SemaChecking.cpp revision a3d727ba775eaecd4fd69e1c90b81732394716a6
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 return shift ? 63 : (1 << IsQuad) - 1; 343 case NeonTypeFlags::Float16: 344 assert(!shift && "cannot shift float types!"); 345 return (4 << IsQuad) - 1; 346 case NeonTypeFlags::Float32: 347 assert(!shift && "cannot shift float types!"); 348 return (2 << IsQuad) - 1; 349 case NeonTypeFlags::Float64: 350 assert(!shift && "cannot shift float types!"); 351 return (1 << IsQuad) - 1; 352 } 353 llvm_unreachable("Invalid NeonTypeFlag!"); 354} 355 356/// getNeonEltType - Return the QualType corresponding to the elements of 357/// the vector type specified by the NeonTypeFlags. This is used to check 358/// the pointer arguments for Neon load/store intrinsics. 359static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) { 360 switch (Flags.getEltType()) { 361 case NeonTypeFlags::Int8: 362 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 363 case NeonTypeFlags::Int16: 364 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 365 case NeonTypeFlags::Int32: 366 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 367 case NeonTypeFlags::Int64: 368 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 369 case NeonTypeFlags::Poly8: 370 return Context.SignedCharTy; 371 case NeonTypeFlags::Poly16: 372 return Context.ShortTy; 373 case NeonTypeFlags::Float16: 374 return Context.UnsignedShortTy; 375 case NeonTypeFlags::Float32: 376 return Context.FloatTy; 377 case NeonTypeFlags::Float64: 378 return Context.DoubleTy; 379 } 380 llvm_unreachable("Invalid NeonTypeFlag!"); 381} 382 383bool Sema::CheckAArch64BuiltinFunctionCall(unsigned BuiltinID, 384 CallExpr *TheCall) { 385 386 llvm::APSInt Result; 387 388 uint64_t mask = 0; 389 unsigned TV = 0; 390 int PtrArgNum = -1; 391 bool HasConstPtr = false; 392 switch (BuiltinID) { 393#define GET_NEON_AARCH64_OVERLOAD_CHECK 394#include "clang/Basic/arm_neon.inc" 395#undef GET_NEON_AARCH64_OVERLOAD_CHECK 396 } 397 398 // For NEON intrinsics which are overloaded on vector element type, validate 399 // the immediate which specifies which variant to emit. 400 unsigned ImmArg = TheCall->getNumArgs() - 1; 401 if (mask) { 402 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 403 return true; 404 405 TV = Result.getLimitedValue(64); 406 if ((TV > 63) || (mask & (1ULL << TV)) == 0) 407 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 408 << TheCall->getArg(ImmArg)->getSourceRange(); 409 } 410 411 if (PtrArgNum >= 0) { 412 // Check that pointer arguments have the specified type. 413 Expr *Arg = TheCall->getArg(PtrArgNum); 414 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 415 Arg = ICE->getSubExpr(); 416 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 417 QualType RHSTy = RHS.get()->getType(); 418 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 419 if (HasConstPtr) 420 EltTy = EltTy.withConst(); 421 QualType LHSTy = Context.getPointerType(EltTy); 422 AssignConvertType ConvTy; 423 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 424 if (RHS.isInvalid()) 425 return true; 426 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 427 RHS.get(), AA_Assigning)) 428 return true; 429 } 430 431 // For NEON intrinsics which take an immediate value as part of the 432 // instruction, range check them here. 433 unsigned i = 0, l = 0, u = 0; 434 switch (BuiltinID) { 435 default: 436 return false; 437#define GET_NEON_AARCH64_IMMEDIATE_CHECK 438#include "clang/Basic/arm_neon.inc" 439#undef GET_NEON_AARCH64_IMMEDIATE_CHECK 440 } 441 ; 442 443 // We can't check the value of a dependent argument. 444 if (TheCall->getArg(i)->isTypeDependent() || 445 TheCall->getArg(i)->isValueDependent()) 446 return false; 447 448 // Check that the immediate argument is actually a constant. 449 if (SemaBuiltinConstantArg(TheCall, i, Result)) 450 return true; 451 452 // Range check against the upper/lower values for this isntruction. 453 unsigned Val = Result.getZExtValue(); 454 if (Val < l || Val > (u + l)) 455 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 456 << l << u + l << TheCall->getArg(i)->getSourceRange(); 457 458 return false; 459} 460 461bool Sema::CheckARMBuiltinExclusiveCall(unsigned BuiltinID, CallExpr *TheCall) { 462 assert((BuiltinID == ARM::BI__builtin_arm_ldrex || 463 BuiltinID == ARM::BI__builtin_arm_strex) && 464 "unexpected ARM builtin"); 465 bool IsLdrex = BuiltinID == ARM::BI__builtin_arm_ldrex; 466 467 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 468 469 // Ensure that we have the proper number of arguments. 470 if (checkArgCount(*this, TheCall, IsLdrex ? 1 : 2)) 471 return true; 472 473 // Inspect the pointer argument of the atomic builtin. This should always be 474 // a pointer type, whose element is an integral scalar or pointer type. 475 // Because it is a pointer type, we don't have to worry about any implicit 476 // casts here. 477 Expr *PointerArg = TheCall->getArg(IsLdrex ? 0 : 1); 478 ExprResult PointerArgRes = DefaultFunctionArrayLvalueConversion(PointerArg); 479 if (PointerArgRes.isInvalid()) 480 return true; 481 PointerArg = PointerArgRes.take(); 482 483 const PointerType *pointerType = PointerArg->getType()->getAs<PointerType>(); 484 if (!pointerType) { 485 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 486 << PointerArg->getType() << PointerArg->getSourceRange(); 487 return true; 488 } 489 490 // ldrex takes a "const volatile T*" and strex takes a "volatile T*". Our next 491 // task is to insert the appropriate casts into the AST. First work out just 492 // what the appropriate type is. 493 QualType ValType = pointerType->getPointeeType(); 494 QualType AddrType = ValType.getUnqualifiedType().withVolatile(); 495 if (IsLdrex) 496 AddrType.addConst(); 497 498 // Issue a warning if the cast is dodgy. 499 CastKind CastNeeded = CK_NoOp; 500 if (!AddrType.isAtLeastAsQualifiedAs(ValType)) { 501 CastNeeded = CK_BitCast; 502 Diag(DRE->getLocStart(), diag::ext_typecheck_convert_discards_qualifiers) 503 << PointerArg->getType() 504 << Context.getPointerType(AddrType) 505 << AA_Passing << PointerArg->getSourceRange(); 506 } 507 508 // Finally, do the cast and replace the argument with the corrected version. 509 AddrType = Context.getPointerType(AddrType); 510 PointerArgRes = ImpCastExprToType(PointerArg, AddrType, CastNeeded); 511 if (PointerArgRes.isInvalid()) 512 return true; 513 PointerArg = PointerArgRes.take(); 514 515 TheCall->setArg(IsLdrex ? 0 : 1, PointerArg); 516 517 // In general, we allow ints, floats and pointers to be loaded and stored. 518 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 519 !ValType->isBlockPointerType() && !ValType->isFloatingType()) { 520 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intfltptr) 521 << PointerArg->getType() << PointerArg->getSourceRange(); 522 return true; 523 } 524 525 // But ARM doesn't have instructions to deal with 128-bit versions. 526 if (Context.getTypeSize(ValType) > 64) { 527 Diag(DRE->getLocStart(), diag::err_atomic_exclusive_builtin_pointer_size) 528 << PointerArg->getType() << PointerArg->getSourceRange(); 529 return true; 530 } 531 532 switch (ValType.getObjCLifetime()) { 533 case Qualifiers::OCL_None: 534 case Qualifiers::OCL_ExplicitNone: 535 // okay 536 break; 537 538 case Qualifiers::OCL_Weak: 539 case Qualifiers::OCL_Strong: 540 case Qualifiers::OCL_Autoreleasing: 541 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 542 << ValType << PointerArg->getSourceRange(); 543 return true; 544 } 545 546 547 if (IsLdrex) { 548 TheCall->setType(ValType); 549 return false; 550 } 551 552 // Initialize the argument to be stored. 553 ExprResult ValArg = TheCall->getArg(0); 554 InitializedEntity Entity = InitializedEntity::InitializeParameter( 555 Context, ValType, /*consume*/ false); 556 ValArg = PerformCopyInitialization(Entity, SourceLocation(), ValArg); 557 if (ValArg.isInvalid()) 558 return true; 559 560 TheCall->setArg(0, ValArg.get()); 561 return false; 562} 563 564bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 565 llvm::APSInt Result; 566 567 if (BuiltinID == ARM::BI__builtin_arm_ldrex || 568 BuiltinID == ARM::BI__builtin_arm_strex) { 569 return CheckARMBuiltinExclusiveCall(BuiltinID, TheCall); 570 } 571 572 uint64_t mask = 0; 573 unsigned TV = 0; 574 int PtrArgNum = -1; 575 bool HasConstPtr = false; 576 switch (BuiltinID) { 577#define GET_NEON_OVERLOAD_CHECK 578#include "clang/Basic/arm_neon.inc" 579#undef GET_NEON_OVERLOAD_CHECK 580 } 581 582 // For NEON intrinsics which are overloaded on vector element type, validate 583 // the immediate which specifies which variant to emit. 584 unsigned ImmArg = TheCall->getNumArgs()-1; 585 if (mask) { 586 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 587 return true; 588 589 TV = Result.getLimitedValue(64); 590 if ((TV > 63) || (mask & (1ULL << TV)) == 0) 591 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 592 << TheCall->getArg(ImmArg)->getSourceRange(); 593 } 594 595 if (PtrArgNum >= 0) { 596 // Check that pointer arguments have the specified type. 597 Expr *Arg = TheCall->getArg(PtrArgNum); 598 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 599 Arg = ICE->getSubExpr(); 600 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 601 QualType RHSTy = RHS.get()->getType(); 602 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 603 if (HasConstPtr) 604 EltTy = EltTy.withConst(); 605 QualType LHSTy = Context.getPointerType(EltTy); 606 AssignConvertType ConvTy; 607 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 608 if (RHS.isInvalid()) 609 return true; 610 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 611 RHS.get(), AA_Assigning)) 612 return true; 613 } 614 615 // For NEON intrinsics which take an immediate value as part of the 616 // instruction, range check them here. 617 unsigned i = 0, l = 0, u = 0; 618 switch (BuiltinID) { 619 default: return false; 620 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 621 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 622 case ARM::BI__builtin_arm_vcvtr_f: 623 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 624#define GET_NEON_IMMEDIATE_CHECK 625#include "clang/Basic/arm_neon.inc" 626#undef GET_NEON_IMMEDIATE_CHECK 627 }; 628 629 // We can't check the value of a dependent argument. 630 if (TheCall->getArg(i)->isTypeDependent() || 631 TheCall->getArg(i)->isValueDependent()) 632 return false; 633 634 // Check that the immediate argument is actually a constant. 635 if (SemaBuiltinConstantArg(TheCall, i, Result)) 636 return true; 637 638 // Range check against the upper/lower values for this isntruction. 639 unsigned Val = Result.getZExtValue(); 640 if (Val < l || Val > (u + l)) 641 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 642 << l << u+l << TheCall->getArg(i)->getSourceRange(); 643 644 // FIXME: VFP Intrinsics should error if VFP not present. 645 return false; 646} 647 648bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 649 unsigned i = 0, l = 0, u = 0; 650 switch (BuiltinID) { 651 default: return false; 652 case Mips::BI__builtin_mips_wrdsp: i = 1; l = 0; u = 63; break; 653 case Mips::BI__builtin_mips_rddsp: i = 0; l = 0; u = 63; break; 654 case Mips::BI__builtin_mips_append: i = 2; l = 0; u = 31; break; 655 case Mips::BI__builtin_mips_balign: i = 2; l = 0; u = 3; break; 656 case Mips::BI__builtin_mips_precr_sra_ph_w: i = 2; l = 0; u = 31; break; 657 case Mips::BI__builtin_mips_precr_sra_r_ph_w: i = 2; l = 0; u = 31; break; 658 case Mips::BI__builtin_mips_prepend: i = 2; l = 0; u = 31; break; 659 }; 660 661 // We can't check the value of a dependent argument. 662 if (TheCall->getArg(i)->isTypeDependent() || 663 TheCall->getArg(i)->isValueDependent()) 664 return false; 665 666 // Check that the immediate argument is actually a constant. 667 llvm::APSInt Result; 668 if (SemaBuiltinConstantArg(TheCall, i, Result)) 669 return true; 670 671 // Range check against the upper/lower values for this instruction. 672 unsigned Val = Result.getZExtValue(); 673 if (Val < l || Val > u) 674 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 675 << l << u << TheCall->getArg(i)->getSourceRange(); 676 677 return false; 678} 679 680/// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo 681/// parameter with the FormatAttr's correct format_idx and firstDataArg. 682/// Returns true when the format fits the function and the FormatStringInfo has 683/// been populated. 684bool Sema::getFormatStringInfo(const FormatAttr *Format, bool IsCXXMember, 685 FormatStringInfo *FSI) { 686 FSI->HasVAListArg = Format->getFirstArg() == 0; 687 FSI->FormatIdx = Format->getFormatIdx() - 1; 688 FSI->FirstDataArg = FSI->HasVAListArg ? 0 : Format->getFirstArg() - 1; 689 690 // The way the format attribute works in GCC, the implicit this argument 691 // of member functions is counted. However, it doesn't appear in our own 692 // lists, so decrement format_idx in that case. 693 if (IsCXXMember) { 694 if(FSI->FormatIdx == 0) 695 return false; 696 --FSI->FormatIdx; 697 if (FSI->FirstDataArg != 0) 698 --FSI->FirstDataArg; 699 } 700 return true; 701} 702 703/// Handles the checks for format strings, non-POD arguments to vararg 704/// functions, and NULL arguments passed to non-NULL parameters. 705void Sema::checkCall(NamedDecl *FDecl, 706 ArrayRef<const Expr *> Args, 707 unsigned NumProtoArgs, 708 bool IsMemberFunction, 709 SourceLocation Loc, 710 SourceRange Range, 711 VariadicCallType CallType) { 712 // FIXME: We should check as much as we can in the template definition. 713 if (CurContext->isDependentContext()) 714 return; 715 716 // Printf and scanf checking. 717 llvm::SmallBitVector CheckedVarArgs; 718 if (FDecl) { 719 for (specific_attr_iterator<FormatAttr> 720 I = FDecl->specific_attr_begin<FormatAttr>(), 721 E = FDecl->specific_attr_end<FormatAttr>(); 722 I != E; ++I) { 723 // Only create vector if there are format attributes. 724 CheckedVarArgs.resize(Args.size()); 725 726 CheckFormatArguments(*I, Args, IsMemberFunction, CallType, Loc, Range, 727 CheckedVarArgs); 728 } 729 } 730 731 // Refuse POD arguments that weren't caught by the format string 732 // checks above. 733 if (CallType != VariadicDoesNotApply) { 734 for (unsigned ArgIdx = NumProtoArgs; ArgIdx < Args.size(); ++ArgIdx) { 735 // Args[ArgIdx] can be null in malformed code. 736 if (const Expr *Arg = Args[ArgIdx]) { 737 if (CheckedVarArgs.empty() || !CheckedVarArgs[ArgIdx]) 738 checkVariadicArgument(Arg, CallType); 739 } 740 } 741 } 742 743 if (FDecl) { 744 for (specific_attr_iterator<NonNullAttr> 745 I = FDecl->specific_attr_begin<NonNullAttr>(), 746 E = FDecl->specific_attr_end<NonNullAttr>(); I != E; ++I) 747 CheckNonNullArguments(*I, Args.data(), Loc); 748 749 // Type safety checking. 750 for (specific_attr_iterator<ArgumentWithTypeTagAttr> 751 i = FDecl->specific_attr_begin<ArgumentWithTypeTagAttr>(), 752 e = FDecl->specific_attr_end<ArgumentWithTypeTagAttr>(); 753 i != e; ++i) { 754 CheckArgumentWithTypeTag(*i, Args.data()); 755 } 756 } 757} 758 759/// CheckConstructorCall - Check a constructor call for correctness and safety 760/// properties not enforced by the C type system. 761void Sema::CheckConstructorCall(FunctionDecl *FDecl, 762 ArrayRef<const Expr *> Args, 763 const FunctionProtoType *Proto, 764 SourceLocation Loc) { 765 VariadicCallType CallType = 766 Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply; 767 checkCall(FDecl, Args, Proto->getNumArgs(), 768 /*IsMemberFunction=*/true, Loc, SourceRange(), CallType); 769} 770 771/// CheckFunctionCall - Check a direct function call for various correctness 772/// and safety properties not strictly enforced by the C type system. 773bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall, 774 const FunctionProtoType *Proto) { 775 bool IsMemberOperatorCall = isa<CXXOperatorCallExpr>(TheCall) && 776 isa<CXXMethodDecl>(FDecl); 777 bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall) || 778 IsMemberOperatorCall; 779 VariadicCallType CallType = getVariadicCallType(FDecl, Proto, 780 TheCall->getCallee()); 781 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 782 Expr** Args = TheCall->getArgs(); 783 unsigned NumArgs = TheCall->getNumArgs(); 784 if (IsMemberOperatorCall) { 785 // If this is a call to a member operator, hide the first argument 786 // from checkCall. 787 // FIXME: Our choice of AST representation here is less than ideal. 788 ++Args; 789 --NumArgs; 790 } 791 checkCall(FDecl, llvm::makeArrayRef<const Expr *>(Args, NumArgs), 792 NumProtoArgs, 793 IsMemberFunction, TheCall->getRParenLoc(), 794 TheCall->getCallee()->getSourceRange(), CallType); 795 796 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 797 // None of the checks below are needed for functions that don't have 798 // simple names (e.g., C++ conversion functions). 799 if (!FnInfo) 800 return false; 801 802 unsigned CMId = FDecl->getMemoryFunctionKind(); 803 if (CMId == 0) 804 return false; 805 806 // Handle memory setting and copying functions. 807 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 808 CheckStrlcpycatArguments(TheCall, FnInfo); 809 else if (CMId == Builtin::BIstrncat) 810 CheckStrncatArguments(TheCall, FnInfo); 811 else 812 CheckMemaccessArguments(TheCall, CMId, FnInfo); 813 814 return false; 815} 816 817bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 818 ArrayRef<const Expr *> Args) { 819 VariadicCallType CallType = 820 Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply; 821 822 checkCall(Method, Args, Method->param_size(), 823 /*IsMemberFunction=*/false, 824 lbrac, Method->getSourceRange(), CallType); 825 826 return false; 827} 828 829bool Sema::CheckPointerCall(NamedDecl *NDecl, CallExpr *TheCall, 830 const FunctionProtoType *Proto) { 831 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 832 if (!V) 833 return false; 834 835 QualType Ty = V->getType(); 836 if (!Ty->isBlockPointerType() && !Ty->isFunctionPointerType()) 837 return false; 838 839 VariadicCallType CallType; 840 if (!Proto || !Proto->isVariadic()) { 841 CallType = VariadicDoesNotApply; 842 } else if (Ty->isBlockPointerType()) { 843 CallType = VariadicBlock; 844 } else { // Ty->isFunctionPointerType() 845 CallType = VariadicFunction; 846 } 847 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 848 849 checkCall(NDecl, 850 llvm::makeArrayRef<const Expr *>(TheCall->getArgs(), 851 TheCall->getNumArgs()), 852 NumProtoArgs, /*IsMemberFunction=*/false, 853 TheCall->getRParenLoc(), 854 TheCall->getCallee()->getSourceRange(), CallType); 855 856 return false; 857} 858 859/// Checks function calls when a FunctionDecl or a NamedDecl is not available, 860/// such as function pointers returned from functions. 861bool Sema::CheckOtherCall(CallExpr *TheCall, const FunctionProtoType *Proto) { 862 VariadicCallType CallType = getVariadicCallType(/*FDecl=*/0, Proto, 863 TheCall->getCallee()); 864 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 865 866 checkCall(/*FDecl=*/0, 867 llvm::makeArrayRef<const Expr *>(TheCall->getArgs(), 868 TheCall->getNumArgs()), 869 NumProtoArgs, /*IsMemberFunction=*/false, 870 TheCall->getRParenLoc(), 871 TheCall->getCallee()->getSourceRange(), CallType); 872 873 return false; 874} 875 876ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 877 AtomicExpr::AtomicOp Op) { 878 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 879 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 880 881 // All these operations take one of the following forms: 882 enum { 883 // C __c11_atomic_init(A *, C) 884 Init, 885 // C __c11_atomic_load(A *, int) 886 Load, 887 // void __atomic_load(A *, CP, int) 888 Copy, 889 // C __c11_atomic_add(A *, M, int) 890 Arithmetic, 891 // C __atomic_exchange_n(A *, CP, int) 892 Xchg, 893 // void __atomic_exchange(A *, C *, CP, int) 894 GNUXchg, 895 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 896 C11CmpXchg, 897 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 898 GNUCmpXchg 899 } Form = Init; 900 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 901 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 902 // where: 903 // C is an appropriate type, 904 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 905 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 906 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 907 // the int parameters are for orderings. 908 909 assert(AtomicExpr::AO__c11_atomic_init == 0 && 910 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 911 && "need to update code for modified C11 atomics"); 912 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 913 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 914 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 915 Op == AtomicExpr::AO__atomic_store_n || 916 Op == AtomicExpr::AO__atomic_exchange_n || 917 Op == AtomicExpr::AO__atomic_compare_exchange_n; 918 bool IsAddSub = false; 919 920 switch (Op) { 921 case AtomicExpr::AO__c11_atomic_init: 922 Form = Init; 923 break; 924 925 case AtomicExpr::AO__c11_atomic_load: 926 case AtomicExpr::AO__atomic_load_n: 927 Form = Load; 928 break; 929 930 case AtomicExpr::AO__c11_atomic_store: 931 case AtomicExpr::AO__atomic_load: 932 case AtomicExpr::AO__atomic_store: 933 case AtomicExpr::AO__atomic_store_n: 934 Form = Copy; 935 break; 936 937 case AtomicExpr::AO__c11_atomic_fetch_add: 938 case AtomicExpr::AO__c11_atomic_fetch_sub: 939 case AtomicExpr::AO__atomic_fetch_add: 940 case AtomicExpr::AO__atomic_fetch_sub: 941 case AtomicExpr::AO__atomic_add_fetch: 942 case AtomicExpr::AO__atomic_sub_fetch: 943 IsAddSub = true; 944 // Fall through. 945 case AtomicExpr::AO__c11_atomic_fetch_and: 946 case AtomicExpr::AO__c11_atomic_fetch_or: 947 case AtomicExpr::AO__c11_atomic_fetch_xor: 948 case AtomicExpr::AO__atomic_fetch_and: 949 case AtomicExpr::AO__atomic_fetch_or: 950 case AtomicExpr::AO__atomic_fetch_xor: 951 case AtomicExpr::AO__atomic_fetch_nand: 952 case AtomicExpr::AO__atomic_and_fetch: 953 case AtomicExpr::AO__atomic_or_fetch: 954 case AtomicExpr::AO__atomic_xor_fetch: 955 case AtomicExpr::AO__atomic_nand_fetch: 956 Form = Arithmetic; 957 break; 958 959 case AtomicExpr::AO__c11_atomic_exchange: 960 case AtomicExpr::AO__atomic_exchange_n: 961 Form = Xchg; 962 break; 963 964 case AtomicExpr::AO__atomic_exchange: 965 Form = GNUXchg; 966 break; 967 968 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 969 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 970 Form = C11CmpXchg; 971 break; 972 973 case AtomicExpr::AO__atomic_compare_exchange: 974 case AtomicExpr::AO__atomic_compare_exchange_n: 975 Form = GNUCmpXchg; 976 break; 977 } 978 979 // Check we have the right number of arguments. 980 if (TheCall->getNumArgs() < NumArgs[Form]) { 981 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 982 << 0 << NumArgs[Form] << TheCall->getNumArgs() 983 << TheCall->getCallee()->getSourceRange(); 984 return ExprError(); 985 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 986 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 987 diag::err_typecheck_call_too_many_args) 988 << 0 << NumArgs[Form] << TheCall->getNumArgs() 989 << TheCall->getCallee()->getSourceRange(); 990 return ExprError(); 991 } 992 993 // Inspect the first argument of the atomic operation. 994 Expr *Ptr = TheCall->getArg(0); 995 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 996 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 997 if (!pointerType) { 998 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 999 << Ptr->getType() << Ptr->getSourceRange(); 1000 return ExprError(); 1001 } 1002 1003 // For a __c11 builtin, this should be a pointer to an _Atomic type. 1004 QualType AtomTy = pointerType->getPointeeType(); // 'A' 1005 QualType ValType = AtomTy; // 'C' 1006 if (IsC11) { 1007 if (!AtomTy->isAtomicType()) { 1008 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 1009 << Ptr->getType() << Ptr->getSourceRange(); 1010 return ExprError(); 1011 } 1012 if (AtomTy.isConstQualified()) { 1013 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_non_const_atomic) 1014 << Ptr->getType() << Ptr->getSourceRange(); 1015 return ExprError(); 1016 } 1017 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 1018 } 1019 1020 // For an arithmetic operation, the implied arithmetic must be well-formed. 1021 if (Form == Arithmetic) { 1022 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 1023 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 1024 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 1025 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 1026 return ExprError(); 1027 } 1028 if (!IsAddSub && !ValType->isIntegerType()) { 1029 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 1030 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 1031 return ExprError(); 1032 } 1033 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 1034 // For __atomic_*_n operations, the value type must be a scalar integral or 1035 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 1036 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 1037 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 1038 return ExprError(); 1039 } 1040 1041 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context) && 1042 !AtomTy->isScalarType()) { 1043 // For GNU atomics, require a trivially-copyable type. This is not part of 1044 // the GNU atomics specification, but we enforce it for sanity. 1045 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 1046 << Ptr->getType() << Ptr->getSourceRange(); 1047 return ExprError(); 1048 } 1049 1050 // FIXME: For any builtin other than a load, the ValType must not be 1051 // const-qualified. 1052 1053 switch (ValType.getObjCLifetime()) { 1054 case Qualifiers::OCL_None: 1055 case Qualifiers::OCL_ExplicitNone: 1056 // okay 1057 break; 1058 1059 case Qualifiers::OCL_Weak: 1060 case Qualifiers::OCL_Strong: 1061 case Qualifiers::OCL_Autoreleasing: 1062 // FIXME: Can this happen? By this point, ValType should be known 1063 // to be trivially copyable. 1064 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1065 << ValType << Ptr->getSourceRange(); 1066 return ExprError(); 1067 } 1068 1069 QualType ResultType = ValType; 1070 if (Form == Copy || Form == GNUXchg || Form == Init) 1071 ResultType = Context.VoidTy; 1072 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 1073 ResultType = Context.BoolTy; 1074 1075 // The type of a parameter passed 'by value'. In the GNU atomics, such 1076 // arguments are actually passed as pointers. 1077 QualType ByValType = ValType; // 'CP' 1078 if (!IsC11 && !IsN) 1079 ByValType = Ptr->getType(); 1080 1081 // The first argument --- the pointer --- has a fixed type; we 1082 // deduce the types of the rest of the arguments accordingly. Walk 1083 // the remaining arguments, converting them to the deduced value type. 1084 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 1085 QualType Ty; 1086 if (i < NumVals[Form] + 1) { 1087 switch (i) { 1088 case 1: 1089 // The second argument is the non-atomic operand. For arithmetic, this 1090 // is always passed by value, and for a compare_exchange it is always 1091 // passed by address. For the rest, GNU uses by-address and C11 uses 1092 // by-value. 1093 assert(Form != Load); 1094 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 1095 Ty = ValType; 1096 else if (Form == Copy || Form == Xchg) 1097 Ty = ByValType; 1098 else if (Form == Arithmetic) 1099 Ty = Context.getPointerDiffType(); 1100 else 1101 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 1102 break; 1103 case 2: 1104 // The third argument to compare_exchange / GNU exchange is a 1105 // (pointer to a) desired value. 1106 Ty = ByValType; 1107 break; 1108 case 3: 1109 // The fourth argument to GNU compare_exchange is a 'weak' flag. 1110 Ty = Context.BoolTy; 1111 break; 1112 } 1113 } else { 1114 // The order(s) are always converted to int. 1115 Ty = Context.IntTy; 1116 } 1117 1118 InitializedEntity Entity = 1119 InitializedEntity::InitializeParameter(Context, Ty, false); 1120 ExprResult Arg = TheCall->getArg(i); 1121 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1122 if (Arg.isInvalid()) 1123 return true; 1124 TheCall->setArg(i, Arg.get()); 1125 } 1126 1127 // Permute the arguments into a 'consistent' order. 1128 SmallVector<Expr*, 5> SubExprs; 1129 SubExprs.push_back(Ptr); 1130 switch (Form) { 1131 case Init: 1132 // Note, AtomicExpr::getVal1() has a special case for this atomic. 1133 SubExprs.push_back(TheCall->getArg(1)); // Val1 1134 break; 1135 case Load: 1136 SubExprs.push_back(TheCall->getArg(1)); // Order 1137 break; 1138 case Copy: 1139 case Arithmetic: 1140 case Xchg: 1141 SubExprs.push_back(TheCall->getArg(2)); // Order 1142 SubExprs.push_back(TheCall->getArg(1)); // Val1 1143 break; 1144 case GNUXchg: 1145 // Note, AtomicExpr::getVal2() has a special case for this atomic. 1146 SubExprs.push_back(TheCall->getArg(3)); // Order 1147 SubExprs.push_back(TheCall->getArg(1)); // Val1 1148 SubExprs.push_back(TheCall->getArg(2)); // Val2 1149 break; 1150 case C11CmpXchg: 1151 SubExprs.push_back(TheCall->getArg(3)); // Order 1152 SubExprs.push_back(TheCall->getArg(1)); // Val1 1153 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 1154 SubExprs.push_back(TheCall->getArg(2)); // Val2 1155 break; 1156 case GNUCmpXchg: 1157 SubExprs.push_back(TheCall->getArg(4)); // Order 1158 SubExprs.push_back(TheCall->getArg(1)); // Val1 1159 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 1160 SubExprs.push_back(TheCall->getArg(2)); // Val2 1161 SubExprs.push_back(TheCall->getArg(3)); // Weak 1162 break; 1163 } 1164 1165 AtomicExpr *AE = new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 1166 SubExprs, ResultType, Op, 1167 TheCall->getRParenLoc()); 1168 1169 if ((Op == AtomicExpr::AO__c11_atomic_load || 1170 (Op == AtomicExpr::AO__c11_atomic_store)) && 1171 Context.AtomicUsesUnsupportedLibcall(AE)) 1172 Diag(AE->getLocStart(), diag::err_atomic_load_store_uses_lib) << 1173 ((Op == AtomicExpr::AO__c11_atomic_load) ? 0 : 1); 1174 1175 return Owned(AE); 1176} 1177 1178 1179/// checkBuiltinArgument - Given a call to a builtin function, perform 1180/// normal type-checking on the given argument, updating the call in 1181/// place. This is useful when a builtin function requires custom 1182/// type-checking for some of its arguments but not necessarily all of 1183/// them. 1184/// 1185/// Returns true on error. 1186static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 1187 FunctionDecl *Fn = E->getDirectCallee(); 1188 assert(Fn && "builtin call without direct callee!"); 1189 1190 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 1191 InitializedEntity Entity = 1192 InitializedEntity::InitializeParameter(S.Context, Param); 1193 1194 ExprResult Arg = E->getArg(0); 1195 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 1196 if (Arg.isInvalid()) 1197 return true; 1198 1199 E->setArg(ArgIndex, Arg.take()); 1200 return false; 1201} 1202 1203/// SemaBuiltinAtomicOverloaded - We have a call to a function like 1204/// __sync_fetch_and_add, which is an overloaded function based on the pointer 1205/// type of its first argument. The main ActOnCallExpr routines have already 1206/// promoted the types of arguments because all of these calls are prototyped as 1207/// void(...). 1208/// 1209/// This function goes through and does final semantic checking for these 1210/// builtins, 1211ExprResult 1212Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 1213 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 1214 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1215 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1216 1217 // Ensure that we have at least one argument to do type inference from. 1218 if (TheCall->getNumArgs() < 1) { 1219 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1220 << 0 << 1 << TheCall->getNumArgs() 1221 << TheCall->getCallee()->getSourceRange(); 1222 return ExprError(); 1223 } 1224 1225 // Inspect the first argument of the atomic builtin. This should always be 1226 // a pointer type, whose element is an integral scalar or pointer type. 1227 // Because it is a pointer type, we don't have to worry about any implicit 1228 // casts here. 1229 // FIXME: We don't allow floating point scalars as input. 1230 Expr *FirstArg = TheCall->getArg(0); 1231 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 1232 if (FirstArgResult.isInvalid()) 1233 return ExprError(); 1234 FirstArg = FirstArgResult.take(); 1235 TheCall->setArg(0, FirstArg); 1236 1237 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 1238 if (!pointerType) { 1239 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 1240 << FirstArg->getType() << FirstArg->getSourceRange(); 1241 return ExprError(); 1242 } 1243 1244 QualType ValType = pointerType->getPointeeType(); 1245 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 1246 !ValType->isBlockPointerType()) { 1247 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 1248 << FirstArg->getType() << FirstArg->getSourceRange(); 1249 return ExprError(); 1250 } 1251 1252 switch (ValType.getObjCLifetime()) { 1253 case Qualifiers::OCL_None: 1254 case Qualifiers::OCL_ExplicitNone: 1255 // okay 1256 break; 1257 1258 case Qualifiers::OCL_Weak: 1259 case Qualifiers::OCL_Strong: 1260 case Qualifiers::OCL_Autoreleasing: 1261 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1262 << ValType << FirstArg->getSourceRange(); 1263 return ExprError(); 1264 } 1265 1266 // Strip any qualifiers off ValType. 1267 ValType = ValType.getUnqualifiedType(); 1268 1269 // The majority of builtins return a value, but a few have special return 1270 // types, so allow them to override appropriately below. 1271 QualType ResultType = ValType; 1272 1273 // We need to figure out which concrete builtin this maps onto. For example, 1274 // __sync_fetch_and_add with a 2 byte object turns into 1275 // __sync_fetch_and_add_2. 1276#define BUILTIN_ROW(x) \ 1277 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 1278 Builtin::BI##x##_8, Builtin::BI##x##_16 } 1279 1280 static const unsigned BuiltinIndices[][5] = { 1281 BUILTIN_ROW(__sync_fetch_and_add), 1282 BUILTIN_ROW(__sync_fetch_and_sub), 1283 BUILTIN_ROW(__sync_fetch_and_or), 1284 BUILTIN_ROW(__sync_fetch_and_and), 1285 BUILTIN_ROW(__sync_fetch_and_xor), 1286 1287 BUILTIN_ROW(__sync_add_and_fetch), 1288 BUILTIN_ROW(__sync_sub_and_fetch), 1289 BUILTIN_ROW(__sync_and_and_fetch), 1290 BUILTIN_ROW(__sync_or_and_fetch), 1291 BUILTIN_ROW(__sync_xor_and_fetch), 1292 1293 BUILTIN_ROW(__sync_val_compare_and_swap), 1294 BUILTIN_ROW(__sync_bool_compare_and_swap), 1295 BUILTIN_ROW(__sync_lock_test_and_set), 1296 BUILTIN_ROW(__sync_lock_release), 1297 BUILTIN_ROW(__sync_swap) 1298 }; 1299#undef BUILTIN_ROW 1300 1301 // Determine the index of the size. 1302 unsigned SizeIndex; 1303 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 1304 case 1: SizeIndex = 0; break; 1305 case 2: SizeIndex = 1; break; 1306 case 4: SizeIndex = 2; break; 1307 case 8: SizeIndex = 3; break; 1308 case 16: SizeIndex = 4; break; 1309 default: 1310 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 1311 << FirstArg->getType() << FirstArg->getSourceRange(); 1312 return ExprError(); 1313 } 1314 1315 // Each of these builtins has one pointer argument, followed by some number of 1316 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 1317 // that we ignore. Find out which row of BuiltinIndices to read from as well 1318 // as the number of fixed args. 1319 unsigned BuiltinID = FDecl->getBuiltinID(); 1320 unsigned BuiltinIndex, NumFixed = 1; 1321 switch (BuiltinID) { 1322 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 1323 case Builtin::BI__sync_fetch_and_add: 1324 case Builtin::BI__sync_fetch_and_add_1: 1325 case Builtin::BI__sync_fetch_and_add_2: 1326 case Builtin::BI__sync_fetch_and_add_4: 1327 case Builtin::BI__sync_fetch_and_add_8: 1328 case Builtin::BI__sync_fetch_and_add_16: 1329 BuiltinIndex = 0; 1330 break; 1331 1332 case Builtin::BI__sync_fetch_and_sub: 1333 case Builtin::BI__sync_fetch_and_sub_1: 1334 case Builtin::BI__sync_fetch_and_sub_2: 1335 case Builtin::BI__sync_fetch_and_sub_4: 1336 case Builtin::BI__sync_fetch_and_sub_8: 1337 case Builtin::BI__sync_fetch_and_sub_16: 1338 BuiltinIndex = 1; 1339 break; 1340 1341 case Builtin::BI__sync_fetch_and_or: 1342 case Builtin::BI__sync_fetch_and_or_1: 1343 case Builtin::BI__sync_fetch_and_or_2: 1344 case Builtin::BI__sync_fetch_and_or_4: 1345 case Builtin::BI__sync_fetch_and_or_8: 1346 case Builtin::BI__sync_fetch_and_or_16: 1347 BuiltinIndex = 2; 1348 break; 1349 1350 case Builtin::BI__sync_fetch_and_and: 1351 case Builtin::BI__sync_fetch_and_and_1: 1352 case Builtin::BI__sync_fetch_and_and_2: 1353 case Builtin::BI__sync_fetch_and_and_4: 1354 case Builtin::BI__sync_fetch_and_and_8: 1355 case Builtin::BI__sync_fetch_and_and_16: 1356 BuiltinIndex = 3; 1357 break; 1358 1359 case Builtin::BI__sync_fetch_and_xor: 1360 case Builtin::BI__sync_fetch_and_xor_1: 1361 case Builtin::BI__sync_fetch_and_xor_2: 1362 case Builtin::BI__sync_fetch_and_xor_4: 1363 case Builtin::BI__sync_fetch_and_xor_8: 1364 case Builtin::BI__sync_fetch_and_xor_16: 1365 BuiltinIndex = 4; 1366 break; 1367 1368 case Builtin::BI__sync_add_and_fetch: 1369 case Builtin::BI__sync_add_and_fetch_1: 1370 case Builtin::BI__sync_add_and_fetch_2: 1371 case Builtin::BI__sync_add_and_fetch_4: 1372 case Builtin::BI__sync_add_and_fetch_8: 1373 case Builtin::BI__sync_add_and_fetch_16: 1374 BuiltinIndex = 5; 1375 break; 1376 1377 case Builtin::BI__sync_sub_and_fetch: 1378 case Builtin::BI__sync_sub_and_fetch_1: 1379 case Builtin::BI__sync_sub_and_fetch_2: 1380 case Builtin::BI__sync_sub_and_fetch_4: 1381 case Builtin::BI__sync_sub_and_fetch_8: 1382 case Builtin::BI__sync_sub_and_fetch_16: 1383 BuiltinIndex = 6; 1384 break; 1385 1386 case Builtin::BI__sync_and_and_fetch: 1387 case Builtin::BI__sync_and_and_fetch_1: 1388 case Builtin::BI__sync_and_and_fetch_2: 1389 case Builtin::BI__sync_and_and_fetch_4: 1390 case Builtin::BI__sync_and_and_fetch_8: 1391 case Builtin::BI__sync_and_and_fetch_16: 1392 BuiltinIndex = 7; 1393 break; 1394 1395 case Builtin::BI__sync_or_and_fetch: 1396 case Builtin::BI__sync_or_and_fetch_1: 1397 case Builtin::BI__sync_or_and_fetch_2: 1398 case Builtin::BI__sync_or_and_fetch_4: 1399 case Builtin::BI__sync_or_and_fetch_8: 1400 case Builtin::BI__sync_or_and_fetch_16: 1401 BuiltinIndex = 8; 1402 break; 1403 1404 case Builtin::BI__sync_xor_and_fetch: 1405 case Builtin::BI__sync_xor_and_fetch_1: 1406 case Builtin::BI__sync_xor_and_fetch_2: 1407 case Builtin::BI__sync_xor_and_fetch_4: 1408 case Builtin::BI__sync_xor_and_fetch_8: 1409 case Builtin::BI__sync_xor_and_fetch_16: 1410 BuiltinIndex = 9; 1411 break; 1412 1413 case Builtin::BI__sync_val_compare_and_swap: 1414 case Builtin::BI__sync_val_compare_and_swap_1: 1415 case Builtin::BI__sync_val_compare_and_swap_2: 1416 case Builtin::BI__sync_val_compare_and_swap_4: 1417 case Builtin::BI__sync_val_compare_and_swap_8: 1418 case Builtin::BI__sync_val_compare_and_swap_16: 1419 BuiltinIndex = 10; 1420 NumFixed = 2; 1421 break; 1422 1423 case Builtin::BI__sync_bool_compare_and_swap: 1424 case Builtin::BI__sync_bool_compare_and_swap_1: 1425 case Builtin::BI__sync_bool_compare_and_swap_2: 1426 case Builtin::BI__sync_bool_compare_and_swap_4: 1427 case Builtin::BI__sync_bool_compare_and_swap_8: 1428 case Builtin::BI__sync_bool_compare_and_swap_16: 1429 BuiltinIndex = 11; 1430 NumFixed = 2; 1431 ResultType = Context.BoolTy; 1432 break; 1433 1434 case Builtin::BI__sync_lock_test_and_set: 1435 case Builtin::BI__sync_lock_test_and_set_1: 1436 case Builtin::BI__sync_lock_test_and_set_2: 1437 case Builtin::BI__sync_lock_test_and_set_4: 1438 case Builtin::BI__sync_lock_test_and_set_8: 1439 case Builtin::BI__sync_lock_test_and_set_16: 1440 BuiltinIndex = 12; 1441 break; 1442 1443 case Builtin::BI__sync_lock_release: 1444 case Builtin::BI__sync_lock_release_1: 1445 case Builtin::BI__sync_lock_release_2: 1446 case Builtin::BI__sync_lock_release_4: 1447 case Builtin::BI__sync_lock_release_8: 1448 case Builtin::BI__sync_lock_release_16: 1449 BuiltinIndex = 13; 1450 NumFixed = 0; 1451 ResultType = Context.VoidTy; 1452 break; 1453 1454 case Builtin::BI__sync_swap: 1455 case Builtin::BI__sync_swap_1: 1456 case Builtin::BI__sync_swap_2: 1457 case Builtin::BI__sync_swap_4: 1458 case Builtin::BI__sync_swap_8: 1459 case Builtin::BI__sync_swap_16: 1460 BuiltinIndex = 14; 1461 break; 1462 } 1463 1464 // Now that we know how many fixed arguments we expect, first check that we 1465 // have at least that many. 1466 if (TheCall->getNumArgs() < 1+NumFixed) { 1467 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1468 << 0 << 1+NumFixed << TheCall->getNumArgs() 1469 << TheCall->getCallee()->getSourceRange(); 1470 return ExprError(); 1471 } 1472 1473 // Get the decl for the concrete builtin from this, we can tell what the 1474 // concrete integer type we should convert to is. 1475 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1476 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1477 FunctionDecl *NewBuiltinDecl; 1478 if (NewBuiltinID == BuiltinID) 1479 NewBuiltinDecl = FDecl; 1480 else { 1481 // Perform builtin lookup to avoid redeclaring it. 1482 DeclarationName DN(&Context.Idents.get(NewBuiltinName)); 1483 LookupResult Res(*this, DN, DRE->getLocStart(), LookupOrdinaryName); 1484 LookupName(Res, TUScope, /*AllowBuiltinCreation=*/true); 1485 assert(Res.getFoundDecl()); 1486 NewBuiltinDecl = dyn_cast<FunctionDecl>(Res.getFoundDecl()); 1487 if (NewBuiltinDecl == 0) 1488 return ExprError(); 1489 } 1490 1491 // The first argument --- the pointer --- has a fixed type; we 1492 // deduce the types of the rest of the arguments accordingly. Walk 1493 // the remaining arguments, converting them to the deduced value type. 1494 for (unsigned i = 0; i != NumFixed; ++i) { 1495 ExprResult Arg = TheCall->getArg(i+1); 1496 1497 // GCC does an implicit conversion to the pointer or integer ValType. This 1498 // can fail in some cases (1i -> int**), check for this error case now. 1499 // Initialize the argument. 1500 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1501 ValType, /*consume*/ false); 1502 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1503 if (Arg.isInvalid()) 1504 return ExprError(); 1505 1506 // Okay, we have something that *can* be converted to the right type. Check 1507 // to see if there is a potentially weird extension going on here. This can 1508 // happen when you do an atomic operation on something like an char* and 1509 // pass in 42. The 42 gets converted to char. This is even more strange 1510 // for things like 45.123 -> char, etc. 1511 // FIXME: Do this check. 1512 TheCall->setArg(i+1, Arg.take()); 1513 } 1514 1515 ASTContext& Context = this->getASTContext(); 1516 1517 // Create a new DeclRefExpr to refer to the new decl. 1518 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1519 Context, 1520 DRE->getQualifierLoc(), 1521 SourceLocation(), 1522 NewBuiltinDecl, 1523 /*enclosing*/ false, 1524 DRE->getLocation(), 1525 Context.BuiltinFnTy, 1526 DRE->getValueKind()); 1527 1528 // Set the callee in the CallExpr. 1529 // FIXME: This loses syntactic information. 1530 QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType()); 1531 ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy, 1532 CK_BuiltinFnToFnPtr); 1533 TheCall->setCallee(PromotedCall.take()); 1534 1535 // Change the result type of the call to match the original value type. This 1536 // is arbitrary, but the codegen for these builtins ins design to handle it 1537 // gracefully. 1538 TheCall->setType(ResultType); 1539 1540 return TheCallResult; 1541} 1542 1543/// CheckObjCString - Checks that the argument to the builtin 1544/// CFString constructor is correct 1545/// Note: It might also make sense to do the UTF-16 conversion here (would 1546/// simplify the backend). 1547bool Sema::CheckObjCString(Expr *Arg) { 1548 Arg = Arg->IgnoreParenCasts(); 1549 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1550 1551 if (!Literal || !Literal->isAscii()) { 1552 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1553 << Arg->getSourceRange(); 1554 return true; 1555 } 1556 1557 if (Literal->containsNonAsciiOrNull()) { 1558 StringRef String = Literal->getString(); 1559 unsigned NumBytes = String.size(); 1560 SmallVector<UTF16, 128> ToBuf(NumBytes); 1561 const UTF8 *FromPtr = (const UTF8 *)String.data(); 1562 UTF16 *ToPtr = &ToBuf[0]; 1563 1564 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1565 &ToPtr, ToPtr + NumBytes, 1566 strictConversion); 1567 // Check for conversion failure. 1568 if (Result != conversionOK) 1569 Diag(Arg->getLocStart(), 1570 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1571 } 1572 return false; 1573} 1574 1575/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1576/// Emit an error and return true on failure, return false on success. 1577bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1578 Expr *Fn = TheCall->getCallee(); 1579 if (TheCall->getNumArgs() > 2) { 1580 Diag(TheCall->getArg(2)->getLocStart(), 1581 diag::err_typecheck_call_too_many_args) 1582 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1583 << Fn->getSourceRange() 1584 << SourceRange(TheCall->getArg(2)->getLocStart(), 1585 (*(TheCall->arg_end()-1))->getLocEnd()); 1586 return true; 1587 } 1588 1589 if (TheCall->getNumArgs() < 2) { 1590 return Diag(TheCall->getLocEnd(), 1591 diag::err_typecheck_call_too_few_args_at_least) 1592 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1593 } 1594 1595 // Type-check the first argument normally. 1596 if (checkBuiltinArgument(*this, TheCall, 0)) 1597 return true; 1598 1599 // Determine whether the current function is variadic or not. 1600 BlockScopeInfo *CurBlock = getCurBlock(); 1601 bool isVariadic; 1602 if (CurBlock) 1603 isVariadic = CurBlock->TheDecl->isVariadic(); 1604 else if (FunctionDecl *FD = getCurFunctionDecl()) 1605 isVariadic = FD->isVariadic(); 1606 else 1607 isVariadic = getCurMethodDecl()->isVariadic(); 1608 1609 if (!isVariadic) { 1610 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1611 return true; 1612 } 1613 1614 // Verify that the second argument to the builtin is the last argument of the 1615 // current function or method. 1616 bool SecondArgIsLastNamedArgument = false; 1617 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1618 1619 // These are valid if SecondArgIsLastNamedArgument is false after the next 1620 // block. 1621 QualType Type; 1622 SourceLocation ParamLoc; 1623 1624 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1625 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1626 // FIXME: This isn't correct for methods (results in bogus warning). 1627 // Get the last formal in the current function. 1628 const ParmVarDecl *LastArg; 1629 if (CurBlock) 1630 LastArg = *(CurBlock->TheDecl->param_end()-1); 1631 else if (FunctionDecl *FD = getCurFunctionDecl()) 1632 LastArg = *(FD->param_end()-1); 1633 else 1634 LastArg = *(getCurMethodDecl()->param_end()-1); 1635 SecondArgIsLastNamedArgument = PV == LastArg; 1636 1637 Type = PV->getType(); 1638 ParamLoc = PV->getLocation(); 1639 } 1640 } 1641 1642 if (!SecondArgIsLastNamedArgument) 1643 Diag(TheCall->getArg(1)->getLocStart(), 1644 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1645 else if (Type->isReferenceType()) { 1646 Diag(Arg->getLocStart(), 1647 diag::warn_va_start_of_reference_type_is_undefined); 1648 Diag(ParamLoc, diag::note_parameter_type) << Type; 1649 } 1650 1651 return false; 1652} 1653 1654/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1655/// friends. This is declared to take (...), so we have to check everything. 1656bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1657 if (TheCall->getNumArgs() < 2) 1658 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1659 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1660 if (TheCall->getNumArgs() > 2) 1661 return Diag(TheCall->getArg(2)->getLocStart(), 1662 diag::err_typecheck_call_too_many_args) 1663 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1664 << SourceRange(TheCall->getArg(2)->getLocStart(), 1665 (*(TheCall->arg_end()-1))->getLocEnd()); 1666 1667 ExprResult OrigArg0 = TheCall->getArg(0); 1668 ExprResult OrigArg1 = TheCall->getArg(1); 1669 1670 // Do standard promotions between the two arguments, returning their common 1671 // type. 1672 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1673 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1674 return true; 1675 1676 // Make sure any conversions are pushed back into the call; this is 1677 // type safe since unordered compare builtins are declared as "_Bool 1678 // foo(...)". 1679 TheCall->setArg(0, OrigArg0.get()); 1680 TheCall->setArg(1, OrigArg1.get()); 1681 1682 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1683 return false; 1684 1685 // If the common type isn't a real floating type, then the arguments were 1686 // invalid for this operation. 1687 if (Res.isNull() || !Res->isRealFloatingType()) 1688 return Diag(OrigArg0.get()->getLocStart(), 1689 diag::err_typecheck_call_invalid_ordered_compare) 1690 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1691 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1692 1693 return false; 1694} 1695 1696/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1697/// __builtin_isnan and friends. This is declared to take (...), so we have 1698/// to check everything. We expect the last argument to be a floating point 1699/// value. 1700bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1701 if (TheCall->getNumArgs() < NumArgs) 1702 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1703 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1704 if (TheCall->getNumArgs() > NumArgs) 1705 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1706 diag::err_typecheck_call_too_many_args) 1707 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1708 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1709 (*(TheCall->arg_end()-1))->getLocEnd()); 1710 1711 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1712 1713 if (OrigArg->isTypeDependent()) 1714 return false; 1715 1716 // This operation requires a non-_Complex floating-point number. 1717 if (!OrigArg->getType()->isRealFloatingType()) 1718 return Diag(OrigArg->getLocStart(), 1719 diag::err_typecheck_call_invalid_unary_fp) 1720 << OrigArg->getType() << OrigArg->getSourceRange(); 1721 1722 // If this is an implicit conversion from float -> double, remove it. 1723 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1724 Expr *CastArg = Cast->getSubExpr(); 1725 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1726 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1727 "promotion from float to double is the only expected cast here"); 1728 Cast->setSubExpr(0); 1729 TheCall->setArg(NumArgs-1, CastArg); 1730 } 1731 } 1732 1733 return false; 1734} 1735 1736/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1737// This is declared to take (...), so we have to check everything. 1738ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1739 if (TheCall->getNumArgs() < 2) 1740 return ExprError(Diag(TheCall->getLocEnd(), 1741 diag::err_typecheck_call_too_few_args_at_least) 1742 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1743 << TheCall->getSourceRange()); 1744 1745 // Determine which of the following types of shufflevector we're checking: 1746 // 1) unary, vector mask: (lhs, mask) 1747 // 2) binary, vector mask: (lhs, rhs, mask) 1748 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1749 QualType resType = TheCall->getArg(0)->getType(); 1750 unsigned numElements = 0; 1751 1752 if (!TheCall->getArg(0)->isTypeDependent() && 1753 !TheCall->getArg(1)->isTypeDependent()) { 1754 QualType LHSType = TheCall->getArg(0)->getType(); 1755 QualType RHSType = TheCall->getArg(1)->getType(); 1756 1757 if (!LHSType->isVectorType() || !RHSType->isVectorType()) 1758 return ExprError(Diag(TheCall->getLocStart(), 1759 diag::err_shufflevector_non_vector) 1760 << SourceRange(TheCall->getArg(0)->getLocStart(), 1761 TheCall->getArg(1)->getLocEnd())); 1762 1763 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1764 unsigned numResElements = TheCall->getNumArgs() - 2; 1765 1766 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1767 // with mask. If so, verify that RHS is an integer vector type with the 1768 // same number of elts as lhs. 1769 if (TheCall->getNumArgs() == 2) { 1770 if (!RHSType->hasIntegerRepresentation() || 1771 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1772 return ExprError(Diag(TheCall->getLocStart(), 1773 diag::err_shufflevector_incompatible_vector) 1774 << SourceRange(TheCall->getArg(1)->getLocStart(), 1775 TheCall->getArg(1)->getLocEnd())); 1776 } else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1777 return ExprError(Diag(TheCall->getLocStart(), 1778 diag::err_shufflevector_incompatible_vector) 1779 << SourceRange(TheCall->getArg(0)->getLocStart(), 1780 TheCall->getArg(1)->getLocEnd())); 1781 } else if (numElements != numResElements) { 1782 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1783 resType = Context.getVectorType(eltType, numResElements, 1784 VectorType::GenericVector); 1785 } 1786 } 1787 1788 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1789 if (TheCall->getArg(i)->isTypeDependent() || 1790 TheCall->getArg(i)->isValueDependent()) 1791 continue; 1792 1793 llvm::APSInt Result(32); 1794 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1795 return ExprError(Diag(TheCall->getLocStart(), 1796 diag::err_shufflevector_nonconstant_argument) 1797 << TheCall->getArg(i)->getSourceRange()); 1798 1799 // Allow -1 which will be translated to undef in the IR. 1800 if (Result.isSigned() && Result.isAllOnesValue()) 1801 continue; 1802 1803 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1804 return ExprError(Diag(TheCall->getLocStart(), 1805 diag::err_shufflevector_argument_too_large) 1806 << TheCall->getArg(i)->getSourceRange()); 1807 } 1808 1809 SmallVector<Expr*, 32> exprs; 1810 1811 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1812 exprs.push_back(TheCall->getArg(i)); 1813 TheCall->setArg(i, 0); 1814 } 1815 1816 return Owned(new (Context) ShuffleVectorExpr(Context, exprs, resType, 1817 TheCall->getCallee()->getLocStart(), 1818 TheCall->getRParenLoc())); 1819} 1820 1821/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1822// This is declared to take (const void*, ...) and can take two 1823// optional constant int args. 1824bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1825 unsigned NumArgs = TheCall->getNumArgs(); 1826 1827 if (NumArgs > 3) 1828 return Diag(TheCall->getLocEnd(), 1829 diag::err_typecheck_call_too_many_args_at_most) 1830 << 0 /*function call*/ << 3 << NumArgs 1831 << TheCall->getSourceRange(); 1832 1833 // Argument 0 is checked for us and the remaining arguments must be 1834 // constant integers. 1835 for (unsigned i = 1; i != NumArgs; ++i) { 1836 Expr *Arg = TheCall->getArg(i); 1837 1838 // We can't check the value of a dependent argument. 1839 if (Arg->isTypeDependent() || Arg->isValueDependent()) 1840 continue; 1841 1842 llvm::APSInt Result; 1843 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1844 return true; 1845 1846 // FIXME: gcc issues a warning and rewrites these to 0. These 1847 // seems especially odd for the third argument since the default 1848 // is 3. 1849 if (i == 1) { 1850 if (Result.getLimitedValue() > 1) 1851 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1852 << "0" << "1" << Arg->getSourceRange(); 1853 } else { 1854 if (Result.getLimitedValue() > 3) 1855 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1856 << "0" << "3" << Arg->getSourceRange(); 1857 } 1858 } 1859 1860 return false; 1861} 1862 1863/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1864/// TheCall is a constant expression. 1865bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1866 llvm::APSInt &Result) { 1867 Expr *Arg = TheCall->getArg(ArgNum); 1868 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1869 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1870 1871 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1872 1873 if (!Arg->isIntegerConstantExpr(Result, Context)) 1874 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1875 << FDecl->getDeclName() << Arg->getSourceRange(); 1876 1877 return false; 1878} 1879 1880/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1881/// int type). This simply type checks that type is one of the defined 1882/// constants (0-3). 1883// For compatibility check 0-3, llvm only handles 0 and 2. 1884bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1885 llvm::APSInt Result; 1886 1887 // We can't check the value of a dependent argument. 1888 if (TheCall->getArg(1)->isTypeDependent() || 1889 TheCall->getArg(1)->isValueDependent()) 1890 return false; 1891 1892 // Check constant-ness first. 1893 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1894 return true; 1895 1896 Expr *Arg = TheCall->getArg(1); 1897 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1898 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1899 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1900 } 1901 1902 return false; 1903} 1904 1905/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1906/// This checks that val is a constant 1. 1907bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1908 Expr *Arg = TheCall->getArg(1); 1909 llvm::APSInt Result; 1910 1911 // TODO: This is less than ideal. Overload this to take a value. 1912 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1913 return true; 1914 1915 if (Result != 1) 1916 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1917 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1918 1919 return false; 1920} 1921 1922namespace { 1923enum StringLiteralCheckType { 1924 SLCT_NotALiteral, 1925 SLCT_UncheckedLiteral, 1926 SLCT_CheckedLiteral 1927}; 1928} 1929 1930// Determine if an expression is a string literal or constant string. 1931// If this function returns false on the arguments to a function expecting a 1932// format string, we will usually need to emit a warning. 1933// True string literals are then checked by CheckFormatString. 1934static StringLiteralCheckType 1935checkFormatStringExpr(Sema &S, const Expr *E, ArrayRef<const Expr *> Args, 1936 bool HasVAListArg, unsigned format_idx, 1937 unsigned firstDataArg, Sema::FormatStringType Type, 1938 Sema::VariadicCallType CallType, bool InFunctionCall, 1939 llvm::SmallBitVector &CheckedVarArgs) { 1940 tryAgain: 1941 if (E->isTypeDependent() || E->isValueDependent()) 1942 return SLCT_NotALiteral; 1943 1944 E = E->IgnoreParenCasts(); 1945 1946 if (E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull)) 1947 // Technically -Wformat-nonliteral does not warn about this case. 1948 // The behavior of printf and friends in this case is implementation 1949 // dependent. Ideally if the format string cannot be null then 1950 // it should have a 'nonnull' attribute in the function prototype. 1951 return SLCT_UncheckedLiteral; 1952 1953 switch (E->getStmtClass()) { 1954 case Stmt::BinaryConditionalOperatorClass: 1955 case Stmt::ConditionalOperatorClass: { 1956 // The expression is a literal if both sub-expressions were, and it was 1957 // completely checked only if both sub-expressions were checked. 1958 const AbstractConditionalOperator *C = 1959 cast<AbstractConditionalOperator>(E); 1960 StringLiteralCheckType Left = 1961 checkFormatStringExpr(S, C->getTrueExpr(), Args, 1962 HasVAListArg, format_idx, firstDataArg, 1963 Type, CallType, InFunctionCall, CheckedVarArgs); 1964 if (Left == SLCT_NotALiteral) 1965 return SLCT_NotALiteral; 1966 StringLiteralCheckType Right = 1967 checkFormatStringExpr(S, C->getFalseExpr(), Args, 1968 HasVAListArg, format_idx, firstDataArg, 1969 Type, CallType, InFunctionCall, CheckedVarArgs); 1970 return Left < Right ? Left : Right; 1971 } 1972 1973 case Stmt::ImplicitCastExprClass: { 1974 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1975 goto tryAgain; 1976 } 1977 1978 case Stmt::OpaqueValueExprClass: 1979 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1980 E = src; 1981 goto tryAgain; 1982 } 1983 return SLCT_NotALiteral; 1984 1985 case Stmt::PredefinedExprClass: 1986 // While __func__, etc., are technically not string literals, they 1987 // cannot contain format specifiers and thus are not a security 1988 // liability. 1989 return SLCT_UncheckedLiteral; 1990 1991 case Stmt::DeclRefExprClass: { 1992 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1993 1994 // As an exception, do not flag errors for variables binding to 1995 // const string literals. 1996 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1997 bool isConstant = false; 1998 QualType T = DR->getType(); 1999 2000 if (const ArrayType *AT = S.Context.getAsArrayType(T)) { 2001 isConstant = AT->getElementType().isConstant(S.Context); 2002 } else if (const PointerType *PT = T->getAs<PointerType>()) { 2003 isConstant = T.isConstant(S.Context) && 2004 PT->getPointeeType().isConstant(S.Context); 2005 } else if (T->isObjCObjectPointerType()) { 2006 // In ObjC, there is usually no "const ObjectPointer" type, 2007 // so don't check if the pointee type is constant. 2008 isConstant = T.isConstant(S.Context); 2009 } 2010 2011 if (isConstant) { 2012 if (const Expr *Init = VD->getAnyInitializer()) { 2013 // Look through initializers like const char c[] = { "foo" } 2014 if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) { 2015 if (InitList->isStringLiteralInit()) 2016 Init = InitList->getInit(0)->IgnoreParenImpCasts(); 2017 } 2018 return checkFormatStringExpr(S, Init, Args, 2019 HasVAListArg, format_idx, 2020 firstDataArg, Type, CallType, 2021 /*InFunctionCall*/false, CheckedVarArgs); 2022 } 2023 } 2024 2025 // For vprintf* functions (i.e., HasVAListArg==true), we add a 2026 // special check to see if the format string is a function parameter 2027 // of the function calling the printf function. If the function 2028 // has an attribute indicating it is a printf-like function, then we 2029 // should suppress warnings concerning non-literals being used in a call 2030 // to a vprintf function. For example: 2031 // 2032 // void 2033 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 2034 // va_list ap; 2035 // va_start(ap, fmt); 2036 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 2037 // ... 2038 // } 2039 if (HasVAListArg) { 2040 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 2041 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 2042 int PVIndex = PV->getFunctionScopeIndex() + 1; 2043 for (specific_attr_iterator<FormatAttr> 2044 i = ND->specific_attr_begin<FormatAttr>(), 2045 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 2046 FormatAttr *PVFormat = *i; 2047 // adjust for implicit parameter 2048 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 2049 if (MD->isInstance()) 2050 ++PVIndex; 2051 // We also check if the formats are compatible. 2052 // We can't pass a 'scanf' string to a 'printf' function. 2053 if (PVIndex == PVFormat->getFormatIdx() && 2054 Type == S.GetFormatStringType(PVFormat)) 2055 return SLCT_UncheckedLiteral; 2056 } 2057 } 2058 } 2059 } 2060 } 2061 2062 return SLCT_NotALiteral; 2063 } 2064 2065 case Stmt::CallExprClass: 2066 case Stmt::CXXMemberCallExprClass: { 2067 const CallExpr *CE = cast<CallExpr>(E); 2068 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 2069 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 2070 unsigned ArgIndex = FA->getFormatIdx(); 2071 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 2072 if (MD->isInstance()) 2073 --ArgIndex; 2074 const Expr *Arg = CE->getArg(ArgIndex - 1); 2075 2076 return checkFormatStringExpr(S, Arg, Args, 2077 HasVAListArg, format_idx, firstDataArg, 2078 Type, CallType, InFunctionCall, 2079 CheckedVarArgs); 2080 } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) { 2081 unsigned BuiltinID = FD->getBuiltinID(); 2082 if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString || 2083 BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) { 2084 const Expr *Arg = CE->getArg(0); 2085 return checkFormatStringExpr(S, Arg, Args, 2086 HasVAListArg, format_idx, 2087 firstDataArg, Type, CallType, 2088 InFunctionCall, CheckedVarArgs); 2089 } 2090 } 2091 } 2092 2093 return SLCT_NotALiteral; 2094 } 2095 case Stmt::ObjCStringLiteralClass: 2096 case Stmt::StringLiteralClass: { 2097 const StringLiteral *StrE = NULL; 2098 2099 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 2100 StrE = ObjCFExpr->getString(); 2101 else 2102 StrE = cast<StringLiteral>(E); 2103 2104 if (StrE) { 2105 S.CheckFormatString(StrE, E, Args, HasVAListArg, format_idx, firstDataArg, 2106 Type, InFunctionCall, CallType, CheckedVarArgs); 2107 return SLCT_CheckedLiteral; 2108 } 2109 2110 return SLCT_NotALiteral; 2111 } 2112 2113 default: 2114 return SLCT_NotALiteral; 2115 } 2116} 2117 2118void 2119Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 2120 const Expr * const *ExprArgs, 2121 SourceLocation CallSiteLoc) { 2122 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 2123 e = NonNull->args_end(); 2124 i != e; ++i) { 2125 const Expr *ArgExpr = ExprArgs[*i]; 2126 2127 // As a special case, transparent unions initialized with zero are 2128 // considered null for the purposes of the nonnull attribute. 2129 if (const RecordType *UT = ArgExpr->getType()->getAsUnionType()) { 2130 if (UT->getDecl()->hasAttr<TransparentUnionAttr>()) 2131 if (const CompoundLiteralExpr *CLE = 2132 dyn_cast<CompoundLiteralExpr>(ArgExpr)) 2133 if (const InitListExpr *ILE = 2134 dyn_cast<InitListExpr>(CLE->getInitializer())) 2135 ArgExpr = ILE->getInit(0); 2136 } 2137 2138 bool Result; 2139 if (ArgExpr->EvaluateAsBooleanCondition(Result, Context) && !Result) 2140 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 2141 } 2142} 2143 2144Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 2145 return llvm::StringSwitch<FormatStringType>(Format->getType()->getName()) 2146 .Case("scanf", FST_Scanf) 2147 .Cases("printf", "printf0", FST_Printf) 2148 .Cases("NSString", "CFString", FST_NSString) 2149 .Case("strftime", FST_Strftime) 2150 .Case("strfmon", FST_Strfmon) 2151 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 2152 .Default(FST_Unknown); 2153} 2154 2155/// CheckFormatArguments - Check calls to printf and scanf (and similar 2156/// functions) for correct use of format strings. 2157/// Returns true if a format string has been fully checked. 2158bool Sema::CheckFormatArguments(const FormatAttr *Format, 2159 ArrayRef<const Expr *> Args, 2160 bool IsCXXMember, 2161 VariadicCallType CallType, 2162 SourceLocation Loc, SourceRange Range, 2163 llvm::SmallBitVector &CheckedVarArgs) { 2164 FormatStringInfo FSI; 2165 if (getFormatStringInfo(Format, IsCXXMember, &FSI)) 2166 return CheckFormatArguments(Args, FSI.HasVAListArg, FSI.FormatIdx, 2167 FSI.FirstDataArg, GetFormatStringType(Format), 2168 CallType, Loc, Range, CheckedVarArgs); 2169 return false; 2170} 2171 2172bool Sema::CheckFormatArguments(ArrayRef<const Expr *> Args, 2173 bool HasVAListArg, unsigned format_idx, 2174 unsigned firstDataArg, FormatStringType Type, 2175 VariadicCallType CallType, 2176 SourceLocation Loc, SourceRange Range, 2177 llvm::SmallBitVector &CheckedVarArgs) { 2178 // CHECK: printf/scanf-like function is called with no format string. 2179 if (format_idx >= Args.size()) { 2180 Diag(Loc, diag::warn_missing_format_string) << Range; 2181 return false; 2182 } 2183 2184 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 2185 2186 // CHECK: format string is not a string literal. 2187 // 2188 // Dynamically generated format strings are difficult to 2189 // automatically vet at compile time. Requiring that format strings 2190 // are string literals: (1) permits the checking of format strings by 2191 // the compiler and thereby (2) can practically remove the source of 2192 // many format string exploits. 2193 2194 // Format string can be either ObjC string (e.g. @"%d") or 2195 // C string (e.g. "%d") 2196 // ObjC string uses the same format specifiers as C string, so we can use 2197 // the same format string checking logic for both ObjC and C strings. 2198 StringLiteralCheckType CT = 2199 checkFormatStringExpr(*this, OrigFormatExpr, Args, HasVAListArg, 2200 format_idx, firstDataArg, Type, CallType, 2201 /*IsFunctionCall*/true, CheckedVarArgs); 2202 if (CT != SLCT_NotALiteral) 2203 // Literal format string found, check done! 2204 return CT == SLCT_CheckedLiteral; 2205 2206 // Strftime is particular as it always uses a single 'time' argument, 2207 // so it is safe to pass a non-literal string. 2208 if (Type == FST_Strftime) 2209 return false; 2210 2211 // Do not emit diag when the string param is a macro expansion and the 2212 // format is either NSString or CFString. This is a hack to prevent 2213 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 2214 // which are usually used in place of NS and CF string literals. 2215 if (Type == FST_NSString && 2216 SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart())) 2217 return false; 2218 2219 // If there are no arguments specified, warn with -Wformat-security, otherwise 2220 // warn only with -Wformat-nonliteral. 2221 if (Args.size() == firstDataArg) 2222 Diag(Args[format_idx]->getLocStart(), 2223 diag::warn_format_nonliteral_noargs) 2224 << OrigFormatExpr->getSourceRange(); 2225 else 2226 Diag(Args[format_idx]->getLocStart(), 2227 diag::warn_format_nonliteral) 2228 << OrigFormatExpr->getSourceRange(); 2229 return false; 2230} 2231 2232namespace { 2233class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 2234protected: 2235 Sema &S; 2236 const StringLiteral *FExpr; 2237 const Expr *OrigFormatExpr; 2238 const unsigned FirstDataArg; 2239 const unsigned NumDataArgs; 2240 const char *Beg; // Start of format string. 2241 const bool HasVAListArg; 2242 ArrayRef<const Expr *> Args; 2243 unsigned FormatIdx; 2244 llvm::SmallBitVector CoveredArgs; 2245 bool usesPositionalArgs; 2246 bool atFirstArg; 2247 bool inFunctionCall; 2248 Sema::VariadicCallType CallType; 2249 llvm::SmallBitVector &CheckedVarArgs; 2250public: 2251 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 2252 const Expr *origFormatExpr, unsigned firstDataArg, 2253 unsigned numDataArgs, const char *beg, bool hasVAListArg, 2254 ArrayRef<const Expr *> Args, 2255 unsigned formatIdx, bool inFunctionCall, 2256 Sema::VariadicCallType callType, 2257 llvm::SmallBitVector &CheckedVarArgs) 2258 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 2259 FirstDataArg(firstDataArg), NumDataArgs(numDataArgs), 2260 Beg(beg), HasVAListArg(hasVAListArg), 2261 Args(Args), FormatIdx(formatIdx), 2262 usesPositionalArgs(false), atFirstArg(true), 2263 inFunctionCall(inFunctionCall), CallType(callType), 2264 CheckedVarArgs(CheckedVarArgs) { 2265 CoveredArgs.resize(numDataArgs); 2266 CoveredArgs.reset(); 2267 } 2268 2269 void DoneProcessing(); 2270 2271 void HandleIncompleteSpecifier(const char *startSpecifier, 2272 unsigned specifierLen); 2273 2274 void HandleInvalidLengthModifier( 2275 const analyze_format_string::FormatSpecifier &FS, 2276 const analyze_format_string::ConversionSpecifier &CS, 2277 const char *startSpecifier, unsigned specifierLen, unsigned DiagID); 2278 2279 void HandleNonStandardLengthModifier( 2280 const analyze_format_string::FormatSpecifier &FS, 2281 const char *startSpecifier, unsigned specifierLen); 2282 2283 void HandleNonStandardConversionSpecifier( 2284 const analyze_format_string::ConversionSpecifier &CS, 2285 const char *startSpecifier, unsigned specifierLen); 2286 2287 virtual void HandlePosition(const char *startPos, unsigned posLen); 2288 2289 virtual void HandleInvalidPosition(const char *startSpecifier, 2290 unsigned specifierLen, 2291 analyze_format_string::PositionContext p); 2292 2293 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 2294 2295 void HandleNullChar(const char *nullCharacter); 2296 2297 template <typename Range> 2298 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 2299 const Expr *ArgumentExpr, 2300 PartialDiagnostic PDiag, 2301 SourceLocation StringLoc, 2302 bool IsStringLocation, Range StringRange, 2303 ArrayRef<FixItHint> Fixit = None); 2304 2305protected: 2306 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 2307 const char *startSpec, 2308 unsigned specifierLen, 2309 const char *csStart, unsigned csLen); 2310 2311 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 2312 const char *startSpec, 2313 unsigned specifierLen); 2314 2315 SourceRange getFormatStringRange(); 2316 CharSourceRange getSpecifierRange(const char *startSpecifier, 2317 unsigned specifierLen); 2318 SourceLocation getLocationOfByte(const char *x); 2319 2320 const Expr *getDataArg(unsigned i) const; 2321 2322 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 2323 const analyze_format_string::ConversionSpecifier &CS, 2324 const char *startSpecifier, unsigned specifierLen, 2325 unsigned argIndex); 2326 2327 template <typename Range> 2328 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 2329 bool IsStringLocation, Range StringRange, 2330 ArrayRef<FixItHint> Fixit = None); 2331 2332 void CheckPositionalAndNonpositionalArgs( 2333 const analyze_format_string::FormatSpecifier *FS); 2334}; 2335} 2336 2337SourceRange CheckFormatHandler::getFormatStringRange() { 2338 return OrigFormatExpr->getSourceRange(); 2339} 2340 2341CharSourceRange CheckFormatHandler:: 2342getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 2343 SourceLocation Start = getLocationOfByte(startSpecifier); 2344 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 2345 2346 // Advance the end SourceLocation by one due to half-open ranges. 2347 End = End.getLocWithOffset(1); 2348 2349 return CharSourceRange::getCharRange(Start, End); 2350} 2351 2352SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 2353 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 2354} 2355 2356void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 2357 unsigned specifierLen){ 2358 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 2359 getLocationOfByte(startSpecifier), 2360 /*IsStringLocation*/true, 2361 getSpecifierRange(startSpecifier, specifierLen)); 2362} 2363 2364void CheckFormatHandler::HandleInvalidLengthModifier( 2365 const analyze_format_string::FormatSpecifier &FS, 2366 const analyze_format_string::ConversionSpecifier &CS, 2367 const char *startSpecifier, unsigned specifierLen, unsigned DiagID) { 2368 using namespace analyze_format_string; 2369 2370 const LengthModifier &LM = FS.getLengthModifier(); 2371 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2372 2373 // See if we know how to fix this length modifier. 2374 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2375 if (FixedLM) { 2376 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2377 getLocationOfByte(LM.getStart()), 2378 /*IsStringLocation*/true, 2379 getSpecifierRange(startSpecifier, specifierLen)); 2380 2381 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2382 << FixedLM->toString() 2383 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2384 2385 } else { 2386 FixItHint Hint; 2387 if (DiagID == diag::warn_format_nonsensical_length) 2388 Hint = FixItHint::CreateRemoval(LMRange); 2389 2390 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2391 getLocationOfByte(LM.getStart()), 2392 /*IsStringLocation*/true, 2393 getSpecifierRange(startSpecifier, specifierLen), 2394 Hint); 2395 } 2396} 2397 2398void CheckFormatHandler::HandleNonStandardLengthModifier( 2399 const analyze_format_string::FormatSpecifier &FS, 2400 const char *startSpecifier, unsigned specifierLen) { 2401 using namespace analyze_format_string; 2402 2403 const LengthModifier &LM = FS.getLengthModifier(); 2404 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2405 2406 // See if we know how to fix this length modifier. 2407 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2408 if (FixedLM) { 2409 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2410 << LM.toString() << 0, 2411 getLocationOfByte(LM.getStart()), 2412 /*IsStringLocation*/true, 2413 getSpecifierRange(startSpecifier, specifierLen)); 2414 2415 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2416 << FixedLM->toString() 2417 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2418 2419 } else { 2420 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2421 << LM.toString() << 0, 2422 getLocationOfByte(LM.getStart()), 2423 /*IsStringLocation*/true, 2424 getSpecifierRange(startSpecifier, specifierLen)); 2425 } 2426} 2427 2428void CheckFormatHandler::HandleNonStandardConversionSpecifier( 2429 const analyze_format_string::ConversionSpecifier &CS, 2430 const char *startSpecifier, unsigned specifierLen) { 2431 using namespace analyze_format_string; 2432 2433 // See if we know how to fix this conversion specifier. 2434 Optional<ConversionSpecifier> FixedCS = CS.getStandardSpecifier(); 2435 if (FixedCS) { 2436 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2437 << CS.toString() << /*conversion specifier*/1, 2438 getLocationOfByte(CS.getStart()), 2439 /*IsStringLocation*/true, 2440 getSpecifierRange(startSpecifier, specifierLen)); 2441 2442 CharSourceRange CSRange = getSpecifierRange(CS.getStart(), CS.getLength()); 2443 S.Diag(getLocationOfByte(CS.getStart()), diag::note_format_fix_specifier) 2444 << FixedCS->toString() 2445 << FixItHint::CreateReplacement(CSRange, FixedCS->toString()); 2446 } else { 2447 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2448 << CS.toString() << /*conversion specifier*/1, 2449 getLocationOfByte(CS.getStart()), 2450 /*IsStringLocation*/true, 2451 getSpecifierRange(startSpecifier, specifierLen)); 2452 } 2453} 2454 2455void CheckFormatHandler::HandlePosition(const char *startPos, 2456 unsigned posLen) { 2457 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 2458 getLocationOfByte(startPos), 2459 /*IsStringLocation*/true, 2460 getSpecifierRange(startPos, posLen)); 2461} 2462 2463void 2464CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 2465 analyze_format_string::PositionContext p) { 2466 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 2467 << (unsigned) p, 2468 getLocationOfByte(startPos), /*IsStringLocation*/true, 2469 getSpecifierRange(startPos, posLen)); 2470} 2471 2472void CheckFormatHandler::HandleZeroPosition(const char *startPos, 2473 unsigned posLen) { 2474 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 2475 getLocationOfByte(startPos), 2476 /*IsStringLocation*/true, 2477 getSpecifierRange(startPos, posLen)); 2478} 2479 2480void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 2481 if (!isa<ObjCStringLiteral>(OrigFormatExpr)) { 2482 // The presence of a null character is likely an error. 2483 EmitFormatDiagnostic( 2484 S.PDiag(diag::warn_printf_format_string_contains_null_char), 2485 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 2486 getFormatStringRange()); 2487 } 2488} 2489 2490// Note that this may return NULL if there was an error parsing or building 2491// one of the argument expressions. 2492const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 2493 return Args[FirstDataArg + i]; 2494} 2495 2496void CheckFormatHandler::DoneProcessing() { 2497 // Does the number of data arguments exceed the number of 2498 // format conversions in the format string? 2499 if (!HasVAListArg) { 2500 // Find any arguments that weren't covered. 2501 CoveredArgs.flip(); 2502 signed notCoveredArg = CoveredArgs.find_first(); 2503 if (notCoveredArg >= 0) { 2504 assert((unsigned)notCoveredArg < NumDataArgs); 2505 if (const Expr *E = getDataArg((unsigned) notCoveredArg)) { 2506 SourceLocation Loc = E->getLocStart(); 2507 if (!S.getSourceManager().isInSystemMacro(Loc)) { 2508 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 2509 Loc, /*IsStringLocation*/false, 2510 getFormatStringRange()); 2511 } 2512 } 2513 } 2514 } 2515} 2516 2517bool 2518CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 2519 SourceLocation Loc, 2520 const char *startSpec, 2521 unsigned specifierLen, 2522 const char *csStart, 2523 unsigned csLen) { 2524 2525 bool keepGoing = true; 2526 if (argIndex < NumDataArgs) { 2527 // Consider the argument coverered, even though the specifier doesn't 2528 // make sense. 2529 CoveredArgs.set(argIndex); 2530 } 2531 else { 2532 // If argIndex exceeds the number of data arguments we 2533 // don't issue a warning because that is just a cascade of warnings (and 2534 // they may have intended '%%' anyway). We don't want to continue processing 2535 // the format string after this point, however, as we will like just get 2536 // gibberish when trying to match arguments. 2537 keepGoing = false; 2538 } 2539 2540 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2541 << StringRef(csStart, csLen), 2542 Loc, /*IsStringLocation*/true, 2543 getSpecifierRange(startSpec, specifierLen)); 2544 2545 return keepGoing; 2546} 2547 2548void 2549CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2550 const char *startSpec, 2551 unsigned specifierLen) { 2552 EmitFormatDiagnostic( 2553 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2554 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2555} 2556 2557bool 2558CheckFormatHandler::CheckNumArgs( 2559 const analyze_format_string::FormatSpecifier &FS, 2560 const analyze_format_string::ConversionSpecifier &CS, 2561 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2562 2563 if (argIndex >= NumDataArgs) { 2564 PartialDiagnostic PDiag = FS.usesPositionalArg() 2565 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2566 << (argIndex+1) << NumDataArgs) 2567 : S.PDiag(diag::warn_printf_insufficient_data_args); 2568 EmitFormatDiagnostic( 2569 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2570 getSpecifierRange(startSpecifier, specifierLen)); 2571 return false; 2572 } 2573 return true; 2574} 2575 2576template<typename Range> 2577void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2578 SourceLocation Loc, 2579 bool IsStringLocation, 2580 Range StringRange, 2581 ArrayRef<FixItHint> FixIt) { 2582 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2583 Loc, IsStringLocation, StringRange, FixIt); 2584} 2585 2586/// \brief If the format string is not within the funcion call, emit a note 2587/// so that the function call and string are in diagnostic messages. 2588/// 2589/// \param InFunctionCall if true, the format string is within the function 2590/// call and only one diagnostic message will be produced. Otherwise, an 2591/// extra note will be emitted pointing to location of the format string. 2592/// 2593/// \param ArgumentExpr the expression that is passed as the format string 2594/// argument in the function call. Used for getting locations when two 2595/// diagnostics are emitted. 2596/// 2597/// \param PDiag the callee should already have provided any strings for the 2598/// diagnostic message. This function only adds locations and fixits 2599/// to diagnostics. 2600/// 2601/// \param Loc primary location for diagnostic. If two diagnostics are 2602/// required, one will be at Loc and a new SourceLocation will be created for 2603/// the other one. 2604/// 2605/// \param IsStringLocation if true, Loc points to the format string should be 2606/// used for the note. Otherwise, Loc points to the argument list and will 2607/// be used with PDiag. 2608/// 2609/// \param StringRange some or all of the string to highlight. This is 2610/// templated so it can accept either a CharSourceRange or a SourceRange. 2611/// 2612/// \param FixIt optional fix it hint for the format string. 2613template<typename Range> 2614void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2615 const Expr *ArgumentExpr, 2616 PartialDiagnostic PDiag, 2617 SourceLocation Loc, 2618 bool IsStringLocation, 2619 Range StringRange, 2620 ArrayRef<FixItHint> FixIt) { 2621 if (InFunctionCall) { 2622 const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag); 2623 D << StringRange; 2624 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2625 I != E; ++I) { 2626 D << *I; 2627 } 2628 } else { 2629 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2630 << ArgumentExpr->getSourceRange(); 2631 2632 const Sema::SemaDiagnosticBuilder &Note = 2633 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2634 diag::note_format_string_defined); 2635 2636 Note << StringRange; 2637 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2638 I != E; ++I) { 2639 Note << *I; 2640 } 2641 } 2642} 2643 2644//===--- CHECK: Printf format string checking ------------------------------===// 2645 2646namespace { 2647class CheckPrintfHandler : public CheckFormatHandler { 2648 bool ObjCContext; 2649public: 2650 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2651 const Expr *origFormatExpr, unsigned firstDataArg, 2652 unsigned numDataArgs, bool isObjC, 2653 const char *beg, bool hasVAListArg, 2654 ArrayRef<const Expr *> Args, 2655 unsigned formatIdx, bool inFunctionCall, 2656 Sema::VariadicCallType CallType, 2657 llvm::SmallBitVector &CheckedVarArgs) 2658 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2659 numDataArgs, beg, hasVAListArg, Args, 2660 formatIdx, inFunctionCall, CallType, CheckedVarArgs), 2661 ObjCContext(isObjC) 2662 {} 2663 2664 2665 bool HandleInvalidPrintfConversionSpecifier( 2666 const analyze_printf::PrintfSpecifier &FS, 2667 const char *startSpecifier, 2668 unsigned specifierLen); 2669 2670 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2671 const char *startSpecifier, 2672 unsigned specifierLen); 2673 bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2674 const char *StartSpecifier, 2675 unsigned SpecifierLen, 2676 const Expr *E); 2677 2678 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2679 const char *startSpecifier, unsigned specifierLen); 2680 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2681 const analyze_printf::OptionalAmount &Amt, 2682 unsigned type, 2683 const char *startSpecifier, unsigned specifierLen); 2684 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2685 const analyze_printf::OptionalFlag &flag, 2686 const char *startSpecifier, unsigned specifierLen); 2687 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2688 const analyze_printf::OptionalFlag &ignoredFlag, 2689 const analyze_printf::OptionalFlag &flag, 2690 const char *startSpecifier, unsigned specifierLen); 2691 bool checkForCStrMembers(const analyze_printf::ArgType &AT, 2692 const Expr *E, const CharSourceRange &CSR); 2693 2694}; 2695} 2696 2697bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2698 const analyze_printf::PrintfSpecifier &FS, 2699 const char *startSpecifier, 2700 unsigned specifierLen) { 2701 const analyze_printf::PrintfConversionSpecifier &CS = 2702 FS.getConversionSpecifier(); 2703 2704 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2705 getLocationOfByte(CS.getStart()), 2706 startSpecifier, specifierLen, 2707 CS.getStart(), CS.getLength()); 2708} 2709 2710bool CheckPrintfHandler::HandleAmount( 2711 const analyze_format_string::OptionalAmount &Amt, 2712 unsigned k, const char *startSpecifier, 2713 unsigned specifierLen) { 2714 2715 if (Amt.hasDataArgument()) { 2716 if (!HasVAListArg) { 2717 unsigned argIndex = Amt.getArgIndex(); 2718 if (argIndex >= NumDataArgs) { 2719 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2720 << k, 2721 getLocationOfByte(Amt.getStart()), 2722 /*IsStringLocation*/true, 2723 getSpecifierRange(startSpecifier, specifierLen)); 2724 // Don't do any more checking. We will just emit 2725 // spurious errors. 2726 return false; 2727 } 2728 2729 // Type check the data argument. It should be an 'int'. 2730 // Although not in conformance with C99, we also allow the argument to be 2731 // an 'unsigned int' as that is a reasonably safe case. GCC also 2732 // doesn't emit a warning for that case. 2733 CoveredArgs.set(argIndex); 2734 const Expr *Arg = getDataArg(argIndex); 2735 if (!Arg) 2736 return false; 2737 2738 QualType T = Arg->getType(); 2739 2740 const analyze_printf::ArgType &AT = Amt.getArgType(S.Context); 2741 assert(AT.isValid()); 2742 2743 if (!AT.matchesType(S.Context, T)) { 2744 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2745 << k << AT.getRepresentativeTypeName(S.Context) 2746 << T << Arg->getSourceRange(), 2747 getLocationOfByte(Amt.getStart()), 2748 /*IsStringLocation*/true, 2749 getSpecifierRange(startSpecifier, specifierLen)); 2750 // Don't do any more checking. We will just emit 2751 // spurious errors. 2752 return false; 2753 } 2754 } 2755 } 2756 return true; 2757} 2758 2759void CheckPrintfHandler::HandleInvalidAmount( 2760 const analyze_printf::PrintfSpecifier &FS, 2761 const analyze_printf::OptionalAmount &Amt, 2762 unsigned type, 2763 const char *startSpecifier, 2764 unsigned specifierLen) { 2765 const analyze_printf::PrintfConversionSpecifier &CS = 2766 FS.getConversionSpecifier(); 2767 2768 FixItHint fixit = 2769 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2770 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2771 Amt.getConstantLength())) 2772 : FixItHint(); 2773 2774 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2775 << type << CS.toString(), 2776 getLocationOfByte(Amt.getStart()), 2777 /*IsStringLocation*/true, 2778 getSpecifierRange(startSpecifier, specifierLen), 2779 fixit); 2780} 2781 2782void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2783 const analyze_printf::OptionalFlag &flag, 2784 const char *startSpecifier, 2785 unsigned specifierLen) { 2786 // Warn about pointless flag with a fixit removal. 2787 const analyze_printf::PrintfConversionSpecifier &CS = 2788 FS.getConversionSpecifier(); 2789 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2790 << flag.toString() << CS.toString(), 2791 getLocationOfByte(flag.getPosition()), 2792 /*IsStringLocation*/true, 2793 getSpecifierRange(startSpecifier, specifierLen), 2794 FixItHint::CreateRemoval( 2795 getSpecifierRange(flag.getPosition(), 1))); 2796} 2797 2798void CheckPrintfHandler::HandleIgnoredFlag( 2799 const analyze_printf::PrintfSpecifier &FS, 2800 const analyze_printf::OptionalFlag &ignoredFlag, 2801 const analyze_printf::OptionalFlag &flag, 2802 const char *startSpecifier, 2803 unsigned specifierLen) { 2804 // Warn about ignored flag with a fixit removal. 2805 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2806 << ignoredFlag.toString() << flag.toString(), 2807 getLocationOfByte(ignoredFlag.getPosition()), 2808 /*IsStringLocation*/true, 2809 getSpecifierRange(startSpecifier, specifierLen), 2810 FixItHint::CreateRemoval( 2811 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2812} 2813 2814// Determines if the specified is a C++ class or struct containing 2815// a member with the specified name and kind (e.g. a CXXMethodDecl named 2816// "c_str()"). 2817template<typename MemberKind> 2818static llvm::SmallPtrSet<MemberKind*, 1> 2819CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) { 2820 const RecordType *RT = Ty->getAs<RecordType>(); 2821 llvm::SmallPtrSet<MemberKind*, 1> Results; 2822 2823 if (!RT) 2824 return Results; 2825 const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl()); 2826 if (!RD) 2827 return Results; 2828 2829 LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(), 2830 Sema::LookupMemberName); 2831 2832 // We just need to include all members of the right kind turned up by the 2833 // filter, at this point. 2834 if (S.LookupQualifiedName(R, RT->getDecl())) 2835 for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) { 2836 NamedDecl *decl = (*I)->getUnderlyingDecl(); 2837 if (MemberKind *FK = dyn_cast<MemberKind>(decl)) 2838 Results.insert(FK); 2839 } 2840 return Results; 2841} 2842 2843// Check if a (w)string was passed when a (w)char* was needed, and offer a 2844// better diagnostic if so. AT is assumed to be valid. 2845// Returns true when a c_str() conversion method is found. 2846bool CheckPrintfHandler::checkForCStrMembers( 2847 const analyze_printf::ArgType &AT, const Expr *E, 2848 const CharSourceRange &CSR) { 2849 typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet; 2850 2851 MethodSet Results = 2852 CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType()); 2853 2854 for (MethodSet::iterator MI = Results.begin(), ME = Results.end(); 2855 MI != ME; ++MI) { 2856 const CXXMethodDecl *Method = *MI; 2857 if (Method->getNumParams() == 0 && 2858 AT.matchesType(S.Context, Method->getResultType())) { 2859 // FIXME: Suggest parens if the expression needs them. 2860 SourceLocation EndLoc = 2861 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()); 2862 S.Diag(E->getLocStart(), diag::note_printf_c_str) 2863 << "c_str()" 2864 << FixItHint::CreateInsertion(EndLoc, ".c_str()"); 2865 return true; 2866 } 2867 } 2868 2869 return false; 2870} 2871 2872bool 2873CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2874 &FS, 2875 const char *startSpecifier, 2876 unsigned specifierLen) { 2877 2878 using namespace analyze_format_string; 2879 using namespace analyze_printf; 2880 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2881 2882 if (FS.consumesDataArgument()) { 2883 if (atFirstArg) { 2884 atFirstArg = false; 2885 usesPositionalArgs = FS.usesPositionalArg(); 2886 } 2887 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2888 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2889 startSpecifier, specifierLen); 2890 return false; 2891 } 2892 } 2893 2894 // First check if the field width, precision, and conversion specifier 2895 // have matching data arguments. 2896 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2897 startSpecifier, specifierLen)) { 2898 return false; 2899 } 2900 2901 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2902 startSpecifier, specifierLen)) { 2903 return false; 2904 } 2905 2906 if (!CS.consumesDataArgument()) { 2907 // FIXME: Technically specifying a precision or field width here 2908 // makes no sense. Worth issuing a warning at some point. 2909 return true; 2910 } 2911 2912 // Consume the argument. 2913 unsigned argIndex = FS.getArgIndex(); 2914 if (argIndex < NumDataArgs) { 2915 // The check to see if the argIndex is valid will come later. 2916 // We set the bit here because we may exit early from this 2917 // function if we encounter some other error. 2918 CoveredArgs.set(argIndex); 2919 } 2920 2921 // Check for using an Objective-C specific conversion specifier 2922 // in a non-ObjC literal. 2923 if (!ObjCContext && CS.isObjCArg()) { 2924 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2925 specifierLen); 2926 } 2927 2928 // Check for invalid use of field width 2929 if (!FS.hasValidFieldWidth()) { 2930 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2931 startSpecifier, specifierLen); 2932 } 2933 2934 // Check for invalid use of precision 2935 if (!FS.hasValidPrecision()) { 2936 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2937 startSpecifier, specifierLen); 2938 } 2939 2940 // Check each flag does not conflict with any other component. 2941 if (!FS.hasValidThousandsGroupingPrefix()) 2942 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2943 if (!FS.hasValidLeadingZeros()) 2944 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2945 if (!FS.hasValidPlusPrefix()) 2946 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2947 if (!FS.hasValidSpacePrefix()) 2948 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2949 if (!FS.hasValidAlternativeForm()) 2950 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2951 if (!FS.hasValidLeftJustified()) 2952 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2953 2954 // Check that flags are not ignored by another flag 2955 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2956 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2957 startSpecifier, specifierLen); 2958 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2959 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2960 startSpecifier, specifierLen); 2961 2962 // Check the length modifier is valid with the given conversion specifier. 2963 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 2964 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2965 diag::warn_format_nonsensical_length); 2966 else if (!FS.hasStandardLengthModifier()) 2967 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 2968 else if (!FS.hasStandardLengthConversionCombination()) 2969 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2970 diag::warn_format_non_standard_conversion_spec); 2971 2972 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2973 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2974 2975 // The remaining checks depend on the data arguments. 2976 if (HasVAListArg) 2977 return true; 2978 2979 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2980 return false; 2981 2982 const Expr *Arg = getDataArg(argIndex); 2983 if (!Arg) 2984 return true; 2985 2986 return checkFormatExpr(FS, startSpecifier, specifierLen, Arg); 2987} 2988 2989static bool requiresParensToAddCast(const Expr *E) { 2990 // FIXME: We should have a general way to reason about operator 2991 // precedence and whether parens are actually needed here. 2992 // Take care of a few common cases where they aren't. 2993 const Expr *Inside = E->IgnoreImpCasts(); 2994 if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside)) 2995 Inside = POE->getSyntacticForm()->IgnoreImpCasts(); 2996 2997 switch (Inside->getStmtClass()) { 2998 case Stmt::ArraySubscriptExprClass: 2999 case Stmt::CallExprClass: 3000 case Stmt::CharacterLiteralClass: 3001 case Stmt::CXXBoolLiteralExprClass: 3002 case Stmt::DeclRefExprClass: 3003 case Stmt::FloatingLiteralClass: 3004 case Stmt::IntegerLiteralClass: 3005 case Stmt::MemberExprClass: 3006 case Stmt::ObjCArrayLiteralClass: 3007 case Stmt::ObjCBoolLiteralExprClass: 3008 case Stmt::ObjCBoxedExprClass: 3009 case Stmt::ObjCDictionaryLiteralClass: 3010 case Stmt::ObjCEncodeExprClass: 3011 case Stmt::ObjCIvarRefExprClass: 3012 case Stmt::ObjCMessageExprClass: 3013 case Stmt::ObjCPropertyRefExprClass: 3014 case Stmt::ObjCStringLiteralClass: 3015 case Stmt::ObjCSubscriptRefExprClass: 3016 case Stmt::ParenExprClass: 3017 case Stmt::StringLiteralClass: 3018 case Stmt::UnaryOperatorClass: 3019 return false; 3020 default: 3021 return true; 3022 } 3023} 3024 3025bool 3026CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 3027 const char *StartSpecifier, 3028 unsigned SpecifierLen, 3029 const Expr *E) { 3030 using namespace analyze_format_string; 3031 using namespace analyze_printf; 3032 // Now type check the data expression that matches the 3033 // format specifier. 3034 const analyze_printf::ArgType &AT = FS.getArgType(S.Context, 3035 ObjCContext); 3036 if (!AT.isValid()) 3037 return true; 3038 3039 QualType ExprTy = E->getType(); 3040 while (const TypeOfExprType *TET = dyn_cast<TypeOfExprType>(ExprTy)) { 3041 ExprTy = TET->getUnderlyingExpr()->getType(); 3042 } 3043 3044 if (AT.matchesType(S.Context, ExprTy)) 3045 return true; 3046 3047 // Look through argument promotions for our error message's reported type. 3048 // This includes the integral and floating promotions, but excludes array 3049 // and function pointer decay; seeing that an argument intended to be a 3050 // string has type 'char [6]' is probably more confusing than 'char *'. 3051 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3052 if (ICE->getCastKind() == CK_IntegralCast || 3053 ICE->getCastKind() == CK_FloatingCast) { 3054 E = ICE->getSubExpr(); 3055 ExprTy = E->getType(); 3056 3057 // Check if we didn't match because of an implicit cast from a 'char' 3058 // or 'short' to an 'int'. This is done because printf is a varargs 3059 // function. 3060 if (ICE->getType() == S.Context.IntTy || 3061 ICE->getType() == S.Context.UnsignedIntTy) { 3062 // All further checking is done on the subexpression. 3063 if (AT.matchesType(S.Context, ExprTy)) 3064 return true; 3065 } 3066 } 3067 } else if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E)) { 3068 // Special case for 'a', which has type 'int' in C. 3069 // Note, however, that we do /not/ want to treat multibyte constants like 3070 // 'MooV' as characters! This form is deprecated but still exists. 3071 if (ExprTy == S.Context.IntTy) 3072 if (llvm::isUIntN(S.Context.getCharWidth(), CL->getValue())) 3073 ExprTy = S.Context.CharTy; 3074 } 3075 3076 // %C in an Objective-C context prints a unichar, not a wchar_t. 3077 // If the argument is an integer of some kind, believe the %C and suggest 3078 // a cast instead of changing the conversion specifier. 3079 QualType IntendedTy = ExprTy; 3080 if (ObjCContext && 3081 FS.getConversionSpecifier().getKind() == ConversionSpecifier::CArg) { 3082 if (ExprTy->isIntegralOrUnscopedEnumerationType() && 3083 !ExprTy->isCharType()) { 3084 // 'unichar' is defined as a typedef of unsigned short, but we should 3085 // prefer using the typedef if it is visible. 3086 IntendedTy = S.Context.UnsignedShortTy; 3087 3088 LookupResult Result(S, &S.Context.Idents.get("unichar"), E->getLocStart(), 3089 Sema::LookupOrdinaryName); 3090 if (S.LookupName(Result, S.getCurScope())) { 3091 NamedDecl *ND = Result.getFoundDecl(); 3092 if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(ND)) 3093 if (TD->getUnderlyingType() == IntendedTy) 3094 IntendedTy = S.Context.getTypedefType(TD); 3095 } 3096 } 3097 } 3098 3099 // Special-case some of Darwin's platform-independence types by suggesting 3100 // casts to primitive types that are known to be large enough. 3101 bool ShouldNotPrintDirectly = false; 3102 if (S.Context.getTargetInfo().getTriple().isOSDarwin()) { 3103 // Use a 'while' to peel off layers of typedefs. 3104 QualType TyTy = IntendedTy; 3105 while (const TypedefType *UserTy = TyTy->getAs<TypedefType>()) { 3106 StringRef Name = UserTy->getDecl()->getName(); 3107 QualType CastTy = llvm::StringSwitch<QualType>(Name) 3108 .Case("NSInteger", S.Context.LongTy) 3109 .Case("NSUInteger", S.Context.UnsignedLongTy) 3110 .Case("SInt32", S.Context.IntTy) 3111 .Case("UInt32", S.Context.UnsignedIntTy) 3112 .Default(QualType()); 3113 3114 if (!CastTy.isNull()) { 3115 ShouldNotPrintDirectly = true; 3116 IntendedTy = CastTy; 3117 break; 3118 } 3119 TyTy = UserTy->desugar(); 3120 } 3121 } 3122 3123 // We may be able to offer a FixItHint if it is a supported type. 3124 PrintfSpecifier fixedFS = FS; 3125 bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(), 3126 S.Context, ObjCContext); 3127 3128 if (success) { 3129 // Get the fix string from the fixed format specifier 3130 SmallString<16> buf; 3131 llvm::raw_svector_ostream os(buf); 3132 fixedFS.toString(os); 3133 3134 CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen); 3135 3136 if (IntendedTy == ExprTy) { 3137 // In this case, the specifier is wrong and should be changed to match 3138 // the argument. 3139 EmitFormatDiagnostic( 3140 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3141 << AT.getRepresentativeTypeName(S.Context) << IntendedTy 3142 << E->getSourceRange(), 3143 E->getLocStart(), 3144 /*IsStringLocation*/false, 3145 SpecRange, 3146 FixItHint::CreateReplacement(SpecRange, os.str())); 3147 3148 } else { 3149 // The canonical type for formatting this value is different from the 3150 // actual type of the expression. (This occurs, for example, with Darwin's 3151 // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but 3152 // should be printed as 'long' for 64-bit compatibility.) 3153 // Rather than emitting a normal format/argument mismatch, we want to 3154 // add a cast to the recommended type (and correct the format string 3155 // if necessary). 3156 SmallString<16> CastBuf; 3157 llvm::raw_svector_ostream CastFix(CastBuf); 3158 CastFix << "("; 3159 IntendedTy.print(CastFix, S.Context.getPrintingPolicy()); 3160 CastFix << ")"; 3161 3162 SmallVector<FixItHint,4> Hints; 3163 if (!AT.matchesType(S.Context, IntendedTy)) 3164 Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str())); 3165 3166 if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) { 3167 // If there's already a cast present, just replace it. 3168 SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc()); 3169 Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str())); 3170 3171 } else if (!requiresParensToAddCast(E)) { 3172 // If the expression has high enough precedence, 3173 // just write the C-style cast. 3174 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 3175 CastFix.str())); 3176 } else { 3177 // Otherwise, add parens around the expression as well as the cast. 3178 CastFix << "("; 3179 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 3180 CastFix.str())); 3181 3182 SourceLocation After = S.PP.getLocForEndOfToken(E->getLocEnd()); 3183 Hints.push_back(FixItHint::CreateInsertion(After, ")")); 3184 } 3185 3186 if (ShouldNotPrintDirectly) { 3187 // The expression has a type that should not be printed directly. 3188 // We extract the name from the typedef because we don't want to show 3189 // the underlying type in the diagnostic. 3190 StringRef Name = cast<TypedefType>(ExprTy)->getDecl()->getName(); 3191 3192 EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast) 3193 << Name << IntendedTy 3194 << E->getSourceRange(), 3195 E->getLocStart(), /*IsStringLocation=*/false, 3196 SpecRange, Hints); 3197 } else { 3198 // In this case, the expression could be printed using a different 3199 // specifier, but we've decided that the specifier is probably correct 3200 // and we should cast instead. Just use the normal warning message. 3201 EmitFormatDiagnostic( 3202 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3203 << AT.getRepresentativeTypeName(S.Context) << ExprTy 3204 << E->getSourceRange(), 3205 E->getLocStart(), /*IsStringLocation*/false, 3206 SpecRange, Hints); 3207 } 3208 } 3209 } else { 3210 const CharSourceRange &CSR = getSpecifierRange(StartSpecifier, 3211 SpecifierLen); 3212 // Since the warning for passing non-POD types to variadic functions 3213 // was deferred until now, we emit a warning for non-POD 3214 // arguments here. 3215 switch (S.isValidVarArgType(ExprTy)) { 3216 case Sema::VAK_Valid: 3217 case Sema::VAK_ValidInCXX11: 3218 EmitFormatDiagnostic( 3219 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3220 << AT.getRepresentativeTypeName(S.Context) << ExprTy 3221 << CSR 3222 << E->getSourceRange(), 3223 E->getLocStart(), /*IsStringLocation*/false, CSR); 3224 break; 3225 3226 case Sema::VAK_Undefined: 3227 EmitFormatDiagnostic( 3228 S.PDiag(diag::warn_non_pod_vararg_with_format_string) 3229 << S.getLangOpts().CPlusPlus11 3230 << ExprTy 3231 << CallType 3232 << AT.getRepresentativeTypeName(S.Context) 3233 << CSR 3234 << E->getSourceRange(), 3235 E->getLocStart(), /*IsStringLocation*/false, CSR); 3236 checkForCStrMembers(AT, E, CSR); 3237 break; 3238 3239 case Sema::VAK_Invalid: 3240 if (ExprTy->isObjCObjectType()) 3241 EmitFormatDiagnostic( 3242 S.PDiag(diag::err_cannot_pass_objc_interface_to_vararg_format) 3243 << S.getLangOpts().CPlusPlus11 3244 << ExprTy 3245 << CallType 3246 << AT.getRepresentativeTypeName(S.Context) 3247 << CSR 3248 << E->getSourceRange(), 3249 E->getLocStart(), /*IsStringLocation*/false, CSR); 3250 else 3251 // FIXME: If this is an initializer list, suggest removing the braces 3252 // or inserting a cast to the target type. 3253 S.Diag(E->getLocStart(), diag::err_cannot_pass_to_vararg_format) 3254 << isa<InitListExpr>(E) << ExprTy << CallType 3255 << AT.getRepresentativeTypeName(S.Context) 3256 << E->getSourceRange(); 3257 break; 3258 } 3259 3260 assert(FirstDataArg + FS.getArgIndex() < CheckedVarArgs.size() && 3261 "format string specifier index out of range"); 3262 CheckedVarArgs[FirstDataArg + FS.getArgIndex()] = true; 3263 } 3264 3265 return true; 3266} 3267 3268//===--- CHECK: Scanf format string checking ------------------------------===// 3269 3270namespace { 3271class CheckScanfHandler : public CheckFormatHandler { 3272public: 3273 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 3274 const Expr *origFormatExpr, unsigned firstDataArg, 3275 unsigned numDataArgs, const char *beg, bool hasVAListArg, 3276 ArrayRef<const Expr *> Args, 3277 unsigned formatIdx, bool inFunctionCall, 3278 Sema::VariadicCallType CallType, 3279 llvm::SmallBitVector &CheckedVarArgs) 3280 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 3281 numDataArgs, beg, hasVAListArg, 3282 Args, formatIdx, inFunctionCall, CallType, 3283 CheckedVarArgs) 3284 {} 3285 3286 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 3287 const char *startSpecifier, 3288 unsigned specifierLen); 3289 3290 bool HandleInvalidScanfConversionSpecifier( 3291 const analyze_scanf::ScanfSpecifier &FS, 3292 const char *startSpecifier, 3293 unsigned specifierLen); 3294 3295 void HandleIncompleteScanList(const char *start, const char *end); 3296}; 3297} 3298 3299void CheckScanfHandler::HandleIncompleteScanList(const char *start, 3300 const char *end) { 3301 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 3302 getLocationOfByte(end), /*IsStringLocation*/true, 3303 getSpecifierRange(start, end - start)); 3304} 3305 3306bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 3307 const analyze_scanf::ScanfSpecifier &FS, 3308 const char *startSpecifier, 3309 unsigned specifierLen) { 3310 3311 const analyze_scanf::ScanfConversionSpecifier &CS = 3312 FS.getConversionSpecifier(); 3313 3314 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 3315 getLocationOfByte(CS.getStart()), 3316 startSpecifier, specifierLen, 3317 CS.getStart(), CS.getLength()); 3318} 3319 3320bool CheckScanfHandler::HandleScanfSpecifier( 3321 const analyze_scanf::ScanfSpecifier &FS, 3322 const char *startSpecifier, 3323 unsigned specifierLen) { 3324 3325 using namespace analyze_scanf; 3326 using namespace analyze_format_string; 3327 3328 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 3329 3330 // Handle case where '%' and '*' don't consume an argument. These shouldn't 3331 // be used to decide if we are using positional arguments consistently. 3332 if (FS.consumesDataArgument()) { 3333 if (atFirstArg) { 3334 atFirstArg = false; 3335 usesPositionalArgs = FS.usesPositionalArg(); 3336 } 3337 else if (usesPositionalArgs != FS.usesPositionalArg()) { 3338 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 3339 startSpecifier, specifierLen); 3340 return false; 3341 } 3342 } 3343 3344 // Check if the field with is non-zero. 3345 const OptionalAmount &Amt = FS.getFieldWidth(); 3346 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 3347 if (Amt.getConstantAmount() == 0) { 3348 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 3349 Amt.getConstantLength()); 3350 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 3351 getLocationOfByte(Amt.getStart()), 3352 /*IsStringLocation*/true, R, 3353 FixItHint::CreateRemoval(R)); 3354 } 3355 } 3356 3357 if (!FS.consumesDataArgument()) { 3358 // FIXME: Technically specifying a precision or field width here 3359 // makes no sense. Worth issuing a warning at some point. 3360 return true; 3361 } 3362 3363 // Consume the argument. 3364 unsigned argIndex = FS.getArgIndex(); 3365 if (argIndex < NumDataArgs) { 3366 // The check to see if the argIndex is valid will come later. 3367 // We set the bit here because we may exit early from this 3368 // function if we encounter some other error. 3369 CoveredArgs.set(argIndex); 3370 } 3371 3372 // Check the length modifier is valid with the given conversion specifier. 3373 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3374 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3375 diag::warn_format_nonsensical_length); 3376 else if (!FS.hasStandardLengthModifier()) 3377 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3378 else if (!FS.hasStandardLengthConversionCombination()) 3379 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3380 diag::warn_format_non_standard_conversion_spec); 3381 3382 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3383 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3384 3385 // The remaining checks depend on the data arguments. 3386 if (HasVAListArg) 3387 return true; 3388 3389 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3390 return false; 3391 3392 // Check that the argument type matches the format specifier. 3393 const Expr *Ex = getDataArg(argIndex); 3394 if (!Ex) 3395 return true; 3396 3397 const analyze_format_string::ArgType &AT = FS.getArgType(S.Context); 3398 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) { 3399 ScanfSpecifier fixedFS = FS; 3400 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 3401 S.Context); 3402 3403 if (success) { 3404 // Get the fix string from the fixed format specifier. 3405 SmallString<128> buf; 3406 llvm::raw_svector_ostream os(buf); 3407 fixedFS.toString(os); 3408 3409 EmitFormatDiagnostic( 3410 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3411 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3412 << Ex->getSourceRange(), 3413 Ex->getLocStart(), 3414 /*IsStringLocation*/false, 3415 getSpecifierRange(startSpecifier, specifierLen), 3416 FixItHint::CreateReplacement( 3417 getSpecifierRange(startSpecifier, specifierLen), 3418 os.str())); 3419 } else { 3420 EmitFormatDiagnostic( 3421 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3422 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3423 << Ex->getSourceRange(), 3424 Ex->getLocStart(), 3425 /*IsStringLocation*/false, 3426 getSpecifierRange(startSpecifier, specifierLen)); 3427 } 3428 } 3429 3430 return true; 3431} 3432 3433void Sema::CheckFormatString(const StringLiteral *FExpr, 3434 const Expr *OrigFormatExpr, 3435 ArrayRef<const Expr *> Args, 3436 bool HasVAListArg, unsigned format_idx, 3437 unsigned firstDataArg, FormatStringType Type, 3438 bool inFunctionCall, VariadicCallType CallType, 3439 llvm::SmallBitVector &CheckedVarArgs) { 3440 3441 // CHECK: is the format string a wide literal? 3442 if (!FExpr->isAscii() && !FExpr->isUTF8()) { 3443 CheckFormatHandler::EmitFormatDiagnostic( 3444 *this, inFunctionCall, Args[format_idx], 3445 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 3446 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3447 return; 3448 } 3449 3450 // Str - The format string. NOTE: this is NOT null-terminated! 3451 StringRef StrRef = FExpr->getString(); 3452 const char *Str = StrRef.data(); 3453 unsigned StrLen = StrRef.size(); 3454 const unsigned numDataArgs = Args.size() - firstDataArg; 3455 3456 // CHECK: empty format string? 3457 if (StrLen == 0 && numDataArgs > 0) { 3458 CheckFormatHandler::EmitFormatDiagnostic( 3459 *this, inFunctionCall, Args[format_idx], 3460 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 3461 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3462 return; 3463 } 3464 3465 if (Type == FST_Printf || Type == FST_NSString) { 3466 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 3467 numDataArgs, (Type == FST_NSString), 3468 Str, HasVAListArg, Args, format_idx, 3469 inFunctionCall, CallType, CheckedVarArgs); 3470 3471 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 3472 getLangOpts(), 3473 Context.getTargetInfo())) 3474 H.DoneProcessing(); 3475 } else if (Type == FST_Scanf) { 3476 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs, 3477 Str, HasVAListArg, Args, format_idx, 3478 inFunctionCall, CallType, CheckedVarArgs); 3479 3480 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 3481 getLangOpts(), 3482 Context.getTargetInfo())) 3483 H.DoneProcessing(); 3484 } // TODO: handle other formats 3485} 3486 3487//===--- CHECK: Standard memory functions ---------------------------------===// 3488 3489/// \brief Determine whether the given type is a dynamic class type (e.g., 3490/// whether it has a vtable). 3491static bool isDynamicClassType(QualType T) { 3492 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 3493 if (CXXRecordDecl *Definition = Record->getDefinition()) 3494 if (Definition->isDynamicClass()) 3495 return true; 3496 3497 return false; 3498} 3499 3500/// \brief If E is a sizeof expression, returns its argument expression, 3501/// otherwise returns NULL. 3502static const Expr *getSizeOfExprArg(const Expr* E) { 3503 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3504 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3505 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 3506 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 3507 3508 return 0; 3509} 3510 3511/// \brief If E is a sizeof expression, returns its argument type. 3512static QualType getSizeOfArgType(const Expr* E) { 3513 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3514 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3515 if (SizeOf->getKind() == clang::UETT_SizeOf) 3516 return SizeOf->getTypeOfArgument(); 3517 3518 return QualType(); 3519} 3520 3521/// \brief Check for dangerous or invalid arguments to memset(). 3522/// 3523/// This issues warnings on known problematic, dangerous or unspecified 3524/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 3525/// function calls. 3526/// 3527/// \param Call The call expression to diagnose. 3528void Sema::CheckMemaccessArguments(const CallExpr *Call, 3529 unsigned BId, 3530 IdentifierInfo *FnName) { 3531 assert(BId != 0); 3532 3533 // It is possible to have a non-standard definition of memset. Validate 3534 // we have enough arguments, and if not, abort further checking. 3535 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 3536 if (Call->getNumArgs() < ExpectedNumArgs) 3537 return; 3538 3539 unsigned LastArg = (BId == Builtin::BImemset || 3540 BId == Builtin::BIstrndup ? 1 : 2); 3541 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 3542 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 3543 3544 // We have special checking when the length is a sizeof expression. 3545 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 3546 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 3547 llvm::FoldingSetNodeID SizeOfArgID; 3548 3549 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 3550 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 3551 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 3552 3553 QualType DestTy = Dest->getType(); 3554 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 3555 QualType PointeeTy = DestPtrTy->getPointeeType(); 3556 3557 // Never warn about void type pointers. This can be used to suppress 3558 // false positives. 3559 if (PointeeTy->isVoidType()) 3560 continue; 3561 3562 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 3563 // actually comparing the expressions for equality. Because computing the 3564 // expression IDs can be expensive, we only do this if the diagnostic is 3565 // enabled. 3566 if (SizeOfArg && 3567 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 3568 SizeOfArg->getExprLoc())) { 3569 // We only compute IDs for expressions if the warning is enabled, and 3570 // cache the sizeof arg's ID. 3571 if (SizeOfArgID == llvm::FoldingSetNodeID()) 3572 SizeOfArg->Profile(SizeOfArgID, Context, true); 3573 llvm::FoldingSetNodeID DestID; 3574 Dest->Profile(DestID, Context, true); 3575 if (DestID == SizeOfArgID) { 3576 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 3577 // over sizeof(src) as well. 3578 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 3579 StringRef ReadableName = FnName->getName(); 3580 3581 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 3582 if (UnaryOp->getOpcode() == UO_AddrOf) 3583 ActionIdx = 1; // If its an address-of operator, just remove it. 3584 if (!PointeeTy->isIncompleteType() && 3585 (Context.getTypeSize(PointeeTy) == Context.getCharWidth())) 3586 ActionIdx = 2; // If the pointee's size is sizeof(char), 3587 // suggest an explicit length. 3588 3589 // If the function is defined as a builtin macro, do not show macro 3590 // expansion. 3591 SourceLocation SL = SizeOfArg->getExprLoc(); 3592 SourceRange DSR = Dest->getSourceRange(); 3593 SourceRange SSR = SizeOfArg->getSourceRange(); 3594 SourceManager &SM = PP.getSourceManager(); 3595 3596 if (SM.isMacroArgExpansion(SL)) { 3597 ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts); 3598 SL = SM.getSpellingLoc(SL); 3599 DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()), 3600 SM.getSpellingLoc(DSR.getEnd())); 3601 SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()), 3602 SM.getSpellingLoc(SSR.getEnd())); 3603 } 3604 3605 DiagRuntimeBehavior(SL, SizeOfArg, 3606 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 3607 << ReadableName 3608 << PointeeTy 3609 << DestTy 3610 << DSR 3611 << SSR); 3612 DiagRuntimeBehavior(SL, SizeOfArg, 3613 PDiag(diag::warn_sizeof_pointer_expr_memaccess_note) 3614 << ActionIdx 3615 << SSR); 3616 3617 break; 3618 } 3619 } 3620 3621 // Also check for cases where the sizeof argument is the exact same 3622 // type as the memory argument, and where it points to a user-defined 3623 // record type. 3624 if (SizeOfArgTy != QualType()) { 3625 if (PointeeTy->isRecordType() && 3626 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 3627 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 3628 PDiag(diag::warn_sizeof_pointer_type_memaccess) 3629 << FnName << SizeOfArgTy << ArgIdx 3630 << PointeeTy << Dest->getSourceRange() 3631 << LenExpr->getSourceRange()); 3632 break; 3633 } 3634 } 3635 3636 // Always complain about dynamic classes. 3637 if (isDynamicClassType(PointeeTy)) { 3638 3639 unsigned OperationType = 0; 3640 // "overwritten" if we're warning about the destination for any call 3641 // but memcmp; otherwise a verb appropriate to the call. 3642 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 3643 if (BId == Builtin::BImemcpy) 3644 OperationType = 1; 3645 else if(BId == Builtin::BImemmove) 3646 OperationType = 2; 3647 else if (BId == Builtin::BImemcmp) 3648 OperationType = 3; 3649 } 3650 3651 DiagRuntimeBehavior( 3652 Dest->getExprLoc(), Dest, 3653 PDiag(diag::warn_dyn_class_memaccess) 3654 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 3655 << FnName << PointeeTy 3656 << OperationType 3657 << Call->getCallee()->getSourceRange()); 3658 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 3659 BId != Builtin::BImemset) 3660 DiagRuntimeBehavior( 3661 Dest->getExprLoc(), Dest, 3662 PDiag(diag::warn_arc_object_memaccess) 3663 << ArgIdx << FnName << PointeeTy 3664 << Call->getCallee()->getSourceRange()); 3665 else 3666 continue; 3667 3668 DiagRuntimeBehavior( 3669 Dest->getExprLoc(), Dest, 3670 PDiag(diag::note_bad_memaccess_silence) 3671 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 3672 break; 3673 } 3674 } 3675} 3676 3677// A little helper routine: ignore addition and subtraction of integer literals. 3678// This intentionally does not ignore all integer constant expressions because 3679// we don't want to remove sizeof(). 3680static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 3681 Ex = Ex->IgnoreParenCasts(); 3682 3683 for (;;) { 3684 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 3685 if (!BO || !BO->isAdditiveOp()) 3686 break; 3687 3688 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 3689 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 3690 3691 if (isa<IntegerLiteral>(RHS)) 3692 Ex = LHS; 3693 else if (isa<IntegerLiteral>(LHS)) 3694 Ex = RHS; 3695 else 3696 break; 3697 } 3698 3699 return Ex; 3700} 3701 3702static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty, 3703 ASTContext &Context) { 3704 // Only handle constant-sized or VLAs, but not flexible members. 3705 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) { 3706 // Only issue the FIXIT for arrays of size > 1. 3707 if (CAT->getSize().getSExtValue() <= 1) 3708 return false; 3709 } else if (!Ty->isVariableArrayType()) { 3710 return false; 3711 } 3712 return true; 3713} 3714 3715// Warn if the user has made the 'size' argument to strlcpy or strlcat 3716// be the size of the source, instead of the destination. 3717void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 3718 IdentifierInfo *FnName) { 3719 3720 // Don't crash if the user has the wrong number of arguments 3721 if (Call->getNumArgs() != 3) 3722 return; 3723 3724 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 3725 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 3726 const Expr *CompareWithSrc = NULL; 3727 3728 // Look for 'strlcpy(dst, x, sizeof(x))' 3729 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 3730 CompareWithSrc = Ex; 3731 else { 3732 // Look for 'strlcpy(dst, x, strlen(x))' 3733 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 3734 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 3735 && SizeCall->getNumArgs() == 1) 3736 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 3737 } 3738 } 3739 3740 if (!CompareWithSrc) 3741 return; 3742 3743 // Determine if the argument to sizeof/strlen is equal to the source 3744 // argument. In principle there's all kinds of things you could do 3745 // here, for instance creating an == expression and evaluating it with 3746 // EvaluateAsBooleanCondition, but this uses a more direct technique: 3747 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 3748 if (!SrcArgDRE) 3749 return; 3750 3751 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 3752 if (!CompareWithSrcDRE || 3753 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 3754 return; 3755 3756 const Expr *OriginalSizeArg = Call->getArg(2); 3757 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 3758 << OriginalSizeArg->getSourceRange() << FnName; 3759 3760 // Output a FIXIT hint if the destination is an array (rather than a 3761 // pointer to an array). This could be enhanced to handle some 3762 // pointers if we know the actual size, like if DstArg is 'array+2' 3763 // we could say 'sizeof(array)-2'. 3764 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 3765 if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context)) 3766 return; 3767 3768 SmallString<128> sizeString; 3769 llvm::raw_svector_ostream OS(sizeString); 3770 OS << "sizeof("; 3771 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3772 OS << ")"; 3773 3774 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 3775 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 3776 OS.str()); 3777} 3778 3779/// Check if two expressions refer to the same declaration. 3780static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 3781 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 3782 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 3783 return D1->getDecl() == D2->getDecl(); 3784 return false; 3785} 3786 3787static const Expr *getStrlenExprArg(const Expr *E) { 3788 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 3789 const FunctionDecl *FD = CE->getDirectCallee(); 3790 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 3791 return 0; 3792 return CE->getArg(0)->IgnoreParenCasts(); 3793 } 3794 return 0; 3795} 3796 3797// Warn on anti-patterns as the 'size' argument to strncat. 3798// The correct size argument should look like following: 3799// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 3800void Sema::CheckStrncatArguments(const CallExpr *CE, 3801 IdentifierInfo *FnName) { 3802 // Don't crash if the user has the wrong number of arguments. 3803 if (CE->getNumArgs() < 3) 3804 return; 3805 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 3806 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 3807 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 3808 3809 // Identify common expressions, which are wrongly used as the size argument 3810 // to strncat and may lead to buffer overflows. 3811 unsigned PatternType = 0; 3812 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 3813 // - sizeof(dst) 3814 if (referToTheSameDecl(SizeOfArg, DstArg)) 3815 PatternType = 1; 3816 // - sizeof(src) 3817 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 3818 PatternType = 2; 3819 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 3820 if (BE->getOpcode() == BO_Sub) { 3821 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 3822 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 3823 // - sizeof(dst) - strlen(dst) 3824 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 3825 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 3826 PatternType = 1; 3827 // - sizeof(src) - (anything) 3828 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 3829 PatternType = 2; 3830 } 3831 } 3832 3833 if (PatternType == 0) 3834 return; 3835 3836 // Generate the diagnostic. 3837 SourceLocation SL = LenArg->getLocStart(); 3838 SourceRange SR = LenArg->getSourceRange(); 3839 SourceManager &SM = PP.getSourceManager(); 3840 3841 // If the function is defined as a builtin macro, do not show macro expansion. 3842 if (SM.isMacroArgExpansion(SL)) { 3843 SL = SM.getSpellingLoc(SL); 3844 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 3845 SM.getSpellingLoc(SR.getEnd())); 3846 } 3847 3848 // Check if the destination is an array (rather than a pointer to an array). 3849 QualType DstTy = DstArg->getType(); 3850 bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy, 3851 Context); 3852 if (!isKnownSizeArray) { 3853 if (PatternType == 1) 3854 Diag(SL, diag::warn_strncat_wrong_size) << SR; 3855 else 3856 Diag(SL, diag::warn_strncat_src_size) << SR; 3857 return; 3858 } 3859 3860 if (PatternType == 1) 3861 Diag(SL, diag::warn_strncat_large_size) << SR; 3862 else 3863 Diag(SL, diag::warn_strncat_src_size) << SR; 3864 3865 SmallString<128> sizeString; 3866 llvm::raw_svector_ostream OS(sizeString); 3867 OS << "sizeof("; 3868 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3869 OS << ") - "; 3870 OS << "strlen("; 3871 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3872 OS << ") - 1"; 3873 3874 Diag(SL, diag::note_strncat_wrong_size) 3875 << FixItHint::CreateReplacement(SR, OS.str()); 3876} 3877 3878//===--- CHECK: Return Address of Stack Variable --------------------------===// 3879 3880static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3881 Decl *ParentDecl); 3882static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3883 Decl *ParentDecl); 3884 3885/// CheckReturnStackAddr - Check if a return statement returns the address 3886/// of a stack variable. 3887void 3888Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3889 SourceLocation ReturnLoc) { 3890 3891 Expr *stackE = 0; 3892 SmallVector<DeclRefExpr *, 8> refVars; 3893 3894 // Perform checking for returned stack addresses, local blocks, 3895 // label addresses or references to temporaries. 3896 if (lhsType->isPointerType() || 3897 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 3898 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 3899 } else if (lhsType->isReferenceType()) { 3900 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 3901 } 3902 3903 if (stackE == 0) 3904 return; // Nothing suspicious was found. 3905 3906 SourceLocation diagLoc; 3907 SourceRange diagRange; 3908 if (refVars.empty()) { 3909 diagLoc = stackE->getLocStart(); 3910 diagRange = stackE->getSourceRange(); 3911 } else { 3912 // We followed through a reference variable. 'stackE' contains the 3913 // problematic expression but we will warn at the return statement pointing 3914 // at the reference variable. We will later display the "trail" of 3915 // reference variables using notes. 3916 diagLoc = refVars[0]->getLocStart(); 3917 diagRange = refVars[0]->getSourceRange(); 3918 } 3919 3920 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 3921 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 3922 : diag::warn_ret_stack_addr) 3923 << DR->getDecl()->getDeclName() << diagRange; 3924 } else if (isa<BlockExpr>(stackE)) { // local block. 3925 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 3926 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 3927 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 3928 } else { // local temporary. 3929 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 3930 : diag::warn_ret_local_temp_addr) 3931 << diagRange; 3932 } 3933 3934 // Display the "trail" of reference variables that we followed until we 3935 // found the problematic expression using notes. 3936 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 3937 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 3938 // If this var binds to another reference var, show the range of the next 3939 // var, otherwise the var binds to the problematic expression, in which case 3940 // show the range of the expression. 3941 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 3942 : stackE->getSourceRange(); 3943 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 3944 << VD->getDeclName() << range; 3945 } 3946} 3947 3948/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 3949/// check if the expression in a return statement evaluates to an address 3950/// to a location on the stack, a local block, an address of a label, or a 3951/// reference to local temporary. The recursion is used to traverse the 3952/// AST of the return expression, with recursion backtracking when we 3953/// encounter a subexpression that (1) clearly does not lead to one of the 3954/// above problematic expressions (2) is something we cannot determine leads to 3955/// a problematic expression based on such local checking. 3956/// 3957/// Both EvalAddr and EvalVal follow through reference variables to evaluate 3958/// the expression that they point to. Such variables are added to the 3959/// 'refVars' vector so that we know what the reference variable "trail" was. 3960/// 3961/// EvalAddr processes expressions that are pointers that are used as 3962/// references (and not L-values). EvalVal handles all other values. 3963/// At the base case of the recursion is a check for the above problematic 3964/// expressions. 3965/// 3966/// This implementation handles: 3967/// 3968/// * pointer-to-pointer casts 3969/// * implicit conversions from array references to pointers 3970/// * taking the address of fields 3971/// * arbitrary interplay between "&" and "*" operators 3972/// * pointer arithmetic from an address of a stack variable 3973/// * taking the address of an array element where the array is on the stack 3974static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3975 Decl *ParentDecl) { 3976 if (E->isTypeDependent()) 3977 return NULL; 3978 3979 // We should only be called for evaluating pointer expressions. 3980 assert((E->getType()->isAnyPointerType() || 3981 E->getType()->isBlockPointerType() || 3982 E->getType()->isObjCQualifiedIdType()) && 3983 "EvalAddr only works on pointers"); 3984 3985 E = E->IgnoreParens(); 3986 3987 // Our "symbolic interpreter" is just a dispatch off the currently 3988 // viewed AST node. We then recursively traverse the AST by calling 3989 // EvalAddr and EvalVal appropriately. 3990 switch (E->getStmtClass()) { 3991 case Stmt::DeclRefExprClass: { 3992 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3993 3994 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3995 // If this is a reference variable, follow through to the expression that 3996 // it points to. 3997 if (V->hasLocalStorage() && 3998 V->getType()->isReferenceType() && V->hasInit()) { 3999 // Add the reference variable to the "trail". 4000 refVars.push_back(DR); 4001 return EvalAddr(V->getInit(), refVars, ParentDecl); 4002 } 4003 4004 return NULL; 4005 } 4006 4007 case Stmt::UnaryOperatorClass: { 4008 // The only unary operator that make sense to handle here 4009 // is AddrOf. All others don't make sense as pointers. 4010 UnaryOperator *U = cast<UnaryOperator>(E); 4011 4012 if (U->getOpcode() == UO_AddrOf) 4013 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 4014 else 4015 return NULL; 4016 } 4017 4018 case Stmt::BinaryOperatorClass: { 4019 // Handle pointer arithmetic. All other binary operators are not valid 4020 // in this context. 4021 BinaryOperator *B = cast<BinaryOperator>(E); 4022 BinaryOperatorKind op = B->getOpcode(); 4023 4024 if (op != BO_Add && op != BO_Sub) 4025 return NULL; 4026 4027 Expr *Base = B->getLHS(); 4028 4029 // Determine which argument is the real pointer base. It could be 4030 // the RHS argument instead of the LHS. 4031 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 4032 4033 assert (Base->getType()->isPointerType()); 4034 return EvalAddr(Base, refVars, ParentDecl); 4035 } 4036 4037 // For conditional operators we need to see if either the LHS or RHS are 4038 // valid DeclRefExpr*s. If one of them is valid, we return it. 4039 case Stmt::ConditionalOperatorClass: { 4040 ConditionalOperator *C = cast<ConditionalOperator>(E); 4041 4042 // Handle the GNU extension for missing LHS. 4043 if (Expr *lhsExpr = C->getLHS()) { 4044 // In C++, we can have a throw-expression, which has 'void' type. 4045 if (!lhsExpr->getType()->isVoidType()) 4046 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 4047 return LHS; 4048 } 4049 4050 // In C++, we can have a throw-expression, which has 'void' type. 4051 if (C->getRHS()->getType()->isVoidType()) 4052 return NULL; 4053 4054 return EvalAddr(C->getRHS(), refVars, ParentDecl); 4055 } 4056 4057 case Stmt::BlockExprClass: 4058 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 4059 return E; // local block. 4060 return NULL; 4061 4062 case Stmt::AddrLabelExprClass: 4063 return E; // address of label. 4064 4065 case Stmt::ExprWithCleanupsClass: 4066 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 4067 ParentDecl); 4068 4069 // For casts, we need to handle conversions from arrays to 4070 // pointer values, and pointer-to-pointer conversions. 4071 case Stmt::ImplicitCastExprClass: 4072 case Stmt::CStyleCastExprClass: 4073 case Stmt::CXXFunctionalCastExprClass: 4074 case Stmt::ObjCBridgedCastExprClass: 4075 case Stmt::CXXStaticCastExprClass: 4076 case Stmt::CXXDynamicCastExprClass: 4077 case Stmt::CXXConstCastExprClass: 4078 case Stmt::CXXReinterpretCastExprClass: { 4079 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 4080 switch (cast<CastExpr>(E)->getCastKind()) { 4081 case CK_BitCast: 4082 case CK_LValueToRValue: 4083 case CK_NoOp: 4084 case CK_BaseToDerived: 4085 case CK_DerivedToBase: 4086 case CK_UncheckedDerivedToBase: 4087 case CK_Dynamic: 4088 case CK_CPointerToObjCPointerCast: 4089 case CK_BlockPointerToObjCPointerCast: 4090 case CK_AnyPointerToBlockPointerCast: 4091 return EvalAddr(SubExpr, refVars, ParentDecl); 4092 4093 case CK_ArrayToPointerDecay: 4094 return EvalVal(SubExpr, refVars, ParentDecl); 4095 4096 default: 4097 return 0; 4098 } 4099 } 4100 4101 case Stmt::MaterializeTemporaryExprClass: 4102 if (Expr *Result = EvalAddr( 4103 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 4104 refVars, ParentDecl)) 4105 return Result; 4106 4107 return E; 4108 4109 // Everything else: we simply don't reason about them. 4110 default: 4111 return NULL; 4112 } 4113} 4114 4115 4116/// EvalVal - This function is complements EvalAddr in the mutual recursion. 4117/// See the comments for EvalAddr for more details. 4118static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 4119 Decl *ParentDecl) { 4120do { 4121 // We should only be called for evaluating non-pointer expressions, or 4122 // expressions with a pointer type that are not used as references but instead 4123 // are l-values (e.g., DeclRefExpr with a pointer type). 4124 4125 // Our "symbolic interpreter" is just a dispatch off the currently 4126 // viewed AST node. We then recursively traverse the AST by calling 4127 // EvalAddr and EvalVal appropriately. 4128 4129 E = E->IgnoreParens(); 4130 switch (E->getStmtClass()) { 4131 case Stmt::ImplicitCastExprClass: { 4132 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 4133 if (IE->getValueKind() == VK_LValue) { 4134 E = IE->getSubExpr(); 4135 continue; 4136 } 4137 return NULL; 4138 } 4139 4140 case Stmt::ExprWithCleanupsClass: 4141 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 4142 4143 case Stmt::DeclRefExprClass: { 4144 // When we hit a DeclRefExpr we are looking at code that refers to a 4145 // variable's name. If it's not a reference variable we check if it has 4146 // local storage within the function, and if so, return the expression. 4147 DeclRefExpr *DR = cast<DeclRefExpr>(E); 4148 4149 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 4150 // Check if it refers to itself, e.g. "int& i = i;". 4151 if (V == ParentDecl) 4152 return DR; 4153 4154 if (V->hasLocalStorage()) { 4155 if (!V->getType()->isReferenceType()) 4156 return DR; 4157 4158 // Reference variable, follow through to the expression that 4159 // it points to. 4160 if (V->hasInit()) { 4161 // Add the reference variable to the "trail". 4162 refVars.push_back(DR); 4163 return EvalVal(V->getInit(), refVars, V); 4164 } 4165 } 4166 } 4167 4168 return NULL; 4169 } 4170 4171 case Stmt::UnaryOperatorClass: { 4172 // The only unary operator that make sense to handle here 4173 // is Deref. All others don't resolve to a "name." This includes 4174 // handling all sorts of rvalues passed to a unary operator. 4175 UnaryOperator *U = cast<UnaryOperator>(E); 4176 4177 if (U->getOpcode() == UO_Deref) 4178 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 4179 4180 return NULL; 4181 } 4182 4183 case Stmt::ArraySubscriptExprClass: { 4184 // Array subscripts are potential references to data on the stack. We 4185 // retrieve the DeclRefExpr* for the array variable if it indeed 4186 // has local storage. 4187 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 4188 } 4189 4190 case Stmt::ConditionalOperatorClass: { 4191 // For conditional operators we need to see if either the LHS or RHS are 4192 // non-NULL Expr's. If one is non-NULL, we return it. 4193 ConditionalOperator *C = cast<ConditionalOperator>(E); 4194 4195 // Handle the GNU extension for missing LHS. 4196 if (Expr *lhsExpr = C->getLHS()) 4197 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 4198 return LHS; 4199 4200 return EvalVal(C->getRHS(), refVars, ParentDecl); 4201 } 4202 4203 // Accesses to members are potential references to data on the stack. 4204 case Stmt::MemberExprClass: { 4205 MemberExpr *M = cast<MemberExpr>(E); 4206 4207 // Check for indirect access. We only want direct field accesses. 4208 if (M->isArrow()) 4209 return NULL; 4210 4211 // Check whether the member type is itself a reference, in which case 4212 // we're not going to refer to the member, but to what the member refers to. 4213 if (M->getMemberDecl()->getType()->isReferenceType()) 4214 return NULL; 4215 4216 return EvalVal(M->getBase(), refVars, ParentDecl); 4217 } 4218 4219 case Stmt::MaterializeTemporaryExprClass: 4220 if (Expr *Result = EvalVal( 4221 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 4222 refVars, ParentDecl)) 4223 return Result; 4224 4225 return E; 4226 4227 default: 4228 // Check that we don't return or take the address of a reference to a 4229 // temporary. This is only useful in C++. 4230 if (!E->isTypeDependent() && E->isRValue()) 4231 return E; 4232 4233 // Everything else: we simply don't reason about them. 4234 return NULL; 4235 } 4236} while (true); 4237} 4238 4239//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 4240 4241/// Check for comparisons of floating point operands using != and ==. 4242/// Issue a warning if these are no self-comparisons, as they are not likely 4243/// to do what the programmer intended. 4244void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 4245 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 4246 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 4247 4248 // Special case: check for x == x (which is OK). 4249 // Do not emit warnings for such cases. 4250 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 4251 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 4252 if (DRL->getDecl() == DRR->getDecl()) 4253 return; 4254 4255 4256 // Special case: check for comparisons against literals that can be exactly 4257 // represented by APFloat. In such cases, do not emit a warning. This 4258 // is a heuristic: often comparison against such literals are used to 4259 // detect if a value in a variable has not changed. This clearly can 4260 // lead to false negatives. 4261 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 4262 if (FLL->isExact()) 4263 return; 4264 } else 4265 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)) 4266 if (FLR->isExact()) 4267 return; 4268 4269 // Check for comparisons with builtin types. 4270 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 4271 if (CL->isBuiltinCall()) 4272 return; 4273 4274 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 4275 if (CR->isBuiltinCall()) 4276 return; 4277 4278 // Emit the diagnostic. 4279 Diag(Loc, diag::warn_floatingpoint_eq) 4280 << LHS->getSourceRange() << RHS->getSourceRange(); 4281} 4282 4283//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 4284//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 4285 4286namespace { 4287 4288/// Structure recording the 'active' range of an integer-valued 4289/// expression. 4290struct IntRange { 4291 /// The number of bits active in the int. 4292 unsigned Width; 4293 4294 /// True if the int is known not to have negative values. 4295 bool NonNegative; 4296 4297 IntRange(unsigned Width, bool NonNegative) 4298 : Width(Width), NonNegative(NonNegative) 4299 {} 4300 4301 /// Returns the range of the bool type. 4302 static IntRange forBoolType() { 4303 return IntRange(1, true); 4304 } 4305 4306 /// Returns the range of an opaque value of the given integral type. 4307 static IntRange forValueOfType(ASTContext &C, QualType T) { 4308 return forValueOfCanonicalType(C, 4309 T->getCanonicalTypeInternal().getTypePtr()); 4310 } 4311 4312 /// Returns the range of an opaque value of a canonical integral type. 4313 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 4314 assert(T->isCanonicalUnqualified()); 4315 4316 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4317 T = VT->getElementType().getTypePtr(); 4318 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4319 T = CT->getElementType().getTypePtr(); 4320 4321 // For enum types, use the known bit width of the enumerators. 4322 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 4323 EnumDecl *Enum = ET->getDecl(); 4324 if (!Enum->isCompleteDefinition()) 4325 return IntRange(C.getIntWidth(QualType(T, 0)), false); 4326 4327 unsigned NumPositive = Enum->getNumPositiveBits(); 4328 unsigned NumNegative = Enum->getNumNegativeBits(); 4329 4330 if (NumNegative == 0) 4331 return IntRange(NumPositive, true/*NonNegative*/); 4332 else 4333 return IntRange(std::max(NumPositive + 1, NumNegative), 4334 false/*NonNegative*/); 4335 } 4336 4337 const BuiltinType *BT = cast<BuiltinType>(T); 4338 assert(BT->isInteger()); 4339 4340 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4341 } 4342 4343 /// Returns the "target" range of a canonical integral type, i.e. 4344 /// the range of values expressible in the type. 4345 /// 4346 /// This matches forValueOfCanonicalType except that enums have the 4347 /// full range of their type, not the range of their enumerators. 4348 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 4349 assert(T->isCanonicalUnqualified()); 4350 4351 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4352 T = VT->getElementType().getTypePtr(); 4353 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4354 T = CT->getElementType().getTypePtr(); 4355 if (const EnumType *ET = dyn_cast<EnumType>(T)) 4356 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 4357 4358 const BuiltinType *BT = cast<BuiltinType>(T); 4359 assert(BT->isInteger()); 4360 4361 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4362 } 4363 4364 /// Returns the supremum of two ranges: i.e. their conservative merge. 4365 static IntRange join(IntRange L, IntRange R) { 4366 return IntRange(std::max(L.Width, R.Width), 4367 L.NonNegative && R.NonNegative); 4368 } 4369 4370 /// Returns the infinum of two ranges: i.e. their aggressive merge. 4371 static IntRange meet(IntRange L, IntRange R) { 4372 return IntRange(std::min(L.Width, R.Width), 4373 L.NonNegative || R.NonNegative); 4374 } 4375}; 4376 4377static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 4378 unsigned MaxWidth) { 4379 if (value.isSigned() && value.isNegative()) 4380 return IntRange(value.getMinSignedBits(), false); 4381 4382 if (value.getBitWidth() > MaxWidth) 4383 value = value.trunc(MaxWidth); 4384 4385 // isNonNegative() just checks the sign bit without considering 4386 // signedness. 4387 return IntRange(value.getActiveBits(), true); 4388} 4389 4390static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 4391 unsigned MaxWidth) { 4392 if (result.isInt()) 4393 return GetValueRange(C, result.getInt(), MaxWidth); 4394 4395 if (result.isVector()) { 4396 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 4397 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 4398 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 4399 R = IntRange::join(R, El); 4400 } 4401 return R; 4402 } 4403 4404 if (result.isComplexInt()) { 4405 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 4406 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 4407 return IntRange::join(R, I); 4408 } 4409 4410 // This can happen with lossless casts to intptr_t of "based" lvalues. 4411 // Assume it might use arbitrary bits. 4412 // FIXME: The only reason we need to pass the type in here is to get 4413 // the sign right on this one case. It would be nice if APValue 4414 // preserved this. 4415 assert(result.isLValue() || result.isAddrLabelDiff()); 4416 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 4417} 4418 4419static QualType GetExprType(Expr *E) { 4420 QualType Ty = E->getType(); 4421 if (const AtomicType *AtomicRHS = Ty->getAs<AtomicType>()) 4422 Ty = AtomicRHS->getValueType(); 4423 return Ty; 4424} 4425 4426/// Pseudo-evaluate the given integer expression, estimating the 4427/// range of values it might take. 4428/// 4429/// \param MaxWidth - the width to which the value will be truncated 4430static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 4431 E = E->IgnoreParens(); 4432 4433 // Try a full evaluation first. 4434 Expr::EvalResult result; 4435 if (E->EvaluateAsRValue(result, C)) 4436 return GetValueRange(C, result.Val, GetExprType(E), MaxWidth); 4437 4438 // I think we only want to look through implicit casts here; if the 4439 // user has an explicit widening cast, we should treat the value as 4440 // being of the new, wider type. 4441 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 4442 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 4443 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 4444 4445 IntRange OutputTypeRange = IntRange::forValueOfType(C, GetExprType(CE)); 4446 4447 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 4448 4449 // Assume that non-integer casts can span the full range of the type. 4450 if (!isIntegerCast) 4451 return OutputTypeRange; 4452 4453 IntRange SubRange 4454 = GetExprRange(C, CE->getSubExpr(), 4455 std::min(MaxWidth, OutputTypeRange.Width)); 4456 4457 // Bail out if the subexpr's range is as wide as the cast type. 4458 if (SubRange.Width >= OutputTypeRange.Width) 4459 return OutputTypeRange; 4460 4461 // Otherwise, we take the smaller width, and we're non-negative if 4462 // either the output type or the subexpr is. 4463 return IntRange(SubRange.Width, 4464 SubRange.NonNegative || OutputTypeRange.NonNegative); 4465 } 4466 4467 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 4468 // If we can fold the condition, just take that operand. 4469 bool CondResult; 4470 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 4471 return GetExprRange(C, CondResult ? CO->getTrueExpr() 4472 : CO->getFalseExpr(), 4473 MaxWidth); 4474 4475 // Otherwise, conservatively merge. 4476 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 4477 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 4478 return IntRange::join(L, R); 4479 } 4480 4481 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4482 switch (BO->getOpcode()) { 4483 4484 // Boolean-valued operations are single-bit and positive. 4485 case BO_LAnd: 4486 case BO_LOr: 4487 case BO_LT: 4488 case BO_GT: 4489 case BO_LE: 4490 case BO_GE: 4491 case BO_EQ: 4492 case BO_NE: 4493 return IntRange::forBoolType(); 4494 4495 // The type of the assignments is the type of the LHS, so the RHS 4496 // is not necessarily the same type. 4497 case BO_MulAssign: 4498 case BO_DivAssign: 4499 case BO_RemAssign: 4500 case BO_AddAssign: 4501 case BO_SubAssign: 4502 case BO_XorAssign: 4503 case BO_OrAssign: 4504 // TODO: bitfields? 4505 return IntRange::forValueOfType(C, GetExprType(E)); 4506 4507 // Simple assignments just pass through the RHS, which will have 4508 // been coerced to the LHS type. 4509 case BO_Assign: 4510 // TODO: bitfields? 4511 return GetExprRange(C, BO->getRHS(), MaxWidth); 4512 4513 // Operations with opaque sources are black-listed. 4514 case BO_PtrMemD: 4515 case BO_PtrMemI: 4516 return IntRange::forValueOfType(C, GetExprType(E)); 4517 4518 // Bitwise-and uses the *infinum* of the two source ranges. 4519 case BO_And: 4520 case BO_AndAssign: 4521 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 4522 GetExprRange(C, BO->getRHS(), MaxWidth)); 4523 4524 // Left shift gets black-listed based on a judgement call. 4525 case BO_Shl: 4526 // ...except that we want to treat '1 << (blah)' as logically 4527 // positive. It's an important idiom. 4528 if (IntegerLiteral *I 4529 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 4530 if (I->getValue() == 1) { 4531 IntRange R = IntRange::forValueOfType(C, GetExprType(E)); 4532 return IntRange(R.Width, /*NonNegative*/ true); 4533 } 4534 } 4535 // fallthrough 4536 4537 case BO_ShlAssign: 4538 return IntRange::forValueOfType(C, GetExprType(E)); 4539 4540 // Right shift by a constant can narrow its left argument. 4541 case BO_Shr: 4542 case BO_ShrAssign: { 4543 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4544 4545 // If the shift amount is a positive constant, drop the width by 4546 // that much. 4547 llvm::APSInt shift; 4548 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 4549 shift.isNonNegative()) { 4550 unsigned zext = shift.getZExtValue(); 4551 if (zext >= L.Width) 4552 L.Width = (L.NonNegative ? 0 : 1); 4553 else 4554 L.Width -= zext; 4555 } 4556 4557 return L; 4558 } 4559 4560 // Comma acts as its right operand. 4561 case BO_Comma: 4562 return GetExprRange(C, BO->getRHS(), MaxWidth); 4563 4564 // Black-list pointer subtractions. 4565 case BO_Sub: 4566 if (BO->getLHS()->getType()->isPointerType()) 4567 return IntRange::forValueOfType(C, GetExprType(E)); 4568 break; 4569 4570 // The width of a division result is mostly determined by the size 4571 // of the LHS. 4572 case BO_Div: { 4573 // Don't 'pre-truncate' the operands. 4574 unsigned opWidth = C.getIntWidth(GetExprType(E)); 4575 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4576 4577 // If the divisor is constant, use that. 4578 llvm::APSInt divisor; 4579 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 4580 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 4581 if (log2 >= L.Width) 4582 L.Width = (L.NonNegative ? 0 : 1); 4583 else 4584 L.Width = std::min(L.Width - log2, MaxWidth); 4585 return L; 4586 } 4587 4588 // Otherwise, just use the LHS's width. 4589 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4590 return IntRange(L.Width, L.NonNegative && R.NonNegative); 4591 } 4592 4593 // The result of a remainder can't be larger than the result of 4594 // either side. 4595 case BO_Rem: { 4596 // Don't 'pre-truncate' the operands. 4597 unsigned opWidth = C.getIntWidth(GetExprType(E)); 4598 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4599 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4600 4601 IntRange meet = IntRange::meet(L, R); 4602 meet.Width = std::min(meet.Width, MaxWidth); 4603 return meet; 4604 } 4605 4606 // The default behavior is okay for these. 4607 case BO_Mul: 4608 case BO_Add: 4609 case BO_Xor: 4610 case BO_Or: 4611 break; 4612 } 4613 4614 // The default case is to treat the operation as if it were closed 4615 // on the narrowest type that encompasses both operands. 4616 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4617 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 4618 return IntRange::join(L, R); 4619 } 4620 4621 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 4622 switch (UO->getOpcode()) { 4623 // Boolean-valued operations are white-listed. 4624 case UO_LNot: 4625 return IntRange::forBoolType(); 4626 4627 // Operations with opaque sources are black-listed. 4628 case UO_Deref: 4629 case UO_AddrOf: // should be impossible 4630 return IntRange::forValueOfType(C, GetExprType(E)); 4631 4632 default: 4633 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 4634 } 4635 } 4636 4637 if (FieldDecl *BitField = E->getSourceBitField()) 4638 return IntRange(BitField->getBitWidthValue(C), 4639 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 4640 4641 return IntRange::forValueOfType(C, GetExprType(E)); 4642} 4643 4644static IntRange GetExprRange(ASTContext &C, Expr *E) { 4645 return GetExprRange(C, E, C.getIntWidth(GetExprType(E))); 4646} 4647 4648/// Checks whether the given value, which currently has the given 4649/// source semantics, has the same value when coerced through the 4650/// target semantics. 4651static bool IsSameFloatAfterCast(const llvm::APFloat &value, 4652 const llvm::fltSemantics &Src, 4653 const llvm::fltSemantics &Tgt) { 4654 llvm::APFloat truncated = value; 4655 4656 bool ignored; 4657 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 4658 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 4659 4660 return truncated.bitwiseIsEqual(value); 4661} 4662 4663/// Checks whether the given value, which currently has the given 4664/// source semantics, has the same value when coerced through the 4665/// target semantics. 4666/// 4667/// The value might be a vector of floats (or a complex number). 4668static bool IsSameFloatAfterCast(const APValue &value, 4669 const llvm::fltSemantics &Src, 4670 const llvm::fltSemantics &Tgt) { 4671 if (value.isFloat()) 4672 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 4673 4674 if (value.isVector()) { 4675 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 4676 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 4677 return false; 4678 return true; 4679 } 4680 4681 assert(value.isComplexFloat()); 4682 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 4683 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 4684} 4685 4686static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 4687 4688static bool IsZero(Sema &S, Expr *E) { 4689 // Suppress cases where we are comparing against an enum constant. 4690 if (const DeclRefExpr *DR = 4691 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 4692 if (isa<EnumConstantDecl>(DR->getDecl())) 4693 return false; 4694 4695 // Suppress cases where the '0' value is expanded from a macro. 4696 if (E->getLocStart().isMacroID()) 4697 return false; 4698 4699 llvm::APSInt Value; 4700 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 4701} 4702 4703static bool HasEnumType(Expr *E) { 4704 // Strip off implicit integral promotions. 4705 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 4706 if (ICE->getCastKind() != CK_IntegralCast && 4707 ICE->getCastKind() != CK_NoOp) 4708 break; 4709 E = ICE->getSubExpr(); 4710 } 4711 4712 return E->getType()->isEnumeralType(); 4713} 4714 4715static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 4716 BinaryOperatorKind op = E->getOpcode(); 4717 if (E->isValueDependent()) 4718 return; 4719 4720 if (op == BO_LT && IsZero(S, E->getRHS())) { 4721 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4722 << "< 0" << "false" << HasEnumType(E->getLHS()) 4723 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4724 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 4725 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4726 << ">= 0" << "true" << HasEnumType(E->getLHS()) 4727 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4728 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 4729 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4730 << "0 >" << "false" << HasEnumType(E->getRHS()) 4731 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4732 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 4733 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4734 << "0 <=" << "true" << HasEnumType(E->getRHS()) 4735 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4736 } 4737} 4738 4739static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E, 4740 Expr *Constant, Expr *Other, 4741 llvm::APSInt Value, 4742 bool RhsConstant) { 4743 // 0 values are handled later by CheckTrivialUnsignedComparison(). 4744 if (Value == 0) 4745 return; 4746 4747 BinaryOperatorKind op = E->getOpcode(); 4748 QualType OtherT = Other->getType(); 4749 QualType ConstantT = Constant->getType(); 4750 QualType CommonT = E->getLHS()->getType(); 4751 if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT)) 4752 return; 4753 assert((OtherT->isIntegerType() && ConstantT->isIntegerType()) 4754 && "comparison with non-integer type"); 4755 4756 bool ConstantSigned = ConstantT->isSignedIntegerType(); 4757 bool CommonSigned = CommonT->isSignedIntegerType(); 4758 4759 bool EqualityOnly = false; 4760 4761 // TODO: Investigate using GetExprRange() to get tighter bounds on 4762 // on the bit ranges. 4763 IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT); 4764 unsigned OtherWidth = OtherRange.Width; 4765 4766 if (CommonSigned) { 4767 // The common type is signed, therefore no signed to unsigned conversion. 4768 if (!OtherRange.NonNegative) { 4769 // Check that the constant is representable in type OtherT. 4770 if (ConstantSigned) { 4771 if (OtherWidth >= Value.getMinSignedBits()) 4772 return; 4773 } else { // !ConstantSigned 4774 if (OtherWidth >= Value.getActiveBits() + 1) 4775 return; 4776 } 4777 } else { // !OtherSigned 4778 // Check that the constant is representable in type OtherT. 4779 // Negative values are out of range. 4780 if (ConstantSigned) { 4781 if (Value.isNonNegative() && OtherWidth >= Value.getActiveBits()) 4782 return; 4783 } else { // !ConstantSigned 4784 if (OtherWidth >= Value.getActiveBits()) 4785 return; 4786 } 4787 } 4788 } else { // !CommonSigned 4789 if (OtherRange.NonNegative) { 4790 if (OtherWidth >= Value.getActiveBits()) 4791 return; 4792 } else if (!OtherRange.NonNegative && !ConstantSigned) { 4793 // Check to see if the constant is representable in OtherT. 4794 if (OtherWidth > Value.getActiveBits()) 4795 return; 4796 // Check to see if the constant is equivalent to a negative value 4797 // cast to CommonT. 4798 if (S.Context.getIntWidth(ConstantT) == S.Context.getIntWidth(CommonT) && 4799 Value.isNegative() && Value.getMinSignedBits() <= OtherWidth) 4800 return; 4801 // The constant value rests between values that OtherT can represent after 4802 // conversion. Relational comparison still works, but equality 4803 // comparisons will be tautological. 4804 EqualityOnly = true; 4805 } else { // OtherSigned && ConstantSigned 4806 assert(0 && "Two signed types converted to unsigned types."); 4807 } 4808 } 4809 4810 bool PositiveConstant = !ConstantSigned || Value.isNonNegative(); 4811 4812 bool IsTrue = true; 4813 if (op == BO_EQ || op == BO_NE) { 4814 IsTrue = op == BO_NE; 4815 } else if (EqualityOnly) { 4816 return; 4817 } else if (RhsConstant) { 4818 if (op == BO_GT || op == BO_GE) 4819 IsTrue = !PositiveConstant; 4820 else // op == BO_LT || op == BO_LE 4821 IsTrue = PositiveConstant; 4822 } else { 4823 if (op == BO_LT || op == BO_LE) 4824 IsTrue = !PositiveConstant; 4825 else // op == BO_GT || op == BO_GE 4826 IsTrue = PositiveConstant; 4827 } 4828 4829 // If this is a comparison to an enum constant, include that 4830 // constant in the diagnostic. 4831 const EnumConstantDecl *ED = 0; 4832 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Constant)) 4833 ED = dyn_cast<EnumConstantDecl>(DR->getDecl()); 4834 4835 SmallString<64> PrettySourceValue; 4836 llvm::raw_svector_ostream OS(PrettySourceValue); 4837 if (ED) 4838 OS << '\'' << *ED << "' (" << Value << ")"; 4839 else 4840 OS << Value; 4841 4842 S.Diag(E->getOperatorLoc(), diag::warn_out_of_range_compare) 4843 << OS.str() << OtherT << IsTrue 4844 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4845} 4846 4847/// Analyze the operands of the given comparison. Implements the 4848/// fallback case from AnalyzeComparison. 4849static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 4850 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4851 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4852} 4853 4854/// \brief Implements -Wsign-compare. 4855/// 4856/// \param E the binary operator to check for warnings 4857static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 4858 // The type the comparison is being performed in. 4859 QualType T = E->getLHS()->getType(); 4860 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 4861 && "comparison with mismatched types"); 4862 if (E->isValueDependent()) 4863 return AnalyzeImpConvsInComparison(S, E); 4864 4865 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 4866 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 4867 4868 bool IsComparisonConstant = false; 4869 4870 // Check whether an integer constant comparison results in a value 4871 // of 'true' or 'false'. 4872 if (T->isIntegralType(S.Context)) { 4873 llvm::APSInt RHSValue; 4874 bool IsRHSIntegralLiteral = 4875 RHS->isIntegerConstantExpr(RHSValue, S.Context); 4876 llvm::APSInt LHSValue; 4877 bool IsLHSIntegralLiteral = 4878 LHS->isIntegerConstantExpr(LHSValue, S.Context); 4879 if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral) 4880 DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true); 4881 else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral) 4882 DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false); 4883 else 4884 IsComparisonConstant = 4885 (IsRHSIntegralLiteral && IsLHSIntegralLiteral); 4886 } else if (!T->hasUnsignedIntegerRepresentation()) 4887 IsComparisonConstant = E->isIntegerConstantExpr(S.Context); 4888 4889 // We don't do anything special if this isn't an unsigned integral 4890 // comparison: we're only interested in integral comparisons, and 4891 // signed comparisons only happen in cases we don't care to warn about. 4892 // 4893 // We also don't care about value-dependent expressions or expressions 4894 // whose result is a constant. 4895 if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant) 4896 return AnalyzeImpConvsInComparison(S, E); 4897 4898 // Check to see if one of the (unmodified) operands is of different 4899 // signedness. 4900 Expr *signedOperand, *unsignedOperand; 4901 if (LHS->getType()->hasSignedIntegerRepresentation()) { 4902 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 4903 "unsigned comparison between two signed integer expressions?"); 4904 signedOperand = LHS; 4905 unsignedOperand = RHS; 4906 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 4907 signedOperand = RHS; 4908 unsignedOperand = LHS; 4909 } else { 4910 CheckTrivialUnsignedComparison(S, E); 4911 return AnalyzeImpConvsInComparison(S, E); 4912 } 4913 4914 // Otherwise, calculate the effective range of the signed operand. 4915 IntRange signedRange = GetExprRange(S.Context, signedOperand); 4916 4917 // Go ahead and analyze implicit conversions in the operands. Note 4918 // that we skip the implicit conversions on both sides. 4919 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 4920 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 4921 4922 // If the signed range is non-negative, -Wsign-compare won't fire, 4923 // but we should still check for comparisons which are always true 4924 // or false. 4925 if (signedRange.NonNegative) 4926 return CheckTrivialUnsignedComparison(S, E); 4927 4928 // For (in)equality comparisons, if the unsigned operand is a 4929 // constant which cannot collide with a overflowed signed operand, 4930 // then reinterpreting the signed operand as unsigned will not 4931 // change the result of the comparison. 4932 if (E->isEqualityOp()) { 4933 unsigned comparisonWidth = S.Context.getIntWidth(T); 4934 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 4935 4936 // We should never be unable to prove that the unsigned operand is 4937 // non-negative. 4938 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 4939 4940 if (unsignedRange.Width < comparisonWidth) 4941 return; 4942 } 4943 4944 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 4945 S.PDiag(diag::warn_mixed_sign_comparison) 4946 << LHS->getType() << RHS->getType() 4947 << LHS->getSourceRange() << RHS->getSourceRange()); 4948} 4949 4950/// Analyzes an attempt to assign the given value to a bitfield. 4951/// 4952/// Returns true if there was something fishy about the attempt. 4953static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 4954 SourceLocation InitLoc) { 4955 assert(Bitfield->isBitField()); 4956 if (Bitfield->isInvalidDecl()) 4957 return false; 4958 4959 // White-list bool bitfields. 4960 if (Bitfield->getType()->isBooleanType()) 4961 return false; 4962 4963 // Ignore value- or type-dependent expressions. 4964 if (Bitfield->getBitWidth()->isValueDependent() || 4965 Bitfield->getBitWidth()->isTypeDependent() || 4966 Init->isValueDependent() || 4967 Init->isTypeDependent()) 4968 return false; 4969 4970 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 4971 4972 llvm::APSInt Value; 4973 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 4974 return false; 4975 4976 unsigned OriginalWidth = Value.getBitWidth(); 4977 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 4978 4979 if (OriginalWidth <= FieldWidth) 4980 return false; 4981 4982 // Compute the value which the bitfield will contain. 4983 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 4984 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 4985 4986 // Check whether the stored value is equal to the original value. 4987 TruncatedValue = TruncatedValue.extend(OriginalWidth); 4988 if (llvm::APSInt::isSameValue(Value, TruncatedValue)) 4989 return false; 4990 4991 // Special-case bitfields of width 1: booleans are naturally 0/1, and 4992 // therefore don't strictly fit into a signed bitfield of width 1. 4993 if (FieldWidth == 1 && Value == 1) 4994 return false; 4995 4996 std::string PrettyValue = Value.toString(10); 4997 std::string PrettyTrunc = TruncatedValue.toString(10); 4998 4999 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 5000 << PrettyValue << PrettyTrunc << OriginalInit->getType() 5001 << Init->getSourceRange(); 5002 5003 return true; 5004} 5005 5006/// Analyze the given simple or compound assignment for warning-worthy 5007/// operations. 5008static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 5009 // Just recurse on the LHS. 5010 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 5011 5012 // We want to recurse on the RHS as normal unless we're assigning to 5013 // a bitfield. 5014 if (FieldDecl *Bitfield = E->getLHS()->getSourceBitField()) { 5015 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 5016 E->getOperatorLoc())) { 5017 // Recurse, ignoring any implicit conversions on the RHS. 5018 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 5019 E->getOperatorLoc()); 5020 } 5021 } 5022 5023 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 5024} 5025 5026/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 5027static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 5028 SourceLocation CContext, unsigned diag, 5029 bool pruneControlFlow = false) { 5030 if (pruneControlFlow) { 5031 S.DiagRuntimeBehavior(E->getExprLoc(), E, 5032 S.PDiag(diag) 5033 << SourceType << T << E->getSourceRange() 5034 << SourceRange(CContext)); 5035 return; 5036 } 5037 S.Diag(E->getExprLoc(), diag) 5038 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 5039} 5040 5041/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 5042static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 5043 SourceLocation CContext, unsigned diag, 5044 bool pruneControlFlow = false) { 5045 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 5046} 5047 5048/// Diagnose an implicit cast from a literal expression. Does not warn when the 5049/// cast wouldn't lose information. 5050void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 5051 SourceLocation CContext) { 5052 // Try to convert the literal exactly to an integer. If we can, don't warn. 5053 bool isExact = false; 5054 const llvm::APFloat &Value = FL->getValue(); 5055 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 5056 T->hasUnsignedIntegerRepresentation()); 5057 if (Value.convertToInteger(IntegerValue, 5058 llvm::APFloat::rmTowardZero, &isExact) 5059 == llvm::APFloat::opOK && isExact) 5060 return; 5061 5062 // FIXME: Force the precision of the source value down so we don't print 5063 // digits which are usually useless (we don't really care here if we 5064 // truncate a digit by accident in edge cases). Ideally, APFloat::toString 5065 // would automatically print the shortest representation, but it's a bit 5066 // tricky to implement. 5067 SmallString<16> PrettySourceValue; 5068 unsigned precision = llvm::APFloat::semanticsPrecision(Value.getSemantics()); 5069 precision = (precision * 59 + 195) / 196; 5070 Value.toString(PrettySourceValue, precision); 5071 5072 SmallString<16> PrettyTargetValue; 5073 if (T->isSpecificBuiltinType(BuiltinType::Bool)) 5074 PrettyTargetValue = IntegerValue == 0 ? "false" : "true"; 5075 else 5076 IntegerValue.toString(PrettyTargetValue); 5077 5078 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 5079 << FL->getType() << T.getUnqualifiedType() << PrettySourceValue 5080 << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext); 5081} 5082 5083std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 5084 if (!Range.Width) return "0"; 5085 5086 llvm::APSInt ValueInRange = Value; 5087 ValueInRange.setIsSigned(!Range.NonNegative); 5088 ValueInRange = ValueInRange.trunc(Range.Width); 5089 return ValueInRange.toString(10); 5090} 5091 5092static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) { 5093 if (!isa<ImplicitCastExpr>(Ex)) 5094 return false; 5095 5096 Expr *InnerE = Ex->IgnoreParenImpCasts(); 5097 const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr(); 5098 const Type *Source = 5099 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 5100 if (Target->isDependentType()) 5101 return false; 5102 5103 const BuiltinType *FloatCandidateBT = 5104 dyn_cast<BuiltinType>(ToBool ? Source : Target); 5105 const Type *BoolCandidateType = ToBool ? Target : Source; 5106 5107 return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) && 5108 FloatCandidateBT && (FloatCandidateBT->isFloatingPoint())); 5109} 5110 5111void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall, 5112 SourceLocation CC) { 5113 unsigned NumArgs = TheCall->getNumArgs(); 5114 for (unsigned i = 0; i < NumArgs; ++i) { 5115 Expr *CurrA = TheCall->getArg(i); 5116 if (!IsImplicitBoolFloatConversion(S, CurrA, true)) 5117 continue; 5118 5119 bool IsSwapped = ((i > 0) && 5120 IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false)); 5121 IsSwapped |= ((i < (NumArgs - 1)) && 5122 IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false)); 5123 if (IsSwapped) { 5124 // Warn on this floating-point to bool conversion. 5125 DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(), 5126 CurrA->getType(), CC, 5127 diag::warn_impcast_floating_point_to_bool); 5128 } 5129 } 5130} 5131 5132void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 5133 SourceLocation CC, bool *ICContext = 0) { 5134 if (E->isTypeDependent() || E->isValueDependent()) return; 5135 5136 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 5137 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 5138 if (Source == Target) return; 5139 if (Target->isDependentType()) return; 5140 5141 // If the conversion context location is invalid don't complain. We also 5142 // don't want to emit a warning if the issue occurs from the expansion of 5143 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 5144 // delay this check as long as possible. Once we detect we are in that 5145 // scenario, we just return. 5146 if (CC.isInvalid()) 5147 return; 5148 5149 // Diagnose implicit casts to bool. 5150 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 5151 if (isa<StringLiteral>(E)) 5152 // Warn on string literal to bool. Checks for string literals in logical 5153 // expressions, for instances, assert(0 && "error here"), is prevented 5154 // by a check in AnalyzeImplicitConversions(). 5155 return DiagnoseImpCast(S, E, T, CC, 5156 diag::warn_impcast_string_literal_to_bool); 5157 if (Source->isFunctionType()) { 5158 // Warn on function to bool. Checks free functions and static member 5159 // functions. Weakly imported functions are excluded from the check, 5160 // since it's common to test their value to check whether the linker 5161 // found a definition for them. 5162 ValueDecl *D = 0; 5163 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 5164 D = R->getDecl(); 5165 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 5166 D = M->getMemberDecl(); 5167 } 5168 5169 if (D && !D->isWeak()) { 5170 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 5171 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 5172 << F << E->getSourceRange() << SourceRange(CC); 5173 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 5174 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 5175 QualType ReturnType; 5176 UnresolvedSet<4> NonTemplateOverloads; 5177 S.tryExprAsCall(*E, ReturnType, NonTemplateOverloads); 5178 if (!ReturnType.isNull() 5179 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 5180 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 5181 << FixItHint::CreateInsertion( 5182 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 5183 return; 5184 } 5185 } 5186 } 5187 } 5188 5189 // Strip vector types. 5190 if (isa<VectorType>(Source)) { 5191 if (!isa<VectorType>(Target)) { 5192 if (S.SourceMgr.isInSystemMacro(CC)) 5193 return; 5194 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 5195 } 5196 5197 // If the vector cast is cast between two vectors of the same size, it is 5198 // a bitcast, not a conversion. 5199 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 5200 return; 5201 5202 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 5203 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 5204 } 5205 5206 // Strip complex types. 5207 if (isa<ComplexType>(Source)) { 5208 if (!isa<ComplexType>(Target)) { 5209 if (S.SourceMgr.isInSystemMacro(CC)) 5210 return; 5211 5212 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 5213 } 5214 5215 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 5216 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 5217 } 5218 5219 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 5220 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 5221 5222 // If the source is floating point... 5223 if (SourceBT && SourceBT->isFloatingPoint()) { 5224 // ...and the target is floating point... 5225 if (TargetBT && TargetBT->isFloatingPoint()) { 5226 // ...then warn if we're dropping FP rank. 5227 5228 // Builtin FP kinds are ordered by increasing FP rank. 5229 if (SourceBT->getKind() > TargetBT->getKind()) { 5230 // Don't warn about float constants that are precisely 5231 // representable in the target type. 5232 Expr::EvalResult result; 5233 if (E->EvaluateAsRValue(result, S.Context)) { 5234 // Value might be a float, a float vector, or a float complex. 5235 if (IsSameFloatAfterCast(result.Val, 5236 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 5237 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 5238 return; 5239 } 5240 5241 if (S.SourceMgr.isInSystemMacro(CC)) 5242 return; 5243 5244 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 5245 } 5246 return; 5247 } 5248 5249 // If the target is integral, always warn. 5250 if (TargetBT && TargetBT->isInteger()) { 5251 if (S.SourceMgr.isInSystemMacro(CC)) 5252 return; 5253 5254 Expr *InnerE = E->IgnoreParenImpCasts(); 5255 // We also want to warn on, e.g., "int i = -1.234" 5256 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 5257 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 5258 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 5259 5260 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 5261 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 5262 } else { 5263 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 5264 } 5265 } 5266 5267 // If the target is bool, warn if expr is a function or method call. 5268 if (Target->isSpecificBuiltinType(BuiltinType::Bool) && 5269 isa<CallExpr>(E)) { 5270 // Check last argument of function call to see if it is an 5271 // implicit cast from a type matching the type the result 5272 // is being cast to. 5273 CallExpr *CEx = cast<CallExpr>(E); 5274 unsigned NumArgs = CEx->getNumArgs(); 5275 if (NumArgs > 0) { 5276 Expr *LastA = CEx->getArg(NumArgs - 1); 5277 Expr *InnerE = LastA->IgnoreParenImpCasts(); 5278 const Type *InnerType = 5279 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 5280 if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) { 5281 // Warn on this floating-point to bool conversion 5282 DiagnoseImpCast(S, E, T, CC, 5283 diag::warn_impcast_floating_point_to_bool); 5284 } 5285 } 5286 } 5287 return; 5288 } 5289 5290 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 5291 == Expr::NPCK_GNUNull) && !Target->isAnyPointerType() 5292 && !Target->isBlockPointerType() && !Target->isMemberPointerType() 5293 && Target->isScalarType() && !Target->isNullPtrType()) { 5294 SourceLocation Loc = E->getSourceRange().getBegin(); 5295 if (Loc.isMacroID()) 5296 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 5297 if (!Loc.isMacroID() || CC.isMacroID()) 5298 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 5299 << T << clang::SourceRange(CC) 5300 << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T)); 5301 } 5302 5303 if (!Source->isIntegerType() || !Target->isIntegerType()) 5304 return; 5305 5306 // TODO: remove this early return once the false positives for constant->bool 5307 // in templates, macros, etc, are reduced or removed. 5308 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 5309 return; 5310 5311 IntRange SourceRange = GetExprRange(S.Context, E); 5312 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 5313 5314 if (SourceRange.Width > TargetRange.Width) { 5315 // If the source is a constant, use a default-on diagnostic. 5316 // TODO: this should happen for bitfield stores, too. 5317 llvm::APSInt Value(32); 5318 if (E->isIntegerConstantExpr(Value, S.Context)) { 5319 if (S.SourceMgr.isInSystemMacro(CC)) 5320 return; 5321 5322 std::string PrettySourceValue = Value.toString(10); 5323 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 5324 5325 S.DiagRuntimeBehavior(E->getExprLoc(), E, 5326 S.PDiag(diag::warn_impcast_integer_precision_constant) 5327 << PrettySourceValue << PrettyTargetValue 5328 << E->getType() << T << E->getSourceRange() 5329 << clang::SourceRange(CC)); 5330 return; 5331 } 5332 5333 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 5334 if (S.SourceMgr.isInSystemMacro(CC)) 5335 return; 5336 5337 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 5338 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 5339 /* pruneControlFlow */ true); 5340 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 5341 } 5342 5343 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 5344 (!TargetRange.NonNegative && SourceRange.NonNegative && 5345 SourceRange.Width == TargetRange.Width)) { 5346 5347 if (S.SourceMgr.isInSystemMacro(CC)) 5348 return; 5349 5350 unsigned DiagID = diag::warn_impcast_integer_sign; 5351 5352 // Traditionally, gcc has warned about this under -Wsign-compare. 5353 // We also want to warn about it in -Wconversion. 5354 // So if -Wconversion is off, use a completely identical diagnostic 5355 // in the sign-compare group. 5356 // The conditional-checking code will 5357 if (ICContext) { 5358 DiagID = diag::warn_impcast_integer_sign_conditional; 5359 *ICContext = true; 5360 } 5361 5362 return DiagnoseImpCast(S, E, T, CC, DiagID); 5363 } 5364 5365 // Diagnose conversions between different enumeration types. 5366 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 5367 // type, to give us better diagnostics. 5368 QualType SourceType = E->getType(); 5369 if (!S.getLangOpts().CPlusPlus) { 5370 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5371 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 5372 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 5373 SourceType = S.Context.getTypeDeclType(Enum); 5374 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 5375 } 5376 } 5377 5378 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 5379 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 5380 if (SourceEnum->getDecl()->hasNameForLinkage() && 5381 TargetEnum->getDecl()->hasNameForLinkage() && 5382 SourceEnum != TargetEnum) { 5383 if (S.SourceMgr.isInSystemMacro(CC)) 5384 return; 5385 5386 return DiagnoseImpCast(S, E, SourceType, T, CC, 5387 diag::warn_impcast_different_enum_types); 5388 } 5389 5390 return; 5391} 5392 5393void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5394 SourceLocation CC, QualType T); 5395 5396void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 5397 SourceLocation CC, bool &ICContext) { 5398 E = E->IgnoreParenImpCasts(); 5399 5400 if (isa<ConditionalOperator>(E)) 5401 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T); 5402 5403 AnalyzeImplicitConversions(S, E, CC); 5404 if (E->getType() != T) 5405 return CheckImplicitConversion(S, E, T, CC, &ICContext); 5406 return; 5407} 5408 5409void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5410 SourceLocation CC, QualType T) { 5411 AnalyzeImplicitConversions(S, E->getCond(), CC); 5412 5413 bool Suspicious = false; 5414 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 5415 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 5416 5417 // If -Wconversion would have warned about either of the candidates 5418 // for a signedness conversion to the context type... 5419 if (!Suspicious) return; 5420 5421 // ...but it's currently ignored... 5422 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 5423 CC)) 5424 return; 5425 5426 // ...then check whether it would have warned about either of the 5427 // candidates for a signedness conversion to the condition type. 5428 if (E->getType() == T) return; 5429 5430 Suspicious = false; 5431 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 5432 E->getType(), CC, &Suspicious); 5433 if (!Suspicious) 5434 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 5435 E->getType(), CC, &Suspicious); 5436} 5437 5438/// AnalyzeImplicitConversions - Find and report any interesting 5439/// implicit conversions in the given expression. There are a couple 5440/// of competing diagnostics here, -Wconversion and -Wsign-compare. 5441void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 5442 QualType T = OrigE->getType(); 5443 Expr *E = OrigE->IgnoreParenImpCasts(); 5444 5445 if (E->isTypeDependent() || E->isValueDependent()) 5446 return; 5447 5448 // For conditional operators, we analyze the arguments as if they 5449 // were being fed directly into the output. 5450 if (isa<ConditionalOperator>(E)) { 5451 ConditionalOperator *CO = cast<ConditionalOperator>(E); 5452 CheckConditionalOperator(S, CO, CC, T); 5453 return; 5454 } 5455 5456 // Check implicit argument conversions for function calls. 5457 if (CallExpr *Call = dyn_cast<CallExpr>(E)) 5458 CheckImplicitArgumentConversions(S, Call, CC); 5459 5460 // Go ahead and check any implicit conversions we might have skipped. 5461 // The non-canonical typecheck is just an optimization; 5462 // CheckImplicitConversion will filter out dead implicit conversions. 5463 if (E->getType() != T) 5464 CheckImplicitConversion(S, E, T, CC); 5465 5466 // Now continue drilling into this expression. 5467 5468 if (PseudoObjectExpr * POE = dyn_cast<PseudoObjectExpr>(E)) { 5469 if (POE->getResultExpr()) 5470 E = POE->getResultExpr(); 5471 } 5472 5473 if (const OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E)) 5474 return AnalyzeImplicitConversions(S, OVE->getSourceExpr(), CC); 5475 5476 // Skip past explicit casts. 5477 if (isa<ExplicitCastExpr>(E)) { 5478 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 5479 return AnalyzeImplicitConversions(S, E, CC); 5480 } 5481 5482 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5483 // Do a somewhat different check with comparison operators. 5484 if (BO->isComparisonOp()) 5485 return AnalyzeComparison(S, BO); 5486 5487 // And with simple assignments. 5488 if (BO->getOpcode() == BO_Assign) 5489 return AnalyzeAssignment(S, BO); 5490 } 5491 5492 // These break the otherwise-useful invariant below. Fortunately, 5493 // we don't really need to recurse into them, because any internal 5494 // expressions should have been analyzed already when they were 5495 // built into statements. 5496 if (isa<StmtExpr>(E)) return; 5497 5498 // Don't descend into unevaluated contexts. 5499 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 5500 5501 // Now just recurse over the expression's children. 5502 CC = E->getExprLoc(); 5503 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 5504 bool IsLogicalOperator = BO && BO->isLogicalOp(); 5505 for (Stmt::child_range I = E->children(); I; ++I) { 5506 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 5507 if (!ChildExpr) 5508 continue; 5509 5510 if (IsLogicalOperator && 5511 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 5512 // Ignore checking string literals that are in logical operators. 5513 continue; 5514 AnalyzeImplicitConversions(S, ChildExpr, CC); 5515 } 5516} 5517 5518} // end anonymous namespace 5519 5520/// Diagnoses "dangerous" implicit conversions within the given 5521/// expression (which is a full expression). Implements -Wconversion 5522/// and -Wsign-compare. 5523/// 5524/// \param CC the "context" location of the implicit conversion, i.e. 5525/// the most location of the syntactic entity requiring the implicit 5526/// conversion 5527void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 5528 // Don't diagnose in unevaluated contexts. 5529 if (isUnevaluatedContext()) 5530 return; 5531 5532 // Don't diagnose for value- or type-dependent expressions. 5533 if (E->isTypeDependent() || E->isValueDependent()) 5534 return; 5535 5536 // Check for array bounds violations in cases where the check isn't triggered 5537 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 5538 // ArraySubscriptExpr is on the RHS of a variable initialization. 5539 CheckArrayAccess(E); 5540 5541 // This is not the right CC for (e.g.) a variable initialization. 5542 AnalyzeImplicitConversions(*this, E, CC); 5543} 5544 5545/// Diagnose when expression is an integer constant expression and its evaluation 5546/// results in integer overflow 5547void Sema::CheckForIntOverflow (Expr *E) { 5548 if (isa<BinaryOperator>(E->IgnoreParens())) { 5549 SmallVector<PartialDiagnosticAt, 4> Diags; 5550 E->EvaluateForOverflow(Context, &Diags); 5551 } 5552} 5553 5554namespace { 5555/// \brief Visitor for expressions which looks for unsequenced operations on the 5556/// same object. 5557class SequenceChecker : public EvaluatedExprVisitor<SequenceChecker> { 5558 typedef EvaluatedExprVisitor<SequenceChecker> Base; 5559 5560 /// \brief A tree of sequenced regions within an expression. Two regions are 5561 /// unsequenced if one is an ancestor or a descendent of the other. When we 5562 /// finish processing an expression with sequencing, such as a comma 5563 /// expression, we fold its tree nodes into its parent, since they are 5564 /// unsequenced with respect to nodes we will visit later. 5565 class SequenceTree { 5566 struct Value { 5567 explicit Value(unsigned Parent) : Parent(Parent), Merged(false) {} 5568 unsigned Parent : 31; 5569 bool Merged : 1; 5570 }; 5571 SmallVector<Value, 8> Values; 5572 5573 public: 5574 /// \brief A region within an expression which may be sequenced with respect 5575 /// to some other region. 5576 class Seq { 5577 explicit Seq(unsigned N) : Index(N) {} 5578 unsigned Index; 5579 friend class SequenceTree; 5580 public: 5581 Seq() : Index(0) {} 5582 }; 5583 5584 SequenceTree() { Values.push_back(Value(0)); } 5585 Seq root() const { return Seq(0); } 5586 5587 /// \brief Create a new sequence of operations, which is an unsequenced 5588 /// subset of \p Parent. This sequence of operations is sequenced with 5589 /// respect to other children of \p Parent. 5590 Seq allocate(Seq Parent) { 5591 Values.push_back(Value(Parent.Index)); 5592 return Seq(Values.size() - 1); 5593 } 5594 5595 /// \brief Merge a sequence of operations into its parent. 5596 void merge(Seq S) { 5597 Values[S.Index].Merged = true; 5598 } 5599 5600 /// \brief Determine whether two operations are unsequenced. This operation 5601 /// is asymmetric: \p Cur should be the more recent sequence, and \p Old 5602 /// should have been merged into its parent as appropriate. 5603 bool isUnsequenced(Seq Cur, Seq Old) { 5604 unsigned C = representative(Cur.Index); 5605 unsigned Target = representative(Old.Index); 5606 while (C >= Target) { 5607 if (C == Target) 5608 return true; 5609 C = Values[C].Parent; 5610 } 5611 return false; 5612 } 5613 5614 private: 5615 /// \brief Pick a representative for a sequence. 5616 unsigned representative(unsigned K) { 5617 if (Values[K].Merged) 5618 // Perform path compression as we go. 5619 return Values[K].Parent = representative(Values[K].Parent); 5620 return K; 5621 } 5622 }; 5623 5624 /// An object for which we can track unsequenced uses. 5625 typedef NamedDecl *Object; 5626 5627 /// Different flavors of object usage which we track. We only track the 5628 /// least-sequenced usage of each kind. 5629 enum UsageKind { 5630 /// A read of an object. Multiple unsequenced reads are OK. 5631 UK_Use, 5632 /// A modification of an object which is sequenced before the value 5633 /// computation of the expression, such as ++n in C++. 5634 UK_ModAsValue, 5635 /// A modification of an object which is not sequenced before the value 5636 /// computation of the expression, such as n++. 5637 UK_ModAsSideEffect, 5638 5639 UK_Count = UK_ModAsSideEffect + 1 5640 }; 5641 5642 struct Usage { 5643 Usage() : Use(0), Seq() {} 5644 Expr *Use; 5645 SequenceTree::Seq Seq; 5646 }; 5647 5648 struct UsageInfo { 5649 UsageInfo() : Diagnosed(false) {} 5650 Usage Uses[UK_Count]; 5651 /// Have we issued a diagnostic for this variable already? 5652 bool Diagnosed; 5653 }; 5654 typedef llvm::SmallDenseMap<Object, UsageInfo, 16> UsageInfoMap; 5655 5656 Sema &SemaRef; 5657 /// Sequenced regions within the expression. 5658 SequenceTree Tree; 5659 /// Declaration modifications and references which we have seen. 5660 UsageInfoMap UsageMap; 5661 /// The region we are currently within. 5662 SequenceTree::Seq Region; 5663 /// Filled in with declarations which were modified as a side-effect 5664 /// (that is, post-increment operations). 5665 SmallVectorImpl<std::pair<Object, Usage> > *ModAsSideEffect; 5666 /// Expressions to check later. We defer checking these to reduce 5667 /// stack usage. 5668 SmallVectorImpl<Expr *> &WorkList; 5669 5670 /// RAII object wrapping the visitation of a sequenced subexpression of an 5671 /// expression. At the end of this process, the side-effects of the evaluation 5672 /// become sequenced with respect to the value computation of the result, so 5673 /// we downgrade any UK_ModAsSideEffect within the evaluation to 5674 /// UK_ModAsValue. 5675 struct SequencedSubexpression { 5676 SequencedSubexpression(SequenceChecker &Self) 5677 : Self(Self), OldModAsSideEffect(Self.ModAsSideEffect) { 5678 Self.ModAsSideEffect = &ModAsSideEffect; 5679 } 5680 ~SequencedSubexpression() { 5681 for (unsigned I = 0, E = ModAsSideEffect.size(); I != E; ++I) { 5682 UsageInfo &U = Self.UsageMap[ModAsSideEffect[I].first]; 5683 U.Uses[UK_ModAsSideEffect] = ModAsSideEffect[I].second; 5684 Self.addUsage(U, ModAsSideEffect[I].first, 5685 ModAsSideEffect[I].second.Use, UK_ModAsValue); 5686 } 5687 Self.ModAsSideEffect = OldModAsSideEffect; 5688 } 5689 5690 SequenceChecker &Self; 5691 SmallVector<std::pair<Object, Usage>, 4> ModAsSideEffect; 5692 SmallVectorImpl<std::pair<Object, Usage> > *OldModAsSideEffect; 5693 }; 5694 5695 /// RAII object wrapping the visitation of a subexpression which we might 5696 /// choose to evaluate as a constant. If any subexpression is evaluated and 5697 /// found to be non-constant, this allows us to suppress the evaluation of 5698 /// the outer expression. 5699 class EvaluationTracker { 5700 public: 5701 EvaluationTracker(SequenceChecker &Self) 5702 : Self(Self), Prev(Self.EvalTracker), EvalOK(true) { 5703 Self.EvalTracker = this; 5704 } 5705 ~EvaluationTracker() { 5706 Self.EvalTracker = Prev; 5707 if (Prev) 5708 Prev->EvalOK &= EvalOK; 5709 } 5710 5711 bool evaluate(const Expr *E, bool &Result) { 5712 if (!EvalOK || E->isValueDependent()) 5713 return false; 5714 EvalOK = E->EvaluateAsBooleanCondition(Result, Self.SemaRef.Context); 5715 return EvalOK; 5716 } 5717 5718 private: 5719 SequenceChecker &Self; 5720 EvaluationTracker *Prev; 5721 bool EvalOK; 5722 } *EvalTracker; 5723 5724 /// \brief Find the object which is produced by the specified expression, 5725 /// if any. 5726 Object getObject(Expr *E, bool Mod) const { 5727 E = E->IgnoreParenCasts(); 5728 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 5729 if (Mod && (UO->getOpcode() == UO_PreInc || UO->getOpcode() == UO_PreDec)) 5730 return getObject(UO->getSubExpr(), Mod); 5731 } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5732 if (BO->getOpcode() == BO_Comma) 5733 return getObject(BO->getRHS(), Mod); 5734 if (Mod && BO->isAssignmentOp()) 5735 return getObject(BO->getLHS(), Mod); 5736 } else if (MemberExpr *ME = dyn_cast<MemberExpr>(E)) { 5737 // FIXME: Check for more interesting cases, like "x.n = ++x.n". 5738 if (isa<CXXThisExpr>(ME->getBase()->IgnoreParenCasts())) 5739 return ME->getMemberDecl(); 5740 } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5741 // FIXME: If this is a reference, map through to its value. 5742 return DRE->getDecl(); 5743 return 0; 5744 } 5745 5746 /// \brief Note that an object was modified or used by an expression. 5747 void addUsage(UsageInfo &UI, Object O, Expr *Ref, UsageKind UK) { 5748 Usage &U = UI.Uses[UK]; 5749 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) { 5750 if (UK == UK_ModAsSideEffect && ModAsSideEffect) 5751 ModAsSideEffect->push_back(std::make_pair(O, U)); 5752 U.Use = Ref; 5753 U.Seq = Region; 5754 } 5755 } 5756 /// \brief Check whether a modification or use conflicts with a prior usage. 5757 void checkUsage(Object O, UsageInfo &UI, Expr *Ref, UsageKind OtherKind, 5758 bool IsModMod) { 5759 if (UI.Diagnosed) 5760 return; 5761 5762 const Usage &U = UI.Uses[OtherKind]; 5763 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) 5764 return; 5765 5766 Expr *Mod = U.Use; 5767 Expr *ModOrUse = Ref; 5768 if (OtherKind == UK_Use) 5769 std::swap(Mod, ModOrUse); 5770 5771 SemaRef.Diag(Mod->getExprLoc(), 5772 IsModMod ? diag::warn_unsequenced_mod_mod 5773 : diag::warn_unsequenced_mod_use) 5774 << O << SourceRange(ModOrUse->getExprLoc()); 5775 UI.Diagnosed = true; 5776 } 5777 5778 void notePreUse(Object O, Expr *Use) { 5779 UsageInfo &U = UsageMap[O]; 5780 // Uses conflict with other modifications. 5781 checkUsage(O, U, Use, UK_ModAsValue, false); 5782 } 5783 void notePostUse(Object O, Expr *Use) { 5784 UsageInfo &U = UsageMap[O]; 5785 checkUsage(O, U, Use, UK_ModAsSideEffect, false); 5786 addUsage(U, O, Use, UK_Use); 5787 } 5788 5789 void notePreMod(Object O, Expr *Mod) { 5790 UsageInfo &U = UsageMap[O]; 5791 // Modifications conflict with other modifications and with uses. 5792 checkUsage(O, U, Mod, UK_ModAsValue, true); 5793 checkUsage(O, U, Mod, UK_Use, false); 5794 } 5795 void notePostMod(Object O, Expr *Use, UsageKind UK) { 5796 UsageInfo &U = UsageMap[O]; 5797 checkUsage(O, U, Use, UK_ModAsSideEffect, true); 5798 addUsage(U, O, Use, UK); 5799 } 5800 5801public: 5802 SequenceChecker(Sema &S, Expr *E, SmallVectorImpl<Expr *> &WorkList) 5803 : Base(S.Context), SemaRef(S), Region(Tree.root()), ModAsSideEffect(0), 5804 WorkList(WorkList), EvalTracker(0) { 5805 Visit(E); 5806 } 5807 5808 void VisitStmt(Stmt *S) { 5809 // Skip all statements which aren't expressions for now. 5810 } 5811 5812 void VisitExpr(Expr *E) { 5813 // By default, just recurse to evaluated subexpressions. 5814 Base::VisitStmt(E); 5815 } 5816 5817 void VisitCastExpr(CastExpr *E) { 5818 Object O = Object(); 5819 if (E->getCastKind() == CK_LValueToRValue) 5820 O = getObject(E->getSubExpr(), false); 5821 5822 if (O) 5823 notePreUse(O, E); 5824 VisitExpr(E); 5825 if (O) 5826 notePostUse(O, E); 5827 } 5828 5829 void VisitBinComma(BinaryOperator *BO) { 5830 // C++11 [expr.comma]p1: 5831 // Every value computation and side effect associated with the left 5832 // expression is sequenced before every value computation and side 5833 // effect associated with the right expression. 5834 SequenceTree::Seq LHS = Tree.allocate(Region); 5835 SequenceTree::Seq RHS = Tree.allocate(Region); 5836 SequenceTree::Seq OldRegion = Region; 5837 5838 { 5839 SequencedSubexpression SeqLHS(*this); 5840 Region = LHS; 5841 Visit(BO->getLHS()); 5842 } 5843 5844 Region = RHS; 5845 Visit(BO->getRHS()); 5846 5847 Region = OldRegion; 5848 5849 // Forget that LHS and RHS are sequenced. They are both unsequenced 5850 // with respect to other stuff. 5851 Tree.merge(LHS); 5852 Tree.merge(RHS); 5853 } 5854 5855 void VisitBinAssign(BinaryOperator *BO) { 5856 // The modification is sequenced after the value computation of the LHS 5857 // and RHS, so check it before inspecting the operands and update the 5858 // map afterwards. 5859 Object O = getObject(BO->getLHS(), true); 5860 if (!O) 5861 return VisitExpr(BO); 5862 5863 notePreMod(O, BO); 5864 5865 // C++11 [expr.ass]p7: 5866 // E1 op= E2 is equivalent to E1 = E1 op E2, except that E1 is evaluated 5867 // only once. 5868 // 5869 // Therefore, for a compound assignment operator, O is considered used 5870 // everywhere except within the evaluation of E1 itself. 5871 if (isa<CompoundAssignOperator>(BO)) 5872 notePreUse(O, BO); 5873 5874 Visit(BO->getLHS()); 5875 5876 if (isa<CompoundAssignOperator>(BO)) 5877 notePostUse(O, BO); 5878 5879 Visit(BO->getRHS()); 5880 5881 // C++11 [expr.ass]p1: 5882 // the assignment is sequenced [...] before the value computation of the 5883 // assignment expression. 5884 // C11 6.5.16/3 has no such rule. 5885 notePostMod(O, BO, SemaRef.getLangOpts().CPlusPlus ? UK_ModAsValue 5886 : UK_ModAsSideEffect); 5887 } 5888 void VisitCompoundAssignOperator(CompoundAssignOperator *CAO) { 5889 VisitBinAssign(CAO); 5890 } 5891 5892 void VisitUnaryPreInc(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 5893 void VisitUnaryPreDec(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 5894 void VisitUnaryPreIncDec(UnaryOperator *UO) { 5895 Object O = getObject(UO->getSubExpr(), true); 5896 if (!O) 5897 return VisitExpr(UO); 5898 5899 notePreMod(O, UO); 5900 Visit(UO->getSubExpr()); 5901 // C++11 [expr.pre.incr]p1: 5902 // the expression ++x is equivalent to x+=1 5903 notePostMod(O, UO, SemaRef.getLangOpts().CPlusPlus ? UK_ModAsValue 5904 : UK_ModAsSideEffect); 5905 } 5906 5907 void VisitUnaryPostInc(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 5908 void VisitUnaryPostDec(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 5909 void VisitUnaryPostIncDec(UnaryOperator *UO) { 5910 Object O = getObject(UO->getSubExpr(), true); 5911 if (!O) 5912 return VisitExpr(UO); 5913 5914 notePreMod(O, UO); 5915 Visit(UO->getSubExpr()); 5916 notePostMod(O, UO, UK_ModAsSideEffect); 5917 } 5918 5919 /// Don't visit the RHS of '&&' or '||' if it might not be evaluated. 5920 void VisitBinLOr(BinaryOperator *BO) { 5921 // The side-effects of the LHS of an '&&' are sequenced before the 5922 // value computation of the RHS, and hence before the value computation 5923 // of the '&&' itself, unless the LHS evaluates to zero. We treat them 5924 // as if they were unconditionally sequenced. 5925 EvaluationTracker Eval(*this); 5926 { 5927 SequencedSubexpression Sequenced(*this); 5928 Visit(BO->getLHS()); 5929 } 5930 5931 bool Result; 5932 if (Eval.evaluate(BO->getLHS(), Result)) { 5933 if (!Result) 5934 Visit(BO->getRHS()); 5935 } else { 5936 // Check for unsequenced operations in the RHS, treating it as an 5937 // entirely separate evaluation. 5938 // 5939 // FIXME: If there are operations in the RHS which are unsequenced 5940 // with respect to operations outside the RHS, and those operations 5941 // are unconditionally evaluated, diagnose them. 5942 WorkList.push_back(BO->getRHS()); 5943 } 5944 } 5945 void VisitBinLAnd(BinaryOperator *BO) { 5946 EvaluationTracker Eval(*this); 5947 { 5948 SequencedSubexpression Sequenced(*this); 5949 Visit(BO->getLHS()); 5950 } 5951 5952 bool Result; 5953 if (Eval.evaluate(BO->getLHS(), Result)) { 5954 if (Result) 5955 Visit(BO->getRHS()); 5956 } else { 5957 WorkList.push_back(BO->getRHS()); 5958 } 5959 } 5960 5961 // Only visit the condition, unless we can be sure which subexpression will 5962 // be chosen. 5963 void VisitAbstractConditionalOperator(AbstractConditionalOperator *CO) { 5964 EvaluationTracker Eval(*this); 5965 { 5966 SequencedSubexpression Sequenced(*this); 5967 Visit(CO->getCond()); 5968 } 5969 5970 bool Result; 5971 if (Eval.evaluate(CO->getCond(), Result)) 5972 Visit(Result ? CO->getTrueExpr() : CO->getFalseExpr()); 5973 else { 5974 WorkList.push_back(CO->getTrueExpr()); 5975 WorkList.push_back(CO->getFalseExpr()); 5976 } 5977 } 5978 5979 void VisitCallExpr(CallExpr *CE) { 5980 // C++11 [intro.execution]p15: 5981 // When calling a function [...], every value computation and side effect 5982 // associated with any argument expression, or with the postfix expression 5983 // designating the called function, is sequenced before execution of every 5984 // expression or statement in the body of the function [and thus before 5985 // the value computation of its result]. 5986 SequencedSubexpression Sequenced(*this); 5987 Base::VisitCallExpr(CE); 5988 5989 // FIXME: CXXNewExpr and CXXDeleteExpr implicitly call functions. 5990 } 5991 5992 void VisitCXXConstructExpr(CXXConstructExpr *CCE) { 5993 // This is a call, so all subexpressions are sequenced before the result. 5994 SequencedSubexpression Sequenced(*this); 5995 5996 if (!CCE->isListInitialization()) 5997 return VisitExpr(CCE); 5998 5999 // In C++11, list initializations are sequenced. 6000 SmallVector<SequenceTree::Seq, 32> Elts; 6001 SequenceTree::Seq Parent = Region; 6002 for (CXXConstructExpr::arg_iterator I = CCE->arg_begin(), 6003 E = CCE->arg_end(); 6004 I != E; ++I) { 6005 Region = Tree.allocate(Parent); 6006 Elts.push_back(Region); 6007 Visit(*I); 6008 } 6009 6010 // Forget that the initializers are sequenced. 6011 Region = Parent; 6012 for (unsigned I = 0; I < Elts.size(); ++I) 6013 Tree.merge(Elts[I]); 6014 } 6015 6016 void VisitInitListExpr(InitListExpr *ILE) { 6017 if (!SemaRef.getLangOpts().CPlusPlus11) 6018 return VisitExpr(ILE); 6019 6020 // In C++11, list initializations are sequenced. 6021 SmallVector<SequenceTree::Seq, 32> Elts; 6022 SequenceTree::Seq Parent = Region; 6023 for (unsigned I = 0; I < ILE->getNumInits(); ++I) { 6024 Expr *E = ILE->getInit(I); 6025 if (!E) continue; 6026 Region = Tree.allocate(Parent); 6027 Elts.push_back(Region); 6028 Visit(E); 6029 } 6030 6031 // Forget that the initializers are sequenced. 6032 Region = Parent; 6033 for (unsigned I = 0; I < Elts.size(); ++I) 6034 Tree.merge(Elts[I]); 6035 } 6036}; 6037} 6038 6039void Sema::CheckUnsequencedOperations(Expr *E) { 6040 SmallVector<Expr *, 8> WorkList; 6041 WorkList.push_back(E); 6042 while (!WorkList.empty()) { 6043 Expr *Item = WorkList.pop_back_val(); 6044 SequenceChecker(*this, Item, WorkList); 6045 } 6046} 6047 6048void Sema::CheckCompletedExpr(Expr *E, SourceLocation CheckLoc, 6049 bool IsConstexpr) { 6050 CheckImplicitConversions(E, CheckLoc); 6051 CheckUnsequencedOperations(E); 6052 if (!IsConstexpr && !E->isValueDependent()) 6053 CheckForIntOverflow(E); 6054} 6055 6056void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 6057 FieldDecl *BitField, 6058 Expr *Init) { 6059 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 6060} 6061 6062/// CheckParmsForFunctionDef - Check that the parameters of the given 6063/// function are appropriate for the definition of a function. This 6064/// takes care of any checks that cannot be performed on the 6065/// declaration itself, e.g., that the types of each of the function 6066/// parameters are complete. 6067bool Sema::CheckParmsForFunctionDef(ParmVarDecl *const *P, 6068 ParmVarDecl *const *PEnd, 6069 bool CheckParameterNames) { 6070 bool HasInvalidParm = false; 6071 for (; P != PEnd; ++P) { 6072 ParmVarDecl *Param = *P; 6073 6074 // C99 6.7.5.3p4: the parameters in a parameter type list in a 6075 // function declarator that is part of a function definition of 6076 // that function shall not have incomplete type. 6077 // 6078 // This is also C++ [dcl.fct]p6. 6079 if (!Param->isInvalidDecl() && 6080 RequireCompleteType(Param->getLocation(), Param->getType(), 6081 diag::err_typecheck_decl_incomplete_type)) { 6082 Param->setInvalidDecl(); 6083 HasInvalidParm = true; 6084 } 6085 6086 // C99 6.9.1p5: If the declarator includes a parameter type list, the 6087 // declaration of each parameter shall include an identifier. 6088 if (CheckParameterNames && 6089 Param->getIdentifier() == 0 && 6090 !Param->isImplicit() && 6091 !getLangOpts().CPlusPlus) 6092 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 6093 6094 // C99 6.7.5.3p12: 6095 // If the function declarator is not part of a definition of that 6096 // function, parameters may have incomplete type and may use the [*] 6097 // notation in their sequences of declarator specifiers to specify 6098 // variable length array types. 6099 QualType PType = Param->getOriginalType(); 6100 while (const ArrayType *AT = Context.getAsArrayType(PType)) { 6101 if (AT->getSizeModifier() == ArrayType::Star) { 6102 // FIXME: This diagnostic should point the '[*]' if source-location 6103 // information is added for it. 6104 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 6105 break; 6106 } 6107 PType= AT->getElementType(); 6108 } 6109 6110 // MSVC destroys objects passed by value in the callee. Therefore a 6111 // function definition which takes such a parameter must be able to call the 6112 // object's destructor. 6113 if (getLangOpts().CPlusPlus && 6114 Context.getTargetInfo().getCXXABI().isArgumentDestroyedByCallee()) { 6115 if (const RecordType *RT = Param->getType()->getAs<RecordType>()) 6116 FinalizeVarWithDestructor(Param, RT); 6117 } 6118 } 6119 6120 return HasInvalidParm; 6121} 6122 6123/// CheckCastAlign - Implements -Wcast-align, which warns when a 6124/// pointer cast increases the alignment requirements. 6125void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 6126 // This is actually a lot of work to potentially be doing on every 6127 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 6128 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 6129 TRange.getBegin()) 6130 == DiagnosticsEngine::Ignored) 6131 return; 6132 6133 // Ignore dependent types. 6134 if (T->isDependentType() || Op->getType()->isDependentType()) 6135 return; 6136 6137 // Require that the destination be a pointer type. 6138 const PointerType *DestPtr = T->getAs<PointerType>(); 6139 if (!DestPtr) return; 6140 6141 // If the destination has alignment 1, we're done. 6142 QualType DestPointee = DestPtr->getPointeeType(); 6143 if (DestPointee->isIncompleteType()) return; 6144 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 6145 if (DestAlign.isOne()) return; 6146 6147 // Require that the source be a pointer type. 6148 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 6149 if (!SrcPtr) return; 6150 QualType SrcPointee = SrcPtr->getPointeeType(); 6151 6152 // Whitelist casts from cv void*. We already implicitly 6153 // whitelisted casts to cv void*, since they have alignment 1. 6154 // Also whitelist casts involving incomplete types, which implicitly 6155 // includes 'void'. 6156 if (SrcPointee->isIncompleteType()) return; 6157 6158 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 6159 if (SrcAlign >= DestAlign) return; 6160 6161 Diag(TRange.getBegin(), diag::warn_cast_align) 6162 << Op->getType() << T 6163 << static_cast<unsigned>(SrcAlign.getQuantity()) 6164 << static_cast<unsigned>(DestAlign.getQuantity()) 6165 << TRange << Op->getSourceRange(); 6166} 6167 6168static const Type* getElementType(const Expr *BaseExpr) { 6169 const Type* EltType = BaseExpr->getType().getTypePtr(); 6170 if (EltType->isAnyPointerType()) 6171 return EltType->getPointeeType().getTypePtr(); 6172 else if (EltType->isArrayType()) 6173 return EltType->getBaseElementTypeUnsafe(); 6174 return EltType; 6175} 6176 6177/// \brief Check whether this array fits the idiom of a size-one tail padded 6178/// array member of a struct. 6179/// 6180/// We avoid emitting out-of-bounds access warnings for such arrays as they are 6181/// commonly used to emulate flexible arrays in C89 code. 6182static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 6183 const NamedDecl *ND) { 6184 if (Size != 1 || !ND) return false; 6185 6186 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 6187 if (!FD) return false; 6188 6189 // Don't consider sizes resulting from macro expansions or template argument 6190 // substitution to form C89 tail-padded arrays. 6191 6192 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 6193 while (TInfo) { 6194 TypeLoc TL = TInfo->getTypeLoc(); 6195 // Look through typedefs. 6196 if (TypedefTypeLoc TTL = TL.getAs<TypedefTypeLoc>()) { 6197 const TypedefNameDecl *TDL = TTL.getTypedefNameDecl(); 6198 TInfo = TDL->getTypeSourceInfo(); 6199 continue; 6200 } 6201 if (ConstantArrayTypeLoc CTL = TL.getAs<ConstantArrayTypeLoc>()) { 6202 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 6203 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 6204 return false; 6205 } 6206 break; 6207 } 6208 6209 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 6210 if (!RD) return false; 6211 if (RD->isUnion()) return false; 6212 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 6213 if (!CRD->isStandardLayout()) return false; 6214 } 6215 6216 // See if this is the last field decl in the record. 6217 const Decl *D = FD; 6218 while ((D = D->getNextDeclInContext())) 6219 if (isa<FieldDecl>(D)) 6220 return false; 6221 return true; 6222} 6223 6224void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 6225 const ArraySubscriptExpr *ASE, 6226 bool AllowOnePastEnd, bool IndexNegated) { 6227 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 6228 if (IndexExpr->isValueDependent()) 6229 return; 6230 6231 const Type *EffectiveType = getElementType(BaseExpr); 6232 BaseExpr = BaseExpr->IgnoreParenCasts(); 6233 const ConstantArrayType *ArrayTy = 6234 Context.getAsConstantArrayType(BaseExpr->getType()); 6235 if (!ArrayTy) 6236 return; 6237 6238 llvm::APSInt index; 6239 if (!IndexExpr->EvaluateAsInt(index, Context)) 6240 return; 6241 if (IndexNegated) 6242 index = -index; 6243 6244 const NamedDecl *ND = NULL; 6245 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 6246 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 6247 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 6248 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 6249 6250 if (index.isUnsigned() || !index.isNegative()) { 6251 llvm::APInt size = ArrayTy->getSize(); 6252 if (!size.isStrictlyPositive()) 6253 return; 6254 6255 const Type* BaseType = getElementType(BaseExpr); 6256 if (BaseType != EffectiveType) { 6257 // Make sure we're comparing apples to apples when comparing index to size 6258 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 6259 uint64_t array_typesize = Context.getTypeSize(BaseType); 6260 // Handle ptrarith_typesize being zero, such as when casting to void* 6261 if (!ptrarith_typesize) ptrarith_typesize = 1; 6262 if (ptrarith_typesize != array_typesize) { 6263 // There's a cast to a different size type involved 6264 uint64_t ratio = array_typesize / ptrarith_typesize; 6265 // TODO: Be smarter about handling cases where array_typesize is not a 6266 // multiple of ptrarith_typesize 6267 if (ptrarith_typesize * ratio == array_typesize) 6268 size *= llvm::APInt(size.getBitWidth(), ratio); 6269 } 6270 } 6271 6272 if (size.getBitWidth() > index.getBitWidth()) 6273 index = index.zext(size.getBitWidth()); 6274 else if (size.getBitWidth() < index.getBitWidth()) 6275 size = size.zext(index.getBitWidth()); 6276 6277 // For array subscripting the index must be less than size, but for pointer 6278 // arithmetic also allow the index (offset) to be equal to size since 6279 // computing the next address after the end of the array is legal and 6280 // commonly done e.g. in C++ iterators and range-based for loops. 6281 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 6282 return; 6283 6284 // Also don't warn for arrays of size 1 which are members of some 6285 // structure. These are often used to approximate flexible arrays in C89 6286 // code. 6287 if (IsTailPaddedMemberArray(*this, size, ND)) 6288 return; 6289 6290 // Suppress the warning if the subscript expression (as identified by the 6291 // ']' location) and the index expression are both from macro expansions 6292 // within a system header. 6293 if (ASE) { 6294 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 6295 ASE->getRBracketLoc()); 6296 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 6297 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 6298 IndexExpr->getLocStart()); 6299 if (SourceMgr.isWrittenInSameFile(RBracketLoc, IndexLoc)) 6300 return; 6301 } 6302 } 6303 6304 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 6305 if (ASE) 6306 DiagID = diag::warn_array_index_exceeds_bounds; 6307 6308 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 6309 PDiag(DiagID) << index.toString(10, true) 6310 << size.toString(10, true) 6311 << (unsigned)size.getLimitedValue(~0U) 6312 << IndexExpr->getSourceRange()); 6313 } else { 6314 unsigned DiagID = diag::warn_array_index_precedes_bounds; 6315 if (!ASE) { 6316 DiagID = diag::warn_ptr_arith_precedes_bounds; 6317 if (index.isNegative()) index = -index; 6318 } 6319 6320 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 6321 PDiag(DiagID) << index.toString(10, true) 6322 << IndexExpr->getSourceRange()); 6323 } 6324 6325 if (!ND) { 6326 // Try harder to find a NamedDecl to point at in the note. 6327 while (const ArraySubscriptExpr *ASE = 6328 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 6329 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 6330 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 6331 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 6332 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 6333 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 6334 } 6335 6336 if (ND) 6337 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 6338 PDiag(diag::note_array_index_out_of_bounds) 6339 << ND->getDeclName()); 6340} 6341 6342void Sema::CheckArrayAccess(const Expr *expr) { 6343 int AllowOnePastEnd = 0; 6344 while (expr) { 6345 expr = expr->IgnoreParenImpCasts(); 6346 switch (expr->getStmtClass()) { 6347 case Stmt::ArraySubscriptExprClass: { 6348 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 6349 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 6350 AllowOnePastEnd > 0); 6351 return; 6352 } 6353 case Stmt::UnaryOperatorClass: { 6354 // Only unwrap the * and & unary operators 6355 const UnaryOperator *UO = cast<UnaryOperator>(expr); 6356 expr = UO->getSubExpr(); 6357 switch (UO->getOpcode()) { 6358 case UO_AddrOf: 6359 AllowOnePastEnd++; 6360 break; 6361 case UO_Deref: 6362 AllowOnePastEnd--; 6363 break; 6364 default: 6365 return; 6366 } 6367 break; 6368 } 6369 case Stmt::ConditionalOperatorClass: { 6370 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 6371 if (const Expr *lhs = cond->getLHS()) 6372 CheckArrayAccess(lhs); 6373 if (const Expr *rhs = cond->getRHS()) 6374 CheckArrayAccess(rhs); 6375 return; 6376 } 6377 default: 6378 return; 6379 } 6380 } 6381} 6382 6383//===--- CHECK: Objective-C retain cycles ----------------------------------// 6384 6385namespace { 6386 struct RetainCycleOwner { 6387 RetainCycleOwner() : Variable(0), Indirect(false) {} 6388 VarDecl *Variable; 6389 SourceRange Range; 6390 SourceLocation Loc; 6391 bool Indirect; 6392 6393 void setLocsFrom(Expr *e) { 6394 Loc = e->getExprLoc(); 6395 Range = e->getSourceRange(); 6396 } 6397 }; 6398} 6399 6400/// Consider whether capturing the given variable can possibly lead to 6401/// a retain cycle. 6402static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 6403 // In ARC, it's captured strongly iff the variable has __strong 6404 // lifetime. In MRR, it's captured strongly if the variable is 6405 // __block and has an appropriate type. 6406 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 6407 return false; 6408 6409 owner.Variable = var; 6410 if (ref) 6411 owner.setLocsFrom(ref); 6412 return true; 6413} 6414 6415static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 6416 while (true) { 6417 e = e->IgnoreParens(); 6418 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 6419 switch (cast->getCastKind()) { 6420 case CK_BitCast: 6421 case CK_LValueBitCast: 6422 case CK_LValueToRValue: 6423 case CK_ARCReclaimReturnedObject: 6424 e = cast->getSubExpr(); 6425 continue; 6426 6427 default: 6428 return false; 6429 } 6430 } 6431 6432 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 6433 ObjCIvarDecl *ivar = ref->getDecl(); 6434 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 6435 return false; 6436 6437 // Try to find a retain cycle in the base. 6438 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 6439 return false; 6440 6441 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 6442 owner.Indirect = true; 6443 return true; 6444 } 6445 6446 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 6447 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 6448 if (!var) return false; 6449 return considerVariable(var, ref, owner); 6450 } 6451 6452 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 6453 if (member->isArrow()) return false; 6454 6455 // Don't count this as an indirect ownership. 6456 e = member->getBase(); 6457 continue; 6458 } 6459 6460 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 6461 // Only pay attention to pseudo-objects on property references. 6462 ObjCPropertyRefExpr *pre 6463 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 6464 ->IgnoreParens()); 6465 if (!pre) return false; 6466 if (pre->isImplicitProperty()) return false; 6467 ObjCPropertyDecl *property = pre->getExplicitProperty(); 6468 if (!property->isRetaining() && 6469 !(property->getPropertyIvarDecl() && 6470 property->getPropertyIvarDecl()->getType() 6471 .getObjCLifetime() == Qualifiers::OCL_Strong)) 6472 return false; 6473 6474 owner.Indirect = true; 6475 if (pre->isSuperReceiver()) { 6476 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 6477 if (!owner.Variable) 6478 return false; 6479 owner.Loc = pre->getLocation(); 6480 owner.Range = pre->getSourceRange(); 6481 return true; 6482 } 6483 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 6484 ->getSourceExpr()); 6485 continue; 6486 } 6487 6488 // Array ivars? 6489 6490 return false; 6491 } 6492} 6493 6494namespace { 6495 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 6496 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 6497 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 6498 Variable(variable), Capturer(0) {} 6499 6500 VarDecl *Variable; 6501 Expr *Capturer; 6502 6503 void VisitDeclRefExpr(DeclRefExpr *ref) { 6504 if (ref->getDecl() == Variable && !Capturer) 6505 Capturer = ref; 6506 } 6507 6508 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 6509 if (Capturer) return; 6510 Visit(ref->getBase()); 6511 if (Capturer && ref->isFreeIvar()) 6512 Capturer = ref; 6513 } 6514 6515 void VisitBlockExpr(BlockExpr *block) { 6516 // Look inside nested blocks 6517 if (block->getBlockDecl()->capturesVariable(Variable)) 6518 Visit(block->getBlockDecl()->getBody()); 6519 } 6520 6521 void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) { 6522 if (Capturer) return; 6523 if (OVE->getSourceExpr()) 6524 Visit(OVE->getSourceExpr()); 6525 } 6526 }; 6527} 6528 6529/// Check whether the given argument is a block which captures a 6530/// variable. 6531static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 6532 assert(owner.Variable && owner.Loc.isValid()); 6533 6534 e = e->IgnoreParenCasts(); 6535 6536 // Look through [^{...} copy] and Block_copy(^{...}). 6537 if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) { 6538 Selector Cmd = ME->getSelector(); 6539 if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") { 6540 e = ME->getInstanceReceiver(); 6541 if (!e) 6542 return 0; 6543 e = e->IgnoreParenCasts(); 6544 } 6545 } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) { 6546 if (CE->getNumArgs() == 1) { 6547 FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl()); 6548 if (Fn) { 6549 const IdentifierInfo *FnI = Fn->getIdentifier(); 6550 if (FnI && FnI->isStr("_Block_copy")) { 6551 e = CE->getArg(0)->IgnoreParenCasts(); 6552 } 6553 } 6554 } 6555 } 6556 6557 BlockExpr *block = dyn_cast<BlockExpr>(e); 6558 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 6559 return 0; 6560 6561 FindCaptureVisitor visitor(S.Context, owner.Variable); 6562 visitor.Visit(block->getBlockDecl()->getBody()); 6563 return visitor.Capturer; 6564} 6565 6566static void diagnoseRetainCycle(Sema &S, Expr *capturer, 6567 RetainCycleOwner &owner) { 6568 assert(capturer); 6569 assert(owner.Variable && owner.Loc.isValid()); 6570 6571 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 6572 << owner.Variable << capturer->getSourceRange(); 6573 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 6574 << owner.Indirect << owner.Range; 6575} 6576 6577/// Check for a keyword selector that starts with the word 'add' or 6578/// 'set'. 6579static bool isSetterLikeSelector(Selector sel) { 6580 if (sel.isUnarySelector()) return false; 6581 6582 StringRef str = sel.getNameForSlot(0); 6583 while (!str.empty() && str.front() == '_') str = str.substr(1); 6584 if (str.startswith("set")) 6585 str = str.substr(3); 6586 else if (str.startswith("add")) { 6587 // Specially whitelist 'addOperationWithBlock:'. 6588 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 6589 return false; 6590 str = str.substr(3); 6591 } 6592 else 6593 return false; 6594 6595 if (str.empty()) return true; 6596 return !isLowercase(str.front()); 6597} 6598 6599/// Check a message send to see if it's likely to cause a retain cycle. 6600void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 6601 // Only check instance methods whose selector looks like a setter. 6602 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 6603 return; 6604 6605 // Try to find a variable that the receiver is strongly owned by. 6606 RetainCycleOwner owner; 6607 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 6608 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 6609 return; 6610 } else { 6611 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 6612 owner.Variable = getCurMethodDecl()->getSelfDecl(); 6613 owner.Loc = msg->getSuperLoc(); 6614 owner.Range = msg->getSuperLoc(); 6615 } 6616 6617 // Check whether the receiver is captured by any of the arguments. 6618 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 6619 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 6620 return diagnoseRetainCycle(*this, capturer, owner); 6621} 6622 6623/// Check a property assign to see if it's likely to cause a retain cycle. 6624void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 6625 RetainCycleOwner owner; 6626 if (!findRetainCycleOwner(*this, receiver, owner)) 6627 return; 6628 6629 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 6630 diagnoseRetainCycle(*this, capturer, owner); 6631} 6632 6633void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) { 6634 RetainCycleOwner Owner; 6635 if (!considerVariable(Var, /*DeclRefExpr=*/0, Owner)) 6636 return; 6637 6638 // Because we don't have an expression for the variable, we have to set the 6639 // location explicitly here. 6640 Owner.Loc = Var->getLocation(); 6641 Owner.Range = Var->getSourceRange(); 6642 6643 if (Expr *Capturer = findCapturingExpr(*this, Init, Owner)) 6644 diagnoseRetainCycle(*this, Capturer, Owner); 6645} 6646 6647static bool checkUnsafeAssignLiteral(Sema &S, SourceLocation Loc, 6648 Expr *RHS, bool isProperty) { 6649 // Check if RHS is an Objective-C object literal, which also can get 6650 // immediately zapped in a weak reference. Note that we explicitly 6651 // allow ObjCStringLiterals, since those are designed to never really die. 6652 RHS = RHS->IgnoreParenImpCasts(); 6653 6654 // This enum needs to match with the 'select' in 6655 // warn_objc_arc_literal_assign (off-by-1). 6656 Sema::ObjCLiteralKind Kind = S.CheckLiteralKind(RHS); 6657 if (Kind == Sema::LK_String || Kind == Sema::LK_None) 6658 return false; 6659 6660 S.Diag(Loc, diag::warn_arc_literal_assign) 6661 << (unsigned) Kind 6662 << (isProperty ? 0 : 1) 6663 << RHS->getSourceRange(); 6664 6665 return true; 6666} 6667 6668static bool checkUnsafeAssignObject(Sema &S, SourceLocation Loc, 6669 Qualifiers::ObjCLifetime LT, 6670 Expr *RHS, bool isProperty) { 6671 // Strip off any implicit cast added to get to the one ARC-specific. 6672 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6673 if (cast->getCastKind() == CK_ARCConsumeObject) { 6674 S.Diag(Loc, diag::warn_arc_retained_assign) 6675 << (LT == Qualifiers::OCL_ExplicitNone) 6676 << (isProperty ? 0 : 1) 6677 << RHS->getSourceRange(); 6678 return true; 6679 } 6680 RHS = cast->getSubExpr(); 6681 } 6682 6683 if (LT == Qualifiers::OCL_Weak && 6684 checkUnsafeAssignLiteral(S, Loc, RHS, isProperty)) 6685 return true; 6686 6687 return false; 6688} 6689 6690bool Sema::checkUnsafeAssigns(SourceLocation Loc, 6691 QualType LHS, Expr *RHS) { 6692 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 6693 6694 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 6695 return false; 6696 6697 if (checkUnsafeAssignObject(*this, Loc, LT, RHS, false)) 6698 return true; 6699 6700 return false; 6701} 6702 6703void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 6704 Expr *LHS, Expr *RHS) { 6705 QualType LHSType; 6706 // PropertyRef on LHS type need be directly obtained from 6707 // its declaration as it has a PsuedoType. 6708 ObjCPropertyRefExpr *PRE 6709 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 6710 if (PRE && !PRE->isImplicitProperty()) { 6711 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6712 if (PD) 6713 LHSType = PD->getType(); 6714 } 6715 6716 if (LHSType.isNull()) 6717 LHSType = LHS->getType(); 6718 6719 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 6720 6721 if (LT == Qualifiers::OCL_Weak) { 6722 DiagnosticsEngine::Level Level = 6723 Diags.getDiagnosticLevel(diag::warn_arc_repeated_use_of_weak, Loc); 6724 if (Level != DiagnosticsEngine::Ignored) 6725 getCurFunction()->markSafeWeakUse(LHS); 6726 } 6727 6728 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 6729 return; 6730 6731 // FIXME. Check for other life times. 6732 if (LT != Qualifiers::OCL_None) 6733 return; 6734 6735 if (PRE) { 6736 if (PRE->isImplicitProperty()) 6737 return; 6738 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6739 if (!PD) 6740 return; 6741 6742 unsigned Attributes = PD->getPropertyAttributes(); 6743 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 6744 // when 'assign' attribute was not explicitly specified 6745 // by user, ignore it and rely on property type itself 6746 // for lifetime info. 6747 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 6748 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 6749 LHSType->isObjCRetainableType()) 6750 return; 6751 6752 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6753 if (cast->getCastKind() == CK_ARCConsumeObject) { 6754 Diag(Loc, diag::warn_arc_retained_property_assign) 6755 << RHS->getSourceRange(); 6756 return; 6757 } 6758 RHS = cast->getSubExpr(); 6759 } 6760 } 6761 else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) { 6762 if (checkUnsafeAssignObject(*this, Loc, Qualifiers::OCL_Weak, RHS, true)) 6763 return; 6764 } 6765 } 6766} 6767 6768//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 6769 6770namespace { 6771bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 6772 SourceLocation StmtLoc, 6773 const NullStmt *Body) { 6774 // Do not warn if the body is a macro that expands to nothing, e.g: 6775 // 6776 // #define CALL(x) 6777 // if (condition) 6778 // CALL(0); 6779 // 6780 if (Body->hasLeadingEmptyMacro()) 6781 return false; 6782 6783 // Get line numbers of statement and body. 6784 bool StmtLineInvalid; 6785 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 6786 &StmtLineInvalid); 6787 if (StmtLineInvalid) 6788 return false; 6789 6790 bool BodyLineInvalid; 6791 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 6792 &BodyLineInvalid); 6793 if (BodyLineInvalid) 6794 return false; 6795 6796 // Warn if null statement and body are on the same line. 6797 if (StmtLine != BodyLine) 6798 return false; 6799 6800 return true; 6801} 6802} // Unnamed namespace 6803 6804void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 6805 const Stmt *Body, 6806 unsigned DiagID) { 6807 // Since this is a syntactic check, don't emit diagnostic for template 6808 // instantiations, this just adds noise. 6809 if (CurrentInstantiationScope) 6810 return; 6811 6812 // The body should be a null statement. 6813 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6814 if (!NBody) 6815 return; 6816 6817 // Do the usual checks. 6818 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6819 return; 6820 6821 Diag(NBody->getSemiLoc(), DiagID); 6822 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6823} 6824 6825void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 6826 const Stmt *PossibleBody) { 6827 assert(!CurrentInstantiationScope); // Ensured by caller 6828 6829 SourceLocation StmtLoc; 6830 const Stmt *Body; 6831 unsigned DiagID; 6832 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 6833 StmtLoc = FS->getRParenLoc(); 6834 Body = FS->getBody(); 6835 DiagID = diag::warn_empty_for_body; 6836 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 6837 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 6838 Body = WS->getBody(); 6839 DiagID = diag::warn_empty_while_body; 6840 } else 6841 return; // Neither `for' nor `while'. 6842 6843 // The body should be a null statement. 6844 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6845 if (!NBody) 6846 return; 6847 6848 // Skip expensive checks if diagnostic is disabled. 6849 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 6850 DiagnosticsEngine::Ignored) 6851 return; 6852 6853 // Do the usual checks. 6854 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6855 return; 6856 6857 // `for(...);' and `while(...);' are popular idioms, so in order to keep 6858 // noise level low, emit diagnostics only if for/while is followed by a 6859 // CompoundStmt, e.g.: 6860 // for (int i = 0; i < n; i++); 6861 // { 6862 // a(i); 6863 // } 6864 // or if for/while is followed by a statement with more indentation 6865 // than for/while itself: 6866 // for (int i = 0; i < n; i++); 6867 // a(i); 6868 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 6869 if (!ProbableTypo) { 6870 bool BodyColInvalid; 6871 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 6872 PossibleBody->getLocStart(), 6873 &BodyColInvalid); 6874 if (BodyColInvalid) 6875 return; 6876 6877 bool StmtColInvalid; 6878 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 6879 S->getLocStart(), 6880 &StmtColInvalid); 6881 if (StmtColInvalid) 6882 return; 6883 6884 if (BodyCol > StmtCol) 6885 ProbableTypo = true; 6886 } 6887 6888 if (ProbableTypo) { 6889 Diag(NBody->getSemiLoc(), DiagID); 6890 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6891 } 6892} 6893 6894//===--- Layout compatibility ----------------------------------------------// 6895 6896namespace { 6897 6898bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2); 6899 6900/// \brief Check if two enumeration types are layout-compatible. 6901bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) { 6902 // C++11 [dcl.enum] p8: 6903 // Two enumeration types are layout-compatible if they have the same 6904 // underlying type. 6905 return ED1->isComplete() && ED2->isComplete() && 6906 C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType()); 6907} 6908 6909/// \brief Check if two fields are layout-compatible. 6910bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) { 6911 if (!isLayoutCompatible(C, Field1->getType(), Field2->getType())) 6912 return false; 6913 6914 if (Field1->isBitField() != Field2->isBitField()) 6915 return false; 6916 6917 if (Field1->isBitField()) { 6918 // Make sure that the bit-fields are the same length. 6919 unsigned Bits1 = Field1->getBitWidthValue(C); 6920 unsigned Bits2 = Field2->getBitWidthValue(C); 6921 6922 if (Bits1 != Bits2) 6923 return false; 6924 } 6925 6926 return true; 6927} 6928 6929/// \brief Check if two standard-layout structs are layout-compatible. 6930/// (C++11 [class.mem] p17) 6931bool isLayoutCompatibleStruct(ASTContext &C, 6932 RecordDecl *RD1, 6933 RecordDecl *RD2) { 6934 // If both records are C++ classes, check that base classes match. 6935 if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) { 6936 // If one of records is a CXXRecordDecl we are in C++ mode, 6937 // thus the other one is a CXXRecordDecl, too. 6938 const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2); 6939 // Check number of base classes. 6940 if (D1CXX->getNumBases() != D2CXX->getNumBases()) 6941 return false; 6942 6943 // Check the base classes. 6944 for (CXXRecordDecl::base_class_const_iterator 6945 Base1 = D1CXX->bases_begin(), 6946 BaseEnd1 = D1CXX->bases_end(), 6947 Base2 = D2CXX->bases_begin(); 6948 Base1 != BaseEnd1; 6949 ++Base1, ++Base2) { 6950 if (!isLayoutCompatible(C, Base1->getType(), Base2->getType())) 6951 return false; 6952 } 6953 } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) { 6954 // If only RD2 is a C++ class, it should have zero base classes. 6955 if (D2CXX->getNumBases() > 0) 6956 return false; 6957 } 6958 6959 // Check the fields. 6960 RecordDecl::field_iterator Field2 = RD2->field_begin(), 6961 Field2End = RD2->field_end(), 6962 Field1 = RD1->field_begin(), 6963 Field1End = RD1->field_end(); 6964 for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) { 6965 if (!isLayoutCompatible(C, *Field1, *Field2)) 6966 return false; 6967 } 6968 if (Field1 != Field1End || Field2 != Field2End) 6969 return false; 6970 6971 return true; 6972} 6973 6974/// \brief Check if two standard-layout unions are layout-compatible. 6975/// (C++11 [class.mem] p18) 6976bool isLayoutCompatibleUnion(ASTContext &C, 6977 RecordDecl *RD1, 6978 RecordDecl *RD2) { 6979 llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields; 6980 for (RecordDecl::field_iterator Field2 = RD2->field_begin(), 6981 Field2End = RD2->field_end(); 6982 Field2 != Field2End; ++Field2) { 6983 UnmatchedFields.insert(*Field2); 6984 } 6985 6986 for (RecordDecl::field_iterator Field1 = RD1->field_begin(), 6987 Field1End = RD1->field_end(); 6988 Field1 != Field1End; ++Field1) { 6989 llvm::SmallPtrSet<FieldDecl *, 8>::iterator 6990 I = UnmatchedFields.begin(), 6991 E = UnmatchedFields.end(); 6992 6993 for ( ; I != E; ++I) { 6994 if (isLayoutCompatible(C, *Field1, *I)) { 6995 bool Result = UnmatchedFields.erase(*I); 6996 (void) Result; 6997 assert(Result); 6998 break; 6999 } 7000 } 7001 if (I == E) 7002 return false; 7003 } 7004 7005 return UnmatchedFields.empty(); 7006} 7007 7008bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) { 7009 if (RD1->isUnion() != RD2->isUnion()) 7010 return false; 7011 7012 if (RD1->isUnion()) 7013 return isLayoutCompatibleUnion(C, RD1, RD2); 7014 else 7015 return isLayoutCompatibleStruct(C, RD1, RD2); 7016} 7017 7018/// \brief Check if two types are layout-compatible in C++11 sense. 7019bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) { 7020 if (T1.isNull() || T2.isNull()) 7021 return false; 7022 7023 // C++11 [basic.types] p11: 7024 // If two types T1 and T2 are the same type, then T1 and T2 are 7025 // layout-compatible types. 7026 if (C.hasSameType(T1, T2)) 7027 return true; 7028 7029 T1 = T1.getCanonicalType().getUnqualifiedType(); 7030 T2 = T2.getCanonicalType().getUnqualifiedType(); 7031 7032 const Type::TypeClass TC1 = T1->getTypeClass(); 7033 const Type::TypeClass TC2 = T2->getTypeClass(); 7034 7035 if (TC1 != TC2) 7036 return false; 7037 7038 if (TC1 == Type::Enum) { 7039 return isLayoutCompatible(C, 7040 cast<EnumType>(T1)->getDecl(), 7041 cast<EnumType>(T2)->getDecl()); 7042 } else if (TC1 == Type::Record) { 7043 if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType()) 7044 return false; 7045 7046 return isLayoutCompatible(C, 7047 cast<RecordType>(T1)->getDecl(), 7048 cast<RecordType>(T2)->getDecl()); 7049 } 7050 7051 return false; 7052} 7053} 7054 7055//===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----// 7056 7057namespace { 7058/// \brief Given a type tag expression find the type tag itself. 7059/// 7060/// \param TypeExpr Type tag expression, as it appears in user's code. 7061/// 7062/// \param VD Declaration of an identifier that appears in a type tag. 7063/// 7064/// \param MagicValue Type tag magic value. 7065bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx, 7066 const ValueDecl **VD, uint64_t *MagicValue) { 7067 while(true) { 7068 if (!TypeExpr) 7069 return false; 7070 7071 TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts(); 7072 7073 switch (TypeExpr->getStmtClass()) { 7074 case Stmt::UnaryOperatorClass: { 7075 const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr); 7076 if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) { 7077 TypeExpr = UO->getSubExpr(); 7078 continue; 7079 } 7080 return false; 7081 } 7082 7083 case Stmt::DeclRefExprClass: { 7084 const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr); 7085 *VD = DRE->getDecl(); 7086 return true; 7087 } 7088 7089 case Stmt::IntegerLiteralClass: { 7090 const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr); 7091 llvm::APInt MagicValueAPInt = IL->getValue(); 7092 if (MagicValueAPInt.getActiveBits() <= 64) { 7093 *MagicValue = MagicValueAPInt.getZExtValue(); 7094 return true; 7095 } else 7096 return false; 7097 } 7098 7099 case Stmt::BinaryConditionalOperatorClass: 7100 case Stmt::ConditionalOperatorClass: { 7101 const AbstractConditionalOperator *ACO = 7102 cast<AbstractConditionalOperator>(TypeExpr); 7103 bool Result; 7104 if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) { 7105 if (Result) 7106 TypeExpr = ACO->getTrueExpr(); 7107 else 7108 TypeExpr = ACO->getFalseExpr(); 7109 continue; 7110 } 7111 return false; 7112 } 7113 7114 case Stmt::BinaryOperatorClass: { 7115 const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr); 7116 if (BO->getOpcode() == BO_Comma) { 7117 TypeExpr = BO->getRHS(); 7118 continue; 7119 } 7120 return false; 7121 } 7122 7123 default: 7124 return false; 7125 } 7126 } 7127} 7128 7129/// \brief Retrieve the C type corresponding to type tag TypeExpr. 7130/// 7131/// \param TypeExpr Expression that specifies a type tag. 7132/// 7133/// \param MagicValues Registered magic values. 7134/// 7135/// \param FoundWrongKind Set to true if a type tag was found, but of a wrong 7136/// kind. 7137/// 7138/// \param TypeInfo Information about the corresponding C type. 7139/// 7140/// \returns true if the corresponding C type was found. 7141bool GetMatchingCType( 7142 const IdentifierInfo *ArgumentKind, 7143 const Expr *TypeExpr, const ASTContext &Ctx, 7144 const llvm::DenseMap<Sema::TypeTagMagicValue, 7145 Sema::TypeTagData> *MagicValues, 7146 bool &FoundWrongKind, 7147 Sema::TypeTagData &TypeInfo) { 7148 FoundWrongKind = false; 7149 7150 // Variable declaration that has type_tag_for_datatype attribute. 7151 const ValueDecl *VD = NULL; 7152 7153 uint64_t MagicValue; 7154 7155 if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue)) 7156 return false; 7157 7158 if (VD) { 7159 for (specific_attr_iterator<TypeTagForDatatypeAttr> 7160 I = VD->specific_attr_begin<TypeTagForDatatypeAttr>(), 7161 E = VD->specific_attr_end<TypeTagForDatatypeAttr>(); 7162 I != E; ++I) { 7163 if (I->getArgumentKind() != ArgumentKind) { 7164 FoundWrongKind = true; 7165 return false; 7166 } 7167 TypeInfo.Type = I->getMatchingCType(); 7168 TypeInfo.LayoutCompatible = I->getLayoutCompatible(); 7169 TypeInfo.MustBeNull = I->getMustBeNull(); 7170 return true; 7171 } 7172 return false; 7173 } 7174 7175 if (!MagicValues) 7176 return false; 7177 7178 llvm::DenseMap<Sema::TypeTagMagicValue, 7179 Sema::TypeTagData>::const_iterator I = 7180 MagicValues->find(std::make_pair(ArgumentKind, MagicValue)); 7181 if (I == MagicValues->end()) 7182 return false; 7183 7184 TypeInfo = I->second; 7185 return true; 7186} 7187} // unnamed namespace 7188 7189void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind, 7190 uint64_t MagicValue, QualType Type, 7191 bool LayoutCompatible, 7192 bool MustBeNull) { 7193 if (!TypeTagForDatatypeMagicValues) 7194 TypeTagForDatatypeMagicValues.reset( 7195 new llvm::DenseMap<TypeTagMagicValue, TypeTagData>); 7196 7197 TypeTagMagicValue Magic(ArgumentKind, MagicValue); 7198 (*TypeTagForDatatypeMagicValues)[Magic] = 7199 TypeTagData(Type, LayoutCompatible, MustBeNull); 7200} 7201 7202namespace { 7203bool IsSameCharType(QualType T1, QualType T2) { 7204 const BuiltinType *BT1 = T1->getAs<BuiltinType>(); 7205 if (!BT1) 7206 return false; 7207 7208 const BuiltinType *BT2 = T2->getAs<BuiltinType>(); 7209 if (!BT2) 7210 return false; 7211 7212 BuiltinType::Kind T1Kind = BT1->getKind(); 7213 BuiltinType::Kind T2Kind = BT2->getKind(); 7214 7215 return (T1Kind == BuiltinType::SChar && T2Kind == BuiltinType::Char_S) || 7216 (T1Kind == BuiltinType::UChar && T2Kind == BuiltinType::Char_U) || 7217 (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) || 7218 (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar); 7219} 7220} // unnamed namespace 7221 7222void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr, 7223 const Expr * const *ExprArgs) { 7224 const IdentifierInfo *ArgumentKind = Attr->getArgumentKind(); 7225 bool IsPointerAttr = Attr->getIsPointer(); 7226 7227 const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()]; 7228 bool FoundWrongKind; 7229 TypeTagData TypeInfo; 7230 if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context, 7231 TypeTagForDatatypeMagicValues.get(), 7232 FoundWrongKind, TypeInfo)) { 7233 if (FoundWrongKind) 7234 Diag(TypeTagExpr->getExprLoc(), 7235 diag::warn_type_tag_for_datatype_wrong_kind) 7236 << TypeTagExpr->getSourceRange(); 7237 return; 7238 } 7239 7240 const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()]; 7241 if (IsPointerAttr) { 7242 // Skip implicit cast of pointer to `void *' (as a function argument). 7243 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr)) 7244 if (ICE->getType()->isVoidPointerType() && 7245 ICE->getCastKind() == CK_BitCast) 7246 ArgumentExpr = ICE->getSubExpr(); 7247 } 7248 QualType ArgumentType = ArgumentExpr->getType(); 7249 7250 // Passing a `void*' pointer shouldn't trigger a warning. 7251 if (IsPointerAttr && ArgumentType->isVoidPointerType()) 7252 return; 7253 7254 if (TypeInfo.MustBeNull) { 7255 // Type tag with matching void type requires a null pointer. 7256 if (!ArgumentExpr->isNullPointerConstant(Context, 7257 Expr::NPC_ValueDependentIsNotNull)) { 7258 Diag(ArgumentExpr->getExprLoc(), 7259 diag::warn_type_safety_null_pointer_required) 7260 << ArgumentKind->getName() 7261 << ArgumentExpr->getSourceRange() 7262 << TypeTagExpr->getSourceRange(); 7263 } 7264 return; 7265 } 7266 7267 QualType RequiredType = TypeInfo.Type; 7268 if (IsPointerAttr) 7269 RequiredType = Context.getPointerType(RequiredType); 7270 7271 bool mismatch = false; 7272 if (!TypeInfo.LayoutCompatible) { 7273 mismatch = !Context.hasSameType(ArgumentType, RequiredType); 7274 7275 // C++11 [basic.fundamental] p1: 7276 // Plain char, signed char, and unsigned char are three distinct types. 7277 // 7278 // But we treat plain `char' as equivalent to `signed char' or `unsigned 7279 // char' depending on the current char signedness mode. 7280 if (mismatch) 7281 if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(), 7282 RequiredType->getPointeeType())) || 7283 (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType))) 7284 mismatch = false; 7285 } else 7286 if (IsPointerAttr) 7287 mismatch = !isLayoutCompatible(Context, 7288 ArgumentType->getPointeeType(), 7289 RequiredType->getPointeeType()); 7290 else 7291 mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType); 7292 7293 if (mismatch) 7294 Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch) 7295 << ArgumentType << ArgumentKind->getName() 7296 << TypeInfo.LayoutCompatible << RequiredType 7297 << ArgumentExpr->getSourceRange() 7298 << TypeTagExpr->getSourceRange(); 7299} 7300