SemaChecking.cpp revision 76517426dc8bf7734c07eefc35171a6bfdba1a2b
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.getLangOptions(), 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(2)); // Val2 641 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 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 DRE->getLocation(), 987 NewBuiltinDecl->getType(), 988 DRE->getValueKind()); 989 990 // Set the callee in the CallExpr. 991 // FIXME: This leaks the original parens and implicit casts. 992 ExprResult PromotedCall = UsualUnaryConversions(NewDRE); 993 if (PromotedCall.isInvalid()) 994 return ExprError(); 995 TheCall->setCallee(PromotedCall.take()); 996 997 // Change the result type of the call to match the original value type. This 998 // is arbitrary, but the codegen for these builtins ins design to handle it 999 // gracefully. 1000 TheCall->setType(ResultType); 1001 1002 return move(TheCallResult); 1003} 1004 1005/// CheckObjCString - Checks that the argument to the builtin 1006/// CFString constructor is correct 1007/// Note: It might also make sense to do the UTF-16 conversion here (would 1008/// simplify the backend). 1009bool Sema::CheckObjCString(Expr *Arg) { 1010 Arg = Arg->IgnoreParenCasts(); 1011 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1012 1013 if (!Literal || !Literal->isAscii()) { 1014 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1015 << Arg->getSourceRange(); 1016 return true; 1017 } 1018 1019 if (Literal->containsNonAsciiOrNull()) { 1020 StringRef String = Literal->getString(); 1021 unsigned NumBytes = String.size(); 1022 SmallVector<UTF16, 128> ToBuf(NumBytes); 1023 const UTF8 *FromPtr = (UTF8 *)String.data(); 1024 UTF16 *ToPtr = &ToBuf[0]; 1025 1026 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1027 &ToPtr, ToPtr + NumBytes, 1028 strictConversion); 1029 // Check for conversion failure. 1030 if (Result != conversionOK) 1031 Diag(Arg->getLocStart(), 1032 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1033 } 1034 return false; 1035} 1036 1037/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1038/// Emit an error and return true on failure, return false on success. 1039bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1040 Expr *Fn = TheCall->getCallee(); 1041 if (TheCall->getNumArgs() > 2) { 1042 Diag(TheCall->getArg(2)->getLocStart(), 1043 diag::err_typecheck_call_too_many_args) 1044 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1045 << Fn->getSourceRange() 1046 << SourceRange(TheCall->getArg(2)->getLocStart(), 1047 (*(TheCall->arg_end()-1))->getLocEnd()); 1048 return true; 1049 } 1050 1051 if (TheCall->getNumArgs() < 2) { 1052 return Diag(TheCall->getLocEnd(), 1053 diag::err_typecheck_call_too_few_args_at_least) 1054 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1055 } 1056 1057 // Type-check the first argument normally. 1058 if (checkBuiltinArgument(*this, TheCall, 0)) 1059 return true; 1060 1061 // Determine whether the current function is variadic or not. 1062 BlockScopeInfo *CurBlock = getCurBlock(); 1063 bool isVariadic; 1064 if (CurBlock) 1065 isVariadic = CurBlock->TheDecl->isVariadic(); 1066 else if (FunctionDecl *FD = getCurFunctionDecl()) 1067 isVariadic = FD->isVariadic(); 1068 else 1069 isVariadic = getCurMethodDecl()->isVariadic(); 1070 1071 if (!isVariadic) { 1072 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1073 return true; 1074 } 1075 1076 // Verify that the second argument to the builtin is the last argument of the 1077 // current function or method. 1078 bool SecondArgIsLastNamedArgument = false; 1079 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1080 1081 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1082 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1083 // FIXME: This isn't correct for methods (results in bogus warning). 1084 // Get the last formal in the current function. 1085 const ParmVarDecl *LastArg; 1086 if (CurBlock) 1087 LastArg = *(CurBlock->TheDecl->param_end()-1); 1088 else if (FunctionDecl *FD = getCurFunctionDecl()) 1089 LastArg = *(FD->param_end()-1); 1090 else 1091 LastArg = *(getCurMethodDecl()->param_end()-1); 1092 SecondArgIsLastNamedArgument = PV == LastArg; 1093 } 1094 } 1095 1096 if (!SecondArgIsLastNamedArgument) 1097 Diag(TheCall->getArg(1)->getLocStart(), 1098 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1099 return false; 1100} 1101 1102/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1103/// friends. This is declared to take (...), so we have to check everything. 1104bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1105 if (TheCall->getNumArgs() < 2) 1106 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1107 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1108 if (TheCall->getNumArgs() > 2) 1109 return Diag(TheCall->getArg(2)->getLocStart(), 1110 diag::err_typecheck_call_too_many_args) 1111 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1112 << SourceRange(TheCall->getArg(2)->getLocStart(), 1113 (*(TheCall->arg_end()-1))->getLocEnd()); 1114 1115 ExprResult OrigArg0 = TheCall->getArg(0); 1116 ExprResult OrigArg1 = TheCall->getArg(1); 1117 1118 // Do standard promotions between the two arguments, returning their common 1119 // type. 1120 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1121 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1122 return true; 1123 1124 // Make sure any conversions are pushed back into the call; this is 1125 // type safe since unordered compare builtins are declared as "_Bool 1126 // foo(...)". 1127 TheCall->setArg(0, OrigArg0.get()); 1128 TheCall->setArg(1, OrigArg1.get()); 1129 1130 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1131 return false; 1132 1133 // If the common type isn't a real floating type, then the arguments were 1134 // invalid for this operation. 1135 if (!Res->isRealFloatingType()) 1136 return Diag(OrigArg0.get()->getLocStart(), 1137 diag::err_typecheck_call_invalid_ordered_compare) 1138 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1139 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1140 1141 return false; 1142} 1143 1144/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1145/// __builtin_isnan and friends. This is declared to take (...), so we have 1146/// to check everything. We expect the last argument to be a floating point 1147/// value. 1148bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1149 if (TheCall->getNumArgs() < NumArgs) 1150 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1151 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1152 if (TheCall->getNumArgs() > NumArgs) 1153 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1154 diag::err_typecheck_call_too_many_args) 1155 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1156 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1157 (*(TheCall->arg_end()-1))->getLocEnd()); 1158 1159 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1160 1161 if (OrigArg->isTypeDependent()) 1162 return false; 1163 1164 // This operation requires a non-_Complex floating-point number. 1165 if (!OrigArg->getType()->isRealFloatingType()) 1166 return Diag(OrigArg->getLocStart(), 1167 diag::err_typecheck_call_invalid_unary_fp) 1168 << OrigArg->getType() << OrigArg->getSourceRange(); 1169 1170 // If this is an implicit conversion from float -> double, remove it. 1171 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1172 Expr *CastArg = Cast->getSubExpr(); 1173 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1174 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1175 "promotion from float to double is the only expected cast here"); 1176 Cast->setSubExpr(0); 1177 TheCall->setArg(NumArgs-1, CastArg); 1178 OrigArg = CastArg; 1179 } 1180 } 1181 1182 return false; 1183} 1184 1185/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1186// This is declared to take (...), so we have to check everything. 1187ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1188 if (TheCall->getNumArgs() < 2) 1189 return ExprError(Diag(TheCall->getLocEnd(), 1190 diag::err_typecheck_call_too_few_args_at_least) 1191 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1192 << TheCall->getSourceRange()); 1193 1194 // Determine which of the following types of shufflevector we're checking: 1195 // 1) unary, vector mask: (lhs, mask) 1196 // 2) binary, vector mask: (lhs, rhs, mask) 1197 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1198 QualType resType = TheCall->getArg(0)->getType(); 1199 unsigned numElements = 0; 1200 1201 if (!TheCall->getArg(0)->isTypeDependent() && 1202 !TheCall->getArg(1)->isTypeDependent()) { 1203 QualType LHSType = TheCall->getArg(0)->getType(); 1204 QualType RHSType = TheCall->getArg(1)->getType(); 1205 1206 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1207 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1208 << SourceRange(TheCall->getArg(0)->getLocStart(), 1209 TheCall->getArg(1)->getLocEnd()); 1210 return ExprError(); 1211 } 1212 1213 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1214 unsigned numResElements = TheCall->getNumArgs() - 2; 1215 1216 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1217 // with mask. If so, verify that RHS is an integer vector type with the 1218 // same number of elts as lhs. 1219 if (TheCall->getNumArgs() == 2) { 1220 if (!RHSType->hasIntegerRepresentation() || 1221 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1222 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1223 << SourceRange(TheCall->getArg(1)->getLocStart(), 1224 TheCall->getArg(1)->getLocEnd()); 1225 numResElements = numElements; 1226 } 1227 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1228 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1229 << SourceRange(TheCall->getArg(0)->getLocStart(), 1230 TheCall->getArg(1)->getLocEnd()); 1231 return ExprError(); 1232 } else if (numElements != numResElements) { 1233 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1234 resType = Context.getVectorType(eltType, numResElements, 1235 VectorType::GenericVector); 1236 } 1237 } 1238 1239 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1240 if (TheCall->getArg(i)->isTypeDependent() || 1241 TheCall->getArg(i)->isValueDependent()) 1242 continue; 1243 1244 llvm::APSInt Result(32); 1245 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1246 return ExprError(Diag(TheCall->getLocStart(), 1247 diag::err_shufflevector_nonconstant_argument) 1248 << TheCall->getArg(i)->getSourceRange()); 1249 1250 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1251 return ExprError(Diag(TheCall->getLocStart(), 1252 diag::err_shufflevector_argument_too_large) 1253 << TheCall->getArg(i)->getSourceRange()); 1254 } 1255 1256 SmallVector<Expr*, 32> exprs; 1257 1258 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1259 exprs.push_back(TheCall->getArg(i)); 1260 TheCall->setArg(i, 0); 1261 } 1262 1263 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 1264 exprs.size(), resType, 1265 TheCall->getCallee()->getLocStart(), 1266 TheCall->getRParenLoc())); 1267} 1268 1269/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1270// This is declared to take (const void*, ...) and can take two 1271// optional constant int args. 1272bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1273 unsigned NumArgs = TheCall->getNumArgs(); 1274 1275 if (NumArgs > 3) 1276 return Diag(TheCall->getLocEnd(), 1277 diag::err_typecheck_call_too_many_args_at_most) 1278 << 0 /*function call*/ << 3 << NumArgs 1279 << TheCall->getSourceRange(); 1280 1281 // Argument 0 is checked for us and the remaining arguments must be 1282 // constant integers. 1283 for (unsigned i = 1; i != NumArgs; ++i) { 1284 Expr *Arg = TheCall->getArg(i); 1285 1286 llvm::APSInt Result; 1287 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1288 return true; 1289 1290 // FIXME: gcc issues a warning and rewrites these to 0. These 1291 // seems especially odd for the third argument since the default 1292 // is 3. 1293 if (i == 1) { 1294 if (Result.getLimitedValue() > 1) 1295 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1296 << "0" << "1" << Arg->getSourceRange(); 1297 } else { 1298 if (Result.getLimitedValue() > 3) 1299 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1300 << "0" << "3" << Arg->getSourceRange(); 1301 } 1302 } 1303 1304 return false; 1305} 1306 1307/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1308/// TheCall is a constant expression. 1309bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1310 llvm::APSInt &Result) { 1311 Expr *Arg = TheCall->getArg(ArgNum); 1312 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1313 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1314 1315 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1316 1317 if (!Arg->isIntegerConstantExpr(Result, Context)) 1318 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1319 << FDecl->getDeclName() << Arg->getSourceRange(); 1320 1321 return false; 1322} 1323 1324/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1325/// int type). This simply type checks that type is one of the defined 1326/// constants (0-3). 1327// For compatibility check 0-3, llvm only handles 0 and 2. 1328bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1329 llvm::APSInt Result; 1330 1331 // Check constant-ness first. 1332 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1333 return true; 1334 1335 Expr *Arg = TheCall->getArg(1); 1336 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1337 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1338 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1339 } 1340 1341 return false; 1342} 1343 1344/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1345/// This checks that val is a constant 1. 1346bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1347 Expr *Arg = TheCall->getArg(1); 1348 llvm::APSInt Result; 1349 1350 // TODO: This is less than ideal. Overload this to take a value. 1351 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1352 return true; 1353 1354 if (Result != 1) 1355 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1356 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1357 1358 return false; 1359} 1360 1361// Handle i > 1 ? "x" : "y", recursively. 1362bool Sema::SemaCheckStringLiteral(const Expr *E, Expr **Args, 1363 unsigned NumArgs, bool HasVAListArg, 1364 unsigned format_idx, unsigned firstDataArg, 1365 FormatStringType Type, bool inFunctionCall) { 1366 tryAgain: 1367 if (E->isTypeDependent() || E->isValueDependent()) 1368 return false; 1369 1370 E = E->IgnoreParenCasts(); 1371 1372 if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull)) 1373 // Technically -Wformat-nonliteral does not warn about this case. 1374 // The behavior of printf and friends in this case is implementation 1375 // dependent. Ideally if the format string cannot be null then 1376 // it should have a 'nonnull' attribute in the function prototype. 1377 return true; 1378 1379 switch (E->getStmtClass()) { 1380 case Stmt::BinaryConditionalOperatorClass: 1381 case Stmt::ConditionalOperatorClass: { 1382 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 1383 return SemaCheckStringLiteral(C->getTrueExpr(), Args, NumArgs, HasVAListArg, 1384 format_idx, firstDataArg, Type, 1385 inFunctionCall) 1386 && SemaCheckStringLiteral(C->getFalseExpr(), Args, NumArgs, HasVAListArg, 1387 format_idx, firstDataArg, Type, 1388 inFunctionCall); 1389 } 1390 1391 case Stmt::ImplicitCastExprClass: { 1392 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1393 goto tryAgain; 1394 } 1395 1396 case Stmt::OpaqueValueExprClass: 1397 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1398 E = src; 1399 goto tryAgain; 1400 } 1401 return false; 1402 1403 case Stmt::PredefinedExprClass: 1404 // While __func__, etc., are technically not string literals, they 1405 // cannot contain format specifiers and thus are not a security 1406 // liability. 1407 return true; 1408 1409 case Stmt::DeclRefExprClass: { 1410 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1411 1412 // As an exception, do not flag errors for variables binding to 1413 // const string literals. 1414 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1415 bool isConstant = false; 1416 QualType T = DR->getType(); 1417 1418 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1419 isConstant = AT->getElementType().isConstant(Context); 1420 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1421 isConstant = T.isConstant(Context) && 1422 PT->getPointeeType().isConstant(Context); 1423 } else if (T->isObjCObjectPointerType()) { 1424 // In ObjC, there is usually no "const ObjectPointer" type, 1425 // so don't check if the pointee type is constant. 1426 isConstant = T.isConstant(Context); 1427 } 1428 1429 if (isConstant) { 1430 if (const Expr *Init = VD->getAnyInitializer()) 1431 return SemaCheckStringLiteral(Init, Args, NumArgs, 1432 HasVAListArg, format_idx, firstDataArg, 1433 Type, /*inFunctionCall*/false); 1434 } 1435 1436 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1437 // special check to see if the format string is a function parameter 1438 // of the function calling the printf function. If the function 1439 // has an attribute indicating it is a printf-like function, then we 1440 // should suppress warnings concerning non-literals being used in a call 1441 // to a vprintf function. For example: 1442 // 1443 // void 1444 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1445 // va_list ap; 1446 // va_start(ap, fmt); 1447 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1448 // ... 1449 // 1450 if (HasVAListArg) { 1451 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 1452 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 1453 int PVIndex = PV->getFunctionScopeIndex() + 1; 1454 for (specific_attr_iterator<FormatAttr> 1455 i = ND->specific_attr_begin<FormatAttr>(), 1456 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 1457 FormatAttr *PVFormat = *i; 1458 // adjust for implicit parameter 1459 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1460 if (MD->isInstance()) 1461 ++PVIndex; 1462 // We also check if the formats are compatible. 1463 // We can't pass a 'scanf' string to a 'printf' function. 1464 if (PVIndex == PVFormat->getFormatIdx() && 1465 Type == GetFormatStringType(PVFormat)) 1466 return true; 1467 } 1468 } 1469 } 1470 } 1471 } 1472 1473 return false; 1474 } 1475 1476 case Stmt::CallExprClass: 1477 case Stmt::CXXMemberCallExprClass: { 1478 const CallExpr *CE = cast<CallExpr>(E); 1479 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 1480 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 1481 unsigned ArgIndex = FA->getFormatIdx(); 1482 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1483 if (MD->isInstance()) 1484 --ArgIndex; 1485 const Expr *Arg = CE->getArg(ArgIndex - 1); 1486 1487 return SemaCheckStringLiteral(Arg, Args, NumArgs, HasVAListArg, 1488 format_idx, firstDataArg, Type, 1489 inFunctionCall); 1490 } 1491 } 1492 1493 return false; 1494 } 1495 case Stmt::ObjCStringLiteralClass: 1496 case Stmt::StringLiteralClass: { 1497 const StringLiteral *StrE = NULL; 1498 1499 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1500 StrE = ObjCFExpr->getString(); 1501 else 1502 StrE = cast<StringLiteral>(E); 1503 1504 if (StrE) { 1505 CheckFormatString(StrE, E, Args, NumArgs, HasVAListArg, format_idx, 1506 firstDataArg, Type, inFunctionCall); 1507 return true; 1508 } 1509 1510 return false; 1511 } 1512 1513 default: 1514 return false; 1515 } 1516} 1517 1518void 1519Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1520 const Expr * const *ExprArgs, 1521 SourceLocation CallSiteLoc) { 1522 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1523 e = NonNull->args_end(); 1524 i != e; ++i) { 1525 const Expr *ArgExpr = ExprArgs[*i]; 1526 if (ArgExpr->isNullPointerConstant(Context, 1527 Expr::NPC_ValueDependentIsNotNull)) 1528 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1529 } 1530} 1531 1532Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 1533 return llvm::StringSwitch<FormatStringType>(Format->getType()) 1534 .Case("scanf", FST_Scanf) 1535 .Cases("printf", "printf0", FST_Printf) 1536 .Cases("NSString", "CFString", FST_NSString) 1537 .Case("strftime", FST_Strftime) 1538 .Case("strfmon", FST_Strfmon) 1539 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 1540 .Default(FST_Unknown); 1541} 1542 1543/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1544/// functions) for correct use of format strings. 1545void Sema::CheckFormatArguments(const FormatAttr *Format, CallExpr *TheCall) { 1546 bool IsCXXMember = false; 1547 // The way the format attribute works in GCC, the implicit this argument 1548 // of member functions is counted. However, it doesn't appear in our own 1549 // lists, so decrement format_idx in that case. 1550 IsCXXMember = isa<CXXMemberCallExpr>(TheCall); 1551 CheckFormatArguments(Format, TheCall->getArgs(), TheCall->getNumArgs(), 1552 IsCXXMember, TheCall->getRParenLoc(), 1553 TheCall->getCallee()->getSourceRange()); 1554} 1555 1556void Sema::CheckFormatArguments(const FormatAttr *Format, Expr **Args, 1557 unsigned NumArgs, bool IsCXXMember, 1558 SourceLocation Loc, SourceRange Range) { 1559 bool HasVAListArg = Format->getFirstArg() == 0; 1560 unsigned format_idx = Format->getFormatIdx() - 1; 1561 unsigned firstDataArg = HasVAListArg ? 0 : Format->getFirstArg() - 1; 1562 if (IsCXXMember) { 1563 if (format_idx == 0) 1564 return; 1565 --format_idx; 1566 if(firstDataArg != 0) 1567 --firstDataArg; 1568 } 1569 CheckFormatArguments(Args, NumArgs, HasVAListArg, format_idx, 1570 firstDataArg, GetFormatStringType(Format), Loc, Range); 1571} 1572 1573void Sema::CheckFormatArguments(Expr **Args, unsigned NumArgs, 1574 bool HasVAListArg, unsigned format_idx, 1575 unsigned firstDataArg, FormatStringType Type, 1576 SourceLocation Loc, SourceRange Range) { 1577 // CHECK: printf/scanf-like function is called with no format string. 1578 if (format_idx >= NumArgs) { 1579 Diag(Loc, diag::warn_missing_format_string) << Range; 1580 return; 1581 } 1582 1583 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1584 1585 // CHECK: format string is not a string literal. 1586 // 1587 // Dynamically generated format strings are difficult to 1588 // automatically vet at compile time. Requiring that format strings 1589 // are string literals: (1) permits the checking of format strings by 1590 // the compiler and thereby (2) can practically remove the source of 1591 // many format string exploits. 1592 1593 // Format string can be either ObjC string (e.g. @"%d") or 1594 // C string (e.g. "%d") 1595 // ObjC string uses the same format specifiers as C string, so we can use 1596 // the same format string checking logic for both ObjC and C strings. 1597 if (SemaCheckStringLiteral(OrigFormatExpr, Args, NumArgs, HasVAListArg, 1598 format_idx, firstDataArg, Type)) 1599 return; // Literal format string found, check done! 1600 1601 // Strftime is particular as it always uses a single 'time' argument, 1602 // so it is safe to pass a non-literal string. 1603 if (Type == FST_Strftime) 1604 return; 1605 1606 // Do not emit diag when the string param is a macro expansion and the 1607 // format is either NSString or CFString. This is a hack to prevent 1608 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 1609 // which are usually used in place of NS and CF string literals. 1610 if (Type == FST_NSString && Args[format_idx]->getLocStart().isMacroID()) 1611 return; 1612 1613 // If there are no arguments specified, warn with -Wformat-security, otherwise 1614 // warn only with -Wformat-nonliteral. 1615 if (NumArgs == format_idx+1) 1616 Diag(Args[format_idx]->getLocStart(), 1617 diag::warn_format_nonliteral_noargs) 1618 << OrigFormatExpr->getSourceRange(); 1619 else 1620 Diag(Args[format_idx]->getLocStart(), 1621 diag::warn_format_nonliteral) 1622 << OrigFormatExpr->getSourceRange(); 1623} 1624 1625namespace { 1626class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1627protected: 1628 Sema &S; 1629 const StringLiteral *FExpr; 1630 const Expr *OrigFormatExpr; 1631 const unsigned FirstDataArg; 1632 const unsigned NumDataArgs; 1633 const bool IsObjCLiteral; 1634 const char *Beg; // Start of format string. 1635 const bool HasVAListArg; 1636 const Expr * const *Args; 1637 const unsigned NumArgs; 1638 unsigned FormatIdx; 1639 llvm::BitVector CoveredArgs; 1640 bool usesPositionalArgs; 1641 bool atFirstArg; 1642 bool inFunctionCall; 1643public: 1644 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1645 const Expr *origFormatExpr, unsigned firstDataArg, 1646 unsigned numDataArgs, bool isObjCLiteral, 1647 const char *beg, bool hasVAListArg, 1648 Expr **args, unsigned numArgs, 1649 unsigned formatIdx, bool inFunctionCall) 1650 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1651 FirstDataArg(firstDataArg), 1652 NumDataArgs(numDataArgs), 1653 IsObjCLiteral(isObjCLiteral), Beg(beg), 1654 HasVAListArg(hasVAListArg), 1655 Args(args), NumArgs(numArgs), FormatIdx(formatIdx), 1656 usesPositionalArgs(false), atFirstArg(true), 1657 inFunctionCall(inFunctionCall) { 1658 CoveredArgs.resize(numDataArgs); 1659 CoveredArgs.reset(); 1660 } 1661 1662 void DoneProcessing(); 1663 1664 void HandleIncompleteSpecifier(const char *startSpecifier, 1665 unsigned specifierLen); 1666 1667 void HandleNonStandardLengthModifier( 1668 const analyze_format_string::LengthModifier &LM, 1669 const char *startSpecifier, unsigned specifierLen); 1670 1671 void HandleNonStandardConversionSpecifier( 1672 const analyze_format_string::ConversionSpecifier &CS, 1673 const char *startSpecifier, unsigned specifierLen); 1674 1675 void HandleNonStandardConversionSpecification( 1676 const analyze_format_string::LengthModifier &LM, 1677 const analyze_format_string::ConversionSpecifier &CS, 1678 const char *startSpecifier, unsigned specifierLen); 1679 1680 virtual void HandleInvalidPosition(const char *startSpecifier, 1681 unsigned specifierLen, 1682 analyze_format_string::PositionContext p); 1683 1684 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1685 1686 void HandleNullChar(const char *nullCharacter); 1687 1688 template <typename Range> 1689 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 1690 const Expr *ArgumentExpr, 1691 PartialDiagnostic PDiag, 1692 SourceLocation StringLoc, 1693 bool IsStringLocation, Range StringRange, 1694 FixItHint Fixit = FixItHint()); 1695 1696protected: 1697 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1698 const char *startSpec, 1699 unsigned specifierLen, 1700 const char *csStart, unsigned csLen); 1701 1702 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 1703 const char *startSpec, 1704 unsigned specifierLen); 1705 1706 SourceRange getFormatStringRange(); 1707 CharSourceRange getSpecifierRange(const char *startSpecifier, 1708 unsigned specifierLen); 1709 SourceLocation getLocationOfByte(const char *x); 1710 1711 const Expr *getDataArg(unsigned i) const; 1712 1713 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1714 const analyze_format_string::ConversionSpecifier &CS, 1715 const char *startSpecifier, unsigned specifierLen, 1716 unsigned argIndex); 1717 1718 template <typename Range> 1719 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 1720 bool IsStringLocation, Range StringRange, 1721 FixItHint Fixit = FixItHint()); 1722 1723 void CheckPositionalAndNonpositionalArgs( 1724 const analyze_format_string::FormatSpecifier *FS); 1725}; 1726} 1727 1728SourceRange CheckFormatHandler::getFormatStringRange() { 1729 return OrigFormatExpr->getSourceRange(); 1730} 1731 1732CharSourceRange CheckFormatHandler:: 1733getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1734 SourceLocation Start = getLocationOfByte(startSpecifier); 1735 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1736 1737 // Advance the end SourceLocation by one due to half-open ranges. 1738 End = End.getLocWithOffset(1); 1739 1740 return CharSourceRange::getCharRange(Start, End); 1741} 1742 1743SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1744 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1745} 1746 1747void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1748 unsigned specifierLen){ 1749 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 1750 getLocationOfByte(startSpecifier), 1751 /*IsStringLocation*/true, 1752 getSpecifierRange(startSpecifier, specifierLen)); 1753} 1754 1755void CheckFormatHandler::HandleNonStandardLengthModifier( 1756 const analyze_format_string::LengthModifier &LM, 1757 const char *startSpecifier, unsigned specifierLen) { 1758 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) << LM.toString() 1759 << "length modifier", 1760 getLocationOfByte(LM.getStart()), 1761 /*IsStringLocation*/true, 1762 getSpecifierRange(startSpecifier, specifierLen)); 1763} 1764 1765void CheckFormatHandler::HandleNonStandardConversionSpecifier( 1766 const analyze_format_string::ConversionSpecifier &CS, 1767 const char *startSpecifier, unsigned specifierLen) { 1768 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) << CS.toString() 1769 << "conversion specifier", 1770 getLocationOfByte(CS.getStart()), 1771 /*IsStringLocation*/true, 1772 getSpecifierRange(startSpecifier, specifierLen)); 1773} 1774 1775void CheckFormatHandler::HandleNonStandardConversionSpecification( 1776 const analyze_format_string::LengthModifier &LM, 1777 const analyze_format_string::ConversionSpecifier &CS, 1778 const char *startSpecifier, unsigned specifierLen) { 1779 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_conversion_spec) 1780 << LM.toString() << CS.toString(), 1781 getLocationOfByte(LM.getStart()), 1782 /*IsStringLocation*/true, 1783 getSpecifierRange(startSpecifier, specifierLen)); 1784} 1785 1786void 1787CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1788 analyze_format_string::PositionContext p) { 1789 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 1790 << (unsigned) p, 1791 getLocationOfByte(startPos), /*IsStringLocation*/true, 1792 getSpecifierRange(startPos, posLen)); 1793} 1794 1795void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1796 unsigned posLen) { 1797 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 1798 getLocationOfByte(startPos), 1799 /*IsStringLocation*/true, 1800 getSpecifierRange(startPos, posLen)); 1801} 1802 1803void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1804 if (!IsObjCLiteral) { 1805 // The presence of a null character is likely an error. 1806 EmitFormatDiagnostic( 1807 S.PDiag(diag::warn_printf_format_string_contains_null_char), 1808 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 1809 getFormatStringRange()); 1810 } 1811} 1812 1813const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1814 return Args[FirstDataArg + i]; 1815} 1816 1817void CheckFormatHandler::DoneProcessing() { 1818 // Does the number of data arguments exceed the number of 1819 // format conversions in the format string? 1820 if (!HasVAListArg) { 1821 // Find any arguments that weren't covered. 1822 CoveredArgs.flip(); 1823 signed notCoveredArg = CoveredArgs.find_first(); 1824 if (notCoveredArg >= 0) { 1825 assert((unsigned)notCoveredArg < NumDataArgs); 1826 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 1827 getDataArg((unsigned) notCoveredArg)->getLocStart(), 1828 /*IsStringLocation*/false, getFormatStringRange()); 1829 } 1830 } 1831} 1832 1833bool 1834CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1835 SourceLocation Loc, 1836 const char *startSpec, 1837 unsigned specifierLen, 1838 const char *csStart, 1839 unsigned csLen) { 1840 1841 bool keepGoing = true; 1842 if (argIndex < NumDataArgs) { 1843 // Consider the argument coverered, even though the specifier doesn't 1844 // make sense. 1845 CoveredArgs.set(argIndex); 1846 } 1847 else { 1848 // If argIndex exceeds the number of data arguments we 1849 // don't issue a warning because that is just a cascade of warnings (and 1850 // they may have intended '%%' anyway). We don't want to continue processing 1851 // the format string after this point, however, as we will like just get 1852 // gibberish when trying to match arguments. 1853 keepGoing = false; 1854 } 1855 1856 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 1857 << StringRef(csStart, csLen), 1858 Loc, /*IsStringLocation*/true, 1859 getSpecifierRange(startSpec, specifierLen)); 1860 1861 return keepGoing; 1862} 1863 1864void 1865CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 1866 const char *startSpec, 1867 unsigned specifierLen) { 1868 EmitFormatDiagnostic( 1869 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 1870 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 1871} 1872 1873bool 1874CheckFormatHandler::CheckNumArgs( 1875 const analyze_format_string::FormatSpecifier &FS, 1876 const analyze_format_string::ConversionSpecifier &CS, 1877 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1878 1879 if (argIndex >= NumDataArgs) { 1880 PartialDiagnostic PDiag = FS.usesPositionalArg() 1881 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 1882 << (argIndex+1) << NumDataArgs) 1883 : S.PDiag(diag::warn_printf_insufficient_data_args); 1884 EmitFormatDiagnostic( 1885 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 1886 getSpecifierRange(startSpecifier, specifierLen)); 1887 return false; 1888 } 1889 return true; 1890} 1891 1892template<typename Range> 1893void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 1894 SourceLocation Loc, 1895 bool IsStringLocation, 1896 Range StringRange, 1897 FixItHint FixIt) { 1898 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 1899 Loc, IsStringLocation, StringRange, FixIt); 1900} 1901 1902/// \brief If the format string is not within the funcion call, emit a note 1903/// so that the function call and string are in diagnostic messages. 1904/// 1905/// \param inFunctionCall if true, the format string is within the function 1906/// call and only one diagnostic message will be produced. Otherwise, an 1907/// extra note will be emitted pointing to location of the format string. 1908/// 1909/// \param ArgumentExpr the expression that is passed as the format string 1910/// argument in the function call. Used for getting locations when two 1911/// diagnostics are emitted. 1912/// 1913/// \param PDiag the callee should already have provided any strings for the 1914/// diagnostic message. This function only adds locations and fixits 1915/// to diagnostics. 1916/// 1917/// \param Loc primary location for diagnostic. If two diagnostics are 1918/// required, one will be at Loc and a new SourceLocation will be created for 1919/// the other one. 1920/// 1921/// \param IsStringLocation if true, Loc points to the format string should be 1922/// used for the note. Otherwise, Loc points to the argument list and will 1923/// be used with PDiag. 1924/// 1925/// \param StringRange some or all of the string to highlight. This is 1926/// templated so it can accept either a CharSourceRange or a SourceRange. 1927/// 1928/// \param Fixit optional fix it hint for the format string. 1929template<typename Range> 1930void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 1931 const Expr *ArgumentExpr, 1932 PartialDiagnostic PDiag, 1933 SourceLocation Loc, 1934 bool IsStringLocation, 1935 Range StringRange, 1936 FixItHint FixIt) { 1937 if (InFunctionCall) 1938 S.Diag(Loc, PDiag) << StringRange << FixIt; 1939 else { 1940 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 1941 << ArgumentExpr->getSourceRange(); 1942 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 1943 diag::note_format_string_defined) 1944 << StringRange << FixIt; 1945 } 1946} 1947 1948//===--- CHECK: Printf format string checking ------------------------------===// 1949 1950namespace { 1951class CheckPrintfHandler : public CheckFormatHandler { 1952public: 1953 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1954 const Expr *origFormatExpr, unsigned firstDataArg, 1955 unsigned numDataArgs, bool isObjCLiteral, 1956 const char *beg, bool hasVAListArg, 1957 Expr **Args, unsigned NumArgs, 1958 unsigned formatIdx, bool inFunctionCall) 1959 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1960 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1961 Args, NumArgs, formatIdx, inFunctionCall) {} 1962 1963 1964 bool HandleInvalidPrintfConversionSpecifier( 1965 const analyze_printf::PrintfSpecifier &FS, 1966 const char *startSpecifier, 1967 unsigned specifierLen); 1968 1969 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1970 const char *startSpecifier, 1971 unsigned specifierLen); 1972 1973 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1974 const char *startSpecifier, unsigned specifierLen); 1975 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1976 const analyze_printf::OptionalAmount &Amt, 1977 unsigned type, 1978 const char *startSpecifier, unsigned specifierLen); 1979 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1980 const analyze_printf::OptionalFlag &flag, 1981 const char *startSpecifier, unsigned specifierLen); 1982 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1983 const analyze_printf::OptionalFlag &ignoredFlag, 1984 const analyze_printf::OptionalFlag &flag, 1985 const char *startSpecifier, unsigned specifierLen); 1986}; 1987} 1988 1989bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1990 const analyze_printf::PrintfSpecifier &FS, 1991 const char *startSpecifier, 1992 unsigned specifierLen) { 1993 const analyze_printf::PrintfConversionSpecifier &CS = 1994 FS.getConversionSpecifier(); 1995 1996 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1997 getLocationOfByte(CS.getStart()), 1998 startSpecifier, specifierLen, 1999 CS.getStart(), CS.getLength()); 2000} 2001 2002bool CheckPrintfHandler::HandleAmount( 2003 const analyze_format_string::OptionalAmount &Amt, 2004 unsigned k, const char *startSpecifier, 2005 unsigned specifierLen) { 2006 2007 if (Amt.hasDataArgument()) { 2008 if (!HasVAListArg) { 2009 unsigned argIndex = Amt.getArgIndex(); 2010 if (argIndex >= NumDataArgs) { 2011 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2012 << k, 2013 getLocationOfByte(Amt.getStart()), 2014 /*IsStringLocation*/true, 2015 getSpecifierRange(startSpecifier, specifierLen)); 2016 // Don't do any more checking. We will just emit 2017 // spurious errors. 2018 return false; 2019 } 2020 2021 // Type check the data argument. It should be an 'int'. 2022 // Although not in conformance with C99, we also allow the argument to be 2023 // an 'unsigned int' as that is a reasonably safe case. GCC also 2024 // doesn't emit a warning for that case. 2025 CoveredArgs.set(argIndex); 2026 const Expr *Arg = getDataArg(argIndex); 2027 QualType T = Arg->getType(); 2028 2029 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 2030 assert(ATR.isValid()); 2031 2032 if (!ATR.matchesType(S.Context, T)) { 2033 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2034 << k << ATR.getRepresentativeTypeName(S.Context) 2035 << T << Arg->getSourceRange(), 2036 getLocationOfByte(Amt.getStart()), 2037 /*IsStringLocation*/true, 2038 getSpecifierRange(startSpecifier, specifierLen)); 2039 // Don't do any more checking. We will just emit 2040 // spurious errors. 2041 return false; 2042 } 2043 } 2044 } 2045 return true; 2046} 2047 2048void CheckPrintfHandler::HandleInvalidAmount( 2049 const analyze_printf::PrintfSpecifier &FS, 2050 const analyze_printf::OptionalAmount &Amt, 2051 unsigned type, 2052 const char *startSpecifier, 2053 unsigned specifierLen) { 2054 const analyze_printf::PrintfConversionSpecifier &CS = 2055 FS.getConversionSpecifier(); 2056 2057 FixItHint fixit = 2058 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2059 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2060 Amt.getConstantLength())) 2061 : FixItHint(); 2062 2063 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2064 << type << CS.toString(), 2065 getLocationOfByte(Amt.getStart()), 2066 /*IsStringLocation*/true, 2067 getSpecifierRange(startSpecifier, specifierLen), 2068 fixit); 2069} 2070 2071void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2072 const analyze_printf::OptionalFlag &flag, 2073 const char *startSpecifier, 2074 unsigned specifierLen) { 2075 // Warn about pointless flag with a fixit removal. 2076 const analyze_printf::PrintfConversionSpecifier &CS = 2077 FS.getConversionSpecifier(); 2078 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2079 << flag.toString() << CS.toString(), 2080 getLocationOfByte(flag.getPosition()), 2081 /*IsStringLocation*/true, 2082 getSpecifierRange(startSpecifier, specifierLen), 2083 FixItHint::CreateRemoval( 2084 getSpecifierRange(flag.getPosition(), 1))); 2085} 2086 2087void CheckPrintfHandler::HandleIgnoredFlag( 2088 const analyze_printf::PrintfSpecifier &FS, 2089 const analyze_printf::OptionalFlag &ignoredFlag, 2090 const analyze_printf::OptionalFlag &flag, 2091 const char *startSpecifier, 2092 unsigned specifierLen) { 2093 // Warn about ignored flag with a fixit removal. 2094 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2095 << ignoredFlag.toString() << flag.toString(), 2096 getLocationOfByte(ignoredFlag.getPosition()), 2097 /*IsStringLocation*/true, 2098 getSpecifierRange(startSpecifier, specifierLen), 2099 FixItHint::CreateRemoval( 2100 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2101} 2102 2103bool 2104CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2105 &FS, 2106 const char *startSpecifier, 2107 unsigned specifierLen) { 2108 2109 using namespace analyze_format_string; 2110 using namespace analyze_printf; 2111 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2112 2113 if (FS.consumesDataArgument()) { 2114 if (atFirstArg) { 2115 atFirstArg = false; 2116 usesPositionalArgs = FS.usesPositionalArg(); 2117 } 2118 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2119 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2120 startSpecifier, specifierLen); 2121 return false; 2122 } 2123 } 2124 2125 // First check if the field width, precision, and conversion specifier 2126 // have matching data arguments. 2127 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2128 startSpecifier, specifierLen)) { 2129 return false; 2130 } 2131 2132 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2133 startSpecifier, specifierLen)) { 2134 return false; 2135 } 2136 2137 if (!CS.consumesDataArgument()) { 2138 // FIXME: Technically specifying a precision or field width here 2139 // makes no sense. Worth issuing a warning at some point. 2140 return true; 2141 } 2142 2143 // Consume the argument. 2144 unsigned argIndex = FS.getArgIndex(); 2145 if (argIndex < NumDataArgs) { 2146 // The check to see if the argIndex is valid will come later. 2147 // We set the bit here because we may exit early from this 2148 // function if we encounter some other error. 2149 CoveredArgs.set(argIndex); 2150 } 2151 2152 // Check for using an Objective-C specific conversion specifier 2153 // in a non-ObjC literal. 2154 if (!IsObjCLiteral && CS.isObjCArg()) { 2155 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2156 specifierLen); 2157 } 2158 2159 // Check for invalid use of field width 2160 if (!FS.hasValidFieldWidth()) { 2161 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2162 startSpecifier, specifierLen); 2163 } 2164 2165 // Check for invalid use of precision 2166 if (!FS.hasValidPrecision()) { 2167 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2168 startSpecifier, specifierLen); 2169 } 2170 2171 // Check each flag does not conflict with any other component. 2172 if (!FS.hasValidThousandsGroupingPrefix()) 2173 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2174 if (!FS.hasValidLeadingZeros()) 2175 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2176 if (!FS.hasValidPlusPrefix()) 2177 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2178 if (!FS.hasValidSpacePrefix()) 2179 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2180 if (!FS.hasValidAlternativeForm()) 2181 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2182 if (!FS.hasValidLeftJustified()) 2183 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2184 2185 // Check that flags are not ignored by another flag 2186 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2187 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2188 startSpecifier, specifierLen); 2189 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2190 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2191 startSpecifier, specifierLen); 2192 2193 // Check the length modifier is valid with the given conversion specifier. 2194 const LengthModifier &LM = FS.getLengthModifier(); 2195 if (!FS.hasValidLengthModifier()) 2196 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2197 << LM.toString() << CS.toString(), 2198 getLocationOfByte(LM.getStart()), 2199 /*IsStringLocation*/true, 2200 getSpecifierRange(startSpecifier, specifierLen), 2201 FixItHint::CreateRemoval( 2202 getSpecifierRange(LM.getStart(), 2203 LM.getLength()))); 2204 if (!FS.hasStandardLengthModifier()) 2205 HandleNonStandardLengthModifier(LM, startSpecifier, specifierLen); 2206 if (!FS.hasStandardConversionSpecifier(S.getLangOptions())) 2207 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2208 if (!FS.hasStandardLengthConversionCombination()) 2209 HandleNonStandardConversionSpecification(LM, CS, startSpecifier, 2210 specifierLen); 2211 2212 // Are we using '%n'? 2213 if (CS.getKind() == ConversionSpecifier::nArg) { 2214 // Issue a warning about this being a possible security issue. 2215 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_write_back), 2216 getLocationOfByte(CS.getStart()), 2217 /*IsStringLocation*/true, 2218 getSpecifierRange(startSpecifier, specifierLen)); 2219 // Continue checking the other format specifiers. 2220 return true; 2221 } 2222 2223 // The remaining checks depend on the data arguments. 2224 if (HasVAListArg) 2225 return true; 2226 2227 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2228 return false; 2229 2230 // Now type check the data expression that matches the 2231 // format specifier. 2232 const Expr *Ex = getDataArg(argIndex); 2233 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context, 2234 IsObjCLiteral); 2235 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2236 // Check if we didn't match because of an implicit cast from a 'char' 2237 // or 'short' to an 'int'. This is done because printf is a varargs 2238 // function. 2239 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 2240 if (ICE->getType() == S.Context.IntTy) { 2241 // All further checking is done on the subexpression. 2242 Ex = ICE->getSubExpr(); 2243 if (ATR.matchesType(S.Context, Ex->getType())) 2244 return true; 2245 } 2246 2247 // We may be able to offer a FixItHint if it is a supported type. 2248 PrintfSpecifier fixedFS = FS; 2249 bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions(), 2250 S.Context, IsObjCLiteral); 2251 2252 if (success) { 2253 // Get the fix string from the fixed format specifier 2254 SmallString<128> buf; 2255 llvm::raw_svector_ostream os(buf); 2256 fixedFS.toString(os); 2257 2258 EmitFormatDiagnostic( 2259 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2260 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2261 << Ex->getSourceRange(), 2262 getLocationOfByte(CS.getStart()), 2263 /*IsStringLocation*/true, 2264 getSpecifierRange(startSpecifier, specifierLen), 2265 FixItHint::CreateReplacement( 2266 getSpecifierRange(startSpecifier, specifierLen), 2267 os.str())); 2268 } 2269 else { 2270 EmitFormatDiagnostic( 2271 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2272 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2273 << getSpecifierRange(startSpecifier, specifierLen) 2274 << Ex->getSourceRange(), 2275 getLocationOfByte(CS.getStart()), 2276 true, 2277 getSpecifierRange(startSpecifier, specifierLen)); 2278 } 2279 } 2280 2281 return true; 2282} 2283 2284//===--- CHECK: Scanf format string checking ------------------------------===// 2285 2286namespace { 2287class CheckScanfHandler : public CheckFormatHandler { 2288public: 2289 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2290 const Expr *origFormatExpr, unsigned firstDataArg, 2291 unsigned numDataArgs, bool isObjCLiteral, 2292 const char *beg, bool hasVAListArg, 2293 Expr **Args, unsigned NumArgs, 2294 unsigned formatIdx, bool inFunctionCall) 2295 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2296 numDataArgs, isObjCLiteral, beg, hasVAListArg, 2297 Args, NumArgs, formatIdx, inFunctionCall) {} 2298 2299 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2300 const char *startSpecifier, 2301 unsigned specifierLen); 2302 2303 bool HandleInvalidScanfConversionSpecifier( 2304 const analyze_scanf::ScanfSpecifier &FS, 2305 const char *startSpecifier, 2306 unsigned specifierLen); 2307 2308 void HandleIncompleteScanList(const char *start, const char *end); 2309}; 2310} 2311 2312void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2313 const char *end) { 2314 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2315 getLocationOfByte(end), /*IsStringLocation*/true, 2316 getSpecifierRange(start, end - start)); 2317} 2318 2319bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2320 const analyze_scanf::ScanfSpecifier &FS, 2321 const char *startSpecifier, 2322 unsigned specifierLen) { 2323 2324 const analyze_scanf::ScanfConversionSpecifier &CS = 2325 FS.getConversionSpecifier(); 2326 2327 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2328 getLocationOfByte(CS.getStart()), 2329 startSpecifier, specifierLen, 2330 CS.getStart(), CS.getLength()); 2331} 2332 2333bool CheckScanfHandler::HandleScanfSpecifier( 2334 const analyze_scanf::ScanfSpecifier &FS, 2335 const char *startSpecifier, 2336 unsigned specifierLen) { 2337 2338 using namespace analyze_scanf; 2339 using namespace analyze_format_string; 2340 2341 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2342 2343 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2344 // be used to decide if we are using positional arguments consistently. 2345 if (FS.consumesDataArgument()) { 2346 if (atFirstArg) { 2347 atFirstArg = false; 2348 usesPositionalArgs = FS.usesPositionalArg(); 2349 } 2350 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2351 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2352 startSpecifier, specifierLen); 2353 return false; 2354 } 2355 } 2356 2357 // Check if the field with is non-zero. 2358 const OptionalAmount &Amt = FS.getFieldWidth(); 2359 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2360 if (Amt.getConstantAmount() == 0) { 2361 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2362 Amt.getConstantLength()); 2363 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 2364 getLocationOfByte(Amt.getStart()), 2365 /*IsStringLocation*/true, R, 2366 FixItHint::CreateRemoval(R)); 2367 } 2368 } 2369 2370 if (!FS.consumesDataArgument()) { 2371 // FIXME: Technically specifying a precision or field width here 2372 // makes no sense. Worth issuing a warning at some point. 2373 return true; 2374 } 2375 2376 // Consume the argument. 2377 unsigned argIndex = FS.getArgIndex(); 2378 if (argIndex < NumDataArgs) { 2379 // The check to see if the argIndex is valid will come later. 2380 // We set the bit here because we may exit early from this 2381 // function if we encounter some other error. 2382 CoveredArgs.set(argIndex); 2383 } 2384 2385 // Check the length modifier is valid with the given conversion specifier. 2386 const LengthModifier &LM = FS.getLengthModifier(); 2387 if (!FS.hasValidLengthModifier()) { 2388 const CharSourceRange &R = getSpecifierRange(LM.getStart(), LM.getLength()); 2389 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2390 << LM.toString() << CS.toString() 2391 << getSpecifierRange(startSpecifier, specifierLen), 2392 getLocationOfByte(LM.getStart()), 2393 /*IsStringLocation*/true, R, 2394 FixItHint::CreateRemoval(R)); 2395 } 2396 2397 if (!FS.hasStandardLengthModifier()) 2398 HandleNonStandardLengthModifier(LM, startSpecifier, specifierLen); 2399 if (!FS.hasStandardConversionSpecifier(S.getLangOptions())) 2400 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2401 if (!FS.hasStandardLengthConversionCombination()) 2402 HandleNonStandardConversionSpecification(LM, CS, startSpecifier, 2403 specifierLen); 2404 2405 // The remaining checks depend on the data arguments. 2406 if (HasVAListArg) 2407 return true; 2408 2409 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2410 return false; 2411 2412 // Check that the argument type matches the format specifier. 2413 const Expr *Ex = getDataArg(argIndex); 2414 const analyze_scanf::ScanfArgTypeResult &ATR = FS.getArgType(S.Context); 2415 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2416 ScanfSpecifier fixedFS = FS; 2417 bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions(), 2418 S.Context); 2419 2420 if (success) { 2421 // Get the fix string from the fixed format specifier. 2422 SmallString<128> buf; 2423 llvm::raw_svector_ostream os(buf); 2424 fixedFS.toString(os); 2425 2426 EmitFormatDiagnostic( 2427 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2428 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2429 << Ex->getSourceRange(), 2430 getLocationOfByte(CS.getStart()), 2431 /*IsStringLocation*/true, 2432 getSpecifierRange(startSpecifier, specifierLen), 2433 FixItHint::CreateReplacement( 2434 getSpecifierRange(startSpecifier, specifierLen), 2435 os.str())); 2436 } else { 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 } 2445 } 2446 2447 return true; 2448} 2449 2450void Sema::CheckFormatString(const StringLiteral *FExpr, 2451 const Expr *OrigFormatExpr, 2452 Expr **Args, unsigned NumArgs, 2453 bool HasVAListArg, unsigned format_idx, 2454 unsigned firstDataArg, FormatStringType Type, 2455 bool inFunctionCall) { 2456 2457 // CHECK: is the format string a wide literal? 2458 if (!FExpr->isAscii()) { 2459 CheckFormatHandler::EmitFormatDiagnostic( 2460 *this, inFunctionCall, Args[format_idx], 2461 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 2462 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2463 return; 2464 } 2465 2466 // Str - The format string. NOTE: this is NOT null-terminated! 2467 StringRef StrRef = FExpr->getString(); 2468 const char *Str = StrRef.data(); 2469 unsigned StrLen = StrRef.size(); 2470 const unsigned numDataArgs = NumArgs - firstDataArg; 2471 2472 // CHECK: empty format string? 2473 if (StrLen == 0 && numDataArgs > 0) { 2474 CheckFormatHandler::EmitFormatDiagnostic( 2475 *this, inFunctionCall, Args[format_idx], 2476 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 2477 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2478 return; 2479 } 2480 2481 if (Type == FST_Printf || Type == FST_NSString) { 2482 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2483 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2484 Str, HasVAListArg, Args, NumArgs, format_idx, 2485 inFunctionCall); 2486 2487 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 2488 getLangOptions())) 2489 H.DoneProcessing(); 2490 } else if (Type == FST_Scanf) { 2491 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2492 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2493 Str, HasVAListArg, Args, NumArgs, format_idx, 2494 inFunctionCall); 2495 2496 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 2497 getLangOptions())) 2498 H.DoneProcessing(); 2499 } // TODO: handle other formats 2500} 2501 2502//===--- CHECK: Standard memory functions ---------------------------------===// 2503 2504/// \brief Determine whether the given type is a dynamic class type (e.g., 2505/// whether it has a vtable). 2506static bool isDynamicClassType(QualType T) { 2507 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2508 if (CXXRecordDecl *Definition = Record->getDefinition()) 2509 if (Definition->isDynamicClass()) 2510 return true; 2511 2512 return false; 2513} 2514 2515/// \brief If E is a sizeof expression, returns its argument expression, 2516/// otherwise returns NULL. 2517static const Expr *getSizeOfExprArg(const Expr* E) { 2518 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2519 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2520 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2521 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2522 2523 return 0; 2524} 2525 2526/// \brief If E is a sizeof expression, returns its argument type. 2527static QualType getSizeOfArgType(const Expr* E) { 2528 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2529 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2530 if (SizeOf->getKind() == clang::UETT_SizeOf) 2531 return SizeOf->getTypeOfArgument(); 2532 2533 return QualType(); 2534} 2535 2536/// \brief Check for dangerous or invalid arguments to memset(). 2537/// 2538/// This issues warnings on known problematic, dangerous or unspecified 2539/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2540/// function calls. 2541/// 2542/// \param Call The call expression to diagnose. 2543void Sema::CheckMemaccessArguments(const CallExpr *Call, 2544 unsigned BId, 2545 IdentifierInfo *FnName) { 2546 assert(BId != 0); 2547 2548 // It is possible to have a non-standard definition of memset. Validate 2549 // we have enough arguments, and if not, abort further checking. 2550 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 2551 if (Call->getNumArgs() < ExpectedNumArgs) 2552 return; 2553 2554 unsigned LastArg = (BId == Builtin::BImemset || 2555 BId == Builtin::BIstrndup ? 1 : 2); 2556 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 2557 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 2558 2559 // We have special checking when the length is a sizeof expression. 2560 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2561 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2562 llvm::FoldingSetNodeID SizeOfArgID; 2563 2564 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2565 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2566 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2567 2568 QualType DestTy = Dest->getType(); 2569 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2570 QualType PointeeTy = DestPtrTy->getPointeeType(); 2571 2572 // Never warn about void type pointers. This can be used to suppress 2573 // false positives. 2574 if (PointeeTy->isVoidType()) 2575 continue; 2576 2577 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2578 // actually comparing the expressions for equality. Because computing the 2579 // expression IDs can be expensive, we only do this if the diagnostic is 2580 // enabled. 2581 if (SizeOfArg && 2582 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2583 SizeOfArg->getExprLoc())) { 2584 // We only compute IDs for expressions if the warning is enabled, and 2585 // cache the sizeof arg's ID. 2586 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2587 SizeOfArg->Profile(SizeOfArgID, Context, true); 2588 llvm::FoldingSetNodeID DestID; 2589 Dest->Profile(DestID, Context, true); 2590 if (DestID == SizeOfArgID) { 2591 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 2592 // over sizeof(src) as well. 2593 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2594 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2595 if (UnaryOp->getOpcode() == UO_AddrOf) 2596 ActionIdx = 1; // If its an address-of operator, just remove it. 2597 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2598 ActionIdx = 2; // If the pointee's size is sizeof(char), 2599 // suggest an explicit length. 2600 unsigned DestSrcSelect = 2601 (BId == Builtin::BIstrndup ? 1 : ArgIdx); 2602 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 2603 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 2604 << FnName << DestSrcSelect << ActionIdx 2605 << Dest->getSourceRange() 2606 << SizeOfArg->getSourceRange()); 2607 break; 2608 } 2609 } 2610 2611 // Also check for cases where the sizeof argument is the exact same 2612 // type as the memory argument, and where it points to a user-defined 2613 // record type. 2614 if (SizeOfArgTy != QualType()) { 2615 if (PointeeTy->isRecordType() && 2616 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 2617 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 2618 PDiag(diag::warn_sizeof_pointer_type_memaccess) 2619 << FnName << SizeOfArgTy << ArgIdx 2620 << PointeeTy << Dest->getSourceRange() 2621 << LenExpr->getSourceRange()); 2622 break; 2623 } 2624 } 2625 2626 // Always complain about dynamic classes. 2627 if (isDynamicClassType(PointeeTy)) { 2628 2629 unsigned OperationType = 0; 2630 // "overwritten" if we're warning about the destination for any call 2631 // but memcmp; otherwise a verb appropriate to the call. 2632 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 2633 if (BId == Builtin::BImemcpy) 2634 OperationType = 1; 2635 else if(BId == Builtin::BImemmove) 2636 OperationType = 2; 2637 else if (BId == Builtin::BImemcmp) 2638 OperationType = 3; 2639 } 2640 2641 DiagRuntimeBehavior( 2642 Dest->getExprLoc(), Dest, 2643 PDiag(diag::warn_dyn_class_memaccess) 2644 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 2645 << FnName << PointeeTy 2646 << OperationType 2647 << Call->getCallee()->getSourceRange()); 2648 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 2649 BId != Builtin::BImemset) 2650 DiagRuntimeBehavior( 2651 Dest->getExprLoc(), Dest, 2652 PDiag(diag::warn_arc_object_memaccess) 2653 << ArgIdx << FnName << PointeeTy 2654 << Call->getCallee()->getSourceRange()); 2655 else 2656 continue; 2657 2658 DiagRuntimeBehavior( 2659 Dest->getExprLoc(), Dest, 2660 PDiag(diag::note_bad_memaccess_silence) 2661 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 2662 break; 2663 } 2664 } 2665} 2666 2667// A little helper routine: ignore addition and subtraction of integer literals. 2668// This intentionally does not ignore all integer constant expressions because 2669// we don't want to remove sizeof(). 2670static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2671 Ex = Ex->IgnoreParenCasts(); 2672 2673 for (;;) { 2674 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2675 if (!BO || !BO->isAdditiveOp()) 2676 break; 2677 2678 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2679 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2680 2681 if (isa<IntegerLiteral>(RHS)) 2682 Ex = LHS; 2683 else if (isa<IntegerLiteral>(LHS)) 2684 Ex = RHS; 2685 else 2686 break; 2687 } 2688 2689 return Ex; 2690} 2691 2692// Warn if the user has made the 'size' argument to strlcpy or strlcat 2693// be the size of the source, instead of the destination. 2694void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2695 IdentifierInfo *FnName) { 2696 2697 // Don't crash if the user has the wrong number of arguments 2698 if (Call->getNumArgs() != 3) 2699 return; 2700 2701 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2702 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2703 const Expr *CompareWithSrc = NULL; 2704 2705 // Look for 'strlcpy(dst, x, sizeof(x))' 2706 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2707 CompareWithSrc = Ex; 2708 else { 2709 // Look for 'strlcpy(dst, x, strlen(x))' 2710 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2711 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 2712 && SizeCall->getNumArgs() == 1) 2713 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2714 } 2715 } 2716 2717 if (!CompareWithSrc) 2718 return; 2719 2720 // Determine if the argument to sizeof/strlen is equal to the source 2721 // argument. In principle there's all kinds of things you could do 2722 // here, for instance creating an == expression and evaluating it with 2723 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2724 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2725 if (!SrcArgDRE) 2726 return; 2727 2728 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2729 if (!CompareWithSrcDRE || 2730 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2731 return; 2732 2733 const Expr *OriginalSizeArg = Call->getArg(2); 2734 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2735 << OriginalSizeArg->getSourceRange() << FnName; 2736 2737 // Output a FIXIT hint if the destination is an array (rather than a 2738 // pointer to an array). This could be enhanced to handle some 2739 // pointers if we know the actual size, like if DstArg is 'array+2' 2740 // we could say 'sizeof(array)-2'. 2741 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2742 QualType DstArgTy = DstArg->getType(); 2743 2744 // Only handle constant-sized or VLAs, but not flexible members. 2745 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2746 // Only issue the FIXIT for arrays of size > 1. 2747 if (CAT->getSize().getSExtValue() <= 1) 2748 return; 2749 } else if (!DstArgTy->isVariableArrayType()) { 2750 return; 2751 } 2752 2753 SmallString<128> sizeString; 2754 llvm::raw_svector_ostream OS(sizeString); 2755 OS << "sizeof("; 2756 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2757 OS << ")"; 2758 2759 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2760 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2761 OS.str()); 2762} 2763 2764/// Check if two expressions refer to the same declaration. 2765static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 2766 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 2767 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 2768 return D1->getDecl() == D2->getDecl(); 2769 return false; 2770} 2771 2772static const Expr *getStrlenExprArg(const Expr *E) { 2773 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 2774 const FunctionDecl *FD = CE->getDirectCallee(); 2775 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 2776 return 0; 2777 return CE->getArg(0)->IgnoreParenCasts(); 2778 } 2779 return 0; 2780} 2781 2782// Warn on anti-patterns as the 'size' argument to strncat. 2783// The correct size argument should look like following: 2784// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 2785void Sema::CheckStrncatArguments(const CallExpr *CE, 2786 IdentifierInfo *FnName) { 2787 // Don't crash if the user has the wrong number of arguments. 2788 if (CE->getNumArgs() < 3) 2789 return; 2790 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 2791 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 2792 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 2793 2794 // Identify common expressions, which are wrongly used as the size argument 2795 // to strncat and may lead to buffer overflows. 2796 unsigned PatternType = 0; 2797 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 2798 // - sizeof(dst) 2799 if (referToTheSameDecl(SizeOfArg, DstArg)) 2800 PatternType = 1; 2801 // - sizeof(src) 2802 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 2803 PatternType = 2; 2804 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 2805 if (BE->getOpcode() == BO_Sub) { 2806 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 2807 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 2808 // - sizeof(dst) - strlen(dst) 2809 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 2810 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 2811 PatternType = 1; 2812 // - sizeof(src) - (anything) 2813 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 2814 PatternType = 2; 2815 } 2816 } 2817 2818 if (PatternType == 0) 2819 return; 2820 2821 // Generate the diagnostic. 2822 SourceLocation SL = LenArg->getLocStart(); 2823 SourceRange SR = LenArg->getSourceRange(); 2824 SourceManager &SM = PP.getSourceManager(); 2825 2826 // If the function is defined as a builtin macro, do not show macro expansion. 2827 if (SM.isMacroArgExpansion(SL)) { 2828 SL = SM.getSpellingLoc(SL); 2829 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 2830 SM.getSpellingLoc(SR.getEnd())); 2831 } 2832 2833 if (PatternType == 1) 2834 Diag(SL, diag::warn_strncat_large_size) << SR; 2835 else 2836 Diag(SL, diag::warn_strncat_src_size) << SR; 2837 2838 // Output a FIXIT hint if the destination is an array (rather than a 2839 // pointer to an array). This could be enhanced to handle some 2840 // pointers if we know the actual size, like if DstArg is 'array+2' 2841 // we could say 'sizeof(array)-2'. 2842 QualType DstArgTy = DstArg->getType(); 2843 2844 // Only handle constant-sized or VLAs, but not flexible members. 2845 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2846 // Only issue the FIXIT for arrays of size > 1. 2847 if (CAT->getSize().getSExtValue() <= 1) 2848 return; 2849 } else if (!DstArgTy->isVariableArrayType()) { 2850 return; 2851 } 2852 2853 SmallString<128> sizeString; 2854 llvm::raw_svector_ostream OS(sizeString); 2855 OS << "sizeof("; 2856 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2857 OS << ") - "; 2858 OS << "strlen("; 2859 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2860 OS << ") - 1"; 2861 2862 Diag(SL, diag::note_strncat_wrong_size) 2863 << FixItHint::CreateReplacement(SR, OS.str()); 2864} 2865 2866//===--- CHECK: Return Address of Stack Variable --------------------------===// 2867 2868static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars); 2869static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars); 2870 2871/// CheckReturnStackAddr - Check if a return statement returns the address 2872/// of a stack variable. 2873void 2874Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 2875 SourceLocation ReturnLoc) { 2876 2877 Expr *stackE = 0; 2878 SmallVector<DeclRefExpr *, 8> refVars; 2879 2880 // Perform checking for returned stack addresses, local blocks, 2881 // label addresses or references to temporaries. 2882 if (lhsType->isPointerType() || 2883 (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 2884 stackE = EvalAddr(RetValExp, refVars); 2885 } else if (lhsType->isReferenceType()) { 2886 stackE = EvalVal(RetValExp, refVars); 2887 } 2888 2889 if (stackE == 0) 2890 return; // Nothing suspicious was found. 2891 2892 SourceLocation diagLoc; 2893 SourceRange diagRange; 2894 if (refVars.empty()) { 2895 diagLoc = stackE->getLocStart(); 2896 diagRange = stackE->getSourceRange(); 2897 } else { 2898 // We followed through a reference variable. 'stackE' contains the 2899 // problematic expression but we will warn at the return statement pointing 2900 // at the reference variable. We will later display the "trail" of 2901 // reference variables using notes. 2902 diagLoc = refVars[0]->getLocStart(); 2903 diagRange = refVars[0]->getSourceRange(); 2904 } 2905 2906 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 2907 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 2908 : diag::warn_ret_stack_addr) 2909 << DR->getDecl()->getDeclName() << diagRange; 2910 } else if (isa<BlockExpr>(stackE)) { // local block. 2911 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 2912 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 2913 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 2914 } else { // local temporary. 2915 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 2916 : diag::warn_ret_local_temp_addr) 2917 << diagRange; 2918 } 2919 2920 // Display the "trail" of reference variables that we followed until we 2921 // found the problematic expression using notes. 2922 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 2923 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 2924 // If this var binds to another reference var, show the range of the next 2925 // var, otherwise the var binds to the problematic expression, in which case 2926 // show the range of the expression. 2927 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 2928 : stackE->getSourceRange(); 2929 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 2930 << VD->getDeclName() << range; 2931 } 2932} 2933 2934/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 2935/// check if the expression in a return statement evaluates to an address 2936/// to a location on the stack, a local block, an address of a label, or a 2937/// reference to local temporary. The recursion is used to traverse the 2938/// AST of the return expression, with recursion backtracking when we 2939/// encounter a subexpression that (1) clearly does not lead to one of the 2940/// above problematic expressions (2) is something we cannot determine leads to 2941/// a problematic expression based on such local checking. 2942/// 2943/// Both EvalAddr and EvalVal follow through reference variables to evaluate 2944/// the expression that they point to. Such variables are added to the 2945/// 'refVars' vector so that we know what the reference variable "trail" was. 2946/// 2947/// EvalAddr processes expressions that are pointers that are used as 2948/// references (and not L-values). EvalVal handles all other values. 2949/// At the base case of the recursion is a check for the above problematic 2950/// expressions. 2951/// 2952/// This implementation handles: 2953/// 2954/// * pointer-to-pointer casts 2955/// * implicit conversions from array references to pointers 2956/// * taking the address of fields 2957/// * arbitrary interplay between "&" and "*" operators 2958/// * pointer arithmetic from an address of a stack variable 2959/// * taking the address of an array element where the array is on the stack 2960static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2961 if (E->isTypeDependent()) 2962 return NULL; 2963 2964 // We should only be called for evaluating pointer expressions. 2965 assert((E->getType()->isAnyPointerType() || 2966 E->getType()->isBlockPointerType() || 2967 E->getType()->isObjCQualifiedIdType()) && 2968 "EvalAddr only works on pointers"); 2969 2970 E = E->IgnoreParens(); 2971 2972 // Our "symbolic interpreter" is just a dispatch off the currently 2973 // viewed AST node. We then recursively traverse the AST by calling 2974 // EvalAddr and EvalVal appropriately. 2975 switch (E->getStmtClass()) { 2976 case Stmt::DeclRefExprClass: { 2977 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2978 2979 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2980 // If this is a reference variable, follow through to the expression that 2981 // it points to. 2982 if (V->hasLocalStorage() && 2983 V->getType()->isReferenceType() && V->hasInit()) { 2984 // Add the reference variable to the "trail". 2985 refVars.push_back(DR); 2986 return EvalAddr(V->getInit(), refVars); 2987 } 2988 2989 return NULL; 2990 } 2991 2992 case Stmt::UnaryOperatorClass: { 2993 // The only unary operator that make sense to handle here 2994 // is AddrOf. All others don't make sense as pointers. 2995 UnaryOperator *U = cast<UnaryOperator>(E); 2996 2997 if (U->getOpcode() == UO_AddrOf) 2998 return EvalVal(U->getSubExpr(), refVars); 2999 else 3000 return NULL; 3001 } 3002 3003 case Stmt::BinaryOperatorClass: { 3004 // Handle pointer arithmetic. All other binary operators are not valid 3005 // in this context. 3006 BinaryOperator *B = cast<BinaryOperator>(E); 3007 BinaryOperatorKind op = B->getOpcode(); 3008 3009 if (op != BO_Add && op != BO_Sub) 3010 return NULL; 3011 3012 Expr *Base = B->getLHS(); 3013 3014 // Determine which argument is the real pointer base. It could be 3015 // the RHS argument instead of the LHS. 3016 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 3017 3018 assert (Base->getType()->isPointerType()); 3019 return EvalAddr(Base, refVars); 3020 } 3021 3022 // For conditional operators we need to see if either the LHS or RHS are 3023 // valid DeclRefExpr*s. If one of them is valid, we return it. 3024 case Stmt::ConditionalOperatorClass: { 3025 ConditionalOperator *C = cast<ConditionalOperator>(E); 3026 3027 // Handle the GNU extension for missing LHS. 3028 if (Expr *lhsExpr = C->getLHS()) { 3029 // In C++, we can have a throw-expression, which has 'void' type. 3030 if (!lhsExpr->getType()->isVoidType()) 3031 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 3032 return LHS; 3033 } 3034 3035 // In C++, we can have a throw-expression, which has 'void' type. 3036 if (C->getRHS()->getType()->isVoidType()) 3037 return NULL; 3038 3039 return EvalAddr(C->getRHS(), refVars); 3040 } 3041 3042 case Stmt::BlockExprClass: 3043 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 3044 return E; // local block. 3045 return NULL; 3046 3047 case Stmt::AddrLabelExprClass: 3048 return E; // address of label. 3049 3050 case Stmt::ExprWithCleanupsClass: 3051 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars); 3052 3053 // For casts, we need to handle conversions from arrays to 3054 // pointer values, and pointer-to-pointer conversions. 3055 case Stmt::ImplicitCastExprClass: 3056 case Stmt::CStyleCastExprClass: 3057 case Stmt::CXXFunctionalCastExprClass: 3058 case Stmt::ObjCBridgedCastExprClass: { 3059 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 3060 QualType T = SubExpr->getType(); 3061 3062 if (cast<CastExpr>(E)->getCastKind() == CK_CopyAndAutoreleaseBlockObject) 3063 return 0; 3064 else if (SubExpr->getType()->isPointerType() || 3065 SubExpr->getType()->isBlockPointerType() || 3066 SubExpr->getType()->isObjCQualifiedIdType()) 3067 return EvalAddr(SubExpr, refVars); 3068 else if (T->isArrayType()) 3069 return EvalVal(SubExpr, refVars); 3070 else 3071 return 0; 3072 } 3073 3074 // C++ casts. For dynamic casts, static casts, and const casts, we 3075 // are always converting from a pointer-to-pointer, so we just blow 3076 // through the cast. In the case the dynamic cast doesn't fail (and 3077 // return NULL), we take the conservative route and report cases 3078 // where we return the address of a stack variable. For Reinterpre 3079 // FIXME: The comment about is wrong; we're not always converting 3080 // from pointer to pointer. I'm guessing that this code should also 3081 // handle references to objects. 3082 case Stmt::CXXStaticCastExprClass: 3083 case Stmt::CXXDynamicCastExprClass: 3084 case Stmt::CXXConstCastExprClass: 3085 case Stmt::CXXReinterpretCastExprClass: { 3086 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 3087 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 3088 return EvalAddr(S, refVars); 3089 else 3090 return NULL; 3091 } 3092 3093 case Stmt::MaterializeTemporaryExprClass: 3094 if (Expr *Result = EvalAddr( 3095 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3096 refVars)) 3097 return Result; 3098 3099 return E; 3100 3101 // Everything else: we simply don't reason about them. 3102 default: 3103 return NULL; 3104 } 3105} 3106 3107 3108/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3109/// See the comments for EvalAddr for more details. 3110static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 3111do { 3112 // We should only be called for evaluating non-pointer expressions, or 3113 // expressions with a pointer type that are not used as references but instead 3114 // are l-values (e.g., DeclRefExpr with a pointer type). 3115 3116 // Our "symbolic interpreter" is just a dispatch off the currently 3117 // viewed AST node. We then recursively traverse the AST by calling 3118 // EvalAddr and EvalVal appropriately. 3119 3120 E = E->IgnoreParens(); 3121 switch (E->getStmtClass()) { 3122 case Stmt::ImplicitCastExprClass: { 3123 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3124 if (IE->getValueKind() == VK_LValue) { 3125 E = IE->getSubExpr(); 3126 continue; 3127 } 3128 return NULL; 3129 } 3130 3131 case Stmt::ExprWithCleanupsClass: 3132 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars); 3133 3134 case Stmt::DeclRefExprClass: { 3135 // When we hit a DeclRefExpr we are looking at code that refers to a 3136 // variable's name. If it's not a reference variable we check if it has 3137 // local storage within the function, and if so, return the expression. 3138 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3139 3140 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3141 if (V->hasLocalStorage()) { 3142 if (!V->getType()->isReferenceType()) 3143 return DR; 3144 3145 // Reference variable, follow through to the expression that 3146 // it points to. 3147 if (V->hasInit()) { 3148 // Add the reference variable to the "trail". 3149 refVars.push_back(DR); 3150 return EvalVal(V->getInit(), refVars); 3151 } 3152 } 3153 3154 return NULL; 3155 } 3156 3157 case Stmt::UnaryOperatorClass: { 3158 // The only unary operator that make sense to handle here 3159 // is Deref. All others don't resolve to a "name." This includes 3160 // handling all sorts of rvalues passed to a unary operator. 3161 UnaryOperator *U = cast<UnaryOperator>(E); 3162 3163 if (U->getOpcode() == UO_Deref) 3164 return EvalAddr(U->getSubExpr(), refVars); 3165 3166 return NULL; 3167 } 3168 3169 case Stmt::ArraySubscriptExprClass: { 3170 // Array subscripts are potential references to data on the stack. We 3171 // retrieve the DeclRefExpr* for the array variable if it indeed 3172 // has local storage. 3173 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 3174 } 3175 3176 case Stmt::ConditionalOperatorClass: { 3177 // For conditional operators we need to see if either the LHS or RHS are 3178 // non-NULL Expr's. If one is non-NULL, we return it. 3179 ConditionalOperator *C = cast<ConditionalOperator>(E); 3180 3181 // Handle the GNU extension for missing LHS. 3182 if (Expr *lhsExpr = C->getLHS()) 3183 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 3184 return LHS; 3185 3186 return EvalVal(C->getRHS(), refVars); 3187 } 3188 3189 // Accesses to members are potential references to data on the stack. 3190 case Stmt::MemberExprClass: { 3191 MemberExpr *M = cast<MemberExpr>(E); 3192 3193 // Check for indirect access. We only want direct field accesses. 3194 if (M->isArrow()) 3195 return NULL; 3196 3197 // Check whether the member type is itself a reference, in which case 3198 // we're not going to refer to the member, but to what the member refers to. 3199 if (M->getMemberDecl()->getType()->isReferenceType()) 3200 return NULL; 3201 3202 return EvalVal(M->getBase(), refVars); 3203 } 3204 3205 case Stmt::MaterializeTemporaryExprClass: 3206 if (Expr *Result = EvalVal( 3207 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3208 refVars)) 3209 return Result; 3210 3211 return E; 3212 3213 default: 3214 // Check that we don't return or take the address of a reference to a 3215 // temporary. This is only useful in C++. 3216 if (!E->isTypeDependent() && E->isRValue()) 3217 return E; 3218 3219 // Everything else: we simply don't reason about them. 3220 return NULL; 3221 } 3222} while (true); 3223} 3224 3225//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3226 3227/// Check for comparisons of floating point operands using != and ==. 3228/// Issue a warning if these are no self-comparisons, as they are not likely 3229/// to do what the programmer intended. 3230void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3231 bool EmitWarning = true; 3232 3233 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3234 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3235 3236 // Special case: check for x == x (which is OK). 3237 // Do not emit warnings for such cases. 3238 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3239 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3240 if (DRL->getDecl() == DRR->getDecl()) 3241 EmitWarning = false; 3242 3243 3244 // Special case: check for comparisons against literals that can be exactly 3245 // represented by APFloat. In such cases, do not emit a warning. This 3246 // is a heuristic: often comparison against such literals are used to 3247 // detect if a value in a variable has not changed. This clearly can 3248 // lead to false negatives. 3249 if (EmitWarning) { 3250 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3251 if (FLL->isExact()) 3252 EmitWarning = false; 3253 } else 3254 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 3255 if (FLR->isExact()) 3256 EmitWarning = false; 3257 } 3258 } 3259 3260 // Check for comparisons with builtin types. 3261 if (EmitWarning) 3262 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3263 if (CL->isBuiltinCall()) 3264 EmitWarning = false; 3265 3266 if (EmitWarning) 3267 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3268 if (CR->isBuiltinCall()) 3269 EmitWarning = false; 3270 3271 // Emit the diagnostic. 3272 if (EmitWarning) 3273 Diag(Loc, diag::warn_floatingpoint_eq) 3274 << LHS->getSourceRange() << RHS->getSourceRange(); 3275} 3276 3277//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3278//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3279 3280namespace { 3281 3282/// Structure recording the 'active' range of an integer-valued 3283/// expression. 3284struct IntRange { 3285 /// The number of bits active in the int. 3286 unsigned Width; 3287 3288 /// True if the int is known not to have negative values. 3289 bool NonNegative; 3290 3291 IntRange(unsigned Width, bool NonNegative) 3292 : Width(Width), NonNegative(NonNegative) 3293 {} 3294 3295 /// Returns the range of the bool type. 3296 static IntRange forBoolType() { 3297 return IntRange(1, true); 3298 } 3299 3300 /// Returns the range of an opaque value of the given integral type. 3301 static IntRange forValueOfType(ASTContext &C, QualType T) { 3302 return forValueOfCanonicalType(C, 3303 T->getCanonicalTypeInternal().getTypePtr()); 3304 } 3305 3306 /// Returns the range of an opaque value of a canonical integral type. 3307 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3308 assert(T->isCanonicalUnqualified()); 3309 3310 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3311 T = VT->getElementType().getTypePtr(); 3312 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3313 T = CT->getElementType().getTypePtr(); 3314 3315 // For enum types, use the known bit width of the enumerators. 3316 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3317 EnumDecl *Enum = ET->getDecl(); 3318 if (!Enum->isCompleteDefinition()) 3319 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3320 3321 unsigned NumPositive = Enum->getNumPositiveBits(); 3322 unsigned NumNegative = Enum->getNumNegativeBits(); 3323 3324 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 3325 } 3326 3327 const BuiltinType *BT = cast<BuiltinType>(T); 3328 assert(BT->isInteger()); 3329 3330 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3331 } 3332 3333 /// Returns the "target" range of a canonical integral type, i.e. 3334 /// the range of values expressible in the type. 3335 /// 3336 /// This matches forValueOfCanonicalType except that enums have the 3337 /// full range of their type, not the range of their enumerators. 3338 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 3339 assert(T->isCanonicalUnqualified()); 3340 3341 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3342 T = VT->getElementType().getTypePtr(); 3343 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3344 T = CT->getElementType().getTypePtr(); 3345 if (const EnumType *ET = dyn_cast<EnumType>(T)) 3346 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 3347 3348 const BuiltinType *BT = cast<BuiltinType>(T); 3349 assert(BT->isInteger()); 3350 3351 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3352 } 3353 3354 /// Returns the supremum of two ranges: i.e. their conservative merge. 3355 static IntRange join(IntRange L, IntRange R) { 3356 return IntRange(std::max(L.Width, R.Width), 3357 L.NonNegative && R.NonNegative); 3358 } 3359 3360 /// Returns the infinum of two ranges: i.e. their aggressive merge. 3361 static IntRange meet(IntRange L, IntRange R) { 3362 return IntRange(std::min(L.Width, R.Width), 3363 L.NonNegative || R.NonNegative); 3364 } 3365}; 3366 3367static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 3368 unsigned MaxWidth) { 3369 if (value.isSigned() && value.isNegative()) 3370 return IntRange(value.getMinSignedBits(), false); 3371 3372 if (value.getBitWidth() > MaxWidth) 3373 value = value.trunc(MaxWidth); 3374 3375 // isNonNegative() just checks the sign bit without considering 3376 // signedness. 3377 return IntRange(value.getActiveBits(), true); 3378} 3379 3380static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 3381 unsigned MaxWidth) { 3382 if (result.isInt()) 3383 return GetValueRange(C, result.getInt(), MaxWidth); 3384 3385 if (result.isVector()) { 3386 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 3387 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 3388 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 3389 R = IntRange::join(R, El); 3390 } 3391 return R; 3392 } 3393 3394 if (result.isComplexInt()) { 3395 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 3396 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 3397 return IntRange::join(R, I); 3398 } 3399 3400 // This can happen with lossless casts to intptr_t of "based" lvalues. 3401 // Assume it might use arbitrary bits. 3402 // FIXME: The only reason we need to pass the type in here is to get 3403 // the sign right on this one case. It would be nice if APValue 3404 // preserved this. 3405 assert(result.isLValue() || result.isAddrLabelDiff()); 3406 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 3407} 3408 3409/// Pseudo-evaluate the given integer expression, estimating the 3410/// range of values it might take. 3411/// 3412/// \param MaxWidth - the width to which the value will be truncated 3413static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 3414 E = E->IgnoreParens(); 3415 3416 // Try a full evaluation first. 3417 Expr::EvalResult result; 3418 if (E->EvaluateAsRValue(result, C)) 3419 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 3420 3421 // I think we only want to look through implicit casts here; if the 3422 // user has an explicit widening cast, we should treat the value as 3423 // being of the new, wider type. 3424 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 3425 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 3426 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 3427 3428 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 3429 3430 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 3431 3432 // Assume that non-integer casts can span the full range of the type. 3433 if (!isIntegerCast) 3434 return OutputTypeRange; 3435 3436 IntRange SubRange 3437 = GetExprRange(C, CE->getSubExpr(), 3438 std::min(MaxWidth, OutputTypeRange.Width)); 3439 3440 // Bail out if the subexpr's range is as wide as the cast type. 3441 if (SubRange.Width >= OutputTypeRange.Width) 3442 return OutputTypeRange; 3443 3444 // Otherwise, we take the smaller width, and we're non-negative if 3445 // either the output type or the subexpr is. 3446 return IntRange(SubRange.Width, 3447 SubRange.NonNegative || OutputTypeRange.NonNegative); 3448 } 3449 3450 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 3451 // If we can fold the condition, just take that operand. 3452 bool CondResult; 3453 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 3454 return GetExprRange(C, CondResult ? CO->getTrueExpr() 3455 : CO->getFalseExpr(), 3456 MaxWidth); 3457 3458 // Otherwise, conservatively merge. 3459 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 3460 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 3461 return IntRange::join(L, R); 3462 } 3463 3464 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3465 switch (BO->getOpcode()) { 3466 3467 // Boolean-valued operations are single-bit and positive. 3468 case BO_LAnd: 3469 case BO_LOr: 3470 case BO_LT: 3471 case BO_GT: 3472 case BO_LE: 3473 case BO_GE: 3474 case BO_EQ: 3475 case BO_NE: 3476 return IntRange::forBoolType(); 3477 3478 // The type of the assignments is the type of the LHS, so the RHS 3479 // is not necessarily the same type. 3480 case BO_MulAssign: 3481 case BO_DivAssign: 3482 case BO_RemAssign: 3483 case BO_AddAssign: 3484 case BO_SubAssign: 3485 case BO_XorAssign: 3486 case BO_OrAssign: 3487 // TODO: bitfields? 3488 return IntRange::forValueOfType(C, E->getType()); 3489 3490 // Simple assignments just pass through the RHS, which will have 3491 // been coerced to the LHS type. 3492 case BO_Assign: 3493 // TODO: bitfields? 3494 return GetExprRange(C, BO->getRHS(), MaxWidth); 3495 3496 // Operations with opaque sources are black-listed. 3497 case BO_PtrMemD: 3498 case BO_PtrMemI: 3499 return IntRange::forValueOfType(C, E->getType()); 3500 3501 // Bitwise-and uses the *infinum* of the two source ranges. 3502 case BO_And: 3503 case BO_AndAssign: 3504 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 3505 GetExprRange(C, BO->getRHS(), MaxWidth)); 3506 3507 // Left shift gets black-listed based on a judgement call. 3508 case BO_Shl: 3509 // ...except that we want to treat '1 << (blah)' as logically 3510 // positive. It's an important idiom. 3511 if (IntegerLiteral *I 3512 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 3513 if (I->getValue() == 1) { 3514 IntRange R = IntRange::forValueOfType(C, E->getType()); 3515 return IntRange(R.Width, /*NonNegative*/ true); 3516 } 3517 } 3518 // fallthrough 3519 3520 case BO_ShlAssign: 3521 return IntRange::forValueOfType(C, E->getType()); 3522 3523 // Right shift by a constant can narrow its left argument. 3524 case BO_Shr: 3525 case BO_ShrAssign: { 3526 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3527 3528 // If the shift amount is a positive constant, drop the width by 3529 // that much. 3530 llvm::APSInt shift; 3531 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 3532 shift.isNonNegative()) { 3533 unsigned zext = shift.getZExtValue(); 3534 if (zext >= L.Width) 3535 L.Width = (L.NonNegative ? 0 : 1); 3536 else 3537 L.Width -= zext; 3538 } 3539 3540 return L; 3541 } 3542 3543 // Comma acts as its right operand. 3544 case BO_Comma: 3545 return GetExprRange(C, BO->getRHS(), MaxWidth); 3546 3547 // Black-list pointer subtractions. 3548 case BO_Sub: 3549 if (BO->getLHS()->getType()->isPointerType()) 3550 return IntRange::forValueOfType(C, E->getType()); 3551 break; 3552 3553 // The width of a division result is mostly determined by the size 3554 // of the LHS. 3555 case BO_Div: { 3556 // Don't 'pre-truncate' the operands. 3557 unsigned opWidth = C.getIntWidth(E->getType()); 3558 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3559 3560 // If the divisor is constant, use that. 3561 llvm::APSInt divisor; 3562 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3563 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3564 if (log2 >= L.Width) 3565 L.Width = (L.NonNegative ? 0 : 1); 3566 else 3567 L.Width = std::min(L.Width - log2, MaxWidth); 3568 return L; 3569 } 3570 3571 // Otherwise, just use the LHS's width. 3572 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3573 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3574 } 3575 3576 // The result of a remainder can't be larger than the result of 3577 // either side. 3578 case BO_Rem: { 3579 // Don't 'pre-truncate' the operands. 3580 unsigned opWidth = C.getIntWidth(E->getType()); 3581 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3582 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3583 3584 IntRange meet = IntRange::meet(L, R); 3585 meet.Width = std::min(meet.Width, MaxWidth); 3586 return meet; 3587 } 3588 3589 // The default behavior is okay for these. 3590 case BO_Mul: 3591 case BO_Add: 3592 case BO_Xor: 3593 case BO_Or: 3594 break; 3595 } 3596 3597 // The default case is to treat the operation as if it were closed 3598 // on the narrowest type that encompasses both operands. 3599 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3600 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 3601 return IntRange::join(L, R); 3602 } 3603 3604 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 3605 switch (UO->getOpcode()) { 3606 // Boolean-valued operations are white-listed. 3607 case UO_LNot: 3608 return IntRange::forBoolType(); 3609 3610 // Operations with opaque sources are black-listed. 3611 case UO_Deref: 3612 case UO_AddrOf: // should be impossible 3613 return IntRange::forValueOfType(C, E->getType()); 3614 3615 default: 3616 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 3617 } 3618 } 3619 3620 if (dyn_cast<OffsetOfExpr>(E)) { 3621 IntRange::forValueOfType(C, E->getType()); 3622 } 3623 3624 if (FieldDecl *BitField = E->getBitField()) 3625 return IntRange(BitField->getBitWidthValue(C), 3626 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 3627 3628 return IntRange::forValueOfType(C, E->getType()); 3629} 3630 3631static IntRange GetExprRange(ASTContext &C, Expr *E) { 3632 return GetExprRange(C, E, C.getIntWidth(E->getType())); 3633} 3634 3635/// Checks whether the given value, which currently has the given 3636/// source semantics, has the same value when coerced through the 3637/// target semantics. 3638static bool IsSameFloatAfterCast(const llvm::APFloat &value, 3639 const llvm::fltSemantics &Src, 3640 const llvm::fltSemantics &Tgt) { 3641 llvm::APFloat truncated = value; 3642 3643 bool ignored; 3644 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 3645 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 3646 3647 return truncated.bitwiseIsEqual(value); 3648} 3649 3650/// Checks whether the given value, which currently has the given 3651/// source semantics, has the same value when coerced through the 3652/// target semantics. 3653/// 3654/// The value might be a vector of floats (or a complex number). 3655static bool IsSameFloatAfterCast(const APValue &value, 3656 const llvm::fltSemantics &Src, 3657 const llvm::fltSemantics &Tgt) { 3658 if (value.isFloat()) 3659 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 3660 3661 if (value.isVector()) { 3662 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 3663 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 3664 return false; 3665 return true; 3666 } 3667 3668 assert(value.isComplexFloat()); 3669 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 3670 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 3671} 3672 3673static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 3674 3675static bool IsZero(Sema &S, Expr *E) { 3676 // Suppress cases where we are comparing against an enum constant. 3677 if (const DeclRefExpr *DR = 3678 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 3679 if (isa<EnumConstantDecl>(DR->getDecl())) 3680 return false; 3681 3682 // Suppress cases where the '0' value is expanded from a macro. 3683 if (E->getLocStart().isMacroID()) 3684 return false; 3685 3686 llvm::APSInt Value; 3687 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 3688} 3689 3690static bool HasEnumType(Expr *E) { 3691 // Strip off implicit integral promotions. 3692 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3693 if (ICE->getCastKind() != CK_IntegralCast && 3694 ICE->getCastKind() != CK_NoOp) 3695 break; 3696 E = ICE->getSubExpr(); 3697 } 3698 3699 return E->getType()->isEnumeralType(); 3700} 3701 3702static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 3703 BinaryOperatorKind op = E->getOpcode(); 3704 if (E->isValueDependent()) 3705 return; 3706 3707 if (op == BO_LT && IsZero(S, E->getRHS())) { 3708 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3709 << "< 0" << "false" << HasEnumType(E->getLHS()) 3710 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3711 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 3712 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3713 << ">= 0" << "true" << HasEnumType(E->getLHS()) 3714 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3715 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 3716 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3717 << "0 >" << "false" << HasEnumType(E->getRHS()) 3718 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3719 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 3720 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3721 << "0 <=" << "true" << HasEnumType(E->getRHS()) 3722 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3723 } 3724} 3725 3726/// Analyze the operands of the given comparison. Implements the 3727/// fallback case from AnalyzeComparison. 3728static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 3729 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3730 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3731} 3732 3733/// \brief Implements -Wsign-compare. 3734/// 3735/// \param E the binary operator to check for warnings 3736static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 3737 // The type the comparison is being performed in. 3738 QualType T = E->getLHS()->getType(); 3739 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 3740 && "comparison with mismatched types"); 3741 3742 // We don't do anything special if this isn't an unsigned integral 3743 // comparison: we're only interested in integral comparisons, and 3744 // signed comparisons only happen in cases we don't care to warn about. 3745 // 3746 // We also don't care about value-dependent expressions or expressions 3747 // whose result is a constant. 3748 if (!T->hasUnsignedIntegerRepresentation() 3749 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 3750 return AnalyzeImpConvsInComparison(S, E); 3751 3752 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 3753 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 3754 3755 // Check to see if one of the (unmodified) operands is of different 3756 // signedness. 3757 Expr *signedOperand, *unsignedOperand; 3758 if (LHS->getType()->hasSignedIntegerRepresentation()) { 3759 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 3760 "unsigned comparison between two signed integer expressions?"); 3761 signedOperand = LHS; 3762 unsignedOperand = RHS; 3763 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 3764 signedOperand = RHS; 3765 unsignedOperand = LHS; 3766 } else { 3767 CheckTrivialUnsignedComparison(S, E); 3768 return AnalyzeImpConvsInComparison(S, E); 3769 } 3770 3771 // Otherwise, calculate the effective range of the signed operand. 3772 IntRange signedRange = GetExprRange(S.Context, signedOperand); 3773 3774 // Go ahead and analyze implicit conversions in the operands. Note 3775 // that we skip the implicit conversions on both sides. 3776 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 3777 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 3778 3779 // If the signed range is non-negative, -Wsign-compare won't fire, 3780 // but we should still check for comparisons which are always true 3781 // or false. 3782 if (signedRange.NonNegative) 3783 return CheckTrivialUnsignedComparison(S, E); 3784 3785 // For (in)equality comparisons, if the unsigned operand is a 3786 // constant which cannot collide with a overflowed signed operand, 3787 // then reinterpreting the signed operand as unsigned will not 3788 // change the result of the comparison. 3789 if (E->isEqualityOp()) { 3790 unsigned comparisonWidth = S.Context.getIntWidth(T); 3791 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3792 3793 // We should never be unable to prove that the unsigned operand is 3794 // non-negative. 3795 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3796 3797 if (unsignedRange.Width < comparisonWidth) 3798 return; 3799 } 3800 3801 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 3802 << LHS->getType() << RHS->getType() 3803 << LHS->getSourceRange() << RHS->getSourceRange(); 3804} 3805 3806/// Analyzes an attempt to assign the given value to a bitfield. 3807/// 3808/// Returns true if there was something fishy about the attempt. 3809static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3810 SourceLocation InitLoc) { 3811 assert(Bitfield->isBitField()); 3812 if (Bitfield->isInvalidDecl()) 3813 return false; 3814 3815 // White-list bool bitfields. 3816 if (Bitfield->getType()->isBooleanType()) 3817 return false; 3818 3819 // Ignore value- or type-dependent expressions. 3820 if (Bitfield->getBitWidth()->isValueDependent() || 3821 Bitfield->getBitWidth()->isTypeDependent() || 3822 Init->isValueDependent() || 3823 Init->isTypeDependent()) 3824 return false; 3825 3826 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3827 3828 llvm::APSInt Value; 3829 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 3830 return false; 3831 3832 unsigned OriginalWidth = Value.getBitWidth(); 3833 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 3834 3835 if (OriginalWidth <= FieldWidth) 3836 return false; 3837 3838 // Compute the value which the bitfield will contain. 3839 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 3840 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 3841 3842 // Check whether the stored value is equal to the original value. 3843 TruncatedValue = TruncatedValue.extend(OriginalWidth); 3844 if (Value == TruncatedValue) 3845 return false; 3846 3847 // Special-case bitfields of width 1: booleans are naturally 0/1, and 3848 // therefore don't strictly fit into a signed bitfield of width 1. 3849 if (FieldWidth == 1 && Value == 1) 3850 return false; 3851 3852 std::string PrettyValue = Value.toString(10); 3853 std::string PrettyTrunc = TruncatedValue.toString(10); 3854 3855 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 3856 << PrettyValue << PrettyTrunc << OriginalInit->getType() 3857 << Init->getSourceRange(); 3858 3859 return true; 3860} 3861 3862/// Analyze the given simple or compound assignment for warning-worthy 3863/// operations. 3864static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 3865 // Just recurse on the LHS. 3866 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3867 3868 // We want to recurse on the RHS as normal unless we're assigning to 3869 // a bitfield. 3870 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 3871 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 3872 E->getOperatorLoc())) { 3873 // Recurse, ignoring any implicit conversions on the RHS. 3874 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 3875 E->getOperatorLoc()); 3876 } 3877 } 3878 3879 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3880} 3881 3882/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3883static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 3884 SourceLocation CContext, unsigned diag, 3885 bool pruneControlFlow = false) { 3886 if (pruneControlFlow) { 3887 S.DiagRuntimeBehavior(E->getExprLoc(), E, 3888 S.PDiag(diag) 3889 << SourceType << T << E->getSourceRange() 3890 << SourceRange(CContext)); 3891 return; 3892 } 3893 S.Diag(E->getExprLoc(), diag) 3894 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 3895} 3896 3897/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3898static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 3899 SourceLocation CContext, unsigned diag, 3900 bool pruneControlFlow = false) { 3901 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 3902} 3903 3904/// Diagnose an implicit cast from a literal expression. Does not warn when the 3905/// cast wouldn't lose information. 3906void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 3907 SourceLocation CContext) { 3908 // Try to convert the literal exactly to an integer. If we can, don't warn. 3909 bool isExact = false; 3910 const llvm::APFloat &Value = FL->getValue(); 3911 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 3912 T->hasUnsignedIntegerRepresentation()); 3913 if (Value.convertToInteger(IntegerValue, 3914 llvm::APFloat::rmTowardZero, &isExact) 3915 == llvm::APFloat::opOK && isExact) 3916 return; 3917 3918 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 3919 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 3920} 3921 3922std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 3923 if (!Range.Width) return "0"; 3924 3925 llvm::APSInt ValueInRange = Value; 3926 ValueInRange.setIsSigned(!Range.NonNegative); 3927 ValueInRange = ValueInRange.trunc(Range.Width); 3928 return ValueInRange.toString(10); 3929} 3930 3931void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 3932 SourceLocation CC, bool *ICContext = 0) { 3933 if (E->isTypeDependent() || E->isValueDependent()) return; 3934 3935 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 3936 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 3937 if (Source == Target) return; 3938 if (Target->isDependentType()) return; 3939 3940 // If the conversion context location is invalid don't complain. We also 3941 // don't want to emit a warning if the issue occurs from the expansion of 3942 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 3943 // delay this check as long as possible. Once we detect we are in that 3944 // scenario, we just return. 3945 if (CC.isInvalid()) 3946 return; 3947 3948 // Diagnose implicit casts to bool. 3949 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 3950 if (isa<StringLiteral>(E)) 3951 // Warn on string literal to bool. Checks for string literals in logical 3952 // expressions, for instances, assert(0 && "error here"), is prevented 3953 // by a check in AnalyzeImplicitConversions(). 3954 return DiagnoseImpCast(S, E, T, CC, 3955 diag::warn_impcast_string_literal_to_bool); 3956 if (Source->isFunctionType()) { 3957 // Warn on function to bool. Checks free functions and static member 3958 // functions. Weakly imported functions are excluded from the check, 3959 // since it's common to test their value to check whether the linker 3960 // found a definition for them. 3961 ValueDecl *D = 0; 3962 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 3963 D = R->getDecl(); 3964 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 3965 D = M->getMemberDecl(); 3966 } 3967 3968 if (D && !D->isWeak()) { 3969 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 3970 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 3971 << F << E->getSourceRange() << SourceRange(CC); 3972 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 3973 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 3974 QualType ReturnType; 3975 UnresolvedSet<4> NonTemplateOverloads; 3976 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 3977 if (!ReturnType.isNull() 3978 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 3979 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 3980 << FixItHint::CreateInsertion( 3981 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 3982 return; 3983 } 3984 } 3985 } 3986 return; // Other casts to bool are not checked. 3987 } 3988 3989 // Strip vector types. 3990 if (isa<VectorType>(Source)) { 3991 if (!isa<VectorType>(Target)) { 3992 if (S.SourceMgr.isInSystemMacro(CC)) 3993 return; 3994 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 3995 } 3996 3997 // If the vector cast is cast between two vectors of the same size, it is 3998 // a bitcast, not a conversion. 3999 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 4000 return; 4001 4002 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 4003 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 4004 } 4005 4006 // Strip complex types. 4007 if (isa<ComplexType>(Source)) { 4008 if (!isa<ComplexType>(Target)) { 4009 if (S.SourceMgr.isInSystemMacro(CC)) 4010 return; 4011 4012 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 4013 } 4014 4015 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 4016 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 4017 } 4018 4019 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 4020 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 4021 4022 // If the source is floating point... 4023 if (SourceBT && SourceBT->isFloatingPoint()) { 4024 // ...and the target is floating point... 4025 if (TargetBT && TargetBT->isFloatingPoint()) { 4026 // ...then warn if we're dropping FP rank. 4027 4028 // Builtin FP kinds are ordered by increasing FP rank. 4029 if (SourceBT->getKind() > TargetBT->getKind()) { 4030 // Don't warn about float constants that are precisely 4031 // representable in the target type. 4032 Expr::EvalResult result; 4033 if (E->EvaluateAsRValue(result, S.Context)) { 4034 // Value might be a float, a float vector, or a float complex. 4035 if (IsSameFloatAfterCast(result.Val, 4036 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 4037 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 4038 return; 4039 } 4040 4041 if (S.SourceMgr.isInSystemMacro(CC)) 4042 return; 4043 4044 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 4045 } 4046 return; 4047 } 4048 4049 // If the target is integral, always warn. 4050 if ((TargetBT && TargetBT->isInteger())) { 4051 if (S.SourceMgr.isInSystemMacro(CC)) 4052 return; 4053 4054 Expr *InnerE = E->IgnoreParenImpCasts(); 4055 // We also want to warn on, e.g., "int i = -1.234" 4056 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 4057 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 4058 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 4059 4060 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 4061 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 4062 } else { 4063 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 4064 } 4065 } 4066 4067 return; 4068 } 4069 4070 if (!Source->isIntegerType() || !Target->isIntegerType()) 4071 return; 4072 4073 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 4074 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 4075 S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer) 4076 << E->getSourceRange() << clang::SourceRange(CC); 4077 return; 4078 } 4079 4080 IntRange SourceRange = GetExprRange(S.Context, E); 4081 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 4082 4083 if (SourceRange.Width > TargetRange.Width) { 4084 // If the source is a constant, use a default-on diagnostic. 4085 // TODO: this should happen for bitfield stores, too. 4086 llvm::APSInt Value(32); 4087 if (E->isIntegerConstantExpr(Value, S.Context)) { 4088 if (S.SourceMgr.isInSystemMacro(CC)) 4089 return; 4090 4091 std::string PrettySourceValue = Value.toString(10); 4092 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 4093 4094 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4095 S.PDiag(diag::warn_impcast_integer_precision_constant) 4096 << PrettySourceValue << PrettyTargetValue 4097 << E->getType() << T << E->getSourceRange() 4098 << clang::SourceRange(CC)); 4099 return; 4100 } 4101 4102 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 4103 if (S.SourceMgr.isInSystemMacro(CC)) 4104 return; 4105 4106 if (SourceRange.Width == 64 && TargetRange.Width == 32) 4107 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 4108 /* pruneControlFlow */ true); 4109 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 4110 } 4111 4112 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 4113 (!TargetRange.NonNegative && SourceRange.NonNegative && 4114 SourceRange.Width == TargetRange.Width)) { 4115 4116 if (S.SourceMgr.isInSystemMacro(CC)) 4117 return; 4118 4119 unsigned DiagID = diag::warn_impcast_integer_sign; 4120 4121 // Traditionally, gcc has warned about this under -Wsign-compare. 4122 // We also want to warn about it in -Wconversion. 4123 // So if -Wconversion is off, use a completely identical diagnostic 4124 // in the sign-compare group. 4125 // The conditional-checking code will 4126 if (ICContext) { 4127 DiagID = diag::warn_impcast_integer_sign_conditional; 4128 *ICContext = true; 4129 } 4130 4131 return DiagnoseImpCast(S, E, T, CC, DiagID); 4132 } 4133 4134 // Diagnose conversions between different enumeration types. 4135 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 4136 // type, to give us better diagnostics. 4137 QualType SourceType = E->getType(); 4138 if (!S.getLangOptions().CPlusPlus) { 4139 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 4140 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 4141 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 4142 SourceType = S.Context.getTypeDeclType(Enum); 4143 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 4144 } 4145 } 4146 4147 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 4148 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 4149 if ((SourceEnum->getDecl()->getIdentifier() || 4150 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 4151 (TargetEnum->getDecl()->getIdentifier() || 4152 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 4153 SourceEnum != TargetEnum) { 4154 if (S.SourceMgr.isInSystemMacro(CC)) 4155 return; 4156 4157 return DiagnoseImpCast(S, E, SourceType, T, CC, 4158 diag::warn_impcast_different_enum_types); 4159 } 4160 4161 return; 4162} 4163 4164void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 4165 4166void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 4167 SourceLocation CC, bool &ICContext) { 4168 E = E->IgnoreParenImpCasts(); 4169 4170 if (isa<ConditionalOperator>(E)) 4171 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 4172 4173 AnalyzeImplicitConversions(S, E, CC); 4174 if (E->getType() != T) 4175 return CheckImplicitConversion(S, E, T, CC, &ICContext); 4176 return; 4177} 4178 4179void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 4180 SourceLocation CC = E->getQuestionLoc(); 4181 4182 AnalyzeImplicitConversions(S, E->getCond(), CC); 4183 4184 bool Suspicious = false; 4185 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 4186 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 4187 4188 // If -Wconversion would have warned about either of the candidates 4189 // for a signedness conversion to the context type... 4190 if (!Suspicious) return; 4191 4192 // ...but it's currently ignored... 4193 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 4194 CC)) 4195 return; 4196 4197 // ...then check whether it would have warned about either of the 4198 // candidates for a signedness conversion to the condition type. 4199 if (E->getType() == T) return; 4200 4201 Suspicious = false; 4202 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 4203 E->getType(), CC, &Suspicious); 4204 if (!Suspicious) 4205 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 4206 E->getType(), CC, &Suspicious); 4207} 4208 4209/// AnalyzeImplicitConversions - Find and report any interesting 4210/// implicit conversions in the given expression. There are a couple 4211/// of competing diagnostics here, -Wconversion and -Wsign-compare. 4212void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 4213 QualType T = OrigE->getType(); 4214 Expr *E = OrigE->IgnoreParenImpCasts(); 4215 4216 if (E->isTypeDependent() || E->isValueDependent()) 4217 return; 4218 4219 // For conditional operators, we analyze the arguments as if they 4220 // were being fed directly into the output. 4221 if (isa<ConditionalOperator>(E)) { 4222 ConditionalOperator *CO = cast<ConditionalOperator>(E); 4223 CheckConditionalOperator(S, CO, T); 4224 return; 4225 } 4226 4227 // Go ahead and check any implicit conversions we might have skipped. 4228 // The non-canonical typecheck is just an optimization; 4229 // CheckImplicitConversion will filter out dead implicit conversions. 4230 if (E->getType() != T) 4231 CheckImplicitConversion(S, E, T, CC); 4232 4233 // Now continue drilling into this expression. 4234 4235 // Skip past explicit casts. 4236 if (isa<ExplicitCastExpr>(E)) { 4237 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 4238 return AnalyzeImplicitConversions(S, E, CC); 4239 } 4240 4241 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4242 // Do a somewhat different check with comparison operators. 4243 if (BO->isComparisonOp()) 4244 return AnalyzeComparison(S, BO); 4245 4246 // And with simple assignments. 4247 if (BO->getOpcode() == BO_Assign) 4248 return AnalyzeAssignment(S, BO); 4249 } 4250 4251 // These break the otherwise-useful invariant below. Fortunately, 4252 // we don't really need to recurse into them, because any internal 4253 // expressions should have been analyzed already when they were 4254 // built into statements. 4255 if (isa<StmtExpr>(E)) return; 4256 4257 // Don't descend into unevaluated contexts. 4258 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 4259 4260 // Now just recurse over the expression's children. 4261 CC = E->getExprLoc(); 4262 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 4263 bool IsLogicalOperator = BO && BO->isLogicalOp(); 4264 for (Stmt::child_range I = E->children(); I; ++I) { 4265 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 4266 if (!ChildExpr) 4267 continue; 4268 4269 if (IsLogicalOperator && 4270 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 4271 // Ignore checking string literals that are in logical operators. 4272 continue; 4273 AnalyzeImplicitConversions(S, ChildExpr, CC); 4274 } 4275} 4276 4277} // end anonymous namespace 4278 4279/// Diagnoses "dangerous" implicit conversions within the given 4280/// expression (which is a full expression). Implements -Wconversion 4281/// and -Wsign-compare. 4282/// 4283/// \param CC the "context" location of the implicit conversion, i.e. 4284/// the most location of the syntactic entity requiring the implicit 4285/// conversion 4286void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 4287 // Don't diagnose in unevaluated contexts. 4288 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 4289 return; 4290 4291 // Don't diagnose for value- or type-dependent expressions. 4292 if (E->isTypeDependent() || E->isValueDependent()) 4293 return; 4294 4295 // Check for array bounds violations in cases where the check isn't triggered 4296 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 4297 // ArraySubscriptExpr is on the RHS of a variable initialization. 4298 CheckArrayAccess(E); 4299 4300 // This is not the right CC for (e.g.) a variable initialization. 4301 AnalyzeImplicitConversions(*this, E, CC); 4302} 4303 4304void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 4305 FieldDecl *BitField, 4306 Expr *Init) { 4307 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 4308} 4309 4310/// CheckParmsForFunctionDef - Check that the parameters of the given 4311/// function are appropriate for the definition of a function. This 4312/// takes care of any checks that cannot be performed on the 4313/// declaration itself, e.g., that the types of each of the function 4314/// parameters are complete. 4315bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 4316 bool CheckParameterNames) { 4317 bool HasInvalidParm = false; 4318 for (; P != PEnd; ++P) { 4319 ParmVarDecl *Param = *P; 4320 4321 // C99 6.7.5.3p4: the parameters in a parameter type list in a 4322 // function declarator that is part of a function definition of 4323 // that function shall not have incomplete type. 4324 // 4325 // This is also C++ [dcl.fct]p6. 4326 if (!Param->isInvalidDecl() && 4327 RequireCompleteType(Param->getLocation(), Param->getType(), 4328 diag::err_typecheck_decl_incomplete_type)) { 4329 Param->setInvalidDecl(); 4330 HasInvalidParm = true; 4331 } 4332 4333 // C99 6.9.1p5: If the declarator includes a parameter type list, the 4334 // declaration of each parameter shall include an identifier. 4335 if (CheckParameterNames && 4336 Param->getIdentifier() == 0 && 4337 !Param->isImplicit() && 4338 !getLangOptions().CPlusPlus) 4339 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 4340 4341 // C99 6.7.5.3p12: 4342 // If the function declarator is not part of a definition of that 4343 // function, parameters may have incomplete type and may use the [*] 4344 // notation in their sequences of declarator specifiers to specify 4345 // variable length array types. 4346 QualType PType = Param->getOriginalType(); 4347 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 4348 if (AT->getSizeModifier() == ArrayType::Star) { 4349 // FIXME: This diagnosic should point the the '[*]' if source-location 4350 // information is added for it. 4351 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 4352 } 4353 } 4354 } 4355 4356 return HasInvalidParm; 4357} 4358 4359/// CheckCastAlign - Implements -Wcast-align, which warns when a 4360/// pointer cast increases the alignment requirements. 4361void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 4362 // This is actually a lot of work to potentially be doing on every 4363 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 4364 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 4365 TRange.getBegin()) 4366 == DiagnosticsEngine::Ignored) 4367 return; 4368 4369 // Ignore dependent types. 4370 if (T->isDependentType() || Op->getType()->isDependentType()) 4371 return; 4372 4373 // Require that the destination be a pointer type. 4374 const PointerType *DestPtr = T->getAs<PointerType>(); 4375 if (!DestPtr) return; 4376 4377 // If the destination has alignment 1, we're done. 4378 QualType DestPointee = DestPtr->getPointeeType(); 4379 if (DestPointee->isIncompleteType()) return; 4380 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 4381 if (DestAlign.isOne()) return; 4382 4383 // Require that the source be a pointer type. 4384 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 4385 if (!SrcPtr) return; 4386 QualType SrcPointee = SrcPtr->getPointeeType(); 4387 4388 // Whitelist casts from cv void*. We already implicitly 4389 // whitelisted casts to cv void*, since they have alignment 1. 4390 // Also whitelist casts involving incomplete types, which implicitly 4391 // includes 'void'. 4392 if (SrcPointee->isIncompleteType()) return; 4393 4394 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 4395 if (SrcAlign >= DestAlign) return; 4396 4397 Diag(TRange.getBegin(), diag::warn_cast_align) 4398 << Op->getType() << T 4399 << static_cast<unsigned>(SrcAlign.getQuantity()) 4400 << static_cast<unsigned>(DestAlign.getQuantity()) 4401 << TRange << Op->getSourceRange(); 4402} 4403 4404static const Type* getElementType(const Expr *BaseExpr) { 4405 const Type* EltType = BaseExpr->getType().getTypePtr(); 4406 if (EltType->isAnyPointerType()) 4407 return EltType->getPointeeType().getTypePtr(); 4408 else if (EltType->isArrayType()) 4409 return EltType->getBaseElementTypeUnsafe(); 4410 return EltType; 4411} 4412 4413/// \brief Check whether this array fits the idiom of a size-one tail padded 4414/// array member of a struct. 4415/// 4416/// We avoid emitting out-of-bounds access warnings for such arrays as they are 4417/// commonly used to emulate flexible arrays in C89 code. 4418static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 4419 const NamedDecl *ND) { 4420 if (Size != 1 || !ND) return false; 4421 4422 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 4423 if (!FD) return false; 4424 4425 // Don't consider sizes resulting from macro expansions or template argument 4426 // substitution to form C89 tail-padded arrays. 4427 ConstantArrayTypeLoc TL = 4428 cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc()); 4429 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr()); 4430 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 4431 return false; 4432 4433 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 4434 if (!RD) return false; 4435 if (RD->isUnion()) return false; 4436 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 4437 if (!CRD->isStandardLayout()) return false; 4438 } 4439 4440 // See if this is the last field decl in the record. 4441 const Decl *D = FD; 4442 while ((D = D->getNextDeclInContext())) 4443 if (isa<FieldDecl>(D)) 4444 return false; 4445 return true; 4446} 4447 4448void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 4449 const ArraySubscriptExpr *ASE, 4450 bool AllowOnePastEnd, bool IndexNegated) { 4451 IndexExpr = IndexExpr->IgnoreParenCasts(); 4452 if (IndexExpr->isValueDependent()) 4453 return; 4454 4455 const Type *EffectiveType = getElementType(BaseExpr); 4456 BaseExpr = BaseExpr->IgnoreParenCasts(); 4457 const ConstantArrayType *ArrayTy = 4458 Context.getAsConstantArrayType(BaseExpr->getType()); 4459 if (!ArrayTy) 4460 return; 4461 4462 llvm::APSInt index; 4463 if (!IndexExpr->EvaluateAsInt(index, Context)) 4464 return; 4465 if (IndexNegated) 4466 index = -index; 4467 4468 const NamedDecl *ND = NULL; 4469 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4470 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4471 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4472 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4473 4474 if (index.isUnsigned() || !index.isNegative()) { 4475 llvm::APInt size = ArrayTy->getSize(); 4476 if (!size.isStrictlyPositive()) 4477 return; 4478 4479 const Type* BaseType = getElementType(BaseExpr); 4480 if (BaseType != EffectiveType) { 4481 // Make sure we're comparing apples to apples when comparing index to size 4482 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 4483 uint64_t array_typesize = Context.getTypeSize(BaseType); 4484 // Handle ptrarith_typesize being zero, such as when casting to void* 4485 if (!ptrarith_typesize) ptrarith_typesize = 1; 4486 if (ptrarith_typesize != array_typesize) { 4487 // There's a cast to a different size type involved 4488 uint64_t ratio = array_typesize / ptrarith_typesize; 4489 // TODO: Be smarter about handling cases where array_typesize is not a 4490 // multiple of ptrarith_typesize 4491 if (ptrarith_typesize * ratio == array_typesize) 4492 size *= llvm::APInt(size.getBitWidth(), ratio); 4493 } 4494 } 4495 4496 if (size.getBitWidth() > index.getBitWidth()) 4497 index = index.sext(size.getBitWidth()); 4498 else if (size.getBitWidth() < index.getBitWidth()) 4499 size = size.sext(index.getBitWidth()); 4500 4501 // For array subscripting the index must be less than size, but for pointer 4502 // arithmetic also allow the index (offset) to be equal to size since 4503 // computing the next address after the end of the array is legal and 4504 // commonly done e.g. in C++ iterators and range-based for loops. 4505 if (AllowOnePastEnd ? index.sle(size) : index.slt(size)) 4506 return; 4507 4508 // Also don't warn for arrays of size 1 which are members of some 4509 // structure. These are often used to approximate flexible arrays in C89 4510 // code. 4511 if (IsTailPaddedMemberArray(*this, size, ND)) 4512 return; 4513 4514 // Suppress the warning if the subscript expression (as identified by the 4515 // ']' location) and the index expression are both from macro expansions 4516 // within a system header. 4517 if (ASE) { 4518 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 4519 ASE->getRBracketLoc()); 4520 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 4521 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 4522 IndexExpr->getLocStart()); 4523 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 4524 return; 4525 } 4526 } 4527 4528 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 4529 if (ASE) 4530 DiagID = diag::warn_array_index_exceeds_bounds; 4531 4532 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4533 PDiag(DiagID) << index.toString(10, true) 4534 << size.toString(10, true) 4535 << (unsigned)size.getLimitedValue(~0U) 4536 << IndexExpr->getSourceRange()); 4537 } else { 4538 unsigned DiagID = diag::warn_array_index_precedes_bounds; 4539 if (!ASE) { 4540 DiagID = diag::warn_ptr_arith_precedes_bounds; 4541 if (index.isNegative()) index = -index; 4542 } 4543 4544 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4545 PDiag(DiagID) << index.toString(10, true) 4546 << IndexExpr->getSourceRange()); 4547 } 4548 4549 if (!ND) { 4550 // Try harder to find a NamedDecl to point at in the note. 4551 while (const ArraySubscriptExpr *ASE = 4552 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 4553 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 4554 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4555 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4556 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4557 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4558 } 4559 4560 if (ND) 4561 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 4562 PDiag(diag::note_array_index_out_of_bounds) 4563 << ND->getDeclName()); 4564} 4565 4566void Sema::CheckArrayAccess(const Expr *expr) { 4567 int AllowOnePastEnd = 0; 4568 while (expr) { 4569 expr = expr->IgnoreParenImpCasts(); 4570 switch (expr->getStmtClass()) { 4571 case Stmt::ArraySubscriptExprClass: { 4572 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 4573 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 4574 AllowOnePastEnd > 0); 4575 return; 4576 } 4577 case Stmt::UnaryOperatorClass: { 4578 // Only unwrap the * and & unary operators 4579 const UnaryOperator *UO = cast<UnaryOperator>(expr); 4580 expr = UO->getSubExpr(); 4581 switch (UO->getOpcode()) { 4582 case UO_AddrOf: 4583 AllowOnePastEnd++; 4584 break; 4585 case UO_Deref: 4586 AllowOnePastEnd--; 4587 break; 4588 default: 4589 return; 4590 } 4591 break; 4592 } 4593 case Stmt::ConditionalOperatorClass: { 4594 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 4595 if (const Expr *lhs = cond->getLHS()) 4596 CheckArrayAccess(lhs); 4597 if (const Expr *rhs = cond->getRHS()) 4598 CheckArrayAccess(rhs); 4599 return; 4600 } 4601 default: 4602 return; 4603 } 4604 } 4605} 4606 4607//===--- CHECK: Objective-C retain cycles ----------------------------------// 4608 4609namespace { 4610 struct RetainCycleOwner { 4611 RetainCycleOwner() : Variable(0), Indirect(false) {} 4612 VarDecl *Variable; 4613 SourceRange Range; 4614 SourceLocation Loc; 4615 bool Indirect; 4616 4617 void setLocsFrom(Expr *e) { 4618 Loc = e->getExprLoc(); 4619 Range = e->getSourceRange(); 4620 } 4621 }; 4622} 4623 4624/// Consider whether capturing the given variable can possibly lead to 4625/// a retain cycle. 4626static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 4627 // In ARC, it's captured strongly iff the variable has __strong 4628 // lifetime. In MRR, it's captured strongly if the variable is 4629 // __block and has an appropriate type. 4630 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4631 return false; 4632 4633 owner.Variable = var; 4634 owner.setLocsFrom(ref); 4635 return true; 4636} 4637 4638static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 4639 while (true) { 4640 e = e->IgnoreParens(); 4641 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 4642 switch (cast->getCastKind()) { 4643 case CK_BitCast: 4644 case CK_LValueBitCast: 4645 case CK_LValueToRValue: 4646 case CK_ARCReclaimReturnedObject: 4647 e = cast->getSubExpr(); 4648 continue; 4649 4650 default: 4651 return false; 4652 } 4653 } 4654 4655 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 4656 ObjCIvarDecl *ivar = ref->getDecl(); 4657 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4658 return false; 4659 4660 // Try to find a retain cycle in the base. 4661 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 4662 return false; 4663 4664 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 4665 owner.Indirect = true; 4666 return true; 4667 } 4668 4669 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 4670 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 4671 if (!var) return false; 4672 return considerVariable(var, ref, owner); 4673 } 4674 4675 if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) { 4676 owner.Variable = ref->getDecl(); 4677 owner.setLocsFrom(ref); 4678 return true; 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 VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) { 4738 if (ref->getDecl() == Variable && !Capturer) 4739 Capturer = ref; 4740 } 4741 4742 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 4743 if (Capturer) return; 4744 Visit(ref->getBase()); 4745 if (Capturer && ref->isFreeIvar()) 4746 Capturer = ref; 4747 } 4748 4749 void VisitBlockExpr(BlockExpr *block) { 4750 // Look inside nested blocks 4751 if (block->getBlockDecl()->capturesVariable(Variable)) 4752 Visit(block->getBlockDecl()->getBody()); 4753 } 4754 }; 4755} 4756 4757/// Check whether the given argument is a block which captures a 4758/// variable. 4759static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 4760 assert(owner.Variable && owner.Loc.isValid()); 4761 4762 e = e->IgnoreParenCasts(); 4763 BlockExpr *block = dyn_cast<BlockExpr>(e); 4764 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 4765 return 0; 4766 4767 FindCaptureVisitor visitor(S.Context, owner.Variable); 4768 visitor.Visit(block->getBlockDecl()->getBody()); 4769 return visitor.Capturer; 4770} 4771 4772static void diagnoseRetainCycle(Sema &S, Expr *capturer, 4773 RetainCycleOwner &owner) { 4774 assert(capturer); 4775 assert(owner.Variable && owner.Loc.isValid()); 4776 4777 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 4778 << owner.Variable << capturer->getSourceRange(); 4779 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 4780 << owner.Indirect << owner.Range; 4781} 4782 4783/// Check for a keyword selector that starts with the word 'add' or 4784/// 'set'. 4785static bool isSetterLikeSelector(Selector sel) { 4786 if (sel.isUnarySelector()) return false; 4787 4788 StringRef str = sel.getNameForSlot(0); 4789 while (!str.empty() && str.front() == '_') str = str.substr(1); 4790 if (str.startswith("set")) 4791 str = str.substr(3); 4792 else if (str.startswith("add")) { 4793 // Specially whitelist 'addOperationWithBlock:'. 4794 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 4795 return false; 4796 str = str.substr(3); 4797 } 4798 else 4799 return false; 4800 4801 if (str.empty()) return true; 4802 return !islower(str.front()); 4803} 4804 4805/// Check a message send to see if it's likely to cause a retain cycle. 4806void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 4807 // Only check instance methods whose selector looks like a setter. 4808 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 4809 return; 4810 4811 // Try to find a variable that the receiver is strongly owned by. 4812 RetainCycleOwner owner; 4813 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 4814 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 4815 return; 4816 } else { 4817 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 4818 owner.Variable = getCurMethodDecl()->getSelfDecl(); 4819 owner.Loc = msg->getSuperLoc(); 4820 owner.Range = msg->getSuperLoc(); 4821 } 4822 4823 // Check whether the receiver is captured by any of the arguments. 4824 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 4825 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 4826 return diagnoseRetainCycle(*this, capturer, owner); 4827} 4828 4829/// Check a property assign to see if it's likely to cause a retain cycle. 4830void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 4831 RetainCycleOwner owner; 4832 if (!findRetainCycleOwner(*this, receiver, owner)) 4833 return; 4834 4835 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 4836 diagnoseRetainCycle(*this, capturer, owner); 4837} 4838 4839bool Sema::checkUnsafeAssigns(SourceLocation Loc, 4840 QualType LHS, Expr *RHS) { 4841 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 4842 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 4843 return false; 4844 // strip off any implicit cast added to get to the one arc-specific 4845 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4846 if (cast->getCastKind() == CK_ARCConsumeObject) { 4847 Diag(Loc, diag::warn_arc_retained_assign) 4848 << (LT == Qualifiers::OCL_ExplicitNone) 4849 << RHS->getSourceRange(); 4850 return true; 4851 } 4852 RHS = cast->getSubExpr(); 4853 } 4854 return false; 4855} 4856 4857void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 4858 Expr *LHS, Expr *RHS) { 4859 QualType LHSType; 4860 // PropertyRef on LHS type need be directly obtained from 4861 // its declaration as it has a PsuedoType. 4862 ObjCPropertyRefExpr *PRE 4863 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 4864 if (PRE && !PRE->isImplicitProperty()) { 4865 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4866 if (PD) 4867 LHSType = PD->getType(); 4868 } 4869 4870 if (LHSType.isNull()) 4871 LHSType = LHS->getType(); 4872 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 4873 return; 4874 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 4875 // FIXME. Check for other life times. 4876 if (LT != Qualifiers::OCL_None) 4877 return; 4878 4879 if (PRE) { 4880 if (PRE->isImplicitProperty()) 4881 return; 4882 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4883 if (!PD) 4884 return; 4885 4886 unsigned Attributes = PD->getPropertyAttributes(); 4887 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 4888 // when 'assign' attribute was not explicitly specified 4889 // by user, ignore it and rely on property type itself 4890 // for lifetime info. 4891 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 4892 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 4893 LHSType->isObjCRetainableType()) 4894 return; 4895 4896 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4897 if (cast->getCastKind() == CK_ARCConsumeObject) { 4898 Diag(Loc, diag::warn_arc_retained_property_assign) 4899 << RHS->getSourceRange(); 4900 return; 4901 } 4902 RHS = cast->getSubExpr(); 4903 } 4904 } 4905 } 4906} 4907 4908//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 4909 4910namespace { 4911bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 4912 SourceLocation StmtLoc, 4913 const NullStmt *Body) { 4914 // Do not warn if the body is a macro that expands to nothing, e.g: 4915 // 4916 // #define CALL(x) 4917 // if (condition) 4918 // CALL(0); 4919 // 4920 if (Body->hasLeadingEmptyMacro()) 4921 return false; 4922 4923 // Get line numbers of statement and body. 4924 bool StmtLineInvalid; 4925 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 4926 &StmtLineInvalid); 4927 if (StmtLineInvalid) 4928 return false; 4929 4930 bool BodyLineInvalid; 4931 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 4932 &BodyLineInvalid); 4933 if (BodyLineInvalid) 4934 return false; 4935 4936 // Warn if null statement and body are on the same line. 4937 if (StmtLine != BodyLine) 4938 return false; 4939 4940 return true; 4941} 4942} // Unnamed namespace 4943 4944void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 4945 const Stmt *Body, 4946 unsigned DiagID) { 4947 // Since this is a syntactic check, don't emit diagnostic for template 4948 // instantiations, this just adds noise. 4949 if (CurrentInstantiationScope) 4950 return; 4951 4952 // The body should be a null statement. 4953 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 4954 if (!NBody) 4955 return; 4956 4957 // Do the usual checks. 4958 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 4959 return; 4960 4961 Diag(NBody->getSemiLoc(), DiagID); 4962 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 4963} 4964 4965void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 4966 const Stmt *PossibleBody) { 4967 assert(!CurrentInstantiationScope); // Ensured by caller 4968 4969 SourceLocation StmtLoc; 4970 const Stmt *Body; 4971 unsigned DiagID; 4972 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 4973 StmtLoc = FS->getRParenLoc(); 4974 Body = FS->getBody(); 4975 DiagID = diag::warn_empty_for_body; 4976 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 4977 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 4978 Body = WS->getBody(); 4979 DiagID = diag::warn_empty_while_body; 4980 } else 4981 return; // Neither `for' nor `while'. 4982 4983 // The body should be a null statement. 4984 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 4985 if (!NBody) 4986 return; 4987 4988 // Skip expensive checks if diagnostic is disabled. 4989 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 4990 DiagnosticsEngine::Ignored) 4991 return; 4992 4993 // Do the usual checks. 4994 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 4995 return; 4996 4997 // `for(...);' and `while(...);' are popular idioms, so in order to keep 4998 // noise level low, emit diagnostics only if for/while is followed by a 4999 // CompoundStmt, e.g.: 5000 // for (int i = 0; i < n; i++); 5001 // { 5002 // a(i); 5003 // } 5004 // or if for/while is followed by a statement with more indentation 5005 // than for/while itself: 5006 // for (int i = 0; i < n; i++); 5007 // a(i); 5008 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 5009 if (!ProbableTypo) { 5010 bool BodyColInvalid; 5011 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 5012 PossibleBody->getLocStart(), 5013 &BodyColInvalid); 5014 if (BodyColInvalid) 5015 return; 5016 5017 bool StmtColInvalid; 5018 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 5019 S->getLocStart(), 5020 &StmtColInvalid); 5021 if (StmtColInvalid) 5022 return; 5023 5024 if (BodyCol > StmtCol) 5025 ProbableTypo = true; 5026 } 5027 5028 if (ProbableTypo) { 5029 Diag(NBody->getSemiLoc(), DiagID); 5030 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5031 } 5032} 5033