SemaChecking.cpp revision e3fbbe9a486c4c8709726d8c79af33140eb02e53
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/BitVector.h" 36#include "llvm/ADT/STLExtras.h" 37#include "llvm/ADT/SmallString.h" 38#include "llvm/Support/ConvertUTF.h" 39#include "llvm/Support/raw_ostream.h" 40#include <limits> 41using namespace clang; 42using namespace sema; 43 44SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 45 unsigned ByteNo) const { 46 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 47 PP.getLangOpts(), PP.getTargetInfo()); 48} 49 50/// Checks that a call expression's argument count is the desired number. 51/// This is useful when doing custom type-checking. Returns true on error. 52static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 53 unsigned argCount = call->getNumArgs(); 54 if (argCount == desiredArgCount) return false; 55 56 if (argCount < desiredArgCount) 57 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 58 << 0 /*function call*/ << desiredArgCount << argCount 59 << call->getSourceRange(); 60 61 // Highlight all the excess arguments. 62 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 63 call->getArg(argCount - 1)->getLocEnd()); 64 65 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 66 << 0 /*function call*/ << desiredArgCount << argCount 67 << call->getArg(1)->getSourceRange(); 68} 69 70/// Check that the first argument to __builtin_annotation is an integer 71/// and the second argument is a non-wide string literal. 72static bool SemaBuiltinAnnotation(Sema &S, CallExpr *TheCall) { 73 if (checkArgCount(S, TheCall, 2)) 74 return true; 75 76 // First argument should be an integer. 77 Expr *ValArg = TheCall->getArg(0); 78 QualType Ty = ValArg->getType(); 79 if (!Ty->isIntegerType()) { 80 S.Diag(ValArg->getLocStart(), diag::err_builtin_annotation_first_arg) 81 << ValArg->getSourceRange(); 82 return true; 83 } 84 85 // Second argument should be a constant string. 86 Expr *StrArg = TheCall->getArg(1)->IgnoreParenCasts(); 87 StringLiteral *Literal = dyn_cast<StringLiteral>(StrArg); 88 if (!Literal || !Literal->isAscii()) { 89 S.Diag(StrArg->getLocStart(), diag::err_builtin_annotation_second_arg) 90 << StrArg->getSourceRange(); 91 return true; 92 } 93 94 TheCall->setType(Ty); 95 return false; 96} 97 98/// Check that the argument to __builtin_addressof is a glvalue, and set the 99/// result type to the corresponding pointer type. 100static bool SemaBuiltinAddressof(Sema &S, CallExpr *TheCall) { 101 if (checkArgCount(S, TheCall, 1)) 102 return true; 103 104 ExprResult Arg(S.Owned(TheCall->getArg(0))); 105 QualType ResultType = S.CheckAddressOfOperand(Arg, TheCall->getLocStart()); 106 if (ResultType.isNull()) 107 return true; 108 109 TheCall->setArg(0, Arg.take()); 110 TheCall->setType(ResultType); 111 return false; 112} 113 114ExprResult 115Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 116 ExprResult TheCallResult(Owned(TheCall)); 117 118 // Find out if any arguments are required to be integer constant expressions. 119 unsigned ICEArguments = 0; 120 ASTContext::GetBuiltinTypeError Error; 121 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 122 if (Error != ASTContext::GE_None) 123 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 124 125 // If any arguments are required to be ICE's, check and diagnose. 126 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 127 // Skip arguments not required to be ICE's. 128 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 129 130 llvm::APSInt Result; 131 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 132 return true; 133 ICEArguments &= ~(1 << ArgNo); 134 } 135 136 switch (BuiltinID) { 137 case Builtin::BI__builtin___CFStringMakeConstantString: 138 assert(TheCall->getNumArgs() == 1 && 139 "Wrong # arguments to builtin CFStringMakeConstantString"); 140 if (CheckObjCString(TheCall->getArg(0))) 141 return ExprError(); 142 break; 143 case Builtin::BI__builtin_stdarg_start: 144 case Builtin::BI__builtin_va_start: 145 if (SemaBuiltinVAStart(TheCall)) 146 return ExprError(); 147 break; 148 case Builtin::BI__builtin_isgreater: 149 case Builtin::BI__builtin_isgreaterequal: 150 case Builtin::BI__builtin_isless: 151 case Builtin::BI__builtin_islessequal: 152 case Builtin::BI__builtin_islessgreater: 153 case Builtin::BI__builtin_isunordered: 154 if (SemaBuiltinUnorderedCompare(TheCall)) 155 return ExprError(); 156 break; 157 case Builtin::BI__builtin_fpclassify: 158 if (SemaBuiltinFPClassification(TheCall, 6)) 159 return ExprError(); 160 break; 161 case Builtin::BI__builtin_isfinite: 162 case Builtin::BI__builtin_isinf: 163 case Builtin::BI__builtin_isinf_sign: 164 case Builtin::BI__builtin_isnan: 165 case Builtin::BI__builtin_isnormal: 166 if (SemaBuiltinFPClassification(TheCall, 1)) 167 return ExprError(); 168 break; 169 case Builtin::BI__builtin_shufflevector: 170 return SemaBuiltinShuffleVector(TheCall); 171 // TheCall will be freed by the smart pointer here, but that's fine, since 172 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 173 case Builtin::BI__builtin_prefetch: 174 if (SemaBuiltinPrefetch(TheCall)) 175 return ExprError(); 176 break; 177 case Builtin::BI__builtin_object_size: 178 if (SemaBuiltinObjectSize(TheCall)) 179 return ExprError(); 180 break; 181 case Builtin::BI__builtin_longjmp: 182 if (SemaBuiltinLongjmp(TheCall)) 183 return ExprError(); 184 break; 185 186 case Builtin::BI__builtin_classify_type: 187 if (checkArgCount(*this, TheCall, 1)) return true; 188 TheCall->setType(Context.IntTy); 189 break; 190 case Builtin::BI__builtin_constant_p: 191 if (checkArgCount(*this, TheCall, 1)) return true; 192 TheCall->setType(Context.IntTy); 193 break; 194 case Builtin::BI__sync_fetch_and_add: 195 case Builtin::BI__sync_fetch_and_add_1: 196 case Builtin::BI__sync_fetch_and_add_2: 197 case Builtin::BI__sync_fetch_and_add_4: 198 case Builtin::BI__sync_fetch_and_add_8: 199 case Builtin::BI__sync_fetch_and_add_16: 200 case Builtin::BI__sync_fetch_and_sub: 201 case Builtin::BI__sync_fetch_and_sub_1: 202 case Builtin::BI__sync_fetch_and_sub_2: 203 case Builtin::BI__sync_fetch_and_sub_4: 204 case Builtin::BI__sync_fetch_and_sub_8: 205 case Builtin::BI__sync_fetch_and_sub_16: 206 case Builtin::BI__sync_fetch_and_or: 207 case Builtin::BI__sync_fetch_and_or_1: 208 case Builtin::BI__sync_fetch_and_or_2: 209 case Builtin::BI__sync_fetch_and_or_4: 210 case Builtin::BI__sync_fetch_and_or_8: 211 case Builtin::BI__sync_fetch_and_or_16: 212 case Builtin::BI__sync_fetch_and_and: 213 case Builtin::BI__sync_fetch_and_and_1: 214 case Builtin::BI__sync_fetch_and_and_2: 215 case Builtin::BI__sync_fetch_and_and_4: 216 case Builtin::BI__sync_fetch_and_and_8: 217 case Builtin::BI__sync_fetch_and_and_16: 218 case Builtin::BI__sync_fetch_and_xor: 219 case Builtin::BI__sync_fetch_and_xor_1: 220 case Builtin::BI__sync_fetch_and_xor_2: 221 case Builtin::BI__sync_fetch_and_xor_4: 222 case Builtin::BI__sync_fetch_and_xor_8: 223 case Builtin::BI__sync_fetch_and_xor_16: 224 case Builtin::BI__sync_add_and_fetch: 225 case Builtin::BI__sync_add_and_fetch_1: 226 case Builtin::BI__sync_add_and_fetch_2: 227 case Builtin::BI__sync_add_and_fetch_4: 228 case Builtin::BI__sync_add_and_fetch_8: 229 case Builtin::BI__sync_add_and_fetch_16: 230 case Builtin::BI__sync_sub_and_fetch: 231 case Builtin::BI__sync_sub_and_fetch_1: 232 case Builtin::BI__sync_sub_and_fetch_2: 233 case Builtin::BI__sync_sub_and_fetch_4: 234 case Builtin::BI__sync_sub_and_fetch_8: 235 case Builtin::BI__sync_sub_and_fetch_16: 236 case Builtin::BI__sync_and_and_fetch: 237 case Builtin::BI__sync_and_and_fetch_1: 238 case Builtin::BI__sync_and_and_fetch_2: 239 case Builtin::BI__sync_and_and_fetch_4: 240 case Builtin::BI__sync_and_and_fetch_8: 241 case Builtin::BI__sync_and_and_fetch_16: 242 case Builtin::BI__sync_or_and_fetch: 243 case Builtin::BI__sync_or_and_fetch_1: 244 case Builtin::BI__sync_or_and_fetch_2: 245 case Builtin::BI__sync_or_and_fetch_4: 246 case Builtin::BI__sync_or_and_fetch_8: 247 case Builtin::BI__sync_or_and_fetch_16: 248 case Builtin::BI__sync_xor_and_fetch: 249 case Builtin::BI__sync_xor_and_fetch_1: 250 case Builtin::BI__sync_xor_and_fetch_2: 251 case Builtin::BI__sync_xor_and_fetch_4: 252 case Builtin::BI__sync_xor_and_fetch_8: 253 case Builtin::BI__sync_xor_and_fetch_16: 254 case Builtin::BI__sync_val_compare_and_swap: 255 case Builtin::BI__sync_val_compare_and_swap_1: 256 case Builtin::BI__sync_val_compare_and_swap_2: 257 case Builtin::BI__sync_val_compare_and_swap_4: 258 case Builtin::BI__sync_val_compare_and_swap_8: 259 case Builtin::BI__sync_val_compare_and_swap_16: 260 case Builtin::BI__sync_bool_compare_and_swap: 261 case Builtin::BI__sync_bool_compare_and_swap_1: 262 case Builtin::BI__sync_bool_compare_and_swap_2: 263 case Builtin::BI__sync_bool_compare_and_swap_4: 264 case Builtin::BI__sync_bool_compare_and_swap_8: 265 case Builtin::BI__sync_bool_compare_and_swap_16: 266 case Builtin::BI__sync_lock_test_and_set: 267 case Builtin::BI__sync_lock_test_and_set_1: 268 case Builtin::BI__sync_lock_test_and_set_2: 269 case Builtin::BI__sync_lock_test_and_set_4: 270 case Builtin::BI__sync_lock_test_and_set_8: 271 case Builtin::BI__sync_lock_test_and_set_16: 272 case Builtin::BI__sync_lock_release: 273 case Builtin::BI__sync_lock_release_1: 274 case Builtin::BI__sync_lock_release_2: 275 case Builtin::BI__sync_lock_release_4: 276 case Builtin::BI__sync_lock_release_8: 277 case Builtin::BI__sync_lock_release_16: 278 case Builtin::BI__sync_swap: 279 case Builtin::BI__sync_swap_1: 280 case Builtin::BI__sync_swap_2: 281 case Builtin::BI__sync_swap_4: 282 case Builtin::BI__sync_swap_8: 283 case Builtin::BI__sync_swap_16: 284 return SemaBuiltinAtomicOverloaded(TheCallResult); 285#define BUILTIN(ID, TYPE, ATTRS) 286#define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \ 287 case Builtin::BI##ID: \ 288 return SemaAtomicOpsOverloaded(TheCallResult, AtomicExpr::AO##ID); 289#include "clang/Basic/Builtins.def" 290 case Builtin::BI__builtin_annotation: 291 if (SemaBuiltinAnnotation(*this, TheCall)) 292 return ExprError(); 293 break; 294 case Builtin::BI__builtin_addressof: 295 if (SemaBuiltinAddressof(*this, TheCall)) 296 return ExprError(); 297 break; 298 } 299 300 // Since the target specific builtins for each arch overlap, only check those 301 // of the arch we are compiling for. 302 if (BuiltinID >= Builtin::FirstTSBuiltin) { 303 switch (Context.getTargetInfo().getTriple().getArch()) { 304 case llvm::Triple::arm: 305 case llvm::Triple::thumb: 306 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 307 return ExprError(); 308 break; 309 case llvm::Triple::mips: 310 case llvm::Triple::mipsel: 311 case llvm::Triple::mips64: 312 case llvm::Triple::mips64el: 313 if (CheckMipsBuiltinFunctionCall(BuiltinID, TheCall)) 314 return ExprError(); 315 break; 316 default: 317 break; 318 } 319 } 320 321 return TheCallResult; 322} 323 324// Get the valid immediate range for the specified NEON type code. 325static unsigned RFT(unsigned t, bool shift = false) { 326 NeonTypeFlags Type(t); 327 int IsQuad = Type.isQuad(); 328 switch (Type.getEltType()) { 329 case NeonTypeFlags::Int8: 330 case NeonTypeFlags::Poly8: 331 return shift ? 7 : (8 << IsQuad) - 1; 332 case NeonTypeFlags::Int16: 333 case NeonTypeFlags::Poly16: 334 return shift ? 15 : (4 << IsQuad) - 1; 335 case NeonTypeFlags::Int32: 336 return shift ? 31 : (2 << IsQuad) - 1; 337 case NeonTypeFlags::Int64: 338 return shift ? 63 : (1 << IsQuad) - 1; 339 case NeonTypeFlags::Float16: 340 assert(!shift && "cannot shift float types!"); 341 return (4 << IsQuad) - 1; 342 case NeonTypeFlags::Float32: 343 assert(!shift && "cannot shift float types!"); 344 return (2 << IsQuad) - 1; 345 } 346 llvm_unreachable("Invalid NeonTypeFlag!"); 347} 348 349/// getNeonEltType - Return the QualType corresponding to the elements of 350/// the vector type specified by the NeonTypeFlags. This is used to check 351/// the pointer arguments for Neon load/store intrinsics. 352static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) { 353 switch (Flags.getEltType()) { 354 case NeonTypeFlags::Int8: 355 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 356 case NeonTypeFlags::Int16: 357 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 358 case NeonTypeFlags::Int32: 359 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 360 case NeonTypeFlags::Int64: 361 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 362 case NeonTypeFlags::Poly8: 363 return Context.SignedCharTy; 364 case NeonTypeFlags::Poly16: 365 return Context.ShortTy; 366 case NeonTypeFlags::Float16: 367 return Context.UnsignedShortTy; 368 case NeonTypeFlags::Float32: 369 return Context.FloatTy; 370 } 371 llvm_unreachable("Invalid NeonTypeFlag!"); 372} 373 374bool Sema::CheckARMBuiltinExclusiveCall(unsigned BuiltinID, CallExpr *TheCall) { 375 assert((BuiltinID == ARM::BI__builtin_arm_ldrex || 376 BuiltinID == ARM::BI__builtin_arm_strex) && 377 "unexpected ARM builtin"); 378 bool IsLdrex = BuiltinID == ARM::BI__builtin_arm_ldrex; 379 380 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 381 382 // Ensure that we have the proper number of arguments. 383 if (checkArgCount(*this, TheCall, IsLdrex ? 1 : 2)) 384 return true; 385 386 // Inspect the pointer argument of the atomic builtin. This should always be 387 // a pointer type, whose element is an integral scalar or pointer type. 388 // Because it is a pointer type, we don't have to worry about any implicit 389 // casts here. 390 Expr *PointerArg = TheCall->getArg(IsLdrex ? 0 : 1); 391 ExprResult PointerArgRes = DefaultFunctionArrayLvalueConversion(PointerArg); 392 if (PointerArgRes.isInvalid()) 393 return true; 394 PointerArg = PointerArgRes.take(); 395 396 const PointerType *pointerType = PointerArg->getType()->getAs<PointerType>(); 397 if (!pointerType) { 398 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 399 << PointerArg->getType() << PointerArg->getSourceRange(); 400 return true; 401 } 402 403 // ldrex takes a "const volatile T*" and strex takes a "volatile T*". Our next 404 // task is to insert the appropriate casts into the AST. First work out just 405 // what the appropriate type is. 406 QualType ValType = pointerType->getPointeeType(); 407 QualType AddrType = ValType.getUnqualifiedType().withVolatile(); 408 if (IsLdrex) 409 AddrType.addConst(); 410 411 // Issue a warning if the cast is dodgy. 412 CastKind CastNeeded = CK_NoOp; 413 if (!AddrType.isAtLeastAsQualifiedAs(ValType)) { 414 CastNeeded = CK_BitCast; 415 Diag(DRE->getLocStart(), diag::ext_typecheck_convert_discards_qualifiers) 416 << PointerArg->getType() 417 << Context.getPointerType(AddrType) 418 << AA_Passing << PointerArg->getSourceRange(); 419 } 420 421 // Finally, do the cast and replace the argument with the corrected version. 422 AddrType = Context.getPointerType(AddrType); 423 PointerArgRes = ImpCastExprToType(PointerArg, AddrType, CastNeeded); 424 if (PointerArgRes.isInvalid()) 425 return true; 426 PointerArg = PointerArgRes.take(); 427 428 TheCall->setArg(IsLdrex ? 0 : 1, PointerArg); 429 430 // In general, we allow ints, floats and pointers to be loaded and stored. 431 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 432 !ValType->isBlockPointerType() && !ValType->isFloatingType()) { 433 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intfltptr) 434 << PointerArg->getType() << PointerArg->getSourceRange(); 435 return true; 436 } 437 438 // But ARM doesn't have instructions to deal with 128-bit versions. 439 if (Context.getTypeSize(ValType) > 64) { 440 Diag(DRE->getLocStart(), diag::err_atomic_exclusive_builtin_pointer_size) 441 << PointerArg->getType() << PointerArg->getSourceRange(); 442 return true; 443 } 444 445 switch (ValType.getObjCLifetime()) { 446 case Qualifiers::OCL_None: 447 case Qualifiers::OCL_ExplicitNone: 448 // okay 449 break; 450 451 case Qualifiers::OCL_Weak: 452 case Qualifiers::OCL_Strong: 453 case Qualifiers::OCL_Autoreleasing: 454 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 455 << ValType << PointerArg->getSourceRange(); 456 return true; 457 } 458 459 460 if (IsLdrex) { 461 TheCall->setType(ValType); 462 return false; 463 } 464 465 // Initialize the argument to be stored. 466 ExprResult ValArg = TheCall->getArg(0); 467 InitializedEntity Entity = InitializedEntity::InitializeParameter( 468 Context, ValType, /*consume*/ false); 469 ValArg = PerformCopyInitialization(Entity, SourceLocation(), ValArg); 470 if (ValArg.isInvalid()) 471 return true; 472 473 TheCall->setArg(0, ValArg.get()); 474 return false; 475} 476 477bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 478 llvm::APSInt Result; 479 480 if (BuiltinID == ARM::BI__builtin_arm_ldrex || 481 BuiltinID == ARM::BI__builtin_arm_strex) { 482 return CheckARMBuiltinExclusiveCall(BuiltinID, TheCall); 483 } 484 485 uint64_t mask = 0; 486 unsigned TV = 0; 487 int PtrArgNum = -1; 488 bool HasConstPtr = false; 489 switch (BuiltinID) { 490#define GET_NEON_OVERLOAD_CHECK 491#include "clang/Basic/arm_neon.inc" 492#undef GET_NEON_OVERLOAD_CHECK 493 } 494 495 // For NEON intrinsics which are overloaded on vector element type, validate 496 // the immediate which specifies which variant to emit. 497 unsigned ImmArg = TheCall->getNumArgs()-1; 498 if (mask) { 499 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 500 return true; 501 502 TV = Result.getLimitedValue(64); 503 if ((TV > 63) || (mask & (1ULL << TV)) == 0) 504 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 505 << TheCall->getArg(ImmArg)->getSourceRange(); 506 } 507 508 if (PtrArgNum >= 0) { 509 // Check that pointer arguments have the specified type. 510 Expr *Arg = TheCall->getArg(PtrArgNum); 511 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 512 Arg = ICE->getSubExpr(); 513 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 514 QualType RHSTy = RHS.get()->getType(); 515 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 516 if (HasConstPtr) 517 EltTy = EltTy.withConst(); 518 QualType LHSTy = Context.getPointerType(EltTy); 519 AssignConvertType ConvTy; 520 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 521 if (RHS.isInvalid()) 522 return true; 523 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 524 RHS.get(), AA_Assigning)) 525 return true; 526 } 527 528 // For NEON intrinsics which take an immediate value as part of the 529 // instruction, range check them here. 530 unsigned i = 0, l = 0, u = 0; 531 switch (BuiltinID) { 532 default: return false; 533 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 534 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 535 case ARM::BI__builtin_arm_vcvtr_f: 536 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 537#define GET_NEON_IMMEDIATE_CHECK 538#include "clang/Basic/arm_neon.inc" 539#undef GET_NEON_IMMEDIATE_CHECK 540 }; 541 542 // We can't check the value of a dependent argument. 543 if (TheCall->getArg(i)->isTypeDependent() || 544 TheCall->getArg(i)->isValueDependent()) 545 return false; 546 547 // Check that the immediate argument is actually a constant. 548 if (SemaBuiltinConstantArg(TheCall, i, Result)) 549 return true; 550 551 // Range check against the upper/lower values for this isntruction. 552 unsigned Val = Result.getZExtValue(); 553 if (Val < l || Val > (u + l)) 554 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 555 << l << u+l << TheCall->getArg(i)->getSourceRange(); 556 557 // FIXME: VFP Intrinsics should error if VFP not present. 558 return false; 559} 560 561bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 562 unsigned i = 0, l = 0, u = 0; 563 switch (BuiltinID) { 564 default: return false; 565 case Mips::BI__builtin_mips_wrdsp: i = 1; l = 0; u = 63; break; 566 case Mips::BI__builtin_mips_rddsp: i = 0; l = 0; u = 63; break; 567 case Mips::BI__builtin_mips_append: i = 2; l = 0; u = 31; break; 568 case Mips::BI__builtin_mips_balign: i = 2; l = 0; u = 3; break; 569 case Mips::BI__builtin_mips_precr_sra_ph_w: i = 2; l = 0; u = 31; break; 570 case Mips::BI__builtin_mips_precr_sra_r_ph_w: i = 2; l = 0; u = 31; break; 571 case Mips::BI__builtin_mips_prepend: i = 2; l = 0; u = 31; break; 572 }; 573 574 // We can't check the value of a dependent argument. 575 if (TheCall->getArg(i)->isTypeDependent() || 576 TheCall->getArg(i)->isValueDependent()) 577 return false; 578 579 // Check that the immediate argument is actually a constant. 580 llvm::APSInt Result; 581 if (SemaBuiltinConstantArg(TheCall, i, Result)) 582 return true; 583 584 // Range check against the upper/lower values for this instruction. 585 unsigned Val = Result.getZExtValue(); 586 if (Val < l || Val > u) 587 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 588 << l << u << TheCall->getArg(i)->getSourceRange(); 589 590 return false; 591} 592 593/// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo 594/// parameter with the FormatAttr's correct format_idx and firstDataArg. 595/// Returns true when the format fits the function and the FormatStringInfo has 596/// been populated. 597bool Sema::getFormatStringInfo(const FormatAttr *Format, bool IsCXXMember, 598 FormatStringInfo *FSI) { 599 FSI->HasVAListArg = Format->getFirstArg() == 0; 600 FSI->FormatIdx = Format->getFormatIdx() - 1; 601 FSI->FirstDataArg = FSI->HasVAListArg ? 0 : Format->getFirstArg() - 1; 602 603 // The way the format attribute works in GCC, the implicit this argument 604 // of member functions is counted. However, it doesn't appear in our own 605 // lists, so decrement format_idx in that case. 606 if (IsCXXMember) { 607 if(FSI->FormatIdx == 0) 608 return false; 609 --FSI->FormatIdx; 610 if (FSI->FirstDataArg != 0) 611 --FSI->FirstDataArg; 612 } 613 return true; 614} 615 616/// Handles the checks for format strings, non-POD arguments to vararg 617/// functions, and NULL arguments passed to non-NULL parameters. 618void Sema::checkCall(NamedDecl *FDecl, 619 ArrayRef<const Expr *> Args, 620 unsigned NumProtoArgs, 621 bool IsMemberFunction, 622 SourceLocation Loc, 623 SourceRange Range, 624 VariadicCallType CallType) { 625 if (CurContext->isDependentContext()) 626 return; 627 628 // Printf and scanf checking. 629 bool HandledFormatString = false; 630 if (FDecl) 631 for (specific_attr_iterator<FormatAttr> 632 I = FDecl->specific_attr_begin<FormatAttr>(), 633 E = FDecl->specific_attr_end<FormatAttr>(); I != E ; ++I) 634 if (CheckFormatArguments(*I, Args, IsMemberFunction, CallType, Loc, 635 Range)) 636 HandledFormatString = true; 637 638 // Refuse POD arguments that weren't caught by the format string 639 // checks above. 640 if (!HandledFormatString && CallType != VariadicDoesNotApply) 641 for (unsigned ArgIdx = NumProtoArgs; ArgIdx < Args.size(); ++ArgIdx) { 642 // Args[ArgIdx] can be null in malformed code. 643 if (const Expr *Arg = Args[ArgIdx]) 644 variadicArgumentPODCheck(Arg, CallType); 645 } 646 647 if (FDecl) { 648 for (specific_attr_iterator<NonNullAttr> 649 I = FDecl->specific_attr_begin<NonNullAttr>(), 650 E = FDecl->specific_attr_end<NonNullAttr>(); I != E; ++I) 651 CheckNonNullArguments(*I, Args.data(), Loc); 652 653 // Type safety checking. 654 for (specific_attr_iterator<ArgumentWithTypeTagAttr> 655 i = FDecl->specific_attr_begin<ArgumentWithTypeTagAttr>(), 656 e = FDecl->specific_attr_end<ArgumentWithTypeTagAttr>(); 657 i != e; ++i) { 658 CheckArgumentWithTypeTag(*i, Args.data()); 659 } 660 } 661} 662 663/// CheckConstructorCall - Check a constructor call for correctness and safety 664/// properties not enforced by the C type system. 665void Sema::CheckConstructorCall(FunctionDecl *FDecl, 666 ArrayRef<const Expr *> Args, 667 const FunctionProtoType *Proto, 668 SourceLocation Loc) { 669 VariadicCallType CallType = 670 Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply; 671 checkCall(FDecl, Args, Proto->getNumArgs(), 672 /*IsMemberFunction=*/true, Loc, SourceRange(), CallType); 673} 674 675/// CheckFunctionCall - Check a direct function call for various correctness 676/// and safety properties not strictly enforced by the C type system. 677bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall, 678 const FunctionProtoType *Proto) { 679 bool IsMemberOperatorCall = isa<CXXOperatorCallExpr>(TheCall) && 680 isa<CXXMethodDecl>(FDecl); 681 bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall) || 682 IsMemberOperatorCall; 683 VariadicCallType CallType = getVariadicCallType(FDecl, Proto, 684 TheCall->getCallee()); 685 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 686 Expr** Args = TheCall->getArgs(); 687 unsigned NumArgs = TheCall->getNumArgs(); 688 if (IsMemberOperatorCall) { 689 // If this is a call to a member operator, hide the first argument 690 // from checkCall. 691 // FIXME: Our choice of AST representation here is less than ideal. 692 ++Args; 693 --NumArgs; 694 } 695 checkCall(FDecl, llvm::makeArrayRef<const Expr *>(Args, NumArgs), 696 NumProtoArgs, 697 IsMemberFunction, TheCall->getRParenLoc(), 698 TheCall->getCallee()->getSourceRange(), CallType); 699 700 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 701 // None of the checks below are needed for functions that don't have 702 // simple names (e.g., C++ conversion functions). 703 if (!FnInfo) 704 return false; 705 706 unsigned CMId = FDecl->getMemoryFunctionKind(); 707 if (CMId == 0) 708 return false; 709 710 // Handle memory setting and copying functions. 711 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 712 CheckStrlcpycatArguments(TheCall, FnInfo); 713 else if (CMId == Builtin::BIstrncat) 714 CheckStrncatArguments(TheCall, FnInfo); 715 else 716 CheckMemaccessArguments(TheCall, CMId, FnInfo); 717 718 return false; 719} 720 721bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 722 ArrayRef<const Expr *> Args) { 723 VariadicCallType CallType = 724 Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply; 725 726 checkCall(Method, Args, Method->param_size(), 727 /*IsMemberFunction=*/false, 728 lbrac, Method->getSourceRange(), CallType); 729 730 return false; 731} 732 733bool Sema::CheckPointerCall(NamedDecl *NDecl, CallExpr *TheCall, 734 const FunctionProtoType *Proto) { 735 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 736 if (!V) 737 return false; 738 739 QualType Ty = V->getType(); 740 if (!Ty->isBlockPointerType() && !Ty->isFunctionPointerType()) 741 return false; 742 743 VariadicCallType CallType; 744 if (!Proto || !Proto->isVariadic()) { 745 CallType = VariadicDoesNotApply; 746 } else if (Ty->isBlockPointerType()) { 747 CallType = VariadicBlock; 748 } else { // Ty->isFunctionPointerType() 749 CallType = VariadicFunction; 750 } 751 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 752 753 checkCall(NDecl, 754 llvm::makeArrayRef<const Expr *>(TheCall->getArgs(), 755 TheCall->getNumArgs()), 756 NumProtoArgs, /*IsMemberFunction=*/false, 757 TheCall->getRParenLoc(), 758 TheCall->getCallee()->getSourceRange(), CallType); 759 760 return false; 761} 762 763/// Checks function calls when a FunctionDecl or a NamedDecl is not available, 764/// such as function pointers returned from functions. 765bool Sema::CheckOtherCall(CallExpr *TheCall, const FunctionProtoType *Proto) { 766 VariadicCallType CallType = getVariadicCallType(/*FDecl=*/0, Proto, 767 TheCall->getCallee()); 768 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 769 770 checkCall(/*FDecl=*/0, 771 llvm::makeArrayRef<const Expr *>(TheCall->getArgs(), 772 TheCall->getNumArgs()), 773 NumProtoArgs, /*IsMemberFunction=*/false, 774 TheCall->getRParenLoc(), 775 TheCall->getCallee()->getSourceRange(), CallType); 776 777 return false; 778} 779 780ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 781 AtomicExpr::AtomicOp Op) { 782 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 783 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 784 785 // All these operations take one of the following forms: 786 enum { 787 // C __c11_atomic_init(A *, C) 788 Init, 789 // C __c11_atomic_load(A *, int) 790 Load, 791 // void __atomic_load(A *, CP, int) 792 Copy, 793 // C __c11_atomic_add(A *, M, int) 794 Arithmetic, 795 // C __atomic_exchange_n(A *, CP, int) 796 Xchg, 797 // void __atomic_exchange(A *, C *, CP, int) 798 GNUXchg, 799 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 800 C11CmpXchg, 801 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 802 GNUCmpXchg 803 } Form = Init; 804 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 805 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 806 // where: 807 // C is an appropriate type, 808 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 809 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 810 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 811 // the int parameters are for orderings. 812 813 assert(AtomicExpr::AO__c11_atomic_init == 0 && 814 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 815 && "need to update code for modified C11 atomics"); 816 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 817 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 818 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 819 Op == AtomicExpr::AO__atomic_store_n || 820 Op == AtomicExpr::AO__atomic_exchange_n || 821 Op == AtomicExpr::AO__atomic_compare_exchange_n; 822 bool IsAddSub = false; 823 824 switch (Op) { 825 case AtomicExpr::AO__c11_atomic_init: 826 Form = Init; 827 break; 828 829 case AtomicExpr::AO__c11_atomic_load: 830 case AtomicExpr::AO__atomic_load_n: 831 Form = Load; 832 break; 833 834 case AtomicExpr::AO__c11_atomic_store: 835 case AtomicExpr::AO__atomic_load: 836 case AtomicExpr::AO__atomic_store: 837 case AtomicExpr::AO__atomic_store_n: 838 Form = Copy; 839 break; 840 841 case AtomicExpr::AO__c11_atomic_fetch_add: 842 case AtomicExpr::AO__c11_atomic_fetch_sub: 843 case AtomicExpr::AO__atomic_fetch_add: 844 case AtomicExpr::AO__atomic_fetch_sub: 845 case AtomicExpr::AO__atomic_add_fetch: 846 case AtomicExpr::AO__atomic_sub_fetch: 847 IsAddSub = true; 848 // Fall through. 849 case AtomicExpr::AO__c11_atomic_fetch_and: 850 case AtomicExpr::AO__c11_atomic_fetch_or: 851 case AtomicExpr::AO__c11_atomic_fetch_xor: 852 case AtomicExpr::AO__atomic_fetch_and: 853 case AtomicExpr::AO__atomic_fetch_or: 854 case AtomicExpr::AO__atomic_fetch_xor: 855 case AtomicExpr::AO__atomic_fetch_nand: 856 case AtomicExpr::AO__atomic_and_fetch: 857 case AtomicExpr::AO__atomic_or_fetch: 858 case AtomicExpr::AO__atomic_xor_fetch: 859 case AtomicExpr::AO__atomic_nand_fetch: 860 Form = Arithmetic; 861 break; 862 863 case AtomicExpr::AO__c11_atomic_exchange: 864 case AtomicExpr::AO__atomic_exchange_n: 865 Form = Xchg; 866 break; 867 868 case AtomicExpr::AO__atomic_exchange: 869 Form = GNUXchg; 870 break; 871 872 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 873 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 874 Form = C11CmpXchg; 875 break; 876 877 case AtomicExpr::AO__atomic_compare_exchange: 878 case AtomicExpr::AO__atomic_compare_exchange_n: 879 Form = GNUCmpXchg; 880 break; 881 } 882 883 // Check we have the right number of arguments. 884 if (TheCall->getNumArgs() < NumArgs[Form]) { 885 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 886 << 0 << NumArgs[Form] << TheCall->getNumArgs() 887 << TheCall->getCallee()->getSourceRange(); 888 return ExprError(); 889 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 890 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 891 diag::err_typecheck_call_too_many_args) 892 << 0 << NumArgs[Form] << TheCall->getNumArgs() 893 << TheCall->getCallee()->getSourceRange(); 894 return ExprError(); 895 } 896 897 // Inspect the first argument of the atomic operation. 898 Expr *Ptr = TheCall->getArg(0); 899 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 900 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 901 if (!pointerType) { 902 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 903 << Ptr->getType() << Ptr->getSourceRange(); 904 return ExprError(); 905 } 906 907 // For a __c11 builtin, this should be a pointer to an _Atomic type. 908 QualType AtomTy = pointerType->getPointeeType(); // 'A' 909 QualType ValType = AtomTy; // 'C' 910 if (IsC11) { 911 if (!AtomTy->isAtomicType()) { 912 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 913 << Ptr->getType() << Ptr->getSourceRange(); 914 return ExprError(); 915 } 916 if (AtomTy.isConstQualified()) { 917 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_non_const_atomic) 918 << Ptr->getType() << Ptr->getSourceRange(); 919 return ExprError(); 920 } 921 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 922 } 923 924 // For an arithmetic operation, the implied arithmetic must be well-formed. 925 if (Form == Arithmetic) { 926 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 927 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 928 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 929 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 930 return ExprError(); 931 } 932 if (!IsAddSub && !ValType->isIntegerType()) { 933 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 934 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 935 return ExprError(); 936 } 937 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 938 // For __atomic_*_n operations, the value type must be a scalar integral or 939 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 940 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 941 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 942 return ExprError(); 943 } 944 945 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context)) { 946 // For GNU atomics, require a trivially-copyable type. This is not part of 947 // the GNU atomics specification, but we enforce it for sanity. 948 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 949 << Ptr->getType() << Ptr->getSourceRange(); 950 return ExprError(); 951 } 952 953 // FIXME: For any builtin other than a load, the ValType must not be 954 // const-qualified. 955 956 switch (ValType.getObjCLifetime()) { 957 case Qualifiers::OCL_None: 958 case Qualifiers::OCL_ExplicitNone: 959 // okay 960 break; 961 962 case Qualifiers::OCL_Weak: 963 case Qualifiers::OCL_Strong: 964 case Qualifiers::OCL_Autoreleasing: 965 // FIXME: Can this happen? By this point, ValType should be known 966 // to be trivially copyable. 967 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 968 << ValType << Ptr->getSourceRange(); 969 return ExprError(); 970 } 971 972 QualType ResultType = ValType; 973 if (Form == Copy || Form == GNUXchg || Form == Init) 974 ResultType = Context.VoidTy; 975 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 976 ResultType = Context.BoolTy; 977 978 // The type of a parameter passed 'by value'. In the GNU atomics, such 979 // arguments are actually passed as pointers. 980 QualType ByValType = ValType; // 'CP' 981 if (!IsC11 && !IsN) 982 ByValType = Ptr->getType(); 983 984 // The first argument --- the pointer --- has a fixed type; we 985 // deduce the types of the rest of the arguments accordingly. Walk 986 // the remaining arguments, converting them to the deduced value type. 987 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 988 QualType Ty; 989 if (i < NumVals[Form] + 1) { 990 switch (i) { 991 case 1: 992 // The second argument is the non-atomic operand. For arithmetic, this 993 // is always passed by value, and for a compare_exchange it is always 994 // passed by address. For the rest, GNU uses by-address and C11 uses 995 // by-value. 996 assert(Form != Load); 997 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 998 Ty = ValType; 999 else if (Form == Copy || Form == Xchg) 1000 Ty = ByValType; 1001 else if (Form == Arithmetic) 1002 Ty = Context.getPointerDiffType(); 1003 else 1004 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 1005 break; 1006 case 2: 1007 // The third argument to compare_exchange / GNU exchange is a 1008 // (pointer to a) desired value. 1009 Ty = ByValType; 1010 break; 1011 case 3: 1012 // The fourth argument to GNU compare_exchange is a 'weak' flag. 1013 Ty = Context.BoolTy; 1014 break; 1015 } 1016 } else { 1017 // The order(s) are always converted to int. 1018 Ty = Context.IntTy; 1019 } 1020 1021 InitializedEntity Entity = 1022 InitializedEntity::InitializeParameter(Context, Ty, false); 1023 ExprResult Arg = TheCall->getArg(i); 1024 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1025 if (Arg.isInvalid()) 1026 return true; 1027 TheCall->setArg(i, Arg.get()); 1028 } 1029 1030 // Permute the arguments into a 'consistent' order. 1031 SmallVector<Expr*, 5> SubExprs; 1032 SubExprs.push_back(Ptr); 1033 switch (Form) { 1034 case Init: 1035 // Note, AtomicExpr::getVal1() has a special case for this atomic. 1036 SubExprs.push_back(TheCall->getArg(1)); // Val1 1037 break; 1038 case Load: 1039 SubExprs.push_back(TheCall->getArg(1)); // Order 1040 break; 1041 case Copy: 1042 case Arithmetic: 1043 case Xchg: 1044 SubExprs.push_back(TheCall->getArg(2)); // Order 1045 SubExprs.push_back(TheCall->getArg(1)); // Val1 1046 break; 1047 case GNUXchg: 1048 // Note, AtomicExpr::getVal2() has a special case for this atomic. 1049 SubExprs.push_back(TheCall->getArg(3)); // Order 1050 SubExprs.push_back(TheCall->getArg(1)); // Val1 1051 SubExprs.push_back(TheCall->getArg(2)); // Val2 1052 break; 1053 case C11CmpXchg: 1054 SubExprs.push_back(TheCall->getArg(3)); // Order 1055 SubExprs.push_back(TheCall->getArg(1)); // Val1 1056 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 1057 SubExprs.push_back(TheCall->getArg(2)); // Val2 1058 break; 1059 case GNUCmpXchg: 1060 SubExprs.push_back(TheCall->getArg(4)); // Order 1061 SubExprs.push_back(TheCall->getArg(1)); // Val1 1062 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 1063 SubExprs.push_back(TheCall->getArg(2)); // Val2 1064 SubExprs.push_back(TheCall->getArg(3)); // Weak 1065 break; 1066 } 1067 1068 AtomicExpr *AE = new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 1069 SubExprs, ResultType, Op, 1070 TheCall->getRParenLoc()); 1071 1072 if ((Op == AtomicExpr::AO__c11_atomic_load || 1073 (Op == AtomicExpr::AO__c11_atomic_store)) && 1074 Context.AtomicUsesUnsupportedLibcall(AE)) 1075 Diag(AE->getLocStart(), diag::err_atomic_load_store_uses_lib) << 1076 ((Op == AtomicExpr::AO__c11_atomic_load) ? 0 : 1); 1077 1078 return Owned(AE); 1079} 1080 1081 1082/// checkBuiltinArgument - Given a call to a builtin function, perform 1083/// normal type-checking on the given argument, updating the call in 1084/// place. This is useful when a builtin function requires custom 1085/// type-checking for some of its arguments but not necessarily all of 1086/// them. 1087/// 1088/// Returns true on error. 1089static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 1090 FunctionDecl *Fn = E->getDirectCallee(); 1091 assert(Fn && "builtin call without direct callee!"); 1092 1093 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 1094 InitializedEntity Entity = 1095 InitializedEntity::InitializeParameter(S.Context, Param); 1096 1097 ExprResult Arg = E->getArg(0); 1098 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 1099 if (Arg.isInvalid()) 1100 return true; 1101 1102 E->setArg(ArgIndex, Arg.take()); 1103 return false; 1104} 1105 1106/// SemaBuiltinAtomicOverloaded - We have a call to a function like 1107/// __sync_fetch_and_add, which is an overloaded function based on the pointer 1108/// type of its first argument. The main ActOnCallExpr routines have already 1109/// promoted the types of arguments because all of these calls are prototyped as 1110/// void(...). 1111/// 1112/// This function goes through and does final semantic checking for these 1113/// builtins, 1114ExprResult 1115Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 1116 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 1117 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1118 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1119 1120 // Ensure that we have at least one argument to do type inference from. 1121 if (TheCall->getNumArgs() < 1) { 1122 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1123 << 0 << 1 << TheCall->getNumArgs() 1124 << TheCall->getCallee()->getSourceRange(); 1125 return ExprError(); 1126 } 1127 1128 // Inspect the first argument of the atomic builtin. This should always be 1129 // a pointer type, whose element is an integral scalar or pointer type. 1130 // Because it is a pointer type, we don't have to worry about any implicit 1131 // casts here. 1132 // FIXME: We don't allow floating point scalars as input. 1133 Expr *FirstArg = TheCall->getArg(0); 1134 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 1135 if (FirstArgResult.isInvalid()) 1136 return ExprError(); 1137 FirstArg = FirstArgResult.take(); 1138 TheCall->setArg(0, FirstArg); 1139 1140 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 1141 if (!pointerType) { 1142 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 1143 << FirstArg->getType() << FirstArg->getSourceRange(); 1144 return ExprError(); 1145 } 1146 1147 QualType ValType = pointerType->getPointeeType(); 1148 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 1149 !ValType->isBlockPointerType()) { 1150 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 1151 << FirstArg->getType() << FirstArg->getSourceRange(); 1152 return ExprError(); 1153 } 1154 1155 switch (ValType.getObjCLifetime()) { 1156 case Qualifiers::OCL_None: 1157 case Qualifiers::OCL_ExplicitNone: 1158 // okay 1159 break; 1160 1161 case Qualifiers::OCL_Weak: 1162 case Qualifiers::OCL_Strong: 1163 case Qualifiers::OCL_Autoreleasing: 1164 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1165 << ValType << FirstArg->getSourceRange(); 1166 return ExprError(); 1167 } 1168 1169 // Strip any qualifiers off ValType. 1170 ValType = ValType.getUnqualifiedType(); 1171 1172 // The majority of builtins return a value, but a few have special return 1173 // types, so allow them to override appropriately below. 1174 QualType ResultType = ValType; 1175 1176 // We need to figure out which concrete builtin this maps onto. For example, 1177 // __sync_fetch_and_add with a 2 byte object turns into 1178 // __sync_fetch_and_add_2. 1179#define BUILTIN_ROW(x) \ 1180 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 1181 Builtin::BI##x##_8, Builtin::BI##x##_16 } 1182 1183 static const unsigned BuiltinIndices[][5] = { 1184 BUILTIN_ROW(__sync_fetch_and_add), 1185 BUILTIN_ROW(__sync_fetch_and_sub), 1186 BUILTIN_ROW(__sync_fetch_and_or), 1187 BUILTIN_ROW(__sync_fetch_and_and), 1188 BUILTIN_ROW(__sync_fetch_and_xor), 1189 1190 BUILTIN_ROW(__sync_add_and_fetch), 1191 BUILTIN_ROW(__sync_sub_and_fetch), 1192 BUILTIN_ROW(__sync_and_and_fetch), 1193 BUILTIN_ROW(__sync_or_and_fetch), 1194 BUILTIN_ROW(__sync_xor_and_fetch), 1195 1196 BUILTIN_ROW(__sync_val_compare_and_swap), 1197 BUILTIN_ROW(__sync_bool_compare_and_swap), 1198 BUILTIN_ROW(__sync_lock_test_and_set), 1199 BUILTIN_ROW(__sync_lock_release), 1200 BUILTIN_ROW(__sync_swap) 1201 }; 1202#undef BUILTIN_ROW 1203 1204 // Determine the index of the size. 1205 unsigned SizeIndex; 1206 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 1207 case 1: SizeIndex = 0; break; 1208 case 2: SizeIndex = 1; break; 1209 case 4: SizeIndex = 2; break; 1210 case 8: SizeIndex = 3; break; 1211 case 16: SizeIndex = 4; break; 1212 default: 1213 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 1214 << FirstArg->getType() << FirstArg->getSourceRange(); 1215 return ExprError(); 1216 } 1217 1218 // Each of these builtins has one pointer argument, followed by some number of 1219 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 1220 // that we ignore. Find out which row of BuiltinIndices to read from as well 1221 // as the number of fixed args. 1222 unsigned BuiltinID = FDecl->getBuiltinID(); 1223 unsigned BuiltinIndex, NumFixed = 1; 1224 switch (BuiltinID) { 1225 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 1226 case Builtin::BI__sync_fetch_and_add: 1227 case Builtin::BI__sync_fetch_and_add_1: 1228 case Builtin::BI__sync_fetch_and_add_2: 1229 case Builtin::BI__sync_fetch_and_add_4: 1230 case Builtin::BI__sync_fetch_and_add_8: 1231 case Builtin::BI__sync_fetch_and_add_16: 1232 BuiltinIndex = 0; 1233 break; 1234 1235 case Builtin::BI__sync_fetch_and_sub: 1236 case Builtin::BI__sync_fetch_and_sub_1: 1237 case Builtin::BI__sync_fetch_and_sub_2: 1238 case Builtin::BI__sync_fetch_and_sub_4: 1239 case Builtin::BI__sync_fetch_and_sub_8: 1240 case Builtin::BI__sync_fetch_and_sub_16: 1241 BuiltinIndex = 1; 1242 break; 1243 1244 case Builtin::BI__sync_fetch_and_or: 1245 case Builtin::BI__sync_fetch_and_or_1: 1246 case Builtin::BI__sync_fetch_and_or_2: 1247 case Builtin::BI__sync_fetch_and_or_4: 1248 case Builtin::BI__sync_fetch_and_or_8: 1249 case Builtin::BI__sync_fetch_and_or_16: 1250 BuiltinIndex = 2; 1251 break; 1252 1253 case Builtin::BI__sync_fetch_and_and: 1254 case Builtin::BI__sync_fetch_and_and_1: 1255 case Builtin::BI__sync_fetch_and_and_2: 1256 case Builtin::BI__sync_fetch_and_and_4: 1257 case Builtin::BI__sync_fetch_and_and_8: 1258 case Builtin::BI__sync_fetch_and_and_16: 1259 BuiltinIndex = 3; 1260 break; 1261 1262 case Builtin::BI__sync_fetch_and_xor: 1263 case Builtin::BI__sync_fetch_and_xor_1: 1264 case Builtin::BI__sync_fetch_and_xor_2: 1265 case Builtin::BI__sync_fetch_and_xor_4: 1266 case Builtin::BI__sync_fetch_and_xor_8: 1267 case Builtin::BI__sync_fetch_and_xor_16: 1268 BuiltinIndex = 4; 1269 break; 1270 1271 case Builtin::BI__sync_add_and_fetch: 1272 case Builtin::BI__sync_add_and_fetch_1: 1273 case Builtin::BI__sync_add_and_fetch_2: 1274 case Builtin::BI__sync_add_and_fetch_4: 1275 case Builtin::BI__sync_add_and_fetch_8: 1276 case Builtin::BI__sync_add_and_fetch_16: 1277 BuiltinIndex = 5; 1278 break; 1279 1280 case Builtin::BI__sync_sub_and_fetch: 1281 case Builtin::BI__sync_sub_and_fetch_1: 1282 case Builtin::BI__sync_sub_and_fetch_2: 1283 case Builtin::BI__sync_sub_and_fetch_4: 1284 case Builtin::BI__sync_sub_and_fetch_8: 1285 case Builtin::BI__sync_sub_and_fetch_16: 1286 BuiltinIndex = 6; 1287 break; 1288 1289 case Builtin::BI__sync_and_and_fetch: 1290 case Builtin::BI__sync_and_and_fetch_1: 1291 case Builtin::BI__sync_and_and_fetch_2: 1292 case Builtin::BI__sync_and_and_fetch_4: 1293 case Builtin::BI__sync_and_and_fetch_8: 1294 case Builtin::BI__sync_and_and_fetch_16: 1295 BuiltinIndex = 7; 1296 break; 1297 1298 case Builtin::BI__sync_or_and_fetch: 1299 case Builtin::BI__sync_or_and_fetch_1: 1300 case Builtin::BI__sync_or_and_fetch_2: 1301 case Builtin::BI__sync_or_and_fetch_4: 1302 case Builtin::BI__sync_or_and_fetch_8: 1303 case Builtin::BI__sync_or_and_fetch_16: 1304 BuiltinIndex = 8; 1305 break; 1306 1307 case Builtin::BI__sync_xor_and_fetch: 1308 case Builtin::BI__sync_xor_and_fetch_1: 1309 case Builtin::BI__sync_xor_and_fetch_2: 1310 case Builtin::BI__sync_xor_and_fetch_4: 1311 case Builtin::BI__sync_xor_and_fetch_8: 1312 case Builtin::BI__sync_xor_and_fetch_16: 1313 BuiltinIndex = 9; 1314 break; 1315 1316 case Builtin::BI__sync_val_compare_and_swap: 1317 case Builtin::BI__sync_val_compare_and_swap_1: 1318 case Builtin::BI__sync_val_compare_and_swap_2: 1319 case Builtin::BI__sync_val_compare_and_swap_4: 1320 case Builtin::BI__sync_val_compare_and_swap_8: 1321 case Builtin::BI__sync_val_compare_and_swap_16: 1322 BuiltinIndex = 10; 1323 NumFixed = 2; 1324 break; 1325 1326 case Builtin::BI__sync_bool_compare_and_swap: 1327 case Builtin::BI__sync_bool_compare_and_swap_1: 1328 case Builtin::BI__sync_bool_compare_and_swap_2: 1329 case Builtin::BI__sync_bool_compare_and_swap_4: 1330 case Builtin::BI__sync_bool_compare_and_swap_8: 1331 case Builtin::BI__sync_bool_compare_and_swap_16: 1332 BuiltinIndex = 11; 1333 NumFixed = 2; 1334 ResultType = Context.BoolTy; 1335 break; 1336 1337 case Builtin::BI__sync_lock_test_and_set: 1338 case Builtin::BI__sync_lock_test_and_set_1: 1339 case Builtin::BI__sync_lock_test_and_set_2: 1340 case Builtin::BI__sync_lock_test_and_set_4: 1341 case Builtin::BI__sync_lock_test_and_set_8: 1342 case Builtin::BI__sync_lock_test_and_set_16: 1343 BuiltinIndex = 12; 1344 break; 1345 1346 case Builtin::BI__sync_lock_release: 1347 case Builtin::BI__sync_lock_release_1: 1348 case Builtin::BI__sync_lock_release_2: 1349 case Builtin::BI__sync_lock_release_4: 1350 case Builtin::BI__sync_lock_release_8: 1351 case Builtin::BI__sync_lock_release_16: 1352 BuiltinIndex = 13; 1353 NumFixed = 0; 1354 ResultType = Context.VoidTy; 1355 break; 1356 1357 case Builtin::BI__sync_swap: 1358 case Builtin::BI__sync_swap_1: 1359 case Builtin::BI__sync_swap_2: 1360 case Builtin::BI__sync_swap_4: 1361 case Builtin::BI__sync_swap_8: 1362 case Builtin::BI__sync_swap_16: 1363 BuiltinIndex = 14; 1364 break; 1365 } 1366 1367 // Now that we know how many fixed arguments we expect, first check that we 1368 // have at least that many. 1369 if (TheCall->getNumArgs() < 1+NumFixed) { 1370 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1371 << 0 << 1+NumFixed << TheCall->getNumArgs() 1372 << TheCall->getCallee()->getSourceRange(); 1373 return ExprError(); 1374 } 1375 1376 // Get the decl for the concrete builtin from this, we can tell what the 1377 // concrete integer type we should convert to is. 1378 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1379 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1380 FunctionDecl *NewBuiltinDecl; 1381 if (NewBuiltinID == BuiltinID) 1382 NewBuiltinDecl = FDecl; 1383 else { 1384 // Perform builtin lookup to avoid redeclaring it. 1385 DeclarationName DN(&Context.Idents.get(NewBuiltinName)); 1386 LookupResult Res(*this, DN, DRE->getLocStart(), LookupOrdinaryName); 1387 LookupName(Res, TUScope, /*AllowBuiltinCreation=*/true); 1388 assert(Res.getFoundDecl()); 1389 NewBuiltinDecl = dyn_cast<FunctionDecl>(Res.getFoundDecl()); 1390 if (NewBuiltinDecl == 0) 1391 return ExprError(); 1392 } 1393 1394 // The first argument --- the pointer --- has a fixed type; we 1395 // deduce the types of the rest of the arguments accordingly. Walk 1396 // the remaining arguments, converting them to the deduced value type. 1397 for (unsigned i = 0; i != NumFixed; ++i) { 1398 ExprResult Arg = TheCall->getArg(i+1); 1399 1400 // GCC does an implicit conversion to the pointer or integer ValType. This 1401 // can fail in some cases (1i -> int**), check for this error case now. 1402 // Initialize the argument. 1403 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1404 ValType, /*consume*/ false); 1405 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1406 if (Arg.isInvalid()) 1407 return ExprError(); 1408 1409 // Okay, we have something that *can* be converted to the right type. Check 1410 // to see if there is a potentially weird extension going on here. This can 1411 // happen when you do an atomic operation on something like an char* and 1412 // pass in 42. The 42 gets converted to char. This is even more strange 1413 // for things like 45.123 -> char, etc. 1414 // FIXME: Do this check. 1415 TheCall->setArg(i+1, Arg.take()); 1416 } 1417 1418 ASTContext& Context = this->getASTContext(); 1419 1420 // Create a new DeclRefExpr to refer to the new decl. 1421 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1422 Context, 1423 DRE->getQualifierLoc(), 1424 SourceLocation(), 1425 NewBuiltinDecl, 1426 /*enclosing*/ false, 1427 DRE->getLocation(), 1428 Context.BuiltinFnTy, 1429 DRE->getValueKind()); 1430 1431 // Set the callee in the CallExpr. 1432 // FIXME: This loses syntactic information. 1433 QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType()); 1434 ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy, 1435 CK_BuiltinFnToFnPtr); 1436 TheCall->setCallee(PromotedCall.take()); 1437 1438 // Change the result type of the call to match the original value type. This 1439 // is arbitrary, but the codegen for these builtins ins design to handle it 1440 // gracefully. 1441 TheCall->setType(ResultType); 1442 1443 return TheCallResult; 1444} 1445 1446/// CheckObjCString - Checks that the argument to the builtin 1447/// CFString constructor is correct 1448/// Note: It might also make sense to do the UTF-16 conversion here (would 1449/// simplify the backend). 1450bool Sema::CheckObjCString(Expr *Arg) { 1451 Arg = Arg->IgnoreParenCasts(); 1452 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1453 1454 if (!Literal || !Literal->isAscii()) { 1455 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1456 << Arg->getSourceRange(); 1457 return true; 1458 } 1459 1460 if (Literal->containsNonAsciiOrNull()) { 1461 StringRef String = Literal->getString(); 1462 unsigned NumBytes = String.size(); 1463 SmallVector<UTF16, 128> ToBuf(NumBytes); 1464 const UTF8 *FromPtr = (const UTF8 *)String.data(); 1465 UTF16 *ToPtr = &ToBuf[0]; 1466 1467 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1468 &ToPtr, ToPtr + NumBytes, 1469 strictConversion); 1470 // Check for conversion failure. 1471 if (Result != conversionOK) 1472 Diag(Arg->getLocStart(), 1473 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1474 } 1475 return false; 1476} 1477 1478/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1479/// Emit an error and return true on failure, return false on success. 1480bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1481 Expr *Fn = TheCall->getCallee(); 1482 if (TheCall->getNumArgs() > 2) { 1483 Diag(TheCall->getArg(2)->getLocStart(), 1484 diag::err_typecheck_call_too_many_args) 1485 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1486 << Fn->getSourceRange() 1487 << SourceRange(TheCall->getArg(2)->getLocStart(), 1488 (*(TheCall->arg_end()-1))->getLocEnd()); 1489 return true; 1490 } 1491 1492 if (TheCall->getNumArgs() < 2) { 1493 return Diag(TheCall->getLocEnd(), 1494 diag::err_typecheck_call_too_few_args_at_least) 1495 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1496 } 1497 1498 // Type-check the first argument normally. 1499 if (checkBuiltinArgument(*this, TheCall, 0)) 1500 return true; 1501 1502 // Determine whether the current function is variadic or not. 1503 BlockScopeInfo *CurBlock = getCurBlock(); 1504 bool isVariadic; 1505 if (CurBlock) 1506 isVariadic = CurBlock->TheDecl->isVariadic(); 1507 else if (FunctionDecl *FD = getCurFunctionDecl()) 1508 isVariadic = FD->isVariadic(); 1509 else 1510 isVariadic = getCurMethodDecl()->isVariadic(); 1511 1512 if (!isVariadic) { 1513 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1514 return true; 1515 } 1516 1517 // Verify that the second argument to the builtin is the last argument of the 1518 // current function or method. 1519 bool SecondArgIsLastNamedArgument = false; 1520 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1521 1522 // These are valid if SecondArgIsLastNamedArgument is false after the next 1523 // block. 1524 QualType Type; 1525 SourceLocation ParamLoc; 1526 1527 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1528 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1529 // FIXME: This isn't correct for methods (results in bogus warning). 1530 // Get the last formal in the current function. 1531 const ParmVarDecl *LastArg; 1532 if (CurBlock) 1533 LastArg = *(CurBlock->TheDecl->param_end()-1); 1534 else if (FunctionDecl *FD = getCurFunctionDecl()) 1535 LastArg = *(FD->param_end()-1); 1536 else 1537 LastArg = *(getCurMethodDecl()->param_end()-1); 1538 SecondArgIsLastNamedArgument = PV == LastArg; 1539 1540 Type = PV->getType(); 1541 ParamLoc = PV->getLocation(); 1542 } 1543 } 1544 1545 if (!SecondArgIsLastNamedArgument) 1546 Diag(TheCall->getArg(1)->getLocStart(), 1547 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1548 else if (Type->isReferenceType()) { 1549 Diag(Arg->getLocStart(), 1550 diag::warn_va_start_of_reference_type_is_undefined); 1551 Diag(ParamLoc, diag::note_parameter_type) << Type; 1552 } 1553 1554 return false; 1555} 1556 1557/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1558/// friends. This is declared to take (...), so we have to check everything. 1559bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1560 if (TheCall->getNumArgs() < 2) 1561 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1562 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1563 if (TheCall->getNumArgs() > 2) 1564 return Diag(TheCall->getArg(2)->getLocStart(), 1565 diag::err_typecheck_call_too_many_args) 1566 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1567 << SourceRange(TheCall->getArg(2)->getLocStart(), 1568 (*(TheCall->arg_end()-1))->getLocEnd()); 1569 1570 ExprResult OrigArg0 = TheCall->getArg(0); 1571 ExprResult OrigArg1 = TheCall->getArg(1); 1572 1573 // Do standard promotions between the two arguments, returning their common 1574 // type. 1575 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1576 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1577 return true; 1578 1579 // Make sure any conversions are pushed back into the call; this is 1580 // type safe since unordered compare builtins are declared as "_Bool 1581 // foo(...)". 1582 TheCall->setArg(0, OrigArg0.get()); 1583 TheCall->setArg(1, OrigArg1.get()); 1584 1585 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1586 return false; 1587 1588 // If the common type isn't a real floating type, then the arguments were 1589 // invalid for this operation. 1590 if (Res.isNull() || !Res->isRealFloatingType()) 1591 return Diag(OrigArg0.get()->getLocStart(), 1592 diag::err_typecheck_call_invalid_ordered_compare) 1593 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1594 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1595 1596 return false; 1597} 1598 1599/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1600/// __builtin_isnan and friends. This is declared to take (...), so we have 1601/// to check everything. We expect the last argument to be a floating point 1602/// value. 1603bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1604 if (TheCall->getNumArgs() < NumArgs) 1605 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1606 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1607 if (TheCall->getNumArgs() > NumArgs) 1608 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1609 diag::err_typecheck_call_too_many_args) 1610 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1611 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1612 (*(TheCall->arg_end()-1))->getLocEnd()); 1613 1614 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1615 1616 if (OrigArg->isTypeDependent()) 1617 return false; 1618 1619 // This operation requires a non-_Complex floating-point number. 1620 if (!OrigArg->getType()->isRealFloatingType()) 1621 return Diag(OrigArg->getLocStart(), 1622 diag::err_typecheck_call_invalid_unary_fp) 1623 << OrigArg->getType() << OrigArg->getSourceRange(); 1624 1625 // If this is an implicit conversion from float -> double, remove it. 1626 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1627 Expr *CastArg = Cast->getSubExpr(); 1628 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1629 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1630 "promotion from float to double is the only expected cast here"); 1631 Cast->setSubExpr(0); 1632 TheCall->setArg(NumArgs-1, CastArg); 1633 } 1634 } 1635 1636 return false; 1637} 1638 1639/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1640// This is declared to take (...), so we have to check everything. 1641ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1642 if (TheCall->getNumArgs() < 2) 1643 return ExprError(Diag(TheCall->getLocEnd(), 1644 diag::err_typecheck_call_too_few_args_at_least) 1645 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1646 << TheCall->getSourceRange()); 1647 1648 // Determine which of the following types of shufflevector we're checking: 1649 // 1) unary, vector mask: (lhs, mask) 1650 // 2) binary, vector mask: (lhs, rhs, mask) 1651 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1652 QualType resType = TheCall->getArg(0)->getType(); 1653 unsigned numElements = 0; 1654 1655 if (!TheCall->getArg(0)->isTypeDependent() && 1656 !TheCall->getArg(1)->isTypeDependent()) { 1657 QualType LHSType = TheCall->getArg(0)->getType(); 1658 QualType RHSType = TheCall->getArg(1)->getType(); 1659 1660 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1661 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1662 << SourceRange(TheCall->getArg(0)->getLocStart(), 1663 TheCall->getArg(1)->getLocEnd()); 1664 return ExprError(); 1665 } 1666 1667 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1668 unsigned numResElements = TheCall->getNumArgs() - 2; 1669 1670 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1671 // with mask. If so, verify that RHS is an integer vector type with the 1672 // same number of elts as lhs. 1673 if (TheCall->getNumArgs() == 2) { 1674 if (!RHSType->hasIntegerRepresentation() || 1675 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1676 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1677 << SourceRange(TheCall->getArg(1)->getLocStart(), 1678 TheCall->getArg(1)->getLocEnd()); 1679 } else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1680 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1681 << SourceRange(TheCall->getArg(0)->getLocStart(), 1682 TheCall->getArg(1)->getLocEnd()); 1683 return ExprError(); 1684 } else if (numElements != numResElements) { 1685 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1686 resType = Context.getVectorType(eltType, numResElements, 1687 VectorType::GenericVector); 1688 } 1689 } 1690 1691 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1692 if (TheCall->getArg(i)->isTypeDependent() || 1693 TheCall->getArg(i)->isValueDependent()) 1694 continue; 1695 1696 llvm::APSInt Result(32); 1697 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1698 return ExprError(Diag(TheCall->getLocStart(), 1699 diag::err_shufflevector_nonconstant_argument) 1700 << TheCall->getArg(i)->getSourceRange()); 1701 1702 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1703 return ExprError(Diag(TheCall->getLocStart(), 1704 diag::err_shufflevector_argument_too_large) 1705 << TheCall->getArg(i)->getSourceRange()); 1706 } 1707 1708 SmallVector<Expr*, 32> exprs; 1709 1710 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1711 exprs.push_back(TheCall->getArg(i)); 1712 TheCall->setArg(i, 0); 1713 } 1714 1715 return Owned(new (Context) ShuffleVectorExpr(Context, exprs, resType, 1716 TheCall->getCallee()->getLocStart(), 1717 TheCall->getRParenLoc())); 1718} 1719 1720/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1721// This is declared to take (const void*, ...) and can take two 1722// optional constant int args. 1723bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1724 unsigned NumArgs = TheCall->getNumArgs(); 1725 1726 if (NumArgs > 3) 1727 return Diag(TheCall->getLocEnd(), 1728 diag::err_typecheck_call_too_many_args_at_most) 1729 << 0 /*function call*/ << 3 << NumArgs 1730 << TheCall->getSourceRange(); 1731 1732 // Argument 0 is checked for us and the remaining arguments must be 1733 // constant integers. 1734 for (unsigned i = 1; i != NumArgs; ++i) { 1735 Expr *Arg = TheCall->getArg(i); 1736 1737 // We can't check the value of a dependent argument. 1738 if (Arg->isTypeDependent() || Arg->isValueDependent()) 1739 continue; 1740 1741 llvm::APSInt Result; 1742 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1743 return true; 1744 1745 // FIXME: gcc issues a warning and rewrites these to 0. These 1746 // seems especially odd for the third argument since the default 1747 // is 3. 1748 if (i == 1) { 1749 if (Result.getLimitedValue() > 1) 1750 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1751 << "0" << "1" << Arg->getSourceRange(); 1752 } else { 1753 if (Result.getLimitedValue() > 3) 1754 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1755 << "0" << "3" << Arg->getSourceRange(); 1756 } 1757 } 1758 1759 return false; 1760} 1761 1762/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1763/// TheCall is a constant expression. 1764bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1765 llvm::APSInt &Result) { 1766 Expr *Arg = TheCall->getArg(ArgNum); 1767 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1768 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1769 1770 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1771 1772 if (!Arg->isIntegerConstantExpr(Result, Context)) 1773 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1774 << FDecl->getDeclName() << Arg->getSourceRange(); 1775 1776 return false; 1777} 1778 1779/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1780/// int type). This simply type checks that type is one of the defined 1781/// constants (0-3). 1782// For compatibility check 0-3, llvm only handles 0 and 2. 1783bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1784 llvm::APSInt Result; 1785 1786 // We can't check the value of a dependent argument. 1787 if (TheCall->getArg(1)->isTypeDependent() || 1788 TheCall->getArg(1)->isValueDependent()) 1789 return false; 1790 1791 // Check constant-ness first. 1792 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1793 return true; 1794 1795 Expr *Arg = TheCall->getArg(1); 1796 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1797 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1798 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1799 } 1800 1801 return false; 1802} 1803 1804/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1805/// This checks that val is a constant 1. 1806bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1807 Expr *Arg = TheCall->getArg(1); 1808 llvm::APSInt Result; 1809 1810 // TODO: This is less than ideal. Overload this to take a value. 1811 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1812 return true; 1813 1814 if (Result != 1) 1815 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1816 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1817 1818 return false; 1819} 1820 1821// Determine if an expression is a string literal or constant string. 1822// If this function returns false on the arguments to a function expecting a 1823// format string, we will usually need to emit a warning. 1824// True string literals are then checked by CheckFormatString. 1825Sema::StringLiteralCheckType 1826Sema::checkFormatStringExpr(const Expr *E, ArrayRef<const Expr *> Args, 1827 bool HasVAListArg, 1828 unsigned format_idx, unsigned firstDataArg, 1829 FormatStringType Type, VariadicCallType CallType, 1830 bool inFunctionCall) { 1831 tryAgain: 1832 if (E->isTypeDependent() || E->isValueDependent()) 1833 return SLCT_NotALiteral; 1834 1835 E = E->IgnoreParenCasts(); 1836 1837 if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull)) 1838 // Technically -Wformat-nonliteral does not warn about this case. 1839 // The behavior of printf and friends in this case is implementation 1840 // dependent. Ideally if the format string cannot be null then 1841 // it should have a 'nonnull' attribute in the function prototype. 1842 return SLCT_CheckedLiteral; 1843 1844 switch (E->getStmtClass()) { 1845 case Stmt::BinaryConditionalOperatorClass: 1846 case Stmt::ConditionalOperatorClass: { 1847 // The expression is a literal if both sub-expressions were, and it was 1848 // completely checked only if both sub-expressions were checked. 1849 const AbstractConditionalOperator *C = 1850 cast<AbstractConditionalOperator>(E); 1851 StringLiteralCheckType Left = 1852 checkFormatStringExpr(C->getTrueExpr(), Args, 1853 HasVAListArg, format_idx, firstDataArg, 1854 Type, CallType, inFunctionCall); 1855 if (Left == SLCT_NotALiteral) 1856 return SLCT_NotALiteral; 1857 StringLiteralCheckType Right = 1858 checkFormatStringExpr(C->getFalseExpr(), Args, 1859 HasVAListArg, format_idx, firstDataArg, 1860 Type, CallType, inFunctionCall); 1861 return Left < Right ? Left : Right; 1862 } 1863 1864 case Stmt::ImplicitCastExprClass: { 1865 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1866 goto tryAgain; 1867 } 1868 1869 case Stmt::OpaqueValueExprClass: 1870 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1871 E = src; 1872 goto tryAgain; 1873 } 1874 return SLCT_NotALiteral; 1875 1876 case Stmt::PredefinedExprClass: 1877 // While __func__, etc., are technically not string literals, they 1878 // cannot contain format specifiers and thus are not a security 1879 // liability. 1880 return SLCT_UncheckedLiteral; 1881 1882 case Stmt::DeclRefExprClass: { 1883 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1884 1885 // As an exception, do not flag errors for variables binding to 1886 // const string literals. 1887 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1888 bool isConstant = false; 1889 QualType T = DR->getType(); 1890 1891 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1892 isConstant = AT->getElementType().isConstant(Context); 1893 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1894 isConstant = T.isConstant(Context) && 1895 PT->getPointeeType().isConstant(Context); 1896 } else if (T->isObjCObjectPointerType()) { 1897 // In ObjC, there is usually no "const ObjectPointer" type, 1898 // so don't check if the pointee type is constant. 1899 isConstant = T.isConstant(Context); 1900 } 1901 1902 if (isConstant) { 1903 if (const Expr *Init = VD->getAnyInitializer()) { 1904 // Look through initializers like const char c[] = { "foo" } 1905 if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) { 1906 if (InitList->isStringLiteralInit()) 1907 Init = InitList->getInit(0)->IgnoreParenImpCasts(); 1908 } 1909 return checkFormatStringExpr(Init, Args, 1910 HasVAListArg, format_idx, 1911 firstDataArg, Type, CallType, 1912 /*inFunctionCall*/false); 1913 } 1914 } 1915 1916 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1917 // special check to see if the format string is a function parameter 1918 // of the function calling the printf function. If the function 1919 // has an attribute indicating it is a printf-like function, then we 1920 // should suppress warnings concerning non-literals being used in a call 1921 // to a vprintf function. For example: 1922 // 1923 // void 1924 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1925 // va_list ap; 1926 // va_start(ap, fmt); 1927 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1928 // ... 1929 // 1930 if (HasVAListArg) { 1931 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 1932 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 1933 int PVIndex = PV->getFunctionScopeIndex() + 1; 1934 for (specific_attr_iterator<FormatAttr> 1935 i = ND->specific_attr_begin<FormatAttr>(), 1936 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 1937 FormatAttr *PVFormat = *i; 1938 // adjust for implicit parameter 1939 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1940 if (MD->isInstance()) 1941 ++PVIndex; 1942 // We also check if the formats are compatible. 1943 // We can't pass a 'scanf' string to a 'printf' function. 1944 if (PVIndex == PVFormat->getFormatIdx() && 1945 Type == GetFormatStringType(PVFormat)) 1946 return SLCT_UncheckedLiteral; 1947 } 1948 } 1949 } 1950 } 1951 } 1952 1953 return SLCT_NotALiteral; 1954 } 1955 1956 case Stmt::CallExprClass: 1957 case Stmt::CXXMemberCallExprClass: { 1958 const CallExpr *CE = cast<CallExpr>(E); 1959 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 1960 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 1961 unsigned ArgIndex = FA->getFormatIdx(); 1962 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1963 if (MD->isInstance()) 1964 --ArgIndex; 1965 const Expr *Arg = CE->getArg(ArgIndex - 1); 1966 1967 return checkFormatStringExpr(Arg, Args, 1968 HasVAListArg, format_idx, firstDataArg, 1969 Type, CallType, inFunctionCall); 1970 } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) { 1971 unsigned BuiltinID = FD->getBuiltinID(); 1972 if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString || 1973 BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) { 1974 const Expr *Arg = CE->getArg(0); 1975 return checkFormatStringExpr(Arg, Args, 1976 HasVAListArg, format_idx, 1977 firstDataArg, Type, CallType, 1978 inFunctionCall); 1979 } 1980 } 1981 } 1982 1983 return SLCT_NotALiteral; 1984 } 1985 case Stmt::ObjCStringLiteralClass: 1986 case Stmt::StringLiteralClass: { 1987 const StringLiteral *StrE = NULL; 1988 1989 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1990 StrE = ObjCFExpr->getString(); 1991 else 1992 StrE = cast<StringLiteral>(E); 1993 1994 if (StrE) { 1995 CheckFormatString(StrE, E, Args, HasVAListArg, format_idx, 1996 firstDataArg, Type, inFunctionCall, CallType); 1997 return SLCT_CheckedLiteral; 1998 } 1999 2000 return SLCT_NotALiteral; 2001 } 2002 2003 default: 2004 return SLCT_NotALiteral; 2005 } 2006} 2007 2008void 2009Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 2010 const Expr * const *ExprArgs, 2011 SourceLocation CallSiteLoc) { 2012 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 2013 e = NonNull->args_end(); 2014 i != e; ++i) { 2015 const Expr *ArgExpr = ExprArgs[*i]; 2016 2017 // As a special case, transparent unions initialized with zero are 2018 // considered null for the purposes of the nonnull attribute. 2019 if (const RecordType *UT = ArgExpr->getType()->getAsUnionType()) { 2020 if (UT->getDecl()->hasAttr<TransparentUnionAttr>()) 2021 if (const CompoundLiteralExpr *CLE = 2022 dyn_cast<CompoundLiteralExpr>(ArgExpr)) 2023 if (const InitListExpr *ILE = 2024 dyn_cast<InitListExpr>(CLE->getInitializer())) 2025 ArgExpr = ILE->getInit(0); 2026 } 2027 2028 bool Result; 2029 if (ArgExpr->EvaluateAsBooleanCondition(Result, Context) && !Result) 2030 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 2031 } 2032} 2033 2034Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 2035 return llvm::StringSwitch<FormatStringType>(Format->getType()) 2036 .Case("scanf", FST_Scanf) 2037 .Cases("printf", "printf0", FST_Printf) 2038 .Cases("NSString", "CFString", FST_NSString) 2039 .Case("strftime", FST_Strftime) 2040 .Case("strfmon", FST_Strfmon) 2041 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 2042 .Default(FST_Unknown); 2043} 2044 2045/// CheckFormatArguments - Check calls to printf and scanf (and similar 2046/// functions) for correct use of format strings. 2047/// Returns true if a format string has been fully checked. 2048bool Sema::CheckFormatArguments(const FormatAttr *Format, 2049 ArrayRef<const Expr *> Args, 2050 bool IsCXXMember, 2051 VariadicCallType CallType, 2052 SourceLocation Loc, SourceRange Range) { 2053 FormatStringInfo FSI; 2054 if (getFormatStringInfo(Format, IsCXXMember, &FSI)) 2055 return CheckFormatArguments(Args, FSI.HasVAListArg, FSI.FormatIdx, 2056 FSI.FirstDataArg, GetFormatStringType(Format), 2057 CallType, Loc, Range); 2058 return false; 2059} 2060 2061bool Sema::CheckFormatArguments(ArrayRef<const Expr *> Args, 2062 bool HasVAListArg, unsigned format_idx, 2063 unsigned firstDataArg, FormatStringType Type, 2064 VariadicCallType CallType, 2065 SourceLocation Loc, SourceRange Range) { 2066 // CHECK: printf/scanf-like function is called with no format string. 2067 if (format_idx >= Args.size()) { 2068 Diag(Loc, diag::warn_missing_format_string) << Range; 2069 return false; 2070 } 2071 2072 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 2073 2074 // CHECK: format string is not a string literal. 2075 // 2076 // Dynamically generated format strings are difficult to 2077 // automatically vet at compile time. Requiring that format strings 2078 // are string literals: (1) permits the checking of format strings by 2079 // the compiler and thereby (2) can practically remove the source of 2080 // many format string exploits. 2081 2082 // Format string can be either ObjC string (e.g. @"%d") or 2083 // C string (e.g. "%d") 2084 // ObjC string uses the same format specifiers as C string, so we can use 2085 // the same format string checking logic for both ObjC and C strings. 2086 StringLiteralCheckType CT = 2087 checkFormatStringExpr(OrigFormatExpr, Args, HasVAListArg, 2088 format_idx, firstDataArg, Type, CallType); 2089 if (CT != SLCT_NotALiteral) 2090 // Literal format string found, check done! 2091 return CT == SLCT_CheckedLiteral; 2092 2093 // Strftime is particular as it always uses a single 'time' argument, 2094 // so it is safe to pass a non-literal string. 2095 if (Type == FST_Strftime) 2096 return false; 2097 2098 // Do not emit diag when the string param is a macro expansion and the 2099 // format is either NSString or CFString. This is a hack to prevent 2100 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 2101 // which are usually used in place of NS and CF string literals. 2102 if (Type == FST_NSString && 2103 SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart())) 2104 return false; 2105 2106 // If there are no arguments specified, warn with -Wformat-security, otherwise 2107 // warn only with -Wformat-nonliteral. 2108 if (Args.size() == firstDataArg) 2109 Diag(Args[format_idx]->getLocStart(), 2110 diag::warn_format_nonliteral_noargs) 2111 << OrigFormatExpr->getSourceRange(); 2112 else 2113 Diag(Args[format_idx]->getLocStart(), 2114 diag::warn_format_nonliteral) 2115 << OrigFormatExpr->getSourceRange(); 2116 return false; 2117} 2118 2119namespace { 2120class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 2121protected: 2122 Sema &S; 2123 const StringLiteral *FExpr; 2124 const Expr *OrigFormatExpr; 2125 const unsigned FirstDataArg; 2126 const unsigned NumDataArgs; 2127 const char *Beg; // Start of format string. 2128 const bool HasVAListArg; 2129 ArrayRef<const Expr *> Args; 2130 unsigned FormatIdx; 2131 llvm::BitVector CoveredArgs; 2132 bool usesPositionalArgs; 2133 bool atFirstArg; 2134 bool inFunctionCall; 2135 Sema::VariadicCallType CallType; 2136public: 2137 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 2138 const Expr *origFormatExpr, unsigned firstDataArg, 2139 unsigned numDataArgs, const char *beg, bool hasVAListArg, 2140 ArrayRef<const Expr *> Args, 2141 unsigned formatIdx, bool inFunctionCall, 2142 Sema::VariadicCallType callType) 2143 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 2144 FirstDataArg(firstDataArg), NumDataArgs(numDataArgs), 2145 Beg(beg), HasVAListArg(hasVAListArg), 2146 Args(Args), FormatIdx(formatIdx), 2147 usesPositionalArgs(false), atFirstArg(true), 2148 inFunctionCall(inFunctionCall), CallType(callType) { 2149 CoveredArgs.resize(numDataArgs); 2150 CoveredArgs.reset(); 2151 } 2152 2153 void DoneProcessing(); 2154 2155 void HandleIncompleteSpecifier(const char *startSpecifier, 2156 unsigned specifierLen); 2157 2158 void HandleInvalidLengthModifier( 2159 const analyze_format_string::FormatSpecifier &FS, 2160 const analyze_format_string::ConversionSpecifier &CS, 2161 const char *startSpecifier, unsigned specifierLen, unsigned DiagID); 2162 2163 void HandleNonStandardLengthModifier( 2164 const analyze_format_string::FormatSpecifier &FS, 2165 const char *startSpecifier, unsigned specifierLen); 2166 2167 void HandleNonStandardConversionSpecifier( 2168 const analyze_format_string::ConversionSpecifier &CS, 2169 const char *startSpecifier, unsigned specifierLen); 2170 2171 virtual void HandlePosition(const char *startPos, unsigned posLen); 2172 2173 virtual void HandleInvalidPosition(const char *startSpecifier, 2174 unsigned specifierLen, 2175 analyze_format_string::PositionContext p); 2176 2177 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 2178 2179 void HandleNullChar(const char *nullCharacter); 2180 2181 template <typename Range> 2182 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 2183 const Expr *ArgumentExpr, 2184 PartialDiagnostic PDiag, 2185 SourceLocation StringLoc, 2186 bool IsStringLocation, Range StringRange, 2187 ArrayRef<FixItHint> Fixit = None); 2188 2189protected: 2190 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 2191 const char *startSpec, 2192 unsigned specifierLen, 2193 const char *csStart, unsigned csLen); 2194 2195 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 2196 const char *startSpec, 2197 unsigned specifierLen); 2198 2199 SourceRange getFormatStringRange(); 2200 CharSourceRange getSpecifierRange(const char *startSpecifier, 2201 unsigned specifierLen); 2202 SourceLocation getLocationOfByte(const char *x); 2203 2204 const Expr *getDataArg(unsigned i) const; 2205 2206 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 2207 const analyze_format_string::ConversionSpecifier &CS, 2208 const char *startSpecifier, unsigned specifierLen, 2209 unsigned argIndex); 2210 2211 template <typename Range> 2212 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 2213 bool IsStringLocation, Range StringRange, 2214 ArrayRef<FixItHint> Fixit = None); 2215 2216 void CheckPositionalAndNonpositionalArgs( 2217 const analyze_format_string::FormatSpecifier *FS); 2218}; 2219} 2220 2221SourceRange CheckFormatHandler::getFormatStringRange() { 2222 return OrigFormatExpr->getSourceRange(); 2223} 2224 2225CharSourceRange CheckFormatHandler:: 2226getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 2227 SourceLocation Start = getLocationOfByte(startSpecifier); 2228 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 2229 2230 // Advance the end SourceLocation by one due to half-open ranges. 2231 End = End.getLocWithOffset(1); 2232 2233 return CharSourceRange::getCharRange(Start, End); 2234} 2235 2236SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 2237 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 2238} 2239 2240void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 2241 unsigned specifierLen){ 2242 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 2243 getLocationOfByte(startSpecifier), 2244 /*IsStringLocation*/true, 2245 getSpecifierRange(startSpecifier, specifierLen)); 2246} 2247 2248void CheckFormatHandler::HandleInvalidLengthModifier( 2249 const analyze_format_string::FormatSpecifier &FS, 2250 const analyze_format_string::ConversionSpecifier &CS, 2251 const char *startSpecifier, unsigned specifierLen, unsigned DiagID) { 2252 using namespace analyze_format_string; 2253 2254 const LengthModifier &LM = FS.getLengthModifier(); 2255 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2256 2257 // See if we know how to fix this length modifier. 2258 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2259 if (FixedLM) { 2260 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2261 getLocationOfByte(LM.getStart()), 2262 /*IsStringLocation*/true, 2263 getSpecifierRange(startSpecifier, specifierLen)); 2264 2265 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2266 << FixedLM->toString() 2267 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2268 2269 } else { 2270 FixItHint Hint; 2271 if (DiagID == diag::warn_format_nonsensical_length) 2272 Hint = FixItHint::CreateRemoval(LMRange); 2273 2274 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2275 getLocationOfByte(LM.getStart()), 2276 /*IsStringLocation*/true, 2277 getSpecifierRange(startSpecifier, specifierLen), 2278 Hint); 2279 } 2280} 2281 2282void CheckFormatHandler::HandleNonStandardLengthModifier( 2283 const analyze_format_string::FormatSpecifier &FS, 2284 const char *startSpecifier, unsigned specifierLen) { 2285 using namespace analyze_format_string; 2286 2287 const LengthModifier &LM = FS.getLengthModifier(); 2288 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2289 2290 // See if we know how to fix this length modifier. 2291 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2292 if (FixedLM) { 2293 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2294 << LM.toString() << 0, 2295 getLocationOfByte(LM.getStart()), 2296 /*IsStringLocation*/true, 2297 getSpecifierRange(startSpecifier, specifierLen)); 2298 2299 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2300 << FixedLM->toString() 2301 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2302 2303 } else { 2304 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2305 << LM.toString() << 0, 2306 getLocationOfByte(LM.getStart()), 2307 /*IsStringLocation*/true, 2308 getSpecifierRange(startSpecifier, specifierLen)); 2309 } 2310} 2311 2312void CheckFormatHandler::HandleNonStandardConversionSpecifier( 2313 const analyze_format_string::ConversionSpecifier &CS, 2314 const char *startSpecifier, unsigned specifierLen) { 2315 using namespace analyze_format_string; 2316 2317 // See if we know how to fix this conversion specifier. 2318 Optional<ConversionSpecifier> FixedCS = CS.getStandardSpecifier(); 2319 if (FixedCS) { 2320 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2321 << CS.toString() << /*conversion specifier*/1, 2322 getLocationOfByte(CS.getStart()), 2323 /*IsStringLocation*/true, 2324 getSpecifierRange(startSpecifier, specifierLen)); 2325 2326 CharSourceRange CSRange = getSpecifierRange(CS.getStart(), CS.getLength()); 2327 S.Diag(getLocationOfByte(CS.getStart()), diag::note_format_fix_specifier) 2328 << FixedCS->toString() 2329 << FixItHint::CreateReplacement(CSRange, FixedCS->toString()); 2330 } else { 2331 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2332 << CS.toString() << /*conversion specifier*/1, 2333 getLocationOfByte(CS.getStart()), 2334 /*IsStringLocation*/true, 2335 getSpecifierRange(startSpecifier, specifierLen)); 2336 } 2337} 2338 2339void CheckFormatHandler::HandlePosition(const char *startPos, 2340 unsigned posLen) { 2341 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 2342 getLocationOfByte(startPos), 2343 /*IsStringLocation*/true, 2344 getSpecifierRange(startPos, posLen)); 2345} 2346 2347void 2348CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 2349 analyze_format_string::PositionContext p) { 2350 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 2351 << (unsigned) p, 2352 getLocationOfByte(startPos), /*IsStringLocation*/true, 2353 getSpecifierRange(startPos, posLen)); 2354} 2355 2356void CheckFormatHandler::HandleZeroPosition(const char *startPos, 2357 unsigned posLen) { 2358 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 2359 getLocationOfByte(startPos), 2360 /*IsStringLocation*/true, 2361 getSpecifierRange(startPos, posLen)); 2362} 2363 2364void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 2365 if (!isa<ObjCStringLiteral>(OrigFormatExpr)) { 2366 // The presence of a null character is likely an error. 2367 EmitFormatDiagnostic( 2368 S.PDiag(diag::warn_printf_format_string_contains_null_char), 2369 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 2370 getFormatStringRange()); 2371 } 2372} 2373 2374// Note that this may return NULL if there was an error parsing or building 2375// one of the argument expressions. 2376const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 2377 return Args[FirstDataArg + i]; 2378} 2379 2380void CheckFormatHandler::DoneProcessing() { 2381 // Does the number of data arguments exceed the number of 2382 // format conversions in the format string? 2383 if (!HasVAListArg) { 2384 // Find any arguments that weren't covered. 2385 CoveredArgs.flip(); 2386 signed notCoveredArg = CoveredArgs.find_first(); 2387 if (notCoveredArg >= 0) { 2388 assert((unsigned)notCoveredArg < NumDataArgs); 2389 if (const Expr *E = getDataArg((unsigned) notCoveredArg)) { 2390 SourceLocation Loc = E->getLocStart(); 2391 if (!S.getSourceManager().isInSystemMacro(Loc)) { 2392 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 2393 Loc, /*IsStringLocation*/false, 2394 getFormatStringRange()); 2395 } 2396 } 2397 } 2398 } 2399} 2400 2401bool 2402CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 2403 SourceLocation Loc, 2404 const char *startSpec, 2405 unsigned specifierLen, 2406 const char *csStart, 2407 unsigned csLen) { 2408 2409 bool keepGoing = true; 2410 if (argIndex < NumDataArgs) { 2411 // Consider the argument coverered, even though the specifier doesn't 2412 // make sense. 2413 CoveredArgs.set(argIndex); 2414 } 2415 else { 2416 // If argIndex exceeds the number of data arguments we 2417 // don't issue a warning because that is just a cascade of warnings (and 2418 // they may have intended '%%' anyway). We don't want to continue processing 2419 // the format string after this point, however, as we will like just get 2420 // gibberish when trying to match arguments. 2421 keepGoing = false; 2422 } 2423 2424 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2425 << StringRef(csStart, csLen), 2426 Loc, /*IsStringLocation*/true, 2427 getSpecifierRange(startSpec, specifierLen)); 2428 2429 return keepGoing; 2430} 2431 2432void 2433CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2434 const char *startSpec, 2435 unsigned specifierLen) { 2436 EmitFormatDiagnostic( 2437 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2438 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2439} 2440 2441bool 2442CheckFormatHandler::CheckNumArgs( 2443 const analyze_format_string::FormatSpecifier &FS, 2444 const analyze_format_string::ConversionSpecifier &CS, 2445 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2446 2447 if (argIndex >= NumDataArgs) { 2448 PartialDiagnostic PDiag = FS.usesPositionalArg() 2449 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2450 << (argIndex+1) << NumDataArgs) 2451 : S.PDiag(diag::warn_printf_insufficient_data_args); 2452 EmitFormatDiagnostic( 2453 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2454 getSpecifierRange(startSpecifier, specifierLen)); 2455 return false; 2456 } 2457 return true; 2458} 2459 2460template<typename Range> 2461void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2462 SourceLocation Loc, 2463 bool IsStringLocation, 2464 Range StringRange, 2465 ArrayRef<FixItHint> FixIt) { 2466 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2467 Loc, IsStringLocation, StringRange, FixIt); 2468} 2469 2470/// \brief If the format string is not within the funcion call, emit a note 2471/// so that the function call and string are in diagnostic messages. 2472/// 2473/// \param InFunctionCall if true, the format string is within the function 2474/// call and only one diagnostic message will be produced. Otherwise, an 2475/// extra note will be emitted pointing to location of the format string. 2476/// 2477/// \param ArgumentExpr the expression that is passed as the format string 2478/// argument in the function call. Used for getting locations when two 2479/// diagnostics are emitted. 2480/// 2481/// \param PDiag the callee should already have provided any strings for the 2482/// diagnostic message. This function only adds locations and fixits 2483/// to diagnostics. 2484/// 2485/// \param Loc primary location for diagnostic. If two diagnostics are 2486/// required, one will be at Loc and a new SourceLocation will be created for 2487/// the other one. 2488/// 2489/// \param IsStringLocation if true, Loc points to the format string should be 2490/// used for the note. Otherwise, Loc points to the argument list and will 2491/// be used with PDiag. 2492/// 2493/// \param StringRange some or all of the string to highlight. This is 2494/// templated so it can accept either a CharSourceRange or a SourceRange. 2495/// 2496/// \param FixIt optional fix it hint for the format string. 2497template<typename Range> 2498void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2499 const Expr *ArgumentExpr, 2500 PartialDiagnostic PDiag, 2501 SourceLocation Loc, 2502 bool IsStringLocation, 2503 Range StringRange, 2504 ArrayRef<FixItHint> FixIt) { 2505 if (InFunctionCall) { 2506 const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag); 2507 D << StringRange; 2508 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2509 I != E; ++I) { 2510 D << *I; 2511 } 2512 } else { 2513 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2514 << ArgumentExpr->getSourceRange(); 2515 2516 const Sema::SemaDiagnosticBuilder &Note = 2517 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2518 diag::note_format_string_defined); 2519 2520 Note << StringRange; 2521 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2522 I != E; ++I) { 2523 Note << *I; 2524 } 2525 } 2526} 2527 2528//===--- CHECK: Printf format string checking ------------------------------===// 2529 2530namespace { 2531class CheckPrintfHandler : public CheckFormatHandler { 2532 bool ObjCContext; 2533public: 2534 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2535 const Expr *origFormatExpr, unsigned firstDataArg, 2536 unsigned numDataArgs, bool isObjC, 2537 const char *beg, bool hasVAListArg, 2538 ArrayRef<const Expr *> Args, 2539 unsigned formatIdx, bool inFunctionCall, 2540 Sema::VariadicCallType CallType) 2541 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2542 numDataArgs, beg, hasVAListArg, Args, 2543 formatIdx, inFunctionCall, CallType), ObjCContext(isObjC) 2544 {} 2545 2546 2547 bool HandleInvalidPrintfConversionSpecifier( 2548 const analyze_printf::PrintfSpecifier &FS, 2549 const char *startSpecifier, 2550 unsigned specifierLen); 2551 2552 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2553 const char *startSpecifier, 2554 unsigned specifierLen); 2555 bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2556 const char *StartSpecifier, 2557 unsigned SpecifierLen, 2558 const Expr *E); 2559 2560 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2561 const char *startSpecifier, unsigned specifierLen); 2562 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2563 const analyze_printf::OptionalAmount &Amt, 2564 unsigned type, 2565 const char *startSpecifier, unsigned specifierLen); 2566 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2567 const analyze_printf::OptionalFlag &flag, 2568 const char *startSpecifier, unsigned specifierLen); 2569 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2570 const analyze_printf::OptionalFlag &ignoredFlag, 2571 const analyze_printf::OptionalFlag &flag, 2572 const char *startSpecifier, unsigned specifierLen); 2573 bool checkForCStrMembers(const analyze_printf::ArgType &AT, 2574 const Expr *E, const CharSourceRange &CSR); 2575 2576}; 2577} 2578 2579bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2580 const analyze_printf::PrintfSpecifier &FS, 2581 const char *startSpecifier, 2582 unsigned specifierLen) { 2583 const analyze_printf::PrintfConversionSpecifier &CS = 2584 FS.getConversionSpecifier(); 2585 2586 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2587 getLocationOfByte(CS.getStart()), 2588 startSpecifier, specifierLen, 2589 CS.getStart(), CS.getLength()); 2590} 2591 2592bool CheckPrintfHandler::HandleAmount( 2593 const analyze_format_string::OptionalAmount &Amt, 2594 unsigned k, const char *startSpecifier, 2595 unsigned specifierLen) { 2596 2597 if (Amt.hasDataArgument()) { 2598 if (!HasVAListArg) { 2599 unsigned argIndex = Amt.getArgIndex(); 2600 if (argIndex >= NumDataArgs) { 2601 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2602 << k, 2603 getLocationOfByte(Amt.getStart()), 2604 /*IsStringLocation*/true, 2605 getSpecifierRange(startSpecifier, specifierLen)); 2606 // Don't do any more checking. We will just emit 2607 // spurious errors. 2608 return false; 2609 } 2610 2611 // Type check the data argument. It should be an 'int'. 2612 // Although not in conformance with C99, we also allow the argument to be 2613 // an 'unsigned int' as that is a reasonably safe case. GCC also 2614 // doesn't emit a warning for that case. 2615 CoveredArgs.set(argIndex); 2616 const Expr *Arg = getDataArg(argIndex); 2617 if (!Arg) 2618 return false; 2619 2620 QualType T = Arg->getType(); 2621 2622 const analyze_printf::ArgType &AT = Amt.getArgType(S.Context); 2623 assert(AT.isValid()); 2624 2625 if (!AT.matchesType(S.Context, T)) { 2626 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2627 << k << AT.getRepresentativeTypeName(S.Context) 2628 << T << Arg->getSourceRange(), 2629 getLocationOfByte(Amt.getStart()), 2630 /*IsStringLocation*/true, 2631 getSpecifierRange(startSpecifier, specifierLen)); 2632 // Don't do any more checking. We will just emit 2633 // spurious errors. 2634 return false; 2635 } 2636 } 2637 } 2638 return true; 2639} 2640 2641void CheckPrintfHandler::HandleInvalidAmount( 2642 const analyze_printf::PrintfSpecifier &FS, 2643 const analyze_printf::OptionalAmount &Amt, 2644 unsigned type, 2645 const char *startSpecifier, 2646 unsigned specifierLen) { 2647 const analyze_printf::PrintfConversionSpecifier &CS = 2648 FS.getConversionSpecifier(); 2649 2650 FixItHint fixit = 2651 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2652 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2653 Amt.getConstantLength())) 2654 : FixItHint(); 2655 2656 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2657 << type << CS.toString(), 2658 getLocationOfByte(Amt.getStart()), 2659 /*IsStringLocation*/true, 2660 getSpecifierRange(startSpecifier, specifierLen), 2661 fixit); 2662} 2663 2664void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2665 const analyze_printf::OptionalFlag &flag, 2666 const char *startSpecifier, 2667 unsigned specifierLen) { 2668 // Warn about pointless flag with a fixit removal. 2669 const analyze_printf::PrintfConversionSpecifier &CS = 2670 FS.getConversionSpecifier(); 2671 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2672 << flag.toString() << CS.toString(), 2673 getLocationOfByte(flag.getPosition()), 2674 /*IsStringLocation*/true, 2675 getSpecifierRange(startSpecifier, specifierLen), 2676 FixItHint::CreateRemoval( 2677 getSpecifierRange(flag.getPosition(), 1))); 2678} 2679 2680void CheckPrintfHandler::HandleIgnoredFlag( 2681 const analyze_printf::PrintfSpecifier &FS, 2682 const analyze_printf::OptionalFlag &ignoredFlag, 2683 const analyze_printf::OptionalFlag &flag, 2684 const char *startSpecifier, 2685 unsigned specifierLen) { 2686 // Warn about ignored flag with a fixit removal. 2687 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2688 << ignoredFlag.toString() << flag.toString(), 2689 getLocationOfByte(ignoredFlag.getPosition()), 2690 /*IsStringLocation*/true, 2691 getSpecifierRange(startSpecifier, specifierLen), 2692 FixItHint::CreateRemoval( 2693 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2694} 2695 2696// Determines if the specified is a C++ class or struct containing 2697// a member with the specified name and kind (e.g. a CXXMethodDecl named 2698// "c_str()"). 2699template<typename MemberKind> 2700static llvm::SmallPtrSet<MemberKind*, 1> 2701CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) { 2702 const RecordType *RT = Ty->getAs<RecordType>(); 2703 llvm::SmallPtrSet<MemberKind*, 1> Results; 2704 2705 if (!RT) 2706 return Results; 2707 const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl()); 2708 if (!RD) 2709 return Results; 2710 2711 LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(), 2712 Sema::LookupMemberName); 2713 2714 // We just need to include all members of the right kind turned up by the 2715 // filter, at this point. 2716 if (S.LookupQualifiedName(R, RT->getDecl())) 2717 for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) { 2718 NamedDecl *decl = (*I)->getUnderlyingDecl(); 2719 if (MemberKind *FK = dyn_cast<MemberKind>(decl)) 2720 Results.insert(FK); 2721 } 2722 return Results; 2723} 2724 2725// Check if a (w)string was passed when a (w)char* was needed, and offer a 2726// better diagnostic if so. AT is assumed to be valid. 2727// Returns true when a c_str() conversion method is found. 2728bool CheckPrintfHandler::checkForCStrMembers( 2729 const analyze_printf::ArgType &AT, const Expr *E, 2730 const CharSourceRange &CSR) { 2731 typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet; 2732 2733 MethodSet Results = 2734 CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType()); 2735 2736 for (MethodSet::iterator MI = Results.begin(), ME = Results.end(); 2737 MI != ME; ++MI) { 2738 const CXXMethodDecl *Method = *MI; 2739 if (Method->getNumParams() == 0 && 2740 AT.matchesType(S.Context, Method->getResultType())) { 2741 // FIXME: Suggest parens if the expression needs them. 2742 SourceLocation EndLoc = 2743 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()); 2744 S.Diag(E->getLocStart(), diag::note_printf_c_str) 2745 << "c_str()" 2746 << FixItHint::CreateInsertion(EndLoc, ".c_str()"); 2747 return true; 2748 } 2749 } 2750 2751 return false; 2752} 2753 2754bool 2755CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2756 &FS, 2757 const char *startSpecifier, 2758 unsigned specifierLen) { 2759 2760 using namespace analyze_format_string; 2761 using namespace analyze_printf; 2762 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2763 2764 if (FS.consumesDataArgument()) { 2765 if (atFirstArg) { 2766 atFirstArg = false; 2767 usesPositionalArgs = FS.usesPositionalArg(); 2768 } 2769 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2770 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2771 startSpecifier, specifierLen); 2772 return false; 2773 } 2774 } 2775 2776 // First check if the field width, precision, and conversion specifier 2777 // have matching data arguments. 2778 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2779 startSpecifier, specifierLen)) { 2780 return false; 2781 } 2782 2783 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2784 startSpecifier, specifierLen)) { 2785 return false; 2786 } 2787 2788 if (!CS.consumesDataArgument()) { 2789 // FIXME: Technically specifying a precision or field width here 2790 // makes no sense. Worth issuing a warning at some point. 2791 return true; 2792 } 2793 2794 // Consume the argument. 2795 unsigned argIndex = FS.getArgIndex(); 2796 if (argIndex < NumDataArgs) { 2797 // The check to see if the argIndex is valid will come later. 2798 // We set the bit here because we may exit early from this 2799 // function if we encounter some other error. 2800 CoveredArgs.set(argIndex); 2801 } 2802 2803 // Check for using an Objective-C specific conversion specifier 2804 // in a non-ObjC literal. 2805 if (!ObjCContext && CS.isObjCArg()) { 2806 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2807 specifierLen); 2808 } 2809 2810 // Check for invalid use of field width 2811 if (!FS.hasValidFieldWidth()) { 2812 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2813 startSpecifier, specifierLen); 2814 } 2815 2816 // Check for invalid use of precision 2817 if (!FS.hasValidPrecision()) { 2818 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2819 startSpecifier, specifierLen); 2820 } 2821 2822 // Check each flag does not conflict with any other component. 2823 if (!FS.hasValidThousandsGroupingPrefix()) 2824 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2825 if (!FS.hasValidLeadingZeros()) 2826 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2827 if (!FS.hasValidPlusPrefix()) 2828 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2829 if (!FS.hasValidSpacePrefix()) 2830 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2831 if (!FS.hasValidAlternativeForm()) 2832 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2833 if (!FS.hasValidLeftJustified()) 2834 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2835 2836 // Check that flags are not ignored by another flag 2837 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2838 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2839 startSpecifier, specifierLen); 2840 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2841 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2842 startSpecifier, specifierLen); 2843 2844 // Check the length modifier is valid with the given conversion specifier. 2845 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 2846 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2847 diag::warn_format_nonsensical_length); 2848 else if (!FS.hasStandardLengthModifier()) 2849 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 2850 else if (!FS.hasStandardLengthConversionCombination()) 2851 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2852 diag::warn_format_non_standard_conversion_spec); 2853 2854 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2855 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2856 2857 // The remaining checks depend on the data arguments. 2858 if (HasVAListArg) 2859 return true; 2860 2861 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2862 return false; 2863 2864 const Expr *Arg = getDataArg(argIndex); 2865 if (!Arg) 2866 return true; 2867 2868 return checkFormatExpr(FS, startSpecifier, specifierLen, Arg); 2869} 2870 2871static bool requiresParensToAddCast(const Expr *E) { 2872 // FIXME: We should have a general way to reason about operator 2873 // precedence and whether parens are actually needed here. 2874 // Take care of a few common cases where they aren't. 2875 const Expr *Inside = E->IgnoreImpCasts(); 2876 if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside)) 2877 Inside = POE->getSyntacticForm()->IgnoreImpCasts(); 2878 2879 switch (Inside->getStmtClass()) { 2880 case Stmt::ArraySubscriptExprClass: 2881 case Stmt::CallExprClass: 2882 case Stmt::CharacterLiteralClass: 2883 case Stmt::CXXBoolLiteralExprClass: 2884 case Stmt::DeclRefExprClass: 2885 case Stmt::FloatingLiteralClass: 2886 case Stmt::IntegerLiteralClass: 2887 case Stmt::MemberExprClass: 2888 case Stmt::ObjCArrayLiteralClass: 2889 case Stmt::ObjCBoolLiteralExprClass: 2890 case Stmt::ObjCBoxedExprClass: 2891 case Stmt::ObjCDictionaryLiteralClass: 2892 case Stmt::ObjCEncodeExprClass: 2893 case Stmt::ObjCIvarRefExprClass: 2894 case Stmt::ObjCMessageExprClass: 2895 case Stmt::ObjCPropertyRefExprClass: 2896 case Stmt::ObjCStringLiteralClass: 2897 case Stmt::ObjCSubscriptRefExprClass: 2898 case Stmt::ParenExprClass: 2899 case Stmt::StringLiteralClass: 2900 case Stmt::UnaryOperatorClass: 2901 return false; 2902 default: 2903 return true; 2904 } 2905} 2906 2907bool 2908CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2909 const char *StartSpecifier, 2910 unsigned SpecifierLen, 2911 const Expr *E) { 2912 using namespace analyze_format_string; 2913 using namespace analyze_printf; 2914 // Now type check the data expression that matches the 2915 // format specifier. 2916 const analyze_printf::ArgType &AT = FS.getArgType(S.Context, 2917 ObjCContext); 2918 if (!AT.isValid()) 2919 return true; 2920 2921 QualType ExprTy = E->getType(); 2922 while (const TypeOfExprType *TET = dyn_cast<TypeOfExprType>(ExprTy)) { 2923 ExprTy = TET->getUnderlyingExpr()->getType(); 2924 } 2925 2926 if (AT.matchesType(S.Context, ExprTy)) 2927 return true; 2928 2929 // Look through argument promotions for our error message's reported type. 2930 // This includes the integral and floating promotions, but excludes array 2931 // and function pointer decay; seeing that an argument intended to be a 2932 // string has type 'char [6]' is probably more confusing than 'char *'. 2933 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2934 if (ICE->getCastKind() == CK_IntegralCast || 2935 ICE->getCastKind() == CK_FloatingCast) { 2936 E = ICE->getSubExpr(); 2937 ExprTy = E->getType(); 2938 2939 // Check if we didn't match because of an implicit cast from a 'char' 2940 // or 'short' to an 'int'. This is done because printf is a varargs 2941 // function. 2942 if (ICE->getType() == S.Context.IntTy || 2943 ICE->getType() == S.Context.UnsignedIntTy) { 2944 // All further checking is done on the subexpression. 2945 if (AT.matchesType(S.Context, ExprTy)) 2946 return true; 2947 } 2948 } 2949 } else if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E)) { 2950 // Special case for 'a', which has type 'int' in C. 2951 // Note, however, that we do /not/ want to treat multibyte constants like 2952 // 'MooV' as characters! This form is deprecated but still exists. 2953 if (ExprTy == S.Context.IntTy) 2954 if (llvm::isUIntN(S.Context.getCharWidth(), CL->getValue())) 2955 ExprTy = S.Context.CharTy; 2956 } 2957 2958 // %C in an Objective-C context prints a unichar, not a wchar_t. 2959 // If the argument is an integer of some kind, believe the %C and suggest 2960 // a cast instead of changing the conversion specifier. 2961 QualType IntendedTy = ExprTy; 2962 if (ObjCContext && 2963 FS.getConversionSpecifier().getKind() == ConversionSpecifier::CArg) { 2964 if (ExprTy->isIntegralOrUnscopedEnumerationType() && 2965 !ExprTy->isCharType()) { 2966 // 'unichar' is defined as a typedef of unsigned short, but we should 2967 // prefer using the typedef if it is visible. 2968 IntendedTy = S.Context.UnsignedShortTy; 2969 2970 LookupResult Result(S, &S.Context.Idents.get("unichar"), E->getLocStart(), 2971 Sema::LookupOrdinaryName); 2972 if (S.LookupName(Result, S.getCurScope())) { 2973 NamedDecl *ND = Result.getFoundDecl(); 2974 if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(ND)) 2975 if (TD->getUnderlyingType() == IntendedTy) 2976 IntendedTy = S.Context.getTypedefType(TD); 2977 } 2978 } 2979 } 2980 2981 // Special-case some of Darwin's platform-independence types by suggesting 2982 // casts to primitive types that are known to be large enough. 2983 bool ShouldNotPrintDirectly = false; 2984 if (S.Context.getTargetInfo().getTriple().isOSDarwin()) { 2985 // Use a 'while' to peel off layers of typedefs. 2986 QualType TyTy = IntendedTy; 2987 while (const TypedefType *UserTy = TyTy->getAs<TypedefType>()) { 2988 StringRef Name = UserTy->getDecl()->getName(); 2989 QualType CastTy = llvm::StringSwitch<QualType>(Name) 2990 .Case("NSInteger", S.Context.LongTy) 2991 .Case("NSUInteger", S.Context.UnsignedLongTy) 2992 .Case("SInt32", S.Context.IntTy) 2993 .Case("UInt32", S.Context.UnsignedIntTy) 2994 .Default(QualType()); 2995 2996 if (!CastTy.isNull()) { 2997 ShouldNotPrintDirectly = true; 2998 IntendedTy = CastTy; 2999 break; 3000 } 3001 TyTy = UserTy->desugar(); 3002 } 3003 } 3004 3005 // We may be able to offer a FixItHint if it is a supported type. 3006 PrintfSpecifier fixedFS = FS; 3007 bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(), 3008 S.Context, ObjCContext); 3009 3010 if (success) { 3011 // Get the fix string from the fixed format specifier 3012 SmallString<16> buf; 3013 llvm::raw_svector_ostream os(buf); 3014 fixedFS.toString(os); 3015 3016 CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen); 3017 3018 if (IntendedTy == ExprTy) { 3019 // In this case, the specifier is wrong and should be changed to match 3020 // the argument. 3021 EmitFormatDiagnostic( 3022 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3023 << AT.getRepresentativeTypeName(S.Context) << IntendedTy 3024 << E->getSourceRange(), 3025 E->getLocStart(), 3026 /*IsStringLocation*/false, 3027 SpecRange, 3028 FixItHint::CreateReplacement(SpecRange, os.str())); 3029 3030 } else { 3031 // The canonical type for formatting this value is different from the 3032 // actual type of the expression. (This occurs, for example, with Darwin's 3033 // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but 3034 // should be printed as 'long' for 64-bit compatibility.) 3035 // Rather than emitting a normal format/argument mismatch, we want to 3036 // add a cast to the recommended type (and correct the format string 3037 // if necessary). 3038 SmallString<16> CastBuf; 3039 llvm::raw_svector_ostream CastFix(CastBuf); 3040 CastFix << "("; 3041 IntendedTy.print(CastFix, S.Context.getPrintingPolicy()); 3042 CastFix << ")"; 3043 3044 SmallVector<FixItHint,4> Hints; 3045 if (!AT.matchesType(S.Context, IntendedTy)) 3046 Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str())); 3047 3048 if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) { 3049 // If there's already a cast present, just replace it. 3050 SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc()); 3051 Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str())); 3052 3053 } else if (!requiresParensToAddCast(E)) { 3054 // If the expression has high enough precedence, 3055 // just write the C-style cast. 3056 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 3057 CastFix.str())); 3058 } else { 3059 // Otherwise, add parens around the expression as well as the cast. 3060 CastFix << "("; 3061 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 3062 CastFix.str())); 3063 3064 SourceLocation After = S.PP.getLocForEndOfToken(E->getLocEnd()); 3065 Hints.push_back(FixItHint::CreateInsertion(After, ")")); 3066 } 3067 3068 if (ShouldNotPrintDirectly) { 3069 // The expression has a type that should not be printed directly. 3070 // We extract the name from the typedef because we don't want to show 3071 // the underlying type in the diagnostic. 3072 StringRef Name = cast<TypedefType>(ExprTy)->getDecl()->getName(); 3073 3074 EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast) 3075 << Name << IntendedTy 3076 << E->getSourceRange(), 3077 E->getLocStart(), /*IsStringLocation=*/false, 3078 SpecRange, Hints); 3079 } else { 3080 // In this case, the expression could be printed using a different 3081 // specifier, but we've decided that the specifier is probably correct 3082 // and we should cast instead. Just use the normal warning message. 3083 EmitFormatDiagnostic( 3084 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3085 << AT.getRepresentativeTypeName(S.Context) << ExprTy 3086 << E->getSourceRange(), 3087 E->getLocStart(), /*IsStringLocation*/false, 3088 SpecRange, Hints); 3089 } 3090 } 3091 } else { 3092 const CharSourceRange &CSR = getSpecifierRange(StartSpecifier, 3093 SpecifierLen); 3094 // Since the warning for passing non-POD types to variadic functions 3095 // was deferred until now, we emit a warning for non-POD 3096 // arguments here. 3097 if (S.isValidVarArgType(ExprTy) == Sema::VAK_Invalid) { 3098 unsigned DiagKind; 3099 if (ExprTy->isObjCObjectType()) 3100 DiagKind = diag::err_cannot_pass_objc_interface_to_vararg_format; 3101 else 3102 DiagKind = diag::warn_non_pod_vararg_with_format_string; 3103 3104 EmitFormatDiagnostic( 3105 S.PDiag(DiagKind) 3106 << S.getLangOpts().CPlusPlus11 3107 << ExprTy 3108 << CallType 3109 << AT.getRepresentativeTypeName(S.Context) 3110 << CSR 3111 << E->getSourceRange(), 3112 E->getLocStart(), /*IsStringLocation*/false, CSR); 3113 3114 checkForCStrMembers(AT, E, CSR); 3115 } else 3116 EmitFormatDiagnostic( 3117 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3118 << AT.getRepresentativeTypeName(S.Context) << ExprTy 3119 << CSR 3120 << E->getSourceRange(), 3121 E->getLocStart(), /*IsStringLocation*/false, CSR); 3122 } 3123 3124 return true; 3125} 3126 3127//===--- CHECK: Scanf format string checking ------------------------------===// 3128 3129namespace { 3130class CheckScanfHandler : public CheckFormatHandler { 3131public: 3132 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 3133 const Expr *origFormatExpr, unsigned firstDataArg, 3134 unsigned numDataArgs, const char *beg, bool hasVAListArg, 3135 ArrayRef<const Expr *> Args, 3136 unsigned formatIdx, bool inFunctionCall, 3137 Sema::VariadicCallType CallType) 3138 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 3139 numDataArgs, beg, hasVAListArg, 3140 Args, formatIdx, inFunctionCall, CallType) 3141 {} 3142 3143 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 3144 const char *startSpecifier, 3145 unsigned specifierLen); 3146 3147 bool HandleInvalidScanfConversionSpecifier( 3148 const analyze_scanf::ScanfSpecifier &FS, 3149 const char *startSpecifier, 3150 unsigned specifierLen); 3151 3152 void HandleIncompleteScanList(const char *start, const char *end); 3153}; 3154} 3155 3156void CheckScanfHandler::HandleIncompleteScanList(const char *start, 3157 const char *end) { 3158 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 3159 getLocationOfByte(end), /*IsStringLocation*/true, 3160 getSpecifierRange(start, end - start)); 3161} 3162 3163bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 3164 const analyze_scanf::ScanfSpecifier &FS, 3165 const char *startSpecifier, 3166 unsigned specifierLen) { 3167 3168 const analyze_scanf::ScanfConversionSpecifier &CS = 3169 FS.getConversionSpecifier(); 3170 3171 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 3172 getLocationOfByte(CS.getStart()), 3173 startSpecifier, specifierLen, 3174 CS.getStart(), CS.getLength()); 3175} 3176 3177bool CheckScanfHandler::HandleScanfSpecifier( 3178 const analyze_scanf::ScanfSpecifier &FS, 3179 const char *startSpecifier, 3180 unsigned specifierLen) { 3181 3182 using namespace analyze_scanf; 3183 using namespace analyze_format_string; 3184 3185 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 3186 3187 // Handle case where '%' and '*' don't consume an argument. These shouldn't 3188 // be used to decide if we are using positional arguments consistently. 3189 if (FS.consumesDataArgument()) { 3190 if (atFirstArg) { 3191 atFirstArg = false; 3192 usesPositionalArgs = FS.usesPositionalArg(); 3193 } 3194 else if (usesPositionalArgs != FS.usesPositionalArg()) { 3195 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 3196 startSpecifier, specifierLen); 3197 return false; 3198 } 3199 } 3200 3201 // Check if the field with is non-zero. 3202 const OptionalAmount &Amt = FS.getFieldWidth(); 3203 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 3204 if (Amt.getConstantAmount() == 0) { 3205 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 3206 Amt.getConstantLength()); 3207 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 3208 getLocationOfByte(Amt.getStart()), 3209 /*IsStringLocation*/true, R, 3210 FixItHint::CreateRemoval(R)); 3211 } 3212 } 3213 3214 if (!FS.consumesDataArgument()) { 3215 // FIXME: Technically specifying a precision or field width here 3216 // makes no sense. Worth issuing a warning at some point. 3217 return true; 3218 } 3219 3220 // Consume the argument. 3221 unsigned argIndex = FS.getArgIndex(); 3222 if (argIndex < NumDataArgs) { 3223 // The check to see if the argIndex is valid will come later. 3224 // We set the bit here because we may exit early from this 3225 // function if we encounter some other error. 3226 CoveredArgs.set(argIndex); 3227 } 3228 3229 // Check the length modifier is valid with the given conversion specifier. 3230 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3231 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3232 diag::warn_format_nonsensical_length); 3233 else if (!FS.hasStandardLengthModifier()) 3234 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3235 else if (!FS.hasStandardLengthConversionCombination()) 3236 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3237 diag::warn_format_non_standard_conversion_spec); 3238 3239 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3240 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3241 3242 // The remaining checks depend on the data arguments. 3243 if (HasVAListArg) 3244 return true; 3245 3246 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3247 return false; 3248 3249 // Check that the argument type matches the format specifier. 3250 const Expr *Ex = getDataArg(argIndex); 3251 if (!Ex) 3252 return true; 3253 3254 const analyze_format_string::ArgType &AT = FS.getArgType(S.Context); 3255 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) { 3256 ScanfSpecifier fixedFS = FS; 3257 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 3258 S.Context); 3259 3260 if (success) { 3261 // Get the fix string from the fixed format specifier. 3262 SmallString<128> buf; 3263 llvm::raw_svector_ostream os(buf); 3264 fixedFS.toString(os); 3265 3266 EmitFormatDiagnostic( 3267 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3268 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3269 << Ex->getSourceRange(), 3270 Ex->getLocStart(), 3271 /*IsStringLocation*/false, 3272 getSpecifierRange(startSpecifier, specifierLen), 3273 FixItHint::CreateReplacement( 3274 getSpecifierRange(startSpecifier, specifierLen), 3275 os.str())); 3276 } else { 3277 EmitFormatDiagnostic( 3278 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3279 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3280 << Ex->getSourceRange(), 3281 Ex->getLocStart(), 3282 /*IsStringLocation*/false, 3283 getSpecifierRange(startSpecifier, specifierLen)); 3284 } 3285 } 3286 3287 return true; 3288} 3289 3290void Sema::CheckFormatString(const StringLiteral *FExpr, 3291 const Expr *OrigFormatExpr, 3292 ArrayRef<const Expr *> Args, 3293 bool HasVAListArg, unsigned format_idx, 3294 unsigned firstDataArg, FormatStringType Type, 3295 bool inFunctionCall, VariadicCallType CallType) { 3296 3297 // CHECK: is the format string a wide literal? 3298 if (!FExpr->isAscii() && !FExpr->isUTF8()) { 3299 CheckFormatHandler::EmitFormatDiagnostic( 3300 *this, inFunctionCall, Args[format_idx], 3301 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 3302 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3303 return; 3304 } 3305 3306 // Str - The format string. NOTE: this is NOT null-terminated! 3307 StringRef StrRef = FExpr->getString(); 3308 const char *Str = StrRef.data(); 3309 unsigned StrLen = StrRef.size(); 3310 const unsigned numDataArgs = Args.size() - firstDataArg; 3311 3312 // CHECK: empty format string? 3313 if (StrLen == 0 && numDataArgs > 0) { 3314 CheckFormatHandler::EmitFormatDiagnostic( 3315 *this, inFunctionCall, Args[format_idx], 3316 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 3317 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3318 return; 3319 } 3320 3321 if (Type == FST_Printf || Type == FST_NSString) { 3322 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 3323 numDataArgs, (Type == FST_NSString), 3324 Str, HasVAListArg, Args, format_idx, 3325 inFunctionCall, CallType); 3326 3327 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 3328 getLangOpts(), 3329 Context.getTargetInfo())) 3330 H.DoneProcessing(); 3331 } else if (Type == FST_Scanf) { 3332 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs, 3333 Str, HasVAListArg, Args, format_idx, 3334 inFunctionCall, CallType); 3335 3336 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 3337 getLangOpts(), 3338 Context.getTargetInfo())) 3339 H.DoneProcessing(); 3340 } // TODO: handle other formats 3341} 3342 3343//===--- CHECK: Standard memory functions ---------------------------------===// 3344 3345/// \brief Determine whether the given type is a dynamic class type (e.g., 3346/// whether it has a vtable). 3347static bool isDynamicClassType(QualType T) { 3348 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 3349 if (CXXRecordDecl *Definition = Record->getDefinition()) 3350 if (Definition->isDynamicClass()) 3351 return true; 3352 3353 return false; 3354} 3355 3356/// \brief If E is a sizeof expression, returns its argument expression, 3357/// otherwise returns NULL. 3358static const Expr *getSizeOfExprArg(const Expr* E) { 3359 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3360 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3361 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 3362 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 3363 3364 return 0; 3365} 3366 3367/// \brief If E is a sizeof expression, returns its argument type. 3368static QualType getSizeOfArgType(const Expr* E) { 3369 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3370 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3371 if (SizeOf->getKind() == clang::UETT_SizeOf) 3372 return SizeOf->getTypeOfArgument(); 3373 3374 return QualType(); 3375} 3376 3377/// \brief Check for dangerous or invalid arguments to memset(). 3378/// 3379/// This issues warnings on known problematic, dangerous or unspecified 3380/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 3381/// function calls. 3382/// 3383/// \param Call The call expression to diagnose. 3384void Sema::CheckMemaccessArguments(const CallExpr *Call, 3385 unsigned BId, 3386 IdentifierInfo *FnName) { 3387 assert(BId != 0); 3388 3389 // It is possible to have a non-standard definition of memset. Validate 3390 // we have enough arguments, and if not, abort further checking. 3391 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 3392 if (Call->getNumArgs() < ExpectedNumArgs) 3393 return; 3394 3395 unsigned LastArg = (BId == Builtin::BImemset || 3396 BId == Builtin::BIstrndup ? 1 : 2); 3397 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 3398 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 3399 3400 // We have special checking when the length is a sizeof expression. 3401 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 3402 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 3403 llvm::FoldingSetNodeID SizeOfArgID; 3404 3405 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 3406 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 3407 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 3408 3409 QualType DestTy = Dest->getType(); 3410 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 3411 QualType PointeeTy = DestPtrTy->getPointeeType(); 3412 3413 // Never warn about void type pointers. This can be used to suppress 3414 // false positives. 3415 if (PointeeTy->isVoidType()) 3416 continue; 3417 3418 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 3419 // actually comparing the expressions for equality. Because computing the 3420 // expression IDs can be expensive, we only do this if the diagnostic is 3421 // enabled. 3422 if (SizeOfArg && 3423 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 3424 SizeOfArg->getExprLoc())) { 3425 // We only compute IDs for expressions if the warning is enabled, and 3426 // cache the sizeof arg's ID. 3427 if (SizeOfArgID == llvm::FoldingSetNodeID()) 3428 SizeOfArg->Profile(SizeOfArgID, Context, true); 3429 llvm::FoldingSetNodeID DestID; 3430 Dest->Profile(DestID, Context, true); 3431 if (DestID == SizeOfArgID) { 3432 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 3433 // over sizeof(src) as well. 3434 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 3435 StringRef ReadableName = FnName->getName(); 3436 3437 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 3438 if (UnaryOp->getOpcode() == UO_AddrOf) 3439 ActionIdx = 1; // If its an address-of operator, just remove it. 3440 if (!PointeeTy->isIncompleteType() && 3441 (Context.getTypeSize(PointeeTy) == Context.getCharWidth())) 3442 ActionIdx = 2; // If the pointee's size is sizeof(char), 3443 // suggest an explicit length. 3444 3445 // If the function is defined as a builtin macro, do not show macro 3446 // expansion. 3447 SourceLocation SL = SizeOfArg->getExprLoc(); 3448 SourceRange DSR = Dest->getSourceRange(); 3449 SourceRange SSR = SizeOfArg->getSourceRange(); 3450 SourceManager &SM = PP.getSourceManager(); 3451 3452 if (SM.isMacroArgExpansion(SL)) { 3453 ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts); 3454 SL = SM.getSpellingLoc(SL); 3455 DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()), 3456 SM.getSpellingLoc(DSR.getEnd())); 3457 SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()), 3458 SM.getSpellingLoc(SSR.getEnd())); 3459 } 3460 3461 DiagRuntimeBehavior(SL, SizeOfArg, 3462 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 3463 << ReadableName 3464 << PointeeTy 3465 << DestTy 3466 << DSR 3467 << SSR); 3468 DiagRuntimeBehavior(SL, SizeOfArg, 3469 PDiag(diag::warn_sizeof_pointer_expr_memaccess_note) 3470 << ActionIdx 3471 << SSR); 3472 3473 break; 3474 } 3475 } 3476 3477 // Also check for cases where the sizeof argument is the exact same 3478 // type as the memory argument, and where it points to a user-defined 3479 // record type. 3480 if (SizeOfArgTy != QualType()) { 3481 if (PointeeTy->isRecordType() && 3482 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 3483 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 3484 PDiag(diag::warn_sizeof_pointer_type_memaccess) 3485 << FnName << SizeOfArgTy << ArgIdx 3486 << PointeeTy << Dest->getSourceRange() 3487 << LenExpr->getSourceRange()); 3488 break; 3489 } 3490 } 3491 3492 // Always complain about dynamic classes. 3493 if (isDynamicClassType(PointeeTy)) { 3494 3495 unsigned OperationType = 0; 3496 // "overwritten" if we're warning about the destination for any call 3497 // but memcmp; otherwise a verb appropriate to the call. 3498 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 3499 if (BId == Builtin::BImemcpy) 3500 OperationType = 1; 3501 else if(BId == Builtin::BImemmove) 3502 OperationType = 2; 3503 else if (BId == Builtin::BImemcmp) 3504 OperationType = 3; 3505 } 3506 3507 DiagRuntimeBehavior( 3508 Dest->getExprLoc(), Dest, 3509 PDiag(diag::warn_dyn_class_memaccess) 3510 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 3511 << FnName << PointeeTy 3512 << OperationType 3513 << Call->getCallee()->getSourceRange()); 3514 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 3515 BId != Builtin::BImemset) 3516 DiagRuntimeBehavior( 3517 Dest->getExprLoc(), Dest, 3518 PDiag(diag::warn_arc_object_memaccess) 3519 << ArgIdx << FnName << PointeeTy 3520 << Call->getCallee()->getSourceRange()); 3521 else 3522 continue; 3523 3524 DiagRuntimeBehavior( 3525 Dest->getExprLoc(), Dest, 3526 PDiag(diag::note_bad_memaccess_silence) 3527 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 3528 break; 3529 } 3530 } 3531} 3532 3533// A little helper routine: ignore addition and subtraction of integer literals. 3534// This intentionally does not ignore all integer constant expressions because 3535// we don't want to remove sizeof(). 3536static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 3537 Ex = Ex->IgnoreParenCasts(); 3538 3539 for (;;) { 3540 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 3541 if (!BO || !BO->isAdditiveOp()) 3542 break; 3543 3544 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 3545 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 3546 3547 if (isa<IntegerLiteral>(RHS)) 3548 Ex = LHS; 3549 else if (isa<IntegerLiteral>(LHS)) 3550 Ex = RHS; 3551 else 3552 break; 3553 } 3554 3555 return Ex; 3556} 3557 3558static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty, 3559 ASTContext &Context) { 3560 // Only handle constant-sized or VLAs, but not flexible members. 3561 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) { 3562 // Only issue the FIXIT for arrays of size > 1. 3563 if (CAT->getSize().getSExtValue() <= 1) 3564 return false; 3565 } else if (!Ty->isVariableArrayType()) { 3566 return false; 3567 } 3568 return true; 3569} 3570 3571// Warn if the user has made the 'size' argument to strlcpy or strlcat 3572// be the size of the source, instead of the destination. 3573void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 3574 IdentifierInfo *FnName) { 3575 3576 // Don't crash if the user has the wrong number of arguments 3577 if (Call->getNumArgs() != 3) 3578 return; 3579 3580 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 3581 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 3582 const Expr *CompareWithSrc = NULL; 3583 3584 // Look for 'strlcpy(dst, x, sizeof(x))' 3585 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 3586 CompareWithSrc = Ex; 3587 else { 3588 // Look for 'strlcpy(dst, x, strlen(x))' 3589 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 3590 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 3591 && SizeCall->getNumArgs() == 1) 3592 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 3593 } 3594 } 3595 3596 if (!CompareWithSrc) 3597 return; 3598 3599 // Determine if the argument to sizeof/strlen is equal to the source 3600 // argument. In principle there's all kinds of things you could do 3601 // here, for instance creating an == expression and evaluating it with 3602 // EvaluateAsBooleanCondition, but this uses a more direct technique: 3603 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 3604 if (!SrcArgDRE) 3605 return; 3606 3607 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 3608 if (!CompareWithSrcDRE || 3609 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 3610 return; 3611 3612 const Expr *OriginalSizeArg = Call->getArg(2); 3613 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 3614 << OriginalSizeArg->getSourceRange() << FnName; 3615 3616 // Output a FIXIT hint if the destination is an array (rather than a 3617 // pointer to an array). This could be enhanced to handle some 3618 // pointers if we know the actual size, like if DstArg is 'array+2' 3619 // we could say 'sizeof(array)-2'. 3620 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 3621 if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context)) 3622 return; 3623 3624 SmallString<128> sizeString; 3625 llvm::raw_svector_ostream OS(sizeString); 3626 OS << "sizeof("; 3627 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3628 OS << ")"; 3629 3630 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 3631 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 3632 OS.str()); 3633} 3634 3635/// Check if two expressions refer to the same declaration. 3636static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 3637 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 3638 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 3639 return D1->getDecl() == D2->getDecl(); 3640 return false; 3641} 3642 3643static const Expr *getStrlenExprArg(const Expr *E) { 3644 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 3645 const FunctionDecl *FD = CE->getDirectCallee(); 3646 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 3647 return 0; 3648 return CE->getArg(0)->IgnoreParenCasts(); 3649 } 3650 return 0; 3651} 3652 3653// Warn on anti-patterns as the 'size' argument to strncat. 3654// The correct size argument should look like following: 3655// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 3656void Sema::CheckStrncatArguments(const CallExpr *CE, 3657 IdentifierInfo *FnName) { 3658 // Don't crash if the user has the wrong number of arguments. 3659 if (CE->getNumArgs() < 3) 3660 return; 3661 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 3662 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 3663 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 3664 3665 // Identify common expressions, which are wrongly used as the size argument 3666 // to strncat and may lead to buffer overflows. 3667 unsigned PatternType = 0; 3668 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 3669 // - sizeof(dst) 3670 if (referToTheSameDecl(SizeOfArg, DstArg)) 3671 PatternType = 1; 3672 // - sizeof(src) 3673 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 3674 PatternType = 2; 3675 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 3676 if (BE->getOpcode() == BO_Sub) { 3677 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 3678 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 3679 // - sizeof(dst) - strlen(dst) 3680 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 3681 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 3682 PatternType = 1; 3683 // - sizeof(src) - (anything) 3684 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 3685 PatternType = 2; 3686 } 3687 } 3688 3689 if (PatternType == 0) 3690 return; 3691 3692 // Generate the diagnostic. 3693 SourceLocation SL = LenArg->getLocStart(); 3694 SourceRange SR = LenArg->getSourceRange(); 3695 SourceManager &SM = PP.getSourceManager(); 3696 3697 // If the function is defined as a builtin macro, do not show macro expansion. 3698 if (SM.isMacroArgExpansion(SL)) { 3699 SL = SM.getSpellingLoc(SL); 3700 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 3701 SM.getSpellingLoc(SR.getEnd())); 3702 } 3703 3704 // Check if the destination is an array (rather than a pointer to an array). 3705 QualType DstTy = DstArg->getType(); 3706 bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy, 3707 Context); 3708 if (!isKnownSizeArray) { 3709 if (PatternType == 1) 3710 Diag(SL, diag::warn_strncat_wrong_size) << SR; 3711 else 3712 Diag(SL, diag::warn_strncat_src_size) << SR; 3713 return; 3714 } 3715 3716 if (PatternType == 1) 3717 Diag(SL, diag::warn_strncat_large_size) << SR; 3718 else 3719 Diag(SL, diag::warn_strncat_src_size) << SR; 3720 3721 SmallString<128> sizeString; 3722 llvm::raw_svector_ostream OS(sizeString); 3723 OS << "sizeof("; 3724 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3725 OS << ") - "; 3726 OS << "strlen("; 3727 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3728 OS << ") - 1"; 3729 3730 Diag(SL, diag::note_strncat_wrong_size) 3731 << FixItHint::CreateReplacement(SR, OS.str()); 3732} 3733 3734//===--- CHECK: Return Address of Stack Variable --------------------------===// 3735 3736static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3737 Decl *ParentDecl); 3738static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3739 Decl *ParentDecl); 3740 3741/// CheckReturnStackAddr - Check if a return statement returns the address 3742/// of a stack variable. 3743void 3744Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3745 SourceLocation ReturnLoc) { 3746 3747 Expr *stackE = 0; 3748 SmallVector<DeclRefExpr *, 8> refVars; 3749 3750 // Perform checking for returned stack addresses, local blocks, 3751 // label addresses or references to temporaries. 3752 if (lhsType->isPointerType() || 3753 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 3754 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 3755 } else if (lhsType->isReferenceType()) { 3756 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 3757 } 3758 3759 if (stackE == 0) 3760 return; // Nothing suspicious was found. 3761 3762 SourceLocation diagLoc; 3763 SourceRange diagRange; 3764 if (refVars.empty()) { 3765 diagLoc = stackE->getLocStart(); 3766 diagRange = stackE->getSourceRange(); 3767 } else { 3768 // We followed through a reference variable. 'stackE' contains the 3769 // problematic expression but we will warn at the return statement pointing 3770 // at the reference variable. We will later display the "trail" of 3771 // reference variables using notes. 3772 diagLoc = refVars[0]->getLocStart(); 3773 diagRange = refVars[0]->getSourceRange(); 3774 } 3775 3776 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 3777 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 3778 : diag::warn_ret_stack_addr) 3779 << DR->getDecl()->getDeclName() << diagRange; 3780 } else if (isa<BlockExpr>(stackE)) { // local block. 3781 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 3782 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 3783 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 3784 } else { // local temporary. 3785 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 3786 : diag::warn_ret_local_temp_addr) 3787 << diagRange; 3788 } 3789 3790 // Display the "trail" of reference variables that we followed until we 3791 // found the problematic expression using notes. 3792 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 3793 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 3794 // If this var binds to another reference var, show the range of the next 3795 // var, otherwise the var binds to the problematic expression, in which case 3796 // show the range of the expression. 3797 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 3798 : stackE->getSourceRange(); 3799 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 3800 << VD->getDeclName() << range; 3801 } 3802} 3803 3804/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 3805/// check if the expression in a return statement evaluates to an address 3806/// to a location on the stack, a local block, an address of a label, or a 3807/// reference to local temporary. The recursion is used to traverse the 3808/// AST of the return expression, with recursion backtracking when we 3809/// encounter a subexpression that (1) clearly does not lead to one of the 3810/// above problematic expressions (2) is something we cannot determine leads to 3811/// a problematic expression based on such local checking. 3812/// 3813/// Both EvalAddr and EvalVal follow through reference variables to evaluate 3814/// the expression that they point to. Such variables are added to the 3815/// 'refVars' vector so that we know what the reference variable "trail" was. 3816/// 3817/// EvalAddr processes expressions that are pointers that are used as 3818/// references (and not L-values). EvalVal handles all other values. 3819/// At the base case of the recursion is a check for the above problematic 3820/// expressions. 3821/// 3822/// This implementation handles: 3823/// 3824/// * pointer-to-pointer casts 3825/// * implicit conversions from array references to pointers 3826/// * taking the address of fields 3827/// * arbitrary interplay between "&" and "*" operators 3828/// * pointer arithmetic from an address of a stack variable 3829/// * taking the address of an array element where the array is on the stack 3830static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3831 Decl *ParentDecl) { 3832 if (E->isTypeDependent()) 3833 return NULL; 3834 3835 // We should only be called for evaluating pointer expressions. 3836 assert((E->getType()->isAnyPointerType() || 3837 E->getType()->isBlockPointerType() || 3838 E->getType()->isObjCQualifiedIdType()) && 3839 "EvalAddr only works on pointers"); 3840 3841 E = E->IgnoreParens(); 3842 3843 // Our "symbolic interpreter" is just a dispatch off the currently 3844 // viewed AST node. We then recursively traverse the AST by calling 3845 // EvalAddr and EvalVal appropriately. 3846 switch (E->getStmtClass()) { 3847 case Stmt::DeclRefExprClass: { 3848 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3849 3850 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3851 // If this is a reference variable, follow through to the expression that 3852 // it points to. 3853 if (V->hasLocalStorage() && 3854 V->getType()->isReferenceType() && V->hasInit()) { 3855 // Add the reference variable to the "trail". 3856 refVars.push_back(DR); 3857 return EvalAddr(V->getInit(), refVars, ParentDecl); 3858 } 3859 3860 return NULL; 3861 } 3862 3863 case Stmt::UnaryOperatorClass: { 3864 // The only unary operator that make sense to handle here 3865 // is AddrOf. All others don't make sense as pointers. 3866 UnaryOperator *U = cast<UnaryOperator>(E); 3867 3868 if (U->getOpcode() == UO_AddrOf) 3869 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 3870 else 3871 return NULL; 3872 } 3873 3874 case Stmt::BinaryOperatorClass: { 3875 // Handle pointer arithmetic. All other binary operators are not valid 3876 // in this context. 3877 BinaryOperator *B = cast<BinaryOperator>(E); 3878 BinaryOperatorKind op = B->getOpcode(); 3879 3880 if (op != BO_Add && op != BO_Sub) 3881 return NULL; 3882 3883 Expr *Base = B->getLHS(); 3884 3885 // Determine which argument is the real pointer base. It could be 3886 // the RHS argument instead of the LHS. 3887 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 3888 3889 assert (Base->getType()->isPointerType()); 3890 return EvalAddr(Base, refVars, ParentDecl); 3891 } 3892 3893 // For conditional operators we need to see if either the LHS or RHS are 3894 // valid DeclRefExpr*s. If one of them is valid, we return it. 3895 case Stmt::ConditionalOperatorClass: { 3896 ConditionalOperator *C = cast<ConditionalOperator>(E); 3897 3898 // Handle the GNU extension for missing LHS. 3899 if (Expr *lhsExpr = C->getLHS()) { 3900 // In C++, we can have a throw-expression, which has 'void' type. 3901 if (!lhsExpr->getType()->isVoidType()) 3902 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 3903 return LHS; 3904 } 3905 3906 // In C++, we can have a throw-expression, which has 'void' type. 3907 if (C->getRHS()->getType()->isVoidType()) 3908 return NULL; 3909 3910 return EvalAddr(C->getRHS(), refVars, ParentDecl); 3911 } 3912 3913 case Stmt::BlockExprClass: 3914 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 3915 return E; // local block. 3916 return NULL; 3917 3918 case Stmt::AddrLabelExprClass: 3919 return E; // address of label. 3920 3921 case Stmt::ExprWithCleanupsClass: 3922 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 3923 ParentDecl); 3924 3925 // For casts, we need to handle conversions from arrays to 3926 // pointer values, and pointer-to-pointer conversions. 3927 case Stmt::ImplicitCastExprClass: 3928 case Stmt::CStyleCastExprClass: 3929 case Stmt::CXXFunctionalCastExprClass: 3930 case Stmt::ObjCBridgedCastExprClass: 3931 case Stmt::CXXStaticCastExprClass: 3932 case Stmt::CXXDynamicCastExprClass: 3933 case Stmt::CXXConstCastExprClass: 3934 case Stmt::CXXReinterpretCastExprClass: { 3935 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 3936 switch (cast<CastExpr>(E)->getCastKind()) { 3937 case CK_BitCast: 3938 case CK_LValueToRValue: 3939 case CK_NoOp: 3940 case CK_BaseToDerived: 3941 case CK_DerivedToBase: 3942 case CK_UncheckedDerivedToBase: 3943 case CK_Dynamic: 3944 case CK_CPointerToObjCPointerCast: 3945 case CK_BlockPointerToObjCPointerCast: 3946 case CK_AnyPointerToBlockPointerCast: 3947 return EvalAddr(SubExpr, refVars, ParentDecl); 3948 3949 case CK_ArrayToPointerDecay: 3950 return EvalVal(SubExpr, refVars, ParentDecl); 3951 3952 default: 3953 return 0; 3954 } 3955 } 3956 3957 case Stmt::MaterializeTemporaryExprClass: 3958 if (Expr *Result = EvalAddr( 3959 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3960 refVars, ParentDecl)) 3961 return Result; 3962 3963 return E; 3964 3965 // Everything else: we simply don't reason about them. 3966 default: 3967 return NULL; 3968 } 3969} 3970 3971 3972/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3973/// See the comments for EvalAddr for more details. 3974static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3975 Decl *ParentDecl) { 3976do { 3977 // We should only be called for evaluating non-pointer expressions, or 3978 // expressions with a pointer type that are not used as references but instead 3979 // are l-values (e.g., DeclRefExpr with a pointer type). 3980 3981 // Our "symbolic interpreter" is just a dispatch off the currently 3982 // viewed AST node. We then recursively traverse the AST by calling 3983 // EvalAddr and EvalVal appropriately. 3984 3985 E = E->IgnoreParens(); 3986 switch (E->getStmtClass()) { 3987 case Stmt::ImplicitCastExprClass: { 3988 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3989 if (IE->getValueKind() == VK_LValue) { 3990 E = IE->getSubExpr(); 3991 continue; 3992 } 3993 return NULL; 3994 } 3995 3996 case Stmt::ExprWithCleanupsClass: 3997 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 3998 3999 case Stmt::DeclRefExprClass: { 4000 // When we hit a DeclRefExpr we are looking at code that refers to a 4001 // variable's name. If it's not a reference variable we check if it has 4002 // local storage within the function, and if so, return the expression. 4003 DeclRefExpr *DR = cast<DeclRefExpr>(E); 4004 4005 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 4006 // Check if it refers to itself, e.g. "int& i = i;". 4007 if (V == ParentDecl) 4008 return DR; 4009 4010 if (V->hasLocalStorage()) { 4011 if (!V->getType()->isReferenceType()) 4012 return DR; 4013 4014 // Reference variable, follow through to the expression that 4015 // it points to. 4016 if (V->hasInit()) { 4017 // Add the reference variable to the "trail". 4018 refVars.push_back(DR); 4019 return EvalVal(V->getInit(), refVars, V); 4020 } 4021 } 4022 } 4023 4024 return NULL; 4025 } 4026 4027 case Stmt::UnaryOperatorClass: { 4028 // The only unary operator that make sense to handle here 4029 // is Deref. All others don't resolve to a "name." This includes 4030 // handling all sorts of rvalues passed to a unary operator. 4031 UnaryOperator *U = cast<UnaryOperator>(E); 4032 4033 if (U->getOpcode() == UO_Deref) 4034 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 4035 4036 return NULL; 4037 } 4038 4039 case Stmt::ArraySubscriptExprClass: { 4040 // Array subscripts are potential references to data on the stack. We 4041 // retrieve the DeclRefExpr* for the array variable if it indeed 4042 // has local storage. 4043 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 4044 } 4045 4046 case Stmt::ConditionalOperatorClass: { 4047 // For conditional operators we need to see if either the LHS or RHS are 4048 // non-NULL Expr's. If one is non-NULL, we return it. 4049 ConditionalOperator *C = cast<ConditionalOperator>(E); 4050 4051 // Handle the GNU extension for missing LHS. 4052 if (Expr *lhsExpr = C->getLHS()) 4053 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 4054 return LHS; 4055 4056 return EvalVal(C->getRHS(), refVars, ParentDecl); 4057 } 4058 4059 // Accesses to members are potential references to data on the stack. 4060 case Stmt::MemberExprClass: { 4061 MemberExpr *M = cast<MemberExpr>(E); 4062 4063 // Check for indirect access. We only want direct field accesses. 4064 if (M->isArrow()) 4065 return NULL; 4066 4067 // Check whether the member type is itself a reference, in which case 4068 // we're not going to refer to the member, but to what the member refers to. 4069 if (M->getMemberDecl()->getType()->isReferenceType()) 4070 return NULL; 4071 4072 return EvalVal(M->getBase(), refVars, ParentDecl); 4073 } 4074 4075 case Stmt::MaterializeTemporaryExprClass: 4076 if (Expr *Result = EvalVal( 4077 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 4078 refVars, ParentDecl)) 4079 return Result; 4080 4081 return E; 4082 4083 default: 4084 // Check that we don't return or take the address of a reference to a 4085 // temporary. This is only useful in C++. 4086 if (!E->isTypeDependent() && E->isRValue()) 4087 return E; 4088 4089 // Everything else: we simply don't reason about them. 4090 return NULL; 4091 } 4092} while (true); 4093} 4094 4095//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 4096 4097/// Check for comparisons of floating point operands using != and ==. 4098/// Issue a warning if these are no self-comparisons, as they are not likely 4099/// to do what the programmer intended. 4100void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 4101 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 4102 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 4103 4104 // Special case: check for x == x (which is OK). 4105 // Do not emit warnings for such cases. 4106 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 4107 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 4108 if (DRL->getDecl() == DRR->getDecl()) 4109 return; 4110 4111 4112 // Special case: check for comparisons against literals that can be exactly 4113 // represented by APFloat. In such cases, do not emit a warning. This 4114 // is a heuristic: often comparison against such literals are used to 4115 // detect if a value in a variable has not changed. This clearly can 4116 // lead to false negatives. 4117 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 4118 if (FLL->isExact()) 4119 return; 4120 } else 4121 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)) 4122 if (FLR->isExact()) 4123 return; 4124 4125 // Check for comparisons with builtin types. 4126 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 4127 if (CL->isBuiltinCall()) 4128 return; 4129 4130 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 4131 if (CR->isBuiltinCall()) 4132 return; 4133 4134 // Emit the diagnostic. 4135 Diag(Loc, diag::warn_floatingpoint_eq) 4136 << LHS->getSourceRange() << RHS->getSourceRange(); 4137} 4138 4139//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 4140//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 4141 4142namespace { 4143 4144/// Structure recording the 'active' range of an integer-valued 4145/// expression. 4146struct IntRange { 4147 /// The number of bits active in the int. 4148 unsigned Width; 4149 4150 /// True if the int is known not to have negative values. 4151 bool NonNegative; 4152 4153 IntRange(unsigned Width, bool NonNegative) 4154 : Width(Width), NonNegative(NonNegative) 4155 {} 4156 4157 /// Returns the range of the bool type. 4158 static IntRange forBoolType() { 4159 return IntRange(1, true); 4160 } 4161 4162 /// Returns the range of an opaque value of the given integral type. 4163 static IntRange forValueOfType(ASTContext &C, QualType T) { 4164 return forValueOfCanonicalType(C, 4165 T->getCanonicalTypeInternal().getTypePtr()); 4166 } 4167 4168 /// Returns the range of an opaque value of a canonical integral type. 4169 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 4170 assert(T->isCanonicalUnqualified()); 4171 4172 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4173 T = VT->getElementType().getTypePtr(); 4174 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4175 T = CT->getElementType().getTypePtr(); 4176 4177 // For enum types, use the known bit width of the enumerators. 4178 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 4179 EnumDecl *Enum = ET->getDecl(); 4180 if (!Enum->isCompleteDefinition()) 4181 return IntRange(C.getIntWidth(QualType(T, 0)), false); 4182 4183 unsigned NumPositive = Enum->getNumPositiveBits(); 4184 unsigned NumNegative = Enum->getNumNegativeBits(); 4185 4186 if (NumNegative == 0) 4187 return IntRange(NumPositive, true/*NonNegative*/); 4188 else 4189 return IntRange(std::max(NumPositive + 1, NumNegative), 4190 false/*NonNegative*/); 4191 } 4192 4193 const BuiltinType *BT = cast<BuiltinType>(T); 4194 assert(BT->isInteger()); 4195 4196 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4197 } 4198 4199 /// Returns the "target" range of a canonical integral type, i.e. 4200 /// the range of values expressible in the type. 4201 /// 4202 /// This matches forValueOfCanonicalType except that enums have the 4203 /// full range of their type, not the range of their enumerators. 4204 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 4205 assert(T->isCanonicalUnqualified()); 4206 4207 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4208 T = VT->getElementType().getTypePtr(); 4209 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4210 T = CT->getElementType().getTypePtr(); 4211 if (const EnumType *ET = dyn_cast<EnumType>(T)) 4212 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 4213 4214 const BuiltinType *BT = cast<BuiltinType>(T); 4215 assert(BT->isInteger()); 4216 4217 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4218 } 4219 4220 /// Returns the supremum of two ranges: i.e. their conservative merge. 4221 static IntRange join(IntRange L, IntRange R) { 4222 return IntRange(std::max(L.Width, R.Width), 4223 L.NonNegative && R.NonNegative); 4224 } 4225 4226 /// Returns the infinum of two ranges: i.e. their aggressive merge. 4227 static IntRange meet(IntRange L, IntRange R) { 4228 return IntRange(std::min(L.Width, R.Width), 4229 L.NonNegative || R.NonNegative); 4230 } 4231}; 4232 4233static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 4234 unsigned MaxWidth) { 4235 if (value.isSigned() && value.isNegative()) 4236 return IntRange(value.getMinSignedBits(), false); 4237 4238 if (value.getBitWidth() > MaxWidth) 4239 value = value.trunc(MaxWidth); 4240 4241 // isNonNegative() just checks the sign bit without considering 4242 // signedness. 4243 return IntRange(value.getActiveBits(), true); 4244} 4245 4246static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 4247 unsigned MaxWidth) { 4248 if (result.isInt()) 4249 return GetValueRange(C, result.getInt(), MaxWidth); 4250 4251 if (result.isVector()) { 4252 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 4253 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 4254 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 4255 R = IntRange::join(R, El); 4256 } 4257 return R; 4258 } 4259 4260 if (result.isComplexInt()) { 4261 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 4262 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 4263 return IntRange::join(R, I); 4264 } 4265 4266 // This can happen with lossless casts to intptr_t of "based" lvalues. 4267 // Assume it might use arbitrary bits. 4268 // FIXME: The only reason we need to pass the type in here is to get 4269 // the sign right on this one case. It would be nice if APValue 4270 // preserved this. 4271 assert(result.isLValue() || result.isAddrLabelDiff()); 4272 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 4273} 4274 4275static QualType GetExprType(Expr *E) { 4276 QualType Ty = E->getType(); 4277 if (const AtomicType *AtomicRHS = Ty->getAs<AtomicType>()) 4278 Ty = AtomicRHS->getValueType(); 4279 return Ty; 4280} 4281 4282/// Pseudo-evaluate the given integer expression, estimating the 4283/// range of values it might take. 4284/// 4285/// \param MaxWidth - the width to which the value will be truncated 4286static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 4287 E = E->IgnoreParens(); 4288 4289 // Try a full evaluation first. 4290 Expr::EvalResult result; 4291 if (E->EvaluateAsRValue(result, C)) 4292 return GetValueRange(C, result.Val, GetExprType(E), MaxWidth); 4293 4294 // I think we only want to look through implicit casts here; if the 4295 // user has an explicit widening cast, we should treat the value as 4296 // being of the new, wider type. 4297 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 4298 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 4299 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 4300 4301 IntRange OutputTypeRange = IntRange::forValueOfType(C, GetExprType(CE)); 4302 4303 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 4304 4305 // Assume that non-integer casts can span the full range of the type. 4306 if (!isIntegerCast) 4307 return OutputTypeRange; 4308 4309 IntRange SubRange 4310 = GetExprRange(C, CE->getSubExpr(), 4311 std::min(MaxWidth, OutputTypeRange.Width)); 4312 4313 // Bail out if the subexpr's range is as wide as the cast type. 4314 if (SubRange.Width >= OutputTypeRange.Width) 4315 return OutputTypeRange; 4316 4317 // Otherwise, we take the smaller width, and we're non-negative if 4318 // either the output type or the subexpr is. 4319 return IntRange(SubRange.Width, 4320 SubRange.NonNegative || OutputTypeRange.NonNegative); 4321 } 4322 4323 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 4324 // If we can fold the condition, just take that operand. 4325 bool CondResult; 4326 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 4327 return GetExprRange(C, CondResult ? CO->getTrueExpr() 4328 : CO->getFalseExpr(), 4329 MaxWidth); 4330 4331 // Otherwise, conservatively merge. 4332 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 4333 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 4334 return IntRange::join(L, R); 4335 } 4336 4337 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4338 switch (BO->getOpcode()) { 4339 4340 // Boolean-valued operations are single-bit and positive. 4341 case BO_LAnd: 4342 case BO_LOr: 4343 case BO_LT: 4344 case BO_GT: 4345 case BO_LE: 4346 case BO_GE: 4347 case BO_EQ: 4348 case BO_NE: 4349 return IntRange::forBoolType(); 4350 4351 // The type of the assignments is the type of the LHS, so the RHS 4352 // is not necessarily the same type. 4353 case BO_MulAssign: 4354 case BO_DivAssign: 4355 case BO_RemAssign: 4356 case BO_AddAssign: 4357 case BO_SubAssign: 4358 case BO_XorAssign: 4359 case BO_OrAssign: 4360 // TODO: bitfields? 4361 return IntRange::forValueOfType(C, GetExprType(E)); 4362 4363 // Simple assignments just pass through the RHS, which will have 4364 // been coerced to the LHS type. 4365 case BO_Assign: 4366 // TODO: bitfields? 4367 return GetExprRange(C, BO->getRHS(), MaxWidth); 4368 4369 // Operations with opaque sources are black-listed. 4370 case BO_PtrMemD: 4371 case BO_PtrMemI: 4372 return IntRange::forValueOfType(C, GetExprType(E)); 4373 4374 // Bitwise-and uses the *infinum* of the two source ranges. 4375 case BO_And: 4376 case BO_AndAssign: 4377 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 4378 GetExprRange(C, BO->getRHS(), MaxWidth)); 4379 4380 // Left shift gets black-listed based on a judgement call. 4381 case BO_Shl: 4382 // ...except that we want to treat '1 << (blah)' as logically 4383 // positive. It's an important idiom. 4384 if (IntegerLiteral *I 4385 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 4386 if (I->getValue() == 1) { 4387 IntRange R = IntRange::forValueOfType(C, GetExprType(E)); 4388 return IntRange(R.Width, /*NonNegative*/ true); 4389 } 4390 } 4391 // fallthrough 4392 4393 case BO_ShlAssign: 4394 return IntRange::forValueOfType(C, GetExprType(E)); 4395 4396 // Right shift by a constant can narrow its left argument. 4397 case BO_Shr: 4398 case BO_ShrAssign: { 4399 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4400 4401 // If the shift amount is a positive constant, drop the width by 4402 // that much. 4403 llvm::APSInt shift; 4404 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 4405 shift.isNonNegative()) { 4406 unsigned zext = shift.getZExtValue(); 4407 if (zext >= L.Width) 4408 L.Width = (L.NonNegative ? 0 : 1); 4409 else 4410 L.Width -= zext; 4411 } 4412 4413 return L; 4414 } 4415 4416 // Comma acts as its right operand. 4417 case BO_Comma: 4418 return GetExprRange(C, BO->getRHS(), MaxWidth); 4419 4420 // Black-list pointer subtractions. 4421 case BO_Sub: 4422 if (BO->getLHS()->getType()->isPointerType()) 4423 return IntRange::forValueOfType(C, GetExprType(E)); 4424 break; 4425 4426 // The width of a division result is mostly determined by the size 4427 // of the LHS. 4428 case BO_Div: { 4429 // Don't 'pre-truncate' the operands. 4430 unsigned opWidth = C.getIntWidth(GetExprType(E)); 4431 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4432 4433 // If the divisor is constant, use that. 4434 llvm::APSInt divisor; 4435 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 4436 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 4437 if (log2 >= L.Width) 4438 L.Width = (L.NonNegative ? 0 : 1); 4439 else 4440 L.Width = std::min(L.Width - log2, MaxWidth); 4441 return L; 4442 } 4443 4444 // Otherwise, just use the LHS's width. 4445 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4446 return IntRange(L.Width, L.NonNegative && R.NonNegative); 4447 } 4448 4449 // The result of a remainder can't be larger than the result of 4450 // either side. 4451 case BO_Rem: { 4452 // Don't 'pre-truncate' the operands. 4453 unsigned opWidth = C.getIntWidth(GetExprType(E)); 4454 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4455 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4456 4457 IntRange meet = IntRange::meet(L, R); 4458 meet.Width = std::min(meet.Width, MaxWidth); 4459 return meet; 4460 } 4461 4462 // The default behavior is okay for these. 4463 case BO_Mul: 4464 case BO_Add: 4465 case BO_Xor: 4466 case BO_Or: 4467 break; 4468 } 4469 4470 // The default case is to treat the operation as if it were closed 4471 // on the narrowest type that encompasses both operands. 4472 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4473 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 4474 return IntRange::join(L, R); 4475 } 4476 4477 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 4478 switch (UO->getOpcode()) { 4479 // Boolean-valued operations are white-listed. 4480 case UO_LNot: 4481 return IntRange::forBoolType(); 4482 4483 // Operations with opaque sources are black-listed. 4484 case UO_Deref: 4485 case UO_AddrOf: // should be impossible 4486 return IntRange::forValueOfType(C, GetExprType(E)); 4487 4488 default: 4489 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 4490 } 4491 } 4492 4493 if (FieldDecl *BitField = E->getSourceBitField()) 4494 return IntRange(BitField->getBitWidthValue(C), 4495 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 4496 4497 return IntRange::forValueOfType(C, GetExprType(E)); 4498} 4499 4500static IntRange GetExprRange(ASTContext &C, Expr *E) { 4501 return GetExprRange(C, E, C.getIntWidth(GetExprType(E))); 4502} 4503 4504/// Checks whether the given value, which currently has the given 4505/// source semantics, has the same value when coerced through the 4506/// target semantics. 4507static bool IsSameFloatAfterCast(const llvm::APFloat &value, 4508 const llvm::fltSemantics &Src, 4509 const llvm::fltSemantics &Tgt) { 4510 llvm::APFloat truncated = value; 4511 4512 bool ignored; 4513 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 4514 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 4515 4516 return truncated.bitwiseIsEqual(value); 4517} 4518 4519/// Checks whether the given value, which currently has the given 4520/// source semantics, has the same value when coerced through the 4521/// target semantics. 4522/// 4523/// The value might be a vector of floats (or a complex number). 4524static bool IsSameFloatAfterCast(const APValue &value, 4525 const llvm::fltSemantics &Src, 4526 const llvm::fltSemantics &Tgt) { 4527 if (value.isFloat()) 4528 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 4529 4530 if (value.isVector()) { 4531 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 4532 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 4533 return false; 4534 return true; 4535 } 4536 4537 assert(value.isComplexFloat()); 4538 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 4539 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 4540} 4541 4542static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 4543 4544static bool IsZero(Sema &S, Expr *E) { 4545 // Suppress cases where we are comparing against an enum constant. 4546 if (const DeclRefExpr *DR = 4547 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 4548 if (isa<EnumConstantDecl>(DR->getDecl())) 4549 return false; 4550 4551 // Suppress cases where the '0' value is expanded from a macro. 4552 if (E->getLocStart().isMacroID()) 4553 return false; 4554 4555 llvm::APSInt Value; 4556 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 4557} 4558 4559static bool HasEnumType(Expr *E) { 4560 // Strip off implicit integral promotions. 4561 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 4562 if (ICE->getCastKind() != CK_IntegralCast && 4563 ICE->getCastKind() != CK_NoOp) 4564 break; 4565 E = ICE->getSubExpr(); 4566 } 4567 4568 return E->getType()->isEnumeralType(); 4569} 4570 4571static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 4572 BinaryOperatorKind op = E->getOpcode(); 4573 if (E->isValueDependent()) 4574 return; 4575 4576 if (op == BO_LT && IsZero(S, E->getRHS())) { 4577 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4578 << "< 0" << "false" << HasEnumType(E->getLHS()) 4579 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4580 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 4581 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4582 << ">= 0" << "true" << HasEnumType(E->getLHS()) 4583 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4584 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 4585 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4586 << "0 >" << "false" << HasEnumType(E->getRHS()) 4587 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4588 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 4589 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4590 << "0 <=" << "true" << HasEnumType(E->getRHS()) 4591 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4592 } 4593} 4594 4595static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E, 4596 Expr *Constant, Expr *Other, 4597 llvm::APSInt Value, 4598 bool RhsConstant) { 4599 // 0 values are handled later by CheckTrivialUnsignedComparison(). 4600 if (Value == 0) 4601 return; 4602 4603 BinaryOperatorKind op = E->getOpcode(); 4604 QualType OtherT = Other->getType(); 4605 QualType ConstantT = Constant->getType(); 4606 QualType CommonT = E->getLHS()->getType(); 4607 if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT)) 4608 return; 4609 assert((OtherT->isIntegerType() && ConstantT->isIntegerType()) 4610 && "comparison with non-integer type"); 4611 4612 bool ConstantSigned = ConstantT->isSignedIntegerType(); 4613 bool CommonSigned = CommonT->isSignedIntegerType(); 4614 4615 bool EqualityOnly = false; 4616 4617 // TODO: Investigate using GetExprRange() to get tighter bounds on 4618 // on the bit ranges. 4619 IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT); 4620 unsigned OtherWidth = OtherRange.Width; 4621 4622 if (CommonSigned) { 4623 // The common type is signed, therefore no signed to unsigned conversion. 4624 if (!OtherRange.NonNegative) { 4625 // Check that the constant is representable in type OtherT. 4626 if (ConstantSigned) { 4627 if (OtherWidth >= Value.getMinSignedBits()) 4628 return; 4629 } else { // !ConstantSigned 4630 if (OtherWidth >= Value.getActiveBits() + 1) 4631 return; 4632 } 4633 } else { // !OtherSigned 4634 // Check that the constant is representable in type OtherT. 4635 // Negative values are out of range. 4636 if (ConstantSigned) { 4637 if (Value.isNonNegative() && OtherWidth >= Value.getActiveBits()) 4638 return; 4639 } else { // !ConstantSigned 4640 if (OtherWidth >= Value.getActiveBits()) 4641 return; 4642 } 4643 } 4644 } else { // !CommonSigned 4645 if (OtherRange.NonNegative) { 4646 if (OtherWidth >= Value.getActiveBits()) 4647 return; 4648 } else if (!OtherRange.NonNegative && !ConstantSigned) { 4649 // Check to see if the constant is representable in OtherT. 4650 if (OtherWidth > Value.getActiveBits()) 4651 return; 4652 // Check to see if the constant is equivalent to a negative value 4653 // cast to CommonT. 4654 if (S.Context.getIntWidth(ConstantT) == S.Context.getIntWidth(CommonT) && 4655 Value.isNegative() && Value.getMinSignedBits() <= OtherWidth) 4656 return; 4657 // The constant value rests between values that OtherT can represent after 4658 // conversion. Relational comparison still works, but equality 4659 // comparisons will be tautological. 4660 EqualityOnly = true; 4661 } else { // OtherSigned && ConstantSigned 4662 assert(0 && "Two signed types converted to unsigned types."); 4663 } 4664 } 4665 4666 bool PositiveConstant = !ConstantSigned || Value.isNonNegative(); 4667 4668 bool IsTrue = true; 4669 if (op == BO_EQ || op == BO_NE) { 4670 IsTrue = op == BO_NE; 4671 } else if (EqualityOnly) { 4672 return; 4673 } else if (RhsConstant) { 4674 if (op == BO_GT || op == BO_GE) 4675 IsTrue = !PositiveConstant; 4676 else // op == BO_LT || op == BO_LE 4677 IsTrue = PositiveConstant; 4678 } else { 4679 if (op == BO_LT || op == BO_LE) 4680 IsTrue = !PositiveConstant; 4681 else // op == BO_GT || op == BO_GE 4682 IsTrue = PositiveConstant; 4683 } 4684 4685 // If this is a comparison to an enum constant, include that 4686 // constant in the diagnostic. 4687 const EnumConstantDecl *ED = 0; 4688 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Constant)) 4689 ED = dyn_cast<EnumConstantDecl>(DR->getDecl()); 4690 4691 SmallString<64> PrettySourceValue; 4692 llvm::raw_svector_ostream OS(PrettySourceValue); 4693 if (ED) 4694 OS << '\'' << *ED << "' (" << Value << ")"; 4695 else 4696 OS << Value; 4697 4698 S.Diag(E->getOperatorLoc(), diag::warn_out_of_range_compare) 4699 << OS.str() << OtherT << IsTrue 4700 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4701} 4702 4703/// Analyze the operands of the given comparison. Implements the 4704/// fallback case from AnalyzeComparison. 4705static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 4706 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4707 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4708} 4709 4710/// \brief Implements -Wsign-compare. 4711/// 4712/// \param E the binary operator to check for warnings 4713static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 4714 // The type the comparison is being performed in. 4715 QualType T = E->getLHS()->getType(); 4716 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 4717 && "comparison with mismatched types"); 4718 if (E->isValueDependent()) 4719 return AnalyzeImpConvsInComparison(S, E); 4720 4721 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 4722 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 4723 4724 bool IsComparisonConstant = false; 4725 4726 // Check whether an integer constant comparison results in a value 4727 // of 'true' or 'false'. 4728 if (T->isIntegralType(S.Context)) { 4729 llvm::APSInt RHSValue; 4730 bool IsRHSIntegralLiteral = 4731 RHS->isIntegerConstantExpr(RHSValue, S.Context); 4732 llvm::APSInt LHSValue; 4733 bool IsLHSIntegralLiteral = 4734 LHS->isIntegerConstantExpr(LHSValue, S.Context); 4735 if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral) 4736 DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true); 4737 else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral) 4738 DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false); 4739 else 4740 IsComparisonConstant = 4741 (IsRHSIntegralLiteral && IsLHSIntegralLiteral); 4742 } else if (!T->hasUnsignedIntegerRepresentation()) 4743 IsComparisonConstant = E->isIntegerConstantExpr(S.Context); 4744 4745 // We don't do anything special if this isn't an unsigned integral 4746 // comparison: we're only interested in integral comparisons, and 4747 // signed comparisons only happen in cases we don't care to warn about. 4748 // 4749 // We also don't care about value-dependent expressions or expressions 4750 // whose result is a constant. 4751 if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant) 4752 return AnalyzeImpConvsInComparison(S, E); 4753 4754 // Check to see if one of the (unmodified) operands is of different 4755 // signedness. 4756 Expr *signedOperand, *unsignedOperand; 4757 if (LHS->getType()->hasSignedIntegerRepresentation()) { 4758 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 4759 "unsigned comparison between two signed integer expressions?"); 4760 signedOperand = LHS; 4761 unsignedOperand = RHS; 4762 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 4763 signedOperand = RHS; 4764 unsignedOperand = LHS; 4765 } else { 4766 CheckTrivialUnsignedComparison(S, E); 4767 return AnalyzeImpConvsInComparison(S, E); 4768 } 4769 4770 // Otherwise, calculate the effective range of the signed operand. 4771 IntRange signedRange = GetExprRange(S.Context, signedOperand); 4772 4773 // Go ahead and analyze implicit conversions in the operands. Note 4774 // that we skip the implicit conversions on both sides. 4775 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 4776 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 4777 4778 // If the signed range is non-negative, -Wsign-compare won't fire, 4779 // but we should still check for comparisons which are always true 4780 // or false. 4781 if (signedRange.NonNegative) 4782 return CheckTrivialUnsignedComparison(S, E); 4783 4784 // For (in)equality comparisons, if the unsigned operand is a 4785 // constant which cannot collide with a overflowed signed operand, 4786 // then reinterpreting the signed operand as unsigned will not 4787 // change the result of the comparison. 4788 if (E->isEqualityOp()) { 4789 unsigned comparisonWidth = S.Context.getIntWidth(T); 4790 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 4791 4792 // We should never be unable to prove that the unsigned operand is 4793 // non-negative. 4794 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 4795 4796 if (unsignedRange.Width < comparisonWidth) 4797 return; 4798 } 4799 4800 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 4801 S.PDiag(diag::warn_mixed_sign_comparison) 4802 << LHS->getType() << RHS->getType() 4803 << LHS->getSourceRange() << RHS->getSourceRange()); 4804} 4805 4806/// Analyzes an attempt to assign the given value to a bitfield. 4807/// 4808/// Returns true if there was something fishy about the attempt. 4809static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 4810 SourceLocation InitLoc) { 4811 assert(Bitfield->isBitField()); 4812 if (Bitfield->isInvalidDecl()) 4813 return false; 4814 4815 // White-list bool bitfields. 4816 if (Bitfield->getType()->isBooleanType()) 4817 return false; 4818 4819 // Ignore value- or type-dependent expressions. 4820 if (Bitfield->getBitWidth()->isValueDependent() || 4821 Bitfield->getBitWidth()->isTypeDependent() || 4822 Init->isValueDependent() || 4823 Init->isTypeDependent()) 4824 return false; 4825 4826 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 4827 4828 llvm::APSInt Value; 4829 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 4830 return false; 4831 4832 unsigned OriginalWidth = Value.getBitWidth(); 4833 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 4834 4835 if (OriginalWidth <= FieldWidth) 4836 return false; 4837 4838 // Compute the value which the bitfield will contain. 4839 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 4840 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 4841 4842 // Check whether the stored value is equal to the original value. 4843 TruncatedValue = TruncatedValue.extend(OriginalWidth); 4844 if (llvm::APSInt::isSameValue(Value, TruncatedValue)) 4845 return false; 4846 4847 // Special-case bitfields of width 1: booleans are naturally 0/1, and 4848 // therefore don't strictly fit into a signed bitfield of width 1. 4849 if (FieldWidth == 1 && Value == 1) 4850 return false; 4851 4852 std::string PrettyValue = Value.toString(10); 4853 std::string PrettyTrunc = TruncatedValue.toString(10); 4854 4855 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 4856 << PrettyValue << PrettyTrunc << OriginalInit->getType() 4857 << Init->getSourceRange(); 4858 4859 return true; 4860} 4861 4862/// Analyze the given simple or compound assignment for warning-worthy 4863/// operations. 4864static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 4865 // Just recurse on the LHS. 4866 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4867 4868 // We want to recurse on the RHS as normal unless we're assigning to 4869 // a bitfield. 4870 if (FieldDecl *Bitfield = E->getLHS()->getSourceBitField()) { 4871 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 4872 E->getOperatorLoc())) { 4873 // Recurse, ignoring any implicit conversions on the RHS. 4874 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 4875 E->getOperatorLoc()); 4876 } 4877 } 4878 4879 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4880} 4881 4882/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4883static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 4884 SourceLocation CContext, unsigned diag, 4885 bool pruneControlFlow = false) { 4886 if (pruneControlFlow) { 4887 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4888 S.PDiag(diag) 4889 << SourceType << T << E->getSourceRange() 4890 << SourceRange(CContext)); 4891 return; 4892 } 4893 S.Diag(E->getExprLoc(), diag) 4894 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 4895} 4896 4897/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4898static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 4899 SourceLocation CContext, unsigned diag, 4900 bool pruneControlFlow = false) { 4901 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 4902} 4903 4904/// Diagnose an implicit cast from a literal expression. Does not warn when the 4905/// cast wouldn't lose information. 4906void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 4907 SourceLocation CContext) { 4908 // Try to convert the literal exactly to an integer. If we can, don't warn. 4909 bool isExact = false; 4910 const llvm::APFloat &Value = FL->getValue(); 4911 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 4912 T->hasUnsignedIntegerRepresentation()); 4913 if (Value.convertToInteger(IntegerValue, 4914 llvm::APFloat::rmTowardZero, &isExact) 4915 == llvm::APFloat::opOK && isExact) 4916 return; 4917 4918 SmallString<16> PrettySourceValue; 4919 Value.toString(PrettySourceValue); 4920 SmallString<16> PrettyTargetValue; 4921 if (T->isSpecificBuiltinType(BuiltinType::Bool)) 4922 PrettyTargetValue = IntegerValue == 0 ? "false" : "true"; 4923 else 4924 IntegerValue.toString(PrettyTargetValue); 4925 4926 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 4927 << FL->getType() << T.getUnqualifiedType() << PrettySourceValue 4928 << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext); 4929} 4930 4931std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 4932 if (!Range.Width) return "0"; 4933 4934 llvm::APSInt ValueInRange = Value; 4935 ValueInRange.setIsSigned(!Range.NonNegative); 4936 ValueInRange = ValueInRange.trunc(Range.Width); 4937 return ValueInRange.toString(10); 4938} 4939 4940static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) { 4941 if (!isa<ImplicitCastExpr>(Ex)) 4942 return false; 4943 4944 Expr *InnerE = Ex->IgnoreParenImpCasts(); 4945 const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr(); 4946 const Type *Source = 4947 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 4948 if (Target->isDependentType()) 4949 return false; 4950 4951 const BuiltinType *FloatCandidateBT = 4952 dyn_cast<BuiltinType>(ToBool ? Source : Target); 4953 const Type *BoolCandidateType = ToBool ? Target : Source; 4954 4955 return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) && 4956 FloatCandidateBT && (FloatCandidateBT->isFloatingPoint())); 4957} 4958 4959void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall, 4960 SourceLocation CC) { 4961 unsigned NumArgs = TheCall->getNumArgs(); 4962 for (unsigned i = 0; i < NumArgs; ++i) { 4963 Expr *CurrA = TheCall->getArg(i); 4964 if (!IsImplicitBoolFloatConversion(S, CurrA, true)) 4965 continue; 4966 4967 bool IsSwapped = ((i > 0) && 4968 IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false)); 4969 IsSwapped |= ((i < (NumArgs - 1)) && 4970 IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false)); 4971 if (IsSwapped) { 4972 // Warn on this floating-point to bool conversion. 4973 DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(), 4974 CurrA->getType(), CC, 4975 diag::warn_impcast_floating_point_to_bool); 4976 } 4977 } 4978} 4979 4980void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 4981 SourceLocation CC, bool *ICContext = 0) { 4982 if (E->isTypeDependent() || E->isValueDependent()) return; 4983 4984 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 4985 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 4986 if (Source == Target) return; 4987 if (Target->isDependentType()) return; 4988 4989 // If the conversion context location is invalid don't complain. We also 4990 // don't want to emit a warning if the issue occurs from the expansion of 4991 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 4992 // delay this check as long as possible. Once we detect we are in that 4993 // scenario, we just return. 4994 if (CC.isInvalid()) 4995 return; 4996 4997 // Diagnose implicit casts to bool. 4998 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 4999 if (isa<StringLiteral>(E)) 5000 // Warn on string literal to bool. Checks for string literals in logical 5001 // expressions, for instances, assert(0 && "error here"), is prevented 5002 // by a check in AnalyzeImplicitConversions(). 5003 return DiagnoseImpCast(S, E, T, CC, 5004 diag::warn_impcast_string_literal_to_bool); 5005 if (Source->isFunctionType()) { 5006 // Warn on function to bool. Checks free functions and static member 5007 // functions. Weakly imported functions are excluded from the check, 5008 // since it's common to test their value to check whether the linker 5009 // found a definition for them. 5010 ValueDecl *D = 0; 5011 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 5012 D = R->getDecl(); 5013 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 5014 D = M->getMemberDecl(); 5015 } 5016 5017 if (D && !D->isWeak()) { 5018 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 5019 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 5020 << F << E->getSourceRange() << SourceRange(CC); 5021 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 5022 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 5023 QualType ReturnType; 5024 UnresolvedSet<4> NonTemplateOverloads; 5025 S.tryExprAsCall(*E, ReturnType, NonTemplateOverloads); 5026 if (!ReturnType.isNull() 5027 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 5028 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 5029 << FixItHint::CreateInsertion( 5030 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 5031 return; 5032 } 5033 } 5034 } 5035 } 5036 5037 // Strip vector types. 5038 if (isa<VectorType>(Source)) { 5039 if (!isa<VectorType>(Target)) { 5040 if (S.SourceMgr.isInSystemMacro(CC)) 5041 return; 5042 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 5043 } 5044 5045 // If the vector cast is cast between two vectors of the same size, it is 5046 // a bitcast, not a conversion. 5047 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 5048 return; 5049 5050 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 5051 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 5052 } 5053 5054 // Strip complex types. 5055 if (isa<ComplexType>(Source)) { 5056 if (!isa<ComplexType>(Target)) { 5057 if (S.SourceMgr.isInSystemMacro(CC)) 5058 return; 5059 5060 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 5061 } 5062 5063 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 5064 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 5065 } 5066 5067 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 5068 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 5069 5070 // If the source is floating point... 5071 if (SourceBT && SourceBT->isFloatingPoint()) { 5072 // ...and the target is floating point... 5073 if (TargetBT && TargetBT->isFloatingPoint()) { 5074 // ...then warn if we're dropping FP rank. 5075 5076 // Builtin FP kinds are ordered by increasing FP rank. 5077 if (SourceBT->getKind() > TargetBT->getKind()) { 5078 // Don't warn about float constants that are precisely 5079 // representable in the target type. 5080 Expr::EvalResult result; 5081 if (E->EvaluateAsRValue(result, S.Context)) { 5082 // Value might be a float, a float vector, or a float complex. 5083 if (IsSameFloatAfterCast(result.Val, 5084 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 5085 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 5086 return; 5087 } 5088 5089 if (S.SourceMgr.isInSystemMacro(CC)) 5090 return; 5091 5092 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 5093 } 5094 return; 5095 } 5096 5097 // If the target is integral, always warn. 5098 if (TargetBT && TargetBT->isInteger()) { 5099 if (S.SourceMgr.isInSystemMacro(CC)) 5100 return; 5101 5102 Expr *InnerE = E->IgnoreParenImpCasts(); 5103 // We also want to warn on, e.g., "int i = -1.234" 5104 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 5105 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 5106 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 5107 5108 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 5109 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 5110 } else { 5111 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 5112 } 5113 } 5114 5115 // If the target is bool, warn if expr is a function or method call. 5116 if (Target->isSpecificBuiltinType(BuiltinType::Bool) && 5117 isa<CallExpr>(E)) { 5118 // Check last argument of function call to see if it is an 5119 // implicit cast from a type matching the type the result 5120 // is being cast to. 5121 CallExpr *CEx = cast<CallExpr>(E); 5122 unsigned NumArgs = CEx->getNumArgs(); 5123 if (NumArgs > 0) { 5124 Expr *LastA = CEx->getArg(NumArgs - 1); 5125 Expr *InnerE = LastA->IgnoreParenImpCasts(); 5126 const Type *InnerType = 5127 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 5128 if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) { 5129 // Warn on this floating-point to bool conversion 5130 DiagnoseImpCast(S, E, T, CC, 5131 diag::warn_impcast_floating_point_to_bool); 5132 } 5133 } 5134 } 5135 return; 5136 } 5137 5138 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 5139 == Expr::NPCK_GNUNull) && !Target->isAnyPointerType() 5140 && !Target->isBlockPointerType() && !Target->isMemberPointerType() 5141 && Target->isScalarType() && !Target->isNullPtrType()) { 5142 SourceLocation Loc = E->getSourceRange().getBegin(); 5143 if (Loc.isMacroID()) 5144 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 5145 if (!Loc.isMacroID() || CC.isMacroID()) 5146 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 5147 << T << clang::SourceRange(CC) 5148 << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T)); 5149 } 5150 5151 if (!Source->isIntegerType() || !Target->isIntegerType()) 5152 return; 5153 5154 // TODO: remove this early return once the false positives for constant->bool 5155 // in templates, macros, etc, are reduced or removed. 5156 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 5157 return; 5158 5159 IntRange SourceRange = GetExprRange(S.Context, E); 5160 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 5161 5162 if (SourceRange.Width > TargetRange.Width) { 5163 // If the source is a constant, use a default-on diagnostic. 5164 // TODO: this should happen for bitfield stores, too. 5165 llvm::APSInt Value(32); 5166 if (E->isIntegerConstantExpr(Value, S.Context)) { 5167 if (S.SourceMgr.isInSystemMacro(CC)) 5168 return; 5169 5170 std::string PrettySourceValue = Value.toString(10); 5171 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 5172 5173 S.DiagRuntimeBehavior(E->getExprLoc(), E, 5174 S.PDiag(diag::warn_impcast_integer_precision_constant) 5175 << PrettySourceValue << PrettyTargetValue 5176 << E->getType() << T << E->getSourceRange() 5177 << clang::SourceRange(CC)); 5178 return; 5179 } 5180 5181 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 5182 if (S.SourceMgr.isInSystemMacro(CC)) 5183 return; 5184 5185 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 5186 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 5187 /* pruneControlFlow */ true); 5188 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 5189 } 5190 5191 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 5192 (!TargetRange.NonNegative && SourceRange.NonNegative && 5193 SourceRange.Width == TargetRange.Width)) { 5194 5195 if (S.SourceMgr.isInSystemMacro(CC)) 5196 return; 5197 5198 unsigned DiagID = diag::warn_impcast_integer_sign; 5199 5200 // Traditionally, gcc has warned about this under -Wsign-compare. 5201 // We also want to warn about it in -Wconversion. 5202 // So if -Wconversion is off, use a completely identical diagnostic 5203 // in the sign-compare group. 5204 // The conditional-checking code will 5205 if (ICContext) { 5206 DiagID = diag::warn_impcast_integer_sign_conditional; 5207 *ICContext = true; 5208 } 5209 5210 return DiagnoseImpCast(S, E, T, CC, DiagID); 5211 } 5212 5213 // Diagnose conversions between different enumeration types. 5214 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 5215 // type, to give us better diagnostics. 5216 QualType SourceType = E->getType(); 5217 if (!S.getLangOpts().CPlusPlus) { 5218 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5219 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 5220 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 5221 SourceType = S.Context.getTypeDeclType(Enum); 5222 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 5223 } 5224 } 5225 5226 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 5227 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 5228 if (SourceEnum->getDecl()->hasNameForLinkage() && 5229 TargetEnum->getDecl()->hasNameForLinkage() && 5230 SourceEnum != TargetEnum) { 5231 if (S.SourceMgr.isInSystemMacro(CC)) 5232 return; 5233 5234 return DiagnoseImpCast(S, E, SourceType, T, CC, 5235 diag::warn_impcast_different_enum_types); 5236 } 5237 5238 return; 5239} 5240 5241void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5242 SourceLocation CC, QualType T); 5243 5244void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 5245 SourceLocation CC, bool &ICContext) { 5246 E = E->IgnoreParenImpCasts(); 5247 5248 if (isa<ConditionalOperator>(E)) 5249 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T); 5250 5251 AnalyzeImplicitConversions(S, E, CC); 5252 if (E->getType() != T) 5253 return CheckImplicitConversion(S, E, T, CC, &ICContext); 5254 return; 5255} 5256 5257void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5258 SourceLocation CC, QualType T) { 5259 AnalyzeImplicitConversions(S, E->getCond(), CC); 5260 5261 bool Suspicious = false; 5262 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 5263 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 5264 5265 // If -Wconversion would have warned about either of the candidates 5266 // for a signedness conversion to the context type... 5267 if (!Suspicious) return; 5268 5269 // ...but it's currently ignored... 5270 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 5271 CC)) 5272 return; 5273 5274 // ...then check whether it would have warned about either of the 5275 // candidates for a signedness conversion to the condition type. 5276 if (E->getType() == T) return; 5277 5278 Suspicious = false; 5279 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 5280 E->getType(), CC, &Suspicious); 5281 if (!Suspicious) 5282 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 5283 E->getType(), CC, &Suspicious); 5284} 5285 5286/// AnalyzeImplicitConversions - Find and report any interesting 5287/// implicit conversions in the given expression. There are a couple 5288/// of competing diagnostics here, -Wconversion and -Wsign-compare. 5289void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 5290 QualType T = OrigE->getType(); 5291 Expr *E = OrigE->IgnoreParenImpCasts(); 5292 5293 if (E->isTypeDependent() || E->isValueDependent()) 5294 return; 5295 5296 // For conditional operators, we analyze the arguments as if they 5297 // were being fed directly into the output. 5298 if (isa<ConditionalOperator>(E)) { 5299 ConditionalOperator *CO = cast<ConditionalOperator>(E); 5300 CheckConditionalOperator(S, CO, CC, T); 5301 return; 5302 } 5303 5304 // Check implicit argument conversions for function calls. 5305 if (CallExpr *Call = dyn_cast<CallExpr>(E)) 5306 CheckImplicitArgumentConversions(S, Call, CC); 5307 5308 // Go ahead and check any implicit conversions we might have skipped. 5309 // The non-canonical typecheck is just an optimization; 5310 // CheckImplicitConversion will filter out dead implicit conversions. 5311 if (E->getType() != T) 5312 CheckImplicitConversion(S, E, T, CC); 5313 5314 // Now continue drilling into this expression. 5315 5316 if (PseudoObjectExpr * POE = dyn_cast<PseudoObjectExpr>(E)) { 5317 if (POE->getResultExpr()) 5318 E = POE->getResultExpr(); 5319 } 5320 5321 if (const OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E)) 5322 return AnalyzeImplicitConversions(S, OVE->getSourceExpr(), CC); 5323 5324 // Skip past explicit casts. 5325 if (isa<ExplicitCastExpr>(E)) { 5326 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 5327 return AnalyzeImplicitConversions(S, E, CC); 5328 } 5329 5330 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5331 // Do a somewhat different check with comparison operators. 5332 if (BO->isComparisonOp()) 5333 return AnalyzeComparison(S, BO); 5334 5335 // And with simple assignments. 5336 if (BO->getOpcode() == BO_Assign) 5337 return AnalyzeAssignment(S, BO); 5338 } 5339 5340 // These break the otherwise-useful invariant below. Fortunately, 5341 // we don't really need to recurse into them, because any internal 5342 // expressions should have been analyzed already when they were 5343 // built into statements. 5344 if (isa<StmtExpr>(E)) return; 5345 5346 // Don't descend into unevaluated contexts. 5347 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 5348 5349 // Now just recurse over the expression's children. 5350 CC = E->getExprLoc(); 5351 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 5352 bool IsLogicalOperator = BO && BO->isLogicalOp(); 5353 for (Stmt::child_range I = E->children(); I; ++I) { 5354 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 5355 if (!ChildExpr) 5356 continue; 5357 5358 if (IsLogicalOperator && 5359 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 5360 // Ignore checking string literals that are in logical operators. 5361 continue; 5362 AnalyzeImplicitConversions(S, ChildExpr, CC); 5363 } 5364} 5365 5366} // end anonymous namespace 5367 5368/// Diagnoses "dangerous" implicit conversions within the given 5369/// expression (which is a full expression). Implements -Wconversion 5370/// and -Wsign-compare. 5371/// 5372/// \param CC the "context" location of the implicit conversion, i.e. 5373/// the most location of the syntactic entity requiring the implicit 5374/// conversion 5375void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 5376 // Don't diagnose in unevaluated contexts. 5377 if (isUnevaluatedContext()) 5378 return; 5379 5380 // Don't diagnose for value- or type-dependent expressions. 5381 if (E->isTypeDependent() || E->isValueDependent()) 5382 return; 5383 5384 // Check for array bounds violations in cases where the check isn't triggered 5385 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 5386 // ArraySubscriptExpr is on the RHS of a variable initialization. 5387 CheckArrayAccess(E); 5388 5389 // This is not the right CC for (e.g.) a variable initialization. 5390 AnalyzeImplicitConversions(*this, E, CC); 5391} 5392 5393/// Diagnose when expression is an integer constant expression and its evaluation 5394/// results in integer overflow 5395void Sema::CheckForIntOverflow (Expr *E) { 5396 if (isa<BinaryOperator>(E->IgnoreParens())) { 5397 llvm::SmallVector<PartialDiagnosticAt, 4> Diags; 5398 E->EvaluateForOverflow(Context, &Diags); 5399 } 5400} 5401 5402namespace { 5403/// \brief Visitor for expressions which looks for unsequenced operations on the 5404/// same object. 5405class SequenceChecker : public EvaluatedExprVisitor<SequenceChecker> { 5406 typedef EvaluatedExprVisitor<SequenceChecker> Base; 5407 5408 /// \brief A tree of sequenced regions within an expression. Two regions are 5409 /// unsequenced if one is an ancestor or a descendent of the other. When we 5410 /// finish processing an expression with sequencing, such as a comma 5411 /// expression, we fold its tree nodes into its parent, since they are 5412 /// unsequenced with respect to nodes we will visit later. 5413 class SequenceTree { 5414 struct Value { 5415 explicit Value(unsigned Parent) : Parent(Parent), Merged(false) {} 5416 unsigned Parent : 31; 5417 bool Merged : 1; 5418 }; 5419 llvm::SmallVector<Value, 8> Values; 5420 5421 public: 5422 /// \brief A region within an expression which may be sequenced with respect 5423 /// to some other region. 5424 class Seq { 5425 explicit Seq(unsigned N) : Index(N) {} 5426 unsigned Index; 5427 friend class SequenceTree; 5428 public: 5429 Seq() : Index(0) {} 5430 }; 5431 5432 SequenceTree() { Values.push_back(Value(0)); } 5433 Seq root() const { return Seq(0); } 5434 5435 /// \brief Create a new sequence of operations, which is an unsequenced 5436 /// subset of \p Parent. This sequence of operations is sequenced with 5437 /// respect to other children of \p Parent. 5438 Seq allocate(Seq Parent) { 5439 Values.push_back(Value(Parent.Index)); 5440 return Seq(Values.size() - 1); 5441 } 5442 5443 /// \brief Merge a sequence of operations into its parent. 5444 void merge(Seq S) { 5445 Values[S.Index].Merged = true; 5446 } 5447 5448 /// \brief Determine whether two operations are unsequenced. This operation 5449 /// is asymmetric: \p Cur should be the more recent sequence, and \p Old 5450 /// should have been merged into its parent as appropriate. 5451 bool isUnsequenced(Seq Cur, Seq Old) { 5452 unsigned C = representative(Cur.Index); 5453 unsigned Target = representative(Old.Index); 5454 while (C >= Target) { 5455 if (C == Target) 5456 return true; 5457 C = Values[C].Parent; 5458 } 5459 return false; 5460 } 5461 5462 private: 5463 /// \brief Pick a representative for a sequence. 5464 unsigned representative(unsigned K) { 5465 if (Values[K].Merged) 5466 // Perform path compression as we go. 5467 return Values[K].Parent = representative(Values[K].Parent); 5468 return K; 5469 } 5470 }; 5471 5472 /// An object for which we can track unsequenced uses. 5473 typedef NamedDecl *Object; 5474 5475 /// Different flavors of object usage which we track. We only track the 5476 /// least-sequenced usage of each kind. 5477 enum UsageKind { 5478 /// A read of an object. Multiple unsequenced reads are OK. 5479 UK_Use, 5480 /// A modification of an object which is sequenced before the value 5481 /// computation of the expression, such as ++n in C++. 5482 UK_ModAsValue, 5483 /// A modification of an object which is not sequenced before the value 5484 /// computation of the expression, such as n++. 5485 UK_ModAsSideEffect, 5486 5487 UK_Count = UK_ModAsSideEffect + 1 5488 }; 5489 5490 struct Usage { 5491 Usage() : Use(0), Seq() {} 5492 Expr *Use; 5493 SequenceTree::Seq Seq; 5494 }; 5495 5496 struct UsageInfo { 5497 UsageInfo() : Diagnosed(false) {} 5498 Usage Uses[UK_Count]; 5499 /// Have we issued a diagnostic for this variable already? 5500 bool Diagnosed; 5501 }; 5502 typedef llvm::SmallDenseMap<Object, UsageInfo, 16> UsageInfoMap; 5503 5504 Sema &SemaRef; 5505 /// Sequenced regions within the expression. 5506 SequenceTree Tree; 5507 /// Declaration modifications and references which we have seen. 5508 UsageInfoMap UsageMap; 5509 /// The region we are currently within. 5510 SequenceTree::Seq Region; 5511 /// Filled in with declarations which were modified as a side-effect 5512 /// (that is, post-increment operations). 5513 llvm::SmallVectorImpl<std::pair<Object, Usage> > *ModAsSideEffect; 5514 /// Expressions to check later. We defer checking these to reduce 5515 /// stack usage. 5516 llvm::SmallVectorImpl<Expr*> &WorkList; 5517 5518 /// RAII object wrapping the visitation of a sequenced subexpression of an 5519 /// expression. At the end of this process, the side-effects of the evaluation 5520 /// become sequenced with respect to the value computation of the result, so 5521 /// we downgrade any UK_ModAsSideEffect within the evaluation to 5522 /// UK_ModAsValue. 5523 struct SequencedSubexpression { 5524 SequencedSubexpression(SequenceChecker &Self) 5525 : Self(Self), OldModAsSideEffect(Self.ModAsSideEffect) { 5526 Self.ModAsSideEffect = &ModAsSideEffect; 5527 } 5528 ~SequencedSubexpression() { 5529 for (unsigned I = 0, E = ModAsSideEffect.size(); I != E; ++I) { 5530 UsageInfo &U = Self.UsageMap[ModAsSideEffect[I].first]; 5531 U.Uses[UK_ModAsSideEffect] = ModAsSideEffect[I].second; 5532 Self.addUsage(U, ModAsSideEffect[I].first, 5533 ModAsSideEffect[I].second.Use, UK_ModAsValue); 5534 } 5535 Self.ModAsSideEffect = OldModAsSideEffect; 5536 } 5537 5538 SequenceChecker &Self; 5539 llvm::SmallVector<std::pair<Object, Usage>, 4> ModAsSideEffect; 5540 llvm::SmallVectorImpl<std::pair<Object, Usage> > *OldModAsSideEffect; 5541 }; 5542 5543 /// RAII object wrapping the visitation of a subexpression which we might 5544 /// choose to evaluate as a constant. If any subexpression is evaluated and 5545 /// found to be non-constant, this allows us to suppress the evaluation of 5546 /// the outer expression. 5547 class EvaluationTracker { 5548 public: 5549 EvaluationTracker(SequenceChecker &Self) 5550 : Self(Self), Prev(Self.EvalTracker), EvalOK(true) { 5551 Self.EvalTracker = this; 5552 } 5553 ~EvaluationTracker() { 5554 Self.EvalTracker = Prev; 5555 if (Prev) 5556 Prev->EvalOK &= EvalOK; 5557 } 5558 5559 bool evaluate(const Expr *E, bool &Result) { 5560 if (!EvalOK || E->isValueDependent()) 5561 return false; 5562 EvalOK = E->EvaluateAsBooleanCondition(Result, Self.SemaRef.Context); 5563 return EvalOK; 5564 } 5565 5566 private: 5567 SequenceChecker &Self; 5568 EvaluationTracker *Prev; 5569 bool EvalOK; 5570 } *EvalTracker; 5571 5572 /// \brief Find the object which is produced by the specified expression, 5573 /// if any. 5574 Object getObject(Expr *E, bool Mod) const { 5575 E = E->IgnoreParenCasts(); 5576 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 5577 if (Mod && (UO->getOpcode() == UO_PreInc || UO->getOpcode() == UO_PreDec)) 5578 return getObject(UO->getSubExpr(), Mod); 5579 } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5580 if (BO->getOpcode() == BO_Comma) 5581 return getObject(BO->getRHS(), Mod); 5582 if (Mod && BO->isAssignmentOp()) 5583 return getObject(BO->getLHS(), Mod); 5584 } else if (MemberExpr *ME = dyn_cast<MemberExpr>(E)) { 5585 // FIXME: Check for more interesting cases, like "x.n = ++x.n". 5586 if (isa<CXXThisExpr>(ME->getBase()->IgnoreParenCasts())) 5587 return ME->getMemberDecl(); 5588 } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5589 // FIXME: If this is a reference, map through to its value. 5590 return DRE->getDecl(); 5591 return 0; 5592 } 5593 5594 /// \brief Note that an object was modified or used by an expression. 5595 void addUsage(UsageInfo &UI, Object O, Expr *Ref, UsageKind UK) { 5596 Usage &U = UI.Uses[UK]; 5597 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) { 5598 if (UK == UK_ModAsSideEffect && ModAsSideEffect) 5599 ModAsSideEffect->push_back(std::make_pair(O, U)); 5600 U.Use = Ref; 5601 U.Seq = Region; 5602 } 5603 } 5604 /// \brief Check whether a modification or use conflicts with a prior usage. 5605 void checkUsage(Object O, UsageInfo &UI, Expr *Ref, UsageKind OtherKind, 5606 bool IsModMod) { 5607 if (UI.Diagnosed) 5608 return; 5609 5610 const Usage &U = UI.Uses[OtherKind]; 5611 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) 5612 return; 5613 5614 Expr *Mod = U.Use; 5615 Expr *ModOrUse = Ref; 5616 if (OtherKind == UK_Use) 5617 std::swap(Mod, ModOrUse); 5618 5619 SemaRef.Diag(Mod->getExprLoc(), 5620 IsModMod ? diag::warn_unsequenced_mod_mod 5621 : diag::warn_unsequenced_mod_use) 5622 << O << SourceRange(ModOrUse->getExprLoc()); 5623 UI.Diagnosed = true; 5624 } 5625 5626 void notePreUse(Object O, Expr *Use) { 5627 UsageInfo &U = UsageMap[O]; 5628 // Uses conflict with other modifications. 5629 checkUsage(O, U, Use, UK_ModAsValue, false); 5630 } 5631 void notePostUse(Object O, Expr *Use) { 5632 UsageInfo &U = UsageMap[O]; 5633 checkUsage(O, U, Use, UK_ModAsSideEffect, false); 5634 addUsage(U, O, Use, UK_Use); 5635 } 5636 5637 void notePreMod(Object O, Expr *Mod) { 5638 UsageInfo &U = UsageMap[O]; 5639 // Modifications conflict with other modifications and with uses. 5640 checkUsage(O, U, Mod, UK_ModAsValue, true); 5641 checkUsage(O, U, Mod, UK_Use, false); 5642 } 5643 void notePostMod(Object O, Expr *Use, UsageKind UK) { 5644 UsageInfo &U = UsageMap[O]; 5645 checkUsage(O, U, Use, UK_ModAsSideEffect, true); 5646 addUsage(U, O, Use, UK); 5647 } 5648 5649public: 5650 SequenceChecker(Sema &S, Expr *E, 5651 llvm::SmallVectorImpl<Expr*> &WorkList) 5652 : Base(S.Context), SemaRef(S), Region(Tree.root()), 5653 ModAsSideEffect(0), WorkList(WorkList), EvalTracker(0) { 5654 Visit(E); 5655 } 5656 5657 void VisitStmt(Stmt *S) { 5658 // Skip all statements which aren't expressions for now. 5659 } 5660 5661 void VisitExpr(Expr *E) { 5662 // By default, just recurse to evaluated subexpressions. 5663 Base::VisitStmt(E); 5664 } 5665 5666 void VisitCastExpr(CastExpr *E) { 5667 Object O = Object(); 5668 if (E->getCastKind() == CK_LValueToRValue) 5669 O = getObject(E->getSubExpr(), false); 5670 5671 if (O) 5672 notePreUse(O, E); 5673 VisitExpr(E); 5674 if (O) 5675 notePostUse(O, E); 5676 } 5677 5678 void VisitBinComma(BinaryOperator *BO) { 5679 // C++11 [expr.comma]p1: 5680 // Every value computation and side effect associated with the left 5681 // expression is sequenced before every value computation and side 5682 // effect associated with the right expression. 5683 SequenceTree::Seq LHS = Tree.allocate(Region); 5684 SequenceTree::Seq RHS = Tree.allocate(Region); 5685 SequenceTree::Seq OldRegion = Region; 5686 5687 { 5688 SequencedSubexpression SeqLHS(*this); 5689 Region = LHS; 5690 Visit(BO->getLHS()); 5691 } 5692 5693 Region = RHS; 5694 Visit(BO->getRHS()); 5695 5696 Region = OldRegion; 5697 5698 // Forget that LHS and RHS are sequenced. They are both unsequenced 5699 // with respect to other stuff. 5700 Tree.merge(LHS); 5701 Tree.merge(RHS); 5702 } 5703 5704 void VisitBinAssign(BinaryOperator *BO) { 5705 // The modification is sequenced after the value computation of the LHS 5706 // and RHS, so check it before inspecting the operands and update the 5707 // map afterwards. 5708 Object O = getObject(BO->getLHS(), true); 5709 if (!O) 5710 return VisitExpr(BO); 5711 5712 notePreMod(O, BO); 5713 5714 // C++11 [expr.ass]p7: 5715 // E1 op= E2 is equivalent to E1 = E1 op E2, except that E1 is evaluated 5716 // only once. 5717 // 5718 // Therefore, for a compound assignment operator, O is considered used 5719 // everywhere except within the evaluation of E1 itself. 5720 if (isa<CompoundAssignOperator>(BO)) 5721 notePreUse(O, BO); 5722 5723 Visit(BO->getLHS()); 5724 5725 if (isa<CompoundAssignOperator>(BO)) 5726 notePostUse(O, BO); 5727 5728 Visit(BO->getRHS()); 5729 5730 // C++11 [expr.ass]p1: 5731 // the assignment is sequenced [...] before the value computation of the 5732 // assignment expression. 5733 // C11 6.5.16/3 has no such rule. 5734 notePostMod(O, BO, SemaRef.getLangOpts().CPlusPlus ? UK_ModAsValue 5735 : UK_ModAsSideEffect); 5736 } 5737 void VisitCompoundAssignOperator(CompoundAssignOperator *CAO) { 5738 VisitBinAssign(CAO); 5739 } 5740 5741 void VisitUnaryPreInc(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 5742 void VisitUnaryPreDec(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 5743 void VisitUnaryPreIncDec(UnaryOperator *UO) { 5744 Object O = getObject(UO->getSubExpr(), true); 5745 if (!O) 5746 return VisitExpr(UO); 5747 5748 notePreMod(O, UO); 5749 Visit(UO->getSubExpr()); 5750 // C++11 [expr.pre.incr]p1: 5751 // the expression ++x is equivalent to x+=1 5752 notePostMod(O, UO, SemaRef.getLangOpts().CPlusPlus ? UK_ModAsValue 5753 : UK_ModAsSideEffect); 5754 } 5755 5756 void VisitUnaryPostInc(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 5757 void VisitUnaryPostDec(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 5758 void VisitUnaryPostIncDec(UnaryOperator *UO) { 5759 Object O = getObject(UO->getSubExpr(), true); 5760 if (!O) 5761 return VisitExpr(UO); 5762 5763 notePreMod(O, UO); 5764 Visit(UO->getSubExpr()); 5765 notePostMod(O, UO, UK_ModAsSideEffect); 5766 } 5767 5768 /// Don't visit the RHS of '&&' or '||' if it might not be evaluated. 5769 void VisitBinLOr(BinaryOperator *BO) { 5770 // The side-effects of the LHS of an '&&' are sequenced before the 5771 // value computation of the RHS, and hence before the value computation 5772 // of the '&&' itself, unless the LHS evaluates to zero. We treat them 5773 // as if they were unconditionally sequenced. 5774 EvaluationTracker Eval(*this); 5775 { 5776 SequencedSubexpression Sequenced(*this); 5777 Visit(BO->getLHS()); 5778 } 5779 5780 bool Result; 5781 if (Eval.evaluate(BO->getLHS(), Result)) { 5782 if (!Result) 5783 Visit(BO->getRHS()); 5784 } else { 5785 // Check for unsequenced operations in the RHS, treating it as an 5786 // entirely separate evaluation. 5787 // 5788 // FIXME: If there are operations in the RHS which are unsequenced 5789 // with respect to operations outside the RHS, and those operations 5790 // are unconditionally evaluated, diagnose them. 5791 WorkList.push_back(BO->getRHS()); 5792 } 5793 } 5794 void VisitBinLAnd(BinaryOperator *BO) { 5795 EvaluationTracker Eval(*this); 5796 { 5797 SequencedSubexpression Sequenced(*this); 5798 Visit(BO->getLHS()); 5799 } 5800 5801 bool Result; 5802 if (Eval.evaluate(BO->getLHS(), Result)) { 5803 if (Result) 5804 Visit(BO->getRHS()); 5805 } else { 5806 WorkList.push_back(BO->getRHS()); 5807 } 5808 } 5809 5810 // Only visit the condition, unless we can be sure which subexpression will 5811 // be chosen. 5812 void VisitAbstractConditionalOperator(AbstractConditionalOperator *CO) { 5813 EvaluationTracker Eval(*this); 5814 { 5815 SequencedSubexpression Sequenced(*this); 5816 Visit(CO->getCond()); 5817 } 5818 5819 bool Result; 5820 if (Eval.evaluate(CO->getCond(), Result)) 5821 Visit(Result ? CO->getTrueExpr() : CO->getFalseExpr()); 5822 else { 5823 WorkList.push_back(CO->getTrueExpr()); 5824 WorkList.push_back(CO->getFalseExpr()); 5825 } 5826 } 5827 5828 void VisitCallExpr(CallExpr *CE) { 5829 // C++11 [intro.execution]p15: 5830 // When calling a function [...], every value computation and side effect 5831 // associated with any argument expression, or with the postfix expression 5832 // designating the called function, is sequenced before execution of every 5833 // expression or statement in the body of the function [and thus before 5834 // the value computation of its result]. 5835 SequencedSubexpression Sequenced(*this); 5836 Base::VisitCallExpr(CE); 5837 5838 // FIXME: CXXNewExpr and CXXDeleteExpr implicitly call functions. 5839 } 5840 5841 void VisitCXXConstructExpr(CXXConstructExpr *CCE) { 5842 // This is a call, so all subexpressions are sequenced before the result. 5843 SequencedSubexpression Sequenced(*this); 5844 5845 if (!CCE->isListInitialization()) 5846 return VisitExpr(CCE); 5847 5848 // In C++11, list initializations are sequenced. 5849 llvm::SmallVector<SequenceTree::Seq, 32> Elts; 5850 SequenceTree::Seq Parent = Region; 5851 for (CXXConstructExpr::arg_iterator I = CCE->arg_begin(), 5852 E = CCE->arg_end(); 5853 I != E; ++I) { 5854 Region = Tree.allocate(Parent); 5855 Elts.push_back(Region); 5856 Visit(*I); 5857 } 5858 5859 // Forget that the initializers are sequenced. 5860 Region = Parent; 5861 for (unsigned I = 0; I < Elts.size(); ++I) 5862 Tree.merge(Elts[I]); 5863 } 5864 5865 void VisitInitListExpr(InitListExpr *ILE) { 5866 if (!SemaRef.getLangOpts().CPlusPlus11) 5867 return VisitExpr(ILE); 5868 5869 // In C++11, list initializations are sequenced. 5870 llvm::SmallVector<SequenceTree::Seq, 32> Elts; 5871 SequenceTree::Seq Parent = Region; 5872 for (unsigned I = 0; I < ILE->getNumInits(); ++I) { 5873 Expr *E = ILE->getInit(I); 5874 if (!E) continue; 5875 Region = Tree.allocate(Parent); 5876 Elts.push_back(Region); 5877 Visit(E); 5878 } 5879 5880 // Forget that the initializers are sequenced. 5881 Region = Parent; 5882 for (unsigned I = 0; I < Elts.size(); ++I) 5883 Tree.merge(Elts[I]); 5884 } 5885}; 5886} 5887 5888void Sema::CheckUnsequencedOperations(Expr *E) { 5889 llvm::SmallVector<Expr*, 8> WorkList; 5890 WorkList.push_back(E); 5891 while (!WorkList.empty()) { 5892 Expr *Item = WorkList.back(); 5893 WorkList.pop_back(); 5894 SequenceChecker(*this, Item, WorkList); 5895 } 5896} 5897 5898void Sema::CheckCompletedExpr(Expr *E, SourceLocation CheckLoc, 5899 bool IsConstexpr) { 5900 CheckImplicitConversions(E, CheckLoc); 5901 CheckUnsequencedOperations(E); 5902 if (!IsConstexpr && !E->isValueDependent()) 5903 CheckForIntOverflow(E); 5904} 5905 5906void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 5907 FieldDecl *BitField, 5908 Expr *Init) { 5909 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 5910} 5911 5912/// CheckParmsForFunctionDef - Check that the parameters of the given 5913/// function are appropriate for the definition of a function. This 5914/// takes care of any checks that cannot be performed on the 5915/// declaration itself, e.g., that the types of each of the function 5916/// parameters are complete. 5917bool Sema::CheckParmsForFunctionDef(ParmVarDecl *const *P, 5918 ParmVarDecl *const *PEnd, 5919 bool CheckParameterNames) { 5920 bool HasInvalidParm = false; 5921 for (; P != PEnd; ++P) { 5922 ParmVarDecl *Param = *P; 5923 5924 // C99 6.7.5.3p4: the parameters in a parameter type list in a 5925 // function declarator that is part of a function definition of 5926 // that function shall not have incomplete type. 5927 // 5928 // This is also C++ [dcl.fct]p6. 5929 if (!Param->isInvalidDecl() && 5930 RequireCompleteType(Param->getLocation(), Param->getType(), 5931 diag::err_typecheck_decl_incomplete_type)) { 5932 Param->setInvalidDecl(); 5933 HasInvalidParm = true; 5934 } 5935 5936 // C99 6.9.1p5: If the declarator includes a parameter type list, the 5937 // declaration of each parameter shall include an identifier. 5938 if (CheckParameterNames && 5939 Param->getIdentifier() == 0 && 5940 !Param->isImplicit() && 5941 !getLangOpts().CPlusPlus) 5942 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 5943 5944 // C99 6.7.5.3p12: 5945 // If the function declarator is not part of a definition of that 5946 // function, parameters may have incomplete type and may use the [*] 5947 // notation in their sequences of declarator specifiers to specify 5948 // variable length array types. 5949 QualType PType = Param->getOriginalType(); 5950 while (const ArrayType *AT = Context.getAsArrayType(PType)) { 5951 if (AT->getSizeModifier() == ArrayType::Star) { 5952 // FIXME: This diagnostic should point the '[*]' if source-location 5953 // information is added for it. 5954 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 5955 break; 5956 } 5957 PType= AT->getElementType(); 5958 } 5959 5960 // MSVC destroys objects passed by value in the callee. Therefore a 5961 // function definition which takes such a parameter must be able to call the 5962 // object's destructor. 5963 if (getLangOpts().CPlusPlus && 5964 Context.getTargetInfo().getCXXABI().isArgumentDestroyedByCallee()) { 5965 if (const RecordType *RT = Param->getType()->getAs<RecordType>()) 5966 FinalizeVarWithDestructor(Param, RT); 5967 } 5968 } 5969 5970 return HasInvalidParm; 5971} 5972 5973/// CheckCastAlign - Implements -Wcast-align, which warns when a 5974/// pointer cast increases the alignment requirements. 5975void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 5976 // This is actually a lot of work to potentially be doing on every 5977 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 5978 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 5979 TRange.getBegin()) 5980 == DiagnosticsEngine::Ignored) 5981 return; 5982 5983 // Ignore dependent types. 5984 if (T->isDependentType() || Op->getType()->isDependentType()) 5985 return; 5986 5987 // Require that the destination be a pointer type. 5988 const PointerType *DestPtr = T->getAs<PointerType>(); 5989 if (!DestPtr) return; 5990 5991 // If the destination has alignment 1, we're done. 5992 QualType DestPointee = DestPtr->getPointeeType(); 5993 if (DestPointee->isIncompleteType()) return; 5994 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 5995 if (DestAlign.isOne()) return; 5996 5997 // Require that the source be a pointer type. 5998 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 5999 if (!SrcPtr) return; 6000 QualType SrcPointee = SrcPtr->getPointeeType(); 6001 6002 // Whitelist casts from cv void*. We already implicitly 6003 // whitelisted casts to cv void*, since they have alignment 1. 6004 // Also whitelist casts involving incomplete types, which implicitly 6005 // includes 'void'. 6006 if (SrcPointee->isIncompleteType()) return; 6007 6008 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 6009 if (SrcAlign >= DestAlign) return; 6010 6011 Diag(TRange.getBegin(), diag::warn_cast_align) 6012 << Op->getType() << T 6013 << static_cast<unsigned>(SrcAlign.getQuantity()) 6014 << static_cast<unsigned>(DestAlign.getQuantity()) 6015 << TRange << Op->getSourceRange(); 6016} 6017 6018static const Type* getElementType(const Expr *BaseExpr) { 6019 const Type* EltType = BaseExpr->getType().getTypePtr(); 6020 if (EltType->isAnyPointerType()) 6021 return EltType->getPointeeType().getTypePtr(); 6022 else if (EltType->isArrayType()) 6023 return EltType->getBaseElementTypeUnsafe(); 6024 return EltType; 6025} 6026 6027/// \brief Check whether this array fits the idiom of a size-one tail padded 6028/// array member of a struct. 6029/// 6030/// We avoid emitting out-of-bounds access warnings for such arrays as they are 6031/// commonly used to emulate flexible arrays in C89 code. 6032static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 6033 const NamedDecl *ND) { 6034 if (Size != 1 || !ND) return false; 6035 6036 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 6037 if (!FD) return false; 6038 6039 // Don't consider sizes resulting from macro expansions or template argument 6040 // substitution to form C89 tail-padded arrays. 6041 6042 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 6043 while (TInfo) { 6044 TypeLoc TL = TInfo->getTypeLoc(); 6045 // Look through typedefs. 6046 if (TypedefTypeLoc TTL = TL.getAs<TypedefTypeLoc>()) { 6047 const TypedefNameDecl *TDL = TTL.getTypedefNameDecl(); 6048 TInfo = TDL->getTypeSourceInfo(); 6049 continue; 6050 } 6051 if (ConstantArrayTypeLoc CTL = TL.getAs<ConstantArrayTypeLoc>()) { 6052 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 6053 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 6054 return false; 6055 } 6056 break; 6057 } 6058 6059 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 6060 if (!RD) return false; 6061 if (RD->isUnion()) return false; 6062 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 6063 if (!CRD->isStandardLayout()) return false; 6064 } 6065 6066 // See if this is the last field decl in the record. 6067 const Decl *D = FD; 6068 while ((D = D->getNextDeclInContext())) 6069 if (isa<FieldDecl>(D)) 6070 return false; 6071 return true; 6072} 6073 6074void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 6075 const ArraySubscriptExpr *ASE, 6076 bool AllowOnePastEnd, bool IndexNegated) { 6077 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 6078 if (IndexExpr->isValueDependent()) 6079 return; 6080 6081 const Type *EffectiveType = getElementType(BaseExpr); 6082 BaseExpr = BaseExpr->IgnoreParenCasts(); 6083 const ConstantArrayType *ArrayTy = 6084 Context.getAsConstantArrayType(BaseExpr->getType()); 6085 if (!ArrayTy) 6086 return; 6087 6088 llvm::APSInt index; 6089 if (!IndexExpr->EvaluateAsInt(index, Context)) 6090 return; 6091 if (IndexNegated) 6092 index = -index; 6093 6094 const NamedDecl *ND = NULL; 6095 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 6096 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 6097 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 6098 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 6099 6100 if (index.isUnsigned() || !index.isNegative()) { 6101 llvm::APInt size = ArrayTy->getSize(); 6102 if (!size.isStrictlyPositive()) 6103 return; 6104 6105 const Type* BaseType = getElementType(BaseExpr); 6106 if (BaseType != EffectiveType) { 6107 // Make sure we're comparing apples to apples when comparing index to size 6108 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 6109 uint64_t array_typesize = Context.getTypeSize(BaseType); 6110 // Handle ptrarith_typesize being zero, such as when casting to void* 6111 if (!ptrarith_typesize) ptrarith_typesize = 1; 6112 if (ptrarith_typesize != array_typesize) { 6113 // There's a cast to a different size type involved 6114 uint64_t ratio = array_typesize / ptrarith_typesize; 6115 // TODO: Be smarter about handling cases where array_typesize is not a 6116 // multiple of ptrarith_typesize 6117 if (ptrarith_typesize * ratio == array_typesize) 6118 size *= llvm::APInt(size.getBitWidth(), ratio); 6119 } 6120 } 6121 6122 if (size.getBitWidth() > index.getBitWidth()) 6123 index = index.zext(size.getBitWidth()); 6124 else if (size.getBitWidth() < index.getBitWidth()) 6125 size = size.zext(index.getBitWidth()); 6126 6127 // For array subscripting the index must be less than size, but for pointer 6128 // arithmetic also allow the index (offset) to be equal to size since 6129 // computing the next address after the end of the array is legal and 6130 // commonly done e.g. in C++ iterators and range-based for loops. 6131 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 6132 return; 6133 6134 // Also don't warn for arrays of size 1 which are members of some 6135 // structure. These are often used to approximate flexible arrays in C89 6136 // code. 6137 if (IsTailPaddedMemberArray(*this, size, ND)) 6138 return; 6139 6140 // Suppress the warning if the subscript expression (as identified by the 6141 // ']' location) and the index expression are both from macro expansions 6142 // within a system header. 6143 if (ASE) { 6144 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 6145 ASE->getRBracketLoc()); 6146 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 6147 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 6148 IndexExpr->getLocStart()); 6149 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 6150 return; 6151 } 6152 } 6153 6154 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 6155 if (ASE) 6156 DiagID = diag::warn_array_index_exceeds_bounds; 6157 6158 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 6159 PDiag(DiagID) << index.toString(10, true) 6160 << size.toString(10, true) 6161 << (unsigned)size.getLimitedValue(~0U) 6162 << IndexExpr->getSourceRange()); 6163 } else { 6164 unsigned DiagID = diag::warn_array_index_precedes_bounds; 6165 if (!ASE) { 6166 DiagID = diag::warn_ptr_arith_precedes_bounds; 6167 if (index.isNegative()) index = -index; 6168 } 6169 6170 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 6171 PDiag(DiagID) << index.toString(10, true) 6172 << IndexExpr->getSourceRange()); 6173 } 6174 6175 if (!ND) { 6176 // Try harder to find a NamedDecl to point at in the note. 6177 while (const ArraySubscriptExpr *ASE = 6178 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 6179 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 6180 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 6181 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 6182 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 6183 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 6184 } 6185 6186 if (ND) 6187 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 6188 PDiag(diag::note_array_index_out_of_bounds) 6189 << ND->getDeclName()); 6190} 6191 6192void Sema::CheckArrayAccess(const Expr *expr) { 6193 int AllowOnePastEnd = 0; 6194 while (expr) { 6195 expr = expr->IgnoreParenImpCasts(); 6196 switch (expr->getStmtClass()) { 6197 case Stmt::ArraySubscriptExprClass: { 6198 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 6199 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 6200 AllowOnePastEnd > 0); 6201 return; 6202 } 6203 case Stmt::UnaryOperatorClass: { 6204 // Only unwrap the * and & unary operators 6205 const UnaryOperator *UO = cast<UnaryOperator>(expr); 6206 expr = UO->getSubExpr(); 6207 switch (UO->getOpcode()) { 6208 case UO_AddrOf: 6209 AllowOnePastEnd++; 6210 break; 6211 case UO_Deref: 6212 AllowOnePastEnd--; 6213 break; 6214 default: 6215 return; 6216 } 6217 break; 6218 } 6219 case Stmt::ConditionalOperatorClass: { 6220 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 6221 if (const Expr *lhs = cond->getLHS()) 6222 CheckArrayAccess(lhs); 6223 if (const Expr *rhs = cond->getRHS()) 6224 CheckArrayAccess(rhs); 6225 return; 6226 } 6227 default: 6228 return; 6229 } 6230 } 6231} 6232 6233//===--- CHECK: Objective-C retain cycles ----------------------------------// 6234 6235namespace { 6236 struct RetainCycleOwner { 6237 RetainCycleOwner() : Variable(0), Indirect(false) {} 6238 VarDecl *Variable; 6239 SourceRange Range; 6240 SourceLocation Loc; 6241 bool Indirect; 6242 6243 void setLocsFrom(Expr *e) { 6244 Loc = e->getExprLoc(); 6245 Range = e->getSourceRange(); 6246 } 6247 }; 6248} 6249 6250/// Consider whether capturing the given variable can possibly lead to 6251/// a retain cycle. 6252static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 6253 // In ARC, it's captured strongly iff the variable has __strong 6254 // lifetime. In MRR, it's captured strongly if the variable is 6255 // __block and has an appropriate type. 6256 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 6257 return false; 6258 6259 owner.Variable = var; 6260 if (ref) 6261 owner.setLocsFrom(ref); 6262 return true; 6263} 6264 6265static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 6266 while (true) { 6267 e = e->IgnoreParens(); 6268 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 6269 switch (cast->getCastKind()) { 6270 case CK_BitCast: 6271 case CK_LValueBitCast: 6272 case CK_LValueToRValue: 6273 case CK_ARCReclaimReturnedObject: 6274 e = cast->getSubExpr(); 6275 continue; 6276 6277 default: 6278 return false; 6279 } 6280 } 6281 6282 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 6283 ObjCIvarDecl *ivar = ref->getDecl(); 6284 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 6285 return false; 6286 6287 // Try to find a retain cycle in the base. 6288 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 6289 return false; 6290 6291 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 6292 owner.Indirect = true; 6293 return true; 6294 } 6295 6296 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 6297 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 6298 if (!var) return false; 6299 return considerVariable(var, ref, owner); 6300 } 6301 6302 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 6303 if (member->isArrow()) return false; 6304 6305 // Don't count this as an indirect ownership. 6306 e = member->getBase(); 6307 continue; 6308 } 6309 6310 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 6311 // Only pay attention to pseudo-objects on property references. 6312 ObjCPropertyRefExpr *pre 6313 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 6314 ->IgnoreParens()); 6315 if (!pre) return false; 6316 if (pre->isImplicitProperty()) return false; 6317 ObjCPropertyDecl *property = pre->getExplicitProperty(); 6318 if (!property->isRetaining() && 6319 !(property->getPropertyIvarDecl() && 6320 property->getPropertyIvarDecl()->getType() 6321 .getObjCLifetime() == Qualifiers::OCL_Strong)) 6322 return false; 6323 6324 owner.Indirect = true; 6325 if (pre->isSuperReceiver()) { 6326 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 6327 if (!owner.Variable) 6328 return false; 6329 owner.Loc = pre->getLocation(); 6330 owner.Range = pre->getSourceRange(); 6331 return true; 6332 } 6333 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 6334 ->getSourceExpr()); 6335 continue; 6336 } 6337 6338 // Array ivars? 6339 6340 return false; 6341 } 6342} 6343 6344namespace { 6345 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 6346 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 6347 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 6348 Variable(variable), Capturer(0) {} 6349 6350 VarDecl *Variable; 6351 Expr *Capturer; 6352 6353 void VisitDeclRefExpr(DeclRefExpr *ref) { 6354 if (ref->getDecl() == Variable && !Capturer) 6355 Capturer = ref; 6356 } 6357 6358 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 6359 if (Capturer) return; 6360 Visit(ref->getBase()); 6361 if (Capturer && ref->isFreeIvar()) 6362 Capturer = ref; 6363 } 6364 6365 void VisitBlockExpr(BlockExpr *block) { 6366 // Look inside nested blocks 6367 if (block->getBlockDecl()->capturesVariable(Variable)) 6368 Visit(block->getBlockDecl()->getBody()); 6369 } 6370 6371 void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) { 6372 if (Capturer) return; 6373 if (OVE->getSourceExpr()) 6374 Visit(OVE->getSourceExpr()); 6375 } 6376 }; 6377} 6378 6379/// Check whether the given argument is a block which captures a 6380/// variable. 6381static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 6382 assert(owner.Variable && owner.Loc.isValid()); 6383 6384 e = e->IgnoreParenCasts(); 6385 6386 // Look through [^{...} copy] and Block_copy(^{...}). 6387 if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) { 6388 Selector Cmd = ME->getSelector(); 6389 if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") { 6390 e = ME->getInstanceReceiver(); 6391 if (!e) 6392 return 0; 6393 e = e->IgnoreParenCasts(); 6394 } 6395 } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) { 6396 if (CE->getNumArgs() == 1) { 6397 FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl()); 6398 if (Fn) { 6399 const IdentifierInfo *FnI = Fn->getIdentifier(); 6400 if (FnI && FnI->isStr("_Block_copy")) { 6401 e = CE->getArg(0)->IgnoreParenCasts(); 6402 } 6403 } 6404 } 6405 } 6406 6407 BlockExpr *block = dyn_cast<BlockExpr>(e); 6408 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 6409 return 0; 6410 6411 FindCaptureVisitor visitor(S.Context, owner.Variable); 6412 visitor.Visit(block->getBlockDecl()->getBody()); 6413 return visitor.Capturer; 6414} 6415 6416static void diagnoseRetainCycle(Sema &S, Expr *capturer, 6417 RetainCycleOwner &owner) { 6418 assert(capturer); 6419 assert(owner.Variable && owner.Loc.isValid()); 6420 6421 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 6422 << owner.Variable << capturer->getSourceRange(); 6423 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 6424 << owner.Indirect << owner.Range; 6425} 6426 6427/// Check for a keyword selector that starts with the word 'add' or 6428/// 'set'. 6429static bool isSetterLikeSelector(Selector sel) { 6430 if (sel.isUnarySelector()) return false; 6431 6432 StringRef str = sel.getNameForSlot(0); 6433 while (!str.empty() && str.front() == '_') str = str.substr(1); 6434 if (str.startswith("set")) 6435 str = str.substr(3); 6436 else if (str.startswith("add")) { 6437 // Specially whitelist 'addOperationWithBlock:'. 6438 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 6439 return false; 6440 str = str.substr(3); 6441 } 6442 else 6443 return false; 6444 6445 if (str.empty()) return true; 6446 return !isLowercase(str.front()); 6447} 6448 6449/// Check a message send to see if it's likely to cause a retain cycle. 6450void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 6451 // Only check instance methods whose selector looks like a setter. 6452 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 6453 return; 6454 6455 // Try to find a variable that the receiver is strongly owned by. 6456 RetainCycleOwner owner; 6457 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 6458 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 6459 return; 6460 } else { 6461 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 6462 owner.Variable = getCurMethodDecl()->getSelfDecl(); 6463 owner.Loc = msg->getSuperLoc(); 6464 owner.Range = msg->getSuperLoc(); 6465 } 6466 6467 // Check whether the receiver is captured by any of the arguments. 6468 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 6469 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 6470 return diagnoseRetainCycle(*this, capturer, owner); 6471} 6472 6473/// Check a property assign to see if it's likely to cause a retain cycle. 6474void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 6475 RetainCycleOwner owner; 6476 if (!findRetainCycleOwner(*this, receiver, owner)) 6477 return; 6478 6479 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 6480 diagnoseRetainCycle(*this, capturer, owner); 6481} 6482 6483void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) { 6484 RetainCycleOwner Owner; 6485 if (!considerVariable(Var, /*DeclRefExpr=*/0, Owner)) 6486 return; 6487 6488 // Because we don't have an expression for the variable, we have to set the 6489 // location explicitly here. 6490 Owner.Loc = Var->getLocation(); 6491 Owner.Range = Var->getSourceRange(); 6492 6493 if (Expr *Capturer = findCapturingExpr(*this, Init, Owner)) 6494 diagnoseRetainCycle(*this, Capturer, Owner); 6495} 6496 6497static bool checkUnsafeAssignLiteral(Sema &S, SourceLocation Loc, 6498 Expr *RHS, bool isProperty) { 6499 // Check if RHS is an Objective-C object literal, which also can get 6500 // immediately zapped in a weak reference. Note that we explicitly 6501 // allow ObjCStringLiterals, since those are designed to never really die. 6502 RHS = RHS->IgnoreParenImpCasts(); 6503 6504 // This enum needs to match with the 'select' in 6505 // warn_objc_arc_literal_assign (off-by-1). 6506 Sema::ObjCLiteralKind Kind = S.CheckLiteralKind(RHS); 6507 if (Kind == Sema::LK_String || Kind == Sema::LK_None) 6508 return false; 6509 6510 S.Diag(Loc, diag::warn_arc_literal_assign) 6511 << (unsigned) Kind 6512 << (isProperty ? 0 : 1) 6513 << RHS->getSourceRange(); 6514 6515 return true; 6516} 6517 6518static bool checkUnsafeAssignObject(Sema &S, SourceLocation Loc, 6519 Qualifiers::ObjCLifetime LT, 6520 Expr *RHS, bool isProperty) { 6521 // Strip off any implicit cast added to get to the one ARC-specific. 6522 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6523 if (cast->getCastKind() == CK_ARCConsumeObject) { 6524 S.Diag(Loc, diag::warn_arc_retained_assign) 6525 << (LT == Qualifiers::OCL_ExplicitNone) 6526 << (isProperty ? 0 : 1) 6527 << RHS->getSourceRange(); 6528 return true; 6529 } 6530 RHS = cast->getSubExpr(); 6531 } 6532 6533 if (LT == Qualifiers::OCL_Weak && 6534 checkUnsafeAssignLiteral(S, Loc, RHS, isProperty)) 6535 return true; 6536 6537 return false; 6538} 6539 6540bool Sema::checkUnsafeAssigns(SourceLocation Loc, 6541 QualType LHS, Expr *RHS) { 6542 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 6543 6544 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 6545 return false; 6546 6547 if (checkUnsafeAssignObject(*this, Loc, LT, RHS, false)) 6548 return true; 6549 6550 return false; 6551} 6552 6553void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 6554 Expr *LHS, Expr *RHS) { 6555 QualType LHSType; 6556 // PropertyRef on LHS type need be directly obtained from 6557 // its declaration as it has a PsuedoType. 6558 ObjCPropertyRefExpr *PRE 6559 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 6560 if (PRE && !PRE->isImplicitProperty()) { 6561 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6562 if (PD) 6563 LHSType = PD->getType(); 6564 } 6565 6566 if (LHSType.isNull()) 6567 LHSType = LHS->getType(); 6568 6569 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 6570 6571 if (LT == Qualifiers::OCL_Weak) { 6572 DiagnosticsEngine::Level Level = 6573 Diags.getDiagnosticLevel(diag::warn_arc_repeated_use_of_weak, Loc); 6574 if (Level != DiagnosticsEngine::Ignored) 6575 getCurFunction()->markSafeWeakUse(LHS); 6576 } 6577 6578 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 6579 return; 6580 6581 // FIXME. Check for other life times. 6582 if (LT != Qualifiers::OCL_None) 6583 return; 6584 6585 if (PRE) { 6586 if (PRE->isImplicitProperty()) 6587 return; 6588 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6589 if (!PD) 6590 return; 6591 6592 unsigned Attributes = PD->getPropertyAttributes(); 6593 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 6594 // when 'assign' attribute was not explicitly specified 6595 // by user, ignore it and rely on property type itself 6596 // for lifetime info. 6597 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 6598 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 6599 LHSType->isObjCRetainableType()) 6600 return; 6601 6602 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6603 if (cast->getCastKind() == CK_ARCConsumeObject) { 6604 Diag(Loc, diag::warn_arc_retained_property_assign) 6605 << RHS->getSourceRange(); 6606 return; 6607 } 6608 RHS = cast->getSubExpr(); 6609 } 6610 } 6611 else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) { 6612 if (checkUnsafeAssignObject(*this, Loc, Qualifiers::OCL_Weak, RHS, true)) 6613 return; 6614 } 6615 } 6616} 6617 6618//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 6619 6620namespace { 6621bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 6622 SourceLocation StmtLoc, 6623 const NullStmt *Body) { 6624 // Do not warn if the body is a macro that expands to nothing, e.g: 6625 // 6626 // #define CALL(x) 6627 // if (condition) 6628 // CALL(0); 6629 // 6630 if (Body->hasLeadingEmptyMacro()) 6631 return false; 6632 6633 // Get line numbers of statement and body. 6634 bool StmtLineInvalid; 6635 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 6636 &StmtLineInvalid); 6637 if (StmtLineInvalid) 6638 return false; 6639 6640 bool BodyLineInvalid; 6641 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 6642 &BodyLineInvalid); 6643 if (BodyLineInvalid) 6644 return false; 6645 6646 // Warn if null statement and body are on the same line. 6647 if (StmtLine != BodyLine) 6648 return false; 6649 6650 return true; 6651} 6652} // Unnamed namespace 6653 6654void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 6655 const Stmt *Body, 6656 unsigned DiagID) { 6657 // Since this is a syntactic check, don't emit diagnostic for template 6658 // instantiations, this just adds noise. 6659 if (CurrentInstantiationScope) 6660 return; 6661 6662 // The body should be a null statement. 6663 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6664 if (!NBody) 6665 return; 6666 6667 // Do the usual checks. 6668 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6669 return; 6670 6671 Diag(NBody->getSemiLoc(), DiagID); 6672 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6673} 6674 6675void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 6676 const Stmt *PossibleBody) { 6677 assert(!CurrentInstantiationScope); // Ensured by caller 6678 6679 SourceLocation StmtLoc; 6680 const Stmt *Body; 6681 unsigned DiagID; 6682 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 6683 StmtLoc = FS->getRParenLoc(); 6684 Body = FS->getBody(); 6685 DiagID = diag::warn_empty_for_body; 6686 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 6687 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 6688 Body = WS->getBody(); 6689 DiagID = diag::warn_empty_while_body; 6690 } else 6691 return; // Neither `for' nor `while'. 6692 6693 // The body should be a null statement. 6694 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6695 if (!NBody) 6696 return; 6697 6698 // Skip expensive checks if diagnostic is disabled. 6699 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 6700 DiagnosticsEngine::Ignored) 6701 return; 6702 6703 // Do the usual checks. 6704 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6705 return; 6706 6707 // `for(...);' and `while(...);' are popular idioms, so in order to keep 6708 // noise level low, emit diagnostics only if for/while is followed by a 6709 // CompoundStmt, e.g.: 6710 // for (int i = 0; i < n; i++); 6711 // { 6712 // a(i); 6713 // } 6714 // or if for/while is followed by a statement with more indentation 6715 // than for/while itself: 6716 // for (int i = 0; i < n; i++); 6717 // a(i); 6718 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 6719 if (!ProbableTypo) { 6720 bool BodyColInvalid; 6721 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 6722 PossibleBody->getLocStart(), 6723 &BodyColInvalid); 6724 if (BodyColInvalid) 6725 return; 6726 6727 bool StmtColInvalid; 6728 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 6729 S->getLocStart(), 6730 &StmtColInvalid); 6731 if (StmtColInvalid) 6732 return; 6733 6734 if (BodyCol > StmtCol) 6735 ProbableTypo = true; 6736 } 6737 6738 if (ProbableTypo) { 6739 Diag(NBody->getSemiLoc(), DiagID); 6740 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6741 } 6742} 6743 6744//===--- Layout compatibility ----------------------------------------------// 6745 6746namespace { 6747 6748bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2); 6749 6750/// \brief Check if two enumeration types are layout-compatible. 6751bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) { 6752 // C++11 [dcl.enum] p8: 6753 // Two enumeration types are layout-compatible if they have the same 6754 // underlying type. 6755 return ED1->isComplete() && ED2->isComplete() && 6756 C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType()); 6757} 6758 6759/// \brief Check if two fields are layout-compatible. 6760bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) { 6761 if (!isLayoutCompatible(C, Field1->getType(), Field2->getType())) 6762 return false; 6763 6764 if (Field1->isBitField() != Field2->isBitField()) 6765 return false; 6766 6767 if (Field1->isBitField()) { 6768 // Make sure that the bit-fields are the same length. 6769 unsigned Bits1 = Field1->getBitWidthValue(C); 6770 unsigned Bits2 = Field2->getBitWidthValue(C); 6771 6772 if (Bits1 != Bits2) 6773 return false; 6774 } 6775 6776 return true; 6777} 6778 6779/// \brief Check if two standard-layout structs are layout-compatible. 6780/// (C++11 [class.mem] p17) 6781bool isLayoutCompatibleStruct(ASTContext &C, 6782 RecordDecl *RD1, 6783 RecordDecl *RD2) { 6784 // If both records are C++ classes, check that base classes match. 6785 if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) { 6786 // If one of records is a CXXRecordDecl we are in C++ mode, 6787 // thus the other one is a CXXRecordDecl, too. 6788 const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2); 6789 // Check number of base classes. 6790 if (D1CXX->getNumBases() != D2CXX->getNumBases()) 6791 return false; 6792 6793 // Check the base classes. 6794 for (CXXRecordDecl::base_class_const_iterator 6795 Base1 = D1CXX->bases_begin(), 6796 BaseEnd1 = D1CXX->bases_end(), 6797 Base2 = D2CXX->bases_begin(); 6798 Base1 != BaseEnd1; 6799 ++Base1, ++Base2) { 6800 if (!isLayoutCompatible(C, Base1->getType(), Base2->getType())) 6801 return false; 6802 } 6803 } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) { 6804 // If only RD2 is a C++ class, it should have zero base classes. 6805 if (D2CXX->getNumBases() > 0) 6806 return false; 6807 } 6808 6809 // Check the fields. 6810 RecordDecl::field_iterator Field2 = RD2->field_begin(), 6811 Field2End = RD2->field_end(), 6812 Field1 = RD1->field_begin(), 6813 Field1End = RD1->field_end(); 6814 for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) { 6815 if (!isLayoutCompatible(C, *Field1, *Field2)) 6816 return false; 6817 } 6818 if (Field1 != Field1End || Field2 != Field2End) 6819 return false; 6820 6821 return true; 6822} 6823 6824/// \brief Check if two standard-layout unions are layout-compatible. 6825/// (C++11 [class.mem] p18) 6826bool isLayoutCompatibleUnion(ASTContext &C, 6827 RecordDecl *RD1, 6828 RecordDecl *RD2) { 6829 llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields; 6830 for (RecordDecl::field_iterator Field2 = RD2->field_begin(), 6831 Field2End = RD2->field_end(); 6832 Field2 != Field2End; ++Field2) { 6833 UnmatchedFields.insert(*Field2); 6834 } 6835 6836 for (RecordDecl::field_iterator Field1 = RD1->field_begin(), 6837 Field1End = RD1->field_end(); 6838 Field1 != Field1End; ++Field1) { 6839 llvm::SmallPtrSet<FieldDecl *, 8>::iterator 6840 I = UnmatchedFields.begin(), 6841 E = UnmatchedFields.end(); 6842 6843 for ( ; I != E; ++I) { 6844 if (isLayoutCompatible(C, *Field1, *I)) { 6845 bool Result = UnmatchedFields.erase(*I); 6846 (void) Result; 6847 assert(Result); 6848 break; 6849 } 6850 } 6851 if (I == E) 6852 return false; 6853 } 6854 6855 return UnmatchedFields.empty(); 6856} 6857 6858bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) { 6859 if (RD1->isUnion() != RD2->isUnion()) 6860 return false; 6861 6862 if (RD1->isUnion()) 6863 return isLayoutCompatibleUnion(C, RD1, RD2); 6864 else 6865 return isLayoutCompatibleStruct(C, RD1, RD2); 6866} 6867 6868/// \brief Check if two types are layout-compatible in C++11 sense. 6869bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) { 6870 if (T1.isNull() || T2.isNull()) 6871 return false; 6872 6873 // C++11 [basic.types] p11: 6874 // If two types T1 and T2 are the same type, then T1 and T2 are 6875 // layout-compatible types. 6876 if (C.hasSameType(T1, T2)) 6877 return true; 6878 6879 T1 = T1.getCanonicalType().getUnqualifiedType(); 6880 T2 = T2.getCanonicalType().getUnqualifiedType(); 6881 6882 const Type::TypeClass TC1 = T1->getTypeClass(); 6883 const Type::TypeClass TC2 = T2->getTypeClass(); 6884 6885 if (TC1 != TC2) 6886 return false; 6887 6888 if (TC1 == Type::Enum) { 6889 return isLayoutCompatible(C, 6890 cast<EnumType>(T1)->getDecl(), 6891 cast<EnumType>(T2)->getDecl()); 6892 } else if (TC1 == Type::Record) { 6893 if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType()) 6894 return false; 6895 6896 return isLayoutCompatible(C, 6897 cast<RecordType>(T1)->getDecl(), 6898 cast<RecordType>(T2)->getDecl()); 6899 } 6900 6901 return false; 6902} 6903} 6904 6905//===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----// 6906 6907namespace { 6908/// \brief Given a type tag expression find the type tag itself. 6909/// 6910/// \param TypeExpr Type tag expression, as it appears in user's code. 6911/// 6912/// \param VD Declaration of an identifier that appears in a type tag. 6913/// 6914/// \param MagicValue Type tag magic value. 6915bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx, 6916 const ValueDecl **VD, uint64_t *MagicValue) { 6917 while(true) { 6918 if (!TypeExpr) 6919 return false; 6920 6921 TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts(); 6922 6923 switch (TypeExpr->getStmtClass()) { 6924 case Stmt::UnaryOperatorClass: { 6925 const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr); 6926 if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) { 6927 TypeExpr = UO->getSubExpr(); 6928 continue; 6929 } 6930 return false; 6931 } 6932 6933 case Stmt::DeclRefExprClass: { 6934 const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr); 6935 *VD = DRE->getDecl(); 6936 return true; 6937 } 6938 6939 case Stmt::IntegerLiteralClass: { 6940 const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr); 6941 llvm::APInt MagicValueAPInt = IL->getValue(); 6942 if (MagicValueAPInt.getActiveBits() <= 64) { 6943 *MagicValue = MagicValueAPInt.getZExtValue(); 6944 return true; 6945 } else 6946 return false; 6947 } 6948 6949 case Stmt::BinaryConditionalOperatorClass: 6950 case Stmt::ConditionalOperatorClass: { 6951 const AbstractConditionalOperator *ACO = 6952 cast<AbstractConditionalOperator>(TypeExpr); 6953 bool Result; 6954 if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) { 6955 if (Result) 6956 TypeExpr = ACO->getTrueExpr(); 6957 else 6958 TypeExpr = ACO->getFalseExpr(); 6959 continue; 6960 } 6961 return false; 6962 } 6963 6964 case Stmt::BinaryOperatorClass: { 6965 const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr); 6966 if (BO->getOpcode() == BO_Comma) { 6967 TypeExpr = BO->getRHS(); 6968 continue; 6969 } 6970 return false; 6971 } 6972 6973 default: 6974 return false; 6975 } 6976 } 6977} 6978 6979/// \brief Retrieve the C type corresponding to type tag TypeExpr. 6980/// 6981/// \param TypeExpr Expression that specifies a type tag. 6982/// 6983/// \param MagicValues Registered magic values. 6984/// 6985/// \param FoundWrongKind Set to true if a type tag was found, but of a wrong 6986/// kind. 6987/// 6988/// \param TypeInfo Information about the corresponding C type. 6989/// 6990/// \returns true if the corresponding C type was found. 6991bool GetMatchingCType( 6992 const IdentifierInfo *ArgumentKind, 6993 const Expr *TypeExpr, const ASTContext &Ctx, 6994 const llvm::DenseMap<Sema::TypeTagMagicValue, 6995 Sema::TypeTagData> *MagicValues, 6996 bool &FoundWrongKind, 6997 Sema::TypeTagData &TypeInfo) { 6998 FoundWrongKind = false; 6999 7000 // Variable declaration that has type_tag_for_datatype attribute. 7001 const ValueDecl *VD = NULL; 7002 7003 uint64_t MagicValue; 7004 7005 if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue)) 7006 return false; 7007 7008 if (VD) { 7009 for (specific_attr_iterator<TypeTagForDatatypeAttr> 7010 I = VD->specific_attr_begin<TypeTagForDatatypeAttr>(), 7011 E = VD->specific_attr_end<TypeTagForDatatypeAttr>(); 7012 I != E; ++I) { 7013 if (I->getArgumentKind() != ArgumentKind) { 7014 FoundWrongKind = true; 7015 return false; 7016 } 7017 TypeInfo.Type = I->getMatchingCType(); 7018 TypeInfo.LayoutCompatible = I->getLayoutCompatible(); 7019 TypeInfo.MustBeNull = I->getMustBeNull(); 7020 return true; 7021 } 7022 return false; 7023 } 7024 7025 if (!MagicValues) 7026 return false; 7027 7028 llvm::DenseMap<Sema::TypeTagMagicValue, 7029 Sema::TypeTagData>::const_iterator I = 7030 MagicValues->find(std::make_pair(ArgumentKind, MagicValue)); 7031 if (I == MagicValues->end()) 7032 return false; 7033 7034 TypeInfo = I->second; 7035 return true; 7036} 7037} // unnamed namespace 7038 7039void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind, 7040 uint64_t MagicValue, QualType Type, 7041 bool LayoutCompatible, 7042 bool MustBeNull) { 7043 if (!TypeTagForDatatypeMagicValues) 7044 TypeTagForDatatypeMagicValues.reset( 7045 new llvm::DenseMap<TypeTagMagicValue, TypeTagData>); 7046 7047 TypeTagMagicValue Magic(ArgumentKind, MagicValue); 7048 (*TypeTagForDatatypeMagicValues)[Magic] = 7049 TypeTagData(Type, LayoutCompatible, MustBeNull); 7050} 7051 7052namespace { 7053bool IsSameCharType(QualType T1, QualType T2) { 7054 const BuiltinType *BT1 = T1->getAs<BuiltinType>(); 7055 if (!BT1) 7056 return false; 7057 7058 const BuiltinType *BT2 = T2->getAs<BuiltinType>(); 7059 if (!BT2) 7060 return false; 7061 7062 BuiltinType::Kind T1Kind = BT1->getKind(); 7063 BuiltinType::Kind T2Kind = BT2->getKind(); 7064 7065 return (T1Kind == BuiltinType::SChar && T2Kind == BuiltinType::Char_S) || 7066 (T1Kind == BuiltinType::UChar && T2Kind == BuiltinType::Char_U) || 7067 (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) || 7068 (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar); 7069} 7070} // unnamed namespace 7071 7072void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr, 7073 const Expr * const *ExprArgs) { 7074 const IdentifierInfo *ArgumentKind = Attr->getArgumentKind(); 7075 bool IsPointerAttr = Attr->getIsPointer(); 7076 7077 const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()]; 7078 bool FoundWrongKind; 7079 TypeTagData TypeInfo; 7080 if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context, 7081 TypeTagForDatatypeMagicValues.get(), 7082 FoundWrongKind, TypeInfo)) { 7083 if (FoundWrongKind) 7084 Diag(TypeTagExpr->getExprLoc(), 7085 diag::warn_type_tag_for_datatype_wrong_kind) 7086 << TypeTagExpr->getSourceRange(); 7087 return; 7088 } 7089 7090 const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()]; 7091 if (IsPointerAttr) { 7092 // Skip implicit cast of pointer to `void *' (as a function argument). 7093 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr)) 7094 if (ICE->getType()->isVoidPointerType() && 7095 ICE->getCastKind() == CK_BitCast) 7096 ArgumentExpr = ICE->getSubExpr(); 7097 } 7098 QualType ArgumentType = ArgumentExpr->getType(); 7099 7100 // Passing a `void*' pointer shouldn't trigger a warning. 7101 if (IsPointerAttr && ArgumentType->isVoidPointerType()) 7102 return; 7103 7104 if (TypeInfo.MustBeNull) { 7105 // Type tag with matching void type requires a null pointer. 7106 if (!ArgumentExpr->isNullPointerConstant(Context, 7107 Expr::NPC_ValueDependentIsNotNull)) { 7108 Diag(ArgumentExpr->getExprLoc(), 7109 diag::warn_type_safety_null_pointer_required) 7110 << ArgumentKind->getName() 7111 << ArgumentExpr->getSourceRange() 7112 << TypeTagExpr->getSourceRange(); 7113 } 7114 return; 7115 } 7116 7117 QualType RequiredType = TypeInfo.Type; 7118 if (IsPointerAttr) 7119 RequiredType = Context.getPointerType(RequiredType); 7120 7121 bool mismatch = false; 7122 if (!TypeInfo.LayoutCompatible) { 7123 mismatch = !Context.hasSameType(ArgumentType, RequiredType); 7124 7125 // C++11 [basic.fundamental] p1: 7126 // Plain char, signed char, and unsigned char are three distinct types. 7127 // 7128 // But we treat plain `char' as equivalent to `signed char' or `unsigned 7129 // char' depending on the current char signedness mode. 7130 if (mismatch) 7131 if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(), 7132 RequiredType->getPointeeType())) || 7133 (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType))) 7134 mismatch = false; 7135 } else 7136 if (IsPointerAttr) 7137 mismatch = !isLayoutCompatible(Context, 7138 ArgumentType->getPointeeType(), 7139 RequiredType->getPointeeType()); 7140 else 7141 mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType); 7142 7143 if (mismatch) 7144 Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch) 7145 << ArgumentType << ArgumentKind->getName() 7146 << TypeInfo.LayoutCompatible << RequiredType 7147 << ArgumentExpr->getSourceRange() 7148 << TypeTagExpr->getSourceRange(); 7149} 7150