SemaChecking.cpp revision 5e25301c24c92a9b7018cee20e524c4eb7192bf0
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements extra semantic analysis beyond what is enforced 11// by the C type system. 12// 13//===----------------------------------------------------------------------===// 14 15#include "clang/Sema/SemaInternal.h" 16#include "clang/AST/ASTContext.h" 17#include "clang/AST/CharUnits.h" 18#include "clang/AST/DeclCXX.h" 19#include "clang/AST/DeclObjC.h" 20#include "clang/AST/EvaluatedExprVisitor.h" 21#include "clang/AST/Expr.h" 22#include "clang/AST/ExprCXX.h" 23#include "clang/AST/ExprObjC.h" 24#include "clang/AST/StmtCXX.h" 25#include "clang/AST/StmtObjC.h" 26#include "clang/Analysis/Analyses/FormatString.h" 27#include "clang/Basic/TargetBuiltins.h" 28#include "clang/Basic/TargetInfo.h" 29#include "clang/Lex/Preprocessor.h" 30#include "clang/Sema/Initialization.h" 31#include "clang/Sema/Lookup.h" 32#include "clang/Sema/ScopeInfo.h" 33#include "clang/Sema/Sema.h" 34#include "llvm/ADT/BitVector.h" 35#include "llvm/ADT/STLExtras.h" 36#include "llvm/ADT/SmallString.h" 37#include "llvm/Support/ConvertUTF.h" 38#include "llvm/Support/raw_ostream.h" 39#include <limits> 40using namespace clang; 41using namespace sema; 42 43SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 44 unsigned ByteNo) const { 45 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 46 PP.getLangOpts(), PP.getTargetInfo()); 47} 48 49/// Checks that a call expression's argument count is the desired number. 50/// This is useful when doing custom type-checking. Returns true on error. 51static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 52 unsigned argCount = call->getNumArgs(); 53 if (argCount == desiredArgCount) return false; 54 55 if (argCount < desiredArgCount) 56 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 57 << 0 /*function call*/ << desiredArgCount << argCount 58 << call->getSourceRange(); 59 60 // Highlight all the excess arguments. 61 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 62 call->getArg(argCount - 1)->getLocEnd()); 63 64 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 65 << 0 /*function call*/ << desiredArgCount << argCount 66 << call->getArg(1)->getSourceRange(); 67} 68 69/// Check that the first argument to __builtin_annotation is an integer 70/// and the second argument is a non-wide string literal. 71static bool SemaBuiltinAnnotation(Sema &S, CallExpr *TheCall) { 72 if (checkArgCount(S, TheCall, 2)) 73 return true; 74 75 // First argument should be an integer. 76 Expr *ValArg = TheCall->getArg(0); 77 QualType Ty = ValArg->getType(); 78 if (!Ty->isIntegerType()) { 79 S.Diag(ValArg->getLocStart(), diag::err_builtin_annotation_first_arg) 80 << ValArg->getSourceRange(); 81 return true; 82 } 83 84 // Second argument should be a constant string. 85 Expr *StrArg = TheCall->getArg(1)->IgnoreParenCasts(); 86 StringLiteral *Literal = dyn_cast<StringLiteral>(StrArg); 87 if (!Literal || !Literal->isAscii()) { 88 S.Diag(StrArg->getLocStart(), diag::err_builtin_annotation_second_arg) 89 << StrArg->getSourceRange(); 90 return true; 91 } 92 93 TheCall->setType(Ty); 94 return false; 95} 96 97ExprResult 98Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 99 ExprResult TheCallResult(Owned(TheCall)); 100 101 // Find out if any arguments are required to be integer constant expressions. 102 unsigned ICEArguments = 0; 103 ASTContext::GetBuiltinTypeError Error; 104 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 105 if (Error != ASTContext::GE_None) 106 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 107 108 // If any arguments are required to be ICE's, check and diagnose. 109 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 110 // Skip arguments not required to be ICE's. 111 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 112 113 llvm::APSInt Result; 114 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 115 return true; 116 ICEArguments &= ~(1 << ArgNo); 117 } 118 119 switch (BuiltinID) { 120 case Builtin::BI__builtin___CFStringMakeConstantString: 121 assert(TheCall->getNumArgs() == 1 && 122 "Wrong # arguments to builtin CFStringMakeConstantString"); 123 if (CheckObjCString(TheCall->getArg(0))) 124 return ExprError(); 125 break; 126 case Builtin::BI__builtin_stdarg_start: 127 case Builtin::BI__builtin_va_start: 128 if (SemaBuiltinVAStart(TheCall)) 129 return ExprError(); 130 break; 131 case Builtin::BI__builtin_isgreater: 132 case Builtin::BI__builtin_isgreaterequal: 133 case Builtin::BI__builtin_isless: 134 case Builtin::BI__builtin_islessequal: 135 case Builtin::BI__builtin_islessgreater: 136 case Builtin::BI__builtin_isunordered: 137 if (SemaBuiltinUnorderedCompare(TheCall)) 138 return ExprError(); 139 break; 140 case Builtin::BI__builtin_fpclassify: 141 if (SemaBuiltinFPClassification(TheCall, 6)) 142 return ExprError(); 143 break; 144 case Builtin::BI__builtin_isfinite: 145 case Builtin::BI__builtin_isinf: 146 case Builtin::BI__builtin_isinf_sign: 147 case Builtin::BI__builtin_isnan: 148 case Builtin::BI__builtin_isnormal: 149 if (SemaBuiltinFPClassification(TheCall, 1)) 150 return ExprError(); 151 break; 152 case Builtin::BI__builtin_shufflevector: 153 return SemaBuiltinShuffleVector(TheCall); 154 // TheCall will be freed by the smart pointer here, but that's fine, since 155 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 156 case Builtin::BI__builtin_prefetch: 157 if (SemaBuiltinPrefetch(TheCall)) 158 return ExprError(); 159 break; 160 case Builtin::BI__builtin_object_size: 161 if (SemaBuiltinObjectSize(TheCall)) 162 return ExprError(); 163 break; 164 case Builtin::BI__builtin_longjmp: 165 if (SemaBuiltinLongjmp(TheCall)) 166 return ExprError(); 167 break; 168 169 case Builtin::BI__builtin_classify_type: 170 if (checkArgCount(*this, TheCall, 1)) return true; 171 TheCall->setType(Context.IntTy); 172 break; 173 case Builtin::BI__builtin_constant_p: 174 if (checkArgCount(*this, TheCall, 1)) return true; 175 TheCall->setType(Context.IntTy); 176 break; 177 case Builtin::BI__sync_fetch_and_add: 178 case Builtin::BI__sync_fetch_and_add_1: 179 case Builtin::BI__sync_fetch_and_add_2: 180 case Builtin::BI__sync_fetch_and_add_4: 181 case Builtin::BI__sync_fetch_and_add_8: 182 case Builtin::BI__sync_fetch_and_add_16: 183 case Builtin::BI__sync_fetch_and_sub: 184 case Builtin::BI__sync_fetch_and_sub_1: 185 case Builtin::BI__sync_fetch_and_sub_2: 186 case Builtin::BI__sync_fetch_and_sub_4: 187 case Builtin::BI__sync_fetch_and_sub_8: 188 case Builtin::BI__sync_fetch_and_sub_16: 189 case Builtin::BI__sync_fetch_and_or: 190 case Builtin::BI__sync_fetch_and_or_1: 191 case Builtin::BI__sync_fetch_and_or_2: 192 case Builtin::BI__sync_fetch_and_or_4: 193 case Builtin::BI__sync_fetch_and_or_8: 194 case Builtin::BI__sync_fetch_and_or_16: 195 case Builtin::BI__sync_fetch_and_and: 196 case Builtin::BI__sync_fetch_and_and_1: 197 case Builtin::BI__sync_fetch_and_and_2: 198 case Builtin::BI__sync_fetch_and_and_4: 199 case Builtin::BI__sync_fetch_and_and_8: 200 case Builtin::BI__sync_fetch_and_and_16: 201 case Builtin::BI__sync_fetch_and_xor: 202 case Builtin::BI__sync_fetch_and_xor_1: 203 case Builtin::BI__sync_fetch_and_xor_2: 204 case Builtin::BI__sync_fetch_and_xor_4: 205 case Builtin::BI__sync_fetch_and_xor_8: 206 case Builtin::BI__sync_fetch_and_xor_16: 207 case Builtin::BI__sync_add_and_fetch: 208 case Builtin::BI__sync_add_and_fetch_1: 209 case Builtin::BI__sync_add_and_fetch_2: 210 case Builtin::BI__sync_add_and_fetch_4: 211 case Builtin::BI__sync_add_and_fetch_8: 212 case Builtin::BI__sync_add_and_fetch_16: 213 case Builtin::BI__sync_sub_and_fetch: 214 case Builtin::BI__sync_sub_and_fetch_1: 215 case Builtin::BI__sync_sub_and_fetch_2: 216 case Builtin::BI__sync_sub_and_fetch_4: 217 case Builtin::BI__sync_sub_and_fetch_8: 218 case Builtin::BI__sync_sub_and_fetch_16: 219 case Builtin::BI__sync_and_and_fetch: 220 case Builtin::BI__sync_and_and_fetch_1: 221 case Builtin::BI__sync_and_and_fetch_2: 222 case Builtin::BI__sync_and_and_fetch_4: 223 case Builtin::BI__sync_and_and_fetch_8: 224 case Builtin::BI__sync_and_and_fetch_16: 225 case Builtin::BI__sync_or_and_fetch: 226 case Builtin::BI__sync_or_and_fetch_1: 227 case Builtin::BI__sync_or_and_fetch_2: 228 case Builtin::BI__sync_or_and_fetch_4: 229 case Builtin::BI__sync_or_and_fetch_8: 230 case Builtin::BI__sync_or_and_fetch_16: 231 case Builtin::BI__sync_xor_and_fetch: 232 case Builtin::BI__sync_xor_and_fetch_1: 233 case Builtin::BI__sync_xor_and_fetch_2: 234 case Builtin::BI__sync_xor_and_fetch_4: 235 case Builtin::BI__sync_xor_and_fetch_8: 236 case Builtin::BI__sync_xor_and_fetch_16: 237 case Builtin::BI__sync_val_compare_and_swap: 238 case Builtin::BI__sync_val_compare_and_swap_1: 239 case Builtin::BI__sync_val_compare_and_swap_2: 240 case Builtin::BI__sync_val_compare_and_swap_4: 241 case Builtin::BI__sync_val_compare_and_swap_8: 242 case Builtin::BI__sync_val_compare_and_swap_16: 243 case Builtin::BI__sync_bool_compare_and_swap: 244 case Builtin::BI__sync_bool_compare_and_swap_1: 245 case Builtin::BI__sync_bool_compare_and_swap_2: 246 case Builtin::BI__sync_bool_compare_and_swap_4: 247 case Builtin::BI__sync_bool_compare_and_swap_8: 248 case Builtin::BI__sync_bool_compare_and_swap_16: 249 case Builtin::BI__sync_lock_test_and_set: 250 case Builtin::BI__sync_lock_test_and_set_1: 251 case Builtin::BI__sync_lock_test_and_set_2: 252 case Builtin::BI__sync_lock_test_and_set_4: 253 case Builtin::BI__sync_lock_test_and_set_8: 254 case Builtin::BI__sync_lock_test_and_set_16: 255 case Builtin::BI__sync_lock_release: 256 case Builtin::BI__sync_lock_release_1: 257 case Builtin::BI__sync_lock_release_2: 258 case Builtin::BI__sync_lock_release_4: 259 case Builtin::BI__sync_lock_release_8: 260 case Builtin::BI__sync_lock_release_16: 261 case Builtin::BI__sync_swap: 262 case Builtin::BI__sync_swap_1: 263 case Builtin::BI__sync_swap_2: 264 case Builtin::BI__sync_swap_4: 265 case Builtin::BI__sync_swap_8: 266 case Builtin::BI__sync_swap_16: 267 return SemaBuiltinAtomicOverloaded(TheCallResult); 268#define BUILTIN(ID, TYPE, ATTRS) 269#define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \ 270 case Builtin::BI##ID: \ 271 return SemaAtomicOpsOverloaded(TheCallResult, AtomicExpr::AO##ID); 272#include "clang/Basic/Builtins.def" 273 case Builtin::BI__builtin_annotation: 274 if (SemaBuiltinAnnotation(*this, TheCall)) 275 return ExprError(); 276 break; 277 } 278 279 // Since the target specific builtins for each arch overlap, only check those 280 // of the arch we are compiling for. 281 if (BuiltinID >= Builtin::FirstTSBuiltin) { 282 switch (Context.getTargetInfo().getTriple().getArch()) { 283 case llvm::Triple::arm: 284 case llvm::Triple::thumb: 285 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 286 return ExprError(); 287 break; 288 case llvm::Triple::mips: 289 case llvm::Triple::mipsel: 290 case llvm::Triple::mips64: 291 case llvm::Triple::mips64el: 292 if (CheckMipsBuiltinFunctionCall(BuiltinID, TheCall)) 293 return ExprError(); 294 break; 295 default: 296 break; 297 } 298 } 299 300 return TheCallResult; 301} 302 303// Get the valid immediate range for the specified NEON type code. 304static unsigned RFT(unsigned t, bool shift = false) { 305 NeonTypeFlags Type(t); 306 int IsQuad = Type.isQuad(); 307 switch (Type.getEltType()) { 308 case NeonTypeFlags::Int8: 309 case NeonTypeFlags::Poly8: 310 return shift ? 7 : (8 << IsQuad) - 1; 311 case NeonTypeFlags::Int16: 312 case NeonTypeFlags::Poly16: 313 return shift ? 15 : (4 << IsQuad) - 1; 314 case NeonTypeFlags::Int32: 315 return shift ? 31 : (2 << IsQuad) - 1; 316 case NeonTypeFlags::Int64: 317 return shift ? 63 : (1 << IsQuad) - 1; 318 case NeonTypeFlags::Float16: 319 assert(!shift && "cannot shift float types!"); 320 return (4 << IsQuad) - 1; 321 case NeonTypeFlags::Float32: 322 assert(!shift && "cannot shift float types!"); 323 return (2 << IsQuad) - 1; 324 } 325 llvm_unreachable("Invalid NeonTypeFlag!"); 326} 327 328/// getNeonEltType - Return the QualType corresponding to the elements of 329/// the vector type specified by the NeonTypeFlags. This is used to check 330/// the pointer arguments for Neon load/store intrinsics. 331static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) { 332 switch (Flags.getEltType()) { 333 case NeonTypeFlags::Int8: 334 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 335 case NeonTypeFlags::Int16: 336 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 337 case NeonTypeFlags::Int32: 338 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 339 case NeonTypeFlags::Int64: 340 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 341 case NeonTypeFlags::Poly8: 342 return Context.SignedCharTy; 343 case NeonTypeFlags::Poly16: 344 return Context.ShortTy; 345 case NeonTypeFlags::Float16: 346 return Context.UnsignedShortTy; 347 case NeonTypeFlags::Float32: 348 return Context.FloatTy; 349 } 350 llvm_unreachable("Invalid NeonTypeFlag!"); 351} 352 353bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 354 llvm::APSInt Result; 355 356 uint64_t mask = 0; 357 unsigned TV = 0; 358 int PtrArgNum = -1; 359 bool HasConstPtr = false; 360 switch (BuiltinID) { 361#define GET_NEON_OVERLOAD_CHECK 362#include "clang/Basic/arm_neon.inc" 363#undef GET_NEON_OVERLOAD_CHECK 364 } 365 366 // For NEON intrinsics which are overloaded on vector element type, validate 367 // the immediate which specifies which variant to emit. 368 unsigned ImmArg = TheCall->getNumArgs()-1; 369 if (mask) { 370 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 371 return true; 372 373 TV = Result.getLimitedValue(64); 374 if ((TV > 63) || (mask & (1ULL << TV)) == 0) 375 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 376 << TheCall->getArg(ImmArg)->getSourceRange(); 377 } 378 379 if (PtrArgNum >= 0) { 380 // Check that pointer arguments have the specified type. 381 Expr *Arg = TheCall->getArg(PtrArgNum); 382 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 383 Arg = ICE->getSubExpr(); 384 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 385 QualType RHSTy = RHS.get()->getType(); 386 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 387 if (HasConstPtr) 388 EltTy = EltTy.withConst(); 389 QualType LHSTy = Context.getPointerType(EltTy); 390 AssignConvertType ConvTy; 391 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 392 if (RHS.isInvalid()) 393 return true; 394 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 395 RHS.get(), AA_Assigning)) 396 return true; 397 } 398 399 // For NEON intrinsics which take an immediate value as part of the 400 // instruction, range check them here. 401 unsigned i = 0, l = 0, u = 0; 402 switch (BuiltinID) { 403 default: return false; 404 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 405 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 406 case ARM::BI__builtin_arm_vcvtr_f: 407 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 408#define GET_NEON_IMMEDIATE_CHECK 409#include "clang/Basic/arm_neon.inc" 410#undef GET_NEON_IMMEDIATE_CHECK 411 }; 412 413 // We can't check the value of a dependent argument. 414 if (TheCall->getArg(i)->isTypeDependent() || 415 TheCall->getArg(i)->isValueDependent()) 416 return false; 417 418 // Check that the immediate argument is actually a constant. 419 if (SemaBuiltinConstantArg(TheCall, i, Result)) 420 return true; 421 422 // Range check against the upper/lower values for this isntruction. 423 unsigned Val = Result.getZExtValue(); 424 if (Val < l || Val > (u + l)) 425 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 426 << l << u+l << TheCall->getArg(i)->getSourceRange(); 427 428 // FIXME: VFP Intrinsics should error if VFP not present. 429 return false; 430} 431 432bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 433 unsigned i = 0, l = 0, u = 0; 434 switch (BuiltinID) { 435 default: return false; 436 case Mips::BI__builtin_mips_wrdsp: i = 1; l = 0; u = 63; break; 437 case Mips::BI__builtin_mips_rddsp: i = 0; l = 0; u = 63; break; 438 case Mips::BI__builtin_mips_append: i = 2; l = 0; u = 31; break; 439 case Mips::BI__builtin_mips_balign: i = 2; l = 0; u = 3; break; 440 case Mips::BI__builtin_mips_precr_sra_ph_w: i = 2; l = 0; u = 31; break; 441 case Mips::BI__builtin_mips_precr_sra_r_ph_w: i = 2; l = 0; u = 31; break; 442 case Mips::BI__builtin_mips_prepend: i = 2; l = 0; u = 31; break; 443 }; 444 445 // We can't check the value of a dependent argument. 446 if (TheCall->getArg(i)->isTypeDependent() || 447 TheCall->getArg(i)->isValueDependent()) 448 return false; 449 450 // Check that the immediate argument is actually a constant. 451 llvm::APSInt Result; 452 if (SemaBuiltinConstantArg(TheCall, i, Result)) 453 return true; 454 455 // Range check against the upper/lower values for this instruction. 456 unsigned Val = Result.getZExtValue(); 457 if (Val < l || Val > u) 458 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 459 << l << u << TheCall->getArg(i)->getSourceRange(); 460 461 return false; 462} 463 464/// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo 465/// parameter with the FormatAttr's correct format_idx and firstDataArg. 466/// Returns true when the format fits the function and the FormatStringInfo has 467/// been populated. 468bool Sema::getFormatStringInfo(const FormatAttr *Format, bool IsCXXMember, 469 FormatStringInfo *FSI) { 470 FSI->HasVAListArg = Format->getFirstArg() == 0; 471 FSI->FormatIdx = Format->getFormatIdx() - 1; 472 FSI->FirstDataArg = FSI->HasVAListArg ? 0 : Format->getFirstArg() - 1; 473 474 // The way the format attribute works in GCC, the implicit this argument 475 // of member functions is counted. However, it doesn't appear in our own 476 // lists, so decrement format_idx in that case. 477 if (IsCXXMember) { 478 if(FSI->FormatIdx == 0) 479 return false; 480 --FSI->FormatIdx; 481 if (FSI->FirstDataArg != 0) 482 --FSI->FirstDataArg; 483 } 484 return true; 485} 486 487/// Handles the checks for format strings, non-POD arguments to vararg 488/// functions, and NULL arguments passed to non-NULL parameters. 489void Sema::checkCall(NamedDecl *FDecl, 490 ArrayRef<const Expr *> Args, 491 unsigned NumProtoArgs, 492 bool IsMemberFunction, 493 SourceLocation Loc, 494 SourceRange Range, 495 VariadicCallType CallType) { 496 if (CurContext->isDependentContext()) 497 return; 498 499 // Printf and scanf checking. 500 bool HandledFormatString = false; 501 for (specific_attr_iterator<FormatAttr> 502 I = FDecl->specific_attr_begin<FormatAttr>(), 503 E = FDecl->specific_attr_end<FormatAttr>(); I != E ; ++I) 504 if (CheckFormatArguments(*I, Args, IsMemberFunction, CallType, Loc, Range)) 505 HandledFormatString = true; 506 507 // Refuse POD arguments that weren't caught by the format string 508 // checks above. 509 if (!HandledFormatString && CallType != VariadicDoesNotApply) 510 for (unsigned ArgIdx = NumProtoArgs; ArgIdx < Args.size(); ++ArgIdx) { 511 // Args[ArgIdx] can be null in malformed code. 512 if (const Expr *Arg = Args[ArgIdx]) 513 variadicArgumentPODCheck(Arg, CallType); 514 } 515 516 for (specific_attr_iterator<NonNullAttr> 517 I = FDecl->specific_attr_begin<NonNullAttr>(), 518 E = FDecl->specific_attr_end<NonNullAttr>(); I != E; ++I) 519 CheckNonNullArguments(*I, Args.data(), Loc); 520 521 // Type safety checking. 522 for (specific_attr_iterator<ArgumentWithTypeTagAttr> 523 i = FDecl->specific_attr_begin<ArgumentWithTypeTagAttr>(), 524 e = FDecl->specific_attr_end<ArgumentWithTypeTagAttr>(); i != e; ++i) { 525 CheckArgumentWithTypeTag(*i, Args.data()); 526 } 527} 528 529/// CheckConstructorCall - Check a constructor call for correctness and safety 530/// properties not enforced by the C type system. 531void Sema::CheckConstructorCall(FunctionDecl *FDecl, 532 ArrayRef<const Expr *> Args, 533 const FunctionProtoType *Proto, 534 SourceLocation Loc) { 535 VariadicCallType CallType = 536 Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply; 537 checkCall(FDecl, Args, Proto->getNumArgs(), 538 /*IsMemberFunction=*/true, Loc, SourceRange(), CallType); 539} 540 541/// CheckFunctionCall - Check a direct function call for various correctness 542/// and safety properties not strictly enforced by the C type system. 543bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall, 544 const FunctionProtoType *Proto) { 545 bool IsMemberOperatorCall = isa<CXXOperatorCallExpr>(TheCall) && 546 isa<CXXMethodDecl>(FDecl); 547 bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall) || 548 IsMemberOperatorCall; 549 VariadicCallType CallType = getVariadicCallType(FDecl, Proto, 550 TheCall->getCallee()); 551 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 552 Expr** Args = TheCall->getArgs(); 553 unsigned NumArgs = TheCall->getNumArgs(); 554 if (IsMemberOperatorCall) { 555 // If this is a call to a member operator, hide the first argument 556 // from checkCall. 557 // FIXME: Our choice of AST representation here is less than ideal. 558 ++Args; 559 --NumArgs; 560 } 561 checkCall(FDecl, llvm::makeArrayRef<const Expr *>(Args, NumArgs), 562 NumProtoArgs, 563 IsMemberFunction, TheCall->getRParenLoc(), 564 TheCall->getCallee()->getSourceRange(), CallType); 565 566 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 567 // None of the checks below are needed for functions that don't have 568 // simple names (e.g., C++ conversion functions). 569 if (!FnInfo) 570 return false; 571 572 unsigned CMId = FDecl->getMemoryFunctionKind(); 573 if (CMId == 0) 574 return false; 575 576 // Handle memory setting and copying functions. 577 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 578 CheckStrlcpycatArguments(TheCall, FnInfo); 579 else if (CMId == Builtin::BIstrncat) 580 CheckStrncatArguments(TheCall, FnInfo); 581 else 582 CheckMemaccessArguments(TheCall, CMId, FnInfo); 583 584 return false; 585} 586 587bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 588 Expr **Args, unsigned NumArgs) { 589 VariadicCallType CallType = 590 Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply; 591 592 checkCall(Method, llvm::makeArrayRef<const Expr *>(Args, NumArgs), 593 Method->param_size(), 594 /*IsMemberFunction=*/false, 595 lbrac, Method->getSourceRange(), CallType); 596 597 return false; 598} 599 600bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall, 601 const FunctionProtoType *Proto) { 602 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 603 if (!V) 604 return false; 605 606 QualType Ty = V->getType(); 607 if (!Ty->isBlockPointerType()) 608 return false; 609 610 VariadicCallType CallType = 611 Proto && Proto->isVariadic() ? VariadicBlock : VariadicDoesNotApply ; 612 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 613 614 checkCall(NDecl, 615 llvm::makeArrayRef<const Expr *>(TheCall->getArgs(), 616 TheCall->getNumArgs()), 617 NumProtoArgs, /*IsMemberFunction=*/false, 618 TheCall->getRParenLoc(), 619 TheCall->getCallee()->getSourceRange(), CallType); 620 621 return false; 622} 623 624ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 625 AtomicExpr::AtomicOp Op) { 626 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 627 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 628 629 // All these operations take one of the following forms: 630 enum { 631 // C __c11_atomic_init(A *, C) 632 Init, 633 // C __c11_atomic_load(A *, int) 634 Load, 635 // void __atomic_load(A *, CP, int) 636 Copy, 637 // C __c11_atomic_add(A *, M, int) 638 Arithmetic, 639 // C __atomic_exchange_n(A *, CP, int) 640 Xchg, 641 // void __atomic_exchange(A *, C *, CP, int) 642 GNUXchg, 643 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 644 C11CmpXchg, 645 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 646 GNUCmpXchg 647 } Form = Init; 648 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 649 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 650 // where: 651 // C is an appropriate type, 652 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 653 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 654 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 655 // the int parameters are for orderings. 656 657 assert(AtomicExpr::AO__c11_atomic_init == 0 && 658 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 659 && "need to update code for modified C11 atomics"); 660 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 661 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 662 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 663 Op == AtomicExpr::AO__atomic_store_n || 664 Op == AtomicExpr::AO__atomic_exchange_n || 665 Op == AtomicExpr::AO__atomic_compare_exchange_n; 666 bool IsAddSub = false; 667 668 switch (Op) { 669 case AtomicExpr::AO__c11_atomic_init: 670 Form = Init; 671 break; 672 673 case AtomicExpr::AO__c11_atomic_load: 674 case AtomicExpr::AO__atomic_load_n: 675 Form = Load; 676 break; 677 678 case AtomicExpr::AO__c11_atomic_store: 679 case AtomicExpr::AO__atomic_load: 680 case AtomicExpr::AO__atomic_store: 681 case AtomicExpr::AO__atomic_store_n: 682 Form = Copy; 683 break; 684 685 case AtomicExpr::AO__c11_atomic_fetch_add: 686 case AtomicExpr::AO__c11_atomic_fetch_sub: 687 case AtomicExpr::AO__atomic_fetch_add: 688 case AtomicExpr::AO__atomic_fetch_sub: 689 case AtomicExpr::AO__atomic_add_fetch: 690 case AtomicExpr::AO__atomic_sub_fetch: 691 IsAddSub = true; 692 // Fall through. 693 case AtomicExpr::AO__c11_atomic_fetch_and: 694 case AtomicExpr::AO__c11_atomic_fetch_or: 695 case AtomicExpr::AO__c11_atomic_fetch_xor: 696 case AtomicExpr::AO__atomic_fetch_and: 697 case AtomicExpr::AO__atomic_fetch_or: 698 case AtomicExpr::AO__atomic_fetch_xor: 699 case AtomicExpr::AO__atomic_fetch_nand: 700 case AtomicExpr::AO__atomic_and_fetch: 701 case AtomicExpr::AO__atomic_or_fetch: 702 case AtomicExpr::AO__atomic_xor_fetch: 703 case AtomicExpr::AO__atomic_nand_fetch: 704 Form = Arithmetic; 705 break; 706 707 case AtomicExpr::AO__c11_atomic_exchange: 708 case AtomicExpr::AO__atomic_exchange_n: 709 Form = Xchg; 710 break; 711 712 case AtomicExpr::AO__atomic_exchange: 713 Form = GNUXchg; 714 break; 715 716 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 717 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 718 Form = C11CmpXchg; 719 break; 720 721 case AtomicExpr::AO__atomic_compare_exchange: 722 case AtomicExpr::AO__atomic_compare_exchange_n: 723 Form = GNUCmpXchg; 724 break; 725 } 726 727 // Check we have the right number of arguments. 728 if (TheCall->getNumArgs() < NumArgs[Form]) { 729 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 730 << 0 << NumArgs[Form] << TheCall->getNumArgs() 731 << TheCall->getCallee()->getSourceRange(); 732 return ExprError(); 733 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 734 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 735 diag::err_typecheck_call_too_many_args) 736 << 0 << NumArgs[Form] << TheCall->getNumArgs() 737 << TheCall->getCallee()->getSourceRange(); 738 return ExprError(); 739 } 740 741 // Inspect the first argument of the atomic operation. 742 Expr *Ptr = TheCall->getArg(0); 743 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 744 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 745 if (!pointerType) { 746 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 747 << Ptr->getType() << Ptr->getSourceRange(); 748 return ExprError(); 749 } 750 751 // For a __c11 builtin, this should be a pointer to an _Atomic type. 752 QualType AtomTy = pointerType->getPointeeType(); // 'A' 753 QualType ValType = AtomTy; // 'C' 754 if (IsC11) { 755 if (!AtomTy->isAtomicType()) { 756 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 757 << Ptr->getType() << Ptr->getSourceRange(); 758 return ExprError(); 759 } 760 if (AtomTy.isConstQualified()) { 761 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_non_const_atomic) 762 << Ptr->getType() << Ptr->getSourceRange(); 763 return ExprError(); 764 } 765 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 766 } 767 768 // For an arithmetic operation, the implied arithmetic must be well-formed. 769 if (Form == Arithmetic) { 770 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 771 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 772 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 773 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 774 return ExprError(); 775 } 776 if (!IsAddSub && !ValType->isIntegerType()) { 777 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 778 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 779 return ExprError(); 780 } 781 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 782 // For __atomic_*_n operations, the value type must be a scalar integral or 783 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 784 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 785 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 786 return ExprError(); 787 } 788 789 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context)) { 790 // For GNU atomics, require a trivially-copyable type. This is not part of 791 // the GNU atomics specification, but we enforce it for sanity. 792 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 793 << Ptr->getType() << Ptr->getSourceRange(); 794 return ExprError(); 795 } 796 797 // FIXME: For any builtin other than a load, the ValType must not be 798 // const-qualified. 799 800 switch (ValType.getObjCLifetime()) { 801 case Qualifiers::OCL_None: 802 case Qualifiers::OCL_ExplicitNone: 803 // okay 804 break; 805 806 case Qualifiers::OCL_Weak: 807 case Qualifiers::OCL_Strong: 808 case Qualifiers::OCL_Autoreleasing: 809 // FIXME: Can this happen? By this point, ValType should be known 810 // to be trivially copyable. 811 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 812 << ValType << Ptr->getSourceRange(); 813 return ExprError(); 814 } 815 816 QualType ResultType = ValType; 817 if (Form == Copy || Form == GNUXchg || Form == Init) 818 ResultType = Context.VoidTy; 819 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 820 ResultType = Context.BoolTy; 821 822 // The type of a parameter passed 'by value'. In the GNU atomics, such 823 // arguments are actually passed as pointers. 824 QualType ByValType = ValType; // 'CP' 825 if (!IsC11 && !IsN) 826 ByValType = Ptr->getType(); 827 828 // The first argument --- the pointer --- has a fixed type; we 829 // deduce the types of the rest of the arguments accordingly. Walk 830 // the remaining arguments, converting them to the deduced value type. 831 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 832 QualType Ty; 833 if (i < NumVals[Form] + 1) { 834 switch (i) { 835 case 1: 836 // The second argument is the non-atomic operand. For arithmetic, this 837 // is always passed by value, and for a compare_exchange it is always 838 // passed by address. For the rest, GNU uses by-address and C11 uses 839 // by-value. 840 assert(Form != Load); 841 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 842 Ty = ValType; 843 else if (Form == Copy || Form == Xchg) 844 Ty = ByValType; 845 else if (Form == Arithmetic) 846 Ty = Context.getPointerDiffType(); 847 else 848 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 849 break; 850 case 2: 851 // The third argument to compare_exchange / GNU exchange is a 852 // (pointer to a) desired value. 853 Ty = ByValType; 854 break; 855 case 3: 856 // The fourth argument to GNU compare_exchange is a 'weak' flag. 857 Ty = Context.BoolTy; 858 break; 859 } 860 } else { 861 // The order(s) are always converted to int. 862 Ty = Context.IntTy; 863 } 864 865 InitializedEntity Entity = 866 InitializedEntity::InitializeParameter(Context, Ty, false); 867 ExprResult Arg = TheCall->getArg(i); 868 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 869 if (Arg.isInvalid()) 870 return true; 871 TheCall->setArg(i, Arg.get()); 872 } 873 874 // Permute the arguments into a 'consistent' order. 875 SmallVector<Expr*, 5> SubExprs; 876 SubExprs.push_back(Ptr); 877 switch (Form) { 878 case Init: 879 // Note, AtomicExpr::getVal1() has a special case for this atomic. 880 SubExprs.push_back(TheCall->getArg(1)); // Val1 881 break; 882 case Load: 883 SubExprs.push_back(TheCall->getArg(1)); // Order 884 break; 885 case Copy: 886 case Arithmetic: 887 case Xchg: 888 SubExprs.push_back(TheCall->getArg(2)); // Order 889 SubExprs.push_back(TheCall->getArg(1)); // Val1 890 break; 891 case GNUXchg: 892 // Note, AtomicExpr::getVal2() has a special case for this atomic. 893 SubExprs.push_back(TheCall->getArg(3)); // Order 894 SubExprs.push_back(TheCall->getArg(1)); // Val1 895 SubExprs.push_back(TheCall->getArg(2)); // Val2 896 break; 897 case C11CmpXchg: 898 SubExprs.push_back(TheCall->getArg(3)); // Order 899 SubExprs.push_back(TheCall->getArg(1)); // Val1 900 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 901 SubExprs.push_back(TheCall->getArg(2)); // Val2 902 break; 903 case GNUCmpXchg: 904 SubExprs.push_back(TheCall->getArg(4)); // Order 905 SubExprs.push_back(TheCall->getArg(1)); // Val1 906 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 907 SubExprs.push_back(TheCall->getArg(2)); // Val2 908 SubExprs.push_back(TheCall->getArg(3)); // Weak 909 break; 910 } 911 912 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 913 SubExprs, ResultType, Op, 914 TheCall->getRParenLoc())); 915} 916 917 918/// checkBuiltinArgument - Given a call to a builtin function, perform 919/// normal type-checking on the given argument, updating the call in 920/// place. This is useful when a builtin function requires custom 921/// type-checking for some of its arguments but not necessarily all of 922/// them. 923/// 924/// Returns true on error. 925static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 926 FunctionDecl *Fn = E->getDirectCallee(); 927 assert(Fn && "builtin call without direct callee!"); 928 929 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 930 InitializedEntity Entity = 931 InitializedEntity::InitializeParameter(S.Context, Param); 932 933 ExprResult Arg = E->getArg(0); 934 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 935 if (Arg.isInvalid()) 936 return true; 937 938 E->setArg(ArgIndex, Arg.take()); 939 return false; 940} 941 942/// SemaBuiltinAtomicOverloaded - We have a call to a function like 943/// __sync_fetch_and_add, which is an overloaded function based on the pointer 944/// type of its first argument. The main ActOnCallExpr routines have already 945/// promoted the types of arguments because all of these calls are prototyped as 946/// void(...). 947/// 948/// This function goes through and does final semantic checking for these 949/// builtins, 950ExprResult 951Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 952 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 953 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 954 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 955 956 // Ensure that we have at least one argument to do type inference from. 957 if (TheCall->getNumArgs() < 1) { 958 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 959 << 0 << 1 << TheCall->getNumArgs() 960 << TheCall->getCallee()->getSourceRange(); 961 return ExprError(); 962 } 963 964 // Inspect the first argument of the atomic builtin. This should always be 965 // a pointer type, whose element is an integral scalar or pointer type. 966 // Because it is a pointer type, we don't have to worry about any implicit 967 // casts here. 968 // FIXME: We don't allow floating point scalars as input. 969 Expr *FirstArg = TheCall->getArg(0); 970 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 971 if (FirstArgResult.isInvalid()) 972 return ExprError(); 973 FirstArg = FirstArgResult.take(); 974 TheCall->setArg(0, FirstArg); 975 976 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 977 if (!pointerType) { 978 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 979 << FirstArg->getType() << FirstArg->getSourceRange(); 980 return ExprError(); 981 } 982 983 QualType ValType = pointerType->getPointeeType(); 984 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 985 !ValType->isBlockPointerType()) { 986 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 987 << FirstArg->getType() << FirstArg->getSourceRange(); 988 return ExprError(); 989 } 990 991 switch (ValType.getObjCLifetime()) { 992 case Qualifiers::OCL_None: 993 case Qualifiers::OCL_ExplicitNone: 994 // okay 995 break; 996 997 case Qualifiers::OCL_Weak: 998 case Qualifiers::OCL_Strong: 999 case Qualifiers::OCL_Autoreleasing: 1000 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1001 << ValType << FirstArg->getSourceRange(); 1002 return ExprError(); 1003 } 1004 1005 // Strip any qualifiers off ValType. 1006 ValType = ValType.getUnqualifiedType(); 1007 1008 // The majority of builtins return a value, but a few have special return 1009 // types, so allow them to override appropriately below. 1010 QualType ResultType = ValType; 1011 1012 // We need to figure out which concrete builtin this maps onto. For example, 1013 // __sync_fetch_and_add with a 2 byte object turns into 1014 // __sync_fetch_and_add_2. 1015#define BUILTIN_ROW(x) \ 1016 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 1017 Builtin::BI##x##_8, Builtin::BI##x##_16 } 1018 1019 static const unsigned BuiltinIndices[][5] = { 1020 BUILTIN_ROW(__sync_fetch_and_add), 1021 BUILTIN_ROW(__sync_fetch_and_sub), 1022 BUILTIN_ROW(__sync_fetch_and_or), 1023 BUILTIN_ROW(__sync_fetch_and_and), 1024 BUILTIN_ROW(__sync_fetch_and_xor), 1025 1026 BUILTIN_ROW(__sync_add_and_fetch), 1027 BUILTIN_ROW(__sync_sub_and_fetch), 1028 BUILTIN_ROW(__sync_and_and_fetch), 1029 BUILTIN_ROW(__sync_or_and_fetch), 1030 BUILTIN_ROW(__sync_xor_and_fetch), 1031 1032 BUILTIN_ROW(__sync_val_compare_and_swap), 1033 BUILTIN_ROW(__sync_bool_compare_and_swap), 1034 BUILTIN_ROW(__sync_lock_test_and_set), 1035 BUILTIN_ROW(__sync_lock_release), 1036 BUILTIN_ROW(__sync_swap) 1037 }; 1038#undef BUILTIN_ROW 1039 1040 // Determine the index of the size. 1041 unsigned SizeIndex; 1042 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 1043 case 1: SizeIndex = 0; break; 1044 case 2: SizeIndex = 1; break; 1045 case 4: SizeIndex = 2; break; 1046 case 8: SizeIndex = 3; break; 1047 case 16: SizeIndex = 4; break; 1048 default: 1049 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 1050 << FirstArg->getType() << FirstArg->getSourceRange(); 1051 return ExprError(); 1052 } 1053 1054 // Each of these builtins has one pointer argument, followed by some number of 1055 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 1056 // that we ignore. Find out which row of BuiltinIndices to read from as well 1057 // as the number of fixed args. 1058 unsigned BuiltinID = FDecl->getBuiltinID(); 1059 unsigned BuiltinIndex, NumFixed = 1; 1060 switch (BuiltinID) { 1061 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 1062 case Builtin::BI__sync_fetch_and_add: 1063 case Builtin::BI__sync_fetch_and_add_1: 1064 case Builtin::BI__sync_fetch_and_add_2: 1065 case Builtin::BI__sync_fetch_and_add_4: 1066 case Builtin::BI__sync_fetch_and_add_8: 1067 case Builtin::BI__sync_fetch_and_add_16: 1068 BuiltinIndex = 0; 1069 break; 1070 1071 case Builtin::BI__sync_fetch_and_sub: 1072 case Builtin::BI__sync_fetch_and_sub_1: 1073 case Builtin::BI__sync_fetch_and_sub_2: 1074 case Builtin::BI__sync_fetch_and_sub_4: 1075 case Builtin::BI__sync_fetch_and_sub_8: 1076 case Builtin::BI__sync_fetch_and_sub_16: 1077 BuiltinIndex = 1; 1078 break; 1079 1080 case Builtin::BI__sync_fetch_and_or: 1081 case Builtin::BI__sync_fetch_and_or_1: 1082 case Builtin::BI__sync_fetch_and_or_2: 1083 case Builtin::BI__sync_fetch_and_or_4: 1084 case Builtin::BI__sync_fetch_and_or_8: 1085 case Builtin::BI__sync_fetch_and_or_16: 1086 BuiltinIndex = 2; 1087 break; 1088 1089 case Builtin::BI__sync_fetch_and_and: 1090 case Builtin::BI__sync_fetch_and_and_1: 1091 case Builtin::BI__sync_fetch_and_and_2: 1092 case Builtin::BI__sync_fetch_and_and_4: 1093 case Builtin::BI__sync_fetch_and_and_8: 1094 case Builtin::BI__sync_fetch_and_and_16: 1095 BuiltinIndex = 3; 1096 break; 1097 1098 case Builtin::BI__sync_fetch_and_xor: 1099 case Builtin::BI__sync_fetch_and_xor_1: 1100 case Builtin::BI__sync_fetch_and_xor_2: 1101 case Builtin::BI__sync_fetch_and_xor_4: 1102 case Builtin::BI__sync_fetch_and_xor_8: 1103 case Builtin::BI__sync_fetch_and_xor_16: 1104 BuiltinIndex = 4; 1105 break; 1106 1107 case Builtin::BI__sync_add_and_fetch: 1108 case Builtin::BI__sync_add_and_fetch_1: 1109 case Builtin::BI__sync_add_and_fetch_2: 1110 case Builtin::BI__sync_add_and_fetch_4: 1111 case Builtin::BI__sync_add_and_fetch_8: 1112 case Builtin::BI__sync_add_and_fetch_16: 1113 BuiltinIndex = 5; 1114 break; 1115 1116 case Builtin::BI__sync_sub_and_fetch: 1117 case Builtin::BI__sync_sub_and_fetch_1: 1118 case Builtin::BI__sync_sub_and_fetch_2: 1119 case Builtin::BI__sync_sub_and_fetch_4: 1120 case Builtin::BI__sync_sub_and_fetch_8: 1121 case Builtin::BI__sync_sub_and_fetch_16: 1122 BuiltinIndex = 6; 1123 break; 1124 1125 case Builtin::BI__sync_and_and_fetch: 1126 case Builtin::BI__sync_and_and_fetch_1: 1127 case Builtin::BI__sync_and_and_fetch_2: 1128 case Builtin::BI__sync_and_and_fetch_4: 1129 case Builtin::BI__sync_and_and_fetch_8: 1130 case Builtin::BI__sync_and_and_fetch_16: 1131 BuiltinIndex = 7; 1132 break; 1133 1134 case Builtin::BI__sync_or_and_fetch: 1135 case Builtin::BI__sync_or_and_fetch_1: 1136 case Builtin::BI__sync_or_and_fetch_2: 1137 case Builtin::BI__sync_or_and_fetch_4: 1138 case Builtin::BI__sync_or_and_fetch_8: 1139 case Builtin::BI__sync_or_and_fetch_16: 1140 BuiltinIndex = 8; 1141 break; 1142 1143 case Builtin::BI__sync_xor_and_fetch: 1144 case Builtin::BI__sync_xor_and_fetch_1: 1145 case Builtin::BI__sync_xor_and_fetch_2: 1146 case Builtin::BI__sync_xor_and_fetch_4: 1147 case Builtin::BI__sync_xor_and_fetch_8: 1148 case Builtin::BI__sync_xor_and_fetch_16: 1149 BuiltinIndex = 9; 1150 break; 1151 1152 case Builtin::BI__sync_val_compare_and_swap: 1153 case Builtin::BI__sync_val_compare_and_swap_1: 1154 case Builtin::BI__sync_val_compare_and_swap_2: 1155 case Builtin::BI__sync_val_compare_and_swap_4: 1156 case Builtin::BI__sync_val_compare_and_swap_8: 1157 case Builtin::BI__sync_val_compare_and_swap_16: 1158 BuiltinIndex = 10; 1159 NumFixed = 2; 1160 break; 1161 1162 case Builtin::BI__sync_bool_compare_and_swap: 1163 case Builtin::BI__sync_bool_compare_and_swap_1: 1164 case Builtin::BI__sync_bool_compare_and_swap_2: 1165 case Builtin::BI__sync_bool_compare_and_swap_4: 1166 case Builtin::BI__sync_bool_compare_and_swap_8: 1167 case Builtin::BI__sync_bool_compare_and_swap_16: 1168 BuiltinIndex = 11; 1169 NumFixed = 2; 1170 ResultType = Context.BoolTy; 1171 break; 1172 1173 case Builtin::BI__sync_lock_test_and_set: 1174 case Builtin::BI__sync_lock_test_and_set_1: 1175 case Builtin::BI__sync_lock_test_and_set_2: 1176 case Builtin::BI__sync_lock_test_and_set_4: 1177 case Builtin::BI__sync_lock_test_and_set_8: 1178 case Builtin::BI__sync_lock_test_and_set_16: 1179 BuiltinIndex = 12; 1180 break; 1181 1182 case Builtin::BI__sync_lock_release: 1183 case Builtin::BI__sync_lock_release_1: 1184 case Builtin::BI__sync_lock_release_2: 1185 case Builtin::BI__sync_lock_release_4: 1186 case Builtin::BI__sync_lock_release_8: 1187 case Builtin::BI__sync_lock_release_16: 1188 BuiltinIndex = 13; 1189 NumFixed = 0; 1190 ResultType = Context.VoidTy; 1191 break; 1192 1193 case Builtin::BI__sync_swap: 1194 case Builtin::BI__sync_swap_1: 1195 case Builtin::BI__sync_swap_2: 1196 case Builtin::BI__sync_swap_4: 1197 case Builtin::BI__sync_swap_8: 1198 case Builtin::BI__sync_swap_16: 1199 BuiltinIndex = 14; 1200 break; 1201 } 1202 1203 // Now that we know how many fixed arguments we expect, first check that we 1204 // have at least that many. 1205 if (TheCall->getNumArgs() < 1+NumFixed) { 1206 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1207 << 0 << 1+NumFixed << TheCall->getNumArgs() 1208 << TheCall->getCallee()->getSourceRange(); 1209 return ExprError(); 1210 } 1211 1212 // Get the decl for the concrete builtin from this, we can tell what the 1213 // concrete integer type we should convert to is. 1214 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1215 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1216 FunctionDecl *NewBuiltinDecl; 1217 if (NewBuiltinID == BuiltinID) 1218 NewBuiltinDecl = FDecl; 1219 else { 1220 // Perform builtin lookup to avoid redeclaring it. 1221 DeclarationName DN(&Context.Idents.get(NewBuiltinName)); 1222 LookupResult Res(*this, DN, DRE->getLocStart(), LookupOrdinaryName); 1223 LookupName(Res, TUScope, /*AllowBuiltinCreation=*/true); 1224 assert(Res.getFoundDecl()); 1225 NewBuiltinDecl = dyn_cast<FunctionDecl>(Res.getFoundDecl()); 1226 if (NewBuiltinDecl == 0) 1227 return ExprError(); 1228 } 1229 1230 // The first argument --- the pointer --- has a fixed type; we 1231 // deduce the types of the rest of the arguments accordingly. Walk 1232 // the remaining arguments, converting them to the deduced value type. 1233 for (unsigned i = 0; i != NumFixed; ++i) { 1234 ExprResult Arg = TheCall->getArg(i+1); 1235 1236 // GCC does an implicit conversion to the pointer or integer ValType. This 1237 // can fail in some cases (1i -> int**), check for this error case now. 1238 // Initialize the argument. 1239 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1240 ValType, /*consume*/ false); 1241 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1242 if (Arg.isInvalid()) 1243 return ExprError(); 1244 1245 // Okay, we have something that *can* be converted to the right type. Check 1246 // to see if there is a potentially weird extension going on here. This can 1247 // happen when you do an atomic operation on something like an char* and 1248 // pass in 42. The 42 gets converted to char. This is even more strange 1249 // for things like 45.123 -> char, etc. 1250 // FIXME: Do this check. 1251 TheCall->setArg(i+1, Arg.take()); 1252 } 1253 1254 ASTContext& Context = this->getASTContext(); 1255 1256 // Create a new DeclRefExpr to refer to the new decl. 1257 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1258 Context, 1259 DRE->getQualifierLoc(), 1260 SourceLocation(), 1261 NewBuiltinDecl, 1262 /*enclosing*/ false, 1263 DRE->getLocation(), 1264 Context.BuiltinFnTy, 1265 DRE->getValueKind()); 1266 1267 // Set the callee in the CallExpr. 1268 // FIXME: This loses syntactic information. 1269 QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType()); 1270 ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy, 1271 CK_BuiltinFnToFnPtr); 1272 TheCall->setCallee(PromotedCall.take()); 1273 1274 // Change the result type of the call to match the original value type. This 1275 // is arbitrary, but the codegen for these builtins ins design to handle it 1276 // gracefully. 1277 TheCall->setType(ResultType); 1278 1279 return TheCallResult; 1280} 1281 1282/// CheckObjCString - Checks that the argument to the builtin 1283/// CFString constructor is correct 1284/// Note: It might also make sense to do the UTF-16 conversion here (would 1285/// simplify the backend). 1286bool Sema::CheckObjCString(Expr *Arg) { 1287 Arg = Arg->IgnoreParenCasts(); 1288 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1289 1290 if (!Literal || !Literal->isAscii()) { 1291 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1292 << Arg->getSourceRange(); 1293 return true; 1294 } 1295 1296 if (Literal->containsNonAsciiOrNull()) { 1297 StringRef String = Literal->getString(); 1298 unsigned NumBytes = String.size(); 1299 SmallVector<UTF16, 128> ToBuf(NumBytes); 1300 const UTF8 *FromPtr = (const UTF8 *)String.data(); 1301 UTF16 *ToPtr = &ToBuf[0]; 1302 1303 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1304 &ToPtr, ToPtr + NumBytes, 1305 strictConversion); 1306 // Check for conversion failure. 1307 if (Result != conversionOK) 1308 Diag(Arg->getLocStart(), 1309 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1310 } 1311 return false; 1312} 1313 1314/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1315/// Emit an error and return true on failure, return false on success. 1316bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1317 Expr *Fn = TheCall->getCallee(); 1318 if (TheCall->getNumArgs() > 2) { 1319 Diag(TheCall->getArg(2)->getLocStart(), 1320 diag::err_typecheck_call_too_many_args) 1321 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1322 << Fn->getSourceRange() 1323 << SourceRange(TheCall->getArg(2)->getLocStart(), 1324 (*(TheCall->arg_end()-1))->getLocEnd()); 1325 return true; 1326 } 1327 1328 if (TheCall->getNumArgs() < 2) { 1329 return Diag(TheCall->getLocEnd(), 1330 diag::err_typecheck_call_too_few_args_at_least) 1331 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1332 } 1333 1334 // Type-check the first argument normally. 1335 if (checkBuiltinArgument(*this, TheCall, 0)) 1336 return true; 1337 1338 // Determine whether the current function is variadic or not. 1339 BlockScopeInfo *CurBlock = getCurBlock(); 1340 bool isVariadic; 1341 if (CurBlock) 1342 isVariadic = CurBlock->TheDecl->isVariadic(); 1343 else if (FunctionDecl *FD = getCurFunctionDecl()) 1344 isVariadic = FD->isVariadic(); 1345 else 1346 isVariadic = getCurMethodDecl()->isVariadic(); 1347 1348 if (!isVariadic) { 1349 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1350 return true; 1351 } 1352 1353 // Verify that the second argument to the builtin is the last argument of the 1354 // current function or method. 1355 bool SecondArgIsLastNamedArgument = false; 1356 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1357 1358 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1359 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1360 // FIXME: This isn't correct for methods (results in bogus warning). 1361 // Get the last formal in the current function. 1362 const ParmVarDecl *LastArg; 1363 if (CurBlock) 1364 LastArg = *(CurBlock->TheDecl->param_end()-1); 1365 else if (FunctionDecl *FD = getCurFunctionDecl()) 1366 LastArg = *(FD->param_end()-1); 1367 else 1368 LastArg = *(getCurMethodDecl()->param_end()-1); 1369 SecondArgIsLastNamedArgument = PV == LastArg; 1370 } 1371 } 1372 1373 if (!SecondArgIsLastNamedArgument) 1374 Diag(TheCall->getArg(1)->getLocStart(), 1375 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1376 return false; 1377} 1378 1379/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1380/// friends. This is declared to take (...), so we have to check everything. 1381bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1382 if (TheCall->getNumArgs() < 2) 1383 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1384 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1385 if (TheCall->getNumArgs() > 2) 1386 return Diag(TheCall->getArg(2)->getLocStart(), 1387 diag::err_typecheck_call_too_many_args) 1388 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1389 << SourceRange(TheCall->getArg(2)->getLocStart(), 1390 (*(TheCall->arg_end()-1))->getLocEnd()); 1391 1392 ExprResult OrigArg0 = TheCall->getArg(0); 1393 ExprResult OrigArg1 = TheCall->getArg(1); 1394 1395 // Do standard promotions between the two arguments, returning their common 1396 // type. 1397 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1398 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1399 return true; 1400 1401 // Make sure any conversions are pushed back into the call; this is 1402 // type safe since unordered compare builtins are declared as "_Bool 1403 // foo(...)". 1404 TheCall->setArg(0, OrigArg0.get()); 1405 TheCall->setArg(1, OrigArg1.get()); 1406 1407 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1408 return false; 1409 1410 // If the common type isn't a real floating type, then the arguments were 1411 // invalid for this operation. 1412 if (Res.isNull() || !Res->isRealFloatingType()) 1413 return Diag(OrigArg0.get()->getLocStart(), 1414 diag::err_typecheck_call_invalid_ordered_compare) 1415 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1416 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1417 1418 return false; 1419} 1420 1421/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1422/// __builtin_isnan and friends. This is declared to take (...), so we have 1423/// to check everything. We expect the last argument to be a floating point 1424/// value. 1425bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1426 if (TheCall->getNumArgs() < NumArgs) 1427 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1428 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1429 if (TheCall->getNumArgs() > NumArgs) 1430 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1431 diag::err_typecheck_call_too_many_args) 1432 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1433 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1434 (*(TheCall->arg_end()-1))->getLocEnd()); 1435 1436 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1437 1438 if (OrigArg->isTypeDependent()) 1439 return false; 1440 1441 // This operation requires a non-_Complex floating-point number. 1442 if (!OrigArg->getType()->isRealFloatingType()) 1443 return Diag(OrigArg->getLocStart(), 1444 diag::err_typecheck_call_invalid_unary_fp) 1445 << OrigArg->getType() << OrigArg->getSourceRange(); 1446 1447 // If this is an implicit conversion from float -> double, remove it. 1448 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1449 Expr *CastArg = Cast->getSubExpr(); 1450 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1451 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1452 "promotion from float to double is the only expected cast here"); 1453 Cast->setSubExpr(0); 1454 TheCall->setArg(NumArgs-1, CastArg); 1455 } 1456 } 1457 1458 return false; 1459} 1460 1461/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1462// This is declared to take (...), so we have to check everything. 1463ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1464 if (TheCall->getNumArgs() < 2) 1465 return ExprError(Diag(TheCall->getLocEnd(), 1466 diag::err_typecheck_call_too_few_args_at_least) 1467 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1468 << TheCall->getSourceRange()); 1469 1470 // Determine which of the following types of shufflevector we're checking: 1471 // 1) unary, vector mask: (lhs, mask) 1472 // 2) binary, vector mask: (lhs, rhs, mask) 1473 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1474 QualType resType = TheCall->getArg(0)->getType(); 1475 unsigned numElements = 0; 1476 1477 if (!TheCall->getArg(0)->isTypeDependent() && 1478 !TheCall->getArg(1)->isTypeDependent()) { 1479 QualType LHSType = TheCall->getArg(0)->getType(); 1480 QualType RHSType = TheCall->getArg(1)->getType(); 1481 1482 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1483 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1484 << SourceRange(TheCall->getArg(0)->getLocStart(), 1485 TheCall->getArg(1)->getLocEnd()); 1486 return ExprError(); 1487 } 1488 1489 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1490 unsigned numResElements = TheCall->getNumArgs() - 2; 1491 1492 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1493 // with mask. If so, verify that RHS is an integer vector type with the 1494 // same number of elts as lhs. 1495 if (TheCall->getNumArgs() == 2) { 1496 if (!RHSType->hasIntegerRepresentation() || 1497 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1498 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1499 << SourceRange(TheCall->getArg(1)->getLocStart(), 1500 TheCall->getArg(1)->getLocEnd()); 1501 numResElements = numElements; 1502 } 1503 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1504 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1505 << SourceRange(TheCall->getArg(0)->getLocStart(), 1506 TheCall->getArg(1)->getLocEnd()); 1507 return ExprError(); 1508 } else if (numElements != numResElements) { 1509 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1510 resType = Context.getVectorType(eltType, numResElements, 1511 VectorType::GenericVector); 1512 } 1513 } 1514 1515 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1516 if (TheCall->getArg(i)->isTypeDependent() || 1517 TheCall->getArg(i)->isValueDependent()) 1518 continue; 1519 1520 llvm::APSInt Result(32); 1521 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1522 return ExprError(Diag(TheCall->getLocStart(), 1523 diag::err_shufflevector_nonconstant_argument) 1524 << TheCall->getArg(i)->getSourceRange()); 1525 1526 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1527 return ExprError(Diag(TheCall->getLocStart(), 1528 diag::err_shufflevector_argument_too_large) 1529 << TheCall->getArg(i)->getSourceRange()); 1530 } 1531 1532 SmallVector<Expr*, 32> exprs; 1533 1534 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1535 exprs.push_back(TheCall->getArg(i)); 1536 TheCall->setArg(i, 0); 1537 } 1538 1539 return Owned(new (Context) ShuffleVectorExpr(Context, exprs, resType, 1540 TheCall->getCallee()->getLocStart(), 1541 TheCall->getRParenLoc())); 1542} 1543 1544/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1545// This is declared to take (const void*, ...) and can take two 1546// optional constant int args. 1547bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1548 unsigned NumArgs = TheCall->getNumArgs(); 1549 1550 if (NumArgs > 3) 1551 return Diag(TheCall->getLocEnd(), 1552 diag::err_typecheck_call_too_many_args_at_most) 1553 << 0 /*function call*/ << 3 << NumArgs 1554 << TheCall->getSourceRange(); 1555 1556 // Argument 0 is checked for us and the remaining arguments must be 1557 // constant integers. 1558 for (unsigned i = 1; i != NumArgs; ++i) { 1559 Expr *Arg = TheCall->getArg(i); 1560 1561 // We can't check the value of a dependent argument. 1562 if (Arg->isTypeDependent() || Arg->isValueDependent()) 1563 continue; 1564 1565 llvm::APSInt Result; 1566 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1567 return true; 1568 1569 // FIXME: gcc issues a warning and rewrites these to 0. These 1570 // seems especially odd for the third argument since the default 1571 // is 3. 1572 if (i == 1) { 1573 if (Result.getLimitedValue() > 1) 1574 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1575 << "0" << "1" << Arg->getSourceRange(); 1576 } else { 1577 if (Result.getLimitedValue() > 3) 1578 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1579 << "0" << "3" << Arg->getSourceRange(); 1580 } 1581 } 1582 1583 return false; 1584} 1585 1586/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1587/// TheCall is a constant expression. 1588bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1589 llvm::APSInt &Result) { 1590 Expr *Arg = TheCall->getArg(ArgNum); 1591 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1592 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1593 1594 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1595 1596 if (!Arg->isIntegerConstantExpr(Result, Context)) 1597 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1598 << FDecl->getDeclName() << Arg->getSourceRange(); 1599 1600 return false; 1601} 1602 1603/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1604/// int type). This simply type checks that type is one of the defined 1605/// constants (0-3). 1606// For compatibility check 0-3, llvm only handles 0 and 2. 1607bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1608 llvm::APSInt Result; 1609 1610 // We can't check the value of a dependent argument. 1611 if (TheCall->getArg(1)->isTypeDependent() || 1612 TheCall->getArg(1)->isValueDependent()) 1613 return false; 1614 1615 // Check constant-ness first. 1616 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1617 return true; 1618 1619 Expr *Arg = TheCall->getArg(1); 1620 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1621 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1622 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1623 } 1624 1625 return false; 1626} 1627 1628/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1629/// This checks that val is a constant 1. 1630bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1631 Expr *Arg = TheCall->getArg(1); 1632 llvm::APSInt Result; 1633 1634 // TODO: This is less than ideal. Overload this to take a value. 1635 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1636 return true; 1637 1638 if (Result != 1) 1639 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1640 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1641 1642 return false; 1643} 1644 1645// Determine if an expression is a string literal or constant string. 1646// If this function returns false on the arguments to a function expecting a 1647// format string, we will usually need to emit a warning. 1648// True string literals are then checked by CheckFormatString. 1649Sema::StringLiteralCheckType 1650Sema::checkFormatStringExpr(const Expr *E, ArrayRef<const Expr *> Args, 1651 bool HasVAListArg, 1652 unsigned format_idx, unsigned firstDataArg, 1653 FormatStringType Type, VariadicCallType CallType, 1654 bool inFunctionCall) { 1655 tryAgain: 1656 if (E->isTypeDependent() || E->isValueDependent()) 1657 return SLCT_NotALiteral; 1658 1659 E = E->IgnoreParenCasts(); 1660 1661 if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull)) 1662 // Technically -Wformat-nonliteral does not warn about this case. 1663 // The behavior of printf and friends in this case is implementation 1664 // dependent. Ideally if the format string cannot be null then 1665 // it should have a 'nonnull' attribute in the function prototype. 1666 return SLCT_CheckedLiteral; 1667 1668 switch (E->getStmtClass()) { 1669 case Stmt::BinaryConditionalOperatorClass: 1670 case Stmt::ConditionalOperatorClass: { 1671 // The expression is a literal if both sub-expressions were, and it was 1672 // completely checked only if both sub-expressions were checked. 1673 const AbstractConditionalOperator *C = 1674 cast<AbstractConditionalOperator>(E); 1675 StringLiteralCheckType Left = 1676 checkFormatStringExpr(C->getTrueExpr(), Args, 1677 HasVAListArg, format_idx, firstDataArg, 1678 Type, CallType, inFunctionCall); 1679 if (Left == SLCT_NotALiteral) 1680 return SLCT_NotALiteral; 1681 StringLiteralCheckType Right = 1682 checkFormatStringExpr(C->getFalseExpr(), Args, 1683 HasVAListArg, format_idx, firstDataArg, 1684 Type, CallType, inFunctionCall); 1685 return Left < Right ? Left : Right; 1686 } 1687 1688 case Stmt::ImplicitCastExprClass: { 1689 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1690 goto tryAgain; 1691 } 1692 1693 case Stmt::OpaqueValueExprClass: 1694 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1695 E = src; 1696 goto tryAgain; 1697 } 1698 return SLCT_NotALiteral; 1699 1700 case Stmt::PredefinedExprClass: 1701 // While __func__, etc., are technically not string literals, they 1702 // cannot contain format specifiers and thus are not a security 1703 // liability. 1704 return SLCT_UncheckedLiteral; 1705 1706 case Stmt::DeclRefExprClass: { 1707 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1708 1709 // As an exception, do not flag errors for variables binding to 1710 // const string literals. 1711 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1712 bool isConstant = false; 1713 QualType T = DR->getType(); 1714 1715 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1716 isConstant = AT->getElementType().isConstant(Context); 1717 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1718 isConstant = T.isConstant(Context) && 1719 PT->getPointeeType().isConstant(Context); 1720 } else if (T->isObjCObjectPointerType()) { 1721 // In ObjC, there is usually no "const ObjectPointer" type, 1722 // so don't check if the pointee type is constant. 1723 isConstant = T.isConstant(Context); 1724 } 1725 1726 if (isConstant) { 1727 if (const Expr *Init = VD->getAnyInitializer()) { 1728 // Look through initializers like const char c[] = { "foo" } 1729 if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) { 1730 if (InitList->isStringLiteralInit()) 1731 Init = InitList->getInit(0)->IgnoreParenImpCasts(); 1732 } 1733 return checkFormatStringExpr(Init, Args, 1734 HasVAListArg, format_idx, 1735 firstDataArg, Type, CallType, 1736 /*inFunctionCall*/false); 1737 } 1738 } 1739 1740 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1741 // special check to see if the format string is a function parameter 1742 // of the function calling the printf function. If the function 1743 // has an attribute indicating it is a printf-like function, then we 1744 // should suppress warnings concerning non-literals being used in a call 1745 // to a vprintf function. For example: 1746 // 1747 // void 1748 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1749 // va_list ap; 1750 // va_start(ap, fmt); 1751 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1752 // ... 1753 // 1754 if (HasVAListArg) { 1755 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 1756 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 1757 int PVIndex = PV->getFunctionScopeIndex() + 1; 1758 for (specific_attr_iterator<FormatAttr> 1759 i = ND->specific_attr_begin<FormatAttr>(), 1760 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 1761 FormatAttr *PVFormat = *i; 1762 // adjust for implicit parameter 1763 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1764 if (MD->isInstance()) 1765 ++PVIndex; 1766 // We also check if the formats are compatible. 1767 // We can't pass a 'scanf' string to a 'printf' function. 1768 if (PVIndex == PVFormat->getFormatIdx() && 1769 Type == GetFormatStringType(PVFormat)) 1770 return SLCT_UncheckedLiteral; 1771 } 1772 } 1773 } 1774 } 1775 } 1776 1777 return SLCT_NotALiteral; 1778 } 1779 1780 case Stmt::CallExprClass: 1781 case Stmt::CXXMemberCallExprClass: { 1782 const CallExpr *CE = cast<CallExpr>(E); 1783 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 1784 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 1785 unsigned ArgIndex = FA->getFormatIdx(); 1786 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1787 if (MD->isInstance()) 1788 --ArgIndex; 1789 const Expr *Arg = CE->getArg(ArgIndex - 1); 1790 1791 return checkFormatStringExpr(Arg, Args, 1792 HasVAListArg, format_idx, firstDataArg, 1793 Type, CallType, inFunctionCall); 1794 } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) { 1795 unsigned BuiltinID = FD->getBuiltinID(); 1796 if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString || 1797 BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) { 1798 const Expr *Arg = CE->getArg(0); 1799 return checkFormatStringExpr(Arg, Args, 1800 HasVAListArg, format_idx, 1801 firstDataArg, Type, CallType, 1802 inFunctionCall); 1803 } 1804 } 1805 } 1806 1807 return SLCT_NotALiteral; 1808 } 1809 case Stmt::ObjCStringLiteralClass: 1810 case Stmt::StringLiteralClass: { 1811 const StringLiteral *StrE = NULL; 1812 1813 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1814 StrE = ObjCFExpr->getString(); 1815 else 1816 StrE = cast<StringLiteral>(E); 1817 1818 if (StrE) { 1819 CheckFormatString(StrE, E, Args, HasVAListArg, format_idx, 1820 firstDataArg, Type, inFunctionCall, CallType); 1821 return SLCT_CheckedLiteral; 1822 } 1823 1824 return SLCT_NotALiteral; 1825 } 1826 1827 default: 1828 return SLCT_NotALiteral; 1829 } 1830} 1831 1832void 1833Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1834 const Expr * const *ExprArgs, 1835 SourceLocation CallSiteLoc) { 1836 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1837 e = NonNull->args_end(); 1838 i != e; ++i) { 1839 const Expr *ArgExpr = ExprArgs[*i]; 1840 1841 // As a special case, transparent unions initialized with zero are 1842 // considered null for the purposes of the nonnull attribute. 1843 if (const RecordType *UT = ArgExpr->getType()->getAsUnionType()) { 1844 if (UT->getDecl()->hasAttr<TransparentUnionAttr>()) 1845 if (const CompoundLiteralExpr *CLE = 1846 dyn_cast<CompoundLiteralExpr>(ArgExpr)) 1847 if (const InitListExpr *ILE = 1848 dyn_cast<InitListExpr>(CLE->getInitializer())) 1849 ArgExpr = ILE->getInit(0); 1850 } 1851 1852 bool Result; 1853 if (ArgExpr->EvaluateAsBooleanCondition(Result, Context) && !Result) 1854 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1855 } 1856} 1857 1858Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 1859 return llvm::StringSwitch<FormatStringType>(Format->getType()) 1860 .Case("scanf", FST_Scanf) 1861 .Cases("printf", "printf0", FST_Printf) 1862 .Cases("NSString", "CFString", FST_NSString) 1863 .Case("strftime", FST_Strftime) 1864 .Case("strfmon", FST_Strfmon) 1865 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 1866 .Default(FST_Unknown); 1867} 1868 1869/// CheckFormatArguments - Check calls to printf and scanf (and similar 1870/// functions) for correct use of format strings. 1871/// Returns true if a format string has been fully checked. 1872bool Sema::CheckFormatArguments(const FormatAttr *Format, 1873 ArrayRef<const Expr *> Args, 1874 bool IsCXXMember, 1875 VariadicCallType CallType, 1876 SourceLocation Loc, SourceRange Range) { 1877 FormatStringInfo FSI; 1878 if (getFormatStringInfo(Format, IsCXXMember, &FSI)) 1879 return CheckFormatArguments(Args, FSI.HasVAListArg, FSI.FormatIdx, 1880 FSI.FirstDataArg, GetFormatStringType(Format), 1881 CallType, Loc, Range); 1882 return false; 1883} 1884 1885bool Sema::CheckFormatArguments(ArrayRef<const Expr *> Args, 1886 bool HasVAListArg, unsigned format_idx, 1887 unsigned firstDataArg, FormatStringType Type, 1888 VariadicCallType CallType, 1889 SourceLocation Loc, SourceRange Range) { 1890 // CHECK: printf/scanf-like function is called with no format string. 1891 if (format_idx >= Args.size()) { 1892 Diag(Loc, diag::warn_missing_format_string) << Range; 1893 return false; 1894 } 1895 1896 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1897 1898 // CHECK: format string is not a string literal. 1899 // 1900 // Dynamically generated format strings are difficult to 1901 // automatically vet at compile time. Requiring that format strings 1902 // are string literals: (1) permits the checking of format strings by 1903 // the compiler and thereby (2) can practically remove the source of 1904 // many format string exploits. 1905 1906 // Format string can be either ObjC string (e.g. @"%d") or 1907 // C string (e.g. "%d") 1908 // ObjC string uses the same format specifiers as C string, so we can use 1909 // the same format string checking logic for both ObjC and C strings. 1910 StringLiteralCheckType CT = 1911 checkFormatStringExpr(OrigFormatExpr, Args, HasVAListArg, 1912 format_idx, firstDataArg, Type, CallType); 1913 if (CT != SLCT_NotALiteral) 1914 // Literal format string found, check done! 1915 return CT == SLCT_CheckedLiteral; 1916 1917 // Strftime is particular as it always uses a single 'time' argument, 1918 // so it is safe to pass a non-literal string. 1919 if (Type == FST_Strftime) 1920 return false; 1921 1922 // Do not emit diag when the string param is a macro expansion and the 1923 // format is either NSString or CFString. This is a hack to prevent 1924 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 1925 // which are usually used in place of NS and CF string literals. 1926 if (Type == FST_NSString && 1927 SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart())) 1928 return false; 1929 1930 // If there are no arguments specified, warn with -Wformat-security, otherwise 1931 // warn only with -Wformat-nonliteral. 1932 if (Args.size() == format_idx+1) 1933 Diag(Args[format_idx]->getLocStart(), 1934 diag::warn_format_nonliteral_noargs) 1935 << OrigFormatExpr->getSourceRange(); 1936 else 1937 Diag(Args[format_idx]->getLocStart(), 1938 diag::warn_format_nonliteral) 1939 << OrigFormatExpr->getSourceRange(); 1940 return false; 1941} 1942 1943namespace { 1944class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1945protected: 1946 Sema &S; 1947 const StringLiteral *FExpr; 1948 const Expr *OrigFormatExpr; 1949 const unsigned FirstDataArg; 1950 const unsigned NumDataArgs; 1951 const char *Beg; // Start of format string. 1952 const bool HasVAListArg; 1953 ArrayRef<const Expr *> Args; 1954 unsigned FormatIdx; 1955 llvm::BitVector CoveredArgs; 1956 bool usesPositionalArgs; 1957 bool atFirstArg; 1958 bool inFunctionCall; 1959 Sema::VariadicCallType CallType; 1960public: 1961 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1962 const Expr *origFormatExpr, unsigned firstDataArg, 1963 unsigned numDataArgs, const char *beg, bool hasVAListArg, 1964 ArrayRef<const Expr *> Args, 1965 unsigned formatIdx, bool inFunctionCall, 1966 Sema::VariadicCallType callType) 1967 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1968 FirstDataArg(firstDataArg), NumDataArgs(numDataArgs), 1969 Beg(beg), HasVAListArg(hasVAListArg), 1970 Args(Args), FormatIdx(formatIdx), 1971 usesPositionalArgs(false), atFirstArg(true), 1972 inFunctionCall(inFunctionCall), CallType(callType) { 1973 CoveredArgs.resize(numDataArgs); 1974 CoveredArgs.reset(); 1975 } 1976 1977 void DoneProcessing(); 1978 1979 void HandleIncompleteSpecifier(const char *startSpecifier, 1980 unsigned specifierLen); 1981 1982 void HandleInvalidLengthModifier( 1983 const analyze_format_string::FormatSpecifier &FS, 1984 const analyze_format_string::ConversionSpecifier &CS, 1985 const char *startSpecifier, unsigned specifierLen, unsigned DiagID); 1986 1987 void HandleNonStandardLengthModifier( 1988 const analyze_format_string::FormatSpecifier &FS, 1989 const char *startSpecifier, unsigned specifierLen); 1990 1991 void HandleNonStandardConversionSpecifier( 1992 const analyze_format_string::ConversionSpecifier &CS, 1993 const char *startSpecifier, unsigned specifierLen); 1994 1995 virtual void HandlePosition(const char *startPos, unsigned posLen); 1996 1997 virtual void HandleInvalidPosition(const char *startSpecifier, 1998 unsigned specifierLen, 1999 analyze_format_string::PositionContext p); 2000 2001 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 2002 2003 void HandleNullChar(const char *nullCharacter); 2004 2005 template <typename Range> 2006 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 2007 const Expr *ArgumentExpr, 2008 PartialDiagnostic PDiag, 2009 SourceLocation StringLoc, 2010 bool IsStringLocation, Range StringRange, 2011 ArrayRef<FixItHint> Fixit = ArrayRef<FixItHint>()); 2012 2013protected: 2014 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 2015 const char *startSpec, 2016 unsigned specifierLen, 2017 const char *csStart, unsigned csLen); 2018 2019 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 2020 const char *startSpec, 2021 unsigned specifierLen); 2022 2023 SourceRange getFormatStringRange(); 2024 CharSourceRange getSpecifierRange(const char *startSpecifier, 2025 unsigned specifierLen); 2026 SourceLocation getLocationOfByte(const char *x); 2027 2028 const Expr *getDataArg(unsigned i) const; 2029 2030 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 2031 const analyze_format_string::ConversionSpecifier &CS, 2032 const char *startSpecifier, unsigned specifierLen, 2033 unsigned argIndex); 2034 2035 template <typename Range> 2036 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 2037 bool IsStringLocation, Range StringRange, 2038 ArrayRef<FixItHint> Fixit = ArrayRef<FixItHint>()); 2039 2040 void CheckPositionalAndNonpositionalArgs( 2041 const analyze_format_string::FormatSpecifier *FS); 2042}; 2043} 2044 2045SourceRange CheckFormatHandler::getFormatStringRange() { 2046 return OrigFormatExpr->getSourceRange(); 2047} 2048 2049CharSourceRange CheckFormatHandler:: 2050getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 2051 SourceLocation Start = getLocationOfByte(startSpecifier); 2052 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 2053 2054 // Advance the end SourceLocation by one due to half-open ranges. 2055 End = End.getLocWithOffset(1); 2056 2057 return CharSourceRange::getCharRange(Start, End); 2058} 2059 2060SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 2061 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 2062} 2063 2064void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 2065 unsigned specifierLen){ 2066 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 2067 getLocationOfByte(startSpecifier), 2068 /*IsStringLocation*/true, 2069 getSpecifierRange(startSpecifier, specifierLen)); 2070} 2071 2072void CheckFormatHandler::HandleInvalidLengthModifier( 2073 const analyze_format_string::FormatSpecifier &FS, 2074 const analyze_format_string::ConversionSpecifier &CS, 2075 const char *startSpecifier, unsigned specifierLen, unsigned DiagID) { 2076 using namespace analyze_format_string; 2077 2078 const LengthModifier &LM = FS.getLengthModifier(); 2079 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2080 2081 // See if we know how to fix this length modifier. 2082 llvm::Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2083 if (FixedLM) { 2084 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2085 getLocationOfByte(LM.getStart()), 2086 /*IsStringLocation*/true, 2087 getSpecifierRange(startSpecifier, specifierLen)); 2088 2089 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2090 << FixedLM->toString() 2091 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2092 2093 } else { 2094 FixItHint Hint; 2095 if (DiagID == diag::warn_format_nonsensical_length) 2096 Hint = FixItHint::CreateRemoval(LMRange); 2097 2098 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2099 getLocationOfByte(LM.getStart()), 2100 /*IsStringLocation*/true, 2101 getSpecifierRange(startSpecifier, specifierLen), 2102 Hint); 2103 } 2104} 2105 2106void CheckFormatHandler::HandleNonStandardLengthModifier( 2107 const analyze_format_string::FormatSpecifier &FS, 2108 const char *startSpecifier, unsigned specifierLen) { 2109 using namespace analyze_format_string; 2110 2111 const LengthModifier &LM = FS.getLengthModifier(); 2112 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2113 2114 // See if we know how to fix this length modifier. 2115 llvm::Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2116 if (FixedLM) { 2117 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2118 << LM.toString() << 0, 2119 getLocationOfByte(LM.getStart()), 2120 /*IsStringLocation*/true, 2121 getSpecifierRange(startSpecifier, specifierLen)); 2122 2123 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2124 << FixedLM->toString() 2125 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2126 2127 } else { 2128 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2129 << LM.toString() << 0, 2130 getLocationOfByte(LM.getStart()), 2131 /*IsStringLocation*/true, 2132 getSpecifierRange(startSpecifier, specifierLen)); 2133 } 2134} 2135 2136void CheckFormatHandler::HandleNonStandardConversionSpecifier( 2137 const analyze_format_string::ConversionSpecifier &CS, 2138 const char *startSpecifier, unsigned specifierLen) { 2139 using namespace analyze_format_string; 2140 2141 // See if we know how to fix this conversion specifier. 2142 llvm::Optional<ConversionSpecifier> FixedCS = CS.getStandardSpecifier(); 2143 if (FixedCS) { 2144 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2145 << CS.toString() << /*conversion specifier*/1, 2146 getLocationOfByte(CS.getStart()), 2147 /*IsStringLocation*/true, 2148 getSpecifierRange(startSpecifier, specifierLen)); 2149 2150 CharSourceRange CSRange = getSpecifierRange(CS.getStart(), CS.getLength()); 2151 S.Diag(getLocationOfByte(CS.getStart()), diag::note_format_fix_specifier) 2152 << FixedCS->toString() 2153 << FixItHint::CreateReplacement(CSRange, FixedCS->toString()); 2154 } else { 2155 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2156 << CS.toString() << /*conversion specifier*/1, 2157 getLocationOfByte(CS.getStart()), 2158 /*IsStringLocation*/true, 2159 getSpecifierRange(startSpecifier, specifierLen)); 2160 } 2161} 2162 2163void CheckFormatHandler::HandlePosition(const char *startPos, 2164 unsigned posLen) { 2165 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 2166 getLocationOfByte(startPos), 2167 /*IsStringLocation*/true, 2168 getSpecifierRange(startPos, posLen)); 2169} 2170 2171void 2172CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 2173 analyze_format_string::PositionContext p) { 2174 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 2175 << (unsigned) p, 2176 getLocationOfByte(startPos), /*IsStringLocation*/true, 2177 getSpecifierRange(startPos, posLen)); 2178} 2179 2180void CheckFormatHandler::HandleZeroPosition(const char *startPos, 2181 unsigned posLen) { 2182 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 2183 getLocationOfByte(startPos), 2184 /*IsStringLocation*/true, 2185 getSpecifierRange(startPos, posLen)); 2186} 2187 2188void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 2189 if (!isa<ObjCStringLiteral>(OrigFormatExpr)) { 2190 // The presence of a null character is likely an error. 2191 EmitFormatDiagnostic( 2192 S.PDiag(diag::warn_printf_format_string_contains_null_char), 2193 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 2194 getFormatStringRange()); 2195 } 2196} 2197 2198// Note that this may return NULL if there was an error parsing or building 2199// one of the argument expressions. 2200const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 2201 return Args[FirstDataArg + i]; 2202} 2203 2204void CheckFormatHandler::DoneProcessing() { 2205 // Does the number of data arguments exceed the number of 2206 // format conversions in the format string? 2207 if (!HasVAListArg) { 2208 // Find any arguments that weren't covered. 2209 CoveredArgs.flip(); 2210 signed notCoveredArg = CoveredArgs.find_first(); 2211 if (notCoveredArg >= 0) { 2212 assert((unsigned)notCoveredArg < NumDataArgs); 2213 if (const Expr *E = getDataArg((unsigned) notCoveredArg)) { 2214 SourceLocation Loc = E->getLocStart(); 2215 if (!S.getSourceManager().isInSystemMacro(Loc)) { 2216 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 2217 Loc, /*IsStringLocation*/false, 2218 getFormatStringRange()); 2219 } 2220 } 2221 } 2222 } 2223} 2224 2225bool 2226CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 2227 SourceLocation Loc, 2228 const char *startSpec, 2229 unsigned specifierLen, 2230 const char *csStart, 2231 unsigned csLen) { 2232 2233 bool keepGoing = true; 2234 if (argIndex < NumDataArgs) { 2235 // Consider the argument coverered, even though the specifier doesn't 2236 // make sense. 2237 CoveredArgs.set(argIndex); 2238 } 2239 else { 2240 // If argIndex exceeds the number of data arguments we 2241 // don't issue a warning because that is just a cascade of warnings (and 2242 // they may have intended '%%' anyway). We don't want to continue processing 2243 // the format string after this point, however, as we will like just get 2244 // gibberish when trying to match arguments. 2245 keepGoing = false; 2246 } 2247 2248 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2249 << StringRef(csStart, csLen), 2250 Loc, /*IsStringLocation*/true, 2251 getSpecifierRange(startSpec, specifierLen)); 2252 2253 return keepGoing; 2254} 2255 2256void 2257CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2258 const char *startSpec, 2259 unsigned specifierLen) { 2260 EmitFormatDiagnostic( 2261 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2262 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2263} 2264 2265bool 2266CheckFormatHandler::CheckNumArgs( 2267 const analyze_format_string::FormatSpecifier &FS, 2268 const analyze_format_string::ConversionSpecifier &CS, 2269 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2270 2271 if (argIndex >= NumDataArgs) { 2272 PartialDiagnostic PDiag = FS.usesPositionalArg() 2273 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2274 << (argIndex+1) << NumDataArgs) 2275 : S.PDiag(diag::warn_printf_insufficient_data_args); 2276 EmitFormatDiagnostic( 2277 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2278 getSpecifierRange(startSpecifier, specifierLen)); 2279 return false; 2280 } 2281 return true; 2282} 2283 2284template<typename Range> 2285void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2286 SourceLocation Loc, 2287 bool IsStringLocation, 2288 Range StringRange, 2289 ArrayRef<FixItHint> FixIt) { 2290 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2291 Loc, IsStringLocation, StringRange, FixIt); 2292} 2293 2294/// \brief If the format string is not within the funcion call, emit a note 2295/// so that the function call and string are in diagnostic messages. 2296/// 2297/// \param InFunctionCall if true, the format string is within the function 2298/// call and only one diagnostic message will be produced. Otherwise, an 2299/// extra note will be emitted pointing to location of the format string. 2300/// 2301/// \param ArgumentExpr the expression that is passed as the format string 2302/// argument in the function call. Used for getting locations when two 2303/// diagnostics are emitted. 2304/// 2305/// \param PDiag the callee should already have provided any strings for the 2306/// diagnostic message. This function only adds locations and fixits 2307/// to diagnostics. 2308/// 2309/// \param Loc primary location for diagnostic. If two diagnostics are 2310/// required, one will be at Loc and a new SourceLocation will be created for 2311/// the other one. 2312/// 2313/// \param IsStringLocation if true, Loc points to the format string should be 2314/// used for the note. Otherwise, Loc points to the argument list and will 2315/// be used with PDiag. 2316/// 2317/// \param StringRange some or all of the string to highlight. This is 2318/// templated so it can accept either a CharSourceRange or a SourceRange. 2319/// 2320/// \param FixIt optional fix it hint for the format string. 2321template<typename Range> 2322void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2323 const Expr *ArgumentExpr, 2324 PartialDiagnostic PDiag, 2325 SourceLocation Loc, 2326 bool IsStringLocation, 2327 Range StringRange, 2328 ArrayRef<FixItHint> FixIt) { 2329 if (InFunctionCall) { 2330 const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag); 2331 D << StringRange; 2332 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2333 I != E; ++I) { 2334 D << *I; 2335 } 2336 } else { 2337 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2338 << ArgumentExpr->getSourceRange(); 2339 2340 const Sema::SemaDiagnosticBuilder &Note = 2341 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2342 diag::note_format_string_defined); 2343 2344 Note << StringRange; 2345 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2346 I != E; ++I) { 2347 Note << *I; 2348 } 2349 } 2350} 2351 2352//===--- CHECK: Printf format string checking ------------------------------===// 2353 2354namespace { 2355class CheckPrintfHandler : public CheckFormatHandler { 2356 bool ObjCContext; 2357public: 2358 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2359 const Expr *origFormatExpr, unsigned firstDataArg, 2360 unsigned numDataArgs, bool isObjC, 2361 const char *beg, bool hasVAListArg, 2362 ArrayRef<const Expr *> Args, 2363 unsigned formatIdx, bool inFunctionCall, 2364 Sema::VariadicCallType CallType) 2365 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2366 numDataArgs, beg, hasVAListArg, Args, 2367 formatIdx, inFunctionCall, CallType), ObjCContext(isObjC) 2368 {} 2369 2370 2371 bool HandleInvalidPrintfConversionSpecifier( 2372 const analyze_printf::PrintfSpecifier &FS, 2373 const char *startSpecifier, 2374 unsigned specifierLen); 2375 2376 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2377 const char *startSpecifier, 2378 unsigned specifierLen); 2379 bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2380 const char *StartSpecifier, 2381 unsigned SpecifierLen, 2382 const Expr *E); 2383 2384 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2385 const char *startSpecifier, unsigned specifierLen); 2386 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2387 const analyze_printf::OptionalAmount &Amt, 2388 unsigned type, 2389 const char *startSpecifier, unsigned specifierLen); 2390 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2391 const analyze_printf::OptionalFlag &flag, 2392 const char *startSpecifier, unsigned specifierLen); 2393 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2394 const analyze_printf::OptionalFlag &ignoredFlag, 2395 const analyze_printf::OptionalFlag &flag, 2396 const char *startSpecifier, unsigned specifierLen); 2397 bool checkForCStrMembers(const analyze_printf::ArgType &AT, 2398 const Expr *E, const CharSourceRange &CSR); 2399 2400}; 2401} 2402 2403bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2404 const analyze_printf::PrintfSpecifier &FS, 2405 const char *startSpecifier, 2406 unsigned specifierLen) { 2407 const analyze_printf::PrintfConversionSpecifier &CS = 2408 FS.getConversionSpecifier(); 2409 2410 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2411 getLocationOfByte(CS.getStart()), 2412 startSpecifier, specifierLen, 2413 CS.getStart(), CS.getLength()); 2414} 2415 2416bool CheckPrintfHandler::HandleAmount( 2417 const analyze_format_string::OptionalAmount &Amt, 2418 unsigned k, const char *startSpecifier, 2419 unsigned specifierLen) { 2420 2421 if (Amt.hasDataArgument()) { 2422 if (!HasVAListArg) { 2423 unsigned argIndex = Amt.getArgIndex(); 2424 if (argIndex >= NumDataArgs) { 2425 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2426 << k, 2427 getLocationOfByte(Amt.getStart()), 2428 /*IsStringLocation*/true, 2429 getSpecifierRange(startSpecifier, specifierLen)); 2430 // Don't do any more checking. We will just emit 2431 // spurious errors. 2432 return false; 2433 } 2434 2435 // Type check the data argument. It should be an 'int'. 2436 // Although not in conformance with C99, we also allow the argument to be 2437 // an 'unsigned int' as that is a reasonably safe case. GCC also 2438 // doesn't emit a warning for that case. 2439 CoveredArgs.set(argIndex); 2440 const Expr *Arg = getDataArg(argIndex); 2441 if (!Arg) 2442 return false; 2443 2444 QualType T = Arg->getType(); 2445 2446 const analyze_printf::ArgType &AT = Amt.getArgType(S.Context); 2447 assert(AT.isValid()); 2448 2449 if (!AT.matchesType(S.Context, T)) { 2450 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2451 << k << AT.getRepresentativeTypeName(S.Context) 2452 << T << Arg->getSourceRange(), 2453 getLocationOfByte(Amt.getStart()), 2454 /*IsStringLocation*/true, 2455 getSpecifierRange(startSpecifier, specifierLen)); 2456 // Don't do any more checking. We will just emit 2457 // spurious errors. 2458 return false; 2459 } 2460 } 2461 } 2462 return true; 2463} 2464 2465void CheckPrintfHandler::HandleInvalidAmount( 2466 const analyze_printf::PrintfSpecifier &FS, 2467 const analyze_printf::OptionalAmount &Amt, 2468 unsigned type, 2469 const char *startSpecifier, 2470 unsigned specifierLen) { 2471 const analyze_printf::PrintfConversionSpecifier &CS = 2472 FS.getConversionSpecifier(); 2473 2474 FixItHint fixit = 2475 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2476 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2477 Amt.getConstantLength())) 2478 : FixItHint(); 2479 2480 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2481 << type << CS.toString(), 2482 getLocationOfByte(Amt.getStart()), 2483 /*IsStringLocation*/true, 2484 getSpecifierRange(startSpecifier, specifierLen), 2485 fixit); 2486} 2487 2488void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2489 const analyze_printf::OptionalFlag &flag, 2490 const char *startSpecifier, 2491 unsigned specifierLen) { 2492 // Warn about pointless flag with a fixit removal. 2493 const analyze_printf::PrintfConversionSpecifier &CS = 2494 FS.getConversionSpecifier(); 2495 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2496 << flag.toString() << CS.toString(), 2497 getLocationOfByte(flag.getPosition()), 2498 /*IsStringLocation*/true, 2499 getSpecifierRange(startSpecifier, specifierLen), 2500 FixItHint::CreateRemoval( 2501 getSpecifierRange(flag.getPosition(), 1))); 2502} 2503 2504void CheckPrintfHandler::HandleIgnoredFlag( 2505 const analyze_printf::PrintfSpecifier &FS, 2506 const analyze_printf::OptionalFlag &ignoredFlag, 2507 const analyze_printf::OptionalFlag &flag, 2508 const char *startSpecifier, 2509 unsigned specifierLen) { 2510 // Warn about ignored flag with a fixit removal. 2511 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2512 << ignoredFlag.toString() << flag.toString(), 2513 getLocationOfByte(ignoredFlag.getPosition()), 2514 /*IsStringLocation*/true, 2515 getSpecifierRange(startSpecifier, specifierLen), 2516 FixItHint::CreateRemoval( 2517 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2518} 2519 2520// Determines if the specified is a C++ class or struct containing 2521// a member with the specified name and kind (e.g. a CXXMethodDecl named 2522// "c_str()"). 2523template<typename MemberKind> 2524static llvm::SmallPtrSet<MemberKind*, 1> 2525CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) { 2526 const RecordType *RT = Ty->getAs<RecordType>(); 2527 llvm::SmallPtrSet<MemberKind*, 1> Results; 2528 2529 if (!RT) 2530 return Results; 2531 const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl()); 2532 if (!RD) 2533 return Results; 2534 2535 LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(), 2536 Sema::LookupMemberName); 2537 2538 // We just need to include all members of the right kind turned up by the 2539 // filter, at this point. 2540 if (S.LookupQualifiedName(R, RT->getDecl())) 2541 for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) { 2542 NamedDecl *decl = (*I)->getUnderlyingDecl(); 2543 if (MemberKind *FK = dyn_cast<MemberKind>(decl)) 2544 Results.insert(FK); 2545 } 2546 return Results; 2547} 2548 2549// Check if a (w)string was passed when a (w)char* was needed, and offer a 2550// better diagnostic if so. AT is assumed to be valid. 2551// Returns true when a c_str() conversion method is found. 2552bool CheckPrintfHandler::checkForCStrMembers( 2553 const analyze_printf::ArgType &AT, const Expr *E, 2554 const CharSourceRange &CSR) { 2555 typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet; 2556 2557 MethodSet Results = 2558 CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType()); 2559 2560 for (MethodSet::iterator MI = Results.begin(), ME = Results.end(); 2561 MI != ME; ++MI) { 2562 const CXXMethodDecl *Method = *MI; 2563 if (Method->getNumParams() == 0 && 2564 AT.matchesType(S.Context, Method->getResultType())) { 2565 // FIXME: Suggest parens if the expression needs them. 2566 SourceLocation EndLoc = 2567 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()); 2568 S.Diag(E->getLocStart(), diag::note_printf_c_str) 2569 << "c_str()" 2570 << FixItHint::CreateInsertion(EndLoc, ".c_str()"); 2571 return true; 2572 } 2573 } 2574 2575 return false; 2576} 2577 2578bool 2579CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2580 &FS, 2581 const char *startSpecifier, 2582 unsigned specifierLen) { 2583 2584 using namespace analyze_format_string; 2585 using namespace analyze_printf; 2586 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2587 2588 if (FS.consumesDataArgument()) { 2589 if (atFirstArg) { 2590 atFirstArg = false; 2591 usesPositionalArgs = FS.usesPositionalArg(); 2592 } 2593 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2594 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2595 startSpecifier, specifierLen); 2596 return false; 2597 } 2598 } 2599 2600 // First check if the field width, precision, and conversion specifier 2601 // have matching data arguments. 2602 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2603 startSpecifier, specifierLen)) { 2604 return false; 2605 } 2606 2607 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2608 startSpecifier, specifierLen)) { 2609 return false; 2610 } 2611 2612 if (!CS.consumesDataArgument()) { 2613 // FIXME: Technically specifying a precision or field width here 2614 // makes no sense. Worth issuing a warning at some point. 2615 return true; 2616 } 2617 2618 // Consume the argument. 2619 unsigned argIndex = FS.getArgIndex(); 2620 if (argIndex < NumDataArgs) { 2621 // The check to see if the argIndex is valid will come later. 2622 // We set the bit here because we may exit early from this 2623 // function if we encounter some other error. 2624 CoveredArgs.set(argIndex); 2625 } 2626 2627 // Check for using an Objective-C specific conversion specifier 2628 // in a non-ObjC literal. 2629 if (!ObjCContext && CS.isObjCArg()) { 2630 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2631 specifierLen); 2632 } 2633 2634 // Check for invalid use of field width 2635 if (!FS.hasValidFieldWidth()) { 2636 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2637 startSpecifier, specifierLen); 2638 } 2639 2640 // Check for invalid use of precision 2641 if (!FS.hasValidPrecision()) { 2642 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2643 startSpecifier, specifierLen); 2644 } 2645 2646 // Check each flag does not conflict with any other component. 2647 if (!FS.hasValidThousandsGroupingPrefix()) 2648 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2649 if (!FS.hasValidLeadingZeros()) 2650 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2651 if (!FS.hasValidPlusPrefix()) 2652 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2653 if (!FS.hasValidSpacePrefix()) 2654 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2655 if (!FS.hasValidAlternativeForm()) 2656 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2657 if (!FS.hasValidLeftJustified()) 2658 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2659 2660 // Check that flags are not ignored by another flag 2661 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2662 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2663 startSpecifier, specifierLen); 2664 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2665 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2666 startSpecifier, specifierLen); 2667 2668 // Check the length modifier is valid with the given conversion specifier. 2669 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 2670 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2671 diag::warn_format_nonsensical_length); 2672 else if (!FS.hasStandardLengthModifier()) 2673 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 2674 else if (!FS.hasStandardLengthConversionCombination()) 2675 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2676 diag::warn_format_non_standard_conversion_spec); 2677 2678 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2679 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2680 2681 // The remaining checks depend on the data arguments. 2682 if (HasVAListArg) 2683 return true; 2684 2685 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2686 return false; 2687 2688 const Expr *Arg = getDataArg(argIndex); 2689 if (!Arg) 2690 return true; 2691 2692 return checkFormatExpr(FS, startSpecifier, specifierLen, Arg); 2693} 2694 2695static bool requiresParensToAddCast(const Expr *E) { 2696 // FIXME: We should have a general way to reason about operator 2697 // precedence and whether parens are actually needed here. 2698 // Take care of a few common cases where they aren't. 2699 const Expr *Inside = E->IgnoreImpCasts(); 2700 if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside)) 2701 Inside = POE->getSyntacticForm()->IgnoreImpCasts(); 2702 2703 switch (Inside->getStmtClass()) { 2704 case Stmt::ArraySubscriptExprClass: 2705 case Stmt::CallExprClass: 2706 case Stmt::CharacterLiteralClass: 2707 case Stmt::CXXBoolLiteralExprClass: 2708 case Stmt::DeclRefExprClass: 2709 case Stmt::FloatingLiteralClass: 2710 case Stmt::IntegerLiteralClass: 2711 case Stmt::MemberExprClass: 2712 case Stmt::ObjCArrayLiteralClass: 2713 case Stmt::ObjCBoolLiteralExprClass: 2714 case Stmt::ObjCBoxedExprClass: 2715 case Stmt::ObjCDictionaryLiteralClass: 2716 case Stmt::ObjCEncodeExprClass: 2717 case Stmt::ObjCIvarRefExprClass: 2718 case Stmt::ObjCMessageExprClass: 2719 case Stmt::ObjCPropertyRefExprClass: 2720 case Stmt::ObjCStringLiteralClass: 2721 case Stmt::ObjCSubscriptRefExprClass: 2722 case Stmt::ParenExprClass: 2723 case Stmt::StringLiteralClass: 2724 case Stmt::UnaryOperatorClass: 2725 return false; 2726 default: 2727 return true; 2728 } 2729} 2730 2731bool 2732CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2733 const char *StartSpecifier, 2734 unsigned SpecifierLen, 2735 const Expr *E) { 2736 using namespace analyze_format_string; 2737 using namespace analyze_printf; 2738 // Now type check the data expression that matches the 2739 // format specifier. 2740 const analyze_printf::ArgType &AT = FS.getArgType(S.Context, 2741 ObjCContext); 2742 if (!AT.isValid()) 2743 return true; 2744 2745 QualType ExprTy = E->getType(); 2746 if (AT.matchesType(S.Context, ExprTy)) 2747 return true; 2748 2749 // Look through argument promotions for our error message's reported type. 2750 // This includes the integral and floating promotions, but excludes array 2751 // and function pointer decay; seeing that an argument intended to be a 2752 // string has type 'char [6]' is probably more confusing than 'char *'. 2753 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2754 if (ICE->getCastKind() == CK_IntegralCast || 2755 ICE->getCastKind() == CK_FloatingCast) { 2756 E = ICE->getSubExpr(); 2757 ExprTy = E->getType(); 2758 2759 // Check if we didn't match because of an implicit cast from a 'char' 2760 // or 'short' to an 'int'. This is done because printf is a varargs 2761 // function. 2762 if (ICE->getType() == S.Context.IntTy || 2763 ICE->getType() == S.Context.UnsignedIntTy) { 2764 // All further checking is done on the subexpression. 2765 if (AT.matchesType(S.Context, ExprTy)) 2766 return true; 2767 } 2768 } 2769 } else if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E)) { 2770 // Special case for 'a', which has type 'int' in C. 2771 // Note, however, that we do /not/ want to treat multibyte constants like 2772 // 'MooV' as characters! This form is deprecated but still exists. 2773 if (ExprTy == S.Context.IntTy) 2774 if (llvm::isUIntN(S.Context.getCharWidth(), CL->getValue())) 2775 ExprTy = S.Context.CharTy; 2776 } 2777 2778 // %C in an Objective-C context prints a unichar, not a wchar_t. 2779 // If the argument is an integer of some kind, believe the %C and suggest 2780 // a cast instead of changing the conversion specifier. 2781 QualType IntendedTy = ExprTy; 2782 if (ObjCContext && 2783 FS.getConversionSpecifier().getKind() == ConversionSpecifier::CArg) { 2784 if (ExprTy->isIntegralOrUnscopedEnumerationType() && 2785 !ExprTy->isCharType()) { 2786 // 'unichar' is defined as a typedef of unsigned short, but we should 2787 // prefer using the typedef if it is visible. 2788 IntendedTy = S.Context.UnsignedShortTy; 2789 2790 LookupResult Result(S, &S.Context.Idents.get("unichar"), E->getLocStart(), 2791 Sema::LookupOrdinaryName); 2792 if (S.LookupName(Result, S.getCurScope())) { 2793 NamedDecl *ND = Result.getFoundDecl(); 2794 if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(ND)) 2795 if (TD->getUnderlyingType() == IntendedTy) 2796 IntendedTy = S.Context.getTypedefType(TD); 2797 } 2798 } 2799 } 2800 2801 // Special-case some of Darwin's platform-independence types by suggesting 2802 // casts to primitive types that are known to be large enough. 2803 bool ShouldNotPrintDirectly = false; 2804 if (S.Context.getTargetInfo().getTriple().isOSDarwin()) { 2805 if (const TypedefType *UserTy = IntendedTy->getAs<TypedefType>()) { 2806 StringRef Name = UserTy->getDecl()->getName(); 2807 QualType CastTy = llvm::StringSwitch<QualType>(Name) 2808 .Case("NSInteger", S.Context.LongTy) 2809 .Case("NSUInteger", S.Context.UnsignedLongTy) 2810 .Case("SInt32", S.Context.IntTy) 2811 .Case("UInt32", S.Context.UnsignedIntTy) 2812 .Default(QualType()); 2813 2814 if (!CastTy.isNull()) { 2815 ShouldNotPrintDirectly = true; 2816 IntendedTy = CastTy; 2817 } 2818 } 2819 } 2820 2821 // We may be able to offer a FixItHint if it is a supported type. 2822 PrintfSpecifier fixedFS = FS; 2823 bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(), 2824 S.Context, ObjCContext); 2825 2826 if (success) { 2827 // Get the fix string from the fixed format specifier 2828 SmallString<16> buf; 2829 llvm::raw_svector_ostream os(buf); 2830 fixedFS.toString(os); 2831 2832 CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen); 2833 2834 if (IntendedTy == ExprTy) { 2835 // In this case, the specifier is wrong and should be changed to match 2836 // the argument. 2837 EmitFormatDiagnostic( 2838 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2839 << AT.getRepresentativeTypeName(S.Context) << IntendedTy 2840 << E->getSourceRange(), 2841 E->getLocStart(), 2842 /*IsStringLocation*/false, 2843 SpecRange, 2844 FixItHint::CreateReplacement(SpecRange, os.str())); 2845 2846 } else { 2847 // The canonical type for formatting this value is different from the 2848 // actual type of the expression. (This occurs, for example, with Darwin's 2849 // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but 2850 // should be printed as 'long' for 64-bit compatibility.) 2851 // Rather than emitting a normal format/argument mismatch, we want to 2852 // add a cast to the recommended type (and correct the format string 2853 // if necessary). 2854 SmallString<16> CastBuf; 2855 llvm::raw_svector_ostream CastFix(CastBuf); 2856 CastFix << "("; 2857 IntendedTy.print(CastFix, S.Context.getPrintingPolicy()); 2858 CastFix << ")"; 2859 2860 SmallVector<FixItHint,4> Hints; 2861 if (!AT.matchesType(S.Context, IntendedTy)) 2862 Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str())); 2863 2864 if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) { 2865 // If there's already a cast present, just replace it. 2866 SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc()); 2867 Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str())); 2868 2869 } else if (!requiresParensToAddCast(E)) { 2870 // If the expression has high enough precedence, 2871 // just write the C-style cast. 2872 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 2873 CastFix.str())); 2874 } else { 2875 // Otherwise, add parens around the expression as well as the cast. 2876 CastFix << "("; 2877 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 2878 CastFix.str())); 2879 2880 SourceLocation After = S.PP.getLocForEndOfToken(E->getLocEnd()); 2881 Hints.push_back(FixItHint::CreateInsertion(After, ")")); 2882 } 2883 2884 if (ShouldNotPrintDirectly) { 2885 // The expression has a type that should not be printed directly. 2886 // We extract the name from the typedef because we don't want to show 2887 // the underlying type in the diagnostic. 2888 StringRef Name = cast<TypedefType>(ExprTy)->getDecl()->getName(); 2889 2890 EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast) 2891 << Name << IntendedTy 2892 << E->getSourceRange(), 2893 E->getLocStart(), /*IsStringLocation=*/false, 2894 SpecRange, Hints); 2895 } else { 2896 // In this case, the expression could be printed using a different 2897 // specifier, but we've decided that the specifier is probably correct 2898 // and we should cast instead. Just use the normal warning message. 2899 EmitFormatDiagnostic( 2900 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2901 << AT.getRepresentativeTypeName(S.Context) << ExprTy 2902 << E->getSourceRange(), 2903 E->getLocStart(), /*IsStringLocation*/false, 2904 SpecRange, Hints); 2905 } 2906 } 2907 } else { 2908 const CharSourceRange &CSR = getSpecifierRange(StartSpecifier, 2909 SpecifierLen); 2910 // Since the warning for passing non-POD types to variadic functions 2911 // was deferred until now, we emit a warning for non-POD 2912 // arguments here. 2913 if (S.isValidVarArgType(ExprTy) == Sema::VAK_Invalid) { 2914 unsigned DiagKind; 2915 if (ExprTy->isObjCObjectType()) 2916 DiagKind = diag::err_cannot_pass_objc_interface_to_vararg_format; 2917 else 2918 DiagKind = diag::warn_non_pod_vararg_with_format_string; 2919 2920 EmitFormatDiagnostic( 2921 S.PDiag(DiagKind) 2922 << S.getLangOpts().CPlusPlus11 2923 << ExprTy 2924 << CallType 2925 << AT.getRepresentativeTypeName(S.Context) 2926 << CSR 2927 << E->getSourceRange(), 2928 E->getLocStart(), /*IsStringLocation*/false, CSR); 2929 2930 checkForCStrMembers(AT, E, CSR); 2931 } else 2932 EmitFormatDiagnostic( 2933 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2934 << AT.getRepresentativeTypeName(S.Context) << ExprTy 2935 << CSR 2936 << E->getSourceRange(), 2937 E->getLocStart(), /*IsStringLocation*/false, CSR); 2938 } 2939 2940 return true; 2941} 2942 2943//===--- CHECK: Scanf format string checking ------------------------------===// 2944 2945namespace { 2946class CheckScanfHandler : public CheckFormatHandler { 2947public: 2948 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2949 const Expr *origFormatExpr, unsigned firstDataArg, 2950 unsigned numDataArgs, const char *beg, bool hasVAListArg, 2951 ArrayRef<const Expr *> Args, 2952 unsigned formatIdx, bool inFunctionCall, 2953 Sema::VariadicCallType CallType) 2954 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2955 numDataArgs, beg, hasVAListArg, 2956 Args, formatIdx, inFunctionCall, CallType) 2957 {} 2958 2959 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2960 const char *startSpecifier, 2961 unsigned specifierLen); 2962 2963 bool HandleInvalidScanfConversionSpecifier( 2964 const analyze_scanf::ScanfSpecifier &FS, 2965 const char *startSpecifier, 2966 unsigned specifierLen); 2967 2968 void HandleIncompleteScanList(const char *start, const char *end); 2969}; 2970} 2971 2972void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2973 const char *end) { 2974 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2975 getLocationOfByte(end), /*IsStringLocation*/true, 2976 getSpecifierRange(start, end - start)); 2977} 2978 2979bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2980 const analyze_scanf::ScanfSpecifier &FS, 2981 const char *startSpecifier, 2982 unsigned specifierLen) { 2983 2984 const analyze_scanf::ScanfConversionSpecifier &CS = 2985 FS.getConversionSpecifier(); 2986 2987 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2988 getLocationOfByte(CS.getStart()), 2989 startSpecifier, specifierLen, 2990 CS.getStart(), CS.getLength()); 2991} 2992 2993bool CheckScanfHandler::HandleScanfSpecifier( 2994 const analyze_scanf::ScanfSpecifier &FS, 2995 const char *startSpecifier, 2996 unsigned specifierLen) { 2997 2998 using namespace analyze_scanf; 2999 using namespace analyze_format_string; 3000 3001 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 3002 3003 // Handle case where '%' and '*' don't consume an argument. These shouldn't 3004 // be used to decide if we are using positional arguments consistently. 3005 if (FS.consumesDataArgument()) { 3006 if (atFirstArg) { 3007 atFirstArg = false; 3008 usesPositionalArgs = FS.usesPositionalArg(); 3009 } 3010 else if (usesPositionalArgs != FS.usesPositionalArg()) { 3011 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 3012 startSpecifier, specifierLen); 3013 return false; 3014 } 3015 } 3016 3017 // Check if the field with is non-zero. 3018 const OptionalAmount &Amt = FS.getFieldWidth(); 3019 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 3020 if (Amt.getConstantAmount() == 0) { 3021 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 3022 Amt.getConstantLength()); 3023 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 3024 getLocationOfByte(Amt.getStart()), 3025 /*IsStringLocation*/true, R, 3026 FixItHint::CreateRemoval(R)); 3027 } 3028 } 3029 3030 if (!FS.consumesDataArgument()) { 3031 // FIXME: Technically specifying a precision or field width here 3032 // makes no sense. Worth issuing a warning at some point. 3033 return true; 3034 } 3035 3036 // Consume the argument. 3037 unsigned argIndex = FS.getArgIndex(); 3038 if (argIndex < NumDataArgs) { 3039 // The check to see if the argIndex is valid will come later. 3040 // We set the bit here because we may exit early from this 3041 // function if we encounter some other error. 3042 CoveredArgs.set(argIndex); 3043 } 3044 3045 // Check the length modifier is valid with the given conversion specifier. 3046 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3047 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3048 diag::warn_format_nonsensical_length); 3049 else if (!FS.hasStandardLengthModifier()) 3050 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3051 else if (!FS.hasStandardLengthConversionCombination()) 3052 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3053 diag::warn_format_non_standard_conversion_spec); 3054 3055 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3056 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3057 3058 // The remaining checks depend on the data arguments. 3059 if (HasVAListArg) 3060 return true; 3061 3062 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3063 return false; 3064 3065 // Check that the argument type matches the format specifier. 3066 const Expr *Ex = getDataArg(argIndex); 3067 if (!Ex) 3068 return true; 3069 3070 const analyze_format_string::ArgType &AT = FS.getArgType(S.Context); 3071 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) { 3072 ScanfSpecifier fixedFS = FS; 3073 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 3074 S.Context); 3075 3076 if (success) { 3077 // Get the fix string from the fixed format specifier. 3078 SmallString<128> buf; 3079 llvm::raw_svector_ostream os(buf); 3080 fixedFS.toString(os); 3081 3082 EmitFormatDiagnostic( 3083 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3084 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3085 << Ex->getSourceRange(), 3086 Ex->getLocStart(), 3087 /*IsStringLocation*/false, 3088 getSpecifierRange(startSpecifier, specifierLen), 3089 FixItHint::CreateReplacement( 3090 getSpecifierRange(startSpecifier, specifierLen), 3091 os.str())); 3092 } else { 3093 EmitFormatDiagnostic( 3094 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3095 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3096 << Ex->getSourceRange(), 3097 Ex->getLocStart(), 3098 /*IsStringLocation*/false, 3099 getSpecifierRange(startSpecifier, specifierLen)); 3100 } 3101 } 3102 3103 return true; 3104} 3105 3106void Sema::CheckFormatString(const StringLiteral *FExpr, 3107 const Expr *OrigFormatExpr, 3108 ArrayRef<const Expr *> Args, 3109 bool HasVAListArg, unsigned format_idx, 3110 unsigned firstDataArg, FormatStringType Type, 3111 bool inFunctionCall, VariadicCallType CallType) { 3112 3113 // CHECK: is the format string a wide literal? 3114 if (!FExpr->isAscii() && !FExpr->isUTF8()) { 3115 CheckFormatHandler::EmitFormatDiagnostic( 3116 *this, inFunctionCall, Args[format_idx], 3117 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 3118 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3119 return; 3120 } 3121 3122 // Str - The format string. NOTE: this is NOT null-terminated! 3123 StringRef StrRef = FExpr->getString(); 3124 const char *Str = StrRef.data(); 3125 unsigned StrLen = StrRef.size(); 3126 const unsigned numDataArgs = Args.size() - firstDataArg; 3127 3128 // CHECK: empty format string? 3129 if (StrLen == 0 && numDataArgs > 0) { 3130 CheckFormatHandler::EmitFormatDiagnostic( 3131 *this, inFunctionCall, Args[format_idx], 3132 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 3133 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3134 return; 3135 } 3136 3137 if (Type == FST_Printf || Type == FST_NSString) { 3138 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 3139 numDataArgs, (Type == FST_NSString), 3140 Str, HasVAListArg, Args, format_idx, 3141 inFunctionCall, CallType); 3142 3143 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 3144 getLangOpts(), 3145 Context.getTargetInfo())) 3146 H.DoneProcessing(); 3147 } else if (Type == FST_Scanf) { 3148 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs, 3149 Str, HasVAListArg, Args, format_idx, 3150 inFunctionCall, CallType); 3151 3152 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 3153 getLangOpts(), 3154 Context.getTargetInfo())) 3155 H.DoneProcessing(); 3156 } // TODO: handle other formats 3157} 3158 3159//===--- CHECK: Standard memory functions ---------------------------------===// 3160 3161/// \brief Determine whether the given type is a dynamic class type (e.g., 3162/// whether it has a vtable). 3163static bool isDynamicClassType(QualType T) { 3164 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 3165 if (CXXRecordDecl *Definition = Record->getDefinition()) 3166 if (Definition->isDynamicClass()) 3167 return true; 3168 3169 return false; 3170} 3171 3172/// \brief If E is a sizeof expression, returns its argument expression, 3173/// otherwise returns NULL. 3174static const Expr *getSizeOfExprArg(const Expr* E) { 3175 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3176 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3177 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 3178 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 3179 3180 return 0; 3181} 3182 3183/// \brief If E is a sizeof expression, returns its argument type. 3184static QualType getSizeOfArgType(const Expr* E) { 3185 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3186 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3187 if (SizeOf->getKind() == clang::UETT_SizeOf) 3188 return SizeOf->getTypeOfArgument(); 3189 3190 return QualType(); 3191} 3192 3193/// \brief Check for dangerous or invalid arguments to memset(). 3194/// 3195/// This issues warnings on known problematic, dangerous or unspecified 3196/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 3197/// function calls. 3198/// 3199/// \param Call The call expression to diagnose. 3200void Sema::CheckMemaccessArguments(const CallExpr *Call, 3201 unsigned BId, 3202 IdentifierInfo *FnName) { 3203 assert(BId != 0); 3204 3205 // It is possible to have a non-standard definition of memset. Validate 3206 // we have enough arguments, and if not, abort further checking. 3207 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 3208 if (Call->getNumArgs() < ExpectedNumArgs) 3209 return; 3210 3211 unsigned LastArg = (BId == Builtin::BImemset || 3212 BId == Builtin::BIstrndup ? 1 : 2); 3213 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 3214 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 3215 3216 // We have special checking when the length is a sizeof expression. 3217 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 3218 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 3219 llvm::FoldingSetNodeID SizeOfArgID; 3220 3221 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 3222 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 3223 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 3224 3225 QualType DestTy = Dest->getType(); 3226 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 3227 QualType PointeeTy = DestPtrTy->getPointeeType(); 3228 3229 // Never warn about void type pointers. This can be used to suppress 3230 // false positives. 3231 if (PointeeTy->isVoidType()) 3232 continue; 3233 3234 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 3235 // actually comparing the expressions for equality. Because computing the 3236 // expression IDs can be expensive, we only do this if the diagnostic is 3237 // enabled. 3238 if (SizeOfArg && 3239 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 3240 SizeOfArg->getExprLoc())) { 3241 // We only compute IDs for expressions if the warning is enabled, and 3242 // cache the sizeof arg's ID. 3243 if (SizeOfArgID == llvm::FoldingSetNodeID()) 3244 SizeOfArg->Profile(SizeOfArgID, Context, true); 3245 llvm::FoldingSetNodeID DestID; 3246 Dest->Profile(DestID, Context, true); 3247 if (DestID == SizeOfArgID) { 3248 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 3249 // over sizeof(src) as well. 3250 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 3251 StringRef ReadableName = FnName->getName(); 3252 3253 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 3254 if (UnaryOp->getOpcode() == UO_AddrOf) 3255 ActionIdx = 1; // If its an address-of operator, just remove it. 3256 if (!PointeeTy->isIncompleteType() && 3257 (Context.getTypeSize(PointeeTy) == Context.getCharWidth())) 3258 ActionIdx = 2; // If the pointee's size is sizeof(char), 3259 // suggest an explicit length. 3260 3261 // If the function is defined as a builtin macro, do not show macro 3262 // expansion. 3263 SourceLocation SL = SizeOfArg->getExprLoc(); 3264 SourceRange DSR = Dest->getSourceRange(); 3265 SourceRange SSR = SizeOfArg->getSourceRange(); 3266 SourceManager &SM = PP.getSourceManager(); 3267 3268 if (SM.isMacroArgExpansion(SL)) { 3269 ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts); 3270 SL = SM.getSpellingLoc(SL); 3271 DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()), 3272 SM.getSpellingLoc(DSR.getEnd())); 3273 SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()), 3274 SM.getSpellingLoc(SSR.getEnd())); 3275 } 3276 3277 DiagRuntimeBehavior(SL, SizeOfArg, 3278 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 3279 << ReadableName 3280 << PointeeTy 3281 << DestTy 3282 << DSR 3283 << SSR); 3284 DiagRuntimeBehavior(SL, SizeOfArg, 3285 PDiag(diag::warn_sizeof_pointer_expr_memaccess_note) 3286 << ActionIdx 3287 << SSR); 3288 3289 break; 3290 } 3291 } 3292 3293 // Also check for cases where the sizeof argument is the exact same 3294 // type as the memory argument, and where it points to a user-defined 3295 // record type. 3296 if (SizeOfArgTy != QualType()) { 3297 if (PointeeTy->isRecordType() && 3298 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 3299 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 3300 PDiag(diag::warn_sizeof_pointer_type_memaccess) 3301 << FnName << SizeOfArgTy << ArgIdx 3302 << PointeeTy << Dest->getSourceRange() 3303 << LenExpr->getSourceRange()); 3304 break; 3305 } 3306 } 3307 3308 // Always complain about dynamic classes. 3309 if (isDynamicClassType(PointeeTy)) { 3310 3311 unsigned OperationType = 0; 3312 // "overwritten" if we're warning about the destination for any call 3313 // but memcmp; otherwise a verb appropriate to the call. 3314 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 3315 if (BId == Builtin::BImemcpy) 3316 OperationType = 1; 3317 else if(BId == Builtin::BImemmove) 3318 OperationType = 2; 3319 else if (BId == Builtin::BImemcmp) 3320 OperationType = 3; 3321 } 3322 3323 DiagRuntimeBehavior( 3324 Dest->getExprLoc(), Dest, 3325 PDiag(diag::warn_dyn_class_memaccess) 3326 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 3327 << FnName << PointeeTy 3328 << OperationType 3329 << Call->getCallee()->getSourceRange()); 3330 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 3331 BId != Builtin::BImemset) 3332 DiagRuntimeBehavior( 3333 Dest->getExprLoc(), Dest, 3334 PDiag(diag::warn_arc_object_memaccess) 3335 << ArgIdx << FnName << PointeeTy 3336 << Call->getCallee()->getSourceRange()); 3337 else 3338 continue; 3339 3340 DiagRuntimeBehavior( 3341 Dest->getExprLoc(), Dest, 3342 PDiag(diag::note_bad_memaccess_silence) 3343 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 3344 break; 3345 } 3346 } 3347} 3348 3349// A little helper routine: ignore addition and subtraction of integer literals. 3350// This intentionally does not ignore all integer constant expressions because 3351// we don't want to remove sizeof(). 3352static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 3353 Ex = Ex->IgnoreParenCasts(); 3354 3355 for (;;) { 3356 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 3357 if (!BO || !BO->isAdditiveOp()) 3358 break; 3359 3360 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 3361 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 3362 3363 if (isa<IntegerLiteral>(RHS)) 3364 Ex = LHS; 3365 else if (isa<IntegerLiteral>(LHS)) 3366 Ex = RHS; 3367 else 3368 break; 3369 } 3370 3371 return Ex; 3372} 3373 3374static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty, 3375 ASTContext &Context) { 3376 // Only handle constant-sized or VLAs, but not flexible members. 3377 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) { 3378 // Only issue the FIXIT for arrays of size > 1. 3379 if (CAT->getSize().getSExtValue() <= 1) 3380 return false; 3381 } else if (!Ty->isVariableArrayType()) { 3382 return false; 3383 } 3384 return true; 3385} 3386 3387// Warn if the user has made the 'size' argument to strlcpy or strlcat 3388// be the size of the source, instead of the destination. 3389void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 3390 IdentifierInfo *FnName) { 3391 3392 // Don't crash if the user has the wrong number of arguments 3393 if (Call->getNumArgs() != 3) 3394 return; 3395 3396 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 3397 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 3398 const Expr *CompareWithSrc = NULL; 3399 3400 // Look for 'strlcpy(dst, x, sizeof(x))' 3401 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 3402 CompareWithSrc = Ex; 3403 else { 3404 // Look for 'strlcpy(dst, x, strlen(x))' 3405 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 3406 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 3407 && SizeCall->getNumArgs() == 1) 3408 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 3409 } 3410 } 3411 3412 if (!CompareWithSrc) 3413 return; 3414 3415 // Determine if the argument to sizeof/strlen is equal to the source 3416 // argument. In principle there's all kinds of things you could do 3417 // here, for instance creating an == expression and evaluating it with 3418 // EvaluateAsBooleanCondition, but this uses a more direct technique: 3419 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 3420 if (!SrcArgDRE) 3421 return; 3422 3423 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 3424 if (!CompareWithSrcDRE || 3425 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 3426 return; 3427 3428 const Expr *OriginalSizeArg = Call->getArg(2); 3429 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 3430 << OriginalSizeArg->getSourceRange() << FnName; 3431 3432 // Output a FIXIT hint if the destination is an array (rather than a 3433 // pointer to an array). This could be enhanced to handle some 3434 // pointers if we know the actual size, like if DstArg is 'array+2' 3435 // we could say 'sizeof(array)-2'. 3436 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 3437 if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context)) 3438 return; 3439 3440 SmallString<128> sizeString; 3441 llvm::raw_svector_ostream OS(sizeString); 3442 OS << "sizeof("; 3443 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3444 OS << ")"; 3445 3446 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 3447 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 3448 OS.str()); 3449} 3450 3451/// Check if two expressions refer to the same declaration. 3452static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 3453 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 3454 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 3455 return D1->getDecl() == D2->getDecl(); 3456 return false; 3457} 3458 3459static const Expr *getStrlenExprArg(const Expr *E) { 3460 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 3461 const FunctionDecl *FD = CE->getDirectCallee(); 3462 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 3463 return 0; 3464 return CE->getArg(0)->IgnoreParenCasts(); 3465 } 3466 return 0; 3467} 3468 3469// Warn on anti-patterns as the 'size' argument to strncat. 3470// The correct size argument should look like following: 3471// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 3472void Sema::CheckStrncatArguments(const CallExpr *CE, 3473 IdentifierInfo *FnName) { 3474 // Don't crash if the user has the wrong number of arguments. 3475 if (CE->getNumArgs() < 3) 3476 return; 3477 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 3478 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 3479 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 3480 3481 // Identify common expressions, which are wrongly used as the size argument 3482 // to strncat and may lead to buffer overflows. 3483 unsigned PatternType = 0; 3484 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 3485 // - sizeof(dst) 3486 if (referToTheSameDecl(SizeOfArg, DstArg)) 3487 PatternType = 1; 3488 // - sizeof(src) 3489 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 3490 PatternType = 2; 3491 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 3492 if (BE->getOpcode() == BO_Sub) { 3493 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 3494 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 3495 // - sizeof(dst) - strlen(dst) 3496 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 3497 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 3498 PatternType = 1; 3499 // - sizeof(src) - (anything) 3500 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 3501 PatternType = 2; 3502 } 3503 } 3504 3505 if (PatternType == 0) 3506 return; 3507 3508 // Generate the diagnostic. 3509 SourceLocation SL = LenArg->getLocStart(); 3510 SourceRange SR = LenArg->getSourceRange(); 3511 SourceManager &SM = PP.getSourceManager(); 3512 3513 // If the function is defined as a builtin macro, do not show macro expansion. 3514 if (SM.isMacroArgExpansion(SL)) { 3515 SL = SM.getSpellingLoc(SL); 3516 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 3517 SM.getSpellingLoc(SR.getEnd())); 3518 } 3519 3520 // Check if the destination is an array (rather than a pointer to an array). 3521 QualType DstTy = DstArg->getType(); 3522 bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy, 3523 Context); 3524 if (!isKnownSizeArray) { 3525 if (PatternType == 1) 3526 Diag(SL, diag::warn_strncat_wrong_size) << SR; 3527 else 3528 Diag(SL, diag::warn_strncat_src_size) << SR; 3529 return; 3530 } 3531 3532 if (PatternType == 1) 3533 Diag(SL, diag::warn_strncat_large_size) << SR; 3534 else 3535 Diag(SL, diag::warn_strncat_src_size) << SR; 3536 3537 SmallString<128> sizeString; 3538 llvm::raw_svector_ostream OS(sizeString); 3539 OS << "sizeof("; 3540 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3541 OS << ") - "; 3542 OS << "strlen("; 3543 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3544 OS << ") - 1"; 3545 3546 Diag(SL, diag::note_strncat_wrong_size) 3547 << FixItHint::CreateReplacement(SR, OS.str()); 3548} 3549 3550//===--- CHECK: Return Address of Stack Variable --------------------------===// 3551 3552static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3553 Decl *ParentDecl); 3554static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3555 Decl *ParentDecl); 3556 3557/// CheckReturnStackAddr - Check if a return statement returns the address 3558/// of a stack variable. 3559void 3560Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3561 SourceLocation ReturnLoc) { 3562 3563 Expr *stackE = 0; 3564 SmallVector<DeclRefExpr *, 8> refVars; 3565 3566 // Perform checking for returned stack addresses, local blocks, 3567 // label addresses or references to temporaries. 3568 if (lhsType->isPointerType() || 3569 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 3570 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 3571 } else if (lhsType->isReferenceType()) { 3572 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 3573 } 3574 3575 if (stackE == 0) 3576 return; // Nothing suspicious was found. 3577 3578 SourceLocation diagLoc; 3579 SourceRange diagRange; 3580 if (refVars.empty()) { 3581 diagLoc = stackE->getLocStart(); 3582 diagRange = stackE->getSourceRange(); 3583 } else { 3584 // We followed through a reference variable. 'stackE' contains the 3585 // problematic expression but we will warn at the return statement pointing 3586 // at the reference variable. We will later display the "trail" of 3587 // reference variables using notes. 3588 diagLoc = refVars[0]->getLocStart(); 3589 diagRange = refVars[0]->getSourceRange(); 3590 } 3591 3592 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 3593 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 3594 : diag::warn_ret_stack_addr) 3595 << DR->getDecl()->getDeclName() << diagRange; 3596 } else if (isa<BlockExpr>(stackE)) { // local block. 3597 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 3598 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 3599 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 3600 } else { // local temporary. 3601 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 3602 : diag::warn_ret_local_temp_addr) 3603 << diagRange; 3604 } 3605 3606 // Display the "trail" of reference variables that we followed until we 3607 // found the problematic expression using notes. 3608 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 3609 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 3610 // If this var binds to another reference var, show the range of the next 3611 // var, otherwise the var binds to the problematic expression, in which case 3612 // show the range of the expression. 3613 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 3614 : stackE->getSourceRange(); 3615 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 3616 << VD->getDeclName() << range; 3617 } 3618} 3619 3620/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 3621/// check if the expression in a return statement evaluates to an address 3622/// to a location on the stack, a local block, an address of a label, or a 3623/// reference to local temporary. The recursion is used to traverse the 3624/// AST of the return expression, with recursion backtracking when we 3625/// encounter a subexpression that (1) clearly does not lead to one of the 3626/// above problematic expressions (2) is something we cannot determine leads to 3627/// a problematic expression based on such local checking. 3628/// 3629/// Both EvalAddr and EvalVal follow through reference variables to evaluate 3630/// the expression that they point to. Such variables are added to the 3631/// 'refVars' vector so that we know what the reference variable "trail" was. 3632/// 3633/// EvalAddr processes expressions that are pointers that are used as 3634/// references (and not L-values). EvalVal handles all other values. 3635/// At the base case of the recursion is a check for the above problematic 3636/// expressions. 3637/// 3638/// This implementation handles: 3639/// 3640/// * pointer-to-pointer casts 3641/// * implicit conversions from array references to pointers 3642/// * taking the address of fields 3643/// * arbitrary interplay between "&" and "*" operators 3644/// * pointer arithmetic from an address of a stack variable 3645/// * taking the address of an array element where the array is on the stack 3646static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3647 Decl *ParentDecl) { 3648 if (E->isTypeDependent()) 3649 return NULL; 3650 3651 // We should only be called for evaluating pointer expressions. 3652 assert((E->getType()->isAnyPointerType() || 3653 E->getType()->isBlockPointerType() || 3654 E->getType()->isObjCQualifiedIdType()) && 3655 "EvalAddr only works on pointers"); 3656 3657 E = E->IgnoreParens(); 3658 3659 // Our "symbolic interpreter" is just a dispatch off the currently 3660 // viewed AST node. We then recursively traverse the AST by calling 3661 // EvalAddr and EvalVal appropriately. 3662 switch (E->getStmtClass()) { 3663 case Stmt::DeclRefExprClass: { 3664 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3665 3666 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3667 // If this is a reference variable, follow through to the expression that 3668 // it points to. 3669 if (V->hasLocalStorage() && 3670 V->getType()->isReferenceType() && V->hasInit()) { 3671 // Add the reference variable to the "trail". 3672 refVars.push_back(DR); 3673 return EvalAddr(V->getInit(), refVars, ParentDecl); 3674 } 3675 3676 return NULL; 3677 } 3678 3679 case Stmt::UnaryOperatorClass: { 3680 // The only unary operator that make sense to handle here 3681 // is AddrOf. All others don't make sense as pointers. 3682 UnaryOperator *U = cast<UnaryOperator>(E); 3683 3684 if (U->getOpcode() == UO_AddrOf) 3685 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 3686 else 3687 return NULL; 3688 } 3689 3690 case Stmt::BinaryOperatorClass: { 3691 // Handle pointer arithmetic. All other binary operators are not valid 3692 // in this context. 3693 BinaryOperator *B = cast<BinaryOperator>(E); 3694 BinaryOperatorKind op = B->getOpcode(); 3695 3696 if (op != BO_Add && op != BO_Sub) 3697 return NULL; 3698 3699 Expr *Base = B->getLHS(); 3700 3701 // Determine which argument is the real pointer base. It could be 3702 // the RHS argument instead of the LHS. 3703 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 3704 3705 assert (Base->getType()->isPointerType()); 3706 return EvalAddr(Base, refVars, ParentDecl); 3707 } 3708 3709 // For conditional operators we need to see if either the LHS or RHS are 3710 // valid DeclRefExpr*s. If one of them is valid, we return it. 3711 case Stmt::ConditionalOperatorClass: { 3712 ConditionalOperator *C = cast<ConditionalOperator>(E); 3713 3714 // Handle the GNU extension for missing LHS. 3715 if (Expr *lhsExpr = C->getLHS()) { 3716 // In C++, we can have a throw-expression, which has 'void' type. 3717 if (!lhsExpr->getType()->isVoidType()) 3718 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 3719 return LHS; 3720 } 3721 3722 // In C++, we can have a throw-expression, which has 'void' type. 3723 if (C->getRHS()->getType()->isVoidType()) 3724 return NULL; 3725 3726 return EvalAddr(C->getRHS(), refVars, ParentDecl); 3727 } 3728 3729 case Stmt::BlockExprClass: 3730 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 3731 return E; // local block. 3732 return NULL; 3733 3734 case Stmt::AddrLabelExprClass: 3735 return E; // address of label. 3736 3737 case Stmt::ExprWithCleanupsClass: 3738 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 3739 ParentDecl); 3740 3741 // For casts, we need to handle conversions from arrays to 3742 // pointer values, and pointer-to-pointer conversions. 3743 case Stmt::ImplicitCastExprClass: 3744 case Stmt::CStyleCastExprClass: 3745 case Stmt::CXXFunctionalCastExprClass: 3746 case Stmt::ObjCBridgedCastExprClass: 3747 case Stmt::CXXStaticCastExprClass: 3748 case Stmt::CXXDynamicCastExprClass: 3749 case Stmt::CXXConstCastExprClass: 3750 case Stmt::CXXReinterpretCastExprClass: { 3751 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 3752 switch (cast<CastExpr>(E)->getCastKind()) { 3753 case CK_BitCast: 3754 case CK_LValueToRValue: 3755 case CK_NoOp: 3756 case CK_BaseToDerived: 3757 case CK_DerivedToBase: 3758 case CK_UncheckedDerivedToBase: 3759 case CK_Dynamic: 3760 case CK_CPointerToObjCPointerCast: 3761 case CK_BlockPointerToObjCPointerCast: 3762 case CK_AnyPointerToBlockPointerCast: 3763 return EvalAddr(SubExpr, refVars, ParentDecl); 3764 3765 case CK_ArrayToPointerDecay: 3766 return EvalVal(SubExpr, refVars, ParentDecl); 3767 3768 default: 3769 return 0; 3770 } 3771 } 3772 3773 case Stmt::MaterializeTemporaryExprClass: 3774 if (Expr *Result = EvalAddr( 3775 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3776 refVars, ParentDecl)) 3777 return Result; 3778 3779 return E; 3780 3781 // Everything else: we simply don't reason about them. 3782 default: 3783 return NULL; 3784 } 3785} 3786 3787 3788/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3789/// See the comments for EvalAddr for more details. 3790static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3791 Decl *ParentDecl) { 3792do { 3793 // We should only be called for evaluating non-pointer expressions, or 3794 // expressions with a pointer type that are not used as references but instead 3795 // are l-values (e.g., DeclRefExpr with a pointer type). 3796 3797 // Our "symbolic interpreter" is just a dispatch off the currently 3798 // viewed AST node. We then recursively traverse the AST by calling 3799 // EvalAddr and EvalVal appropriately. 3800 3801 E = E->IgnoreParens(); 3802 switch (E->getStmtClass()) { 3803 case Stmt::ImplicitCastExprClass: { 3804 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3805 if (IE->getValueKind() == VK_LValue) { 3806 E = IE->getSubExpr(); 3807 continue; 3808 } 3809 return NULL; 3810 } 3811 3812 case Stmt::ExprWithCleanupsClass: 3813 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 3814 3815 case Stmt::DeclRefExprClass: { 3816 // When we hit a DeclRefExpr we are looking at code that refers to a 3817 // variable's name. If it's not a reference variable we check if it has 3818 // local storage within the function, and if so, return the expression. 3819 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3820 3821 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 3822 // Check if it refers to itself, e.g. "int& i = i;". 3823 if (V == ParentDecl) 3824 return DR; 3825 3826 if (V->hasLocalStorage()) { 3827 if (!V->getType()->isReferenceType()) 3828 return DR; 3829 3830 // Reference variable, follow through to the expression that 3831 // it points to. 3832 if (V->hasInit()) { 3833 // Add the reference variable to the "trail". 3834 refVars.push_back(DR); 3835 return EvalVal(V->getInit(), refVars, V); 3836 } 3837 } 3838 } 3839 3840 return NULL; 3841 } 3842 3843 case Stmt::UnaryOperatorClass: { 3844 // The only unary operator that make sense to handle here 3845 // is Deref. All others don't resolve to a "name." This includes 3846 // handling all sorts of rvalues passed to a unary operator. 3847 UnaryOperator *U = cast<UnaryOperator>(E); 3848 3849 if (U->getOpcode() == UO_Deref) 3850 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 3851 3852 return NULL; 3853 } 3854 3855 case Stmt::ArraySubscriptExprClass: { 3856 // Array subscripts are potential references to data on the stack. We 3857 // retrieve the DeclRefExpr* for the array variable if it indeed 3858 // has local storage. 3859 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 3860 } 3861 3862 case Stmt::ConditionalOperatorClass: { 3863 // For conditional operators we need to see if either the LHS or RHS are 3864 // non-NULL Expr's. If one is non-NULL, we return it. 3865 ConditionalOperator *C = cast<ConditionalOperator>(E); 3866 3867 // Handle the GNU extension for missing LHS. 3868 if (Expr *lhsExpr = C->getLHS()) 3869 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 3870 return LHS; 3871 3872 return EvalVal(C->getRHS(), refVars, ParentDecl); 3873 } 3874 3875 // Accesses to members are potential references to data on the stack. 3876 case Stmt::MemberExprClass: { 3877 MemberExpr *M = cast<MemberExpr>(E); 3878 3879 // Check for indirect access. We only want direct field accesses. 3880 if (M->isArrow()) 3881 return NULL; 3882 3883 // Check whether the member type is itself a reference, in which case 3884 // we're not going to refer to the member, but to what the member refers to. 3885 if (M->getMemberDecl()->getType()->isReferenceType()) 3886 return NULL; 3887 3888 return EvalVal(M->getBase(), refVars, ParentDecl); 3889 } 3890 3891 case Stmt::MaterializeTemporaryExprClass: 3892 if (Expr *Result = EvalVal( 3893 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3894 refVars, ParentDecl)) 3895 return Result; 3896 3897 return E; 3898 3899 default: 3900 // Check that we don't return or take the address of a reference to a 3901 // temporary. This is only useful in C++. 3902 if (!E->isTypeDependent() && E->isRValue()) 3903 return E; 3904 3905 // Everything else: we simply don't reason about them. 3906 return NULL; 3907 } 3908} while (true); 3909} 3910 3911//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3912 3913/// Check for comparisons of floating point operands using != and ==. 3914/// Issue a warning if these are no self-comparisons, as they are not likely 3915/// to do what the programmer intended. 3916void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3917 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3918 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3919 3920 // Special case: check for x == x (which is OK). 3921 // Do not emit warnings for such cases. 3922 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3923 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3924 if (DRL->getDecl() == DRR->getDecl()) 3925 return; 3926 3927 3928 // Special case: check for comparisons against literals that can be exactly 3929 // represented by APFloat. In such cases, do not emit a warning. This 3930 // is a heuristic: often comparison against such literals are used to 3931 // detect if a value in a variable has not changed. This clearly can 3932 // lead to false negatives. 3933 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3934 if (FLL->isExact()) 3935 return; 3936 } else 3937 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)) 3938 if (FLR->isExact()) 3939 return; 3940 3941 // Check for comparisons with builtin types. 3942 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3943 if (CL->isBuiltinCall()) 3944 return; 3945 3946 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3947 if (CR->isBuiltinCall()) 3948 return; 3949 3950 // Emit the diagnostic. 3951 Diag(Loc, diag::warn_floatingpoint_eq) 3952 << LHS->getSourceRange() << RHS->getSourceRange(); 3953} 3954 3955//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3956//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3957 3958namespace { 3959 3960/// Structure recording the 'active' range of an integer-valued 3961/// expression. 3962struct IntRange { 3963 /// The number of bits active in the int. 3964 unsigned Width; 3965 3966 /// True if the int is known not to have negative values. 3967 bool NonNegative; 3968 3969 IntRange(unsigned Width, bool NonNegative) 3970 : Width(Width), NonNegative(NonNegative) 3971 {} 3972 3973 /// Returns the range of the bool type. 3974 static IntRange forBoolType() { 3975 return IntRange(1, true); 3976 } 3977 3978 /// Returns the range of an opaque value of the given integral type. 3979 static IntRange forValueOfType(ASTContext &C, QualType T) { 3980 return forValueOfCanonicalType(C, 3981 T->getCanonicalTypeInternal().getTypePtr()); 3982 } 3983 3984 /// Returns the range of an opaque value of a canonical integral type. 3985 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3986 assert(T->isCanonicalUnqualified()); 3987 3988 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3989 T = VT->getElementType().getTypePtr(); 3990 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3991 T = CT->getElementType().getTypePtr(); 3992 3993 // For enum types, use the known bit width of the enumerators. 3994 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3995 EnumDecl *Enum = ET->getDecl(); 3996 if (!Enum->isCompleteDefinition()) 3997 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3998 3999 unsigned NumPositive = Enum->getNumPositiveBits(); 4000 unsigned NumNegative = Enum->getNumNegativeBits(); 4001 4002 if (NumNegative == 0) 4003 return IntRange(NumPositive, true/*NonNegative*/); 4004 else 4005 return IntRange(std::max(NumPositive + 1, NumNegative), 4006 false/*NonNegative*/); 4007 } 4008 4009 const BuiltinType *BT = cast<BuiltinType>(T); 4010 assert(BT->isInteger()); 4011 4012 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4013 } 4014 4015 /// Returns the "target" range of a canonical integral type, i.e. 4016 /// the range of values expressible in the type. 4017 /// 4018 /// This matches forValueOfCanonicalType except that enums have the 4019 /// full range of their type, not the range of their enumerators. 4020 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 4021 assert(T->isCanonicalUnqualified()); 4022 4023 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4024 T = VT->getElementType().getTypePtr(); 4025 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4026 T = CT->getElementType().getTypePtr(); 4027 if (const EnumType *ET = dyn_cast<EnumType>(T)) 4028 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 4029 4030 const BuiltinType *BT = cast<BuiltinType>(T); 4031 assert(BT->isInteger()); 4032 4033 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4034 } 4035 4036 /// Returns the supremum of two ranges: i.e. their conservative merge. 4037 static IntRange join(IntRange L, IntRange R) { 4038 return IntRange(std::max(L.Width, R.Width), 4039 L.NonNegative && R.NonNegative); 4040 } 4041 4042 /// Returns the infinum of two ranges: i.e. their aggressive merge. 4043 static IntRange meet(IntRange L, IntRange R) { 4044 return IntRange(std::min(L.Width, R.Width), 4045 L.NonNegative || R.NonNegative); 4046 } 4047}; 4048 4049static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 4050 unsigned MaxWidth) { 4051 if (value.isSigned() && value.isNegative()) 4052 return IntRange(value.getMinSignedBits(), false); 4053 4054 if (value.getBitWidth() > MaxWidth) 4055 value = value.trunc(MaxWidth); 4056 4057 // isNonNegative() just checks the sign bit without considering 4058 // signedness. 4059 return IntRange(value.getActiveBits(), true); 4060} 4061 4062static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 4063 unsigned MaxWidth) { 4064 if (result.isInt()) 4065 return GetValueRange(C, result.getInt(), MaxWidth); 4066 4067 if (result.isVector()) { 4068 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 4069 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 4070 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 4071 R = IntRange::join(R, El); 4072 } 4073 return R; 4074 } 4075 4076 if (result.isComplexInt()) { 4077 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 4078 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 4079 return IntRange::join(R, I); 4080 } 4081 4082 // This can happen with lossless casts to intptr_t of "based" lvalues. 4083 // Assume it might use arbitrary bits. 4084 // FIXME: The only reason we need to pass the type in here is to get 4085 // the sign right on this one case. It would be nice if APValue 4086 // preserved this. 4087 assert(result.isLValue() || result.isAddrLabelDiff()); 4088 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 4089} 4090 4091/// Pseudo-evaluate the given integer expression, estimating the 4092/// range of values it might take. 4093/// 4094/// \param MaxWidth - the width to which the value will be truncated 4095static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 4096 E = E->IgnoreParens(); 4097 4098 // Try a full evaluation first. 4099 Expr::EvalResult result; 4100 if (E->EvaluateAsRValue(result, C)) 4101 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 4102 4103 // I think we only want to look through implicit casts here; if the 4104 // user has an explicit widening cast, we should treat the value as 4105 // being of the new, wider type. 4106 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 4107 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 4108 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 4109 4110 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 4111 4112 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 4113 4114 // Assume that non-integer casts can span the full range of the type. 4115 if (!isIntegerCast) 4116 return OutputTypeRange; 4117 4118 IntRange SubRange 4119 = GetExprRange(C, CE->getSubExpr(), 4120 std::min(MaxWidth, OutputTypeRange.Width)); 4121 4122 // Bail out if the subexpr's range is as wide as the cast type. 4123 if (SubRange.Width >= OutputTypeRange.Width) 4124 return OutputTypeRange; 4125 4126 // Otherwise, we take the smaller width, and we're non-negative if 4127 // either the output type or the subexpr is. 4128 return IntRange(SubRange.Width, 4129 SubRange.NonNegative || OutputTypeRange.NonNegative); 4130 } 4131 4132 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 4133 // If we can fold the condition, just take that operand. 4134 bool CondResult; 4135 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 4136 return GetExprRange(C, CondResult ? CO->getTrueExpr() 4137 : CO->getFalseExpr(), 4138 MaxWidth); 4139 4140 // Otherwise, conservatively merge. 4141 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 4142 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 4143 return IntRange::join(L, R); 4144 } 4145 4146 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4147 switch (BO->getOpcode()) { 4148 4149 // Boolean-valued operations are single-bit and positive. 4150 case BO_LAnd: 4151 case BO_LOr: 4152 case BO_LT: 4153 case BO_GT: 4154 case BO_LE: 4155 case BO_GE: 4156 case BO_EQ: 4157 case BO_NE: 4158 return IntRange::forBoolType(); 4159 4160 // The type of the assignments is the type of the LHS, so the RHS 4161 // is not necessarily the same type. 4162 case BO_MulAssign: 4163 case BO_DivAssign: 4164 case BO_RemAssign: 4165 case BO_AddAssign: 4166 case BO_SubAssign: 4167 case BO_XorAssign: 4168 case BO_OrAssign: 4169 // TODO: bitfields? 4170 return IntRange::forValueOfType(C, E->getType()); 4171 4172 // Simple assignments just pass through the RHS, which will have 4173 // been coerced to the LHS type. 4174 case BO_Assign: 4175 // TODO: bitfields? 4176 return GetExprRange(C, BO->getRHS(), MaxWidth); 4177 4178 // Operations with opaque sources are black-listed. 4179 case BO_PtrMemD: 4180 case BO_PtrMemI: 4181 return IntRange::forValueOfType(C, E->getType()); 4182 4183 // Bitwise-and uses the *infinum* of the two source ranges. 4184 case BO_And: 4185 case BO_AndAssign: 4186 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 4187 GetExprRange(C, BO->getRHS(), MaxWidth)); 4188 4189 // Left shift gets black-listed based on a judgement call. 4190 case BO_Shl: 4191 // ...except that we want to treat '1 << (blah)' as logically 4192 // positive. It's an important idiom. 4193 if (IntegerLiteral *I 4194 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 4195 if (I->getValue() == 1) { 4196 IntRange R = IntRange::forValueOfType(C, E->getType()); 4197 return IntRange(R.Width, /*NonNegative*/ true); 4198 } 4199 } 4200 // fallthrough 4201 4202 case BO_ShlAssign: 4203 return IntRange::forValueOfType(C, E->getType()); 4204 4205 // Right shift by a constant can narrow its left argument. 4206 case BO_Shr: 4207 case BO_ShrAssign: { 4208 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4209 4210 // If the shift amount is a positive constant, drop the width by 4211 // that much. 4212 llvm::APSInt shift; 4213 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 4214 shift.isNonNegative()) { 4215 unsigned zext = shift.getZExtValue(); 4216 if (zext >= L.Width) 4217 L.Width = (L.NonNegative ? 0 : 1); 4218 else 4219 L.Width -= zext; 4220 } 4221 4222 return L; 4223 } 4224 4225 // Comma acts as its right operand. 4226 case BO_Comma: 4227 return GetExprRange(C, BO->getRHS(), MaxWidth); 4228 4229 // Black-list pointer subtractions. 4230 case BO_Sub: 4231 if (BO->getLHS()->getType()->isPointerType()) 4232 return IntRange::forValueOfType(C, E->getType()); 4233 break; 4234 4235 // The width of a division result is mostly determined by the size 4236 // of the LHS. 4237 case BO_Div: { 4238 // Don't 'pre-truncate' the operands. 4239 unsigned opWidth = C.getIntWidth(E->getType()); 4240 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4241 4242 // If the divisor is constant, use that. 4243 llvm::APSInt divisor; 4244 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 4245 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 4246 if (log2 >= L.Width) 4247 L.Width = (L.NonNegative ? 0 : 1); 4248 else 4249 L.Width = std::min(L.Width - log2, MaxWidth); 4250 return L; 4251 } 4252 4253 // Otherwise, just use the LHS's width. 4254 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4255 return IntRange(L.Width, L.NonNegative && R.NonNegative); 4256 } 4257 4258 // The result of a remainder can't be larger than the result of 4259 // either side. 4260 case BO_Rem: { 4261 // Don't 'pre-truncate' the operands. 4262 unsigned opWidth = C.getIntWidth(E->getType()); 4263 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4264 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4265 4266 IntRange meet = IntRange::meet(L, R); 4267 meet.Width = std::min(meet.Width, MaxWidth); 4268 return meet; 4269 } 4270 4271 // The default behavior is okay for these. 4272 case BO_Mul: 4273 case BO_Add: 4274 case BO_Xor: 4275 case BO_Or: 4276 break; 4277 } 4278 4279 // The default case is to treat the operation as if it were closed 4280 // on the narrowest type that encompasses both operands. 4281 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4282 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 4283 return IntRange::join(L, R); 4284 } 4285 4286 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 4287 switch (UO->getOpcode()) { 4288 // Boolean-valued operations are white-listed. 4289 case UO_LNot: 4290 return IntRange::forBoolType(); 4291 4292 // Operations with opaque sources are black-listed. 4293 case UO_Deref: 4294 case UO_AddrOf: // should be impossible 4295 return IntRange::forValueOfType(C, E->getType()); 4296 4297 default: 4298 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 4299 } 4300 } 4301 4302 if (dyn_cast<OffsetOfExpr>(E)) { 4303 IntRange::forValueOfType(C, E->getType()); 4304 } 4305 4306 if (FieldDecl *BitField = E->getBitField()) 4307 return IntRange(BitField->getBitWidthValue(C), 4308 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 4309 4310 return IntRange::forValueOfType(C, E->getType()); 4311} 4312 4313static IntRange GetExprRange(ASTContext &C, Expr *E) { 4314 return GetExprRange(C, E, C.getIntWidth(E->getType())); 4315} 4316 4317/// Checks whether the given value, which currently has the given 4318/// source semantics, has the same value when coerced through the 4319/// target semantics. 4320static bool IsSameFloatAfterCast(const llvm::APFloat &value, 4321 const llvm::fltSemantics &Src, 4322 const llvm::fltSemantics &Tgt) { 4323 llvm::APFloat truncated = value; 4324 4325 bool ignored; 4326 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 4327 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 4328 4329 return truncated.bitwiseIsEqual(value); 4330} 4331 4332/// Checks whether the given value, which currently has the given 4333/// source semantics, has the same value when coerced through the 4334/// target semantics. 4335/// 4336/// The value might be a vector of floats (or a complex number). 4337static bool IsSameFloatAfterCast(const APValue &value, 4338 const llvm::fltSemantics &Src, 4339 const llvm::fltSemantics &Tgt) { 4340 if (value.isFloat()) 4341 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 4342 4343 if (value.isVector()) { 4344 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 4345 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 4346 return false; 4347 return true; 4348 } 4349 4350 assert(value.isComplexFloat()); 4351 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 4352 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 4353} 4354 4355static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 4356 4357static bool IsZero(Sema &S, Expr *E) { 4358 // Suppress cases where we are comparing against an enum constant. 4359 if (const DeclRefExpr *DR = 4360 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 4361 if (isa<EnumConstantDecl>(DR->getDecl())) 4362 return false; 4363 4364 // Suppress cases where the '0' value is expanded from a macro. 4365 if (E->getLocStart().isMacroID()) 4366 return false; 4367 4368 llvm::APSInt Value; 4369 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 4370} 4371 4372static bool HasEnumType(Expr *E) { 4373 // Strip off implicit integral promotions. 4374 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 4375 if (ICE->getCastKind() != CK_IntegralCast && 4376 ICE->getCastKind() != CK_NoOp) 4377 break; 4378 E = ICE->getSubExpr(); 4379 } 4380 4381 return E->getType()->isEnumeralType(); 4382} 4383 4384static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 4385 BinaryOperatorKind op = E->getOpcode(); 4386 if (E->isValueDependent()) 4387 return; 4388 4389 if (op == BO_LT && IsZero(S, E->getRHS())) { 4390 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4391 << "< 0" << "false" << HasEnumType(E->getLHS()) 4392 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4393 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 4394 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4395 << ">= 0" << "true" << HasEnumType(E->getLHS()) 4396 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4397 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 4398 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4399 << "0 >" << "false" << HasEnumType(E->getRHS()) 4400 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4401 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 4402 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4403 << "0 <=" << "true" << HasEnumType(E->getRHS()) 4404 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4405 } 4406} 4407 4408static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E, 4409 Expr *Constant, Expr *Other, 4410 llvm::APSInt Value, 4411 bool RhsConstant) { 4412 // 0 values are handled later by CheckTrivialUnsignedComparison(). 4413 if (Value == 0) 4414 return; 4415 4416 BinaryOperatorKind op = E->getOpcode(); 4417 QualType OtherT = Other->getType(); 4418 QualType ConstantT = Constant->getType(); 4419 QualType CommonT = E->getLHS()->getType(); 4420 if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT)) 4421 return; 4422 assert((OtherT->isIntegerType() && ConstantT->isIntegerType()) 4423 && "comparison with non-integer type"); 4424 4425 bool ConstantSigned = ConstantT->isSignedIntegerType(); 4426 bool CommonSigned = CommonT->isSignedIntegerType(); 4427 4428 bool EqualityOnly = false; 4429 4430 // TODO: Investigate using GetExprRange() to get tighter bounds on 4431 // on the bit ranges. 4432 IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT); 4433 unsigned OtherWidth = OtherRange.Width; 4434 4435 if (CommonSigned) { 4436 // The common type is signed, therefore no signed to unsigned conversion. 4437 if (!OtherRange.NonNegative) { 4438 // Check that the constant is representable in type OtherT. 4439 if (ConstantSigned) { 4440 if (OtherWidth >= Value.getMinSignedBits()) 4441 return; 4442 } else { // !ConstantSigned 4443 if (OtherWidth >= Value.getActiveBits() + 1) 4444 return; 4445 } 4446 } else { // !OtherSigned 4447 // Check that the constant is representable in type OtherT. 4448 // Negative values are out of range. 4449 if (ConstantSigned) { 4450 if (Value.isNonNegative() && OtherWidth >= Value.getActiveBits()) 4451 return; 4452 } else { // !ConstantSigned 4453 if (OtherWidth >= Value.getActiveBits()) 4454 return; 4455 } 4456 } 4457 } else { // !CommonSigned 4458 if (OtherRange.NonNegative) { 4459 if (OtherWidth >= Value.getActiveBits()) 4460 return; 4461 } else if (!OtherRange.NonNegative && !ConstantSigned) { 4462 // Check to see if the constant is representable in OtherT. 4463 if (OtherWidth > Value.getActiveBits()) 4464 return; 4465 // Check to see if the constant is equivalent to a negative value 4466 // cast to CommonT. 4467 if (S.Context.getIntWidth(ConstantT) == S.Context.getIntWidth(CommonT) && 4468 Value.isNegative() && Value.getMinSignedBits() <= OtherWidth) 4469 return; 4470 // The constant value rests between values that OtherT can represent after 4471 // conversion. Relational comparison still works, but equality 4472 // comparisons will be tautological. 4473 EqualityOnly = true; 4474 } else { // OtherSigned && ConstantSigned 4475 assert(0 && "Two signed types converted to unsigned types."); 4476 } 4477 } 4478 4479 bool PositiveConstant = !ConstantSigned || Value.isNonNegative(); 4480 4481 bool IsTrue = true; 4482 if (op == BO_EQ || op == BO_NE) { 4483 IsTrue = op == BO_NE; 4484 } else if (EqualityOnly) { 4485 return; 4486 } else if (RhsConstant) { 4487 if (op == BO_GT || op == BO_GE) 4488 IsTrue = !PositiveConstant; 4489 else // op == BO_LT || op == BO_LE 4490 IsTrue = PositiveConstant; 4491 } else { 4492 if (op == BO_LT || op == BO_LE) 4493 IsTrue = !PositiveConstant; 4494 else // op == BO_GT || op == BO_GE 4495 IsTrue = PositiveConstant; 4496 } 4497 SmallString<16> PrettySourceValue(Value.toString(10)); 4498 S.Diag(E->getOperatorLoc(), diag::warn_out_of_range_compare) 4499 << PrettySourceValue << OtherT << IsTrue 4500 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4501} 4502 4503/// Analyze the operands of the given comparison. Implements the 4504/// fallback case from AnalyzeComparison. 4505static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 4506 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4507 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4508} 4509 4510/// \brief Implements -Wsign-compare. 4511/// 4512/// \param E the binary operator to check for warnings 4513static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 4514 // The type the comparison is being performed in. 4515 QualType T = E->getLHS()->getType(); 4516 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 4517 && "comparison with mismatched types"); 4518 if (E->isValueDependent()) 4519 return AnalyzeImpConvsInComparison(S, E); 4520 4521 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 4522 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 4523 4524 bool IsComparisonConstant = false; 4525 4526 // Check whether an integer constant comparison results in a value 4527 // of 'true' or 'false'. 4528 if (T->isIntegralType(S.Context)) { 4529 llvm::APSInt RHSValue; 4530 bool IsRHSIntegralLiteral = 4531 RHS->isIntegerConstantExpr(RHSValue, S.Context); 4532 llvm::APSInt LHSValue; 4533 bool IsLHSIntegralLiteral = 4534 LHS->isIntegerConstantExpr(LHSValue, S.Context); 4535 if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral) 4536 DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true); 4537 else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral) 4538 DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false); 4539 else 4540 IsComparisonConstant = 4541 (IsRHSIntegralLiteral && IsLHSIntegralLiteral); 4542 } else if (!T->hasUnsignedIntegerRepresentation()) 4543 IsComparisonConstant = E->isIntegerConstantExpr(S.Context); 4544 4545 // We don't do anything special if this isn't an unsigned integral 4546 // comparison: we're only interested in integral comparisons, and 4547 // signed comparisons only happen in cases we don't care to warn about. 4548 // 4549 // We also don't care about value-dependent expressions or expressions 4550 // whose result is a constant. 4551 if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant) 4552 return AnalyzeImpConvsInComparison(S, E); 4553 4554 // Check to see if one of the (unmodified) operands is of different 4555 // signedness. 4556 Expr *signedOperand, *unsignedOperand; 4557 if (LHS->getType()->hasSignedIntegerRepresentation()) { 4558 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 4559 "unsigned comparison between two signed integer expressions?"); 4560 signedOperand = LHS; 4561 unsignedOperand = RHS; 4562 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 4563 signedOperand = RHS; 4564 unsignedOperand = LHS; 4565 } else { 4566 CheckTrivialUnsignedComparison(S, E); 4567 return AnalyzeImpConvsInComparison(S, E); 4568 } 4569 4570 // Otherwise, calculate the effective range of the signed operand. 4571 IntRange signedRange = GetExprRange(S.Context, signedOperand); 4572 4573 // Go ahead and analyze implicit conversions in the operands. Note 4574 // that we skip the implicit conversions on both sides. 4575 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 4576 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 4577 4578 // If the signed range is non-negative, -Wsign-compare won't fire, 4579 // but we should still check for comparisons which are always true 4580 // or false. 4581 if (signedRange.NonNegative) 4582 return CheckTrivialUnsignedComparison(S, E); 4583 4584 // For (in)equality comparisons, if the unsigned operand is a 4585 // constant which cannot collide with a overflowed signed operand, 4586 // then reinterpreting the signed operand as unsigned will not 4587 // change the result of the comparison. 4588 if (E->isEqualityOp()) { 4589 unsigned comparisonWidth = S.Context.getIntWidth(T); 4590 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 4591 4592 // We should never be unable to prove that the unsigned operand is 4593 // non-negative. 4594 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 4595 4596 if (unsignedRange.Width < comparisonWidth) 4597 return; 4598 } 4599 4600 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 4601 S.PDiag(diag::warn_mixed_sign_comparison) 4602 << LHS->getType() << RHS->getType() 4603 << LHS->getSourceRange() << RHS->getSourceRange()); 4604} 4605 4606/// Analyzes an attempt to assign the given value to a bitfield. 4607/// 4608/// Returns true if there was something fishy about the attempt. 4609static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 4610 SourceLocation InitLoc) { 4611 assert(Bitfield->isBitField()); 4612 if (Bitfield->isInvalidDecl()) 4613 return false; 4614 4615 // White-list bool bitfields. 4616 if (Bitfield->getType()->isBooleanType()) 4617 return false; 4618 4619 // Ignore value- or type-dependent expressions. 4620 if (Bitfield->getBitWidth()->isValueDependent() || 4621 Bitfield->getBitWidth()->isTypeDependent() || 4622 Init->isValueDependent() || 4623 Init->isTypeDependent()) 4624 return false; 4625 4626 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 4627 4628 llvm::APSInt Value; 4629 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 4630 return false; 4631 4632 unsigned OriginalWidth = Value.getBitWidth(); 4633 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 4634 4635 if (OriginalWidth <= FieldWidth) 4636 return false; 4637 4638 // Compute the value which the bitfield will contain. 4639 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 4640 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 4641 4642 // Check whether the stored value is equal to the original value. 4643 TruncatedValue = TruncatedValue.extend(OriginalWidth); 4644 if (llvm::APSInt::isSameValue(Value, TruncatedValue)) 4645 return false; 4646 4647 // Special-case bitfields of width 1: booleans are naturally 0/1, and 4648 // therefore don't strictly fit into a signed bitfield of width 1. 4649 if (FieldWidth == 1 && Value == 1) 4650 return false; 4651 4652 std::string PrettyValue = Value.toString(10); 4653 std::string PrettyTrunc = TruncatedValue.toString(10); 4654 4655 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 4656 << PrettyValue << PrettyTrunc << OriginalInit->getType() 4657 << Init->getSourceRange(); 4658 4659 return true; 4660} 4661 4662/// Analyze the given simple or compound assignment for warning-worthy 4663/// operations. 4664static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 4665 // Just recurse on the LHS. 4666 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4667 4668 // We want to recurse on the RHS as normal unless we're assigning to 4669 // a bitfield. 4670 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 4671 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 4672 E->getOperatorLoc())) { 4673 // Recurse, ignoring any implicit conversions on the RHS. 4674 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 4675 E->getOperatorLoc()); 4676 } 4677 } 4678 4679 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4680} 4681 4682/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4683static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 4684 SourceLocation CContext, unsigned diag, 4685 bool pruneControlFlow = false) { 4686 if (pruneControlFlow) { 4687 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4688 S.PDiag(diag) 4689 << SourceType << T << E->getSourceRange() 4690 << SourceRange(CContext)); 4691 return; 4692 } 4693 S.Diag(E->getExprLoc(), diag) 4694 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 4695} 4696 4697/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4698static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 4699 SourceLocation CContext, unsigned diag, 4700 bool pruneControlFlow = false) { 4701 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 4702} 4703 4704/// Diagnose an implicit cast from a literal expression. Does not warn when the 4705/// cast wouldn't lose information. 4706void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 4707 SourceLocation CContext) { 4708 // Try to convert the literal exactly to an integer. If we can, don't warn. 4709 bool isExact = false; 4710 const llvm::APFloat &Value = FL->getValue(); 4711 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 4712 T->hasUnsignedIntegerRepresentation()); 4713 if (Value.convertToInteger(IntegerValue, 4714 llvm::APFloat::rmTowardZero, &isExact) 4715 == llvm::APFloat::opOK && isExact) 4716 return; 4717 4718 SmallString<16> PrettySourceValue; 4719 Value.toString(PrettySourceValue); 4720 SmallString<16> PrettyTargetValue; 4721 if (T->isSpecificBuiltinType(BuiltinType::Bool)) 4722 PrettyTargetValue = IntegerValue == 0 ? "false" : "true"; 4723 else 4724 IntegerValue.toString(PrettyTargetValue); 4725 4726 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 4727 << FL->getType() << T.getUnqualifiedType() << PrettySourceValue 4728 << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext); 4729} 4730 4731std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 4732 if (!Range.Width) return "0"; 4733 4734 llvm::APSInt ValueInRange = Value; 4735 ValueInRange.setIsSigned(!Range.NonNegative); 4736 ValueInRange = ValueInRange.trunc(Range.Width); 4737 return ValueInRange.toString(10); 4738} 4739 4740static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) { 4741 if (!isa<ImplicitCastExpr>(Ex)) 4742 return false; 4743 4744 Expr *InnerE = Ex->IgnoreParenImpCasts(); 4745 const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr(); 4746 const Type *Source = 4747 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 4748 if (Target->isDependentType()) 4749 return false; 4750 4751 const BuiltinType *FloatCandidateBT = 4752 dyn_cast<BuiltinType>(ToBool ? Source : Target); 4753 const Type *BoolCandidateType = ToBool ? Target : Source; 4754 4755 return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) && 4756 FloatCandidateBT && (FloatCandidateBT->isFloatingPoint())); 4757} 4758 4759void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall, 4760 SourceLocation CC) { 4761 unsigned NumArgs = TheCall->getNumArgs(); 4762 for (unsigned i = 0; i < NumArgs; ++i) { 4763 Expr *CurrA = TheCall->getArg(i); 4764 if (!IsImplicitBoolFloatConversion(S, CurrA, true)) 4765 continue; 4766 4767 bool IsSwapped = ((i > 0) && 4768 IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false)); 4769 IsSwapped |= ((i < (NumArgs - 1)) && 4770 IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false)); 4771 if (IsSwapped) { 4772 // Warn on this floating-point to bool conversion. 4773 DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(), 4774 CurrA->getType(), CC, 4775 diag::warn_impcast_floating_point_to_bool); 4776 } 4777 } 4778} 4779 4780void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 4781 SourceLocation CC, bool *ICContext = 0) { 4782 if (E->isTypeDependent() || E->isValueDependent()) return; 4783 4784 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 4785 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 4786 if (Source == Target) return; 4787 if (Target->isDependentType()) return; 4788 4789 // If the conversion context location is invalid don't complain. We also 4790 // don't want to emit a warning if the issue occurs from the expansion of 4791 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 4792 // delay this check as long as possible. Once we detect we are in that 4793 // scenario, we just return. 4794 if (CC.isInvalid()) 4795 return; 4796 4797 // Diagnose implicit casts to bool. 4798 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 4799 if (isa<StringLiteral>(E)) 4800 // Warn on string literal to bool. Checks for string literals in logical 4801 // expressions, for instances, assert(0 && "error here"), is prevented 4802 // by a check in AnalyzeImplicitConversions(). 4803 return DiagnoseImpCast(S, E, T, CC, 4804 diag::warn_impcast_string_literal_to_bool); 4805 if (Source->isFunctionType()) { 4806 // Warn on function to bool. Checks free functions and static member 4807 // functions. Weakly imported functions are excluded from the check, 4808 // since it's common to test their value to check whether the linker 4809 // found a definition for them. 4810 ValueDecl *D = 0; 4811 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 4812 D = R->getDecl(); 4813 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 4814 D = M->getMemberDecl(); 4815 } 4816 4817 if (D && !D->isWeak()) { 4818 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 4819 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 4820 << F << E->getSourceRange() << SourceRange(CC); 4821 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 4822 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 4823 QualType ReturnType; 4824 UnresolvedSet<4> NonTemplateOverloads; 4825 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 4826 if (!ReturnType.isNull() 4827 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 4828 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 4829 << FixItHint::CreateInsertion( 4830 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 4831 return; 4832 } 4833 } 4834 } 4835 } 4836 4837 // Strip vector types. 4838 if (isa<VectorType>(Source)) { 4839 if (!isa<VectorType>(Target)) { 4840 if (S.SourceMgr.isInSystemMacro(CC)) 4841 return; 4842 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 4843 } 4844 4845 // If the vector cast is cast between two vectors of the same size, it is 4846 // a bitcast, not a conversion. 4847 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 4848 return; 4849 4850 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 4851 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 4852 } 4853 4854 // Strip complex types. 4855 if (isa<ComplexType>(Source)) { 4856 if (!isa<ComplexType>(Target)) { 4857 if (S.SourceMgr.isInSystemMacro(CC)) 4858 return; 4859 4860 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 4861 } 4862 4863 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 4864 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 4865 } 4866 4867 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 4868 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 4869 4870 // If the source is floating point... 4871 if (SourceBT && SourceBT->isFloatingPoint()) { 4872 // ...and the target is floating point... 4873 if (TargetBT && TargetBT->isFloatingPoint()) { 4874 // ...then warn if we're dropping FP rank. 4875 4876 // Builtin FP kinds are ordered by increasing FP rank. 4877 if (SourceBT->getKind() > TargetBT->getKind()) { 4878 // Don't warn about float constants that are precisely 4879 // representable in the target type. 4880 Expr::EvalResult result; 4881 if (E->EvaluateAsRValue(result, S.Context)) { 4882 // Value might be a float, a float vector, or a float complex. 4883 if (IsSameFloatAfterCast(result.Val, 4884 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 4885 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 4886 return; 4887 } 4888 4889 if (S.SourceMgr.isInSystemMacro(CC)) 4890 return; 4891 4892 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 4893 } 4894 return; 4895 } 4896 4897 // If the target is integral, always warn. 4898 if (TargetBT && TargetBT->isInteger()) { 4899 if (S.SourceMgr.isInSystemMacro(CC)) 4900 return; 4901 4902 Expr *InnerE = E->IgnoreParenImpCasts(); 4903 // We also want to warn on, e.g., "int i = -1.234" 4904 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 4905 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 4906 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 4907 4908 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 4909 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 4910 } else { 4911 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 4912 } 4913 } 4914 4915 // If the target is bool, warn if expr is a function or method call. 4916 if (Target->isSpecificBuiltinType(BuiltinType::Bool) && 4917 isa<CallExpr>(E)) { 4918 // Check last argument of function call to see if it is an 4919 // implicit cast from a type matching the type the result 4920 // is being cast to. 4921 CallExpr *CEx = cast<CallExpr>(E); 4922 unsigned NumArgs = CEx->getNumArgs(); 4923 if (NumArgs > 0) { 4924 Expr *LastA = CEx->getArg(NumArgs - 1); 4925 Expr *InnerE = LastA->IgnoreParenImpCasts(); 4926 const Type *InnerType = 4927 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 4928 if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) { 4929 // Warn on this floating-point to bool conversion 4930 DiagnoseImpCast(S, E, T, CC, 4931 diag::warn_impcast_floating_point_to_bool); 4932 } 4933 } 4934 } 4935 return; 4936 } 4937 4938 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 4939 == Expr::NPCK_GNUNull) && !Target->isAnyPointerType() 4940 && !Target->isBlockPointerType() && !Target->isMemberPointerType() 4941 && Target->isScalarType()) { 4942 SourceLocation Loc = E->getSourceRange().getBegin(); 4943 if (Loc.isMacroID()) 4944 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 4945 if (!Loc.isMacroID() || CC.isMacroID()) 4946 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 4947 << T << clang::SourceRange(CC) 4948 << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T)); 4949 } 4950 4951 if (!Source->isIntegerType() || !Target->isIntegerType()) 4952 return; 4953 4954 // TODO: remove this early return once the false positives for constant->bool 4955 // in templates, macros, etc, are reduced or removed. 4956 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 4957 return; 4958 4959 IntRange SourceRange = GetExprRange(S.Context, E); 4960 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 4961 4962 if (SourceRange.Width > TargetRange.Width) { 4963 // If the source is a constant, use a default-on diagnostic. 4964 // TODO: this should happen for bitfield stores, too. 4965 llvm::APSInt Value(32); 4966 if (E->isIntegerConstantExpr(Value, S.Context)) { 4967 if (S.SourceMgr.isInSystemMacro(CC)) 4968 return; 4969 4970 std::string PrettySourceValue = Value.toString(10); 4971 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 4972 4973 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4974 S.PDiag(diag::warn_impcast_integer_precision_constant) 4975 << PrettySourceValue << PrettyTargetValue 4976 << E->getType() << T << E->getSourceRange() 4977 << clang::SourceRange(CC)); 4978 return; 4979 } 4980 4981 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 4982 if (S.SourceMgr.isInSystemMacro(CC)) 4983 return; 4984 4985 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 4986 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 4987 /* pruneControlFlow */ true); 4988 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 4989 } 4990 4991 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 4992 (!TargetRange.NonNegative && SourceRange.NonNegative && 4993 SourceRange.Width == TargetRange.Width)) { 4994 4995 if (S.SourceMgr.isInSystemMacro(CC)) 4996 return; 4997 4998 unsigned DiagID = diag::warn_impcast_integer_sign; 4999 5000 // Traditionally, gcc has warned about this under -Wsign-compare. 5001 // We also want to warn about it in -Wconversion. 5002 // So if -Wconversion is off, use a completely identical diagnostic 5003 // in the sign-compare group. 5004 // The conditional-checking code will 5005 if (ICContext) { 5006 DiagID = diag::warn_impcast_integer_sign_conditional; 5007 *ICContext = true; 5008 } 5009 5010 return DiagnoseImpCast(S, E, T, CC, DiagID); 5011 } 5012 5013 // Diagnose conversions between different enumeration types. 5014 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 5015 // type, to give us better diagnostics. 5016 QualType SourceType = E->getType(); 5017 if (!S.getLangOpts().CPlusPlus) { 5018 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5019 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 5020 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 5021 SourceType = S.Context.getTypeDeclType(Enum); 5022 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 5023 } 5024 } 5025 5026 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 5027 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 5028 if ((SourceEnum->getDecl()->getIdentifier() || 5029 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 5030 (TargetEnum->getDecl()->getIdentifier() || 5031 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 5032 SourceEnum != TargetEnum) { 5033 if (S.SourceMgr.isInSystemMacro(CC)) 5034 return; 5035 5036 return DiagnoseImpCast(S, E, SourceType, T, CC, 5037 diag::warn_impcast_different_enum_types); 5038 } 5039 5040 return; 5041} 5042 5043void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5044 SourceLocation CC, QualType T); 5045 5046void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 5047 SourceLocation CC, bool &ICContext) { 5048 E = E->IgnoreParenImpCasts(); 5049 5050 if (isa<ConditionalOperator>(E)) 5051 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T); 5052 5053 AnalyzeImplicitConversions(S, E, CC); 5054 if (E->getType() != T) 5055 return CheckImplicitConversion(S, E, T, CC, &ICContext); 5056 return; 5057} 5058 5059void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5060 SourceLocation CC, QualType T) { 5061 AnalyzeImplicitConversions(S, E->getCond(), CC); 5062 5063 bool Suspicious = false; 5064 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 5065 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 5066 5067 // If -Wconversion would have warned about either of the candidates 5068 // for a signedness conversion to the context type... 5069 if (!Suspicious) return; 5070 5071 // ...but it's currently ignored... 5072 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 5073 CC)) 5074 return; 5075 5076 // ...then check whether it would have warned about either of the 5077 // candidates for a signedness conversion to the condition type. 5078 if (E->getType() == T) return; 5079 5080 Suspicious = false; 5081 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 5082 E->getType(), CC, &Suspicious); 5083 if (!Suspicious) 5084 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 5085 E->getType(), CC, &Suspicious); 5086} 5087 5088/// AnalyzeImplicitConversions - Find and report any interesting 5089/// implicit conversions in the given expression. There are a couple 5090/// of competing diagnostics here, -Wconversion and -Wsign-compare. 5091void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 5092 QualType T = OrigE->getType(); 5093 Expr *E = OrigE->IgnoreParenImpCasts(); 5094 5095 if (E->isTypeDependent() || E->isValueDependent()) 5096 return; 5097 5098 // For conditional operators, we analyze the arguments as if they 5099 // were being fed directly into the output. 5100 if (isa<ConditionalOperator>(E)) { 5101 ConditionalOperator *CO = cast<ConditionalOperator>(E); 5102 CheckConditionalOperator(S, CO, CC, T); 5103 return; 5104 } 5105 5106 // Check implicit argument conversions for function calls. 5107 if (CallExpr *Call = dyn_cast<CallExpr>(E)) 5108 CheckImplicitArgumentConversions(S, Call, CC); 5109 5110 // Go ahead and check any implicit conversions we might have skipped. 5111 // The non-canonical typecheck is just an optimization; 5112 // CheckImplicitConversion will filter out dead implicit conversions. 5113 if (E->getType() != T) 5114 CheckImplicitConversion(S, E, T, CC); 5115 5116 // Now continue drilling into this expression. 5117 5118 // Skip past explicit casts. 5119 if (isa<ExplicitCastExpr>(E)) { 5120 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 5121 return AnalyzeImplicitConversions(S, E, CC); 5122 } 5123 5124 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5125 // Do a somewhat different check with comparison operators. 5126 if (BO->isComparisonOp()) 5127 return AnalyzeComparison(S, BO); 5128 5129 // And with simple assignments. 5130 if (BO->getOpcode() == BO_Assign) 5131 return AnalyzeAssignment(S, BO); 5132 } 5133 5134 // These break the otherwise-useful invariant below. Fortunately, 5135 // we don't really need to recurse into them, because any internal 5136 // expressions should have been analyzed already when they were 5137 // built into statements. 5138 if (isa<StmtExpr>(E)) return; 5139 5140 // Don't descend into unevaluated contexts. 5141 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 5142 5143 // Now just recurse over the expression's children. 5144 CC = E->getExprLoc(); 5145 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 5146 bool IsLogicalOperator = BO && BO->isLogicalOp(); 5147 for (Stmt::child_range I = E->children(); I; ++I) { 5148 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 5149 if (!ChildExpr) 5150 continue; 5151 5152 if (IsLogicalOperator && 5153 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 5154 // Ignore checking string literals that are in logical operators. 5155 continue; 5156 AnalyzeImplicitConversions(S, ChildExpr, CC); 5157 } 5158} 5159 5160} // end anonymous namespace 5161 5162/// Diagnoses "dangerous" implicit conversions within the given 5163/// expression (which is a full expression). Implements -Wconversion 5164/// and -Wsign-compare. 5165/// 5166/// \param CC the "context" location of the implicit conversion, i.e. 5167/// the most location of the syntactic entity requiring the implicit 5168/// conversion 5169void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 5170 // Don't diagnose in unevaluated contexts. 5171 if (isUnevaluatedContext()) 5172 return; 5173 5174 // Don't diagnose for value- or type-dependent expressions. 5175 if (E->isTypeDependent() || E->isValueDependent()) 5176 return; 5177 5178 // Check for array bounds violations in cases where the check isn't triggered 5179 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 5180 // ArraySubscriptExpr is on the RHS of a variable initialization. 5181 CheckArrayAccess(E); 5182 5183 // This is not the right CC for (e.g.) a variable initialization. 5184 AnalyzeImplicitConversions(*this, E, CC); 5185} 5186 5187/// Diagnose when expression is an integer constant expression and its evaluation 5188/// results in integer overflow 5189void Sema::CheckForIntOverflow (Expr *E) { 5190 if (const BinaryOperator *BExpr = dyn_cast<BinaryOperator>(E->IgnoreParens())) { 5191 unsigned Opc = BExpr->getOpcode(); 5192 if (Opc != BO_Add && Opc != BO_Sub && Opc != BO_Mul) 5193 return; 5194 llvm::SmallVector<PartialDiagnosticAt, 4> Diags; 5195 E->EvaluateForOverflow(Context, &Diags); 5196 } 5197} 5198 5199namespace { 5200/// \brief Visitor for expressions which looks for unsequenced operations on the 5201/// same object. 5202class SequenceChecker : public EvaluatedExprVisitor<SequenceChecker> { 5203 /// \brief A tree of sequenced regions within an expression. Two regions are 5204 /// unsequenced if one is an ancestor or a descendent of the other. When we 5205 /// finish processing an expression with sequencing, such as a comma 5206 /// expression, we fold its tree nodes into its parent, since they are 5207 /// unsequenced with respect to nodes we will visit later. 5208 class SequenceTree { 5209 struct Value { 5210 explicit Value(unsigned Parent) : Parent(Parent), Merged(false) {} 5211 unsigned Parent : 31; 5212 bool Merged : 1; 5213 }; 5214 llvm::SmallVector<Value, 8> Values; 5215 5216 public: 5217 /// \brief A region within an expression which may be sequenced with respect 5218 /// to some other region. 5219 class Seq { 5220 explicit Seq(unsigned N) : Index(N) {} 5221 unsigned Index; 5222 friend class SequenceTree; 5223 public: 5224 Seq() : Index(0) {} 5225 }; 5226 5227 SequenceTree() { Values.push_back(Value(0)); } 5228 Seq root() const { return Seq(0); } 5229 5230 /// \brief Create a new sequence of operations, which is an unsequenced 5231 /// subset of \p Parent. This sequence of operations is sequenced with 5232 /// respect to other children of \p Parent. 5233 Seq allocate(Seq Parent) { 5234 Values.push_back(Value(Parent.Index)); 5235 return Seq(Values.size() - 1); 5236 } 5237 5238 /// \brief Merge a sequence of operations into its parent. 5239 void merge(Seq S) { 5240 Values[S.Index].Merged = true; 5241 } 5242 5243 /// \brief Determine whether two operations are unsequenced. This operation 5244 /// is asymmetric: \p Cur should be the more recent sequence, and \p Old 5245 /// should have been merged into its parent as appropriate. 5246 bool isUnsequenced(Seq Cur, Seq Old) { 5247 unsigned C = representative(Cur.Index); 5248 unsigned Target = representative(Old.Index); 5249 while (C >= Target) { 5250 if (C == Target) 5251 return true; 5252 C = Values[C].Parent; 5253 } 5254 return false; 5255 } 5256 5257 private: 5258 /// \brief Pick a representative for a sequence. 5259 unsigned representative(unsigned K) { 5260 if (Values[K].Merged) 5261 // Perform path compression as we go. 5262 return Values[K].Parent = representative(Values[K].Parent); 5263 return K; 5264 } 5265 }; 5266 5267 /// An object for which we can track unsequenced uses. 5268 typedef NamedDecl *Object; 5269 5270 /// Different flavors of object usage which we track. We only track the 5271 /// least-sequenced usage of each kind. 5272 enum UsageKind { 5273 /// A read of an object. Multiple unsequenced reads are OK. 5274 UK_Use, 5275 /// A modification of an object which is sequenced before the value 5276 /// computation of the expression, such as ++n. 5277 UK_ModAsValue, 5278 /// A modification of an object which is not sequenced before the value 5279 /// computation of the expression, such as n++. 5280 UK_ModAsSideEffect, 5281 5282 UK_Count = UK_ModAsSideEffect + 1 5283 }; 5284 5285 struct Usage { 5286 Usage() : Use(0), Seq() {} 5287 Expr *Use; 5288 SequenceTree::Seq Seq; 5289 }; 5290 5291 struct UsageInfo { 5292 UsageInfo() : Diagnosed(false) {} 5293 Usage Uses[UK_Count]; 5294 /// Have we issued a diagnostic for this variable already? 5295 bool Diagnosed; 5296 }; 5297 typedef llvm::SmallDenseMap<Object, UsageInfo, 16> UsageInfoMap; 5298 5299 Sema &SemaRef; 5300 /// Sequenced regions within the expression. 5301 SequenceTree Tree; 5302 /// Declaration modifications and references which we have seen. 5303 UsageInfoMap UsageMap; 5304 /// The region we are currently within. 5305 SequenceTree::Seq Region; 5306 /// Filled in with declarations which were modified as a side-effect 5307 /// (that is, post-increment operations). 5308 llvm::SmallVectorImpl<std::pair<Object, Usage> > *ModAsSideEffect; 5309 /// Expressions to check later. We defer checking these to reduce 5310 /// stack usage. 5311 llvm::SmallVectorImpl<Expr*> &WorkList; 5312 5313 /// RAII object wrapping the visitation of a sequenced subexpression of an 5314 /// expression. At the end of this process, the side-effects of the evaluation 5315 /// become sequenced with respect to the value computation of the result, so 5316 /// we downgrade any UK_ModAsSideEffect within the evaluation to 5317 /// UK_ModAsValue. 5318 struct SequencedSubexpression { 5319 SequencedSubexpression(SequenceChecker &Self) 5320 : Self(Self), OldModAsSideEffect(Self.ModAsSideEffect) { 5321 Self.ModAsSideEffect = &ModAsSideEffect; 5322 } 5323 ~SequencedSubexpression() { 5324 for (unsigned I = 0, E = ModAsSideEffect.size(); I != E; ++I) { 5325 UsageInfo &U = Self.UsageMap[ModAsSideEffect[I].first]; 5326 U.Uses[UK_ModAsSideEffect] = ModAsSideEffect[I].second; 5327 Self.addUsage(U, ModAsSideEffect[I].first, 5328 ModAsSideEffect[I].second.Use, UK_ModAsValue); 5329 } 5330 Self.ModAsSideEffect = OldModAsSideEffect; 5331 } 5332 5333 SequenceChecker &Self; 5334 llvm::SmallVector<std::pair<Object, Usage>, 4> ModAsSideEffect; 5335 llvm::SmallVectorImpl<std::pair<Object, Usage> > *OldModAsSideEffect; 5336 }; 5337 5338 /// \brief Find the object which is produced by the specified expression, 5339 /// if any. 5340 Object getObject(Expr *E, bool Mod) const { 5341 E = E->IgnoreParenCasts(); 5342 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 5343 if (Mod && (UO->getOpcode() == UO_PreInc || UO->getOpcode() == UO_PreDec)) 5344 return getObject(UO->getSubExpr(), Mod); 5345 } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5346 if (BO->getOpcode() == BO_Comma) 5347 return getObject(BO->getRHS(), Mod); 5348 if (Mod && BO->isAssignmentOp()) 5349 return getObject(BO->getLHS(), Mod); 5350 } else if (MemberExpr *ME = dyn_cast<MemberExpr>(E)) { 5351 // FIXME: Check for more interesting cases, like "x.n = ++x.n". 5352 if (isa<CXXThisExpr>(ME->getBase()->IgnoreParenCasts())) 5353 return ME->getMemberDecl(); 5354 } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5355 // FIXME: If this is a reference, map through to its value. 5356 return DRE->getDecl(); 5357 return 0; 5358 } 5359 5360 /// \brief Note that an object was modified or used by an expression. 5361 void addUsage(UsageInfo &UI, Object O, Expr *Ref, UsageKind UK) { 5362 Usage &U = UI.Uses[UK]; 5363 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) { 5364 if (UK == UK_ModAsSideEffect && ModAsSideEffect) 5365 ModAsSideEffect->push_back(std::make_pair(O, U)); 5366 U.Use = Ref; 5367 U.Seq = Region; 5368 } 5369 } 5370 /// \brief Check whether a modification or use conflicts with a prior usage. 5371 void checkUsage(Object O, UsageInfo &UI, Expr *Ref, UsageKind OtherKind, 5372 bool IsModMod) { 5373 if (UI.Diagnosed) 5374 return; 5375 5376 const Usage &U = UI.Uses[OtherKind]; 5377 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) 5378 return; 5379 5380 Expr *Mod = U.Use; 5381 Expr *ModOrUse = Ref; 5382 if (OtherKind == UK_Use) 5383 std::swap(Mod, ModOrUse); 5384 5385 SemaRef.Diag(Mod->getExprLoc(), 5386 IsModMod ? diag::warn_unsequenced_mod_mod 5387 : diag::warn_unsequenced_mod_use) 5388 << O << SourceRange(ModOrUse->getExprLoc()); 5389 UI.Diagnosed = true; 5390 } 5391 5392 void notePreUse(Object O, Expr *Use) { 5393 UsageInfo &U = UsageMap[O]; 5394 // Uses conflict with other modifications. 5395 checkUsage(O, U, Use, UK_ModAsValue, false); 5396 } 5397 void notePostUse(Object O, Expr *Use) { 5398 UsageInfo &U = UsageMap[O]; 5399 checkUsage(O, U, Use, UK_ModAsSideEffect, false); 5400 addUsage(U, O, Use, UK_Use); 5401 } 5402 5403 void notePreMod(Object O, Expr *Mod) { 5404 UsageInfo &U = UsageMap[O]; 5405 // Modifications conflict with other modifications and with uses. 5406 checkUsage(O, U, Mod, UK_ModAsValue, true); 5407 checkUsage(O, U, Mod, UK_Use, false); 5408 } 5409 void notePostMod(Object O, Expr *Use, UsageKind UK) { 5410 UsageInfo &U = UsageMap[O]; 5411 checkUsage(O, U, Use, UK_ModAsSideEffect, true); 5412 addUsage(U, O, Use, UK); 5413 } 5414 5415public: 5416 SequenceChecker(Sema &S, Expr *E, 5417 llvm::SmallVectorImpl<Expr*> &WorkList) 5418 : EvaluatedExprVisitor<SequenceChecker>(S.Context), SemaRef(S), 5419 Region(Tree.root()), ModAsSideEffect(0), WorkList(WorkList) { 5420 Visit(E); 5421 } 5422 5423 void VisitStmt(Stmt *S) { 5424 // Skip all statements which aren't expressions for now. 5425 } 5426 5427 void VisitExpr(Expr *E) { 5428 // By default, just recurse to evaluated subexpressions. 5429 EvaluatedExprVisitor<SequenceChecker>::VisitStmt(E); 5430 } 5431 5432 void VisitCastExpr(CastExpr *E) { 5433 Object O = Object(); 5434 if (E->getCastKind() == CK_LValueToRValue) 5435 O = getObject(E->getSubExpr(), false); 5436 5437 if (O) 5438 notePreUse(O, E); 5439 VisitExpr(E); 5440 if (O) 5441 notePostUse(O, E); 5442 } 5443 5444 void VisitBinComma(BinaryOperator *BO) { 5445 // C++11 [expr.comma]p1: 5446 // Every value computation and side effect associated with the left 5447 // expression is sequenced before every value computation and side 5448 // effect associated with the right expression. 5449 SequenceTree::Seq LHS = Tree.allocate(Region); 5450 SequenceTree::Seq RHS = Tree.allocate(Region); 5451 SequenceTree::Seq OldRegion = Region; 5452 5453 { 5454 SequencedSubexpression SeqLHS(*this); 5455 Region = LHS; 5456 Visit(BO->getLHS()); 5457 } 5458 5459 Region = RHS; 5460 Visit(BO->getRHS()); 5461 5462 Region = OldRegion; 5463 5464 // Forget that LHS and RHS are sequenced. They are both unsequenced 5465 // with respect to other stuff. 5466 Tree.merge(LHS); 5467 Tree.merge(RHS); 5468 } 5469 5470 void VisitBinAssign(BinaryOperator *BO) { 5471 // The modification is sequenced after the value computation of the LHS 5472 // and RHS, so check it before inspecting the operands and update the 5473 // map afterwards. 5474 Object O = getObject(BO->getLHS(), true); 5475 if (!O) 5476 return VisitExpr(BO); 5477 5478 notePreMod(O, BO); 5479 5480 // C++11 [expr.ass]p7: 5481 // E1 op= E2 is equivalent to E1 = E1 op E2, except that E1 is evaluated 5482 // only once. 5483 // 5484 // Therefore, for a compound assignment operator, O is considered used 5485 // everywhere except within the evaluation of E1 itself. 5486 if (isa<CompoundAssignOperator>(BO)) 5487 notePreUse(O, BO); 5488 5489 Visit(BO->getLHS()); 5490 5491 if (isa<CompoundAssignOperator>(BO)) 5492 notePostUse(O, BO); 5493 5494 Visit(BO->getRHS()); 5495 5496 notePostMod(O, BO, UK_ModAsValue); 5497 } 5498 void VisitCompoundAssignOperator(CompoundAssignOperator *CAO) { 5499 VisitBinAssign(CAO); 5500 } 5501 5502 void VisitUnaryPreInc(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 5503 void VisitUnaryPreDec(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 5504 void VisitUnaryPreIncDec(UnaryOperator *UO) { 5505 Object O = getObject(UO->getSubExpr(), true); 5506 if (!O) 5507 return VisitExpr(UO); 5508 5509 notePreMod(O, UO); 5510 Visit(UO->getSubExpr()); 5511 notePostMod(O, UO, UK_ModAsValue); 5512 } 5513 5514 void VisitUnaryPostInc(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 5515 void VisitUnaryPostDec(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 5516 void VisitUnaryPostIncDec(UnaryOperator *UO) { 5517 Object O = getObject(UO->getSubExpr(), true); 5518 if (!O) 5519 return VisitExpr(UO); 5520 5521 notePreMod(O, UO); 5522 Visit(UO->getSubExpr()); 5523 notePostMod(O, UO, UK_ModAsSideEffect); 5524 } 5525 5526 /// Don't visit the RHS of '&&' or '||' if it might not be evaluated. 5527 void VisitBinLOr(BinaryOperator *BO) { 5528 // The side-effects of the LHS of an '&&' are sequenced before the 5529 // value computation of the RHS, and hence before the value computation 5530 // of the '&&' itself, unless the LHS evaluates to zero. We treat them 5531 // as if they were unconditionally sequenced. 5532 { 5533 SequencedSubexpression Sequenced(*this); 5534 Visit(BO->getLHS()); 5535 } 5536 5537 bool Result; 5538 if (!BO->getLHS()->isValueDependent() && 5539 BO->getLHS()->EvaluateAsBooleanCondition(Result, SemaRef.Context)) { 5540 if (!Result) 5541 Visit(BO->getRHS()); 5542 } else { 5543 // Check for unsequenced operations in the RHS, treating it as an 5544 // entirely separate evaluation. 5545 // 5546 // FIXME: If there are operations in the RHS which are unsequenced 5547 // with respect to operations outside the RHS, and those operations 5548 // are unconditionally evaluated, diagnose them. 5549 WorkList.push_back(BO->getRHS()); 5550 } 5551 } 5552 void VisitBinLAnd(BinaryOperator *BO) { 5553 { 5554 SequencedSubexpression Sequenced(*this); 5555 Visit(BO->getLHS()); 5556 } 5557 5558 bool Result; 5559 if (!BO->getLHS()->isValueDependent() && 5560 BO->getLHS()->EvaluateAsBooleanCondition(Result, SemaRef.Context)) { 5561 if (Result) 5562 Visit(BO->getRHS()); 5563 } else { 5564 WorkList.push_back(BO->getRHS()); 5565 } 5566 } 5567 5568 // Only visit the condition, unless we can be sure which subexpression will 5569 // be chosen. 5570 void VisitAbstractConditionalOperator(AbstractConditionalOperator *CO) { 5571 SequencedSubexpression Sequenced(*this); 5572 Visit(CO->getCond()); 5573 5574 bool Result; 5575 if (!CO->getCond()->isValueDependent() && 5576 CO->getCond()->EvaluateAsBooleanCondition(Result, SemaRef.Context)) 5577 Visit(Result ? CO->getTrueExpr() : CO->getFalseExpr()); 5578 else { 5579 WorkList.push_back(CO->getTrueExpr()); 5580 WorkList.push_back(CO->getFalseExpr()); 5581 } 5582 } 5583 5584 void VisitCXXConstructExpr(CXXConstructExpr *CCE) { 5585 if (!CCE->isListInitialization()) 5586 return VisitExpr(CCE); 5587 5588 // In C++11, list initializations are sequenced. 5589 llvm::SmallVector<SequenceTree::Seq, 32> Elts; 5590 SequenceTree::Seq Parent = Region; 5591 for (CXXConstructExpr::arg_iterator I = CCE->arg_begin(), 5592 E = CCE->arg_end(); 5593 I != E; ++I) { 5594 Region = Tree.allocate(Parent); 5595 Elts.push_back(Region); 5596 Visit(*I); 5597 } 5598 5599 // Forget that the initializers are sequenced. 5600 Region = Parent; 5601 for (unsigned I = 0; I < Elts.size(); ++I) 5602 Tree.merge(Elts[I]); 5603 } 5604 5605 void VisitInitListExpr(InitListExpr *ILE) { 5606 if (!SemaRef.getLangOpts().CPlusPlus11) 5607 return VisitExpr(ILE); 5608 5609 // In C++11, list initializations are sequenced. 5610 llvm::SmallVector<SequenceTree::Seq, 32> Elts; 5611 SequenceTree::Seq Parent = Region; 5612 for (unsigned I = 0; I < ILE->getNumInits(); ++I) { 5613 Expr *E = ILE->getInit(I); 5614 if (!E) continue; 5615 Region = Tree.allocate(Parent); 5616 Elts.push_back(Region); 5617 Visit(E); 5618 } 5619 5620 // Forget that the initializers are sequenced. 5621 Region = Parent; 5622 for (unsigned I = 0; I < Elts.size(); ++I) 5623 Tree.merge(Elts[I]); 5624 } 5625}; 5626} 5627 5628void Sema::CheckUnsequencedOperations(Expr *E) { 5629 llvm::SmallVector<Expr*, 8> WorkList; 5630 WorkList.push_back(E); 5631 while (!WorkList.empty()) { 5632 Expr *Item = WorkList.back(); 5633 WorkList.pop_back(); 5634 SequenceChecker(*this, Item, WorkList); 5635 } 5636} 5637 5638void Sema::CheckCompletedExpr(Expr *E, SourceLocation CheckLoc, 5639 bool IsConstexpr) { 5640 CheckImplicitConversions(E, CheckLoc); 5641 CheckUnsequencedOperations(E); 5642 if (!IsConstexpr && !E->isValueDependent()) 5643 CheckForIntOverflow(E); 5644} 5645 5646void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 5647 FieldDecl *BitField, 5648 Expr *Init) { 5649 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 5650} 5651 5652/// CheckParmsForFunctionDef - Check that the parameters of the given 5653/// function are appropriate for the definition of a function. This 5654/// takes care of any checks that cannot be performed on the 5655/// declaration itself, e.g., that the types of each of the function 5656/// parameters are complete. 5657bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 5658 bool CheckParameterNames) { 5659 bool HasInvalidParm = false; 5660 for (; P != PEnd; ++P) { 5661 ParmVarDecl *Param = *P; 5662 5663 // C99 6.7.5.3p4: the parameters in a parameter type list in a 5664 // function declarator that is part of a function definition of 5665 // that function shall not have incomplete type. 5666 // 5667 // This is also C++ [dcl.fct]p6. 5668 if (!Param->isInvalidDecl() && 5669 RequireCompleteType(Param->getLocation(), Param->getType(), 5670 diag::err_typecheck_decl_incomplete_type)) { 5671 Param->setInvalidDecl(); 5672 HasInvalidParm = true; 5673 } 5674 5675 // C99 6.9.1p5: If the declarator includes a parameter type list, the 5676 // declaration of each parameter shall include an identifier. 5677 if (CheckParameterNames && 5678 Param->getIdentifier() == 0 && 5679 !Param->isImplicit() && 5680 !getLangOpts().CPlusPlus) 5681 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 5682 5683 // C99 6.7.5.3p12: 5684 // If the function declarator is not part of a definition of that 5685 // function, parameters may have incomplete type and may use the [*] 5686 // notation in their sequences of declarator specifiers to specify 5687 // variable length array types. 5688 QualType PType = Param->getOriginalType(); 5689 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 5690 if (AT->getSizeModifier() == ArrayType::Star) { 5691 // FIXME: This diagnosic should point the '[*]' if source-location 5692 // information is added for it. 5693 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 5694 } 5695 } 5696 } 5697 5698 return HasInvalidParm; 5699} 5700 5701/// CheckCastAlign - Implements -Wcast-align, which warns when a 5702/// pointer cast increases the alignment requirements. 5703void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 5704 // This is actually a lot of work to potentially be doing on every 5705 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 5706 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 5707 TRange.getBegin()) 5708 == DiagnosticsEngine::Ignored) 5709 return; 5710 5711 // Ignore dependent types. 5712 if (T->isDependentType() || Op->getType()->isDependentType()) 5713 return; 5714 5715 // Require that the destination be a pointer type. 5716 const PointerType *DestPtr = T->getAs<PointerType>(); 5717 if (!DestPtr) return; 5718 5719 // If the destination has alignment 1, we're done. 5720 QualType DestPointee = DestPtr->getPointeeType(); 5721 if (DestPointee->isIncompleteType()) return; 5722 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 5723 if (DestAlign.isOne()) return; 5724 5725 // Require that the source be a pointer type. 5726 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 5727 if (!SrcPtr) return; 5728 QualType SrcPointee = SrcPtr->getPointeeType(); 5729 5730 // Whitelist casts from cv void*. We already implicitly 5731 // whitelisted casts to cv void*, since they have alignment 1. 5732 // Also whitelist casts involving incomplete types, which implicitly 5733 // includes 'void'. 5734 if (SrcPointee->isIncompleteType()) return; 5735 5736 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 5737 if (SrcAlign >= DestAlign) return; 5738 5739 Diag(TRange.getBegin(), diag::warn_cast_align) 5740 << Op->getType() << T 5741 << static_cast<unsigned>(SrcAlign.getQuantity()) 5742 << static_cast<unsigned>(DestAlign.getQuantity()) 5743 << TRange << Op->getSourceRange(); 5744} 5745 5746static const Type* getElementType(const Expr *BaseExpr) { 5747 const Type* EltType = BaseExpr->getType().getTypePtr(); 5748 if (EltType->isAnyPointerType()) 5749 return EltType->getPointeeType().getTypePtr(); 5750 else if (EltType->isArrayType()) 5751 return EltType->getBaseElementTypeUnsafe(); 5752 return EltType; 5753} 5754 5755/// \brief Check whether this array fits the idiom of a size-one tail padded 5756/// array member of a struct. 5757/// 5758/// We avoid emitting out-of-bounds access warnings for such arrays as they are 5759/// commonly used to emulate flexible arrays in C89 code. 5760static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 5761 const NamedDecl *ND) { 5762 if (Size != 1 || !ND) return false; 5763 5764 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 5765 if (!FD) return false; 5766 5767 // Don't consider sizes resulting from macro expansions or template argument 5768 // substitution to form C89 tail-padded arrays. 5769 5770 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 5771 while (TInfo) { 5772 TypeLoc TL = TInfo->getTypeLoc(); 5773 // Look through typedefs. 5774 const TypedefTypeLoc *TTL = dyn_cast<TypedefTypeLoc>(&TL); 5775 if (TTL) { 5776 const TypedefNameDecl *TDL = TTL->getTypedefNameDecl(); 5777 TInfo = TDL->getTypeSourceInfo(); 5778 continue; 5779 } 5780 if (const ConstantArrayTypeLoc *CTL = dyn_cast<ConstantArrayTypeLoc>(&TL)) { 5781 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL->getSizeExpr()); 5782 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 5783 return false; 5784 } 5785 break; 5786 } 5787 5788 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 5789 if (!RD) return false; 5790 if (RD->isUnion()) return false; 5791 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 5792 if (!CRD->isStandardLayout()) return false; 5793 } 5794 5795 // See if this is the last field decl in the record. 5796 const Decl *D = FD; 5797 while ((D = D->getNextDeclInContext())) 5798 if (isa<FieldDecl>(D)) 5799 return false; 5800 return true; 5801} 5802 5803void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 5804 const ArraySubscriptExpr *ASE, 5805 bool AllowOnePastEnd, bool IndexNegated) { 5806 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 5807 if (IndexExpr->isValueDependent()) 5808 return; 5809 5810 const Type *EffectiveType = getElementType(BaseExpr); 5811 BaseExpr = BaseExpr->IgnoreParenCasts(); 5812 const ConstantArrayType *ArrayTy = 5813 Context.getAsConstantArrayType(BaseExpr->getType()); 5814 if (!ArrayTy) 5815 return; 5816 5817 llvm::APSInt index; 5818 if (!IndexExpr->EvaluateAsInt(index, Context)) 5819 return; 5820 if (IndexNegated) 5821 index = -index; 5822 5823 const NamedDecl *ND = NULL; 5824 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 5825 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 5826 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 5827 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 5828 5829 if (index.isUnsigned() || !index.isNegative()) { 5830 llvm::APInt size = ArrayTy->getSize(); 5831 if (!size.isStrictlyPositive()) 5832 return; 5833 5834 const Type* BaseType = getElementType(BaseExpr); 5835 if (BaseType != EffectiveType) { 5836 // Make sure we're comparing apples to apples when comparing index to size 5837 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 5838 uint64_t array_typesize = Context.getTypeSize(BaseType); 5839 // Handle ptrarith_typesize being zero, such as when casting to void* 5840 if (!ptrarith_typesize) ptrarith_typesize = 1; 5841 if (ptrarith_typesize != array_typesize) { 5842 // There's a cast to a different size type involved 5843 uint64_t ratio = array_typesize / ptrarith_typesize; 5844 // TODO: Be smarter about handling cases where array_typesize is not a 5845 // multiple of ptrarith_typesize 5846 if (ptrarith_typesize * ratio == array_typesize) 5847 size *= llvm::APInt(size.getBitWidth(), ratio); 5848 } 5849 } 5850 5851 if (size.getBitWidth() > index.getBitWidth()) 5852 index = index.zext(size.getBitWidth()); 5853 else if (size.getBitWidth() < index.getBitWidth()) 5854 size = size.zext(index.getBitWidth()); 5855 5856 // For array subscripting the index must be less than size, but for pointer 5857 // arithmetic also allow the index (offset) to be equal to size since 5858 // computing the next address after the end of the array is legal and 5859 // commonly done e.g. in C++ iterators and range-based for loops. 5860 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 5861 return; 5862 5863 // Also don't warn for arrays of size 1 which are members of some 5864 // structure. These are often used to approximate flexible arrays in C89 5865 // code. 5866 if (IsTailPaddedMemberArray(*this, size, ND)) 5867 return; 5868 5869 // Suppress the warning if the subscript expression (as identified by the 5870 // ']' location) and the index expression are both from macro expansions 5871 // within a system header. 5872 if (ASE) { 5873 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 5874 ASE->getRBracketLoc()); 5875 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 5876 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 5877 IndexExpr->getLocStart()); 5878 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 5879 return; 5880 } 5881 } 5882 5883 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 5884 if (ASE) 5885 DiagID = diag::warn_array_index_exceeds_bounds; 5886 5887 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 5888 PDiag(DiagID) << index.toString(10, true) 5889 << size.toString(10, true) 5890 << (unsigned)size.getLimitedValue(~0U) 5891 << IndexExpr->getSourceRange()); 5892 } else { 5893 unsigned DiagID = diag::warn_array_index_precedes_bounds; 5894 if (!ASE) { 5895 DiagID = diag::warn_ptr_arith_precedes_bounds; 5896 if (index.isNegative()) index = -index; 5897 } 5898 5899 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 5900 PDiag(DiagID) << index.toString(10, true) 5901 << IndexExpr->getSourceRange()); 5902 } 5903 5904 if (!ND) { 5905 // Try harder to find a NamedDecl to point at in the note. 5906 while (const ArraySubscriptExpr *ASE = 5907 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 5908 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 5909 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 5910 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 5911 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 5912 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 5913 } 5914 5915 if (ND) 5916 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 5917 PDiag(diag::note_array_index_out_of_bounds) 5918 << ND->getDeclName()); 5919} 5920 5921void Sema::CheckArrayAccess(const Expr *expr) { 5922 int AllowOnePastEnd = 0; 5923 while (expr) { 5924 expr = expr->IgnoreParenImpCasts(); 5925 switch (expr->getStmtClass()) { 5926 case Stmt::ArraySubscriptExprClass: { 5927 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 5928 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 5929 AllowOnePastEnd > 0); 5930 return; 5931 } 5932 case Stmt::UnaryOperatorClass: { 5933 // Only unwrap the * and & unary operators 5934 const UnaryOperator *UO = cast<UnaryOperator>(expr); 5935 expr = UO->getSubExpr(); 5936 switch (UO->getOpcode()) { 5937 case UO_AddrOf: 5938 AllowOnePastEnd++; 5939 break; 5940 case UO_Deref: 5941 AllowOnePastEnd--; 5942 break; 5943 default: 5944 return; 5945 } 5946 break; 5947 } 5948 case Stmt::ConditionalOperatorClass: { 5949 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 5950 if (const Expr *lhs = cond->getLHS()) 5951 CheckArrayAccess(lhs); 5952 if (const Expr *rhs = cond->getRHS()) 5953 CheckArrayAccess(rhs); 5954 return; 5955 } 5956 default: 5957 return; 5958 } 5959 } 5960} 5961 5962//===--- CHECK: Objective-C retain cycles ----------------------------------// 5963 5964namespace { 5965 struct RetainCycleOwner { 5966 RetainCycleOwner() : Variable(0), Indirect(false) {} 5967 VarDecl *Variable; 5968 SourceRange Range; 5969 SourceLocation Loc; 5970 bool Indirect; 5971 5972 void setLocsFrom(Expr *e) { 5973 Loc = e->getExprLoc(); 5974 Range = e->getSourceRange(); 5975 } 5976 }; 5977} 5978 5979/// Consider whether capturing the given variable can possibly lead to 5980/// a retain cycle. 5981static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 5982 // In ARC, it's captured strongly iff the variable has __strong 5983 // lifetime. In MRR, it's captured strongly if the variable is 5984 // __block and has an appropriate type. 5985 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 5986 return false; 5987 5988 owner.Variable = var; 5989 if (ref) 5990 owner.setLocsFrom(ref); 5991 return true; 5992} 5993 5994static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 5995 while (true) { 5996 e = e->IgnoreParens(); 5997 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 5998 switch (cast->getCastKind()) { 5999 case CK_BitCast: 6000 case CK_LValueBitCast: 6001 case CK_LValueToRValue: 6002 case CK_ARCReclaimReturnedObject: 6003 e = cast->getSubExpr(); 6004 continue; 6005 6006 default: 6007 return false; 6008 } 6009 } 6010 6011 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 6012 ObjCIvarDecl *ivar = ref->getDecl(); 6013 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 6014 return false; 6015 6016 // Try to find a retain cycle in the base. 6017 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 6018 return false; 6019 6020 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 6021 owner.Indirect = true; 6022 return true; 6023 } 6024 6025 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 6026 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 6027 if (!var) return false; 6028 return considerVariable(var, ref, owner); 6029 } 6030 6031 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 6032 if (member->isArrow()) return false; 6033 6034 // Don't count this as an indirect ownership. 6035 e = member->getBase(); 6036 continue; 6037 } 6038 6039 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 6040 // Only pay attention to pseudo-objects on property references. 6041 ObjCPropertyRefExpr *pre 6042 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 6043 ->IgnoreParens()); 6044 if (!pre) return false; 6045 if (pre->isImplicitProperty()) return false; 6046 ObjCPropertyDecl *property = pre->getExplicitProperty(); 6047 if (!property->isRetaining() && 6048 !(property->getPropertyIvarDecl() && 6049 property->getPropertyIvarDecl()->getType() 6050 .getObjCLifetime() == Qualifiers::OCL_Strong)) 6051 return false; 6052 6053 owner.Indirect = true; 6054 if (pre->isSuperReceiver()) { 6055 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 6056 if (!owner.Variable) 6057 return false; 6058 owner.Loc = pre->getLocation(); 6059 owner.Range = pre->getSourceRange(); 6060 return true; 6061 } 6062 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 6063 ->getSourceExpr()); 6064 continue; 6065 } 6066 6067 // Array ivars? 6068 6069 return false; 6070 } 6071} 6072 6073namespace { 6074 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 6075 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 6076 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 6077 Variable(variable), Capturer(0) {} 6078 6079 VarDecl *Variable; 6080 Expr *Capturer; 6081 6082 void VisitDeclRefExpr(DeclRefExpr *ref) { 6083 if (ref->getDecl() == Variable && !Capturer) 6084 Capturer = ref; 6085 } 6086 6087 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 6088 if (Capturer) return; 6089 Visit(ref->getBase()); 6090 if (Capturer && ref->isFreeIvar()) 6091 Capturer = ref; 6092 } 6093 6094 void VisitBlockExpr(BlockExpr *block) { 6095 // Look inside nested blocks 6096 if (block->getBlockDecl()->capturesVariable(Variable)) 6097 Visit(block->getBlockDecl()->getBody()); 6098 } 6099 6100 void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) { 6101 if (Capturer) return; 6102 if (OVE->getSourceExpr()) 6103 Visit(OVE->getSourceExpr()); 6104 } 6105 }; 6106} 6107 6108/// Check whether the given argument is a block which captures a 6109/// variable. 6110static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 6111 assert(owner.Variable && owner.Loc.isValid()); 6112 6113 e = e->IgnoreParenCasts(); 6114 6115 // Look through [^{...} copy] and Block_copy(^{...}). 6116 if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) { 6117 Selector Cmd = ME->getSelector(); 6118 if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") { 6119 e = ME->getInstanceReceiver(); 6120 if (!e) 6121 return 0; 6122 e = e->IgnoreParenCasts(); 6123 } 6124 } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) { 6125 if (CE->getNumArgs() == 1) { 6126 FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl()); 6127 if (Fn) { 6128 const IdentifierInfo *FnI = Fn->getIdentifier(); 6129 if (FnI && FnI->isStr("_Block_copy")) { 6130 e = CE->getArg(0)->IgnoreParenCasts(); 6131 } 6132 } 6133 } 6134 } 6135 6136 BlockExpr *block = dyn_cast<BlockExpr>(e); 6137 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 6138 return 0; 6139 6140 FindCaptureVisitor visitor(S.Context, owner.Variable); 6141 visitor.Visit(block->getBlockDecl()->getBody()); 6142 return visitor.Capturer; 6143} 6144 6145static void diagnoseRetainCycle(Sema &S, Expr *capturer, 6146 RetainCycleOwner &owner) { 6147 assert(capturer); 6148 assert(owner.Variable && owner.Loc.isValid()); 6149 6150 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 6151 << owner.Variable << capturer->getSourceRange(); 6152 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 6153 << owner.Indirect << owner.Range; 6154} 6155 6156/// Check for a keyword selector that starts with the word 'add' or 6157/// 'set'. 6158static bool isSetterLikeSelector(Selector sel) { 6159 if (sel.isUnarySelector()) return false; 6160 6161 StringRef str = sel.getNameForSlot(0); 6162 while (!str.empty() && str.front() == '_') str = str.substr(1); 6163 if (str.startswith("set")) 6164 str = str.substr(3); 6165 else if (str.startswith("add")) { 6166 // Specially whitelist 'addOperationWithBlock:'. 6167 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 6168 return false; 6169 str = str.substr(3); 6170 } 6171 else 6172 return false; 6173 6174 if (str.empty()) return true; 6175 return !islower(str.front()); 6176} 6177 6178/// Check a message send to see if it's likely to cause a retain cycle. 6179void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 6180 // Only check instance methods whose selector looks like a setter. 6181 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 6182 return; 6183 6184 // Try to find a variable that the receiver is strongly owned by. 6185 RetainCycleOwner owner; 6186 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 6187 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 6188 return; 6189 } else { 6190 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 6191 owner.Variable = getCurMethodDecl()->getSelfDecl(); 6192 owner.Loc = msg->getSuperLoc(); 6193 owner.Range = msg->getSuperLoc(); 6194 } 6195 6196 // Check whether the receiver is captured by any of the arguments. 6197 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 6198 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 6199 return diagnoseRetainCycle(*this, capturer, owner); 6200} 6201 6202/// Check a property assign to see if it's likely to cause a retain cycle. 6203void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 6204 RetainCycleOwner owner; 6205 if (!findRetainCycleOwner(*this, receiver, owner)) 6206 return; 6207 6208 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 6209 diagnoseRetainCycle(*this, capturer, owner); 6210} 6211 6212void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) { 6213 RetainCycleOwner Owner; 6214 if (!considerVariable(Var, /*DeclRefExpr=*/0, Owner)) 6215 return; 6216 6217 // Because we don't have an expression for the variable, we have to set the 6218 // location explicitly here. 6219 Owner.Loc = Var->getLocation(); 6220 Owner.Range = Var->getSourceRange(); 6221 6222 if (Expr *Capturer = findCapturingExpr(*this, Init, Owner)) 6223 diagnoseRetainCycle(*this, Capturer, Owner); 6224} 6225 6226static bool checkUnsafeAssignLiteral(Sema &S, SourceLocation Loc, 6227 Expr *RHS, bool isProperty) { 6228 // Check if RHS is an Objective-C object literal, which also can get 6229 // immediately zapped in a weak reference. Note that we explicitly 6230 // allow ObjCStringLiterals, since those are designed to never really die. 6231 RHS = RHS->IgnoreParenImpCasts(); 6232 6233 // This enum needs to match with the 'select' in 6234 // warn_objc_arc_literal_assign (off-by-1). 6235 Sema::ObjCLiteralKind Kind = S.CheckLiteralKind(RHS); 6236 if (Kind == Sema::LK_String || Kind == Sema::LK_None) 6237 return false; 6238 6239 S.Diag(Loc, diag::warn_arc_literal_assign) 6240 << (unsigned) Kind 6241 << (isProperty ? 0 : 1) 6242 << RHS->getSourceRange(); 6243 6244 return true; 6245} 6246 6247static bool checkUnsafeAssignObject(Sema &S, SourceLocation Loc, 6248 Qualifiers::ObjCLifetime LT, 6249 Expr *RHS, bool isProperty) { 6250 // Strip off any implicit cast added to get to the one ARC-specific. 6251 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6252 if (cast->getCastKind() == CK_ARCConsumeObject) { 6253 S.Diag(Loc, diag::warn_arc_retained_assign) 6254 << (LT == Qualifiers::OCL_ExplicitNone) 6255 << (isProperty ? 0 : 1) 6256 << RHS->getSourceRange(); 6257 return true; 6258 } 6259 RHS = cast->getSubExpr(); 6260 } 6261 6262 if (LT == Qualifiers::OCL_Weak && 6263 checkUnsafeAssignLiteral(S, Loc, RHS, isProperty)) 6264 return true; 6265 6266 return false; 6267} 6268 6269bool Sema::checkUnsafeAssigns(SourceLocation Loc, 6270 QualType LHS, Expr *RHS) { 6271 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 6272 6273 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 6274 return false; 6275 6276 if (checkUnsafeAssignObject(*this, Loc, LT, RHS, false)) 6277 return true; 6278 6279 return false; 6280} 6281 6282void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 6283 Expr *LHS, Expr *RHS) { 6284 QualType LHSType; 6285 // PropertyRef on LHS type need be directly obtained from 6286 // its declaration as it has a PsuedoType. 6287 ObjCPropertyRefExpr *PRE 6288 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 6289 if (PRE && !PRE->isImplicitProperty()) { 6290 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6291 if (PD) 6292 LHSType = PD->getType(); 6293 } 6294 6295 if (LHSType.isNull()) 6296 LHSType = LHS->getType(); 6297 6298 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 6299 6300 if (LT == Qualifiers::OCL_Weak) { 6301 DiagnosticsEngine::Level Level = 6302 Diags.getDiagnosticLevel(diag::warn_arc_repeated_use_of_weak, Loc); 6303 if (Level != DiagnosticsEngine::Ignored) 6304 getCurFunction()->markSafeWeakUse(LHS); 6305 } 6306 6307 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 6308 return; 6309 6310 // FIXME. Check for other life times. 6311 if (LT != Qualifiers::OCL_None) 6312 return; 6313 6314 if (PRE) { 6315 if (PRE->isImplicitProperty()) 6316 return; 6317 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6318 if (!PD) 6319 return; 6320 6321 unsigned Attributes = PD->getPropertyAttributes(); 6322 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 6323 // when 'assign' attribute was not explicitly specified 6324 // by user, ignore it and rely on property type itself 6325 // for lifetime info. 6326 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 6327 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 6328 LHSType->isObjCRetainableType()) 6329 return; 6330 6331 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6332 if (cast->getCastKind() == CK_ARCConsumeObject) { 6333 Diag(Loc, diag::warn_arc_retained_property_assign) 6334 << RHS->getSourceRange(); 6335 return; 6336 } 6337 RHS = cast->getSubExpr(); 6338 } 6339 } 6340 else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) { 6341 if (checkUnsafeAssignObject(*this, Loc, Qualifiers::OCL_Weak, RHS, true)) 6342 return; 6343 } 6344 } 6345} 6346 6347//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 6348 6349namespace { 6350bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 6351 SourceLocation StmtLoc, 6352 const NullStmt *Body) { 6353 // Do not warn if the body is a macro that expands to nothing, e.g: 6354 // 6355 // #define CALL(x) 6356 // if (condition) 6357 // CALL(0); 6358 // 6359 if (Body->hasLeadingEmptyMacro()) 6360 return false; 6361 6362 // Get line numbers of statement and body. 6363 bool StmtLineInvalid; 6364 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 6365 &StmtLineInvalid); 6366 if (StmtLineInvalid) 6367 return false; 6368 6369 bool BodyLineInvalid; 6370 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 6371 &BodyLineInvalid); 6372 if (BodyLineInvalid) 6373 return false; 6374 6375 // Warn if null statement and body are on the same line. 6376 if (StmtLine != BodyLine) 6377 return false; 6378 6379 return true; 6380} 6381} // Unnamed namespace 6382 6383void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 6384 const Stmt *Body, 6385 unsigned DiagID) { 6386 // Since this is a syntactic check, don't emit diagnostic for template 6387 // instantiations, this just adds noise. 6388 if (CurrentInstantiationScope) 6389 return; 6390 6391 // The body should be a null statement. 6392 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6393 if (!NBody) 6394 return; 6395 6396 // Do the usual checks. 6397 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6398 return; 6399 6400 Diag(NBody->getSemiLoc(), DiagID); 6401 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6402} 6403 6404void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 6405 const Stmt *PossibleBody) { 6406 assert(!CurrentInstantiationScope); // Ensured by caller 6407 6408 SourceLocation StmtLoc; 6409 const Stmt *Body; 6410 unsigned DiagID; 6411 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 6412 StmtLoc = FS->getRParenLoc(); 6413 Body = FS->getBody(); 6414 DiagID = diag::warn_empty_for_body; 6415 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 6416 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 6417 Body = WS->getBody(); 6418 DiagID = diag::warn_empty_while_body; 6419 } else 6420 return; // Neither `for' nor `while'. 6421 6422 // The body should be a null statement. 6423 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6424 if (!NBody) 6425 return; 6426 6427 // Skip expensive checks if diagnostic is disabled. 6428 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 6429 DiagnosticsEngine::Ignored) 6430 return; 6431 6432 // Do the usual checks. 6433 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6434 return; 6435 6436 // `for(...);' and `while(...);' are popular idioms, so in order to keep 6437 // noise level low, emit diagnostics only if for/while is followed by a 6438 // CompoundStmt, e.g.: 6439 // for (int i = 0; i < n; i++); 6440 // { 6441 // a(i); 6442 // } 6443 // or if for/while is followed by a statement with more indentation 6444 // than for/while itself: 6445 // for (int i = 0; i < n; i++); 6446 // a(i); 6447 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 6448 if (!ProbableTypo) { 6449 bool BodyColInvalid; 6450 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 6451 PossibleBody->getLocStart(), 6452 &BodyColInvalid); 6453 if (BodyColInvalid) 6454 return; 6455 6456 bool StmtColInvalid; 6457 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 6458 S->getLocStart(), 6459 &StmtColInvalid); 6460 if (StmtColInvalid) 6461 return; 6462 6463 if (BodyCol > StmtCol) 6464 ProbableTypo = true; 6465 } 6466 6467 if (ProbableTypo) { 6468 Diag(NBody->getSemiLoc(), DiagID); 6469 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6470 } 6471} 6472 6473//===--- Layout compatibility ----------------------------------------------// 6474 6475namespace { 6476 6477bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2); 6478 6479/// \brief Check if two enumeration types are layout-compatible. 6480bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) { 6481 // C++11 [dcl.enum] p8: 6482 // Two enumeration types are layout-compatible if they have the same 6483 // underlying type. 6484 return ED1->isComplete() && ED2->isComplete() && 6485 C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType()); 6486} 6487 6488/// \brief Check if two fields are layout-compatible. 6489bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) { 6490 if (!isLayoutCompatible(C, Field1->getType(), Field2->getType())) 6491 return false; 6492 6493 if (Field1->isBitField() != Field2->isBitField()) 6494 return false; 6495 6496 if (Field1->isBitField()) { 6497 // Make sure that the bit-fields are the same length. 6498 unsigned Bits1 = Field1->getBitWidthValue(C); 6499 unsigned Bits2 = Field2->getBitWidthValue(C); 6500 6501 if (Bits1 != Bits2) 6502 return false; 6503 } 6504 6505 return true; 6506} 6507 6508/// \brief Check if two standard-layout structs are layout-compatible. 6509/// (C++11 [class.mem] p17) 6510bool isLayoutCompatibleStruct(ASTContext &C, 6511 RecordDecl *RD1, 6512 RecordDecl *RD2) { 6513 // If both records are C++ classes, check that base classes match. 6514 if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) { 6515 // If one of records is a CXXRecordDecl we are in C++ mode, 6516 // thus the other one is a CXXRecordDecl, too. 6517 const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2); 6518 // Check number of base classes. 6519 if (D1CXX->getNumBases() != D2CXX->getNumBases()) 6520 return false; 6521 6522 // Check the base classes. 6523 for (CXXRecordDecl::base_class_const_iterator 6524 Base1 = D1CXX->bases_begin(), 6525 BaseEnd1 = D1CXX->bases_end(), 6526 Base2 = D2CXX->bases_begin(); 6527 Base1 != BaseEnd1; 6528 ++Base1, ++Base2) { 6529 if (!isLayoutCompatible(C, Base1->getType(), Base2->getType())) 6530 return false; 6531 } 6532 } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) { 6533 // If only RD2 is a C++ class, it should have zero base classes. 6534 if (D2CXX->getNumBases() > 0) 6535 return false; 6536 } 6537 6538 // Check the fields. 6539 RecordDecl::field_iterator Field2 = RD2->field_begin(), 6540 Field2End = RD2->field_end(), 6541 Field1 = RD1->field_begin(), 6542 Field1End = RD1->field_end(); 6543 for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) { 6544 if (!isLayoutCompatible(C, *Field1, *Field2)) 6545 return false; 6546 } 6547 if (Field1 != Field1End || Field2 != Field2End) 6548 return false; 6549 6550 return true; 6551} 6552 6553/// \brief Check if two standard-layout unions are layout-compatible. 6554/// (C++11 [class.mem] p18) 6555bool isLayoutCompatibleUnion(ASTContext &C, 6556 RecordDecl *RD1, 6557 RecordDecl *RD2) { 6558 llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields; 6559 for (RecordDecl::field_iterator Field2 = RD2->field_begin(), 6560 Field2End = RD2->field_end(); 6561 Field2 != Field2End; ++Field2) { 6562 UnmatchedFields.insert(*Field2); 6563 } 6564 6565 for (RecordDecl::field_iterator Field1 = RD1->field_begin(), 6566 Field1End = RD1->field_end(); 6567 Field1 != Field1End; ++Field1) { 6568 llvm::SmallPtrSet<FieldDecl *, 8>::iterator 6569 I = UnmatchedFields.begin(), 6570 E = UnmatchedFields.end(); 6571 6572 for ( ; I != E; ++I) { 6573 if (isLayoutCompatible(C, *Field1, *I)) { 6574 bool Result = UnmatchedFields.erase(*I); 6575 (void) Result; 6576 assert(Result); 6577 break; 6578 } 6579 } 6580 if (I == E) 6581 return false; 6582 } 6583 6584 return UnmatchedFields.empty(); 6585} 6586 6587bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) { 6588 if (RD1->isUnion() != RD2->isUnion()) 6589 return false; 6590 6591 if (RD1->isUnion()) 6592 return isLayoutCompatibleUnion(C, RD1, RD2); 6593 else 6594 return isLayoutCompatibleStruct(C, RD1, RD2); 6595} 6596 6597/// \brief Check if two types are layout-compatible in C++11 sense. 6598bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) { 6599 if (T1.isNull() || T2.isNull()) 6600 return false; 6601 6602 // C++11 [basic.types] p11: 6603 // If two types T1 and T2 are the same type, then T1 and T2 are 6604 // layout-compatible types. 6605 if (C.hasSameType(T1, T2)) 6606 return true; 6607 6608 T1 = T1.getCanonicalType().getUnqualifiedType(); 6609 T2 = T2.getCanonicalType().getUnqualifiedType(); 6610 6611 const Type::TypeClass TC1 = T1->getTypeClass(); 6612 const Type::TypeClass TC2 = T2->getTypeClass(); 6613 6614 if (TC1 != TC2) 6615 return false; 6616 6617 if (TC1 == Type::Enum) { 6618 return isLayoutCompatible(C, 6619 cast<EnumType>(T1)->getDecl(), 6620 cast<EnumType>(T2)->getDecl()); 6621 } else if (TC1 == Type::Record) { 6622 if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType()) 6623 return false; 6624 6625 return isLayoutCompatible(C, 6626 cast<RecordType>(T1)->getDecl(), 6627 cast<RecordType>(T2)->getDecl()); 6628 } 6629 6630 return false; 6631} 6632} 6633 6634//===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----// 6635 6636namespace { 6637/// \brief Given a type tag expression find the type tag itself. 6638/// 6639/// \param TypeExpr Type tag expression, as it appears in user's code. 6640/// 6641/// \param VD Declaration of an identifier that appears in a type tag. 6642/// 6643/// \param MagicValue Type tag magic value. 6644bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx, 6645 const ValueDecl **VD, uint64_t *MagicValue) { 6646 while(true) { 6647 if (!TypeExpr) 6648 return false; 6649 6650 TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts(); 6651 6652 switch (TypeExpr->getStmtClass()) { 6653 case Stmt::UnaryOperatorClass: { 6654 const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr); 6655 if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) { 6656 TypeExpr = UO->getSubExpr(); 6657 continue; 6658 } 6659 return false; 6660 } 6661 6662 case Stmt::DeclRefExprClass: { 6663 const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr); 6664 *VD = DRE->getDecl(); 6665 return true; 6666 } 6667 6668 case Stmt::IntegerLiteralClass: { 6669 const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr); 6670 llvm::APInt MagicValueAPInt = IL->getValue(); 6671 if (MagicValueAPInt.getActiveBits() <= 64) { 6672 *MagicValue = MagicValueAPInt.getZExtValue(); 6673 return true; 6674 } else 6675 return false; 6676 } 6677 6678 case Stmt::BinaryConditionalOperatorClass: 6679 case Stmt::ConditionalOperatorClass: { 6680 const AbstractConditionalOperator *ACO = 6681 cast<AbstractConditionalOperator>(TypeExpr); 6682 bool Result; 6683 if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) { 6684 if (Result) 6685 TypeExpr = ACO->getTrueExpr(); 6686 else 6687 TypeExpr = ACO->getFalseExpr(); 6688 continue; 6689 } 6690 return false; 6691 } 6692 6693 case Stmt::BinaryOperatorClass: { 6694 const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr); 6695 if (BO->getOpcode() == BO_Comma) { 6696 TypeExpr = BO->getRHS(); 6697 continue; 6698 } 6699 return false; 6700 } 6701 6702 default: 6703 return false; 6704 } 6705 } 6706} 6707 6708/// \brief Retrieve the C type corresponding to type tag TypeExpr. 6709/// 6710/// \param TypeExpr Expression that specifies a type tag. 6711/// 6712/// \param MagicValues Registered magic values. 6713/// 6714/// \param FoundWrongKind Set to true if a type tag was found, but of a wrong 6715/// kind. 6716/// 6717/// \param TypeInfo Information about the corresponding C type. 6718/// 6719/// \returns true if the corresponding C type was found. 6720bool GetMatchingCType( 6721 const IdentifierInfo *ArgumentKind, 6722 const Expr *TypeExpr, const ASTContext &Ctx, 6723 const llvm::DenseMap<Sema::TypeTagMagicValue, 6724 Sema::TypeTagData> *MagicValues, 6725 bool &FoundWrongKind, 6726 Sema::TypeTagData &TypeInfo) { 6727 FoundWrongKind = false; 6728 6729 // Variable declaration that has type_tag_for_datatype attribute. 6730 const ValueDecl *VD = NULL; 6731 6732 uint64_t MagicValue; 6733 6734 if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue)) 6735 return false; 6736 6737 if (VD) { 6738 for (specific_attr_iterator<TypeTagForDatatypeAttr> 6739 I = VD->specific_attr_begin<TypeTagForDatatypeAttr>(), 6740 E = VD->specific_attr_end<TypeTagForDatatypeAttr>(); 6741 I != E; ++I) { 6742 if (I->getArgumentKind() != ArgumentKind) { 6743 FoundWrongKind = true; 6744 return false; 6745 } 6746 TypeInfo.Type = I->getMatchingCType(); 6747 TypeInfo.LayoutCompatible = I->getLayoutCompatible(); 6748 TypeInfo.MustBeNull = I->getMustBeNull(); 6749 return true; 6750 } 6751 return false; 6752 } 6753 6754 if (!MagicValues) 6755 return false; 6756 6757 llvm::DenseMap<Sema::TypeTagMagicValue, 6758 Sema::TypeTagData>::const_iterator I = 6759 MagicValues->find(std::make_pair(ArgumentKind, MagicValue)); 6760 if (I == MagicValues->end()) 6761 return false; 6762 6763 TypeInfo = I->second; 6764 return true; 6765} 6766} // unnamed namespace 6767 6768void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind, 6769 uint64_t MagicValue, QualType Type, 6770 bool LayoutCompatible, 6771 bool MustBeNull) { 6772 if (!TypeTagForDatatypeMagicValues) 6773 TypeTagForDatatypeMagicValues.reset( 6774 new llvm::DenseMap<TypeTagMagicValue, TypeTagData>); 6775 6776 TypeTagMagicValue Magic(ArgumentKind, MagicValue); 6777 (*TypeTagForDatatypeMagicValues)[Magic] = 6778 TypeTagData(Type, LayoutCompatible, MustBeNull); 6779} 6780 6781namespace { 6782bool IsSameCharType(QualType T1, QualType T2) { 6783 const BuiltinType *BT1 = T1->getAs<BuiltinType>(); 6784 if (!BT1) 6785 return false; 6786 6787 const BuiltinType *BT2 = T2->getAs<BuiltinType>(); 6788 if (!BT2) 6789 return false; 6790 6791 BuiltinType::Kind T1Kind = BT1->getKind(); 6792 BuiltinType::Kind T2Kind = BT2->getKind(); 6793 6794 return (T1Kind == BuiltinType::SChar && T2Kind == BuiltinType::Char_S) || 6795 (T1Kind == BuiltinType::UChar && T2Kind == BuiltinType::Char_U) || 6796 (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) || 6797 (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar); 6798} 6799} // unnamed namespace 6800 6801void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr, 6802 const Expr * const *ExprArgs) { 6803 const IdentifierInfo *ArgumentKind = Attr->getArgumentKind(); 6804 bool IsPointerAttr = Attr->getIsPointer(); 6805 6806 const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()]; 6807 bool FoundWrongKind; 6808 TypeTagData TypeInfo; 6809 if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context, 6810 TypeTagForDatatypeMagicValues.get(), 6811 FoundWrongKind, TypeInfo)) { 6812 if (FoundWrongKind) 6813 Diag(TypeTagExpr->getExprLoc(), 6814 diag::warn_type_tag_for_datatype_wrong_kind) 6815 << TypeTagExpr->getSourceRange(); 6816 return; 6817 } 6818 6819 const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()]; 6820 if (IsPointerAttr) { 6821 // Skip implicit cast of pointer to `void *' (as a function argument). 6822 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr)) 6823 if (ICE->getType()->isVoidPointerType() && 6824 ICE->getCastKind() == CK_BitCast) 6825 ArgumentExpr = ICE->getSubExpr(); 6826 } 6827 QualType ArgumentType = ArgumentExpr->getType(); 6828 6829 // Passing a `void*' pointer shouldn't trigger a warning. 6830 if (IsPointerAttr && ArgumentType->isVoidPointerType()) 6831 return; 6832 6833 if (TypeInfo.MustBeNull) { 6834 // Type tag with matching void type requires a null pointer. 6835 if (!ArgumentExpr->isNullPointerConstant(Context, 6836 Expr::NPC_ValueDependentIsNotNull)) { 6837 Diag(ArgumentExpr->getExprLoc(), 6838 diag::warn_type_safety_null_pointer_required) 6839 << ArgumentKind->getName() 6840 << ArgumentExpr->getSourceRange() 6841 << TypeTagExpr->getSourceRange(); 6842 } 6843 return; 6844 } 6845 6846 QualType RequiredType = TypeInfo.Type; 6847 if (IsPointerAttr) 6848 RequiredType = Context.getPointerType(RequiredType); 6849 6850 bool mismatch = false; 6851 if (!TypeInfo.LayoutCompatible) { 6852 mismatch = !Context.hasSameType(ArgumentType, RequiredType); 6853 6854 // C++11 [basic.fundamental] p1: 6855 // Plain char, signed char, and unsigned char are three distinct types. 6856 // 6857 // But we treat plain `char' as equivalent to `signed char' or `unsigned 6858 // char' depending on the current char signedness mode. 6859 if (mismatch) 6860 if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(), 6861 RequiredType->getPointeeType())) || 6862 (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType))) 6863 mismatch = false; 6864 } else 6865 if (IsPointerAttr) 6866 mismatch = !isLayoutCompatible(Context, 6867 ArgumentType->getPointeeType(), 6868 RequiredType->getPointeeType()); 6869 else 6870 mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType); 6871 6872 if (mismatch) 6873 Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch) 6874 << ArgumentType << ArgumentKind->getName() 6875 << TypeInfo.LayoutCompatible << RequiredType 6876 << ArgumentExpr->getSourceRange() 6877 << TypeTagExpr->getSourceRange(); 6878} 6879