesPositionalArg()) { 3001 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 3002 startSpecifier, specifierLen); 3003 return false; 3004 } 3005 } 3006 3007 // First check if the field width, precision, and conversion specifier 3008 // have matching data arguments. 3009 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 3010 startSpecifier, specifierLen)) { 3011 return false; 3012 } 3013 3014 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 3015 startSpecifier, specifierLen)) { 3016 return false; 3017 } 3018 3019 if (!CS.consumesDataArgument()) { 3020 // FIXME: Technically specifying a precision or field width here 3021 // makes no sense. Worth issuing a warning at some point. 3022 return true; 3023 } 3024 3025 // Consume the argument. 3026 unsigned argIndex = FS.getArgIndex(); 3027 if (argIndex < NumDataArgs) { 3028 // The check to see if the argIndex is valid will come later. 3029 // We set the bit here because we may exit early from this 3030 // function if we encounter some other error. 3031 CoveredArgs.set(argIndex); 3032 } 3033 3034 // Check for using an Objective-C specific conversion specifier 3035 // in a non-ObjC literal. 3036 if (!ObjCContext && CS.isObjCArg()) { 3037 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 3038 specifierLen); 3039 } 3040 3041 // Check for invalid use of field width 3042 if (!FS.hasValidFieldWidth()) { 3043 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 3044 startSpecifier, specifierLen); 3045 } 3046 3047 // Check for invalid use of precision 3048 if (!FS.hasValidPrecision()) { 3049 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 3050 startSpecifier, specifierLen); 3051 } 3052 3053 // Check each flag does not conflict with any other component. 3054 if (!FS.hasValidThousandsGroupingPrefix()) 3055 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 3056 if (!FS.hasValidLeadingZeros()) 3057 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 3058 if (!FS.hasValidPlusPrefix()) 3059 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 3060 if (!FS.hasValidSpacePrefix()) 3061 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 3062 if (!FS.hasValidAlternativeForm()) 3063 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 3064 if (!FS.hasValidLeftJustified()) 3065 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 3066 3067 // Check that flags are not ignored by another flag 3068 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 3069 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 3070 startSpecifier, specifierLen); 3071 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 3072 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 3073 startSpecifier, specifierLen); 3074 3075 // Check the length modifier is valid with the given conversion specifier. 3076 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3077 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3078 diag::warn_format_nonsensical_length); 3079 else if (!FS.hasStandardLengthModifier()) 3080 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3081 else if (!FS.hasStandardLengthConversionCombination()) 3082 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3083 diag::warn_format_non_standard_conversion_spec); 3084 3085 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3086 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3087 3088 // The remaining checks depend on the data arguments. 3089 if (HasVAListArg) 3090 return true; 3091 3092 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3093 return false; 3094 3095 const Expr *Arg = getDataArg(argIndex); 3096 if (!Arg) 3097 return true; 3098 3099 return checkFormatExpr(FS, startSpecifier, specifierLen, Arg); 3100 } 3101 3102 static bool requiresParensToAddCast(const Expr *E) { 3103 // FIXME: We should have a general way to reason about operator 3104 // precedence and whether parens are actually needed here. 3105 // Take care of a few common cases where they aren't. 3106 const Expr *Inside = E->IgnoreImpCasts(); 3107 if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside)) 3108 Inside = POE->getSyntacticForm()->IgnoreImpCasts(); 3109 3110 switch (Inside->getStmtClass()) { 3111 case Stmt::ArraySubscriptExprClass: 3112 case Stmt::CallExprClass: 3113 case Stmt::CharacterLiteralClass: 3114 case Stmt::CXXBoolLiteralExprClass: 3115 case Stmt::DeclRefExprClass: 3116 case Stmt::FloatingLiteralClass: 3117 case Stmt::IntegerLiteralClass: 3118 case Stmt::MemberExprClass: 3119 case Stmt::ObjCArrayLiteralClass: 3120 case Stmt::ObjCBoolLiteralExprClass: 3121 case Stmt::ObjCBoxedExprClass: 3122 case Stmt::ObjCDictionaryLiteralClass: 3123 case Stmt::ObjCEncodeExprClass: 3124 case Stmt::ObjCIvarRefExprClass: 3125 case Stmt::ObjCMessageExprClass: 3126 case Stmt::ObjCPropertyRefExprClass: 3127 case Stmt::ObjCStringLiteralClass: 3128 case Stmt::ObjCSubscriptRefExprClass: 3129 case Stmt::ParenExprClass: 3130 case Stmt::StringLiteralClass: 3131 case Stmt::UnaryOperatorClass: 3132 return false; 3133 default: 3134 return true; 3135 } 3136 } 3137 3138 bool 3139 CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 3140 const char *StartSpecifier, 3141 unsigned SpecifierLen, 3142 const Expr *E) { 3143 using namespace analyze_format_string; 3144 using namespace analyze_printf; 3145 // Now type check the data expression that matches the 3146 // format specifier. 3147 const analyze_printf::ArgType &AT = FS.getArgType(S.Context, 3148 ObjCContext); 3149 if (!AT.isValid()) 3150 return true; 3151 3152 QualType ExprTy = E->getType(); 3153 while (const TypeOfExprType *TET = dyn_cast<TypeOfExprType>(ExprTy)) { 3154 ExprTy = TET->getUnderlyingExpr()->getType(); 3155 } 3156 3157 if (AT.matchesType(S.Context, ExprTy)) 3158 return true; 3159 3160 // Look through argument promotions for our error message's reported type. 3161 // This includes the integral and floating promotions, but excludes array 3162 // and function pointer decay; seeing that an argument intended to be a 3163 // string has type 'char [6]' is probably more confusing than 'char *'. 3164 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3165 if (ICE->getCastKind() == CK_IntegralCast || 3166 ICE->getCastKind() == CK_FloatingCast) { 3167 E = ICE->getSubExpr(); 3168 ExprTy = E->getType(); 3169 3170 // Check if we didn't match because of an implicit cast from a 'char' 3171 // or 'short' to an 'int'. This is done because printf is a varargs 3172 // function. 3173 if (ICE->getType() == S.Context.IntTy || 3174 ICE->getType() == S.Context.UnsignedIntTy) { 3175 // All further checking is done on the subexpression. 3176 if (AT.matchesType(S.Context, ExprTy)) 3177 return true; 3178 } 3179 } 3180 } else if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E)) { 3181 // Special case for 'a', which has type 'int' in C. 3182 // Note, however, that we do /not/ want to treat multibyte constants like 3183 // 'MooV' as characters! This form is deprecated but still exists. 3184 if (ExprTy == S.Context.IntTy) 3185 if (llvm::isUIntN(S.Context.getCharWidth(), CL->getValue())) 3186 ExprTy = S.Context.CharTy; 3187 } 3188 3189 // %C in an Objective-C context prints a unichar, not a wchar_t. 3190 // If the argument is an integer of some kind, believe the %C and suggest 3191 // a cast instead of changing the conversion specifier. 3192 QualType IntendedTy = ExprTy; 3193 if (ObjCContext && 3194 FS.getConversionSpecifier().getKind() == ConversionSpecifier::CArg) { 3195 if (ExprTy->isIntegralOrUnscopedEnumerationType() && 3196 !ExprTy->isCharType()) { 3197 // 'unichar' is defined as a typedef of unsigned short, but we should 3198 // prefer using the typedef if it is visible. 3199 IntendedTy = S.Context.UnsignedShortTy; 3200 3201 // While we are here, check if the value is an IntegerLiteral that happens 3202 // to be within the valid range. 3203 if (const IntegerLiteral *IL = dyn_cast<IntegerLiteral>(E)) { 3204 const llvm::APInt &V = IL->getValue(); 3205 if (V.getActiveBits() <= S.Context.getTypeSize(IntendedTy)) 3206 return true; 3207 } 3208 3209 LookupResult Result(S, &S.Context.Idents.get("unichar"), E->getLocStart(), 3210 Sema::LookupOrdinaryName); 3211 if (S.LookupName(Result, S.getCurScope())) { 3212 NamedDecl *ND = Result.getFoundDecl(); 3213 if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(ND)) 3214 if (TD->getUnderlyingType() == IntendedTy) 3215 IntendedTy = S.Context.getTypedefType(TD); 3216 } 3217 } 3218 } 3219 3220 // Special-case some of Darwin's platform-independence types by suggesting 3221 // casts to primitive types that are known to be large enough. 3222 bool ShouldNotPrintDirectly = false; 3223 if (S.Context.getTargetInfo().getTriple().isOSDarwin()) { 3224 // Use a 'while' to peel off layers of typedefs. 3225 QualType TyTy = IntendedTy; 3226 while (const TypedefType *UserTy = TyTy->getAs<TypedefType>()) { 3227 StringRef Name = UserTy->getDecl()->getName(); 3228 QualType CastTy = llvm::StringSwitch<QualType>(Name) 3229 .Case("NSInteger", S.Context.LongTy) 3230 .Case("NSUInteger", S.Context.UnsignedLongTy) 3231 .Case("SInt32", S.Context.IntTy) 3232 .Case("UInt32", S.Context.UnsignedIntTy) 3233 .Default(QualType()); 3234 3235 if (!CastTy.isNull()) { 3236 ShouldNotPrintDirectly = true; 3237 IntendedTy = CastTy; 3238 break; 3239 } 3240 TyTy = UserTy->desugar(); 3241 } 3242 } 3243 3244 // We may be able to offer a FixItHint if it is a supported type. 3245 PrintfSpecifier fixedFS = FS; 3246 bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(), 3247 S.Context, ObjCContext); 3248 3249 if (success) { 3250 // Get the fix string from the fixed format specifier 3251 SmallString<16> buf; 3252 llvm::raw_svector_ostream os(buf); 3253 fixedFS.toString(os); 3254 3255 CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen); 3256 3257 if (IntendedTy == ExprTy) { 3258 // In this case, the specifier is wrong and should be changed to match 3259 // the argument. 3260 EmitFormatDiagnostic( 3261 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3262 << AT.getRepresentativeTypeName(S.Context) << IntendedTy 3263 << E->getSourceRange(), 3264 E->getLocStart(), 3265 /*IsStringLocation*/false, 3266 SpecRange, 3267 FixItHint::CreateReplacement(SpecRange, os.str())); 3268 3269 } else { 3270 // The canonical type for formatting this value is different from the 3271 // actual type of the expression. (This occurs, for example, with Darwin's 3272 // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but 3273 // should be printed as 'long' for 64-bit compatibility.) 3274 // Rather than emitting a normal format/argument mismatch, we want to 3275 // add a cast to the recommended type (and correct the format string 3276 // if necessary). 3277 SmallString<16> CastBuf; 3278 llvm::raw_svector_ostream CastFix(CastBuf); 3279 CastFix << "("; 3280 IntendedTy.print(CastFix, S.Context.getPrintingPolicy()); 3281 CastFix << ")"; 3282 3283 SmallVector<FixItHint,4> Hints; 3284 if (!AT.matchesType(S.Context, IntendedTy)) 3285 Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str())); 3286 3287 if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) { 3288 // If there's already a cast present, just replace it. 3289 SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc()); 3290 Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str())); 3291 3292 } else if (!requiresParensToAddCast(E)) { 3293 // If the expression has high enough precedence, 3294 // just write the C-style cast. 3295 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 3296 CastFix.str())); 3297 } else { 3298 // Otherwise, add parens around the expression as well as the cast. 3299 CastFix << "("; 3300 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 3301 CastFix.str())); 3302 3303 SourceLocation After = S.PP.getLocForEndOfToken(E->getLocEnd()); 3304 Hints.push_back(FixItHint::CreateInsertion(After, ")")); 3305 } 3306 3307 if (ShouldNotPrintDirectly) { 3308 // The expression has a type that should not be printed directly. 3309 // We extract the name from the typedef because we don't want to show 3310 // the underlying type in the diagnostic. 3311 StringRef Name = cast<TypedefType>(ExprTy)->getDecl()->getName(); 3312 3313 EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast) 3314 << Name << IntendedTy 3315 << E->getSourceRange(), 3316 E->getLocStart(), /*IsStringLocation=*/false, 3317 SpecRange, Hints); 3318 } else { 3319 // In this case, the expression could be printed using a different 3320 // specifier, but we've decided that the specifier is probably correct 3321 // and we should cast instead. Just use the normal warning message. 3322 EmitFormatDiagnostic( 3323 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3324 << AT.getRepresentativeTypeName(S.Context) << ExprTy 3325 << E->getSourceRange(), 3326 E->getLocStart(), /*IsStringLocation*/false, 3327 SpecRange, Hints); 3328 } 3329 } 3330 } else { 3331 const CharSourceRange &CSR = getSpecifierRange(StartSpecifier, 3332 SpecifierLen); 3333 // Since the warning for passing non-POD types to variadic functions 3334 // was deferred until now, we emit a warning for non-POD 3335 // arguments here. 3336 switch (S.isValidVarArgType(ExprTy)) { 3337 case Sema::VAK_Valid: 3338 case Sema::VAK_ValidInCXX11: 3339 EmitFormatDiagnostic( 3340 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3341 << AT.getRepresentativeTypeName(S.Context) << ExprTy 3342 << CSR 3343 << E->getSourceRange(), 3344 E->getLocStart(), /*IsStringLocation*/false, CSR); 3345 break; 3346 3347 case Sema::VAK_Undefined: 3348 EmitFormatDiagnostic( 3349 S.PDiag(diag::warn_non_pod_vararg_with_format_string) 3350 << S.getLangOpts().CPlusPlus11 3351 << ExprTy 3352 << CallType 3353 << AT.getRepresentativeTypeName(S.Context) 3354 << CSR 3355 << E->getSourceRange(), 3356 E->getLocStart(), /*IsStringLocation*/false, CSR); 3357 checkForCStrMembers(AT, E); 3358 break; 3359 3360 case Sema::VAK_Invalid: 3361 if (ExprTy->isObjCObjectType()) 3362 EmitFormatDiagnostic( 3363 S.PDiag(diag::err_cannot_pass_objc_interface_to_vararg_format) 3364 << S.getLangOpts().CPlusPlus11 3365 << ExprTy 3366 << CallType 3367 << AT.getRepresentativeTypeName(S.Context) 3368 << CSR 3369 << E->getSourceRange(), 3370 E->getLocStart(), /*IsStringLocation*/false, CSR); 3371 else 3372 // FIXME: If this is an initializer list, suggest removing the braces 3373 // or inserting a cast to the target type. 3374 S.Diag(E->getLocStart(), diag::err_cannot_pass_to_vararg_format) 3375 << isa<InitListExpr>(E) << ExprTy << CallType 3376 << AT.getRepresentativeTypeName(S.Context) 3377 << E->getSourceRange(); 3378 break; 3379 } 3380 3381 assert(FirstDataArg + FS.getArgIndex() < CheckedVarArgs.size() && 3382 "format string specifier index out of range"); 3383 CheckedVarArgs[FirstDataArg + FS.getArgIndex()] = true; 3384 } 3385 3386 return true; 3387 } 3388 3389 //===--- CHECK: Scanf format string checking ------------------------------===// 3390 3391 namespace { 3392 class CheckScanfHandler : public CheckFormatHandler { 3393 public: 3394 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 3395 const Expr *origFormatExpr, unsigned firstDataArg, 3396 unsigned numDataArgs, const char *beg, bool hasVAListArg, 3397 ArrayRef<const Expr *> Args, 3398 unsigned formatIdx, bool inFunctionCall, 3399 Sema::VariadicCallType CallType, 3400 llvm::SmallBitVector &CheckedVarArgs) 3401 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 3402 numDataArgs, beg, hasVAListArg, 3403 Args, formatIdx, inFunctionCall, CallType, 3404 CheckedVarArgs) 3405 {} 3406 3407 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 3408 const char *startSpecifier, 3409 unsigned specifierLen) override; 3410 3411 bool HandleInvalidScanfConversionSpecifier( 3412 const analyze_scanf::ScanfSpecifier &FS, 3413 const char *startSpecifier, 3414 unsigned specifierLen) override; 3415 3416 void HandleIncompleteScanList(const char *start, const char *end) override; 3417 }; 3418 } 3419 3420 void CheckScanfHandler::HandleIncompleteScanList(const char *start, 3421 const char *end) { 3422 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 3423 getLocationOfByte(end), /*IsStringLocation*/true, 3424 getSpecifierRange(start, end - start)); 3425 } 3426 3427 bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 3428 const analyze_scanf::ScanfSpecifier &FS, 3429 const char *startSpecifier, 3430 unsigned specifierLen) { 3431 3432 const analyze_scanf::ScanfConversionSpecifier &CS = 3433 FS.getConversionSpecifier(); 3434 3435 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 3436 getLocationOfByte(CS.getStart()), 3437 startSpecifier, specifierLen, 3438 CS.getStart(), CS.getLength()); 3439 } 3440 3441 bool CheckScanfHandler::HandleScanfSpecifier( 3442 const analyze_scanf::ScanfSpecifier &FS, 3443 const char *startSpecifier, 3444 unsigned specifierLen) { 3445 3446 using namespace analyze_scanf; 3447 using namespace analyze_format_string; 3448 3449 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 3450 3451 // Handle case where '%' and '*' don't consume an argument. These shouldn't 3452 // be used to decide if we are using positional arguments consistently. 3453 if (FS.consumesDataArgument()) { 3454 if (atFirstArg) { 3455 atFirstArg = false; 3456 usesPositionalArgs = FS.usesPositionalArg(); 3457 } 3458 else if (usesPositionalArgs != FS.usesPositionalArg()) { 3459 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 3460 startSpecifier, specifierLen); 3461 return false; 3462 } 3463 } 3464 3465 // Check if the field with is non-zero. 3466 const OptionalAmount &Amt = FS.getFieldWidth(); 3467 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 3468 if (Amt.getConstantAmount() == 0) { 3469 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 3470 Amt.getConstantLength()); 3471 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 3472 getLocationOfByte(Amt.getStart()), 3473 /*IsStringLocation*/true, R, 3474 FixItHint::CreateRemoval(R)); 3475 } 3476 } 3477 3478 if (!FS.consumesDataArgument()) { 3479 // FIXME: Technically specifying a precision or field width here 3480 // makes no sense. Worth issuing a warning at some point. 3481 return true; 3482 } 3483 3484 // Consume the argument. 3485 unsigned argIndex = FS.getArgIndex(); 3486 if (argIndex < NumDataArgs) { 3487 // The check to see if the argIndex is valid will come later. 3488 // We set the bit here because we may exit early from this 3489 // function if we encounter some other error. 3490 CoveredArgs.set(argIndex); 3491 } 3492 3493 // Check the length modifier is valid with the given conversion specifier. 3494 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3495 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3496 diag::warn_format_nonsensical_length); 3497 else if (!FS.hasStandardLengthModifier()) 3498 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3499 else if (!FS.hasStandardLengthConversionCombination()) 3500 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3501 diag::warn_format_non_standard_conversion_spec); 3502 3503 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3504 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3505 3506 // The remaining checks depend on the data arguments. 3507 if (HasVAListArg) 3508 return true; 3509 3510 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3511 return false; 3512 3513 // Check that the argument type matches the format specifier. 3514 const Expr *Ex = getDataArg(argIndex); 3515 if (!Ex) 3516 return true; 3517 3518 const analyze_format_string::ArgType &AT = FS.getArgType(S.Context); 3519 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) { 3520 ScanfSpecifier fixedFS = FS; 3521 bool success = fixedFS.fixType(Ex->getType(), 3522 Ex->IgnoreImpCasts()->getType(), 3523 S.getLangOpts(), S.Context); 3524 3525 if (success) { 3526 // Get the fix string from the fixed format specifier. 3527 SmallString<128> buf; 3528 llvm::raw_svector_ostream os(buf); 3529 fixedFS.toString(os); 3530 3531 EmitFormatDiagnostic( 3532 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3533 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3534 << Ex->getSourceRange(), 3535 Ex->getLocStart(), 3536 /*IsStringLocation*/false, 3537 getSpecifierRange(startSpecifier, specifierLen), 3538 FixItHint::CreateReplacement( 3539 getSpecifierRange(startSpecifier, specifierLen), 3540 os.str())); 3541 } else { 3542 EmitFormatDiagnostic( 3543 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3544 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3545 << Ex->getSourceRange(), 3546 Ex->getLocStart(), 3547 /*IsStringLocation*/false, 3548 getSpecifierRange(startSpecifier, specifierLen)); 3549 } 3550 } 3551 3552 return true; 3553 } 3554 3555 void Sema::CheckFormatString(const StringLiteral *FExpr, 3556 const Expr *OrigFormatExpr, 3557 ArrayRef<const Expr *> Args, 3558 bool HasVAListArg, unsigned format_idx, 3559 unsigned firstDataArg, FormatStringType Type, 3560 bool inFunctionCall, VariadicCallType CallType, 3561 llvm::SmallBitVector &CheckedVarArgs) { 3562 3563 // CHECK: is the format string a wide literal? 3564 if (!FExpr->isAscii() && !FExpr->isUTF8()) { 3565 CheckFormatHandler::EmitFormatDiagnostic( 3566 *this, inFunctionCall, Args[format_idx], 3567 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 3568 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3569 return; 3570 } 3571 3572 // Str - The format string. NOTE: this is NOT null-terminated! 3573 StringRef StrRef = FExpr->getString(); 3574 const char *Str = StrRef.data(); 3575 // Account for cases where the string literal is truncated in a declaration. 3576 const ConstantArrayType *T = Context.getAsConstantArrayType(FExpr->getType()); 3577 assert(T && "String literal not of constant array type!"); 3578 size_t TypeSize = T->getSize().getZExtValue(); 3579 size_t StrLen = std::min(std::max(TypeSize, size_t(1)) - 1, StrRef.size()); 3580 const unsigned numDataArgs = Args.size() - firstDataArg; 3581 3582 // Emit a warning if the string literal is truncated and does not contain an 3583 // embedded null character. 3584 if (TypeSize <= StrRef.size() && 3585 StrRef.substr(0, TypeSize).find('\0') == StringRef::npos) { 3586 CheckFormatHandler::EmitFormatDiagnostic( 3587 *this, inFunctionCall, Args[format_idx], 3588 PDiag(diag::warn_printf_format_string_not_null_terminated), 3589 FExpr->getLocStart(), 3590 /*IsStringLocation=*/true, OrigFormatExpr->getSourceRange()); 3591 return; 3592 } 3593 3594 // CHECK: empty format string? 3595 if (StrLen == 0 && numDataArgs > 0) { 3596 CheckFormatHandler::EmitFormatDiagnostic( 3597 *this, inFunctionCall, Args[format_idx], 3598 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 3599 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3600 return; 3601 } 3602 3603 if (Type == FST_Printf || Type == FST_NSString) { 3604 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 3605 numDataArgs, (Type == FST_NSString), 3606 Str, HasVAListArg, Args, format_idx, 3607 inFunctionCall, CallType, CheckedVarArgs); 3608 3609 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 3610 getLangOpts(), 3611 Context.getTargetInfo())) 3612 H.DoneProcessing(); 3613 } else if (Type == FST_Scanf) { 3614 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs, 3615 Str, HasVAListArg, Args, format_idx, 3616 inFunctionCall, CallType, CheckedVarArgs); 3617 3618 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 3619 getLangOpts(), 3620 Context.getTargetInfo())) 3621 H.DoneProcessing(); 3622 } // TODO: handle other formats 3623 } 3624 3625 //===--- CHECK: Warn on use of wrong absolute value function. -------------===// 3626 3627 // Returns the related absolute value function that is larger, of 0 if one 3628 // does not exist. 3629 static unsigned getLargerAbsoluteValueFunction(unsigned AbsFunction) { 3630 switch (AbsFunction) { 3631 default: 3632 return 0; 3633 3634 case Builtin::BI__builtin_abs: 3635 return Builtin::BI__builtin_labs; 3636 case Builtin::BI__builtin_labs: 3637 return Builtin::BI__builtin_llabs; 3638 case Builtin::BI__builtin_llabs: 3639 return 0; 3640 3641 case Builtin::BI__builtin_fabsf: 3642 return Builtin::BI__builtin_fabs; 3643 case Builtin::BI__builtin_fabs: 3644 return Builtin::BI__builtin_fabsl; 3645 case Builtin::BI__builtin_fabsl: 3646 return 0; 3647 3648 case Builtin::BI__builtin_cabsf: 3649 return Builtin::BI__builtin_cabs; 3650 case Builtin::BI__builtin_cabs: 3651 return Builtin::BI__builtin_cabsl; 3652 case Builtin::BI__builtin_cabsl: 3653 return 0; 3654 3655 case Builtin::BIabs: 3656 return Builtin::BIlabs; 3657 case Builtin::BIlabs: 3658 return Builtin::BIllabs; 3659 case Builtin::BIllabs: 3660 return 0; 3661 3662 case Builtin::BIfabsf: 3663 return Builtin::BIfabs; 3664 case Builtin::BIfabs: 3665 return Builtin::BIfabsl; 3666 case Builtin::BIfabsl: 3667 return 0; 3668 3669 case Builtin::BIcabsf: 3670 return Builtin::BIcabs; 3671 case Builtin::BIcabs: 3672 return Builtin::BIcabsl; 3673 case Builtin::BIcabsl: 3674 return 0; 3675 } 3676 } 3677 3678 // Returns the argument type of the absolute value function. 3679 static QualType getAbsoluteValueArgumentType(ASTContext &Context, 3680 unsigned AbsType) { 3681 if (AbsType == 0) 3682 return QualType(); 3683 3684 ASTContext::GetBuiltinTypeError Error = ASTContext::GE_None; 3685 QualType BuiltinType = Context.GetBuiltinType(AbsType, Error); 3686 if (Error != ASTContext::GE_None) 3687 return QualType(); 3688 3689 const FunctionProtoType *FT = BuiltinType->getAs<FunctionProtoType>(); 3690 if (!FT) 3691 return QualType(); 3692 3693 if (FT->getNumParams() != 1) 3694 return QualType(); 3695 3696 return FT->getParamType(0); 3697 } 3698 3699 // Returns the best absolute value function, or zero, based on type and 3700 // current absolute value function. 3701 static unsigned getBestAbsFunction(ASTContext &Context, QualType ArgType, 3702 unsigned AbsFunctionKind) { 3703 unsigned BestKind = 0; 3704 uint64_t ArgSize = Context.getTypeSize(ArgType); 3705 for (unsigned Kind = AbsFunctionKind; Kind != 0; 3706 Kind = getLargerAbsoluteValueFunction(Kind)) { 3707 QualType ParamType = getAbsoluteValueArgumentType(Context, Kind); 3708 if (Context.getTypeSize(ParamType) >= ArgSize) { 3709 if (BestKind == 0) 3710 BestKind = Kind; 3711 else if (Context.hasSameType(ParamType, ArgType)) { 3712 BestKind = Kind; 3713 break; 3714 } 3715 } 3716 } 3717 return BestKind; 3718 } 3719 3720 enum AbsoluteValueKind { 3721 AVK_Integer, 3722 AVK_Floating, 3723 AVK_Complex 3724 }; 3725 3726 static AbsoluteValueKind getAbsoluteValueKind(QualType T) { 3727 if (T->isIntegralOrEnumerationType()) 3728 return AVK_Integer; 3729 if (T->isRealFloatingType()) 3730 return AVK_Floating; 3731 if (T->isAnyComplexType()) 3732 return AVK_Complex; 3733 3734 llvm_unreachable("Type not integer, floating, or complex"); 3735 } 3736 3737 // Changes the absolute value function to a different type. Preserves whether 3738 // the function is a builtin. 3739 static unsigned changeAbsFunction(unsigned AbsKind, 3740 AbsoluteValueKind ValueKind) { 3741 switch (ValueKind) { 3742 case AVK_Integer: 3743 switch (AbsKind) { 3744 default: 3745 return 0; 3746 case Builtin::BI__builtin_fabsf: 3747 case Builtin::BI__builtin_fabs: 3748 case Builtin::BI__builtin_fabsl: 3749 case Builtin::BI__builtin_cabsf: 3750 case Builtin::BI__builtin_cabs: 3751 case Builtin::BI__builtin_cabsl: 3752 return Builtin::BI__builtin_abs; 3753 case Builtin::BIfabsf: 3754 case Builtin::BIfabs: 3755 case Builtin::BIfabsl: 3756 case Builtin::BIcabsf: 3757 case Builtin::BIcabs: 3758 case Builtin::BIcabsl: 3759 return Builtin::BIabs; 3760 } 3761 case AVK_Floating: 3762 switch (AbsKind) { 3763 default: 3764 return 0; 3765 case Builtin::BI__builtin_abs: 3766 case Builtin::BI__builtin_labs: 3767 case Builtin::BI__builtin_llabs: 3768 case Builtin::BI__builtin_cabsf: 3769 case Builtin::BI__builtin_cabs: 3770 case Builtin::BI__builtin_cabsl: 3771 return Builtin::BI__builtin_fabsf; 3772 case Builtin::BIabs: 3773 case Builtin::BIlabs: 3774 case Builtin::BIllabs: 3775 case Builtin::BIcabsf: 3776 case Builtin::BIcabs: 3777 case Builtin::BIcabsl: 3778 return Builtin::BIfabsf; 3779 } 3780 case AVK_Complex: 3781 switch (AbsKind) { 3782 default: 3783 return 0; 3784 case Builtin::BI__builtin_abs: 3785 case Builtin::BI__builtin_labs: 3786 case Builtin::BI__builtin_llabs: 3787 case Builtin::BI__builtin_fabsf: 3788 case Builtin::BI__builtin_fabs: 3789 case Builtin::BI__builtin_fabsl: 3790 return Builtin::BI__builtin_cabsf; 3791 case Builtin::BIabs: 3792 case Builtin::BIlabs: 3793 case Builtin::BIllabs: 3794 case Builtin::BIfabsf: 3795 case Builtin::BIfabs: 3796 case Builtin::BIfabsl: 3797 return Builtin::BIcabsf; 3798 } 3799 } 3800 llvm_unreachable("Unable to convert function"); 3801 } 3802 3803 static unsigned getAbsoluteValueFunctionKind(const FunctionDecl *FDecl) { 3804 const IdentifierInfo *FnInfo = FDecl->getIdentifier(); 3805 if (!FnInfo) 3806 return 0; 3807 3808 switch (FDecl->getBuiltinID()) { 3809 default: 3810 return 0; 3811 case Builtin::BI__builtin_abs: 3812 case Builtin::BI__builtin_fabs: 3813 case Builtin::BI__builtin_fabsf: 3814 case Builtin::BI__builtin_fabsl: 3815 case Builtin::BI__builtin_labs: 3816 case Builtin::BI__builtin_llabs: 3817 case Builtin::BI__builtin_cabs: 3818 case Builtin::BI__builtin_cabsf: 3819 case Builtin::BI__builtin_cabsl: 3820 case Builtin::BIabs: 3821 case Builtin::BIlabs: 3822 case Builtin::BIllabs: 3823 case Builtin::BIfabs: 3824 case Builtin::BIfabsf: 3825 case Builtin::BIfabsl: 3826 case Builtin::BIcabs: 3827 case Builtin::BIcabsf: 3828 case Builtin::BIcabsl: 3829 return FDecl->getBuiltinID(); 3830 } 3831 llvm_unreachable("Unknown Builtin type"); 3832 } 3833 3834 // If the replacement is valid, emit a note with replacement function. 3835 // Additionally, suggest including the proper header if not already included. 3836 static void emitReplacement(Sema &S, SourceLocation Loc, SourceRange Range, 3837 unsigned AbsKind) { 3838 std::string AbsName = S.Context.BuiltinInfo.GetName(AbsKind); 3839 3840 // Look up absolute value function in TU scope. 3841 DeclarationName DN(&S.Context.Idents.get(AbsName)); 3842 LookupResult R(S, DN, Loc, Sema::LookupAnyName); 3843 R.suppressDiagnostics(); 3844 S.LookupName(R, S.TUScope); 3845 3846 // Skip notes if multiple results found in lookup. 3847 if (!R.empty() && !R.isSingleResult()) 3848 return; 3849 3850 FunctionDecl *FD = 0; 3851 bool FoundFunction = R.isSingleResult(); 3852 // When one result is found, see if it is the correct function. 3853 if (R.isSingleResult()) { 3854 FD = dyn_cast<FunctionDecl>(R.getFoundDecl()); 3855 if (!FD || FD->getBuiltinID() != AbsKind) 3856 return; 3857 } 3858 3859 // Look for local name conflict, prepend "::" as necessary. 3860 R.clear(); 3861 S.LookupName(R, S.getCurScope()); 3862 3863 if (!FoundFunction) { 3864 if (!R.empty()) { 3865 AbsName = "::" + AbsName; 3866 } 3867 } else { // FoundFunction 3868 if (R.isSingleResult()) { 3869 if (R.getFoundDecl() != FD) { 3870 AbsName = "::" + AbsName; 3871 } 3872 } else if (!R.empty()) { 3873 AbsName = "::" + AbsName; 3874 } 3875 } 3876 3877 S.Diag(Loc, diag::note_replace_abs_function) 3878 << AbsName << FixItHint::CreateReplacement(Range, AbsName); 3879 3880 if (!FoundFunction) { 3881 S.Diag(Loc, diag::note_please_include_header) 3882 << S.Context.BuiltinInfo.getHeaderName(AbsKind) 3883 << S.Context.BuiltinInfo.GetName(AbsKind); 3884 } 3885 } 3886 3887 // Warn when using the wrong abs() function. 3888 void Sema::CheckAbsoluteValueFunction(const CallExpr *Call, 3889 const FunctionDecl *FDecl, 3890 IdentifierInfo *FnInfo) { 3891 if (Call->getNumArgs() != 1) 3892 return; 3893 3894 unsigned AbsKind = getAbsoluteValueFunctionKind(FDecl); 3895 if (AbsKind == 0) 3896 return; 3897 3898 QualType ArgType = Call->getArg(0)->IgnoreParenImpCasts()->getType(); 3899 QualType ParamType = Call->getArg(0)->getType(); 3900 3901 // Unsigned types can not be negative. Suggest to drop the absolute value 3902 // function. 3903 if (ArgType->isUnsignedIntegerType()) { 3904 Diag(Call->getExprLoc(), diag::warn_unsigned_abs) << ArgType << ParamType; 3905 Diag(Call->getExprLoc(), diag::note_remove_abs) 3906 << FDecl 3907 << FixItHint::CreateRemoval(Call->getCallee()->getSourceRange()); 3908 return; 3909 } 3910 3911 AbsoluteValueKind ArgValueKind = getAbsoluteValueKind(ArgType); 3912 AbsoluteValueKind ParamValueKind = getAbsoluteValueKind(ParamType); 3913 3914 // The argument and parameter are the same kind. Check if they are the right 3915 // size. 3916 if (ArgValueKind == ParamValueKind) { 3917 if (Context.getTypeSize(ArgType) <= Context.getTypeSize(ParamType)) 3918 return; 3919 3920 unsigned NewAbsKind = getBestAbsFunction(Context, ArgType, AbsKind); 3921 Diag(Call->getExprLoc(), diag::warn_abs_too_small) 3922 << FDecl << ArgType << ParamType; 3923 3924 if (NewAbsKind == 0) 3925 return; 3926 3927 emitReplacement(*this, Call->getExprLoc(), 3928 Call->getCallee()->getSourceRange(), NewAbsKind); 3929 return; 3930 } 3931 3932 // ArgValueKind != ParamValueKind 3933 // The wrong type of absolute value function was used. Attempt to find the 3934 // proper one. 3935 unsigned NewAbsKind = changeAbsFunction(AbsKind, ArgValueKind); 3936 NewAbsKind = getBestAbsFunction(Context, ArgType, NewAbsKind); 3937 if (NewAbsKind == 0) 3938 return; 3939 3940 Diag(Call->getExprLoc(), diag::warn_wrong_absolute_value_type) 3941 << FDecl << ParamValueKind << ArgValueKind; 3942 3943 emitReplacement(*this, Call->getExprLoc(), 3944 Call->getCallee()->getSourceRange(), NewAbsKind); 3945 return; 3946 } 3947 3948 //===--- CHECK: Standard memory functions ---------------------------------===// 3949 3950 /// \brief Takes the expression passed to the size_t parameter of functions 3951 /// such as memcmp, strncat, etc and warns if it's a comparison. 3952 /// 3953 /// This is to catch typos like `if (memcmp(&a, &b, sizeof(a) > 0))`. 3954 static bool CheckMemorySizeofForComparison(Sema &S, const Expr *E, 3955 IdentifierInfo *FnName, 3956 SourceLocation FnLoc, 3957 SourceLocation RParenLoc) { 3958 const BinaryOperator *Size = dyn_cast<BinaryOperator>(E); 3959 if (!Size) 3960 return false; 3961 3962 // if E is binop and op is >, <, >=, <=, ==, &&, ||: 3963 if (!Size->isComparisonOp() && !Size->isEqualityOp() && !Size->isLogicalOp()) 3964 return false; 3965 3966 Preprocessor &PP = S.getPreprocessor(); 3967 SourceRange SizeRange = Size->getSourceRange(); 3968 S.Diag(Size->getOperatorLoc(), diag::warn_memsize_comparison) 3969 << SizeRange << FnName; 3970 S.Diag(FnLoc, diag::warn_memsize_comparison_paren_note) 3971 << FnName 3972 << FixItHint::CreateInsertion( 3973 PP.getLocForEndOfToken(Size->getLHS()->getLocEnd()), 3974 ")") 3975 << FixItHint::CreateRemoval(RParenLoc); 3976 S.Diag(SizeRange.getBegin(), diag::warn_memsize_comparison_cast_note) 3977 << FixItHint::CreateInsertion(SizeRange.getBegin(), "(size_t)(") 3978 << FixItHint::CreateInsertion( 3979 PP.getLocForEndOfToken(SizeRange.getEnd()), ")"); 3980 3981 return true; 3982 } 3983 3984 /// \brief Determine whether the given type is a dynamic class type (e.g., 3985 /// whether it has a vtable). 3986 static bool isDynamicClassType(QualType T) { 3987 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 3988 if (CXXRecordDecl *Definition = Record->getDefinition()) 3989 if (Definition->isDynamicClass()) 3990 return true; 3991 3992 return false; 3993 } 3994 3995 /// \brief If E is a sizeof expression, returns its argument expression, 3996 /// otherwise returns NULL. 3997 static const Expr *getSizeOfExprArg(const Expr* E) { 3998 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3999 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 4000 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 4001 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 4002 4003 return 0; 4004 } 4005 4006 /// \brief If E is a sizeof expression, returns its argument type. 4007 static QualType getSizeOfArgType(const Expr* E) { 4008 if (const UnaryExprOrTypeTraitExpr *SizeOf = 4009 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 4010 if (SizeOf->getKind() == clang::UETT_SizeOf) 4011 return SizeOf->getTypeOfArgument(); 4012 4013 return QualType(); 4014 } 4015 4016 /// \brief Check for dangerous or invalid arguments to memset(). 4017 /// 4018 /// This issues warnings on known problematic, dangerous or unspecified 4019 /// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 4020 /// function calls. 4021 /// 4022 /// \param Call The call expression to diagnose. 4023 void Sema::CheckMemaccessArguments(const CallExpr *Call, 4024 unsigned BId, 4025 IdentifierInfo *FnName) { 4026 assert(BId != 0); 4027 4028 // It is possible to have a non-standard definition of memset. Validate 4029 // we have enough arguments, and if not, abort further checking. 4030 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 4031 if (Call->getNumArgs() < ExpectedNumArgs) 4032 return; 4033 4034 unsigned LastArg = (BId == Builtin::BImemset || 4035 BId == Builtin::BIstrndup ? 1 : 2); 4036 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 4037 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 4038 4039 if (CheckMemorySizeofForComparison(*this, LenExpr, FnName, 4040 Call->getLocStart(), Call->getRParenLoc())) 4041 return; 4042 4043 // We have special checking when the length is a sizeof expression. 4044 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 4045 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 4046 llvm::FoldingSetNodeID SizeOfArgID; 4047 4048 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 4049 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 4050 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 4051 4052 QualType DestTy = Dest->getType(); 4053 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 4054 QualType PointeeTy = DestPtrTy->getPointeeType(); 4055 4056 // Never warn about void type pointers. This can be used to suppress 4057 // false positives. 4058 if (PointeeTy->isVoidType()) 4059 continue; 4060 4061 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 4062 // actually comparing the expressions for equality. Because computing the 4063 // expression IDs can be expensive, we only do this if the diagnostic is 4064 // enabled. 4065 if (SizeOfArg && 4066 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 4067 SizeOfArg->getExprLoc())) { 4068 // We only compute IDs for expressions if the warning is enabled, and 4069 // cache the sizeof arg's ID. 4070 if (SizeOfArgID == llvm::FoldingSetNodeID()) 4071 SizeOfArg->Profile(SizeOfArgID, Context, true); 4072 llvm::FoldingSetNodeID DestID; 4073 Dest->Profile(DestID, Context, true); 4074 if (DestID == SizeOfArgID) { 4075 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 4076 // over sizeof(src) as well. 4077 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 4078 StringRef ReadableName = FnName->getName(); 4079 4080 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 4081 if (UnaryOp->getOpcode() == UO_AddrOf) 4082 ActionIdx = 1; // If its an address-of operator, just remove it. 4083 if (!PointeeTy->isIncompleteType() && 4084 (Context.getTypeSize(PointeeTy) == Context.getCharWidth())) 4085 ActionIdx = 2; // If the pointee's size is sizeof(char), 4086 // suggest an explicit length. 4087 4088 // If the function is defined as a builtin macro, do not show macro 4089 // expansion. 4090 SourceLocation SL = SizeOfArg->getExprLoc(); 4091 SourceRange DSR = Dest->getSourceRange(); 4092 SourceRange SSR = SizeOfArg->getSourceRange(); 4093 SourceManager &SM = PP.getSourceManager(); 4094 4095 if (SM.isMacroArgExpansion(SL)) { 4096 ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts); 4097 SL = SM.getSpellingLoc(SL); 4098 DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()), 4099 SM.getSpellingLoc(DSR.getEnd())); 4100 SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()), 4101 SM.getSpellingLoc(SSR.getEnd())); 4102 } 4103 4104 DiagRuntimeBehavior(SL, SizeOfArg, 4105 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 4106 << ReadableName 4107 << PointeeTy 4108 << DestTy 4109 << DSR 4110 << SSR); 4111 DiagRuntimeBehavior(SL, SizeOfArg, 4112 PDiag(diag::warn_sizeof_pointer_expr_memaccess_note) 4113 << ActionIdx 4114 << SSR); 4115 4116 break; 4117 } 4118 } 4119 4120 // Also check for cases where the sizeof argument is the exact same 4121 // type as the memory argument, and where it points to a user-defined 4122 // record type. 4123 if (SizeOfArgTy != QualType()) { 4124 if (PointeeTy->isRecordType() && 4125 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 4126 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 4127 PDiag(diag::warn_sizeof_pointer_type_memaccess) 4128 << FnName << SizeOfArgTy << ArgIdx 4129 << PointeeTy << Dest->getSourceRange() 4130 << LenExpr->getSourceRange()); 4131 break; 4132 } 4133 } 4134 4135 // Always complain about dynamic classes. 4136 if (isDynamicClassType(PointeeTy)) { 4137 4138 unsigned OperationType = 0; 4139 // "overwritten" if we're warning about the destination for any call 4140 // but memcmp; otherwise a verb appropriate to the call. 4141 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 4142 if (BId == Builtin::BImemcpy) 4143 OperationType = 1; 4144 else if(BId == Builtin::BImemmove) 4145 OperationType = 2; 4146 else if (BId == Builtin::BImemcmp) 4147 OperationType = 3; 4148 } 4149 4150 DiagRuntimeBehavior( 4151 Dest->getExprLoc(), Dest, 4152 PDiag(diag::warn_dyn_class_memaccess) 4153 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 4154 << FnName << PointeeTy 4155 << OperationType 4156 << Call->getCallee()->getSourceRange()); 4157 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 4158 BId != Builtin::BImemset) 4159 DiagRuntimeBehavior( 4160 Dest->getExprLoc(), Dest, 4161 PDiag(diag::warn_arc_object_memaccess) 4162 << ArgIdx << FnName << PointeeTy 4163 << Call->getCallee()->getSourceRange()); 4164 else 4165 continue; 4166 4167 DiagRuntimeBehavior( 4168 Dest->getExprLoc(), Dest, 4169 PDiag(diag::note_bad_memaccess_silence) 4170 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 4171 break; 4172 } 4173 } 4174 } 4175 4176 // A little helper routine: ignore addition and subtraction of integer literals. 4177 // This intentionally does not ignore all integer constant expressions because 4178 // we don't want to remove sizeof(). 4179 static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 4180 Ex = Ex->IgnoreParenCasts(); 4181 4182 for (;;) { 4183 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 4184 if (!BO || !BO->isAdditiveOp()) 4185 break; 4186 4187 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 4188 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 4189 4190 if (isa<IntegerLiteral>(RHS)) 4191 Ex = LHS; 4192 else if (isa<IntegerLiteral>(LHS)) 4193 Ex = RHS; 4194 else 4195 break; 4196 } 4197 4198 return Ex; 4199 } 4200 4201 static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty, 4202 ASTContext &Context) { 4203 // Only handle constant-sized or VLAs, but not flexible members. 4204 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) { 4205 // Only issue the FIXIT for arrays of size > 1. 4206 if (CAT->getSize().getSExtValue() <= 1) 4207 return false; 4208 } else if (!Ty->isVariableArrayType()) { 4209 return false; 4210 } 4211 return true; 4212 } 4213 4214 // Warn if the user has made the 'size' argument to strlcpy or strlcat 4215 // be the size of the source, instead of the destination. 4216 void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 4217 IdentifierInfo *FnName) { 4218 4219 // Don't crash if the user has the wrong number of arguments 4220 if (Call->getNumArgs() != 3) 4221 return; 4222 4223 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 4224 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 4225 const Expr *CompareWithSrc = NULL; 4226 4227 if (CheckMemorySizeofForComparison(*this, SizeArg, FnName, 4228 Call->getLocStart(), Call->getRParenLoc())) 4229 return; 4230 4231 // Look for 'strlcpy(dst, x, sizeof(x))' 4232 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 4233 CompareWithSrc = Ex; 4234 else { 4235 // Look for 'strlcpy(dst, x, strlen(x))' 4236 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 4237 if (SizeCall->getBuiltinCallee() == Builtin::BIstrlen && 4238 SizeCall->getNumArgs() == 1) 4239 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 4240 } 4241 } 4242 4243 if (!CompareWithSrc) 4244 return; 4245 4246 // Determine if the argument to sizeof/strlen is equal to the source 4247 // argument. In principle there's all kinds of things you could do 4248 // here, for instance creating an == expression and evaluating it with 4249 // EvaluateAsBooleanCondition, but this uses a more direct technique: 4250 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 4251 if (!SrcArgDRE) 4252 return; 4253 4254 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 4255 if (!CompareWithSrcDRE || 4256 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 4257 return; 4258 4259 const Expr *OriginalSizeArg = Call->getArg(2); 4260 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 4261 << OriginalSizeArg->getSourceRange() << FnName; 4262 4263 // Output a FIXIT hint if the destination is an array (rather than a 4264 // pointer to an array). This could be enhanced to handle some 4265 // pointers if we know the actual size, like if DstArg is 'array+2' 4266 // we could say 'sizeof(array)-2'. 4267 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 4268 if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context)) 4269 return; 4270 4271 SmallString<128> sizeString; 4272 llvm::raw_svector_ostream OS(sizeString); 4273 OS << "sizeof("; 4274 DstArg->printPretty(OS, 0, getPrintingPolicy()); 4275 OS << ")"; 4276 4277 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 4278 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 4279 OS.str()); 4280 } 4281 4282 /// Check if two expressions refer to the same declaration. 4283 static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 4284 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 4285 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 4286 return D1->getDecl() == D2->getDecl(); 4287 return false; 4288 } 4289 4290 static const Expr *getStrlenExprArg(const Expr *E) { 4291 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 4292 const FunctionDecl *FD = CE->getDirectCallee(); 4293 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 4294 return 0; 4295 return CE->getArg(0)->IgnoreParenCasts(); 4296 } 4297 return 0; 4298 } 4299 4300 // Warn on anti-patterns as the 'size' argument to strncat. 4301 // The correct size argument should look like following: 4302 // strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 4303 void Sema::CheckStrncatArguments(const CallExpr *CE, 4304 IdentifierInfo *FnName) { 4305 // Don't crash if the user has the wrong number of arguments. 4306 if (CE->getNumArgs() < 3) 4307 return; 4308 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 4309 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 4310 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 4311 4312 if (CheckMemorySizeofForComparison(*this, LenArg, FnName, CE->getLocStart(), 4313 CE->getRParenLoc())) 4314 return; 4315 4316 // Identify common expressions, which are wrongly used as the size argument 4317 // to strncat and may lead to buffer overflows. 4318 unsigned PatternType = 0; 4319 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 4320 // - sizeof(dst) 4321 if (referToTheSameDecl(SizeOfArg, DstArg)) 4322 PatternType = 1; 4323 // - sizeof(src) 4324 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 4325 PatternType = 2; 4326 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 4327 if (BE->getOpcode() == BO_Sub) { 4328 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 4329 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 4330 // - sizeof(dst) - strlen(dst) 4331 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 4332 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 4333 PatternType = 1; 4334 // - sizeof(src) - (anything) 4335 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 4336 PatternType = 2; 4337 } 4338 } 4339 4340 if (PatternType == 0) 4341 return; 4342 4343 // Generate the diagnostic. 4344 SourceLocation SL = LenArg->getLocStart(); 4345 SourceRange SR = LenArg->getSourceRange(); 4346 SourceManager &SM = PP.getSourceManager(); 4347 4348 // If the function is defined as a builtin macro, do not show macro expansion. 4349 if (SM.isMacroArgExpansion(SL)) { 4350 SL = SM.getSpellingLoc(SL); 4351 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 4352 SM.getSpellingLoc(SR.getEnd())); 4353 } 4354 4355 // Check if the destination is an array (rather than a pointer to an array). 4356 QualType DstTy = DstArg->getType(); 4357 bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy, 4358 Context); 4359 if (!isKnownSizeArray) { 4360 if (PatternType == 1) 4361 Diag(SL, diag::warn_strncat_wrong_size) << SR; 4362 else 4363 Diag(SL, diag::warn_strncat_src_size) << SR; 4364 return; 4365 } 4366 4367 if (PatternType == 1) 4368 Diag(SL, diag::warn_strncat_large_size) << SR; 4369 else 4370 Diag(SL, diag::warn_strncat_src_size) << SR; 4371 4372 SmallString<128> sizeString; 4373 llvm::raw_svector_ostream OS(sizeString); 4374 OS << "sizeof("; 4375 DstArg->printPretty(OS, 0, getPrintingPolicy()); 4376 OS << ") - "; 4377 OS << "strlen("; 4378 DstArg->printPretty(OS, 0, getPrintingPolicy()); 4379 OS << ") - 1"; 4380 4381 Diag(SL, diag::note_strncat_wrong_size) 4382 << FixItHint::CreateReplacement(SR, OS.str()); 4383 } 4384 4385 //===--- CHECK: Return Address of Stack Variable --------------------------===// 4386 4387 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 4388 Decl *ParentDecl); 4389 static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 4390 Decl *ParentDecl); 4391 4392 /// CheckReturnStackAddr - Check if a return statement returns the address 4393 /// of a stack variable. 4394 static void 4395 CheckReturnStackAddr(Sema &S, Expr *RetValExp, QualType lhsType, 4396 SourceLocation ReturnLoc) { 4397 4398 Expr *stackE = 0; 4399 SmallVector<DeclRefExpr *, 8> refVars; 4400 4401 // Perform checking for returned stack addresses, local blocks, 4402 // label addresses or references to temporaries. 4403 if (lhsType->isPointerType() || 4404 (!S.getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 4405 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 4406 } else if (lhsType->isReferenceType()) { 4407 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 4408 } 4409 4410 if (stackE == 0) 4411 return; // Nothing suspicious was found. 4412 4413 SourceLocation diagLoc; 4414 SourceRange diagRange; 4415 if (refVars.empty()) { 4416 diagLoc = stackE->getLocStart(); 4417 diagRange = stackE->getSourceRange(); 4418 } else { 4419 // We followed through a reference variable. 'stackE' contains the 4420 // problematic expression but we will warn at the return statement pointing 4421 // at the reference variable. We will later display the "trail" of 4422 // reference variables using notes. 4423 diagLoc = refVars[0]->getLocStart(); 4424 diagRange = refVars[0]->getSourceRange(); 4425 } 4426 4427 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 4428 S.Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 4429 : diag::warn_ret_stack_addr) 4430 << DR->getDecl()->getDeclName() << diagRange; 4431 } else if (isa<BlockExpr>(stackE)) { // local block. 4432 S.Diag(diagLoc, diag::err_ret_local_block) << diagRange; 4433 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 4434 S.Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 4435 } else { // local temporary. 4436 S.Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 4437 : diag::warn_ret_local_temp_addr) 4438 << diagRange; 4439 } 4440 4441 // Display the "trail" of reference variables that we followed until we 4442 // found the problematic expression using notes. 4443 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 4444 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 4445 // If this var binds to another reference var, show the range of the next 4446 // var, otherwise the var binds to the problematic expression, in which case 4447 // show the range of the expression. 4448 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 4449 : stackE->getSourceRange(); 4450 S.Diag(VD->getLocation(), diag::note_ref_var_local_bind) 4451 << VD->getDeclName() << range; 4452 } 4453 } 4454 4455 /// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 4456 /// check if the expression in a return statement evaluates to an address 4457 /// to a location on the stack, a local block, an address of a label, or a 4458 /// reference to local temporary. The recursion is used to traverse the 4459 /// AST of the return expression, with recursion backtracking when we 4460 /// encounter a subexpression that (1) clearly does not lead to one of the 4461 /// above problematic expressions (2) is something we cannot determine leads to 4462 /// a problematic expression based on such local checking. 4463 /// 4464 /// Both EvalAddr and EvalVal follow through reference variables to evaluate 4465 /// the expression that they point to. Such variables are added to the 4466 /// 'refVars' vector so that we know what the reference variable "trail" was. 4467 /// 4468 /// EvalAddr processes expressions that are pointers that are used as 4469 /// references (and not L-values). EvalVal handles all other values. 4470 /// At the base case of the recursion is a check for the above problematic 4471 /// expressions. 4472 /// 4473 /// This implementation handles: 4474 /// 4475 /// * pointer-to-pointer casts 4476 /// * implicit conversions from array references to pointers 4477 /// * taking the address of fields 4478 /// * arbitrary interplay between "&" and "*" operators 4479 /// * pointer arithmetic from an address of a stack variable 4480 /// * taking the address of an array element where the array is on the stack 4481 static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 4482 Decl *ParentDecl) { 4483 if (E->isTypeDependent()) 4484 return NULL; 4485 4486 // We should only be called for evaluating pointer expressions. 4487 assert((E->getType()->isAnyPointerType() || 4488 E->getType()->isBlockPointerType() || 4489 E->getType()->isObjCQualifiedIdType()) && 4490 "EvalAddr only works on pointers"); 4491 4492 E = E->IgnoreParens(); 4493 4494 // Our "symbolic interpreter" is just a dispatch off the currently 4495 // viewed AST node. We then recursively traverse the AST by calling 4496 // EvalAddr and EvalVal appropriately. 4497 switch (E->getStmtClass()) { 4498 case Stmt::DeclRefExprClass: { 4499 DeclRefExpr *DR = cast<DeclRefExpr>(E); 4500 4501 // If we leave the immediate function, the lifetime isn't about to end. 4502 if (DR->refersToEnclosingLocal()) 4503 return 0; 4504 4505 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 4506 // If this is a reference variable, follow through to the expression that 4507 // it points to. 4508 if (V->hasLocalStorage() && 4509 V->getType()->isReferenceType() && V->hasInit()) { 4510 // Add the reference variable to the "trail". 4511 refVars.push_back(DR); 4512 return EvalAddr(V->getInit(), refVars, ParentDecl); 4513 } 4514 4515 return NULL; 4516 } 4517 4518 case Stmt::UnaryOperatorClass: { 4519 // The only unary operator that make sense to handle here 4520 // is AddrOf. All others don't make sense as pointers. 4521 UnaryOperator *U = cast<UnaryOperator>(E); 4522 4523 if (U->getOpcode() == UO_AddrOf) 4524 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 4525 else 4526 return NULL; 4527 } 4528 4529 case Stmt::BinaryOperatorClass: { 4530 // Handle pointer arithmetic. All other binary operators are not valid 4531 // in this context. 4532 BinaryOperator *B = cast<BinaryOperator>(E); 4533 BinaryOperatorKind op = B->getOpcode(); 4534 4535 if (op != BO_Add && op != BO_Sub) 4536 return NULL; 4537 4538 Expr *Base = B->getLHS(); 4539 4540 // Determine which argument is the real pointer base. It could be 4541 // the RHS argument instead of the LHS. 4542 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 4543 4544 assert (Base->getType()->isPointerType()); 4545 return EvalAddr(Base, refVars, ParentDecl); 4546 } 4547 4548 // For conditional operators we need to see if either the LHS or RHS are 4549 // valid DeclRefExpr*s. If one of them is valid, we return it. 4550 case Stmt::ConditionalOperatorClass: { 4551 ConditionalOperator *C = cast<ConditionalOperator>(E); 4552 4553 // Handle the GNU extension for missing LHS. 4554 // FIXME: That isn't a ConditionalOperator, so doesn't get here. 4555 if (Expr *LHSExpr = C->getLHS()) { 4556 // In C++, we can have a throw-expression, which has 'void' type. 4557 if (!LHSExpr->getType()->isVoidType()) 4558 if (Expr *LHS = EvalAddr(LHSExpr, refVars, ParentDecl)) 4559 return LHS; 4560 } 4561 4562 // In C++, we can have a throw-expression, which has 'void' type. 4563 if (C->getRHS()->getType()->isVoidType()) 4564 return 0; 4565 4566 return EvalAddr(C->getRHS(), refVars, ParentDecl); 4567 } 4568 4569 case Stmt::BlockExprClass: 4570 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 4571 return E; // local block. 4572 return NULL; 4573 4574 case Stmt::AddrLabelExprClass: 4575 return E; // address of label. 4576 4577 case Stmt::ExprWithCleanupsClass: 4578 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 4579 ParentDecl); 4580 4581 // For casts, we need to handle conversions from arrays to 4582 // pointer values, and pointer-to-pointer conversions. 4583 case Stmt::ImplicitCastExprClass: 4584 case Stmt::CStyleCastExprClass: 4585 case Stmt::CXXFunctionalCastExprClass: 4586 case Stmt::ObjCBridgedCastExprClass: 4587 case Stmt::CXXStaticCastExprClass: 4588 case Stmt::CXXDynamicCastExprClass: 4589 case Stmt::CXXConstCastExprClass: 4590 case Stmt::CXXReinterpretCastExprClass: { 4591 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 4592 switch (cast<CastExpr>(E)->getCastKind()) { 4593 case CK_BitCast: 4594 case CK_LValueToRValue: 4595 case CK_NoOp: 4596 case CK_BaseToDerived: 4597 case CK_DerivedToBase: 4598 case CK_UncheckedDerivedToBase: 4599 case CK_Dynamic: 4600 case CK_CPointerToObjCPointerCast: 4601 case CK_BlockPointerToObjCPointerCast: 4602 case CK_AnyPointerToBlockPointerCast: 4603 return EvalAddr(SubExpr, refVars, ParentDecl); 4604 4605 case CK_ArrayToPointerDecay: 4606 return EvalVal(SubExpr, refVars, ParentDecl); 4607 4608 default: 4609 return 0; 4610 } 4611 } 4612 4613 case Stmt::MaterializeTemporaryExprClass: 4614 if (Expr *Result = EvalAddr( 4615 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 4616 refVars, ParentDecl)) 4617 return Result; 4618 4619 return E; 4620 4621 // Everything else: we simply don't reason about them. 4622 default: 4623 return NULL; 4624 } 4625 } 4626 4627 4628 /// EvalVal - This function is complements EvalAddr in the mutual recursion. 4629 /// See the comments for EvalAddr for more details. 4630 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 4631 Decl *ParentDecl) { 4632 do { 4633 // We should only be called for evaluating non-pointer expressions, or 4634 // expressions with a pointer type that are not used as references but instead 4635 // are l-values (e.g., DeclRefExpr with a pointer type). 4636 4637 // Our "symbolic interpreter" is just a dispatch off the currently 4638 // viewed AST node. We then recursively traverse the AST by calling 4639 // EvalAddr and EvalVal appropriately. 4640 4641 E = E->IgnoreParens(); 4642 switch (E->getStmtClass()) { 4643 case Stmt::ImplicitCastExprClass: { 4644 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 4645 if (IE->getValueKind() == VK_LValue) { 4646 E = IE->getSubExpr(); 4647 continue; 4648 } 4649 return NULL; 4650 } 4651 4652 case Stmt::ExprWithCleanupsClass: 4653 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 4654 4655 case Stmt::DeclRefExprClass: { 4656 // When we hit a DeclRefExpr we are looking at code that refers to a 4657 // variable's name. If it's not a reference variable we check if it has 4658 // local storage within the function, and if so, return the expression. 4659 DeclRefExpr *DR = cast<DeclRefExpr>(E); 4660 4661 // If we leave the immediate function, the lifetime isn't about to end. 4662 if (DR->refersToEnclosingLocal()) 4663 return 0; 4664 4665 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 4666 // Check if it refers to itself, e.g. "int& i = i;". 4667 if (V == ParentDecl) 4668 return DR; 4669 4670 if (V->hasLocalStorage()) { 4671 if (!V->getType()->isReferenceType()) 4672 return DR; 4673 4674 // Reference variable, follow through to the expression that 4675 // it points to. 4676 if (V->hasInit()) { 4677 // Add the reference variable to the "trail". 4678 refVars.push_back(DR); 4679 return EvalVal(V->getInit(), refVars, V); 4680 } 4681 } 4682 } 4683 4684 return NULL; 4685 } 4686 4687 case Stmt::UnaryOperatorClass: { 4688 // The only unary operator that make sense to handle here 4689 // is Deref. All others don't resolve to a "name." This includes 4690 // handling all sorts of rvalues passed to a unary operator. 4691 UnaryOperator *U = cast<UnaryOperator>(E); 4692 4693 if (U->getOpcode() == UO_Deref) 4694 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 4695 4696 return NULL; 4697 } 4698 4699 case Stmt::ArraySubscriptExprClass: { 4700 // Array subscripts are potential references to data on the stack. We 4701 // retrieve the DeclRefExpr* for the array variable if it indeed 4702 // has local storage. 4703 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 4704 } 4705 4706 case Stmt::ConditionalOperatorClass: { 4707 // For conditional operators we need to see if either the LHS or RHS are 4708 // non-NULL Expr's. If one is non-NULL, we return it. 4709 ConditionalOperator *C = cast<ConditionalOperator>(E); 4710 4711 // Handle the GNU extension for missing LHS. 4712 if (Expr *LHSExpr = C->getLHS()) { 4713 // In C++, we can have a throw-expression, which has 'void' type. 4714 if (!LHSExpr->getType()->isVoidType()) 4715 if (Expr *LHS = EvalVal(LHSExpr, refVars, ParentDecl)) 4716 return LHS; 4717 } 4718 4719 // In C++, we can have a throw-expression, which has 'void' type. 4720 if (C->getRHS()->getType()->isVoidType()) 4721 return 0; 4722 4723 return EvalVal(C->getRHS(), refVars, ParentDecl); 4724 } 4725 4726 // Accesses to members are potential references to data on the stack. 4727 case Stmt::MemberExprClass: { 4728 MemberExpr *M = cast<MemberExpr>(E); 4729 4730 // Check for indirect access. We only want direct field accesses. 4731 if (M->isArrow()) 4732 return NULL; 4733 4734 // Check whether the member type is itself a reference, in which case 4735 // we're not going to refer to the member, but to what the member refers to. 4736 if (M->getMemberDecl()->getType()->isReferenceType()) 4737 return NULL; 4738 4739 return EvalVal(M->getBase(), refVars, ParentDecl); 4740 } 4741 4742 case Stmt::MaterializeTemporaryExprClass: 4743 if (Expr *Result = EvalVal( 4744 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 4745 refVars, ParentDecl)) 4746 return Result; 4747 4748 return E; 4749 4750 default: 4751 // Check that we don't return or take the address of a reference to a 4752 // temporary. This is only useful in C++. 4753 if (!E->isTypeDependent() && E->isRValue()) 4754 return E; 4755 4756 // Everything else: we simply don't reason about them. 4757 return NULL; 4758 } 4759 } while (true); 4760 } 4761 4762 void 4763 Sema::CheckReturnValExpr(Expr *RetValExp, QualType lhsType, 4764 SourceLocation ReturnLoc, 4765 bool isObjCMethod, 4766 const AttrVec *Attrs, 4767 const FunctionDecl *FD) { 4768 CheckReturnStackAddr(*this, RetValExp, lhsType, ReturnLoc); 4769 4770 // Check if the return value is null but should not be. 4771 if (Attrs && hasSpecificAttr<ReturnsNonNullAttr>(*Attrs) && 4772 CheckNonNullExpr(*this, RetValExp)) 4773 Diag(ReturnLoc, diag::warn_null_ret) 4774 << (isObjCMethod ? 1 : 0) << RetValExp->getSourceRange(); 4775 4776 // C++11 [basic.stc.dynamic.allocation]p4: 4777 // If an allocation function declared with a non-throwing 4778 // exception-specification fails to allocate storage, it shall return 4779 // a null pointer. Any other allocation function that fails to allocate 4780 // storage shall indicate failure only by throwing an exception [...] 4781 if (FD) { 4782 OverloadedOperatorKind Op = FD->getOverloadedOperator(); 4783 if (Op == OO_New || Op == OO_Array_New) { 4784 const FunctionProtoType *Proto 4785 = FD->getType()->castAs<FunctionProtoType>(); 4786 if (!Proto->isNothrow(Context, /*ResultIfDependent*/true) && 4787 CheckNonNullExpr(*this, RetValExp)) 4788 Diag(ReturnLoc, diag::warn_operator_new_returns_null) 4789 << FD << getLangOpts().CPlusPlus11; 4790 } 4791 } 4792 } 4793 4794 //===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 4795 4796 /// Check for comparisons of floating point operands using != and ==. 4797 /// Issue a warning if these are no self-comparisons, as they are not likely 4798 /// to do what the programmer intended. 4799 void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 4800 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 4801 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 4802 4803 // Special case: check for x == x (which is OK). 4804 // Do not emit warnings for such cases. 4805 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 4806 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 4807 if (DRL->getDecl() == DRR->getDecl()) 4808 return; 4809 4810 4811 // Special case: check for comparisons against literals that can be exactly 4812 // represented by APFloat. In such cases, do not emit a warning. This 4813 // is a heuristic: often comparison against such literals are used to 4814 // detect if a value in a variable has not changed. This clearly can 4815 // lead to false negatives. 4816 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 4817 if (FLL->isExact()) 4818 return; 4819 } else 4820 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)) 4821 if (FLR->isExact()) 4822 return; 4823 4824 // Check for comparisons with builtin types. 4825 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 4826 if (CL->getBuiltinCallee()) 4827 return; 4828 4829 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 4830 if (CR->getBuiltinCallee()) 4831 return; 4832 4833 // Emit the diagnostic. 4834 Diag(Loc, diag::warn_floatingpoint_eq) 4835 << LHS->getSourceRange() << RHS->getSourceRange(); 4836 } 4837 4838 //===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 4839 //===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 4840 4841 namespace { 4842 4843 /// Structure recording the 'active' range of an integer-valued 4844 /// expression. 4845 struct IntRange { 4846 /// The number of bits active in the int. 4847 unsigned Width; 4848 4849 /// True if the int is known not to have negative values. 4850 bool NonNegative; 4851 4852 IntRange(unsigned Width, bool NonNegative) 4853 : Width(Width), NonNegative(NonNegative) 4854 {} 4855 4856 /// Returns the range of the bool type. 4857 static IntRange forBoolType() { 4858 return IntRange(1, true); 4859 } 4860 4861 /// Returns the range of an opaque value of the given integral type. 4862 static IntRange forValueOfType(ASTContext &C, QualType T) { 4863 return forValueOfCanonicalType(C, 4864 T->getCanonicalTypeInternal().getTypePtr()); 4865 } 4866 4867 /// Returns the range of an opaque value of a canonical integral type. 4868 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 4869 assert(T->isCanonicalUnqualified()); 4870 4871 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4872 T = VT->getElementType().getTypePtr(); 4873 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4874 T = CT->getElementType().getTypePtr(); 4875 4876 // For enum types, use the known bit width of the enumerators. 4877 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 4878 EnumDecl *Enum = ET->getDecl(); 4879 if (!Enum->isCompleteDefinition()) 4880 return IntRange(C.getIntWidth(QualType(T, 0)), false); 4881 4882 unsigned NumPositive = Enum->getNumPositiveBits(); 4883 unsigned NumNegative = Enum->getNumNegativeBits(); 4884 4885 if (NumNegative == 0) 4886 return IntRange(NumPositive, true/*NonNegative*/); 4887 else 4888 return IntRange(std::max(NumPositive + 1, NumNegative), 4889 false/*NonNegative*/); 4890 } 4891 4892 const BuiltinType *BT = cast<BuiltinType>(T); 4893 assert(BT->isInteger()); 4894 4895 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4896 } 4897 4898 /// Returns the "target" range of a canonical integral type, i.e. 4899 /// the range of values expressible in the type. 4900 /// 4901 /// This matches forValueOfCanonicalType except that enums have the 4902 /// full range of their type, not the range of their enumerators. 4903 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 4904 assert(T->isCanonicalUnqualified()); 4905 4906 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4907 T = VT->getElementType().getTypePtr(); 4908 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4909 T = CT->getElementType().getTypePtr(); 4910 if (const EnumType *ET = dyn_cast<EnumType>(T)) 4911 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 4912 4913 const BuiltinType *BT = cast<BuiltinType>(T); 4914 assert(BT->isInteger()); 4915 4916 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4917 } 4918 4919 /// Returns the supremum of two ranges: i.e. their conservative merge. 4920 static IntRange join(IntRange L, IntRange R) { 4921 return IntRange(std::max(L.Width, R.Width), 4922 L.NonNegative && R.NonNegative); 4923 } 4924 4925 /// Returns the infinum of two ranges: i.e. their aggressive merge. 4926 static IntRange meet(IntRange L, IntRange R) { 4927 return IntRange(std::min(L.Width, R.Width), 4928 L.NonNegative || R.NonNegative); 4929 } 4930 }; 4931 4932 static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 4933 unsigned MaxWidth) { 4934 if (value.isSigned() && value.isNegative()) 4935 return IntRange(value.getMinSignedBits(), false); 4936 4937 if (value.getBitWidth() > MaxWidth) 4938 value = value.trunc(MaxWidth); 4939 4940 // isNonNegative() just checks the sign bit without considering 4941 // signedness. 4942 return IntRange(value.getActiveBits(), true); 4943 } 4944 4945 static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 4946 unsigned MaxWidth) { 4947 if (result.isInt()) 4948 return GetValueRange(C, result.getInt(), MaxWidth); 4949 4950 if (result.isVector()) { 4951 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 4952 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 4953 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 4954 R = IntRange::join(R, El); 4955 } 4956 return R; 4957 } 4958 4959 if (result.isComplexInt()) { 4960 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 4961 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 4962 return IntRange::join(R, I); 4963 } 4964 4965 // This can happen with lossless casts to intptr_t of "based" lvalues. 4966 // Assume it might use arbitrary bits. 4967 // FIXME: The only reason we need to pass the type in here is to get 4968 // the sign right on this one case. It would be nice if APValue 4969 // preserved this. 4970 assert(result.isLValue() || result.isAddrLabelDiff()); 4971 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 4972 } 4973 4974 static QualType GetExprType(Expr *E) { 4975 QualType Ty = E->getType(); 4976 if (const AtomicType *AtomicRHS = Ty->getAs<AtomicType>()) 4977 Ty = AtomicRHS->getValueType(); 4978 return Ty; 4979 } 4980 4981 /// Pseudo-evaluate the given integer expression, estimating the 4982 /// range of values it might take. 4983 /// 4984 /// \param MaxWidth - the width to which the value will be truncated 4985 static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 4986 E = E->IgnoreParens(); 4987 4988 // Try a full evaluation first. 4989 Expr::EvalResult result; 4990 if (E->EvaluateAsRValue(result, C)) 4991 return GetValueRange(C, result.Val, GetExprType(E), MaxWidth); 4992 4993 // I think we only want to look through implicit casts here; if the 4994 // user has an explicit widening cast, we should treat the value as 4995 // being of the new, wider type. 4996 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 4997 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 4998 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 4999 5000 IntRange OutputTypeRange = IntRange::forValueOfType(C, GetExprType(CE)); 5001 5002 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 5003 5004 // Assume that non-integer casts can span the full range of the type. 5005 if (!isIntegerCast) 5006 return OutputTypeRange; 5007 5008 IntRange SubRange 5009 = GetExprRange(C, CE->getSubExpr(), 5010 std::min(MaxWidth, OutputTypeRange.Width)); 5011 5012 // Bail out if the subexpr's range is as wide as the cast type. 5013 if (SubRange.Width >= OutputTypeRange.Width) 5014 return OutputTypeRange; 5015 5016 // Otherwise, we take the smaller width, and we're non-negative if 5017 // either the output type or the subexpr is. 5018 return IntRange(SubRange.Width, 5019 SubRange.NonNegative || OutputTypeRange.NonNegative); 5020 } 5021 5022 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 5023 // If we can fold the condition, just take that operand. 5024 bool CondResult; 5025 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 5026 return GetExprRange(C, CondResult ? CO->getTrueExpr() 5027 : CO->getFalseExpr(), 5028 MaxWidth); 5029 5030 // Otherwise, conservatively merge. 5031 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 5032 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 5033 return IntRange::join(L, R); 5034 } 5035 5036 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5037 switch (BO->getOpcode()) { 5038 5039 // Boolean-valued operations are single-bit and positive. 5040 case BO_LAnd: 5041 case BO_LOr: 5042 case BO_LT: 5043 case BO_GT: 5044 case BO_LE: 5045 case BO_GE: 5046 case BO_EQ: 5047 case BO_NE: 5048 return IntRange::forBoolType(); 5049 5050 // The type of the assignments is the type of the LHS, so the RHS 5051 // is not necessarily the same type. 5052 case BO_MulAssign: 5053 case BO_DivAssign: 5054 case BO_RemAssign: 5055 case BO_AddAssign: 5056 case BO_SubAssign: 5057 case BO_XorAssign: 5058 case BO_OrAssign: 5059 // TODO: bitfields? 5060 return IntRange::forValueOfType(C, GetExprType(E)); 5061 5062 // Simple assignments just pass through the RHS, which will have 5063 // been coerced to the LHS type. 5064 case BO_Assign: 5065 // TODO: bitfields? 5066 return GetExprRange(C, BO->getRHS(), MaxWidth); 5067 5068 // Operations with opaque sources are black-listed. 5069 case BO_PtrMemD: 5070 case BO_PtrMemI: 5071 return IntRange::forValueOfType(C, GetExprType(E)); 5072 5073 // Bitwise-and uses the *infinum* of the two source ranges. 5074 case BO_And: 5075 case BO_AndAssign: 5076 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 5077 GetExprRange(C, BO->getRHS(), MaxWidth)); 5078 5079 // Left shift gets black-listed based on a judgement call. 5080 case BO_Shl: 5081 // ...except that we want to treat '1 << (blah)' as logically 5082 // positive. It's an important idiom. 5083 if (IntegerLiteral *I 5084 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 5085 if (I->getValue() == 1) { 5086 IntRange R = IntRange::forValueOfType(C, GetExprType(E)); 5087 return IntRange(R.Width, /*NonNegative*/ true); 5088 } 5089 } 5090 // fallthrough 5091 5092 case BO_ShlAssign: 5093 return IntRange::forValueOfType(C, GetExprType(E)); 5094 5095 // Right shift by a constant can narrow its left argument. 5096 case BO_Shr: 5097 case BO_ShrAssign: { 5098 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 5099 5100 // If the shift amount is a positive constant, drop the width by 5101 // that much. 5102 llvm::APSInt shift; 5103 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 5104 shift.isNonNegative()) { 5105 unsigned zext = shift.getZExtValue(); 5106 if (zext >= L.Width) 5107 L.Width = (L.NonNegative ? 0 : 1); 5108 else 5109 L.Width -= zext; 5110 } 5111 5112 return L; 5113 } 5114 5115 // Comma acts as its right operand. 5116 case BO_Comma: 5117 return GetExprRange(C, BO->getRHS(), MaxWidth); 5118 5119 // Black-list pointer subtractions. 5120 case BO_Sub: 5121 if (BO->getLHS()->getType()->isPointerType()) 5122 return IntRange::forValueOfType(C, GetExprType(E)); 5123 break; 5124 5125 // The width of a division result is mostly determined by the size 5126 // of the LHS. 5127 case BO_Div: { 5128 // Don't 'pre-truncate' the operands. 5129 unsigned opWidth = C.getIntWidth(GetExprType(E)); 5130 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 5131 5132 // If the divisor is constant, use that. 5133 llvm::APSInt divisor; 5134 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 5135 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 5136 if (log2 >= L.Width) 5137 L.Width = (L.NonNegative ? 0 : 1); 5138 else 5139 L.Width = std::min(L.Width - log2, MaxWidth); 5140 return L; 5141 } 5142 5143 // Otherwise, just use the LHS's width. 5144 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 5145 return IntRange(L.Width, L.NonNegative && R.NonNegative); 5146 } 5147 5148 // The result of a remainder can't be larger than the result of 5149 // either side. 5150 case BO_Rem: { 5151 // Don't 'pre-truncate' the operands. 5152 unsigned opWidth = C.getIntWidth(GetExprType(E)); 5153 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 5154 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 5155 5156 IntRange meet = IntRange::meet(L, R); 5157 meet.Width = std::min(meet.Width, MaxWidth); 5158 return meet; 5159 } 5160 5161 // The default behavior is okay for these. 5162 case BO_Mul: 5163 case BO_Add: 5164 case BO_Xor: 5165 case BO_Or: 5166 break; 5167 } 5168 5169 // The default case is to treat the operation as if it were closed 5170 // on the narrowest type that encompasses both operands. 5171 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 5172 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 5173 return IntRange::join(L, R); 5174 } 5175 5176 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 5177 switch (UO->getOpcode()) { 5178 // Boolean-valued operations are white-listed. 5179 case UO_LNot: 5180 return IntRange::forBoolType(); 5181 5182 // Operations with opaque sources are black-listed. 5183 case UO_Deref: 5184 case UO_AddrOf: // should be impossible 5185 return IntRange::forValueOfType(C, GetExprType(E)); 5186 5187 default: 5188 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 5189 } 5190 } 5191 5192 if (OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E)) 5193 return GetExprRange(C, OVE->getSourceExpr(), MaxWidth); 5194 5195 if (FieldDecl *BitField = E->getSourceBitField()) 5196 return IntRange(BitField->getBitWidthValue(C), 5197 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 5198 5199 return IntRange::forValueOfType(C, GetExprType(E)); 5200 } 5201 5202 static IntRange GetExprRange(ASTContext &C, Expr *E) { 5203 return GetExprRange(C, E, C.getIntWidth(GetExprType(E))); 5204 } 5205 5206 /// Checks whether the given value, which currently has the given 5207 /// source semantics, has the same value when coerced through the 5208 /// target semantics. 5209 static bool IsSameFloatAfterCast(const llvm::APFloat &value, 5210 const llvm::fltSemantics &Src, 5211 const llvm::fltSemantics &Tgt) { 5212 llvm::APFloat truncated = value; 5213 5214 bool ignored; 5215 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 5216 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 5217 5218 return truncated.bitwiseIsEqual(value); 5219 } 5220 5221 /// Checks whether the given value, which currently has the given 5222 /// source semantics, has the same value when coerced through the 5223 /// target semantics. 5224 /// 5225 /// The value might be a vector of floats (or a complex number). 5226 static bool IsSameFloatAfterCast(const APValue &value, 5227 const llvm::fltSemantics &Src, 5228 const llvm::fltSemantics &Tgt) { 5229 if (value.isFloat()) 5230 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 5231 5232 if (value.isVector()) { 5233 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 5234 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 5235 return false; 5236 return true; 5237 } 5238 5239 assert(value.isComplexFloat()); 5240 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 5241 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 5242 } 5243 5244 static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 5245 5246 static bool IsZero(Sema &S, Expr *E) { 5247 // Suppress cases where we are comparing against an enum constant. 5248 if (const DeclRefExpr *DR = 5249 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 5250 if (isa<EnumConstantDecl>(DR->getDecl())) 5251 return false; 5252 5253 // Suppress cases where the '0' value is expanded from a macro. 5254 if (E->getLocStart().isMacroID()) 5255 return false; 5256 5257 llvm::APSInt Value; 5258 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 5259 } 5260 5261 static bool HasEnumType(Expr *E) { 5262 // Strip off implicit integral promotions. 5263 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 5264 if (ICE->getCastKind() != CK_IntegralCast && 5265 ICE->getCastKind() != CK_NoOp) 5266 break; 5267 E = ICE->getSubExpr(); 5268 } 5269 5270 return E->getType()->isEnumeralType(); 5271 } 5272 5273 static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 5274 // Disable warning in template instantiations. 5275 if (!S.ActiveTemplateInstantiations.empty()) 5276 return; 5277 5278 BinaryOperatorKind op = E->getOpcode(); 5279 if (E->isValueDependent()) 5280 return; 5281 5282 if (op == BO_LT && IsZero(S, E->getRHS())) { 5283 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 5284 << "< 0" << "false" << HasEnumType(E->getLHS()) 5285 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 5286 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 5287 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 5288 << ">= 0" << "true" << HasEnumType(E->getLHS()) 5289 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 5290 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 5291 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 5292 << "0 >" << "false" << HasEnumType(E->getRHS()) 5293 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 5294 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 5295 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 5296 << "0 <=" << "true" << HasEnumType(E->getRHS()) 5297 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 5298 } 5299 } 5300 5301 static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E, 5302 Expr *Constant, Expr *Other, 5303 llvm::APSInt Value, 5304 bool RhsConstant) { 5305 // Disable warning in template instantiations. 5306 if (!S.ActiveTemplateInstantiations.empty()) 5307 return; 5308 5309 // 0 values are handled later by CheckTrivialUnsignedComparison(). 5310 if (Value == 0) 5311 return; 5312 5313 BinaryOperatorKind op = E->getOpcode(); 5314 QualType OtherT = Other->getType(); 5315 QualType ConstantT = Constant->getType(); 5316 QualType CommonT = E->getLHS()->getType(); 5317 if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT)) 5318 return; 5319 assert((OtherT->isIntegerType() && ConstantT->isIntegerType()) 5320 && "comparison with non-integer type"); 5321 5322 bool ConstantSigned = ConstantT->isSignedIntegerType(); 5323 bool CommonSigned = CommonT->isSignedIntegerType(); 5324 5325 bool EqualityOnly = false; 5326 5327 // TODO: Investigate using GetExprRange() to get tighter bounds on 5328 // on the bit ranges. 5329 IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT); 5330 unsigned OtherWidth = OtherRange.Width; 5331 5332 if (CommonSigned) { 5333 // The common type is signed, therefore no signed to unsigned conversion. 5334 if (!OtherRange.NonNegative) { 5335 // Check that the constant is representable in type OtherT. 5336 if (ConstantSigned) { 5337 if (OtherWidth >= Value.getMinSignedBits()) 5338 return; 5339 } else { // !ConstantSigned 5340 if (OtherWidth >= Value.getActiveBits() + 1) 5341 return; 5342 } 5343 } else { // !OtherSigned 5344 // Check that the constant is representable in type OtherT. 5345 // Negative values are out of range. 5346 if (ConstantSigned) { 5347 if (Value.isNonNegative() && OtherWidth >= Value.getActiveBits()) 5348 return; 5349 } else { // !ConstantSigned 5350 if (OtherWidth >= Value.getActiveBits()) 5351 return; 5352 } 5353 } 5354 } else { // !CommonSigned 5355 if (OtherRange.NonNegative) { 5356 if (OtherWidth >= Value.getActiveBits()) 5357 return; 5358 } else if (!OtherRange.NonNegative && !ConstantSigned) { 5359 // Check to see if the constant is representable in OtherT. 5360 if (OtherWidth > Value.getActiveBits()) 5361 return; 5362 // Check to see if the constant is equivalent to a negative value 5363 // cast to CommonT. 5364 if (S.Context.getIntWidth(ConstantT) == S.Context.getIntWidth(CommonT) && 5365 Value.isNegative() && Value.getMinSignedBits() <= OtherWidth) 5366 return; 5367 // The constant value rests between values that OtherT can represent after 5368 // conversion. Relational comparison still works, but equality 5369 // comparisons will be tautological. 5370 EqualityOnly = true; 5371 } else { // OtherSigned && ConstantSigned 5372 assert(0 && "Two signed types converted to unsigned types."); 5373 } 5374 } 5375 5376 bool PositiveConstant = !ConstantSigned || Value.isNonNegative(); 5377 5378 bool IsTrue = true; 5379 if (op == BO_EQ || op == BO_NE) { 5380 IsTrue = op == BO_NE; 5381 } else if (EqualityOnly) { 5382 return; 5383 } else if (RhsConstant) { 5384 if (op == BO_GT || op == BO_GE) 5385 IsTrue = !PositiveConstant; 5386 else // op == BO_LT || op == BO_LE 5387 IsTrue = PositiveConstant; 5388 } else { 5389 if (op == BO_LT || op == BO_LE) 5390 IsTrue = !PositiveConstant; 5391 else // op == BO_GT || op == BO_GE 5392 IsTrue = PositiveConstant; 5393 } 5394 5395 // If this is a comparison to an enum constant, include that 5396 // constant in the diagnostic. 5397 const EnumConstantDecl *ED = 0; 5398 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Constant)) 5399 ED = dyn_cast<EnumConstantDecl>(DR->getDecl()); 5400 5401 SmallString<64> PrettySourceValue; 5402 llvm::raw_svector_ostream OS(PrettySourceValue); 5403 if (ED) 5404 OS << '\'' << *ED << "' (" << Value << ")"; 5405 else 5406 OS << Value; 5407 5408 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 5409 S.PDiag(diag::warn_out_of_range_compare) 5410 << OS.str() << OtherT << IsTrue 5411 << E->getLHS()->getSourceRange() 5412 << E->getRHS()->getSourceRange()); 5413 } 5414 5415 /// Analyze the operands of the given comparison. Implements the 5416 /// fallback case from AnalyzeComparison. 5417 static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 5418 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 5419 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 5420 } 5421 5422 /// \brief Implements -Wsign-compare. 5423 /// 5424 /// \param E the binary operator to check for warnings 5425 static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 5426 // The type the comparison is being performed in. 5427 QualType T = E->getLHS()->getType(); 5428 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 5429 && "comparison with mismatched types"); 5430 if (E->isValueDependent()) 5431 return AnalyzeImpConvsInComparison(S, E); 5432 5433 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 5434 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 5435 5436 bool IsComparisonConstant = false; 5437 5438 // Check whether an integer constant comparison results in a value 5439 // of 'true' or 'false'. 5440 if (T->isIntegralType(S.Context)) { 5441 llvm::APSInt RHSValue; 5442 bool IsRHSIntegralLiteral = 5443 RHS->isIntegerConstantExpr(RHSValue, S.Context); 5444 llvm::APSInt LHSValue; 5445 bool IsLHSIntegralLiteral = 5446 LHS->isIntegerConstantExpr(LHSValue, S.Context); 5447 if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral) 5448 DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true); 5449 else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral) 5450 DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false); 5451 else 5452 IsComparisonConstant = 5453 (IsRHSIntegralLiteral && IsLHSIntegralLiteral); 5454 } else if (!T->hasUnsignedIntegerRepresentation()) 5455 IsComparisonConstant = E->isIntegerConstantExpr(S.Context); 5456 5457 // We don't do anything special if this isn't an unsigned integral 5458 // comparison: we're only interested in integral comparisons, and 5459 // signed comparisons only happen in cases we don't care to warn about. 5460 // 5461 // We also don't care about value-dependent expressions or expressions 5462 // whose result is a constant. 5463 if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant) 5464 return AnalyzeImpConvsInComparison(S, E); 5465 5466 // Check to see if one of the (unmodified) operands is of different 5467 // signedness. 5468 Expr *signedOperand, *unsignedOperand; 5469 if (LHS->getType()->hasSignedIntegerRepresentation()) { 5470 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 5471 "unsigned comparison between two signed integer expressions?"); 5472 signedOperand = LHS; 5473 unsignedOperand = RHS; 5474 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 5475 signedOperand = RHS; 5476 unsignedOperand = LHS; 5477 } else { 5478 CheckTrivialUnsignedComparison(S, E); 5479 return AnalyzeImpConvsInComparison(S, E); 5480 } 5481 5482 // Otherwise, calculate the effective range of the signed operand. 5483 IntRange signedRange = GetExprRange(S.Context, signedOperand); 5484 5485 // Go ahead and analyze implicit conversions in the operands. Note 5486 // that we skip the implicit conversions on both sides. 5487 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 5488 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 5489 5490 // If the signed range is non-negative, -Wsign-compare won't fire, 5491 // but we should still check for comparisons which are always true 5492 // or false. 5493 if (signedRange.NonNegative) 5494 return CheckTrivialUnsignedComparison(S, E); 5495 5496 // For (in)equality comparisons, if the unsigned operand is a 5497 // constant which cannot collide with a overflowed signed operand, 5498 // then reinterpreting the signed operand as unsigned will not 5499 // change the result of the comparison. 5500 if (E->isEqualityOp()) { 5501 unsigned comparisonWidth = S.Context.getIntWidth(T); 5502 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 5503 5504 // We should never be unable to prove that the unsigned operand is 5505 // non-negative. 5506 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 5507 5508 if (unsignedRange.Width < comparisonWidth) 5509 return; 5510 } 5511 5512 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 5513 S.PDiag(diag::warn_mixed_sign_comparison) 5514 << LHS->getType() << RHS->getType() 5515 << LHS->getSourceRange() << RHS->getSourceRange()); 5516 } 5517 5518 /// Analyzes an attempt to assign the given value to a bitfield. 5519 /// 5520 /// Returns true if there was something fishy about the attempt. 5521 static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 5522 SourceLocation InitLoc) { 5523 assert(Bitfield->isBitField()); 5524 if (Bitfield->isInvalidDecl()) 5525 return false; 5526 5527 // White-list bool bitfields. 5528 if (Bitfield->getType()->isBooleanType()) 5529 return false; 5530 5531 // Ignore value- or type-dependent expressions. 5532 if (Bitfield->getBitWidth()->isValueDependent() || 5533 Bitfield->getBitWidth()->isTypeDependent() || 5534 Init->isValueDependent() || 5535 Init->isTypeDependent()) 5536 return false; 5537 5538 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 5539 5540 llvm::APSInt Value; 5541 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 5542 return false; 5543 5544 unsigned OriginalWidth = Value.getBitWidth(); 5545 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 5546 5547 if (OriginalWidth <= FieldWidth) 5548 return false; 5549 5550 // Compute the value which the bitfield will contain. 5551 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 5552 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 5553 5554 // Check whether the stored value is equal to the original value. 5555 TruncatedValue = TruncatedValue.extend(OriginalWidth); 5556 if (llvm::APSInt::isSameValue(Value, TruncatedValue)) 5557 return false; 5558 5559 // Special-case bitfields of width 1: booleans are naturally 0/1, and 5560 // therefore don't strictly fit into a signed bitfield of width 1. 5561 if (FieldWidth == 1 && Value == 1) 5562 return false; 5563 5564 std::string PrettyValue = Value.toString(10); 5565 std::string PrettyTrunc = TruncatedValue.toString(10); 5566 5567 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 5568 << PrettyValue << PrettyTrunc << OriginalInit->getType() 5569 << Init->getSourceRange(); 5570 5571 return true; 5572 } 5573 5574 /// Analyze the given simple or compound assignment for warning-worthy 5575 /// operations. 5576 static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 5577 // Just recurse on the LHS. 5578 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 5579 5580 // We want to recurse on the RHS as normal unless we're assigning to 5581 // a bitfield. 5582 if (FieldDecl *Bitfield = E->getLHS()->getSourceBitField()) { 5583 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 5584 E->getOperatorLoc())) { 5585 // Recurse, ignoring any implicit conversions on the RHS. 5586 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 5587 E->getOperatorLoc()); 5588 } 5589 } 5590 5591 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 5592 } 5593 5594 /// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 5595 static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 5596 SourceLocation CContext, unsigned diag, 5597 bool pruneControlFlow = false) { 5598 if (pruneControlFlow) { 5599 S.DiagRuntimeBehavior(E->getExprLoc(), E, 5600 S.PDiag(diag) 5601 << SourceType << T << E->getSourceRange() 5602 << SourceRange(CContext)); 5603 return; 5604 } 5605 S.Diag(E->getExprLoc(), diag) 5606 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 5607 } 5608 5609 /// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 5610 static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 5611 SourceLocation CContext, unsigned diag, 5612 bool pruneControlFlow = false) { 5613 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 5614 } 5615 5616 /// Diagnose an implicit cast from a literal expression. Does not warn when the 5617 /// cast wouldn't lose information. 5618 void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 5619 SourceLocation CContext) { 5620 // Try to convert the literal exactly to an integer. If we can, don't warn. 5621 bool isExact = false; 5622 const llvm::APFloat &Value = FL->getValue(); 5623 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 5624 T->hasUnsignedIntegerRepresentation()); 5625 if (Value.convertToInteger(IntegerValue, 5626 llvm::APFloat::rmTowardZero, &isExact) 5627 == llvm::APFloat::opOK && isExact) 5628 return; 5629 5630 // FIXME: Force the precision of the source value down so we don't print 5631 // digits which are usually useless (we don't really care here if we 5632 // truncate a digit by accident in edge cases). Ideally, APFloat::toString 5633 // would automatically print the shortest representation, but it's a bit 5634 // tricky to implement. 5635 SmallString<16> PrettySourceValue; 5636 unsigned precision = llvm::APFloat::semanticsPrecision(Value.getSemantics()); 5637 precision = (precision * 59 + 195) / 196; 5638 Value.toString(PrettySourceValue, precision); 5639 5640 SmallString<16> PrettyTargetValue; 5641 if (T->isSpecificBuiltinType(BuiltinType::Bool)) 5642 PrettyTargetValue = IntegerValue == 0 ? "false" : "true"; 5643 else 5644 IntegerValue.toString(PrettyTargetValue); 5645 5646 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 5647 << FL->getType() << T.getUnqualifiedType() << PrettySourceValue 5648 << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext); 5649 } 5650 5651 std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 5652 if (!Range.Width) return "0"; 5653 5654 llvm::APSInt ValueInRange = Value; 5655 ValueInRange.setIsSigned(!Range.NonNegative); 5656 ValueInRange = ValueInRange.trunc(Range.Width); 5657 return ValueInRange.toString(10); 5658 } 5659 5660 static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) { 5661 if (!isa<ImplicitCastExpr>(Ex)) 5662 return false; 5663 5664 Expr *InnerE = Ex->IgnoreParenImpCasts(); 5665 const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr(); 5666 const Type *Source = 5667 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 5668 if (Target->isDependentType()) 5669 return false; 5670 5671 const BuiltinType *FloatCandidateBT = 5672 dyn_cast<BuiltinType>(ToBool ? Source : Target); 5673 const Type *BoolCandidateType = ToBool ? Target : Source; 5674 5675 return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) && 5676 FloatCandidateBT && (FloatCandidateBT->isFloatingPoint())); 5677 } 5678 5679 void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall, 5680 SourceLocation CC) { 5681 unsigned NumArgs = TheCall->getNumArgs(); 5682 for (unsigned i = 0; i < NumArgs; ++i) { 5683 Expr *CurrA = TheCall->getArg(i); 5684 if (!IsImplicitBoolFloatConversion(S, CurrA, true)) 5685 continue; 5686 5687 bool IsSwapped = ((i > 0) && 5688 IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false)); 5689 IsSwapped |= ((i < (NumArgs - 1)) && 5690 IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false)); 5691 if (IsSwapped) { 5692 // Warn on this floating-point to bool conversion. 5693 DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(), 5694 CurrA->getType(), CC, 5695 diag::warn_impcast_floating_point_to_bool); 5696 } 5697 } 5698 } 5699 5700 void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 5701 SourceLocation CC, bool *ICContext = 0) { 5702 if (E->isTypeDependent() || E->isValueDependent()) return; 5703 5704 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 5705 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 5706 if (Source == Target) return; 5707 if (Target->isDependentType()) return; 5708 5709 // If the conversion context location is invalid don't complain. We also 5710 // don't want to emit a warning if the issue occurs from the expansion of 5711 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 5712 // delay this check as long as possible. Once we detect we are in that 5713 // scenario, we just return. 5714 if (CC.isInvalid()) 5715 return; 5716 5717 // Diagnose implicit casts to bool. 5718 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 5719 if (isa<StringLiteral>(E)) 5720 // Warn on string literal to bool. Checks for string literals in logical 5721 // and expressions, for instance, assert(0 && "error here"), are 5722 // prevented by a check in AnalyzeImplicitConversions(). 5723 return DiagnoseImpCast(S, E, T, CC, 5724 diag::warn_impcast_string_literal_to_bool); 5725 if (isa<ObjCStringLiteral>(E) || isa<ObjCArrayLiteral>(E) || 5726 isa<ObjCDictionaryLiteral>(E) || isa<ObjCBoxedExpr>(E)) { 5727 // This covers the literal expressions that evaluate to Objective-C 5728 // objects. 5729 return DiagnoseImpCast(S, E, T, CC, 5730 diag::warn_impcast_objective_c_literal_to_bool); 5731 } 5732 if (Source->isPointerType() || Source->canDecayToPointerType()) { 5733 // Warn on pointer to bool conversion that is always true. 5734 S.DiagnoseAlwaysNonNullPointer(E, Expr::NPCK_NotNull, /*IsEqual*/ false, 5735 SourceRange(CC)); 5736 } 5737 } 5738 5739 // Strip vector types. 5740 if (isa<VectorType>(Source)) { 5741 if (!isa<VectorType>(Target)) { 5742 if (S.SourceMgr.isInSystemMacro(CC)) 5743 return; 5744 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 5745 } 5746 5747 // If the vector cast is cast between two vectors of the same size, it is 5748 // a bitcast, not a conversion. 5749 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 5750 return; 5751 5752 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 5753 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 5754 } 5755 5756 // Strip complex types. 5757 if (isa<ComplexType>(Source)) { 5758 if (!isa<ComplexType>(Target)) { 5759 if (S.SourceMgr.isInSystemMacro(CC)) 5760 return; 5761 5762 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 5763 } 5764 5765 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 5766 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 5767 } 5768 5769 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 5770 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 5771 5772 // If the source is floating point... 5773 if (SourceBT && SourceBT->isFloatingPoint()) { 5774 // ...and the target is floating point... 5775 if (TargetBT && TargetBT->isFloatingPoint()) { 5776 // ...then warn if we're dropping FP rank. 5777 5778 // Builtin FP kinds are ordered by increasing FP rank. 5779 if (SourceBT->getKind() > TargetBT->getKind()) { 5780 // Don't warn about float constants that are precisely 5781 // representable in the target type. 5782 Expr::EvalResult result; 5783 if (E->EvaluateAsRValue(result, S.Context)) { 5784 // Value might be a float, a float vector, or a float complex. 5785 if (IsSameFloatAfterCast(result.Val, 5786 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 5787 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 5788 return; 5789 } 5790 5791 if (S.SourceMgr.isInSystemMacro(CC)) 5792 return; 5793 5794 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 5795 } 5796 return; 5797 } 5798 5799 // If the target is integral, always warn. 5800 if (TargetBT && TargetBT->isInteger()) { 5801 if (S.SourceMgr.isInSystemMacro(CC)) 5802 return; 5803 5804 Expr *InnerE = E->IgnoreParenImpCasts(); 5805 // We also want to warn on, e.g., "int i = -1.234" 5806 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 5807 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 5808 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 5809 5810 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 5811 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 5812 } else { 5813 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 5814 } 5815 } 5816 5817 // If the target is bool, warn if expr is a function or method call. 5818 if (Target->isSpecificBuiltinType(BuiltinType::Bool) && 5819 isa<CallExpr>(E)) { 5820 // Check last argument of function call to see if it is an 5821 // implicit cast from a type matching the type the result 5822 // is being cast to. 5823 CallExpr *CEx = cast<CallExpr>(E); 5824 unsigned NumArgs = CEx->getNumArgs(); 5825 if (NumArgs > 0) { 5826 Expr *LastA = CEx->getArg(NumArgs - 1); 5827 Expr *InnerE = LastA->IgnoreParenImpCasts(); 5828 const Type *InnerType = 5829 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 5830 if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) { 5831 // Warn on this floating-point to bool conversion 5832 DiagnoseImpCast(S, E, T, CC, 5833 diag::warn_impcast_floating_point_to_bool); 5834 } 5835 } 5836 } 5837 return; 5838 } 5839 5840 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 5841 == Expr::NPCK_GNUNull) && !Target->isAnyPointerType() 5842 && !Target->isBlockPointerType() && !Target->isMemberPointerType() 5843 && Target->isScalarType() && !Target->isNullPtrType()) { 5844 SourceLocation Loc = E->getSourceRange().getBegin(); 5845 if (Loc.isMacroID()) 5846 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 5847 if (!Loc.isMacroID() || CC.isMacroID()) 5848 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 5849 << T << clang::SourceRange(CC) 5850 << FixItHint::CreateReplacement(Loc, 5851 S.getFixItZeroLiteralForType(T, Loc)); 5852 } 5853 5854 if (!Source->isIntegerType() || !Target->isIntegerType()) 5855 return; 5856 5857 // TODO: remove this early return once the false positives for constant->bool 5858 // in templates, macros, etc, are reduced or removed. 5859 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 5860 return; 5861 5862 IntRange SourceRange = GetExprRange(S.Context, E); 5863 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 5864 5865 if (SourceRange.Width > TargetRange.Width) { 5866 // If the source is a constant, use a default-on diagnostic. 5867 // TODO: this should happen for bitfield stores, too. 5868 llvm::APSInt Value(32); 5869 if (E->isIntegerConstantExpr(Value, S.Context)) { 5870 if (S.SourceMgr.isInSystemMacro(CC)) 5871 return; 5872 5873 std::string PrettySourceValue = Value.toString(10); 5874 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 5875 5876 S.DiagRuntimeBehavior(E->getExprLoc(), E, 5877 S.PDiag(diag::warn_impcast_integer_precision_constant) 5878 << PrettySourceValue << PrettyTargetValue 5879 << E->getType() << T << E->getSourceRange() 5880 << clang::SourceRange(CC)); 5881 return; 5882 } 5883 5884 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 5885 if (S.SourceMgr.isInSystemMacro(CC)) 5886 return; 5887 5888 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 5889 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 5890 /* pruneControlFlow */ true); 5891 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 5892 } 5893 5894 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 5895 (!TargetRange.NonNegative && SourceRange.NonNegative && 5896 SourceRange.Width == TargetRange.Width)) { 5897 5898 if (S.SourceMgr.isInSystemMacro(CC)) 5899 return; 5900 5901 unsigned DiagID = diag::warn_impcast_integer_sign; 5902 5903 // Traditionally, gcc has warned about this under -Wsign-compare. 5904 // We also want to warn about it in -Wconversion. 5905 // So if -Wconversion is off, use a completely identical diagnostic 5906 // in the sign-compare group. 5907 // The conditional-checking code will 5908 if (ICContext) { 5909 DiagID = diag::warn_impcast_integer_sign_conditional; 5910 *ICContext = true; 5911 } 5912 5913 return DiagnoseImpCast(S, E, T, CC, DiagID); 5914 } 5915 5916 // Diagnose conversions between different enumeration types. 5917 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 5918 // type, to give us better diagnostics. 5919 QualType SourceType = E->getType(); 5920 if (!S.getLangOpts().CPlusPlus) { 5921 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5922 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 5923 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 5924 SourceType = S.Context.getTypeDeclType(Enum); 5925 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 5926 } 5927 } 5928 5929 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 5930 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 5931 if (SourceEnum->getDecl()->hasNameForLinkage() && 5932 TargetEnum->getDecl()->hasNameForLinkage() && 5933 SourceEnum != TargetEnum) { 5934 if (S.SourceMgr.isInSystemMacro(CC)) 5935 return; 5936 5937 return DiagnoseImpCast(S, E, SourceType, T, CC, 5938 diag::warn_impcast_different_enum_types); 5939 } 5940 5941 return; 5942 } 5943 5944 void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5945 SourceLocation CC, QualType T); 5946 5947 void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 5948 SourceLocation CC, bool &ICContext) { 5949 E = E->IgnoreParenImpCasts(); 5950 5951 if (isa<ConditionalOperator>(E)) 5952 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T); 5953 5954 AnalyzeImplicitConversions(S, E, CC); 5955 if (E->getType() != T) 5956 return CheckImplicitConversion(S, E, T, CC, &ICContext); 5957 return; 5958 } 5959 5960 void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5961 SourceLocation CC, QualType T) { 5962 AnalyzeImplicitConversions(S, E->getCond(), CC); 5963 5964 bool Suspicious = false; 5965 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 5966 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 5967 5968 // If -Wconversion would have warned about either of the candidates 5969 // for a signedness conversion to the context type... 5970 if (!Suspicious) return; 5971 5972 // ...but it's currently ignored... 5973 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 5974 CC)) 5975 return; 5976 5977 // ...then check whether it would have warned about either of the 5978 // candidates for a signedness conversion to the condition type. 5979 if (E->getType() == T) return; 5980 5981 Suspicious = false; 5982 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 5983 E->getType(), CC, &Suspicious); 5984 if (!Suspicious) 5985 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 5986 E->getType(), CC, &Suspicious); 5987 } 5988 5989 /// AnalyzeImplicitConversions - Find and report any interesting 5990 /// implicit conversions in the given expression. There are a couple 5991 /// of competing diagnostics here, -Wconversion and -Wsign-compare. 5992 void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 5993 QualType T = OrigE->getType(); 5994 Expr *E = OrigE->IgnoreParenImpCasts(); 5995 5996 if (E->isTypeDependent() || E->isValueDependent()) 5997 return; 5998 5999 // For conditional operators, we analyze the arguments as if they 6000 // were being fed directly into the output. 6001 if (isa<ConditionalOperator>(E)) { 6002 ConditionalOperator *CO = cast<ConditionalOperator>(E); 6003 CheckConditionalOperator(S, CO, CC, T); 6004 return; 6005 } 6006 6007 // Check implicit argument conversions for function calls. 6008 if (CallExpr *Call = dyn_cast<CallExpr>(E)) 6009 CheckImplicitArgumentConversions(S, Call, CC); 6010 6011 // Go ahead and check any implicit conversions we might have skipped. 6012 // The non-canonical typecheck is just an optimization; 6013 // CheckImplicitConversion will filter out dead implicit conversions. 6014 if (E->getType() != T) 6015 CheckImplicitConversion(S, E, T, CC); 6016 6017 // Now continue drilling into this expression. 6018 6019 if (PseudoObjectExpr * POE = dyn_cast<PseudoObjectExpr>(E)) { 6020 if (POE->getResultExpr()) 6021 E = POE->getResultExpr(); 6022 } 6023 6024 if (const OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E)) 6025 return AnalyzeImplicitConversions(S, OVE->getSourceExpr(), CC); 6026 6027 // Skip past explicit casts. 6028 if (isa<ExplicitCastExpr>(E)) { 6029 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 6030 return AnalyzeImplicitConversions(S, E, CC); 6031 } 6032 6033 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 6034 // Do a somewhat different check with comparison operators. 6035 if (BO->isComparisonOp()) 6036 return AnalyzeComparison(S, BO); 6037 6038 // And with simple assignments. 6039 if (BO->getOpcode() == BO_Assign) 6040 return AnalyzeAssignment(S, BO); 6041 } 6042 6043 // These break the otherwise-useful invariant below. Fortunately, 6044 // we don't really need to recurse into them, because any internal 6045 // expressions should have been analyzed already when they were 6046 // built into statements. 6047 if (isa<StmtExpr>(E)) return; 6048 6049 // Don't descend into unevaluated contexts. 6050 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 6051 6052 // Now just recurse over the expression's children. 6053 CC = E->getExprLoc(); 6054 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 6055 bool IsLogicalAndOperator = BO && BO->getOpcode() == BO_LAnd; 6056 for (Stmt::child_range I = E->children(); I; ++I) { 6057 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 6058 if (!ChildExpr) 6059 continue; 6060 6061 if (IsLogicalAndOperator && 6062 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 6063 // Ignore checking string literals that are in logical and operators. 6064 // This is a common pattern for asserts. 6065 continue; 6066 AnalyzeImplicitConversions(S, ChildExpr, CC); 6067 } 6068 } 6069 6070 } // end anonymous namespace 6071 6072 enum { 6073 AddressOf, 6074 FunctionPointer, 6075 ArrayPointer 6076 }; 6077 6078 /// \brief Diagnose pointers that are always non-null. 6079 /// \param E the expression containing the pointer 6080 /// \param NullKind NPCK_NotNull if E is a cast to bool, otherwise, E is 6081 /// compared to a null pointer 6082 /// \param IsEqual True when the comparison is equal to a null pointer 6083 /// \param Range Extra SourceRange to highlight in the diagnostic 6084 void Sema::DiagnoseAlwaysNonNullPointer(Expr *E, 6085 Expr::NullPointerConstantKind NullKind, 6086 bool IsEqual, SourceRange Range) { 6087 6088 // Don't warn inside macros. 6089 if (E->getExprLoc().isMacroID()) 6090 return; 6091 E = E->IgnoreImpCasts(); 6092 6093 const bool IsCompare = NullKind != Expr::NPCK_NotNull; 6094 6095 bool IsAddressOf = false; 6096 6097 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 6098 if (UO->getOpcode() != UO_AddrOf) 6099 return; 6100 IsAddressOf = true; 6101 E = UO->getSubExpr(); 6102 } 6103 6104 // Expect to find a single Decl. Skip anything more complicated. 6105 ValueDecl *D = 0; 6106 if (DeclRefExpr *R = dyn_cast<DeclRefExpr>(E)) { 6107 D = R->getDecl(); 6108 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 6109 D = M->getMemberDecl(); 6110 } 6111 6112 // Weak Decls can be null. 6113 if (!D || D->isWeak()) 6114 return; 6115 6116 QualType T = D->getType(); 6117 const bool IsArray = T->isArrayType(); 6118 const bool IsFunction = T->isFunctionType(); 6119 6120 if (IsAddressOf) { 6121 // Address of function is used to silence the function warning. 6122 if (IsFunction) 6123 return; 6124 // Address of reference can be null. 6125 if (T->isReferenceType()) 6126 return; 6127 } 6128 6129 // Found nothing. 6130 if (!IsAddressOf && !IsFunction && !IsArray) 6131 return; 6132 6133 // Pretty print the expression for the diagnostic. 6134 std::string Str; 6135 llvm::raw_string_ostream S(Str); 6136 E->printPretty(S, 0, getPrintingPolicy()); 6137 6138 unsigned DiagID = IsCompare ? diag::warn_null_pointer_compare 6139 : diag::warn_impcast_pointer_to_bool; 6140 unsigned DiagType; 6141 if (IsAddressOf) 6142 DiagType = AddressOf; 6143 else if (IsFunction) 6144 DiagType = FunctionPointer; 6145 else if (IsArray) 6146 DiagType = ArrayPointer; 6147 else 6148 llvm_unreachable("Could not determine diagnostic."); 6149 Diag(E->getExprLoc(), DiagID) << DiagType << S.str() << E->getSourceRange() 6150 << Range << IsEqual; 6151 6152 if (!IsFunction) 6153 return; 6154 6155 // Suggest '&' to silence the function warning. 6156 Diag(E->getExprLoc(), diag::note_function_warning_silence) 6157 << FixItHint::CreateInsertion(E->getLocStart(), "&"); 6158 6159 // Check to see if '()' fixit should be emitted. 6160 QualType ReturnType; 6161 UnresolvedSet<4> NonTemplateOverloads; 6162 tryExprAsCall(*E, ReturnType, NonTemplateOverloads); 6163 if (ReturnType.isNull()) 6164 return; 6165 6166 if (IsCompare) { 6167 // There are two cases here. If there is null constant, the only suggest 6168 // for a pointer return type. If the null is 0, then suggest if the return 6169 // type is a pointer or an integer type. 6170 if (!ReturnType->isPointerType()) { 6171 if (NullKind == Expr::NPCK_ZeroExpression || 6172 NullKind == Expr::NPCK_ZeroLiteral) { 6173 if (!ReturnType->isIntegerType()) 6174 return; 6175 } else { 6176 return; 6177 } 6178 } 6179 } else { // !IsCompare 6180 // For function to bool, only suggest if the function pointer has bool 6181 // return type. 6182 if (!ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 6183 return; 6184 } 6185 Diag(E->getExprLoc(), diag::note_function_to_function_call) 6186 << FixItHint::CreateInsertion( 6187 getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 6188 } 6189 6190 6191 /// Diagnoses "dangerous" implicit conversions within the given 6192 /// expression (which is a full expression). Implements -Wconversion 6193 /// and -Wsign-compare. 6194 /// 6195 /// \param CC the "context" location of the implicit conversion, i.e. 6196 /// the most location of the syntactic entity requiring the implicit 6197 /// conversion 6198 void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 6199 // Don't diagnose in unevaluated contexts. 6200 if (isUnevaluatedContext()) 6201 return; 6202 6203 // Don't diagnose for value- or type-dependent expressions. 6204 if (E->isTypeDependent() || E->isValueDependent()) 6205 return; 6206 6207 // Check for array bounds violations in cases where the check isn't triggered 6208 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 6209 // ArraySubscriptExpr is on the RHS of a variable initialization. 6210 CheckArrayAccess(E); 6211 6212 // This is not the right CC for (e.g.) a variable initialization. 6213 AnalyzeImplicitConversions(*this, E, CC); 6214 } 6215 6216 /// Diagnose when expression is an integer constant expression and its evaluation 6217 /// results in integer overflow 6218 void Sema::CheckForIntOverflow (Expr *E) { 6219 if (isa<BinaryOperator>(E->IgnoreParens())) 6220 E->EvaluateForOverflow(Context); 6221 } 6222 6223 namespace { 6224 /// \brief Visitor for expressions which looks for unsequenced operations on the 6225 /// same object. 6226 class SequenceChecker : public EvaluatedExprVisitor<SequenceChecker> { 6227 typedef EvaluatedExprVisitor<SequenceChecker> Base; 6228 6229 /// \brief A tree of sequenced regions within an expression. Two regions are 6230 /// unsequenced if one is an ancestor or a descendent of the other. When we 6231 /// finish processing an expression with sequencing, such as a comma 6232 /// expression, we fold its tree nodes into its parent, since they are 6233 /// unsequenced with respect to nodes we will visit later. 6234 class SequenceTree { 6235 struct Value { 6236 explicit Value(unsigned Parent) : Parent(Parent), Merged(false) {} 6237 unsigned Parent : 31; 6238 bool Merged : 1; 6239 }; 6240 SmallVector<Value, 8> Values; 6241 6242 public: 6243 /// \brief A region within an expression which may be sequenced with respect 6244 /// to some other region. 6245 class Seq { 6246 explicit Seq(unsigned N) : Index(N) {} 6247 unsigned Index; 6248 friend class SequenceTree; 6249 public: 6250 Seq() : Index(0) {} 6251 }; 6252 6253 SequenceTree() { Values.push_back(Value(0)); } 6254 Seq root() const { return Seq(0); } 6255 6256 /// \brief Create a new sequence of operations, which is an unsequenced 6257 /// subset of \p Parent. This sequence of operations is sequenced with 6258 /// respect to other children of \p Parent. 6259 Seq allocate(Seq Parent) { 6260 Values.push_back(Value(Parent.Index)); 6261 return Seq(Values.size() - 1); 6262 } 6263 6264 /// \brief Merge a sequence of operations into its parent. 6265 void merge(Seq S) { 6266 Values[S.Index].Merged = true; 6267 } 6268 6269 /// \brief Determine whether two operations are unsequenced. This operation 6270 /// is asymmetric: \p Cur should be the more recent sequence, and \p Old 6271 /// should have been merged into its parent as appropriate. 6272 bool isUnsequenced(Seq Cur, Seq Old) { 6273 unsigned C = representative(Cur.Index); 6274 unsigned Target = representative(Old.Index); 6275 while (C >= Target) { 6276 if (C == Target) 6277 return true; 6278 C = Values[C].Parent; 6279 } 6280 return false; 6281 } 6282 6283 private: 6284 /// \brief Pick a representative for a sequence. 6285 unsigned representative(unsigned K) { 6286 if (Values[K].Merged) 6287 // Perform path compression as we go. 6288 return Values[K].Parent = representative(Values[K].Parent); 6289 return K; 6290 } 6291 }; 6292 6293 /// An object for which we can track unsequenced uses. 6294 typedef NamedDecl *Object; 6295 6296 /// Different flavors of object usage which we track. We only track the 6297 /// least-sequenced usage of each kind. 6298 enum UsageKind { 6299 /// A read of an object. Multiple unsequenced reads are OK. 6300 UK_Use, 6301 /// A modification of an object which is sequenced before the value 6302 /// computation of the expression, such as ++n in C++. 6303 UK_ModAsValue, 6304 /// A modification of an object which is not sequenced before the value 6305 /// computation of the expression, such as n++. 6306 UK_ModAsSideEffect, 6307 6308 UK_Count = UK_ModAsSideEffect + 1 6309 }; 6310 6311 struct Usage { 6312 Usage() : Use(0), Seq() {} 6313 Expr *Use; 6314 SequenceTree::Seq Seq; 6315 }; 6316 6317 struct UsageInfo { 6318 UsageInfo() : Diagnosed(false) {} 6319 Usage Uses[UK_Count]; 6320 /// Have we issued a diagnostic for this variable already? 6321 bool Diagnosed; 6322 }; 6323 typedef llvm::SmallDenseMap<Object, UsageInfo, 16> UsageInfoMap; 6324 6325 Sema &SemaRef; 6326 /// Sequenced regions within the expression. 6327 SequenceTree Tree; 6328 /// Declaration modifications and references which we have seen. 6329 UsageInfoMap UsageMap; 6330 /// The region we are currently within. 6331 SequenceTree::Seq Region; 6332 /// Filled in with declarations which were modified as a side-effect 6333 /// (that is, post-increment operations). 6334 SmallVectorImpl<std::pair<Object, Usage> > *ModAsSideEffect; 6335 /// Expressions to check later. We defer checking these to reduce 6336 /// stack usage. 6337 SmallVectorImpl<Expr *> &WorkList; 6338 6339 /// RAII object wrapping the visitation of a sequenced subexpression of an 6340 /// expression. At the end of this process, the side-effects of the evaluation 6341 /// become sequenced with respect to the value computation of the result, so 6342 /// we downgrade any UK_ModAsSideEffect within the evaluation to 6343 /// UK_ModAsValue. 6344 struct SequencedSubexpression { 6345 SequencedSubexpression(SequenceChecker &Self) 6346 : Self(Self), OldModAsSideEffect(Self.ModAsSideEffect) { 6347 Self.ModAsSideEffect = &ModAsSideEffect; 6348 } 6349 ~SequencedSubexpression() { 6350 for (unsigned I = 0, E = ModAsSideEffect.size(); I != E; ++I) { 6351 UsageInfo &U = Self.UsageMap[ModAsSideEffect[I].first]; 6352 U.Uses[UK_ModAsSideEffect] = ModAsSideEffect[I].second; 6353 Self.addUsage(U, ModAsSideEffect[I].first, 6354 ModAsSideEffect[I].second.Use, UK_ModAsValue); 6355 } 6356 Self.ModAsSideEffect = OldModAsSideEffect; 6357 } 6358 6359 SequenceChecker &Self; 6360 SmallVector<std::pair<Object, Usage>, 4> ModAsSideEffect; 6361 SmallVectorImpl<std::pair<Object, Usage> > *OldModAsSideEffect; 6362 }; 6363 6364 /// RAII object wrapping the visitation of a subexpression which we might 6365 /// choose to evaluate as a constant. If any subexpression is evaluated and 6366 /// found to be non-constant, this allows us to suppress the evaluation of 6367 /// the outer expression. 6368 class EvaluationTracker { 6369 public: 6370 EvaluationTracker(SequenceChecker &Self) 6371 : Self(Self), Prev(Self.EvalTracker), EvalOK(true) { 6372 Self.EvalTracker = this; 6373 } 6374 ~EvaluationTracker() { 6375 Self.EvalTracker = Prev; 6376 if (Prev) 6377 Prev->EvalOK &= EvalOK; 6378 } 6379 6380 bool evaluate(const Expr *E, bool &Result) { 6381 if (!EvalOK || E->isValueDependent()) 6382 return false; 6383 EvalOK = E->EvaluateAsBooleanCondition(Result, Self.SemaRef.Context); 6384 return EvalOK; 6385 } 6386 6387 private: 6388 SequenceChecker &Self; 6389 EvaluationTracker *Prev; 6390 bool EvalOK; 6391 } *EvalTracker; 6392 6393 /// \brief Find the object which is produced by the specified expression, 6394 /// if any. 6395 Object getObject(Expr *E, bool Mod) const { 6396 E = E->IgnoreParenCasts(); 6397 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 6398 if (Mod && (UO->getOpcode() == UO_PreInc || UO->getOpcode() == UO_PreDec)) 6399 return getObject(UO->getSubExpr(), Mod); 6400 } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 6401 if (BO->getOpcode() == BO_Comma) 6402 return getObject(BO->getRHS(), Mod); 6403 if (Mod && BO->isAssignmentOp()) 6404 return getObject(BO->getLHS(), Mod); 6405 } else if (MemberExpr *ME = dyn_cast<MemberExpr>(E)) { 6406 // FIXME: Check for more interesting cases, like "x.n = ++x.n". 6407 if (isa<CXXThisExpr>(ME->getBase()->IgnoreParenCasts())) 6408 return ME->getMemberDecl(); 6409 } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 6410 // FIXME: If this is a reference, map through to its value. 6411 return DRE->getDecl(); 6412 return 0; 6413 } 6414 6415 /// \brief Note that an object was modified or used by an expression. 6416 void addUsage(UsageInfo &UI, Object O, Expr *Ref, UsageKind UK) { 6417 Usage &U = UI.Uses[UK]; 6418 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) { 6419 if (UK == UK_ModAsSideEffect && ModAsSideEffect) 6420 ModAsSideEffect->push_back(std::make_pair(O, U)); 6421 U.Use = Ref; 6422 U.Seq = Region; 6423 } 6424 } 6425 /// \brief Check whether a modification or use conflicts with a prior usage. 6426 void checkUsage(Object O, UsageInfo &UI, Expr *Ref, UsageKind OtherKind, 6427 bool IsModMod) { 6428 if (UI.Diagnosed) 6429 return; 6430 6431 const Usage &U = UI.Uses[OtherKind]; 6432 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) 6433 return; 6434 6435 Expr *Mod = U.Use; 6436 Expr *ModOrUse = Ref; 6437 if (OtherKind == UK_Use) 6438 std::swap(Mod, ModOrUse); 6439 6440 SemaRef.Diag(Mod->getExprLoc(), 6441 IsModMod ? diag::warn_unsequenced_mod_mod 6442 : diag::warn_unsequenced_mod_use) 6443 << O << SourceRange(ModOrUse->getExprLoc()); 6444 UI.Diagnosed = true; 6445 } 6446 6447 void notePreUse(Object O, Expr *Use) { 6448 UsageInfo &U = UsageMap[O]; 6449 // Uses conflict with other modifications. 6450 checkUsage(O, U, Use, UK_ModAsValue, false); 6451 } 6452 void notePostUse(Object O, Expr *Use) { 6453 UsageInfo &U = UsageMap[O]; 6454 checkUsage(O, U, Use, UK_ModAsSideEffect, false); 6455 addUsage(U, O, Use, UK_Use); 6456 } 6457 6458 void notePreMod(Object O, Expr *Mod) { 6459 UsageInfo &U = UsageMap[O]; 6460 // Modifications conflict with other modifications and with uses. 6461 checkUsage(O, U, Mod, UK_ModAsValue, true); 6462 checkUsage(O, U, Mod, UK_Use, false); 6463 } 6464 void notePostMod(Object O, Expr *Use, UsageKind UK) { 6465 UsageInfo &U = UsageMap[O]; 6466 checkUsage(O, U, Use, UK_ModAsSideEffect, true); 6467 addUsage(U, O, Use, UK); 6468 } 6469 6470 public: 6471 SequenceChecker(Sema &S, Expr *E, SmallVectorImpl<Expr *> &WorkList) 6472 : Base(S.Context), SemaRef(S), Region(Tree.root()), ModAsSideEffect(0), 6473 WorkList(WorkList), EvalTracker(0) { 6474 Visit(E); 6475 } 6476 6477 void VisitStmt(Stmt *S) { 6478 // Skip all statements which aren't expressions for now. 6479 } 6480 6481 void VisitExpr(Expr *E) { 6482 // By default, just recurse to evaluated subexpressions. 6483 Base::VisitStmt(E); 6484 } 6485 6486 void VisitCastExpr(CastExpr *E) { 6487 Object O = Object(); 6488 if (E->getCastKind() == CK_LValueToRValue) 6489 O = getObject(E->getSubExpr(), false); 6490 6491 if (O) 6492 notePreUse(O, E); 6493 VisitExpr(E); 6494 if (O) 6495 notePostUse(O, E); 6496 } 6497 6498 void VisitBinComma(BinaryOperator *BO) { 6499 // C++11 [expr.comma]p1: 6500 // Every value computation and side effect associated with the left 6501 // expression is sequenced before every value computation and side 6502 // effect associated with the right expression. 6503 SequenceTree::Seq LHS = Tree.allocate(Region); 6504 SequenceTree::Seq RHS = Tree.allocate(Region); 6505 SequenceTree::Seq OldRegion = Region; 6506 6507 { 6508 SequencedSubexpression SeqLHS(*this); 6509 Region = LHS; 6510 Visit(BO->getLHS()); 6511 } 6512 6513 Region = RHS; 6514 Visit(BO->getRHS()); 6515 6516 Region = OldRegion; 6517 6518 // Forget that LHS and RHS are sequenced. They are both unsequenced 6519 // with respect to other stuff. 6520 Tree.merge(LHS); 6521 Tree.merge(RHS); 6522 } 6523 6524 void VisitBinAssign(BinaryOperator *BO) { 6525 // The modification is sequenced after the value computation of the LHS 6526 // and RHS, so check it before inspecting the operands and update the 6527 // map afterwards. 6528 Object O = getObject(BO->getLHS(), true); 6529 if (!O) 6530 return VisitExpr(BO); 6531 6532 notePreMod(O, BO); 6533 6534 // C++11 [expr.ass]p7: 6535 // E1 op= E2 is equivalent to E1 = E1 op E2, except that E1 is evaluated 6536 // only once. 6537 // 6538 // Therefore, for a compound assignment operator, O is considered used 6539 // everywhere except within the evaluation of E1 itself. 6540 if (isa<CompoundAssignOperator>(BO)) 6541 notePreUse(O, BO); 6542 6543 Visit(BO->getLHS()); 6544 6545 if (isa<CompoundAssignOperator>(BO)) 6546 notePostUse(O, BO); 6547 6548 Visit(BO->getRHS()); 6549 6550 // C++11 [expr.ass]p1: 6551 // the assignment is sequenced [...] before the value computation of the 6552 // assignment expression. 6553 // C11 6.5.16/3 has no such rule. 6554 notePostMod(O, BO, SemaRef.getLangOpts().CPlusPlus ? UK_ModAsValue 6555 : UK_ModAsSideEffect); 6556 } 6557 void VisitCompoundAssignOperator(CompoundAssignOperator *CAO) { 6558 VisitBinAssign(CAO); 6559 } 6560 6561 void VisitUnaryPreInc(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 6562 void VisitUnaryPreDec(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 6563 void VisitUnaryPreIncDec(UnaryOperator *UO) { 6564 Object O = getObject(UO->getSubExpr(), true); 6565 if (!O) 6566 return VisitExpr(UO); 6567 6568 notePreMod(O, UO); 6569 Visit(UO->getSubExpr()); 6570 // C++11 [expr.pre.incr]p1: 6571 // the expression ++x is equivalent to x+=1 6572 notePostMod(O, UO, SemaRef.getLangOpts().CPlusPlus ? UK_ModAsValue 6573 : UK_ModAsSideEffect); 6574 } 6575 6576 void VisitUnaryPostInc(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 6577 void VisitUnaryPostDec(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 6578 void VisitUnaryPostIncDec(UnaryOperator *UO) { 6579 Object O = getObject(UO->getSubExpr(), true); 6580 if (!O) 6581 return VisitExpr(UO); 6582 6583 notePreMod(O, UO); 6584 Visit(UO->getSubExpr()); 6585 notePostMod(O, UO, UK_ModAsSideEffect); 6586 } 6587 6588 /// Don't visit the RHS of '&&' or '||' if it might not be evaluated. 6589 void VisitBinLOr(BinaryOperator *BO) { 6590 // The side-effects of the LHS of an '&&' are sequenced before the 6591 // value computation of the RHS, and hence before the value computation 6592 // of the '&&' itself, unless the LHS evaluates to zero. We treat them 6593 // as if they were unconditionally sequenced. 6594 EvaluationTracker Eval(*this); 6595 { 6596 SequencedSubexpression Sequenced(*this); 6597 Visit(BO->getLHS()); 6598 } 6599 6600 bool Result; 6601 if (Eval.evaluate(BO->getLHS(), Result)) { 6602 if (!Result) 6603 Visit(BO->getRHS()); 6604 } else { 6605 // Check for unsequenced operations in the RHS, treating it as an 6606 // entirely separate evaluation. 6607 // 6608 // FIXME: If there are operations in the RHS which are unsequenced 6609 // with respect to operations outside the RHS, and those operations 6610 // are unconditionally evaluated, diagnose them. 6611 WorkList.push_back(BO->getRHS()); 6612 } 6613 } 6614 void VisitBinLAnd(BinaryOperator *BO) { 6615 EvaluationTracker Eval(*this); 6616 { 6617 SequencedSubexpression Sequenced(*this); 6618 Visit(BO->getLHS()); 6619 } 6620 6621 bool Result; 6622 if (Eval.evaluate(BO->getLHS(), Result)) { 6623 if (Result) 6624 Visit(BO->getRHS()); 6625 } else { 6626 WorkList.push_back(BO->getRHS()); 6627 } 6628 } 6629 6630 // Only visit the condition, unless we can be sure which subexpression will 6631 // be chosen. 6632 void VisitAbstractConditionalOperator(AbstractConditionalOperator *CO) { 6633 EvaluationTracker Eval(*this); 6634 { 6635 SequencedSubexpression Sequenced(*this); 6636 Visit(CO->getCond()); 6637 } 6638 6639 bool Result; 6640 if (Eval.evaluate(CO->getCond(), Result)) 6641 Visit(Result ? CO->getTrueExpr() : CO->getFalseExpr()); 6642 else { 6643 WorkList.push_back(CO->getTrueExpr()); 6644 WorkList.push_back(CO->getFalseExpr()); 6645 } 6646 } 6647 6648 void VisitCallExpr(CallExpr *CE) { 6649 // C++11 [intro.execution]p15: 6650 // When calling a function [...], every value computation and side effect 6651 // associated with any argument expression, or with the postfix expression 6652 // designating the called function, is sequenced before execution of every 6653 // expression or statement in the body of the function [and thus before 6654 // the value computation of its result]. 6655 SequencedSubexpression Sequenced(*this); 6656 Base::VisitCallExpr(CE); 6657 6658 // FIXME: CXXNewExpr and CXXDeleteExpr implicitly call functions. 6659 } 6660 6661 void VisitCXXConstructExpr(CXXConstructExpr *CCE) { 6662 // This is a call, so all subexpressions are sequenced before the result. 6663 SequencedSubexpression Sequenced(*this); 6664 6665 if (!CCE->isListInitialization()) 6666 return VisitExpr(CCE); 6667 6668 // In C++11, list initializations are sequenced. 6669 SmallVector<SequenceTree::Seq, 32> Elts; 6670 SequenceTree::Seq Parent = Region; 6671 for (CXXConstructExpr::arg_iterator I = CCE->arg_begin(), 6672 E = CCE->arg_end(); 6673 I != E; ++I) { 6674 Region = Tree.allocate(Parent); 6675 Elts.push_back(Region); 6676 Visit(*I); 6677 } 6678 6679 // Forget that the initializers are sequenced. 6680 Region = Parent; 6681 for (unsigned I = 0; I < Elts.size(); ++I) 6682 Tree.merge(Elts[I]); 6683 } 6684 6685 void VisitInitListExpr(InitListExpr *ILE) { 6686 if (!SemaRef.getLangOpts().CPlusPlus11) 6687 return VisitExpr(ILE); 6688 6689 // In C++11, list initializations are sequenced. 6690 SmallVector<SequenceTree::Seq, 32> Elts; 6691 SequenceTree::Seq Parent = Region; 6692 for (unsigned I = 0; I < ILE->getNumInits(); ++I) { 6693 Expr *E = ILE->getInit(I); 6694 if (!E) continue; 6695 Region = Tree.allocate(Parent); 6696 Elts.push_back(Region); 6697 Visit(E); 6698 } 6699 6700 // Forget that the initializers are sequenced. 6701 Region = Parent; 6702 for (unsigned I = 0; I < Elts.size(); ++I) 6703 Tree.merge(Elts[I]); 6704 } 6705 }; 6706 } 6707 6708 void Sema::CheckUnsequencedOperations(Expr *E) { 6709 SmallVector<Expr *, 8> WorkList; 6710 WorkList.push_back(E); 6711 while (!WorkList.empty()) { 6712 Expr *Item = WorkList.pop_back_val(); 6713 SequenceChecker(*this, Item, WorkList); 6714 } 6715 } 6716 6717 void Sema::CheckCompletedExpr(Expr *E, SourceLocation CheckLoc, 6718 bool IsConstexpr) { 6719 CheckImplicitConversions(E, CheckLoc); 6720 CheckUnsequencedOperations(E); 6721 if (!IsConstexpr && !E->isValueDependent()) 6722 CheckForIntOverflow(E); 6723 } 6724 6725 void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 6726 FieldDecl *BitField, 6727 Expr *Init) { 6728 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 6729 } 6730 6731 /// CheckParmsForFunctionDef - Check that the parameters of the given 6732 /// function are appropriate for the definition of a function. This 6733 /// takes care of any checks that cannot be performed on the 6734 /// declaration itself, e.g., that the types of each of the function 6735 /// parameters are complete. 6736 bool Sema::CheckParmsForFunctionDef(ParmVarDecl *const *P, 6737 ParmVarDecl *const *PEnd, 6738 bool CheckParameterNames) { 6739 bool HasInvalidParm = false; 6740 for (; P != PEnd; ++P) { 6741 ParmVarDecl *Param = *P; 6742 6743 // C99 6.7.5.3p4: the parameters in a parameter type list in a 6744 // function declarator that is part of a function definition of 6745 // that function shall not have incomplete type. 6746 // 6747 // This is also C++ [dcl.fct]p6. 6748 if (!Param->isInvalidDecl() && 6749 RequireCompleteType(Param->getLocation(), Param->getType(), 6750 diag::err_typecheck_decl_incomplete_type)) { 6751 Param->setInvalidDecl(); 6752 HasInvalidParm = true; 6753 } 6754 6755 // C99 6.9.1p5: If the declarator includes a parameter type list, the 6756 // declaration of each parameter shall include an identifier. 6757 if (CheckParameterNames && 6758 Param->getIdentifier() == 0 && 6759 !Param->isImplicit() && 6760 !getLangOpts().CPlusPlus) 6761 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 6762 6763 // C99 6.7.5.3p12: 6764 // If the function declarator is not part of a definition of that 6765 // function, parameters may have incomplete type and may use the [*] 6766 // notation in their sequences of declarator specifiers to specify 6767 // variable length array types. 6768 QualType PType = Param->getOriginalType(); 6769 while (const ArrayType *AT = Context.getAsArrayType(PType)) { 6770 if (AT->getSizeModifier() == ArrayType::Star) { 6771 // FIXME: This diagnostic should point the '[*]' if source-location 6772 // information is added for it. 6773 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 6774 break; 6775 } 6776 PType= AT->getElementType(); 6777 } 6778 6779 // MSVC destroys objects passed by value in the callee. Therefore a 6780 // function definition which takes such a parameter must be able to call the 6781 // object's destructor. However, we don't perform any direct access check 6782 // on the dtor. 6783 if (getLangOpts().CPlusPlus && Context.getTargetInfo() 6784 .getCXXABI() 6785 .areArgsDestroyedLeftToRightInCallee()) { 6786 if (!Param->isInvalidDecl()) { 6787 if (const RecordType *RT = Param->getType()->getAs<RecordType>()) { 6788 CXXRecordDecl *ClassDecl = cast<CXXRecordDecl>(RT->getDecl()); 6789 if (!ClassDecl->isInvalidDecl() && 6790 !ClassDecl->hasIrrelevantDestructor() && 6791 !ClassDecl->isDependentContext()) { 6792 CXXDestructorDecl *Destructor = LookupDestructor(ClassDecl); 6793 MarkFunctionReferenced(Param->getLocation(), Destructor); 6794 DiagnoseUseOfDecl(Destructor, Param->getLocation()); 6795 } 6796 } 6797 } 6798 } 6799 } 6800 6801 return HasInvalidParm; 6802 } 6803 6804 /// CheckCastAlign - Implements -Wcast-align, which warns when a 6805 /// pointer cast increases the alignment requirements. 6806 void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 6807 // This is actually a lot of work to potentially be doing on every 6808 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 6809 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 6810 TRange.getBegin()) 6811 == DiagnosticsEngine::Ignored) 6812 return; 6813 6814 // Ignore dependent types. 6815 if (T->isDependentType() || Op->getType()->isDependentType()) 6816 return; 6817 6818 // Require that the destination be a pointer type. 6819 const PointerType *DestPtr = T->getAs<PointerType>(); 6820 if (!DestPtr) return; 6821 6822 // If the destination has alignment 1, we're done. 6823 QualType DestPointee = DestPtr->getPointeeType(); 6824 if (DestPointee->isIncompleteType()) return; 6825 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 6826 if (DestAlign.isOne()) return; 6827 6828 // Require that the source be a pointer type. 6829 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 6830 if (!SrcPtr) return; 6831 QualType SrcPointee = SrcPtr->getPointeeType(); 6832 6833 // Whitelist casts from cv void*. We already implicitly 6834 // whitelisted casts to cv void*, since they have alignment 1. 6835 // Also whitelist casts involving incomplete types, which implicitly 6836 // includes 'void'. 6837 if (SrcPointee->isIncompleteType()) return; 6838 6839 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 6840 if (SrcAlign >= DestAlign) return; 6841 6842 Diag(TRange.getBegin(), diag::warn_cast_align) 6843 << Op->getType() << T 6844 << static_cast<unsigned>(SrcAlign.getQuantity()) 6845 << static_cast<unsigned>(DestAlign.getQuantity()) 6846 << TRange << Op->getSourceRange(); 6847 } 6848 6849 static const Type* getElementType(const Expr *BaseExpr) { 6850 const Type* EltType = BaseExpr->getType().getTypePtr(); 6851 if (EltType->isAnyPointerType()) 6852 return EltType->getPointeeType().getTypePtr(); 6853 else if (EltType->isArrayType()) 6854 return EltType->getBaseElementTypeUnsafe(); 6855 return EltType; 6856 } 6857 6858 /// \brief Check whether this array fits the idiom of a size-one tail padded 6859 /// array member of a struct. 6860 /// 6861 /// We avoid emitting out-of-bounds access warnings for such arrays as they are 6862 /// commonly used to emulate flexible arrays in C89 code. 6863 static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 6864 const NamedDecl *ND) { 6865 if (Size != 1 || !ND) return false; 6866 6867 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 6868 if (!FD) return false; 6869 6870 // Don't consider sizes resulting from macro expansions or template argument 6871 // substitution to form C89 tail-padded arrays. 6872 6873 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 6874 while (TInfo) { 6875 TypeLoc TL = TInfo->getTypeLoc(); 6876 // Look through typedefs. 6877 if (TypedefTypeLoc TTL = TL.getAs<TypedefTypeLoc>()) { 6878 const TypedefNameDecl *TDL = TTL.getTypedefNameDecl(); 6879 TInfo = TDL->getTypeSourceInfo(); 6880 continue; 6881 } 6882 if (ConstantArrayTypeLoc CTL = TL.getAs<ConstantArrayTypeLoc>()) { 6883 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 6884 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 6885 return false; 6886 } 6887 break; 6888 } 6889 6890 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 6891 if (!RD) return false; 6892 if (RD->isUnion()) return false; 6893 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 6894 if (!CRD->isStandardLayout()) return false; 6895 } 6896 6897 // See if this is the last field decl in the record. 6898 const Decl *D = FD; 6899 while ((D = D->getNextDeclInContext())) 6900 if (isa<FieldDecl>(D)) 6901 return false; 6902 return true; 6903 } 6904 6905 void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 6906 const ArraySubscriptExpr *ASE, 6907 bool AllowOnePastEnd, bool IndexNegated) { 6908 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 6909 if (IndexExpr->isValueDependent()) 6910 return; 6911 6912 const Type *EffectiveType = getElementType(BaseExpr); 6913 BaseExpr = BaseExpr->IgnoreParenCasts(); 6914 const ConstantArrayType *ArrayTy = 6915 Context.getAsConstantArrayType(BaseExpr->getType()); 6916 if (!ArrayTy) 6917 return; 6918 6919 llvm::APSInt index; 6920 if (!IndexExpr->EvaluateAsInt(index, Context)) 6921 return; 6922 if (IndexNegated) 6923 index = -index; 6924 6925 const NamedDecl *ND = NULL; 6926 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 6927 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 6928 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 6929 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 6930 6931 if (index.isUnsigned() || !index.isNegative()) { 6932 llvm::APInt size = ArrayTy->getSize(); 6933 if (!size.isStrictlyPositive()) 6934 return; 6935 6936 const Type* BaseType = getElementType(BaseExpr); 6937 if (BaseType != EffectiveType) { 6938 // Make sure we're comparing apples to apples when comparing index to size 6939 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 6940 uint64_t array_typesize = Context.getTypeSize(BaseType); 6941 // Handle ptrarith_typesize being zero, such as when casting to void* 6942 if (!ptrarith_typesize) ptrarith_typesize = 1; 6943 if (ptrarith_typesize != array_typesize) { 6944 // There's a cast to a different size type involved 6945 uint64_t ratio = array_typesize / ptrarith_typesize; 6946 // TODO: Be smarter about handling cases where array_typesize is not a 6947 // multiple of ptrarith_typesize 6948 if (ptrarith_typesize * ratio == array_typesize) 6949 size *= llvm::APInt(size.getBitWidth(), ratio); 6950 } 6951 } 6952 6953 if (size.getBitWidth() > index.getBitWidth()) 6954 index = index.zext(size.getBitWidth()); 6955 else if (size.getBitWidth() < index.getBitWidth()) 6956 size = size.zext(index.getBitWidth()); 6957 6958 // For array subscripting the index must be less than size, but for pointer 6959 // arithmetic also allow the index (offset) to be equal to size since 6960 // computing the next address after the end of the array is legal and 6961 // commonly done e.g. in C++ iterators and range-based for loops. 6962 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 6963 return; 6964 6965 // Also don't warn for arrays of size 1 which are members of some 6966 // structure. These are often used to approximate flexible arrays in C89 6967 // code. 6968 if (IsTailPaddedMemberArray(*this, size, ND)) 6969 return; 6970 6971 // Suppress the warning if the subscript expression (as identified by the 6972 // ']' location) and the index expression are both from macro expansions 6973 // within a system header. 6974 if (ASE) { 6975 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 6976 ASE->getRBracketLoc()); 6977 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 6978 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 6979 IndexExpr->getLocStart()); 6980 if (SourceMgr.isWrittenInSameFile(RBracketLoc, IndexLoc)) 6981 return; 6982 } 6983 } 6984 6985 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 6986 if (ASE) 6987 DiagID = diag::warn_array_index_exceeds_bounds; 6988 6989 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 6990 PDiag(DiagID) << index.toString(10, true) 6991 << size.toString(10, true) 6992 << (unsigned)size.getLimitedValue(~0U) 6993 << IndexExpr->getSourceRange()); 6994 } else { 6995 unsigned DiagID = diag::warn_array_index_precedes_bounds; 6996 if (!ASE) { 6997 DiagID = diag::warn_ptr_arith_precedes_bounds; 6998 if (index.isNegative()) index = -index; 6999 } 7000 7001 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 7002 PDiag(DiagID) << index.toString(10, true) 7003 << IndexExpr->getSourceRange()); 7004 } 7005 7006 if (!ND) { 7007 // Try harder to find a NamedDecl to point at in the note. 7008 while (const ArraySubscriptExpr *ASE = 7009 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 7010 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 7011 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 7012 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 7013 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 7014 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 7015 } 7016 7017 if (ND) 7018 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 7019 PDiag(diag::note_array_index_out_of_bounds) 7020 << ND->getDeclName()); 7021 } 7022 7023 void Sema::CheckArrayAccess(const Expr *expr) { 7024 int AllowOnePastEnd = 0; 7025 while (expr) { 7026 expr = expr->IgnoreParenImpCasts(); 7027 switch (expr->getStmtClass()) { 7028 case Stmt::ArraySubscriptExprClass: { 7029 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 7030 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 7031 AllowOnePastEnd > 0); 7032 return; 7033 } 7034 case Stmt::UnaryOperatorClass: { 7035 // Only unwrap the * and & unary operators 7036 const UnaryOperator *UO = cast<UnaryOperator>(expr); 7037 expr = UO->getSubExpr(); 7038 switch (UO->getOpcode()) { 7039 case UO_AddrOf: 7040 AllowOnePastEnd++; 7041 break; 7042 case UO_Deref: 7043 AllowOnePastEnd--; 7044 break; 7045 default: 7046 return; 7047 } 7048 break; 7049 } 7050 case Stmt::ConditionalOperatorClass: { 7051 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 7052 if (const Expr *lhs = cond->getLHS()) 7053 CheckArrayAccess(lhs); 7054 if (const Expr *rhs = cond->getRHS()) 7055 CheckArrayAccess(rhs); 7056 return; 7057 } 7058 default: 7059 return; 7060 } 7061 } 7062 } 7063 7064 //===--- CHECK: Objective-C retain cycles ----------------------------------// 7065 7066 namespace { 7067 struct RetainCycleOwner { 7068 RetainCycleOwner() : Variable(0), Indirect(false) {} 7069 VarDecl *Variable; 7070 SourceRange Range; 7071 SourceLocation Loc; 7072 bool Indirect; 7073 7074 void setLocsFrom(Expr *e) { 7075 Loc = e->getExprLoc(); 7076 Range = e->getSourceRange(); 7077 } 7078 }; 7079 } 7080 7081 /// Consider whether capturing the given variable can possibly lead to 7082 /// a retain cycle. 7083 static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 7084 // In ARC, it's captured strongly iff the variable has __strong 7085 // lifetime. In MRR, it's captured strongly if the variable is 7086 // __block and has an appropriate type. 7087 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 7088 return false; 7089 7090 owner.Variable = var; 7091 if (ref) 7092 owner.setLocsFrom(ref); 7093 return true; 7094 } 7095 7096 static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 7097 while (true) { 7098 e = e->IgnoreParens(); 7099 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 7100 switch (cast->getCastKind()) { 7101 case CK_BitCast: 7102 case CK_LValueBitCast: 7103 case CK_LValueToRValue: 7104 case CK_ARCReclaimReturnedObject: 7105 e = cast->getSubExpr(); 7106 continue; 7107 7108 default: 7109 return false; 7110 } 7111 } 7112 7113 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 7114 ObjCIvarDecl *ivar = ref->getDecl(); 7115 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 7116 return false; 7117 7118 // Try to find a retain cycle in the base. 7119 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 7120 return false; 7121 7122 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 7123 owner.Indirect = true; 7124 return true; 7125 } 7126 7127 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 7128 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 7129 if (!var) return false; 7130 return considerVariable(var, ref, owner); 7131 } 7132 7133 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 7134 if (member->isArrow()) return false; 7135 7136 // Don't count this as an indirect ownership. 7137 e = member->getBase(); 7138 continue; 7139 } 7140 7141 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 7142 // Only pay attention to pseudo-objects on property references. 7143 ObjCPropertyRefExpr *pre 7144 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 7145 ->IgnoreParens()); 7146 if (!pre) return false; 7147 if (pre->isImplicitProperty()) return false; 7148 ObjCPropertyDecl *property = pre->getExplicitProperty(); 7149 if (!property->isRetaining() && 7150 !(property->getPropertyIvarDecl() && 7151 property->getPropertyIvarDecl()->getType() 7152 .getObjCLifetime() == Qualifiers::OCL_Strong)) 7153 return false; 7154 7155 owner.Indirect = true; 7156 if (pre->isSuperReceiver()) { 7157 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 7158 if (!owner.Variable) 7159 return false; 7160 owner.Loc = pre->getLocation(); 7161 owner.Range = pre->getSourceRange(); 7162 return true; 7163 } 7164 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 7165 ->getSourceExpr()); 7166 continue; 7167 } 7168 7169 // Array ivars? 7170 7171 return false; 7172 } 7173 } 7174 7175 namespace { 7176 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 7177 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 7178 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 7179 Variable(variable), Capturer(0) {} 7180 7181 VarDecl *Variable; 7182 Expr *Capturer; 7183 7184 void VisitDeclRefExpr(DeclRefExpr *ref) { 7185 if (ref->getDecl() == Variable && !Capturer) 7186 Capturer = ref; 7187 } 7188 7189 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 7190 if (Capturer) return; 7191 Visit(ref->getBase()); 7192 if (Capturer && ref->isFreeIvar()) 7193 Capturer = ref; 7194 } 7195 7196 void VisitBlockExpr(BlockExpr *block) { 7197 // Look inside nested blocks 7198 if (block->getBlockDecl()->capturesVariable(Variable)) 7199 Visit(block->getBlockDecl()->getBody()); 7200 } 7201 7202 void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) { 7203 if (Capturer) return; 7204 if (OVE->getSourceExpr()) 7205 Visit(OVE->getSourceExpr()); 7206 } 7207 }; 7208 } 7209 7210 /// Check whether the given argument is a block which captures a 7211 /// variable. 7212 static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 7213 assert(owner.Variable && owner.Loc.isValid()); 7214 7215 e = e->IgnoreParenCasts(); 7216 7217 // Look through [^{...} copy] and Block_copy(^{...}). 7218 if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) { 7219 Selector Cmd = ME->getSelector(); 7220 if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") { 7221 e = ME->getInstanceReceiver(); 7222 if (!e) 7223 return 0; 7224 e = e->IgnoreParenCasts(); 7225 } 7226 } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) { 7227 if (CE->getNumArgs() == 1) { 7228 FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl()); 7229 if (Fn) { 7230 const IdentifierInfo *FnI = Fn->getIdentifier(); 7231 if (FnI && FnI->isStr("_Block_copy")) { 7232 e = CE->getArg(0)->IgnoreParenCasts(); 7233 } 7234 } 7235 } 7236 } 7237 7238 BlockExpr *block = dyn_cast<BlockExpr>(e); 7239 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 7240 return 0; 7241 7242 FindCaptureVisitor visitor(S.Context, owner.Variable); 7243 visitor.Visit(block->getBlockDecl()->getBody()); 7244 return visitor.Capturer; 7245 } 7246 7247 static void diagnoseRetainCycle(Sema &S, Expr *capturer, 7248 RetainCycleOwner &owner) { 7249 assert(capturer); 7250 assert(owner.Variable && owner.Loc.isValid()); 7251 7252 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 7253 << owner.Variable << capturer->getSourceRange(); 7254 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 7255 << owner.Indirect << owner.Range; 7256 } 7257 7258 /// Check for a keyword selector that starts with the word 'add' or 7259 /// 'set'. 7260 static bool isSetterLikeSelector(Selector sel) { 7261 if (sel.isUnarySelector()) return false; 7262 7263 StringRef str = sel.getNameForSlot(0); 7264 while (!str.empty() && str.front() == '_') str = str.substr(1); 7265 if (str.startswith("set")) 7266 str = str.substr(3); 7267 else if (str.startswith("add")) { 7268 // Specially whitelist 'addOperationWithBlock:'. 7269 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 7270 return false; 7271 str = str.substr(3); 7272 } 7273 else 7274 return false; 7275 7276 if (str.empty()) return true; 7277 return !isLowercase(str.front()); 7278 } 7279 7280 /// Check a message send to see if it's likely to cause a retain cycle. 7281 void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 7282 // Only check instance methods whose selector looks like a setter. 7283 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 7284 return; 7285 7286 // Try to find a variable that the receiver is strongly owned by. 7287 RetainCycleOwner owner; 7288 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 7289 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 7290 return; 7291 } else { 7292 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 7293 owner.Variable = getCurMethodDecl()->getSelfDecl(); 7294 owner.Loc = msg->getSuperLoc(); 7295 owner.Range = msg->getSuperLoc(); 7296 } 7297 7298 // Check whether the receiver is captured by any of the arguments. 7299 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 7300 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 7301 return diagnoseRetainCycle(*this, capturer, owner); 7302 } 7303 7304 /// Check a property assign to see if it's likely to cause a retain cycle. 7305 void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 7306 RetainCycleOwner owner; 7307 if (!findRetainCycleOwner(*this, receiver, owner)) 7308 return; 7309 7310 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 7311 diagnoseRetainCycle(*this, capturer, owner); 7312 } 7313 7314 void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) { 7315 RetainCycleOwner Owner; 7316 if (!considerVariable(Var, /*DeclRefExpr=*/0, Owner)) 7317 return; 7318 7319 // Because we don't have an expression for the variable, we have to set the 7320 // location explicitly here. 7321 Owner.Loc = Var->getLocation(); 7322 Owner.Range = Var->getSourceRange(); 7323 7324 if (Expr *Capturer = findCapturingExpr(*this, Init, Owner)) 7325 diagnoseRetainCycle(*this, Capturer, Owner); 7326 } 7327 7328 static bool checkUnsafeAssignLiteral(Sema &S, SourceLocation Loc, 7329 Expr *RHS, bool isProperty) { 7330 // Check if RHS is an Objective-C object literal, which also can get 7331 // immediately zapped in a weak reference. Note that we explicitly 7332 // allow ObjCStringLiterals, since those are designed to never really die. 7333 RHS = RHS->IgnoreParenImpCasts(); 7334 7335 // This enum needs to match with the 'select' in 7336 // warn_objc_arc_literal_assign (off-by-1). 7337 Sema::ObjCLiteralKind Kind = S.CheckLiteralKind(RHS); 7338 if (Kind == Sema::LK_String || Kind == Sema::LK_None) 7339 return false; 7340 7341 S.Diag(Loc, diag::warn_arc_literal_assign) 7342 << (unsigned) Kind 7343 << (isProperty ? 0 : 1) 7344 << RHS->getSourceRange(); 7345 7346 return true; 7347 } 7348 7349 static bool checkUnsafeAssignObject(Sema &S, SourceLocation Loc, 7350 Qualifiers::ObjCLifetime LT, 7351 Expr *RHS, bool isProperty) { 7352 // Strip off any implicit cast added to get to the one ARC-specific. 7353 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 7354 if (cast->getCastKind() == CK_ARCConsumeObject) { 7355 S.Diag(Loc, diag::warn_arc_retained_assign) 7356 << (LT == Qualifiers::OCL_ExplicitNone) 7357 << (isProperty ? 0 : 1) 7358 << RHS->getSourceRange(); 7359 return true; 7360 } 7361 RHS = cast->getSubExpr(); 7362 } 7363 7364 if (LT == Qualifiers::OCL_Weak && 7365 checkUnsafeAssignLiteral(S, Loc, RHS, isProperty)) 7366 return true; 7367 7368 return false; 7369 } 7370 7371 bool Sema::checkUnsafeAssigns(SourceLocation Loc, 7372 QualType LHS, Expr *RHS) { 7373 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 7374 7375 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 7376 return false; 7377 7378 if (checkUnsafeAssignObject(*this, Loc, LT, RHS, false)) 7379 return true; 7380 7381 return false; 7382 } 7383 7384 void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 7385 Expr *LHS, Expr *RHS) { 7386 QualType LHSType; 7387 // PropertyRef on LHS type need be directly obtained from 7388 // its declaration as it has a PseudoType. 7389 ObjCPropertyRefExpr *PRE 7390 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 7391 if (PRE && !PRE->isImplicitProperty()) { 7392 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 7393 if (PD) 7394 LHSType = PD->getType(); 7395 } 7396 7397 if (LHSType.isNull()) 7398 LHSType = LHS->getType(); 7399 7400 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 7401 7402 if (LT == Qualifiers::OCL_Weak) { 7403 DiagnosticsEngine::Level Level = 7404 Diags.getDiagnosticLevel(diag::warn_arc_repeated_use_of_weak, Loc); 7405 if (Level != DiagnosticsEngine::Ignored) 7406 getCurFunction()->markSafeWeakUse(LHS); 7407 } 7408 7409 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 7410 return; 7411 7412 // FIXME. Check for other life times. 7413 if (LT != Qualifiers::OCL_None) 7414 return; 7415 7416 if (PRE) { 7417 if (PRE->isImplicitProperty()) 7418 return; 7419 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 7420 if (!PD) 7421 return; 7422 7423 unsigned Attributes = PD->getPropertyAttributes(); 7424 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 7425 // when 'assign' attribute was not explicitly specified 7426 // by user, ignore it and rely on property type itself 7427 // for lifetime info. 7428 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 7429 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 7430 LHSType->isObjCRetainableType()) 7431 return; 7432 7433 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 7434 if (cast->getCastKind() == CK_ARCConsumeObject) { 7435 Diag(Loc, diag::warn_arc_retained_property_assign) 7436 << RHS->getSourceRange(); 7437 return; 7438 } 7439 RHS = cast->getSubExpr(); 7440 } 7441 } 7442 else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) { 7443 if (checkUnsafeAssignObject(*this, Loc, Qualifiers::OCL_Weak, RHS, true)) 7444 return; 7445 } 7446 } 7447 } 7448 7449 //===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 7450 7451 namespace { 7452 bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 7453 SourceLocation StmtLoc, 7454 const NullStmt *Body) { 7455 // Do not warn if the body is a macro that expands to nothing, e.g: 7456 // 7457 // #define CALL(x) 7458 // if (condition) 7459 // CALL(0); 7460 // 7461 if (Body->hasLeadingEmptyMacro()) 7462 return false; 7463 7464 // Get line numbers of statement and body. 7465 bool StmtLineInvalid; 7466 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 7467 &StmtLineInvalid); 7468 if (StmtLineInvalid) 7469 return false; 7470 7471 bool BodyLineInvalid; 7472 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 7473 &BodyLineInvalid); 7474 if (BodyLineInvalid) 7475 return false; 7476 7477 // Warn if null statement and body are on the same line. 7478 if (StmtLine != BodyLine) 7479 return false; 7480 7481 return true; 7482 } 7483 } // Unnamed namespace 7484 7485 void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 7486 const Stmt *Body, 7487 unsigned DiagID) { 7488 // Since this is a syntactic check, don't emit diagnostic for template 7489 // instantiations, this just adds noise. 7490 if (CurrentInstantiationScope) 7491 return; 7492 7493 // The body should be a null statement. 7494 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 7495 if (!NBody) 7496 return; 7497 7498 // Do the usual checks. 7499 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 7500 return; 7501 7502 Diag(NBody->getSemiLoc(), DiagID); 7503 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 7504 } 7505 7506 void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 7507 const Stmt *PossibleBody) { 7508 assert(!CurrentInstantiationScope); // Ensured by caller 7509 7510 SourceLocation StmtLoc; 7511 const Stmt *Body; 7512 unsigned DiagID; 7513 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 7514 StmtLoc = FS->getRParenLoc(); 7515 Body = FS->getBody(); 7516 DiagID = diag::warn_empty_for_body; 7517 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 7518 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 7519 Body = WS->getBody(); 7520 DiagID = diag::warn_empty_while_body; 7521 } else 7522 return; // Neither `for' nor `while'. 7523 7524 // The body should be a null statement. 7525 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 7526 if (!NBody) 7527 return; 7528 7529 // Skip expensive checks if diagnostic is disabled. 7530 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 7531 DiagnosticsEngine::Ignored) 7532 return; 7533 7534 // Do the usual checks. 7535 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 7536 return; 7537 7538 // `for(...);' and `while(...);' are popular idioms, so in order to keep 7539 // noise level low, emit diagnostics only if for/while is followed by a 7540 // CompoundStmt, e.g.: 7541 // for (int i = 0; i < n; i++); 7542 // { 7543 // a(i); 7544 // } 7545 // or if for/while is followed by a statement with more indentation 7546 // than for/while itself: 7547 // for (int i = 0; i < n; i++); 7548 // a(i); 7549 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 7550 if (!ProbableTypo) { 7551 bool BodyColInvalid; 7552 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 7553 PossibleBody->getLocStart(), 7554 &BodyColInvalid); 7555 if (BodyColInvalid) 7556 return; 7557 7558 bool StmtColInvalid; 7559 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 7560 S->getLocStart(), 7561 &StmtColInvalid); 7562 if (StmtColInvalid) 7563 return; 7564 7565 if (BodyCol > StmtCol) 7566 ProbableTypo = true; 7567 } 7568 7569 if (ProbableTypo) { 7570 Diag(NBody->getSemiLoc(), DiagID); 7571 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 7572 } 7573 } 7574 7575 //===--- Layout compatibility ----------------------------------------------// 7576 7577 namespace { 7578 7579 bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2); 7580 7581 /// \brief Check if two enumeration types are layout-compatible. 7582 bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) { 7583 // C++11 [dcl.enum] p8: 7584 // Two enumeration types are layout-compatible if they have the same 7585 // underlying type. 7586 return ED1->isComplete() && ED2->isComplete() && 7587 C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType()); 7588 } 7589 7590 /// \brief Check if two fields are layout-compatible. 7591 bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) { 7592 if (!isLayoutCompatible(C, Field1->getType(), Field2->getType())) 7593 return false; 7594 7595 if (Field1->isBitField() != Field2->isBitField()) 7596 return false; 7597 7598 if (Field1->isBitField()) { 7599 // Make sure that the bit-fields are the same length. 7600 unsigned Bits1 = Field1->getBitWidthValue(C); 7601 unsigned Bits2 = Field2->getBitWidthValue(C); 7602 7603 if (Bits1 != Bits2) 7604 return false; 7605 } 7606 7607 return true; 7608 } 7609 7610 /// \brief Check if two standard-layout structs are layout-compatible. 7611 /// (C++11 [class.mem] p17) 7612 bool isLayoutCompatibleStruct(ASTContext &C, 7613 RecordDecl *RD1, 7614 RecordDecl *RD2) { 7615 // If both records are C++ classes, check that base classes match. 7616 if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) { 7617 // If one of records is a CXXRecordDecl we are in C++ mode, 7618 // thus the other one is a CXXRecordDecl, too. 7619 const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2); 7620 // Check number of base classes. 7621 if (D1CXX->getNumBases() != D2CXX->getNumBases()) 7622 return false; 7623 7624 // Check the base classes. 7625 for (CXXRecordDecl::base_class_const_iterator 7626 Base1 = D1CXX->bases_begin(), 7627 BaseEnd1 = D1CXX->bases_end(), 7628 Base2 = D2CXX->bases_begin(); 7629 Base1 != BaseEnd1; 7630 ++Base1, ++Base2) { 7631 if (!isLayoutCompatible(C, Base1->getType(), Base2->getType())) 7632 return false; 7633 } 7634 } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) { 7635 // If only RD2 is a C++ class, it should have zero base classes. 7636 if (D2CXX->getNumBases() > 0) 7637 return false; 7638 } 7639 7640 // Check the fields. 7641 RecordDecl::field_iterator Field2 = RD2->field_begin(), 7642 Field2End = RD2->field_end(), 7643 Field1 = RD1->field_begin(), 7644 Field1End = RD1->field_end(); 7645 for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) { 7646 if (!isLayoutCompatible(C, *Field1, *Field2)) 7647 return false; 7648 } 7649 if (Field1 != Field1End || Field2 != Field2End) 7650 return false; 7651 7652 return true; 7653 } 7654 7655 /// \brief Check if two standard-layout unions are layout-compatible. 7656 /// (C++11 [class.mem] p18) 7657 bool isLayoutCompatibleUnion(ASTContext &C, 7658 RecordDecl *RD1, 7659 RecordDecl *RD2) { 7660 llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields; 7661 for (auto *Field2 : RD2->fields()) 7662 UnmatchedFields.insert(Field2); 7663 7664 for (auto *Field1 : RD1->fields()) { 7665 llvm::SmallPtrSet<FieldDecl *, 8>::iterator 7666 I = UnmatchedFields.begin(), 7667 E = UnmatchedFields.end(); 7668 7669 for ( ; I != E; ++I) { 7670 if (isLayoutCompatible(C, Field1, *I)) { 7671 bool Result = UnmatchedFields.erase(*I); 7672 (void) Result; 7673 assert(Result); 7674 break; 7675 } 7676 } 7677 if (I == E) 7678 return false; 7679 } 7680 7681 return UnmatchedFields.empty(); 7682 } 7683 7684 bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) { 7685 if (RD1->isUnion() != RD2->isUnion()) 7686 return false; 7687 7688 if (RD1->isUnion()) 7689 return isLayoutCompatibleUnion(C, RD1, RD2); 7690 else 7691 return isLayoutCompatibleStruct(C, RD1, RD2); 7692 } 7693 7694 /// \brief Check if two types are layout-compatible in C++11 sense. 7695 bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) { 7696 if (T1.isNull() || T2.isNull()) 7697 return false; 7698 7699 // C++11 [basic.types] p11: 7700 // If two types T1 and T2 are the same type, then T1 and T2 are 7701 // layout-compatible types. 7702 if (C.hasSameType(T1, T2)) 7703 return true; 7704 7705 T1 = T1.getCanonicalType().getUnqualifiedType(); 7706 T2 = T2.getCanonicalType().getUnqualifiedType(); 7707 7708 const Type::TypeClass TC1 = T1->getTypeClass(); 7709 const Type::TypeClass TC2 = T2->getTypeClass(); 7710 7711 if (TC1 != TC2) 7712 return false; 7713 7714 if (TC1 == Type::Enum) { 7715 return isLayoutCompatible(C, 7716 cast<EnumType>(T1)->getDecl(), 7717 cast<EnumType>(T2)->getDecl()); 7718 } else if (TC1 == Type::Record) { 7719 if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType()) 7720 return false; 7721 7722 return isLayoutCompatible(C, 7723 cast<RecordType>(T1)->getDecl(), 7724 cast<RecordType>(T2)->getDecl()); 7725 } 7726 7727 return false; 7728 } 7729 } 7730 7731 //===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----// 7732 7733 namespace { 7734 /// \brief Given a type tag expression find the type tag itself. 7735 /// 7736 /// \param TypeExpr Type tag expression, as it appears in user's code. 7737 /// 7738 /// \param VD Declaration of an identifier that appears in a type tag. 7739 /// 7740 /// \param MagicValue Type tag magic value. 7741 bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx, 7742 const ValueDecl **VD, uint64_t *MagicValue) { 7743 while(true) { 7744 if (!TypeExpr) 7745 return false; 7746 7747 TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts(); 7748 7749 switch (TypeExpr->getStmtClass()) { 7750 case Stmt::UnaryOperatorClass: { 7751 const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr); 7752 if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) { 7753 TypeExpr = UO->getSubExpr(); 7754 continue; 7755 } 7756 return false; 7757 } 7758 7759 case Stmt::DeclRefExprClass: { 7760 const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr); 7761 *VD = DRE->getDecl(); 7762 return true; 7763 } 7764 7765 case Stmt::IntegerLiteralClass: { 7766 const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr); 7767 llvm::APInt MagicValueAPInt = IL->getValue(); 7768 if (MagicValueAPInt.getActiveBits() <= 64) { 7769 *MagicValue = MagicValueAPInt.getZExtValue(); 7770 return true; 7771 } else 7772 return false; 7773 } 7774 7775 case Stmt::BinaryConditionalOperatorClass: 7776 case Stmt::ConditionalOperatorClass: { 7777 const AbstractConditionalOperator *ACO = 7778 cast<AbstractConditionalOperator>(TypeExpr); 7779 bool Result; 7780 if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) { 7781 if (Result) 7782 TypeExpr = ACO->getTrueExpr(); 7783 else 7784 TypeExpr = ACO->getFalseExpr(); 7785 continue; 7786 } 7787 return false; 7788 } 7789 7790 case Stmt::BinaryOperatorClass: { 7791 const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr); 7792 if (BO->getOpcode() == BO_Comma) { 7793 TypeExpr = BO->getRHS(); 7794 continue; 7795 } 7796 return false; 7797 } 7798 7799 default: 7800 return false; 7801 } 7802 } 7803 } 7804 7805 /// \brief Retrieve the C type corresponding to type tag TypeExpr. 7806 /// 7807 /// \param TypeExpr Expression that specifies a type tag. 7808 /// 7809 /// \param MagicValues Registered magic values. 7810 /// 7811 /// \param FoundWrongKind Set to true if a type tag was found, but of a wrong 7812 /// kind. 7813 /// 7814 /// \param TypeInfo Information about the corresponding C type. 7815 /// 7816 /// \returns true if the corresponding C type was found. 7817 bool GetMatchingCType( 7818 const IdentifierInfo *ArgumentKind, 7819 const Expr *TypeExpr, const ASTContext &Ctx, 7820 const llvm::DenseMap<Sema::TypeTagMagicValue, 7821 Sema::TypeTagData> *MagicValues, 7822 bool &FoundWrongKind, 7823 Sema::TypeTagData &TypeInfo) { 7824 FoundWrongKind = false; 7825 7826 // Variable declaration that has type_tag_for_datatype attribute. 7827 const ValueDecl *VD = NULL; 7828 7829 uint64_t MagicValue; 7830 7831 if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue)) 7832 return false; 7833 7834 if (VD) { 7835 if (TypeTagForDatatypeAttr *I = VD->getAttr<TypeTagForDatatypeAttr>()) { 7836 if (I->getArgumentKind() != ArgumentKind) { 7837 FoundWrongKind = true; 7838 return false; 7839 } 7840 TypeInfo.Type = I->getMatchingCType(); 7841 TypeInfo.LayoutCompatible = I->getLayoutCompatible(); 7842 TypeInfo.MustBeNull = I->getMustBeNull(); 7843 return true; 7844 } 7845 return false; 7846 } 7847 7848 if (!MagicValues) 7849 return false; 7850 7851 llvm::DenseMap<Sema::TypeTagMagicValue, 7852 Sema::TypeTagData>::const_iterator I = 7853 MagicValues->find(std::make_pair(ArgumentKind, MagicValue)); 7854 if (I == MagicValues->end()) 7855 return false; 7856 7857 TypeInfo = I->second; 7858 return true; 7859 } 7860 } // unnamed namespace 7861 7862 void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind, 7863 uint64_t MagicValue, QualType Type, 7864 bool LayoutCompatible, 7865 bool MustBeNull) { 7866 if (!TypeTagForDatatypeMagicValues) 7867 TypeTagForDatatypeMagicValues.reset( 7868 new llvm::DenseMap<TypeTagMagicValue, TypeTagData>); 7869 7870 TypeTagMagicValue Magic(ArgumentKind, MagicValue); 7871 (*TypeTagForDatatypeMagicValues)[Magic] = 7872 TypeTagData(Type, LayoutCompatible, MustBeNull); 7873 } 7874 7875 namespace { 7876 bool IsSameCharType(QualType T1, QualType T2) { 7877 const BuiltinType *BT1 = T1->getAs<BuiltinType>(); 7878 if (!BT1) 7879 return false; 7880 7881 const BuiltinType *BT2 = T2->getAs<BuiltinType>(); 7882 if (!BT2) 7883 return false; 7884 7885 BuiltinType::Kind T1Kind = BT1->getKind(); 7886 BuiltinType::Kind T2Kind = BT2->getKind(); 7887 7888 return (T1Kind == BuiltinType::SChar && T2Kind == BuiltinType::Char_S) || 7889 (T1Kind == BuiltinType::UChar && T2Kind == BuiltinType::Char_U) || 7890 (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) || 7891 (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar); 7892 } 7893 } // unnamed namespace 7894 7895 void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr, 7896 const Expr * const *ExprArgs) { 7897 const IdentifierInfo *ArgumentKind = Attr->getArgumentKind(); 7898 bool IsPointerAttr = Attr->getIsPointer(); 7899 7900 const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()]; 7901 bool FoundWrongKind; 7902 TypeTagData TypeInfo; 7903 if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context, 7904 TypeTagForDatatypeMagicValues.get(), 7905 FoundWrongKind, TypeInfo)) { 7906 if (FoundWrongKind) 7907 Diag(TypeTagExpr->getExprLoc(), 7908 diag::warn_type_tag_for_datatype_wrong_kind) 7909 << TypeTagExpr->getSourceRange(); 7910 return; 7911 } 7912 7913 const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()]; 7914 if (IsPointerAttr) { 7915 // Skip implicit cast of pointer to `void *' (as a function argument). 7916 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr)) 7917 if (ICE->getType()->isVoidPointerType() && 7918 ICE->getCastKind() == CK_BitCast) 7919 ArgumentExpr = ICE->getSubExpr(); 7920 } 7921 QualType ArgumentType = ArgumentExpr->getType(); 7922 7923 // Passing a `void*' pointer shouldn't trigger a warning. 7924 if (IsPointerAttr && ArgumentType->isVoidPointerType()) 7925 return; 7926 7927 if (TypeInfo.MustBeNull) { 7928 // Type tag with matching void type requires a null pointer. 7929 if (!ArgumentExpr->isNullPointerConstant(Context, 7930 Expr::NPC_ValueDependentIsNotNull)) { 7931 Diag(ArgumentExpr->getExprLoc(), 7932 diag::warn_type_safety_null_pointer_required) 7933 << ArgumentKind->getName() 7934 << ArgumentExpr->getSourceRange() 7935 << TypeTagExpr->getSourceRange(); 7936 } 7937 return; 7938 } 7939 7940 QualType RequiredType = TypeInfo.Type; 7941 if (IsPointerAttr) 7942 RequiredType = Context.getPointerType(RequiredType); 7943 7944 bool mismatch = false; 7945 if (!TypeInfo.LayoutCompatible) { 7946 mismatch = !Context.hasSameType(ArgumentType, RequiredType); 7947 7948 // C++11 [basic.fundamental] p1: 7949 // Plain char, signed char, and unsigned char are three distinct types. 7950 // 7951 // But we treat plain `char' as equivalent to `signed char' or `unsigned 7952 // char' depending on the current char signedness mode. 7953 if (mismatch) 7954 if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(), 7955 RequiredType->getPointeeType())) || 7956 (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType))) 7957 mismatch = false; 7958 } else 7959 if (IsPointerAttr) 7960 mismatch = !isLayoutCompatible(Context, 7961 ArgumentType->getPointeeType(), 7962 RequiredType->getPointeeType()); 7963 else 7964 mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType); 7965 7966 if (mismatch) 7967 Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch) 7968 << ArgumentType << ArgumentKind 7969 << TypeInfo.LayoutCompatible << RequiredType 7970 << ArgumentExpr->getSourceRange() 7971 << TypeTagExpr->getSourceRange(); 7972 } 7973 7974