1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements extra semantic analysis beyond what is enforced 11// by the C type system. 12// 13//===----------------------------------------------------------------------===// 14 15#include "clang/Sema/SemaInternal.h" 16#include "clang/AST/ASTContext.h" 17#include "clang/AST/CharUnits.h" 18#include "clang/AST/DeclCXX.h" 19#include "clang/AST/DeclObjC.h" 20#include "clang/AST/EvaluatedExprVisitor.h" 21#include "clang/AST/Expr.h" 22#include "clang/AST/ExprCXX.h" 23#include "clang/AST/ExprObjC.h" 24#include "clang/AST/StmtCXX.h" 25#include "clang/AST/StmtObjC.h" 26#include "clang/Analysis/Analyses/FormatString.h" 27#include "clang/Basic/CharInfo.h" 28#include "clang/Basic/TargetBuiltins.h" 29#include "clang/Basic/TargetInfo.h" 30#include "clang/Lex/Preprocessor.h" 31#include "clang/Sema/Initialization.h" 32#include "clang/Sema/Lookup.h" 33#include "clang/Sema/ScopeInfo.h" 34#include "clang/Sema/Sema.h" 35#include "llvm/ADT/BitVector.h" 36#include "llvm/ADT/STLExtras.h" 37#include "llvm/ADT/SmallString.h" 38#include "llvm/Support/ConvertUTF.h" 39#include "llvm/Support/raw_ostream.h" 40#include <limits> 41using namespace clang; 42using namespace sema; 43 44SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 45 unsigned ByteNo) const { 46 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 47 PP.getLangOpts(), PP.getTargetInfo()); 48} 49 50/// Checks that a call expression's argument count is the desired number. 51/// This is useful when doing custom type-checking. Returns true on error. 52static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 53 unsigned argCount = call->getNumArgs(); 54 if (argCount == desiredArgCount) return false; 55 56 if (argCount < desiredArgCount) 57 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 58 << 0 /*function call*/ << desiredArgCount << argCount 59 << call->getSourceRange(); 60 61 // Highlight all the excess arguments. 62 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 63 call->getArg(argCount - 1)->getLocEnd()); 64 65 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 66 << 0 /*function call*/ << desiredArgCount << argCount 67 << call->getArg(1)->getSourceRange(); 68} 69 70/// Check that the first argument to __builtin_annotation is an integer 71/// and the second argument is a non-wide string literal. 72static bool SemaBuiltinAnnotation(Sema &S, CallExpr *TheCall) { 73 if (checkArgCount(S, TheCall, 2)) 74 return true; 75 76 // First argument should be an integer. 77 Expr *ValArg = TheCall->getArg(0); 78 QualType Ty = ValArg->getType(); 79 if (!Ty->isIntegerType()) { 80 S.Diag(ValArg->getLocStart(), diag::err_builtin_annotation_first_arg) 81 << ValArg->getSourceRange(); 82 return true; 83 } 84 85 // Second argument should be a constant string. 86 Expr *StrArg = TheCall->getArg(1)->IgnoreParenCasts(); 87 StringLiteral *Literal = dyn_cast<StringLiteral>(StrArg); 88 if (!Literal || !Literal->isAscii()) { 89 S.Diag(StrArg->getLocStart(), diag::err_builtin_annotation_second_arg) 90 << StrArg->getSourceRange(); 91 return true; 92 } 93 94 TheCall->setType(Ty); 95 return false; 96} 97 98ExprResult 99Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 100 ExprResult TheCallResult(Owned(TheCall)); 101 102 // Find out if any arguments are required to be integer constant expressions. 103 unsigned ICEArguments = 0; 104 ASTContext::GetBuiltinTypeError Error; 105 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 106 if (Error != ASTContext::GE_None) 107 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 108 109 // If any arguments are required to be ICE's, check and diagnose. 110 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 111 // Skip arguments not required to be ICE's. 112 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 113 114 llvm::APSInt Result; 115 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 116 return true; 117 ICEArguments &= ~(1 << ArgNo); 118 } 119 120 switch (BuiltinID) { 121 case Builtin::BI__builtin___CFStringMakeConstantString: 122 assert(TheCall->getNumArgs() == 1 && 123 "Wrong # arguments to builtin CFStringMakeConstantString"); 124 if (CheckObjCString(TheCall->getArg(0))) 125 return ExprError(); 126 break; 127 case Builtin::BI__builtin_stdarg_start: 128 case Builtin::BI__builtin_va_start: 129 if (SemaBuiltinVAStart(TheCall)) 130 return ExprError(); 131 break; 132 case Builtin::BI__builtin_isgreater: 133 case Builtin::BI__builtin_isgreaterequal: 134 case Builtin::BI__builtin_isless: 135 case Builtin::BI__builtin_islessequal: 136 case Builtin::BI__builtin_islessgreater: 137 case Builtin::BI__builtin_isunordered: 138 if (SemaBuiltinUnorderedCompare(TheCall)) 139 return ExprError(); 140 break; 141 case Builtin::BI__builtin_fpclassify: 142 if (SemaBuiltinFPClassification(TheCall, 6)) 143 return ExprError(); 144 break; 145 case Builtin::BI__builtin_isfinite: 146 case Builtin::BI__builtin_isinf: 147 case Builtin::BI__builtin_isinf_sign: 148 case Builtin::BI__builtin_isnan: 149 case Builtin::BI__builtin_isnormal: 150 if (SemaBuiltinFPClassification(TheCall, 1)) 151 return ExprError(); 152 break; 153 case Builtin::BI__builtin_shufflevector: 154 return SemaBuiltinShuffleVector(TheCall); 155 // TheCall will be freed by the smart pointer here, but that's fine, since 156 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 157 case Builtin::BI__builtin_prefetch: 158 if (SemaBuiltinPrefetch(TheCall)) 159 return ExprError(); 160 break; 161 case Builtin::BI__builtin_object_size: 162 if (SemaBuiltinObjectSize(TheCall)) 163 return ExprError(); 164 break; 165 case Builtin::BI__builtin_longjmp: 166 if (SemaBuiltinLongjmp(TheCall)) 167 return ExprError(); 168 break; 169 170 case Builtin::BI__builtin_classify_type: 171 if (checkArgCount(*this, TheCall, 1)) return true; 172 TheCall->setType(Context.IntTy); 173 break; 174 case Builtin::BI__builtin_constant_p: 175 if (checkArgCount(*this, TheCall, 1)) return true; 176 TheCall->setType(Context.IntTy); 177 break; 178 case Builtin::BI__sync_fetch_and_add: 179 case Builtin::BI__sync_fetch_and_add_1: 180 case Builtin::BI__sync_fetch_and_add_2: 181 case Builtin::BI__sync_fetch_and_add_4: 182 case Builtin::BI__sync_fetch_and_add_8: 183 case Builtin::BI__sync_fetch_and_add_16: 184 case Builtin::BI__sync_fetch_and_sub: 185 case Builtin::BI__sync_fetch_and_sub_1: 186 case Builtin::BI__sync_fetch_and_sub_2: 187 case Builtin::BI__sync_fetch_and_sub_4: 188 case Builtin::BI__sync_fetch_and_sub_8: 189 case Builtin::BI__sync_fetch_and_sub_16: 190 case Builtin::BI__sync_fetch_and_or: 191 case Builtin::BI__sync_fetch_and_or_1: 192 case Builtin::BI__sync_fetch_and_or_2: 193 case Builtin::BI__sync_fetch_and_or_4: 194 case Builtin::BI__sync_fetch_and_or_8: 195 case Builtin::BI__sync_fetch_and_or_16: 196 case Builtin::BI__sync_fetch_and_and: 197 case Builtin::BI__sync_fetch_and_and_1: 198 case Builtin::BI__sync_fetch_and_and_2: 199 case Builtin::BI__sync_fetch_and_and_4: 200 case Builtin::BI__sync_fetch_and_and_8: 201 case Builtin::BI__sync_fetch_and_and_16: 202 case Builtin::BI__sync_fetch_and_xor: 203 case Builtin::BI__sync_fetch_and_xor_1: 204 case Builtin::BI__sync_fetch_and_xor_2: 205 case Builtin::BI__sync_fetch_and_xor_4: 206 case Builtin::BI__sync_fetch_and_xor_8: 207 case Builtin::BI__sync_fetch_and_xor_16: 208 case Builtin::BI__sync_add_and_fetch: 209 case Builtin::BI__sync_add_and_fetch_1: 210 case Builtin::BI__sync_add_and_fetch_2: 211 case Builtin::BI__sync_add_and_fetch_4: 212 case Builtin::BI__sync_add_and_fetch_8: 213 case Builtin::BI__sync_add_and_fetch_16: 214 case Builtin::BI__sync_sub_and_fetch: 215 case Builtin::BI__sync_sub_and_fetch_1: 216 case Builtin::BI__sync_sub_and_fetch_2: 217 case Builtin::BI__sync_sub_and_fetch_4: 218 case Builtin::BI__sync_sub_and_fetch_8: 219 case Builtin::BI__sync_sub_and_fetch_16: 220 case Builtin::BI__sync_and_and_fetch: 221 case Builtin::BI__sync_and_and_fetch_1: 222 case Builtin::BI__sync_and_and_fetch_2: 223 case Builtin::BI__sync_and_and_fetch_4: 224 case Builtin::BI__sync_and_and_fetch_8: 225 case Builtin::BI__sync_and_and_fetch_16: 226 case Builtin::BI__sync_or_and_fetch: 227 case Builtin::BI__sync_or_and_fetch_1: 228 case Builtin::BI__sync_or_and_fetch_2: 229 case Builtin::BI__sync_or_and_fetch_4: 230 case Builtin::BI__sync_or_and_fetch_8: 231 case Builtin::BI__sync_or_and_fetch_16: 232 case Builtin::BI__sync_xor_and_fetch: 233 case Builtin::BI__sync_xor_and_fetch_1: 234 case Builtin::BI__sync_xor_and_fetch_2: 235 case Builtin::BI__sync_xor_and_fetch_4: 236 case Builtin::BI__sync_xor_and_fetch_8: 237 case Builtin::BI__sync_xor_and_fetch_16: 238 case Builtin::BI__sync_val_compare_and_swap: 239 case Builtin::BI__sync_val_compare_and_swap_1: 240 case Builtin::BI__sync_val_compare_and_swap_2: 241 case Builtin::BI__sync_val_compare_and_swap_4: 242 case Builtin::BI__sync_val_compare_and_swap_8: 243 case Builtin::BI__sync_val_compare_and_swap_16: 244 case Builtin::BI__sync_bool_compare_and_swap: 245 case Builtin::BI__sync_bool_compare_and_swap_1: 246 case Builtin::BI__sync_bool_compare_and_swap_2: 247 case Builtin::BI__sync_bool_compare_and_swap_4: 248 case Builtin::BI__sync_bool_compare_and_swap_8: 249 case Builtin::BI__sync_bool_compare_and_swap_16: 250 case Builtin::BI__sync_lock_test_and_set: 251 case Builtin::BI__sync_lock_test_and_set_1: 252 case Builtin::BI__sync_lock_test_and_set_2: 253 case Builtin::BI__sync_lock_test_and_set_4: 254 case Builtin::BI__sync_lock_test_and_set_8: 255 case Builtin::BI__sync_lock_test_and_set_16: 256 case Builtin::BI__sync_lock_release: 257 case Builtin::BI__sync_lock_release_1: 258 case Builtin::BI__sync_lock_release_2: 259 case Builtin::BI__sync_lock_release_4: 260 case Builtin::BI__sync_lock_release_8: 261 case Builtin::BI__sync_lock_release_16: 262 case Builtin::BI__sync_swap: 263 case Builtin::BI__sync_swap_1: 264 case Builtin::BI__sync_swap_2: 265 case Builtin::BI__sync_swap_4: 266 case Builtin::BI__sync_swap_8: 267 case Builtin::BI__sync_swap_16: 268 return SemaBuiltinAtomicOverloaded(TheCallResult); 269#define BUILTIN(ID, TYPE, ATTRS) 270#define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \ 271 case Builtin::BI##ID: \ 272 return SemaAtomicOpsOverloaded(TheCallResult, AtomicExpr::AO##ID); 273#include "clang/Basic/Builtins.def" 274 case Builtin::BI__builtin_annotation: 275 if (SemaBuiltinAnnotation(*this, TheCall)) 276 return ExprError(); 277 break; 278 } 279 280 // Since the target specific builtins for each arch overlap, only check those 281 // of the arch we are compiling for. 282 if (BuiltinID >= Builtin::FirstTSBuiltin) { 283 switch (Context.getTargetInfo().getTriple().getArch()) { 284 case llvm::Triple::arm: 285 case llvm::Triple::thumb: 286 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 287 return ExprError(); 288 break; 289 case llvm::Triple::mips: 290 case llvm::Triple::mipsel: 291 case llvm::Triple::mips64: 292 case llvm::Triple::mips64el: 293 if (CheckMipsBuiltinFunctionCall(BuiltinID, TheCall)) 294 return ExprError(); 295 break; 296 default: 297 break; 298 } 299 } 300 301 return TheCallResult; 302} 303 304// Get the valid immediate range for the specified NEON type code. 305static unsigned RFT(unsigned t, bool shift = false) { 306 NeonTypeFlags Type(t); 307 int IsQuad = Type.isQuad(); 308 switch (Type.getEltType()) { 309 case NeonTypeFlags::Int8: 310 case NeonTypeFlags::Poly8: 311 return shift ? 7 : (8 << IsQuad) - 1; 312 case NeonTypeFlags::Int16: 313 case NeonTypeFlags::Poly16: 314 return shift ? 15 : (4 << IsQuad) - 1; 315 case NeonTypeFlags::Int32: 316 return shift ? 31 : (2 << IsQuad) - 1; 317 case NeonTypeFlags::Int64: 318 return shift ? 63 : (1 << IsQuad) - 1; 319 case NeonTypeFlags::Float16: 320 assert(!shift && "cannot shift float types!"); 321 return (4 << IsQuad) - 1; 322 case NeonTypeFlags::Float32: 323 assert(!shift && "cannot shift float types!"); 324 return (2 << IsQuad) - 1; 325 } 326 llvm_unreachable("Invalid NeonTypeFlag!"); 327} 328 329/// getNeonEltType - Return the QualType corresponding to the elements of 330/// the vector type specified by the NeonTypeFlags. This is used to check 331/// the pointer arguments for Neon load/store intrinsics. 332static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) { 333 switch (Flags.getEltType()) { 334 case NeonTypeFlags::Int8: 335 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 336 case NeonTypeFlags::Int16: 337 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 338 case NeonTypeFlags::Int32: 339 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 340 case NeonTypeFlags::Int64: 341 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 342 case NeonTypeFlags::Poly8: 343 return Context.SignedCharTy; 344 case NeonTypeFlags::Poly16: 345 return Context.ShortTy; 346 case NeonTypeFlags::Float16: 347 return Context.UnsignedShortTy; 348 case NeonTypeFlags::Float32: 349 return Context.FloatTy; 350 } 351 llvm_unreachable("Invalid NeonTypeFlag!"); 352} 353 354bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 355 llvm::APSInt Result; 356 357 uint64_t mask = 0; 358 unsigned TV = 0; 359 int PtrArgNum = -1; 360 bool HasConstPtr = false; 361 switch (BuiltinID) { 362#define GET_NEON_OVERLOAD_CHECK 363#include "clang/Basic/arm_neon.inc" 364#undef GET_NEON_OVERLOAD_CHECK 365 } 366 367 // For NEON intrinsics which are overloaded on vector element type, validate 368 // the immediate which specifies which variant to emit. 369 unsigned ImmArg = TheCall->getNumArgs()-1; 370 if (mask) { 371 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 372 return true; 373 374 TV = Result.getLimitedValue(64); 375 if ((TV > 63) || (mask & (1ULL << TV)) == 0) 376 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 377 << TheCall->getArg(ImmArg)->getSourceRange(); 378 } 379 380 if (PtrArgNum >= 0) { 381 // Check that pointer arguments have the specified type. 382 Expr *Arg = TheCall->getArg(PtrArgNum); 383 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 384 Arg = ICE->getSubExpr(); 385 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 386 QualType RHSTy = RHS.get()->getType(); 387 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 388 if (HasConstPtr) 389 EltTy = EltTy.withConst(); 390 QualType LHSTy = Context.getPointerType(EltTy); 391 AssignConvertType ConvTy; 392 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 393 if (RHS.isInvalid()) 394 return true; 395 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 396 RHS.get(), AA_Assigning)) 397 return true; 398 } 399 400 // For NEON intrinsics which take an immediate value as part of the 401 // instruction, range check them here. 402 unsigned i = 0, l = 0, u = 0; 403 switch (BuiltinID) { 404 default: return false; 405 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 406 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 407 case ARM::BI__builtin_arm_vcvtr_f: 408 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 409#define GET_NEON_IMMEDIATE_CHECK 410#include "clang/Basic/arm_neon.inc" 411#undef GET_NEON_IMMEDIATE_CHECK 412 }; 413 414 // We can't check the value of a dependent argument. 415 if (TheCall->getArg(i)->isTypeDependent() || 416 TheCall->getArg(i)->isValueDependent()) 417 return false; 418 419 // Check that the immediate argument is actually a constant. 420 if (SemaBuiltinConstantArg(TheCall, i, Result)) 421 return true; 422 423 // Range check against the upper/lower values for this isntruction. 424 unsigned Val = Result.getZExtValue(); 425 if (Val < l || Val > (u + l)) 426 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 427 << l << u+l << TheCall->getArg(i)->getSourceRange(); 428 429 // FIXME: VFP Intrinsics should error if VFP not present. 430 return false; 431} 432 433bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 434 unsigned i = 0, l = 0, u = 0; 435 switch (BuiltinID) { 436 default: return false; 437 case Mips::BI__builtin_mips_wrdsp: i = 1; l = 0; u = 63; break; 438 case Mips::BI__builtin_mips_rddsp: i = 0; l = 0; u = 63; break; 439 case Mips::BI__builtin_mips_append: i = 2; l = 0; u = 31; break; 440 case Mips::BI__builtin_mips_balign: i = 2; l = 0; u = 3; break; 441 case Mips::BI__builtin_mips_precr_sra_ph_w: i = 2; l = 0; u = 31; break; 442 case Mips::BI__builtin_mips_precr_sra_r_ph_w: i = 2; l = 0; u = 31; break; 443 case Mips::BI__builtin_mips_prepend: i = 2; l = 0; u = 31; break; 444 }; 445 446 // We can't check the value of a dependent argument. 447 if (TheCall->getArg(i)->isTypeDependent() || 448 TheCall->getArg(i)->isValueDependent()) 449 return false; 450 451 // Check that the immediate argument is actually a constant. 452 llvm::APSInt Result; 453 if (SemaBuiltinConstantArg(TheCall, i, Result)) 454 return true; 455 456 // Range check against the upper/lower values for this instruction. 457 unsigned Val = Result.getZExtValue(); 458 if (Val < l || Val > u) 459 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 460 << l << u << TheCall->getArg(i)->getSourceRange(); 461 462 return false; 463} 464 465/// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo 466/// parameter with the FormatAttr's correct format_idx and firstDataArg. 467/// Returns true when the format fits the function and the FormatStringInfo has 468/// been populated. 469bool Sema::getFormatStringInfo(const FormatAttr *Format, bool IsCXXMember, 470 FormatStringInfo *FSI) { 471 FSI->HasVAListArg = Format->getFirstArg() == 0; 472 FSI->FormatIdx = Format->getFormatIdx() - 1; 473 FSI->FirstDataArg = FSI->HasVAListArg ? 0 : Format->getFirstArg() - 1; 474 475 // The way the format attribute works in GCC, the implicit this argument 476 // of member functions is counted. However, it doesn't appear in our own 477 // lists, so decrement format_idx in that case. 478 if (IsCXXMember) { 479 if(FSI->FormatIdx == 0) 480 return false; 481 --FSI->FormatIdx; 482 if (FSI->FirstDataArg != 0) 483 --FSI->FirstDataArg; 484 } 485 return true; 486} 487 488/// Handles the checks for format strings, non-POD arguments to vararg 489/// functions, and NULL arguments passed to non-NULL parameters. 490void Sema::checkCall(NamedDecl *FDecl, 491 ArrayRef<const Expr *> Args, 492 unsigned NumProtoArgs, 493 bool IsMemberFunction, 494 SourceLocation Loc, 495 SourceRange Range, 496 VariadicCallType CallType) { 497 if (CurContext->isDependentContext()) 498 return; 499 500 // Printf and scanf checking. 501 bool HandledFormatString = false; 502 for (specific_attr_iterator<FormatAttr> 503 I = FDecl->specific_attr_begin<FormatAttr>(), 504 E = FDecl->specific_attr_end<FormatAttr>(); I != E ; ++I) 505 if (CheckFormatArguments(*I, Args, IsMemberFunction, CallType, Loc, Range)) 506 HandledFormatString = true; 507 508 // Refuse POD arguments that weren't caught by the format string 509 // checks above. 510 if (!HandledFormatString && CallType != VariadicDoesNotApply) 511 for (unsigned ArgIdx = NumProtoArgs; ArgIdx < Args.size(); ++ArgIdx) { 512 // Args[ArgIdx] can be null in malformed code. 513 if (const Expr *Arg = Args[ArgIdx]) 514 variadicArgumentPODCheck(Arg, CallType); 515 } 516 517 for (specific_attr_iterator<NonNullAttr> 518 I = FDecl->specific_attr_begin<NonNullAttr>(), 519 E = FDecl->specific_attr_end<NonNullAttr>(); I != E; ++I) 520 CheckNonNullArguments(*I, Args.data(), Loc); 521 522 // Type safety checking. 523 for (specific_attr_iterator<ArgumentWithTypeTagAttr> 524 i = FDecl->specific_attr_begin<ArgumentWithTypeTagAttr>(), 525 e = FDecl->specific_attr_end<ArgumentWithTypeTagAttr>(); i != e; ++i) { 526 CheckArgumentWithTypeTag(*i, Args.data()); 527 } 528} 529 530/// CheckConstructorCall - Check a constructor call for correctness and safety 531/// properties not enforced by the C type system. 532void Sema::CheckConstructorCall(FunctionDecl *FDecl, 533 ArrayRef<const Expr *> Args, 534 const FunctionProtoType *Proto, 535 SourceLocation Loc) { 536 VariadicCallType CallType = 537 Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply; 538 checkCall(FDecl, Args, Proto->getNumArgs(), 539 /*IsMemberFunction=*/true, Loc, SourceRange(), CallType); 540} 541 542/// CheckFunctionCall - Check a direct function call for various correctness 543/// and safety properties not strictly enforced by the C type system. 544bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall, 545 const FunctionProtoType *Proto) { 546 bool IsMemberOperatorCall = isa<CXXOperatorCallExpr>(TheCall) && 547 isa<CXXMethodDecl>(FDecl); 548 bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall) || 549 IsMemberOperatorCall; 550 VariadicCallType CallType = getVariadicCallType(FDecl, Proto, 551 TheCall->getCallee()); 552 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 553 Expr** Args = TheCall->getArgs(); 554 unsigned NumArgs = TheCall->getNumArgs(); 555 if (IsMemberOperatorCall) { 556 // If this is a call to a member operator, hide the first argument 557 // from checkCall. 558 // FIXME: Our choice of AST representation here is less than ideal. 559 ++Args; 560 --NumArgs; 561 } 562 checkCall(FDecl, llvm::makeArrayRef<const Expr *>(Args, NumArgs), 563 NumProtoArgs, 564 IsMemberFunction, TheCall->getRParenLoc(), 565 TheCall->getCallee()->getSourceRange(), CallType); 566 567 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 568 // None of the checks below are needed for functions that don't have 569 // simple names (e.g., C++ conversion functions). 570 if (!FnInfo) 571 return false; 572 573 unsigned CMId = FDecl->getMemoryFunctionKind(); 574 if (CMId == 0) 575 return false; 576 577 // Handle memory setting and copying functions. 578 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 579 CheckStrlcpycatArguments(TheCall, FnInfo); 580 else if (CMId == Builtin::BIstrncat) 581 CheckStrncatArguments(TheCall, FnInfo); 582 else 583 CheckMemaccessArguments(TheCall, CMId, FnInfo); 584 585 return false; 586} 587 588bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 589 ArrayRef<const Expr *> Args) { 590 VariadicCallType CallType = 591 Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply; 592 593 checkCall(Method, Args, 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 = None); 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 = None); 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 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 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 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 // FreeBSD extensions 2628 if (CS.getKind() == ConversionSpecifier::FreeBSDbArg || 2629 CS.getKind() == ConversionSpecifier::FreeBSDDArg) { 2630 // claim the second argument 2631 CoveredArgs.set(argIndex + 1); 2632 2633 // Now type check the data expression that matches the 2634 // format specifier. 2635 const Expr *Ex = getDataArg(argIndex); 2636 const analyze_printf::ArgType &AT = 2637 (CS.getKind() == ConversionSpecifier::FreeBSDbArg) ? 2638 ArgType(S.Context.IntTy) : ArgType::CStrTy; 2639 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) 2640 S.Diag(getLocationOfByte(CS.getStart()), 2641 diag::warn_printf_conversion_argument_type_mismatch) 2642 << AT.getRepresentativeType(S.Context) << Ex->getType() 2643 << getSpecifierRange(startSpecifier, specifierLen) 2644 << Ex->getSourceRange(); 2645 2646 // Now type check the data expression that matches the 2647 // format specifier. 2648 Ex = getDataArg(argIndex + 1); 2649 const analyze_printf::ArgType &AT2 = ArgType::CStrTy; 2650 if (AT2.isValid() && !AT2.matchesType(S.Context, Ex->getType())) 2651 S.Diag(getLocationOfByte(CS.getStart()), 2652 diag::warn_printf_conversion_argument_type_mismatch) 2653 << AT2.getRepresentativeType(S.Context) << Ex->getType() 2654 << getSpecifierRange(startSpecifier, specifierLen) 2655 << Ex->getSourceRange(); 2656 2657 return true; 2658 } 2659 // END OF FREEBSD EXTENSIONS 2660 2661 // Check for using an Objective-C specific conversion specifier 2662 // in a non-ObjC literal. 2663 if (!ObjCContext && CS.isObjCArg()) { 2664 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2665 specifierLen); 2666 } 2667 2668 // Check for invalid use of field width 2669 if (!FS.hasValidFieldWidth()) { 2670 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2671 startSpecifier, specifierLen); 2672 } 2673 2674 // Check for invalid use of precision 2675 if (!FS.hasValidPrecision()) { 2676 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2677 startSpecifier, specifierLen); 2678 } 2679 2680 // Check each flag does not conflict with any other component. 2681 if (!FS.hasValidThousandsGroupingPrefix()) 2682 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2683 if (!FS.hasValidLeadingZeros()) 2684 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2685 if (!FS.hasValidPlusPrefix()) 2686 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2687 if (!FS.hasValidSpacePrefix()) 2688 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2689 if (!FS.hasValidAlternativeForm()) 2690 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2691 if (!FS.hasValidLeftJustified()) 2692 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2693 2694 // Check that flags are not ignored by another flag 2695 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2696 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2697 startSpecifier, specifierLen); 2698 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2699 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2700 startSpecifier, specifierLen); 2701 2702 // Check the length modifier is valid with the given conversion specifier. 2703 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 2704 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2705 diag::warn_format_nonsensical_length); 2706 else if (!FS.hasStandardLengthModifier()) 2707 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 2708 else if (!FS.hasStandardLengthConversionCombination()) 2709 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2710 diag::warn_format_non_standard_conversion_spec); 2711 2712 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2713 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2714 2715 // The remaining checks depend on the data arguments. 2716 if (HasVAListArg) 2717 return true; 2718 2719 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2720 return false; 2721 2722 const Expr *Arg = getDataArg(argIndex); 2723 if (!Arg) 2724 return true; 2725 2726 return checkFormatExpr(FS, startSpecifier, specifierLen, Arg); 2727} 2728 2729static bool requiresParensToAddCast(const Expr *E) { 2730 // FIXME: We should have a general way to reason about operator 2731 // precedence and whether parens are actually needed here. 2732 // Take care of a few common cases where they aren't. 2733 const Expr *Inside = E->IgnoreImpCasts(); 2734 if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside)) 2735 Inside = POE->getSyntacticForm()->IgnoreImpCasts(); 2736 2737 switch (Inside->getStmtClass()) { 2738 case Stmt::ArraySubscriptExprClass: 2739 case Stmt::CallExprClass: 2740 case Stmt::CharacterLiteralClass: 2741 case Stmt::CXXBoolLiteralExprClass: 2742 case Stmt::DeclRefExprClass: 2743 case Stmt::FloatingLiteralClass: 2744 case Stmt::IntegerLiteralClass: 2745 case Stmt::MemberExprClass: 2746 case Stmt::ObjCArrayLiteralClass: 2747 case Stmt::ObjCBoolLiteralExprClass: 2748 case Stmt::ObjCBoxedExprClass: 2749 case Stmt::ObjCDictionaryLiteralClass: 2750 case Stmt::ObjCEncodeExprClass: 2751 case Stmt::ObjCIvarRefExprClass: 2752 case Stmt::ObjCMessageExprClass: 2753 case Stmt::ObjCPropertyRefExprClass: 2754 case Stmt::ObjCStringLiteralClass: 2755 case Stmt::ObjCSubscriptRefExprClass: 2756 case Stmt::ParenExprClass: 2757 case Stmt::StringLiteralClass: 2758 case Stmt::UnaryOperatorClass: 2759 return false; 2760 default: 2761 return true; 2762 } 2763} 2764 2765bool 2766CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2767 const char *StartSpecifier, 2768 unsigned SpecifierLen, 2769 const Expr *E) { 2770 using namespace analyze_format_string; 2771 using namespace analyze_printf; 2772 // Now type check the data expression that matches the 2773 // format specifier. 2774 const analyze_printf::ArgType &AT = FS.getArgType(S.Context, 2775 ObjCContext); 2776 if (!AT.isValid()) 2777 return true; 2778 2779 QualType ExprTy = E->getType(); 2780 while (const TypeOfExprType *TET = dyn_cast<TypeOfExprType>(ExprTy)) { 2781 ExprTy = TET->getUnderlyingExpr()->getType(); 2782 } 2783 2784 if (AT.matchesType(S.Context, ExprTy)) 2785 return true; 2786 2787 // Look through argument promotions for our error message's reported type. 2788 // This includes the integral and floating promotions, but excludes array 2789 // and function pointer decay; seeing that an argument intended to be a 2790 // string has type 'char [6]' is probably more confusing than 'char *'. 2791 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2792 if (ICE->getCastKind() == CK_IntegralCast || 2793 ICE->getCastKind() == CK_FloatingCast) { 2794 E = ICE->getSubExpr(); 2795 ExprTy = E->getType(); 2796 2797 // Check if we didn't match because of an implicit cast from a 'char' 2798 // or 'short' to an 'int'. This is done because printf is a varargs 2799 // function. 2800 if (ICE->getType() == S.Context.IntTy || 2801 ICE->getType() == S.Context.UnsignedIntTy) { 2802 // All further checking is done on the subexpression. 2803 if (AT.matchesType(S.Context, ExprTy)) 2804 return true; 2805 } 2806 } 2807 } else if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E)) { 2808 // Special case for 'a', which has type 'int' in C. 2809 // Note, however, that we do /not/ want to treat multibyte constants like 2810 // 'MooV' as characters! This form is deprecated but still exists. 2811 if (ExprTy == S.Context.IntTy) 2812 if (llvm::isUIntN(S.Context.getCharWidth(), CL->getValue())) 2813 ExprTy = S.Context.CharTy; 2814 } 2815 2816 // %C in an Objective-C context prints a unichar, not a wchar_t. 2817 // If the argument is an integer of some kind, believe the %C and suggest 2818 // a cast instead of changing the conversion specifier. 2819 QualType IntendedTy = ExprTy; 2820 if (ObjCContext && 2821 FS.getConversionSpecifier().getKind() == ConversionSpecifier::CArg) { 2822 if (ExprTy->isIntegralOrUnscopedEnumerationType() && 2823 !ExprTy->isCharType()) { 2824 // 'unichar' is defined as a typedef of unsigned short, but we should 2825 // prefer using the typedef if it is visible. 2826 IntendedTy = S.Context.UnsignedShortTy; 2827 2828 LookupResult Result(S, &S.Context.Idents.get("unichar"), E->getLocStart(), 2829 Sema::LookupOrdinaryName); 2830 if (S.LookupName(Result, S.getCurScope())) { 2831 NamedDecl *ND = Result.getFoundDecl(); 2832 if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(ND)) 2833 if (TD->getUnderlyingType() == IntendedTy) 2834 IntendedTy = S.Context.getTypedefType(TD); 2835 } 2836 } 2837 } 2838 2839 // Special-case some of Darwin's platform-independence types by suggesting 2840 // casts to primitive types that are known to be large enough. 2841 bool ShouldNotPrintDirectly = false; 2842 if (S.Context.getTargetInfo().getTriple().isOSDarwin()) { 2843 // Use a 'while' to peel off layers of typedefs. 2844 QualType TyTy = IntendedTy; 2845 while (const TypedefType *UserTy = TyTy->getAs<TypedefType>()) { 2846 StringRef Name = UserTy->getDecl()->getName(); 2847 QualType CastTy = llvm::StringSwitch<QualType>(Name) 2848 .Case("NSInteger", S.Context.LongTy) 2849 .Case("NSUInteger", S.Context.UnsignedLongTy) 2850 .Case("SInt32", S.Context.IntTy) 2851 .Case("UInt32", S.Context.UnsignedIntTy) 2852 .Default(QualType()); 2853 2854 if (!CastTy.isNull()) { 2855 ShouldNotPrintDirectly = true; 2856 IntendedTy = CastTy; 2857 break; 2858 } 2859 TyTy = UserTy->desugar(); 2860 } 2861 } 2862 2863 // We may be able to offer a FixItHint if it is a supported type. 2864 PrintfSpecifier fixedFS = FS; 2865 bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(), 2866 S.Context, ObjCContext); 2867 2868 if (success) { 2869 // Get the fix string from the fixed format specifier 2870 SmallString<16> buf; 2871 llvm::raw_svector_ostream os(buf); 2872 fixedFS.toString(os); 2873 2874 CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen); 2875 2876 if (IntendedTy == ExprTy) { 2877 // In this case, the specifier is wrong and should be changed to match 2878 // the argument. 2879 EmitFormatDiagnostic( 2880 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2881 << AT.getRepresentativeTypeName(S.Context) << IntendedTy 2882 << E->getSourceRange(), 2883 E->getLocStart(), 2884 /*IsStringLocation*/false, 2885 SpecRange, 2886 FixItHint::CreateReplacement(SpecRange, os.str())); 2887 2888 } else { 2889 // The canonical type for formatting this value is different from the 2890 // actual type of the expression. (This occurs, for example, with Darwin's 2891 // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but 2892 // should be printed as 'long' for 64-bit compatibility.) 2893 // Rather than emitting a normal format/argument mismatch, we want to 2894 // add a cast to the recommended type (and correct the format string 2895 // if necessary). 2896 SmallString<16> CastBuf; 2897 llvm::raw_svector_ostream CastFix(CastBuf); 2898 CastFix << "("; 2899 IntendedTy.print(CastFix, S.Context.getPrintingPolicy()); 2900 CastFix << ")"; 2901 2902 SmallVector<FixItHint,4> Hints; 2903 if (!AT.matchesType(S.Context, IntendedTy)) 2904 Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str())); 2905 2906 if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) { 2907 // If there's already a cast present, just replace it. 2908 SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc()); 2909 Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str())); 2910 2911 } else if (!requiresParensToAddCast(E)) { 2912 // If the expression has high enough precedence, 2913 // just write the C-style cast. 2914 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 2915 CastFix.str())); 2916 } else { 2917 // Otherwise, add parens around the expression as well as the cast. 2918 CastFix << "("; 2919 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 2920 CastFix.str())); 2921 2922 SourceLocation After = S.PP.getLocForEndOfToken(E->getLocEnd()); 2923 Hints.push_back(FixItHint::CreateInsertion(After, ")")); 2924 } 2925 2926 if (ShouldNotPrintDirectly) { 2927 // The expression has a type that should not be printed directly. 2928 // We extract the name from the typedef because we don't want to show 2929 // the underlying type in the diagnostic. 2930 StringRef Name = cast<TypedefType>(ExprTy)->getDecl()->getName(); 2931 2932 EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast) 2933 << Name << IntendedTy 2934 << E->getSourceRange(), 2935 E->getLocStart(), /*IsStringLocation=*/false, 2936 SpecRange, Hints); 2937 } else { 2938 // In this case, the expression could be printed using a different 2939 // specifier, but we've decided that the specifier is probably correct 2940 // and we should cast instead. Just use the normal warning message. 2941 EmitFormatDiagnostic( 2942 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2943 << AT.getRepresentativeTypeName(S.Context) << ExprTy 2944 << E->getSourceRange(), 2945 E->getLocStart(), /*IsStringLocation*/false, 2946 SpecRange, Hints); 2947 } 2948 } 2949 } else { 2950 const CharSourceRange &CSR = getSpecifierRange(StartSpecifier, 2951 SpecifierLen); 2952 // Since the warning for passing non-POD types to variadic functions 2953 // was deferred until now, we emit a warning for non-POD 2954 // arguments here. 2955 if (S.isValidVarArgType(ExprTy) == Sema::VAK_Invalid) { 2956 unsigned DiagKind; 2957 if (ExprTy->isObjCObjectType()) 2958 DiagKind = diag::err_cannot_pass_objc_interface_to_vararg_format; 2959 else 2960 DiagKind = diag::warn_non_pod_vararg_with_format_string; 2961 2962 EmitFormatDiagnostic( 2963 S.PDiag(DiagKind) 2964 << S.getLangOpts().CPlusPlus11 2965 << ExprTy 2966 << CallType 2967 << AT.getRepresentativeTypeName(S.Context) 2968 << CSR 2969 << E->getSourceRange(), 2970 E->getLocStart(), /*IsStringLocation*/false, CSR); 2971 2972 checkForCStrMembers(AT, E, CSR); 2973 } else 2974 EmitFormatDiagnostic( 2975 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2976 << AT.getRepresentativeTypeName(S.Context) << ExprTy 2977 << CSR 2978 << E->getSourceRange(), 2979 E->getLocStart(), /*IsStringLocation*/false, CSR); 2980 } 2981 2982 return true; 2983} 2984 2985//===--- CHECK: Scanf format string checking ------------------------------===// 2986 2987namespace { 2988class CheckScanfHandler : public CheckFormatHandler { 2989public: 2990 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2991 const Expr *origFormatExpr, unsigned firstDataArg, 2992 unsigned numDataArgs, const char *beg, bool hasVAListArg, 2993 ArrayRef<const Expr *> Args, 2994 unsigned formatIdx, bool inFunctionCall, 2995 Sema::VariadicCallType CallType) 2996 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2997 numDataArgs, beg, hasVAListArg, 2998 Args, formatIdx, inFunctionCall, CallType) 2999 {} 3000 3001 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 3002 const char *startSpecifier, 3003 unsigned specifierLen); 3004 3005 bool HandleInvalidScanfConversionSpecifier( 3006 const analyze_scanf::ScanfSpecifier &FS, 3007 const char *startSpecifier, 3008 unsigned specifierLen); 3009 3010 void HandleIncompleteScanList(const char *start, const char *end); 3011}; 3012} 3013 3014void CheckScanfHandler::HandleIncompleteScanList(const char *start, 3015 const char *end) { 3016 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 3017 getLocationOfByte(end), /*IsStringLocation*/true, 3018 getSpecifierRange(start, end - start)); 3019} 3020 3021bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 3022 const analyze_scanf::ScanfSpecifier &FS, 3023 const char *startSpecifier, 3024 unsigned specifierLen) { 3025 3026 const analyze_scanf::ScanfConversionSpecifier &CS = 3027 FS.getConversionSpecifier(); 3028 3029 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 3030 getLocationOfByte(CS.getStart()), 3031 startSpecifier, specifierLen, 3032 CS.getStart(), CS.getLength()); 3033} 3034 3035bool CheckScanfHandler::HandleScanfSpecifier( 3036 const analyze_scanf::ScanfSpecifier &FS, 3037 const char *startSpecifier, 3038 unsigned specifierLen) { 3039 3040 using namespace analyze_scanf; 3041 using namespace analyze_format_string; 3042 3043 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 3044 3045 // Handle case where '%' and '*' don't consume an argument. These shouldn't 3046 // be used to decide if we are using positional arguments consistently. 3047 if (FS.consumesDataArgument()) { 3048 if (atFirstArg) { 3049 atFirstArg = false; 3050 usesPositionalArgs = FS.usesPositionalArg(); 3051 } 3052 else if (usesPositionalArgs != FS.usesPositionalArg()) { 3053 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 3054 startSpecifier, specifierLen); 3055 return false; 3056 } 3057 } 3058 3059 // Check if the field with is non-zero. 3060 const OptionalAmount &Amt = FS.getFieldWidth(); 3061 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 3062 if (Amt.getConstantAmount() == 0) { 3063 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 3064 Amt.getConstantLength()); 3065 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 3066 getLocationOfByte(Amt.getStart()), 3067 /*IsStringLocation*/true, R, 3068 FixItHint::CreateRemoval(R)); 3069 } 3070 } 3071 3072 if (!FS.consumesDataArgument()) { 3073 // FIXME: Technically specifying a precision or field width here 3074 // makes no sense. Worth issuing a warning at some point. 3075 return true; 3076 } 3077 3078 // Consume the argument. 3079 unsigned argIndex = FS.getArgIndex(); 3080 if (argIndex < NumDataArgs) { 3081 // The check to see if the argIndex is valid will come later. 3082 // We set the bit here because we may exit early from this 3083 // function if we encounter some other error. 3084 CoveredArgs.set(argIndex); 3085 } 3086 3087 // Check the length modifier is valid with the given conversion specifier. 3088 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3089 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3090 diag::warn_format_nonsensical_length); 3091 else if (!FS.hasStandardLengthModifier()) 3092 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3093 else if (!FS.hasStandardLengthConversionCombination()) 3094 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3095 diag::warn_format_non_standard_conversion_spec); 3096 3097 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3098 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3099 3100 // The remaining checks depend on the data arguments. 3101 if (HasVAListArg) 3102 return true; 3103 3104 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3105 return false; 3106 3107 // Check that the argument type matches the format specifier. 3108 const Expr *Ex = getDataArg(argIndex); 3109 if (!Ex) 3110 return true; 3111 3112 const analyze_format_string::ArgType &AT = FS.getArgType(S.Context); 3113 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) { 3114 ScanfSpecifier fixedFS = FS; 3115 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 3116 S.Context); 3117 3118 if (success) { 3119 // Get the fix string from the fixed format specifier. 3120 SmallString<128> buf; 3121 llvm::raw_svector_ostream os(buf); 3122 fixedFS.toString(os); 3123 3124 EmitFormatDiagnostic( 3125 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3126 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3127 << Ex->getSourceRange(), 3128 Ex->getLocStart(), 3129 /*IsStringLocation*/false, 3130 getSpecifierRange(startSpecifier, specifierLen), 3131 FixItHint::CreateReplacement( 3132 getSpecifierRange(startSpecifier, specifierLen), 3133 os.str())); 3134 } else { 3135 EmitFormatDiagnostic( 3136 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3137 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3138 << Ex->getSourceRange(), 3139 Ex->getLocStart(), 3140 /*IsStringLocation*/false, 3141 getSpecifierRange(startSpecifier, specifierLen)); 3142 } 3143 } 3144 3145 return true; 3146} 3147 3148void Sema::CheckFormatString(const StringLiteral *FExpr, 3149 const Expr *OrigFormatExpr, 3150 ArrayRef<const Expr *> Args, 3151 bool HasVAListArg, unsigned format_idx, 3152 unsigned firstDataArg, FormatStringType Type, 3153 bool inFunctionCall, VariadicCallType CallType) { 3154 3155 // CHECK: is the format string a wide literal? 3156 if (!FExpr->isAscii() && !FExpr->isUTF8()) { 3157 CheckFormatHandler::EmitFormatDiagnostic( 3158 *this, inFunctionCall, Args[format_idx], 3159 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 3160 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3161 return; 3162 } 3163 3164 // Str - The format string. NOTE: this is NOT null-terminated! 3165 StringRef StrRef = FExpr->getString(); 3166 const char *Str = StrRef.data(); 3167 unsigned StrLen = StrRef.size(); 3168 const unsigned numDataArgs = Args.size() - firstDataArg; 3169 3170 // CHECK: empty format string? 3171 if (StrLen == 0 && numDataArgs > 0) { 3172 CheckFormatHandler::EmitFormatDiagnostic( 3173 *this, inFunctionCall, Args[format_idx], 3174 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 3175 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3176 return; 3177 } 3178 3179 if (Type == FST_Printf || Type == FST_NSString) { 3180 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 3181 numDataArgs, (Type == FST_NSString), 3182 Str, HasVAListArg, Args, format_idx, 3183 inFunctionCall, CallType); 3184 3185 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 3186 getLangOpts(), 3187 Context.getTargetInfo())) 3188 H.DoneProcessing(); 3189 } else if (Type == FST_Scanf) { 3190 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs, 3191 Str, HasVAListArg, Args, format_idx, 3192 inFunctionCall, CallType); 3193 3194 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 3195 getLangOpts(), 3196 Context.getTargetInfo())) 3197 H.DoneProcessing(); 3198 } // TODO: handle other formats 3199} 3200 3201//===--- CHECK: Standard memory functions ---------------------------------===// 3202 3203/// \brief Determine whether the given type is a dynamic class type (e.g., 3204/// whether it has a vtable). 3205static bool isDynamicClassType(QualType T) { 3206 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 3207 if (CXXRecordDecl *Definition = Record->getDefinition()) 3208 if (Definition->isDynamicClass()) 3209 return true; 3210 3211 return false; 3212} 3213 3214/// \brief If E is a sizeof expression, returns its argument expression, 3215/// otherwise returns NULL. 3216static const Expr *getSizeOfExprArg(const Expr* E) { 3217 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3218 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3219 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 3220 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 3221 3222 return 0; 3223} 3224 3225/// \brief If E is a sizeof expression, returns its argument type. 3226static QualType getSizeOfArgType(const Expr* E) { 3227 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3228 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3229 if (SizeOf->getKind() == clang::UETT_SizeOf) 3230 return SizeOf->getTypeOfArgument(); 3231 3232 return QualType(); 3233} 3234 3235/// \brief Check for dangerous or invalid arguments to memset(). 3236/// 3237/// This issues warnings on known problematic, dangerous or unspecified 3238/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 3239/// function calls. 3240/// 3241/// \param Call The call expression to diagnose. 3242void Sema::CheckMemaccessArguments(const CallExpr *Call, 3243 unsigned BId, 3244 IdentifierInfo *FnName) { 3245 assert(BId != 0); 3246 3247 // It is possible to have a non-standard definition of memset. Validate 3248 // we have enough arguments, and if not, abort further checking. 3249 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 3250 if (Call->getNumArgs() < ExpectedNumArgs) 3251 return; 3252 3253 unsigned LastArg = (BId == Builtin::BImemset || 3254 BId == Builtin::BIstrndup ? 1 : 2); 3255 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 3256 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 3257 3258 // We have special checking when the length is a sizeof expression. 3259 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 3260 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 3261 llvm::FoldingSetNodeID SizeOfArgID; 3262 3263 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 3264 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 3265 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 3266 3267 QualType DestTy = Dest->getType(); 3268 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 3269 QualType PointeeTy = DestPtrTy->getPointeeType(); 3270 3271 // Never warn about void type pointers. This can be used to suppress 3272 // false positives. 3273 if (PointeeTy->isVoidType()) 3274 continue; 3275 3276 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 3277 // actually comparing the expressions for equality. Because computing the 3278 // expression IDs can be expensive, we only do this if the diagnostic is 3279 // enabled. 3280 if (SizeOfArg && 3281 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 3282 SizeOfArg->getExprLoc())) { 3283 // We only compute IDs for expressions if the warning is enabled, and 3284 // cache the sizeof arg's ID. 3285 if (SizeOfArgID == llvm::FoldingSetNodeID()) 3286 SizeOfArg->Profile(SizeOfArgID, Context, true); 3287 llvm::FoldingSetNodeID DestID; 3288 Dest->Profile(DestID, Context, true); 3289 if (DestID == SizeOfArgID) { 3290 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 3291 // over sizeof(src) as well. 3292 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 3293 StringRef ReadableName = FnName->getName(); 3294 3295 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 3296 if (UnaryOp->getOpcode() == UO_AddrOf) 3297 ActionIdx = 1; // If its an address-of operator, just remove it. 3298 if (!PointeeTy->isIncompleteType() && 3299 (Context.getTypeSize(PointeeTy) == Context.getCharWidth())) 3300 ActionIdx = 2; // If the pointee's size is sizeof(char), 3301 // suggest an explicit length. 3302 3303 // If the function is defined as a builtin macro, do not show macro 3304 // expansion. 3305 SourceLocation SL = SizeOfArg->getExprLoc(); 3306 SourceRange DSR = Dest->getSourceRange(); 3307 SourceRange SSR = SizeOfArg->getSourceRange(); 3308 SourceManager &SM = PP.getSourceManager(); 3309 3310 if (SM.isMacroArgExpansion(SL)) { 3311 ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts); 3312 SL = SM.getSpellingLoc(SL); 3313 DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()), 3314 SM.getSpellingLoc(DSR.getEnd())); 3315 SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()), 3316 SM.getSpellingLoc(SSR.getEnd())); 3317 } 3318 3319 DiagRuntimeBehavior(SL, SizeOfArg, 3320 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 3321 << ReadableName 3322 << PointeeTy 3323 << DestTy 3324 << DSR 3325 << SSR); 3326 DiagRuntimeBehavior(SL, SizeOfArg, 3327 PDiag(diag::warn_sizeof_pointer_expr_memaccess_note) 3328 << ActionIdx 3329 << SSR); 3330 3331 break; 3332 } 3333 } 3334 3335 // Also check for cases where the sizeof argument is the exact same 3336 // type as the memory argument, and where it points to a user-defined 3337 // record type. 3338 if (SizeOfArgTy != QualType()) { 3339 if (PointeeTy->isRecordType() && 3340 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 3341 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 3342 PDiag(diag::warn_sizeof_pointer_type_memaccess) 3343 << FnName << SizeOfArgTy << ArgIdx 3344 << PointeeTy << Dest->getSourceRange() 3345 << LenExpr->getSourceRange()); 3346 break; 3347 } 3348 } 3349 3350 // Always complain about dynamic classes. 3351 if (isDynamicClassType(PointeeTy)) { 3352 3353 unsigned OperationType = 0; 3354 // "overwritten" if we're warning about the destination for any call 3355 // but memcmp; otherwise a verb appropriate to the call. 3356 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 3357 if (BId == Builtin::BImemcpy) 3358 OperationType = 1; 3359 else if(BId == Builtin::BImemmove) 3360 OperationType = 2; 3361 else if (BId == Builtin::BImemcmp) 3362 OperationType = 3; 3363 } 3364 3365 DiagRuntimeBehavior( 3366 Dest->getExprLoc(), Dest, 3367 PDiag(diag::warn_dyn_class_memaccess) 3368 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 3369 << FnName << PointeeTy 3370 << OperationType 3371 << Call->getCallee()->getSourceRange()); 3372 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 3373 BId != Builtin::BImemset) 3374 DiagRuntimeBehavior( 3375 Dest->getExprLoc(), Dest, 3376 PDiag(diag::warn_arc_object_memaccess) 3377 << ArgIdx << FnName << PointeeTy 3378 << Call->getCallee()->getSourceRange()); 3379 else 3380 continue; 3381 3382 DiagRuntimeBehavior( 3383 Dest->getExprLoc(), Dest, 3384 PDiag(diag::note_bad_memaccess_silence) 3385 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 3386 break; 3387 } 3388 } 3389} 3390 3391// A little helper routine: ignore addition and subtraction of integer literals. 3392// This intentionally does not ignore all integer constant expressions because 3393// we don't want to remove sizeof(). 3394static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 3395 Ex = Ex->IgnoreParenCasts(); 3396 3397 for (;;) { 3398 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 3399 if (!BO || !BO->isAdditiveOp()) 3400 break; 3401 3402 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 3403 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 3404 3405 if (isa<IntegerLiteral>(RHS)) 3406 Ex = LHS; 3407 else if (isa<IntegerLiteral>(LHS)) 3408 Ex = RHS; 3409 else 3410 break; 3411 } 3412 3413 return Ex; 3414} 3415 3416static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty, 3417 ASTContext &Context) { 3418 // Only handle constant-sized or VLAs, but not flexible members. 3419 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) { 3420 // Only issue the FIXIT for arrays of size > 1. 3421 if (CAT->getSize().getSExtValue() <= 1) 3422 return false; 3423 } else if (!Ty->isVariableArrayType()) { 3424 return false; 3425 } 3426 return true; 3427} 3428 3429// Warn if the user has made the 'size' argument to strlcpy or strlcat 3430// be the size of the source, instead of the destination. 3431void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 3432 IdentifierInfo *FnName) { 3433 3434 // Don't crash if the user has the wrong number of arguments 3435 if (Call->getNumArgs() != 3) 3436 return; 3437 3438 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 3439 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 3440 const Expr *CompareWithSrc = NULL; 3441 3442 // Look for 'strlcpy(dst, x, sizeof(x))' 3443 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 3444 CompareWithSrc = Ex; 3445 else { 3446 // Look for 'strlcpy(dst, x, strlen(x))' 3447 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 3448 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 3449 && SizeCall->getNumArgs() == 1) 3450 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 3451 } 3452 } 3453 3454 if (!CompareWithSrc) 3455 return; 3456 3457 // Determine if the argument to sizeof/strlen is equal to the source 3458 // argument. In principle there's all kinds of things you could do 3459 // here, for instance creating an == expression and evaluating it with 3460 // EvaluateAsBooleanCondition, but this uses a more direct technique: 3461 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 3462 if (!SrcArgDRE) 3463 return; 3464 3465 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 3466 if (!CompareWithSrcDRE || 3467 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 3468 return; 3469 3470 const Expr *OriginalSizeArg = Call->getArg(2); 3471 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 3472 << OriginalSizeArg->getSourceRange() << FnName; 3473 3474 // Output a FIXIT hint if the destination is an array (rather than a 3475 // pointer to an array). This could be enhanced to handle some 3476 // pointers if we know the actual size, like if DstArg is 'array+2' 3477 // we could say 'sizeof(array)-2'. 3478 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 3479 if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context)) 3480 return; 3481 3482 SmallString<128> sizeString; 3483 llvm::raw_svector_ostream OS(sizeString); 3484 OS << "sizeof("; 3485 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3486 OS << ")"; 3487 3488 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 3489 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 3490 OS.str()); 3491} 3492 3493/// Check if two expressions refer to the same declaration. 3494static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 3495 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 3496 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 3497 return D1->getDecl() == D2->getDecl(); 3498 return false; 3499} 3500 3501static const Expr *getStrlenExprArg(const Expr *E) { 3502 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 3503 const FunctionDecl *FD = CE->getDirectCallee(); 3504 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 3505 return 0; 3506 return CE->getArg(0)->IgnoreParenCasts(); 3507 } 3508 return 0; 3509} 3510 3511// Warn on anti-patterns as the 'size' argument to strncat. 3512// The correct size argument should look like following: 3513// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 3514void Sema::CheckStrncatArguments(const CallExpr *CE, 3515 IdentifierInfo *FnName) { 3516 // Don't crash if the user has the wrong number of arguments. 3517 if (CE->getNumArgs() < 3) 3518 return; 3519 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 3520 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 3521 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 3522 3523 // Identify common expressions, which are wrongly used as the size argument 3524 // to strncat and may lead to buffer overflows. 3525 unsigned PatternType = 0; 3526 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 3527 // - sizeof(dst) 3528 if (referToTheSameDecl(SizeOfArg, DstArg)) 3529 PatternType = 1; 3530 // - sizeof(src) 3531 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 3532 PatternType = 2; 3533 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 3534 if (BE->getOpcode() == BO_Sub) { 3535 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 3536 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 3537 // - sizeof(dst) - strlen(dst) 3538 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 3539 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 3540 PatternType = 1; 3541 // - sizeof(src) - (anything) 3542 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 3543 PatternType = 2; 3544 } 3545 } 3546 3547 if (PatternType == 0) 3548 return; 3549 3550 // Generate the diagnostic. 3551 SourceLocation SL = LenArg->getLocStart(); 3552 SourceRange SR = LenArg->getSourceRange(); 3553 SourceManager &SM = PP.getSourceManager(); 3554 3555 // If the function is defined as a builtin macro, do not show macro expansion. 3556 if (SM.isMacroArgExpansion(SL)) { 3557 SL = SM.getSpellingLoc(SL); 3558 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 3559 SM.getSpellingLoc(SR.getEnd())); 3560 } 3561 3562 // Check if the destination is an array (rather than a pointer to an array). 3563 QualType DstTy = DstArg->getType(); 3564 bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy, 3565 Context); 3566 if (!isKnownSizeArray) { 3567 if (PatternType == 1) 3568 Diag(SL, diag::warn_strncat_wrong_size) << SR; 3569 else 3570 Diag(SL, diag::warn_strncat_src_size) << SR; 3571 return; 3572 } 3573 3574 if (PatternType == 1) 3575 Diag(SL, diag::warn_strncat_large_size) << SR; 3576 else 3577 Diag(SL, diag::warn_strncat_src_size) << SR; 3578 3579 SmallString<128> sizeString; 3580 llvm::raw_svector_ostream OS(sizeString); 3581 OS << "sizeof("; 3582 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3583 OS << ") - "; 3584 OS << "strlen("; 3585 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3586 OS << ") - 1"; 3587 3588 Diag(SL, diag::note_strncat_wrong_size) 3589 << FixItHint::CreateReplacement(SR, OS.str()); 3590} 3591 3592//===--- CHECK: Return Address of Stack Variable --------------------------===// 3593 3594static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3595 Decl *ParentDecl); 3596static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3597 Decl *ParentDecl); 3598 3599/// CheckReturnStackAddr - Check if a return statement returns the address 3600/// of a stack variable. 3601void 3602Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3603 SourceLocation ReturnLoc) { 3604 3605 Expr *stackE = 0; 3606 SmallVector<DeclRefExpr *, 8> refVars; 3607 3608 // Perform checking for returned stack addresses, local blocks, 3609 // label addresses or references to temporaries. 3610 if (lhsType->isPointerType() || 3611 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 3612 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 3613 } else if (lhsType->isReferenceType()) { 3614 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 3615 } 3616 3617 if (stackE == 0) 3618 return; // Nothing suspicious was found. 3619 3620 SourceLocation diagLoc; 3621 SourceRange diagRange; 3622 if (refVars.empty()) { 3623 diagLoc = stackE->getLocStart(); 3624 diagRange = stackE->getSourceRange(); 3625 } else { 3626 // We followed through a reference variable. 'stackE' contains the 3627 // problematic expression but we will warn at the return statement pointing 3628 // at the reference variable. We will later display the "trail" of 3629 // reference variables using notes. 3630 diagLoc = refVars[0]->getLocStart(); 3631 diagRange = refVars[0]->getSourceRange(); 3632 } 3633 3634 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 3635 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 3636 : diag::warn_ret_stack_addr) 3637 << DR->getDecl()->getDeclName() << diagRange; 3638 } else if (isa<BlockExpr>(stackE)) { // local block. 3639 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 3640 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 3641 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 3642 } else { // local temporary. 3643 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 3644 : diag::warn_ret_local_temp_addr) 3645 << diagRange; 3646 } 3647 3648 // Display the "trail" of reference variables that we followed until we 3649 // found the problematic expression using notes. 3650 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 3651 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 3652 // If this var binds to another reference var, show the range of the next 3653 // var, otherwise the var binds to the problematic expression, in which case 3654 // show the range of the expression. 3655 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 3656 : stackE->getSourceRange(); 3657 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 3658 << VD->getDeclName() << range; 3659 } 3660} 3661 3662/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 3663/// check if the expression in a return statement evaluates to an address 3664/// to a location on the stack, a local block, an address of a label, or a 3665/// reference to local temporary. The recursion is used to traverse the 3666/// AST of the return expression, with recursion backtracking when we 3667/// encounter a subexpression that (1) clearly does not lead to one of the 3668/// above problematic expressions (2) is something we cannot determine leads to 3669/// a problematic expression based on such local checking. 3670/// 3671/// Both EvalAddr and EvalVal follow through reference variables to evaluate 3672/// the expression that they point to. Such variables are added to the 3673/// 'refVars' vector so that we know what the reference variable "trail" was. 3674/// 3675/// EvalAddr processes expressions that are pointers that are used as 3676/// references (and not L-values). EvalVal handles all other values. 3677/// At the base case of the recursion is a check for the above problematic 3678/// expressions. 3679/// 3680/// This implementation handles: 3681/// 3682/// * pointer-to-pointer casts 3683/// * implicit conversions from array references to pointers 3684/// * taking the address of fields 3685/// * arbitrary interplay between "&" and "*" operators 3686/// * pointer arithmetic from an address of a stack variable 3687/// * taking the address of an array element where the array is on the stack 3688static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3689 Decl *ParentDecl) { 3690 if (E->isTypeDependent()) 3691 return NULL; 3692 3693 // We should only be called for evaluating pointer expressions. 3694 assert((E->getType()->isAnyPointerType() || 3695 E->getType()->isBlockPointerType() || 3696 E->getType()->isObjCQualifiedIdType()) && 3697 "EvalAddr only works on pointers"); 3698 3699 E = E->IgnoreParens(); 3700 3701 // Our "symbolic interpreter" is just a dispatch off the currently 3702 // viewed AST node. We then recursively traverse the AST by calling 3703 // EvalAddr and EvalVal appropriately. 3704 switch (E->getStmtClass()) { 3705 case Stmt::DeclRefExprClass: { 3706 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3707 3708 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3709 // If this is a reference variable, follow through to the expression that 3710 // it points to. 3711 if (V->hasLocalStorage() && 3712 V->getType()->isReferenceType() && V->hasInit()) { 3713 // Add the reference variable to the "trail". 3714 refVars.push_back(DR); 3715 return EvalAddr(V->getInit(), refVars, ParentDecl); 3716 } 3717 3718 return NULL; 3719 } 3720 3721 case Stmt::UnaryOperatorClass: { 3722 // The only unary operator that make sense to handle here 3723 // is AddrOf. All others don't make sense as pointers. 3724 UnaryOperator *U = cast<UnaryOperator>(E); 3725 3726 if (U->getOpcode() == UO_AddrOf) 3727 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 3728 else 3729 return NULL; 3730 } 3731 3732 case Stmt::BinaryOperatorClass: { 3733 // Handle pointer arithmetic. All other binary operators are not valid 3734 // in this context. 3735 BinaryOperator *B = cast<BinaryOperator>(E); 3736 BinaryOperatorKind op = B->getOpcode(); 3737 3738 if (op != BO_Add && op != BO_Sub) 3739 return NULL; 3740 3741 Expr *Base = B->getLHS(); 3742 3743 // Determine which argument is the real pointer base. It could be 3744 // the RHS argument instead of the LHS. 3745 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 3746 3747 assert (Base->getType()->isPointerType()); 3748 return EvalAddr(Base, refVars, ParentDecl); 3749 } 3750 3751 // For conditional operators we need to see if either the LHS or RHS are 3752 // valid DeclRefExpr*s. If one of them is valid, we return it. 3753 case Stmt::ConditionalOperatorClass: { 3754 ConditionalOperator *C = cast<ConditionalOperator>(E); 3755 3756 // Handle the GNU extension for missing LHS. 3757 if (Expr *lhsExpr = C->getLHS()) { 3758 // In C++, we can have a throw-expression, which has 'void' type. 3759 if (!lhsExpr->getType()->isVoidType()) 3760 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 3761 return LHS; 3762 } 3763 3764 // In C++, we can have a throw-expression, which has 'void' type. 3765 if (C->getRHS()->getType()->isVoidType()) 3766 return NULL; 3767 3768 return EvalAddr(C->getRHS(), refVars, ParentDecl); 3769 } 3770 3771 case Stmt::BlockExprClass: 3772 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 3773 return E; // local block. 3774 return NULL; 3775 3776 case Stmt::AddrLabelExprClass: 3777 return E; // address of label. 3778 3779 case Stmt::ExprWithCleanupsClass: 3780 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 3781 ParentDecl); 3782 3783 // For casts, we need to handle conversions from arrays to 3784 // pointer values, and pointer-to-pointer conversions. 3785 case Stmt::ImplicitCastExprClass: 3786 case Stmt::CStyleCastExprClass: 3787 case Stmt::CXXFunctionalCastExprClass: 3788 case Stmt::ObjCBridgedCastExprClass: 3789 case Stmt::CXXStaticCastExprClass: 3790 case Stmt::CXXDynamicCastExprClass: 3791 case Stmt::CXXConstCastExprClass: 3792 case Stmt::CXXReinterpretCastExprClass: { 3793 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 3794 switch (cast<CastExpr>(E)->getCastKind()) { 3795 case CK_BitCast: 3796 case CK_LValueToRValue: 3797 case CK_NoOp: 3798 case CK_BaseToDerived: 3799 case CK_DerivedToBase: 3800 case CK_UncheckedDerivedToBase: 3801 case CK_Dynamic: 3802 case CK_CPointerToObjCPointerCast: 3803 case CK_BlockPointerToObjCPointerCast: 3804 case CK_AnyPointerToBlockPointerCast: 3805 return EvalAddr(SubExpr, refVars, ParentDecl); 3806 3807 case CK_ArrayToPointerDecay: 3808 return EvalVal(SubExpr, refVars, ParentDecl); 3809 3810 default: 3811 return 0; 3812 } 3813 } 3814 3815 case Stmt::MaterializeTemporaryExprClass: 3816 if (Expr *Result = EvalAddr( 3817 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3818 refVars, ParentDecl)) 3819 return Result; 3820 3821 return E; 3822 3823 // Everything else: we simply don't reason about them. 3824 default: 3825 return NULL; 3826 } 3827} 3828 3829 3830/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3831/// See the comments for EvalAddr for more details. 3832static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3833 Decl *ParentDecl) { 3834do { 3835 // We should only be called for evaluating non-pointer expressions, or 3836 // expressions with a pointer type that are not used as references but instead 3837 // are l-values (e.g., DeclRefExpr with a pointer type). 3838 3839 // Our "symbolic interpreter" is just a dispatch off the currently 3840 // viewed AST node. We then recursively traverse the AST by calling 3841 // EvalAddr and EvalVal appropriately. 3842 3843 E = E->IgnoreParens(); 3844 switch (E->getStmtClass()) { 3845 case Stmt::ImplicitCastExprClass: { 3846 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3847 if (IE->getValueKind() == VK_LValue) { 3848 E = IE->getSubExpr(); 3849 continue; 3850 } 3851 return NULL; 3852 } 3853 3854 case Stmt::ExprWithCleanupsClass: 3855 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 3856 3857 case Stmt::DeclRefExprClass: { 3858 // When we hit a DeclRefExpr we are looking at code that refers to a 3859 // variable's name. If it's not a reference variable we check if it has 3860 // local storage within the function, and if so, return the expression. 3861 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3862 3863 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 3864 // Check if it refers to itself, e.g. "int& i = i;". 3865 if (V == ParentDecl) 3866 return DR; 3867 3868 if (V->hasLocalStorage()) { 3869 if (!V->getType()->isReferenceType()) 3870 return DR; 3871 3872 // Reference variable, follow through to the expression that 3873 // it points to. 3874 if (V->hasInit()) { 3875 // Add the reference variable to the "trail". 3876 refVars.push_back(DR); 3877 return EvalVal(V->getInit(), refVars, V); 3878 } 3879 } 3880 } 3881 3882 return NULL; 3883 } 3884 3885 case Stmt::UnaryOperatorClass: { 3886 // The only unary operator that make sense to handle here 3887 // is Deref. All others don't resolve to a "name." This includes 3888 // handling all sorts of rvalues passed to a unary operator. 3889 UnaryOperator *U = cast<UnaryOperator>(E); 3890 3891 if (U->getOpcode() == UO_Deref) 3892 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 3893 3894 return NULL; 3895 } 3896 3897 case Stmt::ArraySubscriptExprClass: { 3898 // Array subscripts are potential references to data on the stack. We 3899 // retrieve the DeclRefExpr* for the array variable if it indeed 3900 // has local storage. 3901 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 3902 } 3903 3904 case Stmt::ConditionalOperatorClass: { 3905 // For conditional operators we need to see if either the LHS or RHS are 3906 // non-NULL Expr's. If one is non-NULL, we return it. 3907 ConditionalOperator *C = cast<ConditionalOperator>(E); 3908 3909 // Handle the GNU extension for missing LHS. 3910 if (Expr *lhsExpr = C->getLHS()) 3911 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 3912 return LHS; 3913 3914 return EvalVal(C->getRHS(), refVars, ParentDecl); 3915 } 3916 3917 // Accesses to members are potential references to data on the stack. 3918 case Stmt::MemberExprClass: { 3919 MemberExpr *M = cast<MemberExpr>(E); 3920 3921 // Check for indirect access. We only want direct field accesses. 3922 if (M->isArrow()) 3923 return NULL; 3924 3925 // Check whether the member type is itself a reference, in which case 3926 // we're not going to refer to the member, but to what the member refers to. 3927 if (M->getMemberDecl()->getType()->isReferenceType()) 3928 return NULL; 3929 3930 return EvalVal(M->getBase(), refVars, ParentDecl); 3931 } 3932 3933 case Stmt::MaterializeTemporaryExprClass: 3934 if (Expr *Result = EvalVal( 3935 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3936 refVars, ParentDecl)) 3937 return Result; 3938 3939 return E; 3940 3941 default: 3942 // Check that we don't return or take the address of a reference to a 3943 // temporary. This is only useful in C++. 3944 if (!E->isTypeDependent() && E->isRValue()) 3945 return E; 3946 3947 // Everything else: we simply don't reason about them. 3948 return NULL; 3949 } 3950} while (true); 3951} 3952 3953//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3954 3955/// Check for comparisons of floating point operands using != and ==. 3956/// Issue a warning if these are no self-comparisons, as they are not likely 3957/// to do what the programmer intended. 3958void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3959 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3960 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3961 3962 // Special case: check for x == x (which is OK). 3963 // Do not emit warnings for such cases. 3964 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3965 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3966 if (DRL->getDecl() == DRR->getDecl()) 3967 return; 3968 3969 3970 // Special case: check for comparisons against literals that can be exactly 3971 // represented by APFloat. In such cases, do not emit a warning. This 3972 // is a heuristic: often comparison against such literals are used to 3973 // detect if a value in a variable has not changed. This clearly can 3974 // lead to false negatives. 3975 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3976 if (FLL->isExact()) 3977 return; 3978 } else 3979 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)) 3980 if (FLR->isExact()) 3981 return; 3982 3983 // Check for comparisons with builtin types. 3984 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3985 if (CL->isBuiltinCall()) 3986 return; 3987 3988 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3989 if (CR->isBuiltinCall()) 3990 return; 3991 3992 // Emit the diagnostic. 3993 Diag(Loc, diag::warn_floatingpoint_eq) 3994 << LHS->getSourceRange() << RHS->getSourceRange(); 3995} 3996 3997//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3998//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3999 4000namespace { 4001 4002/// Structure recording the 'active' range of an integer-valued 4003/// expression. 4004struct IntRange { 4005 /// The number of bits active in the int. 4006 unsigned Width; 4007 4008 /// True if the int is known not to have negative values. 4009 bool NonNegative; 4010 4011 IntRange(unsigned Width, bool NonNegative) 4012 : Width(Width), NonNegative(NonNegative) 4013 {} 4014 4015 /// Returns the range of the bool type. 4016 static IntRange forBoolType() { 4017 return IntRange(1, true); 4018 } 4019 4020 /// Returns the range of an opaque value of the given integral type. 4021 static IntRange forValueOfType(ASTContext &C, QualType T) { 4022 return forValueOfCanonicalType(C, 4023 T->getCanonicalTypeInternal().getTypePtr()); 4024 } 4025 4026 /// Returns the range of an opaque value of a canonical integral type. 4027 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 4028 assert(T->isCanonicalUnqualified()); 4029 4030 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4031 T = VT->getElementType().getTypePtr(); 4032 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4033 T = CT->getElementType().getTypePtr(); 4034 4035 // For enum types, use the known bit width of the enumerators. 4036 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 4037 EnumDecl *Enum = ET->getDecl(); 4038 if (!Enum->isCompleteDefinition()) 4039 return IntRange(C.getIntWidth(QualType(T, 0)), false); 4040 4041 unsigned NumPositive = Enum->getNumPositiveBits(); 4042 unsigned NumNegative = Enum->getNumNegativeBits(); 4043 4044 if (NumNegative == 0) 4045 return IntRange(NumPositive, true/*NonNegative*/); 4046 else 4047 return IntRange(std::max(NumPositive + 1, NumNegative), 4048 false/*NonNegative*/); 4049 } 4050 4051 const BuiltinType *BT = cast<BuiltinType>(T); 4052 assert(BT->isInteger()); 4053 4054 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4055 } 4056 4057 /// Returns the "target" range of a canonical integral type, i.e. 4058 /// the range of values expressible in the type. 4059 /// 4060 /// This matches forValueOfCanonicalType except that enums have the 4061 /// full range of their type, not the range of their enumerators. 4062 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 4063 assert(T->isCanonicalUnqualified()); 4064 4065 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4066 T = VT->getElementType().getTypePtr(); 4067 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4068 T = CT->getElementType().getTypePtr(); 4069 if (const EnumType *ET = dyn_cast<EnumType>(T)) 4070 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 4071 4072 const BuiltinType *BT = cast<BuiltinType>(T); 4073 assert(BT->isInteger()); 4074 4075 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4076 } 4077 4078 /// Returns the supremum of two ranges: i.e. their conservative merge. 4079 static IntRange join(IntRange L, IntRange R) { 4080 return IntRange(std::max(L.Width, R.Width), 4081 L.NonNegative && R.NonNegative); 4082 } 4083 4084 /// Returns the infinum of two ranges: i.e. their aggressive merge. 4085 static IntRange meet(IntRange L, IntRange R) { 4086 return IntRange(std::min(L.Width, R.Width), 4087 L.NonNegative || R.NonNegative); 4088 } 4089}; 4090 4091static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 4092 unsigned MaxWidth) { 4093 if (value.isSigned() && value.isNegative()) 4094 return IntRange(value.getMinSignedBits(), false); 4095 4096 if (value.getBitWidth() > MaxWidth) 4097 value = value.trunc(MaxWidth); 4098 4099 // isNonNegative() just checks the sign bit without considering 4100 // signedness. 4101 return IntRange(value.getActiveBits(), true); 4102} 4103 4104static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 4105 unsigned MaxWidth) { 4106 if (result.isInt()) 4107 return GetValueRange(C, result.getInt(), MaxWidth); 4108 4109 if (result.isVector()) { 4110 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 4111 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 4112 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 4113 R = IntRange::join(R, El); 4114 } 4115 return R; 4116 } 4117 4118 if (result.isComplexInt()) { 4119 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 4120 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 4121 return IntRange::join(R, I); 4122 } 4123 4124 // This can happen with lossless casts to intptr_t of "based" lvalues. 4125 // Assume it might use arbitrary bits. 4126 // FIXME: The only reason we need to pass the type in here is to get 4127 // the sign right on this one case. It would be nice if APValue 4128 // preserved this. 4129 assert(result.isLValue() || result.isAddrLabelDiff()); 4130 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 4131} 4132 4133/// Pseudo-evaluate the given integer expression, estimating the 4134/// range of values it might take. 4135/// 4136/// \param MaxWidth - the width to which the value will be truncated 4137static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 4138 E = E->IgnoreParens(); 4139 4140 // Try a full evaluation first. 4141 Expr::EvalResult result; 4142 if (E->EvaluateAsRValue(result, C)) 4143 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 4144 4145 // I think we only want to look through implicit casts here; if the 4146 // user has an explicit widening cast, we should treat the value as 4147 // being of the new, wider type. 4148 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 4149 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 4150 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 4151 4152 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 4153 4154 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 4155 4156 // Assume that non-integer casts can span the full range of the type. 4157 if (!isIntegerCast) 4158 return OutputTypeRange; 4159 4160 IntRange SubRange 4161 = GetExprRange(C, CE->getSubExpr(), 4162 std::min(MaxWidth, OutputTypeRange.Width)); 4163 4164 // Bail out if the subexpr's range is as wide as the cast type. 4165 if (SubRange.Width >= OutputTypeRange.Width) 4166 return OutputTypeRange; 4167 4168 // Otherwise, we take the smaller width, and we're non-negative if 4169 // either the output type or the subexpr is. 4170 return IntRange(SubRange.Width, 4171 SubRange.NonNegative || OutputTypeRange.NonNegative); 4172 } 4173 4174 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 4175 // If we can fold the condition, just take that operand. 4176 bool CondResult; 4177 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 4178 return GetExprRange(C, CondResult ? CO->getTrueExpr() 4179 : CO->getFalseExpr(), 4180 MaxWidth); 4181 4182 // Otherwise, conservatively merge. 4183 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 4184 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 4185 return IntRange::join(L, R); 4186 } 4187 4188 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4189 switch (BO->getOpcode()) { 4190 4191 // Boolean-valued operations are single-bit and positive. 4192 case BO_LAnd: 4193 case BO_LOr: 4194 case BO_LT: 4195 case BO_GT: 4196 case BO_LE: 4197 case BO_GE: 4198 case BO_EQ: 4199 case BO_NE: 4200 return IntRange::forBoolType(); 4201 4202 // The type of the assignments is the type of the LHS, so the RHS 4203 // is not necessarily the same type. 4204 case BO_MulAssign: 4205 case BO_DivAssign: 4206 case BO_RemAssign: 4207 case BO_AddAssign: 4208 case BO_SubAssign: 4209 case BO_XorAssign: 4210 case BO_OrAssign: 4211 // TODO: bitfields? 4212 return IntRange::forValueOfType(C, E->getType()); 4213 4214 // Simple assignments just pass through the RHS, which will have 4215 // been coerced to the LHS type. 4216 case BO_Assign: 4217 // TODO: bitfields? 4218 return GetExprRange(C, BO->getRHS(), MaxWidth); 4219 4220 // Operations with opaque sources are black-listed. 4221 case BO_PtrMemD: 4222 case BO_PtrMemI: 4223 return IntRange::forValueOfType(C, E->getType()); 4224 4225 // Bitwise-and uses the *infinum* of the two source ranges. 4226 case BO_And: 4227 case BO_AndAssign: 4228 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 4229 GetExprRange(C, BO->getRHS(), MaxWidth)); 4230 4231 // Left shift gets black-listed based on a judgement call. 4232 case BO_Shl: 4233 // ...except that we want to treat '1 << (blah)' as logically 4234 // positive. It's an important idiom. 4235 if (IntegerLiteral *I 4236 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 4237 if (I->getValue() == 1) { 4238 IntRange R = IntRange::forValueOfType(C, E->getType()); 4239 return IntRange(R.Width, /*NonNegative*/ true); 4240 } 4241 } 4242 // fallthrough 4243 4244 case BO_ShlAssign: 4245 return IntRange::forValueOfType(C, E->getType()); 4246 4247 // Right shift by a constant can narrow its left argument. 4248 case BO_Shr: 4249 case BO_ShrAssign: { 4250 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4251 4252 // If the shift amount is a positive constant, drop the width by 4253 // that much. 4254 llvm::APSInt shift; 4255 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 4256 shift.isNonNegative()) { 4257 unsigned zext = shift.getZExtValue(); 4258 if (zext >= L.Width) 4259 L.Width = (L.NonNegative ? 0 : 1); 4260 else 4261 L.Width -= zext; 4262 } 4263 4264 return L; 4265 } 4266 4267 // Comma acts as its right operand. 4268 case BO_Comma: 4269 return GetExprRange(C, BO->getRHS(), MaxWidth); 4270 4271 // Black-list pointer subtractions. 4272 case BO_Sub: 4273 if (BO->getLHS()->getType()->isPointerType()) 4274 return IntRange::forValueOfType(C, E->getType()); 4275 break; 4276 4277 // The width of a division result is mostly determined by the size 4278 // of the LHS. 4279 case BO_Div: { 4280 // Don't 'pre-truncate' the operands. 4281 unsigned opWidth = C.getIntWidth(E->getType()); 4282 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4283 4284 // If the divisor is constant, use that. 4285 llvm::APSInt divisor; 4286 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 4287 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 4288 if (log2 >= L.Width) 4289 L.Width = (L.NonNegative ? 0 : 1); 4290 else 4291 L.Width = std::min(L.Width - log2, MaxWidth); 4292 return L; 4293 } 4294 4295 // Otherwise, just use the LHS's width. 4296 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4297 return IntRange(L.Width, L.NonNegative && R.NonNegative); 4298 } 4299 4300 // The result of a remainder can't be larger than the result of 4301 // either side. 4302 case BO_Rem: { 4303 // Don't 'pre-truncate' the operands. 4304 unsigned opWidth = C.getIntWidth(E->getType()); 4305 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4306 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4307 4308 IntRange meet = IntRange::meet(L, R); 4309 meet.Width = std::min(meet.Width, MaxWidth); 4310 return meet; 4311 } 4312 4313 // The default behavior is okay for these. 4314 case BO_Mul: 4315 case BO_Add: 4316 case BO_Xor: 4317 case BO_Or: 4318 break; 4319 } 4320 4321 // The default case is to treat the operation as if it were closed 4322 // on the narrowest type that encompasses both operands. 4323 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4324 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 4325 return IntRange::join(L, R); 4326 } 4327 4328 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 4329 switch (UO->getOpcode()) { 4330 // Boolean-valued operations are white-listed. 4331 case UO_LNot: 4332 return IntRange::forBoolType(); 4333 4334 // Operations with opaque sources are black-listed. 4335 case UO_Deref: 4336 case UO_AddrOf: // should be impossible 4337 return IntRange::forValueOfType(C, E->getType()); 4338 4339 default: 4340 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 4341 } 4342 } 4343 4344 if (dyn_cast<OffsetOfExpr>(E)) { 4345 IntRange::forValueOfType(C, E->getType()); 4346 } 4347 4348 if (FieldDecl *BitField = E->getSourceBitField()) 4349 return IntRange(BitField->getBitWidthValue(C), 4350 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 4351 4352 return IntRange::forValueOfType(C, E->getType()); 4353} 4354 4355static IntRange GetExprRange(ASTContext &C, Expr *E) { 4356 return GetExprRange(C, E, C.getIntWidth(E->getType())); 4357} 4358 4359/// Checks whether the given value, which currently has the given 4360/// source semantics, has the same value when coerced through the 4361/// target semantics. 4362static bool IsSameFloatAfterCast(const llvm::APFloat &value, 4363 const llvm::fltSemantics &Src, 4364 const llvm::fltSemantics &Tgt) { 4365 llvm::APFloat truncated = value; 4366 4367 bool ignored; 4368 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 4369 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 4370 4371 return truncated.bitwiseIsEqual(value); 4372} 4373 4374/// Checks whether the given value, which currently has the given 4375/// source semantics, has the same value when coerced through the 4376/// target semantics. 4377/// 4378/// The value might be a vector of floats (or a complex number). 4379static bool IsSameFloatAfterCast(const APValue &value, 4380 const llvm::fltSemantics &Src, 4381 const llvm::fltSemantics &Tgt) { 4382 if (value.isFloat()) 4383 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 4384 4385 if (value.isVector()) { 4386 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 4387 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 4388 return false; 4389 return true; 4390 } 4391 4392 assert(value.isComplexFloat()); 4393 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 4394 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 4395} 4396 4397static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 4398 4399static bool IsZero(Sema &S, Expr *E) { 4400 // Suppress cases where we are comparing against an enum constant. 4401 if (const DeclRefExpr *DR = 4402 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 4403 if (isa<EnumConstantDecl>(DR->getDecl())) 4404 return false; 4405 4406 // Suppress cases where the '0' value is expanded from a macro. 4407 if (E->getLocStart().isMacroID()) 4408 return false; 4409 4410 llvm::APSInt Value; 4411 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 4412} 4413 4414static bool HasEnumType(Expr *E) { 4415 // Strip off implicit integral promotions. 4416 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 4417 if (ICE->getCastKind() != CK_IntegralCast && 4418 ICE->getCastKind() != CK_NoOp) 4419 break; 4420 E = ICE->getSubExpr(); 4421 } 4422 4423 return E->getType()->isEnumeralType(); 4424} 4425 4426static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 4427 BinaryOperatorKind op = E->getOpcode(); 4428 if (E->isValueDependent()) 4429 return; 4430 4431 if (op == BO_LT && IsZero(S, E->getRHS())) { 4432 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4433 << "< 0" << "false" << HasEnumType(E->getLHS()) 4434 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4435 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 4436 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4437 << ">= 0" << "true" << HasEnumType(E->getLHS()) 4438 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4439 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 4440 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4441 << "0 >" << "false" << HasEnumType(E->getRHS()) 4442 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4443 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 4444 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4445 << "0 <=" << "true" << HasEnumType(E->getRHS()) 4446 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4447 } 4448} 4449 4450static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E, 4451 Expr *Constant, Expr *Other, 4452 llvm::APSInt Value, 4453 bool RhsConstant) { 4454 // 0 values are handled later by CheckTrivialUnsignedComparison(). 4455 if (Value == 0) 4456 return; 4457 4458 BinaryOperatorKind op = E->getOpcode(); 4459 QualType OtherT = Other->getType(); 4460 QualType ConstantT = Constant->getType(); 4461 QualType CommonT = E->getLHS()->getType(); 4462 if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT)) 4463 return; 4464 assert((OtherT->isIntegerType() && ConstantT->isIntegerType()) 4465 && "comparison with non-integer type"); 4466 4467 bool ConstantSigned = ConstantT->isSignedIntegerType(); 4468 bool CommonSigned = CommonT->isSignedIntegerType(); 4469 4470 bool EqualityOnly = false; 4471 4472 // TODO: Investigate using GetExprRange() to get tighter bounds on 4473 // on the bit ranges. 4474 IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT); 4475 unsigned OtherWidth = OtherRange.Width; 4476 4477 if (CommonSigned) { 4478 // The common type is signed, therefore no signed to unsigned conversion. 4479 if (!OtherRange.NonNegative) { 4480 // Check that the constant is representable in type OtherT. 4481 if (ConstantSigned) { 4482 if (OtherWidth >= Value.getMinSignedBits()) 4483 return; 4484 } else { // !ConstantSigned 4485 if (OtherWidth >= Value.getActiveBits() + 1) 4486 return; 4487 } 4488 } else { // !OtherSigned 4489 // Check that the constant is representable in type OtherT. 4490 // Negative values are out of range. 4491 if (ConstantSigned) { 4492 if (Value.isNonNegative() && OtherWidth >= Value.getActiveBits()) 4493 return; 4494 } else { // !ConstantSigned 4495 if (OtherWidth >= Value.getActiveBits()) 4496 return; 4497 } 4498 } 4499 } else { // !CommonSigned 4500 if (OtherRange.NonNegative) { 4501 if (OtherWidth >= Value.getActiveBits()) 4502 return; 4503 } else if (!OtherRange.NonNegative && !ConstantSigned) { 4504 // Check to see if the constant is representable in OtherT. 4505 if (OtherWidth > Value.getActiveBits()) 4506 return; 4507 // Check to see if the constant is equivalent to a negative value 4508 // cast to CommonT. 4509 if (S.Context.getIntWidth(ConstantT) == S.Context.getIntWidth(CommonT) && 4510 Value.isNegative() && Value.getMinSignedBits() <= OtherWidth) 4511 return; 4512 // The constant value rests between values that OtherT can represent after 4513 // conversion. Relational comparison still works, but equality 4514 // comparisons will be tautological. 4515 EqualityOnly = true; 4516 } else { // OtherSigned && ConstantSigned 4517 assert(0 && "Two signed types converted to unsigned types."); 4518 } 4519 } 4520 4521 bool PositiveConstant = !ConstantSigned || Value.isNonNegative(); 4522 4523 bool IsTrue = true; 4524 if (op == BO_EQ || op == BO_NE) { 4525 IsTrue = op == BO_NE; 4526 } else if (EqualityOnly) { 4527 return; 4528 } else if (RhsConstant) { 4529 if (op == BO_GT || op == BO_GE) 4530 IsTrue = !PositiveConstant; 4531 else // op == BO_LT || op == BO_LE 4532 IsTrue = PositiveConstant; 4533 } else { 4534 if (op == BO_LT || op == BO_LE) 4535 IsTrue = !PositiveConstant; 4536 else // op == BO_GT || op == BO_GE 4537 IsTrue = PositiveConstant; 4538 } 4539 4540 // If this is a comparison to an enum constant, include that 4541 // constant in the diagnostic. 4542 const EnumConstantDecl *ED = 0; 4543 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Constant)) 4544 ED = dyn_cast<EnumConstantDecl>(DR->getDecl()); 4545 4546 SmallString<64> PrettySourceValue; 4547 llvm::raw_svector_ostream OS(PrettySourceValue); 4548 if (ED) 4549 OS << '\'' << *ED << "' (" << Value << ")"; 4550 else 4551 OS << Value; 4552 4553 S.Diag(E->getOperatorLoc(), diag::warn_out_of_range_compare) 4554 << OS.str() << OtherT << IsTrue 4555 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4556} 4557 4558/// Analyze the operands of the given comparison. Implements the 4559/// fallback case from AnalyzeComparison. 4560static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 4561 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4562 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4563} 4564 4565/// \brief Implements -Wsign-compare. 4566/// 4567/// \param E the binary operator to check for warnings 4568static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 4569 // The type the comparison is being performed in. 4570 QualType T = E->getLHS()->getType(); 4571 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 4572 && "comparison with mismatched types"); 4573 if (E->isValueDependent()) 4574 return AnalyzeImpConvsInComparison(S, E); 4575 4576 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 4577 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 4578 4579 bool IsComparisonConstant = false; 4580 4581 // Check whether an integer constant comparison results in a value 4582 // of 'true' or 'false'. 4583 if (T->isIntegralType(S.Context)) { 4584 llvm::APSInt RHSValue; 4585 bool IsRHSIntegralLiteral = 4586 RHS->isIntegerConstantExpr(RHSValue, S.Context); 4587 llvm::APSInt LHSValue; 4588 bool IsLHSIntegralLiteral = 4589 LHS->isIntegerConstantExpr(LHSValue, S.Context); 4590 if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral) 4591 DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true); 4592 else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral) 4593 DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false); 4594 else 4595 IsComparisonConstant = 4596 (IsRHSIntegralLiteral && IsLHSIntegralLiteral); 4597 } else if (!T->hasUnsignedIntegerRepresentation()) 4598 IsComparisonConstant = E->isIntegerConstantExpr(S.Context); 4599 4600 // We don't do anything special if this isn't an unsigned integral 4601 // comparison: we're only interested in integral comparisons, and 4602 // signed comparisons only happen in cases we don't care to warn about. 4603 // 4604 // We also don't care about value-dependent expressions or expressions 4605 // whose result is a constant. 4606 if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant) 4607 return AnalyzeImpConvsInComparison(S, E); 4608 4609 // Check to see if one of the (unmodified) operands is of different 4610 // signedness. 4611 Expr *signedOperand, *unsignedOperand; 4612 if (LHS->getType()->hasSignedIntegerRepresentation()) { 4613 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 4614 "unsigned comparison between two signed integer expressions?"); 4615 signedOperand = LHS; 4616 unsignedOperand = RHS; 4617 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 4618 signedOperand = RHS; 4619 unsignedOperand = LHS; 4620 } else { 4621 CheckTrivialUnsignedComparison(S, E); 4622 return AnalyzeImpConvsInComparison(S, E); 4623 } 4624 4625 // Otherwise, calculate the effective range of the signed operand. 4626 IntRange signedRange = GetExprRange(S.Context, signedOperand); 4627 4628 // Go ahead and analyze implicit conversions in the operands. Note 4629 // that we skip the implicit conversions on both sides. 4630 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 4631 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 4632 4633 // If the signed range is non-negative, -Wsign-compare won't fire, 4634 // but we should still check for comparisons which are always true 4635 // or false. 4636 if (signedRange.NonNegative) 4637 return CheckTrivialUnsignedComparison(S, E); 4638 4639 // For (in)equality comparisons, if the unsigned operand is a 4640 // constant which cannot collide with a overflowed signed operand, 4641 // then reinterpreting the signed operand as unsigned will not 4642 // change the result of the comparison. 4643 if (E->isEqualityOp()) { 4644 unsigned comparisonWidth = S.Context.getIntWidth(T); 4645 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 4646 4647 // We should never be unable to prove that the unsigned operand is 4648 // non-negative. 4649 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 4650 4651 if (unsignedRange.Width < comparisonWidth) 4652 return; 4653 } 4654 4655 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 4656 S.PDiag(diag::warn_mixed_sign_comparison) 4657 << LHS->getType() << RHS->getType() 4658 << LHS->getSourceRange() << RHS->getSourceRange()); 4659} 4660 4661/// Analyzes an attempt to assign the given value to a bitfield. 4662/// 4663/// Returns true if there was something fishy about the attempt. 4664static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 4665 SourceLocation InitLoc) { 4666 assert(Bitfield->isBitField()); 4667 if (Bitfield->isInvalidDecl()) 4668 return false; 4669 4670 // White-list bool bitfields. 4671 if (Bitfield->getType()->isBooleanType()) 4672 return false; 4673 4674 // Ignore value- or type-dependent expressions. 4675 if (Bitfield->getBitWidth()->isValueDependent() || 4676 Bitfield->getBitWidth()->isTypeDependent() || 4677 Init->isValueDependent() || 4678 Init->isTypeDependent()) 4679 return false; 4680 4681 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 4682 4683 llvm::APSInt Value; 4684 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 4685 return false; 4686 4687 unsigned OriginalWidth = Value.getBitWidth(); 4688 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 4689 4690 if (OriginalWidth <= FieldWidth) 4691 return false; 4692 4693 // Compute the value which the bitfield will contain. 4694 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 4695 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 4696 4697 // Check whether the stored value is equal to the original value. 4698 TruncatedValue = TruncatedValue.extend(OriginalWidth); 4699 if (llvm::APSInt::isSameValue(Value, TruncatedValue)) 4700 return false; 4701 4702 // Special-case bitfields of width 1: booleans are naturally 0/1, and 4703 // therefore don't strictly fit into a signed bitfield of width 1. 4704 if (FieldWidth == 1 && Value == 1) 4705 return false; 4706 4707 std::string PrettyValue = Value.toString(10); 4708 std::string PrettyTrunc = TruncatedValue.toString(10); 4709 4710 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 4711 << PrettyValue << PrettyTrunc << OriginalInit->getType() 4712 << Init->getSourceRange(); 4713 4714 return true; 4715} 4716 4717/// Analyze the given simple or compound assignment for warning-worthy 4718/// operations. 4719static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 4720 // Just recurse on the LHS. 4721 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4722 4723 // We want to recurse on the RHS as normal unless we're assigning to 4724 // a bitfield. 4725 if (FieldDecl *Bitfield = E->getLHS()->getSourceBitField()) { 4726 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 4727 E->getOperatorLoc())) { 4728 // Recurse, ignoring any implicit conversions on the RHS. 4729 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 4730 E->getOperatorLoc()); 4731 } 4732 } 4733 4734 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4735} 4736 4737/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4738static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 4739 SourceLocation CContext, unsigned diag, 4740 bool pruneControlFlow = false) { 4741 if (pruneControlFlow) { 4742 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4743 S.PDiag(diag) 4744 << SourceType << T << E->getSourceRange() 4745 << SourceRange(CContext)); 4746 return; 4747 } 4748 S.Diag(E->getExprLoc(), diag) 4749 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 4750} 4751 4752/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4753static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 4754 SourceLocation CContext, unsigned diag, 4755 bool pruneControlFlow = false) { 4756 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 4757} 4758 4759/// Diagnose an implicit cast from a literal expression. Does not warn when the 4760/// cast wouldn't lose information. 4761void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 4762 SourceLocation CContext) { 4763 // Try to convert the literal exactly to an integer. If we can, don't warn. 4764 bool isExact = false; 4765 const llvm::APFloat &Value = FL->getValue(); 4766 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 4767 T->hasUnsignedIntegerRepresentation()); 4768 if (Value.convertToInteger(IntegerValue, 4769 llvm::APFloat::rmTowardZero, &isExact) 4770 == llvm::APFloat::opOK && isExact) 4771 return; 4772 4773 SmallString<16> PrettySourceValue; 4774 Value.toString(PrettySourceValue); 4775 SmallString<16> PrettyTargetValue; 4776 if (T->isSpecificBuiltinType(BuiltinType::Bool)) 4777 PrettyTargetValue = IntegerValue == 0 ? "false" : "true"; 4778 else 4779 IntegerValue.toString(PrettyTargetValue); 4780 4781 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 4782 << FL->getType() << T.getUnqualifiedType() << PrettySourceValue 4783 << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext); 4784} 4785 4786std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 4787 if (!Range.Width) return "0"; 4788 4789 llvm::APSInt ValueInRange = Value; 4790 ValueInRange.setIsSigned(!Range.NonNegative); 4791 ValueInRange = ValueInRange.trunc(Range.Width); 4792 return ValueInRange.toString(10); 4793} 4794 4795static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) { 4796 if (!isa<ImplicitCastExpr>(Ex)) 4797 return false; 4798 4799 Expr *InnerE = Ex->IgnoreParenImpCasts(); 4800 const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr(); 4801 const Type *Source = 4802 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 4803 if (Target->isDependentType()) 4804 return false; 4805 4806 const BuiltinType *FloatCandidateBT = 4807 dyn_cast<BuiltinType>(ToBool ? Source : Target); 4808 const Type *BoolCandidateType = ToBool ? Target : Source; 4809 4810 return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) && 4811 FloatCandidateBT && (FloatCandidateBT->isFloatingPoint())); 4812} 4813 4814void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall, 4815 SourceLocation CC) { 4816 unsigned NumArgs = TheCall->getNumArgs(); 4817 for (unsigned i = 0; i < NumArgs; ++i) { 4818 Expr *CurrA = TheCall->getArg(i); 4819 if (!IsImplicitBoolFloatConversion(S, CurrA, true)) 4820 continue; 4821 4822 bool IsSwapped = ((i > 0) && 4823 IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false)); 4824 IsSwapped |= ((i < (NumArgs - 1)) && 4825 IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false)); 4826 if (IsSwapped) { 4827 // Warn on this floating-point to bool conversion. 4828 DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(), 4829 CurrA->getType(), CC, 4830 diag::warn_impcast_floating_point_to_bool); 4831 } 4832 } 4833} 4834 4835void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 4836 SourceLocation CC, bool *ICContext = 0) { 4837 if (E->isTypeDependent() || E->isValueDependent()) return; 4838 4839 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 4840 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 4841 if (Source == Target) return; 4842 if (Target->isDependentType()) return; 4843 4844 // If the conversion context location is invalid don't complain. We also 4845 // don't want to emit a warning if the issue occurs from the expansion of 4846 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 4847 // delay this check as long as possible. Once we detect we are in that 4848 // scenario, we just return. 4849 if (CC.isInvalid()) 4850 return; 4851 4852 // Diagnose implicit casts to bool. 4853 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 4854 if (isa<StringLiteral>(E)) 4855 // Warn on string literal to bool. Checks for string literals in logical 4856 // expressions, for instances, assert(0 && "error here"), is prevented 4857 // by a check in AnalyzeImplicitConversions(). 4858 return DiagnoseImpCast(S, E, T, CC, 4859 diag::warn_impcast_string_literal_to_bool); 4860 if (Source->isFunctionType()) { 4861 // Warn on function to bool. Checks free functions and static member 4862 // functions. Weakly imported functions are excluded from the check, 4863 // since it's common to test their value to check whether the linker 4864 // found a definition for them. 4865 ValueDecl *D = 0; 4866 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 4867 D = R->getDecl(); 4868 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 4869 D = M->getMemberDecl(); 4870 } 4871 4872 if (D && !D->isWeak()) { 4873 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 4874 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 4875 << F << E->getSourceRange() << SourceRange(CC); 4876 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 4877 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 4878 QualType ReturnType; 4879 UnresolvedSet<4> NonTemplateOverloads; 4880 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 4881 if (!ReturnType.isNull() 4882 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 4883 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 4884 << FixItHint::CreateInsertion( 4885 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 4886 return; 4887 } 4888 } 4889 } 4890 } 4891 4892 // Strip vector types. 4893 if (isa<VectorType>(Source)) { 4894 if (!isa<VectorType>(Target)) { 4895 if (S.SourceMgr.isInSystemMacro(CC)) 4896 return; 4897 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 4898 } 4899 4900 // If the vector cast is cast between two vectors of the same size, it is 4901 // a bitcast, not a conversion. 4902 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 4903 return; 4904 4905 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 4906 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 4907 } 4908 4909 // Strip complex types. 4910 if (isa<ComplexType>(Source)) { 4911 if (!isa<ComplexType>(Target)) { 4912 if (S.SourceMgr.isInSystemMacro(CC)) 4913 return; 4914 4915 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 4916 } 4917 4918 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 4919 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 4920 } 4921 4922 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 4923 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 4924 4925 // If the source is floating point... 4926 if (SourceBT && SourceBT->isFloatingPoint()) { 4927 // ...and the target is floating point... 4928 if (TargetBT && TargetBT->isFloatingPoint()) { 4929 // ...then warn if we're dropping FP rank. 4930 4931 // Builtin FP kinds are ordered by increasing FP rank. 4932 if (SourceBT->getKind() > TargetBT->getKind()) { 4933 // Don't warn about float constants that are precisely 4934 // representable in the target type. 4935 Expr::EvalResult result; 4936 if (E->EvaluateAsRValue(result, S.Context)) { 4937 // Value might be a float, a float vector, or a float complex. 4938 if (IsSameFloatAfterCast(result.Val, 4939 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 4940 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 4941 return; 4942 } 4943 4944 if (S.SourceMgr.isInSystemMacro(CC)) 4945 return; 4946 4947 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 4948 } 4949 return; 4950 } 4951 4952 // If the target is integral, always warn. 4953 if (TargetBT && TargetBT->isInteger()) { 4954 if (S.SourceMgr.isInSystemMacro(CC)) 4955 return; 4956 4957 Expr *InnerE = E->IgnoreParenImpCasts(); 4958 // We also want to warn on, e.g., "int i = -1.234" 4959 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 4960 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 4961 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 4962 4963 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 4964 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 4965 } else { 4966 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 4967 } 4968 } 4969 4970 // If the target is bool, warn if expr is a function or method call. 4971 if (Target->isSpecificBuiltinType(BuiltinType::Bool) && 4972 isa<CallExpr>(E)) { 4973 // Check last argument of function call to see if it is an 4974 // implicit cast from a type matching the type the result 4975 // is being cast to. 4976 CallExpr *CEx = cast<CallExpr>(E); 4977 unsigned NumArgs = CEx->getNumArgs(); 4978 if (NumArgs > 0) { 4979 Expr *LastA = CEx->getArg(NumArgs - 1); 4980 Expr *InnerE = LastA->IgnoreParenImpCasts(); 4981 const Type *InnerType = 4982 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 4983 if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) { 4984 // Warn on this floating-point to bool conversion 4985 DiagnoseImpCast(S, E, T, CC, 4986 diag::warn_impcast_floating_point_to_bool); 4987 } 4988 } 4989 } 4990 return; 4991 } 4992 4993 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 4994 == Expr::NPCK_GNUNull) && !Target->isAnyPointerType() 4995 && !Target->isBlockPointerType() && !Target->isMemberPointerType() 4996 && Target->isScalarType() && !Target->isNullPtrType()) { 4997 SourceLocation Loc = E->getSourceRange().getBegin(); 4998 if (Loc.isMacroID()) 4999 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 5000 if (!Loc.isMacroID() || CC.isMacroID()) 5001 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 5002 << T << clang::SourceRange(CC) 5003 << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T)); 5004 } 5005 5006 if (!Source->isIntegerType() || !Target->isIntegerType()) 5007 return; 5008 5009 // TODO: remove this early return once the false positives for constant->bool 5010 // in templates, macros, etc, are reduced or removed. 5011 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 5012 return; 5013 5014 IntRange SourceRange = GetExprRange(S.Context, E); 5015 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 5016 5017 if (SourceRange.Width > TargetRange.Width) { 5018 // If the source is a constant, use a default-on diagnostic. 5019 // TODO: this should happen for bitfield stores, too. 5020 llvm::APSInt Value(32); 5021 if (E->isIntegerConstantExpr(Value, S.Context)) { 5022 if (S.SourceMgr.isInSystemMacro(CC)) 5023 return; 5024 5025 std::string PrettySourceValue = Value.toString(10); 5026 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 5027 5028 S.DiagRuntimeBehavior(E->getExprLoc(), E, 5029 S.PDiag(diag::warn_impcast_integer_precision_constant) 5030 << PrettySourceValue << PrettyTargetValue 5031 << E->getType() << T << E->getSourceRange() 5032 << clang::SourceRange(CC)); 5033 return; 5034 } 5035 5036 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 5037 if (S.SourceMgr.isInSystemMacro(CC)) 5038 return; 5039 5040 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 5041 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 5042 /* pruneControlFlow */ true); 5043 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 5044 } 5045 5046 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 5047 (!TargetRange.NonNegative && SourceRange.NonNegative && 5048 SourceRange.Width == TargetRange.Width)) { 5049 5050 if (S.SourceMgr.isInSystemMacro(CC)) 5051 return; 5052 5053 unsigned DiagID = diag::warn_impcast_integer_sign; 5054 5055 // Traditionally, gcc has warned about this under -Wsign-compare. 5056 // We also want to warn about it in -Wconversion. 5057 // So if -Wconversion is off, use a completely identical diagnostic 5058 // in the sign-compare group. 5059 // The conditional-checking code will 5060 if (ICContext) { 5061 DiagID = diag::warn_impcast_integer_sign_conditional; 5062 *ICContext = true; 5063 } 5064 5065 return DiagnoseImpCast(S, E, T, CC, DiagID); 5066 } 5067 5068 // Diagnose conversions between different enumeration types. 5069 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 5070 // type, to give us better diagnostics. 5071 QualType SourceType = E->getType(); 5072 if (!S.getLangOpts().CPlusPlus) { 5073 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5074 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 5075 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 5076 SourceType = S.Context.getTypeDeclType(Enum); 5077 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 5078 } 5079 } 5080 5081 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 5082 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 5083 if (SourceEnum->getDecl()->hasNameForLinkage() && 5084 TargetEnum->getDecl()->hasNameForLinkage() && 5085 SourceEnum != TargetEnum) { 5086 if (S.SourceMgr.isInSystemMacro(CC)) 5087 return; 5088 5089 return DiagnoseImpCast(S, E, SourceType, T, CC, 5090 diag::warn_impcast_different_enum_types); 5091 } 5092 5093 return; 5094} 5095 5096void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5097 SourceLocation CC, QualType T); 5098 5099void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 5100 SourceLocation CC, bool &ICContext) { 5101 E = E->IgnoreParenImpCasts(); 5102 5103 if (isa<ConditionalOperator>(E)) 5104 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T); 5105 5106 AnalyzeImplicitConversions(S, E, CC); 5107 if (E->getType() != T) 5108 return CheckImplicitConversion(S, E, T, CC, &ICContext); 5109 return; 5110} 5111 5112void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5113 SourceLocation CC, QualType T) { 5114 AnalyzeImplicitConversions(S, E->getCond(), CC); 5115 5116 bool Suspicious = false; 5117 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 5118 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 5119 5120 // If -Wconversion would have warned about either of the candidates 5121 // for a signedness conversion to the context type... 5122 if (!Suspicious) return; 5123 5124 // ...but it's currently ignored... 5125 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 5126 CC)) 5127 return; 5128 5129 // ...then check whether it would have warned about either of the 5130 // candidates for a signedness conversion to the condition type. 5131 if (E->getType() == T) return; 5132 5133 Suspicious = false; 5134 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 5135 E->getType(), CC, &Suspicious); 5136 if (!Suspicious) 5137 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 5138 E->getType(), CC, &Suspicious); 5139} 5140 5141/// AnalyzeImplicitConversions - Find and report any interesting 5142/// implicit conversions in the given expression. There are a couple 5143/// of competing diagnostics here, -Wconversion and -Wsign-compare. 5144void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 5145 QualType T = OrigE->getType(); 5146 Expr *E = OrigE->IgnoreParenImpCasts(); 5147 5148 if (E->isTypeDependent() || E->isValueDependent()) 5149 return; 5150 5151 // For conditional operators, we analyze the arguments as if they 5152 // were being fed directly into the output. 5153 if (isa<ConditionalOperator>(E)) { 5154 ConditionalOperator *CO = cast<ConditionalOperator>(E); 5155 CheckConditionalOperator(S, CO, CC, T); 5156 return; 5157 } 5158 5159 // Check implicit argument conversions for function calls. 5160 if (CallExpr *Call = dyn_cast<CallExpr>(E)) 5161 CheckImplicitArgumentConversions(S, Call, CC); 5162 5163 // Go ahead and check any implicit conversions we might have skipped. 5164 // The non-canonical typecheck is just an optimization; 5165 // CheckImplicitConversion will filter out dead implicit conversions. 5166 if (E->getType() != T) 5167 CheckImplicitConversion(S, E, T, CC); 5168 5169 // Now continue drilling into this expression. 5170 5171 // Skip past explicit casts. 5172 if (isa<ExplicitCastExpr>(E)) { 5173 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 5174 return AnalyzeImplicitConversions(S, E, CC); 5175 } 5176 5177 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5178 // Do a somewhat different check with comparison operators. 5179 if (BO->isComparisonOp()) 5180 return AnalyzeComparison(S, BO); 5181 5182 // And with simple assignments. 5183 if (BO->getOpcode() == BO_Assign) 5184 return AnalyzeAssignment(S, BO); 5185 } 5186 5187 // These break the otherwise-useful invariant below. Fortunately, 5188 // we don't really need to recurse into them, because any internal 5189 // expressions should have been analyzed already when they were 5190 // built into statements. 5191 if (isa<StmtExpr>(E)) return; 5192 5193 // Don't descend into unevaluated contexts. 5194 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 5195 5196 // Now just recurse over the expression's children. 5197 CC = E->getExprLoc(); 5198 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 5199 bool IsLogicalOperator = BO && BO->isLogicalOp(); 5200 for (Stmt::child_range I = E->children(); I; ++I) { 5201 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 5202 if (!ChildExpr) 5203 continue; 5204 5205 if (IsLogicalOperator && 5206 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 5207 // Ignore checking string literals that are in logical operators. 5208 continue; 5209 AnalyzeImplicitConversions(S, ChildExpr, CC); 5210 } 5211} 5212 5213} // end anonymous namespace 5214 5215/// Diagnoses "dangerous" implicit conversions within the given 5216/// expression (which is a full expression). Implements -Wconversion 5217/// and -Wsign-compare. 5218/// 5219/// \param CC the "context" location of the implicit conversion, i.e. 5220/// the most location of the syntactic entity requiring the implicit 5221/// conversion 5222void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 5223 // Don't diagnose in unevaluated contexts. 5224 if (isUnevaluatedContext()) 5225 return; 5226 5227 // Don't diagnose for value- or type-dependent expressions. 5228 if (E->isTypeDependent() || E->isValueDependent()) 5229 return; 5230 5231 // Check for array bounds violations in cases where the check isn't triggered 5232 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 5233 // ArraySubscriptExpr is on the RHS of a variable initialization. 5234 CheckArrayAccess(E); 5235 5236 // This is not the right CC for (e.g.) a variable initialization. 5237 AnalyzeImplicitConversions(*this, E, CC); 5238} 5239 5240/// Diagnose when expression is an integer constant expression and its evaluation 5241/// results in integer overflow 5242void Sema::CheckForIntOverflow (Expr *E) { 5243 if (isa<BinaryOperator>(E->IgnoreParens())) { 5244 llvm::SmallVector<PartialDiagnosticAt, 4> Diags; 5245 E->EvaluateForOverflow(Context, &Diags); 5246 } 5247} 5248 5249namespace { 5250/// \brief Visitor for expressions which looks for unsequenced operations on the 5251/// same object. 5252class SequenceChecker : public EvaluatedExprVisitor<SequenceChecker> { 5253 /// \brief A tree of sequenced regions within an expression. Two regions are 5254 /// unsequenced if one is an ancestor or a descendent of the other. When we 5255 /// finish processing an expression with sequencing, such as a comma 5256 /// expression, we fold its tree nodes into its parent, since they are 5257 /// unsequenced with respect to nodes we will visit later. 5258 class SequenceTree { 5259 struct Value { 5260 explicit Value(unsigned Parent) : Parent(Parent), Merged(false) {} 5261 unsigned Parent : 31; 5262 bool Merged : 1; 5263 }; 5264 llvm::SmallVector<Value, 8> Values; 5265 5266 public: 5267 /// \brief A region within an expression which may be sequenced with respect 5268 /// to some other region. 5269 class Seq { 5270 explicit Seq(unsigned N) : Index(N) {} 5271 unsigned Index; 5272 friend class SequenceTree; 5273 public: 5274 Seq() : Index(0) {} 5275 }; 5276 5277 SequenceTree() { Values.push_back(Value(0)); } 5278 Seq root() const { return Seq(0); } 5279 5280 /// \brief Create a new sequence of operations, which is an unsequenced 5281 /// subset of \p Parent. This sequence of operations is sequenced with 5282 /// respect to other children of \p Parent. 5283 Seq allocate(Seq Parent) { 5284 Values.push_back(Value(Parent.Index)); 5285 return Seq(Values.size() - 1); 5286 } 5287 5288 /// \brief Merge a sequence of operations into its parent. 5289 void merge(Seq S) { 5290 Values[S.Index].Merged = true; 5291 } 5292 5293 /// \brief Determine whether two operations are unsequenced. This operation 5294 /// is asymmetric: \p Cur should be the more recent sequence, and \p Old 5295 /// should have been merged into its parent as appropriate. 5296 bool isUnsequenced(Seq Cur, Seq Old) { 5297 unsigned C = representative(Cur.Index); 5298 unsigned Target = representative(Old.Index); 5299 while (C >= Target) { 5300 if (C == Target) 5301 return true; 5302 C = Values[C].Parent; 5303 } 5304 return false; 5305 } 5306 5307 private: 5308 /// \brief Pick a representative for a sequence. 5309 unsigned representative(unsigned K) { 5310 if (Values[K].Merged) 5311 // Perform path compression as we go. 5312 return Values[K].Parent = representative(Values[K].Parent); 5313 return K; 5314 } 5315 }; 5316 5317 /// An object for which we can track unsequenced uses. 5318 typedef NamedDecl *Object; 5319 5320 /// Different flavors of object usage which we track. We only track the 5321 /// least-sequenced usage of each kind. 5322 enum UsageKind { 5323 /// A read of an object. Multiple unsequenced reads are OK. 5324 UK_Use, 5325 /// A modification of an object which is sequenced before the value 5326 /// computation of the expression, such as ++n. 5327 UK_ModAsValue, 5328 /// A modification of an object which is not sequenced before the value 5329 /// computation of the expression, such as n++. 5330 UK_ModAsSideEffect, 5331 5332 UK_Count = UK_ModAsSideEffect + 1 5333 }; 5334 5335 struct Usage { 5336 Usage() : Use(0), Seq() {} 5337 Expr *Use; 5338 SequenceTree::Seq Seq; 5339 }; 5340 5341 struct UsageInfo { 5342 UsageInfo() : Diagnosed(false) {} 5343 Usage Uses[UK_Count]; 5344 /// Have we issued a diagnostic for this variable already? 5345 bool Diagnosed; 5346 }; 5347 typedef llvm::SmallDenseMap<Object, UsageInfo, 16> UsageInfoMap; 5348 5349 Sema &SemaRef; 5350 /// Sequenced regions within the expression. 5351 SequenceTree Tree; 5352 /// Declaration modifications and references which we have seen. 5353 UsageInfoMap UsageMap; 5354 /// The region we are currently within. 5355 SequenceTree::Seq Region; 5356 /// Filled in with declarations which were modified as a side-effect 5357 /// (that is, post-increment operations). 5358 llvm::SmallVectorImpl<std::pair<Object, Usage> > *ModAsSideEffect; 5359 /// Expressions to check later. We defer checking these to reduce 5360 /// stack usage. 5361 llvm::SmallVectorImpl<Expr*> &WorkList; 5362 5363 /// RAII object wrapping the visitation of a sequenced subexpression of an 5364 /// expression. At the end of this process, the side-effects of the evaluation 5365 /// become sequenced with respect to the value computation of the result, so 5366 /// we downgrade any UK_ModAsSideEffect within the evaluation to 5367 /// UK_ModAsValue. 5368 struct SequencedSubexpression { 5369 SequencedSubexpression(SequenceChecker &Self) 5370 : Self(Self), OldModAsSideEffect(Self.ModAsSideEffect) { 5371 Self.ModAsSideEffect = &ModAsSideEffect; 5372 } 5373 ~SequencedSubexpression() { 5374 for (unsigned I = 0, E = ModAsSideEffect.size(); I != E; ++I) { 5375 UsageInfo &U = Self.UsageMap[ModAsSideEffect[I].first]; 5376 U.Uses[UK_ModAsSideEffect] = ModAsSideEffect[I].second; 5377 Self.addUsage(U, ModAsSideEffect[I].first, 5378 ModAsSideEffect[I].second.Use, UK_ModAsValue); 5379 } 5380 Self.ModAsSideEffect = OldModAsSideEffect; 5381 } 5382 5383 SequenceChecker &Self; 5384 llvm::SmallVector<std::pair<Object, Usage>, 4> ModAsSideEffect; 5385 llvm::SmallVectorImpl<std::pair<Object, Usage> > *OldModAsSideEffect; 5386 }; 5387 5388 /// \brief Find the object which is produced by the specified expression, 5389 /// if any. 5390 Object getObject(Expr *E, bool Mod) const { 5391 E = E->IgnoreParenCasts(); 5392 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 5393 if (Mod && (UO->getOpcode() == UO_PreInc || UO->getOpcode() == UO_PreDec)) 5394 return getObject(UO->getSubExpr(), Mod); 5395 } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5396 if (BO->getOpcode() == BO_Comma) 5397 return getObject(BO->getRHS(), Mod); 5398 if (Mod && BO->isAssignmentOp()) 5399 return getObject(BO->getLHS(), Mod); 5400 } else if (MemberExpr *ME = dyn_cast<MemberExpr>(E)) { 5401 // FIXME: Check for more interesting cases, like "x.n = ++x.n". 5402 if (isa<CXXThisExpr>(ME->getBase()->IgnoreParenCasts())) 5403 return ME->getMemberDecl(); 5404 } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5405 // FIXME: If this is a reference, map through to its value. 5406 return DRE->getDecl(); 5407 return 0; 5408 } 5409 5410 /// \brief Note that an object was modified or used by an expression. 5411 void addUsage(UsageInfo &UI, Object O, Expr *Ref, UsageKind UK) { 5412 Usage &U = UI.Uses[UK]; 5413 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) { 5414 if (UK == UK_ModAsSideEffect && ModAsSideEffect) 5415 ModAsSideEffect->push_back(std::make_pair(O, U)); 5416 U.Use = Ref; 5417 U.Seq = Region; 5418 } 5419 } 5420 /// \brief Check whether a modification or use conflicts with a prior usage. 5421 void checkUsage(Object O, UsageInfo &UI, Expr *Ref, UsageKind OtherKind, 5422 bool IsModMod) { 5423 if (UI.Diagnosed) 5424 return; 5425 5426 const Usage &U = UI.Uses[OtherKind]; 5427 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) 5428 return; 5429 5430 Expr *Mod = U.Use; 5431 Expr *ModOrUse = Ref; 5432 if (OtherKind == UK_Use) 5433 std::swap(Mod, ModOrUse); 5434 5435 SemaRef.Diag(Mod->getExprLoc(), 5436 IsModMod ? diag::warn_unsequenced_mod_mod 5437 : diag::warn_unsequenced_mod_use) 5438 << O << SourceRange(ModOrUse->getExprLoc()); 5439 UI.Diagnosed = true; 5440 } 5441 5442 void notePreUse(Object O, Expr *Use) { 5443 UsageInfo &U = UsageMap[O]; 5444 // Uses conflict with other modifications. 5445 checkUsage(O, U, Use, UK_ModAsValue, false); 5446 } 5447 void notePostUse(Object O, Expr *Use) { 5448 UsageInfo &U = UsageMap[O]; 5449 checkUsage(O, U, Use, UK_ModAsSideEffect, false); 5450 addUsage(U, O, Use, UK_Use); 5451 } 5452 5453 void notePreMod(Object O, Expr *Mod) { 5454 UsageInfo &U = UsageMap[O]; 5455 // Modifications conflict with other modifications and with uses. 5456 checkUsage(O, U, Mod, UK_ModAsValue, true); 5457 checkUsage(O, U, Mod, UK_Use, false); 5458 } 5459 void notePostMod(Object O, Expr *Use, UsageKind UK) { 5460 UsageInfo &U = UsageMap[O]; 5461 checkUsage(O, U, Use, UK_ModAsSideEffect, true); 5462 addUsage(U, O, Use, UK); 5463 } 5464 5465public: 5466 SequenceChecker(Sema &S, Expr *E, 5467 llvm::SmallVectorImpl<Expr*> &WorkList) 5468 : EvaluatedExprVisitor<SequenceChecker>(S.Context), SemaRef(S), 5469 Region(Tree.root()), ModAsSideEffect(0), WorkList(WorkList) { 5470 Visit(E); 5471 } 5472 5473 void VisitStmt(Stmt *S) { 5474 // Skip all statements which aren't expressions for now. 5475 } 5476 5477 void VisitExpr(Expr *E) { 5478 // By default, just recurse to evaluated subexpressions. 5479 EvaluatedExprVisitor<SequenceChecker>::VisitStmt(E); 5480 } 5481 5482 void VisitCastExpr(CastExpr *E) { 5483 Object O = Object(); 5484 if (E->getCastKind() == CK_LValueToRValue) 5485 O = getObject(E->getSubExpr(), false); 5486 5487 if (O) 5488 notePreUse(O, E); 5489 VisitExpr(E); 5490 if (O) 5491 notePostUse(O, E); 5492 } 5493 5494 void VisitBinComma(BinaryOperator *BO) { 5495 // C++11 [expr.comma]p1: 5496 // Every value computation and side effect associated with the left 5497 // expression is sequenced before every value computation and side 5498 // effect associated with the right expression. 5499 SequenceTree::Seq LHS = Tree.allocate(Region); 5500 SequenceTree::Seq RHS = Tree.allocate(Region); 5501 SequenceTree::Seq OldRegion = Region; 5502 5503 { 5504 SequencedSubexpression SeqLHS(*this); 5505 Region = LHS; 5506 Visit(BO->getLHS()); 5507 } 5508 5509 Region = RHS; 5510 Visit(BO->getRHS()); 5511 5512 Region = OldRegion; 5513 5514 // Forget that LHS and RHS are sequenced. They are both unsequenced 5515 // with respect to other stuff. 5516 Tree.merge(LHS); 5517 Tree.merge(RHS); 5518 } 5519 5520 void VisitBinAssign(BinaryOperator *BO) { 5521 // The modification is sequenced after the value computation of the LHS 5522 // and RHS, so check it before inspecting the operands and update the 5523 // map afterwards. 5524 Object O = getObject(BO->getLHS(), true); 5525 if (!O) 5526 return VisitExpr(BO); 5527 5528 notePreMod(O, BO); 5529 5530 // C++11 [expr.ass]p7: 5531 // E1 op= E2 is equivalent to E1 = E1 op E2, except that E1 is evaluated 5532 // only once. 5533 // 5534 // Therefore, for a compound assignment operator, O is considered used 5535 // everywhere except within the evaluation of E1 itself. 5536 if (isa<CompoundAssignOperator>(BO)) 5537 notePreUse(O, BO); 5538 5539 Visit(BO->getLHS()); 5540 5541 if (isa<CompoundAssignOperator>(BO)) 5542 notePostUse(O, BO); 5543 5544 Visit(BO->getRHS()); 5545 5546 notePostMod(O, BO, UK_ModAsValue); 5547 } 5548 void VisitCompoundAssignOperator(CompoundAssignOperator *CAO) { 5549 VisitBinAssign(CAO); 5550 } 5551 5552 void VisitUnaryPreInc(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 5553 void VisitUnaryPreDec(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 5554 void VisitUnaryPreIncDec(UnaryOperator *UO) { 5555 Object O = getObject(UO->getSubExpr(), true); 5556 if (!O) 5557 return VisitExpr(UO); 5558 5559 notePreMod(O, UO); 5560 Visit(UO->getSubExpr()); 5561 notePostMod(O, UO, UK_ModAsValue); 5562 } 5563 5564 void VisitUnaryPostInc(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 5565 void VisitUnaryPostDec(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 5566 void VisitUnaryPostIncDec(UnaryOperator *UO) { 5567 Object O = getObject(UO->getSubExpr(), true); 5568 if (!O) 5569 return VisitExpr(UO); 5570 5571 notePreMod(O, UO); 5572 Visit(UO->getSubExpr()); 5573 notePostMod(O, UO, UK_ModAsSideEffect); 5574 } 5575 5576 /// Don't visit the RHS of '&&' or '||' if it might not be evaluated. 5577 void VisitBinLOr(BinaryOperator *BO) { 5578 // The side-effects of the LHS of an '&&' are sequenced before the 5579 // value computation of the RHS, and hence before the value computation 5580 // of the '&&' itself, unless the LHS evaluates to zero. We treat them 5581 // as if they were unconditionally sequenced. 5582 { 5583 SequencedSubexpression Sequenced(*this); 5584 Visit(BO->getLHS()); 5585 } 5586 5587 bool Result; 5588 if (!BO->getLHS()->isValueDependent() && 5589 BO->getLHS()->EvaluateAsBooleanCondition(Result, SemaRef.Context)) { 5590 if (!Result) 5591 Visit(BO->getRHS()); 5592 } else { 5593 // Check for unsequenced operations in the RHS, treating it as an 5594 // entirely separate evaluation. 5595 // 5596 // FIXME: If there are operations in the RHS which are unsequenced 5597 // with respect to operations outside the RHS, and those operations 5598 // are unconditionally evaluated, diagnose them. 5599 WorkList.push_back(BO->getRHS()); 5600 } 5601 } 5602 void VisitBinLAnd(BinaryOperator *BO) { 5603 { 5604 SequencedSubexpression Sequenced(*this); 5605 Visit(BO->getLHS()); 5606 } 5607 5608 bool Result; 5609 if (!BO->getLHS()->isValueDependent() && 5610 BO->getLHS()->EvaluateAsBooleanCondition(Result, SemaRef.Context)) { 5611 if (Result) 5612 Visit(BO->getRHS()); 5613 } else { 5614 WorkList.push_back(BO->getRHS()); 5615 } 5616 } 5617 5618 // Only visit the condition, unless we can be sure which subexpression will 5619 // be chosen. 5620 void VisitAbstractConditionalOperator(AbstractConditionalOperator *CO) { 5621 SequencedSubexpression Sequenced(*this); 5622 Visit(CO->getCond()); 5623 5624 bool Result; 5625 if (!CO->getCond()->isValueDependent() && 5626 CO->getCond()->EvaluateAsBooleanCondition(Result, SemaRef.Context)) 5627 Visit(Result ? CO->getTrueExpr() : CO->getFalseExpr()); 5628 else { 5629 WorkList.push_back(CO->getTrueExpr()); 5630 WorkList.push_back(CO->getFalseExpr()); 5631 } 5632 } 5633 5634 void VisitCXXConstructExpr(CXXConstructExpr *CCE) { 5635 if (!CCE->isListInitialization()) 5636 return VisitExpr(CCE); 5637 5638 // In C++11, list initializations are sequenced. 5639 llvm::SmallVector<SequenceTree::Seq, 32> Elts; 5640 SequenceTree::Seq Parent = Region; 5641 for (CXXConstructExpr::arg_iterator I = CCE->arg_begin(), 5642 E = CCE->arg_end(); 5643 I != E; ++I) { 5644 Region = Tree.allocate(Parent); 5645 Elts.push_back(Region); 5646 Visit(*I); 5647 } 5648 5649 // Forget that the initializers are sequenced. 5650 Region = Parent; 5651 for (unsigned I = 0; I < Elts.size(); ++I) 5652 Tree.merge(Elts[I]); 5653 } 5654 5655 void VisitInitListExpr(InitListExpr *ILE) { 5656 if (!SemaRef.getLangOpts().CPlusPlus11) 5657 return VisitExpr(ILE); 5658 5659 // In C++11, list initializations are sequenced. 5660 llvm::SmallVector<SequenceTree::Seq, 32> Elts; 5661 SequenceTree::Seq Parent = Region; 5662 for (unsigned I = 0; I < ILE->getNumInits(); ++I) { 5663 Expr *E = ILE->getInit(I); 5664 if (!E) continue; 5665 Region = Tree.allocate(Parent); 5666 Elts.push_back(Region); 5667 Visit(E); 5668 } 5669 5670 // Forget that the initializers are sequenced. 5671 Region = Parent; 5672 for (unsigned I = 0; I < Elts.size(); ++I) 5673 Tree.merge(Elts[I]); 5674 } 5675}; 5676} 5677 5678void Sema::CheckUnsequencedOperations(Expr *E) { 5679 llvm::SmallVector<Expr*, 8> WorkList; 5680 WorkList.push_back(E); 5681 while (!WorkList.empty()) { 5682 Expr *Item = WorkList.back(); 5683 WorkList.pop_back(); 5684 SequenceChecker(*this, Item, WorkList); 5685 } 5686} 5687 5688void Sema::CheckCompletedExpr(Expr *E, SourceLocation CheckLoc, 5689 bool IsConstexpr) { 5690 CheckImplicitConversions(E, CheckLoc); 5691 CheckUnsequencedOperations(E); 5692 if (!IsConstexpr && !E->isValueDependent()) 5693 CheckForIntOverflow(E); 5694} 5695 5696void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 5697 FieldDecl *BitField, 5698 Expr *Init) { 5699 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 5700} 5701 5702/// CheckParmsForFunctionDef - Check that the parameters of the given 5703/// function are appropriate for the definition of a function. This 5704/// takes care of any checks that cannot be performed on the 5705/// declaration itself, e.g., that the types of each of the function 5706/// parameters are complete. 5707bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 5708 bool CheckParameterNames) { 5709 bool HasInvalidParm = false; 5710 for (; P != PEnd; ++P) { 5711 ParmVarDecl *Param = *P; 5712 5713 // C99 6.7.5.3p4: the parameters in a parameter type list in a 5714 // function declarator that is part of a function definition of 5715 // that function shall not have incomplete type. 5716 // 5717 // This is also C++ [dcl.fct]p6. 5718 if (!Param->isInvalidDecl() && 5719 RequireCompleteType(Param->getLocation(), Param->getType(), 5720 diag::err_typecheck_decl_incomplete_type)) { 5721 Param->setInvalidDecl(); 5722 HasInvalidParm = true; 5723 } 5724 5725 // C99 6.9.1p5: If the declarator includes a parameter type list, the 5726 // declaration of each parameter shall include an identifier. 5727 if (CheckParameterNames && 5728 Param->getIdentifier() == 0 && 5729 !Param->isImplicit() && 5730 !getLangOpts().CPlusPlus) 5731 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 5732 5733 // C99 6.7.5.3p12: 5734 // If the function declarator is not part of a definition of that 5735 // function, parameters may have incomplete type and may use the [*] 5736 // notation in their sequences of declarator specifiers to specify 5737 // variable length array types. 5738 QualType PType = Param->getOriginalType(); 5739 while (const ArrayType *AT = Context.getAsArrayType(PType)) { 5740 if (AT->getSizeModifier() == ArrayType::Star) { 5741 // FIXME: This diagnostic should point the '[*]' if source-location 5742 // information is added for it. 5743 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 5744 break; 5745 } 5746 PType= AT->getElementType(); 5747 } 5748 } 5749 5750 return HasInvalidParm; 5751} 5752 5753/// CheckCastAlign - Implements -Wcast-align, which warns when a 5754/// pointer cast increases the alignment requirements. 5755void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 5756 // This is actually a lot of work to potentially be doing on every 5757 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 5758 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 5759 TRange.getBegin()) 5760 == DiagnosticsEngine::Ignored) 5761 return; 5762 5763 // Ignore dependent types. 5764 if (T->isDependentType() || Op->getType()->isDependentType()) 5765 return; 5766 5767 // Require that the destination be a pointer type. 5768 const PointerType *DestPtr = T->getAs<PointerType>(); 5769 if (!DestPtr) return; 5770 5771 // If the destination has alignment 1, we're done. 5772 QualType DestPointee = DestPtr->getPointeeType(); 5773 if (DestPointee->isIncompleteType()) return; 5774 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 5775 if (DestAlign.isOne()) return; 5776 5777 // Require that the source be a pointer type. 5778 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 5779 if (!SrcPtr) return; 5780 QualType SrcPointee = SrcPtr->getPointeeType(); 5781 5782 // Whitelist casts from cv void*. We already implicitly 5783 // whitelisted casts to cv void*, since they have alignment 1. 5784 // Also whitelist casts involving incomplete types, which implicitly 5785 // includes 'void'. 5786 if (SrcPointee->isIncompleteType()) return; 5787 5788 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 5789 if (SrcAlign >= DestAlign) return; 5790 5791 Diag(TRange.getBegin(), diag::warn_cast_align) 5792 << Op->getType() << T 5793 << static_cast<unsigned>(SrcAlign.getQuantity()) 5794 << static_cast<unsigned>(DestAlign.getQuantity()) 5795 << TRange << Op->getSourceRange(); 5796} 5797 5798static const Type* getElementType(const Expr *BaseExpr) { 5799 const Type* EltType = BaseExpr->getType().getTypePtr(); 5800 if (EltType->isAnyPointerType()) 5801 return EltType->getPointeeType().getTypePtr(); 5802 else if (EltType->isArrayType()) 5803 return EltType->getBaseElementTypeUnsafe(); 5804 return EltType; 5805} 5806 5807/// \brief Check whether this array fits the idiom of a size-one tail padded 5808/// array member of a struct. 5809/// 5810/// We avoid emitting out-of-bounds access warnings for such arrays as they are 5811/// commonly used to emulate flexible arrays in C89 code. 5812static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 5813 const NamedDecl *ND) { 5814 if (Size != 1 || !ND) return false; 5815 5816 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 5817 if (!FD) return false; 5818 5819 // Don't consider sizes resulting from macro expansions or template argument 5820 // substitution to form C89 tail-padded arrays. 5821 5822 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 5823 while (TInfo) { 5824 TypeLoc TL = TInfo->getTypeLoc(); 5825 // Look through typedefs. 5826 if (TypedefTypeLoc TTL = TL.getAs<TypedefTypeLoc>()) { 5827 const TypedefNameDecl *TDL = TTL.getTypedefNameDecl(); 5828 TInfo = TDL->getTypeSourceInfo(); 5829 continue; 5830 } 5831 if (ConstantArrayTypeLoc CTL = TL.getAs<ConstantArrayTypeLoc>()) { 5832 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 5833 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 5834 return false; 5835 } 5836 break; 5837 } 5838 5839 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 5840 if (!RD) return false; 5841 if (RD->isUnion()) return false; 5842 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 5843 if (!CRD->isStandardLayout()) return false; 5844 } 5845 5846 // See if this is the last field decl in the record. 5847 const Decl *D = FD; 5848 while ((D = D->getNextDeclInContext())) 5849 if (isa<FieldDecl>(D)) 5850 return false; 5851 return true; 5852} 5853 5854void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 5855 const ArraySubscriptExpr *ASE, 5856 bool AllowOnePastEnd, bool IndexNegated) { 5857 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 5858 if (IndexExpr->isValueDependent()) 5859 return; 5860 5861 const Type *EffectiveType = getElementType(BaseExpr); 5862 BaseExpr = BaseExpr->IgnoreParenCasts(); 5863 const ConstantArrayType *ArrayTy = 5864 Context.getAsConstantArrayType(BaseExpr->getType()); 5865 if (!ArrayTy) 5866 return; 5867 5868 llvm::APSInt index; 5869 if (!IndexExpr->EvaluateAsInt(index, Context)) 5870 return; 5871 if (IndexNegated) 5872 index = -index; 5873 5874 const NamedDecl *ND = NULL; 5875 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 5876 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 5877 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 5878 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 5879 5880 if (index.isUnsigned() || !index.isNegative()) { 5881 llvm::APInt size = ArrayTy->getSize(); 5882 if (!size.isStrictlyPositive()) 5883 return; 5884 5885 const Type* BaseType = getElementType(BaseExpr); 5886 if (BaseType != EffectiveType) { 5887 // Make sure we're comparing apples to apples when comparing index to size 5888 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 5889 uint64_t array_typesize = Context.getTypeSize(BaseType); 5890 // Handle ptrarith_typesize being zero, such as when casting to void* 5891 if (!ptrarith_typesize) ptrarith_typesize = 1; 5892 if (ptrarith_typesize != array_typesize) { 5893 // There's a cast to a different size type involved 5894 uint64_t ratio = array_typesize / ptrarith_typesize; 5895 // TODO: Be smarter about handling cases where array_typesize is not a 5896 // multiple of ptrarith_typesize 5897 if (ptrarith_typesize * ratio == array_typesize) 5898 size *= llvm::APInt(size.getBitWidth(), ratio); 5899 } 5900 } 5901 5902 if (size.getBitWidth() > index.getBitWidth()) 5903 index = index.zext(size.getBitWidth()); 5904 else if (size.getBitWidth() < index.getBitWidth()) 5905 size = size.zext(index.getBitWidth()); 5906 5907 // For array subscripting the index must be less than size, but for pointer 5908 // arithmetic also allow the index (offset) to be equal to size since 5909 // computing the next address after the end of the array is legal and 5910 // commonly done e.g. in C++ iterators and range-based for loops. 5911 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 5912 return; 5913 5914 // Also don't warn for arrays of size 1 which are members of some 5915 // structure. These are often used to approximate flexible arrays in C89 5916 // code. 5917 if (IsTailPaddedMemberArray(*this, size, ND)) 5918 return; 5919 5920 // Suppress the warning if the subscript expression (as identified by the 5921 // ']' location) and the index expression are both from macro expansions 5922 // within a system header. 5923 if (ASE) { 5924 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 5925 ASE->getRBracketLoc()); 5926 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 5927 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 5928 IndexExpr->getLocStart()); 5929 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 5930 return; 5931 } 5932 } 5933 5934 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 5935 if (ASE) 5936 DiagID = diag::warn_array_index_exceeds_bounds; 5937 5938 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 5939 PDiag(DiagID) << index.toString(10, true) 5940 << size.toString(10, true) 5941 << (unsigned)size.getLimitedValue(~0U) 5942 << IndexExpr->getSourceRange()); 5943 } else { 5944 unsigned DiagID = diag::warn_array_index_precedes_bounds; 5945 if (!ASE) { 5946 DiagID = diag::warn_ptr_arith_precedes_bounds; 5947 if (index.isNegative()) index = -index; 5948 } 5949 5950 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 5951 PDiag(DiagID) << index.toString(10, true) 5952 << IndexExpr->getSourceRange()); 5953 } 5954 5955 if (!ND) { 5956 // Try harder to find a NamedDecl to point at in the note. 5957 while (const ArraySubscriptExpr *ASE = 5958 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 5959 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 5960 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 5961 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 5962 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 5963 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 5964 } 5965 5966 if (ND) 5967 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 5968 PDiag(diag::note_array_index_out_of_bounds) 5969 << ND->getDeclName()); 5970} 5971 5972void Sema::CheckArrayAccess(const Expr *expr) { 5973 int AllowOnePastEnd = 0; 5974 while (expr) { 5975 expr = expr->IgnoreParenImpCasts(); 5976 switch (expr->getStmtClass()) { 5977 case Stmt::ArraySubscriptExprClass: { 5978 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 5979 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 5980 AllowOnePastEnd > 0); 5981 return; 5982 } 5983 case Stmt::UnaryOperatorClass: { 5984 // Only unwrap the * and & unary operators 5985 const UnaryOperator *UO = cast<UnaryOperator>(expr); 5986 expr = UO->getSubExpr(); 5987 switch (UO->getOpcode()) { 5988 case UO_AddrOf: 5989 AllowOnePastEnd++; 5990 break; 5991 case UO_Deref: 5992 AllowOnePastEnd--; 5993 break; 5994 default: 5995 return; 5996 } 5997 break; 5998 } 5999 case Stmt::ConditionalOperatorClass: { 6000 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 6001 if (const Expr *lhs = cond->getLHS()) 6002 CheckArrayAccess(lhs); 6003 if (const Expr *rhs = cond->getRHS()) 6004 CheckArrayAccess(rhs); 6005 return; 6006 } 6007 default: 6008 return; 6009 } 6010 } 6011} 6012 6013//===--- CHECK: Objective-C retain cycles ----------------------------------// 6014 6015namespace { 6016 struct RetainCycleOwner { 6017 RetainCycleOwner() : Variable(0), Indirect(false) {} 6018 VarDecl *Variable; 6019 SourceRange Range; 6020 SourceLocation Loc; 6021 bool Indirect; 6022 6023 void setLocsFrom(Expr *e) { 6024 Loc = e->getExprLoc(); 6025 Range = e->getSourceRange(); 6026 } 6027 }; 6028} 6029 6030/// Consider whether capturing the given variable can possibly lead to 6031/// a retain cycle. 6032static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 6033 // In ARC, it's captured strongly iff the variable has __strong 6034 // lifetime. In MRR, it's captured strongly if the variable is 6035 // __block and has an appropriate type. 6036 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 6037 return false; 6038 6039 owner.Variable = var; 6040 if (ref) 6041 owner.setLocsFrom(ref); 6042 return true; 6043} 6044 6045static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 6046 while (true) { 6047 e = e->IgnoreParens(); 6048 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 6049 switch (cast->getCastKind()) { 6050 case CK_BitCast: 6051 case CK_LValueBitCast: 6052 case CK_LValueToRValue: 6053 case CK_ARCReclaimReturnedObject: 6054 e = cast->getSubExpr(); 6055 continue; 6056 6057 default: 6058 return false; 6059 } 6060 } 6061 6062 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 6063 ObjCIvarDecl *ivar = ref->getDecl(); 6064 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 6065 return false; 6066 6067 // Try to find a retain cycle in the base. 6068 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 6069 return false; 6070 6071 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 6072 owner.Indirect = true; 6073 return true; 6074 } 6075 6076 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 6077 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 6078 if (!var) return false; 6079 return considerVariable(var, ref, owner); 6080 } 6081 6082 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 6083 if (member->isArrow()) return false; 6084 6085 // Don't count this as an indirect ownership. 6086 e = member->getBase(); 6087 continue; 6088 } 6089 6090 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 6091 // Only pay attention to pseudo-objects on property references. 6092 ObjCPropertyRefExpr *pre 6093 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 6094 ->IgnoreParens()); 6095 if (!pre) return false; 6096 if (pre->isImplicitProperty()) return false; 6097 ObjCPropertyDecl *property = pre->getExplicitProperty(); 6098 if (!property->isRetaining() && 6099 !(property->getPropertyIvarDecl() && 6100 property->getPropertyIvarDecl()->getType() 6101 .getObjCLifetime() == Qualifiers::OCL_Strong)) 6102 return false; 6103 6104 owner.Indirect = true; 6105 if (pre->isSuperReceiver()) { 6106 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 6107 if (!owner.Variable) 6108 return false; 6109 owner.Loc = pre->getLocation(); 6110 owner.Range = pre->getSourceRange(); 6111 return true; 6112 } 6113 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 6114 ->getSourceExpr()); 6115 continue; 6116 } 6117 6118 // Array ivars? 6119 6120 return false; 6121 } 6122} 6123 6124namespace { 6125 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 6126 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 6127 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 6128 Variable(variable), Capturer(0) {} 6129 6130 VarDecl *Variable; 6131 Expr *Capturer; 6132 6133 void VisitDeclRefExpr(DeclRefExpr *ref) { 6134 if (ref->getDecl() == Variable && !Capturer) 6135 Capturer = ref; 6136 } 6137 6138 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 6139 if (Capturer) return; 6140 Visit(ref->getBase()); 6141 if (Capturer && ref->isFreeIvar()) 6142 Capturer = ref; 6143 } 6144 6145 void VisitBlockExpr(BlockExpr *block) { 6146 // Look inside nested blocks 6147 if (block->getBlockDecl()->capturesVariable(Variable)) 6148 Visit(block->getBlockDecl()->getBody()); 6149 } 6150 6151 void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) { 6152 if (Capturer) return; 6153 if (OVE->getSourceExpr()) 6154 Visit(OVE->getSourceExpr()); 6155 } 6156 }; 6157} 6158 6159/// Check whether the given argument is a block which captures a 6160/// variable. 6161static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 6162 assert(owner.Variable && owner.Loc.isValid()); 6163 6164 e = e->IgnoreParenCasts(); 6165 6166 // Look through [^{...} copy] and Block_copy(^{...}). 6167 if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) { 6168 Selector Cmd = ME->getSelector(); 6169 if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") { 6170 e = ME->getInstanceReceiver(); 6171 if (!e) 6172 return 0; 6173 e = e->IgnoreParenCasts(); 6174 } 6175 } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) { 6176 if (CE->getNumArgs() == 1) { 6177 FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl()); 6178 if (Fn) { 6179 const IdentifierInfo *FnI = Fn->getIdentifier(); 6180 if (FnI && FnI->isStr("_Block_copy")) { 6181 e = CE->getArg(0)->IgnoreParenCasts(); 6182 } 6183 } 6184 } 6185 } 6186 6187 BlockExpr *block = dyn_cast<BlockExpr>(e); 6188 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 6189 return 0; 6190 6191 FindCaptureVisitor visitor(S.Context, owner.Variable); 6192 visitor.Visit(block->getBlockDecl()->getBody()); 6193 return visitor.Capturer; 6194} 6195 6196static void diagnoseRetainCycle(Sema &S, Expr *capturer, 6197 RetainCycleOwner &owner) { 6198 assert(capturer); 6199 assert(owner.Variable && owner.Loc.isValid()); 6200 6201 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 6202 << owner.Variable << capturer->getSourceRange(); 6203 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 6204 << owner.Indirect << owner.Range; 6205} 6206 6207/// Check for a keyword selector that starts with the word 'add' or 6208/// 'set'. 6209static bool isSetterLikeSelector(Selector sel) { 6210 if (sel.isUnarySelector()) return false; 6211 6212 StringRef str = sel.getNameForSlot(0); 6213 while (!str.empty() && str.front() == '_') str = str.substr(1); 6214 if (str.startswith("set")) 6215 str = str.substr(3); 6216 else if (str.startswith("add")) { 6217 // Specially whitelist 'addOperationWithBlock:'. 6218 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 6219 return false; 6220 str = str.substr(3); 6221 } 6222 else 6223 return false; 6224 6225 if (str.empty()) return true; 6226 return !isLowercase(str.front()); 6227} 6228 6229/// Check a message send to see if it's likely to cause a retain cycle. 6230void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 6231 // Only check instance methods whose selector looks like a setter. 6232 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 6233 return; 6234 6235 // Try to find a variable that the receiver is strongly owned by. 6236 RetainCycleOwner owner; 6237 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 6238 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 6239 return; 6240 } else { 6241 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 6242 owner.Variable = getCurMethodDecl()->getSelfDecl(); 6243 owner.Loc = msg->getSuperLoc(); 6244 owner.Range = msg->getSuperLoc(); 6245 } 6246 6247 // Check whether the receiver is captured by any of the arguments. 6248 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 6249 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 6250 return diagnoseRetainCycle(*this, capturer, owner); 6251} 6252 6253/// Check a property assign to see if it's likely to cause a retain cycle. 6254void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 6255 RetainCycleOwner owner; 6256 if (!findRetainCycleOwner(*this, receiver, owner)) 6257 return; 6258 6259 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 6260 diagnoseRetainCycle(*this, capturer, owner); 6261} 6262 6263void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) { 6264 RetainCycleOwner Owner; 6265 if (!considerVariable(Var, /*DeclRefExpr=*/0, Owner)) 6266 return; 6267 6268 // Because we don't have an expression for the variable, we have to set the 6269 // location explicitly here. 6270 Owner.Loc = Var->getLocation(); 6271 Owner.Range = Var->getSourceRange(); 6272 6273 if (Expr *Capturer = findCapturingExpr(*this, Init, Owner)) 6274 diagnoseRetainCycle(*this, Capturer, Owner); 6275} 6276 6277static bool checkUnsafeAssignLiteral(Sema &S, SourceLocation Loc, 6278 Expr *RHS, bool isProperty) { 6279 // Check if RHS is an Objective-C object literal, which also can get 6280 // immediately zapped in a weak reference. Note that we explicitly 6281 // allow ObjCStringLiterals, since those are designed to never really die. 6282 RHS = RHS->IgnoreParenImpCasts(); 6283 6284 // This enum needs to match with the 'select' in 6285 // warn_objc_arc_literal_assign (off-by-1). 6286 Sema::ObjCLiteralKind Kind = S.CheckLiteralKind(RHS); 6287 if (Kind == Sema::LK_String || Kind == Sema::LK_None) 6288 return false; 6289 6290 S.Diag(Loc, diag::warn_arc_literal_assign) 6291 << (unsigned) Kind 6292 << (isProperty ? 0 : 1) 6293 << RHS->getSourceRange(); 6294 6295 return true; 6296} 6297 6298static bool checkUnsafeAssignObject(Sema &S, SourceLocation Loc, 6299 Qualifiers::ObjCLifetime LT, 6300 Expr *RHS, bool isProperty) { 6301 // Strip off any implicit cast added to get to the one ARC-specific. 6302 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6303 if (cast->getCastKind() == CK_ARCConsumeObject) { 6304 S.Diag(Loc, diag::warn_arc_retained_assign) 6305 << (LT == Qualifiers::OCL_ExplicitNone) 6306 << (isProperty ? 0 : 1) 6307 << RHS->getSourceRange(); 6308 return true; 6309 } 6310 RHS = cast->getSubExpr(); 6311 } 6312 6313 if (LT == Qualifiers::OCL_Weak && 6314 checkUnsafeAssignLiteral(S, Loc, RHS, isProperty)) 6315 return true; 6316 6317 return false; 6318} 6319 6320bool Sema::checkUnsafeAssigns(SourceLocation Loc, 6321 QualType LHS, Expr *RHS) { 6322 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 6323 6324 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 6325 return false; 6326 6327 if (checkUnsafeAssignObject(*this, Loc, LT, RHS, false)) 6328 return true; 6329 6330 return false; 6331} 6332 6333void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 6334 Expr *LHS, Expr *RHS) { 6335 QualType LHSType; 6336 // PropertyRef on LHS type need be directly obtained from 6337 // its declaration as it has a PsuedoType. 6338 ObjCPropertyRefExpr *PRE 6339 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 6340 if (PRE && !PRE->isImplicitProperty()) { 6341 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6342 if (PD) 6343 LHSType = PD->getType(); 6344 } 6345 6346 if (LHSType.isNull()) 6347 LHSType = LHS->getType(); 6348 6349 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 6350 6351 if (LT == Qualifiers::OCL_Weak) { 6352 DiagnosticsEngine::Level Level = 6353 Diags.getDiagnosticLevel(diag::warn_arc_repeated_use_of_weak, Loc); 6354 if (Level != DiagnosticsEngine::Ignored) 6355 getCurFunction()->markSafeWeakUse(LHS); 6356 } 6357 6358 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 6359 return; 6360 6361 // FIXME. Check for other life times. 6362 if (LT != Qualifiers::OCL_None) 6363 return; 6364 6365 if (PRE) { 6366 if (PRE->isImplicitProperty()) 6367 return; 6368 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6369 if (!PD) 6370 return; 6371 6372 unsigned Attributes = PD->getPropertyAttributes(); 6373 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 6374 // when 'assign' attribute was not explicitly specified 6375 // by user, ignore it and rely on property type itself 6376 // for lifetime info. 6377 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 6378 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 6379 LHSType->isObjCRetainableType()) 6380 return; 6381 6382 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6383 if (cast->getCastKind() == CK_ARCConsumeObject) { 6384 Diag(Loc, diag::warn_arc_retained_property_assign) 6385 << RHS->getSourceRange(); 6386 return; 6387 } 6388 RHS = cast->getSubExpr(); 6389 } 6390 } 6391 else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) { 6392 if (checkUnsafeAssignObject(*this, Loc, Qualifiers::OCL_Weak, RHS, true)) 6393 return; 6394 } 6395 } 6396} 6397 6398//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 6399 6400namespace { 6401bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 6402 SourceLocation StmtLoc, 6403 const NullStmt *Body) { 6404 // Do not warn if the body is a macro that expands to nothing, e.g: 6405 // 6406 // #define CALL(x) 6407 // if (condition) 6408 // CALL(0); 6409 // 6410 if (Body->hasLeadingEmptyMacro()) 6411 return false; 6412 6413 // Get line numbers of statement and body. 6414 bool StmtLineInvalid; 6415 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 6416 &StmtLineInvalid); 6417 if (StmtLineInvalid) 6418 return false; 6419 6420 bool BodyLineInvalid; 6421 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 6422 &BodyLineInvalid); 6423 if (BodyLineInvalid) 6424 return false; 6425 6426 // Warn if null statement and body are on the same line. 6427 if (StmtLine != BodyLine) 6428 return false; 6429 6430 return true; 6431} 6432} // Unnamed namespace 6433 6434void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 6435 const Stmt *Body, 6436 unsigned DiagID) { 6437 // Since this is a syntactic check, don't emit diagnostic for template 6438 // instantiations, this just adds noise. 6439 if (CurrentInstantiationScope) 6440 return; 6441 6442 // The body should be a null statement. 6443 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6444 if (!NBody) 6445 return; 6446 6447 // Do the usual checks. 6448 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6449 return; 6450 6451 Diag(NBody->getSemiLoc(), DiagID); 6452 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6453} 6454 6455void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 6456 const Stmt *PossibleBody) { 6457 assert(!CurrentInstantiationScope); // Ensured by caller 6458 6459 SourceLocation StmtLoc; 6460 const Stmt *Body; 6461 unsigned DiagID; 6462 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 6463 StmtLoc = FS->getRParenLoc(); 6464 Body = FS->getBody(); 6465 DiagID = diag::warn_empty_for_body; 6466 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 6467 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 6468 Body = WS->getBody(); 6469 DiagID = diag::warn_empty_while_body; 6470 } else 6471 return; // Neither `for' nor `while'. 6472 6473 // The body should be a null statement. 6474 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6475 if (!NBody) 6476 return; 6477 6478 // Skip expensive checks if diagnostic is disabled. 6479 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 6480 DiagnosticsEngine::Ignored) 6481 return; 6482 6483 // Do the usual checks. 6484 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6485 return; 6486 6487 // `for(...);' and `while(...);' are popular idioms, so in order to keep 6488 // noise level low, emit diagnostics only if for/while is followed by a 6489 // CompoundStmt, e.g.: 6490 // for (int i = 0; i < n; i++); 6491 // { 6492 // a(i); 6493 // } 6494 // or if for/while is followed by a statement with more indentation 6495 // than for/while itself: 6496 // for (int i = 0; i < n; i++); 6497 // a(i); 6498 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 6499 if (!ProbableTypo) { 6500 bool BodyColInvalid; 6501 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 6502 PossibleBody->getLocStart(), 6503 &BodyColInvalid); 6504 if (BodyColInvalid) 6505 return; 6506 6507 bool StmtColInvalid; 6508 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 6509 S->getLocStart(), 6510 &StmtColInvalid); 6511 if (StmtColInvalid) 6512 return; 6513 6514 if (BodyCol > StmtCol) 6515 ProbableTypo = true; 6516 } 6517 6518 if (ProbableTypo) { 6519 Diag(NBody->getSemiLoc(), DiagID); 6520 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6521 } 6522} 6523 6524//===--- Layout compatibility ----------------------------------------------// 6525 6526namespace { 6527 6528bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2); 6529 6530/// \brief Check if two enumeration types are layout-compatible. 6531bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) { 6532 // C++11 [dcl.enum] p8: 6533 // Two enumeration types are layout-compatible if they have the same 6534 // underlying type. 6535 return ED1->isComplete() && ED2->isComplete() && 6536 C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType()); 6537} 6538 6539/// \brief Check if two fields are layout-compatible. 6540bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) { 6541 if (!isLayoutCompatible(C, Field1->getType(), Field2->getType())) 6542 return false; 6543 6544 if (Field1->isBitField() != Field2->isBitField()) 6545 return false; 6546 6547 if (Field1->isBitField()) { 6548 // Make sure that the bit-fields are the same length. 6549 unsigned Bits1 = Field1->getBitWidthValue(C); 6550 unsigned Bits2 = Field2->getBitWidthValue(C); 6551 6552 if (Bits1 != Bits2) 6553 return false; 6554 } 6555 6556 return true; 6557} 6558 6559/// \brief Check if two standard-layout structs are layout-compatible. 6560/// (C++11 [class.mem] p17) 6561bool isLayoutCompatibleStruct(ASTContext &C, 6562 RecordDecl *RD1, 6563 RecordDecl *RD2) { 6564 // If both records are C++ classes, check that base classes match. 6565 if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) { 6566 // If one of records is a CXXRecordDecl we are in C++ mode, 6567 // thus the other one is a CXXRecordDecl, too. 6568 const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2); 6569 // Check number of base classes. 6570 if (D1CXX->getNumBases() != D2CXX->getNumBases()) 6571 return false; 6572 6573 // Check the base classes. 6574 for (CXXRecordDecl::base_class_const_iterator 6575 Base1 = D1CXX->bases_begin(), 6576 BaseEnd1 = D1CXX->bases_end(), 6577 Base2 = D2CXX->bases_begin(); 6578 Base1 != BaseEnd1; 6579 ++Base1, ++Base2) { 6580 if (!isLayoutCompatible(C, Base1->getType(), Base2->getType())) 6581 return false; 6582 } 6583 } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) { 6584 // If only RD2 is a C++ class, it should have zero base classes. 6585 if (D2CXX->getNumBases() > 0) 6586 return false; 6587 } 6588 6589 // Check the fields. 6590 RecordDecl::field_iterator Field2 = RD2->field_begin(), 6591 Field2End = RD2->field_end(), 6592 Field1 = RD1->field_begin(), 6593 Field1End = RD1->field_end(); 6594 for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) { 6595 if (!isLayoutCompatible(C, *Field1, *Field2)) 6596 return false; 6597 } 6598 if (Field1 != Field1End || Field2 != Field2End) 6599 return false; 6600 6601 return true; 6602} 6603 6604/// \brief Check if two standard-layout unions are layout-compatible. 6605/// (C++11 [class.mem] p18) 6606bool isLayoutCompatibleUnion(ASTContext &C, 6607 RecordDecl *RD1, 6608 RecordDecl *RD2) { 6609 llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields; 6610 for (RecordDecl::field_iterator Field2 = RD2->field_begin(), 6611 Field2End = RD2->field_end(); 6612 Field2 != Field2End; ++Field2) { 6613 UnmatchedFields.insert(*Field2); 6614 } 6615 6616 for (RecordDecl::field_iterator Field1 = RD1->field_begin(), 6617 Field1End = RD1->field_end(); 6618 Field1 != Field1End; ++Field1) { 6619 llvm::SmallPtrSet<FieldDecl *, 8>::iterator 6620 I = UnmatchedFields.begin(), 6621 E = UnmatchedFields.end(); 6622 6623 for ( ; I != E; ++I) { 6624 if (isLayoutCompatible(C, *Field1, *I)) { 6625 bool Result = UnmatchedFields.erase(*I); 6626 (void) Result; 6627 assert(Result); 6628 break; 6629 } 6630 } 6631 if (I == E) 6632 return false; 6633 } 6634 6635 return UnmatchedFields.empty(); 6636} 6637 6638bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) { 6639 if (RD1->isUnion() != RD2->isUnion()) 6640 return false; 6641 6642 if (RD1->isUnion()) 6643 return isLayoutCompatibleUnion(C, RD1, RD2); 6644 else 6645 return isLayoutCompatibleStruct(C, RD1, RD2); 6646} 6647 6648/// \brief Check if two types are layout-compatible in C++11 sense. 6649bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) { 6650 if (T1.isNull() || T2.isNull()) 6651 return false; 6652 6653 // C++11 [basic.types] p11: 6654 // If two types T1 and T2 are the same type, then T1 and T2 are 6655 // layout-compatible types. 6656 if (C.hasSameType(T1, T2)) 6657 return true; 6658 6659 T1 = T1.getCanonicalType().getUnqualifiedType(); 6660 T2 = T2.getCanonicalType().getUnqualifiedType(); 6661 6662 const Type::TypeClass TC1 = T1->getTypeClass(); 6663 const Type::TypeClass TC2 = T2->getTypeClass(); 6664 6665 if (TC1 != TC2) 6666 return false; 6667 6668 if (TC1 == Type::Enum) { 6669 return isLayoutCompatible(C, 6670 cast<EnumType>(T1)->getDecl(), 6671 cast<EnumType>(T2)->getDecl()); 6672 } else if (TC1 == Type::Record) { 6673 if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType()) 6674 return false; 6675 6676 return isLayoutCompatible(C, 6677 cast<RecordType>(T1)->getDecl(), 6678 cast<RecordType>(T2)->getDecl()); 6679 } 6680 6681 return false; 6682} 6683} 6684 6685//===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----// 6686 6687namespace { 6688/// \brief Given a type tag expression find the type tag itself. 6689/// 6690/// \param TypeExpr Type tag expression, as it appears in user's code. 6691/// 6692/// \param VD Declaration of an identifier that appears in a type tag. 6693/// 6694/// \param MagicValue Type tag magic value. 6695bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx, 6696 const ValueDecl **VD, uint64_t *MagicValue) { 6697 while(true) { 6698 if (!TypeExpr) 6699 return false; 6700 6701 TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts(); 6702 6703 switch (TypeExpr->getStmtClass()) { 6704 case Stmt::UnaryOperatorClass: { 6705 const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr); 6706 if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) { 6707 TypeExpr = UO->getSubExpr(); 6708 continue; 6709 } 6710 return false; 6711 } 6712 6713 case Stmt::DeclRefExprClass: { 6714 const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr); 6715 *VD = DRE->getDecl(); 6716 return true; 6717 } 6718 6719 case Stmt::IntegerLiteralClass: { 6720 const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr); 6721 llvm::APInt MagicValueAPInt = IL->getValue(); 6722 if (MagicValueAPInt.getActiveBits() <= 64) { 6723 *MagicValue = MagicValueAPInt.getZExtValue(); 6724 return true; 6725 } else 6726 return false; 6727 } 6728 6729 case Stmt::BinaryConditionalOperatorClass: 6730 case Stmt::ConditionalOperatorClass: { 6731 const AbstractConditionalOperator *ACO = 6732 cast<AbstractConditionalOperator>(TypeExpr); 6733 bool Result; 6734 if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) { 6735 if (Result) 6736 TypeExpr = ACO->getTrueExpr(); 6737 else 6738 TypeExpr = ACO->getFalseExpr(); 6739 continue; 6740 } 6741 return false; 6742 } 6743 6744 case Stmt::BinaryOperatorClass: { 6745 const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr); 6746 if (BO->getOpcode() == BO_Comma) { 6747 TypeExpr = BO->getRHS(); 6748 continue; 6749 } 6750 return false; 6751 } 6752 6753 default: 6754 return false; 6755 } 6756 } 6757} 6758 6759/// \brief Retrieve the C type corresponding to type tag TypeExpr. 6760/// 6761/// \param TypeExpr Expression that specifies a type tag. 6762/// 6763/// \param MagicValues Registered magic values. 6764/// 6765/// \param FoundWrongKind Set to true if a type tag was found, but of a wrong 6766/// kind. 6767/// 6768/// \param TypeInfo Information about the corresponding C type. 6769/// 6770/// \returns true if the corresponding C type was found. 6771bool GetMatchingCType( 6772 const IdentifierInfo *ArgumentKind, 6773 const Expr *TypeExpr, const ASTContext &Ctx, 6774 const llvm::DenseMap<Sema::TypeTagMagicValue, 6775 Sema::TypeTagData> *MagicValues, 6776 bool &FoundWrongKind, 6777 Sema::TypeTagData &TypeInfo) { 6778 FoundWrongKind = false; 6779 6780 // Variable declaration that has type_tag_for_datatype attribute. 6781 const ValueDecl *VD = NULL; 6782 6783 uint64_t MagicValue; 6784 6785 if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue)) 6786 return false; 6787 6788 if (VD) { 6789 for (specific_attr_iterator<TypeTagForDatatypeAttr> 6790 I = VD->specific_attr_begin<TypeTagForDatatypeAttr>(), 6791 E = VD->specific_attr_end<TypeTagForDatatypeAttr>(); 6792 I != E; ++I) { 6793 if (I->getArgumentKind() != ArgumentKind) { 6794 FoundWrongKind = true; 6795 return false; 6796 } 6797 TypeInfo.Type = I->getMatchingCType(); 6798 TypeInfo.LayoutCompatible = I->getLayoutCompatible(); 6799 TypeInfo.MustBeNull = I->getMustBeNull(); 6800 return true; 6801 } 6802 return false; 6803 } 6804 6805 if (!MagicValues) 6806 return false; 6807 6808 llvm::DenseMap<Sema::TypeTagMagicValue, 6809 Sema::TypeTagData>::const_iterator I = 6810 MagicValues->find(std::make_pair(ArgumentKind, MagicValue)); 6811 if (I == MagicValues->end()) 6812 return false; 6813 6814 TypeInfo = I->second; 6815 return true; 6816} 6817} // unnamed namespace 6818 6819void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind, 6820 uint64_t MagicValue, QualType Type, 6821 bool LayoutCompatible, 6822 bool MustBeNull) { 6823 if (!TypeTagForDatatypeMagicValues) 6824 TypeTagForDatatypeMagicValues.reset( 6825 new llvm::DenseMap<TypeTagMagicValue, TypeTagData>); 6826 6827 TypeTagMagicValue Magic(ArgumentKind, MagicValue); 6828 (*TypeTagForDatatypeMagicValues)[Magic] = 6829 TypeTagData(Type, LayoutCompatible, MustBeNull); 6830} 6831 6832namespace { 6833bool IsSameCharType(QualType T1, QualType T2) { 6834 const BuiltinType *BT1 = T1->getAs<BuiltinType>(); 6835 if (!BT1) 6836 return false; 6837 6838 const BuiltinType *BT2 = T2->getAs<BuiltinType>(); 6839 if (!BT2) 6840 return false; 6841 6842 BuiltinType::Kind T1Kind = BT1->getKind(); 6843 BuiltinType::Kind T2Kind = BT2->getKind(); 6844 6845 return (T1Kind == BuiltinType::SChar && T2Kind == BuiltinType::Char_S) || 6846 (T1Kind == BuiltinType::UChar && T2Kind == BuiltinType::Char_U) || 6847 (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) || 6848 (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar); 6849} 6850} // unnamed namespace 6851 6852void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr, 6853 const Expr * const *ExprArgs) { 6854 const IdentifierInfo *ArgumentKind = Attr->getArgumentKind(); 6855 bool IsPointerAttr = Attr->getIsPointer(); 6856 6857 const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()]; 6858 bool FoundWrongKind; 6859 TypeTagData TypeInfo; 6860 if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context, 6861 TypeTagForDatatypeMagicValues.get(), 6862 FoundWrongKind, TypeInfo)) { 6863 if (FoundWrongKind) 6864 Diag(TypeTagExpr->getExprLoc(), 6865 diag::warn_type_tag_for_datatype_wrong_kind) 6866 << TypeTagExpr->getSourceRange(); 6867 return; 6868 } 6869 6870 const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()]; 6871 if (IsPointerAttr) { 6872 // Skip implicit cast of pointer to `void *' (as a function argument). 6873 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr)) 6874 if (ICE->getType()->isVoidPointerType() && 6875 ICE->getCastKind() == CK_BitCast) 6876 ArgumentExpr = ICE->getSubExpr(); 6877 } 6878 QualType ArgumentType = ArgumentExpr->getType(); 6879 6880 // Passing a `void*' pointer shouldn't trigger a warning. 6881 if (IsPointerAttr && ArgumentType->isVoidPointerType()) 6882 return; 6883 6884 if (TypeInfo.MustBeNull) { 6885 // Type tag with matching void type requires a null pointer. 6886 if (!ArgumentExpr->isNullPointerConstant(Context, 6887 Expr::NPC_ValueDependentIsNotNull)) { 6888 Diag(ArgumentExpr->getExprLoc(), 6889 diag::warn_type_safety_null_pointer_required) 6890 << ArgumentKind->getName() 6891 << ArgumentExpr->getSourceRange() 6892 << TypeTagExpr->getSourceRange(); 6893 } 6894 return; 6895 } 6896 6897 QualType RequiredType = TypeInfo.Type; 6898 if (IsPointerAttr) 6899 RequiredType = Context.getPointerType(RequiredType); 6900 6901 bool mismatch = false; 6902 if (!TypeInfo.LayoutCompatible) { 6903 mismatch = !Context.hasSameType(ArgumentType, RequiredType); 6904 6905 // C++11 [basic.fundamental] p1: 6906 // Plain char, signed char, and unsigned char are three distinct types. 6907 // 6908 // But we treat plain `char' as equivalent to `signed char' or `unsigned 6909 // char' depending on the current char signedness mode. 6910 if (mismatch) 6911 if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(), 6912 RequiredType->getPointeeType())) || 6913 (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType))) 6914 mismatch = false; 6915 } else 6916 if (IsPointerAttr) 6917 mismatch = !isLayoutCompatible(Context, 6918 ArgumentType->getPointeeType(), 6919 RequiredType->getPointeeType()); 6920 else 6921 mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType); 6922 6923 if (mismatch) 6924 Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch) 6925 << ArgumentType << ArgumentKind->getName() 6926 << TypeInfo.LayoutCompatible << RequiredType 6927 << ArgumentExpr->getSourceRange() 6928 << TypeTagExpr->getSourceRange(); 6929} 6930