SemaChecking.cpp revision 624bb5e59dbcc24efeee7dff12c9b48d2b5077e9
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements extra semantic analysis beyond what is enforced 11// by the C type system. 12// 13//===----------------------------------------------------------------------===// 14 15#include "clang/Sema/SemaInternal.h" 16#include "clang/AST/ASTContext.h" 17#include "clang/AST/CharUnits.h" 18#include "clang/AST/DeclCXX.h" 19#include "clang/AST/DeclObjC.h" 20#include "clang/AST/EvaluatedExprVisitor.h" 21#include "clang/AST/Expr.h" 22#include "clang/AST/ExprCXX.h" 23#include "clang/AST/ExprObjC.h" 24#include "clang/AST/StmtCXX.h" 25#include "clang/AST/StmtObjC.h" 26#include "clang/Analysis/Analyses/FormatString.h" 27#include "clang/Basic/CharInfo.h" 28#include "clang/Basic/TargetBuiltins.h" 29#include "clang/Basic/TargetInfo.h" 30#include "clang/Lex/Preprocessor.h" 31#include "clang/Sema/Initialization.h" 32#include "clang/Sema/Lookup.h" 33#include "clang/Sema/ScopeInfo.h" 34#include "clang/Sema/Sema.h" 35#include "llvm/ADT/SmallBitVector.h" 36#include "llvm/ADT/SmallString.h" 37#include "llvm/ADT/STLExtras.h" 38#include "llvm/Support/ConvertUTF.h" 39#include "llvm/Support/raw_ostream.h" 40#include <limits> 41using namespace clang; 42using namespace sema; 43 44SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 45 unsigned ByteNo) const { 46 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 47 PP.getLangOpts(), PP.getTargetInfo()); 48} 49 50/// Checks that a call expression's argument count is the desired number. 51/// This is useful when doing custom type-checking. Returns true on error. 52static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 53 unsigned argCount = call->getNumArgs(); 54 if (argCount == desiredArgCount) return false; 55 56 if (argCount < desiredArgCount) 57 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 58 << 0 /*function call*/ << desiredArgCount << argCount 59 << call->getSourceRange(); 60 61 // Highlight all the excess arguments. 62 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 63 call->getArg(argCount - 1)->getLocEnd()); 64 65 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 66 << 0 /*function call*/ << desiredArgCount << argCount 67 << call->getArg(1)->getSourceRange(); 68} 69 70/// Check that the first argument to __builtin_annotation is an integer 71/// and the second argument is a non-wide string literal. 72static bool SemaBuiltinAnnotation(Sema &S, CallExpr *TheCall) { 73 if (checkArgCount(S, TheCall, 2)) 74 return true; 75 76 // First argument should be an integer. 77 Expr *ValArg = TheCall->getArg(0); 78 QualType Ty = ValArg->getType(); 79 if (!Ty->isIntegerType()) { 80 S.Diag(ValArg->getLocStart(), diag::err_builtin_annotation_first_arg) 81 << ValArg->getSourceRange(); 82 return true; 83 } 84 85 // Second argument should be a constant string. 86 Expr *StrArg = TheCall->getArg(1)->IgnoreParenCasts(); 87 StringLiteral *Literal = dyn_cast<StringLiteral>(StrArg); 88 if (!Literal || !Literal->isAscii()) { 89 S.Diag(StrArg->getLocStart(), diag::err_builtin_annotation_second_arg) 90 << StrArg->getSourceRange(); 91 return true; 92 } 93 94 TheCall->setType(Ty); 95 return false; 96} 97 98/// Check that the argument to __builtin_addressof is a glvalue, and set the 99/// result type to the corresponding pointer type. 100static bool SemaBuiltinAddressof(Sema &S, CallExpr *TheCall) { 101 if (checkArgCount(S, TheCall, 1)) 102 return true; 103 104 ExprResult Arg(S.Owned(TheCall->getArg(0))); 105 QualType ResultType = S.CheckAddressOfOperand(Arg, TheCall->getLocStart()); 106 if (ResultType.isNull()) 107 return true; 108 109 TheCall->setArg(0, Arg.take()); 110 TheCall->setType(ResultType); 111 return false; 112} 113 114ExprResult 115Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 116 ExprResult TheCallResult(Owned(TheCall)); 117 118 // Find out if any arguments are required to be integer constant expressions. 119 unsigned ICEArguments = 0; 120 ASTContext::GetBuiltinTypeError Error; 121 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 122 if (Error != ASTContext::GE_None) 123 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 124 125 // If any arguments are required to be ICE's, check and diagnose. 126 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 127 // Skip arguments not required to be ICE's. 128 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 129 130 llvm::APSInt Result; 131 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 132 return true; 133 ICEArguments &= ~(1 << ArgNo); 134 } 135 136 switch (BuiltinID) { 137 case Builtin::BI__builtin___CFStringMakeConstantString: 138 assert(TheCall->getNumArgs() == 1 && 139 "Wrong # arguments to builtin CFStringMakeConstantString"); 140 if (CheckObjCString(TheCall->getArg(0))) 141 return ExprError(); 142 break; 143 case Builtin::BI__builtin_stdarg_start: 144 case Builtin::BI__builtin_va_start: 145 if (SemaBuiltinVAStart(TheCall)) 146 return ExprError(); 147 break; 148 case Builtin::BI__builtin_isgreater: 149 case Builtin::BI__builtin_isgreaterequal: 150 case Builtin::BI__builtin_isless: 151 case Builtin::BI__builtin_islessequal: 152 case Builtin::BI__builtin_islessgreater: 153 case Builtin::BI__builtin_isunordered: 154 if (SemaBuiltinUnorderedCompare(TheCall)) 155 return ExprError(); 156 break; 157 case Builtin::BI__builtin_fpclassify: 158 if (SemaBuiltinFPClassification(TheCall, 6)) 159 return ExprError(); 160 break; 161 case Builtin::BI__builtin_isfinite: 162 case Builtin::BI__builtin_isinf: 163 case Builtin::BI__builtin_isinf_sign: 164 case Builtin::BI__builtin_isnan: 165 case Builtin::BI__builtin_isnormal: 166 if (SemaBuiltinFPClassification(TheCall, 1)) 167 return ExprError(); 168 break; 169 case Builtin::BI__builtin_shufflevector: 170 return SemaBuiltinShuffleVector(TheCall); 171 // TheCall will be freed by the smart pointer here, but that's fine, since 172 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 173 case Builtin::BI__builtin_prefetch: 174 if (SemaBuiltinPrefetch(TheCall)) 175 return ExprError(); 176 break; 177 case Builtin::BI__builtin_object_size: 178 if (SemaBuiltinObjectSize(TheCall)) 179 return ExprError(); 180 break; 181 case Builtin::BI__builtin_longjmp: 182 if (SemaBuiltinLongjmp(TheCall)) 183 return ExprError(); 184 break; 185 186 case Builtin::BI__builtin_classify_type: 187 if (checkArgCount(*this, TheCall, 1)) return true; 188 TheCall->setType(Context.IntTy); 189 break; 190 case Builtin::BI__builtin_constant_p: 191 if (checkArgCount(*this, TheCall, 1)) return true; 192 TheCall->setType(Context.IntTy); 193 break; 194 case Builtin::BI__sync_fetch_and_add: 195 case Builtin::BI__sync_fetch_and_add_1: 196 case Builtin::BI__sync_fetch_and_add_2: 197 case Builtin::BI__sync_fetch_and_add_4: 198 case Builtin::BI__sync_fetch_and_add_8: 199 case Builtin::BI__sync_fetch_and_add_16: 200 case Builtin::BI__sync_fetch_and_sub: 201 case Builtin::BI__sync_fetch_and_sub_1: 202 case Builtin::BI__sync_fetch_and_sub_2: 203 case Builtin::BI__sync_fetch_and_sub_4: 204 case Builtin::BI__sync_fetch_and_sub_8: 205 case Builtin::BI__sync_fetch_and_sub_16: 206 case Builtin::BI__sync_fetch_and_or: 207 case Builtin::BI__sync_fetch_and_or_1: 208 case Builtin::BI__sync_fetch_and_or_2: 209 case Builtin::BI__sync_fetch_and_or_4: 210 case Builtin::BI__sync_fetch_and_or_8: 211 case Builtin::BI__sync_fetch_and_or_16: 212 case Builtin::BI__sync_fetch_and_and: 213 case Builtin::BI__sync_fetch_and_and_1: 214 case Builtin::BI__sync_fetch_and_and_2: 215 case Builtin::BI__sync_fetch_and_and_4: 216 case Builtin::BI__sync_fetch_and_and_8: 217 case Builtin::BI__sync_fetch_and_and_16: 218 case Builtin::BI__sync_fetch_and_xor: 219 case Builtin::BI__sync_fetch_and_xor_1: 220 case Builtin::BI__sync_fetch_and_xor_2: 221 case Builtin::BI__sync_fetch_and_xor_4: 222 case Builtin::BI__sync_fetch_and_xor_8: 223 case Builtin::BI__sync_fetch_and_xor_16: 224 case Builtin::BI__sync_add_and_fetch: 225 case Builtin::BI__sync_add_and_fetch_1: 226 case Builtin::BI__sync_add_and_fetch_2: 227 case Builtin::BI__sync_add_and_fetch_4: 228 case Builtin::BI__sync_add_and_fetch_8: 229 case Builtin::BI__sync_add_and_fetch_16: 230 case Builtin::BI__sync_sub_and_fetch: 231 case Builtin::BI__sync_sub_and_fetch_1: 232 case Builtin::BI__sync_sub_and_fetch_2: 233 case Builtin::BI__sync_sub_and_fetch_4: 234 case Builtin::BI__sync_sub_and_fetch_8: 235 case Builtin::BI__sync_sub_and_fetch_16: 236 case Builtin::BI__sync_and_and_fetch: 237 case Builtin::BI__sync_and_and_fetch_1: 238 case Builtin::BI__sync_and_and_fetch_2: 239 case Builtin::BI__sync_and_and_fetch_4: 240 case Builtin::BI__sync_and_and_fetch_8: 241 case Builtin::BI__sync_and_and_fetch_16: 242 case Builtin::BI__sync_or_and_fetch: 243 case Builtin::BI__sync_or_and_fetch_1: 244 case Builtin::BI__sync_or_and_fetch_2: 245 case Builtin::BI__sync_or_and_fetch_4: 246 case Builtin::BI__sync_or_and_fetch_8: 247 case Builtin::BI__sync_or_and_fetch_16: 248 case Builtin::BI__sync_xor_and_fetch: 249 case Builtin::BI__sync_xor_and_fetch_1: 250 case Builtin::BI__sync_xor_and_fetch_2: 251 case Builtin::BI__sync_xor_and_fetch_4: 252 case Builtin::BI__sync_xor_and_fetch_8: 253 case Builtin::BI__sync_xor_and_fetch_16: 254 case Builtin::BI__sync_val_compare_and_swap: 255 case Builtin::BI__sync_val_compare_and_swap_1: 256 case Builtin::BI__sync_val_compare_and_swap_2: 257 case Builtin::BI__sync_val_compare_and_swap_4: 258 case Builtin::BI__sync_val_compare_and_swap_8: 259 case Builtin::BI__sync_val_compare_and_swap_16: 260 case Builtin::BI__sync_bool_compare_and_swap: 261 case Builtin::BI__sync_bool_compare_and_swap_1: 262 case Builtin::BI__sync_bool_compare_and_swap_2: 263 case Builtin::BI__sync_bool_compare_and_swap_4: 264 case Builtin::BI__sync_bool_compare_and_swap_8: 265 case Builtin::BI__sync_bool_compare_and_swap_16: 266 case Builtin::BI__sync_lock_test_and_set: 267 case Builtin::BI__sync_lock_test_and_set_1: 268 case Builtin::BI__sync_lock_test_and_set_2: 269 case Builtin::BI__sync_lock_test_and_set_4: 270 case Builtin::BI__sync_lock_test_and_set_8: 271 case Builtin::BI__sync_lock_test_and_set_16: 272 case Builtin::BI__sync_lock_release: 273 case Builtin::BI__sync_lock_release_1: 274 case Builtin::BI__sync_lock_release_2: 275 case Builtin::BI__sync_lock_release_4: 276 case Builtin::BI__sync_lock_release_8: 277 case Builtin::BI__sync_lock_release_16: 278 case Builtin::BI__sync_swap: 279 case Builtin::BI__sync_swap_1: 280 case Builtin::BI__sync_swap_2: 281 case Builtin::BI__sync_swap_4: 282 case Builtin::BI__sync_swap_8: 283 case Builtin::BI__sync_swap_16: 284 return SemaBuiltinAtomicOverloaded(TheCallResult); 285#define BUILTIN(ID, TYPE, ATTRS) 286#define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \ 287 case Builtin::BI##ID: \ 288 return SemaAtomicOpsOverloaded(TheCallResult, AtomicExpr::AO##ID); 289#include "clang/Basic/Builtins.def" 290 case Builtin::BI__builtin_annotation: 291 if (SemaBuiltinAnnotation(*this, TheCall)) 292 return ExprError(); 293 break; 294 case Builtin::BI__builtin_addressof: 295 if (SemaBuiltinAddressof(*this, TheCall)) 296 return ExprError(); 297 break; 298 } 299 300 // Since the target specific builtins for each arch overlap, only check those 301 // of the arch we are compiling for. 302 if (BuiltinID >= Builtin::FirstTSBuiltin) { 303 switch (Context.getTargetInfo().getTriple().getArch()) { 304 case llvm::Triple::arm: 305 case llvm::Triple::thumb: 306 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 307 return ExprError(); 308 break; 309 case llvm::Triple::aarch64: 310 if (CheckAArch64BuiltinFunctionCall(BuiltinID, TheCall)) 311 return ExprError(); 312 break; 313 case llvm::Triple::mips: 314 case llvm::Triple::mipsel: 315 case llvm::Triple::mips64: 316 case llvm::Triple::mips64el: 317 if (CheckMipsBuiltinFunctionCall(BuiltinID, TheCall)) 318 return ExprError(); 319 break; 320 default: 321 break; 322 } 323 } 324 325 return TheCallResult; 326} 327 328// Get the valid immediate range for the specified NEON type code. 329static unsigned RFT(unsigned t, bool shift = false) { 330 NeonTypeFlags Type(t); 331 int IsQuad = Type.isQuad(); 332 switch (Type.getEltType()) { 333 case NeonTypeFlags::Int8: 334 case NeonTypeFlags::Poly8: 335 return shift ? 7 : (8 << IsQuad) - 1; 336 case NeonTypeFlags::Int16: 337 case NeonTypeFlags::Poly16: 338 return shift ? 15 : (4 << IsQuad) - 1; 339 case NeonTypeFlags::Int32: 340 return shift ? 31 : (2 << IsQuad) - 1; 341 case NeonTypeFlags::Int64: 342 case NeonTypeFlags::Poly64: 343 return shift ? 63 : (1 << IsQuad) - 1; 344 case NeonTypeFlags::Float16: 345 assert(!shift && "cannot shift float types!"); 346 return (4 << IsQuad) - 1; 347 case NeonTypeFlags::Float32: 348 assert(!shift && "cannot shift float types!"); 349 return (2 << IsQuad) - 1; 350 case NeonTypeFlags::Float64: 351 assert(!shift && "cannot shift float types!"); 352 return (1 << IsQuad) - 1; 353 } 354 llvm_unreachable("Invalid NeonTypeFlag!"); 355} 356 357/// getNeonEltType - Return the QualType corresponding to the elements of 358/// the vector type specified by the NeonTypeFlags. This is used to check 359/// the pointer arguments for Neon load/store intrinsics. 360static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context, 361 bool IsAArch64) { 362 switch (Flags.getEltType()) { 363 case NeonTypeFlags::Int8: 364 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 365 case NeonTypeFlags::Int16: 366 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 367 case NeonTypeFlags::Int32: 368 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 369 case NeonTypeFlags::Int64: 370 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 371 case NeonTypeFlags::Poly8: 372 return IsAArch64 ? Context.UnsignedCharTy : Context.SignedCharTy; 373 case NeonTypeFlags::Poly16: 374 return IsAArch64 ? Context.UnsignedShortTy : Context.ShortTy; 375 case NeonTypeFlags::Poly64: 376 return Context.UnsignedLongLongTy; 377 case NeonTypeFlags::Float16: 378 return Context.HalfTy; 379 case NeonTypeFlags::Float32: 380 return Context.FloatTy; 381 case NeonTypeFlags::Float64: 382 return Context.DoubleTy; 383 } 384 llvm_unreachable("Invalid NeonTypeFlag!"); 385} 386 387bool Sema::CheckAArch64BuiltinFunctionCall(unsigned BuiltinID, 388 CallExpr *TheCall) { 389 390 llvm::APSInt Result; 391 392 uint64_t mask = 0; 393 unsigned TV = 0; 394 int PtrArgNum = -1; 395 bool HasConstPtr = false; 396 switch (BuiltinID) { 397#define GET_NEON_AARCH64_OVERLOAD_CHECK 398#include "clang/Basic/arm_neon.inc" 399#undef GET_NEON_AARCH64_OVERLOAD_CHECK 400 } 401 402 // For NEON intrinsics which are overloaded on vector element type, validate 403 // the immediate which specifies which variant to emit. 404 unsigned ImmArg = TheCall->getNumArgs() - 1; 405 if (mask) { 406 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 407 return true; 408 409 TV = Result.getLimitedValue(64); 410 if ((TV > 63) || (mask & (1ULL << TV)) == 0) 411 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 412 << TheCall->getArg(ImmArg)->getSourceRange(); 413 } 414 415 if (PtrArgNum >= 0) { 416 // Check that pointer arguments have the specified type. 417 Expr *Arg = TheCall->getArg(PtrArgNum); 418 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 419 Arg = ICE->getSubExpr(); 420 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 421 QualType RHSTy = RHS.get()->getType(); 422 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context, true); 423 if (HasConstPtr) 424 EltTy = EltTy.withConst(); 425 QualType LHSTy = Context.getPointerType(EltTy); 426 AssignConvertType ConvTy; 427 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 428 if (RHS.isInvalid()) 429 return true; 430 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 431 RHS.get(), AA_Assigning)) 432 return true; 433 } 434 435 // For NEON intrinsics which take an immediate value as part of the 436 // instruction, range check them here. 437 unsigned i = 0, l = 0, u = 0; 438 switch (BuiltinID) { 439 default: 440 return false; 441#define GET_NEON_AARCH64_IMMEDIATE_CHECK 442#include "clang/Basic/arm_neon.inc" 443#undef GET_NEON_AARCH64_IMMEDIATE_CHECK 444 } 445 ; 446 447 // We can't check the value of a dependent argument. 448 if (TheCall->getArg(i)->isTypeDependent() || 449 TheCall->getArg(i)->isValueDependent()) 450 return false; 451 452 // Check that the immediate argument is actually a constant. 453 if (SemaBuiltinConstantArg(TheCall, i, Result)) 454 return true; 455 456 // Range check against the upper/lower values for this isntruction. 457 unsigned Val = Result.getZExtValue(); 458 if (Val < l || Val > (u + l)) 459 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 460 << l << u + l << TheCall->getArg(i)->getSourceRange(); 461 462 return false; 463} 464 465bool Sema::CheckARMBuiltinExclusiveCall(unsigned BuiltinID, CallExpr *TheCall) { 466 assert((BuiltinID == ARM::BI__builtin_arm_ldrex || 467 BuiltinID == ARM::BI__builtin_arm_strex) && 468 "unexpected ARM builtin"); 469 bool IsLdrex = BuiltinID == ARM::BI__builtin_arm_ldrex; 470 471 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 472 473 // Ensure that we have the proper number of arguments. 474 if (checkArgCount(*this, TheCall, IsLdrex ? 1 : 2)) 475 return true; 476 477 // Inspect the pointer argument of the atomic builtin. This should always be 478 // a pointer type, whose element is an integral scalar or pointer type. 479 // Because it is a pointer type, we don't have to worry about any implicit 480 // casts here. 481 Expr *PointerArg = TheCall->getArg(IsLdrex ? 0 : 1); 482 ExprResult PointerArgRes = DefaultFunctionArrayLvalueConversion(PointerArg); 483 if (PointerArgRes.isInvalid()) 484 return true; 485 PointerArg = PointerArgRes.take(); 486 487 const PointerType *pointerType = PointerArg->getType()->getAs<PointerType>(); 488 if (!pointerType) { 489 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 490 << PointerArg->getType() << PointerArg->getSourceRange(); 491 return true; 492 } 493 494 // ldrex takes a "const volatile T*" and strex takes a "volatile T*". Our next 495 // task is to insert the appropriate casts into the AST. First work out just 496 // what the appropriate type is. 497 QualType ValType = pointerType->getPointeeType(); 498 QualType AddrType = ValType.getUnqualifiedType().withVolatile(); 499 if (IsLdrex) 500 AddrType.addConst(); 501 502 // Issue a warning if the cast is dodgy. 503 CastKind CastNeeded = CK_NoOp; 504 if (!AddrType.isAtLeastAsQualifiedAs(ValType)) { 505 CastNeeded = CK_BitCast; 506 Diag(DRE->getLocStart(), diag::ext_typecheck_convert_discards_qualifiers) 507 << PointerArg->getType() 508 << Context.getPointerType(AddrType) 509 << AA_Passing << PointerArg->getSourceRange(); 510 } 511 512 // Finally, do the cast and replace the argument with the corrected version. 513 AddrType = Context.getPointerType(AddrType); 514 PointerArgRes = ImpCastExprToType(PointerArg, AddrType, CastNeeded); 515 if (PointerArgRes.isInvalid()) 516 return true; 517 PointerArg = PointerArgRes.take(); 518 519 TheCall->setArg(IsLdrex ? 0 : 1, PointerArg); 520 521 // In general, we allow ints, floats and pointers to be loaded and stored. 522 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 523 !ValType->isBlockPointerType() && !ValType->isFloatingType()) { 524 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intfltptr) 525 << PointerArg->getType() << PointerArg->getSourceRange(); 526 return true; 527 } 528 529 // But ARM doesn't have instructions to deal with 128-bit versions. 530 if (Context.getTypeSize(ValType) > 64) { 531 Diag(DRE->getLocStart(), diag::err_atomic_exclusive_builtin_pointer_size) 532 << PointerArg->getType() << PointerArg->getSourceRange(); 533 return true; 534 } 535 536 switch (ValType.getObjCLifetime()) { 537 case Qualifiers::OCL_None: 538 case Qualifiers::OCL_ExplicitNone: 539 // okay 540 break; 541 542 case Qualifiers::OCL_Weak: 543 case Qualifiers::OCL_Strong: 544 case Qualifiers::OCL_Autoreleasing: 545 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 546 << ValType << PointerArg->getSourceRange(); 547 return true; 548 } 549 550 551 if (IsLdrex) { 552 TheCall->setType(ValType); 553 return false; 554 } 555 556 // Initialize the argument to be stored. 557 ExprResult ValArg = TheCall->getArg(0); 558 InitializedEntity Entity = InitializedEntity::InitializeParameter( 559 Context, ValType, /*consume*/ false); 560 ValArg = PerformCopyInitialization(Entity, SourceLocation(), ValArg); 561 if (ValArg.isInvalid()) 562 return true; 563 TheCall->setArg(0, ValArg.get()); 564 565 // __builtin_arm_strex always returns an int. It's marked as such in the .def, 566 // but the custom checker bypasses all default analysis. 567 TheCall->setType(Context.IntTy); 568 return false; 569} 570 571bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 572 llvm::APSInt Result; 573 574 if (BuiltinID == ARM::BI__builtin_arm_ldrex || 575 BuiltinID == ARM::BI__builtin_arm_strex) { 576 return CheckARMBuiltinExclusiveCall(BuiltinID, TheCall); 577 } 578 579 uint64_t mask = 0; 580 unsigned TV = 0; 581 int PtrArgNum = -1; 582 bool HasConstPtr = false; 583 switch (BuiltinID) { 584#define GET_NEON_OVERLOAD_CHECK 585#include "clang/Basic/arm_neon.inc" 586#undef GET_NEON_OVERLOAD_CHECK 587 } 588 589 // For NEON intrinsics which are overloaded on vector element type, validate 590 // the immediate which specifies which variant to emit. 591 unsigned ImmArg = TheCall->getNumArgs()-1; 592 if (mask) { 593 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 594 return true; 595 596 TV = Result.getLimitedValue(64); 597 if ((TV > 63) || (mask & (1ULL << TV)) == 0) 598 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 599 << TheCall->getArg(ImmArg)->getSourceRange(); 600 } 601 602 if (PtrArgNum >= 0) { 603 // Check that pointer arguments have the specified type. 604 Expr *Arg = TheCall->getArg(PtrArgNum); 605 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 606 Arg = ICE->getSubExpr(); 607 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 608 QualType RHSTy = RHS.get()->getType(); 609 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context, false); 610 if (HasConstPtr) 611 EltTy = EltTy.withConst(); 612 QualType LHSTy = Context.getPointerType(EltTy); 613 AssignConvertType ConvTy; 614 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 615 if (RHS.isInvalid()) 616 return true; 617 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 618 RHS.get(), AA_Assigning)) 619 return true; 620 } 621 622 // For NEON intrinsics which take an immediate value as part of the 623 // instruction, range check them here. 624 unsigned i = 0, l = 0, u = 0; 625 switch (BuiltinID) { 626 default: return false; 627 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 628 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 629 case ARM::BI__builtin_arm_vcvtr_f: 630 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 631 case ARM::BI__builtin_arm_dmb: 632 case ARM::BI__builtin_arm_dsb: l = 0; u = 15; break; 633#define GET_NEON_IMMEDIATE_CHECK 634#include "clang/Basic/arm_neon.inc" 635#undef GET_NEON_IMMEDIATE_CHECK 636 }; 637 638 // We can't check the value of a dependent argument. 639 if (TheCall->getArg(i)->isTypeDependent() || 640 TheCall->getArg(i)->isValueDependent()) 641 return false; 642 643 // Check that the immediate argument is actually a constant. 644 if (SemaBuiltinConstantArg(TheCall, i, Result)) 645 return true; 646 647 // Range check against the upper/lower values for this isntruction. 648 unsigned Val = Result.getZExtValue(); 649 if (Val < l || Val > (u + l)) 650 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 651 << l << u+l << TheCall->getArg(i)->getSourceRange(); 652 653 // FIXME: VFP Intrinsics should error if VFP not present. 654 return false; 655} 656 657bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 658 unsigned i = 0, l = 0, u = 0; 659 switch (BuiltinID) { 660 default: return false; 661 case Mips::BI__builtin_mips_wrdsp: i = 1; l = 0; u = 63; break; 662 case Mips::BI__builtin_mips_rddsp: i = 0; l = 0; u = 63; break; 663 case Mips::BI__builtin_mips_append: i = 2; l = 0; u = 31; break; 664 case Mips::BI__builtin_mips_balign: i = 2; l = 0; u = 3; break; 665 case Mips::BI__builtin_mips_precr_sra_ph_w: i = 2; l = 0; u = 31; break; 666 case Mips::BI__builtin_mips_precr_sra_r_ph_w: i = 2; l = 0; u = 31; break; 667 case Mips::BI__builtin_mips_prepend: i = 2; l = 0; u = 31; break; 668 }; 669 670 // We can't check the value of a dependent argument. 671 if (TheCall->getArg(i)->isTypeDependent() || 672 TheCall->getArg(i)->isValueDependent()) 673 return false; 674 675 // Check that the immediate argument is actually a constant. 676 llvm::APSInt Result; 677 if (SemaBuiltinConstantArg(TheCall, i, Result)) 678 return true; 679 680 // Range check against the upper/lower values for this instruction. 681 unsigned Val = Result.getZExtValue(); 682 if (Val < l || Val > u) 683 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 684 << l << u << TheCall->getArg(i)->getSourceRange(); 685 686 return false; 687} 688 689/// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo 690/// parameter with the FormatAttr's correct format_idx and firstDataArg. 691/// Returns true when the format fits the function and the FormatStringInfo has 692/// been populated. 693bool Sema::getFormatStringInfo(const FormatAttr *Format, bool IsCXXMember, 694 FormatStringInfo *FSI) { 695 FSI->HasVAListArg = Format->getFirstArg() == 0; 696 FSI->FormatIdx = Format->getFormatIdx() - 1; 697 FSI->FirstDataArg = FSI->HasVAListArg ? 0 : Format->getFirstArg() - 1; 698 699 // The way the format attribute works in GCC, the implicit this argument 700 // of member functions is counted. However, it doesn't appear in our own 701 // lists, so decrement format_idx in that case. 702 if (IsCXXMember) { 703 if(FSI->FormatIdx == 0) 704 return false; 705 --FSI->FormatIdx; 706 if (FSI->FirstDataArg != 0) 707 --FSI->FirstDataArg; 708 } 709 return true; 710} 711 712/// Handles the checks for format strings, non-POD arguments to vararg 713/// functions, and NULL arguments passed to non-NULL parameters. 714void Sema::checkCall(NamedDecl *FDecl, 715 ArrayRef<const Expr *> Args, 716 unsigned NumProtoArgs, 717 bool IsMemberFunction, 718 SourceLocation Loc, 719 SourceRange Range, 720 VariadicCallType CallType) { 721 // FIXME: We should check as much as we can in the template definition. 722 if (CurContext->isDependentContext()) 723 return; 724 725 // Printf and scanf checking. 726 llvm::SmallBitVector CheckedVarArgs; 727 if (FDecl) { 728 for (specific_attr_iterator<FormatAttr> 729 I = FDecl->specific_attr_begin<FormatAttr>(), 730 E = FDecl->specific_attr_end<FormatAttr>(); 731 I != E; ++I) { 732 // Only create vector if there are format attributes. 733 CheckedVarArgs.resize(Args.size()); 734 735 CheckFormatArguments(*I, Args, IsMemberFunction, CallType, Loc, Range, 736 CheckedVarArgs); 737 } 738 } 739 740 // Refuse POD arguments that weren't caught by the format string 741 // checks above. 742 if (CallType != VariadicDoesNotApply) { 743 for (unsigned ArgIdx = NumProtoArgs; ArgIdx < Args.size(); ++ArgIdx) { 744 // Args[ArgIdx] can be null in malformed code. 745 if (const Expr *Arg = Args[ArgIdx]) { 746 if (CheckedVarArgs.empty() || !CheckedVarArgs[ArgIdx]) 747 checkVariadicArgument(Arg, CallType); 748 } 749 } 750 } 751 752 if (FDecl) { 753 for (specific_attr_iterator<NonNullAttr> 754 I = FDecl->specific_attr_begin<NonNullAttr>(), 755 E = FDecl->specific_attr_end<NonNullAttr>(); I != E; ++I) 756 CheckNonNullArguments(*I, Args.data(), Loc); 757 758 // Type safety checking. 759 for (specific_attr_iterator<ArgumentWithTypeTagAttr> 760 i = FDecl->specific_attr_begin<ArgumentWithTypeTagAttr>(), 761 e = FDecl->specific_attr_end<ArgumentWithTypeTagAttr>(); 762 i != e; ++i) { 763 CheckArgumentWithTypeTag(*i, Args.data()); 764 } 765 } 766} 767 768/// CheckConstructorCall - Check a constructor call for correctness and safety 769/// properties not enforced by the C type system. 770void Sema::CheckConstructorCall(FunctionDecl *FDecl, 771 ArrayRef<const Expr *> Args, 772 const FunctionProtoType *Proto, 773 SourceLocation Loc) { 774 VariadicCallType CallType = 775 Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply; 776 checkCall(FDecl, Args, Proto->getNumArgs(), 777 /*IsMemberFunction=*/true, Loc, SourceRange(), CallType); 778} 779 780/// CheckFunctionCall - Check a direct function call for various correctness 781/// and safety properties not strictly enforced by the C type system. 782bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall, 783 const FunctionProtoType *Proto) { 784 bool IsMemberOperatorCall = isa<CXXOperatorCallExpr>(TheCall) && 785 isa<CXXMethodDecl>(FDecl); 786 bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall) || 787 IsMemberOperatorCall; 788 VariadicCallType CallType = getVariadicCallType(FDecl, Proto, 789 TheCall->getCallee()); 790 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 791 Expr** Args = TheCall->getArgs(); 792 unsigned NumArgs = TheCall->getNumArgs(); 793 if (IsMemberOperatorCall) { 794 // If this is a call to a member operator, hide the first argument 795 // from checkCall. 796 // FIXME: Our choice of AST representation here is less than ideal. 797 ++Args; 798 --NumArgs; 799 } 800 checkCall(FDecl, llvm::makeArrayRef<const Expr *>(Args, NumArgs), 801 NumProtoArgs, 802 IsMemberFunction, TheCall->getRParenLoc(), 803 TheCall->getCallee()->getSourceRange(), CallType); 804 805 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 806 // None of the checks below are needed for functions that don't have 807 // simple names (e.g., C++ conversion functions). 808 if (!FnInfo) 809 return false; 810 811 unsigned CMId = FDecl->getMemoryFunctionKind(); 812 if (CMId == 0) 813 return false; 814 815 // Handle memory setting and copying functions. 816 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 817 CheckStrlcpycatArguments(TheCall, FnInfo); 818 else if (CMId == Builtin::BIstrncat) 819 CheckStrncatArguments(TheCall, FnInfo); 820 else 821 CheckMemaccessArguments(TheCall, CMId, FnInfo); 822 823 return false; 824} 825 826bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 827 ArrayRef<const Expr *> Args) { 828 VariadicCallType CallType = 829 Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply; 830 831 checkCall(Method, Args, Method->param_size(), 832 /*IsMemberFunction=*/false, 833 lbrac, Method->getSourceRange(), CallType); 834 835 return false; 836} 837 838bool Sema::CheckPointerCall(NamedDecl *NDecl, CallExpr *TheCall, 839 const FunctionProtoType *Proto) { 840 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 841 if (!V) 842 return false; 843 844 QualType Ty = V->getType(); 845 if (!Ty->isBlockPointerType() && !Ty->isFunctionPointerType()) 846 return false; 847 848 VariadicCallType CallType; 849 if (!Proto || !Proto->isVariadic()) { 850 CallType = VariadicDoesNotApply; 851 } else if (Ty->isBlockPointerType()) { 852 CallType = VariadicBlock; 853 } else { // Ty->isFunctionPointerType() 854 CallType = VariadicFunction; 855 } 856 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 857 858 checkCall(NDecl, 859 llvm::makeArrayRef<const Expr *>(TheCall->getArgs(), 860 TheCall->getNumArgs()), 861 NumProtoArgs, /*IsMemberFunction=*/false, 862 TheCall->getRParenLoc(), 863 TheCall->getCallee()->getSourceRange(), CallType); 864 865 return false; 866} 867 868/// Checks function calls when a FunctionDecl or a NamedDecl is not available, 869/// such as function pointers returned from functions. 870bool Sema::CheckOtherCall(CallExpr *TheCall, const FunctionProtoType *Proto) { 871 VariadicCallType CallType = getVariadicCallType(/*FDecl=*/0, Proto, 872 TheCall->getCallee()); 873 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 874 875 checkCall(/*FDecl=*/0, 876 llvm::makeArrayRef<const Expr *>(TheCall->getArgs(), 877 TheCall->getNumArgs()), 878 NumProtoArgs, /*IsMemberFunction=*/false, 879 TheCall->getRParenLoc(), 880 TheCall->getCallee()->getSourceRange(), CallType); 881 882 return false; 883} 884 885ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 886 AtomicExpr::AtomicOp Op) { 887 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 888 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 889 890 // All these operations take one of the following forms: 891 enum { 892 // C __c11_atomic_init(A *, C) 893 Init, 894 // C __c11_atomic_load(A *, int) 895 Load, 896 // void __atomic_load(A *, CP, int) 897 Copy, 898 // C __c11_atomic_add(A *, M, int) 899 Arithmetic, 900 // C __atomic_exchange_n(A *, CP, int) 901 Xchg, 902 // void __atomic_exchange(A *, C *, CP, int) 903 GNUXchg, 904 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 905 C11CmpXchg, 906 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 907 GNUCmpXchg 908 } Form = Init; 909 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 910 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 911 // where: 912 // C is an appropriate type, 913 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 914 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 915 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 916 // the int parameters are for orderings. 917 918 assert(AtomicExpr::AO__c11_atomic_init == 0 && 919 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 920 && "need to update code for modified C11 atomics"); 921 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 922 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 923 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 924 Op == AtomicExpr::AO__atomic_store_n || 925 Op == AtomicExpr::AO__atomic_exchange_n || 926 Op == AtomicExpr::AO__atomic_compare_exchange_n; 927 bool IsAddSub = false; 928 929 switch (Op) { 930 case AtomicExpr::AO__c11_atomic_init: 931 Form = Init; 932 break; 933 934 case AtomicExpr::AO__c11_atomic_load: 935 case AtomicExpr::AO__atomic_load_n: 936 Form = Load; 937 break; 938 939 case AtomicExpr::AO__c11_atomic_store: 940 case AtomicExpr::AO__atomic_load: 941 case AtomicExpr::AO__atomic_store: 942 case AtomicExpr::AO__atomic_store_n: 943 Form = Copy; 944 break; 945 946 case AtomicExpr::AO__c11_atomic_fetch_add: 947 case AtomicExpr::AO__c11_atomic_fetch_sub: 948 case AtomicExpr::AO__atomic_fetch_add: 949 case AtomicExpr::AO__atomic_fetch_sub: 950 case AtomicExpr::AO__atomic_add_fetch: 951 case AtomicExpr::AO__atomic_sub_fetch: 952 IsAddSub = true; 953 // Fall through. 954 case AtomicExpr::AO__c11_atomic_fetch_and: 955 case AtomicExpr::AO__c11_atomic_fetch_or: 956 case AtomicExpr::AO__c11_atomic_fetch_xor: 957 case AtomicExpr::AO__atomic_fetch_and: 958 case AtomicExpr::AO__atomic_fetch_or: 959 case AtomicExpr::AO__atomic_fetch_xor: 960 case AtomicExpr::AO__atomic_fetch_nand: 961 case AtomicExpr::AO__atomic_and_fetch: 962 case AtomicExpr::AO__atomic_or_fetch: 963 case AtomicExpr::AO__atomic_xor_fetch: 964 case AtomicExpr::AO__atomic_nand_fetch: 965 Form = Arithmetic; 966 break; 967 968 case AtomicExpr::AO__c11_atomic_exchange: 969 case AtomicExpr::AO__atomic_exchange_n: 970 Form = Xchg; 971 break; 972 973 case AtomicExpr::AO__atomic_exchange: 974 Form = GNUXchg; 975 break; 976 977 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 978 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 979 Form = C11CmpXchg; 980 break; 981 982 case AtomicExpr::AO__atomic_compare_exchange: 983 case AtomicExpr::AO__atomic_compare_exchange_n: 984 Form = GNUCmpXchg; 985 break; 986 } 987 988 // Check we have the right number of arguments. 989 if (TheCall->getNumArgs() < NumArgs[Form]) { 990 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 991 << 0 << NumArgs[Form] << TheCall->getNumArgs() 992 << TheCall->getCallee()->getSourceRange(); 993 return ExprError(); 994 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 995 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 996 diag::err_typecheck_call_too_many_args) 997 << 0 << NumArgs[Form] << TheCall->getNumArgs() 998 << TheCall->getCallee()->getSourceRange(); 999 return ExprError(); 1000 } 1001 1002 // Inspect the first argument of the atomic operation. 1003 Expr *Ptr = TheCall->getArg(0); 1004 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 1005 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 1006 if (!pointerType) { 1007 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 1008 << Ptr->getType() << Ptr->getSourceRange(); 1009 return ExprError(); 1010 } 1011 1012 // For a __c11 builtin, this should be a pointer to an _Atomic type. 1013 QualType AtomTy = pointerType->getPointeeType(); // 'A' 1014 QualType ValType = AtomTy; // 'C' 1015 if (IsC11) { 1016 if (!AtomTy->isAtomicType()) { 1017 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 1018 << Ptr->getType() << Ptr->getSourceRange(); 1019 return ExprError(); 1020 } 1021 if (AtomTy.isConstQualified()) { 1022 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_non_const_atomic) 1023 << Ptr->getType() << Ptr->getSourceRange(); 1024 return ExprError(); 1025 } 1026 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 1027 } 1028 1029 // For an arithmetic operation, the implied arithmetic must be well-formed. 1030 if (Form == Arithmetic) { 1031 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 1032 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 1033 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 1034 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 1035 return ExprError(); 1036 } 1037 if (!IsAddSub && !ValType->isIntegerType()) { 1038 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 1039 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 1040 return ExprError(); 1041 } 1042 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 1043 // For __atomic_*_n operations, the value type must be a scalar integral or 1044 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 1045 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 1046 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 1047 return ExprError(); 1048 } 1049 1050 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context) && 1051 !AtomTy->isScalarType()) { 1052 // For GNU atomics, require a trivially-copyable type. This is not part of 1053 // the GNU atomics specification, but we enforce it for sanity. 1054 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 1055 << Ptr->getType() << Ptr->getSourceRange(); 1056 return ExprError(); 1057 } 1058 1059 // FIXME: For any builtin other than a load, the ValType must not be 1060 // const-qualified. 1061 1062 switch (ValType.getObjCLifetime()) { 1063 case Qualifiers::OCL_None: 1064 case Qualifiers::OCL_ExplicitNone: 1065 // okay 1066 break; 1067 1068 case Qualifiers::OCL_Weak: 1069 case Qualifiers::OCL_Strong: 1070 case Qualifiers::OCL_Autoreleasing: 1071 // FIXME: Can this happen? By this point, ValType should be known 1072 // to be trivially copyable. 1073 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1074 << ValType << Ptr->getSourceRange(); 1075 return ExprError(); 1076 } 1077 1078 QualType ResultType = ValType; 1079 if (Form == Copy || Form == GNUXchg || Form == Init) 1080 ResultType = Context.VoidTy; 1081 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 1082 ResultType = Context.BoolTy; 1083 1084 // The type of a parameter passed 'by value'. In the GNU atomics, such 1085 // arguments are actually passed as pointers. 1086 QualType ByValType = ValType; // 'CP' 1087 if (!IsC11 && !IsN) 1088 ByValType = Ptr->getType(); 1089 1090 // The first argument --- the pointer --- has a fixed type; we 1091 // deduce the types of the rest of the arguments accordingly. Walk 1092 // the remaining arguments, converting them to the deduced value type. 1093 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 1094 QualType Ty; 1095 if (i < NumVals[Form] + 1) { 1096 switch (i) { 1097 case 1: 1098 // The second argument is the non-atomic operand. For arithmetic, this 1099 // is always passed by value, and for a compare_exchange it is always 1100 // passed by address. For the rest, GNU uses by-address and C11 uses 1101 // by-value. 1102 assert(Form != Load); 1103 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 1104 Ty = ValType; 1105 else if (Form == Copy || Form == Xchg) 1106 Ty = ByValType; 1107 else if (Form == Arithmetic) 1108 Ty = Context.getPointerDiffType(); 1109 else 1110 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 1111 break; 1112 case 2: 1113 // The third argument to compare_exchange / GNU exchange is a 1114 // (pointer to a) desired value. 1115 Ty = ByValType; 1116 break; 1117 case 3: 1118 // The fourth argument to GNU compare_exchange is a 'weak' flag. 1119 Ty = Context.BoolTy; 1120 break; 1121 } 1122 } else { 1123 // The order(s) are always converted to int. 1124 Ty = Context.IntTy; 1125 } 1126 1127 InitializedEntity Entity = 1128 InitializedEntity::InitializeParameter(Context, Ty, false); 1129 ExprResult Arg = TheCall->getArg(i); 1130 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1131 if (Arg.isInvalid()) 1132 return true; 1133 TheCall->setArg(i, Arg.get()); 1134 } 1135 1136 // Permute the arguments into a 'consistent' order. 1137 SmallVector<Expr*, 5> SubExprs; 1138 SubExprs.push_back(Ptr); 1139 switch (Form) { 1140 case Init: 1141 // Note, AtomicExpr::getVal1() has a special case for this atomic. 1142 SubExprs.push_back(TheCall->getArg(1)); // Val1 1143 break; 1144 case Load: 1145 SubExprs.push_back(TheCall->getArg(1)); // Order 1146 break; 1147 case Copy: 1148 case Arithmetic: 1149 case Xchg: 1150 SubExprs.push_back(TheCall->getArg(2)); // Order 1151 SubExprs.push_back(TheCall->getArg(1)); // Val1 1152 break; 1153 case GNUXchg: 1154 // Note, AtomicExpr::getVal2() has a special case for this atomic. 1155 SubExprs.push_back(TheCall->getArg(3)); // Order 1156 SubExprs.push_back(TheCall->getArg(1)); // Val1 1157 SubExprs.push_back(TheCall->getArg(2)); // Val2 1158 break; 1159 case C11CmpXchg: 1160 SubExprs.push_back(TheCall->getArg(3)); // Order 1161 SubExprs.push_back(TheCall->getArg(1)); // Val1 1162 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 1163 SubExprs.push_back(TheCall->getArg(2)); // Val2 1164 break; 1165 case GNUCmpXchg: 1166 SubExprs.push_back(TheCall->getArg(4)); // Order 1167 SubExprs.push_back(TheCall->getArg(1)); // Val1 1168 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 1169 SubExprs.push_back(TheCall->getArg(2)); // Val2 1170 SubExprs.push_back(TheCall->getArg(3)); // Weak 1171 break; 1172 } 1173 1174 AtomicExpr *AE = new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 1175 SubExprs, ResultType, Op, 1176 TheCall->getRParenLoc()); 1177 1178 if ((Op == AtomicExpr::AO__c11_atomic_load || 1179 (Op == AtomicExpr::AO__c11_atomic_store)) && 1180 Context.AtomicUsesUnsupportedLibcall(AE)) 1181 Diag(AE->getLocStart(), diag::err_atomic_load_store_uses_lib) << 1182 ((Op == AtomicExpr::AO__c11_atomic_load) ? 0 : 1); 1183 1184 return Owned(AE); 1185} 1186 1187 1188/// checkBuiltinArgument - Given a call to a builtin function, perform 1189/// normal type-checking on the given argument, updating the call in 1190/// place. This is useful when a builtin function requires custom 1191/// type-checking for some of its arguments but not necessarily all of 1192/// them. 1193/// 1194/// Returns true on error. 1195static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 1196 FunctionDecl *Fn = E->getDirectCallee(); 1197 assert(Fn && "builtin call without direct callee!"); 1198 1199 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 1200 InitializedEntity Entity = 1201 InitializedEntity::InitializeParameter(S.Context, Param); 1202 1203 ExprResult Arg = E->getArg(0); 1204 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 1205 if (Arg.isInvalid()) 1206 return true; 1207 1208 E->setArg(ArgIndex, Arg.take()); 1209 return false; 1210} 1211 1212/// SemaBuiltinAtomicOverloaded - We have a call to a function like 1213/// __sync_fetch_and_add, which is an overloaded function based on the pointer 1214/// type of its first argument. The main ActOnCallExpr routines have already 1215/// promoted the types of arguments because all of these calls are prototyped as 1216/// void(...). 1217/// 1218/// This function goes through and does final semantic checking for these 1219/// builtins, 1220ExprResult 1221Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 1222 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 1223 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1224 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1225 1226 // Ensure that we have at least one argument to do type inference from. 1227 if (TheCall->getNumArgs() < 1) { 1228 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1229 << 0 << 1 << TheCall->getNumArgs() 1230 << TheCall->getCallee()->getSourceRange(); 1231 return ExprError(); 1232 } 1233 1234 // Inspect the first argument of the atomic builtin. This should always be 1235 // a pointer type, whose element is an integral scalar or pointer type. 1236 // Because it is a pointer type, we don't have to worry about any implicit 1237 // casts here. 1238 // FIXME: We don't allow floating point scalars as input. 1239 Expr *FirstArg = TheCall->getArg(0); 1240 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 1241 if (FirstArgResult.isInvalid()) 1242 return ExprError(); 1243 FirstArg = FirstArgResult.take(); 1244 TheCall->setArg(0, FirstArg); 1245 1246 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 1247 if (!pointerType) { 1248 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 1249 << FirstArg->getType() << FirstArg->getSourceRange(); 1250 return ExprError(); 1251 } 1252 1253 QualType ValType = pointerType->getPointeeType(); 1254 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 1255 !ValType->isBlockPointerType()) { 1256 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 1257 << FirstArg->getType() << FirstArg->getSourceRange(); 1258 return ExprError(); 1259 } 1260 1261 switch (ValType.getObjCLifetime()) { 1262 case Qualifiers::OCL_None: 1263 case Qualifiers::OCL_ExplicitNone: 1264 // okay 1265 break; 1266 1267 case Qualifiers::OCL_Weak: 1268 case Qualifiers::OCL_Strong: 1269 case Qualifiers::OCL_Autoreleasing: 1270 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1271 << ValType << FirstArg->getSourceRange(); 1272 return ExprError(); 1273 } 1274 1275 // Strip any qualifiers off ValType. 1276 ValType = ValType.getUnqualifiedType(); 1277 1278 // The majority of builtins return a value, but a few have special return 1279 // types, so allow them to override appropriately below. 1280 QualType ResultType = ValType; 1281 1282 // We need to figure out which concrete builtin this maps onto. For example, 1283 // __sync_fetch_and_add with a 2 byte object turns into 1284 // __sync_fetch_and_add_2. 1285#define BUILTIN_ROW(x) \ 1286 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 1287 Builtin::BI##x##_8, Builtin::BI##x##_16 } 1288 1289 static const unsigned BuiltinIndices[][5] = { 1290 BUILTIN_ROW(__sync_fetch_and_add), 1291 BUILTIN_ROW(__sync_fetch_and_sub), 1292 BUILTIN_ROW(__sync_fetch_and_or), 1293 BUILTIN_ROW(__sync_fetch_and_and), 1294 BUILTIN_ROW(__sync_fetch_and_xor), 1295 1296 BUILTIN_ROW(__sync_add_and_fetch), 1297 BUILTIN_ROW(__sync_sub_and_fetch), 1298 BUILTIN_ROW(__sync_and_and_fetch), 1299 BUILTIN_ROW(__sync_or_and_fetch), 1300 BUILTIN_ROW(__sync_xor_and_fetch), 1301 1302 BUILTIN_ROW(__sync_val_compare_and_swap), 1303 BUILTIN_ROW(__sync_bool_compare_and_swap), 1304 BUILTIN_ROW(__sync_lock_test_and_set), 1305 BUILTIN_ROW(__sync_lock_release), 1306 BUILTIN_ROW(__sync_swap) 1307 }; 1308#undef BUILTIN_ROW 1309 1310 // Determine the index of the size. 1311 unsigned SizeIndex; 1312 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 1313 case 1: SizeIndex = 0; break; 1314 case 2: SizeIndex = 1; break; 1315 case 4: SizeIndex = 2; break; 1316 case 8: SizeIndex = 3; break; 1317 case 16: SizeIndex = 4; break; 1318 default: 1319 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 1320 << FirstArg->getType() << FirstArg->getSourceRange(); 1321 return ExprError(); 1322 } 1323 1324 // Each of these builtins has one pointer argument, followed by some number of 1325 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 1326 // that we ignore. Find out which row of BuiltinIndices to read from as well 1327 // as the number of fixed args. 1328 unsigned BuiltinID = FDecl->getBuiltinID(); 1329 unsigned BuiltinIndex, NumFixed = 1; 1330 switch (BuiltinID) { 1331 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 1332 case Builtin::BI__sync_fetch_and_add: 1333 case Builtin::BI__sync_fetch_and_add_1: 1334 case Builtin::BI__sync_fetch_and_add_2: 1335 case Builtin::BI__sync_fetch_and_add_4: 1336 case Builtin::BI__sync_fetch_and_add_8: 1337 case Builtin::BI__sync_fetch_and_add_16: 1338 BuiltinIndex = 0; 1339 break; 1340 1341 case Builtin::BI__sync_fetch_and_sub: 1342 case Builtin::BI__sync_fetch_and_sub_1: 1343 case Builtin::BI__sync_fetch_and_sub_2: 1344 case Builtin::BI__sync_fetch_and_sub_4: 1345 case Builtin::BI__sync_fetch_and_sub_8: 1346 case Builtin::BI__sync_fetch_and_sub_16: 1347 BuiltinIndex = 1; 1348 break; 1349 1350 case Builtin::BI__sync_fetch_and_or: 1351 case Builtin::BI__sync_fetch_and_or_1: 1352 case Builtin::BI__sync_fetch_and_or_2: 1353 case Builtin::BI__sync_fetch_and_or_4: 1354 case Builtin::BI__sync_fetch_and_or_8: 1355 case Builtin::BI__sync_fetch_and_or_16: 1356 BuiltinIndex = 2; 1357 break; 1358 1359 case Builtin::BI__sync_fetch_and_and: 1360 case Builtin::BI__sync_fetch_and_and_1: 1361 case Builtin::BI__sync_fetch_and_and_2: 1362 case Builtin::BI__sync_fetch_and_and_4: 1363 case Builtin::BI__sync_fetch_and_and_8: 1364 case Builtin::BI__sync_fetch_and_and_16: 1365 BuiltinIndex = 3; 1366 break; 1367 1368 case Builtin::BI__sync_fetch_and_xor: 1369 case Builtin::BI__sync_fetch_and_xor_1: 1370 case Builtin::BI__sync_fetch_and_xor_2: 1371 case Builtin::BI__sync_fetch_and_xor_4: 1372 case Builtin::BI__sync_fetch_and_xor_8: 1373 case Builtin::BI__sync_fetch_and_xor_16: 1374 BuiltinIndex = 4; 1375 break; 1376 1377 case Builtin::BI__sync_add_and_fetch: 1378 case Builtin::BI__sync_add_and_fetch_1: 1379 case Builtin::BI__sync_add_and_fetch_2: 1380 case Builtin::BI__sync_add_and_fetch_4: 1381 case Builtin::BI__sync_add_and_fetch_8: 1382 case Builtin::BI__sync_add_and_fetch_16: 1383 BuiltinIndex = 5; 1384 break; 1385 1386 case Builtin::BI__sync_sub_and_fetch: 1387 case Builtin::BI__sync_sub_and_fetch_1: 1388 case Builtin::BI__sync_sub_and_fetch_2: 1389 case Builtin::BI__sync_sub_and_fetch_4: 1390 case Builtin::BI__sync_sub_and_fetch_8: 1391 case Builtin::BI__sync_sub_and_fetch_16: 1392 BuiltinIndex = 6; 1393 break; 1394 1395 case Builtin::BI__sync_and_and_fetch: 1396 case Builtin::BI__sync_and_and_fetch_1: 1397 case Builtin::BI__sync_and_and_fetch_2: 1398 case Builtin::BI__sync_and_and_fetch_4: 1399 case Builtin::BI__sync_and_and_fetch_8: 1400 case Builtin::BI__sync_and_and_fetch_16: 1401 BuiltinIndex = 7; 1402 break; 1403 1404 case Builtin::BI__sync_or_and_fetch: 1405 case Builtin::BI__sync_or_and_fetch_1: 1406 case Builtin::BI__sync_or_and_fetch_2: 1407 case Builtin::BI__sync_or_and_fetch_4: 1408 case Builtin::BI__sync_or_and_fetch_8: 1409 case Builtin::BI__sync_or_and_fetch_16: 1410 BuiltinIndex = 8; 1411 break; 1412 1413 case Builtin::BI__sync_xor_and_fetch: 1414 case Builtin::BI__sync_xor_and_fetch_1: 1415 case Builtin::BI__sync_xor_and_fetch_2: 1416 case Builtin::BI__sync_xor_and_fetch_4: 1417 case Builtin::BI__sync_xor_and_fetch_8: 1418 case Builtin::BI__sync_xor_and_fetch_16: 1419 BuiltinIndex = 9; 1420 break; 1421 1422 case Builtin::BI__sync_val_compare_and_swap: 1423 case Builtin::BI__sync_val_compare_and_swap_1: 1424 case Builtin::BI__sync_val_compare_and_swap_2: 1425 case Builtin::BI__sync_val_compare_and_swap_4: 1426 case Builtin::BI__sync_val_compare_and_swap_8: 1427 case Builtin::BI__sync_val_compare_and_swap_16: 1428 BuiltinIndex = 10; 1429 NumFixed = 2; 1430 break; 1431 1432 case Builtin::BI__sync_bool_compare_and_swap: 1433 case Builtin::BI__sync_bool_compare_and_swap_1: 1434 case Builtin::BI__sync_bool_compare_and_swap_2: 1435 case Builtin::BI__sync_bool_compare_and_swap_4: 1436 case Builtin::BI__sync_bool_compare_and_swap_8: 1437 case Builtin::BI__sync_bool_compare_and_swap_16: 1438 BuiltinIndex = 11; 1439 NumFixed = 2; 1440 ResultType = Context.BoolTy; 1441 break; 1442 1443 case Builtin::BI__sync_lock_test_and_set: 1444 case Builtin::BI__sync_lock_test_and_set_1: 1445 case Builtin::BI__sync_lock_test_and_set_2: 1446 case Builtin::BI__sync_lock_test_and_set_4: 1447 case Builtin::BI__sync_lock_test_and_set_8: 1448 case Builtin::BI__sync_lock_test_and_set_16: 1449 BuiltinIndex = 12; 1450 break; 1451 1452 case Builtin::BI__sync_lock_release: 1453 case Builtin::BI__sync_lock_release_1: 1454 case Builtin::BI__sync_lock_release_2: 1455 case Builtin::BI__sync_lock_release_4: 1456 case Builtin::BI__sync_lock_release_8: 1457 case Builtin::BI__sync_lock_release_16: 1458 BuiltinIndex = 13; 1459 NumFixed = 0; 1460 ResultType = Context.VoidTy; 1461 break; 1462 1463 case Builtin::BI__sync_swap: 1464 case Builtin::BI__sync_swap_1: 1465 case Builtin::BI__sync_swap_2: 1466 case Builtin::BI__sync_swap_4: 1467 case Builtin::BI__sync_swap_8: 1468 case Builtin::BI__sync_swap_16: 1469 BuiltinIndex = 14; 1470 break; 1471 } 1472 1473 // Now that we know how many fixed arguments we expect, first check that we 1474 // have at least that many. 1475 if (TheCall->getNumArgs() < 1+NumFixed) { 1476 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1477 << 0 << 1+NumFixed << TheCall->getNumArgs() 1478 << TheCall->getCallee()->getSourceRange(); 1479 return ExprError(); 1480 } 1481 1482 // Get the decl for the concrete builtin from this, we can tell what the 1483 // concrete integer type we should convert to is. 1484 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1485 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1486 FunctionDecl *NewBuiltinDecl; 1487 if (NewBuiltinID == BuiltinID) 1488 NewBuiltinDecl = FDecl; 1489 else { 1490 // Perform builtin lookup to avoid redeclaring it. 1491 DeclarationName DN(&Context.Idents.get(NewBuiltinName)); 1492 LookupResult Res(*this, DN, DRE->getLocStart(), LookupOrdinaryName); 1493 LookupName(Res, TUScope, /*AllowBuiltinCreation=*/true); 1494 assert(Res.getFoundDecl()); 1495 NewBuiltinDecl = dyn_cast<FunctionDecl>(Res.getFoundDecl()); 1496 if (NewBuiltinDecl == 0) 1497 return ExprError(); 1498 } 1499 1500 // The first argument --- the pointer --- has a fixed type; we 1501 // deduce the types of the rest of the arguments accordingly. Walk 1502 // the remaining arguments, converting them to the deduced value type. 1503 for (unsigned i = 0; i != NumFixed; ++i) { 1504 ExprResult Arg = TheCall->getArg(i+1); 1505 1506 // GCC does an implicit conversion to the pointer or integer ValType. This 1507 // can fail in some cases (1i -> int**), check for this error case now. 1508 // Initialize the argument. 1509 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1510 ValType, /*consume*/ false); 1511 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1512 if (Arg.isInvalid()) 1513 return ExprError(); 1514 1515 // Okay, we have something that *can* be converted to the right type. Check 1516 // to see if there is a potentially weird extension going on here. This can 1517 // happen when you do an atomic operation on something like an char* and 1518 // pass in 42. The 42 gets converted to char. This is even more strange 1519 // for things like 45.123 -> char, etc. 1520 // FIXME: Do this check. 1521 TheCall->setArg(i+1, Arg.take()); 1522 } 1523 1524 ASTContext& Context = this->getASTContext(); 1525 1526 // Create a new DeclRefExpr to refer to the new decl. 1527 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1528 Context, 1529 DRE->getQualifierLoc(), 1530 SourceLocation(), 1531 NewBuiltinDecl, 1532 /*enclosing*/ false, 1533 DRE->getLocation(), 1534 Context.BuiltinFnTy, 1535 DRE->getValueKind()); 1536 1537 // Set the callee in the CallExpr. 1538 // FIXME: This loses syntactic information. 1539 QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType()); 1540 ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy, 1541 CK_BuiltinFnToFnPtr); 1542 TheCall->setCallee(PromotedCall.take()); 1543 1544 // Change the result type of the call to match the original value type. This 1545 // is arbitrary, but the codegen for these builtins ins design to handle it 1546 // gracefully. 1547 TheCall->setType(ResultType); 1548 1549 return TheCallResult; 1550} 1551 1552/// CheckObjCString - Checks that the argument to the builtin 1553/// CFString constructor is correct 1554/// Note: It might also make sense to do the UTF-16 conversion here (would 1555/// simplify the backend). 1556bool Sema::CheckObjCString(Expr *Arg) { 1557 Arg = Arg->IgnoreParenCasts(); 1558 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1559 1560 if (!Literal || !Literal->isAscii()) { 1561 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1562 << Arg->getSourceRange(); 1563 return true; 1564 } 1565 1566 if (Literal->containsNonAsciiOrNull()) { 1567 StringRef String = Literal->getString(); 1568 unsigned NumBytes = String.size(); 1569 SmallVector<UTF16, 128> ToBuf(NumBytes); 1570 const UTF8 *FromPtr = (const UTF8 *)String.data(); 1571 UTF16 *ToPtr = &ToBuf[0]; 1572 1573 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1574 &ToPtr, ToPtr + NumBytes, 1575 strictConversion); 1576 // Check for conversion failure. 1577 if (Result != conversionOK) 1578 Diag(Arg->getLocStart(), 1579 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1580 } 1581 return false; 1582} 1583 1584/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1585/// Emit an error and return true on failure, return false on success. 1586bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1587 Expr *Fn = TheCall->getCallee(); 1588 if (TheCall->getNumArgs() > 2) { 1589 Diag(TheCall->getArg(2)->getLocStart(), 1590 diag::err_typecheck_call_too_many_args) 1591 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1592 << Fn->getSourceRange() 1593 << SourceRange(TheCall->getArg(2)->getLocStart(), 1594 (*(TheCall->arg_end()-1))->getLocEnd()); 1595 return true; 1596 } 1597 1598 if (TheCall->getNumArgs() < 2) { 1599 return Diag(TheCall->getLocEnd(), 1600 diag::err_typecheck_call_too_few_args_at_least) 1601 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1602 } 1603 1604 // Type-check the first argument normally. 1605 if (checkBuiltinArgument(*this, TheCall, 0)) 1606 return true; 1607 1608 // Determine whether the current function is variadic or not. 1609 BlockScopeInfo *CurBlock = getCurBlock(); 1610 bool isVariadic; 1611 if (CurBlock) 1612 isVariadic = CurBlock->TheDecl->isVariadic(); 1613 else if (FunctionDecl *FD = getCurFunctionDecl()) 1614 isVariadic = FD->isVariadic(); 1615 else 1616 isVariadic = getCurMethodDecl()->isVariadic(); 1617 1618 if (!isVariadic) { 1619 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1620 return true; 1621 } 1622 1623 // Verify that the second argument to the builtin is the last argument of the 1624 // current function or method. 1625 bool SecondArgIsLastNamedArgument = false; 1626 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1627 1628 // These are valid if SecondArgIsLastNamedArgument is false after the next 1629 // block. 1630 QualType Type; 1631 SourceLocation ParamLoc; 1632 1633 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1634 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1635 // FIXME: This isn't correct for methods (results in bogus warning). 1636 // Get the last formal in the current function. 1637 const ParmVarDecl *LastArg; 1638 if (CurBlock) 1639 LastArg = *(CurBlock->TheDecl->param_end()-1); 1640 else if (FunctionDecl *FD = getCurFunctionDecl()) 1641 LastArg = *(FD->param_end()-1); 1642 else 1643 LastArg = *(getCurMethodDecl()->param_end()-1); 1644 SecondArgIsLastNamedArgument = PV == LastArg; 1645 1646 Type = PV->getType(); 1647 ParamLoc = PV->getLocation(); 1648 } 1649 } 1650 1651 if (!SecondArgIsLastNamedArgument) 1652 Diag(TheCall->getArg(1)->getLocStart(), 1653 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1654 else if (Type->isReferenceType()) { 1655 Diag(Arg->getLocStart(), 1656 diag::warn_va_start_of_reference_type_is_undefined); 1657 Diag(ParamLoc, diag::note_parameter_type) << Type; 1658 } 1659 1660 TheCall->setType(Context.VoidTy); 1661 return false; 1662} 1663 1664/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1665/// friends. This is declared to take (...), so we have to check everything. 1666bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1667 if (TheCall->getNumArgs() < 2) 1668 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1669 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1670 if (TheCall->getNumArgs() > 2) 1671 return Diag(TheCall->getArg(2)->getLocStart(), 1672 diag::err_typecheck_call_too_many_args) 1673 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1674 << SourceRange(TheCall->getArg(2)->getLocStart(), 1675 (*(TheCall->arg_end()-1))->getLocEnd()); 1676 1677 ExprResult OrigArg0 = TheCall->getArg(0); 1678 ExprResult OrigArg1 = TheCall->getArg(1); 1679 1680 // Do standard promotions between the two arguments, returning their common 1681 // type. 1682 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1683 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1684 return true; 1685 1686 // Make sure any conversions are pushed back into the call; this is 1687 // type safe since unordered compare builtins are declared as "_Bool 1688 // foo(...)". 1689 TheCall->setArg(0, OrigArg0.get()); 1690 TheCall->setArg(1, OrigArg1.get()); 1691 1692 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1693 return false; 1694 1695 // If the common type isn't a real floating type, then the arguments were 1696 // invalid for this operation. 1697 if (Res.isNull() || !Res->isRealFloatingType()) 1698 return Diag(OrigArg0.get()->getLocStart(), 1699 diag::err_typecheck_call_invalid_ordered_compare) 1700 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1701 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1702 1703 return false; 1704} 1705 1706/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1707/// __builtin_isnan and friends. This is declared to take (...), so we have 1708/// to check everything. We expect the last argument to be a floating point 1709/// value. 1710bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1711 if (TheCall->getNumArgs() < NumArgs) 1712 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1713 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1714 if (TheCall->getNumArgs() > NumArgs) 1715 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1716 diag::err_typecheck_call_too_many_args) 1717 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1718 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1719 (*(TheCall->arg_end()-1))->getLocEnd()); 1720 1721 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1722 1723 if (OrigArg->isTypeDependent()) 1724 return false; 1725 1726 // This operation requires a non-_Complex floating-point number. 1727 if (!OrigArg->getType()->isRealFloatingType()) 1728 return Diag(OrigArg->getLocStart(), 1729 diag::err_typecheck_call_invalid_unary_fp) 1730 << OrigArg->getType() << OrigArg->getSourceRange(); 1731 1732 // If this is an implicit conversion from float -> double, remove it. 1733 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1734 Expr *CastArg = Cast->getSubExpr(); 1735 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1736 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1737 "promotion from float to double is the only expected cast here"); 1738 Cast->setSubExpr(0); 1739 TheCall->setArg(NumArgs-1, CastArg); 1740 } 1741 } 1742 1743 return false; 1744} 1745 1746/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1747// This is declared to take (...), so we have to check everything. 1748ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1749 if (TheCall->getNumArgs() < 2) 1750 return ExprError(Diag(TheCall->getLocEnd(), 1751 diag::err_typecheck_call_too_few_args_at_least) 1752 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1753 << TheCall->getSourceRange()); 1754 1755 // Determine which of the following types of shufflevector we're checking: 1756 // 1) unary, vector mask: (lhs, mask) 1757 // 2) binary, vector mask: (lhs, rhs, mask) 1758 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1759 QualType resType = TheCall->getArg(0)->getType(); 1760 unsigned numElements = 0; 1761 1762 if (!TheCall->getArg(0)->isTypeDependent() && 1763 !TheCall->getArg(1)->isTypeDependent()) { 1764 QualType LHSType = TheCall->getArg(0)->getType(); 1765 QualType RHSType = TheCall->getArg(1)->getType(); 1766 1767 if (!LHSType->isVectorType() || !RHSType->isVectorType()) 1768 return ExprError(Diag(TheCall->getLocStart(), 1769 diag::err_shufflevector_non_vector) 1770 << SourceRange(TheCall->getArg(0)->getLocStart(), 1771 TheCall->getArg(1)->getLocEnd())); 1772 1773 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1774 unsigned numResElements = TheCall->getNumArgs() - 2; 1775 1776 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1777 // with mask. If so, verify that RHS is an integer vector type with the 1778 // same number of elts as lhs. 1779 if (TheCall->getNumArgs() == 2) { 1780 if (!RHSType->hasIntegerRepresentation() || 1781 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1782 return ExprError(Diag(TheCall->getLocStart(), 1783 diag::err_shufflevector_incompatible_vector) 1784 << SourceRange(TheCall->getArg(1)->getLocStart(), 1785 TheCall->getArg(1)->getLocEnd())); 1786 } else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1787 return ExprError(Diag(TheCall->getLocStart(), 1788 diag::err_shufflevector_incompatible_vector) 1789 << SourceRange(TheCall->getArg(0)->getLocStart(), 1790 TheCall->getArg(1)->getLocEnd())); 1791 } else if (numElements != numResElements) { 1792 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1793 resType = Context.getVectorType(eltType, numResElements, 1794 VectorType::GenericVector); 1795 } 1796 } 1797 1798 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1799 if (TheCall->getArg(i)->isTypeDependent() || 1800 TheCall->getArg(i)->isValueDependent()) 1801 continue; 1802 1803 llvm::APSInt Result(32); 1804 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1805 return ExprError(Diag(TheCall->getLocStart(), 1806 diag::err_shufflevector_nonconstant_argument) 1807 << TheCall->getArg(i)->getSourceRange()); 1808 1809 // Allow -1 which will be translated to undef in the IR. 1810 if (Result.isSigned() && Result.isAllOnesValue()) 1811 continue; 1812 1813 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1814 return ExprError(Diag(TheCall->getLocStart(), 1815 diag::err_shufflevector_argument_too_large) 1816 << TheCall->getArg(i)->getSourceRange()); 1817 } 1818 1819 SmallVector<Expr*, 32> exprs; 1820 1821 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1822 exprs.push_back(TheCall->getArg(i)); 1823 TheCall->setArg(i, 0); 1824 } 1825 1826 return Owned(new (Context) ShuffleVectorExpr(Context, exprs, resType, 1827 TheCall->getCallee()->getLocStart(), 1828 TheCall->getRParenLoc())); 1829} 1830 1831/// SemaConvertVectorExpr - Handle __builtin_convertvector 1832ExprResult Sema::SemaConvertVectorExpr(Expr *E, TypeSourceInfo *TInfo, 1833 SourceLocation BuiltinLoc, 1834 SourceLocation RParenLoc) { 1835 ExprValueKind VK = VK_RValue; 1836 ExprObjectKind OK = OK_Ordinary; 1837 QualType DstTy = TInfo->getType(); 1838 QualType SrcTy = E->getType(); 1839 1840 if (!SrcTy->isVectorType() && !SrcTy->isDependentType()) 1841 return ExprError(Diag(BuiltinLoc, 1842 diag::err_convertvector_non_vector) 1843 << E->getSourceRange()); 1844 if (!DstTy->isVectorType() && !DstTy->isDependentType()) 1845 return ExprError(Diag(BuiltinLoc, 1846 diag::err_convertvector_non_vector_type)); 1847 1848 if (!SrcTy->isDependentType() && !DstTy->isDependentType()) { 1849 unsigned SrcElts = SrcTy->getAs<VectorType>()->getNumElements(); 1850 unsigned DstElts = DstTy->getAs<VectorType>()->getNumElements(); 1851 if (SrcElts != DstElts) 1852 return ExprError(Diag(BuiltinLoc, 1853 diag::err_convertvector_incompatible_vector) 1854 << E->getSourceRange()); 1855 } 1856 1857 return Owned(new (Context) ConvertVectorExpr(E, TInfo, DstTy, VK, OK, 1858 BuiltinLoc, RParenLoc)); 1859 1860} 1861 1862/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1863// This is declared to take (const void*, ...) and can take two 1864// optional constant int args. 1865bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1866 unsigned NumArgs = TheCall->getNumArgs(); 1867 1868 if (NumArgs > 3) 1869 return Diag(TheCall->getLocEnd(), 1870 diag::err_typecheck_call_too_many_args_at_most) 1871 << 0 /*function call*/ << 3 << NumArgs 1872 << TheCall->getSourceRange(); 1873 1874 // Argument 0 is checked for us and the remaining arguments must be 1875 // constant integers. 1876 for (unsigned i = 1; i != NumArgs; ++i) { 1877 Expr *Arg = TheCall->getArg(i); 1878 1879 // We can't check the value of a dependent argument. 1880 if (Arg->isTypeDependent() || Arg->isValueDependent()) 1881 continue; 1882 1883 llvm::APSInt Result; 1884 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1885 return true; 1886 1887 // FIXME: gcc issues a warning and rewrites these to 0. These 1888 // seems especially odd for the third argument since the default 1889 // is 3. 1890 if (i == 1) { 1891 if (Result.getLimitedValue() > 1) 1892 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1893 << "0" << "1" << Arg->getSourceRange(); 1894 } else { 1895 if (Result.getLimitedValue() > 3) 1896 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1897 << "0" << "3" << Arg->getSourceRange(); 1898 } 1899 } 1900 1901 return false; 1902} 1903 1904/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1905/// TheCall is a constant expression. 1906bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1907 llvm::APSInt &Result) { 1908 Expr *Arg = TheCall->getArg(ArgNum); 1909 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1910 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1911 1912 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1913 1914 if (!Arg->isIntegerConstantExpr(Result, Context)) 1915 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1916 << FDecl->getDeclName() << Arg->getSourceRange(); 1917 1918 return false; 1919} 1920 1921/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1922/// int type). This simply type checks that type is one of the defined 1923/// constants (0-3). 1924// For compatibility check 0-3, llvm only handles 0 and 2. 1925bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1926 llvm::APSInt Result; 1927 1928 // We can't check the value of a dependent argument. 1929 if (TheCall->getArg(1)->isTypeDependent() || 1930 TheCall->getArg(1)->isValueDependent()) 1931 return false; 1932 1933 // Check constant-ness first. 1934 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1935 return true; 1936 1937 Expr *Arg = TheCall->getArg(1); 1938 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1939 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1940 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1941 } 1942 1943 return false; 1944} 1945 1946/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1947/// This checks that val is a constant 1. 1948bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1949 Expr *Arg = TheCall->getArg(1); 1950 llvm::APSInt Result; 1951 1952 // TODO: This is less than ideal. Overload this to take a value. 1953 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1954 return true; 1955 1956 if (Result != 1) 1957 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1958 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1959 1960 return false; 1961} 1962 1963namespace { 1964enum StringLiteralCheckType { 1965 SLCT_NotALiteral, 1966 SLCT_UncheckedLiteral, 1967 SLCT_CheckedLiteral 1968}; 1969} 1970 1971// Determine if an expression is a string literal or constant string. 1972// If this function returns false on the arguments to a function expecting a 1973// format string, we will usually need to emit a warning. 1974// True string literals are then checked by CheckFormatString. 1975static StringLiteralCheckType 1976checkFormatStringExpr(Sema &S, const Expr *E, ArrayRef<const Expr *> Args, 1977 bool HasVAListArg, unsigned format_idx, 1978 unsigned firstDataArg, Sema::FormatStringType Type, 1979 Sema::VariadicCallType CallType, bool InFunctionCall, 1980 llvm::SmallBitVector &CheckedVarArgs) { 1981 tryAgain: 1982 if (E->isTypeDependent() || E->isValueDependent()) 1983 return SLCT_NotALiteral; 1984 1985 E = E->IgnoreParenCasts(); 1986 1987 if (E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull)) 1988 // Technically -Wformat-nonliteral does not warn about this case. 1989 // The behavior of printf and friends in this case is implementation 1990 // dependent. Ideally if the format string cannot be null then 1991 // it should have a 'nonnull' attribute in the function prototype. 1992 return SLCT_UncheckedLiteral; 1993 1994 switch (E->getStmtClass()) { 1995 case Stmt::BinaryConditionalOperatorClass: 1996 case Stmt::ConditionalOperatorClass: { 1997 // The expression is a literal if both sub-expressions were, and it was 1998 // completely checked only if both sub-expressions were checked. 1999 const AbstractConditionalOperator *C = 2000 cast<AbstractConditionalOperator>(E); 2001 StringLiteralCheckType Left = 2002 checkFormatStringExpr(S, C->getTrueExpr(), Args, 2003 HasVAListArg, format_idx, firstDataArg, 2004 Type, CallType, InFunctionCall, CheckedVarArgs); 2005 if (Left == SLCT_NotALiteral) 2006 return SLCT_NotALiteral; 2007 StringLiteralCheckType Right = 2008 checkFormatStringExpr(S, C->getFalseExpr(), Args, 2009 HasVAListArg, format_idx, firstDataArg, 2010 Type, CallType, InFunctionCall, CheckedVarArgs); 2011 return Left < Right ? Left : Right; 2012 } 2013 2014 case Stmt::ImplicitCastExprClass: { 2015 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 2016 goto tryAgain; 2017 } 2018 2019 case Stmt::OpaqueValueExprClass: 2020 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 2021 E = src; 2022 goto tryAgain; 2023 } 2024 return SLCT_NotALiteral; 2025 2026 case Stmt::PredefinedExprClass: 2027 // While __func__, etc., are technically not string literals, they 2028 // cannot contain format specifiers and thus are not a security 2029 // liability. 2030 return SLCT_UncheckedLiteral; 2031 2032 case Stmt::DeclRefExprClass: { 2033 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 2034 2035 // As an exception, do not flag errors for variables binding to 2036 // const string literals. 2037 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 2038 bool isConstant = false; 2039 QualType T = DR->getType(); 2040 2041 if (const ArrayType *AT = S.Context.getAsArrayType(T)) { 2042 isConstant = AT->getElementType().isConstant(S.Context); 2043 } else if (const PointerType *PT = T->getAs<PointerType>()) { 2044 isConstant = T.isConstant(S.Context) && 2045 PT->getPointeeType().isConstant(S.Context); 2046 } else if (T->isObjCObjectPointerType()) { 2047 // In ObjC, there is usually no "const ObjectPointer" type, 2048 // so don't check if the pointee type is constant. 2049 isConstant = T.isConstant(S.Context); 2050 } 2051 2052 if (isConstant) { 2053 if (const Expr *Init = VD->getAnyInitializer()) { 2054 // Look through initializers like const char c[] = { "foo" } 2055 if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) { 2056 if (InitList->isStringLiteralInit()) 2057 Init = InitList->getInit(0)->IgnoreParenImpCasts(); 2058 } 2059 return checkFormatStringExpr(S, Init, Args, 2060 HasVAListArg, format_idx, 2061 firstDataArg, Type, CallType, 2062 /*InFunctionCall*/false, CheckedVarArgs); 2063 } 2064 } 2065 2066 // For vprintf* functions (i.e., HasVAListArg==true), we add a 2067 // special check to see if the format string is a function parameter 2068 // of the function calling the printf function. If the function 2069 // has an attribute indicating it is a printf-like function, then we 2070 // should suppress warnings concerning non-literals being used in a call 2071 // to a vprintf function. For example: 2072 // 2073 // void 2074 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 2075 // va_list ap; 2076 // va_start(ap, fmt); 2077 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 2078 // ... 2079 // } 2080 if (HasVAListArg) { 2081 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 2082 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 2083 int PVIndex = PV->getFunctionScopeIndex() + 1; 2084 for (specific_attr_iterator<FormatAttr> 2085 i = ND->specific_attr_begin<FormatAttr>(), 2086 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 2087 FormatAttr *PVFormat = *i; 2088 // adjust for implicit parameter 2089 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 2090 if (MD->isInstance()) 2091 ++PVIndex; 2092 // We also check if the formats are compatible. 2093 // We can't pass a 'scanf' string to a 'printf' function. 2094 if (PVIndex == PVFormat->getFormatIdx() && 2095 Type == S.GetFormatStringType(PVFormat)) 2096 return SLCT_UncheckedLiteral; 2097 } 2098 } 2099 } 2100 } 2101 } 2102 2103 return SLCT_NotALiteral; 2104 } 2105 2106 case Stmt::CallExprClass: 2107 case Stmt::CXXMemberCallExprClass: { 2108 const CallExpr *CE = cast<CallExpr>(E); 2109 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 2110 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 2111 unsigned ArgIndex = FA->getFormatIdx(); 2112 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 2113 if (MD->isInstance()) 2114 --ArgIndex; 2115 const Expr *Arg = CE->getArg(ArgIndex - 1); 2116 2117 return checkFormatStringExpr(S, Arg, Args, 2118 HasVAListArg, format_idx, firstDataArg, 2119 Type, CallType, InFunctionCall, 2120 CheckedVarArgs); 2121 } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) { 2122 unsigned BuiltinID = FD->getBuiltinID(); 2123 if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString || 2124 BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) { 2125 const Expr *Arg = CE->getArg(0); 2126 return checkFormatStringExpr(S, Arg, Args, 2127 HasVAListArg, format_idx, 2128 firstDataArg, Type, CallType, 2129 InFunctionCall, CheckedVarArgs); 2130 } 2131 } 2132 } 2133 2134 return SLCT_NotALiteral; 2135 } 2136 2137 case Stmt::ObjCMessageExprClass: { 2138 const ObjCMessageExpr *ME = cast<ObjCMessageExpr>(E); 2139 if (const ObjCMethodDecl *MDecl = ME->getMethodDecl()) { 2140 if (const NamedDecl *ND = dyn_cast<NamedDecl>(MDecl)) { 2141 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 2142 unsigned ArgIndex = FA->getFormatIdx(); 2143 if (ArgIndex <= ME->getNumArgs()) { 2144 const Expr *Arg = ME->getArg(ArgIndex-1); 2145 return checkFormatStringExpr(S, Arg, Args, 2146 HasVAListArg, format_idx, 2147 firstDataArg, Type, CallType, 2148 InFunctionCall, CheckedVarArgs); 2149 } 2150 } 2151 } 2152 } 2153 2154 return SLCT_NotALiteral; 2155 } 2156 2157 case Stmt::ObjCStringLiteralClass: 2158 case Stmt::StringLiteralClass: { 2159 const StringLiteral *StrE = NULL; 2160 2161 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 2162 StrE = ObjCFExpr->getString(); 2163 else 2164 StrE = cast<StringLiteral>(E); 2165 2166 if (StrE) { 2167 S.CheckFormatString(StrE, E, Args, HasVAListArg, format_idx, firstDataArg, 2168 Type, InFunctionCall, CallType, CheckedVarArgs); 2169 return SLCT_CheckedLiteral; 2170 } 2171 2172 return SLCT_NotALiteral; 2173 } 2174 2175 default: 2176 return SLCT_NotALiteral; 2177 } 2178} 2179 2180void 2181Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 2182 const Expr * const *ExprArgs, 2183 SourceLocation CallSiteLoc) { 2184 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 2185 e = NonNull->args_end(); 2186 i != e; ++i) { 2187 const Expr *ArgExpr = ExprArgs[*i]; 2188 2189 // As a special case, transparent unions initialized with zero are 2190 // considered null for the purposes of the nonnull attribute. 2191 if (const RecordType *UT = ArgExpr->getType()->getAsUnionType()) { 2192 if (UT->getDecl()->hasAttr<TransparentUnionAttr>()) 2193 if (const CompoundLiteralExpr *CLE = 2194 dyn_cast<CompoundLiteralExpr>(ArgExpr)) 2195 if (const InitListExpr *ILE = 2196 dyn_cast<InitListExpr>(CLE->getInitializer())) 2197 ArgExpr = ILE->getInit(0); 2198 } 2199 2200 bool Result; 2201 if (ArgExpr->EvaluateAsBooleanCondition(Result, Context) && !Result) 2202 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 2203 } 2204} 2205 2206Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 2207 return llvm::StringSwitch<FormatStringType>(Format->getType()->getName()) 2208 .Case("scanf", FST_Scanf) 2209 .Cases("printf", "printf0", FST_Printf) 2210 .Cases("NSString", "CFString", FST_NSString) 2211 .Case("strftime", FST_Strftime) 2212 .Case("strfmon", FST_Strfmon) 2213 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 2214 .Default(FST_Unknown); 2215} 2216 2217/// CheckFormatArguments - Check calls to printf and scanf (and similar 2218/// functions) for correct use of format strings. 2219/// Returns true if a format string has been fully checked. 2220bool Sema::CheckFormatArguments(const FormatAttr *Format, 2221 ArrayRef<const Expr *> Args, 2222 bool IsCXXMember, 2223 VariadicCallType CallType, 2224 SourceLocation Loc, SourceRange Range, 2225 llvm::SmallBitVector &CheckedVarArgs) { 2226 FormatStringInfo FSI; 2227 if (getFormatStringInfo(Format, IsCXXMember, &FSI)) 2228 return CheckFormatArguments(Args, FSI.HasVAListArg, FSI.FormatIdx, 2229 FSI.FirstDataArg, GetFormatStringType(Format), 2230 CallType, Loc, Range, CheckedVarArgs); 2231 return false; 2232} 2233 2234bool Sema::CheckFormatArguments(ArrayRef<const Expr *> Args, 2235 bool HasVAListArg, unsigned format_idx, 2236 unsigned firstDataArg, FormatStringType Type, 2237 VariadicCallType CallType, 2238 SourceLocation Loc, SourceRange Range, 2239 llvm::SmallBitVector &CheckedVarArgs) { 2240 // CHECK: printf/scanf-like function is called with no format string. 2241 if (format_idx >= Args.size()) { 2242 Diag(Loc, diag::warn_missing_format_string) << Range; 2243 return false; 2244 } 2245 2246 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 2247 2248 // CHECK: format string is not a string literal. 2249 // 2250 // Dynamically generated format strings are difficult to 2251 // automatically vet at compile time. Requiring that format strings 2252 // are string literals: (1) permits the checking of format strings by 2253 // the compiler and thereby (2) can practically remove the source of 2254 // many format string exploits. 2255 2256 // Format string can be either ObjC string (e.g. @"%d") or 2257 // C string (e.g. "%d") 2258 // ObjC string uses the same format specifiers as C string, so we can use 2259 // the same format string checking logic for both ObjC and C strings. 2260 StringLiteralCheckType CT = 2261 checkFormatStringExpr(*this, OrigFormatExpr, Args, HasVAListArg, 2262 format_idx, firstDataArg, Type, CallType, 2263 /*IsFunctionCall*/true, CheckedVarArgs); 2264 if (CT != SLCT_NotALiteral) 2265 // Literal format string found, check done! 2266 return CT == SLCT_CheckedLiteral; 2267 2268 // Strftime is particular as it always uses a single 'time' argument, 2269 // so it is safe to pass a non-literal string. 2270 if (Type == FST_Strftime) 2271 return false; 2272 2273 // Do not emit diag when the string param is a macro expansion and the 2274 // format is either NSString or CFString. This is a hack to prevent 2275 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 2276 // which are usually used in place of NS and CF string literals. 2277 if (Type == FST_NSString && 2278 SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart())) 2279 return false; 2280 2281 // If there are no arguments specified, warn with -Wformat-security, otherwise 2282 // warn only with -Wformat-nonliteral. 2283 if (Args.size() == firstDataArg) 2284 Diag(Args[format_idx]->getLocStart(), 2285 diag::warn_format_nonliteral_noargs) 2286 << OrigFormatExpr->getSourceRange(); 2287 else 2288 Diag(Args[format_idx]->getLocStart(), 2289 diag::warn_format_nonliteral) 2290 << OrigFormatExpr->getSourceRange(); 2291 return false; 2292} 2293 2294namespace { 2295class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 2296protected: 2297 Sema &S; 2298 const StringLiteral *FExpr; 2299 const Expr *OrigFormatExpr; 2300 const unsigned FirstDataArg; 2301 const unsigned NumDataArgs; 2302 const char *Beg; // Start of format string. 2303 const bool HasVAListArg; 2304 ArrayRef<const Expr *> Args; 2305 unsigned FormatIdx; 2306 llvm::SmallBitVector CoveredArgs; 2307 bool usesPositionalArgs; 2308 bool atFirstArg; 2309 bool inFunctionCall; 2310 Sema::VariadicCallType CallType; 2311 llvm::SmallBitVector &CheckedVarArgs; 2312public: 2313 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 2314 const Expr *origFormatExpr, unsigned firstDataArg, 2315 unsigned numDataArgs, const char *beg, bool hasVAListArg, 2316 ArrayRef<const Expr *> Args, 2317 unsigned formatIdx, bool inFunctionCall, 2318 Sema::VariadicCallType callType, 2319 llvm::SmallBitVector &CheckedVarArgs) 2320 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 2321 FirstDataArg(firstDataArg), NumDataArgs(numDataArgs), 2322 Beg(beg), HasVAListArg(hasVAListArg), 2323 Args(Args), FormatIdx(formatIdx), 2324 usesPositionalArgs(false), atFirstArg(true), 2325 inFunctionCall(inFunctionCall), CallType(callType), 2326 CheckedVarArgs(CheckedVarArgs) { 2327 CoveredArgs.resize(numDataArgs); 2328 CoveredArgs.reset(); 2329 } 2330 2331 void DoneProcessing(); 2332 2333 void HandleIncompleteSpecifier(const char *startSpecifier, 2334 unsigned specifierLen); 2335 2336 void HandleInvalidLengthModifier( 2337 const analyze_format_string::FormatSpecifier &FS, 2338 const analyze_format_string::ConversionSpecifier &CS, 2339 const char *startSpecifier, unsigned specifierLen, unsigned DiagID); 2340 2341 void HandleNonStandardLengthModifier( 2342 const analyze_format_string::FormatSpecifier &FS, 2343 const char *startSpecifier, unsigned specifierLen); 2344 2345 void HandleNonStandardConversionSpecifier( 2346 const analyze_format_string::ConversionSpecifier &CS, 2347 const char *startSpecifier, unsigned specifierLen); 2348 2349 virtual void HandlePosition(const char *startPos, unsigned posLen); 2350 2351 virtual void HandleInvalidPosition(const char *startSpecifier, 2352 unsigned specifierLen, 2353 analyze_format_string::PositionContext p); 2354 2355 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 2356 2357 void HandleNullChar(const char *nullCharacter); 2358 2359 template <typename Range> 2360 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 2361 const Expr *ArgumentExpr, 2362 PartialDiagnostic PDiag, 2363 SourceLocation StringLoc, 2364 bool IsStringLocation, Range StringRange, 2365 ArrayRef<FixItHint> Fixit = None); 2366 2367protected: 2368 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 2369 const char *startSpec, 2370 unsigned specifierLen, 2371 const char *csStart, unsigned csLen); 2372 2373 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 2374 const char *startSpec, 2375 unsigned specifierLen); 2376 2377 SourceRange getFormatStringRange(); 2378 CharSourceRange getSpecifierRange(const char *startSpecifier, 2379 unsigned specifierLen); 2380 SourceLocation getLocationOfByte(const char *x); 2381 2382 const Expr *getDataArg(unsigned i) const; 2383 2384 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 2385 const analyze_format_string::ConversionSpecifier &CS, 2386 const char *startSpecifier, unsigned specifierLen, 2387 unsigned argIndex); 2388 2389 template <typename Range> 2390 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 2391 bool IsStringLocation, Range StringRange, 2392 ArrayRef<FixItHint> Fixit = None); 2393 2394 void CheckPositionalAndNonpositionalArgs( 2395 const analyze_format_string::FormatSpecifier *FS); 2396}; 2397} 2398 2399SourceRange CheckFormatHandler::getFormatStringRange() { 2400 return OrigFormatExpr->getSourceRange(); 2401} 2402 2403CharSourceRange CheckFormatHandler:: 2404getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 2405 SourceLocation Start = getLocationOfByte(startSpecifier); 2406 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 2407 2408 // Advance the end SourceLocation by one due to half-open ranges. 2409 End = End.getLocWithOffset(1); 2410 2411 return CharSourceRange::getCharRange(Start, End); 2412} 2413 2414SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 2415 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 2416} 2417 2418void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 2419 unsigned specifierLen){ 2420 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 2421 getLocationOfByte(startSpecifier), 2422 /*IsStringLocation*/true, 2423 getSpecifierRange(startSpecifier, specifierLen)); 2424} 2425 2426void CheckFormatHandler::HandleInvalidLengthModifier( 2427 const analyze_format_string::FormatSpecifier &FS, 2428 const analyze_format_string::ConversionSpecifier &CS, 2429 const char *startSpecifier, unsigned specifierLen, unsigned DiagID) { 2430 using namespace analyze_format_string; 2431 2432 const LengthModifier &LM = FS.getLengthModifier(); 2433 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2434 2435 // See if we know how to fix this length modifier. 2436 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2437 if (FixedLM) { 2438 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2439 getLocationOfByte(LM.getStart()), 2440 /*IsStringLocation*/true, 2441 getSpecifierRange(startSpecifier, specifierLen)); 2442 2443 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2444 << FixedLM->toString() 2445 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2446 2447 } else { 2448 FixItHint Hint; 2449 if (DiagID == diag::warn_format_nonsensical_length) 2450 Hint = FixItHint::CreateRemoval(LMRange); 2451 2452 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2453 getLocationOfByte(LM.getStart()), 2454 /*IsStringLocation*/true, 2455 getSpecifierRange(startSpecifier, specifierLen), 2456 Hint); 2457 } 2458} 2459 2460void CheckFormatHandler::HandleNonStandardLengthModifier( 2461 const analyze_format_string::FormatSpecifier &FS, 2462 const char *startSpecifier, unsigned specifierLen) { 2463 using namespace analyze_format_string; 2464 2465 const LengthModifier &LM = FS.getLengthModifier(); 2466 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2467 2468 // See if we know how to fix this length modifier. 2469 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2470 if (FixedLM) { 2471 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2472 << LM.toString() << 0, 2473 getLocationOfByte(LM.getStart()), 2474 /*IsStringLocation*/true, 2475 getSpecifierRange(startSpecifier, specifierLen)); 2476 2477 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2478 << FixedLM->toString() 2479 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2480 2481 } else { 2482 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2483 << LM.toString() << 0, 2484 getLocationOfByte(LM.getStart()), 2485 /*IsStringLocation*/true, 2486 getSpecifierRange(startSpecifier, specifierLen)); 2487 } 2488} 2489 2490void CheckFormatHandler::HandleNonStandardConversionSpecifier( 2491 const analyze_format_string::ConversionSpecifier &CS, 2492 const char *startSpecifier, unsigned specifierLen) { 2493 using namespace analyze_format_string; 2494 2495 // See if we know how to fix this conversion specifier. 2496 Optional<ConversionSpecifier> FixedCS = CS.getStandardSpecifier(); 2497 if (FixedCS) { 2498 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2499 << CS.toString() << /*conversion specifier*/1, 2500 getLocationOfByte(CS.getStart()), 2501 /*IsStringLocation*/true, 2502 getSpecifierRange(startSpecifier, specifierLen)); 2503 2504 CharSourceRange CSRange = getSpecifierRange(CS.getStart(), CS.getLength()); 2505 S.Diag(getLocationOfByte(CS.getStart()), diag::note_format_fix_specifier) 2506 << FixedCS->toString() 2507 << FixItHint::CreateReplacement(CSRange, FixedCS->toString()); 2508 } else { 2509 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2510 << CS.toString() << /*conversion specifier*/1, 2511 getLocationOfByte(CS.getStart()), 2512 /*IsStringLocation*/true, 2513 getSpecifierRange(startSpecifier, specifierLen)); 2514 } 2515} 2516 2517void CheckFormatHandler::HandlePosition(const char *startPos, 2518 unsigned posLen) { 2519 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 2520 getLocationOfByte(startPos), 2521 /*IsStringLocation*/true, 2522 getSpecifierRange(startPos, posLen)); 2523} 2524 2525void 2526CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 2527 analyze_format_string::PositionContext p) { 2528 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 2529 << (unsigned) p, 2530 getLocationOfByte(startPos), /*IsStringLocation*/true, 2531 getSpecifierRange(startPos, posLen)); 2532} 2533 2534void CheckFormatHandler::HandleZeroPosition(const char *startPos, 2535 unsigned posLen) { 2536 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 2537 getLocationOfByte(startPos), 2538 /*IsStringLocation*/true, 2539 getSpecifierRange(startPos, posLen)); 2540} 2541 2542void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 2543 if (!isa<ObjCStringLiteral>(OrigFormatExpr)) { 2544 // The presence of a null character is likely an error. 2545 EmitFormatDiagnostic( 2546 S.PDiag(diag::warn_printf_format_string_contains_null_char), 2547 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 2548 getFormatStringRange()); 2549 } 2550} 2551 2552// Note that this may return NULL if there was an error parsing or building 2553// one of the argument expressions. 2554const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 2555 return Args[FirstDataArg + i]; 2556} 2557 2558void CheckFormatHandler::DoneProcessing() { 2559 // Does the number of data arguments exceed the number of 2560 // format conversions in the format string? 2561 if (!HasVAListArg) { 2562 // Find any arguments that weren't covered. 2563 CoveredArgs.flip(); 2564 signed notCoveredArg = CoveredArgs.find_first(); 2565 if (notCoveredArg >= 0) { 2566 assert((unsigned)notCoveredArg < NumDataArgs); 2567 if (const Expr *E = getDataArg((unsigned) notCoveredArg)) { 2568 SourceLocation Loc = E->getLocStart(); 2569 if (!S.getSourceManager().isInSystemMacro(Loc)) { 2570 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 2571 Loc, /*IsStringLocation*/false, 2572 getFormatStringRange()); 2573 } 2574 } 2575 } 2576 } 2577} 2578 2579bool 2580CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 2581 SourceLocation Loc, 2582 const char *startSpec, 2583 unsigned specifierLen, 2584 const char *csStart, 2585 unsigned csLen) { 2586 2587 bool keepGoing = true; 2588 if (argIndex < NumDataArgs) { 2589 // Consider the argument coverered, even though the specifier doesn't 2590 // make sense. 2591 CoveredArgs.set(argIndex); 2592 } 2593 else { 2594 // If argIndex exceeds the number of data arguments we 2595 // don't issue a warning because that is just a cascade of warnings (and 2596 // they may have intended '%%' anyway). We don't want to continue processing 2597 // the format string after this point, however, as we will like just get 2598 // gibberish when trying to match arguments. 2599 keepGoing = false; 2600 } 2601 2602 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2603 << StringRef(csStart, csLen), 2604 Loc, /*IsStringLocation*/true, 2605 getSpecifierRange(startSpec, specifierLen)); 2606 2607 return keepGoing; 2608} 2609 2610void 2611CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2612 const char *startSpec, 2613 unsigned specifierLen) { 2614 EmitFormatDiagnostic( 2615 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2616 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2617} 2618 2619bool 2620CheckFormatHandler::CheckNumArgs( 2621 const analyze_format_string::FormatSpecifier &FS, 2622 const analyze_format_string::ConversionSpecifier &CS, 2623 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2624 2625 if (argIndex >= NumDataArgs) { 2626 PartialDiagnostic PDiag = FS.usesPositionalArg() 2627 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2628 << (argIndex+1) << NumDataArgs) 2629 : S.PDiag(diag::warn_printf_insufficient_data_args); 2630 EmitFormatDiagnostic( 2631 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2632 getSpecifierRange(startSpecifier, specifierLen)); 2633 return false; 2634 } 2635 return true; 2636} 2637 2638template<typename Range> 2639void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2640 SourceLocation Loc, 2641 bool IsStringLocation, 2642 Range StringRange, 2643 ArrayRef<FixItHint> FixIt) { 2644 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2645 Loc, IsStringLocation, StringRange, FixIt); 2646} 2647 2648/// \brief If the format string is not within the funcion call, emit a note 2649/// so that the function call and string are in diagnostic messages. 2650/// 2651/// \param InFunctionCall if true, the format string is within the function 2652/// call and only one diagnostic message will be produced. Otherwise, an 2653/// extra note will be emitted pointing to location of the format string. 2654/// 2655/// \param ArgumentExpr the expression that is passed as the format string 2656/// argument in the function call. Used for getting locations when two 2657/// diagnostics are emitted. 2658/// 2659/// \param PDiag the callee should already have provided any strings for the 2660/// diagnostic message. This function only adds locations and fixits 2661/// to diagnostics. 2662/// 2663/// \param Loc primary location for diagnostic. If two diagnostics are 2664/// required, one will be at Loc and a new SourceLocation will be created for 2665/// the other one. 2666/// 2667/// \param IsStringLocation if true, Loc points to the format string should be 2668/// used for the note. Otherwise, Loc points to the argument list and will 2669/// be used with PDiag. 2670/// 2671/// \param StringRange some or all of the string to highlight. This is 2672/// templated so it can accept either a CharSourceRange or a SourceRange. 2673/// 2674/// \param FixIt optional fix it hint for the format string. 2675template<typename Range> 2676void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2677 const Expr *ArgumentExpr, 2678 PartialDiagnostic PDiag, 2679 SourceLocation Loc, 2680 bool IsStringLocation, 2681 Range StringRange, 2682 ArrayRef<FixItHint> FixIt) { 2683 if (InFunctionCall) { 2684 const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag); 2685 D << StringRange; 2686 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2687 I != E; ++I) { 2688 D << *I; 2689 } 2690 } else { 2691 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2692 << ArgumentExpr->getSourceRange(); 2693 2694 const Sema::SemaDiagnosticBuilder &Note = 2695 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2696 diag::note_format_string_defined); 2697 2698 Note << StringRange; 2699 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2700 I != E; ++I) { 2701 Note << *I; 2702 } 2703 } 2704} 2705 2706//===--- CHECK: Printf format string checking ------------------------------===// 2707 2708namespace { 2709class CheckPrintfHandler : public CheckFormatHandler { 2710 bool ObjCContext; 2711public: 2712 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2713 const Expr *origFormatExpr, unsigned firstDataArg, 2714 unsigned numDataArgs, bool isObjC, 2715 const char *beg, bool hasVAListArg, 2716 ArrayRef<const Expr *> Args, 2717 unsigned formatIdx, bool inFunctionCall, 2718 Sema::VariadicCallType CallType, 2719 llvm::SmallBitVector &CheckedVarArgs) 2720 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2721 numDataArgs, beg, hasVAListArg, Args, 2722 formatIdx, inFunctionCall, CallType, CheckedVarArgs), 2723 ObjCContext(isObjC) 2724 {} 2725 2726 2727 bool HandleInvalidPrintfConversionSpecifier( 2728 const analyze_printf::PrintfSpecifier &FS, 2729 const char *startSpecifier, 2730 unsigned specifierLen); 2731 2732 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2733 const char *startSpecifier, 2734 unsigned specifierLen); 2735 bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2736 const char *StartSpecifier, 2737 unsigned SpecifierLen, 2738 const Expr *E); 2739 2740 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2741 const char *startSpecifier, unsigned specifierLen); 2742 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2743 const analyze_printf::OptionalAmount &Amt, 2744 unsigned type, 2745 const char *startSpecifier, unsigned specifierLen); 2746 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2747 const analyze_printf::OptionalFlag &flag, 2748 const char *startSpecifier, unsigned specifierLen); 2749 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2750 const analyze_printf::OptionalFlag &ignoredFlag, 2751 const analyze_printf::OptionalFlag &flag, 2752 const char *startSpecifier, unsigned specifierLen); 2753 bool checkForCStrMembers(const analyze_printf::ArgType &AT, 2754 const Expr *E, const CharSourceRange &CSR); 2755 2756}; 2757} 2758 2759bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2760 const analyze_printf::PrintfSpecifier &FS, 2761 const char *startSpecifier, 2762 unsigned specifierLen) { 2763 const analyze_printf::PrintfConversionSpecifier &CS = 2764 FS.getConversionSpecifier(); 2765 2766 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2767 getLocationOfByte(CS.getStart()), 2768 startSpecifier, specifierLen, 2769 CS.getStart(), CS.getLength()); 2770} 2771 2772bool CheckPrintfHandler::HandleAmount( 2773 const analyze_format_string::OptionalAmount &Amt, 2774 unsigned k, const char *startSpecifier, 2775 unsigned specifierLen) { 2776 2777 if (Amt.hasDataArgument()) { 2778 if (!HasVAListArg) { 2779 unsigned argIndex = Amt.getArgIndex(); 2780 if (argIndex >= NumDataArgs) { 2781 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2782 << k, 2783 getLocationOfByte(Amt.getStart()), 2784 /*IsStringLocation*/true, 2785 getSpecifierRange(startSpecifier, specifierLen)); 2786 // Don't do any more checking. We will just emit 2787 // spurious errors. 2788 return false; 2789 } 2790 2791 // Type check the data argument. It should be an 'int'. 2792 // Although not in conformance with C99, we also allow the argument to be 2793 // an 'unsigned int' as that is a reasonably safe case. GCC also 2794 // doesn't emit a warning for that case. 2795 CoveredArgs.set(argIndex); 2796 const Expr *Arg = getDataArg(argIndex); 2797 if (!Arg) 2798 return false; 2799 2800 QualType T = Arg->getType(); 2801 2802 const analyze_printf::ArgType &AT = Amt.getArgType(S.Context); 2803 assert(AT.isValid()); 2804 2805 if (!AT.matchesType(S.Context, T)) { 2806 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2807 << k << AT.getRepresentativeTypeName(S.Context) 2808 << T << Arg->getSourceRange(), 2809 getLocationOfByte(Amt.getStart()), 2810 /*IsStringLocation*/true, 2811 getSpecifierRange(startSpecifier, specifierLen)); 2812 // Don't do any more checking. We will just emit 2813 // spurious errors. 2814 return false; 2815 } 2816 } 2817 } 2818 return true; 2819} 2820 2821void CheckPrintfHandler::HandleInvalidAmount( 2822 const analyze_printf::PrintfSpecifier &FS, 2823 const analyze_printf::OptionalAmount &Amt, 2824 unsigned type, 2825 const char *startSpecifier, 2826 unsigned specifierLen) { 2827 const analyze_printf::PrintfConversionSpecifier &CS = 2828 FS.getConversionSpecifier(); 2829 2830 FixItHint fixit = 2831 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2832 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2833 Amt.getConstantLength())) 2834 : FixItHint(); 2835 2836 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2837 << type << CS.toString(), 2838 getLocationOfByte(Amt.getStart()), 2839 /*IsStringLocation*/true, 2840 getSpecifierRange(startSpecifier, specifierLen), 2841 fixit); 2842} 2843 2844void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2845 const analyze_printf::OptionalFlag &flag, 2846 const char *startSpecifier, 2847 unsigned specifierLen) { 2848 // Warn about pointless flag with a fixit removal. 2849 const analyze_printf::PrintfConversionSpecifier &CS = 2850 FS.getConversionSpecifier(); 2851 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2852 << flag.toString() << CS.toString(), 2853 getLocationOfByte(flag.getPosition()), 2854 /*IsStringLocation*/true, 2855 getSpecifierRange(startSpecifier, specifierLen), 2856 FixItHint::CreateRemoval( 2857 getSpecifierRange(flag.getPosition(), 1))); 2858} 2859 2860void CheckPrintfHandler::HandleIgnoredFlag( 2861 const analyze_printf::PrintfSpecifier &FS, 2862 const analyze_printf::OptionalFlag &ignoredFlag, 2863 const analyze_printf::OptionalFlag &flag, 2864 const char *startSpecifier, 2865 unsigned specifierLen) { 2866 // Warn about ignored flag with a fixit removal. 2867 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2868 << ignoredFlag.toString() << flag.toString(), 2869 getLocationOfByte(ignoredFlag.getPosition()), 2870 /*IsStringLocation*/true, 2871 getSpecifierRange(startSpecifier, specifierLen), 2872 FixItHint::CreateRemoval( 2873 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2874} 2875 2876// Determines if the specified is a C++ class or struct containing 2877// a member with the specified name and kind (e.g. a CXXMethodDecl named 2878// "c_str()"). 2879template<typename MemberKind> 2880static llvm::SmallPtrSet<MemberKind*, 1> 2881CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) { 2882 const RecordType *RT = Ty->getAs<RecordType>(); 2883 llvm::SmallPtrSet<MemberKind*, 1> Results; 2884 2885 if (!RT) 2886 return Results; 2887 const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl()); 2888 if (!RD) 2889 return Results; 2890 2891 LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(), 2892 Sema::LookupMemberName); 2893 2894 // We just need to include all members of the right kind turned up by the 2895 // filter, at this point. 2896 if (S.LookupQualifiedName(R, RT->getDecl())) 2897 for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) { 2898 NamedDecl *decl = (*I)->getUnderlyingDecl(); 2899 if (MemberKind *FK = dyn_cast<MemberKind>(decl)) 2900 Results.insert(FK); 2901 } 2902 return Results; 2903} 2904 2905// Check if a (w)string was passed when a (w)char* was needed, and offer a 2906// better diagnostic if so. AT is assumed to be valid. 2907// Returns true when a c_str() conversion method is found. 2908bool CheckPrintfHandler::checkForCStrMembers( 2909 const analyze_printf::ArgType &AT, const Expr *E, 2910 const CharSourceRange &CSR) { 2911 typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet; 2912 2913 MethodSet Results = 2914 CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType()); 2915 2916 for (MethodSet::iterator MI = Results.begin(), ME = Results.end(); 2917 MI != ME; ++MI) { 2918 const CXXMethodDecl *Method = *MI; 2919 if (Method->getNumParams() == 0 && 2920 AT.matchesType(S.Context, Method->getResultType())) { 2921 // FIXME: Suggest parens if the expression needs them. 2922 SourceLocation EndLoc = 2923 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()); 2924 S.Diag(E->getLocStart(), diag::note_printf_c_str) 2925 << "c_str()" 2926 << FixItHint::CreateInsertion(EndLoc, ".c_str()"); 2927 return true; 2928 } 2929 } 2930 2931 return false; 2932} 2933 2934bool 2935CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2936 &FS, 2937 const char *startSpecifier, 2938 unsigned specifierLen) { 2939 2940 using namespace analyze_format_string; 2941 using namespace analyze_printf; 2942 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2943 2944 if (FS.consumesDataArgument()) { 2945 if (atFirstArg) { 2946 atFirstArg = false; 2947 usesPositionalArgs = FS.usesPositionalArg(); 2948 } 2949 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2950 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2951 startSpecifier, specifierLen); 2952 return false; 2953 } 2954 } 2955 2956 // First check if the field width, precision, and conversion specifier 2957 // have matching data arguments. 2958 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2959 startSpecifier, specifierLen)) { 2960 return false; 2961 } 2962 2963 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2964 startSpecifier, specifierLen)) { 2965 return false; 2966 } 2967 2968 if (!CS.consumesDataArgument()) { 2969 // FIXME: Technically specifying a precision or field width here 2970 // makes no sense. Worth issuing a warning at some point. 2971 return true; 2972 } 2973 2974 // Consume the argument. 2975 unsigned argIndex = FS.getArgIndex(); 2976 if (argIndex < NumDataArgs) { 2977 // The check to see if the argIndex is valid will come later. 2978 // We set the bit here because we may exit early from this 2979 // function if we encounter some other error. 2980 CoveredArgs.set(argIndex); 2981 } 2982 2983 // Check for using an Objective-C specific conversion specifier 2984 // in a non-ObjC literal. 2985 if (!ObjCContext && CS.isObjCArg()) { 2986 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2987 specifierLen); 2988 } 2989 2990 // Check for invalid use of field width 2991 if (!FS.hasValidFieldWidth()) { 2992 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2993 startSpecifier, specifierLen); 2994 } 2995 2996 // Check for invalid use of precision 2997 if (!FS.hasValidPrecision()) { 2998 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2999 startSpecifier, specifierLen); 3000 } 3001 3002 // Check each flag does not conflict with any other component. 3003 if (!FS.hasValidThousandsGroupingPrefix()) 3004 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 3005 if (!FS.hasValidLeadingZeros()) 3006 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 3007 if (!FS.hasValidPlusPrefix()) 3008 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 3009 if (!FS.hasValidSpacePrefix()) 3010 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 3011 if (!FS.hasValidAlternativeForm()) 3012 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 3013 if (!FS.hasValidLeftJustified()) 3014 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 3015 3016 // Check that flags are not ignored by another flag 3017 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 3018 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 3019 startSpecifier, specifierLen); 3020 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 3021 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 3022 startSpecifier, specifierLen); 3023 3024 // Check the length modifier is valid with the given conversion specifier. 3025 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3026 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3027 diag::warn_format_nonsensical_length); 3028 else if (!FS.hasStandardLengthModifier()) 3029 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3030 else if (!FS.hasStandardLengthConversionCombination()) 3031 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3032 diag::warn_format_non_standard_conversion_spec); 3033 3034 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3035 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3036 3037 // The remaining checks depend on the data arguments. 3038 if (HasVAListArg) 3039 return true; 3040 3041 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3042 return false; 3043 3044 const Expr *Arg = getDataArg(argIndex); 3045 if (!Arg) 3046 return true; 3047 3048 return checkFormatExpr(FS, startSpecifier, specifierLen, Arg); 3049} 3050 3051static bool requiresParensToAddCast(const Expr *E) { 3052 // FIXME: We should have a general way to reason about operator 3053 // precedence and whether parens are actually needed here. 3054 // Take care of a few common cases where they aren't. 3055 const Expr *Inside = E->IgnoreImpCasts(); 3056 if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside)) 3057 Inside = POE->getSyntacticForm()->IgnoreImpCasts(); 3058 3059 switch (Inside->getStmtClass()) { 3060 case Stmt::ArraySubscriptExprClass: 3061 case Stmt::CallExprClass: 3062 case Stmt::CharacterLiteralClass: 3063 case Stmt::CXXBoolLiteralExprClass: 3064 case Stmt::DeclRefExprClass: 3065 case Stmt::FloatingLiteralClass: 3066 case Stmt::IntegerLiteralClass: 3067 case Stmt::MemberExprClass: 3068 case Stmt::ObjCArrayLiteralClass: 3069 case Stmt::ObjCBoolLiteralExprClass: 3070 case Stmt::ObjCBoxedExprClass: 3071 case Stmt::ObjCDictionaryLiteralClass: 3072 case Stmt::ObjCEncodeExprClass: 3073 case Stmt::ObjCIvarRefExprClass: 3074 case Stmt::ObjCMessageExprClass: 3075 case Stmt::ObjCPropertyRefExprClass: 3076 case Stmt::ObjCStringLiteralClass: 3077 case Stmt::ObjCSubscriptRefExprClass: 3078 case Stmt::ParenExprClass: 3079 case Stmt::StringLiteralClass: 3080 case Stmt::UnaryOperatorClass: 3081 return false; 3082 default: 3083 return true; 3084 } 3085} 3086 3087bool 3088CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 3089 const char *StartSpecifier, 3090 unsigned SpecifierLen, 3091 const Expr *E) { 3092 using namespace analyze_format_string; 3093 using namespace analyze_printf; 3094 // Now type check the data expression that matches the 3095 // format specifier. 3096 const analyze_printf::ArgType &AT = FS.getArgType(S.Context, 3097 ObjCContext); 3098 if (!AT.isValid()) 3099 return true; 3100 3101 QualType ExprTy = E->getType(); 3102 while (const TypeOfExprType *TET = dyn_cast<TypeOfExprType>(ExprTy)) { 3103 ExprTy = TET->getUnderlyingExpr()->getType(); 3104 } 3105 3106 if (AT.matchesType(S.Context, ExprTy)) 3107 return true; 3108 3109 // Look through argument promotions for our error message's reported type. 3110 // This includes the integral and floating promotions, but excludes array 3111 // and function pointer decay; seeing that an argument intended to be a 3112 // string has type 'char [6]' is probably more confusing than 'char *'. 3113 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3114 if (ICE->getCastKind() == CK_IntegralCast || 3115 ICE->getCastKind() == CK_FloatingCast) { 3116 E = ICE->getSubExpr(); 3117 ExprTy = E->getType(); 3118 3119 // Check if we didn't match because of an implicit cast from a 'char' 3120 // or 'short' to an 'int'. This is done because printf is a varargs 3121 // function. 3122 if (ICE->getType() == S.Context.IntTy || 3123 ICE->getType() == S.Context.UnsignedIntTy) { 3124 // All further checking is done on the subexpression. 3125 if (AT.matchesType(S.Context, ExprTy)) 3126 return true; 3127 } 3128 } 3129 } else if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E)) { 3130 // Special case for 'a', which has type 'int' in C. 3131 // Note, however, that we do /not/ want to treat multibyte constants like 3132 // 'MooV' as characters! This form is deprecated but still exists. 3133 if (ExprTy == S.Context.IntTy) 3134 if (llvm::isUIntN(S.Context.getCharWidth(), CL->getValue())) 3135 ExprTy = S.Context.CharTy; 3136 } 3137 3138 // %C in an Objective-C context prints a unichar, not a wchar_t. 3139 // If the argument is an integer of some kind, believe the %C and suggest 3140 // a cast instead of changing the conversion specifier. 3141 QualType IntendedTy = ExprTy; 3142 if (ObjCContext && 3143 FS.getConversionSpecifier().getKind() == ConversionSpecifier::CArg) { 3144 if (ExprTy->isIntegralOrUnscopedEnumerationType() && 3145 !ExprTy->isCharType()) { 3146 // 'unichar' is defined as a typedef of unsigned short, but we should 3147 // prefer using the typedef if it is visible. 3148 IntendedTy = S.Context.UnsignedShortTy; 3149 3150 // While we are here, check if the value is an IntegerLiteral that happens 3151 // to be within the valid range. 3152 if (const IntegerLiteral *IL = dyn_cast<IntegerLiteral>(E)) { 3153 const llvm::APInt &V = IL->getValue(); 3154 if (V.getActiveBits() <= S.Context.getTypeSize(IntendedTy)) 3155 return true; 3156 } 3157 3158 LookupResult Result(S, &S.Context.Idents.get("unichar"), E->getLocStart(), 3159 Sema::LookupOrdinaryName); 3160 if (S.LookupName(Result, S.getCurScope())) { 3161 NamedDecl *ND = Result.getFoundDecl(); 3162 if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(ND)) 3163 if (TD->getUnderlyingType() == IntendedTy) 3164 IntendedTy = S.Context.getTypedefType(TD); 3165 } 3166 } 3167 } 3168 3169 // Special-case some of Darwin's platform-independence types by suggesting 3170 // casts to primitive types that are known to be large enough. 3171 bool ShouldNotPrintDirectly = false; 3172 if (S.Context.getTargetInfo().getTriple().isOSDarwin()) { 3173 // Use a 'while' to peel off layers of typedefs. 3174 QualType TyTy = IntendedTy; 3175 while (const TypedefType *UserTy = TyTy->getAs<TypedefType>()) { 3176 StringRef Name = UserTy->getDecl()->getName(); 3177 QualType CastTy = llvm::StringSwitch<QualType>(Name) 3178 .Case("NSInteger", S.Context.LongTy) 3179 .Case("NSUInteger", S.Context.UnsignedLongTy) 3180 .Case("SInt32", S.Context.IntTy) 3181 .Case("UInt32", S.Context.UnsignedIntTy) 3182 .Default(QualType()); 3183 3184 if (!CastTy.isNull()) { 3185 ShouldNotPrintDirectly = true; 3186 IntendedTy = CastTy; 3187 break; 3188 } 3189 TyTy = UserTy->desugar(); 3190 } 3191 } 3192 3193 // We may be able to offer a FixItHint if it is a supported type. 3194 PrintfSpecifier fixedFS = FS; 3195 bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(), 3196 S.Context, ObjCContext); 3197 3198 if (success) { 3199 // Get the fix string from the fixed format specifier 3200 SmallString<16> buf; 3201 llvm::raw_svector_ostream os(buf); 3202 fixedFS.toString(os); 3203 3204 CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen); 3205 3206 if (IntendedTy == ExprTy) { 3207 // In this case, the specifier is wrong and should be changed to match 3208 // the argument. 3209 EmitFormatDiagnostic( 3210 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3211 << AT.getRepresentativeTypeName(S.Context) << IntendedTy 3212 << E->getSourceRange(), 3213 E->getLocStart(), 3214 /*IsStringLocation*/false, 3215 SpecRange, 3216 FixItHint::CreateReplacement(SpecRange, os.str())); 3217 3218 } else { 3219 // The canonical type for formatting this value is different from the 3220 // actual type of the expression. (This occurs, for example, with Darwin's 3221 // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but 3222 // should be printed as 'long' for 64-bit compatibility.) 3223 // Rather than emitting a normal format/argument mismatch, we want to 3224 // add a cast to the recommended type (and correct the format string 3225 // if necessary). 3226 SmallString<16> CastBuf; 3227 llvm::raw_svector_ostream CastFix(CastBuf); 3228 CastFix << "("; 3229 IntendedTy.print(CastFix, S.Context.getPrintingPolicy()); 3230 CastFix << ")"; 3231 3232 SmallVector<FixItHint,4> Hints; 3233 if (!AT.matchesType(S.Context, IntendedTy)) 3234 Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str())); 3235 3236 if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) { 3237 // If there's already a cast present, just replace it. 3238 SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc()); 3239 Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str())); 3240 3241 } else if (!requiresParensToAddCast(E)) { 3242 // If the expression has high enough precedence, 3243 // just write the C-style cast. 3244 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 3245 CastFix.str())); 3246 } else { 3247 // Otherwise, add parens around the expression as well as the cast. 3248 CastFix << "("; 3249 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 3250 CastFix.str())); 3251 3252 SourceLocation After = S.PP.getLocForEndOfToken(E->getLocEnd()); 3253 Hints.push_back(FixItHint::CreateInsertion(After, ")")); 3254 } 3255 3256 if (ShouldNotPrintDirectly) { 3257 // The expression has a type that should not be printed directly. 3258 // We extract the name from the typedef because we don't want to show 3259 // the underlying type in the diagnostic. 3260 StringRef Name = cast<TypedefType>(ExprTy)->getDecl()->getName(); 3261 3262 EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast) 3263 << Name << IntendedTy 3264 << E->getSourceRange(), 3265 E->getLocStart(), /*IsStringLocation=*/false, 3266 SpecRange, Hints); 3267 } else { 3268 // In this case, the expression could be printed using a different 3269 // specifier, but we've decided that the specifier is probably correct 3270 // and we should cast instead. Just use the normal warning message. 3271 EmitFormatDiagnostic( 3272 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3273 << AT.getRepresentativeTypeName(S.Context) << ExprTy 3274 << E->getSourceRange(), 3275 E->getLocStart(), /*IsStringLocation*/false, 3276 SpecRange, Hints); 3277 } 3278 } 3279 } else { 3280 const CharSourceRange &CSR = getSpecifierRange(StartSpecifier, 3281 SpecifierLen); 3282 // Since the warning for passing non-POD types to variadic functions 3283 // was deferred until now, we emit a warning for non-POD 3284 // arguments here. 3285 switch (S.isValidVarArgType(ExprTy)) { 3286 case Sema::VAK_Valid: 3287 case Sema::VAK_ValidInCXX11: 3288 EmitFormatDiagnostic( 3289 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3290 << AT.getRepresentativeTypeName(S.Context) << ExprTy 3291 << CSR 3292 << E->getSourceRange(), 3293 E->getLocStart(), /*IsStringLocation*/false, CSR); 3294 break; 3295 3296 case Sema::VAK_Undefined: 3297 EmitFormatDiagnostic( 3298 S.PDiag(diag::warn_non_pod_vararg_with_format_string) 3299 << S.getLangOpts().CPlusPlus11 3300 << ExprTy 3301 << CallType 3302 << AT.getRepresentativeTypeName(S.Context) 3303 << CSR 3304 << E->getSourceRange(), 3305 E->getLocStart(), /*IsStringLocation*/false, CSR); 3306 checkForCStrMembers(AT, E, CSR); 3307 break; 3308 3309 case Sema::VAK_Invalid: 3310 if (ExprTy->isObjCObjectType()) 3311 EmitFormatDiagnostic( 3312 S.PDiag(diag::err_cannot_pass_objc_interface_to_vararg_format) 3313 << S.getLangOpts().CPlusPlus11 3314 << ExprTy 3315 << CallType 3316 << AT.getRepresentativeTypeName(S.Context) 3317 << CSR 3318 << E->getSourceRange(), 3319 E->getLocStart(), /*IsStringLocation*/false, CSR); 3320 else 3321 // FIXME: If this is an initializer list, suggest removing the braces 3322 // or inserting a cast to the target type. 3323 S.Diag(E->getLocStart(), diag::err_cannot_pass_to_vararg_format) 3324 << isa<InitListExpr>(E) << ExprTy << CallType 3325 << AT.getRepresentativeTypeName(S.Context) 3326 << E->getSourceRange(); 3327 break; 3328 } 3329 3330 assert(FirstDataArg + FS.getArgIndex() < CheckedVarArgs.size() && 3331 "format string specifier index out of range"); 3332 CheckedVarArgs[FirstDataArg + FS.getArgIndex()] = true; 3333 } 3334 3335 return true; 3336} 3337 3338//===--- CHECK: Scanf format string checking ------------------------------===// 3339 3340namespace { 3341class CheckScanfHandler : public CheckFormatHandler { 3342public: 3343 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 3344 const Expr *origFormatExpr, unsigned firstDataArg, 3345 unsigned numDataArgs, const char *beg, bool hasVAListArg, 3346 ArrayRef<const Expr *> Args, 3347 unsigned formatIdx, bool inFunctionCall, 3348 Sema::VariadicCallType CallType, 3349 llvm::SmallBitVector &CheckedVarArgs) 3350 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 3351 numDataArgs, beg, hasVAListArg, 3352 Args, formatIdx, inFunctionCall, CallType, 3353 CheckedVarArgs) 3354 {} 3355 3356 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 3357 const char *startSpecifier, 3358 unsigned specifierLen); 3359 3360 bool HandleInvalidScanfConversionSpecifier( 3361 const analyze_scanf::ScanfSpecifier &FS, 3362 const char *startSpecifier, 3363 unsigned specifierLen); 3364 3365 void HandleIncompleteScanList(const char *start, const char *end); 3366}; 3367} 3368 3369void CheckScanfHandler::HandleIncompleteScanList(const char *start, 3370 const char *end) { 3371 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 3372 getLocationOfByte(end), /*IsStringLocation*/true, 3373 getSpecifierRange(start, end - start)); 3374} 3375 3376bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 3377 const analyze_scanf::ScanfSpecifier &FS, 3378 const char *startSpecifier, 3379 unsigned specifierLen) { 3380 3381 const analyze_scanf::ScanfConversionSpecifier &CS = 3382 FS.getConversionSpecifier(); 3383 3384 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 3385 getLocationOfByte(CS.getStart()), 3386 startSpecifier, specifierLen, 3387 CS.getStart(), CS.getLength()); 3388} 3389 3390bool CheckScanfHandler::HandleScanfSpecifier( 3391 const analyze_scanf::ScanfSpecifier &FS, 3392 const char *startSpecifier, 3393 unsigned specifierLen) { 3394 3395 using namespace analyze_scanf; 3396 using namespace analyze_format_string; 3397 3398 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 3399 3400 // Handle case where '%' and '*' don't consume an argument. These shouldn't 3401 // be used to decide if we are using positional arguments consistently. 3402 if (FS.consumesDataArgument()) { 3403 if (atFirstArg) { 3404 atFirstArg = false; 3405 usesPositionalArgs = FS.usesPositionalArg(); 3406 } 3407 else if (usesPositionalArgs != FS.usesPositionalArg()) { 3408 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 3409 startSpecifier, specifierLen); 3410 return false; 3411 } 3412 } 3413 3414 // Check if the field with is non-zero. 3415 const OptionalAmount &Amt = FS.getFieldWidth(); 3416 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 3417 if (Amt.getConstantAmount() == 0) { 3418 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 3419 Amt.getConstantLength()); 3420 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 3421 getLocationOfByte(Amt.getStart()), 3422 /*IsStringLocation*/true, R, 3423 FixItHint::CreateRemoval(R)); 3424 } 3425 } 3426 3427 if (!FS.consumesDataArgument()) { 3428 // FIXME: Technically specifying a precision or field width here 3429 // makes no sense. Worth issuing a warning at some point. 3430 return true; 3431 } 3432 3433 // Consume the argument. 3434 unsigned argIndex = FS.getArgIndex(); 3435 if (argIndex < NumDataArgs) { 3436 // The check to see if the argIndex is valid will come later. 3437 // We set the bit here because we may exit early from this 3438 // function if we encounter some other error. 3439 CoveredArgs.set(argIndex); 3440 } 3441 3442 // Check the length modifier is valid with the given conversion specifier. 3443 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3444 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3445 diag::warn_format_nonsensical_length); 3446 else if (!FS.hasStandardLengthModifier()) 3447 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3448 else if (!FS.hasStandardLengthConversionCombination()) 3449 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3450 diag::warn_format_non_standard_conversion_spec); 3451 3452 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3453 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3454 3455 // The remaining checks depend on the data arguments. 3456 if (HasVAListArg) 3457 return true; 3458 3459 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3460 return false; 3461 3462 // Check that the argument type matches the format specifier. 3463 const Expr *Ex = getDataArg(argIndex); 3464 if (!Ex) 3465 return true; 3466 3467 const analyze_format_string::ArgType &AT = FS.getArgType(S.Context); 3468 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) { 3469 ScanfSpecifier fixedFS = FS; 3470 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 3471 S.Context); 3472 3473 if (success) { 3474 // Get the fix string from the fixed format specifier. 3475 SmallString<128> buf; 3476 llvm::raw_svector_ostream os(buf); 3477 fixedFS.toString(os); 3478 3479 EmitFormatDiagnostic( 3480 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3481 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3482 << Ex->getSourceRange(), 3483 Ex->getLocStart(), 3484 /*IsStringLocation*/false, 3485 getSpecifierRange(startSpecifier, specifierLen), 3486 FixItHint::CreateReplacement( 3487 getSpecifierRange(startSpecifier, specifierLen), 3488 os.str())); 3489 } else { 3490 EmitFormatDiagnostic( 3491 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3492 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3493 << Ex->getSourceRange(), 3494 Ex->getLocStart(), 3495 /*IsStringLocation*/false, 3496 getSpecifierRange(startSpecifier, specifierLen)); 3497 } 3498 } 3499 3500 return true; 3501} 3502 3503void Sema::CheckFormatString(const StringLiteral *FExpr, 3504 const Expr *OrigFormatExpr, 3505 ArrayRef<const Expr *> Args, 3506 bool HasVAListArg, unsigned format_idx, 3507 unsigned firstDataArg, FormatStringType Type, 3508 bool inFunctionCall, VariadicCallType CallType, 3509 llvm::SmallBitVector &CheckedVarArgs) { 3510 3511 // CHECK: is the format string a wide literal? 3512 if (!FExpr->isAscii() && !FExpr->isUTF8()) { 3513 CheckFormatHandler::EmitFormatDiagnostic( 3514 *this, inFunctionCall, Args[format_idx], 3515 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 3516 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3517 return; 3518 } 3519 3520 // Str - The format string. NOTE: this is NOT null-terminated! 3521 StringRef StrRef = FExpr->getString(); 3522 const char *Str = StrRef.data(); 3523 unsigned StrLen = StrRef.size(); 3524 const unsigned numDataArgs = Args.size() - firstDataArg; 3525 3526 // CHECK: empty format string? 3527 if (StrLen == 0 && numDataArgs > 0) { 3528 CheckFormatHandler::EmitFormatDiagnostic( 3529 *this, inFunctionCall, Args[format_idx], 3530 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 3531 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3532 return; 3533 } 3534 3535 if (Type == FST_Printf || Type == FST_NSString) { 3536 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 3537 numDataArgs, (Type == FST_NSString), 3538 Str, HasVAListArg, Args, format_idx, 3539 inFunctionCall, CallType, CheckedVarArgs); 3540 3541 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 3542 getLangOpts(), 3543 Context.getTargetInfo())) 3544 H.DoneProcessing(); 3545 } else if (Type == FST_Scanf) { 3546 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs, 3547 Str, HasVAListArg, Args, format_idx, 3548 inFunctionCall, CallType, CheckedVarArgs); 3549 3550 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 3551 getLangOpts(), 3552 Context.getTargetInfo())) 3553 H.DoneProcessing(); 3554 } // TODO: handle other formats 3555} 3556 3557//===--- CHECK: Standard memory functions ---------------------------------===// 3558 3559/// \brief Determine whether the given type is a dynamic class type (e.g., 3560/// whether it has a vtable). 3561static bool isDynamicClassType(QualType T) { 3562 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 3563 if (CXXRecordDecl *Definition = Record->getDefinition()) 3564 if (Definition->isDynamicClass()) 3565 return true; 3566 3567 return false; 3568} 3569 3570/// \brief If E is a sizeof expression, returns its argument expression, 3571/// otherwise returns NULL. 3572static const Expr *getSizeOfExprArg(const Expr* E) { 3573 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3574 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3575 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 3576 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 3577 3578 return 0; 3579} 3580 3581/// \brief If E is a sizeof expression, returns its argument type. 3582static QualType getSizeOfArgType(const Expr* E) { 3583 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3584 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3585 if (SizeOf->getKind() == clang::UETT_SizeOf) 3586 return SizeOf->getTypeOfArgument(); 3587 3588 return QualType(); 3589} 3590 3591/// \brief Check for dangerous or invalid arguments to memset(). 3592/// 3593/// This issues warnings on known problematic, dangerous or unspecified 3594/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 3595/// function calls. 3596/// 3597/// \param Call The call expression to diagnose. 3598void Sema::CheckMemaccessArguments(const CallExpr *Call, 3599 unsigned BId, 3600 IdentifierInfo *FnName) { 3601 assert(BId != 0); 3602 3603 // It is possible to have a non-standard definition of memset. Validate 3604 // we have enough arguments, and if not, abort further checking. 3605 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 3606 if (Call->getNumArgs() < ExpectedNumArgs) 3607 return; 3608 3609 unsigned LastArg = (BId == Builtin::BImemset || 3610 BId == Builtin::BIstrndup ? 1 : 2); 3611 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 3612 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 3613 3614 // We have special checking when the length is a sizeof expression. 3615 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 3616 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 3617 llvm::FoldingSetNodeID SizeOfArgID; 3618 3619 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 3620 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 3621 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 3622 3623 QualType DestTy = Dest->getType(); 3624 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 3625 QualType PointeeTy = DestPtrTy->getPointeeType(); 3626 3627 // Never warn about void type pointers. This can be used to suppress 3628 // false positives. 3629 if (PointeeTy->isVoidType()) 3630 continue; 3631 3632 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 3633 // actually comparing the expressions for equality. Because computing the 3634 // expression IDs can be expensive, we only do this if the diagnostic is 3635 // enabled. 3636 if (SizeOfArg && 3637 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 3638 SizeOfArg->getExprLoc())) { 3639 // We only compute IDs for expressions if the warning is enabled, and 3640 // cache the sizeof arg's ID. 3641 if (SizeOfArgID == llvm::FoldingSetNodeID()) 3642 SizeOfArg->Profile(SizeOfArgID, Context, true); 3643 llvm::FoldingSetNodeID DestID; 3644 Dest->Profile(DestID, Context, true); 3645 if (DestID == SizeOfArgID) { 3646 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 3647 // over sizeof(src) as well. 3648 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 3649 StringRef ReadableName = FnName->getName(); 3650 3651 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 3652 if (UnaryOp->getOpcode() == UO_AddrOf) 3653 ActionIdx = 1; // If its an address-of operator, just remove it. 3654 if (!PointeeTy->isIncompleteType() && 3655 (Context.getTypeSize(PointeeTy) == Context.getCharWidth())) 3656 ActionIdx = 2; // If the pointee's size is sizeof(char), 3657 // suggest an explicit length. 3658 3659 // If the function is defined as a builtin macro, do not show macro 3660 // expansion. 3661 SourceLocation SL = SizeOfArg->getExprLoc(); 3662 SourceRange DSR = Dest->getSourceRange(); 3663 SourceRange SSR = SizeOfArg->getSourceRange(); 3664 SourceManager &SM = PP.getSourceManager(); 3665 3666 if (SM.isMacroArgExpansion(SL)) { 3667 ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts); 3668 SL = SM.getSpellingLoc(SL); 3669 DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()), 3670 SM.getSpellingLoc(DSR.getEnd())); 3671 SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()), 3672 SM.getSpellingLoc(SSR.getEnd())); 3673 } 3674 3675 DiagRuntimeBehavior(SL, SizeOfArg, 3676 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 3677 << ReadableName 3678 << PointeeTy 3679 << DestTy 3680 << DSR 3681 << SSR); 3682 DiagRuntimeBehavior(SL, SizeOfArg, 3683 PDiag(diag::warn_sizeof_pointer_expr_memaccess_note) 3684 << ActionIdx 3685 << SSR); 3686 3687 break; 3688 } 3689 } 3690 3691 // Also check for cases where the sizeof argument is the exact same 3692 // type as the memory argument, and where it points to a user-defined 3693 // record type. 3694 if (SizeOfArgTy != QualType()) { 3695 if (PointeeTy->isRecordType() && 3696 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 3697 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 3698 PDiag(diag::warn_sizeof_pointer_type_memaccess) 3699 << FnName << SizeOfArgTy << ArgIdx 3700 << PointeeTy << Dest->getSourceRange() 3701 << LenExpr->getSourceRange()); 3702 break; 3703 } 3704 } 3705 3706 // Always complain about dynamic classes. 3707 if (isDynamicClassType(PointeeTy)) { 3708 3709 unsigned OperationType = 0; 3710 // "overwritten" if we're warning about the destination for any call 3711 // but memcmp; otherwise a verb appropriate to the call. 3712 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 3713 if (BId == Builtin::BImemcpy) 3714 OperationType = 1; 3715 else if(BId == Builtin::BImemmove) 3716 OperationType = 2; 3717 else if (BId == Builtin::BImemcmp) 3718 OperationType = 3; 3719 } 3720 3721 DiagRuntimeBehavior( 3722 Dest->getExprLoc(), Dest, 3723 PDiag(diag::warn_dyn_class_memaccess) 3724 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 3725 << FnName << PointeeTy 3726 << OperationType 3727 << Call->getCallee()->getSourceRange()); 3728 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 3729 BId != Builtin::BImemset) 3730 DiagRuntimeBehavior( 3731 Dest->getExprLoc(), Dest, 3732 PDiag(diag::warn_arc_object_memaccess) 3733 << ArgIdx << FnName << PointeeTy 3734 << Call->getCallee()->getSourceRange()); 3735 else 3736 continue; 3737 3738 DiagRuntimeBehavior( 3739 Dest->getExprLoc(), Dest, 3740 PDiag(diag::note_bad_memaccess_silence) 3741 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 3742 break; 3743 } 3744 } 3745} 3746 3747// A little helper routine: ignore addition and subtraction of integer literals. 3748// This intentionally does not ignore all integer constant expressions because 3749// we don't want to remove sizeof(). 3750static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 3751 Ex = Ex->IgnoreParenCasts(); 3752 3753 for (;;) { 3754 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 3755 if (!BO || !BO->isAdditiveOp()) 3756 break; 3757 3758 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 3759 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 3760 3761 if (isa<IntegerLiteral>(RHS)) 3762 Ex = LHS; 3763 else if (isa<IntegerLiteral>(LHS)) 3764 Ex = RHS; 3765 else 3766 break; 3767 } 3768 3769 return Ex; 3770} 3771 3772static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty, 3773 ASTContext &Context) { 3774 // Only handle constant-sized or VLAs, but not flexible members. 3775 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) { 3776 // Only issue the FIXIT for arrays of size > 1. 3777 if (CAT->getSize().getSExtValue() <= 1) 3778 return false; 3779 } else if (!Ty->isVariableArrayType()) { 3780 return false; 3781 } 3782 return true; 3783} 3784 3785// Warn if the user has made the 'size' argument to strlcpy or strlcat 3786// be the size of the source, instead of the destination. 3787void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 3788 IdentifierInfo *FnName) { 3789 3790 // Don't crash if the user has the wrong number of arguments 3791 if (Call->getNumArgs() != 3) 3792 return; 3793 3794 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 3795 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 3796 const Expr *CompareWithSrc = NULL; 3797 3798 // Look for 'strlcpy(dst, x, sizeof(x))' 3799 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 3800 CompareWithSrc = Ex; 3801 else { 3802 // Look for 'strlcpy(dst, x, strlen(x))' 3803 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 3804 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 3805 && SizeCall->getNumArgs() == 1) 3806 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 3807 } 3808 } 3809 3810 if (!CompareWithSrc) 3811 return; 3812 3813 // Determine if the argument to sizeof/strlen is equal to the source 3814 // argument. In principle there's all kinds of things you could do 3815 // here, for instance creating an == expression and evaluating it with 3816 // EvaluateAsBooleanCondition, but this uses a more direct technique: 3817 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 3818 if (!SrcArgDRE) 3819 return; 3820 3821 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 3822 if (!CompareWithSrcDRE || 3823 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 3824 return; 3825 3826 const Expr *OriginalSizeArg = Call->getArg(2); 3827 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 3828 << OriginalSizeArg->getSourceRange() << FnName; 3829 3830 // Output a FIXIT hint if the destination is an array (rather than a 3831 // pointer to an array). This could be enhanced to handle some 3832 // pointers if we know the actual size, like if DstArg is 'array+2' 3833 // we could say 'sizeof(array)-2'. 3834 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 3835 if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context)) 3836 return; 3837 3838 SmallString<128> sizeString; 3839 llvm::raw_svector_ostream OS(sizeString); 3840 OS << "sizeof("; 3841 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3842 OS << ")"; 3843 3844 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 3845 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 3846 OS.str()); 3847} 3848 3849/// Check if two expressions refer to the same declaration. 3850static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 3851 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 3852 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 3853 return D1->getDecl() == D2->getDecl(); 3854 return false; 3855} 3856 3857static const Expr *getStrlenExprArg(const Expr *E) { 3858 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 3859 const FunctionDecl *FD = CE->getDirectCallee(); 3860 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 3861 return 0; 3862 return CE->getArg(0)->IgnoreParenCasts(); 3863 } 3864 return 0; 3865} 3866 3867// Warn on anti-patterns as the 'size' argument to strncat. 3868// The correct size argument should look like following: 3869// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 3870void Sema::CheckStrncatArguments(const CallExpr *CE, 3871 IdentifierInfo *FnName) { 3872 // Don't crash if the user has the wrong number of arguments. 3873 if (CE->getNumArgs() < 3) 3874 return; 3875 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 3876 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 3877 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 3878 3879 // Identify common expressions, which are wrongly used as the size argument 3880 // to strncat and may lead to buffer overflows. 3881 unsigned PatternType = 0; 3882 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 3883 // - sizeof(dst) 3884 if (referToTheSameDecl(SizeOfArg, DstArg)) 3885 PatternType = 1; 3886 // - sizeof(src) 3887 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 3888 PatternType = 2; 3889 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 3890 if (BE->getOpcode() == BO_Sub) { 3891 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 3892 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 3893 // - sizeof(dst) - strlen(dst) 3894 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 3895 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 3896 PatternType = 1; 3897 // - sizeof(src) - (anything) 3898 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 3899 PatternType = 2; 3900 } 3901 } 3902 3903 if (PatternType == 0) 3904 return; 3905 3906 // Generate the diagnostic. 3907 SourceLocation SL = LenArg->getLocStart(); 3908 SourceRange SR = LenArg->getSourceRange(); 3909 SourceManager &SM = PP.getSourceManager(); 3910 3911 // If the function is defined as a builtin macro, do not show macro expansion. 3912 if (SM.isMacroArgExpansion(SL)) { 3913 SL = SM.getSpellingLoc(SL); 3914 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 3915 SM.getSpellingLoc(SR.getEnd())); 3916 } 3917 3918 // Check if the destination is an array (rather than a pointer to an array). 3919 QualType DstTy = DstArg->getType(); 3920 bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy, 3921 Context); 3922 if (!isKnownSizeArray) { 3923 if (PatternType == 1) 3924 Diag(SL, diag::warn_strncat_wrong_size) << SR; 3925 else 3926 Diag(SL, diag::warn_strncat_src_size) << SR; 3927 return; 3928 } 3929 3930 if (PatternType == 1) 3931 Diag(SL, diag::warn_strncat_large_size) << SR; 3932 else 3933 Diag(SL, diag::warn_strncat_src_size) << SR; 3934 3935 SmallString<128> sizeString; 3936 llvm::raw_svector_ostream OS(sizeString); 3937 OS << "sizeof("; 3938 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3939 OS << ") - "; 3940 OS << "strlen("; 3941 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3942 OS << ") - 1"; 3943 3944 Diag(SL, diag::note_strncat_wrong_size) 3945 << FixItHint::CreateReplacement(SR, OS.str()); 3946} 3947 3948//===--- CHECK: Return Address of Stack Variable --------------------------===// 3949 3950static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3951 Decl *ParentDecl); 3952static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3953 Decl *ParentDecl); 3954 3955/// CheckReturnStackAddr - Check if a return statement returns the address 3956/// of a stack variable. 3957void 3958Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3959 SourceLocation ReturnLoc) { 3960 3961 Expr *stackE = 0; 3962 SmallVector<DeclRefExpr *, 8> refVars; 3963 3964 // Perform checking for returned stack addresses, local blocks, 3965 // label addresses or references to temporaries. 3966 if (lhsType->isPointerType() || 3967 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 3968 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 3969 } else if (lhsType->isReferenceType()) { 3970 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 3971 } 3972 3973 if (stackE == 0) 3974 return; // Nothing suspicious was found. 3975 3976 SourceLocation diagLoc; 3977 SourceRange diagRange; 3978 if (refVars.empty()) { 3979 diagLoc = stackE->getLocStart(); 3980 diagRange = stackE->getSourceRange(); 3981 } else { 3982 // We followed through a reference variable. 'stackE' contains the 3983 // problematic expression but we will warn at the return statement pointing 3984 // at the reference variable. We will later display the "trail" of 3985 // reference variables using notes. 3986 diagLoc = refVars[0]->getLocStart(); 3987 diagRange = refVars[0]->getSourceRange(); 3988 } 3989 3990 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 3991 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 3992 : diag::warn_ret_stack_addr) 3993 << DR->getDecl()->getDeclName() << diagRange; 3994 } else if (isa<BlockExpr>(stackE)) { // local block. 3995 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 3996 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 3997 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 3998 } else { // local temporary. 3999 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 4000 : diag::warn_ret_local_temp_addr) 4001 << diagRange; 4002 } 4003 4004 // Display the "trail" of reference variables that we followed until we 4005 // found the problematic expression using notes. 4006 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 4007 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 4008 // If this var binds to another reference var, show the range of the next 4009 // var, otherwise the var binds to the problematic expression, in which case 4010 // show the range of the expression. 4011 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 4012 : stackE->getSourceRange(); 4013 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 4014 << VD->getDeclName() << range; 4015 } 4016} 4017 4018/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 4019/// check if the expression in a return statement evaluates to an address 4020/// to a location on the stack, a local block, an address of a label, or a 4021/// reference to local temporary. The recursion is used to traverse the 4022/// AST of the return expression, with recursion backtracking when we 4023/// encounter a subexpression that (1) clearly does not lead to one of the 4024/// above problematic expressions (2) is something we cannot determine leads to 4025/// a problematic expression based on such local checking. 4026/// 4027/// Both EvalAddr and EvalVal follow through reference variables to evaluate 4028/// the expression that they point to. Such variables are added to the 4029/// 'refVars' vector so that we know what the reference variable "trail" was. 4030/// 4031/// EvalAddr processes expressions that are pointers that are used as 4032/// references (and not L-values). EvalVal handles all other values. 4033/// At the base case of the recursion is a check for the above problematic 4034/// expressions. 4035/// 4036/// This implementation handles: 4037/// 4038/// * pointer-to-pointer casts 4039/// * implicit conversions from array references to pointers 4040/// * taking the address of fields 4041/// * arbitrary interplay between "&" and "*" operators 4042/// * pointer arithmetic from an address of a stack variable 4043/// * taking the address of an array element where the array is on the stack 4044static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 4045 Decl *ParentDecl) { 4046 if (E->isTypeDependent()) 4047 return NULL; 4048 4049 // We should only be called for evaluating pointer expressions. 4050 assert((E->getType()->isAnyPointerType() || 4051 E->getType()->isBlockPointerType() || 4052 E->getType()->isObjCQualifiedIdType()) && 4053 "EvalAddr only works on pointers"); 4054 4055 E = E->IgnoreParens(); 4056 4057 // Our "symbolic interpreter" is just a dispatch off the currently 4058 // viewed AST node. We then recursively traverse the AST by calling 4059 // EvalAddr and EvalVal appropriately. 4060 switch (E->getStmtClass()) { 4061 case Stmt::DeclRefExprClass: { 4062 DeclRefExpr *DR = cast<DeclRefExpr>(E); 4063 4064 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 4065 // If this is a reference variable, follow through to the expression that 4066 // it points to. 4067 if (V->hasLocalStorage() && 4068 V->getType()->isReferenceType() && V->hasInit()) { 4069 // Add the reference variable to the "trail". 4070 refVars.push_back(DR); 4071 return EvalAddr(V->getInit(), refVars, ParentDecl); 4072 } 4073 4074 return NULL; 4075 } 4076 4077 case Stmt::UnaryOperatorClass: { 4078 // The only unary operator that make sense to handle here 4079 // is AddrOf. All others don't make sense as pointers. 4080 UnaryOperator *U = cast<UnaryOperator>(E); 4081 4082 if (U->getOpcode() == UO_AddrOf) 4083 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 4084 else 4085 return NULL; 4086 } 4087 4088 case Stmt::BinaryOperatorClass: { 4089 // Handle pointer arithmetic. All other binary operators are not valid 4090 // in this context. 4091 BinaryOperator *B = cast<BinaryOperator>(E); 4092 BinaryOperatorKind op = B->getOpcode(); 4093 4094 if (op != BO_Add && op != BO_Sub) 4095 return NULL; 4096 4097 Expr *Base = B->getLHS(); 4098 4099 // Determine which argument is the real pointer base. It could be 4100 // the RHS argument instead of the LHS. 4101 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 4102 4103 assert (Base->getType()->isPointerType()); 4104 return EvalAddr(Base, refVars, ParentDecl); 4105 } 4106 4107 // For conditional operators we need to see if either the LHS or RHS are 4108 // valid DeclRefExpr*s. If one of them is valid, we return it. 4109 case Stmt::ConditionalOperatorClass: { 4110 ConditionalOperator *C = cast<ConditionalOperator>(E); 4111 4112 // Handle the GNU extension for missing LHS. 4113 if (Expr *lhsExpr = C->getLHS()) { 4114 // In C++, we can have a throw-expression, which has 'void' type. 4115 if (!lhsExpr->getType()->isVoidType()) 4116 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 4117 return LHS; 4118 } 4119 4120 // In C++, we can have a throw-expression, which has 'void' type. 4121 if (C->getRHS()->getType()->isVoidType()) 4122 return NULL; 4123 4124 return EvalAddr(C->getRHS(), refVars, ParentDecl); 4125 } 4126 4127 case Stmt::BlockExprClass: 4128 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 4129 return E; // local block. 4130 return NULL; 4131 4132 case Stmt::AddrLabelExprClass: 4133 return E; // address of label. 4134 4135 case Stmt::ExprWithCleanupsClass: 4136 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 4137 ParentDecl); 4138 4139 // For casts, we need to handle conversions from arrays to 4140 // pointer values, and pointer-to-pointer conversions. 4141 case Stmt::ImplicitCastExprClass: 4142 case Stmt::CStyleCastExprClass: 4143 case Stmt::CXXFunctionalCastExprClass: 4144 case Stmt::ObjCBridgedCastExprClass: 4145 case Stmt::CXXStaticCastExprClass: 4146 case Stmt::CXXDynamicCastExprClass: 4147 case Stmt::CXXConstCastExprClass: 4148 case Stmt::CXXReinterpretCastExprClass: { 4149 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 4150 switch (cast<CastExpr>(E)->getCastKind()) { 4151 case CK_BitCast: 4152 case CK_LValueToRValue: 4153 case CK_NoOp: 4154 case CK_BaseToDerived: 4155 case CK_DerivedToBase: 4156 case CK_UncheckedDerivedToBase: 4157 case CK_Dynamic: 4158 case CK_CPointerToObjCPointerCast: 4159 case CK_BlockPointerToObjCPointerCast: 4160 case CK_AnyPointerToBlockPointerCast: 4161 return EvalAddr(SubExpr, refVars, ParentDecl); 4162 4163 case CK_ArrayToPointerDecay: 4164 return EvalVal(SubExpr, refVars, ParentDecl); 4165 4166 default: 4167 return 0; 4168 } 4169 } 4170 4171 case Stmt::MaterializeTemporaryExprClass: 4172 if (Expr *Result = EvalAddr( 4173 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 4174 refVars, ParentDecl)) 4175 return Result; 4176 4177 return E; 4178 4179 // Everything else: we simply don't reason about them. 4180 default: 4181 return NULL; 4182 } 4183} 4184 4185 4186/// EvalVal - This function is complements EvalAddr in the mutual recursion. 4187/// See the comments for EvalAddr for more details. 4188static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 4189 Decl *ParentDecl) { 4190do { 4191 // We should only be called for evaluating non-pointer expressions, or 4192 // expressions with a pointer type that are not used as references but instead 4193 // are l-values (e.g., DeclRefExpr with a pointer type). 4194 4195 // Our "symbolic interpreter" is just a dispatch off the currently 4196 // viewed AST node. We then recursively traverse the AST by calling 4197 // EvalAddr and EvalVal appropriately. 4198 4199 E = E->IgnoreParens(); 4200 switch (E->getStmtClass()) { 4201 case Stmt::ImplicitCastExprClass: { 4202 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 4203 if (IE->getValueKind() == VK_LValue) { 4204 E = IE->getSubExpr(); 4205 continue; 4206 } 4207 return NULL; 4208 } 4209 4210 case Stmt::ExprWithCleanupsClass: 4211 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 4212 4213 case Stmt::DeclRefExprClass: { 4214 // When we hit a DeclRefExpr we are looking at code that refers to a 4215 // variable's name. If it's not a reference variable we check if it has 4216 // local storage within the function, and if so, return the expression. 4217 DeclRefExpr *DR = cast<DeclRefExpr>(E); 4218 4219 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 4220 // Check if it refers to itself, e.g. "int& i = i;". 4221 if (V == ParentDecl) 4222 return DR; 4223 4224 if (V->hasLocalStorage()) { 4225 if (!V->getType()->isReferenceType()) 4226 return DR; 4227 4228 // Reference variable, follow through to the expression that 4229 // it points to. 4230 if (V->hasInit()) { 4231 // Add the reference variable to the "trail". 4232 refVars.push_back(DR); 4233 return EvalVal(V->getInit(), refVars, V); 4234 } 4235 } 4236 } 4237 4238 return NULL; 4239 } 4240 4241 case Stmt::UnaryOperatorClass: { 4242 // The only unary operator that make sense to handle here 4243 // is Deref. All others don't resolve to a "name." This includes 4244 // handling all sorts of rvalues passed to a unary operator. 4245 UnaryOperator *U = cast<UnaryOperator>(E); 4246 4247 if (U->getOpcode() == UO_Deref) 4248 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 4249 4250 return NULL; 4251 } 4252 4253 case Stmt::ArraySubscriptExprClass: { 4254 // Array subscripts are potential references to data on the stack. We 4255 // retrieve the DeclRefExpr* for the array variable if it indeed 4256 // has local storage. 4257 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 4258 } 4259 4260 case Stmt::ConditionalOperatorClass: { 4261 // For conditional operators we need to see if either the LHS or RHS are 4262 // non-NULL Expr's. If one is non-NULL, we return it. 4263 ConditionalOperator *C = cast<ConditionalOperator>(E); 4264 4265 // Handle the GNU extension for missing LHS. 4266 if (Expr *lhsExpr = C->getLHS()) 4267 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 4268 return LHS; 4269 4270 return EvalVal(C->getRHS(), refVars, ParentDecl); 4271 } 4272 4273 // Accesses to members are potential references to data on the stack. 4274 case Stmt::MemberExprClass: { 4275 MemberExpr *M = cast<MemberExpr>(E); 4276 4277 // Check for indirect access. We only want direct field accesses. 4278 if (M->isArrow()) 4279 return NULL; 4280 4281 // Check whether the member type is itself a reference, in which case 4282 // we're not going to refer to the member, but to what the member refers to. 4283 if (M->getMemberDecl()->getType()->isReferenceType()) 4284 return NULL; 4285 4286 return EvalVal(M->getBase(), refVars, ParentDecl); 4287 } 4288 4289 case Stmt::MaterializeTemporaryExprClass: 4290 if (Expr *Result = EvalVal( 4291 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 4292 refVars, ParentDecl)) 4293 return Result; 4294 4295 return E; 4296 4297 default: 4298 // Check that we don't return or take the address of a reference to a 4299 // temporary. This is only useful in C++. 4300 if (!E->isTypeDependent() && E->isRValue()) 4301 return E; 4302 4303 // Everything else: we simply don't reason about them. 4304 return NULL; 4305 } 4306} while (true); 4307} 4308 4309//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 4310 4311/// Check for comparisons of floating point operands using != and ==. 4312/// Issue a warning if these are no self-comparisons, as they are not likely 4313/// to do what the programmer intended. 4314void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 4315 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 4316 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 4317 4318 // Special case: check for x == x (which is OK). 4319 // Do not emit warnings for such cases. 4320 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 4321 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 4322 if (DRL->getDecl() == DRR->getDecl()) 4323 return; 4324 4325 4326 // Special case: check for comparisons against literals that can be exactly 4327 // represented by APFloat. In such cases, do not emit a warning. This 4328 // is a heuristic: often comparison against such literals are used to 4329 // detect if a value in a variable has not changed. This clearly can 4330 // lead to false negatives. 4331 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 4332 if (FLL->isExact()) 4333 return; 4334 } else 4335 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)) 4336 if (FLR->isExact()) 4337 return; 4338 4339 // Check for comparisons with builtin types. 4340 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 4341 if (CL->isBuiltinCall()) 4342 return; 4343 4344 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 4345 if (CR->isBuiltinCall()) 4346 return; 4347 4348 // Emit the diagnostic. 4349 Diag(Loc, diag::warn_floatingpoint_eq) 4350 << LHS->getSourceRange() << RHS->getSourceRange(); 4351} 4352 4353//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 4354//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 4355 4356namespace { 4357 4358/// Structure recording the 'active' range of an integer-valued 4359/// expression. 4360struct IntRange { 4361 /// The number of bits active in the int. 4362 unsigned Width; 4363 4364 /// True if the int is known not to have negative values. 4365 bool NonNegative; 4366 4367 IntRange(unsigned Width, bool NonNegative) 4368 : Width(Width), NonNegative(NonNegative) 4369 {} 4370 4371 /// Returns the range of the bool type. 4372 static IntRange forBoolType() { 4373 return IntRange(1, true); 4374 } 4375 4376 /// Returns the range of an opaque value of the given integral type. 4377 static IntRange forValueOfType(ASTContext &C, QualType T) { 4378 return forValueOfCanonicalType(C, 4379 T->getCanonicalTypeInternal().getTypePtr()); 4380 } 4381 4382 /// Returns the range of an opaque value of a canonical integral type. 4383 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 4384 assert(T->isCanonicalUnqualified()); 4385 4386 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4387 T = VT->getElementType().getTypePtr(); 4388 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4389 T = CT->getElementType().getTypePtr(); 4390 4391 // For enum types, use the known bit width of the enumerators. 4392 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 4393 EnumDecl *Enum = ET->getDecl(); 4394 if (!Enum->isCompleteDefinition()) 4395 return IntRange(C.getIntWidth(QualType(T, 0)), false); 4396 4397 unsigned NumPositive = Enum->getNumPositiveBits(); 4398 unsigned NumNegative = Enum->getNumNegativeBits(); 4399 4400 if (NumNegative == 0) 4401 return IntRange(NumPositive, true/*NonNegative*/); 4402 else 4403 return IntRange(std::max(NumPositive + 1, NumNegative), 4404 false/*NonNegative*/); 4405 } 4406 4407 const BuiltinType *BT = cast<BuiltinType>(T); 4408 assert(BT->isInteger()); 4409 4410 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4411 } 4412 4413 /// Returns the "target" range of a canonical integral type, i.e. 4414 /// the range of values expressible in the type. 4415 /// 4416 /// This matches forValueOfCanonicalType except that enums have the 4417 /// full range of their type, not the range of their enumerators. 4418 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 4419 assert(T->isCanonicalUnqualified()); 4420 4421 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4422 T = VT->getElementType().getTypePtr(); 4423 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4424 T = CT->getElementType().getTypePtr(); 4425 if (const EnumType *ET = dyn_cast<EnumType>(T)) 4426 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 4427 4428 const BuiltinType *BT = cast<BuiltinType>(T); 4429 assert(BT->isInteger()); 4430 4431 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4432 } 4433 4434 /// Returns the supremum of two ranges: i.e. their conservative merge. 4435 static IntRange join(IntRange L, IntRange R) { 4436 return IntRange(std::max(L.Width, R.Width), 4437 L.NonNegative && R.NonNegative); 4438 } 4439 4440 /// Returns the infinum of two ranges: i.e. their aggressive merge. 4441 static IntRange meet(IntRange L, IntRange R) { 4442 return IntRange(std::min(L.Width, R.Width), 4443 L.NonNegative || R.NonNegative); 4444 } 4445}; 4446 4447static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 4448 unsigned MaxWidth) { 4449 if (value.isSigned() && value.isNegative()) 4450 return IntRange(value.getMinSignedBits(), false); 4451 4452 if (value.getBitWidth() > MaxWidth) 4453 value = value.trunc(MaxWidth); 4454 4455 // isNonNegative() just checks the sign bit without considering 4456 // signedness. 4457 return IntRange(value.getActiveBits(), true); 4458} 4459 4460static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 4461 unsigned MaxWidth) { 4462 if (result.isInt()) 4463 return GetValueRange(C, result.getInt(), MaxWidth); 4464 4465 if (result.isVector()) { 4466 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 4467 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 4468 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 4469 R = IntRange::join(R, El); 4470 } 4471 return R; 4472 } 4473 4474 if (result.isComplexInt()) { 4475 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 4476 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 4477 return IntRange::join(R, I); 4478 } 4479 4480 // This can happen with lossless casts to intptr_t of "based" lvalues. 4481 // Assume it might use arbitrary bits. 4482 // FIXME: The only reason we need to pass the type in here is to get 4483 // the sign right on this one case. It would be nice if APValue 4484 // preserved this. 4485 assert(result.isLValue() || result.isAddrLabelDiff()); 4486 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 4487} 4488 4489static QualType GetExprType(Expr *E) { 4490 QualType Ty = E->getType(); 4491 if (const AtomicType *AtomicRHS = Ty->getAs<AtomicType>()) 4492 Ty = AtomicRHS->getValueType(); 4493 return Ty; 4494} 4495 4496/// Pseudo-evaluate the given integer expression, estimating the 4497/// range of values it might take. 4498/// 4499/// \param MaxWidth - the width to which the value will be truncated 4500static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 4501 E = E->IgnoreParens(); 4502 4503 // Try a full evaluation first. 4504 Expr::EvalResult result; 4505 if (E->EvaluateAsRValue(result, C)) 4506 return GetValueRange(C, result.Val, GetExprType(E), MaxWidth); 4507 4508 // I think we only want to look through implicit casts here; if the 4509 // user has an explicit widening cast, we should treat the value as 4510 // being of the new, wider type. 4511 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 4512 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 4513 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 4514 4515 IntRange OutputTypeRange = IntRange::forValueOfType(C, GetExprType(CE)); 4516 4517 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 4518 4519 // Assume that non-integer casts can span the full range of the type. 4520 if (!isIntegerCast) 4521 return OutputTypeRange; 4522 4523 IntRange SubRange 4524 = GetExprRange(C, CE->getSubExpr(), 4525 std::min(MaxWidth, OutputTypeRange.Width)); 4526 4527 // Bail out if the subexpr's range is as wide as the cast type. 4528 if (SubRange.Width >= OutputTypeRange.Width) 4529 return OutputTypeRange; 4530 4531 // Otherwise, we take the smaller width, and we're non-negative if 4532 // either the output type or the subexpr is. 4533 return IntRange(SubRange.Width, 4534 SubRange.NonNegative || OutputTypeRange.NonNegative); 4535 } 4536 4537 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 4538 // If we can fold the condition, just take that operand. 4539 bool CondResult; 4540 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 4541 return GetExprRange(C, CondResult ? CO->getTrueExpr() 4542 : CO->getFalseExpr(), 4543 MaxWidth); 4544 4545 // Otherwise, conservatively merge. 4546 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 4547 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 4548 return IntRange::join(L, R); 4549 } 4550 4551 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4552 switch (BO->getOpcode()) { 4553 4554 // Boolean-valued operations are single-bit and positive. 4555 case BO_LAnd: 4556 case BO_LOr: 4557 case BO_LT: 4558 case BO_GT: 4559 case BO_LE: 4560 case BO_GE: 4561 case BO_EQ: 4562 case BO_NE: 4563 return IntRange::forBoolType(); 4564 4565 // The type of the assignments is the type of the LHS, so the RHS 4566 // is not necessarily the same type. 4567 case BO_MulAssign: 4568 case BO_DivAssign: 4569 case BO_RemAssign: 4570 case BO_AddAssign: 4571 case BO_SubAssign: 4572 case BO_XorAssign: 4573 case BO_OrAssign: 4574 // TODO: bitfields? 4575 return IntRange::forValueOfType(C, GetExprType(E)); 4576 4577 // Simple assignments just pass through the RHS, which will have 4578 // been coerced to the LHS type. 4579 case BO_Assign: 4580 // TODO: bitfields? 4581 return GetExprRange(C, BO->getRHS(), MaxWidth); 4582 4583 // Operations with opaque sources are black-listed. 4584 case BO_PtrMemD: 4585 case BO_PtrMemI: 4586 return IntRange::forValueOfType(C, GetExprType(E)); 4587 4588 // Bitwise-and uses the *infinum* of the two source ranges. 4589 case BO_And: 4590 case BO_AndAssign: 4591 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 4592 GetExprRange(C, BO->getRHS(), MaxWidth)); 4593 4594 // Left shift gets black-listed based on a judgement call. 4595 case BO_Shl: 4596 // ...except that we want to treat '1 << (blah)' as logically 4597 // positive. It's an important idiom. 4598 if (IntegerLiteral *I 4599 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 4600 if (I->getValue() == 1) { 4601 IntRange R = IntRange::forValueOfType(C, GetExprType(E)); 4602 return IntRange(R.Width, /*NonNegative*/ true); 4603 } 4604 } 4605 // fallthrough 4606 4607 case BO_ShlAssign: 4608 return IntRange::forValueOfType(C, GetExprType(E)); 4609 4610 // Right shift by a constant can narrow its left argument. 4611 case BO_Shr: 4612 case BO_ShrAssign: { 4613 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4614 4615 // If the shift amount is a positive constant, drop the width by 4616 // that much. 4617 llvm::APSInt shift; 4618 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 4619 shift.isNonNegative()) { 4620 unsigned zext = shift.getZExtValue(); 4621 if (zext >= L.Width) 4622 L.Width = (L.NonNegative ? 0 : 1); 4623 else 4624 L.Width -= zext; 4625 } 4626 4627 return L; 4628 } 4629 4630 // Comma acts as its right operand. 4631 case BO_Comma: 4632 return GetExprRange(C, BO->getRHS(), MaxWidth); 4633 4634 // Black-list pointer subtractions. 4635 case BO_Sub: 4636 if (BO->getLHS()->getType()->isPointerType()) 4637 return IntRange::forValueOfType(C, GetExprType(E)); 4638 break; 4639 4640 // The width of a division result is mostly determined by the size 4641 // of the LHS. 4642 case BO_Div: { 4643 // Don't 'pre-truncate' the operands. 4644 unsigned opWidth = C.getIntWidth(GetExprType(E)); 4645 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4646 4647 // If the divisor is constant, use that. 4648 llvm::APSInt divisor; 4649 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 4650 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 4651 if (log2 >= L.Width) 4652 L.Width = (L.NonNegative ? 0 : 1); 4653 else 4654 L.Width = std::min(L.Width - log2, MaxWidth); 4655 return L; 4656 } 4657 4658 // Otherwise, just use the LHS's width. 4659 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4660 return IntRange(L.Width, L.NonNegative && R.NonNegative); 4661 } 4662 4663 // The result of a remainder can't be larger than the result of 4664 // either side. 4665 case BO_Rem: { 4666 // Don't 'pre-truncate' the operands. 4667 unsigned opWidth = C.getIntWidth(GetExprType(E)); 4668 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4669 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4670 4671 IntRange meet = IntRange::meet(L, R); 4672 meet.Width = std::min(meet.Width, MaxWidth); 4673 return meet; 4674 } 4675 4676 // The default behavior is okay for these. 4677 case BO_Mul: 4678 case BO_Add: 4679 case BO_Xor: 4680 case BO_Or: 4681 break; 4682 } 4683 4684 // The default case is to treat the operation as if it were closed 4685 // on the narrowest type that encompasses both operands. 4686 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4687 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 4688 return IntRange::join(L, R); 4689 } 4690 4691 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 4692 switch (UO->getOpcode()) { 4693 // Boolean-valued operations are white-listed. 4694 case UO_LNot: 4695 return IntRange::forBoolType(); 4696 4697 // Operations with opaque sources are black-listed. 4698 case UO_Deref: 4699 case UO_AddrOf: // should be impossible 4700 return IntRange::forValueOfType(C, GetExprType(E)); 4701 4702 default: 4703 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 4704 } 4705 } 4706 4707 if (OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E)) 4708 return GetExprRange(C, OVE->getSourceExpr(), MaxWidth); 4709 4710 if (FieldDecl *BitField = E->getSourceBitField()) 4711 return IntRange(BitField->getBitWidthValue(C), 4712 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 4713 4714 return IntRange::forValueOfType(C, GetExprType(E)); 4715} 4716 4717static IntRange GetExprRange(ASTContext &C, Expr *E) { 4718 return GetExprRange(C, E, C.getIntWidth(GetExprType(E))); 4719} 4720 4721/// Checks whether the given value, which currently has the given 4722/// source semantics, has the same value when coerced through the 4723/// target semantics. 4724static bool IsSameFloatAfterCast(const llvm::APFloat &value, 4725 const llvm::fltSemantics &Src, 4726 const llvm::fltSemantics &Tgt) { 4727 llvm::APFloat truncated = value; 4728 4729 bool ignored; 4730 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 4731 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 4732 4733 return truncated.bitwiseIsEqual(value); 4734} 4735 4736/// Checks whether the given value, which currently has the given 4737/// source semantics, has the same value when coerced through the 4738/// target semantics. 4739/// 4740/// The value might be a vector of floats (or a complex number). 4741static bool IsSameFloatAfterCast(const APValue &value, 4742 const llvm::fltSemantics &Src, 4743 const llvm::fltSemantics &Tgt) { 4744 if (value.isFloat()) 4745 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 4746 4747 if (value.isVector()) { 4748 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 4749 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 4750 return false; 4751 return true; 4752 } 4753 4754 assert(value.isComplexFloat()); 4755 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 4756 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 4757} 4758 4759static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 4760 4761static bool IsZero(Sema &S, Expr *E) { 4762 // Suppress cases where we are comparing against an enum constant. 4763 if (const DeclRefExpr *DR = 4764 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 4765 if (isa<EnumConstantDecl>(DR->getDecl())) 4766 return false; 4767 4768 // Suppress cases where the '0' value is expanded from a macro. 4769 if (E->getLocStart().isMacroID()) 4770 return false; 4771 4772 llvm::APSInt Value; 4773 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 4774} 4775 4776static bool HasEnumType(Expr *E) { 4777 // Strip off implicit integral promotions. 4778 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 4779 if (ICE->getCastKind() != CK_IntegralCast && 4780 ICE->getCastKind() != CK_NoOp) 4781 break; 4782 E = ICE->getSubExpr(); 4783 } 4784 4785 return E->getType()->isEnumeralType(); 4786} 4787 4788static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 4789 // Disable warning in template instantiations. 4790 if (!S.ActiveTemplateInstantiations.empty()) 4791 return; 4792 4793 BinaryOperatorKind op = E->getOpcode(); 4794 if (E->isValueDependent()) 4795 return; 4796 4797 if (op == BO_LT && IsZero(S, E->getRHS())) { 4798 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4799 << "< 0" << "false" << HasEnumType(E->getLHS()) 4800 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4801 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 4802 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4803 << ">= 0" << "true" << HasEnumType(E->getLHS()) 4804 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4805 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 4806 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4807 << "0 >" << "false" << HasEnumType(E->getRHS()) 4808 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4809 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 4810 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4811 << "0 <=" << "true" << HasEnumType(E->getRHS()) 4812 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4813 } 4814} 4815 4816static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E, 4817 Expr *Constant, Expr *Other, 4818 llvm::APSInt Value, 4819 bool RhsConstant) { 4820 // Disable warning in template instantiations. 4821 if (!S.ActiveTemplateInstantiations.empty()) 4822 return; 4823 4824 // 0 values are handled later by CheckTrivialUnsignedComparison(). 4825 if (Value == 0) 4826 return; 4827 4828 BinaryOperatorKind op = E->getOpcode(); 4829 QualType OtherT = Other->getType(); 4830 QualType ConstantT = Constant->getType(); 4831 QualType CommonT = E->getLHS()->getType(); 4832 if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT)) 4833 return; 4834 assert((OtherT->isIntegerType() && ConstantT->isIntegerType()) 4835 && "comparison with non-integer type"); 4836 4837 bool ConstantSigned = ConstantT->isSignedIntegerType(); 4838 bool CommonSigned = CommonT->isSignedIntegerType(); 4839 4840 bool EqualityOnly = false; 4841 4842 // TODO: Investigate using GetExprRange() to get tighter bounds on 4843 // on the bit ranges. 4844 IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT); 4845 unsigned OtherWidth = OtherRange.Width; 4846 4847 if (CommonSigned) { 4848 // The common type is signed, therefore no signed to unsigned conversion. 4849 if (!OtherRange.NonNegative) { 4850 // Check that the constant is representable in type OtherT. 4851 if (ConstantSigned) { 4852 if (OtherWidth >= Value.getMinSignedBits()) 4853 return; 4854 } else { // !ConstantSigned 4855 if (OtherWidth >= Value.getActiveBits() + 1) 4856 return; 4857 } 4858 } else { // !OtherSigned 4859 // Check that the constant is representable in type OtherT. 4860 // Negative values are out of range. 4861 if (ConstantSigned) { 4862 if (Value.isNonNegative() && OtherWidth >= Value.getActiveBits()) 4863 return; 4864 } else { // !ConstantSigned 4865 if (OtherWidth >= Value.getActiveBits()) 4866 return; 4867 } 4868 } 4869 } else { // !CommonSigned 4870 if (OtherRange.NonNegative) { 4871 if (OtherWidth >= Value.getActiveBits()) 4872 return; 4873 } else if (!OtherRange.NonNegative && !ConstantSigned) { 4874 // Check to see if the constant is representable in OtherT. 4875 if (OtherWidth > Value.getActiveBits()) 4876 return; 4877 // Check to see if the constant is equivalent to a negative value 4878 // cast to CommonT. 4879 if (S.Context.getIntWidth(ConstantT) == S.Context.getIntWidth(CommonT) && 4880 Value.isNegative() && Value.getMinSignedBits() <= OtherWidth) 4881 return; 4882 // The constant value rests between values that OtherT can represent after 4883 // conversion. Relational comparison still works, but equality 4884 // comparisons will be tautological. 4885 EqualityOnly = true; 4886 } else { // OtherSigned && ConstantSigned 4887 assert(0 && "Two signed types converted to unsigned types."); 4888 } 4889 } 4890 4891 bool PositiveConstant = !ConstantSigned || Value.isNonNegative(); 4892 4893 bool IsTrue = true; 4894 if (op == BO_EQ || op == BO_NE) { 4895 IsTrue = op == BO_NE; 4896 } else if (EqualityOnly) { 4897 return; 4898 } else if (RhsConstant) { 4899 if (op == BO_GT || op == BO_GE) 4900 IsTrue = !PositiveConstant; 4901 else // op == BO_LT || op == BO_LE 4902 IsTrue = PositiveConstant; 4903 } else { 4904 if (op == BO_LT || op == BO_LE) 4905 IsTrue = !PositiveConstant; 4906 else // op == BO_GT || op == BO_GE 4907 IsTrue = PositiveConstant; 4908 } 4909 4910 // If this is a comparison to an enum constant, include that 4911 // constant in the diagnostic. 4912 const EnumConstantDecl *ED = 0; 4913 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Constant)) 4914 ED = dyn_cast<EnumConstantDecl>(DR->getDecl()); 4915 4916 SmallString<64> PrettySourceValue; 4917 llvm::raw_svector_ostream OS(PrettySourceValue); 4918 if (ED) 4919 OS << '\'' << *ED << "' (" << Value << ")"; 4920 else 4921 OS << Value; 4922 4923 S.Diag(E->getOperatorLoc(), diag::warn_out_of_range_compare) 4924 << OS.str() << OtherT << IsTrue 4925 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4926} 4927 4928/// Analyze the operands of the given comparison. Implements the 4929/// fallback case from AnalyzeComparison. 4930static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 4931 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4932 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4933} 4934 4935/// \brief Implements -Wsign-compare. 4936/// 4937/// \param E the binary operator to check for warnings 4938static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 4939 // The type the comparison is being performed in. 4940 QualType T = E->getLHS()->getType(); 4941 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 4942 && "comparison with mismatched types"); 4943 if (E->isValueDependent()) 4944 return AnalyzeImpConvsInComparison(S, E); 4945 4946 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 4947 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 4948 4949 bool IsComparisonConstant = false; 4950 4951 // Check whether an integer constant comparison results in a value 4952 // of 'true' or 'false'. 4953 if (T->isIntegralType(S.Context)) { 4954 llvm::APSInt RHSValue; 4955 bool IsRHSIntegralLiteral = 4956 RHS->isIntegerConstantExpr(RHSValue, S.Context); 4957 llvm::APSInt LHSValue; 4958 bool IsLHSIntegralLiteral = 4959 LHS->isIntegerConstantExpr(LHSValue, S.Context); 4960 if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral) 4961 DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true); 4962 else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral) 4963 DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false); 4964 else 4965 IsComparisonConstant = 4966 (IsRHSIntegralLiteral && IsLHSIntegralLiteral); 4967 } else if (!T->hasUnsignedIntegerRepresentation()) 4968 IsComparisonConstant = E->isIntegerConstantExpr(S.Context); 4969 4970 // We don't do anything special if this isn't an unsigned integral 4971 // comparison: we're only interested in integral comparisons, and 4972 // signed comparisons only happen in cases we don't care to warn about. 4973 // 4974 // We also don't care about value-dependent expressions or expressions 4975 // whose result is a constant. 4976 if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant) 4977 return AnalyzeImpConvsInComparison(S, E); 4978 4979 // Check to see if one of the (unmodified) operands is of different 4980 // signedness. 4981 Expr *signedOperand, *unsignedOperand; 4982 if (LHS->getType()->hasSignedIntegerRepresentation()) { 4983 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 4984 "unsigned comparison between two signed integer expressions?"); 4985 signedOperand = LHS; 4986 unsignedOperand = RHS; 4987 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 4988 signedOperand = RHS; 4989 unsignedOperand = LHS; 4990 } else { 4991 CheckTrivialUnsignedComparison(S, E); 4992 return AnalyzeImpConvsInComparison(S, E); 4993 } 4994 4995 // Otherwise, calculate the effective range of the signed operand. 4996 IntRange signedRange = GetExprRange(S.Context, signedOperand); 4997 4998 // Go ahead and analyze implicit conversions in the operands. Note 4999 // that we skip the implicit conversions on both sides. 5000 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 5001 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 5002 5003 // If the signed range is non-negative, -Wsign-compare won't fire, 5004 // but we should still check for comparisons which are always true 5005 // or false. 5006 if (signedRange.NonNegative) 5007 return CheckTrivialUnsignedComparison(S, E); 5008 5009 // For (in)equality comparisons, if the unsigned operand is a 5010 // constant which cannot collide with a overflowed signed operand, 5011 // then reinterpreting the signed operand as unsigned will not 5012 // change the result of the comparison. 5013 if (E->isEqualityOp()) { 5014 unsigned comparisonWidth = S.Context.getIntWidth(T); 5015 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 5016 5017 // We should never be unable to prove that the unsigned operand is 5018 // non-negative. 5019 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 5020 5021 if (unsignedRange.Width < comparisonWidth) 5022 return; 5023 } 5024 5025 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 5026 S.PDiag(diag::warn_mixed_sign_comparison) 5027 << LHS->getType() << RHS->getType() 5028 << LHS->getSourceRange() << RHS->getSourceRange()); 5029} 5030 5031/// Analyzes an attempt to assign the given value to a bitfield. 5032/// 5033/// Returns true if there was something fishy about the attempt. 5034static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 5035 SourceLocation InitLoc) { 5036 assert(Bitfield->isBitField()); 5037 if (Bitfield->isInvalidDecl()) 5038 return false; 5039 5040 // White-list bool bitfields. 5041 if (Bitfield->getType()->isBooleanType()) 5042 return false; 5043 5044 // Ignore value- or type-dependent expressions. 5045 if (Bitfield->getBitWidth()->isValueDependent() || 5046 Bitfield->getBitWidth()->isTypeDependent() || 5047 Init->isValueDependent() || 5048 Init->isTypeDependent()) 5049 return false; 5050 5051 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 5052 5053 llvm::APSInt Value; 5054 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 5055 return false; 5056 5057 unsigned OriginalWidth = Value.getBitWidth(); 5058 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 5059 5060 if (OriginalWidth <= FieldWidth) 5061 return false; 5062 5063 // Compute the value which the bitfield will contain. 5064 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 5065 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 5066 5067 // Check whether the stored value is equal to the original value. 5068 TruncatedValue = TruncatedValue.extend(OriginalWidth); 5069 if (llvm::APSInt::isSameValue(Value, TruncatedValue)) 5070 return false; 5071 5072 // Special-case bitfields of width 1: booleans are naturally 0/1, and 5073 // therefore don't strictly fit into a signed bitfield of width 1. 5074 if (FieldWidth == 1 && Value == 1) 5075 return false; 5076 5077 std::string PrettyValue = Value.toString(10); 5078 std::string PrettyTrunc = TruncatedValue.toString(10); 5079 5080 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 5081 << PrettyValue << PrettyTrunc << OriginalInit->getType() 5082 << Init->getSourceRange(); 5083 5084 return true; 5085} 5086 5087/// Analyze the given simple or compound assignment for warning-worthy 5088/// operations. 5089static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 5090 // Just recurse on the LHS. 5091 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 5092 5093 // We want to recurse on the RHS as normal unless we're assigning to 5094 // a bitfield. 5095 if (FieldDecl *Bitfield = E->getLHS()->getSourceBitField()) { 5096 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 5097 E->getOperatorLoc())) { 5098 // Recurse, ignoring any implicit conversions on the RHS. 5099 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 5100 E->getOperatorLoc()); 5101 } 5102 } 5103 5104 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 5105} 5106 5107/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 5108static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 5109 SourceLocation CContext, unsigned diag, 5110 bool pruneControlFlow = false) { 5111 if (pruneControlFlow) { 5112 S.DiagRuntimeBehavior(E->getExprLoc(), E, 5113 S.PDiag(diag) 5114 << SourceType << T << E->getSourceRange() 5115 << SourceRange(CContext)); 5116 return; 5117 } 5118 S.Diag(E->getExprLoc(), diag) 5119 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 5120} 5121 5122/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 5123static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 5124 SourceLocation CContext, unsigned diag, 5125 bool pruneControlFlow = false) { 5126 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 5127} 5128 5129/// Diagnose an implicit cast from a literal expression. Does not warn when the 5130/// cast wouldn't lose information. 5131void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 5132 SourceLocation CContext) { 5133 // Try to convert the literal exactly to an integer. If we can, don't warn. 5134 bool isExact = false; 5135 const llvm::APFloat &Value = FL->getValue(); 5136 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 5137 T->hasUnsignedIntegerRepresentation()); 5138 if (Value.convertToInteger(IntegerValue, 5139 llvm::APFloat::rmTowardZero, &isExact) 5140 == llvm::APFloat::opOK && isExact) 5141 return; 5142 5143 // FIXME: Force the precision of the source value down so we don't print 5144 // digits which are usually useless (we don't really care here if we 5145 // truncate a digit by accident in edge cases). Ideally, APFloat::toString 5146 // would automatically print the shortest representation, but it's a bit 5147 // tricky to implement. 5148 SmallString<16> PrettySourceValue; 5149 unsigned precision = llvm::APFloat::semanticsPrecision(Value.getSemantics()); 5150 precision = (precision * 59 + 195) / 196; 5151 Value.toString(PrettySourceValue, precision); 5152 5153 SmallString<16> PrettyTargetValue; 5154 if (T->isSpecificBuiltinType(BuiltinType::Bool)) 5155 PrettyTargetValue = IntegerValue == 0 ? "false" : "true"; 5156 else 5157 IntegerValue.toString(PrettyTargetValue); 5158 5159 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 5160 << FL->getType() << T.getUnqualifiedType() << PrettySourceValue 5161 << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext); 5162} 5163 5164std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 5165 if (!Range.Width) return "0"; 5166 5167 llvm::APSInt ValueInRange = Value; 5168 ValueInRange.setIsSigned(!Range.NonNegative); 5169 ValueInRange = ValueInRange.trunc(Range.Width); 5170 return ValueInRange.toString(10); 5171} 5172 5173static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) { 5174 if (!isa<ImplicitCastExpr>(Ex)) 5175 return false; 5176 5177 Expr *InnerE = Ex->IgnoreParenImpCasts(); 5178 const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr(); 5179 const Type *Source = 5180 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 5181 if (Target->isDependentType()) 5182 return false; 5183 5184 const BuiltinType *FloatCandidateBT = 5185 dyn_cast<BuiltinType>(ToBool ? Source : Target); 5186 const Type *BoolCandidateType = ToBool ? Target : Source; 5187 5188 return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) && 5189 FloatCandidateBT && (FloatCandidateBT->isFloatingPoint())); 5190} 5191 5192void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall, 5193 SourceLocation CC) { 5194 unsigned NumArgs = TheCall->getNumArgs(); 5195 for (unsigned i = 0; i < NumArgs; ++i) { 5196 Expr *CurrA = TheCall->getArg(i); 5197 if (!IsImplicitBoolFloatConversion(S, CurrA, true)) 5198 continue; 5199 5200 bool IsSwapped = ((i > 0) && 5201 IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false)); 5202 IsSwapped |= ((i < (NumArgs - 1)) && 5203 IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false)); 5204 if (IsSwapped) { 5205 // Warn on this floating-point to bool conversion. 5206 DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(), 5207 CurrA->getType(), CC, 5208 diag::warn_impcast_floating_point_to_bool); 5209 } 5210 } 5211} 5212 5213void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 5214 SourceLocation CC, bool *ICContext = 0) { 5215 if (E->isTypeDependent() || E->isValueDependent()) return; 5216 5217 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 5218 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 5219 if (Source == Target) return; 5220 if (Target->isDependentType()) return; 5221 5222 // If the conversion context location is invalid don't complain. We also 5223 // don't want to emit a warning if the issue occurs from the expansion of 5224 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 5225 // delay this check as long as possible. Once we detect we are in that 5226 // scenario, we just return. 5227 if (CC.isInvalid()) 5228 return; 5229 5230 // Diagnose implicit casts to bool. 5231 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 5232 if (isa<StringLiteral>(E)) 5233 // Warn on string literal to bool. Checks for string literals in logical 5234 // expressions, for instances, assert(0 && "error here"), is prevented 5235 // by a check in AnalyzeImplicitConversions(). 5236 return DiagnoseImpCast(S, E, T, CC, 5237 diag::warn_impcast_string_literal_to_bool); 5238 if (Source->isFunctionType()) { 5239 // Warn on function to bool. Checks free functions and static member 5240 // functions. Weakly imported functions are excluded from the check, 5241 // since it's common to test their value to check whether the linker 5242 // found a definition for them. 5243 ValueDecl *D = 0; 5244 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 5245 D = R->getDecl(); 5246 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 5247 D = M->getMemberDecl(); 5248 } 5249 5250 if (D && !D->isWeak()) { 5251 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 5252 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 5253 << F << E->getSourceRange() << SourceRange(CC); 5254 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 5255 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 5256 QualType ReturnType; 5257 UnresolvedSet<4> NonTemplateOverloads; 5258 S.tryExprAsCall(*E, ReturnType, NonTemplateOverloads); 5259 if (!ReturnType.isNull() 5260 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 5261 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 5262 << FixItHint::CreateInsertion( 5263 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 5264 return; 5265 } 5266 } 5267 } 5268 } 5269 5270 // Strip vector types. 5271 if (isa<VectorType>(Source)) { 5272 if (!isa<VectorType>(Target)) { 5273 if (S.SourceMgr.isInSystemMacro(CC)) 5274 return; 5275 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 5276 } 5277 5278 // If the vector cast is cast between two vectors of the same size, it is 5279 // a bitcast, not a conversion. 5280 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 5281 return; 5282 5283 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 5284 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 5285 } 5286 5287 // Strip complex types. 5288 if (isa<ComplexType>(Source)) { 5289 if (!isa<ComplexType>(Target)) { 5290 if (S.SourceMgr.isInSystemMacro(CC)) 5291 return; 5292 5293 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 5294 } 5295 5296 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 5297 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 5298 } 5299 5300 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 5301 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 5302 5303 // If the source is floating point... 5304 if (SourceBT && SourceBT->isFloatingPoint()) { 5305 // ...and the target is floating point... 5306 if (TargetBT && TargetBT->isFloatingPoint()) { 5307 // ...then warn if we're dropping FP rank. 5308 5309 // Builtin FP kinds are ordered by increasing FP rank. 5310 if (SourceBT->getKind() > TargetBT->getKind()) { 5311 // Don't warn about float constants that are precisely 5312 // representable in the target type. 5313 Expr::EvalResult result; 5314 if (E->EvaluateAsRValue(result, S.Context)) { 5315 // Value might be a float, a float vector, or a float complex. 5316 if (IsSameFloatAfterCast(result.Val, 5317 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 5318 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 5319 return; 5320 } 5321 5322 if (S.SourceMgr.isInSystemMacro(CC)) 5323 return; 5324 5325 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 5326 } 5327 return; 5328 } 5329 5330 // If the target is integral, always warn. 5331 if (TargetBT && TargetBT->isInteger()) { 5332 if (S.SourceMgr.isInSystemMacro(CC)) 5333 return; 5334 5335 Expr *InnerE = E->IgnoreParenImpCasts(); 5336 // We also want to warn on, e.g., "int i = -1.234" 5337 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 5338 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 5339 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 5340 5341 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 5342 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 5343 } else { 5344 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 5345 } 5346 } 5347 5348 // If the target is bool, warn if expr is a function or method call. 5349 if (Target->isSpecificBuiltinType(BuiltinType::Bool) && 5350 isa<CallExpr>(E)) { 5351 // Check last argument of function call to see if it is an 5352 // implicit cast from a type matching the type the result 5353 // is being cast to. 5354 CallExpr *CEx = cast<CallExpr>(E); 5355 unsigned NumArgs = CEx->getNumArgs(); 5356 if (NumArgs > 0) { 5357 Expr *LastA = CEx->getArg(NumArgs - 1); 5358 Expr *InnerE = LastA->IgnoreParenImpCasts(); 5359 const Type *InnerType = 5360 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 5361 if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) { 5362 // Warn on this floating-point to bool conversion 5363 DiagnoseImpCast(S, E, T, CC, 5364 diag::warn_impcast_floating_point_to_bool); 5365 } 5366 } 5367 } 5368 return; 5369 } 5370 5371 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 5372 == Expr::NPCK_GNUNull) && !Target->isAnyPointerType() 5373 && !Target->isBlockPointerType() && !Target->isMemberPointerType() 5374 && Target->isScalarType() && !Target->isNullPtrType()) { 5375 SourceLocation Loc = E->getSourceRange().getBegin(); 5376 if (Loc.isMacroID()) 5377 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 5378 if (!Loc.isMacroID() || CC.isMacroID()) 5379 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 5380 << T << clang::SourceRange(CC) 5381 << FixItHint::CreateReplacement(Loc, 5382 S.getFixItZeroLiteralForType(T, Loc)); 5383 } 5384 5385 if (!Source->isIntegerType() || !Target->isIntegerType()) 5386 return; 5387 5388 // TODO: remove this early return once the false positives for constant->bool 5389 // in templates, macros, etc, are reduced or removed. 5390 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 5391 return; 5392 5393 IntRange SourceRange = GetExprRange(S.Context, E); 5394 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 5395 5396 if (SourceRange.Width > TargetRange.Width) { 5397 // If the source is a constant, use a default-on diagnostic. 5398 // TODO: this should happen for bitfield stores, too. 5399 llvm::APSInt Value(32); 5400 if (E->isIntegerConstantExpr(Value, S.Context)) { 5401 if (S.SourceMgr.isInSystemMacro(CC)) 5402 return; 5403 5404 std::string PrettySourceValue = Value.toString(10); 5405 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 5406 5407 S.DiagRuntimeBehavior(E->getExprLoc(), E, 5408 S.PDiag(diag::warn_impcast_integer_precision_constant) 5409 << PrettySourceValue << PrettyTargetValue 5410 << E->getType() << T << E->getSourceRange() 5411 << clang::SourceRange(CC)); 5412 return; 5413 } 5414 5415 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 5416 if (S.SourceMgr.isInSystemMacro(CC)) 5417 return; 5418 5419 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 5420 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 5421 /* pruneControlFlow */ true); 5422 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 5423 } 5424 5425 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 5426 (!TargetRange.NonNegative && SourceRange.NonNegative && 5427 SourceRange.Width == TargetRange.Width)) { 5428 5429 if (S.SourceMgr.isInSystemMacro(CC)) 5430 return; 5431 5432 unsigned DiagID = diag::warn_impcast_integer_sign; 5433 5434 // Traditionally, gcc has warned about this under -Wsign-compare. 5435 // We also want to warn about it in -Wconversion. 5436 // So if -Wconversion is off, use a completely identical diagnostic 5437 // in the sign-compare group. 5438 // The conditional-checking code will 5439 if (ICContext) { 5440 DiagID = diag::warn_impcast_integer_sign_conditional; 5441 *ICContext = true; 5442 } 5443 5444 return DiagnoseImpCast(S, E, T, CC, DiagID); 5445 } 5446 5447 // Diagnose conversions between different enumeration types. 5448 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 5449 // type, to give us better diagnostics. 5450 QualType SourceType = E->getType(); 5451 if (!S.getLangOpts().CPlusPlus) { 5452 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5453 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 5454 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 5455 SourceType = S.Context.getTypeDeclType(Enum); 5456 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 5457 } 5458 } 5459 5460 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 5461 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 5462 if (SourceEnum->getDecl()->hasNameForLinkage() && 5463 TargetEnum->getDecl()->hasNameForLinkage() && 5464 SourceEnum != TargetEnum) { 5465 if (S.SourceMgr.isInSystemMacro(CC)) 5466 return; 5467 5468 return DiagnoseImpCast(S, E, SourceType, T, CC, 5469 diag::warn_impcast_different_enum_types); 5470 } 5471 5472 return; 5473} 5474 5475void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5476 SourceLocation CC, QualType T); 5477 5478void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 5479 SourceLocation CC, bool &ICContext) { 5480 E = E->IgnoreParenImpCasts(); 5481 5482 if (isa<ConditionalOperator>(E)) 5483 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T); 5484 5485 AnalyzeImplicitConversions(S, E, CC); 5486 if (E->getType() != T) 5487 return CheckImplicitConversion(S, E, T, CC, &ICContext); 5488 return; 5489} 5490 5491void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5492 SourceLocation CC, QualType T) { 5493 AnalyzeImplicitConversions(S, E->getCond(), CC); 5494 5495 bool Suspicious = false; 5496 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 5497 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 5498 5499 // If -Wconversion would have warned about either of the candidates 5500 // for a signedness conversion to the context type... 5501 if (!Suspicious) return; 5502 5503 // ...but it's currently ignored... 5504 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 5505 CC)) 5506 return; 5507 5508 // ...then check whether it would have warned about either of the 5509 // candidates for a signedness conversion to the condition type. 5510 if (E->getType() == T) return; 5511 5512 Suspicious = false; 5513 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 5514 E->getType(), CC, &Suspicious); 5515 if (!Suspicious) 5516 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 5517 E->getType(), CC, &Suspicious); 5518} 5519 5520/// AnalyzeImplicitConversions - Find and report any interesting 5521/// implicit conversions in the given expression. There are a couple 5522/// of competing diagnostics here, -Wconversion and -Wsign-compare. 5523void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 5524 QualType T = OrigE->getType(); 5525 Expr *E = OrigE->IgnoreParenImpCasts(); 5526 5527 if (E->isTypeDependent() || E->isValueDependent()) 5528 return; 5529 5530 // For conditional operators, we analyze the arguments as if they 5531 // were being fed directly into the output. 5532 if (isa<ConditionalOperator>(E)) { 5533 ConditionalOperator *CO = cast<ConditionalOperator>(E); 5534 CheckConditionalOperator(S, CO, CC, T); 5535 return; 5536 } 5537 5538 // Check implicit argument conversions for function calls. 5539 if (CallExpr *Call = dyn_cast<CallExpr>(E)) 5540 CheckImplicitArgumentConversions(S, Call, CC); 5541 5542 // Go ahead and check any implicit conversions we might have skipped. 5543 // The non-canonical typecheck is just an optimization; 5544 // CheckImplicitConversion will filter out dead implicit conversions. 5545 if (E->getType() != T) 5546 CheckImplicitConversion(S, E, T, CC); 5547 5548 // Now continue drilling into this expression. 5549 5550 if (PseudoObjectExpr * POE = dyn_cast<PseudoObjectExpr>(E)) { 5551 if (POE->getResultExpr()) 5552 E = POE->getResultExpr(); 5553 } 5554 5555 if (const OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E)) 5556 return AnalyzeImplicitConversions(S, OVE->getSourceExpr(), CC); 5557 5558 // Skip past explicit casts. 5559 if (isa<ExplicitCastExpr>(E)) { 5560 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 5561 return AnalyzeImplicitConversions(S, E, CC); 5562 } 5563 5564 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5565 // Do a somewhat different check with comparison operators. 5566 if (BO->isComparisonOp()) 5567 return AnalyzeComparison(S, BO); 5568 5569 // And with simple assignments. 5570 if (BO->getOpcode() == BO_Assign) 5571 return AnalyzeAssignment(S, BO); 5572 } 5573 5574 // These break the otherwise-useful invariant below. Fortunately, 5575 // we don't really need to recurse into them, because any internal 5576 // expressions should have been analyzed already when they were 5577 // built into statements. 5578 if (isa<StmtExpr>(E)) return; 5579 5580 // Don't descend into unevaluated contexts. 5581 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 5582 5583 // Now just recurse over the expression's children. 5584 CC = E->getExprLoc(); 5585 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 5586 bool IsLogicalOperator = BO && BO->isLogicalOp(); 5587 for (Stmt::child_range I = E->children(); I; ++I) { 5588 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 5589 if (!ChildExpr) 5590 continue; 5591 5592 if (IsLogicalOperator && 5593 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 5594 // Ignore checking string literals that are in logical operators. 5595 continue; 5596 AnalyzeImplicitConversions(S, ChildExpr, CC); 5597 } 5598} 5599 5600} // end anonymous namespace 5601 5602/// Diagnoses "dangerous" implicit conversions within the given 5603/// expression (which is a full expression). Implements -Wconversion 5604/// and -Wsign-compare. 5605/// 5606/// \param CC the "context" location of the implicit conversion, i.e. 5607/// the most location of the syntactic entity requiring the implicit 5608/// conversion 5609void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 5610 // Don't diagnose in unevaluated contexts. 5611 if (isUnevaluatedContext()) 5612 return; 5613 5614 // Don't diagnose for value- or type-dependent expressions. 5615 if (E->isTypeDependent() || E->isValueDependent()) 5616 return; 5617 5618 // Check for array bounds violations in cases where the check isn't triggered 5619 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 5620 // ArraySubscriptExpr is on the RHS of a variable initialization. 5621 CheckArrayAccess(E); 5622 5623 // This is not the right CC for (e.g.) a variable initialization. 5624 AnalyzeImplicitConversions(*this, E, CC); 5625} 5626 5627/// Diagnose when expression is an integer constant expression and its evaluation 5628/// results in integer overflow 5629void Sema::CheckForIntOverflow (Expr *E) { 5630 if (isa<BinaryOperator>(E->IgnoreParens())) 5631 E->EvaluateForOverflow(Context); 5632} 5633 5634namespace { 5635/// \brief Visitor for expressions which looks for unsequenced operations on the 5636/// same object. 5637class SequenceChecker : public EvaluatedExprVisitor<SequenceChecker> { 5638 typedef EvaluatedExprVisitor<SequenceChecker> Base; 5639 5640 /// \brief A tree of sequenced regions within an expression. Two regions are 5641 /// unsequenced if one is an ancestor or a descendent of the other. When we 5642 /// finish processing an expression with sequencing, such as a comma 5643 /// expression, we fold its tree nodes into its parent, since they are 5644 /// unsequenced with respect to nodes we will visit later. 5645 class SequenceTree { 5646 struct Value { 5647 explicit Value(unsigned Parent) : Parent(Parent), Merged(false) {} 5648 unsigned Parent : 31; 5649 bool Merged : 1; 5650 }; 5651 SmallVector<Value, 8> Values; 5652 5653 public: 5654 /// \brief A region within an expression which may be sequenced with respect 5655 /// to some other region. 5656 class Seq { 5657 explicit Seq(unsigned N) : Index(N) {} 5658 unsigned Index; 5659 friend class SequenceTree; 5660 public: 5661 Seq() : Index(0) {} 5662 }; 5663 5664 SequenceTree() { Values.push_back(Value(0)); } 5665 Seq root() const { return Seq(0); } 5666 5667 /// \brief Create a new sequence of operations, which is an unsequenced 5668 /// subset of \p Parent. This sequence of operations is sequenced with 5669 /// respect to other children of \p Parent. 5670 Seq allocate(Seq Parent) { 5671 Values.push_back(Value(Parent.Index)); 5672 return Seq(Values.size() - 1); 5673 } 5674 5675 /// \brief Merge a sequence of operations into its parent. 5676 void merge(Seq S) { 5677 Values[S.Index].Merged = true; 5678 } 5679 5680 /// \brief Determine whether two operations are unsequenced. This operation 5681 /// is asymmetric: \p Cur should be the more recent sequence, and \p Old 5682 /// should have been merged into its parent as appropriate. 5683 bool isUnsequenced(Seq Cur, Seq Old) { 5684 unsigned C = representative(Cur.Index); 5685 unsigned Target = representative(Old.Index); 5686 while (C >= Target) { 5687 if (C == Target) 5688 return true; 5689 C = Values[C].Parent; 5690 } 5691 return false; 5692 } 5693 5694 private: 5695 /// \brief Pick a representative for a sequence. 5696 unsigned representative(unsigned K) { 5697 if (Values[K].Merged) 5698 // Perform path compression as we go. 5699 return Values[K].Parent = representative(Values[K].Parent); 5700 return K; 5701 } 5702 }; 5703 5704 /// An object for which we can track unsequenced uses. 5705 typedef NamedDecl *Object; 5706 5707 /// Different flavors of object usage which we track. We only track the 5708 /// least-sequenced usage of each kind. 5709 enum UsageKind { 5710 /// A read of an object. Multiple unsequenced reads are OK. 5711 UK_Use, 5712 /// A modification of an object which is sequenced before the value 5713 /// computation of the expression, such as ++n in C++. 5714 UK_ModAsValue, 5715 /// A modification of an object which is not sequenced before the value 5716 /// computation of the expression, such as n++. 5717 UK_ModAsSideEffect, 5718 5719 UK_Count = UK_ModAsSideEffect + 1 5720 }; 5721 5722 struct Usage { 5723 Usage() : Use(0), Seq() {} 5724 Expr *Use; 5725 SequenceTree::Seq Seq; 5726 }; 5727 5728 struct UsageInfo { 5729 UsageInfo() : Diagnosed(false) {} 5730 Usage Uses[UK_Count]; 5731 /// Have we issued a diagnostic for this variable already? 5732 bool Diagnosed; 5733 }; 5734 typedef llvm::SmallDenseMap<Object, UsageInfo, 16> UsageInfoMap; 5735 5736 Sema &SemaRef; 5737 /// Sequenced regions within the expression. 5738 SequenceTree Tree; 5739 /// Declaration modifications and references which we have seen. 5740 UsageInfoMap UsageMap; 5741 /// The region we are currently within. 5742 SequenceTree::Seq Region; 5743 /// Filled in with declarations which were modified as a side-effect 5744 /// (that is, post-increment operations). 5745 SmallVectorImpl<std::pair<Object, Usage> > *ModAsSideEffect; 5746 /// Expressions to check later. We defer checking these to reduce 5747 /// stack usage. 5748 SmallVectorImpl<Expr *> &WorkList; 5749 5750 /// RAII object wrapping the visitation of a sequenced subexpression of an 5751 /// expression. At the end of this process, the side-effects of the evaluation 5752 /// become sequenced with respect to the value computation of the result, so 5753 /// we downgrade any UK_ModAsSideEffect within the evaluation to 5754 /// UK_ModAsValue. 5755 struct SequencedSubexpression { 5756 SequencedSubexpression(SequenceChecker &Self) 5757 : Self(Self), OldModAsSideEffect(Self.ModAsSideEffect) { 5758 Self.ModAsSideEffect = &ModAsSideEffect; 5759 } 5760 ~SequencedSubexpression() { 5761 for (unsigned I = 0, E = ModAsSideEffect.size(); I != E; ++I) { 5762 UsageInfo &U = Self.UsageMap[ModAsSideEffect[I].first]; 5763 U.Uses[UK_ModAsSideEffect] = ModAsSideEffect[I].second; 5764 Self.addUsage(U, ModAsSideEffect[I].first, 5765 ModAsSideEffect[I].second.Use, UK_ModAsValue); 5766 } 5767 Self.ModAsSideEffect = OldModAsSideEffect; 5768 } 5769 5770 SequenceChecker &Self; 5771 SmallVector<std::pair<Object, Usage>, 4> ModAsSideEffect; 5772 SmallVectorImpl<std::pair<Object, Usage> > *OldModAsSideEffect; 5773 }; 5774 5775 /// RAII object wrapping the visitation of a subexpression which we might 5776 /// choose to evaluate as a constant. If any subexpression is evaluated and 5777 /// found to be non-constant, this allows us to suppress the evaluation of 5778 /// the outer expression. 5779 class EvaluationTracker { 5780 public: 5781 EvaluationTracker(SequenceChecker &Self) 5782 : Self(Self), Prev(Self.EvalTracker), EvalOK(true) { 5783 Self.EvalTracker = this; 5784 } 5785 ~EvaluationTracker() { 5786 Self.EvalTracker = Prev; 5787 if (Prev) 5788 Prev->EvalOK &= EvalOK; 5789 } 5790 5791 bool evaluate(const Expr *E, bool &Result) { 5792 if (!EvalOK || E->isValueDependent()) 5793 return false; 5794 EvalOK = E->EvaluateAsBooleanCondition(Result, Self.SemaRef.Context); 5795 return EvalOK; 5796 } 5797 5798 private: 5799 SequenceChecker &Self; 5800 EvaluationTracker *Prev; 5801 bool EvalOK; 5802 } *EvalTracker; 5803 5804 /// \brief Find the object which is produced by the specified expression, 5805 /// if any. 5806 Object getObject(Expr *E, bool Mod) const { 5807 E = E->IgnoreParenCasts(); 5808 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 5809 if (Mod && (UO->getOpcode() == UO_PreInc || UO->getOpcode() == UO_PreDec)) 5810 return getObject(UO->getSubExpr(), Mod); 5811 } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5812 if (BO->getOpcode() == BO_Comma) 5813 return getObject(BO->getRHS(), Mod); 5814 if (Mod && BO->isAssignmentOp()) 5815 return getObject(BO->getLHS(), Mod); 5816 } else if (MemberExpr *ME = dyn_cast<MemberExpr>(E)) { 5817 // FIXME: Check for more interesting cases, like "x.n = ++x.n". 5818 if (isa<CXXThisExpr>(ME->getBase()->IgnoreParenCasts())) 5819 return ME->getMemberDecl(); 5820 } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5821 // FIXME: If this is a reference, map through to its value. 5822 return DRE->getDecl(); 5823 return 0; 5824 } 5825 5826 /// \brief Note that an object was modified or used by an expression. 5827 void addUsage(UsageInfo &UI, Object O, Expr *Ref, UsageKind UK) { 5828 Usage &U = UI.Uses[UK]; 5829 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) { 5830 if (UK == UK_ModAsSideEffect && ModAsSideEffect) 5831 ModAsSideEffect->push_back(std::make_pair(O, U)); 5832 U.Use = Ref; 5833 U.Seq = Region; 5834 } 5835 } 5836 /// \brief Check whether a modification or use conflicts with a prior usage. 5837 void checkUsage(Object O, UsageInfo &UI, Expr *Ref, UsageKind OtherKind, 5838 bool IsModMod) { 5839 if (UI.Diagnosed) 5840 return; 5841 5842 const Usage &U = UI.Uses[OtherKind]; 5843 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) 5844 return; 5845 5846 Expr *Mod = U.Use; 5847 Expr *ModOrUse = Ref; 5848 if (OtherKind == UK_Use) 5849 std::swap(Mod, ModOrUse); 5850 5851 SemaRef.Diag(Mod->getExprLoc(), 5852 IsModMod ? diag::warn_unsequenced_mod_mod 5853 : diag::warn_unsequenced_mod_use) 5854 << O << SourceRange(ModOrUse->getExprLoc()); 5855 UI.Diagnosed = true; 5856 } 5857 5858 void notePreUse(Object O, Expr *Use) { 5859 UsageInfo &U = UsageMap[O]; 5860 // Uses conflict with other modifications. 5861 checkUsage(O, U, Use, UK_ModAsValue, false); 5862 } 5863 void notePostUse(Object O, Expr *Use) { 5864 UsageInfo &U = UsageMap[O]; 5865 checkUsage(O, U, Use, UK_ModAsSideEffect, false); 5866 addUsage(U, O, Use, UK_Use); 5867 } 5868 5869 void notePreMod(Object O, Expr *Mod) { 5870 UsageInfo &U = UsageMap[O]; 5871 // Modifications conflict with other modifications and with uses. 5872 checkUsage(O, U, Mod, UK_ModAsValue, true); 5873 checkUsage(O, U, Mod, UK_Use, false); 5874 } 5875 void notePostMod(Object O, Expr *Use, UsageKind UK) { 5876 UsageInfo &U = UsageMap[O]; 5877 checkUsage(O, U, Use, UK_ModAsSideEffect, true); 5878 addUsage(U, O, Use, UK); 5879 } 5880 5881public: 5882 SequenceChecker(Sema &S, Expr *E, SmallVectorImpl<Expr *> &WorkList) 5883 : Base(S.Context), SemaRef(S), Region(Tree.root()), ModAsSideEffect(0), 5884 WorkList(WorkList), EvalTracker(0) { 5885 Visit(E); 5886 } 5887 5888 void VisitStmt(Stmt *S) { 5889 // Skip all statements which aren't expressions for now. 5890 } 5891 5892 void VisitExpr(Expr *E) { 5893 // By default, just recurse to evaluated subexpressions. 5894 Base::VisitStmt(E); 5895 } 5896 5897 void VisitCastExpr(CastExpr *E) { 5898 Object O = Object(); 5899 if (E->getCastKind() == CK_LValueToRValue) 5900 O = getObject(E->getSubExpr(), false); 5901 5902 if (O) 5903 notePreUse(O, E); 5904 VisitExpr(E); 5905 if (O) 5906 notePostUse(O, E); 5907 } 5908 5909 void VisitBinComma(BinaryOperator *BO) { 5910 // C++11 [expr.comma]p1: 5911 // Every value computation and side effect associated with the left 5912 // expression is sequenced before every value computation and side 5913 // effect associated with the right expression. 5914 SequenceTree::Seq LHS = Tree.allocate(Region); 5915 SequenceTree::Seq RHS = Tree.allocate(Region); 5916 SequenceTree::Seq OldRegion = Region; 5917 5918 { 5919 SequencedSubexpression SeqLHS(*this); 5920 Region = LHS; 5921 Visit(BO->getLHS()); 5922 } 5923 5924 Region = RHS; 5925 Visit(BO->getRHS()); 5926 5927 Region = OldRegion; 5928 5929 // Forget that LHS and RHS are sequenced. They are both unsequenced 5930 // with respect to other stuff. 5931 Tree.merge(LHS); 5932 Tree.merge(RHS); 5933 } 5934 5935 void VisitBinAssign(BinaryOperator *BO) { 5936 // The modification is sequenced after the value computation of the LHS 5937 // and RHS, so check it before inspecting the operands and update the 5938 // map afterwards. 5939 Object O = getObject(BO->getLHS(), true); 5940 if (!O) 5941 return VisitExpr(BO); 5942 5943 notePreMod(O, BO); 5944 5945 // C++11 [expr.ass]p7: 5946 // E1 op= E2 is equivalent to E1 = E1 op E2, except that E1 is evaluated 5947 // only once. 5948 // 5949 // Therefore, for a compound assignment operator, O is considered used 5950 // everywhere except within the evaluation of E1 itself. 5951 if (isa<CompoundAssignOperator>(BO)) 5952 notePreUse(O, BO); 5953 5954 Visit(BO->getLHS()); 5955 5956 if (isa<CompoundAssignOperator>(BO)) 5957 notePostUse(O, BO); 5958 5959 Visit(BO->getRHS()); 5960 5961 // C++11 [expr.ass]p1: 5962 // the assignment is sequenced [...] before the value computation of the 5963 // assignment expression. 5964 // C11 6.5.16/3 has no such rule. 5965 notePostMod(O, BO, SemaRef.getLangOpts().CPlusPlus ? UK_ModAsValue 5966 : UK_ModAsSideEffect); 5967 } 5968 void VisitCompoundAssignOperator(CompoundAssignOperator *CAO) { 5969 VisitBinAssign(CAO); 5970 } 5971 5972 void VisitUnaryPreInc(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 5973 void VisitUnaryPreDec(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 5974 void VisitUnaryPreIncDec(UnaryOperator *UO) { 5975 Object O = getObject(UO->getSubExpr(), true); 5976 if (!O) 5977 return VisitExpr(UO); 5978 5979 notePreMod(O, UO); 5980 Visit(UO->getSubExpr()); 5981 // C++11 [expr.pre.incr]p1: 5982 // the expression ++x is equivalent to x+=1 5983 notePostMod(O, UO, SemaRef.getLangOpts().CPlusPlus ? UK_ModAsValue 5984 : UK_ModAsSideEffect); 5985 } 5986 5987 void VisitUnaryPostInc(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 5988 void VisitUnaryPostDec(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 5989 void VisitUnaryPostIncDec(UnaryOperator *UO) { 5990 Object O = getObject(UO->getSubExpr(), true); 5991 if (!O) 5992 return VisitExpr(UO); 5993 5994 notePreMod(O, UO); 5995 Visit(UO->getSubExpr()); 5996 notePostMod(O, UO, UK_ModAsSideEffect); 5997 } 5998 5999 /// Don't visit the RHS of '&&' or '||' if it might not be evaluated. 6000 void VisitBinLOr(BinaryOperator *BO) { 6001 // The side-effects of the LHS of an '&&' are sequenced before the 6002 // value computation of the RHS, and hence before the value computation 6003 // of the '&&' itself, unless the LHS evaluates to zero. We treat them 6004 // as if they were unconditionally sequenced. 6005 EvaluationTracker Eval(*this); 6006 { 6007 SequencedSubexpression Sequenced(*this); 6008 Visit(BO->getLHS()); 6009 } 6010 6011 bool Result; 6012 if (Eval.evaluate(BO->getLHS(), Result)) { 6013 if (!Result) 6014 Visit(BO->getRHS()); 6015 } else { 6016 // Check for unsequenced operations in the RHS, treating it as an 6017 // entirely separate evaluation. 6018 // 6019 // FIXME: If there are operations in the RHS which are unsequenced 6020 // with respect to operations outside the RHS, and those operations 6021 // are unconditionally evaluated, diagnose them. 6022 WorkList.push_back(BO->getRHS()); 6023 } 6024 } 6025 void VisitBinLAnd(BinaryOperator *BO) { 6026 EvaluationTracker Eval(*this); 6027 { 6028 SequencedSubexpression Sequenced(*this); 6029 Visit(BO->getLHS()); 6030 } 6031 6032 bool Result; 6033 if (Eval.evaluate(BO->getLHS(), Result)) { 6034 if (Result) 6035 Visit(BO->getRHS()); 6036 } else { 6037 WorkList.push_back(BO->getRHS()); 6038 } 6039 } 6040 6041 // Only visit the condition, unless we can be sure which subexpression will 6042 // be chosen. 6043 void VisitAbstractConditionalOperator(AbstractConditionalOperator *CO) { 6044 EvaluationTracker Eval(*this); 6045 { 6046 SequencedSubexpression Sequenced(*this); 6047 Visit(CO->getCond()); 6048 } 6049 6050 bool Result; 6051 if (Eval.evaluate(CO->getCond(), Result)) 6052 Visit(Result ? CO->getTrueExpr() : CO->getFalseExpr()); 6053 else { 6054 WorkList.push_back(CO->getTrueExpr()); 6055 WorkList.push_back(CO->getFalseExpr()); 6056 } 6057 } 6058 6059 void VisitCallExpr(CallExpr *CE) { 6060 // C++11 [intro.execution]p15: 6061 // When calling a function [...], every value computation and side effect 6062 // associated with any argument expression, or with the postfix expression 6063 // designating the called function, is sequenced before execution of every 6064 // expression or statement in the body of the function [and thus before 6065 // the value computation of its result]. 6066 SequencedSubexpression Sequenced(*this); 6067 Base::VisitCallExpr(CE); 6068 6069 // FIXME: CXXNewExpr and CXXDeleteExpr implicitly call functions. 6070 } 6071 6072 void VisitCXXConstructExpr(CXXConstructExpr *CCE) { 6073 // This is a call, so all subexpressions are sequenced before the result. 6074 SequencedSubexpression Sequenced(*this); 6075 6076 if (!CCE->isListInitialization()) 6077 return VisitExpr(CCE); 6078 6079 // In C++11, list initializations are sequenced. 6080 SmallVector<SequenceTree::Seq, 32> Elts; 6081 SequenceTree::Seq Parent = Region; 6082 for (CXXConstructExpr::arg_iterator I = CCE->arg_begin(), 6083 E = CCE->arg_end(); 6084 I != E; ++I) { 6085 Region = Tree.allocate(Parent); 6086 Elts.push_back(Region); 6087 Visit(*I); 6088 } 6089 6090 // Forget that the initializers are sequenced. 6091 Region = Parent; 6092 for (unsigned I = 0; I < Elts.size(); ++I) 6093 Tree.merge(Elts[I]); 6094 } 6095 6096 void VisitInitListExpr(InitListExpr *ILE) { 6097 if (!SemaRef.getLangOpts().CPlusPlus11) 6098 return VisitExpr(ILE); 6099 6100 // In C++11, list initializations are sequenced. 6101 SmallVector<SequenceTree::Seq, 32> Elts; 6102 SequenceTree::Seq Parent = Region; 6103 for (unsigned I = 0; I < ILE->getNumInits(); ++I) { 6104 Expr *E = ILE->getInit(I); 6105 if (!E) continue; 6106 Region = Tree.allocate(Parent); 6107 Elts.push_back(Region); 6108 Visit(E); 6109 } 6110 6111 // Forget that the initializers are sequenced. 6112 Region = Parent; 6113 for (unsigned I = 0; I < Elts.size(); ++I) 6114 Tree.merge(Elts[I]); 6115 } 6116}; 6117} 6118 6119void Sema::CheckUnsequencedOperations(Expr *E) { 6120 SmallVector<Expr *, 8> WorkList; 6121 WorkList.push_back(E); 6122 while (!WorkList.empty()) { 6123 Expr *Item = WorkList.pop_back_val(); 6124 SequenceChecker(*this, Item, WorkList); 6125 } 6126} 6127 6128void Sema::CheckCompletedExpr(Expr *E, SourceLocation CheckLoc, 6129 bool IsConstexpr) { 6130 CheckImplicitConversions(E, CheckLoc); 6131 CheckUnsequencedOperations(E); 6132 if (!IsConstexpr && !E->isValueDependent()) 6133 CheckForIntOverflow(E); 6134} 6135 6136void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 6137 FieldDecl *BitField, 6138 Expr *Init) { 6139 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 6140} 6141 6142/// CheckParmsForFunctionDef - Check that the parameters of the given 6143/// function are appropriate for the definition of a function. This 6144/// takes care of any checks that cannot be performed on the 6145/// declaration itself, e.g., that the types of each of the function 6146/// parameters are complete. 6147bool Sema::CheckParmsForFunctionDef(ParmVarDecl *const *P, 6148 ParmVarDecl *const *PEnd, 6149 bool CheckParameterNames) { 6150 bool HasInvalidParm = false; 6151 for (; P != PEnd; ++P) { 6152 ParmVarDecl *Param = *P; 6153 6154 // C99 6.7.5.3p4: the parameters in a parameter type list in a 6155 // function declarator that is part of a function definition of 6156 // that function shall not have incomplete type. 6157 // 6158 // This is also C++ [dcl.fct]p6. 6159 if (!Param->isInvalidDecl() && 6160 RequireCompleteType(Param->getLocation(), Param->getType(), 6161 diag::err_typecheck_decl_incomplete_type)) { 6162 Param->setInvalidDecl(); 6163 HasInvalidParm = true; 6164 } 6165 6166 // C99 6.9.1p5: If the declarator includes a parameter type list, the 6167 // declaration of each parameter shall include an identifier. 6168 if (CheckParameterNames && 6169 Param->getIdentifier() == 0 && 6170 !Param->isImplicit() && 6171 !getLangOpts().CPlusPlus) 6172 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 6173 6174 // C99 6.7.5.3p12: 6175 // If the function declarator is not part of a definition of that 6176 // function, parameters may have incomplete type and may use the [*] 6177 // notation in their sequences of declarator specifiers to specify 6178 // variable length array types. 6179 QualType PType = Param->getOriginalType(); 6180 while (const ArrayType *AT = Context.getAsArrayType(PType)) { 6181 if (AT->getSizeModifier() == ArrayType::Star) { 6182 // FIXME: This diagnostic should point the '[*]' if source-location 6183 // information is added for it. 6184 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 6185 break; 6186 } 6187 PType= AT->getElementType(); 6188 } 6189 6190 // MSVC destroys objects passed by value in the callee. Therefore a 6191 // function definition which takes such a parameter must be able to call the 6192 // object's destructor. 6193 if (getLangOpts().CPlusPlus && 6194 Context.getTargetInfo().getCXXABI().isArgumentDestroyedByCallee()) { 6195 if (const RecordType *RT = Param->getType()->getAs<RecordType>()) 6196 FinalizeVarWithDestructor(Param, RT); 6197 } 6198 } 6199 6200 return HasInvalidParm; 6201} 6202 6203/// CheckCastAlign - Implements -Wcast-align, which warns when a 6204/// pointer cast increases the alignment requirements. 6205void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 6206 // This is actually a lot of work to potentially be doing on every 6207 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 6208 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 6209 TRange.getBegin()) 6210 == DiagnosticsEngine::Ignored) 6211 return; 6212 6213 // Ignore dependent types. 6214 if (T->isDependentType() || Op->getType()->isDependentType()) 6215 return; 6216 6217 // Require that the destination be a pointer type. 6218 const PointerType *DestPtr = T->getAs<PointerType>(); 6219 if (!DestPtr) return; 6220 6221 // If the destination has alignment 1, we're done. 6222 QualType DestPointee = DestPtr->getPointeeType(); 6223 if (DestPointee->isIncompleteType()) return; 6224 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 6225 if (DestAlign.isOne()) return; 6226 6227 // Require that the source be a pointer type. 6228 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 6229 if (!SrcPtr) return; 6230 QualType SrcPointee = SrcPtr->getPointeeType(); 6231 6232 // Whitelist casts from cv void*. We already implicitly 6233 // whitelisted casts to cv void*, since they have alignment 1. 6234 // Also whitelist casts involving incomplete types, which implicitly 6235 // includes 'void'. 6236 if (SrcPointee->isIncompleteType()) return; 6237 6238 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 6239 if (SrcAlign >= DestAlign) return; 6240 6241 Diag(TRange.getBegin(), diag::warn_cast_align) 6242 << Op->getType() << T 6243 << static_cast<unsigned>(SrcAlign.getQuantity()) 6244 << static_cast<unsigned>(DestAlign.getQuantity()) 6245 << TRange << Op->getSourceRange(); 6246} 6247 6248static const Type* getElementType(const Expr *BaseExpr) { 6249 const Type* EltType = BaseExpr->getType().getTypePtr(); 6250 if (EltType->isAnyPointerType()) 6251 return EltType->getPointeeType().getTypePtr(); 6252 else if (EltType->isArrayType()) 6253 return EltType->getBaseElementTypeUnsafe(); 6254 return EltType; 6255} 6256 6257/// \brief Check whether this array fits the idiom of a size-one tail padded 6258/// array member of a struct. 6259/// 6260/// We avoid emitting out-of-bounds access warnings for such arrays as they are 6261/// commonly used to emulate flexible arrays in C89 code. 6262static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 6263 const NamedDecl *ND) { 6264 if (Size != 1 || !ND) return false; 6265 6266 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 6267 if (!FD) return false; 6268 6269 // Don't consider sizes resulting from macro expansions or template argument 6270 // substitution to form C89 tail-padded arrays. 6271 6272 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 6273 while (TInfo) { 6274 TypeLoc TL = TInfo->getTypeLoc(); 6275 // Look through typedefs. 6276 if (TypedefTypeLoc TTL = TL.getAs<TypedefTypeLoc>()) { 6277 const TypedefNameDecl *TDL = TTL.getTypedefNameDecl(); 6278 TInfo = TDL->getTypeSourceInfo(); 6279 continue; 6280 } 6281 if (ConstantArrayTypeLoc CTL = TL.getAs<ConstantArrayTypeLoc>()) { 6282 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 6283 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 6284 return false; 6285 } 6286 break; 6287 } 6288 6289 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 6290 if (!RD) return false; 6291 if (RD->isUnion()) return false; 6292 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 6293 if (!CRD->isStandardLayout()) return false; 6294 } 6295 6296 // See if this is the last field decl in the record. 6297 const Decl *D = FD; 6298 while ((D = D->getNextDeclInContext())) 6299 if (isa<FieldDecl>(D)) 6300 return false; 6301 return true; 6302} 6303 6304void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 6305 const ArraySubscriptExpr *ASE, 6306 bool AllowOnePastEnd, bool IndexNegated) { 6307 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 6308 if (IndexExpr->isValueDependent()) 6309 return; 6310 6311 const Type *EffectiveType = getElementType(BaseExpr); 6312 BaseExpr = BaseExpr->IgnoreParenCasts(); 6313 const ConstantArrayType *ArrayTy = 6314 Context.getAsConstantArrayType(BaseExpr->getType()); 6315 if (!ArrayTy) 6316 return; 6317 6318 llvm::APSInt index; 6319 if (!IndexExpr->EvaluateAsInt(index, Context)) 6320 return; 6321 if (IndexNegated) 6322 index = -index; 6323 6324 const NamedDecl *ND = NULL; 6325 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 6326 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 6327 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 6328 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 6329 6330 if (index.isUnsigned() || !index.isNegative()) { 6331 llvm::APInt size = ArrayTy->getSize(); 6332 if (!size.isStrictlyPositive()) 6333 return; 6334 6335 const Type* BaseType = getElementType(BaseExpr); 6336 if (BaseType != EffectiveType) { 6337 // Make sure we're comparing apples to apples when comparing index to size 6338 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 6339 uint64_t array_typesize = Context.getTypeSize(BaseType); 6340 // Handle ptrarith_typesize being zero, such as when casting to void* 6341 if (!ptrarith_typesize) ptrarith_typesize = 1; 6342 if (ptrarith_typesize != array_typesize) { 6343 // There's a cast to a different size type involved 6344 uint64_t ratio = array_typesize / ptrarith_typesize; 6345 // TODO: Be smarter about handling cases where array_typesize is not a 6346 // multiple of ptrarith_typesize 6347 if (ptrarith_typesize * ratio == array_typesize) 6348 size *= llvm::APInt(size.getBitWidth(), ratio); 6349 } 6350 } 6351 6352 if (size.getBitWidth() > index.getBitWidth()) 6353 index = index.zext(size.getBitWidth()); 6354 else if (size.getBitWidth() < index.getBitWidth()) 6355 size = size.zext(index.getBitWidth()); 6356 6357 // For array subscripting the index must be less than size, but for pointer 6358 // arithmetic also allow the index (offset) to be equal to size since 6359 // computing the next address after the end of the array is legal and 6360 // commonly done e.g. in C++ iterators and range-based for loops. 6361 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 6362 return; 6363 6364 // Also don't warn for arrays of size 1 which are members of some 6365 // structure. These are often used to approximate flexible arrays in C89 6366 // code. 6367 if (IsTailPaddedMemberArray(*this, size, ND)) 6368 return; 6369 6370 // Suppress the warning if the subscript expression (as identified by the 6371 // ']' location) and the index expression are both from macro expansions 6372 // within a system header. 6373 if (ASE) { 6374 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 6375 ASE->getRBracketLoc()); 6376 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 6377 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 6378 IndexExpr->getLocStart()); 6379 if (SourceMgr.isWrittenInSameFile(RBracketLoc, IndexLoc)) 6380 return; 6381 } 6382 } 6383 6384 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 6385 if (ASE) 6386 DiagID = diag::warn_array_index_exceeds_bounds; 6387 6388 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 6389 PDiag(DiagID) << index.toString(10, true) 6390 << size.toString(10, true) 6391 << (unsigned)size.getLimitedValue(~0U) 6392 << IndexExpr->getSourceRange()); 6393 } else { 6394 unsigned DiagID = diag::warn_array_index_precedes_bounds; 6395 if (!ASE) { 6396 DiagID = diag::warn_ptr_arith_precedes_bounds; 6397 if (index.isNegative()) index = -index; 6398 } 6399 6400 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 6401 PDiag(DiagID) << index.toString(10, true) 6402 << IndexExpr->getSourceRange()); 6403 } 6404 6405 if (!ND) { 6406 // Try harder to find a NamedDecl to point at in the note. 6407 while (const ArraySubscriptExpr *ASE = 6408 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 6409 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 6410 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 6411 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 6412 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 6413 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 6414 } 6415 6416 if (ND) 6417 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 6418 PDiag(diag::note_array_index_out_of_bounds) 6419 << ND->getDeclName()); 6420} 6421 6422void Sema::CheckArrayAccess(const Expr *expr) { 6423 int AllowOnePastEnd = 0; 6424 while (expr) { 6425 expr = expr->IgnoreParenImpCasts(); 6426 switch (expr->getStmtClass()) { 6427 case Stmt::ArraySubscriptExprClass: { 6428 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 6429 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 6430 AllowOnePastEnd > 0); 6431 return; 6432 } 6433 case Stmt::UnaryOperatorClass: { 6434 // Only unwrap the * and & unary operators 6435 const UnaryOperator *UO = cast<UnaryOperator>(expr); 6436 expr = UO->getSubExpr(); 6437 switch (UO->getOpcode()) { 6438 case UO_AddrOf: 6439 AllowOnePastEnd++; 6440 break; 6441 case UO_Deref: 6442 AllowOnePastEnd--; 6443 break; 6444 default: 6445 return; 6446 } 6447 break; 6448 } 6449 case Stmt::ConditionalOperatorClass: { 6450 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 6451 if (const Expr *lhs = cond->getLHS()) 6452 CheckArrayAccess(lhs); 6453 if (const Expr *rhs = cond->getRHS()) 6454 CheckArrayAccess(rhs); 6455 return; 6456 } 6457 default: 6458 return; 6459 } 6460 } 6461} 6462 6463//===--- CHECK: Objective-C retain cycles ----------------------------------// 6464 6465namespace { 6466 struct RetainCycleOwner { 6467 RetainCycleOwner() : Variable(0), Indirect(false) {} 6468 VarDecl *Variable; 6469 SourceRange Range; 6470 SourceLocation Loc; 6471 bool Indirect; 6472 6473 void setLocsFrom(Expr *e) { 6474 Loc = e->getExprLoc(); 6475 Range = e->getSourceRange(); 6476 } 6477 }; 6478} 6479 6480/// Consider whether capturing the given variable can possibly lead to 6481/// a retain cycle. 6482static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 6483 // In ARC, it's captured strongly iff the variable has __strong 6484 // lifetime. In MRR, it's captured strongly if the variable is 6485 // __block and has an appropriate type. 6486 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 6487 return false; 6488 6489 owner.Variable = var; 6490 if (ref) 6491 owner.setLocsFrom(ref); 6492 return true; 6493} 6494 6495static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 6496 while (true) { 6497 e = e->IgnoreParens(); 6498 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 6499 switch (cast->getCastKind()) { 6500 case CK_BitCast: 6501 case CK_LValueBitCast: 6502 case CK_LValueToRValue: 6503 case CK_ARCReclaimReturnedObject: 6504 e = cast->getSubExpr(); 6505 continue; 6506 6507 default: 6508 return false; 6509 } 6510 } 6511 6512 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 6513 ObjCIvarDecl *ivar = ref->getDecl(); 6514 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 6515 return false; 6516 6517 // Try to find a retain cycle in the base. 6518 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 6519 return false; 6520 6521 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 6522 owner.Indirect = true; 6523 return true; 6524 } 6525 6526 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 6527 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 6528 if (!var) return false; 6529 return considerVariable(var, ref, owner); 6530 } 6531 6532 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 6533 if (member->isArrow()) return false; 6534 6535 // Don't count this as an indirect ownership. 6536 e = member->getBase(); 6537 continue; 6538 } 6539 6540 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 6541 // Only pay attention to pseudo-objects on property references. 6542 ObjCPropertyRefExpr *pre 6543 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 6544 ->IgnoreParens()); 6545 if (!pre) return false; 6546 if (pre->isImplicitProperty()) return false; 6547 ObjCPropertyDecl *property = pre->getExplicitProperty(); 6548 if (!property->isRetaining() && 6549 !(property->getPropertyIvarDecl() && 6550 property->getPropertyIvarDecl()->getType() 6551 .getObjCLifetime() == Qualifiers::OCL_Strong)) 6552 return false; 6553 6554 owner.Indirect = true; 6555 if (pre->isSuperReceiver()) { 6556 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 6557 if (!owner.Variable) 6558 return false; 6559 owner.Loc = pre->getLocation(); 6560 owner.Range = pre->getSourceRange(); 6561 return true; 6562 } 6563 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 6564 ->getSourceExpr()); 6565 continue; 6566 } 6567 6568 // Array ivars? 6569 6570 return false; 6571 } 6572} 6573 6574namespace { 6575 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 6576 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 6577 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 6578 Variable(variable), Capturer(0) {} 6579 6580 VarDecl *Variable; 6581 Expr *Capturer; 6582 6583 void VisitDeclRefExpr(DeclRefExpr *ref) { 6584 if (ref->getDecl() == Variable && !Capturer) 6585 Capturer = ref; 6586 } 6587 6588 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 6589 if (Capturer) return; 6590 Visit(ref->getBase()); 6591 if (Capturer && ref->isFreeIvar()) 6592 Capturer = ref; 6593 } 6594 6595 void VisitBlockExpr(BlockExpr *block) { 6596 // Look inside nested blocks 6597 if (block->getBlockDecl()->capturesVariable(Variable)) 6598 Visit(block->getBlockDecl()->getBody()); 6599 } 6600 6601 void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) { 6602 if (Capturer) return; 6603 if (OVE->getSourceExpr()) 6604 Visit(OVE->getSourceExpr()); 6605 } 6606 }; 6607} 6608 6609/// Check whether the given argument is a block which captures a 6610/// variable. 6611static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 6612 assert(owner.Variable && owner.Loc.isValid()); 6613 6614 e = e->IgnoreParenCasts(); 6615 6616 // Look through [^{...} copy] and Block_copy(^{...}). 6617 if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) { 6618 Selector Cmd = ME->getSelector(); 6619 if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") { 6620 e = ME->getInstanceReceiver(); 6621 if (!e) 6622 return 0; 6623 e = e->IgnoreParenCasts(); 6624 } 6625 } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) { 6626 if (CE->getNumArgs() == 1) { 6627 FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl()); 6628 if (Fn) { 6629 const IdentifierInfo *FnI = Fn->getIdentifier(); 6630 if (FnI && FnI->isStr("_Block_copy")) { 6631 e = CE->getArg(0)->IgnoreParenCasts(); 6632 } 6633 } 6634 } 6635 } 6636 6637 BlockExpr *block = dyn_cast<BlockExpr>(e); 6638 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 6639 return 0; 6640 6641 FindCaptureVisitor visitor(S.Context, owner.Variable); 6642 visitor.Visit(block->getBlockDecl()->getBody()); 6643 return visitor.Capturer; 6644} 6645 6646static void diagnoseRetainCycle(Sema &S, Expr *capturer, 6647 RetainCycleOwner &owner) { 6648 assert(capturer); 6649 assert(owner.Variable && owner.Loc.isValid()); 6650 6651 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 6652 << owner.Variable << capturer->getSourceRange(); 6653 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 6654 << owner.Indirect << owner.Range; 6655} 6656 6657/// Check for a keyword selector that starts with the word 'add' or 6658/// 'set'. 6659static bool isSetterLikeSelector(Selector sel) { 6660 if (sel.isUnarySelector()) return false; 6661 6662 StringRef str = sel.getNameForSlot(0); 6663 while (!str.empty() && str.front() == '_') str = str.substr(1); 6664 if (str.startswith("set")) 6665 str = str.substr(3); 6666 else if (str.startswith("add")) { 6667 // Specially whitelist 'addOperationWithBlock:'. 6668 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 6669 return false; 6670 str = str.substr(3); 6671 } 6672 else 6673 return false; 6674 6675 if (str.empty()) return true; 6676 return !isLowercase(str.front()); 6677} 6678 6679/// Check a message send to see if it's likely to cause a retain cycle. 6680void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 6681 // Only check instance methods whose selector looks like a setter. 6682 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 6683 return; 6684 6685 // Try to find a variable that the receiver is strongly owned by. 6686 RetainCycleOwner owner; 6687 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 6688 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 6689 return; 6690 } else { 6691 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 6692 owner.Variable = getCurMethodDecl()->getSelfDecl(); 6693 owner.Loc = msg->getSuperLoc(); 6694 owner.Range = msg->getSuperLoc(); 6695 } 6696 6697 // Check whether the receiver is captured by any of the arguments. 6698 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 6699 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 6700 return diagnoseRetainCycle(*this, capturer, owner); 6701} 6702 6703/// Check a property assign to see if it's likely to cause a retain cycle. 6704void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 6705 RetainCycleOwner owner; 6706 if (!findRetainCycleOwner(*this, receiver, owner)) 6707 return; 6708 6709 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 6710 diagnoseRetainCycle(*this, capturer, owner); 6711} 6712 6713void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) { 6714 RetainCycleOwner Owner; 6715 if (!considerVariable(Var, /*DeclRefExpr=*/0, Owner)) 6716 return; 6717 6718 // Because we don't have an expression for the variable, we have to set the 6719 // location explicitly here. 6720 Owner.Loc = Var->getLocation(); 6721 Owner.Range = Var->getSourceRange(); 6722 6723 if (Expr *Capturer = findCapturingExpr(*this, Init, Owner)) 6724 diagnoseRetainCycle(*this, Capturer, Owner); 6725} 6726 6727static bool checkUnsafeAssignLiteral(Sema &S, SourceLocation Loc, 6728 Expr *RHS, bool isProperty) { 6729 // Check if RHS is an Objective-C object literal, which also can get 6730 // immediately zapped in a weak reference. Note that we explicitly 6731 // allow ObjCStringLiterals, since those are designed to never really die. 6732 RHS = RHS->IgnoreParenImpCasts(); 6733 6734 // This enum needs to match with the 'select' in 6735 // warn_objc_arc_literal_assign (off-by-1). 6736 Sema::ObjCLiteralKind Kind = S.CheckLiteralKind(RHS); 6737 if (Kind == Sema::LK_String || Kind == Sema::LK_None) 6738 return false; 6739 6740 S.Diag(Loc, diag::warn_arc_literal_assign) 6741 << (unsigned) Kind 6742 << (isProperty ? 0 : 1) 6743 << RHS->getSourceRange(); 6744 6745 return true; 6746} 6747 6748static bool checkUnsafeAssignObject(Sema &S, SourceLocation Loc, 6749 Qualifiers::ObjCLifetime LT, 6750 Expr *RHS, bool isProperty) { 6751 // Strip off any implicit cast added to get to the one ARC-specific. 6752 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6753 if (cast->getCastKind() == CK_ARCConsumeObject) { 6754 S.Diag(Loc, diag::warn_arc_retained_assign) 6755 << (LT == Qualifiers::OCL_ExplicitNone) 6756 << (isProperty ? 0 : 1) 6757 << RHS->getSourceRange(); 6758 return true; 6759 } 6760 RHS = cast->getSubExpr(); 6761 } 6762 6763 if (LT == Qualifiers::OCL_Weak && 6764 checkUnsafeAssignLiteral(S, Loc, RHS, isProperty)) 6765 return true; 6766 6767 return false; 6768} 6769 6770bool Sema::checkUnsafeAssigns(SourceLocation Loc, 6771 QualType LHS, Expr *RHS) { 6772 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 6773 6774 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 6775 return false; 6776 6777 if (checkUnsafeAssignObject(*this, Loc, LT, RHS, false)) 6778 return true; 6779 6780 return false; 6781} 6782 6783void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 6784 Expr *LHS, Expr *RHS) { 6785 QualType LHSType; 6786 // PropertyRef on LHS type need be directly obtained from 6787 // its declaration as it has a PsuedoType. 6788 ObjCPropertyRefExpr *PRE 6789 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 6790 if (PRE && !PRE->isImplicitProperty()) { 6791 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6792 if (PD) 6793 LHSType = PD->getType(); 6794 } 6795 6796 if (LHSType.isNull()) 6797 LHSType = LHS->getType(); 6798 6799 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 6800 6801 if (LT == Qualifiers::OCL_Weak) { 6802 DiagnosticsEngine::Level Level = 6803 Diags.getDiagnosticLevel(diag::warn_arc_repeated_use_of_weak, Loc); 6804 if (Level != DiagnosticsEngine::Ignored) 6805 getCurFunction()->markSafeWeakUse(LHS); 6806 } 6807 6808 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 6809 return; 6810 6811 // FIXME. Check for other life times. 6812 if (LT != Qualifiers::OCL_None) 6813 return; 6814 6815 if (PRE) { 6816 if (PRE->isImplicitProperty()) 6817 return; 6818 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6819 if (!PD) 6820 return; 6821 6822 unsigned Attributes = PD->getPropertyAttributes(); 6823 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 6824 // when 'assign' attribute was not explicitly specified 6825 // by user, ignore it and rely on property type itself 6826 // for lifetime info. 6827 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 6828 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 6829 LHSType->isObjCRetainableType()) 6830 return; 6831 6832 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6833 if (cast->getCastKind() == CK_ARCConsumeObject) { 6834 Diag(Loc, diag::warn_arc_retained_property_assign) 6835 << RHS->getSourceRange(); 6836 return; 6837 } 6838 RHS = cast->getSubExpr(); 6839 } 6840 } 6841 else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) { 6842 if (checkUnsafeAssignObject(*this, Loc, Qualifiers::OCL_Weak, RHS, true)) 6843 return; 6844 } 6845 } 6846} 6847 6848//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 6849 6850namespace { 6851bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 6852 SourceLocation StmtLoc, 6853 const NullStmt *Body) { 6854 // Do not warn if the body is a macro that expands to nothing, e.g: 6855 // 6856 // #define CALL(x) 6857 // if (condition) 6858 // CALL(0); 6859 // 6860 if (Body->hasLeadingEmptyMacro()) 6861 return false; 6862 6863 // Get line numbers of statement and body. 6864 bool StmtLineInvalid; 6865 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 6866 &StmtLineInvalid); 6867 if (StmtLineInvalid) 6868 return false; 6869 6870 bool BodyLineInvalid; 6871 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 6872 &BodyLineInvalid); 6873 if (BodyLineInvalid) 6874 return false; 6875 6876 // Warn if null statement and body are on the same line. 6877 if (StmtLine != BodyLine) 6878 return false; 6879 6880 return true; 6881} 6882} // Unnamed namespace 6883 6884void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 6885 const Stmt *Body, 6886 unsigned DiagID) { 6887 // Since this is a syntactic check, don't emit diagnostic for template 6888 // instantiations, this just adds noise. 6889 if (CurrentInstantiationScope) 6890 return; 6891 6892 // The body should be a null statement. 6893 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6894 if (!NBody) 6895 return; 6896 6897 // Do the usual checks. 6898 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6899 return; 6900 6901 Diag(NBody->getSemiLoc(), DiagID); 6902 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6903} 6904 6905void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 6906 const Stmt *PossibleBody) { 6907 assert(!CurrentInstantiationScope); // Ensured by caller 6908 6909 SourceLocation StmtLoc; 6910 const Stmt *Body; 6911 unsigned DiagID; 6912 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 6913 StmtLoc = FS->getRParenLoc(); 6914 Body = FS->getBody(); 6915 DiagID = diag::warn_empty_for_body; 6916 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 6917 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 6918 Body = WS->getBody(); 6919 DiagID = diag::warn_empty_while_body; 6920 } else 6921 return; // Neither `for' nor `while'. 6922 6923 // The body should be a null statement. 6924 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6925 if (!NBody) 6926 return; 6927 6928 // Skip expensive checks if diagnostic is disabled. 6929 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 6930 DiagnosticsEngine::Ignored) 6931 return; 6932 6933 // Do the usual checks. 6934 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6935 return; 6936 6937 // `for(...);' and `while(...);' are popular idioms, so in order to keep 6938 // noise level low, emit diagnostics only if for/while is followed by a 6939 // CompoundStmt, e.g.: 6940 // for (int i = 0; i < n; i++); 6941 // { 6942 // a(i); 6943 // } 6944 // or if for/while is followed by a statement with more indentation 6945 // than for/while itself: 6946 // for (int i = 0; i < n; i++); 6947 // a(i); 6948 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 6949 if (!ProbableTypo) { 6950 bool BodyColInvalid; 6951 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 6952 PossibleBody->getLocStart(), 6953 &BodyColInvalid); 6954 if (BodyColInvalid) 6955 return; 6956 6957 bool StmtColInvalid; 6958 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 6959 S->getLocStart(), 6960 &StmtColInvalid); 6961 if (StmtColInvalid) 6962 return; 6963 6964 if (BodyCol > StmtCol) 6965 ProbableTypo = true; 6966 } 6967 6968 if (ProbableTypo) { 6969 Diag(NBody->getSemiLoc(), DiagID); 6970 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6971 } 6972} 6973 6974//===--- Layout compatibility ----------------------------------------------// 6975 6976namespace { 6977 6978bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2); 6979 6980/// \brief Check if two enumeration types are layout-compatible. 6981bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) { 6982 // C++11 [dcl.enum] p8: 6983 // Two enumeration types are layout-compatible if they have the same 6984 // underlying type. 6985 return ED1->isComplete() && ED2->isComplete() && 6986 C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType()); 6987} 6988 6989/// \brief Check if two fields are layout-compatible. 6990bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) { 6991 if (!isLayoutCompatible(C, Field1->getType(), Field2->getType())) 6992 return false; 6993 6994 if (Field1->isBitField() != Field2->isBitField()) 6995 return false; 6996 6997 if (Field1->isBitField()) { 6998 // Make sure that the bit-fields are the same length. 6999 unsigned Bits1 = Field1->getBitWidthValue(C); 7000 unsigned Bits2 = Field2->getBitWidthValue(C); 7001 7002 if (Bits1 != Bits2) 7003 return false; 7004 } 7005 7006 return true; 7007} 7008 7009/// \brief Check if two standard-layout structs are layout-compatible. 7010/// (C++11 [class.mem] p17) 7011bool isLayoutCompatibleStruct(ASTContext &C, 7012 RecordDecl *RD1, 7013 RecordDecl *RD2) { 7014 // If both records are C++ classes, check that base classes match. 7015 if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) { 7016 // If one of records is a CXXRecordDecl we are in C++ mode, 7017 // thus the other one is a CXXRecordDecl, too. 7018 const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2); 7019 // Check number of base classes. 7020 if (D1CXX->getNumBases() != D2CXX->getNumBases()) 7021 return false; 7022 7023 // Check the base classes. 7024 for (CXXRecordDecl::base_class_const_iterator 7025 Base1 = D1CXX->bases_begin(), 7026 BaseEnd1 = D1CXX->bases_end(), 7027 Base2 = D2CXX->bases_begin(); 7028 Base1 != BaseEnd1; 7029 ++Base1, ++Base2) { 7030 if (!isLayoutCompatible(C, Base1->getType(), Base2->getType())) 7031 return false; 7032 } 7033 } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) { 7034 // If only RD2 is a C++ class, it should have zero base classes. 7035 if (D2CXX->getNumBases() > 0) 7036 return false; 7037 } 7038 7039 // Check the fields. 7040 RecordDecl::field_iterator Field2 = RD2->field_begin(), 7041 Field2End = RD2->field_end(), 7042 Field1 = RD1->field_begin(), 7043 Field1End = RD1->field_end(); 7044 for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) { 7045 if (!isLayoutCompatible(C, *Field1, *Field2)) 7046 return false; 7047 } 7048 if (Field1 != Field1End || Field2 != Field2End) 7049 return false; 7050 7051 return true; 7052} 7053 7054/// \brief Check if two standard-layout unions are layout-compatible. 7055/// (C++11 [class.mem] p18) 7056bool isLayoutCompatibleUnion(ASTContext &C, 7057 RecordDecl *RD1, 7058 RecordDecl *RD2) { 7059 llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields; 7060 for (RecordDecl::field_iterator Field2 = RD2->field_begin(), 7061 Field2End = RD2->field_end(); 7062 Field2 != Field2End; ++Field2) { 7063 UnmatchedFields.insert(*Field2); 7064 } 7065 7066 for (RecordDecl::field_iterator Field1 = RD1->field_begin(), 7067 Field1End = RD1->field_end(); 7068 Field1 != Field1End; ++Field1) { 7069 llvm::SmallPtrSet<FieldDecl *, 8>::iterator 7070 I = UnmatchedFields.begin(), 7071 E = UnmatchedFields.end(); 7072 7073 for ( ; I != E; ++I) { 7074 if (isLayoutCompatible(C, *Field1, *I)) { 7075 bool Result = UnmatchedFields.erase(*I); 7076 (void) Result; 7077 assert(Result); 7078 break; 7079 } 7080 } 7081 if (I == E) 7082 return false; 7083 } 7084 7085 return UnmatchedFields.empty(); 7086} 7087 7088bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) { 7089 if (RD1->isUnion() != RD2->isUnion()) 7090 return false; 7091 7092 if (RD1->isUnion()) 7093 return isLayoutCompatibleUnion(C, RD1, RD2); 7094 else 7095 return isLayoutCompatibleStruct(C, RD1, RD2); 7096} 7097 7098/// \brief Check if two types are layout-compatible in C++11 sense. 7099bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) { 7100 if (T1.isNull() || T2.isNull()) 7101 return false; 7102 7103 // C++11 [basic.types] p11: 7104 // If two types T1 and T2 are the same type, then T1 and T2 are 7105 // layout-compatible types. 7106 if (C.hasSameType(T1, T2)) 7107 return true; 7108 7109 T1 = T1.getCanonicalType().getUnqualifiedType(); 7110 T2 = T2.getCanonicalType().getUnqualifiedType(); 7111 7112 const Type::TypeClass TC1 = T1->getTypeClass(); 7113 const Type::TypeClass TC2 = T2->getTypeClass(); 7114 7115 if (TC1 != TC2) 7116 return false; 7117 7118 if (TC1 == Type::Enum) { 7119 return isLayoutCompatible(C, 7120 cast<EnumType>(T1)->getDecl(), 7121 cast<EnumType>(T2)->getDecl()); 7122 } else if (TC1 == Type::Record) { 7123 if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType()) 7124 return false; 7125 7126 return isLayoutCompatible(C, 7127 cast<RecordType>(T1)->getDecl(), 7128 cast<RecordType>(T2)->getDecl()); 7129 } 7130 7131 return false; 7132} 7133} 7134 7135//===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----// 7136 7137namespace { 7138/// \brief Given a type tag expression find the type tag itself. 7139/// 7140/// \param TypeExpr Type tag expression, as it appears in user's code. 7141/// 7142/// \param VD Declaration of an identifier that appears in a type tag. 7143/// 7144/// \param MagicValue Type tag magic value. 7145bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx, 7146 const ValueDecl **VD, uint64_t *MagicValue) { 7147 while(true) { 7148 if (!TypeExpr) 7149 return false; 7150 7151 TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts(); 7152 7153 switch (TypeExpr->getStmtClass()) { 7154 case Stmt::UnaryOperatorClass: { 7155 const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr); 7156 if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) { 7157 TypeExpr = UO->getSubExpr(); 7158 continue; 7159 } 7160 return false; 7161 } 7162 7163 case Stmt::DeclRefExprClass: { 7164 const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr); 7165 *VD = DRE->getDecl(); 7166 return true; 7167 } 7168 7169 case Stmt::IntegerLiteralClass: { 7170 const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr); 7171 llvm::APInt MagicValueAPInt = IL->getValue(); 7172 if (MagicValueAPInt.getActiveBits() <= 64) { 7173 *MagicValue = MagicValueAPInt.getZExtValue(); 7174 return true; 7175 } else 7176 return false; 7177 } 7178 7179 case Stmt::BinaryConditionalOperatorClass: 7180 case Stmt::ConditionalOperatorClass: { 7181 const AbstractConditionalOperator *ACO = 7182 cast<AbstractConditionalOperator>(TypeExpr); 7183 bool Result; 7184 if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) { 7185 if (Result) 7186 TypeExpr = ACO->getTrueExpr(); 7187 else 7188 TypeExpr = ACO->getFalseExpr(); 7189 continue; 7190 } 7191 return false; 7192 } 7193 7194 case Stmt::BinaryOperatorClass: { 7195 const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr); 7196 if (BO->getOpcode() == BO_Comma) { 7197 TypeExpr = BO->getRHS(); 7198 continue; 7199 } 7200 return false; 7201 } 7202 7203 default: 7204 return false; 7205 } 7206 } 7207} 7208 7209/// \brief Retrieve the C type corresponding to type tag TypeExpr. 7210/// 7211/// \param TypeExpr Expression that specifies a type tag. 7212/// 7213/// \param MagicValues Registered magic values. 7214/// 7215/// \param FoundWrongKind Set to true if a type tag was found, but of a wrong 7216/// kind. 7217/// 7218/// \param TypeInfo Information about the corresponding C type. 7219/// 7220/// \returns true if the corresponding C type was found. 7221bool GetMatchingCType( 7222 const IdentifierInfo *ArgumentKind, 7223 const Expr *TypeExpr, const ASTContext &Ctx, 7224 const llvm::DenseMap<Sema::TypeTagMagicValue, 7225 Sema::TypeTagData> *MagicValues, 7226 bool &FoundWrongKind, 7227 Sema::TypeTagData &TypeInfo) { 7228 FoundWrongKind = false; 7229 7230 // Variable declaration that has type_tag_for_datatype attribute. 7231 const ValueDecl *VD = NULL; 7232 7233 uint64_t MagicValue; 7234 7235 if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue)) 7236 return false; 7237 7238 if (VD) { 7239 for (specific_attr_iterator<TypeTagForDatatypeAttr> 7240 I = VD->specific_attr_begin<TypeTagForDatatypeAttr>(), 7241 E = VD->specific_attr_end<TypeTagForDatatypeAttr>(); 7242 I != E; ++I) { 7243 if (I->getArgumentKind() != ArgumentKind) { 7244 FoundWrongKind = true; 7245 return false; 7246 } 7247 TypeInfo.Type = I->getMatchingCType(); 7248 TypeInfo.LayoutCompatible = I->getLayoutCompatible(); 7249 TypeInfo.MustBeNull = I->getMustBeNull(); 7250 return true; 7251 } 7252 return false; 7253 } 7254 7255 if (!MagicValues) 7256 return false; 7257 7258 llvm::DenseMap<Sema::TypeTagMagicValue, 7259 Sema::TypeTagData>::const_iterator I = 7260 MagicValues->find(std::make_pair(ArgumentKind, MagicValue)); 7261 if (I == MagicValues->end()) 7262 return false; 7263 7264 TypeInfo = I->second; 7265 return true; 7266} 7267} // unnamed namespace 7268 7269void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind, 7270 uint64_t MagicValue, QualType Type, 7271 bool LayoutCompatible, 7272 bool MustBeNull) { 7273 if (!TypeTagForDatatypeMagicValues) 7274 TypeTagForDatatypeMagicValues.reset( 7275 new llvm::DenseMap<TypeTagMagicValue, TypeTagData>); 7276 7277 TypeTagMagicValue Magic(ArgumentKind, MagicValue); 7278 (*TypeTagForDatatypeMagicValues)[Magic] = 7279 TypeTagData(Type, LayoutCompatible, MustBeNull); 7280} 7281 7282namespace { 7283bool IsSameCharType(QualType T1, QualType T2) { 7284 const BuiltinType *BT1 = T1->getAs<BuiltinType>(); 7285 if (!BT1) 7286 return false; 7287 7288 const BuiltinType *BT2 = T2->getAs<BuiltinType>(); 7289 if (!BT2) 7290 return false; 7291 7292 BuiltinType::Kind T1Kind = BT1->getKind(); 7293 BuiltinType::Kind T2Kind = BT2->getKind(); 7294 7295 return (T1Kind == BuiltinType::SChar && T2Kind == BuiltinType::Char_S) || 7296 (T1Kind == BuiltinType::UChar && T2Kind == BuiltinType::Char_U) || 7297 (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) || 7298 (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar); 7299} 7300} // unnamed namespace 7301 7302void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr, 7303 const Expr * const *ExprArgs) { 7304 const IdentifierInfo *ArgumentKind = Attr->getArgumentKind(); 7305 bool IsPointerAttr = Attr->getIsPointer(); 7306 7307 const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()]; 7308 bool FoundWrongKind; 7309 TypeTagData TypeInfo; 7310 if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context, 7311 TypeTagForDatatypeMagicValues.get(), 7312 FoundWrongKind, TypeInfo)) { 7313 if (FoundWrongKind) 7314 Diag(TypeTagExpr->getExprLoc(), 7315 diag::warn_type_tag_for_datatype_wrong_kind) 7316 << TypeTagExpr->getSourceRange(); 7317 return; 7318 } 7319 7320 const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()]; 7321 if (IsPointerAttr) { 7322 // Skip implicit cast of pointer to `void *' (as a function argument). 7323 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr)) 7324 if (ICE->getType()->isVoidPointerType() && 7325 ICE->getCastKind() == CK_BitCast) 7326 ArgumentExpr = ICE->getSubExpr(); 7327 } 7328 QualType ArgumentType = ArgumentExpr->getType(); 7329 7330 // Passing a `void*' pointer shouldn't trigger a warning. 7331 if (IsPointerAttr && ArgumentType->isVoidPointerType()) 7332 return; 7333 7334 if (TypeInfo.MustBeNull) { 7335 // Type tag with matching void type requires a null pointer. 7336 if (!ArgumentExpr->isNullPointerConstant(Context, 7337 Expr::NPC_ValueDependentIsNotNull)) { 7338 Diag(ArgumentExpr->getExprLoc(), 7339 diag::warn_type_safety_null_pointer_required) 7340 << ArgumentKind->getName() 7341 << ArgumentExpr->getSourceRange() 7342 << TypeTagExpr->getSourceRange(); 7343 } 7344 return; 7345 } 7346 7347 QualType RequiredType = TypeInfo.Type; 7348 if (IsPointerAttr) 7349 RequiredType = Context.getPointerType(RequiredType); 7350 7351 bool mismatch = false; 7352 if (!TypeInfo.LayoutCompatible) { 7353 mismatch = !Context.hasSameType(ArgumentType, RequiredType); 7354 7355 // C++11 [basic.fundamental] p1: 7356 // Plain char, signed char, and unsigned char are three distinct types. 7357 // 7358 // But we treat plain `char' as equivalent to `signed char' or `unsigned 7359 // char' depending on the current char signedness mode. 7360 if (mismatch) 7361 if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(), 7362 RequiredType->getPointeeType())) || 7363 (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType))) 7364 mismatch = false; 7365 } else 7366 if (IsPointerAttr) 7367 mismatch = !isLayoutCompatible(Context, 7368 ArgumentType->getPointeeType(), 7369 RequiredType->getPointeeType()); 7370 else 7371 mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType); 7372 7373 if (mismatch) 7374 Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch) 7375 << ArgumentType << ArgumentKind->getName() 7376 << TypeInfo.LayoutCompatible << RequiredType 7377 << ArgumentExpr->getSourceRange() 7378 << TypeTagExpr->getSourceRange(); 7379} 7380