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