SemaChecking.cpp revision 2ebb98a78471f2039af4321bae657e7daacc2a62
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/Initialization.h" 16#include "clang/Sema/Sema.h" 17#include "clang/Sema/SemaInternal.h" 18#include "clang/Sema/Initialization.h" 19#include "clang/Sema/ScopeInfo.h" 20#include "clang/Analysis/Analyses/FormatString.h" 21#include "clang/AST/ASTContext.h" 22#include "clang/AST/CharUnits.h" 23#include "clang/AST/DeclCXX.h" 24#include "clang/AST/DeclObjC.h" 25#include "clang/AST/ExprCXX.h" 26#include "clang/AST/ExprObjC.h" 27#include "clang/AST/EvaluatedExprVisitor.h" 28#include "clang/AST/DeclObjC.h" 29#include "clang/AST/StmtCXX.h" 30#include "clang/AST/StmtObjC.h" 31#include "clang/Lex/Preprocessor.h" 32#include "llvm/ADT/BitVector.h" 33#include "llvm/ADT/SmallString.h" 34#include "llvm/ADT/STLExtras.h" 35#include "llvm/Support/raw_ostream.h" 36#include "clang/Basic/TargetBuiltins.h" 37#include "clang/Basic/TargetInfo.h" 38#include "clang/Basic/ConvertUTF.h" 39#include <limits> 40using namespace clang; 41using namespace sema; 42 43SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 44 unsigned ByteNo) const { 45 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 46 PP.getLangOpts(), PP.getTargetInfo()); 47} 48 49/// Checks that a call expression's argument count is the desired number. 50/// This is useful when doing custom type-checking. Returns true on error. 51static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 52 unsigned argCount = call->getNumArgs(); 53 if (argCount == desiredArgCount) return false; 54 55 if (argCount < desiredArgCount) 56 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 57 << 0 /*function call*/ << desiredArgCount << argCount 58 << call->getSourceRange(); 59 60 // Highlight all the excess arguments. 61 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 62 call->getArg(argCount - 1)->getLocEnd()); 63 64 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 65 << 0 /*function call*/ << desiredArgCount << argCount 66 << call->getArg(1)->getSourceRange(); 67} 68 69/// CheckBuiltinAnnotationString - Checks that string argument to the builtin 70/// annotation is a non wide string literal. 71static bool CheckBuiltinAnnotationString(Sema &S, Expr *Arg) { 72 Arg = Arg->IgnoreParenCasts(); 73 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 74 if (!Literal || !Literal->isAscii()) { 75 S.Diag(Arg->getLocStart(), diag::err_builtin_annotation_not_string_constant) 76 << Arg->getSourceRange(); 77 return true; 78 } 79 return false; 80} 81 82ExprResult 83Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 84 ExprResult TheCallResult(Owned(TheCall)); 85 86 // Find out if any arguments are required to be integer constant expressions. 87 unsigned ICEArguments = 0; 88 ASTContext::GetBuiltinTypeError Error; 89 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 90 if (Error != ASTContext::GE_None) 91 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 92 93 // If any arguments are required to be ICE's, check and diagnose. 94 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 95 // Skip arguments not required to be ICE's. 96 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 97 98 llvm::APSInt Result; 99 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 100 return true; 101 ICEArguments &= ~(1 << ArgNo); 102 } 103 104 switch (BuiltinID) { 105 case Builtin::BI__builtin___CFStringMakeConstantString: 106 assert(TheCall->getNumArgs() == 1 && 107 "Wrong # arguments to builtin CFStringMakeConstantString"); 108 if (CheckObjCString(TheCall->getArg(0))) 109 return ExprError(); 110 break; 111 case Builtin::BI__builtin_stdarg_start: 112 case Builtin::BI__builtin_va_start: 113 if (SemaBuiltinVAStart(TheCall)) 114 return ExprError(); 115 break; 116 case Builtin::BI__builtin_isgreater: 117 case Builtin::BI__builtin_isgreaterequal: 118 case Builtin::BI__builtin_isless: 119 case Builtin::BI__builtin_islessequal: 120 case Builtin::BI__builtin_islessgreater: 121 case Builtin::BI__builtin_isunordered: 122 if (SemaBuiltinUnorderedCompare(TheCall)) 123 return ExprError(); 124 break; 125 case Builtin::BI__builtin_fpclassify: 126 if (SemaBuiltinFPClassification(TheCall, 6)) 127 return ExprError(); 128 break; 129 case Builtin::BI__builtin_isfinite: 130 case Builtin::BI__builtin_isinf: 131 case Builtin::BI__builtin_isinf_sign: 132 case Builtin::BI__builtin_isnan: 133 case Builtin::BI__builtin_isnormal: 134 if (SemaBuiltinFPClassification(TheCall, 1)) 135 return ExprError(); 136 break; 137 case Builtin::BI__builtin_shufflevector: 138 return SemaBuiltinShuffleVector(TheCall); 139 // TheCall will be freed by the smart pointer here, but that's fine, since 140 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 141 case Builtin::BI__builtin_prefetch: 142 if (SemaBuiltinPrefetch(TheCall)) 143 return ExprError(); 144 break; 145 case Builtin::BI__builtin_object_size: 146 if (SemaBuiltinObjectSize(TheCall)) 147 return ExprError(); 148 break; 149 case Builtin::BI__builtin_longjmp: 150 if (SemaBuiltinLongjmp(TheCall)) 151 return ExprError(); 152 break; 153 154 case Builtin::BI__builtin_classify_type: 155 if (checkArgCount(*this, TheCall, 1)) return true; 156 TheCall->setType(Context.IntTy); 157 break; 158 case Builtin::BI__builtin_constant_p: 159 if (checkArgCount(*this, TheCall, 1)) return true; 160 TheCall->setType(Context.IntTy); 161 break; 162 case Builtin::BI__sync_fetch_and_add: 163 case Builtin::BI__sync_fetch_and_add_1: 164 case Builtin::BI__sync_fetch_and_add_2: 165 case Builtin::BI__sync_fetch_and_add_4: 166 case Builtin::BI__sync_fetch_and_add_8: 167 case Builtin::BI__sync_fetch_and_add_16: 168 case Builtin::BI__sync_fetch_and_sub: 169 case Builtin::BI__sync_fetch_and_sub_1: 170 case Builtin::BI__sync_fetch_and_sub_2: 171 case Builtin::BI__sync_fetch_and_sub_4: 172 case Builtin::BI__sync_fetch_and_sub_8: 173 case Builtin::BI__sync_fetch_and_sub_16: 174 case Builtin::BI__sync_fetch_and_or: 175 case Builtin::BI__sync_fetch_and_or_1: 176 case Builtin::BI__sync_fetch_and_or_2: 177 case Builtin::BI__sync_fetch_and_or_4: 178 case Builtin::BI__sync_fetch_and_or_8: 179 case Builtin::BI__sync_fetch_and_or_16: 180 case Builtin::BI__sync_fetch_and_and: 181 case Builtin::BI__sync_fetch_and_and_1: 182 case Builtin::BI__sync_fetch_and_and_2: 183 case Builtin::BI__sync_fetch_and_and_4: 184 case Builtin::BI__sync_fetch_and_and_8: 185 case Builtin::BI__sync_fetch_and_and_16: 186 case Builtin::BI__sync_fetch_and_xor: 187 case Builtin::BI__sync_fetch_and_xor_1: 188 case Builtin::BI__sync_fetch_and_xor_2: 189 case Builtin::BI__sync_fetch_and_xor_4: 190 case Builtin::BI__sync_fetch_and_xor_8: 191 case Builtin::BI__sync_fetch_and_xor_16: 192 case Builtin::BI__sync_add_and_fetch: 193 case Builtin::BI__sync_add_and_fetch_1: 194 case Builtin::BI__sync_add_and_fetch_2: 195 case Builtin::BI__sync_add_and_fetch_4: 196 case Builtin::BI__sync_add_and_fetch_8: 197 case Builtin::BI__sync_add_and_fetch_16: 198 case Builtin::BI__sync_sub_and_fetch: 199 case Builtin::BI__sync_sub_and_fetch_1: 200 case Builtin::BI__sync_sub_and_fetch_2: 201 case Builtin::BI__sync_sub_and_fetch_4: 202 case Builtin::BI__sync_sub_and_fetch_8: 203 case Builtin::BI__sync_sub_and_fetch_16: 204 case Builtin::BI__sync_and_and_fetch: 205 case Builtin::BI__sync_and_and_fetch_1: 206 case Builtin::BI__sync_and_and_fetch_2: 207 case Builtin::BI__sync_and_and_fetch_4: 208 case Builtin::BI__sync_and_and_fetch_8: 209 case Builtin::BI__sync_and_and_fetch_16: 210 case Builtin::BI__sync_or_and_fetch: 211 case Builtin::BI__sync_or_and_fetch_1: 212 case Builtin::BI__sync_or_and_fetch_2: 213 case Builtin::BI__sync_or_and_fetch_4: 214 case Builtin::BI__sync_or_and_fetch_8: 215 case Builtin::BI__sync_or_and_fetch_16: 216 case Builtin::BI__sync_xor_and_fetch: 217 case Builtin::BI__sync_xor_and_fetch_1: 218 case Builtin::BI__sync_xor_and_fetch_2: 219 case Builtin::BI__sync_xor_and_fetch_4: 220 case Builtin::BI__sync_xor_and_fetch_8: 221 case Builtin::BI__sync_xor_and_fetch_16: 222 case Builtin::BI__sync_val_compare_and_swap: 223 case Builtin::BI__sync_val_compare_and_swap_1: 224 case Builtin::BI__sync_val_compare_and_swap_2: 225 case Builtin::BI__sync_val_compare_and_swap_4: 226 case Builtin::BI__sync_val_compare_and_swap_8: 227 case Builtin::BI__sync_val_compare_and_swap_16: 228 case Builtin::BI__sync_bool_compare_and_swap: 229 case Builtin::BI__sync_bool_compare_and_swap_1: 230 case Builtin::BI__sync_bool_compare_and_swap_2: 231 case Builtin::BI__sync_bool_compare_and_swap_4: 232 case Builtin::BI__sync_bool_compare_and_swap_8: 233 case Builtin::BI__sync_bool_compare_and_swap_16: 234 case Builtin::BI__sync_lock_test_and_set: 235 case Builtin::BI__sync_lock_test_and_set_1: 236 case Builtin::BI__sync_lock_test_and_set_2: 237 case Builtin::BI__sync_lock_test_and_set_4: 238 case Builtin::BI__sync_lock_test_and_set_8: 239 case Builtin::BI__sync_lock_test_and_set_16: 240 case Builtin::BI__sync_lock_release: 241 case Builtin::BI__sync_lock_release_1: 242 case Builtin::BI__sync_lock_release_2: 243 case Builtin::BI__sync_lock_release_4: 244 case Builtin::BI__sync_lock_release_8: 245 case Builtin::BI__sync_lock_release_16: 246 case Builtin::BI__sync_swap: 247 case Builtin::BI__sync_swap_1: 248 case Builtin::BI__sync_swap_2: 249 case Builtin::BI__sync_swap_4: 250 case Builtin::BI__sync_swap_8: 251 case Builtin::BI__sync_swap_16: 252 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 253 case Builtin::BI__atomic_load: 254 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Load); 255 case Builtin::BI__atomic_store: 256 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Store); 257 case Builtin::BI__atomic_init: 258 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Init); 259 case Builtin::BI__atomic_exchange: 260 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xchg); 261 case Builtin::BI__atomic_compare_exchange_strong: 262 return SemaAtomicOpsOverloaded(move(TheCallResult), 263 AtomicExpr::CmpXchgStrong); 264 case Builtin::BI__atomic_compare_exchange_weak: 265 return SemaAtomicOpsOverloaded(move(TheCallResult), 266 AtomicExpr::CmpXchgWeak); 267 case Builtin::BI__atomic_fetch_add: 268 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Add); 269 case Builtin::BI__atomic_fetch_sub: 270 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Sub); 271 case Builtin::BI__atomic_fetch_and: 272 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::And); 273 case Builtin::BI__atomic_fetch_or: 274 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Or); 275 case Builtin::BI__atomic_fetch_xor: 276 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xor); 277 case Builtin::BI__builtin_annotation: 278 if (CheckBuiltinAnnotationString(*this, TheCall->getArg(1))) 279 return ExprError(); 280 break; 281 } 282 283 // Since the target specific builtins for each arch overlap, only check those 284 // of the arch we are compiling for. 285 if (BuiltinID >= Builtin::FirstTSBuiltin) { 286 switch (Context.getTargetInfo().getTriple().getArch()) { 287 case llvm::Triple::arm: 288 case llvm::Triple::thumb: 289 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 290 return ExprError(); 291 break; 292 default: 293 break; 294 } 295 } 296 297 return move(TheCallResult); 298} 299 300// Get the valid immediate range for the specified NEON type code. 301static unsigned RFT(unsigned t, bool shift = false) { 302 NeonTypeFlags Type(t); 303 int IsQuad = Type.isQuad(); 304 switch (Type.getEltType()) { 305 case NeonTypeFlags::Int8: 306 case NeonTypeFlags::Poly8: 307 return shift ? 7 : (8 << IsQuad) - 1; 308 case NeonTypeFlags::Int16: 309 case NeonTypeFlags::Poly16: 310 return shift ? 15 : (4 << IsQuad) - 1; 311 case NeonTypeFlags::Int32: 312 return shift ? 31 : (2 << IsQuad) - 1; 313 case NeonTypeFlags::Int64: 314 return shift ? 63 : (1 << IsQuad) - 1; 315 case NeonTypeFlags::Float16: 316 assert(!shift && "cannot shift float types!"); 317 return (4 << IsQuad) - 1; 318 case NeonTypeFlags::Float32: 319 assert(!shift && "cannot shift float types!"); 320 return (2 << IsQuad) - 1; 321 } 322 llvm_unreachable("Invalid NeonTypeFlag!"); 323} 324 325/// getNeonEltType - Return the QualType corresponding to the elements of 326/// the vector type specified by the NeonTypeFlags. This is used to check 327/// the pointer arguments for Neon load/store intrinsics. 328static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) { 329 switch (Flags.getEltType()) { 330 case NeonTypeFlags::Int8: 331 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 332 case NeonTypeFlags::Int16: 333 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 334 case NeonTypeFlags::Int32: 335 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 336 case NeonTypeFlags::Int64: 337 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 338 case NeonTypeFlags::Poly8: 339 return Context.SignedCharTy; 340 case NeonTypeFlags::Poly16: 341 return Context.ShortTy; 342 case NeonTypeFlags::Float16: 343 return Context.UnsignedShortTy; 344 case NeonTypeFlags::Float32: 345 return Context.FloatTy; 346 } 347 llvm_unreachable("Invalid NeonTypeFlag!"); 348} 349 350bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 351 llvm::APSInt Result; 352 353 unsigned mask = 0; 354 unsigned TV = 0; 355 int PtrArgNum = -1; 356 bool HasConstPtr = false; 357 switch (BuiltinID) { 358#define GET_NEON_OVERLOAD_CHECK 359#include "clang/Basic/arm_neon.inc" 360#undef GET_NEON_OVERLOAD_CHECK 361 } 362 363 // For NEON intrinsics which are overloaded on vector element type, validate 364 // the immediate which specifies which variant to emit. 365 unsigned ImmArg = TheCall->getNumArgs()-1; 366 if (mask) { 367 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 368 return true; 369 370 TV = Result.getLimitedValue(64); 371 if ((TV > 63) || (mask & (1 << TV)) == 0) 372 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 373 << TheCall->getArg(ImmArg)->getSourceRange(); 374 } 375 376 if (PtrArgNum >= 0) { 377 // Check that pointer arguments have the specified type. 378 Expr *Arg = TheCall->getArg(PtrArgNum); 379 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 380 Arg = ICE->getSubExpr(); 381 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 382 QualType RHSTy = RHS.get()->getType(); 383 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 384 if (HasConstPtr) 385 EltTy = EltTy.withConst(); 386 QualType LHSTy = Context.getPointerType(EltTy); 387 AssignConvertType ConvTy; 388 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 389 if (RHS.isInvalid()) 390 return true; 391 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 392 RHS.get(), AA_Assigning)) 393 return true; 394 } 395 396 // For NEON intrinsics which take an immediate value as part of the 397 // instruction, range check them here. 398 unsigned i = 0, l = 0, u = 0; 399 switch (BuiltinID) { 400 default: return false; 401 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 402 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 403 case ARM::BI__builtin_arm_vcvtr_f: 404 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 405#define GET_NEON_IMMEDIATE_CHECK 406#include "clang/Basic/arm_neon.inc" 407#undef GET_NEON_IMMEDIATE_CHECK 408 }; 409 410 // Check that the immediate argument is actually a constant. 411 if (SemaBuiltinConstantArg(TheCall, i, Result)) 412 return true; 413 414 // Range check against the upper/lower values for this isntruction. 415 unsigned Val = Result.getZExtValue(); 416 if (Val < l || Val > (u + l)) 417 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 418 << l << u+l << TheCall->getArg(i)->getSourceRange(); 419 420 // FIXME: VFP Intrinsics should error if VFP not present. 421 return false; 422} 423 424/// CheckFunctionCall - Check a direct function call for various correctness 425/// and safety properties not strictly enforced by the C type system. 426bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 427 // Get the IdentifierInfo* for the called function. 428 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 429 430 // None of the checks below are needed for functions that don't have 431 // simple names (e.g., C++ conversion functions). 432 if (!FnInfo) 433 return false; 434 435 // FIXME: This mechanism should be abstracted to be less fragile and 436 // more efficient. For example, just map function ids to custom 437 // handlers. 438 439 // Printf and scanf checking. 440 for (specific_attr_iterator<FormatAttr> 441 i = FDecl->specific_attr_begin<FormatAttr>(), 442 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 443 CheckFormatArguments(*i, TheCall); 444 } 445 446 for (specific_attr_iterator<NonNullAttr> 447 i = FDecl->specific_attr_begin<NonNullAttr>(), 448 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 449 CheckNonNullArguments(*i, TheCall->getArgs(), 450 TheCall->getCallee()->getLocStart()); 451 } 452 453 unsigned CMId = FDecl->getMemoryFunctionKind(); 454 if (CMId == 0) 455 return false; 456 457 // Handle memory setting and copying functions. 458 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 459 CheckStrlcpycatArguments(TheCall, FnInfo); 460 else if (CMId == Builtin::BIstrncat) 461 CheckStrncatArguments(TheCall, FnInfo); 462 else 463 CheckMemaccessArguments(TheCall, CMId, FnInfo); 464 465 return false; 466} 467 468bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 469 Expr **Args, unsigned NumArgs) { 470 for (specific_attr_iterator<FormatAttr> 471 i = Method->specific_attr_begin<FormatAttr>(), 472 e = Method->specific_attr_end<FormatAttr>(); i != e ; ++i) { 473 474 CheckFormatArguments(*i, Args, NumArgs, false, lbrac, 475 Method->getSourceRange()); 476 } 477 478 // diagnose nonnull arguments. 479 for (specific_attr_iterator<NonNullAttr> 480 i = Method->specific_attr_begin<NonNullAttr>(), 481 e = Method->specific_attr_end<NonNullAttr>(); i != e; ++i) { 482 CheckNonNullArguments(*i, Args, lbrac); 483 } 484 485 return false; 486} 487 488bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 489 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 490 if (!V) 491 return false; 492 493 QualType Ty = V->getType(); 494 if (!Ty->isBlockPointerType()) 495 return false; 496 497 // format string checking. 498 for (specific_attr_iterator<FormatAttr> 499 i = NDecl->specific_attr_begin<FormatAttr>(), 500 e = NDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 501 CheckFormatArguments(*i, TheCall); 502 } 503 504 return false; 505} 506 507ExprResult 508Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, AtomicExpr::AtomicOp Op) { 509 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 510 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 511 512 // All these operations take one of the following four forms: 513 // T __atomic_load(_Atomic(T)*, int) (loads) 514 // T* __atomic_add(_Atomic(T*)*, ptrdiff_t, int) (pointer add/sub) 515 // int __atomic_compare_exchange_strong(_Atomic(T)*, T*, T, int, int) 516 // (cmpxchg) 517 // T __atomic_exchange(_Atomic(T)*, T, int) (everything else) 518 // where T is an appropriate type, and the int paremeterss are for orderings. 519 unsigned NumVals = 1; 520 unsigned NumOrders = 1; 521 if (Op == AtomicExpr::Load) { 522 NumVals = 0; 523 } else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) { 524 NumVals = 2; 525 NumOrders = 2; 526 } 527 if (Op == AtomicExpr::Init) 528 NumOrders = 0; 529 530 if (TheCall->getNumArgs() < NumVals+NumOrders+1) { 531 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 532 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs() 533 << TheCall->getCallee()->getSourceRange(); 534 return ExprError(); 535 } else if (TheCall->getNumArgs() > NumVals+NumOrders+1) { 536 Diag(TheCall->getArg(NumVals+NumOrders+1)->getLocStart(), 537 diag::err_typecheck_call_too_many_args) 538 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs() 539 << TheCall->getCallee()->getSourceRange(); 540 return ExprError(); 541 } 542 543 // Inspect the first argument of the atomic operation. This should always be 544 // a pointer to an _Atomic type. 545 Expr *Ptr = TheCall->getArg(0); 546 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 547 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 548 if (!pointerType) { 549 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 550 << Ptr->getType() << Ptr->getSourceRange(); 551 return ExprError(); 552 } 553 554 QualType AtomTy = pointerType->getPointeeType(); 555 if (!AtomTy->isAtomicType()) { 556 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 557 << Ptr->getType() << Ptr->getSourceRange(); 558 return ExprError(); 559 } 560 QualType ValType = AtomTy->getAs<AtomicType>()->getValueType(); 561 562 if ((Op == AtomicExpr::Add || Op == AtomicExpr::Sub) && 563 !ValType->isIntegerType() && !ValType->isPointerType()) { 564 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 565 << Ptr->getType() << Ptr->getSourceRange(); 566 return ExprError(); 567 } 568 569 if (!ValType->isIntegerType() && 570 (Op == AtomicExpr::And || Op == AtomicExpr::Or || Op == AtomicExpr::Xor)){ 571 Diag(DRE->getLocStart(), diag::err_atomic_op_logical_needs_atomic_int) 572 << Ptr->getType() << Ptr->getSourceRange(); 573 return ExprError(); 574 } 575 576 switch (ValType.getObjCLifetime()) { 577 case Qualifiers::OCL_None: 578 case Qualifiers::OCL_ExplicitNone: 579 // okay 580 break; 581 582 case Qualifiers::OCL_Weak: 583 case Qualifiers::OCL_Strong: 584 case Qualifiers::OCL_Autoreleasing: 585 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 586 << ValType << Ptr->getSourceRange(); 587 return ExprError(); 588 } 589 590 QualType ResultType = ValType; 591 if (Op == AtomicExpr::Store || Op == AtomicExpr::Init) 592 ResultType = Context.VoidTy; 593 else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) 594 ResultType = Context.BoolTy; 595 596 // The first argument --- the pointer --- has a fixed type; we 597 // deduce the types of the rest of the arguments accordingly. Walk 598 // the remaining arguments, converting them to the deduced value type. 599 for (unsigned i = 1; i != NumVals+NumOrders+1; ++i) { 600 ExprResult Arg = TheCall->getArg(i); 601 QualType Ty; 602 if (i < NumVals+1) { 603 // The second argument to a cmpxchg is a pointer to the data which will 604 // be exchanged. The second argument to a pointer add/subtract is the 605 // amount to add/subtract, which must be a ptrdiff_t. The third 606 // argument to a cmpxchg and the second argument in all other cases 607 // is the type of the value. 608 if (i == 1 && (Op == AtomicExpr::CmpXchgWeak || 609 Op == AtomicExpr::CmpXchgStrong)) 610 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 611 else if (!ValType->isIntegerType() && 612 (Op == AtomicExpr::Add || Op == AtomicExpr::Sub)) 613 Ty = Context.getPointerDiffType(); 614 else 615 Ty = ValType; 616 } else { 617 // The order(s) are always converted to int. 618 Ty = Context.IntTy; 619 } 620 InitializedEntity Entity = 621 InitializedEntity::InitializeParameter(Context, Ty, false); 622 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 623 if (Arg.isInvalid()) 624 return true; 625 TheCall->setArg(i, Arg.get()); 626 } 627 628 SmallVector<Expr*, 5> SubExprs; 629 SubExprs.push_back(Ptr); 630 if (Op == AtomicExpr::Load) { 631 SubExprs.push_back(TheCall->getArg(1)); // Order 632 } else if (Op == AtomicExpr::Init) { 633 SubExprs.push_back(TheCall->getArg(1)); // Val1 634 } else if (Op != AtomicExpr::CmpXchgWeak && Op != AtomicExpr::CmpXchgStrong) { 635 SubExprs.push_back(TheCall->getArg(2)); // Order 636 SubExprs.push_back(TheCall->getArg(1)); // Val1 637 } else { 638 SubExprs.push_back(TheCall->getArg(3)); // Order 639 SubExprs.push_back(TheCall->getArg(1)); // Val1 640 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 641 SubExprs.push_back(TheCall->getArg(2)); // Val2 642 } 643 644 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 645 SubExprs.data(), SubExprs.size(), 646 ResultType, Op, 647 TheCall->getRParenLoc())); 648} 649 650 651/// checkBuiltinArgument - Given a call to a builtin function, perform 652/// normal type-checking on the given argument, updating the call in 653/// place. This is useful when a builtin function requires custom 654/// type-checking for some of its arguments but not necessarily all of 655/// them. 656/// 657/// Returns true on error. 658static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 659 FunctionDecl *Fn = E->getDirectCallee(); 660 assert(Fn && "builtin call without direct callee!"); 661 662 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 663 InitializedEntity Entity = 664 InitializedEntity::InitializeParameter(S.Context, Param); 665 666 ExprResult Arg = E->getArg(0); 667 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 668 if (Arg.isInvalid()) 669 return true; 670 671 E->setArg(ArgIndex, Arg.take()); 672 return false; 673} 674 675/// SemaBuiltinAtomicOverloaded - We have a call to a function like 676/// __sync_fetch_and_add, which is an overloaded function based on the pointer 677/// type of its first argument. The main ActOnCallExpr routines have already 678/// promoted the types of arguments because all of these calls are prototyped as 679/// void(...). 680/// 681/// This function goes through and does final semantic checking for these 682/// builtins, 683ExprResult 684Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 685 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 686 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 687 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 688 689 // Ensure that we have at least one argument to do type inference from. 690 if (TheCall->getNumArgs() < 1) { 691 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 692 << 0 << 1 << TheCall->getNumArgs() 693 << TheCall->getCallee()->getSourceRange(); 694 return ExprError(); 695 } 696 697 // Inspect the first argument of the atomic builtin. This should always be 698 // a pointer type, whose element is an integral scalar or pointer type. 699 // Because it is a pointer type, we don't have to worry about any implicit 700 // casts here. 701 // FIXME: We don't allow floating point scalars as input. 702 Expr *FirstArg = TheCall->getArg(0); 703 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 704 if (FirstArgResult.isInvalid()) 705 return ExprError(); 706 FirstArg = FirstArgResult.take(); 707 TheCall->setArg(0, FirstArg); 708 709 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 710 if (!pointerType) { 711 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 712 << FirstArg->getType() << FirstArg->getSourceRange(); 713 return ExprError(); 714 } 715 716 QualType ValType = pointerType->getPointeeType(); 717 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 718 !ValType->isBlockPointerType()) { 719 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 720 << FirstArg->getType() << FirstArg->getSourceRange(); 721 return ExprError(); 722 } 723 724 switch (ValType.getObjCLifetime()) { 725 case Qualifiers::OCL_None: 726 case Qualifiers::OCL_ExplicitNone: 727 // okay 728 break; 729 730 case Qualifiers::OCL_Weak: 731 case Qualifiers::OCL_Strong: 732 case Qualifiers::OCL_Autoreleasing: 733 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 734 << ValType << FirstArg->getSourceRange(); 735 return ExprError(); 736 } 737 738 // Strip any qualifiers off ValType. 739 ValType = ValType.getUnqualifiedType(); 740 741 // The majority of builtins return a value, but a few have special return 742 // types, so allow them to override appropriately below. 743 QualType ResultType = ValType; 744 745 // We need to figure out which concrete builtin this maps onto. For example, 746 // __sync_fetch_and_add with a 2 byte object turns into 747 // __sync_fetch_and_add_2. 748#define BUILTIN_ROW(x) \ 749 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 750 Builtin::BI##x##_8, Builtin::BI##x##_16 } 751 752 static const unsigned BuiltinIndices[][5] = { 753 BUILTIN_ROW(__sync_fetch_and_add), 754 BUILTIN_ROW(__sync_fetch_and_sub), 755 BUILTIN_ROW(__sync_fetch_and_or), 756 BUILTIN_ROW(__sync_fetch_and_and), 757 BUILTIN_ROW(__sync_fetch_and_xor), 758 759 BUILTIN_ROW(__sync_add_and_fetch), 760 BUILTIN_ROW(__sync_sub_and_fetch), 761 BUILTIN_ROW(__sync_and_and_fetch), 762 BUILTIN_ROW(__sync_or_and_fetch), 763 BUILTIN_ROW(__sync_xor_and_fetch), 764 765 BUILTIN_ROW(__sync_val_compare_and_swap), 766 BUILTIN_ROW(__sync_bool_compare_and_swap), 767 BUILTIN_ROW(__sync_lock_test_and_set), 768 BUILTIN_ROW(__sync_lock_release), 769 BUILTIN_ROW(__sync_swap) 770 }; 771#undef BUILTIN_ROW 772 773 // Determine the index of the size. 774 unsigned SizeIndex; 775 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 776 case 1: SizeIndex = 0; break; 777 case 2: SizeIndex = 1; break; 778 case 4: SizeIndex = 2; break; 779 case 8: SizeIndex = 3; break; 780 case 16: SizeIndex = 4; break; 781 default: 782 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 783 << FirstArg->getType() << FirstArg->getSourceRange(); 784 return ExprError(); 785 } 786 787 // Each of these builtins has one pointer argument, followed by some number of 788 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 789 // that we ignore. Find out which row of BuiltinIndices to read from as well 790 // as the number of fixed args. 791 unsigned BuiltinID = FDecl->getBuiltinID(); 792 unsigned BuiltinIndex, NumFixed = 1; 793 switch (BuiltinID) { 794 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 795 case Builtin::BI__sync_fetch_and_add: 796 case Builtin::BI__sync_fetch_and_add_1: 797 case Builtin::BI__sync_fetch_and_add_2: 798 case Builtin::BI__sync_fetch_and_add_4: 799 case Builtin::BI__sync_fetch_and_add_8: 800 case Builtin::BI__sync_fetch_and_add_16: 801 BuiltinIndex = 0; 802 break; 803 804 case Builtin::BI__sync_fetch_and_sub: 805 case Builtin::BI__sync_fetch_and_sub_1: 806 case Builtin::BI__sync_fetch_and_sub_2: 807 case Builtin::BI__sync_fetch_and_sub_4: 808 case Builtin::BI__sync_fetch_and_sub_8: 809 case Builtin::BI__sync_fetch_and_sub_16: 810 BuiltinIndex = 1; 811 break; 812 813 case Builtin::BI__sync_fetch_and_or: 814 case Builtin::BI__sync_fetch_and_or_1: 815 case Builtin::BI__sync_fetch_and_or_2: 816 case Builtin::BI__sync_fetch_and_or_4: 817 case Builtin::BI__sync_fetch_and_or_8: 818 case Builtin::BI__sync_fetch_and_or_16: 819 BuiltinIndex = 2; 820 break; 821 822 case Builtin::BI__sync_fetch_and_and: 823 case Builtin::BI__sync_fetch_and_and_1: 824 case Builtin::BI__sync_fetch_and_and_2: 825 case Builtin::BI__sync_fetch_and_and_4: 826 case Builtin::BI__sync_fetch_and_and_8: 827 case Builtin::BI__sync_fetch_and_and_16: 828 BuiltinIndex = 3; 829 break; 830 831 case Builtin::BI__sync_fetch_and_xor: 832 case Builtin::BI__sync_fetch_and_xor_1: 833 case Builtin::BI__sync_fetch_and_xor_2: 834 case Builtin::BI__sync_fetch_and_xor_4: 835 case Builtin::BI__sync_fetch_and_xor_8: 836 case Builtin::BI__sync_fetch_and_xor_16: 837 BuiltinIndex = 4; 838 break; 839 840 case Builtin::BI__sync_add_and_fetch: 841 case Builtin::BI__sync_add_and_fetch_1: 842 case Builtin::BI__sync_add_and_fetch_2: 843 case Builtin::BI__sync_add_and_fetch_4: 844 case Builtin::BI__sync_add_and_fetch_8: 845 case Builtin::BI__sync_add_and_fetch_16: 846 BuiltinIndex = 5; 847 break; 848 849 case Builtin::BI__sync_sub_and_fetch: 850 case Builtin::BI__sync_sub_and_fetch_1: 851 case Builtin::BI__sync_sub_and_fetch_2: 852 case Builtin::BI__sync_sub_and_fetch_4: 853 case Builtin::BI__sync_sub_and_fetch_8: 854 case Builtin::BI__sync_sub_and_fetch_16: 855 BuiltinIndex = 6; 856 break; 857 858 case Builtin::BI__sync_and_and_fetch: 859 case Builtin::BI__sync_and_and_fetch_1: 860 case Builtin::BI__sync_and_and_fetch_2: 861 case Builtin::BI__sync_and_and_fetch_4: 862 case Builtin::BI__sync_and_and_fetch_8: 863 case Builtin::BI__sync_and_and_fetch_16: 864 BuiltinIndex = 7; 865 break; 866 867 case Builtin::BI__sync_or_and_fetch: 868 case Builtin::BI__sync_or_and_fetch_1: 869 case Builtin::BI__sync_or_and_fetch_2: 870 case Builtin::BI__sync_or_and_fetch_4: 871 case Builtin::BI__sync_or_and_fetch_8: 872 case Builtin::BI__sync_or_and_fetch_16: 873 BuiltinIndex = 8; 874 break; 875 876 case Builtin::BI__sync_xor_and_fetch: 877 case Builtin::BI__sync_xor_and_fetch_1: 878 case Builtin::BI__sync_xor_and_fetch_2: 879 case Builtin::BI__sync_xor_and_fetch_4: 880 case Builtin::BI__sync_xor_and_fetch_8: 881 case Builtin::BI__sync_xor_and_fetch_16: 882 BuiltinIndex = 9; 883 break; 884 885 case Builtin::BI__sync_val_compare_and_swap: 886 case Builtin::BI__sync_val_compare_and_swap_1: 887 case Builtin::BI__sync_val_compare_and_swap_2: 888 case Builtin::BI__sync_val_compare_and_swap_4: 889 case Builtin::BI__sync_val_compare_and_swap_8: 890 case Builtin::BI__sync_val_compare_and_swap_16: 891 BuiltinIndex = 10; 892 NumFixed = 2; 893 break; 894 895 case Builtin::BI__sync_bool_compare_and_swap: 896 case Builtin::BI__sync_bool_compare_and_swap_1: 897 case Builtin::BI__sync_bool_compare_and_swap_2: 898 case Builtin::BI__sync_bool_compare_and_swap_4: 899 case Builtin::BI__sync_bool_compare_and_swap_8: 900 case Builtin::BI__sync_bool_compare_and_swap_16: 901 BuiltinIndex = 11; 902 NumFixed = 2; 903 ResultType = Context.BoolTy; 904 break; 905 906 case Builtin::BI__sync_lock_test_and_set: 907 case Builtin::BI__sync_lock_test_and_set_1: 908 case Builtin::BI__sync_lock_test_and_set_2: 909 case Builtin::BI__sync_lock_test_and_set_4: 910 case Builtin::BI__sync_lock_test_and_set_8: 911 case Builtin::BI__sync_lock_test_and_set_16: 912 BuiltinIndex = 12; 913 break; 914 915 case Builtin::BI__sync_lock_release: 916 case Builtin::BI__sync_lock_release_1: 917 case Builtin::BI__sync_lock_release_2: 918 case Builtin::BI__sync_lock_release_4: 919 case Builtin::BI__sync_lock_release_8: 920 case Builtin::BI__sync_lock_release_16: 921 BuiltinIndex = 13; 922 NumFixed = 0; 923 ResultType = Context.VoidTy; 924 break; 925 926 case Builtin::BI__sync_swap: 927 case Builtin::BI__sync_swap_1: 928 case Builtin::BI__sync_swap_2: 929 case Builtin::BI__sync_swap_4: 930 case Builtin::BI__sync_swap_8: 931 case Builtin::BI__sync_swap_16: 932 BuiltinIndex = 14; 933 break; 934 } 935 936 // Now that we know how many fixed arguments we expect, first check that we 937 // have at least that many. 938 if (TheCall->getNumArgs() < 1+NumFixed) { 939 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 940 << 0 << 1+NumFixed << TheCall->getNumArgs() 941 << TheCall->getCallee()->getSourceRange(); 942 return ExprError(); 943 } 944 945 // Get the decl for the concrete builtin from this, we can tell what the 946 // concrete integer type we should convert to is. 947 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 948 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 949 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 950 FunctionDecl *NewBuiltinDecl = 951 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 952 TUScope, false, DRE->getLocStart())); 953 954 // The first argument --- the pointer --- has a fixed type; we 955 // deduce the types of the rest of the arguments accordingly. Walk 956 // the remaining arguments, converting them to the deduced value type. 957 for (unsigned i = 0; i != NumFixed; ++i) { 958 ExprResult Arg = TheCall->getArg(i+1); 959 960 // GCC does an implicit conversion to the pointer or integer ValType. This 961 // can fail in some cases (1i -> int**), check for this error case now. 962 // Initialize the argument. 963 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 964 ValType, /*consume*/ false); 965 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 966 if (Arg.isInvalid()) 967 return ExprError(); 968 969 // Okay, we have something that *can* be converted to the right type. Check 970 // to see if there is a potentially weird extension going on here. This can 971 // happen when you do an atomic operation on something like an char* and 972 // pass in 42. The 42 gets converted to char. This is even more strange 973 // for things like 45.123 -> char, etc. 974 // FIXME: Do this check. 975 TheCall->setArg(i+1, Arg.take()); 976 } 977 978 ASTContext& Context = this->getASTContext(); 979 980 // Create a new DeclRefExpr to refer to the new decl. 981 DeclRefExpr* NewDRE = DeclRefExpr::Create( 982 Context, 983 DRE->getQualifierLoc(), 984 SourceLocation(), 985 NewBuiltinDecl, 986 /*enclosing*/ false, 987 DRE->getLocation(), 988 NewBuiltinDecl->getType(), 989 DRE->getValueKind()); 990 991 // Set the callee in the CallExpr. 992 // FIXME: This leaks the original parens and implicit casts. 993 ExprResult PromotedCall = UsualUnaryConversions(NewDRE); 994 if (PromotedCall.isInvalid()) 995 return ExprError(); 996 TheCall->setCallee(PromotedCall.take()); 997 998 // Change the result type of the call to match the original value type. This 999 // is arbitrary, but the codegen for these builtins ins design to handle it 1000 // gracefully. 1001 TheCall->setType(ResultType); 1002 1003 return move(TheCallResult); 1004} 1005 1006/// CheckObjCString - Checks that the argument to the builtin 1007/// CFString constructor is correct 1008/// Note: It might also make sense to do the UTF-16 conversion here (would 1009/// simplify the backend). 1010bool Sema::CheckObjCString(Expr *Arg) { 1011 Arg = Arg->IgnoreParenCasts(); 1012 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1013 1014 if (!Literal || !Literal->isAscii()) { 1015 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1016 << Arg->getSourceRange(); 1017 return true; 1018 } 1019 1020 if (Literal->containsNonAsciiOrNull()) { 1021 StringRef String = Literal->getString(); 1022 unsigned NumBytes = String.size(); 1023 SmallVector<UTF16, 128> ToBuf(NumBytes); 1024 const UTF8 *FromPtr = (UTF8 *)String.data(); 1025 UTF16 *ToPtr = &ToBuf[0]; 1026 1027 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1028 &ToPtr, ToPtr + NumBytes, 1029 strictConversion); 1030 // Check for conversion failure. 1031 if (Result != conversionOK) 1032 Diag(Arg->getLocStart(), 1033 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1034 } 1035 return false; 1036} 1037 1038/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1039/// Emit an error and return true on failure, return false on success. 1040bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1041 Expr *Fn = TheCall->getCallee(); 1042 if (TheCall->getNumArgs() > 2) { 1043 Diag(TheCall->getArg(2)->getLocStart(), 1044 diag::err_typecheck_call_too_many_args) 1045 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1046 << Fn->getSourceRange() 1047 << SourceRange(TheCall->getArg(2)->getLocStart(), 1048 (*(TheCall->arg_end()-1))->getLocEnd()); 1049 return true; 1050 } 1051 1052 if (TheCall->getNumArgs() < 2) { 1053 return Diag(TheCall->getLocEnd(), 1054 diag::err_typecheck_call_too_few_args_at_least) 1055 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1056 } 1057 1058 // Type-check the first argument normally. 1059 if (checkBuiltinArgument(*this, TheCall, 0)) 1060 return true; 1061 1062 // Determine whether the current function is variadic or not. 1063 BlockScopeInfo *CurBlock = getCurBlock(); 1064 bool isVariadic; 1065 if (CurBlock) 1066 isVariadic = CurBlock->TheDecl->isVariadic(); 1067 else if (FunctionDecl *FD = getCurFunctionDecl()) 1068 isVariadic = FD->isVariadic(); 1069 else 1070 isVariadic = getCurMethodDecl()->isVariadic(); 1071 1072 if (!isVariadic) { 1073 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1074 return true; 1075 } 1076 1077 // Verify that the second argument to the builtin is the last argument of the 1078 // current function or method. 1079 bool SecondArgIsLastNamedArgument = false; 1080 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1081 1082 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1083 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1084 // FIXME: This isn't correct for methods (results in bogus warning). 1085 // Get the last formal in the current function. 1086 const ParmVarDecl *LastArg; 1087 if (CurBlock) 1088 LastArg = *(CurBlock->TheDecl->param_end()-1); 1089 else if (FunctionDecl *FD = getCurFunctionDecl()) 1090 LastArg = *(FD->param_end()-1); 1091 else 1092 LastArg = *(getCurMethodDecl()->param_end()-1); 1093 SecondArgIsLastNamedArgument = PV == LastArg; 1094 } 1095 } 1096 1097 if (!SecondArgIsLastNamedArgument) 1098 Diag(TheCall->getArg(1)->getLocStart(), 1099 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1100 return false; 1101} 1102 1103/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1104/// friends. This is declared to take (...), so we have to check everything. 1105bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1106 if (TheCall->getNumArgs() < 2) 1107 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1108 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1109 if (TheCall->getNumArgs() > 2) 1110 return Diag(TheCall->getArg(2)->getLocStart(), 1111 diag::err_typecheck_call_too_many_args) 1112 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1113 << SourceRange(TheCall->getArg(2)->getLocStart(), 1114 (*(TheCall->arg_end()-1))->getLocEnd()); 1115 1116 ExprResult OrigArg0 = TheCall->getArg(0); 1117 ExprResult OrigArg1 = TheCall->getArg(1); 1118 1119 // Do standard promotions between the two arguments, returning their common 1120 // type. 1121 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1122 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1123 return true; 1124 1125 // Make sure any conversions are pushed back into the call; this is 1126 // type safe since unordered compare builtins are declared as "_Bool 1127 // foo(...)". 1128 TheCall->setArg(0, OrigArg0.get()); 1129 TheCall->setArg(1, OrigArg1.get()); 1130 1131 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1132 return false; 1133 1134 // If the common type isn't a real floating type, then the arguments were 1135 // invalid for this operation. 1136 if (!Res->isRealFloatingType()) 1137 return Diag(OrigArg0.get()->getLocStart(), 1138 diag::err_typecheck_call_invalid_ordered_compare) 1139 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1140 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1141 1142 return false; 1143} 1144 1145/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1146/// __builtin_isnan and friends. This is declared to take (...), so we have 1147/// to check everything. We expect the last argument to be a floating point 1148/// value. 1149bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1150 if (TheCall->getNumArgs() < NumArgs) 1151 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1152 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1153 if (TheCall->getNumArgs() > NumArgs) 1154 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1155 diag::err_typecheck_call_too_many_args) 1156 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1157 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1158 (*(TheCall->arg_end()-1))->getLocEnd()); 1159 1160 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1161 1162 if (OrigArg->isTypeDependent()) 1163 return false; 1164 1165 // This operation requires a non-_Complex floating-point number. 1166 if (!OrigArg->getType()->isRealFloatingType()) 1167 return Diag(OrigArg->getLocStart(), 1168 diag::err_typecheck_call_invalid_unary_fp) 1169 << OrigArg->getType() << OrigArg->getSourceRange(); 1170 1171 // If this is an implicit conversion from float -> double, remove it. 1172 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1173 Expr *CastArg = Cast->getSubExpr(); 1174 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1175 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1176 "promotion from float to double is the only expected cast here"); 1177 Cast->setSubExpr(0); 1178 TheCall->setArg(NumArgs-1, CastArg); 1179 OrigArg = CastArg; 1180 } 1181 } 1182 1183 return false; 1184} 1185 1186/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1187// This is declared to take (...), so we have to check everything. 1188ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1189 if (TheCall->getNumArgs() < 2) 1190 return ExprError(Diag(TheCall->getLocEnd(), 1191 diag::err_typecheck_call_too_few_args_at_least) 1192 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1193 << TheCall->getSourceRange()); 1194 1195 // Determine which of the following types of shufflevector we're checking: 1196 // 1) unary, vector mask: (lhs, mask) 1197 // 2) binary, vector mask: (lhs, rhs, mask) 1198 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1199 QualType resType = TheCall->getArg(0)->getType(); 1200 unsigned numElements = 0; 1201 1202 if (!TheCall->getArg(0)->isTypeDependent() && 1203 !TheCall->getArg(1)->isTypeDependent()) { 1204 QualType LHSType = TheCall->getArg(0)->getType(); 1205 QualType RHSType = TheCall->getArg(1)->getType(); 1206 1207 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1208 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1209 << SourceRange(TheCall->getArg(0)->getLocStart(), 1210 TheCall->getArg(1)->getLocEnd()); 1211 return ExprError(); 1212 } 1213 1214 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1215 unsigned numResElements = TheCall->getNumArgs() - 2; 1216 1217 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1218 // with mask. If so, verify that RHS is an integer vector type with the 1219 // same number of elts as lhs. 1220 if (TheCall->getNumArgs() == 2) { 1221 if (!RHSType->hasIntegerRepresentation() || 1222 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1223 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1224 << SourceRange(TheCall->getArg(1)->getLocStart(), 1225 TheCall->getArg(1)->getLocEnd()); 1226 numResElements = numElements; 1227 } 1228 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1229 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1230 << SourceRange(TheCall->getArg(0)->getLocStart(), 1231 TheCall->getArg(1)->getLocEnd()); 1232 return ExprError(); 1233 } else if (numElements != numResElements) { 1234 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1235 resType = Context.getVectorType(eltType, numResElements, 1236 VectorType::GenericVector); 1237 } 1238 } 1239 1240 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1241 if (TheCall->getArg(i)->isTypeDependent() || 1242 TheCall->getArg(i)->isValueDependent()) 1243 continue; 1244 1245 llvm::APSInt Result(32); 1246 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1247 return ExprError(Diag(TheCall->getLocStart(), 1248 diag::err_shufflevector_nonconstant_argument) 1249 << TheCall->getArg(i)->getSourceRange()); 1250 1251 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1252 return ExprError(Diag(TheCall->getLocStart(), 1253 diag::err_shufflevector_argument_too_large) 1254 << TheCall->getArg(i)->getSourceRange()); 1255 } 1256 1257 SmallVector<Expr*, 32> exprs; 1258 1259 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1260 exprs.push_back(TheCall->getArg(i)); 1261 TheCall->setArg(i, 0); 1262 } 1263 1264 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 1265 exprs.size(), resType, 1266 TheCall->getCallee()->getLocStart(), 1267 TheCall->getRParenLoc())); 1268} 1269 1270/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1271// This is declared to take (const void*, ...) and can take two 1272// optional constant int args. 1273bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1274 unsigned NumArgs = TheCall->getNumArgs(); 1275 1276 if (NumArgs > 3) 1277 return Diag(TheCall->getLocEnd(), 1278 diag::err_typecheck_call_too_many_args_at_most) 1279 << 0 /*function call*/ << 3 << NumArgs 1280 << TheCall->getSourceRange(); 1281 1282 // Argument 0 is checked for us and the remaining arguments must be 1283 // constant integers. 1284 for (unsigned i = 1; i != NumArgs; ++i) { 1285 Expr *Arg = TheCall->getArg(i); 1286 1287 llvm::APSInt Result; 1288 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1289 return true; 1290 1291 // FIXME: gcc issues a warning and rewrites these to 0. These 1292 // seems especially odd for the third argument since the default 1293 // is 3. 1294 if (i == 1) { 1295 if (Result.getLimitedValue() > 1) 1296 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1297 << "0" << "1" << Arg->getSourceRange(); 1298 } else { 1299 if (Result.getLimitedValue() > 3) 1300 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1301 << "0" << "3" << Arg->getSourceRange(); 1302 } 1303 } 1304 1305 return false; 1306} 1307 1308/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1309/// TheCall is a constant expression. 1310bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1311 llvm::APSInt &Result) { 1312 Expr *Arg = TheCall->getArg(ArgNum); 1313 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1314 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1315 1316 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1317 1318 if (!Arg->isIntegerConstantExpr(Result, Context)) 1319 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1320 << FDecl->getDeclName() << Arg->getSourceRange(); 1321 1322 return false; 1323} 1324 1325/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1326/// int type). This simply type checks that type is one of the defined 1327/// constants (0-3). 1328// For compatibility check 0-3, llvm only handles 0 and 2. 1329bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1330 llvm::APSInt Result; 1331 1332 // Check constant-ness first. 1333 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1334 return true; 1335 1336 Expr *Arg = TheCall->getArg(1); 1337 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1338 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1339 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1340 } 1341 1342 return false; 1343} 1344 1345/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1346/// This checks that val is a constant 1. 1347bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1348 Expr *Arg = TheCall->getArg(1); 1349 llvm::APSInt Result; 1350 1351 // TODO: This is less than ideal. Overload this to take a value. 1352 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1353 return true; 1354 1355 if (Result != 1) 1356 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1357 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1358 1359 return false; 1360} 1361 1362// Handle i > 1 ? "x" : "y", recursively. 1363bool Sema::SemaCheckStringLiteral(const Expr *E, Expr **Args, 1364 unsigned NumArgs, bool HasVAListArg, 1365 unsigned format_idx, unsigned firstDataArg, 1366 FormatStringType Type, bool inFunctionCall) { 1367 tryAgain: 1368 if (E->isTypeDependent() || E->isValueDependent()) 1369 return false; 1370 1371 E = E->IgnoreParenCasts(); 1372 1373 if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull)) 1374 // Technically -Wformat-nonliteral does not warn about this case. 1375 // The behavior of printf and friends in this case is implementation 1376 // dependent. Ideally if the format string cannot be null then 1377 // it should have a 'nonnull' attribute in the function prototype. 1378 return true; 1379 1380 switch (E->getStmtClass()) { 1381 case Stmt::BinaryConditionalOperatorClass: 1382 case Stmt::ConditionalOperatorClass: { 1383 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 1384 return SemaCheckStringLiteral(C->getTrueExpr(), Args, NumArgs, HasVAListArg, 1385 format_idx, firstDataArg, Type, 1386 inFunctionCall) 1387 && SemaCheckStringLiteral(C->getFalseExpr(), Args, NumArgs, HasVAListArg, 1388 format_idx, firstDataArg, Type, 1389 inFunctionCall); 1390 } 1391 1392 case Stmt::ImplicitCastExprClass: { 1393 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1394 goto tryAgain; 1395 } 1396 1397 case Stmt::OpaqueValueExprClass: 1398 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1399 E = src; 1400 goto tryAgain; 1401 } 1402 return false; 1403 1404 case Stmt::PredefinedExprClass: 1405 // While __func__, etc., are technically not string literals, they 1406 // cannot contain format specifiers and thus are not a security 1407 // liability. 1408 return true; 1409 1410 case Stmt::DeclRefExprClass: { 1411 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1412 1413 // As an exception, do not flag errors for variables binding to 1414 // const string literals. 1415 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1416 bool isConstant = false; 1417 QualType T = DR->getType(); 1418 1419 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1420 isConstant = AT->getElementType().isConstant(Context); 1421 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1422 isConstant = T.isConstant(Context) && 1423 PT->getPointeeType().isConstant(Context); 1424 } else if (T->isObjCObjectPointerType()) { 1425 // In ObjC, there is usually no "const ObjectPointer" type, 1426 // so don't check if the pointee type is constant. 1427 isConstant = T.isConstant(Context); 1428 } 1429 1430 if (isConstant) { 1431 if (const Expr *Init = VD->getAnyInitializer()) 1432 return SemaCheckStringLiteral(Init, Args, NumArgs, 1433 HasVAListArg, format_idx, firstDataArg, 1434 Type, /*inFunctionCall*/false); 1435 } 1436 1437 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1438 // special check to see if the format string is a function parameter 1439 // of the function calling the printf function. If the function 1440 // has an attribute indicating it is a printf-like function, then we 1441 // should suppress warnings concerning non-literals being used in a call 1442 // to a vprintf function. For example: 1443 // 1444 // void 1445 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1446 // va_list ap; 1447 // va_start(ap, fmt); 1448 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1449 // ... 1450 // 1451 if (HasVAListArg) { 1452 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 1453 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 1454 int PVIndex = PV->getFunctionScopeIndex() + 1; 1455 for (specific_attr_iterator<FormatAttr> 1456 i = ND->specific_attr_begin<FormatAttr>(), 1457 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 1458 FormatAttr *PVFormat = *i; 1459 // adjust for implicit parameter 1460 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1461 if (MD->isInstance()) 1462 ++PVIndex; 1463 // We also check if the formats are compatible. 1464 // We can't pass a 'scanf' string to a 'printf' function. 1465 if (PVIndex == PVFormat->getFormatIdx() && 1466 Type == GetFormatStringType(PVFormat)) 1467 return true; 1468 } 1469 } 1470 } 1471 } 1472 } 1473 1474 return false; 1475 } 1476 1477 case Stmt::CallExprClass: 1478 case Stmt::CXXMemberCallExprClass: { 1479 const CallExpr *CE = cast<CallExpr>(E); 1480 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 1481 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 1482 unsigned ArgIndex = FA->getFormatIdx(); 1483 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1484 if (MD->isInstance()) 1485 --ArgIndex; 1486 const Expr *Arg = CE->getArg(ArgIndex - 1); 1487 1488 return SemaCheckStringLiteral(Arg, Args, NumArgs, HasVAListArg, 1489 format_idx, firstDataArg, Type, 1490 inFunctionCall); 1491 } 1492 } 1493 1494 return false; 1495 } 1496 case Stmt::ObjCStringLiteralClass: 1497 case Stmt::StringLiteralClass: { 1498 const StringLiteral *StrE = NULL; 1499 1500 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1501 StrE = ObjCFExpr->getString(); 1502 else 1503 StrE = cast<StringLiteral>(E); 1504 1505 if (StrE) { 1506 CheckFormatString(StrE, E, Args, NumArgs, HasVAListArg, format_idx, 1507 firstDataArg, Type, inFunctionCall); 1508 return true; 1509 } 1510 1511 return false; 1512 } 1513 1514 default: 1515 return false; 1516 } 1517} 1518 1519void 1520Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1521 const Expr * const *ExprArgs, 1522 SourceLocation CallSiteLoc) { 1523 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1524 e = NonNull->args_end(); 1525 i != e; ++i) { 1526 const Expr *ArgExpr = ExprArgs[*i]; 1527 if (ArgExpr->isNullPointerConstant(Context, 1528 Expr::NPC_ValueDependentIsNotNull)) 1529 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1530 } 1531} 1532 1533Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 1534 return llvm::StringSwitch<FormatStringType>(Format->getType()) 1535 .Case("scanf", FST_Scanf) 1536 .Cases("printf", "printf0", FST_Printf) 1537 .Cases("NSString", "CFString", FST_NSString) 1538 .Case("strftime", FST_Strftime) 1539 .Case("strfmon", FST_Strfmon) 1540 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 1541 .Default(FST_Unknown); 1542} 1543 1544/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1545/// functions) for correct use of format strings. 1546void Sema::CheckFormatArguments(const FormatAttr *Format, CallExpr *TheCall) { 1547 bool IsCXXMember = false; 1548 // The way the format attribute works in GCC, the implicit this argument 1549 // of member functions is counted. However, it doesn't appear in our own 1550 // lists, so decrement format_idx in that case. 1551 IsCXXMember = isa<CXXMemberCallExpr>(TheCall); 1552 CheckFormatArguments(Format, TheCall->getArgs(), TheCall->getNumArgs(), 1553 IsCXXMember, TheCall->getRParenLoc(), 1554 TheCall->getCallee()->getSourceRange()); 1555} 1556 1557void Sema::CheckFormatArguments(const FormatAttr *Format, Expr **Args, 1558 unsigned NumArgs, bool IsCXXMember, 1559 SourceLocation Loc, SourceRange Range) { 1560 bool HasVAListArg = Format->getFirstArg() == 0; 1561 unsigned format_idx = Format->getFormatIdx() - 1; 1562 unsigned firstDataArg = HasVAListArg ? 0 : Format->getFirstArg() - 1; 1563 if (IsCXXMember) { 1564 if (format_idx == 0) 1565 return; 1566 --format_idx; 1567 if(firstDataArg != 0) 1568 --firstDataArg; 1569 } 1570 CheckFormatArguments(Args, NumArgs, HasVAListArg, format_idx, 1571 firstDataArg, GetFormatStringType(Format), Loc, Range); 1572} 1573 1574void Sema::CheckFormatArguments(Expr **Args, unsigned NumArgs, 1575 bool HasVAListArg, unsigned format_idx, 1576 unsigned firstDataArg, FormatStringType Type, 1577 SourceLocation Loc, SourceRange Range) { 1578 // CHECK: printf/scanf-like function is called with no format string. 1579 if (format_idx >= NumArgs) { 1580 Diag(Loc, diag::warn_missing_format_string) << Range; 1581 return; 1582 } 1583 1584 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1585 1586 // CHECK: format string is not a string literal. 1587 // 1588 // Dynamically generated format strings are difficult to 1589 // automatically vet at compile time. Requiring that format strings 1590 // are string literals: (1) permits the checking of format strings by 1591 // the compiler and thereby (2) can practically remove the source of 1592 // many format string exploits. 1593 1594 // Format string can be either ObjC string (e.g. @"%d") or 1595 // C string (e.g. "%d") 1596 // ObjC string uses the same format specifiers as C string, so we can use 1597 // the same format string checking logic for both ObjC and C strings. 1598 if (SemaCheckStringLiteral(OrigFormatExpr, Args, NumArgs, HasVAListArg, 1599 format_idx, firstDataArg, Type)) 1600 return; // Literal format string found, check done! 1601 1602 // Strftime is particular as it always uses a single 'time' argument, 1603 // so it is safe to pass a non-literal string. 1604 if (Type == FST_Strftime) 1605 return; 1606 1607 // Do not emit diag when the string param is a macro expansion and the 1608 // format is either NSString or CFString. This is a hack to prevent 1609 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 1610 // which are usually used in place of NS and CF string literals. 1611 if (Type == FST_NSString && Args[format_idx]->getLocStart().isMacroID()) 1612 return; 1613 1614 // If there are no arguments specified, warn with -Wformat-security, otherwise 1615 // warn only with -Wformat-nonliteral. 1616 if (NumArgs == format_idx+1) 1617 Diag(Args[format_idx]->getLocStart(), 1618 diag::warn_format_nonliteral_noargs) 1619 << OrigFormatExpr->getSourceRange(); 1620 else 1621 Diag(Args[format_idx]->getLocStart(), 1622 diag::warn_format_nonliteral) 1623 << OrigFormatExpr->getSourceRange(); 1624} 1625 1626namespace { 1627class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1628protected: 1629 Sema &S; 1630 const StringLiteral *FExpr; 1631 const Expr *OrigFormatExpr; 1632 const unsigned FirstDataArg; 1633 const unsigned NumDataArgs; 1634 const bool IsObjCLiteral; 1635 const char *Beg; // Start of format string. 1636 const bool HasVAListArg; 1637 const Expr * const *Args; 1638 const unsigned NumArgs; 1639 unsigned FormatIdx; 1640 llvm::BitVector CoveredArgs; 1641 bool usesPositionalArgs; 1642 bool atFirstArg; 1643 bool inFunctionCall; 1644public: 1645 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1646 const Expr *origFormatExpr, unsigned firstDataArg, 1647 unsigned numDataArgs, bool isObjCLiteral, 1648 const char *beg, bool hasVAListArg, 1649 Expr **args, unsigned numArgs, 1650 unsigned formatIdx, bool inFunctionCall) 1651 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1652 FirstDataArg(firstDataArg), 1653 NumDataArgs(numDataArgs), 1654 IsObjCLiteral(isObjCLiteral), Beg(beg), 1655 HasVAListArg(hasVAListArg), 1656 Args(args), NumArgs(numArgs), FormatIdx(formatIdx), 1657 usesPositionalArgs(false), atFirstArg(true), 1658 inFunctionCall(inFunctionCall) { 1659 CoveredArgs.resize(numDataArgs); 1660 CoveredArgs.reset(); 1661 } 1662 1663 void DoneProcessing(); 1664 1665 void HandleIncompleteSpecifier(const char *startSpecifier, 1666 unsigned specifierLen); 1667 1668 void HandleNonStandardLengthModifier( 1669 const analyze_format_string::LengthModifier &LM, 1670 const char *startSpecifier, unsigned specifierLen); 1671 1672 void HandleNonStandardConversionSpecifier( 1673 const analyze_format_string::ConversionSpecifier &CS, 1674 const char *startSpecifier, unsigned specifierLen); 1675 1676 void HandleNonStandardConversionSpecification( 1677 const analyze_format_string::LengthModifier &LM, 1678 const analyze_format_string::ConversionSpecifier &CS, 1679 const char *startSpecifier, unsigned specifierLen); 1680 1681 virtual void HandlePosition(const char *startPos, unsigned posLen); 1682 1683 virtual void HandleInvalidPosition(const char *startSpecifier, 1684 unsigned specifierLen, 1685 analyze_format_string::PositionContext p); 1686 1687 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1688 1689 void HandleNullChar(const char *nullCharacter); 1690 1691 template <typename Range> 1692 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 1693 const Expr *ArgumentExpr, 1694 PartialDiagnostic PDiag, 1695 SourceLocation StringLoc, 1696 bool IsStringLocation, Range StringRange, 1697 FixItHint Fixit = FixItHint()); 1698 1699protected: 1700 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1701 const char *startSpec, 1702 unsigned specifierLen, 1703 const char *csStart, unsigned csLen); 1704 1705 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 1706 const char *startSpec, 1707 unsigned specifierLen); 1708 1709 SourceRange getFormatStringRange(); 1710 CharSourceRange getSpecifierRange(const char *startSpecifier, 1711 unsigned specifierLen); 1712 SourceLocation getLocationOfByte(const char *x); 1713 1714 const Expr *getDataArg(unsigned i) const; 1715 1716 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1717 const analyze_format_string::ConversionSpecifier &CS, 1718 const char *startSpecifier, unsigned specifierLen, 1719 unsigned argIndex); 1720 1721 template <typename Range> 1722 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 1723 bool IsStringLocation, Range StringRange, 1724 FixItHint Fixit = FixItHint()); 1725 1726 void CheckPositionalAndNonpositionalArgs( 1727 const analyze_format_string::FormatSpecifier *FS); 1728}; 1729} 1730 1731SourceRange CheckFormatHandler::getFormatStringRange() { 1732 return OrigFormatExpr->getSourceRange(); 1733} 1734 1735CharSourceRange CheckFormatHandler:: 1736getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1737 SourceLocation Start = getLocationOfByte(startSpecifier); 1738 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1739 1740 // Advance the end SourceLocation by one due to half-open ranges. 1741 End = End.getLocWithOffset(1); 1742 1743 return CharSourceRange::getCharRange(Start, End); 1744} 1745 1746SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1747 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1748} 1749 1750void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1751 unsigned specifierLen){ 1752 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 1753 getLocationOfByte(startSpecifier), 1754 /*IsStringLocation*/true, 1755 getSpecifierRange(startSpecifier, specifierLen)); 1756} 1757 1758void CheckFormatHandler::HandleNonStandardLengthModifier( 1759 const analyze_format_string::LengthModifier &LM, 1760 const char *startSpecifier, unsigned specifierLen) { 1761 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) << LM.toString() 1762 << 0, 1763 getLocationOfByte(LM.getStart()), 1764 /*IsStringLocation*/true, 1765 getSpecifierRange(startSpecifier, specifierLen)); 1766} 1767 1768void CheckFormatHandler::HandleNonStandardConversionSpecifier( 1769 const analyze_format_string::ConversionSpecifier &CS, 1770 const char *startSpecifier, unsigned specifierLen) { 1771 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) << CS.toString() 1772 << 1, 1773 getLocationOfByte(CS.getStart()), 1774 /*IsStringLocation*/true, 1775 getSpecifierRange(startSpecifier, specifierLen)); 1776} 1777 1778void CheckFormatHandler::HandleNonStandardConversionSpecification( 1779 const analyze_format_string::LengthModifier &LM, 1780 const analyze_format_string::ConversionSpecifier &CS, 1781 const char *startSpecifier, unsigned specifierLen) { 1782 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_conversion_spec) 1783 << LM.toString() << CS.toString(), 1784 getLocationOfByte(LM.getStart()), 1785 /*IsStringLocation*/true, 1786 getSpecifierRange(startSpecifier, specifierLen)); 1787} 1788 1789void CheckFormatHandler::HandlePosition(const char *startPos, 1790 unsigned posLen) { 1791 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 1792 getLocationOfByte(startPos), 1793 /*IsStringLocation*/true, 1794 getSpecifierRange(startPos, posLen)); 1795} 1796 1797void 1798CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1799 analyze_format_string::PositionContext p) { 1800 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 1801 << (unsigned) p, 1802 getLocationOfByte(startPos), /*IsStringLocation*/true, 1803 getSpecifierRange(startPos, posLen)); 1804} 1805 1806void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1807 unsigned posLen) { 1808 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 1809 getLocationOfByte(startPos), 1810 /*IsStringLocation*/true, 1811 getSpecifierRange(startPos, posLen)); 1812} 1813 1814void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1815 if (!IsObjCLiteral) { 1816 // The presence of a null character is likely an error. 1817 EmitFormatDiagnostic( 1818 S.PDiag(diag::warn_printf_format_string_contains_null_char), 1819 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 1820 getFormatStringRange()); 1821 } 1822} 1823 1824const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1825 return Args[FirstDataArg + i]; 1826} 1827 1828void CheckFormatHandler::DoneProcessing() { 1829 // Does the number of data arguments exceed the number of 1830 // format conversions in the format string? 1831 if (!HasVAListArg) { 1832 // Find any arguments that weren't covered. 1833 CoveredArgs.flip(); 1834 signed notCoveredArg = CoveredArgs.find_first(); 1835 if (notCoveredArg >= 0) { 1836 assert((unsigned)notCoveredArg < NumDataArgs); 1837 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 1838 getDataArg((unsigned) notCoveredArg)->getLocStart(), 1839 /*IsStringLocation*/false, getFormatStringRange()); 1840 } 1841 } 1842} 1843 1844bool 1845CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1846 SourceLocation Loc, 1847 const char *startSpec, 1848 unsigned specifierLen, 1849 const char *csStart, 1850 unsigned csLen) { 1851 1852 bool keepGoing = true; 1853 if (argIndex < NumDataArgs) { 1854 // Consider the argument coverered, even though the specifier doesn't 1855 // make sense. 1856 CoveredArgs.set(argIndex); 1857 } 1858 else { 1859 // If argIndex exceeds the number of data arguments we 1860 // don't issue a warning because that is just a cascade of warnings (and 1861 // they may have intended '%%' anyway). We don't want to continue processing 1862 // the format string after this point, however, as we will like just get 1863 // gibberish when trying to match arguments. 1864 keepGoing = false; 1865 } 1866 1867 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 1868 << StringRef(csStart, csLen), 1869 Loc, /*IsStringLocation*/true, 1870 getSpecifierRange(startSpec, specifierLen)); 1871 1872 return keepGoing; 1873} 1874 1875void 1876CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 1877 const char *startSpec, 1878 unsigned specifierLen) { 1879 EmitFormatDiagnostic( 1880 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 1881 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 1882} 1883 1884bool 1885CheckFormatHandler::CheckNumArgs( 1886 const analyze_format_string::FormatSpecifier &FS, 1887 const analyze_format_string::ConversionSpecifier &CS, 1888 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1889 1890 if (argIndex >= NumDataArgs) { 1891 PartialDiagnostic PDiag = FS.usesPositionalArg() 1892 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 1893 << (argIndex+1) << NumDataArgs) 1894 : S.PDiag(diag::warn_printf_insufficient_data_args); 1895 EmitFormatDiagnostic( 1896 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 1897 getSpecifierRange(startSpecifier, specifierLen)); 1898 return false; 1899 } 1900 return true; 1901} 1902 1903template<typename Range> 1904void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 1905 SourceLocation Loc, 1906 bool IsStringLocation, 1907 Range StringRange, 1908 FixItHint FixIt) { 1909 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 1910 Loc, IsStringLocation, StringRange, FixIt); 1911} 1912 1913/// \brief If the format string is not within the funcion call, emit a note 1914/// so that the function call and string are in diagnostic messages. 1915/// 1916/// \param inFunctionCall if true, the format string is within the function 1917/// call and only one diagnostic message will be produced. Otherwise, an 1918/// extra note will be emitted pointing to location of the format string. 1919/// 1920/// \param ArgumentExpr the expression that is passed as the format string 1921/// argument in the function call. Used for getting locations when two 1922/// diagnostics are emitted. 1923/// 1924/// \param PDiag the callee should already have provided any strings for the 1925/// diagnostic message. This function only adds locations and fixits 1926/// to diagnostics. 1927/// 1928/// \param Loc primary location for diagnostic. If two diagnostics are 1929/// required, one will be at Loc and a new SourceLocation will be created for 1930/// the other one. 1931/// 1932/// \param IsStringLocation if true, Loc points to the format string should be 1933/// used for the note. Otherwise, Loc points to the argument list and will 1934/// be used with PDiag. 1935/// 1936/// \param StringRange some or all of the string to highlight. This is 1937/// templated so it can accept either a CharSourceRange or a SourceRange. 1938/// 1939/// \param Fixit optional fix it hint for the format string. 1940template<typename Range> 1941void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 1942 const Expr *ArgumentExpr, 1943 PartialDiagnostic PDiag, 1944 SourceLocation Loc, 1945 bool IsStringLocation, 1946 Range StringRange, 1947 FixItHint FixIt) { 1948 if (InFunctionCall) 1949 S.Diag(Loc, PDiag) << StringRange << FixIt; 1950 else { 1951 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 1952 << ArgumentExpr->getSourceRange(); 1953 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 1954 diag::note_format_string_defined) 1955 << StringRange << FixIt; 1956 } 1957} 1958 1959//===--- CHECK: Printf format string checking ------------------------------===// 1960 1961namespace { 1962class CheckPrintfHandler : public CheckFormatHandler { 1963public: 1964 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1965 const Expr *origFormatExpr, unsigned firstDataArg, 1966 unsigned numDataArgs, bool isObjCLiteral, 1967 const char *beg, bool hasVAListArg, 1968 Expr **Args, unsigned NumArgs, 1969 unsigned formatIdx, bool inFunctionCall) 1970 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1971 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1972 Args, NumArgs, formatIdx, inFunctionCall) {} 1973 1974 1975 bool HandleInvalidPrintfConversionSpecifier( 1976 const analyze_printf::PrintfSpecifier &FS, 1977 const char *startSpecifier, 1978 unsigned specifierLen); 1979 1980 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1981 const char *startSpecifier, 1982 unsigned specifierLen); 1983 1984 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1985 const char *startSpecifier, unsigned specifierLen); 1986 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1987 const analyze_printf::OptionalAmount &Amt, 1988 unsigned type, 1989 const char *startSpecifier, unsigned specifierLen); 1990 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1991 const analyze_printf::OptionalFlag &flag, 1992 const char *startSpecifier, unsigned specifierLen); 1993 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1994 const analyze_printf::OptionalFlag &ignoredFlag, 1995 const analyze_printf::OptionalFlag &flag, 1996 const char *startSpecifier, unsigned specifierLen); 1997}; 1998} 1999 2000bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2001 const analyze_printf::PrintfSpecifier &FS, 2002 const char *startSpecifier, 2003 unsigned specifierLen) { 2004 const analyze_printf::PrintfConversionSpecifier &CS = 2005 FS.getConversionSpecifier(); 2006 2007 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2008 getLocationOfByte(CS.getStart()), 2009 startSpecifier, specifierLen, 2010 CS.getStart(), CS.getLength()); 2011} 2012 2013bool CheckPrintfHandler::HandleAmount( 2014 const analyze_format_string::OptionalAmount &Amt, 2015 unsigned k, const char *startSpecifier, 2016 unsigned specifierLen) { 2017 2018 if (Amt.hasDataArgument()) { 2019 if (!HasVAListArg) { 2020 unsigned argIndex = Amt.getArgIndex(); 2021 if (argIndex >= NumDataArgs) { 2022 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2023 << k, 2024 getLocationOfByte(Amt.getStart()), 2025 /*IsStringLocation*/true, 2026 getSpecifierRange(startSpecifier, specifierLen)); 2027 // Don't do any more checking. We will just emit 2028 // spurious errors. 2029 return false; 2030 } 2031 2032 // Type check the data argument. It should be an 'int'. 2033 // Although not in conformance with C99, we also allow the argument to be 2034 // an 'unsigned int' as that is a reasonably safe case. GCC also 2035 // doesn't emit a warning for that case. 2036 CoveredArgs.set(argIndex); 2037 const Expr *Arg = getDataArg(argIndex); 2038 QualType T = Arg->getType(); 2039 2040 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 2041 assert(ATR.isValid()); 2042 2043 if (!ATR.matchesType(S.Context, T)) { 2044 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2045 << k << ATR.getRepresentativeTypeName(S.Context) 2046 << T << Arg->getSourceRange(), 2047 getLocationOfByte(Amt.getStart()), 2048 /*IsStringLocation*/true, 2049 getSpecifierRange(startSpecifier, specifierLen)); 2050 // Don't do any more checking. We will just emit 2051 // spurious errors. 2052 return false; 2053 } 2054 } 2055 } 2056 return true; 2057} 2058 2059void CheckPrintfHandler::HandleInvalidAmount( 2060 const analyze_printf::PrintfSpecifier &FS, 2061 const analyze_printf::OptionalAmount &Amt, 2062 unsigned type, 2063 const char *startSpecifier, 2064 unsigned specifierLen) { 2065 const analyze_printf::PrintfConversionSpecifier &CS = 2066 FS.getConversionSpecifier(); 2067 2068 FixItHint fixit = 2069 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2070 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2071 Amt.getConstantLength())) 2072 : FixItHint(); 2073 2074 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2075 << type << CS.toString(), 2076 getLocationOfByte(Amt.getStart()), 2077 /*IsStringLocation*/true, 2078 getSpecifierRange(startSpecifier, specifierLen), 2079 fixit); 2080} 2081 2082void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2083 const analyze_printf::OptionalFlag &flag, 2084 const char *startSpecifier, 2085 unsigned specifierLen) { 2086 // Warn about pointless flag with a fixit removal. 2087 const analyze_printf::PrintfConversionSpecifier &CS = 2088 FS.getConversionSpecifier(); 2089 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2090 << flag.toString() << CS.toString(), 2091 getLocationOfByte(flag.getPosition()), 2092 /*IsStringLocation*/true, 2093 getSpecifierRange(startSpecifier, specifierLen), 2094 FixItHint::CreateRemoval( 2095 getSpecifierRange(flag.getPosition(), 1))); 2096} 2097 2098void CheckPrintfHandler::HandleIgnoredFlag( 2099 const analyze_printf::PrintfSpecifier &FS, 2100 const analyze_printf::OptionalFlag &ignoredFlag, 2101 const analyze_printf::OptionalFlag &flag, 2102 const char *startSpecifier, 2103 unsigned specifierLen) { 2104 // Warn about ignored flag with a fixit removal. 2105 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2106 << ignoredFlag.toString() << flag.toString(), 2107 getLocationOfByte(ignoredFlag.getPosition()), 2108 /*IsStringLocation*/true, 2109 getSpecifierRange(startSpecifier, specifierLen), 2110 FixItHint::CreateRemoval( 2111 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2112} 2113 2114bool 2115CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2116 &FS, 2117 const char *startSpecifier, 2118 unsigned specifierLen) { 2119 2120 using namespace analyze_format_string; 2121 using namespace analyze_printf; 2122 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2123 2124 if (FS.consumesDataArgument()) { 2125 if (atFirstArg) { 2126 atFirstArg = false; 2127 usesPositionalArgs = FS.usesPositionalArg(); 2128 } 2129 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2130 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2131 startSpecifier, specifierLen); 2132 return false; 2133 } 2134 } 2135 2136 // First check if the field width, precision, and conversion specifier 2137 // have matching data arguments. 2138 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2139 startSpecifier, specifierLen)) { 2140 return false; 2141 } 2142 2143 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2144 startSpecifier, specifierLen)) { 2145 return false; 2146 } 2147 2148 if (!CS.consumesDataArgument()) { 2149 // FIXME: Technically specifying a precision or field width here 2150 // makes no sense. Worth issuing a warning at some point. 2151 return true; 2152 } 2153 2154 // Consume the argument. 2155 unsigned argIndex = FS.getArgIndex(); 2156 if (argIndex < NumDataArgs) { 2157 // The check to see if the argIndex is valid will come later. 2158 // We set the bit here because we may exit early from this 2159 // function if we encounter some other error. 2160 CoveredArgs.set(argIndex); 2161 } 2162 2163 // Check for using an Objective-C specific conversion specifier 2164 // in a non-ObjC literal. 2165 if (!IsObjCLiteral && CS.isObjCArg()) { 2166 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2167 specifierLen); 2168 } 2169 2170 // Check for invalid use of field width 2171 if (!FS.hasValidFieldWidth()) { 2172 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2173 startSpecifier, specifierLen); 2174 } 2175 2176 // Check for invalid use of precision 2177 if (!FS.hasValidPrecision()) { 2178 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2179 startSpecifier, specifierLen); 2180 } 2181 2182 // Check each flag does not conflict with any other component. 2183 if (!FS.hasValidThousandsGroupingPrefix()) 2184 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2185 if (!FS.hasValidLeadingZeros()) 2186 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2187 if (!FS.hasValidPlusPrefix()) 2188 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2189 if (!FS.hasValidSpacePrefix()) 2190 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2191 if (!FS.hasValidAlternativeForm()) 2192 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2193 if (!FS.hasValidLeftJustified()) 2194 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2195 2196 // Check that flags are not ignored by another flag 2197 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2198 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2199 startSpecifier, specifierLen); 2200 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2201 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2202 startSpecifier, specifierLen); 2203 2204 // Check the length modifier is valid with the given conversion specifier. 2205 const LengthModifier &LM = FS.getLengthModifier(); 2206 if (!FS.hasValidLengthModifier()) 2207 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2208 << LM.toString() << CS.toString(), 2209 getLocationOfByte(LM.getStart()), 2210 /*IsStringLocation*/true, 2211 getSpecifierRange(startSpecifier, specifierLen), 2212 FixItHint::CreateRemoval( 2213 getSpecifierRange(LM.getStart(), 2214 LM.getLength()))); 2215 if (!FS.hasStandardLengthModifier()) 2216 HandleNonStandardLengthModifier(LM, startSpecifier, specifierLen); 2217 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2218 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2219 if (!FS.hasStandardLengthConversionCombination()) 2220 HandleNonStandardConversionSpecification(LM, CS, startSpecifier, 2221 specifierLen); 2222 2223 // Are we using '%n'? 2224 if (CS.getKind() == ConversionSpecifier::nArg) { 2225 // Issue a warning about this being a possible security issue. 2226 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_write_back), 2227 getLocationOfByte(CS.getStart()), 2228 /*IsStringLocation*/true, 2229 getSpecifierRange(startSpecifier, specifierLen)); 2230 // Continue checking the other format specifiers. 2231 return true; 2232 } 2233 2234 // The remaining checks depend on the data arguments. 2235 if (HasVAListArg) 2236 return true; 2237 2238 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2239 return false; 2240 2241 // Now type check the data expression that matches the 2242 // format specifier. 2243 const Expr *Ex = getDataArg(argIndex); 2244 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context, 2245 IsObjCLiteral); 2246 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2247 // Check if we didn't match because of an implicit cast from a 'char' 2248 // or 'short' to an 'int'. This is done because printf is a varargs 2249 // function. 2250 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 2251 if (ICE->getType() == S.Context.IntTy) { 2252 // All further checking is done on the subexpression. 2253 Ex = ICE->getSubExpr(); 2254 if (ATR.matchesType(S.Context, Ex->getType())) 2255 return true; 2256 } 2257 2258 // We may be able to offer a FixItHint if it is a supported type. 2259 PrintfSpecifier fixedFS = FS; 2260 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 2261 S.Context, IsObjCLiteral); 2262 2263 if (success) { 2264 // Get the fix string from the fixed format specifier 2265 SmallString<128> buf; 2266 llvm::raw_svector_ostream os(buf); 2267 fixedFS.toString(os); 2268 2269 EmitFormatDiagnostic( 2270 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2271 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2272 << Ex->getSourceRange(), 2273 getLocationOfByte(CS.getStart()), 2274 /*IsStringLocation*/true, 2275 getSpecifierRange(startSpecifier, specifierLen), 2276 FixItHint::CreateReplacement( 2277 getSpecifierRange(startSpecifier, specifierLen), 2278 os.str())); 2279 } 2280 else { 2281 EmitFormatDiagnostic( 2282 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2283 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2284 << getSpecifierRange(startSpecifier, specifierLen) 2285 << Ex->getSourceRange(), 2286 getLocationOfByte(CS.getStart()), 2287 true, 2288 getSpecifierRange(startSpecifier, specifierLen)); 2289 } 2290 } 2291 2292 return true; 2293} 2294 2295//===--- CHECK: Scanf format string checking ------------------------------===// 2296 2297namespace { 2298class CheckScanfHandler : public CheckFormatHandler { 2299public: 2300 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2301 const Expr *origFormatExpr, unsigned firstDataArg, 2302 unsigned numDataArgs, bool isObjCLiteral, 2303 const char *beg, bool hasVAListArg, 2304 Expr **Args, unsigned NumArgs, 2305 unsigned formatIdx, bool inFunctionCall) 2306 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2307 numDataArgs, isObjCLiteral, beg, hasVAListArg, 2308 Args, NumArgs, formatIdx, inFunctionCall) {} 2309 2310 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2311 const char *startSpecifier, 2312 unsigned specifierLen); 2313 2314 bool HandleInvalidScanfConversionSpecifier( 2315 const analyze_scanf::ScanfSpecifier &FS, 2316 const char *startSpecifier, 2317 unsigned specifierLen); 2318 2319 void HandleIncompleteScanList(const char *start, const char *end); 2320}; 2321} 2322 2323void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2324 const char *end) { 2325 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2326 getLocationOfByte(end), /*IsStringLocation*/true, 2327 getSpecifierRange(start, end - start)); 2328} 2329 2330bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2331 const analyze_scanf::ScanfSpecifier &FS, 2332 const char *startSpecifier, 2333 unsigned specifierLen) { 2334 2335 const analyze_scanf::ScanfConversionSpecifier &CS = 2336 FS.getConversionSpecifier(); 2337 2338 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2339 getLocationOfByte(CS.getStart()), 2340 startSpecifier, specifierLen, 2341 CS.getStart(), CS.getLength()); 2342} 2343 2344bool CheckScanfHandler::HandleScanfSpecifier( 2345 const analyze_scanf::ScanfSpecifier &FS, 2346 const char *startSpecifier, 2347 unsigned specifierLen) { 2348 2349 using namespace analyze_scanf; 2350 using namespace analyze_format_string; 2351 2352 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2353 2354 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2355 // be used to decide if we are using positional arguments consistently. 2356 if (FS.consumesDataArgument()) { 2357 if (atFirstArg) { 2358 atFirstArg = false; 2359 usesPositionalArgs = FS.usesPositionalArg(); 2360 } 2361 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2362 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2363 startSpecifier, specifierLen); 2364 return false; 2365 } 2366 } 2367 2368 // Check if the field with is non-zero. 2369 const OptionalAmount &Amt = FS.getFieldWidth(); 2370 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2371 if (Amt.getConstantAmount() == 0) { 2372 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2373 Amt.getConstantLength()); 2374 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 2375 getLocationOfByte(Amt.getStart()), 2376 /*IsStringLocation*/true, R, 2377 FixItHint::CreateRemoval(R)); 2378 } 2379 } 2380 2381 if (!FS.consumesDataArgument()) { 2382 // FIXME: Technically specifying a precision or field width here 2383 // makes no sense. Worth issuing a warning at some point. 2384 return true; 2385 } 2386 2387 // Consume the argument. 2388 unsigned argIndex = FS.getArgIndex(); 2389 if (argIndex < NumDataArgs) { 2390 // The check to see if the argIndex is valid will come later. 2391 // We set the bit here because we may exit early from this 2392 // function if we encounter some other error. 2393 CoveredArgs.set(argIndex); 2394 } 2395 2396 // Check the length modifier is valid with the given conversion specifier. 2397 const LengthModifier &LM = FS.getLengthModifier(); 2398 if (!FS.hasValidLengthModifier()) { 2399 const CharSourceRange &R = getSpecifierRange(LM.getStart(), LM.getLength()); 2400 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2401 << LM.toString() << CS.toString() 2402 << getSpecifierRange(startSpecifier, specifierLen), 2403 getLocationOfByte(LM.getStart()), 2404 /*IsStringLocation*/true, R, 2405 FixItHint::CreateRemoval(R)); 2406 } 2407 2408 if (!FS.hasStandardLengthModifier()) 2409 HandleNonStandardLengthModifier(LM, startSpecifier, specifierLen); 2410 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2411 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2412 if (!FS.hasStandardLengthConversionCombination()) 2413 HandleNonStandardConversionSpecification(LM, CS, startSpecifier, 2414 specifierLen); 2415 2416 // The remaining checks depend on the data arguments. 2417 if (HasVAListArg) 2418 return true; 2419 2420 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2421 return false; 2422 2423 // Check that the argument type matches the format specifier. 2424 const Expr *Ex = getDataArg(argIndex); 2425 const analyze_scanf::ScanfArgTypeResult &ATR = FS.getArgType(S.Context); 2426 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2427 ScanfSpecifier fixedFS = FS; 2428 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 2429 S.Context); 2430 2431 if (success) { 2432 // Get the fix string from the fixed format specifier. 2433 SmallString<128> buf; 2434 llvm::raw_svector_ostream os(buf); 2435 fixedFS.toString(os); 2436 2437 EmitFormatDiagnostic( 2438 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2439 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2440 << Ex->getSourceRange(), 2441 getLocationOfByte(CS.getStart()), 2442 /*IsStringLocation*/true, 2443 getSpecifierRange(startSpecifier, specifierLen), 2444 FixItHint::CreateReplacement( 2445 getSpecifierRange(startSpecifier, specifierLen), 2446 os.str())); 2447 } else { 2448 EmitFormatDiagnostic( 2449 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2450 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2451 << Ex->getSourceRange(), 2452 getLocationOfByte(CS.getStart()), 2453 /*IsStringLocation*/true, 2454 getSpecifierRange(startSpecifier, specifierLen)); 2455 } 2456 } 2457 2458 return true; 2459} 2460 2461void Sema::CheckFormatString(const StringLiteral *FExpr, 2462 const Expr *OrigFormatExpr, 2463 Expr **Args, unsigned NumArgs, 2464 bool HasVAListArg, unsigned format_idx, 2465 unsigned firstDataArg, FormatStringType Type, 2466 bool inFunctionCall) { 2467 2468 // CHECK: is the format string a wide literal? 2469 if (!FExpr->isAscii()) { 2470 CheckFormatHandler::EmitFormatDiagnostic( 2471 *this, inFunctionCall, Args[format_idx], 2472 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 2473 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2474 return; 2475 } 2476 2477 // Str - The format string. NOTE: this is NOT null-terminated! 2478 StringRef StrRef = FExpr->getString(); 2479 const char *Str = StrRef.data(); 2480 unsigned StrLen = StrRef.size(); 2481 const unsigned numDataArgs = NumArgs - firstDataArg; 2482 2483 // CHECK: empty format string? 2484 if (StrLen == 0 && numDataArgs > 0) { 2485 CheckFormatHandler::EmitFormatDiagnostic( 2486 *this, inFunctionCall, Args[format_idx], 2487 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 2488 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2489 return; 2490 } 2491 2492 if (Type == FST_Printf || Type == FST_NSString) { 2493 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2494 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2495 Str, HasVAListArg, Args, NumArgs, format_idx, 2496 inFunctionCall); 2497 2498 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 2499 getLangOpts())) 2500 H.DoneProcessing(); 2501 } else if (Type == FST_Scanf) { 2502 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2503 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2504 Str, HasVAListArg, Args, NumArgs, format_idx, 2505 inFunctionCall); 2506 2507 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 2508 getLangOpts())) 2509 H.DoneProcessing(); 2510 } // TODO: handle other formats 2511} 2512 2513//===--- CHECK: Standard memory functions ---------------------------------===// 2514 2515/// \brief Determine whether the given type is a dynamic class type (e.g., 2516/// whether it has a vtable). 2517static bool isDynamicClassType(QualType T) { 2518 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2519 if (CXXRecordDecl *Definition = Record->getDefinition()) 2520 if (Definition->isDynamicClass()) 2521 return true; 2522 2523 return false; 2524} 2525 2526/// \brief If E is a sizeof expression, returns its argument expression, 2527/// otherwise returns NULL. 2528static const Expr *getSizeOfExprArg(const Expr* E) { 2529 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2530 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2531 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2532 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2533 2534 return 0; 2535} 2536 2537/// \brief If E is a sizeof expression, returns its argument type. 2538static QualType getSizeOfArgType(const Expr* E) { 2539 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2540 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2541 if (SizeOf->getKind() == clang::UETT_SizeOf) 2542 return SizeOf->getTypeOfArgument(); 2543 2544 return QualType(); 2545} 2546 2547/// \brief Check for dangerous or invalid arguments to memset(). 2548/// 2549/// This issues warnings on known problematic, dangerous or unspecified 2550/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2551/// function calls. 2552/// 2553/// \param Call The call expression to diagnose. 2554void Sema::CheckMemaccessArguments(const CallExpr *Call, 2555 unsigned BId, 2556 IdentifierInfo *FnName) { 2557 assert(BId != 0); 2558 2559 // It is possible to have a non-standard definition of memset. Validate 2560 // we have enough arguments, and if not, abort further checking. 2561 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 2562 if (Call->getNumArgs() < ExpectedNumArgs) 2563 return; 2564 2565 unsigned LastArg = (BId == Builtin::BImemset || 2566 BId == Builtin::BIstrndup ? 1 : 2); 2567 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 2568 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 2569 2570 // We have special checking when the length is a sizeof expression. 2571 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2572 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2573 llvm::FoldingSetNodeID SizeOfArgID; 2574 2575 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2576 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2577 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2578 2579 QualType DestTy = Dest->getType(); 2580 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2581 QualType PointeeTy = DestPtrTy->getPointeeType(); 2582 2583 // Never warn about void type pointers. This can be used to suppress 2584 // false positives. 2585 if (PointeeTy->isVoidType()) 2586 continue; 2587 2588 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2589 // actually comparing the expressions for equality. Because computing the 2590 // expression IDs can be expensive, we only do this if the diagnostic is 2591 // enabled. 2592 if (SizeOfArg && 2593 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2594 SizeOfArg->getExprLoc())) { 2595 // We only compute IDs for expressions if the warning is enabled, and 2596 // cache the sizeof arg's ID. 2597 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2598 SizeOfArg->Profile(SizeOfArgID, Context, true); 2599 llvm::FoldingSetNodeID DestID; 2600 Dest->Profile(DestID, Context, true); 2601 if (DestID == SizeOfArgID) { 2602 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 2603 // over sizeof(src) as well. 2604 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2605 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2606 if (UnaryOp->getOpcode() == UO_AddrOf) 2607 ActionIdx = 1; // If its an address-of operator, just remove it. 2608 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2609 ActionIdx = 2; // If the pointee's size is sizeof(char), 2610 // suggest an explicit length. 2611 unsigned DestSrcSelect = 2612 (BId == Builtin::BIstrndup ? 1 : ArgIdx); 2613 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 2614 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 2615 << FnName << DestSrcSelect << ActionIdx 2616 << Dest->getSourceRange() 2617 << SizeOfArg->getSourceRange()); 2618 break; 2619 } 2620 } 2621 2622 // Also check for cases where the sizeof argument is the exact same 2623 // type as the memory argument, and where it points to a user-defined 2624 // record type. 2625 if (SizeOfArgTy != QualType()) { 2626 if (PointeeTy->isRecordType() && 2627 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 2628 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 2629 PDiag(diag::warn_sizeof_pointer_type_memaccess) 2630 << FnName << SizeOfArgTy << ArgIdx 2631 << PointeeTy << Dest->getSourceRange() 2632 << LenExpr->getSourceRange()); 2633 break; 2634 } 2635 } 2636 2637 // Always complain about dynamic classes. 2638 if (isDynamicClassType(PointeeTy)) { 2639 2640 unsigned OperationType = 0; 2641 // "overwritten" if we're warning about the destination for any call 2642 // but memcmp; otherwise a verb appropriate to the call. 2643 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 2644 if (BId == Builtin::BImemcpy) 2645 OperationType = 1; 2646 else if(BId == Builtin::BImemmove) 2647 OperationType = 2; 2648 else if (BId == Builtin::BImemcmp) 2649 OperationType = 3; 2650 } 2651 2652 DiagRuntimeBehavior( 2653 Dest->getExprLoc(), Dest, 2654 PDiag(diag::warn_dyn_class_memaccess) 2655 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 2656 << FnName << PointeeTy 2657 << OperationType 2658 << Call->getCallee()->getSourceRange()); 2659 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 2660 BId != Builtin::BImemset) 2661 DiagRuntimeBehavior( 2662 Dest->getExprLoc(), Dest, 2663 PDiag(diag::warn_arc_object_memaccess) 2664 << ArgIdx << FnName << PointeeTy 2665 << Call->getCallee()->getSourceRange()); 2666 else 2667 continue; 2668 2669 DiagRuntimeBehavior( 2670 Dest->getExprLoc(), Dest, 2671 PDiag(diag::note_bad_memaccess_silence) 2672 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 2673 break; 2674 } 2675 } 2676} 2677 2678// A little helper routine: ignore addition and subtraction of integer literals. 2679// This intentionally does not ignore all integer constant expressions because 2680// we don't want to remove sizeof(). 2681static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2682 Ex = Ex->IgnoreParenCasts(); 2683 2684 for (;;) { 2685 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2686 if (!BO || !BO->isAdditiveOp()) 2687 break; 2688 2689 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2690 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2691 2692 if (isa<IntegerLiteral>(RHS)) 2693 Ex = LHS; 2694 else if (isa<IntegerLiteral>(LHS)) 2695 Ex = RHS; 2696 else 2697 break; 2698 } 2699 2700 return Ex; 2701} 2702 2703// Warn if the user has made the 'size' argument to strlcpy or strlcat 2704// be the size of the source, instead of the destination. 2705void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2706 IdentifierInfo *FnName) { 2707 2708 // Don't crash if the user has the wrong number of arguments 2709 if (Call->getNumArgs() != 3) 2710 return; 2711 2712 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2713 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2714 const Expr *CompareWithSrc = NULL; 2715 2716 // Look for 'strlcpy(dst, x, sizeof(x))' 2717 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2718 CompareWithSrc = Ex; 2719 else { 2720 // Look for 'strlcpy(dst, x, strlen(x))' 2721 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2722 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 2723 && SizeCall->getNumArgs() == 1) 2724 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2725 } 2726 } 2727 2728 if (!CompareWithSrc) 2729 return; 2730 2731 // Determine if the argument to sizeof/strlen is equal to the source 2732 // argument. In principle there's all kinds of things you could do 2733 // here, for instance creating an == expression and evaluating it with 2734 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2735 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2736 if (!SrcArgDRE) 2737 return; 2738 2739 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2740 if (!CompareWithSrcDRE || 2741 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2742 return; 2743 2744 const Expr *OriginalSizeArg = Call->getArg(2); 2745 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2746 << OriginalSizeArg->getSourceRange() << FnName; 2747 2748 // Output a FIXIT hint if the destination is an array (rather than a 2749 // pointer to an array). This could be enhanced to handle some 2750 // pointers if we know the actual size, like if DstArg is 'array+2' 2751 // we could say 'sizeof(array)-2'. 2752 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2753 QualType DstArgTy = DstArg->getType(); 2754 2755 // Only handle constant-sized or VLAs, but not flexible members. 2756 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2757 // Only issue the FIXIT for arrays of size > 1. 2758 if (CAT->getSize().getSExtValue() <= 1) 2759 return; 2760 } else if (!DstArgTy->isVariableArrayType()) { 2761 return; 2762 } 2763 2764 SmallString<128> sizeString; 2765 llvm::raw_svector_ostream OS(sizeString); 2766 OS << "sizeof("; 2767 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2768 OS << ")"; 2769 2770 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2771 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2772 OS.str()); 2773} 2774 2775/// Check if two expressions refer to the same declaration. 2776static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 2777 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 2778 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 2779 return D1->getDecl() == D2->getDecl(); 2780 return false; 2781} 2782 2783static const Expr *getStrlenExprArg(const Expr *E) { 2784 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 2785 const FunctionDecl *FD = CE->getDirectCallee(); 2786 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 2787 return 0; 2788 return CE->getArg(0)->IgnoreParenCasts(); 2789 } 2790 return 0; 2791} 2792 2793// Warn on anti-patterns as the 'size' argument to strncat. 2794// The correct size argument should look like following: 2795// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 2796void Sema::CheckStrncatArguments(const CallExpr *CE, 2797 IdentifierInfo *FnName) { 2798 // Don't crash if the user has the wrong number of arguments. 2799 if (CE->getNumArgs() < 3) 2800 return; 2801 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 2802 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 2803 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 2804 2805 // Identify common expressions, which are wrongly used as the size argument 2806 // to strncat and may lead to buffer overflows. 2807 unsigned PatternType = 0; 2808 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 2809 // - sizeof(dst) 2810 if (referToTheSameDecl(SizeOfArg, DstArg)) 2811 PatternType = 1; 2812 // - sizeof(src) 2813 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 2814 PatternType = 2; 2815 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 2816 if (BE->getOpcode() == BO_Sub) { 2817 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 2818 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 2819 // - sizeof(dst) - strlen(dst) 2820 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 2821 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 2822 PatternType = 1; 2823 // - sizeof(src) - (anything) 2824 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 2825 PatternType = 2; 2826 } 2827 } 2828 2829 if (PatternType == 0) 2830 return; 2831 2832 // Generate the diagnostic. 2833 SourceLocation SL = LenArg->getLocStart(); 2834 SourceRange SR = LenArg->getSourceRange(); 2835 SourceManager &SM = PP.getSourceManager(); 2836 2837 // If the function is defined as a builtin macro, do not show macro expansion. 2838 if (SM.isMacroArgExpansion(SL)) { 2839 SL = SM.getSpellingLoc(SL); 2840 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 2841 SM.getSpellingLoc(SR.getEnd())); 2842 } 2843 2844 if (PatternType == 1) 2845 Diag(SL, diag::warn_strncat_large_size) << SR; 2846 else 2847 Diag(SL, diag::warn_strncat_src_size) << SR; 2848 2849 // Output a FIXIT hint if the destination is an array (rather than a 2850 // pointer to an array). This could be enhanced to handle some 2851 // pointers if we know the actual size, like if DstArg is 'array+2' 2852 // we could say 'sizeof(array)-2'. 2853 QualType DstArgTy = DstArg->getType(); 2854 2855 // Only handle constant-sized or VLAs, but not flexible members. 2856 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2857 // Only issue the FIXIT for arrays of size > 1. 2858 if (CAT->getSize().getSExtValue() <= 1) 2859 return; 2860 } else if (!DstArgTy->isVariableArrayType()) { 2861 return; 2862 } 2863 2864 SmallString<128> sizeString; 2865 llvm::raw_svector_ostream OS(sizeString); 2866 OS << "sizeof("; 2867 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2868 OS << ") - "; 2869 OS << "strlen("; 2870 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2871 OS << ") - 1"; 2872 2873 Diag(SL, diag::note_strncat_wrong_size) 2874 << FixItHint::CreateReplacement(SR, OS.str()); 2875} 2876 2877//===--- CHECK: Return Address of Stack Variable --------------------------===// 2878 2879static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars); 2880static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars); 2881 2882/// CheckReturnStackAddr - Check if a return statement returns the address 2883/// of a stack variable. 2884void 2885Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 2886 SourceLocation ReturnLoc) { 2887 2888 Expr *stackE = 0; 2889 SmallVector<DeclRefExpr *, 8> refVars; 2890 2891 // Perform checking for returned stack addresses, local blocks, 2892 // label addresses or references to temporaries. 2893 if (lhsType->isPointerType() || 2894 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 2895 stackE = EvalAddr(RetValExp, refVars); 2896 } else if (lhsType->isReferenceType()) { 2897 stackE = EvalVal(RetValExp, refVars); 2898 } 2899 2900 if (stackE == 0) 2901 return; // Nothing suspicious was found. 2902 2903 SourceLocation diagLoc; 2904 SourceRange diagRange; 2905 if (refVars.empty()) { 2906 diagLoc = stackE->getLocStart(); 2907 diagRange = stackE->getSourceRange(); 2908 } else { 2909 // We followed through a reference variable. 'stackE' contains the 2910 // problematic expression but we will warn at the return statement pointing 2911 // at the reference variable. We will later display the "trail" of 2912 // reference variables using notes. 2913 diagLoc = refVars[0]->getLocStart(); 2914 diagRange = refVars[0]->getSourceRange(); 2915 } 2916 2917 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 2918 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 2919 : diag::warn_ret_stack_addr) 2920 << DR->getDecl()->getDeclName() << diagRange; 2921 } else if (isa<BlockExpr>(stackE)) { // local block. 2922 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 2923 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 2924 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 2925 } else { // local temporary. 2926 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 2927 : diag::warn_ret_local_temp_addr) 2928 << diagRange; 2929 } 2930 2931 // Display the "trail" of reference variables that we followed until we 2932 // found the problematic expression using notes. 2933 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 2934 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 2935 // If this var binds to another reference var, show the range of the next 2936 // var, otherwise the var binds to the problematic expression, in which case 2937 // show the range of the expression. 2938 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 2939 : stackE->getSourceRange(); 2940 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 2941 << VD->getDeclName() << range; 2942 } 2943} 2944 2945/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 2946/// check if the expression in a return statement evaluates to an address 2947/// to a location on the stack, a local block, an address of a label, or a 2948/// reference to local temporary. The recursion is used to traverse the 2949/// AST of the return expression, with recursion backtracking when we 2950/// encounter a subexpression that (1) clearly does not lead to one of the 2951/// above problematic expressions (2) is something we cannot determine leads to 2952/// a problematic expression based on such local checking. 2953/// 2954/// Both EvalAddr and EvalVal follow through reference variables to evaluate 2955/// the expression that they point to. Such variables are added to the 2956/// 'refVars' vector so that we know what the reference variable "trail" was. 2957/// 2958/// EvalAddr processes expressions that are pointers that are used as 2959/// references (and not L-values). EvalVal handles all other values. 2960/// At the base case of the recursion is a check for the above problematic 2961/// expressions. 2962/// 2963/// This implementation handles: 2964/// 2965/// * pointer-to-pointer casts 2966/// * implicit conversions from array references to pointers 2967/// * taking the address of fields 2968/// * arbitrary interplay between "&" and "*" operators 2969/// * pointer arithmetic from an address of a stack variable 2970/// * taking the address of an array element where the array is on the stack 2971static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2972 if (E->isTypeDependent()) 2973 return NULL; 2974 2975 // We should only be called for evaluating pointer expressions. 2976 assert((E->getType()->isAnyPointerType() || 2977 E->getType()->isBlockPointerType() || 2978 E->getType()->isObjCQualifiedIdType()) && 2979 "EvalAddr only works on pointers"); 2980 2981 E = E->IgnoreParens(); 2982 2983 // Our "symbolic interpreter" is just a dispatch off the currently 2984 // viewed AST node. We then recursively traverse the AST by calling 2985 // EvalAddr and EvalVal appropriately. 2986 switch (E->getStmtClass()) { 2987 case Stmt::DeclRefExprClass: { 2988 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2989 2990 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2991 // If this is a reference variable, follow through to the expression that 2992 // it points to. 2993 if (V->hasLocalStorage() && 2994 V->getType()->isReferenceType() && V->hasInit()) { 2995 // Add the reference variable to the "trail". 2996 refVars.push_back(DR); 2997 return EvalAddr(V->getInit(), refVars); 2998 } 2999 3000 return NULL; 3001 } 3002 3003 case Stmt::UnaryOperatorClass: { 3004 // The only unary operator that make sense to handle here 3005 // is AddrOf. All others don't make sense as pointers. 3006 UnaryOperator *U = cast<UnaryOperator>(E); 3007 3008 if (U->getOpcode() == UO_AddrOf) 3009 return EvalVal(U->getSubExpr(), refVars); 3010 else 3011 return NULL; 3012 } 3013 3014 case Stmt::BinaryOperatorClass: { 3015 // Handle pointer arithmetic. All other binary operators are not valid 3016 // in this context. 3017 BinaryOperator *B = cast<BinaryOperator>(E); 3018 BinaryOperatorKind op = B->getOpcode(); 3019 3020 if (op != BO_Add && op != BO_Sub) 3021 return NULL; 3022 3023 Expr *Base = B->getLHS(); 3024 3025 // Determine which argument is the real pointer base. It could be 3026 // the RHS argument instead of the LHS. 3027 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 3028 3029 assert (Base->getType()->isPointerType()); 3030 return EvalAddr(Base, refVars); 3031 } 3032 3033 // For conditional operators we need to see if either the LHS or RHS are 3034 // valid DeclRefExpr*s. If one of them is valid, we return it. 3035 case Stmt::ConditionalOperatorClass: { 3036 ConditionalOperator *C = cast<ConditionalOperator>(E); 3037 3038 // Handle the GNU extension for missing LHS. 3039 if (Expr *lhsExpr = C->getLHS()) { 3040 // In C++, we can have a throw-expression, which has 'void' type. 3041 if (!lhsExpr->getType()->isVoidType()) 3042 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 3043 return LHS; 3044 } 3045 3046 // In C++, we can have a throw-expression, which has 'void' type. 3047 if (C->getRHS()->getType()->isVoidType()) 3048 return NULL; 3049 3050 return EvalAddr(C->getRHS(), refVars); 3051 } 3052 3053 case Stmt::BlockExprClass: 3054 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 3055 return E; // local block. 3056 return NULL; 3057 3058 case Stmt::AddrLabelExprClass: 3059 return E; // address of label. 3060 3061 case Stmt::ExprWithCleanupsClass: 3062 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars); 3063 3064 // For casts, we need to handle conversions from arrays to 3065 // pointer values, and pointer-to-pointer conversions. 3066 case Stmt::ImplicitCastExprClass: 3067 case Stmt::CStyleCastExprClass: 3068 case Stmt::CXXFunctionalCastExprClass: 3069 case Stmt::ObjCBridgedCastExprClass: 3070 case Stmt::CXXStaticCastExprClass: 3071 case Stmt::CXXDynamicCastExprClass: 3072 case Stmt::CXXConstCastExprClass: 3073 case Stmt::CXXReinterpretCastExprClass: { 3074 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 3075 switch (cast<CastExpr>(E)->getCastKind()) { 3076 case CK_BitCast: 3077 case CK_LValueToRValue: 3078 case CK_NoOp: 3079 case CK_BaseToDerived: 3080 case CK_DerivedToBase: 3081 case CK_UncheckedDerivedToBase: 3082 case CK_Dynamic: 3083 case CK_CPointerToObjCPointerCast: 3084 case CK_BlockPointerToObjCPointerCast: 3085 case CK_AnyPointerToBlockPointerCast: 3086 return EvalAddr(SubExpr, refVars); 3087 3088 case CK_ArrayToPointerDecay: 3089 return EvalVal(SubExpr, refVars); 3090 3091 default: 3092 return 0; 3093 } 3094 } 3095 3096 case Stmt::MaterializeTemporaryExprClass: 3097 if (Expr *Result = EvalAddr( 3098 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3099 refVars)) 3100 return Result; 3101 3102 return E; 3103 3104 // Everything else: we simply don't reason about them. 3105 default: 3106 return NULL; 3107 } 3108} 3109 3110 3111/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3112/// See the comments for EvalAddr for more details. 3113static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 3114do { 3115 // We should only be called for evaluating non-pointer expressions, or 3116 // expressions with a pointer type that are not used as references but instead 3117 // are l-values (e.g., DeclRefExpr with a pointer type). 3118 3119 // Our "symbolic interpreter" is just a dispatch off the currently 3120 // viewed AST node. We then recursively traverse the AST by calling 3121 // EvalAddr and EvalVal appropriately. 3122 3123 E = E->IgnoreParens(); 3124 switch (E->getStmtClass()) { 3125 case Stmt::ImplicitCastExprClass: { 3126 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3127 if (IE->getValueKind() == VK_LValue) { 3128 E = IE->getSubExpr(); 3129 continue; 3130 } 3131 return NULL; 3132 } 3133 3134 case Stmt::ExprWithCleanupsClass: 3135 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars); 3136 3137 case Stmt::DeclRefExprClass: { 3138 // When we hit a DeclRefExpr we are looking at code that refers to a 3139 // variable's name. If it's not a reference variable we check if it has 3140 // local storage within the function, and if so, return the expression. 3141 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3142 3143 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3144 if (V->hasLocalStorage()) { 3145 if (!V->getType()->isReferenceType()) 3146 return DR; 3147 3148 // Reference variable, follow through to the expression that 3149 // it points to. 3150 if (V->hasInit()) { 3151 // Add the reference variable to the "trail". 3152 refVars.push_back(DR); 3153 return EvalVal(V->getInit(), refVars); 3154 } 3155 } 3156 3157 return NULL; 3158 } 3159 3160 case Stmt::UnaryOperatorClass: { 3161 // The only unary operator that make sense to handle here 3162 // is Deref. All others don't resolve to a "name." This includes 3163 // handling all sorts of rvalues passed to a unary operator. 3164 UnaryOperator *U = cast<UnaryOperator>(E); 3165 3166 if (U->getOpcode() == UO_Deref) 3167 return EvalAddr(U->getSubExpr(), refVars); 3168 3169 return NULL; 3170 } 3171 3172 case Stmt::ArraySubscriptExprClass: { 3173 // Array subscripts are potential references to data on the stack. We 3174 // retrieve the DeclRefExpr* for the array variable if it indeed 3175 // has local storage. 3176 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 3177 } 3178 3179 case Stmt::ConditionalOperatorClass: { 3180 // For conditional operators we need to see if either the LHS or RHS are 3181 // non-NULL Expr's. If one is non-NULL, we return it. 3182 ConditionalOperator *C = cast<ConditionalOperator>(E); 3183 3184 // Handle the GNU extension for missing LHS. 3185 if (Expr *lhsExpr = C->getLHS()) 3186 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 3187 return LHS; 3188 3189 return EvalVal(C->getRHS(), refVars); 3190 } 3191 3192 // Accesses to members are potential references to data on the stack. 3193 case Stmt::MemberExprClass: { 3194 MemberExpr *M = cast<MemberExpr>(E); 3195 3196 // Check for indirect access. We only want direct field accesses. 3197 if (M->isArrow()) 3198 return NULL; 3199 3200 // Check whether the member type is itself a reference, in which case 3201 // we're not going to refer to the member, but to what the member refers to. 3202 if (M->getMemberDecl()->getType()->isReferenceType()) 3203 return NULL; 3204 3205 return EvalVal(M->getBase(), refVars); 3206 } 3207 3208 case Stmt::MaterializeTemporaryExprClass: 3209 if (Expr *Result = EvalVal( 3210 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3211 refVars)) 3212 return Result; 3213 3214 return E; 3215 3216 default: 3217 // Check that we don't return or take the address of a reference to a 3218 // temporary. This is only useful in C++. 3219 if (!E->isTypeDependent() && E->isRValue()) 3220 return E; 3221 3222 // Everything else: we simply don't reason about them. 3223 return NULL; 3224 } 3225} while (true); 3226} 3227 3228//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3229 3230/// Check for comparisons of floating point operands using != and ==. 3231/// Issue a warning if these are no self-comparisons, as they are not likely 3232/// to do what the programmer intended. 3233void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3234 bool EmitWarning = true; 3235 3236 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3237 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3238 3239 // Special case: check for x == x (which is OK). 3240 // Do not emit warnings for such cases. 3241 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3242 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3243 if (DRL->getDecl() == DRR->getDecl()) 3244 EmitWarning = false; 3245 3246 3247 // Special case: check for comparisons against literals that can be exactly 3248 // represented by APFloat. In such cases, do not emit a warning. This 3249 // is a heuristic: often comparison against such literals are used to 3250 // detect if a value in a variable has not changed. This clearly can 3251 // lead to false negatives. 3252 if (EmitWarning) { 3253 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3254 if (FLL->isExact()) 3255 EmitWarning = false; 3256 } else 3257 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 3258 if (FLR->isExact()) 3259 EmitWarning = false; 3260 } 3261 } 3262 3263 // Check for comparisons with builtin types. 3264 if (EmitWarning) 3265 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3266 if (CL->isBuiltinCall()) 3267 EmitWarning = false; 3268 3269 if (EmitWarning) 3270 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3271 if (CR->isBuiltinCall()) 3272 EmitWarning = false; 3273 3274 // Emit the diagnostic. 3275 if (EmitWarning) 3276 Diag(Loc, diag::warn_floatingpoint_eq) 3277 << LHS->getSourceRange() << RHS->getSourceRange(); 3278} 3279 3280//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3281//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3282 3283namespace { 3284 3285/// Structure recording the 'active' range of an integer-valued 3286/// expression. 3287struct IntRange { 3288 /// The number of bits active in the int. 3289 unsigned Width; 3290 3291 /// True if the int is known not to have negative values. 3292 bool NonNegative; 3293 3294 IntRange(unsigned Width, bool NonNegative) 3295 : Width(Width), NonNegative(NonNegative) 3296 {} 3297 3298 /// Returns the range of the bool type. 3299 static IntRange forBoolType() { 3300 return IntRange(1, true); 3301 } 3302 3303 /// Returns the range of an opaque value of the given integral type. 3304 static IntRange forValueOfType(ASTContext &C, QualType T) { 3305 return forValueOfCanonicalType(C, 3306 T->getCanonicalTypeInternal().getTypePtr()); 3307 } 3308 3309 /// Returns the range of an opaque value of a canonical integral type. 3310 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3311 assert(T->isCanonicalUnqualified()); 3312 3313 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3314 T = VT->getElementType().getTypePtr(); 3315 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3316 T = CT->getElementType().getTypePtr(); 3317 3318 // For enum types, use the known bit width of the enumerators. 3319 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3320 EnumDecl *Enum = ET->getDecl(); 3321 if (!Enum->isCompleteDefinition()) 3322 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3323 3324 unsigned NumPositive = Enum->getNumPositiveBits(); 3325 unsigned NumNegative = Enum->getNumNegativeBits(); 3326 3327 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 3328 } 3329 3330 const BuiltinType *BT = cast<BuiltinType>(T); 3331 assert(BT->isInteger()); 3332 3333 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3334 } 3335 3336 /// Returns the "target" range of a canonical integral type, i.e. 3337 /// the range of values expressible in the type. 3338 /// 3339 /// This matches forValueOfCanonicalType except that enums have the 3340 /// full range of their type, not the range of their enumerators. 3341 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 3342 assert(T->isCanonicalUnqualified()); 3343 3344 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3345 T = VT->getElementType().getTypePtr(); 3346 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3347 T = CT->getElementType().getTypePtr(); 3348 if (const EnumType *ET = dyn_cast<EnumType>(T)) 3349 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 3350 3351 const BuiltinType *BT = cast<BuiltinType>(T); 3352 assert(BT->isInteger()); 3353 3354 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3355 } 3356 3357 /// Returns the supremum of two ranges: i.e. their conservative merge. 3358 static IntRange join(IntRange L, IntRange R) { 3359 return IntRange(std::max(L.Width, R.Width), 3360 L.NonNegative && R.NonNegative); 3361 } 3362 3363 /// Returns the infinum of two ranges: i.e. their aggressive merge. 3364 static IntRange meet(IntRange L, IntRange R) { 3365 return IntRange(std::min(L.Width, R.Width), 3366 L.NonNegative || R.NonNegative); 3367 } 3368}; 3369 3370static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 3371 unsigned MaxWidth) { 3372 if (value.isSigned() && value.isNegative()) 3373 return IntRange(value.getMinSignedBits(), false); 3374 3375 if (value.getBitWidth() > MaxWidth) 3376 value = value.trunc(MaxWidth); 3377 3378 // isNonNegative() just checks the sign bit without considering 3379 // signedness. 3380 return IntRange(value.getActiveBits(), true); 3381} 3382 3383static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 3384 unsigned MaxWidth) { 3385 if (result.isInt()) 3386 return GetValueRange(C, result.getInt(), MaxWidth); 3387 3388 if (result.isVector()) { 3389 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 3390 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 3391 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 3392 R = IntRange::join(R, El); 3393 } 3394 return R; 3395 } 3396 3397 if (result.isComplexInt()) { 3398 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 3399 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 3400 return IntRange::join(R, I); 3401 } 3402 3403 // This can happen with lossless casts to intptr_t of "based" lvalues. 3404 // Assume it might use arbitrary bits. 3405 // FIXME: The only reason we need to pass the type in here is to get 3406 // the sign right on this one case. It would be nice if APValue 3407 // preserved this. 3408 assert(result.isLValue() || result.isAddrLabelDiff()); 3409 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 3410} 3411 3412/// Pseudo-evaluate the given integer expression, estimating the 3413/// range of values it might take. 3414/// 3415/// \param MaxWidth - the width to which the value will be truncated 3416static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 3417 E = E->IgnoreParens(); 3418 3419 // Try a full evaluation first. 3420 Expr::EvalResult result; 3421 if (E->EvaluateAsRValue(result, C)) 3422 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 3423 3424 // I think we only want to look through implicit casts here; if the 3425 // user has an explicit widening cast, we should treat the value as 3426 // being of the new, wider type. 3427 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 3428 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 3429 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 3430 3431 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 3432 3433 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 3434 3435 // Assume that non-integer casts can span the full range of the type. 3436 if (!isIntegerCast) 3437 return OutputTypeRange; 3438 3439 IntRange SubRange 3440 = GetExprRange(C, CE->getSubExpr(), 3441 std::min(MaxWidth, OutputTypeRange.Width)); 3442 3443 // Bail out if the subexpr's range is as wide as the cast type. 3444 if (SubRange.Width >= OutputTypeRange.Width) 3445 return OutputTypeRange; 3446 3447 // Otherwise, we take the smaller width, and we're non-negative if 3448 // either the output type or the subexpr is. 3449 return IntRange(SubRange.Width, 3450 SubRange.NonNegative || OutputTypeRange.NonNegative); 3451 } 3452 3453 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 3454 // If we can fold the condition, just take that operand. 3455 bool CondResult; 3456 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 3457 return GetExprRange(C, CondResult ? CO->getTrueExpr() 3458 : CO->getFalseExpr(), 3459 MaxWidth); 3460 3461 // Otherwise, conservatively merge. 3462 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 3463 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 3464 return IntRange::join(L, R); 3465 } 3466 3467 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3468 switch (BO->getOpcode()) { 3469 3470 // Boolean-valued operations are single-bit and positive. 3471 case BO_LAnd: 3472 case BO_LOr: 3473 case BO_LT: 3474 case BO_GT: 3475 case BO_LE: 3476 case BO_GE: 3477 case BO_EQ: 3478 case BO_NE: 3479 return IntRange::forBoolType(); 3480 3481 // The type of the assignments is the type of the LHS, so the RHS 3482 // is not necessarily the same type. 3483 case BO_MulAssign: 3484 case BO_DivAssign: 3485 case BO_RemAssign: 3486 case BO_AddAssign: 3487 case BO_SubAssign: 3488 case BO_XorAssign: 3489 case BO_OrAssign: 3490 // TODO: bitfields? 3491 return IntRange::forValueOfType(C, E->getType()); 3492 3493 // Simple assignments just pass through the RHS, which will have 3494 // been coerced to the LHS type. 3495 case BO_Assign: 3496 // TODO: bitfields? 3497 return GetExprRange(C, BO->getRHS(), MaxWidth); 3498 3499 // Operations with opaque sources are black-listed. 3500 case BO_PtrMemD: 3501 case BO_PtrMemI: 3502 return IntRange::forValueOfType(C, E->getType()); 3503 3504 // Bitwise-and uses the *infinum* of the two source ranges. 3505 case BO_And: 3506 case BO_AndAssign: 3507 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 3508 GetExprRange(C, BO->getRHS(), MaxWidth)); 3509 3510 // Left shift gets black-listed based on a judgement call. 3511 case BO_Shl: 3512 // ...except that we want to treat '1 << (blah)' as logically 3513 // positive. It's an important idiom. 3514 if (IntegerLiteral *I 3515 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 3516 if (I->getValue() == 1) { 3517 IntRange R = IntRange::forValueOfType(C, E->getType()); 3518 return IntRange(R.Width, /*NonNegative*/ true); 3519 } 3520 } 3521 // fallthrough 3522 3523 case BO_ShlAssign: 3524 return IntRange::forValueOfType(C, E->getType()); 3525 3526 // Right shift by a constant can narrow its left argument. 3527 case BO_Shr: 3528 case BO_ShrAssign: { 3529 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3530 3531 // If the shift amount is a positive constant, drop the width by 3532 // that much. 3533 llvm::APSInt shift; 3534 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 3535 shift.isNonNegative()) { 3536 unsigned zext = shift.getZExtValue(); 3537 if (zext >= L.Width) 3538 L.Width = (L.NonNegative ? 0 : 1); 3539 else 3540 L.Width -= zext; 3541 } 3542 3543 return L; 3544 } 3545 3546 // Comma acts as its right operand. 3547 case BO_Comma: 3548 return GetExprRange(C, BO->getRHS(), MaxWidth); 3549 3550 // Black-list pointer subtractions. 3551 case BO_Sub: 3552 if (BO->getLHS()->getType()->isPointerType()) 3553 return IntRange::forValueOfType(C, E->getType()); 3554 break; 3555 3556 // The width of a division result is mostly determined by the size 3557 // of the LHS. 3558 case BO_Div: { 3559 // Don't 'pre-truncate' the operands. 3560 unsigned opWidth = C.getIntWidth(E->getType()); 3561 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3562 3563 // If the divisor is constant, use that. 3564 llvm::APSInt divisor; 3565 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3566 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3567 if (log2 >= L.Width) 3568 L.Width = (L.NonNegative ? 0 : 1); 3569 else 3570 L.Width = std::min(L.Width - log2, MaxWidth); 3571 return L; 3572 } 3573 3574 // Otherwise, just use the LHS's width. 3575 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3576 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3577 } 3578 3579 // The result of a remainder can't be larger than the result of 3580 // either side. 3581 case BO_Rem: { 3582 // Don't 'pre-truncate' the operands. 3583 unsigned opWidth = C.getIntWidth(E->getType()); 3584 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3585 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3586 3587 IntRange meet = IntRange::meet(L, R); 3588 meet.Width = std::min(meet.Width, MaxWidth); 3589 return meet; 3590 } 3591 3592 // The default behavior is okay for these. 3593 case BO_Mul: 3594 case BO_Add: 3595 case BO_Xor: 3596 case BO_Or: 3597 break; 3598 } 3599 3600 // The default case is to treat the operation as if it were closed 3601 // on the narrowest type that encompasses both operands. 3602 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3603 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 3604 return IntRange::join(L, R); 3605 } 3606 3607 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 3608 switch (UO->getOpcode()) { 3609 // Boolean-valued operations are white-listed. 3610 case UO_LNot: 3611 return IntRange::forBoolType(); 3612 3613 // Operations with opaque sources are black-listed. 3614 case UO_Deref: 3615 case UO_AddrOf: // should be impossible 3616 return IntRange::forValueOfType(C, E->getType()); 3617 3618 default: 3619 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 3620 } 3621 } 3622 3623 if (dyn_cast<OffsetOfExpr>(E)) { 3624 IntRange::forValueOfType(C, E->getType()); 3625 } 3626 3627 if (FieldDecl *BitField = E->getBitField()) 3628 return IntRange(BitField->getBitWidthValue(C), 3629 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 3630 3631 return IntRange::forValueOfType(C, E->getType()); 3632} 3633 3634static IntRange GetExprRange(ASTContext &C, Expr *E) { 3635 return GetExprRange(C, E, C.getIntWidth(E->getType())); 3636} 3637 3638/// Checks whether the given value, which currently has the given 3639/// source semantics, has the same value when coerced through the 3640/// target semantics. 3641static bool IsSameFloatAfterCast(const llvm::APFloat &value, 3642 const llvm::fltSemantics &Src, 3643 const llvm::fltSemantics &Tgt) { 3644 llvm::APFloat truncated = value; 3645 3646 bool ignored; 3647 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 3648 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 3649 3650 return truncated.bitwiseIsEqual(value); 3651} 3652 3653/// Checks whether the given value, which currently has the given 3654/// source semantics, has the same value when coerced through the 3655/// target semantics. 3656/// 3657/// The value might be a vector of floats (or a complex number). 3658static bool IsSameFloatAfterCast(const APValue &value, 3659 const llvm::fltSemantics &Src, 3660 const llvm::fltSemantics &Tgt) { 3661 if (value.isFloat()) 3662 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 3663 3664 if (value.isVector()) { 3665 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 3666 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 3667 return false; 3668 return true; 3669 } 3670 3671 assert(value.isComplexFloat()); 3672 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 3673 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 3674} 3675 3676static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 3677 3678static bool IsZero(Sema &S, Expr *E) { 3679 // Suppress cases where we are comparing against an enum constant. 3680 if (const DeclRefExpr *DR = 3681 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 3682 if (isa<EnumConstantDecl>(DR->getDecl())) 3683 return false; 3684 3685 // Suppress cases where the '0' value is expanded from a macro. 3686 if (E->getLocStart().isMacroID()) 3687 return false; 3688 3689 llvm::APSInt Value; 3690 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 3691} 3692 3693static bool HasEnumType(Expr *E) { 3694 // Strip off implicit integral promotions. 3695 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3696 if (ICE->getCastKind() != CK_IntegralCast && 3697 ICE->getCastKind() != CK_NoOp) 3698 break; 3699 E = ICE->getSubExpr(); 3700 } 3701 3702 return E->getType()->isEnumeralType(); 3703} 3704 3705static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 3706 BinaryOperatorKind op = E->getOpcode(); 3707 if (E->isValueDependent()) 3708 return; 3709 3710 if (op == BO_LT && IsZero(S, E->getRHS())) { 3711 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3712 << "< 0" << "false" << HasEnumType(E->getLHS()) 3713 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3714 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 3715 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3716 << ">= 0" << "true" << HasEnumType(E->getLHS()) 3717 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3718 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 3719 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3720 << "0 >" << "false" << HasEnumType(E->getRHS()) 3721 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3722 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 3723 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3724 << "0 <=" << "true" << HasEnumType(E->getRHS()) 3725 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3726 } 3727} 3728 3729/// Analyze the operands of the given comparison. Implements the 3730/// fallback case from AnalyzeComparison. 3731static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 3732 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3733 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3734} 3735 3736/// \brief Implements -Wsign-compare. 3737/// 3738/// \param E the binary operator to check for warnings 3739static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 3740 // The type the comparison is being performed in. 3741 QualType T = E->getLHS()->getType(); 3742 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 3743 && "comparison with mismatched types"); 3744 3745 // We don't do anything special if this isn't an unsigned integral 3746 // comparison: we're only interested in integral comparisons, and 3747 // signed comparisons only happen in cases we don't care to warn about. 3748 // 3749 // We also don't care about value-dependent expressions or expressions 3750 // whose result is a constant. 3751 if (!T->hasUnsignedIntegerRepresentation() 3752 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 3753 return AnalyzeImpConvsInComparison(S, E); 3754 3755 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 3756 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 3757 3758 // Check to see if one of the (unmodified) operands is of different 3759 // signedness. 3760 Expr *signedOperand, *unsignedOperand; 3761 if (LHS->getType()->hasSignedIntegerRepresentation()) { 3762 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 3763 "unsigned comparison between two signed integer expressions?"); 3764 signedOperand = LHS; 3765 unsignedOperand = RHS; 3766 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 3767 signedOperand = RHS; 3768 unsignedOperand = LHS; 3769 } else { 3770 CheckTrivialUnsignedComparison(S, E); 3771 return AnalyzeImpConvsInComparison(S, E); 3772 } 3773 3774 // Otherwise, calculate the effective range of the signed operand. 3775 IntRange signedRange = GetExprRange(S.Context, signedOperand); 3776 3777 // Go ahead and analyze implicit conversions in the operands. Note 3778 // that we skip the implicit conversions on both sides. 3779 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 3780 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 3781 3782 // If the signed range is non-negative, -Wsign-compare won't fire, 3783 // but we should still check for comparisons which are always true 3784 // or false. 3785 if (signedRange.NonNegative) 3786 return CheckTrivialUnsignedComparison(S, E); 3787 3788 // For (in)equality comparisons, if the unsigned operand is a 3789 // constant which cannot collide with a overflowed signed operand, 3790 // then reinterpreting the signed operand as unsigned will not 3791 // change the result of the comparison. 3792 if (E->isEqualityOp()) { 3793 unsigned comparisonWidth = S.Context.getIntWidth(T); 3794 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3795 3796 // We should never be unable to prove that the unsigned operand is 3797 // non-negative. 3798 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3799 3800 if (unsignedRange.Width < comparisonWidth) 3801 return; 3802 } 3803 3804 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 3805 << LHS->getType() << RHS->getType() 3806 << LHS->getSourceRange() << RHS->getSourceRange(); 3807} 3808 3809/// Analyzes an attempt to assign the given value to a bitfield. 3810/// 3811/// Returns true if there was something fishy about the attempt. 3812static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3813 SourceLocation InitLoc) { 3814 assert(Bitfield->isBitField()); 3815 if (Bitfield->isInvalidDecl()) 3816 return false; 3817 3818 // White-list bool bitfields. 3819 if (Bitfield->getType()->isBooleanType()) 3820 return false; 3821 3822 // Ignore value- or type-dependent expressions. 3823 if (Bitfield->getBitWidth()->isValueDependent() || 3824 Bitfield->getBitWidth()->isTypeDependent() || 3825 Init->isValueDependent() || 3826 Init->isTypeDependent()) 3827 return false; 3828 3829 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3830 3831 llvm::APSInt Value; 3832 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 3833 return false; 3834 3835 unsigned OriginalWidth = Value.getBitWidth(); 3836 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 3837 3838 if (OriginalWidth <= FieldWidth) 3839 return false; 3840 3841 // Compute the value which the bitfield will contain. 3842 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 3843 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 3844 3845 // Check whether the stored value is equal to the original value. 3846 TruncatedValue = TruncatedValue.extend(OriginalWidth); 3847 if (Value == TruncatedValue) 3848 return false; 3849 3850 // Special-case bitfields of width 1: booleans are naturally 0/1, and 3851 // therefore don't strictly fit into a signed bitfield of width 1. 3852 if (FieldWidth == 1 && Value == 1) 3853 return false; 3854 3855 std::string PrettyValue = Value.toString(10); 3856 std::string PrettyTrunc = TruncatedValue.toString(10); 3857 3858 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 3859 << PrettyValue << PrettyTrunc << OriginalInit->getType() 3860 << Init->getSourceRange(); 3861 3862 return true; 3863} 3864 3865/// Analyze the given simple or compound assignment for warning-worthy 3866/// operations. 3867static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 3868 // Just recurse on the LHS. 3869 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3870 3871 // We want to recurse on the RHS as normal unless we're assigning to 3872 // a bitfield. 3873 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 3874 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 3875 E->getOperatorLoc())) { 3876 // Recurse, ignoring any implicit conversions on the RHS. 3877 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 3878 E->getOperatorLoc()); 3879 } 3880 } 3881 3882 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3883} 3884 3885/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3886static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 3887 SourceLocation CContext, unsigned diag, 3888 bool pruneControlFlow = false) { 3889 if (pruneControlFlow) { 3890 S.DiagRuntimeBehavior(E->getExprLoc(), E, 3891 S.PDiag(diag) 3892 << SourceType << T << E->getSourceRange() 3893 << SourceRange(CContext)); 3894 return; 3895 } 3896 S.Diag(E->getExprLoc(), diag) 3897 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 3898} 3899 3900/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3901static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 3902 SourceLocation CContext, unsigned diag, 3903 bool pruneControlFlow = false) { 3904 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 3905} 3906 3907/// Diagnose an implicit cast from a literal expression. Does not warn when the 3908/// cast wouldn't lose information. 3909void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 3910 SourceLocation CContext) { 3911 // Try to convert the literal exactly to an integer. If we can, don't warn. 3912 bool isExact = false; 3913 const llvm::APFloat &Value = FL->getValue(); 3914 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 3915 T->hasUnsignedIntegerRepresentation()); 3916 if (Value.convertToInteger(IntegerValue, 3917 llvm::APFloat::rmTowardZero, &isExact) 3918 == llvm::APFloat::opOK && isExact) 3919 return; 3920 3921 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 3922 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 3923} 3924 3925std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 3926 if (!Range.Width) return "0"; 3927 3928 llvm::APSInt ValueInRange = Value; 3929 ValueInRange.setIsSigned(!Range.NonNegative); 3930 ValueInRange = ValueInRange.trunc(Range.Width); 3931 return ValueInRange.toString(10); 3932} 3933 3934void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 3935 SourceLocation CC, bool *ICContext = 0) { 3936 if (E->isTypeDependent() || E->isValueDependent()) return; 3937 3938 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 3939 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 3940 if (Source == Target) return; 3941 if (Target->isDependentType()) return; 3942 3943 // If the conversion context location is invalid don't complain. We also 3944 // don't want to emit a warning if the issue occurs from the expansion of 3945 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 3946 // delay this check as long as possible. Once we detect we are in that 3947 // scenario, we just return. 3948 if (CC.isInvalid()) 3949 return; 3950 3951 // Diagnose implicit casts to bool. 3952 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 3953 if (isa<StringLiteral>(E)) 3954 // Warn on string literal to bool. Checks for string literals in logical 3955 // expressions, for instances, assert(0 && "error here"), is prevented 3956 // by a check in AnalyzeImplicitConversions(). 3957 return DiagnoseImpCast(S, E, T, CC, 3958 diag::warn_impcast_string_literal_to_bool); 3959 if (Source->isFunctionType()) { 3960 // Warn on function to bool. Checks free functions and static member 3961 // functions. Weakly imported functions are excluded from the check, 3962 // since it's common to test their value to check whether the linker 3963 // found a definition for them. 3964 ValueDecl *D = 0; 3965 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 3966 D = R->getDecl(); 3967 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 3968 D = M->getMemberDecl(); 3969 } 3970 3971 if (D && !D->isWeak()) { 3972 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 3973 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 3974 << F << E->getSourceRange() << SourceRange(CC); 3975 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 3976 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 3977 QualType ReturnType; 3978 UnresolvedSet<4> NonTemplateOverloads; 3979 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 3980 if (!ReturnType.isNull() 3981 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 3982 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 3983 << FixItHint::CreateInsertion( 3984 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 3985 return; 3986 } 3987 } 3988 } 3989 return; // Other casts to bool are not checked. 3990 } 3991 3992 // Strip vector types. 3993 if (isa<VectorType>(Source)) { 3994 if (!isa<VectorType>(Target)) { 3995 if (S.SourceMgr.isInSystemMacro(CC)) 3996 return; 3997 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 3998 } 3999 4000 // If the vector cast is cast between two vectors of the same size, it is 4001 // a bitcast, not a conversion. 4002 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 4003 return; 4004 4005 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 4006 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 4007 } 4008 4009 // Strip complex types. 4010 if (isa<ComplexType>(Source)) { 4011 if (!isa<ComplexType>(Target)) { 4012 if (S.SourceMgr.isInSystemMacro(CC)) 4013 return; 4014 4015 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 4016 } 4017 4018 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 4019 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 4020 } 4021 4022 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 4023 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 4024 4025 // If the source is floating point... 4026 if (SourceBT && SourceBT->isFloatingPoint()) { 4027 // ...and the target is floating point... 4028 if (TargetBT && TargetBT->isFloatingPoint()) { 4029 // ...then warn if we're dropping FP rank. 4030 4031 // Builtin FP kinds are ordered by increasing FP rank. 4032 if (SourceBT->getKind() > TargetBT->getKind()) { 4033 // Don't warn about float constants that are precisely 4034 // representable in the target type. 4035 Expr::EvalResult result; 4036 if (E->EvaluateAsRValue(result, S.Context)) { 4037 // Value might be a float, a float vector, or a float complex. 4038 if (IsSameFloatAfterCast(result.Val, 4039 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 4040 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 4041 return; 4042 } 4043 4044 if (S.SourceMgr.isInSystemMacro(CC)) 4045 return; 4046 4047 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 4048 } 4049 return; 4050 } 4051 4052 // If the target is integral, always warn. 4053 if ((TargetBT && TargetBT->isInteger())) { 4054 if (S.SourceMgr.isInSystemMacro(CC)) 4055 return; 4056 4057 Expr *InnerE = E->IgnoreParenImpCasts(); 4058 // We also want to warn on, e.g., "int i = -1.234" 4059 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 4060 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 4061 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 4062 4063 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 4064 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 4065 } else { 4066 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 4067 } 4068 } 4069 4070 return; 4071 } 4072 4073 if (!Source->isIntegerType() || !Target->isIntegerType()) 4074 return; 4075 4076 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 4077 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 4078 SourceLocation Loc = E->getSourceRange().getBegin(); 4079 if (Loc.isMacroID()) 4080 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 4081 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 4082 << T << Loc << clang::SourceRange(CC); 4083 return; 4084 } 4085 4086 IntRange SourceRange = GetExprRange(S.Context, E); 4087 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 4088 4089 if (SourceRange.Width > TargetRange.Width) { 4090 // If the source is a constant, use a default-on diagnostic. 4091 // TODO: this should happen for bitfield stores, too. 4092 llvm::APSInt Value(32); 4093 if (E->isIntegerConstantExpr(Value, S.Context)) { 4094 if (S.SourceMgr.isInSystemMacro(CC)) 4095 return; 4096 4097 std::string PrettySourceValue = Value.toString(10); 4098 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 4099 4100 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4101 S.PDiag(diag::warn_impcast_integer_precision_constant) 4102 << PrettySourceValue << PrettyTargetValue 4103 << E->getType() << T << E->getSourceRange() 4104 << clang::SourceRange(CC)); 4105 return; 4106 } 4107 4108 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 4109 if (S.SourceMgr.isInSystemMacro(CC)) 4110 return; 4111 4112 if (SourceRange.Width == 64 && TargetRange.Width == 32) 4113 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 4114 /* pruneControlFlow */ true); 4115 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 4116 } 4117 4118 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 4119 (!TargetRange.NonNegative && SourceRange.NonNegative && 4120 SourceRange.Width == TargetRange.Width)) { 4121 4122 if (S.SourceMgr.isInSystemMacro(CC)) 4123 return; 4124 4125 unsigned DiagID = diag::warn_impcast_integer_sign; 4126 4127 // Traditionally, gcc has warned about this under -Wsign-compare. 4128 // We also want to warn about it in -Wconversion. 4129 // So if -Wconversion is off, use a completely identical diagnostic 4130 // in the sign-compare group. 4131 // The conditional-checking code will 4132 if (ICContext) { 4133 DiagID = diag::warn_impcast_integer_sign_conditional; 4134 *ICContext = true; 4135 } 4136 4137 return DiagnoseImpCast(S, E, T, CC, DiagID); 4138 } 4139 4140 // Diagnose conversions between different enumeration types. 4141 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 4142 // type, to give us better diagnostics. 4143 QualType SourceType = E->getType(); 4144 if (!S.getLangOpts().CPlusPlus) { 4145 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 4146 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 4147 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 4148 SourceType = S.Context.getTypeDeclType(Enum); 4149 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 4150 } 4151 } 4152 4153 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 4154 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 4155 if ((SourceEnum->getDecl()->getIdentifier() || 4156 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 4157 (TargetEnum->getDecl()->getIdentifier() || 4158 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 4159 SourceEnum != TargetEnum) { 4160 if (S.SourceMgr.isInSystemMacro(CC)) 4161 return; 4162 4163 return DiagnoseImpCast(S, E, SourceType, T, CC, 4164 diag::warn_impcast_different_enum_types); 4165 } 4166 4167 return; 4168} 4169 4170void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 4171 4172void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 4173 SourceLocation CC, bool &ICContext) { 4174 E = E->IgnoreParenImpCasts(); 4175 4176 if (isa<ConditionalOperator>(E)) 4177 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 4178 4179 AnalyzeImplicitConversions(S, E, CC); 4180 if (E->getType() != T) 4181 return CheckImplicitConversion(S, E, T, CC, &ICContext); 4182 return; 4183} 4184 4185void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 4186 SourceLocation CC = E->getQuestionLoc(); 4187 4188 AnalyzeImplicitConversions(S, E->getCond(), CC); 4189 4190 bool Suspicious = false; 4191 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 4192 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 4193 4194 // If -Wconversion would have warned about either of the candidates 4195 // for a signedness conversion to the context type... 4196 if (!Suspicious) return; 4197 4198 // ...but it's currently ignored... 4199 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 4200 CC)) 4201 return; 4202 4203 // ...then check whether it would have warned about either of the 4204 // candidates for a signedness conversion to the condition type. 4205 if (E->getType() == T) return; 4206 4207 Suspicious = false; 4208 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 4209 E->getType(), CC, &Suspicious); 4210 if (!Suspicious) 4211 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 4212 E->getType(), CC, &Suspicious); 4213} 4214 4215/// AnalyzeImplicitConversions - Find and report any interesting 4216/// implicit conversions in the given expression. There are a couple 4217/// of competing diagnostics here, -Wconversion and -Wsign-compare. 4218void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 4219 QualType T = OrigE->getType(); 4220 Expr *E = OrigE->IgnoreParenImpCasts(); 4221 4222 if (E->isTypeDependent() || E->isValueDependent()) 4223 return; 4224 4225 // For conditional operators, we analyze the arguments as if they 4226 // were being fed directly into the output. 4227 if (isa<ConditionalOperator>(E)) { 4228 ConditionalOperator *CO = cast<ConditionalOperator>(E); 4229 CheckConditionalOperator(S, CO, T); 4230 return; 4231 } 4232 4233 // Go ahead and check any implicit conversions we might have skipped. 4234 // The non-canonical typecheck is just an optimization; 4235 // CheckImplicitConversion will filter out dead implicit conversions. 4236 if (E->getType() != T) 4237 CheckImplicitConversion(S, E, T, CC); 4238 4239 // Now continue drilling into this expression. 4240 4241 // Skip past explicit casts. 4242 if (isa<ExplicitCastExpr>(E)) { 4243 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 4244 return AnalyzeImplicitConversions(S, E, CC); 4245 } 4246 4247 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4248 // Do a somewhat different check with comparison operators. 4249 if (BO->isComparisonOp()) 4250 return AnalyzeComparison(S, BO); 4251 4252 // And with simple assignments. 4253 if (BO->getOpcode() == BO_Assign) 4254 return AnalyzeAssignment(S, BO); 4255 } 4256 4257 // These break the otherwise-useful invariant below. Fortunately, 4258 // we don't really need to recurse into them, because any internal 4259 // expressions should have been analyzed already when they were 4260 // built into statements. 4261 if (isa<StmtExpr>(E)) return; 4262 4263 // Don't descend into unevaluated contexts. 4264 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 4265 4266 // Now just recurse over the expression's children. 4267 CC = E->getExprLoc(); 4268 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 4269 bool IsLogicalOperator = BO && BO->isLogicalOp(); 4270 for (Stmt::child_range I = E->children(); I; ++I) { 4271 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 4272 if (!ChildExpr) 4273 continue; 4274 4275 if (IsLogicalOperator && 4276 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 4277 // Ignore checking string literals that are in logical operators. 4278 continue; 4279 AnalyzeImplicitConversions(S, ChildExpr, CC); 4280 } 4281} 4282 4283} // end anonymous namespace 4284 4285/// Diagnoses "dangerous" implicit conversions within the given 4286/// expression (which is a full expression). Implements -Wconversion 4287/// and -Wsign-compare. 4288/// 4289/// \param CC the "context" location of the implicit conversion, i.e. 4290/// the most location of the syntactic entity requiring the implicit 4291/// conversion 4292void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 4293 // Don't diagnose in unevaluated contexts. 4294 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 4295 return; 4296 4297 // Don't diagnose for value- or type-dependent expressions. 4298 if (E->isTypeDependent() || E->isValueDependent()) 4299 return; 4300 4301 // Check for array bounds violations in cases where the check isn't triggered 4302 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 4303 // ArraySubscriptExpr is on the RHS of a variable initialization. 4304 CheckArrayAccess(E); 4305 4306 // This is not the right CC for (e.g.) a variable initialization. 4307 AnalyzeImplicitConversions(*this, E, CC); 4308} 4309 4310void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 4311 FieldDecl *BitField, 4312 Expr *Init) { 4313 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 4314} 4315 4316/// CheckParmsForFunctionDef - Check that the parameters of the given 4317/// function are appropriate for the definition of a function. This 4318/// takes care of any checks that cannot be performed on the 4319/// declaration itself, e.g., that the types of each of the function 4320/// parameters are complete. 4321bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 4322 bool CheckParameterNames) { 4323 bool HasInvalidParm = false; 4324 for (; P != PEnd; ++P) { 4325 ParmVarDecl *Param = *P; 4326 4327 // C99 6.7.5.3p4: the parameters in a parameter type list in a 4328 // function declarator that is part of a function definition of 4329 // that function shall not have incomplete type. 4330 // 4331 // This is also C++ [dcl.fct]p6. 4332 if (!Param->isInvalidDecl() && 4333 RequireCompleteType(Param->getLocation(), Param->getType(), 4334 diag::err_typecheck_decl_incomplete_type)) { 4335 Param->setInvalidDecl(); 4336 HasInvalidParm = true; 4337 } 4338 4339 // C99 6.9.1p5: If the declarator includes a parameter type list, the 4340 // declaration of each parameter shall include an identifier. 4341 if (CheckParameterNames && 4342 Param->getIdentifier() == 0 && 4343 !Param->isImplicit() && 4344 !getLangOpts().CPlusPlus) 4345 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 4346 4347 // C99 6.7.5.3p12: 4348 // If the function declarator is not part of a definition of that 4349 // function, parameters may have incomplete type and may use the [*] 4350 // notation in their sequences of declarator specifiers to specify 4351 // variable length array types. 4352 QualType PType = Param->getOriginalType(); 4353 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 4354 if (AT->getSizeModifier() == ArrayType::Star) { 4355 // FIXME: This diagnosic should point the the '[*]' if source-location 4356 // information is added for it. 4357 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 4358 } 4359 } 4360 } 4361 4362 return HasInvalidParm; 4363} 4364 4365/// CheckCastAlign - Implements -Wcast-align, which warns when a 4366/// pointer cast increases the alignment requirements. 4367void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 4368 // This is actually a lot of work to potentially be doing on every 4369 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 4370 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 4371 TRange.getBegin()) 4372 == DiagnosticsEngine::Ignored) 4373 return; 4374 4375 // Ignore dependent types. 4376 if (T->isDependentType() || Op->getType()->isDependentType()) 4377 return; 4378 4379 // Require that the destination be a pointer type. 4380 const PointerType *DestPtr = T->getAs<PointerType>(); 4381 if (!DestPtr) return; 4382 4383 // If the destination has alignment 1, we're done. 4384 QualType DestPointee = DestPtr->getPointeeType(); 4385 if (DestPointee->isIncompleteType()) return; 4386 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 4387 if (DestAlign.isOne()) return; 4388 4389 // Require that the source be a pointer type. 4390 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 4391 if (!SrcPtr) return; 4392 QualType SrcPointee = SrcPtr->getPointeeType(); 4393 4394 // Whitelist casts from cv void*. We already implicitly 4395 // whitelisted casts to cv void*, since they have alignment 1. 4396 // Also whitelist casts involving incomplete types, which implicitly 4397 // includes 'void'. 4398 if (SrcPointee->isIncompleteType()) return; 4399 4400 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 4401 if (SrcAlign >= DestAlign) return; 4402 4403 Diag(TRange.getBegin(), diag::warn_cast_align) 4404 << Op->getType() << T 4405 << static_cast<unsigned>(SrcAlign.getQuantity()) 4406 << static_cast<unsigned>(DestAlign.getQuantity()) 4407 << TRange << Op->getSourceRange(); 4408} 4409 4410static const Type* getElementType(const Expr *BaseExpr) { 4411 const Type* EltType = BaseExpr->getType().getTypePtr(); 4412 if (EltType->isAnyPointerType()) 4413 return EltType->getPointeeType().getTypePtr(); 4414 else if (EltType->isArrayType()) 4415 return EltType->getBaseElementTypeUnsafe(); 4416 return EltType; 4417} 4418 4419/// \brief Check whether this array fits the idiom of a size-one tail padded 4420/// array member of a struct. 4421/// 4422/// We avoid emitting out-of-bounds access warnings for such arrays as they are 4423/// commonly used to emulate flexible arrays in C89 code. 4424static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 4425 const NamedDecl *ND) { 4426 if (Size != 1 || !ND) return false; 4427 4428 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 4429 if (!FD) return false; 4430 4431 // Don't consider sizes resulting from macro expansions or template argument 4432 // substitution to form C89 tail-padded arrays. 4433 ConstantArrayTypeLoc TL = 4434 cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc()); 4435 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr()); 4436 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 4437 return false; 4438 4439 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 4440 if (!RD) return false; 4441 if (RD->isUnion()) return false; 4442 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 4443 if (!CRD->isStandardLayout()) return false; 4444 } 4445 4446 // See if this is the last field decl in the record. 4447 const Decl *D = FD; 4448 while ((D = D->getNextDeclInContext())) 4449 if (isa<FieldDecl>(D)) 4450 return false; 4451 return true; 4452} 4453 4454void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 4455 const ArraySubscriptExpr *ASE, 4456 bool AllowOnePastEnd, bool IndexNegated) { 4457 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 4458 if (IndexExpr->isValueDependent()) 4459 return; 4460 4461 const Type *EffectiveType = getElementType(BaseExpr); 4462 BaseExpr = BaseExpr->IgnoreParenCasts(); 4463 const ConstantArrayType *ArrayTy = 4464 Context.getAsConstantArrayType(BaseExpr->getType()); 4465 if (!ArrayTy) 4466 return; 4467 4468 llvm::APSInt index; 4469 if (!IndexExpr->EvaluateAsInt(index, Context)) 4470 return; 4471 if (IndexNegated) 4472 index = -index; 4473 4474 const NamedDecl *ND = NULL; 4475 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4476 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4477 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4478 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4479 4480 if (index.isUnsigned() || !index.isNegative()) { 4481 llvm::APInt size = ArrayTy->getSize(); 4482 if (!size.isStrictlyPositive()) 4483 return; 4484 4485 const Type* BaseType = getElementType(BaseExpr); 4486 if (BaseType != EffectiveType) { 4487 // Make sure we're comparing apples to apples when comparing index to size 4488 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 4489 uint64_t array_typesize = Context.getTypeSize(BaseType); 4490 // Handle ptrarith_typesize being zero, such as when casting to void* 4491 if (!ptrarith_typesize) ptrarith_typesize = 1; 4492 if (ptrarith_typesize != array_typesize) { 4493 // There's a cast to a different size type involved 4494 uint64_t ratio = array_typesize / ptrarith_typesize; 4495 // TODO: Be smarter about handling cases where array_typesize is not a 4496 // multiple of ptrarith_typesize 4497 if (ptrarith_typesize * ratio == array_typesize) 4498 size *= llvm::APInt(size.getBitWidth(), ratio); 4499 } 4500 } 4501 4502 if (size.getBitWidth() > index.getBitWidth()) 4503 index = index.zext(size.getBitWidth()); 4504 else if (size.getBitWidth() < index.getBitWidth()) 4505 size = size.zext(index.getBitWidth()); 4506 4507 // For array subscripting the index must be less than size, but for pointer 4508 // arithmetic also allow the index (offset) to be equal to size since 4509 // computing the next address after the end of the array is legal and 4510 // commonly done e.g. in C++ iterators and range-based for loops. 4511 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 4512 return; 4513 4514 // Also don't warn for arrays of size 1 which are members of some 4515 // structure. These are often used to approximate flexible arrays in C89 4516 // code. 4517 if (IsTailPaddedMemberArray(*this, size, ND)) 4518 return; 4519 4520 // Suppress the warning if the subscript expression (as identified by the 4521 // ']' location) and the index expression are both from macro expansions 4522 // within a system header. 4523 if (ASE) { 4524 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 4525 ASE->getRBracketLoc()); 4526 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 4527 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 4528 IndexExpr->getLocStart()); 4529 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 4530 return; 4531 } 4532 } 4533 4534 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 4535 if (ASE) 4536 DiagID = diag::warn_array_index_exceeds_bounds; 4537 4538 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4539 PDiag(DiagID) << index.toString(10, true) 4540 << size.toString(10, true) 4541 << (unsigned)size.getLimitedValue(~0U) 4542 << IndexExpr->getSourceRange()); 4543 } else { 4544 unsigned DiagID = diag::warn_array_index_precedes_bounds; 4545 if (!ASE) { 4546 DiagID = diag::warn_ptr_arith_precedes_bounds; 4547 if (index.isNegative()) index = -index; 4548 } 4549 4550 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4551 PDiag(DiagID) << index.toString(10, true) 4552 << IndexExpr->getSourceRange()); 4553 } 4554 4555 if (!ND) { 4556 // Try harder to find a NamedDecl to point at in the note. 4557 while (const ArraySubscriptExpr *ASE = 4558 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 4559 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 4560 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4561 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4562 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4563 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4564 } 4565 4566 if (ND) 4567 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 4568 PDiag(diag::note_array_index_out_of_bounds) 4569 << ND->getDeclName()); 4570} 4571 4572void Sema::CheckArrayAccess(const Expr *expr) { 4573 int AllowOnePastEnd = 0; 4574 while (expr) { 4575 expr = expr->IgnoreParenImpCasts(); 4576 switch (expr->getStmtClass()) { 4577 case Stmt::ArraySubscriptExprClass: { 4578 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 4579 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 4580 AllowOnePastEnd > 0); 4581 return; 4582 } 4583 case Stmt::UnaryOperatorClass: { 4584 // Only unwrap the * and & unary operators 4585 const UnaryOperator *UO = cast<UnaryOperator>(expr); 4586 expr = UO->getSubExpr(); 4587 switch (UO->getOpcode()) { 4588 case UO_AddrOf: 4589 AllowOnePastEnd++; 4590 break; 4591 case UO_Deref: 4592 AllowOnePastEnd--; 4593 break; 4594 default: 4595 return; 4596 } 4597 break; 4598 } 4599 case Stmt::ConditionalOperatorClass: { 4600 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 4601 if (const Expr *lhs = cond->getLHS()) 4602 CheckArrayAccess(lhs); 4603 if (const Expr *rhs = cond->getRHS()) 4604 CheckArrayAccess(rhs); 4605 return; 4606 } 4607 default: 4608 return; 4609 } 4610 } 4611} 4612 4613//===--- CHECK: Objective-C retain cycles ----------------------------------// 4614 4615namespace { 4616 struct RetainCycleOwner { 4617 RetainCycleOwner() : Variable(0), Indirect(false) {} 4618 VarDecl *Variable; 4619 SourceRange Range; 4620 SourceLocation Loc; 4621 bool Indirect; 4622 4623 void setLocsFrom(Expr *e) { 4624 Loc = e->getExprLoc(); 4625 Range = e->getSourceRange(); 4626 } 4627 }; 4628} 4629 4630/// Consider whether capturing the given variable can possibly lead to 4631/// a retain cycle. 4632static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 4633 // In ARC, it's captured strongly iff the variable has __strong 4634 // lifetime. In MRR, it's captured strongly if the variable is 4635 // __block and has an appropriate type. 4636 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4637 return false; 4638 4639 owner.Variable = var; 4640 owner.setLocsFrom(ref); 4641 return true; 4642} 4643 4644static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 4645 while (true) { 4646 e = e->IgnoreParens(); 4647 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 4648 switch (cast->getCastKind()) { 4649 case CK_BitCast: 4650 case CK_LValueBitCast: 4651 case CK_LValueToRValue: 4652 case CK_ARCReclaimReturnedObject: 4653 e = cast->getSubExpr(); 4654 continue; 4655 4656 default: 4657 return false; 4658 } 4659 } 4660 4661 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 4662 ObjCIvarDecl *ivar = ref->getDecl(); 4663 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4664 return false; 4665 4666 // Try to find a retain cycle in the base. 4667 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 4668 return false; 4669 4670 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 4671 owner.Indirect = true; 4672 return true; 4673 } 4674 4675 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 4676 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 4677 if (!var) return false; 4678 return considerVariable(var, ref, owner); 4679 } 4680 4681 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 4682 if (member->isArrow()) return false; 4683 4684 // Don't count this as an indirect ownership. 4685 e = member->getBase(); 4686 continue; 4687 } 4688 4689 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 4690 // Only pay attention to pseudo-objects on property references. 4691 ObjCPropertyRefExpr *pre 4692 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 4693 ->IgnoreParens()); 4694 if (!pre) return false; 4695 if (pre->isImplicitProperty()) return false; 4696 ObjCPropertyDecl *property = pre->getExplicitProperty(); 4697 if (!property->isRetaining() && 4698 !(property->getPropertyIvarDecl() && 4699 property->getPropertyIvarDecl()->getType() 4700 .getObjCLifetime() == Qualifiers::OCL_Strong)) 4701 return false; 4702 4703 owner.Indirect = true; 4704 if (pre->isSuperReceiver()) { 4705 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 4706 if (!owner.Variable) 4707 return false; 4708 owner.Loc = pre->getLocation(); 4709 owner.Range = pre->getSourceRange(); 4710 return true; 4711 } 4712 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 4713 ->getSourceExpr()); 4714 continue; 4715 } 4716 4717 // Array ivars? 4718 4719 return false; 4720 } 4721} 4722 4723namespace { 4724 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 4725 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 4726 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 4727 Variable(variable), Capturer(0) {} 4728 4729 VarDecl *Variable; 4730 Expr *Capturer; 4731 4732 void VisitDeclRefExpr(DeclRefExpr *ref) { 4733 if (ref->getDecl() == Variable && !Capturer) 4734 Capturer = ref; 4735 } 4736 4737 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 4738 if (Capturer) return; 4739 Visit(ref->getBase()); 4740 if (Capturer && ref->isFreeIvar()) 4741 Capturer = ref; 4742 } 4743 4744 void VisitBlockExpr(BlockExpr *block) { 4745 // Look inside nested blocks 4746 if (block->getBlockDecl()->capturesVariable(Variable)) 4747 Visit(block->getBlockDecl()->getBody()); 4748 } 4749 }; 4750} 4751 4752/// Check whether the given argument is a block which captures a 4753/// variable. 4754static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 4755 assert(owner.Variable && owner.Loc.isValid()); 4756 4757 e = e->IgnoreParenCasts(); 4758 BlockExpr *block = dyn_cast<BlockExpr>(e); 4759 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 4760 return 0; 4761 4762 FindCaptureVisitor visitor(S.Context, owner.Variable); 4763 visitor.Visit(block->getBlockDecl()->getBody()); 4764 return visitor.Capturer; 4765} 4766 4767static void diagnoseRetainCycle(Sema &S, Expr *capturer, 4768 RetainCycleOwner &owner) { 4769 assert(capturer); 4770 assert(owner.Variable && owner.Loc.isValid()); 4771 4772 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 4773 << owner.Variable << capturer->getSourceRange(); 4774 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 4775 << owner.Indirect << owner.Range; 4776} 4777 4778/// Check for a keyword selector that starts with the word 'add' or 4779/// 'set'. 4780static bool isSetterLikeSelector(Selector sel) { 4781 if (sel.isUnarySelector()) return false; 4782 4783 StringRef str = sel.getNameForSlot(0); 4784 while (!str.empty() && str.front() == '_') str = str.substr(1); 4785 if (str.startswith("set")) 4786 str = str.substr(3); 4787 else if (str.startswith("add")) { 4788 // Specially whitelist 'addOperationWithBlock:'. 4789 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 4790 return false; 4791 str = str.substr(3); 4792 } 4793 else 4794 return false; 4795 4796 if (str.empty()) return true; 4797 return !islower(str.front()); 4798} 4799 4800/// Check a message send to see if it's likely to cause a retain cycle. 4801void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 4802 // Only check instance methods whose selector looks like a setter. 4803 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 4804 return; 4805 4806 // Try to find a variable that the receiver is strongly owned by. 4807 RetainCycleOwner owner; 4808 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 4809 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 4810 return; 4811 } else { 4812 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 4813 owner.Variable = getCurMethodDecl()->getSelfDecl(); 4814 owner.Loc = msg->getSuperLoc(); 4815 owner.Range = msg->getSuperLoc(); 4816 } 4817 4818 // Check whether the receiver is captured by any of the arguments. 4819 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 4820 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 4821 return diagnoseRetainCycle(*this, capturer, owner); 4822} 4823 4824/// Check a property assign to see if it's likely to cause a retain cycle. 4825void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 4826 RetainCycleOwner owner; 4827 if (!findRetainCycleOwner(*this, receiver, owner)) 4828 return; 4829 4830 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 4831 diagnoseRetainCycle(*this, capturer, owner); 4832} 4833 4834bool Sema::checkUnsafeAssigns(SourceLocation Loc, 4835 QualType LHS, Expr *RHS) { 4836 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 4837 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 4838 return false; 4839 // strip off any implicit cast added to get to the one arc-specific 4840 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4841 if (cast->getCastKind() == CK_ARCConsumeObject) { 4842 Diag(Loc, diag::warn_arc_retained_assign) 4843 << (LT == Qualifiers::OCL_ExplicitNone) 4844 << RHS->getSourceRange(); 4845 return true; 4846 } 4847 RHS = cast->getSubExpr(); 4848 } 4849 return false; 4850} 4851 4852void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 4853 Expr *LHS, Expr *RHS) { 4854 QualType LHSType; 4855 // PropertyRef on LHS type need be directly obtained from 4856 // its declaration as it has a PsuedoType. 4857 ObjCPropertyRefExpr *PRE 4858 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 4859 if (PRE && !PRE->isImplicitProperty()) { 4860 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4861 if (PD) 4862 LHSType = PD->getType(); 4863 } 4864 4865 if (LHSType.isNull()) 4866 LHSType = LHS->getType(); 4867 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 4868 return; 4869 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 4870 // FIXME. Check for other life times. 4871 if (LT != Qualifiers::OCL_None) 4872 return; 4873 4874 if (PRE) { 4875 if (PRE->isImplicitProperty()) 4876 return; 4877 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4878 if (!PD) 4879 return; 4880 4881 unsigned Attributes = PD->getPropertyAttributes(); 4882 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 4883 // when 'assign' attribute was not explicitly specified 4884 // by user, ignore it and rely on property type itself 4885 // for lifetime info. 4886 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 4887 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 4888 LHSType->isObjCRetainableType()) 4889 return; 4890 4891 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4892 if (cast->getCastKind() == CK_ARCConsumeObject) { 4893 Diag(Loc, diag::warn_arc_retained_property_assign) 4894 << RHS->getSourceRange(); 4895 return; 4896 } 4897 RHS = cast->getSubExpr(); 4898 } 4899 } 4900 } 4901} 4902 4903//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 4904 4905namespace { 4906bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 4907 SourceLocation StmtLoc, 4908 const NullStmt *Body) { 4909 // Do not warn if the body is a macro that expands to nothing, e.g: 4910 // 4911 // #define CALL(x) 4912 // if (condition) 4913 // CALL(0); 4914 // 4915 if (Body->hasLeadingEmptyMacro()) 4916 return false; 4917 4918 // Get line numbers of statement and body. 4919 bool StmtLineInvalid; 4920 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 4921 &StmtLineInvalid); 4922 if (StmtLineInvalid) 4923 return false; 4924 4925 bool BodyLineInvalid; 4926 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 4927 &BodyLineInvalid); 4928 if (BodyLineInvalid) 4929 return false; 4930 4931 // Warn if null statement and body are on the same line. 4932 if (StmtLine != BodyLine) 4933 return false; 4934 4935 return true; 4936} 4937} // Unnamed namespace 4938 4939void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 4940 const Stmt *Body, 4941 unsigned DiagID) { 4942 // Since this is a syntactic check, don't emit diagnostic for template 4943 // instantiations, this just adds noise. 4944 if (CurrentInstantiationScope) 4945 return; 4946 4947 // The body should be a null statement. 4948 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 4949 if (!NBody) 4950 return; 4951 4952 // Do the usual checks. 4953 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 4954 return; 4955 4956 Diag(NBody->getSemiLoc(), DiagID); 4957 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 4958} 4959 4960void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 4961 const Stmt *PossibleBody) { 4962 assert(!CurrentInstantiationScope); // Ensured by caller 4963 4964 SourceLocation StmtLoc; 4965 const Stmt *Body; 4966 unsigned DiagID; 4967 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 4968 StmtLoc = FS->getRParenLoc(); 4969 Body = FS->getBody(); 4970 DiagID = diag::warn_empty_for_body; 4971 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 4972 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 4973 Body = WS->getBody(); 4974 DiagID = diag::warn_empty_while_body; 4975 } else 4976 return; // Neither `for' nor `while'. 4977 4978 // The body should be a null statement. 4979 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 4980 if (!NBody) 4981 return; 4982 4983 // Skip expensive checks if diagnostic is disabled. 4984 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 4985 DiagnosticsEngine::Ignored) 4986 return; 4987 4988 // Do the usual checks. 4989 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 4990 return; 4991 4992 // `for(...);' and `while(...);' are popular idioms, so in order to keep 4993 // noise level low, emit diagnostics only if for/while is followed by a 4994 // CompoundStmt, e.g.: 4995 // for (int i = 0; i < n; i++); 4996 // { 4997 // a(i); 4998 // } 4999 // or if for/while is followed by a statement with more indentation 5000 // than for/while itself: 5001 // for (int i = 0; i < n; i++); 5002 // a(i); 5003 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 5004 if (!ProbableTypo) { 5005 bool BodyColInvalid; 5006 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 5007 PossibleBody->getLocStart(), 5008 &BodyColInvalid); 5009 if (BodyColInvalid) 5010 return; 5011 5012 bool StmtColInvalid; 5013 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 5014 S->getLocStart(), 5015 &StmtColInvalid); 5016 if (StmtColInvalid) 5017 return; 5018 5019 if (BodyCol > StmtCol) 5020 ProbableTypo = true; 5021 } 5022 5023 if (ProbableTypo) { 5024 Diag(NBody->getSemiLoc(), DiagID); 5025 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5026 } 5027} 5028